Groove-type Recognition of Chlamydiaceae-specific Lipopolysaccharide Antigen by a Family of Antibodies Possessing an Unusual Variable Heavy Chain N-Linked Glycan*

Background: S25-26 displays remarkable specificity and avidity toward the unusual inner core LPS from Chlamydia. Results: Liganded and unliganded structures reveal bound antigen and significant ordered N-glycosylation on the variable heavy chain. Conclusion: Groove-type paratope recognizes the antigen in a manner distinct from all other Chlamydia-specific antibodies. Significance: Structural analysis of germ-line-coded antibodies provides insight to anaphylaxis induced by α-galactose (αGal) epitopes on therapeutic antibodies. The structure of the antigen binding fragment of mAb S25-26, determined to 1.95 Å resolution in complex with the Chlamydiaceae family-specific trisaccharide antigen Kdo(2→8)Kdo(2→4)Kdo (Kdo = 3-deoxy-α-d-manno-oct-2-ulopyranosonic acid), displays a germ-line-coded paratope that differs significantly from previously characterized Chlamydiaceae-specific mAbs despite being raised against the identical immunogen. Unlike the terminal Kdo recognition pocket that promotes cross-reactivity in S25-2-type antibodies, S25-26 and the closely related S25-23 utilize a groove composed of germ-line residues to recognize the entire trisaccharide antigen and so confer strict specificity. Interest in S25-23 was sparked by its rare high μm affinity and strict specificity for the family-specific trisaccharide antigen; however, only the related antibody S25-26 proved amenable to crystallization. The structures of three unliganded forms of S25-26 have a labile complementary-determining region H3 adjacent to significant glycosylation of the variable heavy chain on asparagine 85 in Framework Region 3. Analysis of the glycan reveals a heterogeneous mixture with a common root structure that contains an unusually high number of terminal αGal-Gal moieties. One of the few reported structures of glycosylated mAbs containing these epitopes is the therapeutic antibody Cetuximab; however, unlike Cetuximab, one of the unliganded structures in S25-26 shows significant order in the glycan with appropriate electron density for nine residues. The elucidation of the three-dimensional structure of an αGal-containing N-linked glycan on a mAb variable heavy chain has potential clinical interest, as it has been implicated in allergic response in patients receiving therapeutic antibodies.

The ability of the humoral immune system to recognize nonself relies on high levels of diversity in antibody combining sites, which stems from three major sources. The first is from the recombination of the VDJ (variable/diversity/joining) gene segments that code for the heavy chain and VJ gene segments that code for the light chain. The second is from the combinatorial potential of the heavy and light chains, which can profoundly influence the nature of the combining site typically found at their interface. The third is due to the phenomenon of "affinity maturation," where B cells can be stimulated by activated T-cells to subject their heavy and light chain genes to high levels of somatic mutation in an effort to produce mutant antibodies with higher avidity (1)(2)(3)(4).
Carbohydrate antigens are particularly interesting as they generally cannot induce affinity maturation (5)(6)(7)(8), and it has been suggested that the antibody repertoire has evolved to contain gene segments suited for immediate recognition of carbohydrate mono-, oligo-, and polysaccharides associated with pathogens (10 -12).
Although the pool of potential antigens is seemingly limitless, mammalian genomes code for a finite number of gene segments that are under constant pressure to confer a survival advantage, and the germ-line has evolved to code for an anti-body repertoire that can recognize common as well as new pathogens.
The correspondence between the amino acid residues in the heavy chain VDJ and light chain VJ gene segments and the nature of the combining site has come from three-dimensional structural studies, which have shown that antigen is bound via a cluster of polypeptide loops called "complementarity determining regions" (CDRs), 3 three on each chain labeled L1, L2, L3, H1, H2, and H3. The L1, L2, and L3 CDRs are coded by the light chain V gene, whereas H1 and H2 are coded by the heavy chain V gene. Significantly, CDR H3 is coded largely by the heavy chain D and J gene segments.
There has been considerable speculation about the relative importance of the six CDRs. Evidence of the highly evolved nature of the germ-line response to carbohydrate antigens and of the potential relationship between the V, D, and J gene segments is provided by the phenomenon of "V region restriction," where a given antigen will evoke a population of antibodies constructed from a small number of V genes combined with a number of different D and J genes (10,(12)(13)(14).
One set of studies that focus on the recognition of LPS from Chlamydia has provided clear corroboration of this hypothesis. Chlamydiaceae is a family of Gram-negative intracellular pathogenic bacteria, and Chlamydia trachomatis is a leading cause of sexually transmitted disease and blindness with an estimated 40 million active cases of trachoma (15). Chlamydial species possess an interesting truncated LPS that consists solely of an inner core composed of a few Kdo (3-deoxy-␣-D-manno-oct-2ulopyranosonic acid) residues and lipid A (16,17). All Chlamydiaceae contain the unique trisaccharide Kdo(238)Kdo-(234)Kdo (16, 18 -20), and Chlamydia psittaci contains a species-specific Kdo(234)Kdo(234)Kdo trisaccharide and a Kdo(238)[Kdo(234)]Kdo(234)Kdo tetrasaccharide antigen (21). The structures of oligosaccharide bisphosphates were isolated from the LPS of a recombinant Escherichia coli strain expressing the gene gseA (Kdo transferase) of C. psittaci 6BC. The lack of an effective vaccine against Chlamydia has spurred considerable efforts to generate high affinity antibodies against the antigenic LPS (22)(23)(24).
Significantly, the V region common to S25-2 type antibodies provides specific recognition of a single (usually terminal) Kdo residue (11,14,25,28,32) that confers to the combining site tremendous cross-reactive potential. This cross-reactivity is tempered by the nature of CDR H3, where an H3 directed away from the combining site has diminished influence over antigen binding, whereas one that infringes on the combining site can dictate fine specificity by allowing only a subset of antigens to bind. S25-2 itself has a short outward-leaning CDR H3 and is cross-reactive toward many antigens (14,30).
In contrast, the members of the S25-23 family of antibodies are specific for the full-length Kdo(238)Kdo(234)Kdo trisaccharide antigen, with no observable cross-reactivity with Kdo mono-or disaccharides or with the Kdo(234)Kdo(234)Kdo trisaccharide antigen (18,20,31). Furthermore, this group of antibodies proved to be recalcitrant to structure determination, with S25-23 itself eluding all attempts to crystallize it over an 18-year period.
The efforts to determine the three-dimensional structures of the S25-23-type antibodies were prompted first by their striking difference in germ-line gene usage from S25-2-type antibodies, their high affinity for antigen (20), and by their complete lack of cross-reactive potential, all of which require that S25-26 displays a distinct mechanism for antigen recognition. We now report binding data and crystal structures of liganded and unliganded S25-26 Fab as a step toward elucidating the recognition mechanism of S25-23-type antibodies.

EXPERIMENTAL PROCEDURES
ELISA and Isothermic Microcalorimetry (ITC)-Relative avidities of mAbs S25-23, S25-26, S25-2, and S25-39 were determined by ELISA as described (25,26). BSA-neoglycoconjugates of oligosaccharides isolated from recombinant bacteria (33,34) and from synthetic origin (18,31,(35)(36)(37)(38)(39) were prepared as published (28,40). The antigens were coated at 2 and 20 pmol/well calculated for the amount of immobilized oligosaccharide ligand. Affinities of S25-23 and S25-26 were measured with ITC experiments using a MicroCal iTC200 titration calorimeter (GE Healthcare). The mAbs were dialyzed against PBS, pH 7.4, and the concentrations were determined by UV measurements assuming A 280 ϭ 1.35 for a 1-mg/ml solution. The protein concentrations of S25-23 and S25-26 were 6 and 8 mg/ml, respectively. The concentrations in the measurement cell were corrected during the fitting routine such that the number of binding sites was 2.0. The ligands Kdo(238)-Kdo(234)Kdo(236)GlcNAc4P(136)GlcNAc1P (referred to as pentasaccharide bisphosphate (PSBP)) and the synthetic Kdo(238)Kdo-(234)Kdo O-allyl trisaccharide were dissolved at a concentration of 0.1 mM for S25-23 and at 0.3 and 0.52 mM for S25-26. The ligands were dissolved into the protein dialysis buffer before loading into the syringe of the microcalorimeter. All solutions were thoroughly degassed before loading. After temperature equilibration to 25°C, 19 injections, 2 l each of 4-s duration with 3-min equilibration times between injections, were recorded. Stirring speed was set to 1000 rpm. Data were corrected for heat of dilution by measuring the same number of buffer injections and subtraction from the sample data set. Data analysis was performed by fitting the data to the model of 1 set of binding sites using the ITC analysis software of Origin Version 7 SR4 (OriginLab Corp., Northampton, MA).
Germ-line Gene Usage Analysis-The nucleotide sequences of the variable region were analyzed with the IMGT/V-quest and junctional analysis web applications (41,42) to determine the murine germ-line gene segments from which the S25-23 and S25-26 antibodies were derived.
Fab Preparation and Crystallization-The S25-26 mAb was isolated from ascites grown in female Balb/c mice (Harlan Laboratories, Inc.). Fab fragments were prepared by digestion of the intact IgG with papain (Sigma). IgG was dialyzed into 20 mM HEPES (Sigma), pH 7.5, diluted to a concentration of 0.8 mg/ml, 2 mM EDTA (Sigma), and 3 mM DTT (Sigma). The digestion reaction was carried out at room temperature using a papain-to-IgG ratio of 1:200 (in mg) for 3 h. The reaction was quenched by the addition of 20 mM iodoacetamide (Sigma) and dialyzed overnight into 20 mM HEPES buffer, pH 7.5. The Fab fragment was purified by cation-exchange chromatography on a Shodex CM-825 column (Phenomenex) using a linear gradient of 0.0 to 0.5 M NaCl in 20 mM HEPES, pH 7.4.
Data Collection, Molecular Replacement, and Structure Refinement-Crystallization conditions for the liganded structure already contained appropriate levels of cryoprotectant. The crystal was flash-frozen to Ϫ160°C using an Oxford Cryostream 700 crystal cooler (Oxford Cryosystems). Initial data sets were collected on a Rigaku R-AXIS IVϩϩ area detector (Rigaku, Japan) coupled to a MM-002 x-ray generator with Osmic "blue" optics (Rigaku Americas, The Woodlands, TX) and processed using Crystal Clear/d*trek (Rigaku). These data were solved by molecular replacement using Phaser (43) with the variable fragment (Fv) from NNA7 (PDB code 1T2Q) and constant domains from S25-39 (PDB code 3OKD) as search models. Higher resolution data from liganded crystals were collected at the Canadian Macromolecular Crystallography Facility on beamline 08ID-1 (CMCF-ID) of the Canadian Light Source (Saskatoon, SK, Canada) at 0.979 Å wavelength with a Marmosaic CCD300 detector. Manual fitting of -A-weighted F o Ϫ F c and 2F o Ϫ F c electron density maps was carried out with Coot (44). The crystal for unliganded structure 1 contained appropriate levels of cryoprotectant. Crystals for unliganded structures 2 and 3 were dehydrated in a 16°C room until concentration of cryoprotectant (calcium acetate and PEG 3350 respectively) reached appropriate levels. All data collection for unliganded crystals were collected at the Canadian Light Source and solved using the Fv and the constant domains of the liganded S25-26 structure as a model and processed using HKL2000 (HKL Research Inc.). Restrained refinement and translation, libration, and screw (TLS) refinement was carried out using REFMAC5 (45,46). All stereo figures and r.m.s.d. calculations presented in this paper were made using Seto-Ribbon. 4 Electrostatic potential surface figure was made using Chimera molecular visualization software (47). Marvin v5.7.0, from ChemAxon was used for drawing chemical structures. The ribbon diagram of S25-2 and S25-26 was made using PyMOL molecular graphics software, v1.5.0.4, Schrödinger (48).
Hydrophilic interaction liquid chromatography-ultra performance liquid chromatography was carried out on an ethylene-bridged hybrid glycan 1.7 M 2.1 ϫ 150-mm column (Waters) on an Acquity UPLC (Waters) equipped with a Waters temperature control module and a Waters Acquity fluorescence detector. Solvent A was 50 mM formic acid adjusted to pH 4.4 with ammonia solution. Solvent B was acetonitrile. The column temperature was set to 25°C. The following conditions were used: 30 min run using a linear gradient of 30 -47% A at 0.56 ml/min in 23 min followed by 47-70% A and finally reverting to 30% A to complete the run. Samples were injected in 70% acetonitrile. Fluorescence was measured at 420 nm with excitation at 330 nm. The system was calibrated using an external standard of hydrolyzed and 2-aminobenzamidelabeled glucose oligomers to create a dextran ladder, with retention times of all identified peaks expressed as glucose units, as described previously (49). Weak anion exchange HPLC to separate the N-glycans by charge was carried out as detailed in Royle et al. (49) with a fetuin N-glycan standard as reference.
Exoglycosidase digestions were carried out on 10% aliquots of the total labeled N-glycan pools in accordance with methods previously described (49)  After incubation, enzymes were removed by filtration through a protein binding EZ filter (Millipore Corp., Bedford, MA). All exoglycosidase enzymes were obtained from Prozyme.
Electrospray Mass Spectrometry-Aqueous non-labeled glycan samples were cleaned with a Nafion membrane (Aldrich) (52) and after about 1 h were diluted into Milli-Q water containing 0.1 M ammonium phosphate (4 l, to form phosphate adducts of the neutral glycans) and methanol (5 l). Negative ion electrospray mass spectrometry was performed with a Waters Synapt G2 quadrupole-time-of-flight mass spectrometer (Waters MS Technologies, Manchester, UK) with nanospray sample introduction as described earlier (53). Spectral interpretation was as described in the reference by Harvey et al. (54).

Characterization of Chlamydiaceae Family-specific Antibodies through ELISA and ITC-
The avidities of selected S25-23and S25-2-type antibodies to various chlamydial LPS structures ( Fig. 1A) were determined by ELISA using Kdo oligosaccharides that were isolated from deacylated LPS of recombinant bacteria, which expressed chlamydial Kdo transferases. These ligands contained the lipid A backbone in addition to Kdo. To determine the cross-reactive potential, these binding assays were performed at low (2 pmol/well) and high (20 pmol/well) coated antigen concentrations (Table 1), and the results were compared with previously published binding data using oligosaccharides of synthetic origin without the lipid A backbone (18,25,26,31). ITC experiments carried out for S25-23 and S25-26 mAbs (Fig. 1B) revealed affinities ( Table 2)  X-ray Diffraction Data Collection, Solution, and Refinement-Data collection and final refinement statistics for liganded and unliganded structures of monoclonal antibody S25-26 are given in Table 3. Data were collected to 1.95 Å resolution for crystals of S25-26 grown in the presence of Kdo(238)Kdo-(234)Kdo O-allyl trisaccharide (Fig. 1C) in a trigonal unit cell with systematic absences indicating space group P3 1 21 or P3 2 21 with an R sym of 0.064. The structure was solved in space group P3 2 21 by molecular replacement with one molecule in the asymmetric unit.
Excellent electron density could be observed for the entire trisaccharide ligand (Fig. 1D) and for all amino acid residues in the combining site. Appropriate electron density was seen for the remainder of the protein, with the exception of Framework Region 3 of the V L chain and residues 127-138 and 181-193 on the heavy chain constant domain. Data were collected from three different crystal forms of unliganded S25-26. Unliganded 1 is orthorhombic P2 1 2 1 2 1 with data collected to 2.75 Å resolution with an R sym of 0.087. Unliganded 2 and 3 are both monoclinic C2 with data collected to 2.09 and 2.35 Å resolution, with an R sym of 0.114 and 0.071, respectively. The unliganded forms have 2, 2, and 3 molecules per asymmetric unit, respectively.
The V H domain of molecule 3 in Unliganded 3 had significantly higher thermal motion, which manifested as poor electron density in solvent-exposed regions, particularly in CDR H2 and the Framework Region 3 (an area also commonly referred to as "CDR H4" for its high variability compared with other framework regions). A few of the terminal amino acid residues were also disordered in the heavy chain of this domain and were not included in the final model.
N-Linked Glycosylation-Inspection of the electron density maps of all structures of S25-26 found that each molecule contained an N-linked glycosylation site on Asn(H)-85 (residues are identified as H or L to denote the heavy chain and light chain, respectively) with clear density at the outset of refinement for a single covalently bound N-acetylglucosaminyl (GlcNAc) residue.
N-Glycan structures were determined by a combination of HPLC, exoglycosidase digestions, and negative ion electrospray mass spectrometry. All major glycans were found to have corefucosylated biantennary complex structures with antennae terminating in Gal or ␣Gal-␤Gal or glycolylneuraminic acid (Neu5Gc)-␣Gal (Fig. 2, A-C). The determined composition of the N-linked glycans is shown in Tables 4 and 5 Table 5), of which just more than half (31.1% of the total) contain the fucosylated biantennary glycan with antennae terminating in ␣Gal and Neu5Gc (FA2G2Ga (3)[3]1S1 structure; Peak 18 in Table 5, Fig. 2A). The electrospray spectrum showed the same major glycans, but the minor compounds were obscured by peaks from contaminating compounds (Fig. 2B). Weak anion exchange-HPLC analysis indicated predominantly neutral (47.3%) and mono-sialylated (44.1%) glycan content, with a relatively low abundance (8.34%) of di-sialylated residues (Fig. 2C). No tri-or tetra-sialylated antennary structures were detected.
With the knowledge of the composition of the major glycans, a significant number of sugar residues could be modeled into the electron density of Unliganded 3 structure (Fig. 2, D-F). The N-linked mono-antennary oligosaccharide FA1G [3]1Ga(3)1 was successfully modeled into electron density of molecule 1 of Unliganded 3 ( Table 5, Fig. 2D). No electron density corresponding to ordered carbohydrate residues could be seen beyond the mannose core structures in molecules 2 and 3 (Fig.  2, E and F). Unambiguous electron density beyond the initial GlcNAc residue could not be seen in other crystal forms of S25-26.
Germ-line Gene Usage and Sequence Comparison-The amino acid sequences for the variable regions of two S25-2-type and the two S25-23-type antibodies S25-23 and S25-26 are given in Table 6. The germ-line gene segment designations for the CDR loops are shown in Table 7

S25-26 Uses a Recognition Strategy Fundamentally Different
from S25-2-Sequence analysis of the antibodies produced by immunization with Kdo(238)Kdo(234)Kdo(236)GlcNAc tetrasaccharide conjugated to BSA (31) showed two major groupings of V gene usage (27), where most belonged to the S25-2 group with a smaller number belonging to the S25-23 group (Tables 6 and 7). The differences in the sequences corresponding to the combining sites were so striking that it was clear that the two groups must use markedly different strategies
The cross-reactive behavior of S25-2-type antibodies was associated with a deep pocket specific for the Kdo monosaccharide that dominated binding (Fig. 4B) and could accommodate Kdo or Kdo-like moieties from different antigens (11,14,18,22,25,26,28,30,32,55). The specificity of each mAb in this group was shown to be governed largely by CDRs H3-coded by the heavy chain D and J gene segments, and different S25-2-type mAbs were identified that recognized variants and fragments of the chlamydial LPS antigens.
In contrast, the combining site of S25-26 has a groove along the heavy and light chain interface in which the Kdo(238)-Kdo(234)Kdo trisaccharide antigen lies in an extended conformation (Fig. 1B) such that each of the three Kdo residues have a comparable number of interactions to protein (Fig. 3, C and D).
Although S25-26 has approximately the same avidity toward the Kdo(238)Kdo(234)Kdo trisaccharide antigen as the S25-2-type antibody S25-39 (Table 1), the starkly different pattern of hydrogen bonding explains the specific nature of S25-26 over the cross-reactive nature S25-39. Both antibodies use direct hydrogen bonds between protein and carbohydrate as well as bridging water molecules. The total number of direct hydrogen bonds to antigen is approximately the same with 13 for S25-26 and 14 for S25-39 (26); however, although S25-26 makes 5, 3, and 5 hydrogen bonds, respectively to KdoI, There is also a striking difference in the utilization of bridging water molecules in the two structures, where S25-26 relies more heavily on water molecules to form a complementary combining site (Fig. 3, C and D). Although S25-39 forms 7 bridging interactions to KdoII and KdoIII from 6 water molecules, S25-26 forms 16 bridging interactions to all three sugar residues from 10 water molecules.
S25-26 displays another feature lacking in S25-2-type antibodies, which is multiple hydrophobic contacts between the protein and the back face of a Kdo residue. Fig. 4A shows that KdoII is packed against the aromatic ring of Trp(H)-52 with a separation of 3.65-3.85 Å. Hydrophobic interactions from tryptophan, tyrosine, and phenylalanine rings to the back face of carbohydrate residues have been observed in other structures of carbohydrate-antibody complexes (56,57) including the first such structure published (58), and have been shown to favor antigen binding through entropic gains (59).
Finally, the groove-type architecture of S25-26 allows the antigen to lie across the combining site such that recognition takes place not only through the identity of each sugar residue but through the length of the glycosidic linkage (Figs. 1B and 3,  C and D). In contrast, the pocket type architecture of the S25-2-type antibodies allows for effective stereochemical recognition of just the terminal residue, so that S25-39, for example, binds not only the family-specific Kdo(238)Kdo(234)Kdo trisaccharide antigen but the Kdo(234)Kdo(234)Kdo trisaccharide antigen from C. psittaci as well.
There are three distinct examples of S25-2-type antibodies that are specific for the Kdo(234)Kdo(234)Kdo trisaccharide antigen; however, in these cases specific recognition of the antigen develops through separate affinity maturation events that positioned a phenylalanine residue into the combining site such that a hydrophobic stacking interaction to the middle Kdo occurs, simultaneously restricting the conformational space to exclude other antigens (11,60).
Comparison to S25-23-The remarkable specificity of S25-23 toward the family-specific Kdo(238)Kdo(234)Kdo trisaccharide antigen led to it becoming the best characterized Chlamydiaceae family-specific antibody in the literature (18,20,23,31,35,61), but no crystals could be generated despite thousands of trials over the last 18 years. S25-23 and S25-26 have almost identical sequence with few mutations in the combining site (Table 6), and so S25-23 is likely to bind the antigen in a fashion similar to S25-26.
Surface plasmon resonance studies of S25-23 revealed a most unusual carbohydrate-specific antibody with high M affinity of the Fab toward the Kdo trisaccharide antigen (20). This result is confirmed with ITC, where K D values of 9.90 ϫ 10 Ϫ8 and 6.64 ϫ 10 Ϫ8 M were obtained for ligands PSBP and Kdo(238)-Kdo(234)Kdo O-allyl trisaccharide, respectively (Table 2, Fig.  1B). The K D values obtained for S25-26 for the same ligands were significantly lower at 1.55 ϫ 10 Ϫ6 and 7.81 ϫ 10 Ϫ7 M, respectively. The ELISA binding studies on the intact antibodies show the same trend. This difference in binding is remarkable given that the only relevant change in the combining site is the mutation on S25-26 of Asp(H)-58 to glutamate, which should allow S25-23 to form a direct hydrogen bond to the C4 hydroxyl of KdoII in place of a bridging water molecule (Fig.  3C).
A higher affinity of S25-23 over S25-26 toward the PSBP antigen (i.e. the trisaccharide antigen bound to the glucosamine backbone sugar residues of the lipid A moiety; Fig. 1 and Table  1) is observed at low concentrations; however, the molecular basis is not readily apparent, as there are no amino acid differences between the combining sites of S25-23 and S25-26 close to the lipid A attachment point. Unfortunately, diffraction quality crystals could not be obtained for S25-26 Fab in the presence of the pentasaccharide antigen.
The difficulty experienced in trying to crystallize S25-23 is particularly striking when considering that the homologous S25-26 crystallized in multiple space groups and unit cells. One clear factor in the relative ability of these antibodies to crystallize is residue Thr(L)-27A in S25-23, which is a tyrosine residue in S25-26 that participates in direct and/or water-bridged crystal contacts in all structures. This tyrosine residue is not coded in the germ-line and presumably arose from somatic hypermutation.
S25-23-type Antibodies Share V Gene Segments with Other Carbohydrate-specific mAbs-The S25-23-type antibodies utilize VL and V H genes similar to N-type blood group-specific antibody NNA7 (Table 8) that binds glycoprotein structures terminating in sialic acid residues (62). Superposition of mAb NNA7 with liganded S25-26 showed similar combining sites (Fig. 4C) with the exception of CDR H3. Interestingly, the two mAbs arise from the same heavy chain V gene, although NNA7 does not possess the heavy chain variable domain glycosylation site due to an Asn(H)-85 to glutamate mutation. Lewis Y-specific antibody BR96 (77) has a V L gene identical to S25-26 and displays the same high mobility about the C L :C H domains (63) as seen in the various structures of S25-26.
Glycan analysis of various antibodies demonstrated a wide heterogeneity of structures that was shown to be influenced by the expression system (68), albeit N-glycans with terminal ␣Gal reported here have not yet been found from antibodies derived from ascites fluid. Studies with the therapeutic antibody Cetuximab have shown that expression in murine cells can have N-acetylneuraminic acid (Neu5Ac) replaced by Neu5Gc (glycolylneuraminic acid) or varying amounts of terminal ␣Gal (69). The latter observation is of considerable clinical significance because of the creation of the so-called ␣Gal-epitope, which is involved in graft versus host reactions and has been   proposed to play an important role in the development of allergic responses and anaphylactic shock during treatment with therapeutic mAbs expressed in murine cells (69,70). A summary of the N-glycan structures found on Cetuximab compared with S25-26 is shown in Fig. 5. Beyond such dramatic consequences N-glycosylation has been shown to influence biological, physiochemical, and pharmacokinetic properties of antibodies. Therefore, knowledge about the primary and the three-dimensional glycan structures found in glycoproteins in general and in antibodies in particular is of considerable importance.
Approximately 10 -15% of IgG is glycosylated on the variable regions (71,72). Like most N-linked glycans, the carbohydrate structures are heterogeneous, which would be expected to hinder the formation of a crystal lattice, and great difficulty was experienced with liganded and unliganded crystallization trials of S25-23 Fab. Interestingly, all S25-23-type antibodies possess the same glycosylation site at Asn(H)-85. Although the biolog-   ical role of the glycosylation site on S25-23-type antibodies has not yet been explored, the N-glycans are positioned to modulate antibody-antigen binding events. Such a phenomenon was first demonstrated with hemophilia patient-derived mAb LE2E9 specific to the Factor VIII, where mutants lacking the N-linked glycans were reported to be 40% more potent in their inhibition of factor VIII (72,73).
Ordered VH Glycans within the Crystal Lattice of Unliganded 3-The predominant N-glycosylation structure FA1G [3]1Ga(3)1 could be modeled into the heavy chain of molecule 1 of Unliganded 3 (Fig. 2D) with poor density observed for the terminal galactose residues likely due to glycan heterogeneity and/or high thermal motion. The high observed average B-factors for the terminal Gal residues are also consistent with the low occupancy of these residues. The modeled N-glycan represents ϳ61% of the total structures determined. Significantly, the terminal sugar residues of the modeled glycans make contact to neighboring Fabs in the crystal lattice, which not only assists these flexible glycan structures to become ordered but may be key to crystallization.
In contrast to molecule 1, density beyond Man residues of the N-glycans could not be observed for molecules 2 and 3 of Unliganded 3 (Fig. 2, E and F). Nevertheless, the nine residue N-gly-can structure modeled for molecule 1 is one of the largest observed ordered in a Fab crystal structure to date.
Comparison with Germ-line Gene Segments-Although these antibodies experienced somatic hypermutation in both the light and heavy chain genes, determining the likely identity of the germ-line ancestors and hence germ-line paratopes is straightforward. As was found for S25-2, most of the residues involved in direct recognition of antigen for S25-26 and S25-23 are of germ-line origin, with the identities of only two residues different. The mutation of Ser(H)-31 to threonine likely has minimal impact on antigen recognition; however, the mutation of Met(H)-96 to the more flexible arginine allows significant new interactions to the central Kdo hydroxyl groups.
The modest number of mutations in the antibody-combining site is evidence of the presence in the germ-line of gene sequences coding for additional paratopes specific for Kdobased oligosaccharides. The apparent lower avidity of the germ-line combining site would, of course, be mitigated by the highly multivalent IgM. Interestingly, the mutation of MH96R in S25-26 is also present in S25-23. This change in amino acid identity is accomplished by a single point mutation in the gene, which significantly increases the avidity of these antibodies for the Kdo trisaccharide.
Conclusions-Given the general inability of carbohydratespecific antibodies to undergo somatic hypermutation in the absence of a peptide or protein carrier, it would be expected that some germ-line antibodies display a significant degree of cross-reactivity or polyspecificity to increase the number of potential antigens recognized, whereas others have evolved to recognize an epitope on a common pathogen in a specific manner. The former behavior is represented by S25-2-type antibodies, whereas S25-23-type antibodies are highly specific for the Kdo(238)Kdo(234)Kdo trisaccharide antigen. The structural and germ-line analysis of S25-26 demonstrates the molecular basis for a second germ-linecoded immunological response to Chlamydiaceae trisaccharide antigen.
The existence of at least two distinct germ-line antibody responses toward Kdo oligosaccharides reveals the importance of rapid recognition and response to LPS inner core. As observed from the crystal structure of S25-26, it is clear that the S25-23-type antibodies form combining site groove structures that allow for multiple hydrogen bond interactions to all three Kdo residues. Second, simultaneous recognition of KdoI and KdoIII is dependent on linkage length, cannot occur with the Kdo(234)Kdo(234)Kdo trisaccharide, and precludes the recognition of Kdo mono-or disaccharides.
Furthermore, all S25-26-type antibodies possess a glycosylation site covalently bound to Asn(H)-85. We have determined the N-linked glycans composition of S25-26 and a remarkable nine carbohydrate residues could be modeled on the glycosylation site in molecule 1 of Unliganded structure 3. Experiments are under way to determine the effect of glycosylation on antibody affinity.