- Email: [email protected]
The Crystal Structure of a Fab Fragment to the Melanoma-Associated GD2 Ganglioside
JOURNAL OF STRUCTURAL BIOLOGY ARTICLE NO. SB973857
119, 6–16 (1997)
The Crystal Structure of a Fab Fragment to the Melanoma-Associated GD2 Ganglioside Susan L. Pichla, Ramachandran Murali,1 and Roger M. Burnett2 The Wistar Institute, 3601 Spruce Street, Philadelphia, Pennsylvania 19104 Received July 31, 1996, and in revised form February 20, 1997
cell signaling or cell-to-substrate interactions that regulate cell growth. Normally, melanocytes, which are pigment-forming cells, express GM3 as the most prominent ganglioside. As the transformation to a cancerous cell commences in melanoma, the distribution and type of gangliosides on the cell surface start to change. This alteration is due to the triggering of glycosyltransferases that add additional sugars to GM3 to form gangliosides such as GD2, 9-O acetylated GD3, GM2, and GD3 (Ravindranath et al., 1989). As these gangliosides are highly expressed on malignant melanoma cells (Thurin et al., 1987), their appearance on the cell surface signals the progression of the disease. The GD2 and GD3 gangliosides are being used as target antigens for immunotherapy as an alternative to other treatments, such as radiation therapy, that have not been very effective in treating melanoma in its later stages (Herlyn and Koprowski, 1988; Tai et al., 1985; Thurin et al., 1987; Iliopoulos et al., 1989). Monoclonal antibodies can be used in immunotherapy against melanoma to identify and eliminate micrometastases and so prevent further spread of the disease (Herlyn and Koprowski, 1988). The ME36.1 mouse monoclonal antibody (IgG2a) recognizes the GD2 ganglioside and, with lower affinity, the GD3 ganglioside (Fig. 1). ME36.1 can inhibit tumor growth and metastatic spread of melanoma in nude mice (Iliopoulos et al., 1989). Clinical trials with the antibody in 13 patients with melanoma gave encouraging results, and 1 patient had complete remission for over a year (Iliopoulos et al., 1989). Structural studies on ME36.1 were undertaken to understand the specific interactions between the antibody and the therapeutically targeted GD2 ganglioside. Moreover, it has been recognized that rodent antibodies such as the ME36.1 mouse antibody, when used as an immunotherapeutic, elicit an immune response that results in the rapid clearance of the antibody from the human patient (Mayworth and Quinta´ns, 1990; Waldmann, 1991). Thus, stud-
The GD2 ganglioside is a cell-surface component that appears on the surface of metastatic melanoma cells and is a marker for the progression of the disease. The ME36.1 monoclonal antibody binds to the GD2 ganglioside and has shown potential as a therapeutic antibody. ME36.1 is a possible alternative therapy to radiation, which is often ineffective in late-stage melanoma. The crystal structure of the Fab fragment of ME36.1 has been determined using molecular replacement and refined to an R factor of 20.4% at 2.8 Å resolution. The model has good geometry with root-mean-square deviations of 0.008 Å from ideal bond lengths and 1.7° from ideal bond angles. The crystal structure of the ME36.1 Fab shows that its complementarity determining region forms a groove-shaped binding site rather than the pocket-type observed in other sugar binding Fabs. Molecular modeling has placed a four-residue sugar, representative of GD2, in the antigen binding site. The GD2 sugar moiety is stabilized by a network of hydrogen bonds that define the specificity of ME36.1 toward its antigen. r 1997 Academic Press INTRODUCTION
Cell-surface carbohydrates are important as recognition elements as well as markers for disease, and much effort has been devoted to understanding the specificity and affinity of proteins for carbohydrates. Carbohydrates have great inherent flexibility, which allows them to adopt a wide range of stereoisomers (Vyas, 1991; Quiocho, 1986). Gangliosides are a particular class of cell-surface glycolipids that are distinguished from other glycolipids by the presence of sialic acid residues and that are expressed on most cells. Although the function of gangliosides is not clearly known, they appear to be involved in cell-to1 Current address: Department of Pharmacology, The Medical School of the University of Pennsylvania, Philadelphia, PA 19104. 2 To whom correspondence should be addressed. Fax: (215)-8983868. E-mail: [email protected].
1047-8477/97 $25.00 Copyright r 1997 by Academic Press All rights of reproduction in any form reserved.
6
7
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE
FIG. 1. Chemical formulas for the GD2 and GD3 gangliosides. Both consist of carbohydrate chains forming a head group, which is linked to a ceramide group (Cer) that lies in the outer cell membrane. The only difference between the two glycolipids is a branched N-acetylgalactosamine (GalNAc) in the GD2 sugar.
ies are under way to ‘‘humanize’’ the ME36.1 antibody by grafting the mouse complementarity determining region (CDR) loops, which define the antigen binding site, onto a human antibody framework. This potentially increases the effectiveness of the antibody in immunotherapy by making it less immunogenic. However, attempts to humanize the antibody have so far decreased its binding affinity (Dr. G. A. Luckenbach, personal communication). Since the framework residues surrounding the CDR loops are important for the correct folding of these loops (Chothia et al., 1989), a crystal structure for ME36.1 Fab should reveal which framework residues are important for loop conformation. These residues could then be altered in the variable region of the humanized antibody so that the correct conformation of the CDR graft is maintained and the affinity of the antibody for its antigen is not reduced. Crystals of the native ME36.1 Fab were obtained, a complete diffraction data set to 2.8 Å resolution was collected, and the structure was solved. A model of the GD2 carbohydrate (Fig. 2) then was used to investigate the types of interactions that could occur between the side chain residues in the binding site of the Fab and the sugar. The results suggest a mode of binding that is consistent with the types of interactions seen in other protein–carbohydrate structures. MATERIALS AND METHODS Data Collection The purification and crystallization of the ME36.1 Fab fragment and attempts at cocrystallization of the Fab with the extracellular portion of GD2 have been described previously (Pichla et al., 1995). The modified form of GD2 was produced by enzymatically removing its ceramide group to produce a soluble antigen. Crystals of the Fab were grown in 30% PEG 4000, 1% MPD, and Tris buffer at pH 7.5. The crystals belong to the monoclinic space group P21, with unit cell dimensions a 5 37.6 Å, b 5 94.1 Å, c 5 67.4 Å, and b 5 101.0°. Diffraction data sets were collected from six native crystals on a Siemens-Nicolet X100-A
FIG. 2. Schematic of the four-residue sugar used in the modeling to represent GD2. The unusual a2–8 linkage between the two sialic acid (NeuNAc) residues occurs between the C2 of one sugar and the C8 of the hydroxyl in the other. The galactose (Gal), which is attached to the ceramide group in GD2 (Fig. 1), links the second sialic acid residue and the N-acetylgalactosamine (GalNAc).
multiwire area detector. The X-ray source was a Rigaku RU200 rotating-anode generator, equipped with a double mirror focusing system (Supper) and operated at 40 kV and 65 mA. The crystals diffract to 2.5 Å resolution but, as the intensity falls off sharply after 2.8 Å, data collection was limited to the lower resolution. The data collection statistics are presented in Table I. Structure Determination The structure of Fab ME36.1 was solved by molecular replacement using the X-PLOR program package (Bru¨nger, 1992a) to carry out the rotational and translational searches. The Fab portion of the galactan binding J539 Fab structure (Suh et al., 1986) was used as a search model to determine the orientation and position of the ME36.1 Fab in the unknown crystal environment. The J539 Fab was chosen as the search model due to its high sequence similarity to ME36.1 in the heavy (VH) and light chain (VL) variable regions, with 42% homology for VH and 74% homology for VL. Both Fab molecules have a k constant region in
TABLE I ME36.1 Data Collection Statistics as a Function of Resolutiona Resolution limit (Å)
Number of observed reflections
Number of unique reflections
Completeness (%)
R-symb (%)
8.00 5.00 4.50 4.00 3.40 3.00 2.80 Total
1181 2505 2111 3144 6075 6330 4012 33 212
462 841 733 1137 2417 2827 2021 14 677
90.4 98.6 98.0 97.8 98.3 97.6 95.4 96.1
8.6 9.7 9.2 9.6 10.5 12.1 14.7 10.4
a Six data sets, collected on a Siemens-Nicolet X100-A area detector and reduced and merged with XDS (Kabsch, 1988). b R-sym 5 o 0 I 2 ,I . 0 /oI. I, observed intensity and ,I. the average intensity of multiple measurements of symmetry related reflections.
8
PICHLA, MURALI, AND BURNETT
the light chain (CL). The heavy chain constant region (CH1) of J539 belongs to a class different from that of ME36.1. A real-space rotation was performed with X-PLOR using the variable and constant domains of the model separately. The whole Fab also was used, taking advantage of the option in the program to vary the ‘‘elbow angle’’ between the variable and the constant domains. The J539 search model was placed in an orthorhombic unit cell with dimensions 120 3 120 3 150 Å, so that the cell was twice the size of the Fab molecule in each direction. Structure factors were calculated for the model, and the Patterson map corresponding to the crystal was calculated by fast Fourier transformation on a 1.0-Å grid. The radius of integration for the rotation function was 45 Å. The orientations of the search model were sampled using Lattman (1972) angles: u2 5 0–720°, u2 5 0–90°, u1 5 0–360° at a sampling interval of 2.5°. The highest 1000 peaks from the Patterson search at 6.0–3.0 Å (I $ 2s) were listed and sorted according to peak height. Those rotations that were within 10° from one another were clustered and the condensed set of orientations was used to provide starting points for Patterson correlation refinement (PC refinement) (Bru¨nger, 1990). The model orientations that gave the highest correlation value after PC refinement were similar for the constant and variable domains and for the whole Fab molecule (Table II). In all three cases, the highest correlation value obtained was well above that for other possible orientations. The translation search was also performed with each of the three oriented and PC-refined molecules. The origin along the twofold axis in space group P21 is arbitrary; therefore, the translation search was limited to the x–z plane. Sampling intervals of 0.125 Å in x and z were used in the grid search, which was restricted to the asymmetric unit (0 # x # 1.0, 0 # z # 0.5). The highest values obtained in the translation searches for the constant and variable regions and for the whole Fab were 7.6, 9.3, and 9.5 s above background, respectively. These maxima correspond to similar translations for all three search models (Table II). The final correlation value for the whole Fab was 0.367, with an R factor of 44% [R factor 5 (ohkl 0 Fo 2 kFc 0 )/ohkl 0 Fo 0 , where Fo and Fc are the observed and calculated structure factors, respectively]. The correctly oriented and translated whole J539 Fab was used as the basis for further refinement. The first step was to perform molecular dynamics refinement with a molecular model based on the J539 Fab, against the ME36.1 diffraction data using reflections from 6.0 to 3.0 Å resolution (7884 reflections with Fo $ 2.0 s). Since the sequences for the J539 and ME36.1 were so similar, the side chain residues of the variable (VL and VH) domains of J539 were not removed during the first round of refinement. As the CH1 domains belong to different classes, their side chains were changed to polyalanine residues to avoid any bias. The CDR loops were also removed in
TABLE II Molecular Replacement Solution for ME36.1 Search model searcha
Rotation VL–VH CL–CH Whole Fab
Correlation coefficient u1 u2 u3 231.4° 39.6° 74.6° 237.8° 34.0° 75.1° 233.3° 37.0° 76.7°
Translation searchb Dx 0.075 VL–VH 0.137 CH–CL Whole Fab 0.150 a b
Dy 0.0 0.0 0.0
Dz 0.313 0.263 0.275
Rotations are given in Eulerian angles. Translations are given in fractional coordinates.
0.137 0.114 0.227 0.221 0.205 0.367
the initial model. At this early stage of refinement, the isotropic temperature factors were set to a constant value of 19 Å2. The model was prepared for molecular dynamics by first minimizing the total energy, while removing all inappropriate contacts, by subjecting the model to one round of positional refinement of the individual atoms with 140 steps of Powell conjugate gradient minimization (Bru¨nger et al., 1987, 1990). The R factor for the model dropped from its initial value of 44 to 28%. The model was then further refined by simulated annealing (SA) using a slow-cooling protocol. This refinement step was accomplished by first raising the model to a starting temperature of 3000 K, cooling it to 300 K in 50 integration steps of 0.025 psec, and performing a further round of positional refinement. The R factor after SA refinement was 24%. After the initial refinement, 2Fo–Fc difference maps were calculated with the restricted 6.0- to 3.0-Å data (Fo $ 2s). The previously omitted residues within the CDR loops, and the side chains of the variable (VL and VH) and CH1 regions, were manually built into the map using the ME36.1 sequence only where the electron density was clearly interpretable. This model was improved by alternating cycles of refinement and manual rebuilding using the program FRODO (Jones, 1982) on an Evans and Sutherland PS350 vector graphics display. Isotropic temperature factors (B factors) were kept constant for the first rounds of manual rebuilding and refinement until most of the residues had been placed in the model, after which they were refined by alternating rounds of minimization and restrained B-factor refinement to an R factor of 20.9%. At this stage, the resolution limits were extended to include data from 8.0 to 2.8 Å resolution, and 10 355 reflections (Fo $ 2s) were used for several more rounds of refinement. A test set, consisting of 10% of the observed data (1035 reflections), then was extracted from the reflection file for the calculation of R free (Bru¨nger, 1992b). Any bias was removed by one round of simulated annealing using the remaining reflections. At this stage, the R factor was 19.0%, and R free was 33.1%. Refinement was then alternated with model-building in 2Fo–Fc difference maps calculated using 20.0- to 2.8-Å data (Fo $ 2s). The Engh and Huber (1991) topology and parameter files also were introduced. This resulted in an improvement in the stereochemistry of the model, with a substantial decrease in R free (31.0%) and an increase in the R factor (21.1%). Assignment of the water molecules in the Fab structure was not attempted until the Fab model was well-refined to an R factor of 21%, with good stereochemistry indicated by the rms deviations of bond lengths and angles and by the Ramachandran plot. The water molecules were located by using the program PEAKMAX in CCP4 (1994) to select those electron-dense peaks that were .3s above background in an Fo–Fc difference map. The program WATPEAK (CCP4, 1994) then selected only those that were within 3.6 Å from an atom in the Fab molecule, giving a total of 99 water molecules. These candidates were examined graphically and those that did not have good density in 2Fo–Fc difference maps (at least 2s), along with hydrogen bond interactions with side chain or main chain atoms, were deleted to leave 45 waters. The model then was subjected to another round of refinement with the values of occupancy and B factor for the water molecules set to unity and 15 Å2. Examination of the model after refinement showed that most of the water molecules maintained both good density and their hydrogen bond interactions. The exceptions were deleted, leaving 34 water molecules in the structure. The elbow angle for the ME36.1 Fab was determined using the program O (Jones et al., 1991) to find the rotation matrix for the least-squares superposition of the heavy chain variable domain atoms onto those of the light chain variable domain, and similarly for the constant domains, and then calculating the angle between the two rotation axes.
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE Modeling of a Sugar in the ME36.1 Antigen Binding Site A four-residue model sugar (Fig. 2), representative of GD2 (Fig. 1), was built using Quanta (1989) on a Silicon Graphics workstation. The sugar contains two sialic acid residues (NeuNAc), galactose (Gal), and an N-acetylgalactosamine (GalNAc) residue. The sugar was minimized to a local energy minimum by conjugate gradient minimization adapted for CHARMM (Brooks et al., 1987) to obtain a sterically correct starting model with which the ME36.1 binding site could be explored. The starting model for the sugar was adjusted to conform to the antigen binding site by manually placing it in several positions in the middle of the site and evaluating the fit. The conformation of the sugar was then adjusted to maximize favorable interactions, while avoiding molecular overlaps. One position that created extensive hydrogen bond interactions between the sugar and the side chain residues in the binding site was found. The position and molecular interactions were optimized further by several cycles of conjugate gradient minimization without imposing constraints on either the sugar or the Fab molecules. RESULTS
The ME36.1 Fab structure was determined by molecular replacement and refined to 2.8 Å resolution. The refinement statistics are shown in Table III. The final R factor of the model is 20.4% for 10 355 reflections (Fo $ 2s) in the resolution range 8.0–2.8 Å. The final R free value is 32.6%. The corresponding values for R factor and R free using the complete diffraction data without sigma cutoff (14 677 reflections) were 21.1 and 32.8%. The root-mean-square deviation from ideality in bond lengths is 0.008 Å and for bond angles is 1.7°. The average B factors for each domain are VL 22.9 Å2, VH 24.7 Å2, CL 24.4 Å2, and CH 25.1 Å2. The electron densities for the entire light chain and the variable region of the heavy chain are of good quality. The density for the heavy chain constant region is poorer, so that two loop regions in the CH1 domain (H129–H135, H158– H161) could not be fitted to the electron density and are presumed to be highly disordered. Disorder in TABLE III Summary of Refinement of ME36.1 Fab at 2.8 Å Resolution Resolution range R factor (%)a R free (%)b rms bond lengthc rms bond anglec rms improper anglesc rms dihedral anglesc Temperature factors (Å2) Main chain Side chain
8.0–2.8 Å 20.4 32.6 0.008 Å 1.7° 1.4° 28.0° 23.6 24.0
a R factor 5 (o hkl 0 Fo 2 kFc 0 ) /ohkl 0 Fo 0 , where Fo and Fc are the observed and calculated structure factors, respectively. b R free is the R factor for a test set of reflections, comprising 10% of the data, that was omitted from refinement. c Deviations from ideal stereochemistry using parameters of Engh and Huber (1991).
9
these regions has been observed for other Fab structures (Stanfield et al., 1990; Pokkuluri et al., 1994). Apart from these gaps, all but five amino acid side chains (Ser H136, Leu H139, Ser H157, Val H170, Lys H209) could be placed in density. Most of these residues are solvent exposed or occur at the beginning or the end of a gap. A Ramachandran plot (Ramachandran and Sasisekharan, 1968) of the (f, c) dihedral angles (not shown), prepared using the program PROCHECK (Laskowski et al., 1993), indicated that 99% of the 411 residues (213 in the light chain, 198 in the heavy chain) fall in allowed regions. Three of the five residues in the disallowed region are in the loop regions (Thr L50, Asn H54, Asn H156) and 2 lie at the end of the heavy chain (His H200, Pro H201). The ME36.1 Fab structure (Fig. 3) has the immunoglobulin fold common to all Fab structures. The value of 149.6° for the elbow angle between the variable and constant domains falls within the range for other Fab structures (Suh et al., 1986). The antigen binding site is formed by the six CDR loops H1, H2, H3, L1, L2, and L3 (Fig. 3). The antigen binding site is approximately 20 Å long (perpendicular to the VH:VL interface), 10 Å wide, and 8 Å deep. The dimensions are consistent with a groove-type binding site (Cisar et al., 1975) (Fig. 4a), as opposed to the pocket-like binding site seen in the sugar binding Salmonella O-antigen binding Fab (Cygler et al., 1991) (Fig. 4b). The L3 and H3 loops are in the middle of the binding site and form the bottom of the groove. The binding site contains eight aromatic residues, four phenylalanines (Phe L95, Phe L97, Phe H64, Phe H101), and four tyrosines (Tyr L93, Tyr H32, Tyr H60, Tyr H95). Phe L95, Phe L97, and Phe H101 form an aromatic pocket in the binding site, with Phe L97 at the bottom of this pocket (Fig. 5). A high proportion of the other antigen binding site residues are neutral amino acids, but there are several charged residues that occur within and immediately surrounding the binding site (His H35, Asp H50, His L33, Arg L90). In order to characterize the binding of ME36.1 to GD2, crystallization of the complex was attempted. This was dependent on obtaining a soluble form of the GD2 antigen, which was accomplished by enzymatic digestion of the ganglioside with ceramide glycanase to cleave the ceramide group from the antigenic carbohydrate portion of GD2 (Li and Li, 1989; Pichla et al., 1995). Initial cocrystallization trials of the Fab with a soluble form of GD2 produced thin plate-like crystals, but additional trials failed to yield crystals with a size sufficient for diffraction. Attempts were also made to soak the soluble GD2 into native Fab crystals. Addition of small molar quantities of the carbohydrate over several days to
10
PICHLA, MURALI, AND BURNETT
FIG. 3. The overall folding of the ME36.1 Fab molecule. The orientation is parallel to the VH:VL pseudo-twofold axis. The ribbon diagram was created with RIBBON (Priestle, 1988). The heavy chain (yellow) contains the VH and CH1 domains. The light chain (blue) contains the VL and CL domains. The complementarity determining region (CDR) loops at the top of the molecule are labeled (H1, H2, H3, L1, L2, L3). The L3 and H3 loops form the bottom of a groove in the antigen binding site that is 20 Å long, 10 Å wide, and 8 Å deep. Each domain has the immunoglobulin fold found in all antibody molecules. The elbow angle between the variable and constant domains is 149.6°. FIG. 4. (a) The antigen binding region of ME36.1, looking down onto the binding site perpendicular to the VH:VL pseudo-twofold axis. The protein surface is displayed in shades of green (for convex surfaces) to gray (for concave surfaces). The modeled sugar is shown in the binding site in stick representation with yellow carbons, red oxygens, and blue nitrogens. The groove-type binding site of the ME36.1 Fab runs across the upper surface, from lower right almost to the upper left of the molecule. This topography is in contrast to the (b) pocket-type binding site of the Salmonella O-antigen binding Fab Se155-4 (Cygler et al., 1991), where the conformation of the antigenic sugar is known from the crystal structure of the complex. Both figures were created with GRASP (Nicholls et al., 1991).
the solution containing the native Fab crystals produced visible etching. The crystals also lasted longer and diffracted X-rays to a slightly higher resolution than the unsoaked crystals. Both observations indicate that structural changes were occurring in the crystals. However, difference maps, phased with the native ME36.1 model, had no interpretable density in the antigen binding site indicating that the sugar was present. In the absence of experimental results for the complex, the binding of the sugar in the antigen binding site was modeled, guided by struc-
tural information from other known protein–sugar crystal structures. The objective was to elucidate the potential sugar contacts and the analysis does not claim to provide a definitive description of the complex. The carbohydrate portion of the GD2 ganglioside (GalNAcb1 = 4Gal[3 ; 2aNeuNAc8 ; 2aNeuNAc]b1 = 4Glcb1 = 1Cer) (Fig. 1) contains five sugar residues (Thurin et al., 1987): two negatively charged, sialic acid residues (NeuNAc), a galactose residue (Gal), an N-actetylgalactosamine residue
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE
11
FIG. 5. The complementarity determining region. The stereo diagram of the CDR loops was created with SETOR (Evans, 1993). The molecule (rotated ,60° clockwise from Fig. 4a) is viewed looking down onto the antigen binding site. The heavy chain CDR loops (H1, H2, H3) are yellow, and the light chain loops (L1, L2, L3) are blue. The binding site is highly aromatic, with four phenylalanines (Phe H64, Phe H101, Phe L95, Phe L97) and four tyrosine residues (Tyr H32, Tyr H60, Tyr H95, Tyr L93). An aromatic pocket in the binding site is created by residues Phe L95, Phe L97, and Phe H101. Although the amino acids in the binding site are mostly neutral, several are charged (Asp H50, Arg L90, His H35, His L33). FIG. 6. Stereo view of the sugar in the binding site of the ME36.1 Fab. The figure was produced with the program SETOR (Evans, 1993). The molecule is rotated ,180° from Fig. 5. Only the CDR loops, and the residues forming hydrogen bonds with the sugar are displayed. The CDR loops are colored as in Fig. 5. The sugar (green) binds along the longest dimension of the Ab binding site, perpendicular to the pseudo-twofold axis of the VH:VL interface. The seven residues that form hydrogen bonds with the sugar (Asn L31, Ser L49, Arg L90, Tyr L93, Ser H100, Thr H33, Asn H52) are displayed in red.
(GalNAc), and a glucose residue (Glc) that links to the ceramide group. Several observations identify the sialic residues as major contributors to the specific affinity of ME36.1 for GD2. This simplification made the modeling study a feasible task. The ME36.1 antibody has the highest affinity for GD2, but also cross-reacts with lower affinity to GD3 (NeuNAca2 = 8NeuNAca2 = 3Galb1 = 4Glcb1 = 1Cer) (Thurin et al., 1987). The two sugars are identical except that GD2 contains a branched N-acetylgalactosamine (GalNAc) (Fig. 1). For the ME36.1 antibody to cross-react with GD2 and GD3, and with no other ganglioside that contains a termi-
nal sialic acid (NeuAc) residue, ME36.1 must be interacting with both sialic acid residues. The higher affinity for the GD2 ganglioside is most likely due to the additional interactions (hydrogen bonding, van der Waals, and stacking) provided by the GalNAc residue. Therefore, the sugar used as the model consisted of the two sialic acid residues, the galactose residue, and GalNAc. Other complex gangliosides contain sugars with a2–8-linked sialic acid residues but the additional sugars are linked to the GalNAc and have a longer branched chain than GD2, which presumably prevents them from binding to ME36.1.
12
PICHLA, MURALI, AND BURNETT
There were several clues to where the sugar should be placed in the antigen binding site. It is known that iodination of the tyrosines in ME36.1 antibody destroys its ability to bind to GD2, suggesting that at least one tyrosine is involved in binding (Dr. Dorothy Herlyn, personal communication). As there is no evidence to suggest that the galactose residue is involved in any interactions with the antibody, this could be modeled away from the binding site. The H3 and L3 loops were targeted as the major loops involved in binding since they are known to be important for specificity (Davies and Padlan, 1990; Stanfield et al., 1993), and a tyrosine (Tyr L93) is on the L3 loop. The J539 Fab used in the molecular replacement search binds to galactan residues. As the sequence of the L1 loop is identical in J539 and ME36.1, this loop may be important in sugar binding. It is noteworthy that carbohydrate binding proteins often have two distinct globular domains that form a cleft in which the sugar binds (Quiocho, 1986). This arrangement is analogous to that in antibodies, where the variable light and heavy domains form an antigen binding site in the shape of a cavity or groove. The preceding observations established four main criteria for modeling: (1) residues from the H3, L3, and L1 loops should interact with the sugar; (2) a tyrosine should interact with the sugar; (3) the sugar should bind in the groove between the variable light and heavy domains; (4) the galactose residue would lie away from the binding site. An obstacle in modeling the binding of an antigen in a Fab binding site is that it is difficult to predict whether it will occur via an induced fit or a lock-andkey mechanism. It seems most likely that the flexible GD2 sugar will undergo a conformational change as it adopts its antigenic conformation. Moreover, in the cases where structures of both a native Fab and its complex are available, so that movements in the antigen binding site can be observed, at least a small movement occurs (Pokkuluri et al., 1994). In other cases, a significant movement of the loops occurs (Stanfield et al., 1993; Rini et al., 1992), and in several cases the VL and VH domains even move with respect to one another upon antigen binding (Stanfield et al., 1993; Herron et al., 1991; Bhat et al., 1990). Given this spectrum of possible movements, it seemed reasonable to assume that both the Fab and the sugar move during complex formation. Thus, binding was modeled by first minimizing the sugar to a starting model free of steric hindrance. The sugar then was moved manually around the CDR to find possible positions for the antigen binding site. The sugar conformation was adjusted at each position to avoid overlap with protein side chains. One particularly favorable position was found where
several hydrogen bonds were formed between the sugar and the residues in the binding site. Conjugate gradient minimization on the complex (the sugar and the Fab), without constraints on either molecule, then optimized the interactions at this position. Initially, it was thought that a sugar residue should be placed in the aromatic pocket of the binding site (Fig. 5). However, it was immediately clear from the modeling that this could not occur without moving the protein backbone and making a major rearrangement of the side chain residues to avoid steric interference. The best fit for the sugar was obtained by placing it in the groove in the middle of the binding site that runs perpendicular to the pseudo-twofold axis relating the VH and VL domains. In this position, extensive hydrogen bond interactions would occur between the sugar and the side chain residues from five of the six CDR loops of the Fab. Seven residues in the antigen binding site would have hydrogen bond interactions with the sugar (Fig. 6). The terminal sialic acid residue would accept three hydrogen bonds from surrounding side chain residues. The O7 oxygen of the glycerol moiety of the sugar would accept a hydrogen bond from Asn H52 of the H2 loop, Tyr L93 of the L3 loop donating a hydrogen bond to the O4 ring oxygen, and Thr H33 of the H1 loop would donate a hydrogen bond to the negatively charged carboxyl of the sugar. The O4 ring oxygen of the second sialic acid residue would accept a hydrogen bond from Ser H100 of the H3 loop. The carbonyl of the N-methylacetamide of this same sialic acid would accept a hydrogen bond from the Arg L90 of the L3 loop. The GalNAc residue would interact with two surrounding residues. The O4 ring oxygen of GalNAc would form hydrogen bonds with the framework residue, Ser L49, of the L2 loop, and Asn L31 would donate a hydrogen bond to the carbonyl oxygen of the amide group of this same sugar. In this configuration, the galactose residue would project from the binding site into the solvent and not interact with the Fab. After the final minimization, the root-mean-square deviation for all atoms of the variable region of the Fab from their original positions was 0.07 Å. The total buried surface area, determined using a probe radius of 1.7 Å, was 413 Å2 for the Fab and 526 Å2 for GD2. Placement of the water molecules in the ME36.1 Fab structure was not attempted until the Fab model was well-refined to a reasonable R factor and showed good stereochemistry. At the comparatively low resolution of 2.8 Å, very strict criteria were employed to define the set of 34 water molecules in the final structure. In this set, 2 water molecules lie in the antigen binding site of the native Fab structure, where they are close to several residues in-
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE
volved in binding the sugar. One water forms a hydrogen bond with Asp H50, which is in the vicinity of residue H52 that interacts with the sugar. The other water forms hydrogen bonds with Ser L91 and Ser L30, which are both next to two residues interacting with the sugar (Asn L31, Arg L90) (data not shown). The locations of the water molecules are such that they could remain in the binding site to coordinate hydrogen bond interactions between the sugar and the Fab, thus increasing the stability of the complex. However, as only a slight change in the torsion angles for the sugar would cause these waters to be displaced, this possibility cannot be excluded. The model for the complex explains the higher affinity of ME36.1 for GD2 than for GD3, which is due to the manner in which the sugar is anchored in the binding site. On one end of the binding groove, a sialic acid residue forms hydrogen bonds with residues from the H1 and H2 loops. At the other end, the GalNAc residue is involved with the L1 and L3 loops and is in close proximity to the residue in the L1 loop (Ser L30) that hydrogen bonds to a water molecule. This water molecule could coordinate hydrogen bonding between the L1 loop and the GD2 sugar. Thus, the presence of the GalNAc residue in GD2 not only contributes additional contacts with ME36.1, but also limits the conformational flexibility of GD2, which may allow it to maintain a more stable antigenic conformation. DISCUSSION
Comparative studies of antibody structures and their sequences have shown that the main chain conformations for five (L1, L2, L3, H1, H2) of the six CDR loops are limited to only a few possible spatial arrangements (Chothia et al., 1989). That is, each loop has a distinct conformation, but the same loop can be very similar in different Fabs. The conformations of the ME36.1 Fab antigen binding loops L1, L2 and H1, H2 correspond well to the canonical forms corresponding to their sequences (Chothia et al., 1989). However, the L3 loop sequence and structure are distinct from other Fab structures reported. Both the L3 loop and the H3 loop, which has no known canonical structure, occur close to the recombination site for the variable and joining regions of the L and H chains (Roitt et al., 1989). This explains the high variability in the sequences of these loops compared to the others, and it is reasonable to assume that their conformations would be equally as variable (Kabat et al., 1983). Attempts to humanize the ME36.1 Ab were unsuccessful in producing a high-affinity Ab to GD2. Once the mouse ME36.1 native Fab structure was available, it was used to identify residues in the human
13
framework region that could be important in maintaining the affinity for GD2. However, the structure did not provide additional clues beyond those that were already intuitively obvious (Dr. Jose Saldanha, personal communication). For example, several residues in both the light and heavy chain framework regions of the human Ab were changed to the mouse sequence, as they were thought important for maintaining correct loop conformations (Chothia et al., 1989). Other residues that were thought to influence the packing of the VH–VL interface were changed, as the interaction of these two domains affects the spatial arrangement of the CDR loops. In addition, framework residues that might contact antigen and residues putatively identified as glycosylation sites were also examined. Several mouse antibodies have successfully been humanized while maintaining high antigen specificity (Kolbinger et al., 1993), but inexplicably, others cannot be humanized. The modeling of the sugar in the binding site of the ME36.1 Fab structure was based on the types of interactions observed in other protein–carbohydrate structures (Vyas, 1991; Quiocho, 1986; Bundle and Young, 1992). Most of the interactions between the protein and carbohydrate are due to extensive hydrogen bonding and van der Waals forces (Vyas, 1991; Quiocho, 1986). Additionally, stacking interactions, carboxylates, and water molecules have been implicated in protein–carbohydrate binding. In most of the non-Fab protein–carbohydrate structures determined so far, the charged carboxylates of aspartate and glutamate and the uncharged polar asparagine and arginine residues are the most frequent acceptors of hydrogen bonds from sugar hydroxyl groups. Main chain polar groups also form hydrogen bonds with sugar residues (Vyas, 1991). All of these hydrogen bonding interactions facilitate the formation of van der Waals interactions by enabling polar residues in the binding region to approach the sugar (Vyas, 1991). Two antibody–carbohydrate structures have been solved to date, Se155-4–Salmonella O-antigen complex (Cygler et al., 1991) and BR96– nLey complex (Jeffrey et al., 1995). In contrast to the non-Fab protein–carbohydrate structures, the interaction of the sugar with the Fab occurs mostly through hydrogen bonding to tyrosine and tryptophan and to main chain atoms. In the ME36.1 binding site, the interactions of the sugar include charged planar groups (Arg L90, His L33), uncharged polar groups (Ser H100, Asn H52, Thr H33, Asn L31), and an aromatic residue (Tyr L93). The interactions are a combination of those seen in non-Fab–carbohydrate structures and those in Fab– carbohydrate structures. As the binding site of ME36.1 is highly aromatic, consistent with the high frequency of aromatic residues seen in antibody
14
PICHLA, MURALI, AND BURNETT
binding sites (Padlan, 1990), it is conceivable that more aromatic residues are involved in binding GD2 than were modeled. However, as these residues lie in the aromatic pocket, a significant rearrangement of the side chains or loops would be required upon sugar binding, which seems unlikely. In protein–carbohydrate complexes, aromatic residues are frequently involved in stacking interactions with the nonpolar face of the sugar ring, with tryptophan being the most common (Vyas, 1991). These stacking interactions confer specificity by constraining the sugar to a particular conformation by sterically hindering other configurations. In contrast, the aromatic residues in the two known Fab– carbohydrate structures are involved in hydrogen bond interactions rather than stacking interactions. The structure of the J539 galactan binding Fab (Suh et al., 1986), which was used as the search model to solve ME36.1, is only known in its native form and not as a complex with its antigen. The binding site of J539 is also highly aromatic, with two tryptophans and four tyrosine residues. Labeling experiments with deoxyfluorogalactoside and a modeling study with a galactan sugar suggested that antigen binding involves two solvent exposed typtophans (Suh et al., 1986). The interactions of the sole aromatic residue (Tyr L93) in the ME36.1 binding site with the GD2 sugar model are similar to those in other Fab–carbohydrate structures. Tyr L93 forms a hydrogen bond with the sugar, rather than being involved in stacking interactions. There are no tryptophans in the ME36.1 CDR loops, but other aromatic residues in the binding site (four tyrosines and four phenylalanines) could make stacking interactions with the GD2 sugar similar to those observed with tryptophan. Indeed, there are several tryptophan residues in the framework regions of ME36.1 that are in close proximity to the antigen binding site, and it is conceivable that these are involved in antigen binding. It is not uncommon for framework residues to be involved in antigen binding (Bentley et al., 1990). However, this is unlikely in the case of ME36.1 since changing human framework residues to tryptophans where these occur in the mouse sequence did not increase the affinity (Jose Saldanha, personal communication). Moreover, a significant rearrangement of the binding site would be necessary for the tryptophan residues to interact with GD2. Water molecules provide another mechanism for the formation of hydrogen bonds between sugars and protein side chain residues. In the case of antibody– antigen interactions, most if not all water molecules are displaced in an antibody binding site upon antigen binding (Davies and Padlan, 1990). However, the Se155-4–Salmonella O-antigen complex has a critical water molecule in the antigen binding
site that coordinates four hydrogen bonds between one of the sugar residues and several CDR side chain residues (Cygler et al., 1991). The two water molecules in the native ME36.1 Fab binding site lie close to several residues that interact with the sugar model and are in a position where they can coordinate hydrogen bonds between the sugar and side chain residues. However, a small change in the conformation of the sugar could just as easily cause water displacement upon antigen binding. Although the positions of the two water molecules are consistent with the modeled position of the sugar, and indeed lend it support, it is difficult to make a definitive statement about their role in antigen binding. It has been proposed that carbohydrate binding antibodies have either groove- or pocket-type binding sites (Cisar et al., 1975). The BR96 antibody has a deep pocket in its binding site that is 12 Å in diameter and 10 Å deep (Jeffrey et al., 1995). The Se155-4 antibody has a more shallow binding site, 7 Å in diameter and 8 Å deep, but clearly shows a pocket-type binding site (Fig. 4b) (Cygler et al., 1991). In comparison, the ME36.1 binding site is approximately 10 Å wide, 8 Å deep, and 20 Å long and creates a groove-type binding site that is long enough to accommodate the carbohydrate chain of GD2 (Fig. 4a). The contact area in the binding site for ME36.1 is 413 Å2, which is consistent with the value for the deep pocket of BR96 (422 Å2) and somewhat more than that for the shallow pocket of Se155-4 (304 Å2). The molecular model for the interaction of GD2 with ME36.1 Fab is consistent with all types of protein–carbohydrate interactions that have been observed and so presents a reasonable model to explain binding. All four criteria set for the modeling were satisfied: the L3, H3, and L1 loops interact with the sugar; binding involves a tyrosine; the sugar lies in a groove between the domains; and the galactose residue is outside the binding site and does not interact with the Fab. Although previous attempts to obtain cocrystals of ME36.1 Fab–GD2 complex were unsuccessful, trials are continuing. It will be interesting to see whether the interactions of the Fab–GD2 resemble those of the other two Fab–carbohydrate structures or encompass the wider range of interactions observed in other protein–carbohydrate structures. It will also be interesting to see how closely the model predicts the true interaction of the ME36.1 antibody with GD2. The atomic coordinates for both the ME36.1 Fab crystal structure and the GD2 model have been deposited in the Brookhaven Protein Data Bank (PDB Accession No. 1PSK).
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE We thank Drs. Albrecht Luckenbach and Maria Kordowicz (Merck KGaA, Darmstadt, Germany) for supplying generous quantities of purified ME36.1 monoclonal antibody and GD2 ganglioside. We also thank Dr. Jose Saldanha and Dr. Mary Bendig (MRC Collaborative Centre, London, UK) for helpful discussions about the humanization of the antibody. This research was supported by a grant to R.M.B. from the National Institute of Allergy and Infectious Diseases (AI 17270) and by Wistar Training Grant CA-09171 (S.L.P.). Further support was provided by E. Merck, Darmstadt, and the Wistar Cancer Center (CA 10815). REFERENCES Bentley, G. A., Boulot, G., Riottot, M. M., and Poljak, R. J. (1990) Three-dimensional structure of an idiotope-anti-idiotope complex, Nature 348, 254–257. Bhat, T. N., Bentley, G. A., Fischmann, T. O., Boulot, G., and Poljak, R. J. (1990) Small rearrangements in structures of Fv and Fab fragments of antibody D1.3 on antigen binding, Nature 347, 483–485. Brooks, B. R., Bruccoleri, R. E., Olafson, B. D., States, D. J., Swaminathan, S., and Karplus, M. (1983) CHARMM: A Program for macromolecular energy, minimization, and dynamics calculations, J. Comput. Chem. 4, 187–217. Bru¨nger, A. T. (1990) Extension of molecular replacement: A new search strategy based on Patterson correlation refinement, Acta Crystallogr. A46, 46–57. Bru¨nger, A. T. (1992a) X-PLOR Version 3.1: A system for X-ray Crystallography and NMR, Yale University, New Haven, CT. Bru¨nger, A. T. (1992b) Free R value: A novel statistical quantity for assessing the accuracy of crystal structures, Nature 355, 472–475. Bru¨nger, A. T., Krukowski, A., and Erickson, J. W. (1990) Slowcooling protocols for crystallographic refinement by simulated annealing, Acta Crystallogr. A46, 585–593. Bru¨nger, A. T., Kuriyan, J., and Karplus, M. (1987) Crystallographic R factor refinement by molecular dynamics, Science 235, 458–460. Bundle, D. R., and Young, N. M. (1992) Carbohydrate–protein interactions in antibodies and lectins, Current Opinion Struct. Biol. 2, 666–673. Collaborative Computational Project, Number 4. (1994) The CCP4 Suite: Programs for Protein Crystallography, Acta Crystallogr. D50, 760–763. Chothia, C., Lesk, A. M., Tramontano, A., Levitt, M., Smith-Gill, S. J., Air, G., Sheriff, S., Padlan, E. A., Davies, D., Tulip, W. R., Colman, P. M., Spinelli, S., Alzari, P. M., and Poljak, R. J. (1989) Conformations of immunoglobulin hypervariable regions, Nature 342, 877–883. Cisar, J., Kabat, E. A., Dorner, M. M., and Liao, J. (1975) Binding properties of immunoglobulin combining sites specific for terminal or nonterminal antigenic determinants in dextran, J. Exp. Med. 142, 435–459. Cygler, M., Rose, D. R., and Bundle, D. R. (1991) Recognition of a cell-surface oligosaccharide of pathogenic Salmonella by an antibody Fab fragment, Science 253, 442–445. Davies, D. R., and Padlan, E. A. (1990) Antibody–antigen complexes, Annu. Rev. Biochem. 59, 439–473. Engh, R. A., and Huber, R. (1991) Accurate bond and angle parameters for X-ray protein structure refinement, Acta Crystallogr. A47, 392–400. Evans, S. V. (1993) SETOR: Hardware-lighted three-dimensional solid model representations of macromolecules, J. Mol. Graphics 11, 134–138. Herlyn, M., and Koprowski, H. (1988) Melanoma antigens: Immu-
15
nological and biological characterization and clinical significance. Annu. Rev. Immunol. 6, 283–308. Herron, J. N., He, X. M., Ballard, D. W., Blier, P. R., Pace, P. E., Bothwell, A. L. M., Voss, E. W., Jr., and Edmundson, A. B. (1991) An autoantibody to single-stranded DNA: Comparison of the three-dimensional structures of the unliganded Fab and a deoxynucleotide–Fab complex, Proteins 11, 159–175. Iliopoulos, D., Ernst, C., Steplewski, Z., Jambrosic, J. A., Rodeck, U., Herlyn, M., Clark, W. H., Jr., Koprowski, H., and Herlyn, D. (1989) Inhibition of metastases of a human melanoma xenograft by monoclonal antibody to the GD2/GD3 gangliosides, J. Natl. Cancer Inst. 81, 440–444. Jeffrey, P. D., Bajorath, J., Chang, C. Y., Yelton, D., Hellstro¨m, I., Hellstro¨m, K. E., and Sheriff, S. (1995) The X-ray structure of an anti-tumour antibody in complex with antigen, Nature Struct. Biol. 2, 466–471. Jones, T. A. (1982). FRODO: A graphics fitting program for macromolecules, in Sayre, D. (Ed.), Computational Crystallography, pp. 303–317, Clarendon Press, Oxford. Jones, T. A., Zou, J.-Y., Cowan, S. W., and Kjeldgaard, M. (1991) Improved methods for building protein models in electron density maps and the location of errors in these models, Acta Crystallogr. A47, 110–119. Kabsch, W. (1988) Automatic indexing of rotation diffraction patterns. J. Appl. Crystallogr. 21, 67–71. Kabat, E. A., Wu, T. T., Reid-Miller, M., Perry, H. M., and Gottesman, K. S. (1987) Sequences of Proteins of Immunological Interest, 4th ed., Public Health Service, NIH, Washington, DC. Kolbinger, F., Saldanha, J., Hardman, N., and Bendig, M. M. (1993) Humanization of a mouse anti-human IgE antibody: A potential therapeutic for IgE-mediated allergies, Protein Eng. 6, 971–980. Laskowski, R. A., MacArthur, M. W., Moss, D. S., and Thornton, J. M. (1993) PROCHECK: A program to check the stereochemical quality of protein structures, J. Appl. Crystallogr. 26, 283–291. Lattman, E. E. (1972) Optimal sampling of the rotation function, Acta Crystallogr. B28, 1065–1068. Li, Y.-T., and Li, S.-C. (1989) Ceramide glycanase from leech, Hirudo medicinalis, and earthworm, Lumbricus terrestris, Methods Enzymol. 179, 479–487. Mayworth, R. D., and Quinta´ns, J. (1990) Designer and catalytic antibodies, N. Engl. J. Med. 323, 173–178. Nicholls, A., Sharp, K. A., and Honig, B. (1991) Protein folding and association: Insights from the interfacial and thermodynamic properties of hydrocarbons, Proteins: Struct. Funct. Genet. 11, 281–296. Padlan, E. A. (1990) On the nature of antibody combining sites: Unusual structural features that may confer on these sites an enhanced capacity for binding ligands, Proteins 7, 112–124. Pichla, S. L., Murali, R., and Burnett, R. M. (1995) Preliminary crystallographic data for an Fab to the melanoma-associated GD2 ganglioside, and the purification of a soluble form of this antigen, Acta Crystallogr. D51, 124–126. Pokkuluri, P. R., Bouthillier, F., Li, Y., Kuderova, A., Lee, J., and Cygler, M. (1994) Preparation, characterization and crystallization of an antibody Fab fragment that recognizes RNA, J. Mol. Biol. 243, 283–297. Priestle, J. P. (1988) RIBBON: A stereo cartoon drawing program for proteins, J. Appl. Crystallogr. 21, 572–576. Quanta (1989) Version 2.1, Polygen Corp., Waltham, Massachusetts. Quiocho, F. A. (1986) Carbohydrate-binding proteins: Tertiary
16
PICHLA, MURALI, AND BURNETT
structures and protein-sugar interactions, Annu. Rev. Biochem. 55, 287–315. Ramachandran, G. N., and Sasisekharan, V. (1968) Conformation of polypeptides and proteins, Adv. Protein Chem. 23, 283–438. Ravindranath, M. H., Tsuchida, T., and Irie, R. F. (1989) Diversity of ganglioside expression in human melanoma, in Oettgen, H. F. (Ed.), Gangliosides and Cancer, pp. 79–91, VCH, New York. Rini, J. M., Schulze-Gahmen, U., and Wilson, I. A. (1992) Structural evidence for induced fit as a mechanism for antibodyantigen recognition, Science 255, 959–965. Roitt, I. M., Brostoff, J., and Male, D. K. (1989) Immunology, 2nd ed., Gower Medical, New York. Stanfield, R. L., Fieser, T. M., Lerner, R. A., and Wilson, I. A. (1990) Crystal structures of an antibody to a peptide and its complex with peptide antigen at 2.8 Å, Science 248, 712–719. Stanfield, R. L., Takimoto-Kamimura, M., Rini, J. M., Profy, A. T., and Wilson, I. A. (1993) Major antigen-induced domain rearrangements in an antibody, Structure 1, 83–93.
Suh, S. W., Bhat, T. N., Navia, M. A., Cohen, G. H., Rao, D. N., Rudikoff, S., and Davies, D. R. (1986) The galactan-binding immunoglobulin Fab J539: An X-ray diffraction study at 2.6-Å resolution, Proteins 1, 74–80. Tai, T., Cahan, L. D., Tsuchida, T., Saxton, R. E., Irie, R. F., and Morton, D. L. (1985) Immunogenicity of melanoma-associated gangliosides in cancer patients, Int. J. Cancer 35, 607–612. Thurin, J., Thurin, M., Kimoto, Y., Herlyn, M., Lubeck, M. D., Elder, D. E., Smereczynska, M., Karlsson, K.-A., Clark, W. M., Jr., Steplewski, Z., and Koprowski, H. (1987) Monoclonal antibody-defined correlations in melanoma between levels of GD2 and GD3 antigens and antibody-mediated cytotoxicity, Cancer Res. 47, 1229–1233. Vyas, N. K. (1991) Atomic features of protein–carbohydrate interactions, Current Opinion Struct. Biol. 1, 732–740. Waldmann, T. A. (1991) Monoclonal antibodies in diagnosis and therapy, Science 252, 1657–1662.
119, 6–16 (1997)
The Crystal Structure of a Fab Fragment to the Melanoma-Associated GD2 Ganglioside Susan L. Pichla, Ramachandran Murali,1 and Roger M. Burnett2 The Wistar Institute, 3601 Spruce Street, Philadelphia, Pennsylvania 19104 Received July 31, 1996, and in revised form February 20, 1997
cell signaling or cell-to-substrate interactions that regulate cell growth. Normally, melanocytes, which are pigment-forming cells, express GM3 as the most prominent ganglioside. As the transformation to a cancerous cell commences in melanoma, the distribution and type of gangliosides on the cell surface start to change. This alteration is due to the triggering of glycosyltransferases that add additional sugars to GM3 to form gangliosides such as GD2, 9-O acetylated GD3, GM2, and GD3 (Ravindranath et al., 1989). As these gangliosides are highly expressed on malignant melanoma cells (Thurin et al., 1987), their appearance on the cell surface signals the progression of the disease. The GD2 and GD3 gangliosides are being used as target antigens for immunotherapy as an alternative to other treatments, such as radiation therapy, that have not been very effective in treating melanoma in its later stages (Herlyn and Koprowski, 1988; Tai et al., 1985; Thurin et al., 1987; Iliopoulos et al., 1989). Monoclonal antibodies can be used in immunotherapy against melanoma to identify and eliminate micrometastases and so prevent further spread of the disease (Herlyn and Koprowski, 1988). The ME36.1 mouse monoclonal antibody (IgG2a) recognizes the GD2 ganglioside and, with lower affinity, the GD3 ganglioside (Fig. 1). ME36.1 can inhibit tumor growth and metastatic spread of melanoma in nude mice (Iliopoulos et al., 1989). Clinical trials with the antibody in 13 patients with melanoma gave encouraging results, and 1 patient had complete remission for over a year (Iliopoulos et al., 1989). Structural studies on ME36.1 were undertaken to understand the specific interactions between the antibody and the therapeutically targeted GD2 ganglioside. Moreover, it has been recognized that rodent antibodies such as the ME36.1 mouse antibody, when used as an immunotherapeutic, elicit an immune response that results in the rapid clearance of the antibody from the human patient (Mayworth and Quinta´ns, 1990; Waldmann, 1991). Thus, stud-
The GD2 ganglioside is a cell-surface component that appears on the surface of metastatic melanoma cells and is a marker for the progression of the disease. The ME36.1 monoclonal antibody binds to the GD2 ganglioside and has shown potential as a therapeutic antibody. ME36.1 is a possible alternative therapy to radiation, which is often ineffective in late-stage melanoma. The crystal structure of the Fab fragment of ME36.1 has been determined using molecular replacement and refined to an R factor of 20.4% at 2.8 Å resolution. The model has good geometry with root-mean-square deviations of 0.008 Å from ideal bond lengths and 1.7° from ideal bond angles. The crystal structure of the ME36.1 Fab shows that its complementarity determining region forms a groove-shaped binding site rather than the pocket-type observed in other sugar binding Fabs. Molecular modeling has placed a four-residue sugar, representative of GD2, in the antigen binding site. The GD2 sugar moiety is stabilized by a network of hydrogen bonds that define the specificity of ME36.1 toward its antigen. r 1997 Academic Press INTRODUCTION
Cell-surface carbohydrates are important as recognition elements as well as markers for disease, and much effort has been devoted to understanding the specificity and affinity of proteins for carbohydrates. Carbohydrates have great inherent flexibility, which allows them to adopt a wide range of stereoisomers (Vyas, 1991; Quiocho, 1986). Gangliosides are a particular class of cell-surface glycolipids that are distinguished from other glycolipids by the presence of sialic acid residues and that are expressed on most cells. Although the function of gangliosides is not clearly known, they appear to be involved in cell-to1 Current address: Department of Pharmacology, The Medical School of the University of Pennsylvania, Philadelphia, PA 19104. 2 To whom correspondence should be addressed. Fax: (215)-8983868. E-mail: [email protected].
1047-8477/97 $25.00 Copyright r 1997 by Academic Press All rights of reproduction in any form reserved.
6
7
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE
FIG. 1. Chemical formulas for the GD2 and GD3 gangliosides. Both consist of carbohydrate chains forming a head group, which is linked to a ceramide group (Cer) that lies in the outer cell membrane. The only difference between the two glycolipids is a branched N-acetylgalactosamine (GalNAc) in the GD2 sugar.
ies are under way to ‘‘humanize’’ the ME36.1 antibody by grafting the mouse complementarity determining region (CDR) loops, which define the antigen binding site, onto a human antibody framework. This potentially increases the effectiveness of the antibody in immunotherapy by making it less immunogenic. However, attempts to humanize the antibody have so far decreased its binding affinity (Dr. G. A. Luckenbach, personal communication). Since the framework residues surrounding the CDR loops are important for the correct folding of these loops (Chothia et al., 1989), a crystal structure for ME36.1 Fab should reveal which framework residues are important for loop conformation. These residues could then be altered in the variable region of the humanized antibody so that the correct conformation of the CDR graft is maintained and the affinity of the antibody for its antigen is not reduced. Crystals of the native ME36.1 Fab were obtained, a complete diffraction data set to 2.8 Å resolution was collected, and the structure was solved. A model of the GD2 carbohydrate (Fig. 2) then was used to investigate the types of interactions that could occur between the side chain residues in the binding site of the Fab and the sugar. The results suggest a mode of binding that is consistent with the types of interactions seen in other protein–carbohydrate structures. MATERIALS AND METHODS Data Collection The purification and crystallization of the ME36.1 Fab fragment and attempts at cocrystallization of the Fab with the extracellular portion of GD2 have been described previously (Pichla et al., 1995). The modified form of GD2 was produced by enzymatically removing its ceramide group to produce a soluble antigen. Crystals of the Fab were grown in 30% PEG 4000, 1% MPD, and Tris buffer at pH 7.5. The crystals belong to the monoclinic space group P21, with unit cell dimensions a 5 37.6 Å, b 5 94.1 Å, c 5 67.4 Å, and b 5 101.0°. Diffraction data sets were collected from six native crystals on a Siemens-Nicolet X100-A
FIG. 2. Schematic of the four-residue sugar used in the modeling to represent GD2. The unusual a2–8 linkage between the two sialic acid (NeuNAc) residues occurs between the C2 of one sugar and the C8 of the hydroxyl in the other. The galactose (Gal), which is attached to the ceramide group in GD2 (Fig. 1), links the second sialic acid residue and the N-acetylgalactosamine (GalNAc).
multiwire area detector. The X-ray source was a Rigaku RU200 rotating-anode generator, equipped with a double mirror focusing system (Supper) and operated at 40 kV and 65 mA. The crystals diffract to 2.5 Å resolution but, as the intensity falls off sharply after 2.8 Å, data collection was limited to the lower resolution. The data collection statistics are presented in Table I. Structure Determination The structure of Fab ME36.1 was solved by molecular replacement using the X-PLOR program package (Bru¨nger, 1992a) to carry out the rotational and translational searches. The Fab portion of the galactan binding J539 Fab structure (Suh et al., 1986) was used as a search model to determine the orientation and position of the ME36.1 Fab in the unknown crystal environment. The J539 Fab was chosen as the search model due to its high sequence similarity to ME36.1 in the heavy (VH) and light chain (VL) variable regions, with 42% homology for VH and 74% homology for VL. Both Fab molecules have a k constant region in
TABLE I ME36.1 Data Collection Statistics as a Function of Resolutiona Resolution limit (Å)
Number of observed reflections
Number of unique reflections
Completeness (%)
R-symb (%)
8.00 5.00 4.50 4.00 3.40 3.00 2.80 Total
1181 2505 2111 3144 6075 6330 4012 33 212
462 841 733 1137 2417 2827 2021 14 677
90.4 98.6 98.0 97.8 98.3 97.6 95.4 96.1
8.6 9.7 9.2 9.6 10.5 12.1 14.7 10.4
a Six data sets, collected on a Siemens-Nicolet X100-A area detector and reduced and merged with XDS (Kabsch, 1988). b R-sym 5 o 0 I 2 ,I . 0 /oI. I, observed intensity and ,I. the average intensity of multiple measurements of symmetry related reflections.
8
PICHLA, MURALI, AND BURNETT
the light chain (CL). The heavy chain constant region (CH1) of J539 belongs to a class different from that of ME36.1. A real-space rotation was performed with X-PLOR using the variable and constant domains of the model separately. The whole Fab also was used, taking advantage of the option in the program to vary the ‘‘elbow angle’’ between the variable and the constant domains. The J539 search model was placed in an orthorhombic unit cell with dimensions 120 3 120 3 150 Å, so that the cell was twice the size of the Fab molecule in each direction. Structure factors were calculated for the model, and the Patterson map corresponding to the crystal was calculated by fast Fourier transformation on a 1.0-Å grid. The radius of integration for the rotation function was 45 Å. The orientations of the search model were sampled using Lattman (1972) angles: u2 5 0–720°, u2 5 0–90°, u1 5 0–360° at a sampling interval of 2.5°. The highest 1000 peaks from the Patterson search at 6.0–3.0 Å (I $ 2s) were listed and sorted according to peak height. Those rotations that were within 10° from one another were clustered and the condensed set of orientations was used to provide starting points for Patterson correlation refinement (PC refinement) (Bru¨nger, 1990). The model orientations that gave the highest correlation value after PC refinement were similar for the constant and variable domains and for the whole Fab molecule (Table II). In all three cases, the highest correlation value obtained was well above that for other possible orientations. The translation search was also performed with each of the three oriented and PC-refined molecules. The origin along the twofold axis in space group P21 is arbitrary; therefore, the translation search was limited to the x–z plane. Sampling intervals of 0.125 Å in x and z were used in the grid search, which was restricted to the asymmetric unit (0 # x # 1.0, 0 # z # 0.5). The highest values obtained in the translation searches for the constant and variable regions and for the whole Fab were 7.6, 9.3, and 9.5 s above background, respectively. These maxima correspond to similar translations for all three search models (Table II). The final correlation value for the whole Fab was 0.367, with an R factor of 44% [R factor 5 (ohkl 0 Fo 2 kFc 0 )/ohkl 0 Fo 0 , where Fo and Fc are the observed and calculated structure factors, respectively]. The correctly oriented and translated whole J539 Fab was used as the basis for further refinement. The first step was to perform molecular dynamics refinement with a molecular model based on the J539 Fab, against the ME36.1 diffraction data using reflections from 6.0 to 3.0 Å resolution (7884 reflections with Fo $ 2.0 s). Since the sequences for the J539 and ME36.1 were so similar, the side chain residues of the variable (VL and VH) domains of J539 were not removed during the first round of refinement. As the CH1 domains belong to different classes, their side chains were changed to polyalanine residues to avoid any bias. The CDR loops were also removed in
TABLE II Molecular Replacement Solution for ME36.1 Search model searcha
Rotation VL–VH CL–CH Whole Fab
Correlation coefficient u1 u2 u3 231.4° 39.6° 74.6° 237.8° 34.0° 75.1° 233.3° 37.0° 76.7°
Translation searchb Dx 0.075 VL–VH 0.137 CH–CL Whole Fab 0.150 a b
Dy 0.0 0.0 0.0
Dz 0.313 0.263 0.275
Rotations are given in Eulerian angles. Translations are given in fractional coordinates.
0.137 0.114 0.227 0.221 0.205 0.367
the initial model. At this early stage of refinement, the isotropic temperature factors were set to a constant value of 19 Å2. The model was prepared for molecular dynamics by first minimizing the total energy, while removing all inappropriate contacts, by subjecting the model to one round of positional refinement of the individual atoms with 140 steps of Powell conjugate gradient minimization (Bru¨nger et al., 1987, 1990). The R factor for the model dropped from its initial value of 44 to 28%. The model was then further refined by simulated annealing (SA) using a slow-cooling protocol. This refinement step was accomplished by first raising the model to a starting temperature of 3000 K, cooling it to 300 K in 50 integration steps of 0.025 psec, and performing a further round of positional refinement. The R factor after SA refinement was 24%. After the initial refinement, 2Fo–Fc difference maps were calculated with the restricted 6.0- to 3.0-Å data (Fo $ 2s). The previously omitted residues within the CDR loops, and the side chains of the variable (VL and VH) and CH1 regions, were manually built into the map using the ME36.1 sequence only where the electron density was clearly interpretable. This model was improved by alternating cycles of refinement and manual rebuilding using the program FRODO (Jones, 1982) on an Evans and Sutherland PS350 vector graphics display. Isotropic temperature factors (B factors) were kept constant for the first rounds of manual rebuilding and refinement until most of the residues had been placed in the model, after which they were refined by alternating rounds of minimization and restrained B-factor refinement to an R factor of 20.9%. At this stage, the resolution limits were extended to include data from 8.0 to 2.8 Å resolution, and 10 355 reflections (Fo $ 2s) were used for several more rounds of refinement. A test set, consisting of 10% of the observed data (1035 reflections), then was extracted from the reflection file for the calculation of R free (Bru¨nger, 1992b). Any bias was removed by one round of simulated annealing using the remaining reflections. At this stage, the R factor was 19.0%, and R free was 33.1%. Refinement was then alternated with model-building in 2Fo–Fc difference maps calculated using 20.0- to 2.8-Å data (Fo $ 2s). The Engh and Huber (1991) topology and parameter files also were introduced. This resulted in an improvement in the stereochemistry of the model, with a substantial decrease in R free (31.0%) and an increase in the R factor (21.1%). Assignment of the water molecules in the Fab structure was not attempted until the Fab model was well-refined to an R factor of 21%, with good stereochemistry indicated by the rms deviations of bond lengths and angles and by the Ramachandran plot. The water molecules were located by using the program PEAKMAX in CCP4 (1994) to select those electron-dense peaks that were .3s above background in an Fo–Fc difference map. The program WATPEAK (CCP4, 1994) then selected only those that were within 3.6 Å from an atom in the Fab molecule, giving a total of 99 water molecules. These candidates were examined graphically and those that did not have good density in 2Fo–Fc difference maps (at least 2s), along with hydrogen bond interactions with side chain or main chain atoms, were deleted to leave 45 waters. The model then was subjected to another round of refinement with the values of occupancy and B factor for the water molecules set to unity and 15 Å2. Examination of the model after refinement showed that most of the water molecules maintained both good density and their hydrogen bond interactions. The exceptions were deleted, leaving 34 water molecules in the structure. The elbow angle for the ME36.1 Fab was determined using the program O (Jones et al., 1991) to find the rotation matrix for the least-squares superposition of the heavy chain variable domain atoms onto those of the light chain variable domain, and similarly for the constant domains, and then calculating the angle between the two rotation axes.
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE Modeling of a Sugar in the ME36.1 Antigen Binding Site A four-residue model sugar (Fig. 2), representative of GD2 (Fig. 1), was built using Quanta (1989) on a Silicon Graphics workstation. The sugar contains two sialic acid residues (NeuNAc), galactose (Gal), and an N-acetylgalactosamine (GalNAc) residue. The sugar was minimized to a local energy minimum by conjugate gradient minimization adapted for CHARMM (Brooks et al., 1987) to obtain a sterically correct starting model with which the ME36.1 binding site could be explored. The starting model for the sugar was adjusted to conform to the antigen binding site by manually placing it in several positions in the middle of the site and evaluating the fit. The conformation of the sugar was then adjusted to maximize favorable interactions, while avoiding molecular overlaps. One position that created extensive hydrogen bond interactions between the sugar and the side chain residues in the binding site was found. The position and molecular interactions were optimized further by several cycles of conjugate gradient minimization without imposing constraints on either the sugar or the Fab molecules. RESULTS
The ME36.1 Fab structure was determined by molecular replacement and refined to 2.8 Å resolution. The refinement statistics are shown in Table III. The final R factor of the model is 20.4% for 10 355 reflections (Fo $ 2s) in the resolution range 8.0–2.8 Å. The final R free value is 32.6%. The corresponding values for R factor and R free using the complete diffraction data without sigma cutoff (14 677 reflections) were 21.1 and 32.8%. The root-mean-square deviation from ideality in bond lengths is 0.008 Å and for bond angles is 1.7°. The average B factors for each domain are VL 22.9 Å2, VH 24.7 Å2, CL 24.4 Å2, and CH 25.1 Å2. The electron densities for the entire light chain and the variable region of the heavy chain are of good quality. The density for the heavy chain constant region is poorer, so that two loop regions in the CH1 domain (H129–H135, H158– H161) could not be fitted to the electron density and are presumed to be highly disordered. Disorder in TABLE III Summary of Refinement of ME36.1 Fab at 2.8 Å Resolution Resolution range R factor (%)a R free (%)b rms bond lengthc rms bond anglec rms improper anglesc rms dihedral anglesc Temperature factors (Å2) Main chain Side chain
8.0–2.8 Å 20.4 32.6 0.008 Å 1.7° 1.4° 28.0° 23.6 24.0
a R factor 5 (o hkl 0 Fo 2 kFc 0 ) /ohkl 0 Fo 0 , where Fo and Fc are the observed and calculated structure factors, respectively. b R free is the R factor for a test set of reflections, comprising 10% of the data, that was omitted from refinement. c Deviations from ideal stereochemistry using parameters of Engh and Huber (1991).
9
these regions has been observed for other Fab structures (Stanfield et al., 1990; Pokkuluri et al., 1994). Apart from these gaps, all but five amino acid side chains (Ser H136, Leu H139, Ser H157, Val H170, Lys H209) could be placed in density. Most of these residues are solvent exposed or occur at the beginning or the end of a gap. A Ramachandran plot (Ramachandran and Sasisekharan, 1968) of the (f, c) dihedral angles (not shown), prepared using the program PROCHECK (Laskowski et al., 1993), indicated that 99% of the 411 residues (213 in the light chain, 198 in the heavy chain) fall in allowed regions. Three of the five residues in the disallowed region are in the loop regions (Thr L50, Asn H54, Asn H156) and 2 lie at the end of the heavy chain (His H200, Pro H201). The ME36.1 Fab structure (Fig. 3) has the immunoglobulin fold common to all Fab structures. The value of 149.6° for the elbow angle between the variable and constant domains falls within the range for other Fab structures (Suh et al., 1986). The antigen binding site is formed by the six CDR loops H1, H2, H3, L1, L2, and L3 (Fig. 3). The antigen binding site is approximately 20 Å long (perpendicular to the VH:VL interface), 10 Å wide, and 8 Å deep. The dimensions are consistent with a groove-type binding site (Cisar et al., 1975) (Fig. 4a), as opposed to the pocket-like binding site seen in the sugar binding Salmonella O-antigen binding Fab (Cygler et al., 1991) (Fig. 4b). The L3 and H3 loops are in the middle of the binding site and form the bottom of the groove. The binding site contains eight aromatic residues, four phenylalanines (Phe L95, Phe L97, Phe H64, Phe H101), and four tyrosines (Tyr L93, Tyr H32, Tyr H60, Tyr H95). Phe L95, Phe L97, and Phe H101 form an aromatic pocket in the binding site, with Phe L97 at the bottom of this pocket (Fig. 5). A high proportion of the other antigen binding site residues are neutral amino acids, but there are several charged residues that occur within and immediately surrounding the binding site (His H35, Asp H50, His L33, Arg L90). In order to characterize the binding of ME36.1 to GD2, crystallization of the complex was attempted. This was dependent on obtaining a soluble form of the GD2 antigen, which was accomplished by enzymatic digestion of the ganglioside with ceramide glycanase to cleave the ceramide group from the antigenic carbohydrate portion of GD2 (Li and Li, 1989; Pichla et al., 1995). Initial cocrystallization trials of the Fab with a soluble form of GD2 produced thin plate-like crystals, but additional trials failed to yield crystals with a size sufficient for diffraction. Attempts were also made to soak the soluble GD2 into native Fab crystals. Addition of small molar quantities of the carbohydrate over several days to
10
PICHLA, MURALI, AND BURNETT
FIG. 3. The overall folding of the ME36.1 Fab molecule. The orientation is parallel to the VH:VL pseudo-twofold axis. The ribbon diagram was created with RIBBON (Priestle, 1988). The heavy chain (yellow) contains the VH and CH1 domains. The light chain (blue) contains the VL and CL domains. The complementarity determining region (CDR) loops at the top of the molecule are labeled (H1, H2, H3, L1, L2, L3). The L3 and H3 loops form the bottom of a groove in the antigen binding site that is 20 Å long, 10 Å wide, and 8 Å deep. Each domain has the immunoglobulin fold found in all antibody molecules. The elbow angle between the variable and constant domains is 149.6°. FIG. 4. (a) The antigen binding region of ME36.1, looking down onto the binding site perpendicular to the VH:VL pseudo-twofold axis. The protein surface is displayed in shades of green (for convex surfaces) to gray (for concave surfaces). The modeled sugar is shown in the binding site in stick representation with yellow carbons, red oxygens, and blue nitrogens. The groove-type binding site of the ME36.1 Fab runs across the upper surface, from lower right almost to the upper left of the molecule. This topography is in contrast to the (b) pocket-type binding site of the Salmonella O-antigen binding Fab Se155-4 (Cygler et al., 1991), where the conformation of the antigenic sugar is known from the crystal structure of the complex. Both figures were created with GRASP (Nicholls et al., 1991).
the solution containing the native Fab crystals produced visible etching. The crystals also lasted longer and diffracted X-rays to a slightly higher resolution than the unsoaked crystals. Both observations indicate that structural changes were occurring in the crystals. However, difference maps, phased with the native ME36.1 model, had no interpretable density in the antigen binding site indicating that the sugar was present. In the absence of experimental results for the complex, the binding of the sugar in the antigen binding site was modeled, guided by struc-
tural information from other known protein–sugar crystal structures. The objective was to elucidate the potential sugar contacts and the analysis does not claim to provide a definitive description of the complex. The carbohydrate portion of the GD2 ganglioside (GalNAcb1 = 4Gal[3 ; 2aNeuNAc8 ; 2aNeuNAc]b1 = 4Glcb1 = 1Cer) (Fig. 1) contains five sugar residues (Thurin et al., 1987): two negatively charged, sialic acid residues (NeuNAc), a galactose residue (Gal), an N-actetylgalactosamine residue
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE
11
FIG. 5. The complementarity determining region. The stereo diagram of the CDR loops was created with SETOR (Evans, 1993). The molecule (rotated ,60° clockwise from Fig. 4a) is viewed looking down onto the antigen binding site. The heavy chain CDR loops (H1, H2, H3) are yellow, and the light chain loops (L1, L2, L3) are blue. The binding site is highly aromatic, with four phenylalanines (Phe H64, Phe H101, Phe L95, Phe L97) and four tyrosine residues (Tyr H32, Tyr H60, Tyr H95, Tyr L93). An aromatic pocket in the binding site is created by residues Phe L95, Phe L97, and Phe H101. Although the amino acids in the binding site are mostly neutral, several are charged (Asp H50, Arg L90, His H35, His L33). FIG. 6. Stereo view of the sugar in the binding site of the ME36.1 Fab. The figure was produced with the program SETOR (Evans, 1993). The molecule is rotated ,180° from Fig. 5. Only the CDR loops, and the residues forming hydrogen bonds with the sugar are displayed. The CDR loops are colored as in Fig. 5. The sugar (green) binds along the longest dimension of the Ab binding site, perpendicular to the pseudo-twofold axis of the VH:VL interface. The seven residues that form hydrogen bonds with the sugar (Asn L31, Ser L49, Arg L90, Tyr L93, Ser H100, Thr H33, Asn H52) are displayed in red.
(GalNAc), and a glucose residue (Glc) that links to the ceramide group. Several observations identify the sialic residues as major contributors to the specific affinity of ME36.1 for GD2. This simplification made the modeling study a feasible task. The ME36.1 antibody has the highest affinity for GD2, but also cross-reacts with lower affinity to GD3 (NeuNAca2 = 8NeuNAca2 = 3Galb1 = 4Glcb1 = 1Cer) (Thurin et al., 1987). The two sugars are identical except that GD2 contains a branched N-acetylgalactosamine (GalNAc) (Fig. 1). For the ME36.1 antibody to cross-react with GD2 and GD3, and with no other ganglioside that contains a termi-
nal sialic acid (NeuAc) residue, ME36.1 must be interacting with both sialic acid residues. The higher affinity for the GD2 ganglioside is most likely due to the additional interactions (hydrogen bonding, van der Waals, and stacking) provided by the GalNAc residue. Therefore, the sugar used as the model consisted of the two sialic acid residues, the galactose residue, and GalNAc. Other complex gangliosides contain sugars with a2–8-linked sialic acid residues but the additional sugars are linked to the GalNAc and have a longer branched chain than GD2, which presumably prevents them from binding to ME36.1.
12
PICHLA, MURALI, AND BURNETT
There were several clues to where the sugar should be placed in the antigen binding site. It is known that iodination of the tyrosines in ME36.1 antibody destroys its ability to bind to GD2, suggesting that at least one tyrosine is involved in binding (Dr. Dorothy Herlyn, personal communication). As there is no evidence to suggest that the galactose residue is involved in any interactions with the antibody, this could be modeled away from the binding site. The H3 and L3 loops were targeted as the major loops involved in binding since they are known to be important for specificity (Davies and Padlan, 1990; Stanfield et al., 1993), and a tyrosine (Tyr L93) is on the L3 loop. The J539 Fab used in the molecular replacement search binds to galactan residues. As the sequence of the L1 loop is identical in J539 and ME36.1, this loop may be important in sugar binding. It is noteworthy that carbohydrate binding proteins often have two distinct globular domains that form a cleft in which the sugar binds (Quiocho, 1986). This arrangement is analogous to that in antibodies, where the variable light and heavy domains form an antigen binding site in the shape of a cavity or groove. The preceding observations established four main criteria for modeling: (1) residues from the H3, L3, and L1 loops should interact with the sugar; (2) a tyrosine should interact with the sugar; (3) the sugar should bind in the groove between the variable light and heavy domains; (4) the galactose residue would lie away from the binding site. An obstacle in modeling the binding of an antigen in a Fab binding site is that it is difficult to predict whether it will occur via an induced fit or a lock-andkey mechanism. It seems most likely that the flexible GD2 sugar will undergo a conformational change as it adopts its antigenic conformation. Moreover, in the cases where structures of both a native Fab and its complex are available, so that movements in the antigen binding site can be observed, at least a small movement occurs (Pokkuluri et al., 1994). In other cases, a significant movement of the loops occurs (Stanfield et al., 1993; Rini et al., 1992), and in several cases the VL and VH domains even move with respect to one another upon antigen binding (Stanfield et al., 1993; Herron et al., 1991; Bhat et al., 1990). Given this spectrum of possible movements, it seemed reasonable to assume that both the Fab and the sugar move during complex formation. Thus, binding was modeled by first minimizing the sugar to a starting model free of steric hindrance. The sugar then was moved manually around the CDR to find possible positions for the antigen binding site. The sugar conformation was adjusted at each position to avoid overlap with protein side chains. One particularly favorable position was found where
several hydrogen bonds were formed between the sugar and the residues in the binding site. Conjugate gradient minimization on the complex (the sugar and the Fab), without constraints on either molecule, then optimized the interactions at this position. Initially, it was thought that a sugar residue should be placed in the aromatic pocket of the binding site (Fig. 5). However, it was immediately clear from the modeling that this could not occur without moving the protein backbone and making a major rearrangement of the side chain residues to avoid steric interference. The best fit for the sugar was obtained by placing it in the groove in the middle of the binding site that runs perpendicular to the pseudo-twofold axis relating the VH and VL domains. In this position, extensive hydrogen bond interactions would occur between the sugar and the side chain residues from five of the six CDR loops of the Fab. Seven residues in the antigen binding site would have hydrogen bond interactions with the sugar (Fig. 6). The terminal sialic acid residue would accept three hydrogen bonds from surrounding side chain residues. The O7 oxygen of the glycerol moiety of the sugar would accept a hydrogen bond from Asn H52 of the H2 loop, Tyr L93 of the L3 loop donating a hydrogen bond to the O4 ring oxygen, and Thr H33 of the H1 loop would donate a hydrogen bond to the negatively charged carboxyl of the sugar. The O4 ring oxygen of the second sialic acid residue would accept a hydrogen bond from Ser H100 of the H3 loop. The carbonyl of the N-methylacetamide of this same sialic acid would accept a hydrogen bond from the Arg L90 of the L3 loop. The GalNAc residue would interact with two surrounding residues. The O4 ring oxygen of GalNAc would form hydrogen bonds with the framework residue, Ser L49, of the L2 loop, and Asn L31 would donate a hydrogen bond to the carbonyl oxygen of the amide group of this same sugar. In this configuration, the galactose residue would project from the binding site into the solvent and not interact with the Fab. After the final minimization, the root-mean-square deviation for all atoms of the variable region of the Fab from their original positions was 0.07 Å. The total buried surface area, determined using a probe radius of 1.7 Å, was 413 Å2 for the Fab and 526 Å2 for GD2. Placement of the water molecules in the ME36.1 Fab structure was not attempted until the Fab model was well-refined to a reasonable R factor and showed good stereochemistry. At the comparatively low resolution of 2.8 Å, very strict criteria were employed to define the set of 34 water molecules in the final structure. In this set, 2 water molecules lie in the antigen binding site of the native Fab structure, where they are close to several residues in-
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE
volved in binding the sugar. One water forms a hydrogen bond with Asp H50, which is in the vicinity of residue H52 that interacts with the sugar. The other water forms hydrogen bonds with Ser L91 and Ser L30, which are both next to two residues interacting with the sugar (Asn L31, Arg L90) (data not shown). The locations of the water molecules are such that they could remain in the binding site to coordinate hydrogen bond interactions between the sugar and the Fab, thus increasing the stability of the complex. However, as only a slight change in the torsion angles for the sugar would cause these waters to be displaced, this possibility cannot be excluded. The model for the complex explains the higher affinity of ME36.1 for GD2 than for GD3, which is due to the manner in which the sugar is anchored in the binding site. On one end of the binding groove, a sialic acid residue forms hydrogen bonds with residues from the H1 and H2 loops. At the other end, the GalNAc residue is involved with the L1 and L3 loops and is in close proximity to the residue in the L1 loop (Ser L30) that hydrogen bonds to a water molecule. This water molecule could coordinate hydrogen bonding between the L1 loop and the GD2 sugar. Thus, the presence of the GalNAc residue in GD2 not only contributes additional contacts with ME36.1, but also limits the conformational flexibility of GD2, which may allow it to maintain a more stable antigenic conformation. DISCUSSION
Comparative studies of antibody structures and their sequences have shown that the main chain conformations for five (L1, L2, L3, H1, H2) of the six CDR loops are limited to only a few possible spatial arrangements (Chothia et al., 1989). That is, each loop has a distinct conformation, but the same loop can be very similar in different Fabs. The conformations of the ME36.1 Fab antigen binding loops L1, L2 and H1, H2 correspond well to the canonical forms corresponding to their sequences (Chothia et al., 1989). However, the L3 loop sequence and structure are distinct from other Fab structures reported. Both the L3 loop and the H3 loop, which has no known canonical structure, occur close to the recombination site for the variable and joining regions of the L and H chains (Roitt et al., 1989). This explains the high variability in the sequences of these loops compared to the others, and it is reasonable to assume that their conformations would be equally as variable (Kabat et al., 1983). Attempts to humanize the ME36.1 Ab were unsuccessful in producing a high-affinity Ab to GD2. Once the mouse ME36.1 native Fab structure was available, it was used to identify residues in the human
13
framework region that could be important in maintaining the affinity for GD2. However, the structure did not provide additional clues beyond those that were already intuitively obvious (Dr. Jose Saldanha, personal communication). For example, several residues in both the light and heavy chain framework regions of the human Ab were changed to the mouse sequence, as they were thought important for maintaining correct loop conformations (Chothia et al., 1989). Other residues that were thought to influence the packing of the VH–VL interface were changed, as the interaction of these two domains affects the spatial arrangement of the CDR loops. In addition, framework residues that might contact antigen and residues putatively identified as glycosylation sites were also examined. Several mouse antibodies have successfully been humanized while maintaining high antigen specificity (Kolbinger et al., 1993), but inexplicably, others cannot be humanized. The modeling of the sugar in the binding site of the ME36.1 Fab structure was based on the types of interactions observed in other protein–carbohydrate structures (Vyas, 1991; Quiocho, 1986; Bundle and Young, 1992). Most of the interactions between the protein and carbohydrate are due to extensive hydrogen bonding and van der Waals forces (Vyas, 1991; Quiocho, 1986). Additionally, stacking interactions, carboxylates, and water molecules have been implicated in protein–carbohydrate binding. In most of the non-Fab protein–carbohydrate structures determined so far, the charged carboxylates of aspartate and glutamate and the uncharged polar asparagine and arginine residues are the most frequent acceptors of hydrogen bonds from sugar hydroxyl groups. Main chain polar groups also form hydrogen bonds with sugar residues (Vyas, 1991). All of these hydrogen bonding interactions facilitate the formation of van der Waals interactions by enabling polar residues in the binding region to approach the sugar (Vyas, 1991). Two antibody–carbohydrate structures have been solved to date, Se155-4–Salmonella O-antigen complex (Cygler et al., 1991) and BR96– nLey complex (Jeffrey et al., 1995). In contrast to the non-Fab protein–carbohydrate structures, the interaction of the sugar with the Fab occurs mostly through hydrogen bonding to tyrosine and tryptophan and to main chain atoms. In the ME36.1 binding site, the interactions of the sugar include charged planar groups (Arg L90, His L33), uncharged polar groups (Ser H100, Asn H52, Thr H33, Asn L31), and an aromatic residue (Tyr L93). The interactions are a combination of those seen in non-Fab–carbohydrate structures and those in Fab– carbohydrate structures. As the binding site of ME36.1 is highly aromatic, consistent with the high frequency of aromatic residues seen in antibody
14
PICHLA, MURALI, AND BURNETT
binding sites (Padlan, 1990), it is conceivable that more aromatic residues are involved in binding GD2 than were modeled. However, as these residues lie in the aromatic pocket, a significant rearrangement of the side chains or loops would be required upon sugar binding, which seems unlikely. In protein–carbohydrate complexes, aromatic residues are frequently involved in stacking interactions with the nonpolar face of the sugar ring, with tryptophan being the most common (Vyas, 1991). These stacking interactions confer specificity by constraining the sugar to a particular conformation by sterically hindering other configurations. In contrast, the aromatic residues in the two known Fab– carbohydrate structures are involved in hydrogen bond interactions rather than stacking interactions. The structure of the J539 galactan binding Fab (Suh et al., 1986), which was used as the search model to solve ME36.1, is only known in its native form and not as a complex with its antigen. The binding site of J539 is also highly aromatic, with two tryptophans and four tyrosine residues. Labeling experiments with deoxyfluorogalactoside and a modeling study with a galactan sugar suggested that antigen binding involves two solvent exposed typtophans (Suh et al., 1986). The interactions of the sole aromatic residue (Tyr L93) in the ME36.1 binding site with the GD2 sugar model are similar to those in other Fab–carbohydrate structures. Tyr L93 forms a hydrogen bond with the sugar, rather than being involved in stacking interactions. There are no tryptophans in the ME36.1 CDR loops, but other aromatic residues in the binding site (four tyrosines and four phenylalanines) could make stacking interactions with the GD2 sugar similar to those observed with tryptophan. Indeed, there are several tryptophan residues in the framework regions of ME36.1 that are in close proximity to the antigen binding site, and it is conceivable that these are involved in antigen binding. It is not uncommon for framework residues to be involved in antigen binding (Bentley et al., 1990). However, this is unlikely in the case of ME36.1 since changing human framework residues to tryptophans where these occur in the mouse sequence did not increase the affinity (Jose Saldanha, personal communication). Moreover, a significant rearrangement of the binding site would be necessary for the tryptophan residues to interact with GD2. Water molecules provide another mechanism for the formation of hydrogen bonds between sugars and protein side chain residues. In the case of antibody– antigen interactions, most if not all water molecules are displaced in an antibody binding site upon antigen binding (Davies and Padlan, 1990). However, the Se155-4–Salmonella O-antigen complex has a critical water molecule in the antigen binding
site that coordinates four hydrogen bonds between one of the sugar residues and several CDR side chain residues (Cygler et al., 1991). The two water molecules in the native ME36.1 Fab binding site lie close to several residues that interact with the sugar model and are in a position where they can coordinate hydrogen bonds between the sugar and side chain residues. However, a small change in the conformation of the sugar could just as easily cause water displacement upon antigen binding. Although the positions of the two water molecules are consistent with the modeled position of the sugar, and indeed lend it support, it is difficult to make a definitive statement about their role in antigen binding. It has been proposed that carbohydrate binding antibodies have either groove- or pocket-type binding sites (Cisar et al., 1975). The BR96 antibody has a deep pocket in its binding site that is 12 Å in diameter and 10 Å deep (Jeffrey et al., 1995). The Se155-4 antibody has a more shallow binding site, 7 Å in diameter and 8 Å deep, but clearly shows a pocket-type binding site (Fig. 4b) (Cygler et al., 1991). In comparison, the ME36.1 binding site is approximately 10 Å wide, 8 Å deep, and 20 Å long and creates a groove-type binding site that is long enough to accommodate the carbohydrate chain of GD2 (Fig. 4a). The contact area in the binding site for ME36.1 is 413 Å2, which is consistent with the value for the deep pocket of BR96 (422 Å2) and somewhat more than that for the shallow pocket of Se155-4 (304 Å2). The molecular model for the interaction of GD2 with ME36.1 Fab is consistent with all types of protein–carbohydrate interactions that have been observed and so presents a reasonable model to explain binding. All four criteria set for the modeling were satisfied: the L3, H3, and L1 loops interact with the sugar; binding involves a tyrosine; the sugar lies in a groove between the domains; and the galactose residue is outside the binding site and does not interact with the Fab. Although previous attempts to obtain cocrystals of ME36.1 Fab–GD2 complex were unsuccessful, trials are continuing. It will be interesting to see whether the interactions of the Fab–GD2 resemble those of the other two Fab–carbohydrate structures or encompass the wider range of interactions observed in other protein–carbohydrate structures. It will also be interesting to see how closely the model predicts the true interaction of the ME36.1 antibody with GD2. The atomic coordinates for both the ME36.1 Fab crystal structure and the GD2 model have been deposited in the Brookhaven Protein Data Bank (PDB Accession No. 1PSK).
FAB TO THE MELANOMA-ASSOCIATED GD2 GANGLIOSIDE We thank Drs. Albrecht Luckenbach and Maria Kordowicz (Merck KGaA, Darmstadt, Germany) for supplying generous quantities of purified ME36.1 monoclonal antibody and GD2 ganglioside. We also thank Dr. Jose Saldanha and Dr. Mary Bendig (MRC Collaborative Centre, London, UK) for helpful discussions about the humanization of the antibody. This research was supported by a grant to R.M.B. from the National Institute of Allergy and Infectious Diseases (AI 17270) and by Wistar Training Grant CA-09171 (S.L.P.). Further support was provided by E. Merck, Darmstadt, and the Wistar Cancer Center (CA 10815). REFERENCES Bentley, G. A., Boulot, G., Riottot, M. M., and Poljak, R. J. (1990) Three-dimensional structure of an idiotope-anti-idiotope complex, Nature 348, 254–257. Bhat, T. N., Bentley, G. A., Fischmann, T. O., Boulot, G., and Poljak, R. J. (1990) Small rearrangements in structures of Fv and Fab fragments of antibody D1.3 on antigen binding, Nature 347, 483–485. Brooks, B. R., Bruccoleri, R. E., Olafson, B. D., States, D. J., Swaminathan, S., and Karplus, M. (1983) CHARMM: A Program for macromolecular energy, minimization, and dynamics calculations, J. Comput. Chem. 4, 187–217. Bru¨nger, A. T. (1990) Extension of molecular replacement: A new search strategy based on Patterson correlation refinement, Acta Crystallogr. A46, 46–57. Bru¨nger, A. T. (1992a) X-PLOR Version 3.1: A system for X-ray Crystallography and NMR, Yale University, New Haven, CT. Bru¨nger, A. T. (1992b) Free R value: A novel statistical quantity for assessing the accuracy of crystal structures, Nature 355, 472–475. Bru¨nger, A. T., Krukowski, A., and Erickson, J. W. (1990) Slowcooling protocols for crystallographic refinement by simulated annealing, Acta Crystallogr. A46, 585–593. Bru¨nger, A. T., Kuriyan, J., and Karplus, M. (1987) Crystallographic R factor refinement by molecular dynamics, Science 235, 458–460. Bundle, D. R., and Young, N. M. (1992) Carbohydrate–protein interactions in antibodies and lectins, Current Opinion Struct. Biol. 2, 666–673. Collaborative Computational Project, Number 4. (1994) The CCP4 Suite: Programs for Protein Crystallography, Acta Crystallogr. D50, 760–763. Chothia, C., Lesk, A. M., Tramontano, A., Levitt, M., Smith-Gill, S. J., Air, G., Sheriff, S., Padlan, E. A., Davies, D., Tulip, W. R., Colman, P. M., Spinelli, S., Alzari, P. M., and Poljak, R. J. (1989) Conformations of immunoglobulin hypervariable regions, Nature 342, 877–883. Cisar, J., Kabat, E. A., Dorner, M. M., and Liao, J. (1975) Binding properties of immunoglobulin combining sites specific for terminal or nonterminal antigenic determinants in dextran, J. Exp. Med. 142, 435–459. Cygler, M., Rose, D. R., and Bundle, D. R. (1991) Recognition of a cell-surface oligosaccharide of pathogenic Salmonella by an antibody Fab fragment, Science 253, 442–445. Davies, D. R., and Padlan, E. A. (1990) Antibody–antigen complexes, Annu. Rev. Biochem. 59, 439–473. Engh, R. A., and Huber, R. (1991) Accurate bond and angle parameters for X-ray protein structure refinement, Acta Crystallogr. A47, 392–400. Evans, S. V. (1993) SETOR: Hardware-lighted three-dimensional solid model representations of macromolecules, J. Mol. Graphics 11, 134–138. Herlyn, M., and Koprowski, H. (1988) Melanoma antigens: Immu-
15
nological and biological characterization and clinical significance. Annu. Rev. Immunol. 6, 283–308. Herron, J. N., He, X. M., Ballard, D. W., Blier, P. R., Pace, P. E., Bothwell, A. L. M., Voss, E. W., Jr., and Edmundson, A. B. (1991) An autoantibody to single-stranded DNA: Comparison of the three-dimensional structures of the unliganded Fab and a deoxynucleotide–Fab complex, Proteins 11, 159–175. Iliopoulos, D., Ernst, C., Steplewski, Z., Jambrosic, J. A., Rodeck, U., Herlyn, M., Clark, W. H., Jr., Koprowski, H., and Herlyn, D. (1989) Inhibition of metastases of a human melanoma xenograft by monoclonal antibody to the GD2/GD3 gangliosides, J. Natl. Cancer Inst. 81, 440–444. Jeffrey, P. D., Bajorath, J., Chang, C. Y., Yelton, D., Hellstro¨m, I., Hellstro¨m, K. E., and Sheriff, S. (1995) The X-ray structure of an anti-tumour antibody in complex with antigen, Nature Struct. Biol. 2, 466–471. Jones, T. A. (1982). FRODO: A graphics fitting program for macromolecules, in Sayre, D. (Ed.), Computational Crystallography, pp. 303–317, Clarendon Press, Oxford. Jones, T. A., Zou, J.-Y., Cowan, S. W., and Kjeldgaard, M. (1991) Improved methods for building protein models in electron density maps and the location of errors in these models, Acta Crystallogr. A47, 110–119. Kabsch, W. (1988) Automatic indexing of rotation diffraction patterns. J. Appl. Crystallogr. 21, 67–71. Kabat, E. A., Wu, T. T., Reid-Miller, M., Perry, H. M., and Gottesman, K. S. (1987) Sequences of Proteins of Immunological Interest, 4th ed., Public Health Service, NIH, Washington, DC. Kolbinger, F., Saldanha, J., Hardman, N., and Bendig, M. M. (1993) Humanization of a mouse anti-human IgE antibody: A potential therapeutic for IgE-mediated allergies, Protein Eng. 6, 971–980. Laskowski, R. A., MacArthur, M. W., Moss, D. S., and Thornton, J. M. (1993) PROCHECK: A program to check the stereochemical quality of protein structures, J. Appl. Crystallogr. 26, 283–291. Lattman, E. E. (1972) Optimal sampling of the rotation function, Acta Crystallogr. B28, 1065–1068. Li, Y.-T., and Li, S.-C. (1989) Ceramide glycanase from leech, Hirudo medicinalis, and earthworm, Lumbricus terrestris, Methods Enzymol. 179, 479–487. Mayworth, R. D., and Quinta´ns, J. (1990) Designer and catalytic antibodies, N. Engl. J. Med. 323, 173–178. Nicholls, A., Sharp, K. A., and Honig, B. (1991) Protein folding and association: Insights from the interfacial and thermodynamic properties of hydrocarbons, Proteins: Struct. Funct. Genet. 11, 281–296. Padlan, E. A. (1990) On the nature of antibody combining sites: Unusual structural features that may confer on these sites an enhanced capacity for binding ligands, Proteins 7, 112–124. Pichla, S. L., Murali, R., and Burnett, R. M. (1995) Preliminary crystallographic data for an Fab to the melanoma-associated GD2 ganglioside, and the purification of a soluble form of this antigen, Acta Crystallogr. D51, 124–126. Pokkuluri, P. R., Bouthillier, F., Li, Y., Kuderova, A., Lee, J., and Cygler, M. (1994) Preparation, characterization and crystallization of an antibody Fab fragment that recognizes RNA, J. Mol. Biol. 243, 283–297. Priestle, J. P. (1988) RIBBON: A stereo cartoon drawing program for proteins, J. Appl. Crystallogr. 21, 572–576. Quanta (1989) Version 2.1, Polygen Corp., Waltham, Massachusetts. Quiocho, F. A. (1986) Carbohydrate-binding proteins: Tertiary
16
PICHLA, MURALI, AND BURNETT
structures and protein-sugar interactions, Annu. Rev. Biochem. 55, 287–315. Ramachandran, G. N., and Sasisekharan, V. (1968) Conformation of polypeptides and proteins, Adv. Protein Chem. 23, 283–438. Ravindranath, M. H., Tsuchida, T., and Irie, R. F. (1989) Diversity of ganglioside expression in human melanoma, in Oettgen, H. F. (Ed.), Gangliosides and Cancer, pp. 79–91, VCH, New York. Rini, J. M., Schulze-Gahmen, U., and Wilson, I. A. (1992) Structural evidence for induced fit as a mechanism for antibodyantigen recognition, Science 255, 959–965. Roitt, I. M., Brostoff, J., and Male, D. K. (1989) Immunology, 2nd ed., Gower Medical, New York. Stanfield, R. L., Fieser, T. M., Lerner, R. A., and Wilson, I. A. (1990) Crystal structures of an antibody to a peptide and its complex with peptide antigen at 2.8 Å, Science 248, 712–719. Stanfield, R. L., Takimoto-Kamimura, M., Rini, J. M., Profy, A. T., and Wilson, I. A. (1993) Major antigen-induced domain rearrangements in an antibody, Structure 1, 83–93.
Suh, S. W., Bhat, T. N., Navia, M. A., Cohen, G. H., Rao, D. N., Rudikoff, S., and Davies, D. R. (1986) The galactan-binding immunoglobulin Fab J539: An X-ray diffraction study at 2.6-Å resolution, Proteins 1, 74–80. Tai, T., Cahan, L. D., Tsuchida, T., Saxton, R. E., Irie, R. F., and Morton, D. L. (1985) Immunogenicity of melanoma-associated gangliosides in cancer patients, Int. J. Cancer 35, 607–612. Thurin, J., Thurin, M., Kimoto, Y., Herlyn, M., Lubeck, M. D., Elder, D. E., Smereczynska, M., Karlsson, K.-A., Clark, W. M., Jr., Steplewski, Z., and Koprowski, H. (1987) Monoclonal antibody-defined correlations in melanoma between levels of GD2 and GD3 antigens and antibody-mediated cytotoxicity, Cancer Res. 47, 1229–1233. Vyas, N. K. (1991) Atomic features of protein–carbohydrate interactions, Current Opinion Struct. Biol. 1, 732–740. Waldmann, T. A. (1991) Monoclonal antibodies in diagnosis and therapy, Science 252, 1657–1662.