Mode of Binding of Anti-P-glycoprotein Antibody MRK-16 to Its Antigen

Monoclonal antibody MRK-16 recognizes a discontinuous extracellular epitope on the multidrug resistance-associated ATP-binding cassette transporter, P-glycoprotein. The atomic basis for specificity of this antibody is of interest because of its potential as a modulator of P-glycoprotein activity. The crystal structure of Fab MRK-16 is reported to a resolution of 2.8 Å. A structure for a portion of the epitope was derived by comparison to regions of solved structures with similar primary sequence. This has permitted a proposal for the mode of binding of the peptide epitope to the antibody, in which the peptide makes specific contacts with complementarity-determining regions H1, H2, and H3 from the heavy chain and L3 from the light chain. These interactions are consistent with epitope mapping studies and with the observation that MRK-16 is specific for human class I P-glycoprotein. This result identifies side chains in MRK-16 that would be amenable to alteration in antibody engineering experiments to derive improved multidrug resistance inhibitors for clinical use during chemotherapy. In particular, Arg-H97 contacts both Glu-746 and Asp-744 of the peptide, Arg-L96 contacts Asp-743, and Thr-H33 interacts with Thr-747. All of these epitope residues were implicated in mediating specificity by epitope mapping studies.

A highly conserved membrane protein known as P-glycoprotein is overexpressed in many multidrug-resistant cancer cell lines (1). The detection of P-glycoprotein in cells refractory to chemotherapy suggests that its presence may be important in the development of multidrug resistance during tumor progression (2). P-glycoprotein has been proposed to mediate multidrug resistance by deriving energy from ATP and transporting drugs out of the cell (3,4).
Enhanced expression of P-glycoprotein can be found in tumors arising both from tissues that normally express high levels of P-glycoprotein and those that express low levels. Tissues that normally express low levels seem to develop the capacity to express higher levels either during the development of tumors or subsequent to their exposure to chemotherapeutic drugs. In the clinic it has been observed that, in some instances, the success of chemotherapeutic treatment correlates with the level of P-glycoprotein expression; tumors expressing high levels of P-glycoprotein respond poorly to chemotherapy, whereas tumors with low expression respond favorably (5)(6)(7). P-glycoprotein is a member of the ATP-binding cassette family of membrane-integral transporters (8). Based on its amino acid sequence, P-glycoprotein is expected to consist of two structurally similar halves, each of which has six putative transmembrane helical segments (9), although there is some evidence for other conformations (10). The transmembrane regions are connected by relatively short loops. Extracellular loop I connects transmembrane segments 1 and 2, extracellular loop II connects 3 and 4, and loop III connects 5 and 6, etc. Thus, loops I and IV connect structurally related transmembrane segments, the first two of each half of the molecule, on the extracellular side of the membrane. Transmembrane segments 6 and 12 are each followed by a cytoplasmic ATP-hydrolyzing domains of about 350 residues in length, characterized by an ATP-binding cassette motif. P-glycoprotein is found in three isoforms in rodents and two in human, termed class I and III by sequence comparison to the rodent forms. Class I P-glycoprotein is associated with the multidrug resistance phenotype, and class III is not (11).
Many questions about P-glycoprotein remain unanswered. Neither its natural role in the cell nor the mechanism by which it transports a wide variety of molecules out of the cell is well understood. A three-dimensional structure of the entire molecule, or even of structural domains, would be valuable in addressing many of the unanswered questions. Monoclonal antibodies raised against various domains of P-glycoprotein may prove to be useful tools in biophysical studies of portions of P-glycoprotein, in particular those regions of the molecule that may be sensitive to conformational changes. This paper discusses the three-dimensional structure of the antigen-binding fragment (Fab) of antibody MRK-16 and proposes a basis for recognition of part of its epitope. Several monoclonal antibodies have been raised against P-glycoprotein (12,13). MRK-16 is of particular interest since it has been shown to interfere with the function of P-glycoprotein (14). The finding that MRK-16 treatment selectively inhibits the growth of multidrug-resistant cells raises the possibility that immunotherapy could be used to treat multidrug-resistant human cancer cells (15,16).
MRK-16 is a mouse IgG2a() monoclonal antibody generated against adriamycin-resistant human myelogenous leukemia K-562 cells expressing P-glycoprotein (14). The antibody binds to a discontinuous epitope on the extracellular side of P-glycoprotein (extracellular loops I and IV) (12). MRK-16 has been shown to recognize only the human class I isoform of P-glyco-protein and hence can be used as a specific marker for human multidrug resistance-associated P-glycoprotein (12). Table I summarizes the sequence comparison of regions corresponding to the MRK-16 epitopes across species. Substitution of key amino acids in different isoforms of P-glycoprotein, within as well as across species, is observed. For example, residues Asp-110 from loop I and Arg-741, Glu-746, Thr-747, and Arg-749 from loop IV in human class I isoforms are substituted or deleted (Glu-746 deletion, Thr-747, and Arg-749 substituted by Val and Gln, respectively) in the class III human P-glycoprotein isoform. Other changes in these positions are found in the equivalent epitope sequence of the other three classes of hamster and mouse P-glycoprotein. Thus the loop IV sequence confers the specificity of MRK-16 to class I P-glycoprotein, and the structure of the antibody should shed light on the defining interactions.

EXPERIMENTAL PROCEDURES
The preparation and crystallization of Fab MRK-16 has been reported (17). X-ray diffraction data were measured on a twin multi-wire system (San Diego Multiwire Systems) mounted on a Rigaku RU200 generator operating at 40 kV, 150 mA. The unit cell of the crystal, space group P2 1 , a ϭ 54.49 Å, b ϭ 67.77 Å, c ϭ 117.18 Å, ␣ ϭ ␥ ϭ 90°, ␤ ϭ 97.61°, indicates the presence of 2 Fab molecules in the asymmetric unit. A summary of the data collection is presented in Tables II and III. Graphical inspection and interactive model building was carried out using the program O on a Silicon Graphics workstation (18). The structure of the complex was refined with the simulated annealing protocol implemented in X-plor (19,20) using the parameters of Engh and Huber (21). The x-ray amplitude weight was calculated using the check protocol implemented in X-plor; the weight applied was one-third of the calculated amplitude weight.
Molecular replacement calculations were performed with data in the resolution range 10 -4 Å with X-plor. As a result of trials with a number of Fab models from the Brookhaven Protein Data Base (22), the best solution was obtained with the variable and constant domains of the Fab from the IgA, McPC603 (Protein Data Bank code 1MCP) (23). The elbow angle between the pseudo-2-fold axes of the constant and variable domains was rotated by 10°from the value in the starting model.
The initial rotation was refined by Patterson correlation analysis. The constant and variable portions of each of the heavy and light chains were treated as independent bodies. The x and z translations were determined independently for each of the variable and constant domains. The relative y translation for each constant domain with respect to the corresponding variable domain was found by one-dimensional translation searches, and finally, the entire second Fab was positioned relative to the first by the same procedure.
The overall orientation of the structure of the complex that was assembled from molecular replacement was subjected to a rigid body minimization in eight rigid groups using X-plor with data in the 8 to 3-Å resolution range. The packing of the structure was inspected visually. No steric clashes were observed that could not be corrected by minor structural adjustments. The R factor after this stage was 42%, at which point the MRK-16 amino acid sequence was applied to the model. Data to the limit of 2.8 Å were incorporated into the subsequent refinement.
Eight cycles of interactive model building followed by refinement were carried out before arriving at the current refined model. For the first four cycles, the model was subjected to 160 cycles of Powell mini-mization followed by simulated annealing with from 3000 to 300 K with a time step of 0.005 s and ending with 40 cycles of positional refinement. The final four cycles consisted of minimization at room temperature only. Restrained individual B-factor refinement was carried out before computing electron density maps for interpretation.
The initial model was built into F o Ϫ F c "omit" maps from which 10% of the coordinates were omitted at a time from the phase calculation. The refinement process was monitored by the use of both the conventional R factor and R free (24). Simulated annealing omit maps were used as a guide throughout the process of refinement. Non-crystallographic symmetry restraints were imposed at one stage, resulting in a map of improved quality and phases. However, the non-crystallographic symmetry broke down locally near the elbow regions and in CDR 1 loops L1, H2, and framework region L-FR3. The overall r.m.s. deviation between the two molecules is 0.2 Å, but locally the r.m.s. deviations varied between 1 and 2 Å near the elbow regions (residues 111-115) in both the heavy and the light chains, up to 0.5-1.0 Å in the loops and up to 3 Å in the framework regions. The final coordinates and structure factor data have been deposited in the Brookhaven Protein Data Base (22).
Several program packages were used in analyzing the various structural aspects. Stereochemistry was checked with program PROCHECK (25). The molecular surface area was calculated with program MS (26) using a 1.7-Å probe and extended van der Waals radii to enable comparisons with other antibodies.
Atomic structures deposited in the Protein Data Base were searched for stretches of primary sequence closely similar to the loop IV epitope, 2 in order to deduce likely secondary structural motifs for this peptide. As all occurrences of this sequence took on essentially the same secondary structure (see "Results" and Table IV for details), a model of the loop IV peptide was obtained from the consensus by extracting the corresponding loop from the cytochrome P-450 model (residues Ala-242-Ala-246) (27) and regularizing the stereochemistry with program O. The likely resulting structure was compared visually to the MRK-16-binding site with the programs O and SETOR (28). This procedure was intended to obtain a qualitative evaluation of the overall complementarity of the combining site and the epitope and to identify possible interacting pairs of residues in the respective structures. The assumptions involved in forming the peptide structure and the speculative nature of the docking exercise did not justify the application of any quantitative optimization of the complex model. Table III summarizes the results of the refinement process. Those residues with the highest B factors generally reside  within the exposed loop regions and occasionally adopt unfavorable main chain torsional angles. A representation of the entire structure is shown in Fig. 1. Two copies of the asymmetric unit, each of which contains two Fab molecules, are shown to display the crystal packing (discussed below). The geometry of the main chain dihedral angles is presented as a plot in Fig. 2 (29). 83% of the residues of the complex are clustered in the energetically favored regions of the plot. However, three residues, light chain (L)-51 from complementarity-determining region (CDR)-L2 of both the molecules, and L-67 of molecule 1, are in the dissallowed regions, although L-67 does not have well defined electron density (antibody numbering scheme of Kabat et al. (30) is used throughout this paper). Residues Leu-198 -Leu-202 have been deleted from both the molecules in the final model due to poor electron density.

RESULTS
Based on a pair-wise comparison of the two crystallographi-cally independent Fab molecules, only small changes in the relative disposition of the variable light and variable heavy chain domains are seen. The variable light domain of the two molecules was superimposed, and then rotation and translation required to superimpose the variable heavy domains of the two molecules were calculated using program ALIGN (23). The rotation for variable light-variable heavy is 168°for molecule 1 and 170°for molecule 2 and that for the constant light and constant heavy chain 1 domain interface is 173°and 174°for molecules 1 and 2, respectively. However, the elbow angles, i.e.  L-CDR1, although somewhat flexible, seems to be most closely related to type 4. For comparison, the backbone dihedral angles for L-CDR1 in both molecules and those of the standard type 4 are shown in Table V. The variation may be due to the modest resolution of this structure or to crystal contacts (see below). The heavy chain CDR3 loops have recently been analyzed in some detail (33,34). The MRK-16 H-CDR3 is of moderate length (n ϭ 10) with an extended base and additional bulge conformation as illustrated by Fig. 3b in Shirai et al. (33). Here, a ␤-bulge at the C-terminal base of the CDR links Trp-H103 with the backbone at residue H101 (Fig. 4a). The primary sequence of the H-CDR3 of MRK-16 is quite distinctive (residues H92-H103: CARYYRYEAWFASW), particularly by the charged residues Arg-H97 and Glu-H99. The C-terminal end bears some resemblance to that of the anti-single strand DNA antibody BV04-01 (35), which is the same length.
Several interesting points have emerged from examining the crystal contacts. The orientations of the two molecules in the asymmetric unit differ by a rotation of only 13°. Thus, in combination with crystallographic symmetry, two different modes of packing are observed as follows: head-to-head packing, where the variable domain of a symmetry-related molecule 2 packs against the variable domain of molecule 1, and headto-tail packing, where the constant domain (tail) of molecule 2 packs against the variable domain (head) of molecule 1 (see Fig.  1). A detailed analysis of the various crystal contacts carried out using program CRISPACK (36) shows that residues from the variable domain of molecule 2 make several van der Waals contacts with the residues in the variable domain of molecule 1: residues L28 and L30 of CDR L1 of molecule 1 make about 11 van der Waals contacts and two hydrogen bond interactions with residues H55, H56, and H57 and framework residue L81 of molecule 2.
A final, unusual feature in this structure is the visualization of the constant region loop H129 -H141 (Fig. 4b). This segment is rarely ordered in other Fab crystal structures (37) and has been implicated as a mediator of flexibility between segments of the antibody molecule (38). Here, the loop is stabilized by crystal contacts to a neighboring variable region framework region and by an inter-chain disulfide bond between Cys-H128 and the C-terminal Cys-L214.
Epitope Peptide Conformation and Interaction-The Protein Data Bank (22) was searched for segments of structure with primary sequences similar to the essential portion of the loop IV peptide (Asp-Pro-Glu-Thr-Lys). Six occurrences found in four distinct folds in the search had essentially the same type 1 ␤-turn structure (Table IV). A search with a longer stretch, Arg-Ile-Asp-Asp-Pro-Glu-Thr-Lys-Arg, identified a very similar secondary structure in residues 200 -208 of the lysine-, arginine-, ornithine-binding protein (2LAO, Arg-Lys-Asp-Asp-Thr-Glu-Leu-Lys-Ala, (39)). In addition, the secondary structure prediction routine BTURNPRED (40,41) strongly suggested a nonspecific ␤-turn for the region Ile-Asp-Asp-Pro and for the subsequent stretch, Asp-Pro-Glu-Thr. This motif was used to construct a likely model for the core of the loop IV peptide: residues 743-747 (Asp-Asp-Pro-Glu-Thr), the segment most strongly implicated in binding to MRK-16.
The MRK-16 antigen-binding site consists of a non-polar cleft with two polar pockets (Fig. 3a). These pockets, with positive electrostatic potential provided by Arg residues H97 and L96, are positioned a good distance apart (11 Å) to receive the Glu and Asp residues of the peptide (side chain separation 10 Å in the type 1 ␤-turn model), whereas a hydrophobic patch between them can accommodate the Thr and Pro side chains of the epitope. With the modification of the torsion angles of the side chain of Arg-H97, it could be positioned to contact both Asp-744 and Glu-746 of the peptide, whereas Asp-743 was proximate to Arg-L96, without altering the standard ␤-turn conformation of the peptide or any other portion of the binding site. This proposed mode of binding is illustrated in Fig. 3b.
In this binding mode, the buried surface area can be estimated to be at least 300 Å 2 for both the peptide and the Fab. About 80% of the latter derives from heavy chain residues. This is much smaller than in other Fab-peptide complexes, typically 450 -600 Å 2 . This low value may be either due to the incomplete modeling of the peptide residues or to the absence of the portion of the discontinuous epitope corresponding to loop I of P-glycoprotein.
In addition to the principal interactions described above, there is the potential for further comparatively minor interactions with an epitope peptide extending in the C-terminal direction with the H2 CDR, residues Asn-H56 and Tyr-H58. Extension in the N-terminal direction of the epitope is less clear but may contact the L3 CDR, residues His-L93 and Ser-L92. The loop IV epitope in the proposed conformation sits along one side of the cleft contacting largely the heavy chain CDRs (detailed below). The comparatively small buried surface area predicted by the model, along with the involvement of  Many of the essential amino acids in the epitope sequence of MRK-16 are charged residues, suggesting that electrostatic interactions play a major role in binding. In the complex proposed here involving the core of the loop IV epitope (DDPET, residues 743-747), the key residue occupying the central por-tion of the peptide, Glu-746, would make a salt bridge with Arg-H97 and could make contacts with Thr-H33 and Ser-H52A, as well as extensive van der Waals interactions with Tyr-H95 (Fig. 3b). Thr-747 probably makes hydrogen bond contacts with one or more of Thr-H33, Ser-H52, and Ser-H52A, as well as van der Waals interactions with heavy chain CDR2. On the Nterminal side of Glu-746, Pro-745 sits in a pocket surrounded by Tyr-L32, Trp-H100A, and Tyr-H95. Asp-744, in turn, can interact with the side chains of Arg-H97 and, possibly, Tyr-H98. An appealing aspect to this conformation is that Asp-744, another residue implicated to be important in binding, is nicely positioned to interact with Arg-L96. Comparison of Two Molecules in the Asymmetric Unit-Much of the difference between the two crystallographically independent molecules is in the flexibility of the CDR loops L1 and H3. Residues L27E to L30 in CDR L1 and H95 and H96 in CDR H3 are poorly defined in molecule 2. As it is these residues from the heavy chain that are proposed to make contacts with the peptide in the liganded molecule, this difference may reflect an intrinsic flexibility within the site that could play a role in epitope binding. There are a few significant differences in the framework regions (L-FR3) between the two molecules, especially residues L66 to L69, which has well defined densities in both molecules. However, the pair-wise root mean square deviations between the two molecules in this region is 3 Å. Residue L66 is a key determinant of the conformation of the L1 CDR. Thus, this may also reflect a degree of flexibility in the binding site.
Comparison with Other Fab-Peptide Complexes-The data base of solved structures of Fabs consists of a large number of unliganded Fabs and a much smaller number complexed with a variety of antigens that includes small haptens, carbohydrates, proteins, DNA, and peptides. These structures have established our understanding of epitopes and shape complementarity of the binding pockets, as well as the contributions of charge-charge interactions, hydrogen bonds, and van der Waals contacts to the overall specificity of the antigen-antibody interactions (42)(43)(44).
Anti-peptide antibodies are of particular interest since short peptide fragments can be used to elicit antibodies that frequently recognize the folded proteins from which the peptide sequence was derived. Structures of several anti-peptide antibodies, Fab 17/9 (1HIM) ( (50), and Fab TE33 (1TET) (51), have been reported in complexed form. In the complexes, the peptides have been shown to adopt either a type I (as in 17/9 and MN12H2) proposed here for the loop IV peptide or a type II ␤-turn (as in B13I2 and TE33). The extensive participation of heavy chain CDRs, especially H2 and H3, in binding is commonly observed. DISCUSSION The primary interest in this study is the mode of binding of the epitope structure to the MRK-16-binding site. The model for this interaction is based on the following information: the structure of the binding site derived from the crystallographic result; the conformation of the epitope peptide from the data base analysis; previous epitope mapping data identifying important residues in the epitope responsible for antigen recognition (12); and comparison of the sequences of the epitope region in a number of P-glycoproteins, some of which bind and others that do not bind MRK-16. The results of a model of the interaction should be consistent with all these data. To assess the complementarity of the combining site with the major epitope peptide, an analysis of the likely structure of the peptide was necessary. The resulting proposal for a type I ␤-turn motif is strongly supported by the consistent identification of this secondary structure motif among stretches of similar primary sequence in other crystal structures, and by the common occurrence of this motif in the epitopes in peptideantibody complexes. This peptide structure shows a high degree of electrostatic complementarity with the binding site of MRK-16, permitting the identification of a number of potential specific contacts.
Epitope mapping (12) has identified residues Asp-743, Asp-744, Glu-746, and Thr-747, in particular, to be involved in specific contacts with MRK-16, consistent with the specificity of the antibody for human class I P-glycoprotein. The model proposed here would predict that the major determinants of the specificity for this part of the epitope are residues Arg-H97, Arg-L96, and Thr-H33. An alteration of Glu-746 to Lys, as seen in hamster class II Pgp (Table I), would clearly disrupt the interaction with Arg-H97, as would the change of Asp-744 to Pro. The deletion of residue 746, as is seen in several examples in Table I, would either disrupt the secondary structure of the loop, on which the fitting is crucially dependent, or position a hydrophobic side chain in the vicinity of Arg-H97. The proposed interaction of Asp-743 with Arg-L96 would explain the lack of tolerance of an alteration of position 743 to Thr or Gly. The substitution of Thr-747 to Val in Table I would have the effect of breaking interactions with Thr-H33 or the Ser residues. Therefore, the binding proposal presented here is consistent with the epitope mapping studies and with the important residues implicated from sequence comparison in mediating MRK-16 specificity.
This study sheds light on the mode of action of MRK-16, an antibody shown to be a modulator of P-glycoprotein-induced multidrug resistance (15,16). By identifying residues in the antibody that are likely to be responsible for epitope binding, this structure could provide a useful basis for the design of more effective, clinically useful P-glycoprotein inhibitors. Moreover, antibody-based molecules have proven to be valuable tools in the crystallographic study of difficult molecules, notably the membrane-associated cytochrome c oxidase (52,53). Thus, MRK-16, or variants based on the results described here, will be useful in the structural analysis of the P-glycoprotein molecule itself.