Isotype choice for chimeric antibodies affects binding properties.

Construction of a series of chimeric antibodies (murine variable region and human constant region) derived from the murine antibody BIRR1, which recognizes intercellular adhesion molecule 1 (ICAM-1), has revealed differences in the relative binding abilities of the chimeric antibody to antigen. The chimeric antibodies show a ranking of their ability to compete with BIRR1 for antigen on the surface of cells with the order BIRR1 = cIgG1 (100%) > cIgG4 (30%) > cIgG2 (10%) as demonstrated by solid-phase competitive enzyme-linked immunosorbent assay. Papain digestion yielded Fab fragments that were purified to homogeneity. Competitive enzyme-linked immunosorbent assay showed that the chimeric and murine Fab binding constants were equivalent. A solution-phase binding assay (analyzed by size exclusion high performance liquid chromatography) between the intact mAbs and recombinant soluble ICAM-1 further established that the binding constants involving the Fab arms of the two antibodies were equivalent. In summary, the murine and chimeric anti-ICAM-1 antibodies bind cellular ICAM-1 with equivalent affinities but with differing avidities.

There is an increasing interest in the use of monoclonal antibodies in the diagnosis and treatment of human diseases.
Many of these antibodies are of murine origin. As a consequence, human anti-murine antibody responses have been observed when these antibodies were used (LoBuglio et al., 1989;Reynolds et al., 1989) (reviewed by Dillman (1990)) and questions have arisen as to the dose and number of treatments that can be given. Alternative mAb treatment schemes have been proposed and tested. These include the use of conventional immunosuppressants at subtoxic low doses to reduce the ability of the immune system to mount a response to the xenogenic antibody (Chatenoud et al., 1986); switching to a second antibody of a different isotype and idiotype recognizing the same antigen (Chatenoud, 1986;Jonker and Den Brok, 1987); inducing non-reactivity in the patient to the incoming antibody by pre-or co-administration of an anti-CD4 antibody to ablate the T, response (Benjamin et al., 1986;Mathieson et al., 1990;Jin et al., 1991); inducing classical tolerance by the administration of small doses of non-aggregated antibody or alternatively injecting very large doses of antibody, which appears to also re-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
5 To whom correspondence should be addressed: Research and Development Center, Boehringer Ingelheim Pharmaceuticals, Inc., 900 Ridgebury Rd., P. 0. Box 368, Ridgefield, CT 06877-0368. Fax: 203-791-6468. sult in a reduction in the immune response (Sears et al., 1984;Khazaeli et al., 1988); or co-administering interferony with the antibody, which appears, at least in one study (Blottiere et al., 1991), to delay the immune response to the murine mAb and to restrict the response, when it does arise, to the isotype.
One approach to reducing the human anti-murine antibody response involves replacing as much as possible of the nonhuman sequence with equivalent human sequence (Steplewski et al., 1988;LoBuglio et al., 1989). "Humanized" antibodies have been shown to be considerably less immunogenic than the original mouse antibody (reviewed by Adair (1992)). Humanization has been achieved in two ways. In the first instance, the constant regions of a particular mAb can be humanized by replacing the DNA sequences from the mouse constant heavy and light chain genes with suitable sequences from human antibody genes. The resultant gene is a chimera with mouse variable and human constant regions (Morrison et al., 1984). The aim of such a modification is to maintain the binding affinity of the murine mAb (since the murine variable domain is left intact and can fold independently) while minimizing the potential immunological problems. As an alternative, "CDR' grafting" involves replacing the constant region and the framework residues that position the CDRs in the variable region of the murine antibody by their human equivalents (Jones et al., 1986). CDR grafting creates a molecule that contains the minimum amount of murine-encoded material while retaining antigen binding specificity. In some cases, however, this approach has the potential disadvantage of associated loss in antibody affinity (see, e.g., Riechmann et al. (1988); reviewed by Adair (1992)).
We have prepared chimeric derivatives of the murine antiintercellular adhesion molecule 1 (ICA"1) mAb BIRRl (Smith et Cosimi et al., 1990) using human I&,, I@,, and IgG4 constant regions. I C A " 1 is a member of the supergene family expressed on a variety of cell types Rothlein et al., 1986;Marlin and Springer, 1987;Staunton et al., 1988Staunton et al., ,1990Wawryk et al., 1989) and has been shown to be a ligand for the neutrophil-endothelial cell receptors LFA-1 (CDlldCD18) and Mac-1 (CDllWCD18) Dustin et ai., 1986;Marlin and Springer, 1987;Smith et al., 1988;Diamond et al., 1990). BIRRl has been shown to have beneficial effects in non human primates with renal allografts (Cosimi et al., 1990). In this study, the antigen binding ability of three different chimeric isotypes was examined. The chigion; ICAM-1, intercellular adhesion molecule 1; SICAM-1, soluble The abbreviations used are: CDR, complementarity-determining re-ICA"1; mAb, monoclonal antibody; PBS, phosphate-buffered saline; PAGE, polyacrylamide gel electrophoresis; HPLC, high performance liquid chromatography; BSA, bovine serum albumin; ELISA, enzymelinked immunosorbent assay. meric cIgG, was able to compete equivalently with BIRR1 for binding to antigen on cells, confirming that the binding properties of the parental mAb was preserved in the formation of the chimeric antibody. Unexpectedly, however, the cIgG, and cIgG4 antibodies showed decreased competitive abilities. We show in this paper that, while the binding site affinity of cIgG, is the same as BIRR1, the avidity of the antibody is markedly reduced. This observation may have wider consequences for the design of chimeric and humanized antibodies to cell surface antigens.

MATERIALS AND METHODS
Reagents-Biotin-N-hydroxysuccinimide, streptavidin-p-galactosidase conjugate, and p-nitrophenyl-p-u-galactopyranoside were purchased from Life Technologies, Inc. A bicinchoninic acid protein determination kit (BCA-1) was purchased from Sigma. An ImmunoPure Fab preparation kit was purchased from Pierce Chemical Co. All other reagents were of the highest grade available and were obtained from Fisher or Sigma.
Chimeric Gene Construction and Dansfection-Basic molecular biology procedures were as described by Sambrook et al. (1989). COS cell expression was as described by Whittle et al. (1987). Chinese hamster ovary transfections and cell culture were performed as described by Bebbington (1991).
Full-length cDNA was prepared by using oligo(dT) to prime first strand synthesis from mRNA prepared using the guanidiniumLiC1 extraction procedure. After methylation and ligation to EcoRI linkers, the cDNA library was cloned into Escherichia coli (E. coli) in pSP64. E. coli colonies containing either heavy or light chain genes were identified by screening using oligonucleotide 5'-TCCAGATGTTMCTGCTCAC for the light chain, which is complementary to a sequence in the mouse K constant (Ck) region, and by using a 980-base pair BamHI-EcoRI restriction fragment of a previously isolated mouse IgG' ,, constant region clone for the heavy chain. DNA sequences for the 5'-untranslated regions, signal, sequences and variable regions of full-length cDNAs were obtained.
The chimeric light chain sequence was assembled from three fragments: a 397-base pair EcoRI-SfaNI fragment coding for the 5"untranslated region from the cDNA, the signal sequence, and the majority of the light chain variable region; an oligonucleotide adapter, which codes for the remainder of the 3' region of the variable region from the SfaNI site and the 5' residues of the human constant region up to and including a unique NarI site that had been previously engineered into the human Ck gene at the third to fiRh codons so as not to alter the coding potential'; and the human Ck gene as an NarI-EcoRI fragment. The ligated light chain gene was inserted into pEE6-hCMV-neo (Stephens and Cockett, 1989) to give pAL7. Chimeric IgGz and IgG, genes were assembled from three fragments: a 424-base pair EcoRI-Bum1 fragment coding for the 5'-untranslated region from the cDNA, the signal sequence and the majority of the heavy variable region sequence; an oligonucleotide adapter, which codes for the remainder of the 3' region of the variable region from the Bum1 site up to and including a unique HindIII site that had been previously engineered into the first two amino acids of the constant region (Whittle et al., 1987); and the IgGz or IgG, constant regions as HindIII-BamHI fragments. The ligated heavy chain genes were inserted into pEE6-hCMV-gpt to give pAL8 and pAL9, respectively. The chimeric IgG, heavy chain gene was assembled by excising the DNA coding for the heavy chain signal and variable region sequences along with the first 5 amino acids of the CH1 domain from pAL9 as a HindIII-ApaI fragment and inserting the sequence 5' to the IgG, heavy chain constant region to the HindIII and ApaI sites in pE1001, an EE6 hCMV-gpt vector previously modified to contain the IgG, constant region 3' to the hCMV promoter, and to remove the ApaI site in the gpt gene: to give pJA200. The first 5 amino acids of the CHI domains of IgG, and IgG, are identical, so this cloning procedure does not introduce any novel sequence motifs.
Clones were isolated after transformation into E. coli, and the linker and junction sequences for all of the constructions were confirmed by DNA sequencing. Stable cell lines were prepared by transfecting pAL7 into CHO-K1 cells by the calcium phosphate precipitation procedure (reviewed by Bebbington (1991)). A neomycin-resistant cell line secreting light chain was selected and re-transfected with pJA200, pAL8, or ' N. Whittle, unpublished data. J. S. Emtage, unpublished data. pAL9. Cell lines producing 3.8, 6.9, and 4.2 pg/106 cell424 h of cIgG,, cIgG,, and cIgG,, respectively, were used for antibody production.
Antibody Purification-Hybridoma cell line R6.5 D6 was generated as previously described (Smith et al., 1988). BIRRl (IgG,,) mAb was produced by culturing the hybridoma in antibiotic-free Dulbecco's modified Eagle's medium supplemented with glutamine and 5% heat-inactivated fetal calf serum. Cell lines expressing chimeric antibodies were cultured in serum-free medium; in each case, the antibody in the culture supernatant was purified by affinity chromatography using protein A-Sepharose (Colcher et al., 1989) and concentrated by ultra-filtration. The pH was adjusted to 6.5 and the protein dialyzed into phosphatebuffered saline (PBS). A small amount of dimerlaggregate was removed by size exclusion chromatography using a Superdex-200 prep grade column (Pharmacia LKB Biotechnology Inc.) eluting with 64 m M sodium phosphate, 86 m NaC1, pH 6, buffer. Purity and correct assembly of the antibody was tested by reducing and non-reducing SDS-polyacrylamide (SDS-PAGE) as described by Laemmli (1970), and high performance liquid chromatography (HPLC) gel filtration. Identity was confirmed by N-terminal amino acid sequencing for the light chain and amino acid composition analysis. Quantitative amino acid analysis was also used to determine accurate extinction coefficients at 280 nm for each of the chimeric antibodies, allowing accurate protein determinations to be made easily in purified preparations.
Preparation of BIRR1 and cIgG, Fab Fragments-The fragments were generated according to the standard protocol accompanying the Pierce ImmunoPure Fab preparation kit with only slight modifications.
This method involved overnight incubation at 38 "C with cysteine-activated immobilized papain for fragment cleavage, followed by protein A chromatography for purification. mAb and Fab buffer solutions were changed and protein concentrated using a Centricon-30 ultrafiltration device (Amicon, Beverly, MA). Final samples were exhaustively dialyzed against 64 m~ sodium phosphate, 86 m~ NaCI, pH 6, using Spectro/por tubing (M, 6000-8000 cut-off) (Spectrum Medical Industries, Los Angeles, CA) and filtered using a 0.2-pm, low protein binding, sterile filter (Gelman Sciences, Ann Arbor, MI). Protein concentrations were determined using a bicinchoninic acid protein assay (BCA-1, Sigma procedure TPRO-562).
Competitive ELZSA Binding Studies-Wells of microtiter plates were coated with ICAM-1-bearing B lymphoblastoid cell line JY by the method of Noms et al. (1991). All additions to the wells, titering of samples, and determinations of optical density were done with a Biomek 1000 Automated Laboratory Workstation (Beckman, Palo Alto, CA) controlled by a Hewlett-Packard Vectra ES/12 personal computer (Hewlett-Packard Co., Cupertino, CA). All plates were blocked prior to use with a 2% BSA, 0.1% sodium azide, PBS, pH 7.2 solution (BSA-BS). The microtiter plates were washed between steps with a 2 m M MgCl,, 5 m sodium phosphate, 3 m~ potassium phosphate, 140 m NaCI, 0.1% sodium azide, pH 7.0 solution using an Ultrawash2 Microplate Washer (Dynatech Laboratories, Chantilly, VA). Biotinylated BIRRl mAb (prepared by the method of Goding (1980)) was combined with native mAb in BSA-BS so that the final molar concentration ratios were (10-8/10"), (10-7/10-7), (10-7/10-8), (lO"/lO-'), (10-7/10-10), and (10-7/0), respectively; for Fab fragments, (5 x 10-'/5 x (5 x 10-'/4 x lo"?, and (10-7/0), respectively. Each sample was then titered by a factor of 1:3 with BSA-BS so that 5 logs in concentration could be analyzed. Six 100-pl aliquots of the titered samples were then transferred into the wells of the microtiter plates for each mAb concentration and incubated overnight at 4 "C. The following day (-20 h) the wells were washed with BSA-BS and incubated with 100 pl of a 1:500 dilution of streptavidin-p-galactosidase conjugate in BSA-BS. Plates were agitated for approximately 3 h and then washed with PBS, pH 7.2. 100 pl of a 0.5 mg/ml p-nitrophenyl-/3-u-galactopyranoside, PBS solution, pH 7.2, was then added to each well and incubated with agitation for about 2 h. The absorbance of the enzymatic product p-nitrophenol at 405 nm was then determined for each well. HPLC-IA Binding Studies-ICAM-1 is available in soluble form (SICAM) (Marlin, 1990) and was provided by Dr. Steve Marlin and Dr. Richard Shansky (Boehringer Ingelheim Pharmaceuticals, Inc.). Binding between the mAbs and SICAM-1 was characterized by size exclusion chromatography (HPLC immunoassay, HPLC-U), monitoring protein at 220 nm. BIRRl and cIgG, samples were prepared by combining SICAM-1 with the respective mAb in PBS, pH 7.0, at the same concentration (lod M). The association reaction was allowed to proceed a t room temperature for 4 h. The samples were then titered with PBS, pH 7.0, a t a ratio of 2:3 down to a concentration of 1.7 x lo-' M. The dissociation reaction was allowed to proceed for 2 days at 4 "C. Concentrations of bound and unbound antibody were determined by peak area and peak height analyses of HPLC elution profiles.
Datu Analysis-ELISA and HPLC-IA data files containing measure- (w/v) acrylamide gel. Samples were applied in the same order to each gel. Lane I, 4 pg of chimeric B72.3 IgG, marker (Kinget al., 1991); lunes 2 and 10, molecular size marker mixture comprising phosphorylase 6 (94 kDa), bovine serum albumin (67 kDa), ovalbumin (43 kDa), carbonic anhydrase (30 kDa), and soybean trypsin inhibitor (20 m a ) ; lunes 3-9, chimeric IgG, a t loading of 8, 4, 2, 1, 0.5, 0.25, and 0.125 pg, respectively; lunes 11-17, chimeric IgG, a t loadings of 8.4.2, 1, 0.5,0.25, and 0.125 pg, respectively. Gels were run at 30 mA and stained with Coomassie Blue. Biotinylated BIRR1 was then added a t a fixed concentration and after further incubation binding of biotinylated antibody was revealed using streptavidin-p-galactosidase and p-nitrophenyl-p-D-galactopyranoside. In each panel inhibitory ability of the chimeric antibodies is compared to that of BIRRl measured in the same experiment. The x axis shows concentration of these unlabeled antibodies. 100 40 ments of optical density were analyzed using the version 6.08 SAS statistical software system (SAS Institute Inc., Cary, NC) on an IBM 4381 mainframe computer running W C M S (IBM Corp., Armonk, NY). Data analysis was performed by applying nonlinear regression techniques to selected models using the Marquardt iterative method. Parameters were estimated with 95% confidence intervals and residual analysis conducted. SAS/Graph was used to display the data overlaid with the theoretical curve/surface.

RESULTS
Protein Characterizations-Reducing SDS-PAGE of all of the chimeric antibodies showed a single band of the expected heavy and light chains at approximately 55 and 28 kDa, respectively. Under non-reducing conditions, fully assembled antibody was seen in each case. The cIgG, mAb showed an additional minor band (l0-15%) of monomeric cIgG, at about 80 kDa, similar to that seen with all other chimeric and natural IgG, antibodies (Angal et al., 1993) (Fig. 1). The biochemical explanation for the double band observed in the chimeric IgGz preparation is unknown (Fig. 1). The integrity of the cIgG, molecule was assessed by DNA sequence analysis of the assembled genes to confirm that the double banding seen in the non-reducing SDS-PAGE was not due to sequence heterogeneity of the cloned gene, by partial N-terminal amino acid sequencing of the light chain and by total amino acid composition analysis. Using size exclusion HPLC, all antibody preparations were observed to be a single species with less than 1% aggregate at 280 nm. Protein stability for all of the chimeric antibodies was measured by storage at 6 mg/ml at 4 "C or -70 "C for up to 1 month followed by SDS-PAGE and HPLC analysis and antigen binding assay. There was no evidence for alteration of structure or activity of the chimeric antibodies by such analyses (data not shown), suggesting that the material as produced by the cells was in a stable form.
BIRRl and cIgG, Fab fragments prepared by papain digestion were observed to be a single species by size exclusion HPLC a t 280 nm. Analysis by non-reducing SDS-PAGE showed a band a t 50 kDa (>go%) with a minor band a t 25 kDa. Reducing SDS-PAGE gels displayed one band at 25 kDa with slight microheterogeneity in each preparation. This heterogeneity is probably due to some digestion above the disulfide bonds yielding fragments (510%) that dissociate upon incubation with SDS as exhibited in the non-reducing analysis (not shown). The two Fab preparations were indistinguishable using these analytical techniques.
Characterization of d b Binding by ELZSA-Each of the chimeric antibodies reacted as strongly as the murine antibody with a panel of anti-idiotype monoclonal antibodies raised against BIRRl (Rothlein et al., 1993), while nonspecific human immunoglobulins of the appropriate class did not react. These data suggested that the antigen binding site from the murine antibody has been transferred to the chimeric antibodies. Competitive binding assays, however, indicated that while the cIgG, antibody competed equally for antigen with the murine antibody ( Fig. 2A ), the cIgG, and cIgG, did not compete as well (Fig.  2, B and C), with the cIgG, and cIgG, showing approximately 40 and 10% relative potency, respectively.
In an attempt to further differentiate and quantitate the differences in antigen binding among the different antibodies, comprehensive data sets were generated using a JY cell ELISA assay (see "Appendix" for the development of mathematical models describing antibodylantigen binding). The absorbance curve for biotinylated BIRRl a t low mAb concentrations displays simple binding behavior. This segment of the data is fit nicely by the one-site model with A,,, and Amin being well defined (Fig. 3, curve a ) . Since the curve breaks in the lo-' M range (KRlo' "I), the antibody is presumed to be binding in a cooperative bivalent manner and this site is defined as being minimally composed of two ICAM-1s. At higher mAb concentrations, however, absorption increases and the one-site model gives a poor fit due primarily to the uncertainty in A,,, (Fig. 3,   curve b). Consequently, a two-site model was developed to allow for tight binding at low concentrations (site 1: cooperative bivalent, and monovalent a t high mAb concentrations) and for weaker binding at higher concentrations (site 2: monovalent only). The two-site model results in a good fit for the data at low and high concentrations (Fig. 3, curve c). It should be noted that although KRI is determined with excellent confidence limits, data at very high antibody concentrations are required for obtaining a good fit for KR2. Since the BIRRl and chimeric mAbs are expected to display both monovalent and bivalent binding with fixed JY cells, this two-site model was adopted for analysis of the mAb competitive ELISA data.
The competitive binding data taken for BIRRl are shown in Fig. 4. This experimental design results in statistically significant estimates for Amin (low concentration data), p (intermedi- ate concentrations wrapping around the face of the surface), and A, , , (high concentration data). In addition, the reporter mAb is evaluated in each experiment and serves as an indicator for the degree of competition with which the sample mAbs were challenged. The binding behavior along the reporter mAb axis was similar to the data observed for biotinylated BIRRl in Fig.  3. The simulated fit projected a surface that conformed closely with the data. The fitted values for BIRRl and cIgG, given in Table I indicated that the bivalent binding displayed by the chimeric IgG4 was about 30% that of the murine antibody, while the monovalent binding was statistically equivalent. KR1 for the biotinylated BIRRl reporter was (8.6 2 0.6) x loR M-' and (1.05 2 0.08) x lo9 M" in the BIRRl and C I S , experiments, respectively, and indicated that the binding constants for BIRRl and cIgG, were determined under comparable competitive conditions. Characterization of Fab Binding by ELISA-To address whether the loss of binding could be attributed to affinity differences, Fab fragments were prepared for the BIRRl and cIgG, antibodies. Competitive J Y cell ELISA experiments were carried out with the Fab fragments to measure each antibody's affinity. Biotinylated BIRRl intact mAb was used in each experiment as the reporter. The data were fit to response surfaces (not shown) defined by a two-site model (see "Appendix") and resulted in comparable values for each fragment (see Table I). These data suggest that the binding site affinity has not been perturbed for the CIS,, but that the avidity of the antibody has been altered as disclosed by the binding of the whole antibody to cell surface antigen. The binding constant HPLC immunoassay experiment was carried out by combining mAb and SICAM at a ratio of 1:l and titering the reactions from M to as low as 2 x lo-' M. The model used to fit the data (not shown) allowed for mAb-antigen complex ratios of 1:l and 1:2.

DISCUSSION
This paper describes the production of IgG,, I&,, and IgG, human-mouse chimeric forms of the mouse monoclonal anti-ICAM-1 antibody BIRRl, some of the physical properties of these chimeric molecules, and a comparison of their antigen binding ability with that of the native mouse antibody.
With regard to the physical nature of the chimeric antibodies, the cIgG, has, in common with normal human IgG, and all other reported chimeric mouse-human and fully humanized IgG, antibodies analyzed to date, a proportion of the molecules in which it is believed the hinge disulfide bonds do not form, leading to a tetrameric (bivalent) antibody that is non-covalently linked between the heavy chains (Angal et al. (1993), and references therein). This feature of human IgG, has not been associated with marked difference in binding avidity in other situations when chimeric or humanized isotypes have been compared (Shaw et al., 1988;Colcher et al., 1989;Hardman et al., 1989;Hutzell et al., 1991;Shearman et al., 1991)., The significance of the double-banded appearance of the cIgG, is not clear. The SDS-PAGE data suggest that there are two variants of the molecule which appear be stable forms of the antibody. Natural IgG, and other mouse-human chimeric and fully humanized IgG, antibodies have not previously shown this phenomenon (Bruggemann et al., 1988); and in the case of the cIgG, we cannot exclude the possibility that this feature of the antibody contributes to the effect on binding avidity. Integrity of the mouse variable region in the chimeric molecules has been confirmed by DNA sequence analysis of the assembled genes, by DNA sequence of the constant regions for the IgG, molecule to confirm the double banding seen in the non-reducing SDS-PAGE was not due to sequence heterogeneity of the cloned gene, by partial N-terminal amino acid sequencing of the light chain, and by total amino acid composition analysis for all of the chimeric antibody proteins. Furthermore, these chimeric molecules react with a set of monoclonal anti-idiotypic antibodies that were raised against the parent murine anti-ICA"1. The data suggest that the conformation of the mouse variable region in the chimeric molecules remains intact.
Since time-resolved fluorescence depolarization studies of antigen-antibody complexes have shown that the flexibility of the hinge region for these human constant regions also decreases in the order IgG, > I&, > IgG, (Oi et al, 1984;Dangl et al., 1988;Schneider et al., 1988;Tan et al., 1990), it was postulated that the decreased ability to bind cellular ICA"1 was reflective of restricted Fab arm movement resulting from decreased hinge region flexibility. In order to test this hypothesis, we have compared the binding abilities of BIRRl and cIgG, with cellbound, as well as solubilized, ICA"1 in an attempt to differentiate affinity associated with the mouse variable region from avidity, which additionally depends upon a mouse versus a human hinge region. ICAM-1 is an up-regulated cell surface marker. On live cells the density of antigen will increase from a situation where monovalent binding predominates to one where there will be a combination of monovalent and bivalent occupancy of the antigen binding sites by the antibody, depending on dosage and pharmacokinetics. Since regulation of ICA"1 is a dynamic phenomena, it would be difficult to analyze the type and degree of binding with living cells in vivo or in vitro. With fured JY cells, however, the constitutively expressed I C A " 1 density and distribution are fured such that monovalent and bivalent interactions can be observed separately as a function of antibody concentration. Cooperative bivalent binding, for example, is a strong antigenlantibody interaction usually observed at low antibody concentrations, i.e. lo-* to lo-" M. Since this type of interaction depends upon the ability of the Fab arms to assume a particular spatial arrangement dictated by proximal ICAM-1 molecules, bivalent binding is descriptive of antibody avidity. Monovalent binding, on the other hand, is a much weaker interaction since only one of the Fab arms is involved with the binding site at any given time. Thus, mAb monovalent interactions are descriptive of binding affinity and should be observed in a concentration range expected for Fab fragments, i.e.
to lo-' M. In contrast to bivalent binding, monovalent interactions are for the most part independent of hinge region flexibility. Since BIRRl and cIgG, have the same mouse variable regions, a comparison of their respective binding curves should reveal equivalent binding affinities at the higher mAb concentrations even though their binding avidities at the lower mAb concentrations are very different.
Binding response surfaces were generated in all of the competitive ELISA experiments (e.g. Fig. 4). A comparison of the binding constants indicates that while the K,, value for cIgG, is only 30% that for the murine IgG,,, the K,, values descriptive of monovalent binding are statistically equivalent. In addition, the constants obtained for the Fab ELISA data also argue for equivalent monovalent binding behavior. For a visual comparison of mAb and Fab binding, representative slices of the raw data from their respective response surfaces at a fured reporter concentration are shown in Fig. 5. While the cIgG, mAb displays a decrease in cooperative bivalent binding ability relative to the murine analog, the Fab isotherms are nearly identical (see also Table I). These data suggest that although the affinities of the two mAbs are equivalent, their avidities are markedly different.
In summary, the constants obtained in the competitive ELISA experiments are consistent with monovalent and cooperative bivalent binding for the mAbs and monovalent binding for the Fab fragments (Mason and Williams, 1986). Since a reporter is not required for the mAb solution assay, the monovalent and non-cooperative bivalent binding observed in the HPLC immunoassay is a direct measurement of the mAb' s affinity. Thus, the mAb and Fab competitive ELISA (indirect) and the mAb HPLC-JA (direct) independently demonstrate equal affinities between the two antibodies. These results establish that BIRRl and cIgG, bind I C A " 1 with equivalent affinities but with inequivalent avidities.
A recent publication suggests that the subtype of the IgG contributes the the avidity of an antibody to its multimeric antigen (Horgan, 1993). This finding is consistent with what we report here. There are two possible explanations for the differential binding results for the chimeric IgGs presented. First, the constant regions of the molecules could be affecting assay results by interacting with Fc receptor on the J Y cell. We think ' that this is unlikely, since we have been unable to detect any differences in the binding via the Fc of human IgG,, I&,, or IgG, to JY cells5 and FcR1, the only Fc receptor to bind monomeric antibody and not normally found in human B-cell-derived lines (reviewed by Burton and Woof (1992)). The second explanation, and the one that we favor, is that the antigen binding results reflect differences in avidity imposed on the molecules by the different constant region structures. For bivalent binding to occur, flexibility is required in the hinge and switch regions for rotation and elbow bending of the Fab arms (Valentine and Green, 1967;Romans et al., 1977;Jackson et al., 1983;Nezlin, 1990;Schumaker et al., 1991;Borrebaeck, 1992). Since constant regions have been shown to differ in the segmental flexibility of their hinge regions, the potential exists for antibodies with the same binding site but different constant regions to have differing abilities for spatially binding the antibody to the antigen site. In particular, the reported hinge flexibility of the human IgGs correlates with the retention of binding activity noted here for the chimeric antibodies, in that the IgG, with the most flexible hinge has the best binding avidity, while the IgG2 with the least flexibility has the lowest binding avidity.
In conclusion, the engineering of chimeric antibodies by exchanging human constant regions for mouse constant regions not only alters the effector functions of the antibody, but can also alter its antigen binding properties. Thus, while the ability or inability of an IgG isotype to mediate biological events may be paramount in determining therapeutic efficacy, the ability to bind cell surface antigens must also be considered in the design and characterization of chimeric antibodies. M. K. Robinson, unpublished results.

Acknowledgments-We thank Rowena Reedman and Alan Lyons for ham Roberts and Carol GoRon for construction of cell linea, and Karen
providing expression vectors containing the chimeric antibodies, Gra-Proudfoot and Janet Deistung for protein purification. We also thank Tobin Cammett for the biotinylation procedure of BIRR1, Rod Deleon for the JY cell culture, Karen Laszlo Hrabcsak for the preparation of JY cell ELISA microtiter plates, and Paul McGoff for assistance with the chimeric purification and HPLC experiments. We also thank Tapon Roy for advice concerning the statistical treatment of the data.

APPENDIX
Since a cell-based ELISA involves solidlliquid phase reactions, the binding kinetics cannot be strictly characterized using equilibrium relationships for solutions. For example, onrates will be slower as a result of diffusion-limited mass transport of the antibody in solution to the antigen on the cell surface; in addition, off-rates will also be slower due to the relatively high concentrations of the antigen at the cell surface (Nygren and Stenberg (19891, and references therein). Nevertheless, it is precisely this cell surface antibody/antigen interaction that is of interest in considering the in vivo therapeutic aspects of these antibodies. Consequently, we have developed binding models based on equilibrium expressions (Pincus and Rendell, 1981;Mason and Williams, 1980) and we note that K is an apparent binding constant. The models used herein postulate that antigen sites are uniformly distributed on the cell surface and that the binding of one antibody does not affect the binding of another antibody.
Reporter mAb /Direct Binding One-site Model-In the simplest case, binding is between an antibody and a population of identical binding sites. The equilibrium equation for a one-site model is where R is the reporter antibody, G is a free antigen, RG is the antibody-antigen complex, and KR is the apparent binding constant. An association equilibrium expression may be written for the complex formation as where [GI, is the total concentration of antigen binding sites, OR, is the fraction of sites complexed with an antibody and 8, is the fraction of sites free. Noting that OR, + 8, = 1 and solving for OR,, we obtain Equation 3. may be rewritten to relate absorbance to the fraction of bound antigen sites as defined by equilibrium relationships. Substituting for OR,, we obtain Equation 5.
In this paper, R is biotinylated BIRR1, A is the experimentally determined absorbance of the enzymatic product p-nitrophenol at 405 nm, A,, is the maximum absorbance determined under saturating conditions, and Amin is the minimum or background absorbance due to nonspecific binding of streptavidin-P-D-galactosidase with the plate wells. Equation 5 is sigmoid in nature with A,,, and Amin being the fitted upper and lower asymptotes, respectively. lho-site Model-The high concentration data in Fig. 3 can be fit to a model that is characterized by two sites with different binding affinities. 1) Site 1 is considered to be made up of at least two I C A " 1 antigens and is capable of both cooperative bivalent binding (one mAb/site; KRl) and monovalent (one or two mAbs/site; KR2). 2) Site 2 consists of a single I C A " 1 molecule and displays only monovalent binding (one mAb/site; KR2). The equilibrium binding isotherm for a two-site model is described by Equation 6, where the fractions of total binding determinants for site 1 and site 2 are given by p and (1p), respectively. p = eC-, + OR,_, + eR2,_, and (1 -I*) = oc-2 + eRG-, (Eq. 6 ) ( E a 7)

Substitution of Equation 7 into Equation 4
gives the expression used for fitting the mAb direct binding ELISA.

Reporter + Sample mAb /Competitive Binding: lloo-site Model
The equations developed for competitive binding are derived in a manner similar to direct binding. The competition of reporter and sample mAbs for site 1 (two 1CA"ls) will result in one of six possibilities: 1) a free site (no binding), 2) the bivalent binding of R, 3) monovalent R, 4) bivalent sample S, 5) monovalent s, and 6) the monovalent R and s. For site 2 (one ICAM-1) the possibilities are: 1) free site, 2) monovalent R, and 3) monovalent S. The mass balance equations for reactions involving a reporter and a sample antibody are: (1 -= ec-2 + e,,_, + e,,_, where the sample antibody is designated by S. Substituting the appropriate equilibrium expressions into Equation 8 and rearranging, the fraction of antigenic sites bound by biotinylated BIRR1 is where Kl and K, are the apparent binding constants for bivalent and monovalent interactions, respectively. Substitution of Equation 9 into Equation 4 gives the expression used for fitting the mAb competitive ELISA data.

Iluo-site Model
In this ELISA, an intact antibody (bivalent) was used as the reporter and an Fab (monovalent) evaluated as the sample. Since there should be no difference in affinity between the Fab and ICA"1 at site 1 or 2, K,, and X,, were fit as one binding constant. The following equations were derived for a bivalent site 1 and a monovalent site 2 in a manner similar to competitive mAb/mAb ELISA (see above).
KR and K, are the apparent binding constants for the biotinylated mAb and sample Fab, respectively. Substitution of Equation 11 into Equation 4 gives the expression used for fitting the competitive Fab ELISA data. As in the other ELISAs discussed above, the affinity constants for the Fab fragments derived from this competition assay should be viewed as apparent binding constants.