Structure and humanization of a rat monoclonal Fab to human ...

Protein Engineering vol.9 no.7 pp.623-628, 1996

Structure and humanization of a rat monoclonal Fab to humaninterleukin-5

William J.Cook1, Leigh J.Walter2, Nicholas J.Murgolo3,Chuan-Chu Chou4, Mary Petro4, Paul J.Zavodny4,Satwant K.Narula4, Lata Ramanathan3, Paul P.Trotta3-6

and Tattanahalli L.Nagabhushan5

Department of Pathology, University of Alabama at Birmingham. 506Kracke Building, 619 South 19th Street, Birmingham. AL 35233-6823,^Center for Macromolecular Crystallography, University of Alabama atBirmingham, Birmingham, AL 35294 and Departments of StructuralChemistry, 4Immunology and 'Biotechnology Development, Schering-Plough Research Institute, Kenilworth, NJ 07033, USA

'Present address: PharmaGenics, Inc., Allendale, NJ 07401, USA

'To whom correspondence should be addressed

The X-ray crystal structure of a rat monoclonal Fab JES1-39D10, raised against recombinant human interleukin-5,has been determined with the use of molecular replacementtechniques and refined at 2.7 A resolution by simulatedannealing. The overall structure is similar to a murine FabHyHEL-10 that is specific for hen egg white lysozyme. Aninteresting feature of the structure is the presence of leucineresidues to support the HI complementarity-determiningregion (CDR) loop. To our knowledge this is the first Fabcrystal structure containing this unusual HI loop supportpattern. The activity of three humanized versions of 39D10is explained by analysis of Fv interface residues and HIsupport patterns of 39D10 and the human template HLL.Keywords: antibody/humanization/interleukin-5/X-ray crystal-lography

IntroductionMonoclonal antibodies have become important in the treatmentand diagnosis of human disease (for a review see Bach et ai,1993). For clinical use, human antibodies are preferred, but itis generally difficult to produce them. The production of rator mouse antibodies is usually straightforward and thereforeantibodies against specific targets are often produced inanimals. A drawback to the use of animal antibodies in manis their potential antigenicity. Several studies have shown that'humanization' of non-human antibodies -provides a way tominimize this antibody response, while retaining the highlyspecific and tight binding characteristic of the antigen-antibodyinteraction (for reviews see Winter and Milstein, 1991; Winterand Harris, 1993). Typically, humanization involves graftingthe antigen-recognizing complementarity determining regions(CDRs) from an animal antibody onto a human antibodyframework. However, it has been demonstrated that this stepalone is not always sufficient and that other key residuesoutside the hypervariable regions may play an important rolein antigen binding, whether through direct antigen interaction(Amit etal., 1986; Padlan etai, 1989) or by providingstructural support for a CDR (Chothia etai, 1989; Novotnyetal., 1990; Tramontano etal., 1990; Tempest etai, 1991;Foote and Winter, 1992; Presta etai, 1993; Schildbachetai, 1993).

Interleukin-5 (IL-5) is a hematopoietic growth factor that isresponsible for eosinophil differentiation (Campbell etai,1987; Yokota etai, 1987). As a specific cytokine for eosino-philpoiesis, IL-5 plays an important role in diseases associatedwith increased eosinophils, such as asthma (Sanderson, 1992;Takatsu, 1992). Therefore, specific antibodies against IL-5 maybe useful therapeutic agents in treating diseases caused by IL-5-induced eosinophils (for a review see Chand and Sofia,1993). Using a guinea pig model of asthma, Mauser etal.(1993) showed that the rat monoclonal antibody TRFK-5against murine IL-5 blocked airway hyper-responsiveness andeosinophil infiltration. Recently TRFK-5 has been shown toinhibit virus-induced airway hyper-responsiveness to histaminein guinea-pigs (van Oosterhout etai, 1995) and to blockairway hyper-responsiveness in monkeys (Mauser et al., 1995).The rat antibody JES1-39D10, which has been raised againsthuman IL-5 (Denburg etai, 1991), is a neutralizing antibodythat blocks the IL-5-dependent growth of eosinophil-basophilcolonies in culture. Antibody 39D10 is specific for human IL-5 and does not cross-react with mouse IL-5 in an indirectELISA (Denburg et ai, 1991). A humanized version of 39D10has recently been shown to inhibit pulmonary eosinophiliain monkeys with an extended biological duration (Eaganetai, 1995).

While considerable progress in the successful humanizationof an antibody can be made prior to knowledge of its three-dimensional (3-D) structure, based on our present understand-ing of other antibody structures, success is more likely if theprecise structure of the target antigen-binding site is known.To aid in our efforts to derive a humanized version of 39D10,we undertook the determination of the crystal structure of the39D10 Fab fragment. The structure allowed a more accurateinterpretation of the mutagenesis experiments designed to graftthe antigen-binding site onto a human antibody framework.The structure of human IL-5 (Milburn etai, 1993) may allowthe prediction of which residues in the CDR loops of theantibody are important for IL-5 recognition and binding,although the epitope of 39D10 has not yet been determined.

Materials and methodsCloning of 39D10Cloning, expression, purification and humanization of 39D10have been described extensively elsewhere (Chou et ai, 1993).Briefly, two oligonucleotides were designed corresponding tothe N-terminus of the heavy chain of a known rat antibody(IgG2a CAMPATH-1 YTH 34.5HL; Reichmann etai, 1988)and the C-terminus of the rat immunoglobulin G2a heavy-chain constant region (Bruggemann etai, 1986). Severalrestriction sites were introduced at the 5' end of the oligonucleo-tides to facilitate cloning. Another two oligonucleotides weredesigned corresponding to the untranslated region adjacent tothe initiation codon (Hellman et ai, 1985) and the 3'-untrans-lated region of a published rat immunoglobulin kappalight-chain cDNA (Hellman etai, 1985). Using these oligonu-

© Oxford University Press 623

Dow

nloaded from https://academ

ic.oup.com/peds/article/9/7/623/1527838 by guest on 10 August 2022

WJ.Cook et al

cleotides as primers, cDNA fragments of the complete lightchain and a truncated heavy chain of the 39D10 antibodywere isolated by polymerase chain reactions (PCRs), cloned,sequenced and confirmed by comparison with other clonesgenerated using PCR primers designed from different regionsof the cDNAs. The deduced amino acid sequence of the lightchain was verified by the sequencing of the first 15 N-terminalresidues. N-terminal sequencing of the heavy chain wasunsuccessful despite repeated efforts, possibly due to derivatiz-ation and blocking of the N-terminus, which is commonlyfound with immunoglobulins. Cloned 39D10 was expressedin COS cells and its IL-5 binding was confirmed by Westernblot and human IL-5 specific ELISA (Chou etal, 1993).

Proteolytic generation and purification of 39D10 FabfragmentThe hybridoma cell line producing human IL-5-neutralizingmonoclonal antibody 39D10 was produced as described byDenburg etal. (1991). Isotyping of 39D10, using a kit fromZymed (San Francisco, CA), revealed that the heavy chainwas an IgG^ isotype and the light chain was a kappa isotype.The antibody was purified from hybridoma supematants byprotein-G affinity chromatography. It was dialyzed against50 mM sodium phosphate buffer, pH 7.4 containing 0.15 MNaCl (PBS), 10 mM EDTA and 25 mM 2-mercaptoethanoland concentrated to lOmg/ml by ultrafiltration. It was thendigested with papain at a 1:100 ratio of enzyme to substratefor 10 min at 37°C. The reaction was quenched by the additionof an active site inhibitor E-64 [/ranj-epoxysuccinyl-L-leucyl-amido(4-guanidino)butane] to a final concentration of 20 mM.It was dialyzed extensively against PBS and diluted with anequal volume of 1.5 M glycine, 3.0 M NaCl, pH 8.9. Tenmilligrams of papain digest were applied to a protein-G agarosecolumn (1 ml). The flow-through fraction which was enrichedin the Fab fragment was subjected to gel filtration chromato-graphy on a Sephacryl S-100 column (1.6X100 cm) that hadbeen equilibrated in PBS which aided the removal of highermolecular weight impurities. The Fab fragment eluted as apeak with a molecular weight of 50.3 kDa. Further purificationwas effected by cation-exchange chromatography on an S-Sepharose column (2 ml) that had been equilibrated in 10 mMsodium acetate, pH 5.0. The Fab fragment was eluted with 25column volumes of a 0-0.5 M NaCl gradient in the equilibrationbuffer. The eluted Fab fragment was dialyzed against phos-phate-buffered saline and stored at — 20°C prior to crystalliza-tion experiments. The purified Fab fragment had a bindingaffinity equivalent to intact IgG using a previously describedELISA protocol (Chou et al, 1993) and the KA was determinedas 22 pM at 25°C by BIAcore (unpublished data). It wasjudged to be >98% pure based upon SDS-PAGE and aminoacid analysis.

Crystallization and data collectionNumerous attempts to grow crystals of a 39D10-IL-5 complexwere unsuccessful. Crystals of 39D10 Fab were initiallyobtained by vapor-diffusion equilibration using a kit fromHampton Crystal that uses a modified sparse matrix samplingmethod (Jancarik and Kim, 1991). Each screen, which consistsof 50 sets of conditions, was run at 23°C and 4°C. After 2days at 4°C, one set of conditions (no.22 in the screen) usingPEG-4000 and 0.2 M sodium acetate gave small aggregatesof orthorhombic crystals. Refinement of these conditions wasattempted by systematically varying the pH, the PEG-4000concentration and the salt. Unfortunately, the crystals always

624

grew as aggregated plates and needles. In order to grow largesingle crystals suitable for X-ray diffraction, a macroseedingtechnique was used. To obtain seeds, crystalline aggregateswere crushed with a needle and stirred; another drop was thenstreaked with the needle. A number of small perfect crystalsgenerally grew along the needle trail through the drop. Thesesingle crystals (0.05X0.02X0.03 mm) were washed in a stabil-izing solution of 20% PEG-4000 and 0.2 M sodium acetate in0.1 M Tris buffer, pH 8.2 and transferred to 1 (il dropscontaining 10 mg of protein ml in the stabilizing solution. Thedrops containing the seed crystals were then equilibratedagainst the stabilizing solution and crystals with dimensionsup to 0.6X0.4X0.2 mm grew after 2-4 days.

X-ray diffraction data were collected using a Nicolet X-100A area detector. The data collection was carried out at23°C using Cu Ka radiation from a Rigaku RU-300 rotatinganode generator operating at 40 kV and 100 mA. The indexingand integration of the intensity data were done with theXENGEN processing programs (Howard et al., 1987). Of thepotential solutions generated by the indexing program, onehad angles and axial lengths consistent with either a trigonalor hexagonal space group. The cell parameters refined to a =b = 72.7 A and c = 181.1 A. Comparison of the integratedintensities of potentially equivalent reflections based on Lauesymmetry 3m (3ml or 3lm) and 6/m gave Rsym values of0.085, 0.269 and 0.258 respectively, thus identifying the pointgroup symmetry as 3ml. The systematic absence of reflections00/ with / =£ 3n indicated either space group P3|21 or itsenantiomorph P3221. Based on the volume of this cell and amolecular weight for the Fab of 46 675 Da, the calculatedvalue of Vm (Matthews, 1968) for six molecules per cell is2.96 A3/Da, which is in the usual range for proteins. There isone molecule in the asymmetric unit and the solvent volumefraction is ~58%. To obtain a complete data set with multiplemeasurements of all reflections, three data sets were collectedfrom three crystals. For all data sets the detector to crystaldistance was 22 cm and the detector 28 value was 20°.Oscillation frames covered 0.25° and were measured for 450 s.Approximately 400 frames of data were collected on eachcrystal. The crystals are stable to X-rays at room temperaturefor ~48 h and then begin to show steady deterioration. A totalof 62 351 reflections to 2.7 A resolution were processed andmerged into 13 687 unique reflections (86% complete). TheRsym value (based on intensities) was 8.14.

Structure determinationThe crystal structure was solved using the molecular replace-ment routines in XPLOR (Brunger, 1990). The search modelwas the 3.0 A crystal structure of the anti-lysozyme FabHyHEL-10 (PDB accession number 3HFM) (Padlan etal,1989). This antibody has 214 amino acids in the light chainand 215 in the heavy chain; there is ~69% sequence identityoverall with 39D10. All 429 residues from the crystal structurewere used as the search model. A series of cross-rotationfunctions was calculated using a radius of integration of 45 Aand data between 12 and 4 A (F < 2oF). The elbow angle ofHyHEL-10 was varied from the starting value in incrementsof 10° from 0 to 50° and the peaks of each search weresubjected to the Patterson correlation (PC) refinement proced-ure described by Brunger (1990). The entire molecule, the twodomains and, finally, the individual heavy and light chains ineach domain were treated as independent rigid bodies inthe PC refinement. This procedure produced a clear single

Dow

nloaded from https://academ

ic.oup.com/peds/article/9/7/623/1527838 by guest on 10 August 2022

Structure of a rat Fab to human IL-5

Table I. Summary of the refinement statistics

Resolution (A) Number of reflections % complete

6.00-4.474.47-3.833.83-3.453.45-3.183.18-2.982.98-2.832.83-2.70

203518671800167615271256794

0.1460.1500.1680 1940.2240.2500.286

96.594.091.789.486.582.776.6

R is defined as U\\FO\ -

maximum peak in the rotation search corresponding to achange in the elbow angle of 30° from the original HyHEL-10 model.

The model was rotated by the angles corresponding to thePC refinement solution and the translation function fromXPLOR was used with data from 12 to 4 A for each of thetwo enantiomorphic space groups. For space group P3|21 thehighest T value was 0.533 (Ta = 0.020) with the next highestpeak only 0.339. For space group P3221 the correspondingvalues were 0.261 (Ta = 0.019) and 0.258. The /f-factors foreach of these solutions, using data from 8 to 4 A, were 0.399and 0.486 respectively. These results clearly indicated that thecorrect space group is P3121.

RefinementThe coordinates were refined using the simulated annealingroutines in XPLOR (Brunger etal, 1987). Rigid body refine-ment of the best solution using 8 to 3 A data gave an fl-factorof 0.401. At this point all non-identical residues were replacedwith Ala residues and simulated annealing was performedusing data from 6 to 3 A with F > 2a(F). The fl-factordropped to 0.258. Several rounds of model building using2FO-FC maps allowed us to replace all of the Ala substitutionswith the correct residues. During the rebuilding process, severalof the loops in the heavy chain were omitted from therefinement and then modeled based on the 2FO—FC maps. Theresolution was gradually extended to 2.7 A and individualresidues were rebuilt as necessary from 2FO—FC maps usingcomputer graphics (Jones, 1978). Data above 6 A were notincluded in the refinement, since they would be particularlyinfluenced by disordered solvent. The C-terminal residue inthe light chain (Cys214) and the last two residues in the heavychain (GIu217 and Cys218) could not be identified in electrondensity maps.

The /?-factor at 2.7 A resolution using isotropic tempera-ture factors of 15.0 A2 for all atoms was 0.195. Individualthermal parameters were introduced with restraints on thedifferences between temperature factors of connected atoms.The final model includes 249 residues from both chains with3254 protein atoms. The R-factor, based on 10 964 reflectionsin the range 2.7 A =£ d =£ 6.0 A with intensities exceeding2.0<J, is 0.176. The /?-index for all reflections in this range(12 364 reflections) with no sigma cut-off is 0.216. Table Igives the /?-factor tabulated as a function of resolution. Thefinal 2FO-FC electron density map is generally of good quality(Figure 1). The root mean square deviations between Ccc atomsin the final model from those in the original search model of3HFM are 1.3 A for the variable domains (based on 219residues) and 1.6 A for the constant domains (based on 205residues). The two major differences are in the H3 loop, whichis shorter in 3HFM and the loop involving residues 134-136

Fig. 1. Stereo view of the H3 CDR loop in 39D10 with the associatedelectron density The 2FO—FQ map was calculated with phases based on therefined structure after removing VH residues 97-104. The typical ion pairobserved at the base of the H3 loop is labeled (Chothia et al., 1989;Wu et al, 1993). This figure was prepared using FRODO (Jones, 1978).

in the heavy chain, where the electron density is very poorfor 39D10.

Results and discussionOverall structureSuperposition of the ^-factor curve with theoretical curves fordifferent mean positional errors gives an estimated error of0.25-0.35 A in the atomic coordinates (Luzzati, 1952). Thefinal coordinates of 39D10 deviate from ideal bond lengthsand angles by 0.024 A and 4.7° respectively. The structurewas comparable to the 2.7 A resolution structures as determinedwith the Procheck program (version 3.0, Oxford Molecular Inc.;Laskowski et al, 1993). A Ramachandran plot (Ramachandranetal, 1963) of the <)>, y torsion angles for 39D10 showsseveral non-glycine residues with high energy conformations.In the light chain these residues are Ala51 and Ala68, whilein the heavy chain they are Thrl31 and Serl93. All of theseresidues occur in loops between strands of P-sheets. ResidueAla51, which occurs in the L-2 CDR loop, has torsion anglesthat are commonly seen for residues at this position in otherFabs. The torsion angles for Ala68 are also typical for residuesin this position, but this residue is almost always glycine inother Fabs. The strain in this region probably also explainsthe unusual torsion angles for Ser67 (<}> = 175°, y = 139°).Residue Thrl31 in the heavy chain is in a loop that has thepoorest electron density in the entire structure. Although themain chain tracing is relatively clear, there is very poor densityfor the side chains of residues 131-136 of the heavy chain.The same is true for the loop involving Serl93, although ingeneral the side chain density is reasonable for the residues inthis region.

The average fl-factor for all protein atoms is 22.3 A2. Ingeneral, there is a good correlation between the fi-factor andthe secondary structure of the chain. The highest temperaturefactors in the light chain occur at the C-terminus and in loopsinvolving residues 30-32, 92-93, 168-169 and 201-203, whilethe highest in the heavy chain occur around residues 130-136and 192-194 and at the N- and C-termini.

Fab 39D10 shows the typical antibody fold (Figure 2).Rotation angles based on the superpositioning of VL onto VH

625

Dow

nloaded from https://academ

ic.oup.com/peds/article/9/7/623/1527838 by guest on 10 August 2022

WJ.Cook el aL

Fig. 2. Drawing of Fab 39DI0 based on the Ccc positions. The heavy chain is green, the light chain is blue and the CDR loops are silver. Figures 2 and 3were prepared using Ribbons 2.0 (Carson, 1987).

Table II. Quaternary structure of 39D10

Elbow angle CHI-CL Translation vanable Translation constant

176.1° 174.7°Amino acid residues used in overlapVH V L

10-25 11-2633-40 32-3946-49 45-4866-71 61-6677-97 70-90

105-113 97-105

167.9° o.n A 1.80 A 1.91 A

cHi120-126139-158166-169176-185194-202209-212

CL111-117130-149159-162171-180190-198205-208

Calculations of the interdomain configuration were done with the program ELBOW wntten by D.Fremont and T.O.Yeates by overlapping Cnl onto CL andVH onto VL. Structurally conserved residues shown in the Table were used to determine the rotation and translation components relating the four domains aswell as the angles of the Fab. The translation vectors measure the distance from the pseudo-2-fold to the ideal 2-fold axes of each domain. The closestdistance relating the variable and constant pseudo-2-folds is defined as dmM. The elbow angle is the angle formed by the two pseudo rotation vectors relatingthe light and heavy chain in the constant and vanable regions. The two other angles are the pseudo-2-fold rotation angles that describe the relative dispositionof the light and heavy chain for each domain.

and CL and CH1 are given in Table II. The Fab is almostcompletely extended and there are no contacts between thevariable and constant portions of the sequence. To our know-ledge, 39D10 is the first rat Fab whose crystal structure hasbeen determined. However, superposition of the structure onother Fab structures gives typical r.m.s.d.s compared to otherFab structures. Fab 39D10 shows a high sequence homologywith two murine Fabs (D1.3 and HyHel-10) that are specificfor hen egg white lysozyme. Overall, both show similar tertiarystructures to 39D1O, although the D1.3 Fab shows moresimilarity in the CDR loops. The r.m.s.d between the Caatoms in 39D10 and D1.3 is only 1.1 A, based on residues1-213 in the light chain and residues 1-216 in the heavychain. The only major differences occur in the loop involvingresidues 132-137 in the heavy chain, which is the poorestdefined region in the structure of 39D10.

CDR loopsTable HI gives the sequence of 39D10 Fab and defines thestructural (Chothia et al, 1989) and Kabat (Kabat et al, 1991)regions. The CDR loops in the light chain of Fab 39D10correspond to canonical structure 2 for loop LI and structure1 for L2 and L3. For the heavy chain loops, HI and H2

each correspond to canonical structure 1 in their respectivecategories (Chothia etal., 1992). Loop H3 is relatively longand contains a type I' turn involving TyrlOO and GlylOl(Figure 1). With the exception of Asp 104, the remainder ofthe H3 loop residues are in an extended conformation. Thereare six aromatic residues in this loop; four of them (Tyr402,Phe403, Tyr405 and Trp406) are important for hydrophobicpacking against aromatic residues in the light chain. The sidechains of Tyr399 and Tyr400 project into the cavity formedby the six CDR loops and may form part of the IL-5 interactionsurface (Wuetal, 1993).

Structure in relation to humanization experimentsA detailed description of the humanized version design hasbeen presented elsewhere (Chou etal, 1993). Humanizationexperiments were initiated prior to knowledge of the 39D10X-ray structure. Briefly, the design strategy included thedetermination of the 39D10 sequence, construction of a homo-logy model of the 39D10 Fv region from the 1F19 crystalstructure (Lascombe etal, 1989), selection of appropriatehuman templates [HIL for VH and LAY for VL joined to ahuman IgG4 CH1-CH2-CH3 (plasmid p24BRH ATCC57413)/kappa CL (plasmid HuCK ATCC59173) clone] and determina-

626

Dow

nloaded from https://academ

ic.oup.com/peds/article/9/7/623/1527838 by guest on 10 August 2022

Structure of a rat Fab to human IL-5

Table III. Sequence of 39D10 Fab

Heavy Chain

1 10 20 30 40 50 60

I I I I I I IEVKLI^SGGGLVQPSQTLSLTCTVSGLSLTSNSVNHIRQPPGKGLEWMGLIWSNGDTDYN

-.###HI90

1 2 070 80 90 100 110

I I I I I ISAIKSRLSISRDTSKSQVFLKMNSLQSEDTAMYFCAREYY0Y7DYWGQGVMVTVSSAETT##### .„„...„-

H3130 140 ISO 160 170 180

I I I I I IAPSVYPLAPGTAIJCSNSMVTLGCLVKQyFPEPVTVTHNSGALSSGVHTFPAVLQSGLYTL

190 200 210 218

I I I ITSSVTVPSSTWSSQAVTCNVXHPASSTKVDIOUVPREC

Ufht Chain

1 10 20 30 40 50 60

I I I I I I IDIQMTOSPASLSASLGETISIECLASEQISSYIiAWYQQKPGKSPQLLIYGANSLQTQVPS

LI L270 80 90 100 110 120

I I I I I IWSGSGSATQYSLKISSMQPEDEGDYFCQQS YKFPNTFGAGTKLELKRADAAPTVSIFPP

L3130 140 150 160 170 180

I I I I I ISTEQLATGGA3VVCI>INNFYPRDISVXWKID<rrERHDGVLDSVTDQDSKDSTYSMSSTLS

190 200 210 214

I I I ILTKADTESHNLYTCEWHKTSSSPWKSFNRNEC

(-) indicates a structural CDR loop, (#) indicates a Kabat region and ( = )indicates overlap of a structural CDR loop and Kabat region.

tion of the animal residues to retain. The animal residuesselected for modification included residues from the structuraland Kabat CDRs as well as those supporting or proximal tothe CDR loops, as judged from the homology model. TableIV highlights sequence differences between three humanizedversions (CMX5-1, CMX5-2 and CMX5-5).

The crystal structure of HIL has been determined at 2.8 Aresolution (F.A.Saul and R.J.Poljak, unpublished; Brookhavenentry 8FAB). A comparison of the VH regions of 39D10 andHIL shows large differences predominantly in the H3 CDRloop, which is six residues longer in HIL than in 39D10.

Overall, the r.m.s.d. between 115 VH Ca atoms is 1.1 A,omitting extra residues in the CDR loop. The structure of theLAY antibody is not known.

Competitive binding ELISA assays (Chou etai, 1993)demonstrated that CMX5-2 and CMX5-5 had lower affinitiesfor IL-5 than 39D10, with Kd values that were 1.4 times theKd for 39D10. Version CMX5-1 had poorer affinity still; itsKd was 3.3 times the Kd for 39D010. Some of this differencein affinity could be attributed to inappropriate support of theHI loop. HI loop support residues suggested as critical bycomparison of the HIL VH (human template) and 39D10 VH

structures are 24, 27, 29 and 78 (Figure 3). In 39D10 theseresidues are Val, Leu, Leu and Val, while in the HIL antibodythey are Ala, Phe, Phe and Leu. Many examples of the HTLsequence and similar sequences are known at VH 24, 27, 29and 78. However, only one other antibody sequence is knownthat contains Val, Leu, Leu and Val at these positions and itis also a rat antibody (Abrams et al., 1993). To our knowledgethis is the first antibody structure elucidated with this HIsupport pattern. The structure of 39D10 shows that there isnot enough room for Leu at position 78 in VH, because theextra methylene would disrupt HI support. On the other hand,the structure of HTL suggests that there is not enough roomfor Val at position 24 because the extra methyl groups mightdisrupt HI support. This explains why preservation of thesefour residues as either all human or all animal is necessary tomaintain high affinity. It also helps explain the higher relativeaffinity of CMX5-2 and CMX5-5, which contain the HTLpattern, versus CMX5-1, which contains only one residuefrom the HIL pattern. Additional humanized versions wereconstructed with the animal HI support pattern at positions24, 27, 29 and 78 that had the same affinity for IL-5 as 39D10(Celltech Corp., 1995).

Summary

The crystal structure of 39D10 Fab provides molecular explana-tions for relative affinity and expression levels of somehumanized variants initially designed by homology modeling.Furthermore, the structure provides a basis by which frameworkresidues influencing CDR conformations may be determined.On the basis of our current understanding of the structure of39D10, we have achieved humanized versions that are ~92%human sequence and still retain the same affinity for IL-5 asthe animal antibody.

Table IV. Sequence differences between Fab 39D10, human templates and humanized versions

VL domain 13 22 24 34 53 54 55 56 68 71 73 105

39DI0LAYCMX5-ICMX5-2CMX5-5

A EV TV TV TV T

LQ

AN

ST

LR

QE

TA

AG

YFFFF

DDD

VH domain 23 24 27 28 29 30 37 49 73 76 77 78 80 94

39DI0HILCMX5-1CMX5-2CMX5-5

SAAAA

TIIII

VA

AA

LF

FF

ST

TT

LF

FF

TS

SS

IV

GAAAA

TNNNN

SR

QTTTT

VLLLL

LMMMM

FY

YY

(.) indicates that the residues are identical with 39D10 at this position.

627

Dow

nloaded from https://academ

ic.oup.com/peds/article/9/7/623/1527838 by guest on 10 August 2022

WJ.Cook et al.

Fig. 3. Drawing of the HI loops in 39D10 (green) and HIL (pink) based on the Ca positions. The Ca backbones have been superimposed by a least-squaresprocedure. Residues 24, 27, 29 and 78 are yellow for 39DI0 and blue for HIL.

ReferencesAbrams,J.S., Dalie.B., Le.H.V., Miller.K., Murgolo,N.J., Nguyen.H.,

Pearce.M, Tindall.S. and Zavodny,P.J. (1993) International Patent WO93/17106.

Amit,A.C, Mariuzza.R.A., Phillips,S.E.V. and Poljak.R.J. (1986) Science,233, 747-753.

Bach,J.F., Fracchia,G.N. and Chatenoud,L. (1993) Immunol. Today, 14,

Bruggemann,M., Free,J., Diamond,A., Howard,.!., Cobbold.S. andWaldmann,H. (1986) Proc. Natl Acad. Sci. USA, 83, 6075-6079.

Bfunger,A.T. (1990) Ada Crystallogr. Sect. A, 46, 46-57.Bfunger,A.T., Kuriyan,J. and Karplus,M. (1987) Science, 235, 458-460.Campbell.H.D., Tucker, W.Q.J., Hort.Y., Martinson.M.E., Mayo.C,

Clutterbuck,E.J., Sanderson,C.J. and Young.I.G. (1987) Proc. Natl Acad.Sci. USA, 84, 6629-6633.

Carson,M. (1987) J. Mol. Graph., 5, 103-106.Celltech Corp. (1995) UK Patent Application 9412230.6.Chand,N. and Sofia.R.D. (1993) Drugs Today, 29, 477^t85.Chothia,C. el al. (1989) Nature, 342, 877-883.Chothia,C, Lesk,A.M., Gherardi.E., TomlinsonJ.M., Walter.G., Marks.J.D.,

Liewelyn,M.B. and Winter,G. (1992) J. Mol. Biol., Ill, 799-817.Chou,C.-C, Murgolo,N.J., AbramsJ.S., Jehn,C.-H., Petro.M.E., SilverJ.E.,

Tindall,S., Windsor.W.T. and Zavodny,P.J. (1993) International PatentWO93/16184.

DenburgJ.A., SilverJ.E. and AbramsJ.S. (1991) Blood, 77, 1462-1468.Egan,R.W., Athwahl.D., Chou.C.C, Emtage,S., Jehn.C.H., Kung.T.T.,

Mauser,P.J., Murgolo.NJ. and Bodmer,M.W. (1995) Int. Arch. AllergyImmunol., 107, 321-322.

Foote,J. and Winter,G. (1992) J. Mol. Biol., 224, 487^199.Hellman,L., Engstrom.A., Bennich.H. and Pettersson.U. (1985) 40, 107-114.Howard,A.J., Gilliland.G.L., Finzel.B.C, Poulos,T.L., Ohlendorf.D.H. and

Salemme.F.R. (1987) J. Appl. Crystallogr., 20, 383-387.Jancarik,J. amd Kim,S.-H. (1991) J. Appl. Crystallogr., 24, 409-411.Jones.T.A. (1978) J. Appl. Crystallogr., 11, 268-272.Kabat,E.A., Wu,T.T., Perry.H.M., Gottesman.K.S. and Foller,C. (1991)

Sequences of Proteins of Immunological Interest, 5th edn. NIH Publication,Bethesda, MD: National Institute of Health, 91-3242.

Lascombe,M.-B., Alzari.P.M., Boulot,G., Saludjian.R, Tougard,P., Berek,C,Haba,S., Rosen,E.M., Nisonoff,A. and Poljak,R.J. (1989) Proc. Natl Acad.Sci. USA, 86, 607-611.

Laskowski,R.A., MacArthur.M.W., Moss.D.S. and Thornton.J.M. (1993) J.Appl. Crystallogr., 26, 283-291.

Luzzati.V. (1952) Acta Crystallogr., 5, 802-810.Matthews.B.W. (1968) J. Mol. Biol., 33, 491-497.Mauser.PJ. etal. (1993) Am. Rev. Respirat. Dis., 148, 1623-1627.Mauser.P.J., Pitman.A.M., Fernandez.X., Foran,S.K.. Adams.G.K.,

Kreutner.W., Egan,R.W. and Chapman.R.W. (1995) Am. J. Respirat. Crit.Care Med., 152, 467-472.

Milburn.M.V, Hassell.A.M., Lambert.M.H.. Jordan.S.R., Proudfoot.A.E.I..Graber,P. and Wells,T.N.C. (1993) Nature, 363, 172-176.

Novotny,J., Bruccoleri.R.E. and Haber,E. (1990) Proteins, 7, 93-98.Padlan.E.A., Silverton.E.W., Sheriff.S., Cohen,G.H., Smith-Gill.S.J. and

Davies,D.R. (1989) Proc. Natl Acad. Sci. USA, 86, 5938-5942.Presta,L.G., Lahr.S.J., Shields.R.L., Porter.J.P., Gorman,C.M., Fendly.B.M.

and Jardieu.P.M. (1993) J. Immunol., 151, 2623-2632.Ramachandran.G.N., Ramakrishnan.C. and Sasisekharan.V. (1963) J. Mol.

Biol., 7, 95-99.Reichmann.L., Clark.M., Waldmann.H. and Winter,G. (1988) Nature. 332,

323-327.Sanderson.CJ. (1992) Blood. 79, 3101-3109.SchildbachJ.F.. Near,R.I., Bruccoleri.R.E., Haber,E., Jeffrey.P.D.. Novotny.J.,

Sheriff,S. and Margolies.M.N. (1993) Protein Sci., 2, 206-214.Takatsu,K. (1992) Curr. Opin. Immunol.. 4. 299-306.Tempest.P.R., Brenner.P., Lambert.M., Taylor.G., Furze.J.M., Carr.F.J. and

Harris,W.J. (1991) Biotechnology, 9, 266-271.Tramontano,A., Chothia.C. and Lesk.A.M. (1990)7. Mol. Biol., 215, 175-182.van Oosterhout.A.J., van Ark.I., Folkerts.G., van der Linde.H.J.,

Savelkoul.H.F., Verheyen.A.K. and Nijkamp,F.P. (1995) Am. J. Respirat.Crit. Care Med., 151, 177-183.

Winter.G. and Harris,W.J. (1993) Immunol. Today, 14, 243-246.Winter,G. and Milstein.C. (1991) Nature, 349, 293-299.Wu.T.T., Johnson.G. and Kabat.E.A. (1993) Proteins, 16. 1-7.Yokota,T. etal. (1987) Proc. Natl Acad. Sci. USA, 84, 7388-7392.

Received September I, 1995; revised November 27, 1995; accepted January20, 1996

628

Dow

nloaded from https://academ

ic.oup.com/peds/article/9/7/623/1527838 by guest on 10 August 2022