Structure of a protective epitope of group BStreptococcus type III capsular polysaccharideFilippo Carbonia, Roberto Adamoa,1, Monica Fabbrinia, Riccardo De Riccoa, Vittorio Cattaneoa, Barbara Brogionia,Daniele Veggia, Vittoria Pintoa, Irene Passalacquaa, Davide Oldrinia, Rino Rappuolia,2, Enrico Malitoa,Immaculada y Ros Margarita,1, and Francesco Bertia,1,2

aGSK Vaccines, 53100 Siena, Italy

Contributed by Rino Rappuoli, March 27, 2017 (sent for review February 3, 2017; reviewed by Dennis L. Kasper and Robert J. Woods)

Despite substantial progress in the prevention of group B Strepto-coccus (GBS) disease with the introduction of intrapartum antibioticprophylaxis, this pathogen remains a leading cause of neonatalinfection. Capsular polysaccharide conjugate vaccines have beentested in phase I/II clinical studies, showing promise for further de-velopment. Mapping of epitopes recognized by protective anti-bodies is crucial for understanding the mechanism of action ofvaccines and for enabling antigen design. In this study, we reportthe structure of the epitope recognized by a monoclonal antibodywith opsonophagocytic activity and representative of the protectiveresponse against type III GBS polysaccharide. The structure and theatomic-level interactions were determined by saturation transferdifference (STD)-NMR and X-ray crystallography using oligosaccha-rides obtained by synthetic and depolymerization procedures. TheGBS PSIII epitope is made by six sugars. Four of them derive fromtwo adjacent repeating units of the PSIII backbone and two of themfrom the branched galactose–sialic acid disaccharide contained inthis sequence. The sialic acid residue establishes direct binding in-teractions with the functional antibody. The crystal structure pro-vides insight into the molecular basis of antibody–carbohydrateinteractions and confirms that the conformational epitope is notrequired for antigen recognition. Understanding the structural basisof immune recognition of capsular polysaccharide epitopes can aidin the design of novel glycoconjugate vaccines.

group B Streptococcus | capsular polysaccharide | antigen

Bacterial cell surface carbohydrates are the interface of mul-tiple host interactions and have been targeted to develop

highly efficacious glycoconjugate vaccines against severe infec-tions caused by Streptococcus pneumoniae, Haemophilus influ-enzae type b, and Neisseria meningitidis (1, 2). Glycoconjugatevaccines against other important pathogens are under clinical orpreclinical development (2).The mapping of polysaccharide (PS) epitopes recognized by

functional antibodies mediating protection from infection iscrucial for understanding the mechanism of action of this type ofvaccine. In many cases, the antigenic determinants of the im-munological properties of PS and the structural details of theminimal epitope targeted by specific functional antibodies areunknown. Structural biology has been commonly practiced in thelast decade for the characterization of protein antigen–antibodyinteractions (3). However, it has been less applied to carbohy-drate antigens, in part because of the well-known difficulty ofcrystallizing carbohydrates.Minimal epitopes can be composed of short, defined glycans

comprising 2–3 monosaccharides, as for the β-(1→2) mannans ofthe Candida albicans cell wall (4), Vibrio cholerae O1 (5), Shigellaflexneri variant Y (6), and Salmonella (7) O-antigens, or a tetra-saccharide, as for the repeating unit (RU) of S. pneumoniae type14 PS (Pn14) (8, 9), and even six sugar residues, as in the case ofS. flexneri serotype 2a O-antigen (10). In contrast, the type III PSof Streptococcus agalactiae (group B Streptococcus, GBS) has beenproposed as a prototype of a unique length-dependent confor-mational epitope (11).

GBS is an encapsulated Gram-positive β-hemolytic pathogencausing neonatal sepsis and meningitis, particularly in infants bornto mothers carrying the bacteria (12). The GBS capsular PS isconstituted by multiple RUs (from ∼50 up to 300 per polymer) offour to seven monosaccharides shaped to form a backbone and oneor two side chains. Ten serotypes presenting a unique pattern ofglycosidic linkages have been identified and their primary struc-tures elucidated (13). Three monosaccharides (β-D-glucopyranose,β-D-Glc; β-D-galactopyranose, β-D-Gal; and β-D-N-acetylglucosamine,β-D-GlcNAc) are present in all of the described serotypes, andsialic acid (α-N-acetyl-neuraminic acid, NeuNAc) is always foundat the terminus of one chain (13). Maternal concentrations of IgGdirected to the different GBS capsular serotypes inversely corre-late with the risk of newborn infection (14), suggesting that anti-bodies able to cross the placenta can confer serotype-specificinfant protection (15). PS conjugates of different serotypes elicitantibodies that mediate GBS type-specific opsonophagocytic kill-ing (OPK) in functional assays and protection against GBS chal-lenge in neonate mice (16–20). Furthermore, Ia, Ib, II, III, and Vmonovalent vaccines, as well as a trivalent combination (Ia, Ib,and III), have been proven to be safe and immunogenic in

Significance

This article describes the characterization of the antigenic de-terminant of the capsular polysaccharide from the clinicallyrelevant serotype III of group B Streptococcus (GBS). NMR andX-ray crystallography have been applied to elucidate the in-teraction of type III GBS oligosaccharides obtained by syntheticand depolymerization procedures of the bacterial poly-saccharide with a functional monoclonal antibody. A Fab–GBSoligosaccharide complex structure has been solved at highresolution (2.7 Å). The results demonstrate the existence of asialic acid-dependent functional epitope of GBS that is fullycontained within four consecutive sugars deriving from thetype III GBS polysaccharide backbone and one branched di-saccharide present in this sequence. This finding has implica-tions for the development of vaccines against GBS infection.

Author contributions: R.A., R.R., I.y.R.M., and F.B. designed research; F.C., M.F., R.D.R.,V.C., B.B., D.V., V.P., I.P., D.O., and E.M. performed research; F.C., R.A., M.F., R.D.R., R.R.,E.M., I.y.R.M., and F.B. analyzed data; and F.C., R.A., R.R., E.M., I.y.R.M., and F.B. wrotethe paper.

Reviewers: D.L.K., Harvard Medical School; and R.J.W., University of Georgia.

Conflict of interest statement: All authors are employees of GSK Vaccines (known asNovartis Vaccines at the time of the study).

Freely available online through the PNAS open access option.

Data deposition: The crystal structure has been deposited in the Protein Data Bank, www.pdb.org (PDB ID code 5M63) (deposition ID D_1200001992).1R.A., I.y.R.M., and F.B. contributed equally to this work.2To whom correspondence may be addressed. Email: rino.x.rappuoli@gsk.com orfrancesco.x.berti@gsk.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1701885114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1701885114 PNAS | May 9, 2017 | vol. 114 | no. 19 | 5017–5022

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

23,

202

0

nonpregnant and pregnant women, and elicited maternal anti-bodies were efficiently transferred to neonates (21–23).GBS strains belonging to the serotype III (GBSIII) are epide-

miologically the most relevant in neonatal infections (24). Pio-neering immunological studies aimed at elucidating type III-specificantigenic determinants indicated the presence of the terminalNeuNAc residue as essential for the elicitation of protective an-tibodies (25). Subsequent studies revealed that sera from rabbitsimmunized with GBSIII bacteria and from humans vaccinatedwith PSIII conjugates contained two types of anticarbohydrateantibodies—that is, a major population recognizing the native PSbut not its incomplete core antigen derivative lacking sialic acid(corresponding to the S. pneumoniae PS group 14, Pn14), and aminor variable population reacting with both the native PSIII andthe core antigen (26, 27). Remarkably, both types of human PSIII-induced antibodies were shown to mediate GBSIII OPK, whereasantibodies elicited by Pn14 (desialylated PSIII) did not recognizeGBSIII bacteria and therefore did not mediate GBS OPK (27).Studies using 13C NMR spectroscopy highlighted ring-linkage

signal displacements in the core versus the native PS, suggestingthat NeuNAc residues exert a specific control over the conformationof the native PS (26, 28). Molecular dynamics simulations confirmeda more flexible and disordered structure for desialylated PSIII andsuggested that native PSIII could form extended helical structureswhere each turn was made by more than four RUs (29–32). PSIIIfragments smaller than four RUs appeared as weak inhibitors of thebinding of native PSIII to its specific antibodies and failed to elicitan efficient immune response following conjugation (33, 34). Basedon the above observations, it was concluded that the native PSIIIforms a sialic acid-dependent conformational epitope that is es-sential for the elicitation and recognition of functional antibodies,and the length dependency of this conformational epitope was as-cribed to its localization on extended helices within a random coilstructure (33). According to the proposed model, the side chainNeuNAc moiety would exercise a remote control over immuno-logical determinants of the PS backbone, without being directly in-volved in molecular interactions as part of the epitope (34).Structural glycobiology studies using bacterial oligosaccharides

in complex with functional monoclonal antibodies (mAbs) ableto protect against the target pathogen could represent the mostdirect methodology to gain information on PS immunologicaldeterminants at the atomic level. However, due to the high po-larity of the hydroxyl groups and flexibility of the carbohydratestructures, crystallization of carbohydrate–protein complexes is achallenging task (35). To date, only a very limited number ofcarbohydrate antigen–Fab complexes, in comparison with a largenumber of protein antigen–Fab complexes, have been resolvedby X-ray crystallography (5–7, 10). As an alternative, a combi-nation of techniques [NMR, surface plasmon resonance (SPR),ELISA, etc.] is typically used to gain insights into carbohydrate–protein interactions (36, 37).In the present study, we used saturation transfer difference–NMR

(STD-NMR) in conjunction with X-ray crystallography to investigatethe interaction of GBSIII oligosaccharides obtained by synthetic anddepolymerization procedures with a protective antibody.

ResultsSelection and Immunochemical Characterization of a Functional Anti-PSIII Rabbit mAb. To investigate the binding of GBS PSIII tofunctional antibodies at the atomic level and further decipher therole of the NeuNAc moiety, we first generated a rabbit anti-PSIIImAb capable of mediating GBS OPK. Similar to a previous report(26), rabbits immunized with native PSIII conjugated to geneticallydetoxified diphtheria toxin (CRM197) developed two types of anti-bodies recognizing the native PS but differing in their capacity tobind the PSIII desialylated core antigen. Indeed, inhibition ELISAexperiments showed that binding of IgG from rabbit serato immobilized PSIII unconjugated (SI Appendix, Fig. S1) or

conjugated to human serum albumin (HSA) (38) (Fig. 1A) couldbe partially blocked by preincubation with soluble Pn14 or desia-lylated PSIII. Pn14 also partially inhibited antibody-mediatedGBSIII OPK (Fig. 1B), confirming functional activity of bothtypes of NeuNAc-dependent and -independent rabbit antibodies.Similar results were described previously in the case of humanresponses to GBSIII (27).An IgG1 mAb was selected by hybridoma cell line technology

(NVS-1-19-5), and its activity was ascertained by OPKA (OPKtiter, 1.03 μg/mL). The selected monoclonal was representative ofthe typical response to type III PS and had the properties de-scribed in the literature (33, 34). Complete inhibition by nativePSIII and absence of inhibition by Pn14 in ELISA and OPKAexperiments (Fig. 1 A and B) confirmed that the epitope recog-nized by this mAb was NeuNAc-dependent. A Fab fragment ofmAb NVS-1-19-5 was prepared by papain proteolytic cleavage,and its high-affinity binding toward the GBS PSIII was measuredby SPR (KD estimated with the Fab 4 × 10−8 M).The NeuNAc-Gal covalent link in the lateral chain of the PSIII

RU is the most chemically labile glycosidic bond within the PS, andcomplete loss of NeuNAc residues or reduction of their carboxylicgroups results in PS structures incapable of inducing functionalantibodies (25, 26). To investigate the potential of the obtainedmAb for probing the integrity of the PSIII antigen, we subjectedPSIII-CRM197 to mild acid hydrolysis for different incubation timesand obtained conjugates with decreasing sialylation levels of 100,60–70, 20–30 and <5%, as estimated by NMR analysis (SI Ap-pendix, Fig. S2 and Table S1). The resulting glycoconjugates werecharacterized for their structural integrity and saccharide/proteincontent (SI Appendix, Fig. S3 and Table S2). ELISA experimentsusing immobilized native PSIII as the antigen and soluble native orpartially desialylated PSIII-CRM197 as the competitor showed de-creasing binding of the mAb to conjugates with increasing desia-lylation levels (SI Appendix, Fig. S4A). The critical role of theNeuNAc in elicitation of functional antibodies was confirmed byimmunizing female mice with three doses of the different glyco-conjugates, followed by sera analysis by ELISA for quantification

Fig. 1. PSIII antibody specificity and functional activity toward PSIII. In-hibition experiments with PSIII and Pn14 of rabbit anti-PSIII polyclonal sera(pAb) and mAb in (A) ELISA, binding to immobilized PSIII conjugated to HSA,and (B) OPK. (C) Competitive SPR of the binding between the rabbit Fab andPSIII. Different length fragments were used as inhibitors. PSIII and thePn14 were used as positive and negative controls, respectively.

5018 | www.pnas.org/cgi/doi/10.1073/pnas.1701885114 Carboni et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

23,

202

0

of anti-PSIII antibodies, and by OPKA to assess antibody func-tional activity. Lower levels of sialylation resulted in diminishedrecognition of HSA-conjugated native PSIII and decreasing in vitrokilling of GBS bacteria (SI Appendix, Fig. S4 B and C).

Selection of PSIII Glycan Fragments for Structural Studies. Antibodieselicited in mice by PSIII have been shown to bind oligosaccharidefragments in a length-dependent manner (33, 34, 39). Using acompetitive ELISA, it was demonstrated that at least two un-modified RUs were necessary for suboptimal binding to a GBSIII-specific mouse mAb. Inhibition with 3–7 RU PSIII fragmentsshowed a moderate increase and became much higher above thisnumber of RUs (29). SPR experiments using a Fab of the samemAb confirmed that the affinity (KD) was comparable between2 and 6–7 RUs, it increased by threefold from 6–7 RUs to 20 RUs,and it remained constant beyond this point (33).To investigate more in detail the epitope recognized by the se-

lected functional mAb, we generated a set of PSIII oligosaccharidefragments by partial de–N-acetylation followed by nitrosation (40).Improvements in the purification process of the PSIII fragments byusing an anionic exchange HPLC in place of size exclusion HPLC,as described by Paoletti et al. (41), allowed for the isolationof fragments with a degree of polymerization (DP) in the range of2–15 with a more defined composition (SI Appendix, Fig. S5).These fragments were composed of a modified RU and a variablenumber of unmodified RUs (Fig. 2). The identity of these oligo-saccharides was established by 1H NMR, and the ratio between thesignals of the generated hydrated aldehyde group and the H-3e(i.e., equatorial) position of NeuNAc or the H-2 position of Glcwas used to determine the oligosaccharide length (SI Appendix,Fig. S6). The relative ratio of monosaccharide components in theobtained oligosaccharides was ascertained by high-performanceanion-exchange chromatography coupled with pulsed ampero-metric detection (HPAEC-PAD) (SI Appendix, Table S3). MALDI-TOFMS spectra in negative mode confirmed (SI Appendix, Fig. S7)the structure of the first two peaks eluted on the analytical MonoQanionic exchange column as DP2 and DP3. Both were composed ofhomogeneous glycans, whereas longer oligosaccharides containedmixtures of glycans and are hereafter indicated as average DP(avDP). A single PSIII RU was chemically synthesized by conven-tional carbohydrate chemistry procedures (42). Given that the fiveresidues composing the RU can be arranged according to threedifferent sequences, three alternative glycans were assembled withan end-terminal sugar bearing a linker for possible conjugation (Fig.2). After deprotection, the defined glycans were purified on sizeexclusion chromatography and characterized by NMR and MS(procedures for synthesis and characterization data are reportedelsewhere) (42).

Length-dependent recognition of the different fragments wasconfirmed by competitive ELISA with the selected rabbit mAb(SI Appendix, Fig. S8) (33). Inhibition of mAb binding increasedtwo logs between DP1 and DP2 and then slightly increased an-other 1.5 log, changing the PS size from DP2 to DP13, and be-came five-log higher when PSIII was used as an inhibitor (Fig. 1Cand SI Appendix, Table S4). To exclude the effect of the bivalentIgG interaction on the avidity, we performed a competitive SPRassay where different oligosaccharide fragments (DP1–13 range)were tested as competitors for the binding of soluble Fab fragmentto PSIII conjugated to HSA immobilized on the chip. Overall, twomajor populations of inhibitors, DP ≥ 2 and DP < 2, were dif-ferentiated. DP ≥ 2 oligosaccharides showed asymptotically in-creasing affinity up to the native PSIII, with only a 2-log differencebetween native PSIII and DP2. Concerning the single RUs, a 2-logdifference was observed between DP2 and the synthetic DP1 RU1 and 2 had a β-Glc-(1→6)–β-GlcNAc branching; conversely, avery weak inhibition was detected for the linear structure 3, sug-gesting that the arrangement of sugars in the RU impacts Abrecognition and the ramification point is relevant for binding.The binding affinities of the native PSIII and the DP2 fragment

to the rabbit Fab were compared by conjugating the two PSstructures to CRM197 and immobilizing them on an SPR chip. Thetwo binding constants determined according to a 1:1 fitting modeldiffered by less than 10-fold (SI Appendix, Fig. S9 and Table S5)(33). Overall, this analysis confirmed a length-dependent affinityof anti-PSIII GBS antibody but also indicated that DP2 containsthe PSIII portion necessary for high-affinity antibody binding andcould be used for further structural analysis.

Identification of the PSIII Antigenic Determinant by STD-NMR. Tomap the interactions of PSIII oligosaccharides with the protectivemAb, STD-NMR studies were undertaken (43). STD difference(STDD)-NMR spectra were derived by subtracting the STD-NMRspectrum of the glycan in the bound state with the mAb (ligand/protein 15–50:1 molar ratio) to the reference spectrum in theunbound state. The DP3 STDD spectrum was superimposable tothe one obtained for the DP2–mAb complex (SI Appendix, Fig.S10), highlighting that the antibody was recognizing identical re-gions in the two fragments. Due to their higher affinity constantsand larger size, experiments carried out on longer fragments(5.5 and 8 RUs) resulted in lower signal intensities that did notallow unequivocal detection of saturation transfer effects.Notably, most of the 1H NMR resonances that could un-

equivocally be assigned to the DP2–mAb complex were related tothe α-NeuNAc-(2→3)–β-GalB-(1→4) residues (Fig. 3A), indicatinga direct involvement of this branch in antibody binding, in additionto the Glc and Gal residues of the backbone. In the DP2 oligo-saccharide, it was not possible to differentiate the NAc group ofNeuNAc and GlcNAc. To better discriminate the exact positionsengaged in antibody interaction, the synthesized DP1 was com-plexed with the mAb and analyzed by STDD-NMR. As shown inFig. 3 B and C, well-recognizable resonances of the branched RU1 were recovered in the STDD spectrum with medium–high sat-uration intensity (>50%). The highest transfer of saturation wasobserved for H-5 (3.85 ppm), H-4 (3.70 ppm), H-3 (2.76 and1.82 ppm for equatorial and axial, respectively), and H-6(3.65 ppm) of NeuNAc (Fig. 3B), confirming that NeuNAc isthe major anchoring site for mAb interaction. A similar pattern ofsignals related to the NeuNAc was obtained with units 3 (Fig. 3)and 2 (SI Appendix, Fig. S11). Further, the involvement of theNeuNAc N-acetyl group, not clearly distinguishable from the oneof the GlcNAc residue in the DP2, was unveiled in the syntheticprobes at 2.0 ppm. When the desialylated tetrasaccharide [β-Gal-(1→4)–β-GlcNAc-(1→3)–β-Gal-(1→4)–α/β-Glc] was analyzed inthe presence of the mAb, no transfer of saturation was observed.A high STD signal was also observed for the Glc H-2 position at3.37 ppm in the DP2 and DP1 RUs 1 and 2 (SI Appendix, Fig. S11)

Fig. 2. Chemical structure of GBS PSIII and glycan probes used in this study. Thered highlighted GlcNAc residue is used as the referencemonosaccharide to outlinethe three possible sugar sequences related to PSIII RU. The 2,5-anhydri-D-mannoseresidue obtained by chemical depolymerization of PSIII is indicated in blue.

Carboni et al. PNAS | May 9, 2017 | vol. 114 | no. 19 | 5019

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

23,

202

0

in complex with the mAb, whereas no recovery was seen for thisposition in the linear RU 3, where the Glc moiety is located farfrom the NeuNAc (SI Appendix and Fig. 3 B and C). Previousconformational studies revealed that the orientation of the Glcbackbone residue was affected by removal of sialic acid, suggestingspatial proximity of the two residues (32). The transfer of satu-ration to the H-2 of Glc indicated that mAb binding of this residueis determined by its intrinsic location in the sugar sequence.Owing to the better resolution of the STD-NMR spectra of the

synthetic glycans, saturation transfer was more precisely quantifiedfor some of the protons involved in binding. Experiments at in-creasing saturation times from 0.5 up to 5.0 s were performed toavoid possible bias in the calculation of STD effects due to dif-ferent proton longitudinal relaxation times (T1) or intramolecularspin diffusion within the bound state (SI Appendix, Fig. S12). Inthe synthetic glycan 1, an STD relative effect of 100%, 96%, and70% was measured for H-5, H-3eq signals of NeuNAc, and H-2 ofGlc, respectively.Taken together, these results unambiguously showed that the side

chain NeuNAc and Gal residues, in conjunction with the Gal andGlc backbone sugars of the RU, directly interact with the mAb.

Structural Studies of the PSIII–Fab Complex by X-Ray Crystallography.To further elucidate the basis for the PSIII–mAb interaction at theatomic level, we used X-ray crystallography to determine the 3Dstructure of DP2 in complex with the Fab NVS-1-19-5. Crystals ofthe complex belong to space group C2221, contain two copies of a1:1 DP2–Fab complex in the asymmetric unit (ASU), and diffractto a resolution of 2.7 Å. The structure was solved by molecularreplacement, using as template input model the coordinates of arabbit Fab [Protein Data Bank (PDB) ID code 4JO1] with which

Fab NVS-1-19-5 shares 79% sequence identity. Excellent electrondensities were observed for the Fab portions and allowed themodeling of residues Gln20–Ser245 of the heavy (H) chain andVal24–Cys238 of the light (L) chain. Additional difference densitieswere observed in front of the Fab complementarity determiningregions (CDRs), and these were modeled with the DP2 oligosac-charide (SI Appendix, Fig. S13). Of the total 10 monosaccharideresidues of DP2, 6 could be modeled with very good fitting in clearelectron density maps and in both copies of the complex, whereasthe terminal saccharides residues (B″, C″, D″, and E″) had low sigmaelectron densities for one of the copies of the complex only. Forcompleteness, the full DP2 was included in the final refined model,although the terminal residues had low sigma electron density signalsand subsequently high B-factors, both of which reflect uncertainty intheir exact position. Importantly, the stereochemistry and geometryof final refined DP2 molecules were validated using Privateer (SIAppendix, Table S6). The coordinates of the complex were refined tofinal Rwork/Rfree values of 23.6%/28.2% (SI Appendix, Table S7).Structural superimposition of the two copies of the complex presentin the ASU resulted in very low carbon alpha root mean squaredeviation (Cα rmsd) values (∼0.5 �Å; the two structures are essen-tially identical), and their bound DP2 conformations were also highlysimilar; therefore, one copy will be used for the description ofthe structure.The Fab fragment binds the glycan with an extended interface

(577 Å2) formed by two adjacent pockets that are located betweenthe L and H chain of the Fab (Fig. 4 A and B and SI Appendix, Fig.S14). Most of the observed contacts with the Fab involve a singlePS RU, whereas the sugar residues of the consecutive RU depart

Fig. 3. GBS PSIII epitope mapping. (A) Protons interacting with the mAbidentified by STD-NMR and X-ray crystallography. (B) STDD and (C) related 1HNMR spectra of the desialylated tetrasaccharide [β-Gal-(1→4)–β-GlcNAc-(1→3)–β-Gal-(1→4)–α/β-Glc in green], the linear 3 (blue) and branched 1 (red)RUs, and the DP2 fragment (black). Proton positions receiving saturationafter irradiation of the protein are indicated. ‡ indicates the CH2 signal ofthe Tris buffer; * refers to signals related to the protein.

Fig. 4. Crystal structure of DP2 bound to Fab NVS-1-19-5. (A, Left) The Fab isdepicted with surfaces and the DP2 with sticks. Fab residues involved in di-rect binding with DP2 (the paratope) are colored in yellow, and the bindingpockets are labeled. (A, Right) Surface electrostatic potential distribution ofthe Fab, oriented as on Left. (B) Two views (Left and Right) of the large andsmall pockets where the DP2 binds. (C) Details of the interactions betweenDP2 and Fab. Carbon atoms of the DP2 backbone are colored in yellow, andthose belonging to the branches are in green, whereas carbon atoms of theFab are colored in light and dark gray for the L and H chain, respectively.Nitrogen and oxygen atoms are colored in blue and red, respectively, andwater molecules are shown as blue spheres.

5020 | www.pnas.org/cgi/doi/10.1073/pnas.1701885114 Carboni et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

23,

202

0

from the binding pockets, with the exception of GlcNAc-A″, whichis directly bound to Fab amino acid residues. The larger pockethosts the NeuNAc of the RU branch that interacts with multiplehydrophilic residues of the CDRs from both the H and L chains.Specifically, Arg56 of the L chain forms a hydrogen bond with theN-acetyl group of NeuNAc-C ′, and His127 interacts with its O-4(Fig. 4C). Additional weaker interactions of NeuNAc-C ′ occurthrough a water molecule bridge between the carboxylic group ofthe sugar and Asn74 of the H chain and the CH-π interactionbetween Tyr52 of the L chain and the sialic NAc methyl group (Fig.4C). The GlcNAc-A″ residue appears involved in the binding to thesmaller anchoring pocket of the Fab through at least two hydrogenbonds formed by the L chain Arg56 and Lys74 with the NAc groupand the O-3 position of the sugar, respectively. The NAc group isalso stabilized by a CH-π interaction with Tyr126 (Fig. 4C). In-terestingly, the guanidine group of Arg56 is located within H-bonddistance from the NAc carbonyl groups of both NeuNAc-C″ andGlcNAc-A″, whose methyl groups are at the same time favorablyoriented to make CH-π interactions with the phenyl rings ofTyr52 of the L chain and Tyr126 of the H chain, respectively. Thisarrangement allows for a peculiar symmetric network of interac-tions involving the NeuNAc of one RU and the GlcNAc of aconsecutive unit (Fig. 4C). Complete de–N-acetylation of theseresidues resulted in loss of Fab recognition as demonstrated bycompetitive SPR (SI Appendix, Fig. S15). Inspection of the glycanstructure revealed how Gal-E′, although positioned outside of thetwo binding pockets, interacts with side chain atoms of Asn53 of theL chain and of Tyr126 of the H chain. The side chain ofTyr126 seems also favorably positioned to engage H-bond–medi-ated interactions with the O-3 position of Glc-D′. Interestingly,GlcNAc-A′ that was modified into 2,5-anhydro-D-mannose duringthe depolymerization reaction appears to be far from the anchoringpocket; hence, its chemical manipulation during depolymerizationdid not interfere with Fab binding. The DP2–Fab complex crystalstructure also revealed how NeuNAc-C′ and Glc-D′ are in closecontact, with a calculated distance between the O-9 atom ofNeuNAc and the O-6 of Glc of 2.8 Å (Fig. 4C). This suggests thathydrogen bonds between these two positions might be furtherstabilized upon binding to the Fab. O-acetylation of the NeuNAcglycerol chain was noticed in PSIII from GBS clinical isolates, andit affects inhibition of neutrophil suppression and virulence (44, 45).We observed that the mAb interacts with the O-7 of NeuNActhrough an H2O molecule, and dimensionally, this space could alsohost an acetyl group. This could explain the data from the literaturereporting that O-acetylation does not interfere with protection (46).In summary, our structure clearly showed that the side chain

NeuNAc and the backbone Gal and Glc of the first RU, as well asthe backbone GlcNAc of the consecutive RU, interact with mul-tiple hydrophilic residues of the CDRs from both the H and the Lchain, whereas the NeuNAc glycerol moiety is involved in internalinteractions rather than in direct binding to the Fab.

DiscussionConjugate vaccines have been one of the major developments ofthe last 40 y. The nature of the repetitive protective epitopespresent in the PS has been difficult to define, mostly because ofthe difficulty in obtaining crystals from PSs. In many instances, ithas been speculated that the protective epitopes would be formedfrom distant residues that would become close in a helical struc-ture assumed by the PS (28, 29).In this study, we determined the molecular structure of a pro-

tective epitope of GBS PSIII. Based on NMR simulation studiesindicating the formation of several potential helical conformations(29, 32), where sialylated side chains are arranged on the exteriorsurface of the helix, the protective epitope dependent on a con-formational structure has been proposed for GBS PSIII. Inmarked contrast, the data reported in this paper show that theprotective epitope is linear and overlaps two RUs of the PS.

In the proposed conformational epitope structure, the PSIIIhelix structure was supposed to be stabilized by the presence ofspecific interactions between the side chain of the sialic acid andthe backbone glucosyl and galactosyl residues, influencing theorientation of the side chain and stabilizing the conformation ofthe backbone, and required a carbohydrate chain length of at leastfour pentasaccharide RUs to form a conformational epitopeequivalent to one present in the native PSIII (33, 34). Shortercarbohydrate chains were thought to be subjected to conforma-tional changes so that in the absence of specific interactions theoverall helical structure would collapse (29). The need for a helicalepitope for antibody recognition was hypothesized consideringthat GBS PSIII showed a length-dependent binding of specificmonoclonal IgGs (33). Zou et al. (33) observed that KD measuredby SPR for monovalent binding remained virtually constant fromtwo to seven RUs, which demonstrated that the epitope optimi-zation was occurring from two to seven RUs, whereas beyond 20RUs this phenomenon would overlap with the multivalent expo-sition of stabilized epitopes. Our competitive ELISA and SPRdata indicate that the functional mAb characterized here sharessimilar features to the ones used by Zou et al. (33) in that it ex-hibits a similar length-dependent affinity. However, this length-dependent affinity could also be explained by a multivalencyeffect instead of the need for a helical conformation.Johnson et al. showed by STD-NMR experiments that NeuNAc

was not involved in binding to a monoclonal IgM (47), indicatingthe existence of antibodies recognizing the PS backbone. Molec-ular dynamics simulations confirmed that sialic acid plays a role inmaintaining the helical structure and led to the hypothesis that thisPS conformation was pivotal for antibody binding (29).Our results demonstrate the existence of a sialic acid-dependent

functional epitope of GBS that is fully contained within fourconsecutive sugars deriving from the PSIII backbone and onebranched disaccharide present in this sequence. The nature of themapped epitope fully explains the well-known critical role of thesialic acid and, as schematically shown in Fig. 5, does not require aconformational epitope (29, 33, 34). However, our study does notrule out the possibility that mAbs like those described by Johnsonet al. (47) may recognize different regions of GBS PSIII.Our findings have profound implications in understanding

the protective immunity against type III GBS and in the designof conjugate vaccines and suggest that the nature of protectiveepitopes should be investigated in all PS vaccines, especiallythose where the helical structure has been proposed (29).Carbohydrates present a higher level of structural complexitycompared with other classes on natural biopolymers such aspolypeptides or polynucleotides, because of the presence ofα- or β-glycosidic bonds and the occurrence of connectionsat different positions of the ring that create ramifications.

Fig. 5. Schematic representation of GBS PSIII–Fab interaction according to(A) the helical model previously proposed (29–34) and (B) the model thatfully explains the well-known critical role of NeuNAc.

Carboni et al. PNAS | May 9, 2017 | vol. 114 | no. 19 | 5021

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

23,

202

0

These unique features may result in spatial motifs relevant for anti-body recognition and specificity even in the presence of a limitednumber of sugar residues, as it has been demonstrated here for GBSPSIII. In this manner, the hypothesized need for the human immunesystem to recognize GBS PS by means of length-dependent helicalstructures, thus avoiding recognition of self-antigens containing sialicacid, would need to be reconsidered (34). The existence of helicalconformational epitopes, which would imply the necessity of using along PS antigen, has been considered a challenge when applyingthese technologies to certain pathogens (2). The elucidation andunderstanding of the structural base for the immune recognition ofcarbohydrate epitopes reported here paves the way for designingmodern glycoconjugate vaccines with short oligosaccharides obtainedby synthetic, chemoenzymatic, or bioengineering methods (48).

Materials and MethodsSI Appendix, Materials and Methods feature additional information to thatprovided here.

The clone producing the rabbitmAbanti-PSIII NVS-1-19-5was obtained usinghybridoma technology by EPITOMICS Inc., and the relative Fab was prepared byusing a “Fab Preparation Kit” (Pierce). The GBS PSIII fragments were preparedby deamination (40), and different chain-length oligosaccharides were sepa-rated by anionic exchange chromatography. NMR experiments were carriedout on a Bruker 500 MHz NMR instrument, and the X-ray diffraction data werecollected at the European Synchrotron Radiation Facility (ESRF).

ACKNOWLEDGMENTS. We thank Matthew James Bottomley of the GSK groupof companies for useful discussions and critical reading of the manuscript. Thisstudy was sponsored by Novartis Vaccines, now part of the GSK group ofcompanies, which was involved in all stages of the study conduct and analysis.

1. Avci FY, Kasper DL (2010) How bacterial carbohydrates influence the adaptive im-mune system. Annu Rev Immunol 28:107–130.

2. Costantino P, Rappuoli R, Berti F (2011) The design of semi-synthetic and syntheticglycoconjugate vaccines. Expert Opin Drug Discov 6:1045–1066.

3. Anish C, Schumann B, Pereira CL, Seeberger PH (2014) Chemical biology approaches todesigning defined carbohydrate vaccines. Chem Biol 21:38–50.

4. Lipinski T, et al. (2012) A β-mannan trisaccharide conjugate vaccine aids clearance ofCandida albicans in immunocompromised rabbits. Vaccine 30:6263–6269.

5. Villeneuve S, et al. (2000) Crystal structure of an anti-carbohydrate antibody directedagainst Vibrio cholerae O1 in complex with antigen: Molecular basis for serotypespecificity. Proc Natl Acad Sci USA 97:8433–8438.

6. Vyas NK, et al. (2002) Molecular recognition of oligosaccharide epitopes by amonoclonal Fab specific for Shigella flexneri Y lipopolysaccharide: X-ray structuresand thermodynamics. Biochemistry 41:13575–13586.

7. Cygler M, Wu S, Zdanov A, Bundle DR, Rose DR (1993) Recognition of a carbohydrateantigenic determinant of Salmonella by an antibody. Biochem Soc Trans 21:437–441.

8. Safari D, et al. (2008) Identification of the smallest structure capable of evoking op-sonophagocytic antibodies against Streptococcus pneumoniae type 14. Infect Immun76:4615–4623.

9. Safari D, et al. (2011) The immune response to group B streptococcus type III capsularpolysaccharide is directed to the -Glc-GlcNAc-Gal- backbone epitope. Glycoconj J 28:557–562.

10. Vulliez-Le Normand B, et al. (2008) Structures of synthetic O-antigen fragments fromserotype 2a Shigella flexneri in complex with a protective monoclonal antibody. ProcNatl Acad Sci USA 105:9976–9981.

11. Jennings HJ (2012) The role of sialic acid in the formation of protective conforma-tional bacterial polysaccharide epitopes. Anticarbohydrate Antibodies, eds Kosma P,Muller-Loennies S (Springer, Vienna), pp 55–73.

12. Nuccitelli A, Rinaudo CD, Maione D (2015) Group B Streptococcus vaccine: State of theart. Ther Adv Vaccines 3:76–90.

13. Berti F, et al. (2014) Structure of the type IX group B Streptococcus capsular polysaccharideand its evolutionary relationship with types V and VII. J Biol Chem 289:23437–23448.

14. Baker CJ, Kasper DL (1976) Correlation of maternal antibody deficiency with sus-ceptibility to neonatal group B streptococcal infection. N Engl J Med 294:753–756.

15. Munoz FM, Ferrieri P (2013) Group B Streptococcus vaccination in pregnancy: Movingtoward a global maternal immunization program. Vaccine 31:D46–D51.

16. Paoletti LC, et al. (1992) Effects of chain length on the immunogenicity in rabbits ofgroup B Streptococcus type III oligosaccharide-tetanus toxoid conjugates. J Clin Invest89:203–209.

17. Paoletti LC, Kennedy RC, Chanh TC, Kasper DL (1996) Immunogenicity of group BStreptococcus type III polysaccharide-tetanus toxoid vaccine in baboons. Infect Immun64:677–679.

18. Wessels MR, et al. (1990) Immunogenicity in animals of a polysaccharide-proteinconjugate vaccine against type III group B Streptococcus. J Clin Invest 86:1428–1433.

19. Kasper DL, Baker CJ, Galdes B, Katzenellenbogen E, Jennings HJ (1983) Immuno-chemical analysis and immunogenicity of the type II group B streptococcal capsularpolysaccharide. J Clin Invest 72:260–269.

20. Baker CJ, Rench MA, Paoletti LC, Edwards MS (2007) Dose-response to type V group Bstreptococcal polysaccharide-tetanus toxoid conjugate vaccine in healthy adults.Vaccine 25:55–63.

21. Edwards MS (2008) Group B streptococcal conjugate vaccine: A timely concept forwhich the time has come. Hum Vaccin 4:444–448.

22. Donders GG, et al. (2016)Maternal immunizationwith an investigational trivalent groupB streptococcal vaccine: A randomized controlled trial. Obstet Gynecol 127:213–221.

23. Clarke ET, et al. (2013) Polysaccharide-specific memory B cells generated by conjugatevaccines in humans conform to the CD27+IgG+ isotype-switched memory B cell phe-notype and require contact-dependent signals from bystander T cells activated bybacterial proteins to differentiate into plasma cells. J Immunol 191:6071–6083.

24. Edmond KM, et al. (2012) Group B streptococcal disease in infants aged younger than3 months: Systematic review and meta-analysis. Lancet 379:547–556.

25. Kasper DL, et al. (1979) Immunodeterminant specificity of human immunity to type IIIgroup B streptococcus. J Exp Med 149:327–339.

26. Jennings HJ, Lugowski C, Kasper DL (1981) Conformational aspects critical to theimmunospecificity of the type III group B streptococcal polysaccharide. Biochemistry20:4511–4518.

27. Guttormsen H-K, et al. (2002) Type III group B streptococcal polysaccharide inducesantibodies that cross-react with Streptococcus pneumoniae type 14. Infect Immun 70:1724–1738.

28. Jennings HJ, et al. (1984) Structure, conformation and immunology of sialic acidcontaining polysaccharides of human pathogenic bacteria. Pure Appl Chem 56:893–905.

29. González-Outeiriño J, Kadirvelraj R, Woods RJ (2005) Structural elucidation of type IIIgroup B Streptococcus capsular polysaccharide using molecular dynamics simulations:The role of sialic acid. Carbohydr Res 340:1007–1018.

30. Kadirvelraj R, et al. (2006) Understanding the bacterial polysaccharide antigenicity ofStreptococcus agalactiae versus Streptococcus pneumoniae. Proc Natl Acad Sci USA103:8149–8154.

31. Woods R, Yongye A (2012) Computational techniques applied to defining carbohy-drate antigenicity. Anticarbohydrate Antibodies, eds Kosma P, Müller-Loennies S(Springer, Vienna), pp 361–383.

32. Brisson JR, et al. (1997) NMR and molecular dynamics studies of the conformationalepitope of the type III group B Streptococcus capsular polysaccharide and derivatives.Biochemistry 36:3278–3292.

33. Zou W, et al. (1999) Conformational epitope of the type III group B Streptococcuscapsular polysaccharide. J Immunol 163:820–825.

34. Zou W, Jennings HJ (2001) The conformational epitope of type III group B Strepto-coccus capsular polysaccharide. Adv Exp Med Biol 491:473–484.

35. Bush CA, Martin-Pastor M, Imberty A (1999) Structure and conformation of complexcarbohydrates of glycoproteins, glycolipids, and bacterial polysaccharides. Annu RevBiophys Biomol Struct 28:269–293.

36. Oberli MA, et al. (2010) Molecular analysis of carbohydrate-antibody interactions: Casestudy using a Bacillus anthracis tetrasaccharide. J Am Chem Soc 132:10239–10241.

37. Coelho H, et al. (2015) The quest for anticancer vaccines: Deciphering the fine-epitopespecificity of cancer-related monoclonal antibodies by combining microarray screen-ing and Saturation Transfer Difference NMR. J Am Chem Soc 137:12438–12441.

38. Bhushan R, Anthony BF, Frasch CE (1998) Estimation of group B streptococcus type IIIpolysaccharide-specific antibody concentrations in human sera is antigen dependent.Infect Immun 66:5848–5853.

39. Wessels MR, Muñoz A, Kasper DL (1987) A model of high-affinity antibody binding totype III group B Streptococcus capsular polysaccharide. Proc Natl Acad Sci USA 84:9170–9174.

40. Michon F, et al. (2006) Group B streptococcal type II and III conjugate vaccines:Physicochemical properties that influence immunogenicity. Clin Vaccine Immunol 13:936–943.

41. Paoletti LC, et al. (1990) An oligosaccharide-tetanus toxoid conjugate vaccine againsttype III group B Streptococcus. J Biol Chem 265:18278–18283.

42. Cattaneo V, et al. (2016) Synthesis of group B Streptococcus type III polysaccharidefragments for evaluation of their interactions with monoclonal antibodies. Pure &Appl Chem, www.degruyter.com/view/j/pac.ahead-of-print/pac-2016-0918/pac-2016-0918.xml.

43. Meyer B, Peters T (2003) NMR spectroscopy techniques for screening and identifyingligand binding to protein receptors. Angew Chem Int Ed Engl 42:864–890.

44. Lewis AL, Nizet V, Varki A (2004) Discovery and characterization of sialic acidO-acetylation in group B Streptococcus. Proc Natl Acad Sci USA 101:11123–11128.

45. Weiman S, et al. (2010) O-Acetylation of sialic acid on group B Streptococcus inhibitsneutrophil suppression and virulence. Biochem J 428:163–168.

46. Pannaraj PS, et al. (2009) Group B streptococcal conjugate vaccines elicit functionalantibodies independent of strain O-acetylation. Vaccine 27:4452–4456.

47. Johnson MA, et al. (2003) NMR studies of carbohydrates and carbohydrate-mimeticpeptides recognized by an anti-group B Streptococcus antibody. J Biol Chem 278:24740–24752.

48. Berti F, Adamo R (2013) Recent mechanistic insights on glycoconjugate vaccines andfuture perspectives. ACS Chem Biol 8:1653–1663.

5022 | www.pnas.org/cgi/doi/10.1073/pnas.1701885114 Carboni et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

23,

202

0