Structure activity relationships of monocytechemoattractant proteins in complex with ablocking antibody

Carl Reid, Mia Rushe, Matthew Jarpe,Herman van Vlijmen, Brian Dolinski, Fang Qian,Teresa G. Cachero, Hernan Cuervo, Milka Yanachkova,Chioma Nwankwo, Xin Wang, Natalie Etienne,Ellen Garber, Veronique Bailly, Antonin de Fougerolles1

and P.Ann Boriack-Sjodin2

Department of Research, Biogen Idec, Inc., 12 Cambridge Center,Cambridge, MA 02142, USA

1Present address: Department of Research, Alnylam Pharmaceuticals,Inc., 300 Third Street, Cambridge, MA 02142, USA

2To whom correspondence should be addressed.E-mail: ann.boriack-sjodin@biogenidec.com

Monocyte chemoattractant proteins (MCPs) are cytokinesthat direct immune cells bearing appropriate receptors tosites of inflammation or injury and are therefore attractivetherapeutic targets for inhibitory molecules. 11K2 is ablocking mouse monoclonal antibody active againstseveral human and murine MCPs. A 2.5 A structure of theFab fragment of this antibody in complex with humanMCP-1 has been solved. The Fab blocks CCR2 receptorbinding to MCP-1 through an adjacent but distinct bindingsite. The orientation of the Fab indicates that a single MCP-1 dimer will bind two 11K2 antibodies. Several key residueson the antibody and on human MCPs were predicted to beinvolved in antibody selectivity. Mutational analysis ofthese residues confirms their involvement in the antibody–chemokine interaction. In addition to mutations thatdecreased or disrupted binding, one antibody mutation re-sulted in a 70-fold increase in affinity for human MCP-2. Akey residue missing in human MCP-3, a chemokine notrecognized by the antibody, was identified and engineer-ing the preferred residue into the chemokine conferredbinding to the antibody.Keywords: Antibody engineering/monocyte chemoattractantproteins/protein-antibody complex/protein therapeutics/x-raycrystallography

Introduction

Chemokines are proinflammatory molecules that coordinatethe host immune response by promoting leukocyte migrationto the site of infection or injury (Rossi and Zlotnik, 2000;Zlotnik and Yoshie, 2000). These molecules are thought tobind to and cluster on glycosaminoglycans, resulting in alocally high concentration of chemokine which triggersrecruitment of leukocytes bearing chemokine receptors (Rot,1993; Chakravarty et al., 1998; Proudfoot et al., 2003). Thechemokine superfamily is currently comprised of more than50 members that can be classified based on the arrangement oftheir cysteine residues into four subfamilies: CXC, C, CX3Cand CC (Rollins, 1997). Monocyte chemoattractant proteins(MCPs) are members of the CC family of chemokines with

five isoforms in humans: MCP-1-4 and eotaxin. Human MCPsbind to G-protein coupled receptors CCR1, CCR2, CCR3,CCR5 and CCR10, with multiple MCPs binding to multipleCCRs.

The structures of many chemokines have been solved byprotein crystallography and NMR spectroscopy, includingMCP-1, MCP-2 and MCP-3 (Handel and Domaille, 1996;Kim et al., 1996; Lubkowski et al., 1997; Meunier et al.,1997; Blaszczyk et al., 2000). The structures are extremelysimilar to each other and share the same topology as related CCchemokines. The structures consist of an N-terminal loop re-gion, a three-strand antiparallel b sheet forming part of a‘Greek key’ motif and a C-terminal a helix which overlaysthe b sheet. The structure also contains two disulfide bridges.The crystal structure of MCP-2 contains a pyroglutamic acid atthe N-terminus which increases the chemotactic activity of thecytokine (Blaszczyk et al., 2000). A monomeric mutant ofMCP-1 (Pro58!Ala) has also been solved in complex withM3, a herpesvirus decoy receptor (Alexander et al., 2002).

MCP-1 is one of the most studied members of the chemokinefamily and has been shown to be a potential interventionpoint for the treatment of a variety of diseases, includingmultiple sclerosis (Sorensen et al., 2004), rheumatoid arthritis(Hayashida et al., 2001), atherosclerosis (Kusano et al., 2004)and insulin-resistant diabetes (Sartipy and Loskutoff, 2003).The MCP pathway has been validated through numerousanimal models of disease using MCP-1 knockouts (Guet al., 1998; Lu et al., 1998), CCR2 knockouts (Kuriharaet al., 1997; Boring et al., 1998; Dawson et al., 1999), ormonoclonal antibodies against MCP-1 or CCR2 (Fujinakaet al., 1997; Karpus and Kennedy, 1997; Lloyd et al., 1997;Kimura et al., 1998; Campbell et al., 1999; Furukawa et al.,1999; Ono et al., 1999; Schneider et al., 1999). Becausemembers of the MCP family bind and share multiple receptors,it has been speculated that an inhibitor that blocks two or moreMCP isoforms would be a strong therapeutic candidate.

11K2 is a monoclonal antibody (mAb) raised against humanMCP-1 (hMCP-1) that also binds with high affinity to hMCP-2and to murine MCP-1 (mMCP-1) and mMCP-5. This mAb hasproven effective in treating murine models of atherosclerosis(Lutgens et al., 2005) and ulcerative colitis (de Fougerolles,A.R., Mencarelli, A., Sprague, A.G., Cachero, T.G., Jarpe, M.,Kotelianski, V.K., Rennert, P.J. and Fiorucci, S. in prepara-tion). It is thought that the ability of this mAb to inhibit theactivity of mMCP-1 and mMCP-5 contributes to its activity inthese models. In order to determine the basis for the high affi-nityandbroadselectivityof11K2,weundertook astructuralana-lysis of hMCP-1 with the Fab fragment of 11K2 (11K2Fab).

Materials and methods

MCP-1 synthesis and purificationThe 76 amino acid peptide of hMCP-1 was synthesized byestablished automated protocols on an Applied Biosystems

Protein Engineering, Design & Selection vol. 19 no. 7 pp. 317–324, 2006Published online May 8, 2006 doi:10.1093/protein/gzl015

� The Author 2006. Published by Oxford University Press. All rights reserved. 317For Permissions, please e-mail: journals.permissions@oxfordjournals.org

433A Peptide Synthesizer using Fmoc-protecting aminoacids and HBTU/HOBt/DIEA-mediated activation in DMF(Fields et al., 1991). Peptide cleavage/deprotection was acco-mplished with mixture of 91% TFA, 3% H2O, 3% EDT,3% TIS for 3.5 h at room temperature. Soluble peptide wasprecipitated, rinsed with cold ethyl ether, then dissolved in95% aqueous acetic acid and lyophilized. The protein waspurified by reverse-phase HPLC (>98% purity) on a Varian210 HPLC. MALDI-MS of unfolded, purified protein wasperformed on an Applied Biosystems STR Voyager massspectrophotometer. The purified MCP-1 synthetic peptidewas lyophilized and then redissolved at 1 mg/ml in 2 Mguanidine–HCl, 100 mM Tris, pH 8.0, containing 8 mM cys-teine and 1 mM cystine. The protein was allowed to fold for48 h at room temperature while rocking gently. The foldedMCP-1 protein was then purified by reverse-phase HPLC,lyophilized and finally redissolved at 10 mg/ml in 25 mMTris, pH 8.4. About 90–95% of the protein was refolded prop-erly as measured by mass spectrometry and chromatographicanalysis and the protein was tested in a chemotaxis assay andfound to be active.

MCP expression and purificationIn order to efficiently make small amounts of a large number ofMCP mutants, an Escherichia coli expression system utilizingthe refolding methods of the synthetic protein was used. TheMCP mutants listed in Tables II and III were made by site-directed mutagenesis using the QuikChange Site-DirectedMutagenesis Kit (Stratagene, catalog no. 200518). Mutagene-sis reactions were performed according to the manufacturersprotocol using 50 ng of template and 150 ng each of oligo-nucleotide primers. Oligonucleotide primers used to generatethe MCP mutants can be found in the Supplementary dataavailable at PEDS online. Native hMCP-1, hMCP-2, hMCP-3 and MCP mutants were expressed in E.coli using the pET11vector (Novagen) encoding the mature MCP sequence in eitherBL21 (DE3) or Rosetta (DE3) cells. For some mutants SM102,a plasmid encoding Arg-AAG/A tRNA, was used to facilitateexpression. Cells were lysed and inclusion bodies were isolatedby a differential centrifugation-detergent wash procedure.The washed inclusion bodies were solubilized in 8 M urea,0.5 mM EDTA, 10 mM DTT and purified using reverse-phasechromatography. The resulting product was lyophilized andrefolded according to the procedure above. Owing to expres-sion difficulties, hMCP-3 Thr56Lys was synthesized (CellSciences) and then refolded.

Antibody mutation, expression and purification11K2 was generated from hybridomas and purified by affinitychromatography using Protein A resin. The Fab fragment of11K2 was generated by digestion with papain. Briefly, 200 mgof 11K2 monoclonal antibody were incubated in digestionbuffer (20 mM sodium phosphate, pH 7.2, 10 mM EDTA and20 mM Cysteine–HCl) in the presence of 5 ml of immobilizedpapain beads (Pierce) at 37�C for 18 h with shaking. Thedigested antibody was diluted with 4 vol of high salt bindingbuffer (1.5 M glycine, 3 M sodium chloride) to increase theaffinity of murine IgG1 for Protein A. The solution was loadedonto a 10 ml Protein A column to deplete the digested Fcfragments. This process was performed three times. The Fabfragment of 11K2 antibody was identified by SDS–PAGE anddialyzed against phosphate-buffered saline. The purified 11K2

Fab was further characterized by mass spectrometry and foundto be the expected mass.

Antibody mutants were designed using plasmids pCR060(light chain) and pCR101 (heavy chain). Oligonucleotideprimers used to generate the 11K2 mutants can be found in theSupplementary data available at PEDS online. Antibodymutants were expressed in HEK293e cells by co-transfectionof plasmid DNAs encoding the heavy and light chains. Lightchain mutants were co-expressed with wild-type heavy chain,and heavy chain mutants were co-expressed with the wild-typelight chain. All antibodies were expressed as chimeric Fabfragments bearing murine variable regions fused to humanconstant (IgG1, kappa) regions. Transfections were performedin 10 cm dishes using Effectene reagent according to the man-ufacturer’s protocol (Qiagen catalog no. 301427). Supernatantscontaining secreted antibodies were harvested 3 days post-transfection.

Crystallization and structure determinationSynthetic hMCP-1 and 11K2Fab were mixed in a 1:1 molarratio and concentrated to 8 mg/ml. Two microliters of theprotein was mixed with an equal amount of reservoir buffercontaining 13% PEG 4000, 100 mM HEPES, pH 7.5, 30 mMglycyl-glycyl-glycine using vapor diffusion techniques. Crys-tals typically appeared after 1–2 days with microseeding andgrew to full size (0.1 · 0.1 · 0.1 mm) in 1–2 weeks. Crystalswere briefly incubated in 15% PEG 4000, 100 mM HEPES,pH 7.5, 30 mM glycyl-glycyl-glycine, and 25% glycerol andflash frozen in liquid nitrogen prior to data collection. Datawere measured at 100�K at Brookhaven National Laboratory(beamline X4A) using a Quantum4 CCD detector (ADSC)at wavelength 0.97945 A. Data processing was performedusing Denzo, and data reduction was performed using Scale-pack (Otwinowski and Minor, 1997). The crystals contain oneMCP-1 monomer–11K2Fab complex per asymmetric unit andbelong to space group C2221 (a = 86.4 s, b = 89.1 s, c = 176.2s). Molecular replacement was performed using AMoRe(Navaza, 1994) using a model of 11K2Fab generated by homo-logy modeling with Modeller (Sali and Blundell, 1993) usingthe light chain of 184.1 [anti-OspA; PDB code 1OSP (Li et al.,1997)] and the heavy chain of 5G9 [anti-human tissue factor;PDB code 1FGN (Huang et al., 1998)] as the templates. Thetop solution in both the rotation and translation searches wasdetermined to be the correct solution after rigid body refine-ment of the individual Fab domains resulted in a 8.8% pointdrop in the Rfree value and electron density maps clearly re-vealed the position of the MCP-1 monomer. Model buildingwas performed using O (Jones et al., 1991) iteratively withsimulated annealing refinement in CNX (Brunger et al., 1998)with bulk solvent corrections and anisotropic B-factor scalingprotocols to achieve the final structure. The final protein modelhas been refined at 2.5 s resolution to a final R-value of 0.219and free R-value of 0.277 (Table I). The Ramachandran plotshows that 85.6% of all residues are in the most favoredregions and no residues are in disallowed regions. Thecoordinates have been deposited at the Protein Data Bank(2BDN).

Affinity measurementsThe solution phase equilibrium dissociation constants of 11K2Fabs binding to MCPs were determined using the KinExA3000 instrument (Sapidyne Instruments, Boise, ID, USA).

318318

C.Reid et al.

MCP-1 human IgG1–Fc fusion protein was coated onto 98 mpolymethylmethacrylate beads. These beads were packed intoa small glass column in the instrument, and purified 11K2Fabor supernatants from Fab expressing cells were flowed overthe beads. The captured Fab was detected with a Cy-5 labeledanti-murine F (ab)02 antibody (Jackson ImmunoResearch,West Grove, PA, USA). The signal detected was proportionalto the amount of free Fab that flowed over the column.

To measure the solution phase binding affinity, Fab wasincubated with varying concentrations of purified antigen. Themixtures were incubated at room temperature for 3 h to allowthem to come to a steady state. The amount of free Fab wasmeasured using the KinExA. The resulting data were fit to aquadratic equation using the KinExA software and the dis-sociation constants were determined as well as the percenterror from the least squares fit.

First, the binding of purified wild-type 11K2Fab to purifiedwild-type and mutant MCPs was measured. However, it wasunfeasible to purify the large number of Fabs expressed inmammalian cells in order to measure the affinity of mutant11K2Fabs to MCPs. The KinExA method measures the loss ofunbound Fab in solution as a function of the concentration ofantigen. For this reason unpurified supernatants of mammaliancells expressing Fabs were able to be used rather than purifiedprotein. To validate this approach the affinities of purified11K2Fab and unpurified supernatant containing 11K2Fab towild-type hMCP-1 were compared. The two values, 4.6 and7.5 pM, respectively, are not significantly different.

Chemotaxis inhibitionTHP-1 human monocytic cells were grown in RPMI, 10% FBSwith 4 mM L-glutamine. To measure chemotaxis, ChemoTXplates from Neuroprobe were used. Chemokines and 11K2Fabs

were incubated together for 30 min at 37�C to reach equi-librium. The chemokine, with or without Fab, was added to thebottom chamber of the plate. A filter was placed on top and160 000 THP-1 cells were added to the top of the filter. Theplates were incubated for 4 h at 37�C in a tissue cultureincubator. After the incubation the cells and media wereremoved from the top of the plate, the plate was centrifugedbriefly and the filter was removed. The cells that migrated intothe lower chamber were quantified with Promega Cell Titerusing a standard curve of THP-1 cells. The percent inhibitionwas calculated as the decrease in the number of cells migratedin the presence of Fab compared with the chemokine alone.

Results and discussion

Overall structure of MCP-1–11K2FabTo understand the cross-reactivity of antibody–MCP binding,we determined the crystal structure of synthetic hMCP-1 com-plexed to the Fab fragment of the 11K2 mAb (Table I). Thestructure was solved using data from a single crystal bymolecular replacement with a model of the Fab as the searchmodel. The position, orientation and stoichiometry of MCP-1were clear in the electron density maps calculated after rigidbody and positional refinement of the Fab fragment. The asym-metric unit of the crystal contains one MCP monomer and oneFab molecule (Fig. 1A).

In previous structures of MCP molecules, the chemokine is adimer with the N-terminal residues of the protein forming thedimer interface. In this structure, the MCP-1 dimer is generatedusing crystallographic symmetry. Based on the structure, theMCP-1 dimer should bind to two 11K2 antibodies in solution,as the orientation of the Fab fragments does not allow forbivalent binding of a single antibody (Fig. 1B). This resultis supported by light scattering studies in which two antibodiesappear to bind one MCP-1 dimer, and by ELISA experimentsshowing that MCP-1 and MCP-2 can bridge two 11K2 mAbstogether (data not shown).

Superposition of the complexed MCP-1 dimer with nativestructures [PDB codes 1DOL and 1DOK (Lubkowski et al.,1997)] shows that there are few changes to the overall structureof the MCP-1 monomer (CA RMSD = 0.93 and 0.96 s overresidues 4–70 for 1DOL and 1DOK, respectively), but theorientation of the monomers to each other has been slightlyaltered. Figure 1C shows that the orientation of the secondMCP-1 monomer is within the variation seen with uncom-plexed MCP-1 X-ray structures. 11K2 does not bind to theresidues involved in dimer formation. Therefore, the changesin orientation of the second monomer relative to the first mo-lecule are presumably owing to small local changes of thestructures or effects of crystal packing rather than the bindingof the antibody. The variability of dimer orientation in MCP-1structures solved by X-ray and NMR has been commented onpreviously (Lubkowski et al., 1997).

11K2Fab binds to MCP-1 at a relatively flat region of thechemokine. The antibody binds to MCP-1 in a region consist-ing of the end of b1, the b1– b2 loop, the beginning of b2, theloop between b3 and a1, and the a1 helix. The formation of thecomplex buries �1550 s

2 of solvent exposed surface area and�15% of the MCP-1 surface is involved in the interface. Mostof the contacts with MCP-1 are made through CDRs H1 andH3 of the heavy chain and CDRs L1 and L3 of the light chain.The center of the interface is dominated by hydrophobic

Table I. Data collection and refinement statistics

hMCP-1–11K2Fab

Data collectionSpace group C2221

Cell dimensionsa, b, c (A) 86.36, 89.10, 176.24a, b, g (�) 90.0, 90.0, 90.0

Resolution (A) 50–2.5 (2.59–2.5)a

Rsymb 0.090 (0.436)

I/sI 21.2 (3.3)Completeness (%) 99.8 (99.9)Redundancy 6.2 (5.6)

RefinementResolution (A) 35–2.5No. reflections 22739Rwork/Rfree

c 0.219/0.277No. of atoms

Protein 3822Water 43

B-factorsProtein 46.6Water 34.8

RMSDsBond lengths (A) 0.007Bond angles (�) 1.376

aNumbers in parentheses refer to outer resolution shell.bRsym = S j I� <I>j/ S I, where I is the integrated intensity of a given reflection.cR-value = S jF(obs) – F(calc)/ S F(obs), where F(calc) represents the structurefactor amplitudes obtained from back transformation of the model. Ten percentof the data were used in calculating Rfree.

319319

MCP-1–11K2Fab structural analysis and mutagenesis

interactions. Phe101 of the antibody heavy chain (Phe101H) isimbedded in a hydrophobic pocket of MCP-1 composed ofArg30, Thr32, Glu39, Val41, Pro55 and Met64 (Fig. 2Aand B). The pocket is created by the movement of Arg30about chi3 from its native conformation. Other than Arg30,very little movement of the MCP-1 residue side chains isneeded to accommodate the insertion of the phenylalanine.

The periphery of the interface is composed of hydrophilicsalt bridge pairs and hydrogen bond interactions. Lys56 of

MCP-1 is involved in a salt bridge–hydrogen bond interactionwith Asp52 of the heavy chain (Asp52H) and makes an addi-tional interaction with the Lys30H backbone carbonyl oxygen(Fig. 2C). Asp65 and Asp68 of the chemokine are forminghydrogen bonds with Arg32 of the light chain (Arg32L)(Fig. 2D). Glu39 makes a hydrogen bond with Thr32H andstacks against Arg98H (Fig. 2E). Multiple water-mediatedand backbone hydrogen bonds are also made between thetwo molecules.

Fig. 1. Ribbon diagrams of hMCP-1 complexed to 11K2. (A) A stereoview of the asymmetric unit of the hMCP-1:11K2 crystal (MCP-1, green; 11K2 heavychain, blue; 11K2 light chain, yellow). The color scheme presented here is maintained in all ribbon figures. The structural elements of MCP-1 and the CDRchains of the Fab have been labeled. (B) The MCP-1 dimer and resultant Fab molecules generated by crystallographic symmetry (MCP-1, pink; 11K2 heavychain, cyan; 11K2 light chain, orange). The two views differ by �90� rotation. The orientation of the Fab molecules precludes binding of the MCP-1 dimericmolecule by a single 11K2 antibody. (C) A superposition of MCP-1 dimers solved by X-ray crystallography (MCP-1:11K2, green and magenta; 1DOL, grey;1DOK, cyan). Only one monomer of the dimer was superimposed, showing the variation in the orientation of the remaining monomer. All structuralrepresentations were made using Ribbons (Carson, 1997).

320320

C.Reid et al.

How 11K2 blocks CCR2 bindingThe binding site of CCR2 on MCP-1 has been mapped to aregion of the chemokine containing Tyr13, Arg24, Lys35,Lys38, Lys49 (Fig. 3) (Hemmerich et al., 1999). These resi-dues form a discontinuous binding site that is mostly distinctfrom the binding site of 11K2. However, Lys38 is in van derWaals contact with the heavy chain of 11K2Fab. Given theclose proximity of the 11K2 binding surface with the CCR2binding surface, it is likely that the activity of 11K2 is owing tosteric blockage of the MCP-1–CCR2 complex as opposed todirectly overlapping binding sites.

To confirm that the binding of 11K2 functionally blockshMCP-1 and hMCP-2, we tested the effect on chemotaxis(Fig. 4). Human THP-1 monocytic cells migrated in responseto hMCP-1, and less robustly to hMCP-2. Pre-incubating thechemokines with 11K2Fab resulted in the inhibition ofchemotaxis. As expected the concentration of antibody atwhich 50% of the chemotaxis signal is inhibited is roughlyequal to the concentration of chemokine used in the assay,since the concentrations of both chemokines that resulted inmeasurable chemotaxis are well above the KD for the bindinginteraction of the Fab with the chemokines.

Fig. 2. Highlighted interactions between hMCP-1 and 11K2Fab. (A) Phe101H and surrounding residues in the complex. (B) Omit electron density map of thisregion contoured to 3s. (C) Lys56 forms two hydrogen bonds with 11K2. (D) Negatively charged residues at the C-terminus of hMCP-1 interact with Arg32L of11K2. (E) Glu39 interacts with 11K2 through polar interactions.

Fig. 3. Comparison of CCR2 and 11K2Fab binding sites. hMCP-1 monomeris shown as a CPK representation. Residues of MCP-1 involved in CCR2binding are shown in red. The binding site of 11K2Fab on hMCP-1 does notoverlap extensively with the CCR2 binding site, but blocks CCR2 bindingowing to steric interference.

321321

MCP-1–11K2Fab structural analysis and mutagenesis

Mapping 11K2 specificityThe residues involved in the binding of Phe101H are identicalbetween hMCP-1, hMCP-2 and hMCP-3; all residue differ-ences between the three chemokines at the interface of thecomplex are located at the periphery of the interaction(Fig. 5). This was surprising, since hMCP-1 and hMCP-2bind to 11K2Fab with subnanomolar affinity, while no bindingis detected to hMCP-3. Therefore, peripheral residues arelikely involved in the specificity of these interactions. Addi-tionally, there are more residue differences between hMCP-1

and hMCP-2, the molecules that bind to 11K2 with highaffinity, than between hMCP-1 and hMCP-3 or hMCP-2 andhMCP-3, which differ greatly in binding the antibody(Table II). This suggests that specific interactions by one ormore residues of hMCP-1 and hMCP-2 with the antibodyaccount for the tight binding of 11K2 to these chemokines.This hypothesis was tested by constructing a series of muta-tions to the MCP molecules and to 11K2Fab and then theaffinity of the interactions was determined (Tables IIand III). These mutation experiments are now described insections labeled by the location in the chemokines or 11K2.

Lysine 56 (MCP-1)It was speculated that Lys56 of MCP-1 is important to 11K2binding owing to the fact that most high-affinity MCP isoforms(hMCP-1, hMCP-2, mMCP-1 and mMCP-5) have a Lys at thisposition, whereas the isoforms with low affinity [hMCP-3,mMCP-2, rat MCP-1 (rMCP-1)] have different residue types(Thr, Thr and Asn, respectively). In this structure Lys56 formstwo hydrogen bonds with the side chain of Asp52H and thebackbone carbonyl of Lys30H of 11K2Fab. Mutation of Lys56to Asn resulted in a 160-fold reduction in affinity, implyingthat this residue plays a key role in the selectivity profile ofthe antibody. Interestingly, mutation of Lys56 to Arg resultedin a 14-fold loss in activity (Table II). Therefore, the specific

Fig. 4. Inhibition of chemotaxis in response to 2.3 nM hMCP-1 or 56 nMhMCP-2 by 11K2Fab.

Fig. 5. Sequence comparison of MCP chemokines. (A) Sequence alignment of MCP chemokines (h, human; m, mouse; r, rat). (B) CPK representation of hMCP-1 monomer. Residues of hMCP-1 that are at the complex interface have been color coded: cyan, residues identical between hMCP-1, -2 and -3; yellow, residuesdifferent between hMCP-1 and -2; orange, residues different between hMCP-1 and -3; red, residues different between hMCP-1, -2 and -3.

322322

C.Reid et al.

size and conformation of Lys, not just the positive charge, areimportant in this specific interaction.

Since the residue at position 56 of the chemokine is likely tobe important in determining the selectivity against hMCP-3,this hypothesis was tested by mutating residue 56 on hMCP-3.As hypothesized, mutation of hMCP-3 Thr56 to lysineincreased MCP-3 binding for 11K2Fab with an affinity of1.7 nM for mutant hMCP-3 Thr56Lys versus no detectablebinding to native hMCP-3. Therefore, while this result high-lights the importance of Lys56 for the interaction with 11K2, italso underscores the importance of other interface residues,since the mutant MCP-3 (Thr56Lys) still has 4- and 350-foldlower affinity for 11K2 than hMCP-2 (430 pM) and hMCP-1(4.6 pM), respectively.

Phenylalanine 101 (11K2 heavy chain)At the center of the interface, Phe101H is inserted into ahydrophobic pocket of MCP-1. We mutated residues of MCP-1and 11K2 to determine the importance of this phenylalanineand nearby residues that were expected to affect this inter-action. Not surprisingly, mutation of Phe101H to arginine

resulted in a complete loss of binding. Introducing a bulkierresidue (Val100H Phe) near Phe101H also results in a 16 500-fold decrease in affinity (Table III).

To further explore this area of interaction we introducedseveral mutations in MCP-1. Replacing Glu39 with lysinedisrupts the charge–charge interaction across the interfacewith Arg98H and introduces a longer side chain (Fig. 2E).Mutation of Val41 to lysine introduces a flexible and chargedside chain into an area in the center of the hydrophobic pocketthat is tightly packed and has no compensating negativecharge. As a result these MCP-1 mutants have >1000-foldaffinity losses (Glu39Lys, Val41Lys; Table II). An additionalMCP-1 mutant, Pro55Phe, would not refold under the standardprotocols indicating the stability of this variant protein wascompromised. Therefore, the packing of this area is importantfor the stability of MCP-1 itself and is critical for the highaffinity of the MCP-1–11K2Fab complex.

MCP-1 C-terminusResidues involved in 11K2 binding at the C-terminus offerparticularly interesting targets for mutagenesis. Some ofthese residues are different between hMCP-1 and hMCP-2,and they might explain the slight differences in affinity be-tween hMCP-1 and hMCP-2 to 11K2. In hMCP-1, Asp65 andAsp68 form salt bridges with Arg32L of 11K2 (Fig. 2D).Interestingly, Asp65 is a lysine residue in hMCP-2 (andmMCP-1) and thus cannot form the same salt bridge. Mutationof Asp65 to lysine decreased affinity of this mutant by only2.5-fold over wild-type. Because the side chain of Asp65 issolvent exposed, mutation to lysine may be relatively easilyaccommodated and results in a modest decrease on MCP-1binding. Mutation of Lys65 to an aspartic acid in hMCP-2increased affinity of hMCP-2 to 11K2 3.5-fold, which isconsistent with the corresponding affinity decrease of theAsp65Lys mutation in hMCP-1 (Table II).

The salt bridge triad between Asp65 and Asp68 in hMCP-1and Arg32L in 11K2 was further probed by making mutationsin 11K2Fab. It was predicted that changing Arg32L to glu-tamic acid would decrease hMCP-1 binding because of a lossof the salt bridge, but perhaps increase hMCP-2 bindingbecause a new salt bridge could be formed with Lys65. AnArg32L to tryptophan change was speculated to decrease11K2Fab binding to both hMCP-1 and hMCP-2 because inaddition to removing the salt bridge, a very bulky sidechain was introduced. As predicted, the Arg32L Glu mutationhad a very large effect on hMCP-1 binding (�10 000-foldaffinity reduction), but did not affect hMCP-2 binding dramati-cally (2.5-fold affinity reduction). Changing Arg32L to tryp-tophan decreased hMCP-1 binding, although only 6-fold, andsurprisingly increased hMCP-2 binding 70-fold. The complexwas able to accommodate the large tryptophan residue withoutseverely disrupting binding, and in the case of hMCP-2 theintroduced tryptophan picked up a significant interaction,perhaps with the side chain of Lys65.

Glutamate 39 (MCP-1)The side chain of Glu39 in MCP-1 is inserted in a shallowpocket on the 11K2 heavy chain, and contacts the side chain ofArg98H on 11K2. It was hypothesized that reversing thecharges on one or the other binding partner would destroythe binding, but that the charge-reversed mutants wouldbind one another with higher affinity than either mutant to

Table III. Mutant panel and binding affinitiesa

Antibodyb hMCP-1 hMCP-2 hMCP-1E39K

hMCP-1D65K

11K2 7.5 (3.8%)c 428* (2.3%) 82 600* (4%) 13* (6.3%)Phe101H Arg No bindingd NDe ND NDThr32H Ala 920 (10.9%) ND ND NDAsp52H Ala 151 (3.6%) ND ND NDAsp52H Asn 261 (4.4%) ND ND NDVal100H Phe 124 000 (6.6%) ND ND NDArg98H Asp 7736 (3.3%) ND 146 (5.3%) NDArg98H Asn 98 000 (4.4%) ND >2 000 000 NDArg32L Glu 45 460 (2.8%) 1076 (2.7%) ND 641 (3.1%)Arg32L Trp 46 (5.4%) 6.2 ND 5 (3.9%)

aAffinities are in pM.bAll measurements except those marked with ‘*’ were performed using antibodysupernatants.cNumbers in parentheses are the percent error of the curve fits.dThis Fab produced no signal when tested in the KinExA for binding to MCP-1Fc beads, therefore no affinity could be measured.eNot determined.

Table II. Affinity measurements of native and mutant MCPs with

wild-type 11K2

Antigena Binding affinity with 11K2 (pM)

Native hMCP-1 4.6 (5%)b

Native hMCP-2 428 (2.3%)Native hMCP-3 >2 000 000Native mMCP-1 243 (2.5%)Native mMCP-5 1500 (4.6%)hMCP-1 Glu39Lys 82 600 (4%)hMCP-1 Lys56Asn 751 (10.3%)hMCP-1 Lys56Arg 66 (5.6%)hMCP-1 Asp65Lys 13 (6.3%)hMCP-1 Arg30Ala 305 (6.7%)hMCP-1 Val41Lys 6190 (10%)hMCP-2 Lys65Asp 125 (8.9%)hMCP-3 Thr56Lys 1787 (5.1%)

aAll measurements were performed using purified Fab and purified chemokines.bNumbers in parentheses are the percent error of the curve fits.

323323

MCP-1–11K2Fab structural analysis and mutagenesis

the wild-type partner. Mutating the 11K2Fab residue Arg98Hto aspartate caused a three order of magnitude loss in bindingto wild-type hMCP-1 (Table III). Mutating the MCP-1 residueGlu39 to lysine decreased the binding to wild-type 11K2Fab byfour orders of magnitude (Table II). However, much of thebinding affinity was restored when these two mutant proteinswere measured against one another (Table III; Fig. 6).

Implications for therapeutic useThe 11K2 mAb has some unique properties that make it worthyof detailed structural study. Although it was raised in normalmice against hMCP-1, the mAb recognizes several membersof the MCP family across species. Without any special effortsto break tolerance to self proteins, the immunized mouse pro-duced a mAb that reacted against two mouse proteins withhigh affinity. 11K2Fab binds to mMCP-1 with an affinity of�300 pM and mMCP-5 with an affinity of �1.5 nM (Table II).

The chemokine system is characterized by redundancy andpromiscuity, with multiple chemokines binding to the samereceptors and individual chemokines capable of binding andactivating multiple receptors. This redundancy may complicatechemokine blocking strategies from having clinical efficacy.One strategy to overcome this limitation is to design inhibitorsthat are specific for more than one receptor, or for more thanone chemokine. In the ApoE�/� murine atherosclerosis model,transcript profiling identified both mMCP-1 and mMCP-5 asupregulated in diseased tissue. It was hypothesized that block-ing both of these chemokines may be important in controllingdisease (Lutgens et al., 2005). The unexpected modulation ofbinding affinity for hMCP-1 and hMCP-2 with the Arg32L Trpmutation of 11K2 illustrates the potential power of antibodyengineering in modulating the binding of pan-antibodies totheir target antigens.

Acknowledgements

We thank Charles R. Mackay for the initial 11K2 mAb, Rich Tizard and theBiogen Idec DNA sequencing lab for oligonucleotide synthesis and DNAsequencing, Ami Horne for assistance with crystallization, RandyAbramowitz and the staff at X4A for help in data collection, and LauraSilvian and Alan Corin for helpful discussions.

ReferencesAlexander,J.M., Nelson,C.A., van Berkel,V., Lau,E.K., Studts,J.M., Brett,T.J.,

Speck,S.H., Handel,T.M., Virgin,H.W. and Fremont,D.H. (2002) Cell, 111,343–356.

Blaszczyk,J., Coillie,E.V., Proost,P., Damme,J.V., Opdenakker,G.,Bujacz,G.D., Wang,J.M. and Ji,X. (2000) Biochemistry, 39, 14075–14081.

Boring,L., Gosling,J., Cleary,M. and Charo,I.F. (1998) Nature, 394, 894–897.Brunger,A.T. et al. (1998) Acta Crystallogr., D54, 905–921.

Campbell,E.M., Charo,I.F., Kunkel,S.L., Strieter,R.M., Boring,L., Gosling,J.and Lukacs,N.W. (1999) J.Immunol., 163, 2160–2167.

Carson,M. (1997) J. Appl. Crystallogr., 24, 958–961.Chakravarty,L., Rogers,L., Quach,T., Breckenridge,S. and Kolattukudy,P.E.

(1998) J. Biol. Chem., 273, 29641–29647.Dawson,T.C., Kuziel,W.A., Osahar,T.A. and Maeda,N. (1999) Atherosclerosis,

143, 205–211.Fields,C.G., Lloyd,D.H., Macdonald,R.L., Otteson,K.M. and Noble,R.L. (1991)

Pept. Res., 4, 95–101.Fujinaka,H., Yamamoto,T., Takeya,M., Feng,L., Kawasaki,K., Yaoita,E.,

Kondo,D., Wilson,C.B., Uchiyama,M. and Kihara,I. (1997) J. Am. Soc.Nephrol., 8, 1174–1178.

Furukawa,Y., Matsumori,A., Ohashi,N., Shioi,T., Ono,K., Harada,A.,Matsushima,K. and Sasayama,S. (1999) Circ. Res., 84, 306–314.

Gu,L., Okada,Y., Clinton,S.K., Gerard,C., Sukhova,G.K., Libby,P. andRollins,B.J. (1998) Mol. Cell, 2, 275–281.

Handel,T.M. and Domaille,P.J. (1996) Biochemistry, 35, 6569–6584.Hayashida,K., Nanki,T., Girschick,H., Yavuz,S., Ochi,T. and Lipsky,P.E.

(2001) Arthritis Res., 3, 118–126.Hemmerich,S. et al. (1999) Biochemistry, 38, 13013–13025.Huang,M., Syed,R., Stura,E.A., Stone,M.J., Stefanko,R.S., Ruf,W.,

Edgington,T.S. and Wilson,I.A. (1998) J. Mol. Biol., 275, 873–894.Jones,T.A., Zou,J.Y. and Cowan,S.W. Kjeldgaard. (1991). Acta Crystallogr. A,

47, 110–119.Karpus,W.J. and Kennedy,K.J. (1997) J. Leukoc. Biol., 62, 681–687.Kim,K.-S., Rajarathnam,K, Clark-Lewis,I. and Sykes,B.D. (1996) FEBS Lett.,

395, 277–282.Kimura,H. et al. (1998) Lab. Invest., 78, 571–581.Kurihara,T., Warr,G., Loy,J. and Bravo,R. (1997) J. Exp.Med.,186, 1757–1762.Kusano,K.F. et al. (2004) Circ. J., 68, 671–676.Li,H., Dunn,J.J., Luft,B.J. and Lawson,C.L. (1997) Proc. Natl Acad. Sci. USA,

94, 3584–3589.Lloyd,C.M., Minto,A.W., Dorf,M.E., Proudfoot,A., Wells,T.N., Salant,D.J. and

Gutierrez-Ramos,J.C. (1997) J. Exp. Med., 185, 1371–1380.Lu,B., Rutledge,B.J., Gu,L., Fiorillo,J., Lukacs,N.W., Kunkel,S.L., North,R.,

Gerard,C. and Rollins,B.J. (1998) J. Exp. Med., 187, 601–608.Lubkowski,J., Bujacz,G., Boque,L., Domaille,P.J., Handel,T.M. and

Wlodawer,A. (1997) Nat. Struct. Biol., 4, 64–69.Lutgens,E. et al. (2005) Circulation, 111, 3443–3452.Meunier,S., Bernassau,J.-M., Guillemot,J.-C., Ferrara,P. and Darbon,H. (1997)

Biochemistry, 36, 4412–4422.Navaza,J. (1994) Acta Crystallogr. A, 50, 157–163.Ono,K., Matsumori,A., Furukawa,Y., Igata,H., Shioi,T., Matsushima,K. and

Sasayama,S. (1999) Lab. Invest., 79, 195–203.Otwinowski,Z.and Minor,W. (1997) Methods Enzymol., 276, 307–326.Proudfoot,A.E., Handel,T.M., Johnson,Z., Lau,E.K., LiWang,P., Clark-

Lewis,I., Borlat,F., Wells,T.N. and Kosco-Vilbois,M.H. (2003) Proc. NatlAcad. Sci. USA, 100, 1885–1890.

Rollins,B.J. (1997) Blood, 90, 909–928.Rossi,D. and Zlotnik,A. (2000) Annu. Rev. Immunol., 18, 217–242.Rot,A. (1993) Eur. J. Immunol., 23, 303–306.Sali,A. and Blundell,T.L. (1993) J. Mol. Biol., 234, 779–815.Sartipy,P. and Loskutoff,D.J. (2003) Proc. Natl Acad. Sci. USA, 100,

7265–7270.Schneider,A., Panzer,U., Zahner,G., Wenzel,U., Wolf,G., Thaiss,F.,

Helmchen,U. and Stahl,R.A. (1999) Kidney Int., 56, 135–144.Sorensen,T.L., Ransohoff,R.M., Strieter,R.M. and Sellebjerg,F. (2004)

Eur. J. Neurol., 11, 445–449.Zlotnik,A.and Yoshie,O. (2000) Immunity, 12, 121–127.

Received November 29, 2005; revised March 2, 2006;accepted March 15, 2006

Fig. 6. Solution phase binding of wild-type and mutant 11K2 to wild-typeand mutant hMCP-1.

324324

C.Reid et al.