- Email: [email protected]
The Crystal Structure of an Anti-CEA scFv Diabody Assembled from T84.66 scFvs in VL-to-VH Orientation: Implications for Diabody Flexibility
doi:10.1016/S0022-2836(02)01428-6
J. Mol. Biol. (2003) 326, 341–351
COMMUNICATION
The Crystal Structure of an Anti-CEA scFv Diabody Assembled from T84.66 scFvs in VL-to-VH Orientation: Implications for Diabody Flexibility Jennifer A. Carmichael1,2, Barbara E. Power1, Thomas P. J. Garrett1 Paul J. Yazaki3, John E. Shively3, Andrew A. Raubischek3 Anna M. Wu3 and Peter J. Hudson1* 1 CSIRO Health Sciences and Nutrition, 343 Royal Parade Parkville, Vic. 3052, Australia 2
Department of Biochemistry La Trobe University, Bundoora Vic. 3083, Australia 3
Department of Molecular Biology, Beckman Research Institute of the City of Hope Duarte, CA 91010, USA
Diabodies (scFv dimers) are small, bivalent antibody mimetics of approximately 55 kDa in size that possess rapid in vivo targeting pharmacokinetics compared to the intact parent antibody, and may prove highly suitable for imaging and therapeutic applications. Here, we describe T84.66Di, the first diabody crystal structure in which the scFvs comprise V domains linked in the VL-to-VH orientation. The structure was deter˚ resolution. The T84.66Di mined by X-ray diffraction analysis to 2.6 A scFv was constructed from the anti-carcinoembryonic antigen (anti-CEA) antibody T84.66 variable domains connected by an eight residue peptide linker to provide flexibility between Fv modules and promote dimer formation with bivalent affinity to the cell-surface target, CEA. Therefore, it was surprising to observe a close association of some Fv module complementarity-determining regions in the T84.66 diabody crystal, especially compared to other diabody structures all of which are linked in the opposite VH-to-VL orientation. The differences between the arrangement of Fv modules in the T84.66Di VL-to-VH linked diabody structure compared to the crystal structure of L5MK16 and other proposed VH-to-VL linked diabodies has been investigated and their potential for flexibility discussed. The comparison between VH-to-VL and VL-to-VH linked diabodies revealed in this study represents a limited repertoire of possible diabody Fv orientations, but one that reveals the potential flexibility of these molecules. This analysis therefore provides some signposts that may impact on future molecular designs for these therapeutic molecules with respect to diabody flexibility and avidity. q 2003 Elsevier Science Ltd. All rights reserved
*Corresponding author
Keywords: CEA; structure; dimers; diabody; flexibility
Present address: Dr T.P.J. Garrett, The Walter and Eliza Hall Institute of Medical Research, Royal Parade, Parkville, Vic. 3050, Australia. Abbreviations used: CEA, carcinoembryonic antigen; anti-CEA, antibody to carcinoembryonic antigen; Fab, antibody fragment produced by proteolysis; Ig, immunoglobulin; NA, influenza virus neuraminidase; scFv, single chain Fv molecule; T84.66Di, anti-CEA T84.66 diabody; VH, variable region from antibody heavy chain; VL, variable region from antibody light chain; CDR, complementarity-determining region. E-mail address of the corresponding author: [email protected]
Recombinant antibodies and their fragments are becoming the modern paradigm for the design of high-affinity, protein-based targeting reagents1,2 and now represent over 25% of all proteins undergoing clinical trials.3,4 Recombinant antibodies also capture a significant share of the US$6 billion £ 109 per annum immunodiagnostic market, from in vitro immunoassays to in vivo imaging reagents.4 Intact antibodies provide high target binding specificity but their use in rapid tumour targeting and in vivo imaging is limited by slow tissue penetration, long circulating half-lives and often undesirable effector functions.3 Small monovalent antibody fragments (scFv and Fab) exhibit good tissue penetration but their lack of avidity results in faster off rates and
0022-2836/03/$ - see front matter q 2003 Elsevier Science Ltd. All rights reserved
342
rapid clearance. Several laboratories have therefore engineered Fab or scFv molecules into dimers or larger multimers to produce high-avidity reagents of optimal size (50 –110 kDa) for in vivo imaging and therapeutics.5 – 8 Such designs have included chemical cross-linking strategies and a variety of recombinant fusions using adhesive protein domains or peptides. Recently, bivalent designs such as , 55 kDa diabodies (scFv dimers) and , 80 kDa minibodies, which retain the avidity of the parent antibody, have been shown to have ideal in vivo pharmacokinetics for tumour imaging.9 – 13 ScFv molecules can be engineered with the V domains in either VH-to-VL or VL-to-VH orientation and the length of the polypeptide linker joining the VH and VL domains varied to ensure the resulting scFv forms stable monomers (, 27 kDa), or associates into dimers (diabodies) or larger multimers (triabodies/tetrabodies).1,2,8 For example, an NC10 (anti-neuraminidase) scFv molecule with a VH-to-VL linker of three to 12 residues cannot fold into a functional monomeric Fv module and instead associates with a second scFv molecule to form a bivalent dimer (diabody, , 55 kDa). Reducing the linker length below three residues forces scFv association into trimers (triabodies, , 80 kDa) or tetramers (, 108 kDa).14 The parent anti-CEA T84.66 antibody used for this structural analysis has been engineered extensively since its discovery.15 Phase I clinical trials of a chimeric (mouse/human) antibody have demonstrated that the whole antibody is effective in tumour imaging and therapeutics.16,17 Engineered scFv monomers and diabodies of the T84.66 antibody have been constructed using a variety of linker lengths18,19 as well as an scFv-CH3 dimer (minibody).20 The T84.66 diabody (T84.66Di) used for this structure analysis comprised two identical scFvs, each with a relatively long eight residue linker (GGGSGGGG).21 The T84.66Di scFvs associate together almost exclusively into diabodies (scFv dimers) during expression in mammalian cell cultures.21 Both the T84.66Di (, 55 kDa) and the (, 80 kDa) T84.66 minibody show more rapid tumour uptake, high tumour activity at earlier time-points and faster blood clearance rates, compared to the whole (, 150 kDa) T84.66 antibody.9,20,22 The crystal structure of one other diabody has been determined, an anti-phospholipase Cd1 (antiPL Cd1) diabody, L5MK1623 (PDB accession no. 1LMK). In the L5MK16 structure, the V domains of the scFv are linked in VH-to-VL orientation and although only one linker polypeptide was resolved in the electron density maps, this confirmed the true diabody status of this molecule. Two other scFv diabody structures have been proposed on the basis of structural data, both with V domains linked in VH-to-VL orientation. The first proposed diabody structure was an NC10 scFv complexed to its target antigen, influenza neuraminidase,24 in which the Fv modules were associated in a back-
T84.66 anti-CEA Diabody Crystal Structure
to-back diabody conformation by crystallographic symmetry (PDB accession no. 1NMC). The second proposed diabody structure was MFE-2325,26 (PDB accession. no. 1QOK), is presented as a back-toback Fv dimer in the crystal structure. The MFE-23 and T84.66Di are both anti-CEA antibody fragments but the binding of each to the CEA antigen has been localized to different sites.25,27 In neither NC10 nor MFE-23 scFv structures were linker residues visible and either a monomeric or dimeric structure was possible, but both were described by the authors as potential diabodies, given the scFv linker lengths could span the required distance between the C terminus of the VH domain to the N terminus of the VL domain to form this dimer. This study presents the solution of the first diabody structure comprised of scFvs linked in VL-toVH orientation and indeed is the first confirmed diabody structure of a cancer-targeting diabody. We compare the T84.66Di structure to the diabody crystal structure of L5MK16 and the two proposed diabody back-to-back structures, all three of which have the V domains of the scFvs linked in the opposite, VH-to-VL, orientation. The implications, for diabody flexibility and presentation of the complementarity-determining regions (CDRs), dependent on Fv module orientations, are discussed.
Structure description The anti-CEA T84.66 diabody comprised two scFv molecules of identical sequence (scFvA and scFvB). Each scFv comprised 240 residues, constructed as VL (L1-L107)—linker polypeptide (eight residues GGGSGGGG)—VH (H1 to H112), there was no C-terminal tag-tail.21 Thus, the antiCEA T84.66 diabody model consisted of 480 residues, resulting from the stable association of two scFv molecules, and was resolved structurally as an asymmetric dimer of two Fv modules. The data and refinement statistics are shown in Table 1 and a Ca trace of the protein chains is shown in Figure 1. All the main-chain and side-chain atoms were included in the model, including 71 water molecules and three sulphate ions (present in the crystallization precipitant solution) and were well defined except for some of the linker residues of the scFvA chain which had disconnected density at 1s above the average density (Figure 1(a)). This was not unexpected, as these residues were fully exposed to solvent in the crystal. The majority of the side-chain atoms lie in good density but some of the larger solventexposed surface residues were not in clear density and had high B-values. The clear electron density for the eight residue (GGGSGGGG) linkers of the T84.66Di, particularly in the scFvB (Figure 1(b)), confirms this molecule as a true diabody. The present model of T84.66Di has an Rfree of 29.7 and
343
T84.66 anti-CEA Diabody Crystal Structure
Table 1. Crystallographic data and refinement statistics A. Data collection and processing ˚) Resolution (A Unique measured reflections (overall/last shell) Completeness (overall/last shell) Redundancy (% $ 2 refl.) (overall/last shell) Rsymd or Rmergee (overall/last shell) Average I/s(I) (overall/last shell) Space group Resolution range of data scaled Unit cell constants for merged data ˚) a (A ˚) b (A ˚) c (A B. Molecular replacement structure solution First translation Second translation After rigid body refinement C. Final model statisticsg Amino acid residues ˚ 2) Average B-value (A Number of protein atoms Number of solvent molecules Rworkh/Rfreei (overall) D. RMS deviations from ideal geometry ˚ 2) B-value main chain (A ˚ 2) B-value side-chain (A ˚) Bonds: target s: main chain ¼ 2.0, side-chain ¼ 2.5 (A Angles: target s: main chain ¼ 2.5, side-chain ¼ 3.0 (deg.) Dihedrals (deg.) ˚) sA coordinate error (A ˚) Luzzati coordinate error (A
bant04a
bant1lb
bant04 þ bant1lc
2.5 15,066/819 86.2/44.2 52.4/18.5 0.075/0.280 24.3/3.5 P41212 20 –2.5
2.4 17,949/1470 90.1/71.2 62.3/49.3 0.052/0.526 17.9/1.5 P41212 100– 2.4
2.5 18,992/14.64 93.2/72.0 62.7/49.2 0.085/0.5431 20.0/1.6 P41212 20–2.5
64.3 64.3 248.0 (S/s)f 6.4 8.8
64.3 64.3 247.9 cc 26.8 31.9 44.4 AVL 111 33.9
64.3 64.3 248.0 R-value 51.8 50.3 46.2 Aln 8 79.5
A chain 2.15 2.90
B chain 3.33 4.10
All 240 40.8 3636 71 0.215/0.298 All 2.74 3.50 0.007 1.8 27.1 0.36 0.35
AVH 121 36.7
BVL 111 37.2
Bln 8 57.2
BVH 121 49.1
A. The T84 66 diabody was expressed and purified as described by Yazaki.21 The protein (15 mg/ml) was crystallized using the hanging-drop, vapor-diffusion method over a well solution of 0.1 M Hepes (pH 7.5), 0.1 M NaCl and 1.6 M NH4(SO4)2 (1 ml in a 2 ml well). The crystallization drop was prepared on a silanised cover-slip by mixing 1 ml of protein solution with 2 ml of well solution and equilibrated at 21 8C. Typical crystals had the dimensions 0.5 mm £ 0.1 mm £ 0.1 mm and were obtained after about ten days. ˚ ) from a rotating anode Siemens AG generator and Data from two crystals were collected using Ni filtered Cu Ka radiation (1.5418 A R-axis IV image plate detector with a microcapillary focussing optic.50 The oscillation images indicated that the crystals had tetragonal symmetry with one molecule in the asymmetric unit and a solvent content of 49% (v/v).51 All data were processed using the HKL program suite52 using DENZO for integration, then scaled and merged with SCALEPACK. B. Phasing was accomplished by molecular replacement using AMoRe53 and the Fv module from 1GGI54 (Fv with 86% identity, and 78% of structure used after excising loops and truncating non-identical residue side-chains to Ala) as a starting model. The space group was confirmed as P41212 by comparing the correlation coefficients and R-values of the translation functions calculated for each tetragonal space group enantiomer. C. Model building was carried out using O55 and refinement with CNS56 using standard protocols.56,57 Non-crystallographic restraints were used in the early stages of model building but not used in refinement nor in the final model adjustments, the free R-value58 (Rfree fraction ¼ 0.1) was monitored throughout refinement. All main-chain and side-chain atoms were included in the model. Solvent-exposed residues on the surface of the structure with side-chains that could not be placed in full density included: Ser A/BVL22, Arg A/BVL68, Lys A/BVL103, Lys A/BVL111, Gln A/BVH64, Glu A/BVH42, Arg AVL24, Lys AVH27c; but were modeled with subsequently high B-factors. The disparity between the Rwork and Rfree values is the result of the inclusion of the second set of data during refinement and the selection of a new set of reflections derived from the merged data for use as the free data set. Attempts were made to reduce model bias by the use of simulated annealing to perturb the structure as described by Hodel.58 This was sufficient to ensure that the Rfree was returned to previous levels (data not shown) but some model bias still appears be present in the ˚ , with 15,856 unique reflections above a 1s amplitude working data set. The data resolution range used for refinement was 20–2.6 A cutoff, these data restrictions were used due to the low redundacncy and average I/s(I) values calculated for the highest-resolution data shells. D. A Ramachandran plot28 analysis was performed using PROCHECK29 and indicates that 97.3% of all residues fall in the allowed region with 80.1% in the most favored region. This PROCHECK analysis also demonstrated that all stereochemical parameters are better than expected for the given resolution, all other model statistics were calculated using CNS. a First data set collected and used for initial molecular replacement solution. b Second data set collected. c Merged data from first and second data sets used for later stages of model refinement. d Rsym. e Rmerge. f Peak height/noise. g ˚ , with 15,856 unique reflections above a 1s amplitude cutoff. Data resolution range used for refinement 20–2.6 A h R-value of data (90%) used in refinement. i Cross-validated R-value calculated using 10% of data chosen randomly and excluded from refinement.
˚ Rwork of 21.5 in the resolution range 20– 2.6 A (Table 1). Only three residues appeared in the disallowed region of a Ramachandran plot28 calculated in
PROCHECK.29 Firstly, Tyr scFvAVH97 (AVH97) in loop H3 that had good density for both the main and side-chain atoms but had phi/psi values (67.0/ 2 106.4). This Tyr AVH97 OH is within
344
T84.66 anti-CEA Diabody Crystal Structure
Figure 1. T84.66Di Ca trace ribbon diagram. The T84.66Di model contained two scFvs, each consisting of a VL domain linked via an eight residue peptide to a VH domain with the two scFvs associating into a dimer, shown here as a Ca trace ribbon diagram rendered using MOLSCRIPT48 and Raster3D.49 The VH domains are shown as aqua with the CDRs highlighted as dark blue, the VL domains are shown as mauve with the CDRs highlighted in red, the linker residues are shown in green. The view is down the pseudo 2-fold axis of the Fv module at the bottom of the Figure, which comprises scFvA VL and scFvB VH. The insets show the available density at 1s above the mean density for: (a) the scFvA VL C-terminal Lys, the linker residues and N-terminal VH Glu; (b) the scFvB VL C-terminal Lys, the linker residues and N-terminal VH Glu; (c) the closest inter scFv interface approach of the VL chains, scFvA with blue density and scFvB with green density. The insets were rendered using DINO (http://www.dino3d.org).
hydrogen bonding distance of Tyr AVH33 OH, which is C-terminal to the CDR-H1 loop of the same domain. This contact may be responsible for the twisting of this residue away from ideality. The Tyr BVH97 Gly BVH96 peptide was flipped compared to the same bond in the scFvA chain and had reasonable phi/psi values (2 87.0/11.6). The two other residues present in the disallowed region of the Ramachandran plot were Ala AVL51 (phi/psi ¼ 69.0/ 2 47.7) and BVL51 (phi/psi ¼ 73.7/ 2 43.2) in the CDR-L2 loop; residues in this loop have been found to show distorted geometry in other antibody structures.30 The second and third light chain CDR loops (L2 and L3) and the first and second heavy chain CDR loops (H1 and H2) in the T84.66Di are characteristic of the previously described canonical CDR
structure repertoire.31 – 33 The L1 loop of T84.66Di contained 11 residues and, although not corresponding in length to any canonical L1 structure, structurally resembled L1 variant 3 (L1-3, with a loop length of 13 residues) in the regions proximal to the supporting b-strands. However, the two residues missing from the loop tip caused a twist in the loop tip and disruption of the hydrogen bonds Ile30 O· · ·N Phe31 and Ile30 O· · ·N Gly32 expected for this L1-3 canonical variant. The third heavy chain CDR is highly variable in the antibody repertoire and canonical classifications are based on the residues proximal to the framework b-strands. These are defined by Morea et al.34 as bulged and unbulged, or kinked and extended, respectively, by Shirai et al.35; here, we use the Morea notation. In T84.66Di, the CDR-H3
345
T84.66 anti-CEA Diabody Crystal Structure
is bulged, as predicted for a CDR with residue H101 being Ala and residue 94 being Pro. Further, the stabilizing H-bond formation between the NE1 of Trp H103 and the O of Met H100d, was present, as expected for a bulged structure. The Pro H94 residue forms part of a hydrophobic cluster containing Phe H27, Val H99, Val H2, Leu H4, and the aromatic ring of Tyr H101. Pro H94, however, did not participate in a main-chain H-bond within the CDR-H3 loop, thus the CDR-H3 torso as described by Morea34 and Shirai35 is deficient in one stabilizing H-bond, but this may be compensated by the hydrophobic clustering around the Pro and the greater rigidity of this side-chain. The T84.66Di is the only known crystal structure of an antibody VH domain in the Kabat36 database with Pro in this position. The tip of the T84.66Di-H3 loop, consisting of residues 98 to H100a was bent over and stabilized by a H-bond made between the Ser H100 OG and Ala H100c O. The loop residues Phe H95, Tyr H100b, Ala H100c, Met H100d and Trp H103 also participated in Fv domain interface ˚ ). This close contacts (contacts distance , 3.5 A may have forced the folded tip conformation of the T84.66Di-H3 but, as the same conformation is present in the different crystallographic environments of each Fv, it is more likely that this is a stable structure in solution for this diabody. Together, the novel L1 loop with its truncation and the folded tip of the H3 loop in T84.66Di provide for a slightly flatter antigen-binding site. It is conceivable that this may represent a conformation present when antigen is bound, enforced by the novel domain arrangements seen in the crystal structure.
Immunoglobulin contact residues Within Fv domains, the residues in T84.66Di ˚ , occur, as that were involved in contacts , 3.5 A expected, mainly at the VLVH interface. These residues made mainly buried hydrophobic contacts stabilizing the interface between the VH and VL and were consistent with those identified by Chothia & Novotny37,38 as commonly occurring domain interface residues. Between Fv domains, the closest residues were between the loops following the C strand in both light chain segments (Figure 1(c)). The Ser AVL56 OG and Gly BVL57 O were separated by less than ˚ , indicating that a steric clash would occur if 4A these loops were to approach any closer. Crystal contacts were made between T84.66Di and five symmetry-related molecules, totaling 85 ˚ ). The H-bonds and other close contacts (, 3.5 A most extensive contacts were made to the molecule related by symmetry operator 2x; 2y; 2z þ 1=2; ˚ , 22 of these totaling 40 close contacts of , 3.5 A involved the CDR loops, some of which are described above. In T84.66Di, the more extended T84.66Di scFvA linker made no crystal contact and was exposed to solvent, accounting for the
weaker electron density for this linker, while the scFvB linker made 19 contacts to a symmetry molecule stabilizing this loop and providing clear electron density to trace the polypeptide chain (Figure 1(a) and (b), respectively). These crystal contacts appear responsible for the orientation of the Fv modules towards each other in the crystal. The Fv module orientations may therefore be at be at a local extreme, constrained by the crystal packing, as any further twisting of the domains toward each other would to lead to steric clashes between the CDRs and other loop residues.
Diabody comparisons and flexibility analysis The Fv orientation in T84.66Di was compared to other diabody structures (Table 2; Figure 2) and the differences in the Fv domain orientations examined. The L5MK16 was the first and only other confirmed diabody crystal structure solved,23 and differs significantly from T84.66Di in that the V domains were linked in the VH-to-VL orientation,23 and that only one of the linkers was visible in electron density. The two other scFv crystal structures used in this comparison, MFE-2325,26 and NC10,24,39 were observed with Fv modules associated in a back-to-back orientation. Both the NC10 and MFE-23 scFv structures could be reconstructed as diabodies with an eight residue linker, using observed crystallographic symmetry and contacts, although no linker residue was visible in either structure.24,40 Indeed, for MFE-23, the authors propose that the back-to-back dimers are representative of the dimers observed in solution studies of this and other scFvs with long (. 12 residue) linkers.26,39,41,42 The four diabodies, two confirmed diabody crystal structures (T84.66Di and L5MK16) and two proposed diabody crystal structures (NC102 and MFE-232) were compared by geometric alignment, as described in the legend to Table 2, of the results of this analysis and illustrated as stereo diagrams of the respective molecules as a Ca traces in Figure 2. Our analysis used a generic core coordinate set constructed by Gelfand et al.43 and referred to here as the Gelfand Fv coordinate set. This Gelfand Fv coordinate set provides a simple and standard method for comparative Fv alignments that we suggest could be adopted for future comparison of other scFv dimers and higher-order multimers. The results of the analysis are presented in Table 2 and Figure 2, and reveal some obvious differences between the VL-to-VH linked T84.66Di and the VH-to-VL linked diabodies. With respect to the angles made between the Y-axis (pseudo 2-fold axes in each Fv module) and the Fv centerto-center axis (Y – O – O2 and Y2 – O2 –O angles) the two proposed VH-to-VL linked diabodies (NC102 and MFE-232) were necessarily identical, as they were constructed using crystallographic symmetry (Table 2). The L5MK16 VH-to-VL linked diabody,
346
T84.66 anti-CEA Diabody Crystal Structure
Table 2. Geometric analysis utilizing the Gelfand core coordinate set Measure ˚ )a Distance O –O2 (A ˚ )b Distance VL36–VH91 (Fv1) (A ˚ )b Distance VL36–VH91 (Fv2) (A ˚ )c Linker spans (scFvA) (A ˚ )c Linker spans (scFvB) (A Linker residues ˚ /residue) Linker scFvA (A ˚ /residue) Linker scFvB (A Angle X –O– O2 (deg.)d Angle X2–O2 –O (deg.)d Inter-axial X angle (deg.)e Angle Y– O–O2 (deg.)f Angle O2 –O–Y2 (deg.)f Inter-axial Y angle (deg.)g Torsion angle (deg.)h
L5MK16
T84.66Di
NC102
MFE-232
36.4 14.0 13.8 15.7 16.6 6 2.6 2.8 70.2 71.7 238.1 115.6 116.5 52.1 2139
36.1 14.2 14.1 20.8 15.1 8 2.6(3.0) 1.9(2.2) 71.5 66.8 241.7 72.5 101.5 26.0 283
29.1 13.7 13.7 24.0 24.1 8 3.0 3.0 53.4 53.4 273.3 135.8 135.8 91.6 2137
26.1 14.2 14.2 21.5 21.5 8 2.7 2.7 87.6 87.6 24.8 177.4 177.4 174.8 2172
The analysis performed here is an extension of that used by Perisic23 to describe the differences in domain arrangement between the VH-to-VL linked L5MK16 diabody crystal structure (PDB accesion no. 1LMK) and a theoretical model of a VH-to-VL linked dimer of the D1.3 antibody Fv.59 This comparison uses vectors in a similar way to analyze the differences between the domain arrangements of the VL-to-VH linked T84.66Di and the L5MK16 diabody crystal structures as well as two other potential diabodies based on the back to back arrangement of Fv domains seen in the crystal structures of NC10 and MFE-23. A geometrically invariant core, or set of framework residues for the Fv module of antibodies constructed by Gelfand;43,60 were used as the reference molecule as an attempt to remove bias from the selection of the Fv for least-squares alignment and the coordinate system constructed by the same authors used as the vector system for geometric analysis. The Y-axis of the Gelfand coordinate system extends from the origin, or centroid of the invariant core, out along the pseudo 2-fold axis between the VL and VH domains. The X-axis then aligns with a vector bridging the VL and VH domains, the Z-axis then bisects the two V domains. The Gelfand coordinate axes are thus aligned to features of the variable domain Fvs such as the pseudo-2-fold and the CDR faces, and are generally applicable to the study of variable region and Fv module orientations. For the analysis presented here, one Fv module (Fv1) of a diabody was aligned to the Gelfand coordinates and the transformation required to align the Fv1 to the unaligned Fv (Fv2) for each diabody was determined. This transformation was then used to rotate and translate a copy of the Gelfand coordinate system axes (described by O, X, Y and Z) to an identical position in the second Fv (Fv2), the new axes described by O2, X2, Y2 and Z2. A further axis between the origins of the Fv1 (O) and Fv2 (O2) coordinate axes was then constructed. The VH-to-VL linked diabodies were aligned to the Gelfand coordinates using VL-toVL and VH-to-VH alignment using a subset of core coordinates common to the VH and VL domains. Due to the reversal of the linker connectivity in the T84.66Di (VL-to-VH linked), this molecule was aligned VL-to-VH and VH-to-VL with the Gelfand coordinates, using the same set of common V domain coordinates. The domain arrangements in each diabody could then be quantified. a Center-to-center distance (O– O2 axis length). b V domain separation using highly conserved framework residues as anchor points within each Fv. c Linker span length using the Ca atom of the last residue in the VH or VL (S H112 or K/R L107) and the atom in the first residue of the VL or VH (D/E L1 or E/Q H1) as anchor points within each chain of the diabody. d Orientation of the CDR faces with respect to the center-to-center axis described by angles X– O–O2 and O –O2 –X2. e Orientation of the CDR faces with respect to one another or inter axial X angle ¼ (angle X–O –O2) þ (angle X2–O2–O) –1808. f Orientation of the psuedo 2-folds of the Fvs with respect to the center-to-center axis described by angles Y –O– O2 and O–O2–Y2. g Orientation of the pseudo 2-folds in each diabody to each other described by an inter-axial Y angle ¼ (angle Y –O– O2) þ (angle Y2 –O2–O) –1808. h Torsion angle made by each Y-axis pair to the O– O2 axis (Y –O– O2–Y2) derived from O and used for comparison to the Y elbow angles described by Lawrence.44
however, also shows near symmetry with less than a degree difference for the two angles made between the Y-axis and the Fv center-to-center axis. In contrast, the T84.66Di VL-to-VH linked diabody was highly asymmetric with a difference of 298 for the same angles. The VH-to-VL linked diabodies all exhibit similar, relatively high torsion angles for the Y-axes and center-to-center axes (Y-axes torsion angles), so that the Y-axes are rotated away from each other by 2 139 to 2 1728, whereas the T84.66Di VL-to-VH linked diabody had a much smaller Y-axis torsion angle of 2 838 and the pseudo 2-folds are almost perpendicular to each other when viewed down the Fv center-to-center (O –O2) axis. Comparison of the distances between the Fv centers (O to O2 axis length) of the four diabodies showed that the L5MK16 diabody had the greatest ˚ ) between the two Fv centers separation (36.4 A
but, conversely, had the shortest linker length. The distance between the Fv centers of the T84.66Di ˚ ) was similar to the L5MK16 diabody. (36.1 A The proposed diabody distances between the Fv ˚ for the ˚ for the NC102 and 26.1 A centers, 29.1 A MFE-232, were much shorter. Perhaps indicating the possibility that these proposed diabodies may not be realistic in vivo. The paths of both scFv linkers of the T84.66Di were visible (cf. only one linker resolved clearly in the L5MK16 diabody structure). The distances spanned by the linkers were calculated between the Ca atoms of the designated C-terminal residues (Ser112 and Lys107) and N-terminal residues in the V domains of each scFv. The longest linker spans were observed for the proposed diabodies while the shortest span was shown in the T84.66Di ˚ ) with connected density for the (scFvB, 15.1 A linker path. The T84.66Di scFvA linker spanned
T84.66 anti-CEA Diabody Crystal Structure
347
Figure 2. Stereo diagrams of Fv dimers or diabodies. Stereo renderings of the four different diabodies; shown as a Ca trace with the light chain CDRs rendered pale blue and the heavy chain CDRs in red, the linkers of the known dia˚ . WebLab bodies are rendered in yellow. The selection of axes length in X, Y and Z were chosen arbitrarily as 40 A Viewer Lite (http://molsim.vei.co.uk/weblab/) was used to render the Figure. For further details, see the legend to Table 2.
348 ˚ with clear but disconnected density for 20.8 A the linker path, and a more disordered (higher B-factors) and extended structure that was exposed to solvent. The linker span length values gave a ˚ , 2.6 A ˚, longest per residue linker length of 2.8 A ˚ and 2.7 A ˚ for the L5MK16, T84.66Di, NC102 3.0 A and MFE-232 diabodies, respectively, indicating that NC102 had the most strained linkage. With a ˚ per residue linker length possible theoretical 3.8 A for a fully extended peptide, however, the NC102 structure was still well within the possibility of diabody connectivity. The flexibility inherent in diabody Fv module orientations has been demonstrated elegantly in electron micrographs of an NC10 diabody, containing a five residue linker, complexed to an antiidiotype 3-2G12 Fab fragment.44 In these electron micrographs, a variable angle between the antiidiotype 3-2G12 Fabs, complexed to each NC10 diabody Fv, was correlated to an elbow angle between the two antigen-binding sites of the diabody. The elbow angle ranged from 608 to 1808 with a normal distribution around a mean of 1228. A modelling analysis of the L5MK16 diabody, also with a five residue linker, gave rise to similar conclusions about the allowed flexibility.45 In the present work an inter-axial Y angle was correlated with the Lawrence elbow angle and the 91.68 inter-axial Y angle calculated for the NC102 proposed diabody is within the expected range.44 The inter-axial Y angles of 528 and 174.88 calculated for the other VH-to-VL linked diabodies, L5MK16, and MFE-232 respectively, also fall within or were close to the angle ranges described by Lawrence44 and by Holliger.45 In contrast, the T84.66Di inter-axial Y angle (2 6.08) is very different from these reported values and shows that the Fv modules are angled towards each other in this alternately linked diabody. One conclusion, on the basis of contact residues and asymmetric Fv alignment, is that the T84.66Di VL-to-VH linked diabody exhibits greater flexibility than is allowed for the VH-to-VL linked diabodies.. Supporting this, all the VH-to-VL linked diabodies showed some symmetry in their pseudo 2-fold arrangements (necessarily for the proposed diabodies) and the angles of the pseudo 2-folds to the center-to-center axis are all obtuse. Thus, asymmetry and increased flexibility might be a characteristic of VL-to-VH linking, although the evidence is limited by the small number of structures studied and crystal packing constraints. Also, the prediction that VL-to-VH linked diabodies with a negative inter-axial Y angle and an apparent occlusion of the antigen-binding sites (and subsequent reduction in affinity) are a preferred orientation is unlikely, since analysis of T84.66Di in complex with antigen demonstrates that this diabody is capable of binding two molecules of CEA (monomeric molecular mass, 180 kDa) simultaneously in solution.18 Further evidence supporting the flexibility of diabodies in solution is provided by characteri-
T84.66 anti-CEA Diabody Crystal Structure
sation of a T84.66 diabody scFv with a short linker peptide and containing a Cys residue at the C terminus.46 This Cys-diabody exists exclusively as a disulfide-bonded dimer, indicating that even though the C termini of the scFv modules are over ˚ apart in the T84.66Di crystal structure (42 A ˚ 60 A in the L5MK16 diabody), in solution the VL-to-VH linked T84.66Di could swivel to allow close association of the C termini. Further analysis of domain contacts in diabody crystal structures and models, with a variety of linker lengths, will be important in designing functional avidity, For example, if VL-to-VH linked diabodies are inherently restricted in the movement of the Fv modules, then the mutation of some inter-domain loop residues may be required to prevent this constraint. Alternatively, loop residues could be designed to stabilize the Fv modules in a particular orientation to minimize entropy loss upon bivalent binding. This could include either disulphide stabilization or reversible cross-linkers that could be designed to occlude one binding site until the targeted site has been reached and then the second binding site could be “unleashed”.
Conclusion The T84.66Di is a potentially important therapeutic molecule for the diagnosis and treatment of some cancers.9,18,46 We anticipate that molecular designs based on the T84.66 diabody structure may enhance flexibility and stability in vivo and thereby improve functional affinity. Indeed, it is possible to now design stabilized T84.66 diabodies that have a preferred binding orientation for binding two adjacent, asymmetric antigens on the cell surface. Perhaps more importantly, the knowledge of antigen-binding orientation can determine preferred fusion sites on the T84.66 diabody, since optimal clinical formulation may require conjugation to a range of effector molecules, such as radioisotopes for imaging, enzymes for prodrug therapy, toxins for targeted cell-killing, viruses for gene therapy or lipids for improved systemic delivery. Indeed, molecular designs will direct many efficient therapeutic uses of diabodies, including T84.66. In one important example, the problems of renal retention in diabody-directed radioimmunotherapy have been reduced using novel metabolizable chelate linkers.47 Finally, our comparative analysis of diabodies utilised the Gelfand Fv coordinate system (Table 2 and Figure 2) and we suggest that this be adopted as a standard method for future comparison of other scFv dimers and higher order multimers.
Acknowledgements This work was supported as part of a PhD scholarship (J.A.C) jointly administered by LaTrobe
349
T84.66 anti-CEA Diabody Crystal Structure
University and The Biomolecular Research Institute. The T84.66Di coordinates have been deposited with the RSCB (PDB, accession no. 1MOE). 15.
References 1. Todorovska, A., Dolezal, O., Kortt, A. A., Hoogenboom, H. R. & Hudson, P. J. (2001). Design and application of diabodies, triabodies and tetrabodies for cancer targeting. J. Immunol. Methods, 248, 47 – 66. 2. Tomlinson, I. & Holliger, P. (2000). Methods for generating multivalent and bispecific antibody fragments. Methods Enzymol. 326, 461– 479. 3. Carter, P. (2001). Improving the efficacy of antibodybased cancer therapies. Nature Rev. Cancer. 1, 118 – 129. 4. Hudson, P. J. & Souriau, C. (2001). Recombinant antibodies for cancer diagnosis and therapy. Expert Opin. Biol. Ther. 1, 845– 855. 5. Casey, J. L., Napier, M. P., King, D. J., Pedley, R. B., Chaplin, L. C., Weir, N. et al. (2002). Tumour targeting of humanised cross-linked divalent-Fab0 antibody fragments: a clinical phase I/II study. Br. J. Cancer, 86, 1401 –1410. 6. Willuda, J., Kubetzko, S., Waibel, R., Schubiger, P. A., Zangemeister-Wittke, U. & Pluckthun, A. (2001). Tumor targeting of mono-, di-, and tetravalent antip185(HER-2) miniantibodies multimerized by selfassociating peptides. J. Biol. Chem. 276, 14385– 14392. 7. Plu¨ckthun, A. & Pack, P. (1997). New protein engineering approaches to multivalent and bispecific antibody fragments. Immunotechnology, 3, 83 – 105. 8. Kortt, A. A., Dolezal, O., Power, B. E. & Hudson, P. J. (2001). Dimeric and trimeric antibodies: high avidity scFvs for cancer targeting. Biomol. Eng. 18, 95 – 108. 9. Yazaki, P. J., Wu, A. M., Tsai, S. W., Williams, L. E., Ikler, D. N., Wong, J. Y. et al. (2001). Tumor targeting of radiometal labeled anti-CEA recombinant T84.66 diabody and T84.66 minibody: comparison to radioiodinated fragments. Bioconjug. Chem. 12, 220– 228. 10. Tahtis, K., Lee, F. T., Smyth, F. E., Power, B. E., Renner, C., Brechbiel, M. W. et al. (2001). Biodistribution properties of (111)indium-labeled C-functionalized trans-cyclohexyl diethylenetriaminepentaacetic acid humanized 3S193 diabody and F(ab0 )(2) constructs in a breast carcinoma xenograft model. Clin. Cancer Res. 7, 1061– 1072. 11. Power, B. E., Caine, J. M., Burns, J. E., Shapira, D. R., Hattarki, M. K., Tahtis, K. et al. (2001). Construction, expression and characterisation of a single-chain diabody derived from a humanised anti-Lewis Y cancer targeting antibody using a heat-inducible bacterial secretion vector. Cancer Immunol. Immunother. 50, 241– 250. 12. Nielsen, U. B., Adams, G. P., Weiner, L. M. & Marks, J. D. (2000). Targeting of bivalent anti-ErbB2 diabody antibody fragments to tumor cells is independent of the intrinsic antibody affinity. Cancer Res. 60, 6434–6440. 13. Adams, G. P., Shaller, C. C., Chappell, L. L., Wu, C., Horak, E. M., Simmons, H. H. et al. (2000). Delivery of the alpha-emitting radioisotope bismuth-213 to solid tumors via single-chain Fv and diabody molecules. Nucl. Med. Biol. 27, 339– 346. 14. Dolezal, O., Pearce, L. A., Lawrence, L. J., McCoy, A. J., Hudson, P. J. & Kortt, A. A. (2000). ScFv multi-
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
mers of the anti-neuraminidase antibody NC10: shortening of the linker in single-chain Fv fragment assembled in V(L) to V(H) orientation drives the formation of dimers, trimers, tetramers and higher molecular mass multimers. Protein Eng. 13, 565–574. Neumaier, M., Shively, L., Chen, F. S., Gaida, F. J., Ilgen, C., Paxton, R. J. et al. (1990). Cloning of the genes for T84.66, an antibody that has a high specificity and affinity for carcinoembryonic antigen, and expression of chimeric human/mouse T84.66 genes in myeloma and chinese hamster ovary cells. Cancer Res. 50, 2128– 2134. Wong, J. Y., Williams, L. E., Yamauchi, D. M., OdomMaryon, T., Esteban, J. M., Neumaier, M. et al. (1995). Initial experience evaluating 90 yttrium-radiolabeled anti-carcinoembryonic antigen chimeric T84.66 in a phase I radioimmunotherapy trial. Cancer Res. 55, 5929s – 5934s. Wong, J. Y., Chu, D. Z., Yamauchi, D., OdomMaryon, T. L., Williams, L. E., Liu, A. et al. (1998). Dose escalation trial of indium-111-labeled anticarcinoembryonic antigen chimeric monoclonal antibody (chimeric T84.66) in presurgical colorectal cancer patients. J. Nucl. Med. 39, 2097– 2104. Wu, A. M., Williams, L. E., Zieran, L., Padma, A., Sherman, M., Bebb, G. G. et al. (1999). Anticarcinoembryonic antigen (CEA) diabody for rapid tumor targeting and imaging. Tumor Targeting, 4, 47 – 58. Wu, A. M., Chen, W., Raubitschek, A., Williams, L. E., Neumaier, M., Fischer, R. et al. (1996). Tumor localization of anti-CEA single-chain Fvs: improved targeting by non-covalent dimers. Immunotechnology, 2, 21 – 36. Hu, S. Z., Shively, L., Raubitschek, A. A., Sherman, M., Williams, L. E., Wong, J. Y. et al. (1996). Minibody: a novel engineered anti-carcinoembryonic antigen antibody fragment (single-chain Fv-CH3) which exhibits rapid, high-level targeting of xenografts. Cancer Res. 56, 3055– 3061. Yazaki, P. J., Shively, L., Clark, C., Cheung, C., Le, W., Szpikowska, B. et al. (2001). Mammalian expression and hollow fiber bioreactor production of recombinant anti-CEA diabody and minibody for clinical applications. J. Immunol. Methods, 253, 195– 208. Wu, A. M., Yazaki, P. J., Tsai, S. w., Nguyen, K., Anderson, A. L., McCarthy, D. W. et al. (2000). Highresolution microPET imaging of carcinoembryonic antigen-positive xenografts by using a copper-64labeled engineered antibody fragment. Proc. Natl Acad. Sci. USA, 97, 8495– 8500. Perisic, O., Webb, P. A., Holliger, P., Winter, G. & Williams, R. L. (1994). Crystal structure of a diabody, a bivalent antibody fragment. Structure, 2, 1217 –1226. Malby, R. L., McCoy, A. J., Kortt, A. A., Hudson, P. J. & Colman, P. M. (1998). Three-dimensional structures of single-chain Fv – neuraminidase complexes. J. Mol. Biol. 279, 901– 910. Boehm, M. K., Corper, A. L., Wan, T., Sohi, M. K., Sutton, B. J., Thornton, J. D. et al. (2000). Crystal structure of the anti-(carcinoembryonic antigen) single-chain Fv antibody MFE-23 and a model for antigen binding based on intermolecular contacts. Biochem. J. 346, 519–528. Lee, Y. C., Boehm, M. K., Chester, K. A., Begent, R. H. & Perkins, S. J. (2002). Reversible dimer formation and stability of the anti-tumour single-chain Fv
350
27.
28.
29.
30.
31. 32.
33. 34.
35. 36.
37.
38.
39.
40.
41.
42.
antibody MFE-23 by neutron scattering. Analytical ultracentrifugation, and NMR and FT-IR spectroscopy. J. Mol. Biol. 320, 107– 127. Murakami, M., Kuroki, M., Arakawa, F., Kuwahara, M., Oikawa, S., Nakazato, H. & Matsuoka, Y. (1995). A reference of the GOLD classification of monoclonal antibodies against carcinoembryonic antigen to the domain structure of the carcinoembryonic antigen molecule. Hybridoma, 14, 19 – 28. Ramakrishnan, C. & Ramachandran, G. N. (1965). Stereochemical criteria for polypeptide and protein chain conformations. II. Allowed conformations for a pair of peptide units. Biophys. J. 5, 909– 933. Laskowski, R. A., MacArthur, M. W., Moss, D. S. & Thornton, J. M. (1993). PROCHECK: a program to check the stereochemicai quality of protein structures. J. Appl. Crystallog. 26, 283– 291. Fan, Z. C., Shan, L., Guddat, L. W., He, X. M., Gray, W. R., Raison, R. L. & Edmundson, A. B. (1992). Three-dimensional structure of an Fv from a human IgM immunoglobulin. J. Mol. Biol. 228, 188– 207. Chothia, C. & Lesk, A. M. (1987). Canonical structures for the hypervariable regions of immunoglobulins. J. Mol. Biol. 196, 901–917. Chothia, C., Lesk, A. M., Gherardi, E., Tomlinson, I. M., Walter, G., Marks, J. D. et al. (1992). Structural repertoire of the human VH segments. J. Mol. Biol. 227, 799– 817. Al-Lazikani, B., Lesk, A. M. & Chothia, C. (1997). Standard conformations for the canonical structures of immunoglobulins. J. Mol. Biol. 273, 927– 948. Morea, V., Tramontano, A., Rustici, M., Chothia, C. & Lesk, A. M. (1998). Conformations of the third hypervariable region in the VH domain of immunoglobulins. J. Mol. Biol. 275, 269–294. Shirai, H., Kidera, A. & Nakamura, H. (1996). Structural classification of CDR-H3 in antibodies. FEBS Letters, 399, 1 – 8. Kabat, E. A., Wu, T. T., Perry, H. M., Gottesman, K. S. & Foeller, C. (1991). Sequences of Proteins of Immunological Interest NIH, Publ. no. 91-3242, 5th edit., DHHS, PHS, National Institutes of Health, Bethseda, MD. Chothia, C., Novotny, J., Bruccoleri, R. & Karplus, M. (1985). Domain association in immunoglobulin molecules. The packing of variable domains. J. Mol. Biol. 186, 651– 663. Novotny, J. & Haber, E. (1985). Structural invariants of antigen binding: comparison of immunoglobulin VL-VH and VL-VL domain dimers. Proc. Natl Acad. Sci. USA, 82, 4592– 4596. Kortt, A. A., Malby, R. L., Caldwell, J. B., Gruen, L. C., Ivancic, N., Lawrence, M. C. et al. (1994). Recombinant anti-sialidase single-chain variable fragment antibody. Characterization, formation of dimer and higher-molecular-mass multimers and the solution of the crystal structure of the single-chain variable fragment/sialidase complex. Eur. J. Biochem. 221, 151– 157. Boehm, M. K. & Perkins, S. J. (2000). Structural models for carcinoembryonic antigen and its complex with the single-chain Fv antibody molecule MFE23. FEBS Letters, 475, 11 – 16. Whitlow, M., Filpula, D., Rollence, M. L., Feng, S. L. & Wood, J. F. (1994). Multivalent Fvs: characterization of single-chain Fv oligomers and preparation of a bispecific Fv. Protein Eng. 7, 1017– 1026. Kuttner, G., Keitel, T., Giessmann, E., Wessner, H., Scholz, C. & Hohne, W. (1998). A phage library-
T84.66 anti-CEA Diabody Crystal Structure
43.
44.
45.
46.
47. 48.
49. 50.
51. 52. 53. 54.
55.
56.
57.
58.
derived single-chain Fv fragment in complex with turkey egg-white lysozyme: characterization, crystallization and preliminary X-ray analysis. Mol. Immunol. 35, 189– 194. Gelfand, I. M., Kister, A. E., Kulikowski, C. & Stoyanov, O. (1998). Geometric invariant core for the V(L) and V(H) domains of immunoglobulin molecules. Protein Eng. 11, 1015– 10125. Lawrence, L. J., Kortt, A. A., Iliades, P., Tulloch, P. A. & Hudson, P. J. (1998). Orientation of antigen binding sites in dimeric and trimeric single chain Fv antibody fragments. FEBS Letters, 425, 479– 484. Holliger, P., Brissinck, J., Williams, R. L., Thielemans, K. & Winter, G. (1996). Specific killing of lymphoma cells by cytotoxic T-cells mediated by a bispecific diabody. Protein Eng. 9, 299– 305. Li, L., Olafsen, T., Anderson, A. L., Wu, A., Raubitschek, A. A. & Shively, J. E. (2002). Reduction of kidney uptake in radiometal labeled peptide linkers conjugated to recombinant antibody fragments. Site-specific conjugation of DOTA-peptides to a Cys-diabody. Bioconjug. Chem. 13, 985– 995. Harris, L. J., Skaletsky, E. & McPherson, A. (1998). Crystallographic structure of an intact IgG1 monoclonal antibody. J. Mol. Biol. 275, 861– 872. Esnouf, R. M. (1997). An extensively modified version of MolScript that includes greatly enhanced coloring capabilities. J. Mol. Graph. Model. 15, 132– 134. Merritt, E. A. & Murphy, M. E. P. (1994). Raster3D version 2.0—a program for photorealistic molecular graphics. Acta Crystallog. sect. D, 50, 869– 873. Varghese, J. N., van Donkelaar, A., Hrmova, M., Fincher, G. A., Baliac, D. X. & Barnia, Z. (1999). Structure of an exoglucanase complexed with conduritol B epoxide from a 50m crystal using monocapillary optics. Acta Crystallog. sect. A, 55, 329– 330. Matthews, B. W. (1985). Determination of protein molecular weight, hydration, and packing from crystal density. Methods Enzymol. 114, 176–187. Otwinowski, Z. & Minor, W. (1996). Processing of X-ray diffraction data collected in oscillation mode. Methods Enzymol. 276, 307– 326. Navaza, J. (1994). AMoRe: an automated package for molecular replacement. Acta Crystallog. sect. A, 50, 157– 163. Rini, J. M., Stanfield, R. L., Stura, E. A., Salinas, P. A., Profy, A. T. & Wilson, I. A. (1993). Crystal structure of an HIV-1 neutralizing antibody 50.1 in complex with its V3 loop peptide antigen. Proc. Natl Acad. Sci. USA, 90, 6325– 6329. Jones, T. A., Zou, J. Y., Cowan, S. W. & Kjeldgaard, M. (1991). Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Crystallog. sect. A, 47, 110 – 119. Bru¨nger, A. T., Adams, P. D., Clore, G. M., DeLano, W. L., Gros, P., Grosse-Kunstleve, R. W. et al. (1998). Crystallography and NMR system: a new software suite for macromolecular structure determination. Acta Crystallog. sect. D, 54, 905– 921. Adams, P. D., Pannu, N. S., Read, R. J. & Bru¨nger, A. T. (1997). Cross-validated maximum likelihood enhances crystallographic simulated annealing refinement. Proc. Natl Acad. Sci. USA, 94, 5018– 5023. Hodel, A., Kim, S.-H. & Bru¨nger, A. T. (1992). Model bias in macromolecular crystal structures. Acta Crystallog. sect. A, 48, 851– 858.
T84.66 anti-CEA Diabody Crystal Structure
59. Holliger, P., Prospero, T. & Winter, G. (1993). “Diabodies”: small bivalent and bispecific antibody fragments. Proc. Natl Acad. Sci. USA, 90, 6444– 6448. 60. Gelfand, I. M. & Kister, A. E. (1997). A very limited number of keywords (main patterns) describes all
351
sequences of the human variable heavy (VH) and kappa (Vkappa) domains. Proc. Natl Acad. Sci. USA, 94, 12562– 12567.
Edited by I. Wilson (Received 3 September 2002; received in revised form 3 December 2002; accepted 3 December 2002)
J. Mol. Biol. (2003) 326, 341–351
COMMUNICATION
The Crystal Structure of an Anti-CEA scFv Diabody Assembled from T84.66 scFvs in VL-to-VH Orientation: Implications for Diabody Flexibility Jennifer A. Carmichael1,2, Barbara E. Power1, Thomas P. J. Garrett1 Paul J. Yazaki3, John E. Shively3, Andrew A. Raubischek3 Anna M. Wu3 and Peter J. Hudson1* 1 CSIRO Health Sciences and Nutrition, 343 Royal Parade Parkville, Vic. 3052, Australia 2
Department of Biochemistry La Trobe University, Bundoora Vic. 3083, Australia 3
Department of Molecular Biology, Beckman Research Institute of the City of Hope Duarte, CA 91010, USA
Diabodies (scFv dimers) are small, bivalent antibody mimetics of approximately 55 kDa in size that possess rapid in vivo targeting pharmacokinetics compared to the intact parent antibody, and may prove highly suitable for imaging and therapeutic applications. Here, we describe T84.66Di, the first diabody crystal structure in which the scFvs comprise V domains linked in the VL-to-VH orientation. The structure was deter˚ resolution. The T84.66Di mined by X-ray diffraction analysis to 2.6 A scFv was constructed from the anti-carcinoembryonic antigen (anti-CEA) antibody T84.66 variable domains connected by an eight residue peptide linker to provide flexibility between Fv modules and promote dimer formation with bivalent affinity to the cell-surface target, CEA. Therefore, it was surprising to observe a close association of some Fv module complementarity-determining regions in the T84.66 diabody crystal, especially compared to other diabody structures all of which are linked in the opposite VH-to-VL orientation. The differences between the arrangement of Fv modules in the T84.66Di VL-to-VH linked diabody structure compared to the crystal structure of L5MK16 and other proposed VH-to-VL linked diabodies has been investigated and their potential for flexibility discussed. The comparison between VH-to-VL and VL-to-VH linked diabodies revealed in this study represents a limited repertoire of possible diabody Fv orientations, but one that reveals the potential flexibility of these molecules. This analysis therefore provides some signposts that may impact on future molecular designs for these therapeutic molecules with respect to diabody flexibility and avidity. q 2003 Elsevier Science Ltd. All rights reserved
*Corresponding author
Keywords: CEA; structure; dimers; diabody; flexibility
Present address: Dr T.P.J. Garrett, The Walter and Eliza Hall Institute of Medical Research, Royal Parade, Parkville, Vic. 3050, Australia. Abbreviations used: CEA, carcinoembryonic antigen; anti-CEA, antibody to carcinoembryonic antigen; Fab, antibody fragment produced by proteolysis; Ig, immunoglobulin; NA, influenza virus neuraminidase; scFv, single chain Fv molecule; T84.66Di, anti-CEA T84.66 diabody; VH, variable region from antibody heavy chain; VL, variable region from antibody light chain; CDR, complementarity-determining region. E-mail address of the corresponding author: [email protected]
Recombinant antibodies and their fragments are becoming the modern paradigm for the design of high-affinity, protein-based targeting reagents1,2 and now represent over 25% of all proteins undergoing clinical trials.3,4 Recombinant antibodies also capture a significant share of the US$6 billion £ 109 per annum immunodiagnostic market, from in vitro immunoassays to in vivo imaging reagents.4 Intact antibodies provide high target binding specificity but their use in rapid tumour targeting and in vivo imaging is limited by slow tissue penetration, long circulating half-lives and often undesirable effector functions.3 Small monovalent antibody fragments (scFv and Fab) exhibit good tissue penetration but their lack of avidity results in faster off rates and
0022-2836/03/$ - see front matter q 2003 Elsevier Science Ltd. All rights reserved
342
rapid clearance. Several laboratories have therefore engineered Fab or scFv molecules into dimers or larger multimers to produce high-avidity reagents of optimal size (50 –110 kDa) for in vivo imaging and therapeutics.5 – 8 Such designs have included chemical cross-linking strategies and a variety of recombinant fusions using adhesive protein domains or peptides. Recently, bivalent designs such as , 55 kDa diabodies (scFv dimers) and , 80 kDa minibodies, which retain the avidity of the parent antibody, have been shown to have ideal in vivo pharmacokinetics for tumour imaging.9 – 13 ScFv molecules can be engineered with the V domains in either VH-to-VL or VL-to-VH orientation and the length of the polypeptide linker joining the VH and VL domains varied to ensure the resulting scFv forms stable monomers (, 27 kDa), or associates into dimers (diabodies) or larger multimers (triabodies/tetrabodies).1,2,8 For example, an NC10 (anti-neuraminidase) scFv molecule with a VH-to-VL linker of three to 12 residues cannot fold into a functional monomeric Fv module and instead associates with a second scFv molecule to form a bivalent dimer (diabody, , 55 kDa). Reducing the linker length below three residues forces scFv association into trimers (triabodies, , 80 kDa) or tetramers (, 108 kDa).14 The parent anti-CEA T84.66 antibody used for this structural analysis has been engineered extensively since its discovery.15 Phase I clinical trials of a chimeric (mouse/human) antibody have demonstrated that the whole antibody is effective in tumour imaging and therapeutics.16,17 Engineered scFv monomers and diabodies of the T84.66 antibody have been constructed using a variety of linker lengths18,19 as well as an scFv-CH3 dimer (minibody).20 The T84.66 diabody (T84.66Di) used for this structure analysis comprised two identical scFvs, each with a relatively long eight residue linker (GGGSGGGG).21 The T84.66Di scFvs associate together almost exclusively into diabodies (scFv dimers) during expression in mammalian cell cultures.21 Both the T84.66Di (, 55 kDa) and the (, 80 kDa) T84.66 minibody show more rapid tumour uptake, high tumour activity at earlier time-points and faster blood clearance rates, compared to the whole (, 150 kDa) T84.66 antibody.9,20,22 The crystal structure of one other diabody has been determined, an anti-phospholipase Cd1 (antiPL Cd1) diabody, L5MK1623 (PDB accession no. 1LMK). In the L5MK16 structure, the V domains of the scFv are linked in VH-to-VL orientation and although only one linker polypeptide was resolved in the electron density maps, this confirmed the true diabody status of this molecule. Two other scFv diabody structures have been proposed on the basis of structural data, both with V domains linked in VH-to-VL orientation. The first proposed diabody structure was an NC10 scFv complexed to its target antigen, influenza neuraminidase,24 in which the Fv modules were associated in a back-
T84.66 anti-CEA Diabody Crystal Structure
to-back diabody conformation by crystallographic symmetry (PDB accession no. 1NMC). The second proposed diabody structure was MFE-2325,26 (PDB accession. no. 1QOK), is presented as a back-toback Fv dimer in the crystal structure. The MFE-23 and T84.66Di are both anti-CEA antibody fragments but the binding of each to the CEA antigen has been localized to different sites.25,27 In neither NC10 nor MFE-23 scFv structures were linker residues visible and either a monomeric or dimeric structure was possible, but both were described by the authors as potential diabodies, given the scFv linker lengths could span the required distance between the C terminus of the VH domain to the N terminus of the VL domain to form this dimer. This study presents the solution of the first diabody structure comprised of scFvs linked in VL-toVH orientation and indeed is the first confirmed diabody structure of a cancer-targeting diabody. We compare the T84.66Di structure to the diabody crystal structure of L5MK16 and the two proposed diabody back-to-back structures, all three of which have the V domains of the scFvs linked in the opposite, VH-to-VL, orientation. The implications, for diabody flexibility and presentation of the complementarity-determining regions (CDRs), dependent on Fv module orientations, are discussed.
Structure description The anti-CEA T84.66 diabody comprised two scFv molecules of identical sequence (scFvA and scFvB). Each scFv comprised 240 residues, constructed as VL (L1-L107)—linker polypeptide (eight residues GGGSGGGG)—VH (H1 to H112), there was no C-terminal tag-tail.21 Thus, the antiCEA T84.66 diabody model consisted of 480 residues, resulting from the stable association of two scFv molecules, and was resolved structurally as an asymmetric dimer of two Fv modules. The data and refinement statistics are shown in Table 1 and a Ca trace of the protein chains is shown in Figure 1. All the main-chain and side-chain atoms were included in the model, including 71 water molecules and three sulphate ions (present in the crystallization precipitant solution) and were well defined except for some of the linker residues of the scFvA chain which had disconnected density at 1s above the average density (Figure 1(a)). This was not unexpected, as these residues were fully exposed to solvent in the crystal. The majority of the side-chain atoms lie in good density but some of the larger solventexposed surface residues were not in clear density and had high B-values. The clear electron density for the eight residue (GGGSGGGG) linkers of the T84.66Di, particularly in the scFvB (Figure 1(b)), confirms this molecule as a true diabody. The present model of T84.66Di has an Rfree of 29.7 and
343
T84.66 anti-CEA Diabody Crystal Structure
Table 1. Crystallographic data and refinement statistics A. Data collection and processing ˚) Resolution (A Unique measured reflections (overall/last shell) Completeness (overall/last shell) Redundancy (% $ 2 refl.) (overall/last shell) Rsymd or Rmergee (overall/last shell) Average I/s(I) (overall/last shell) Space group Resolution range of data scaled Unit cell constants for merged data ˚) a (A ˚) b (A ˚) c (A B. Molecular replacement structure solution First translation Second translation After rigid body refinement C. Final model statisticsg Amino acid residues ˚ 2) Average B-value (A Number of protein atoms Number of solvent molecules Rworkh/Rfreei (overall) D. RMS deviations from ideal geometry ˚ 2) B-value main chain (A ˚ 2) B-value side-chain (A ˚) Bonds: target s: main chain ¼ 2.0, side-chain ¼ 2.5 (A Angles: target s: main chain ¼ 2.5, side-chain ¼ 3.0 (deg.) Dihedrals (deg.) ˚) sA coordinate error (A ˚) Luzzati coordinate error (A
bant04a
bant1lb
bant04 þ bant1lc
2.5 15,066/819 86.2/44.2 52.4/18.5 0.075/0.280 24.3/3.5 P41212 20 –2.5
2.4 17,949/1470 90.1/71.2 62.3/49.3 0.052/0.526 17.9/1.5 P41212 100– 2.4
2.5 18,992/14.64 93.2/72.0 62.7/49.2 0.085/0.5431 20.0/1.6 P41212 20–2.5
64.3 64.3 248.0 (S/s)f 6.4 8.8
64.3 64.3 247.9 cc 26.8 31.9 44.4 AVL 111 33.9
64.3 64.3 248.0 R-value 51.8 50.3 46.2 Aln 8 79.5
A chain 2.15 2.90
B chain 3.33 4.10
All 240 40.8 3636 71 0.215/0.298 All 2.74 3.50 0.007 1.8 27.1 0.36 0.35
AVH 121 36.7
BVL 111 37.2
Bln 8 57.2
BVH 121 49.1
A. The T84 66 diabody was expressed and purified as described by Yazaki.21 The protein (15 mg/ml) was crystallized using the hanging-drop, vapor-diffusion method over a well solution of 0.1 M Hepes (pH 7.5), 0.1 M NaCl and 1.6 M NH4(SO4)2 (1 ml in a 2 ml well). The crystallization drop was prepared on a silanised cover-slip by mixing 1 ml of protein solution with 2 ml of well solution and equilibrated at 21 8C. Typical crystals had the dimensions 0.5 mm £ 0.1 mm £ 0.1 mm and were obtained after about ten days. ˚ ) from a rotating anode Siemens AG generator and Data from two crystals were collected using Ni filtered Cu Ka radiation (1.5418 A R-axis IV image plate detector with a microcapillary focussing optic.50 The oscillation images indicated that the crystals had tetragonal symmetry with one molecule in the asymmetric unit and a solvent content of 49% (v/v).51 All data were processed using the HKL program suite52 using DENZO for integration, then scaled and merged with SCALEPACK. B. Phasing was accomplished by molecular replacement using AMoRe53 and the Fv module from 1GGI54 (Fv with 86% identity, and 78% of structure used after excising loops and truncating non-identical residue side-chains to Ala) as a starting model. The space group was confirmed as P41212 by comparing the correlation coefficients and R-values of the translation functions calculated for each tetragonal space group enantiomer. C. Model building was carried out using O55 and refinement with CNS56 using standard protocols.56,57 Non-crystallographic restraints were used in the early stages of model building but not used in refinement nor in the final model adjustments, the free R-value58 (Rfree fraction ¼ 0.1) was monitored throughout refinement. All main-chain and side-chain atoms were included in the model. Solvent-exposed residues on the surface of the structure with side-chains that could not be placed in full density included: Ser A/BVL22, Arg A/BVL68, Lys A/BVL103, Lys A/BVL111, Gln A/BVH64, Glu A/BVH42, Arg AVL24, Lys AVH27c; but were modeled with subsequently high B-factors. The disparity between the Rwork and Rfree values is the result of the inclusion of the second set of data during refinement and the selection of a new set of reflections derived from the merged data for use as the free data set. Attempts were made to reduce model bias by the use of simulated annealing to perturb the structure as described by Hodel.58 This was sufficient to ensure that the Rfree was returned to previous levels (data not shown) but some model bias still appears be present in the ˚ , with 15,856 unique reflections above a 1s amplitude working data set. The data resolution range used for refinement was 20–2.6 A cutoff, these data restrictions were used due to the low redundacncy and average I/s(I) values calculated for the highest-resolution data shells. D. A Ramachandran plot28 analysis was performed using PROCHECK29 and indicates that 97.3% of all residues fall in the allowed region with 80.1% in the most favored region. This PROCHECK analysis also demonstrated that all stereochemical parameters are better than expected for the given resolution, all other model statistics were calculated using CNS. a First data set collected and used for initial molecular replacement solution. b Second data set collected. c Merged data from first and second data sets used for later stages of model refinement. d Rsym. e Rmerge. f Peak height/noise. g ˚ , with 15,856 unique reflections above a 1s amplitude cutoff. Data resolution range used for refinement 20–2.6 A h R-value of data (90%) used in refinement. i Cross-validated R-value calculated using 10% of data chosen randomly and excluded from refinement.
˚ Rwork of 21.5 in the resolution range 20– 2.6 A (Table 1). Only three residues appeared in the disallowed region of a Ramachandran plot28 calculated in
PROCHECK.29 Firstly, Tyr scFvAVH97 (AVH97) in loop H3 that had good density for both the main and side-chain atoms but had phi/psi values (67.0/ 2 106.4). This Tyr AVH97 OH is within
344
T84.66 anti-CEA Diabody Crystal Structure
Figure 1. T84.66Di Ca trace ribbon diagram. The T84.66Di model contained two scFvs, each consisting of a VL domain linked via an eight residue peptide to a VH domain with the two scFvs associating into a dimer, shown here as a Ca trace ribbon diagram rendered using MOLSCRIPT48 and Raster3D.49 The VH domains are shown as aqua with the CDRs highlighted as dark blue, the VL domains are shown as mauve with the CDRs highlighted in red, the linker residues are shown in green. The view is down the pseudo 2-fold axis of the Fv module at the bottom of the Figure, which comprises scFvA VL and scFvB VH. The insets show the available density at 1s above the mean density for: (a) the scFvA VL C-terminal Lys, the linker residues and N-terminal VH Glu; (b) the scFvB VL C-terminal Lys, the linker residues and N-terminal VH Glu; (c) the closest inter scFv interface approach of the VL chains, scFvA with blue density and scFvB with green density. The insets were rendered using DINO (http://www.dino3d.org).
hydrogen bonding distance of Tyr AVH33 OH, which is C-terminal to the CDR-H1 loop of the same domain. This contact may be responsible for the twisting of this residue away from ideality. The Tyr BVH97 Gly BVH96 peptide was flipped compared to the same bond in the scFvA chain and had reasonable phi/psi values (2 87.0/11.6). The two other residues present in the disallowed region of the Ramachandran plot were Ala AVL51 (phi/psi ¼ 69.0/ 2 47.7) and BVL51 (phi/psi ¼ 73.7/ 2 43.2) in the CDR-L2 loop; residues in this loop have been found to show distorted geometry in other antibody structures.30 The second and third light chain CDR loops (L2 and L3) and the first and second heavy chain CDR loops (H1 and H2) in the T84.66Di are characteristic of the previously described canonical CDR
structure repertoire.31 – 33 The L1 loop of T84.66Di contained 11 residues and, although not corresponding in length to any canonical L1 structure, structurally resembled L1 variant 3 (L1-3, with a loop length of 13 residues) in the regions proximal to the supporting b-strands. However, the two residues missing from the loop tip caused a twist in the loop tip and disruption of the hydrogen bonds Ile30 O· · ·N Phe31 and Ile30 O· · ·N Gly32 expected for this L1-3 canonical variant. The third heavy chain CDR is highly variable in the antibody repertoire and canonical classifications are based on the residues proximal to the framework b-strands. These are defined by Morea et al.34 as bulged and unbulged, or kinked and extended, respectively, by Shirai et al.35; here, we use the Morea notation. In T84.66Di, the CDR-H3
345
T84.66 anti-CEA Diabody Crystal Structure
is bulged, as predicted for a CDR with residue H101 being Ala and residue 94 being Pro. Further, the stabilizing H-bond formation between the NE1 of Trp H103 and the O of Met H100d, was present, as expected for a bulged structure. The Pro H94 residue forms part of a hydrophobic cluster containing Phe H27, Val H99, Val H2, Leu H4, and the aromatic ring of Tyr H101. Pro H94, however, did not participate in a main-chain H-bond within the CDR-H3 loop, thus the CDR-H3 torso as described by Morea34 and Shirai35 is deficient in one stabilizing H-bond, but this may be compensated by the hydrophobic clustering around the Pro and the greater rigidity of this side-chain. The T84.66Di is the only known crystal structure of an antibody VH domain in the Kabat36 database with Pro in this position. The tip of the T84.66Di-H3 loop, consisting of residues 98 to H100a was bent over and stabilized by a H-bond made between the Ser H100 OG and Ala H100c O. The loop residues Phe H95, Tyr H100b, Ala H100c, Met H100d and Trp H103 also participated in Fv domain interface ˚ ). This close contacts (contacts distance , 3.5 A may have forced the folded tip conformation of the T84.66Di-H3 but, as the same conformation is present in the different crystallographic environments of each Fv, it is more likely that this is a stable structure in solution for this diabody. Together, the novel L1 loop with its truncation and the folded tip of the H3 loop in T84.66Di provide for a slightly flatter antigen-binding site. It is conceivable that this may represent a conformation present when antigen is bound, enforced by the novel domain arrangements seen in the crystal structure.
Immunoglobulin contact residues Within Fv domains, the residues in T84.66Di ˚ , occur, as that were involved in contacts , 3.5 A expected, mainly at the VLVH interface. These residues made mainly buried hydrophobic contacts stabilizing the interface between the VH and VL and were consistent with those identified by Chothia & Novotny37,38 as commonly occurring domain interface residues. Between Fv domains, the closest residues were between the loops following the C strand in both light chain segments (Figure 1(c)). The Ser AVL56 OG and Gly BVL57 O were separated by less than ˚ , indicating that a steric clash would occur if 4A these loops were to approach any closer. Crystal contacts were made between T84.66Di and five symmetry-related molecules, totaling 85 ˚ ). The H-bonds and other close contacts (, 3.5 A most extensive contacts were made to the molecule related by symmetry operator 2x; 2y; 2z þ 1=2; ˚ , 22 of these totaling 40 close contacts of , 3.5 A involved the CDR loops, some of which are described above. In T84.66Di, the more extended T84.66Di scFvA linker made no crystal contact and was exposed to solvent, accounting for the
weaker electron density for this linker, while the scFvB linker made 19 contacts to a symmetry molecule stabilizing this loop and providing clear electron density to trace the polypeptide chain (Figure 1(a) and (b), respectively). These crystal contacts appear responsible for the orientation of the Fv modules towards each other in the crystal. The Fv module orientations may therefore be at be at a local extreme, constrained by the crystal packing, as any further twisting of the domains toward each other would to lead to steric clashes between the CDRs and other loop residues.
Diabody comparisons and flexibility analysis The Fv orientation in T84.66Di was compared to other diabody structures (Table 2; Figure 2) and the differences in the Fv domain orientations examined. The L5MK16 was the first and only other confirmed diabody crystal structure solved,23 and differs significantly from T84.66Di in that the V domains were linked in the VH-to-VL orientation,23 and that only one of the linkers was visible in electron density. The two other scFv crystal structures used in this comparison, MFE-2325,26 and NC10,24,39 were observed with Fv modules associated in a back-to-back orientation. Both the NC10 and MFE-23 scFv structures could be reconstructed as diabodies with an eight residue linker, using observed crystallographic symmetry and contacts, although no linker residue was visible in either structure.24,40 Indeed, for MFE-23, the authors propose that the back-to-back dimers are representative of the dimers observed in solution studies of this and other scFvs with long (. 12 residue) linkers.26,39,41,42 The four diabodies, two confirmed diabody crystal structures (T84.66Di and L5MK16) and two proposed diabody crystal structures (NC102 and MFE-232) were compared by geometric alignment, as described in the legend to Table 2, of the results of this analysis and illustrated as stereo diagrams of the respective molecules as a Ca traces in Figure 2. Our analysis used a generic core coordinate set constructed by Gelfand et al.43 and referred to here as the Gelfand Fv coordinate set. This Gelfand Fv coordinate set provides a simple and standard method for comparative Fv alignments that we suggest could be adopted for future comparison of other scFv dimers and higher-order multimers. The results of the analysis are presented in Table 2 and Figure 2, and reveal some obvious differences between the VL-to-VH linked T84.66Di and the VH-to-VL linked diabodies. With respect to the angles made between the Y-axis (pseudo 2-fold axes in each Fv module) and the Fv centerto-center axis (Y – O – O2 and Y2 – O2 –O angles) the two proposed VH-to-VL linked diabodies (NC102 and MFE-232) were necessarily identical, as they were constructed using crystallographic symmetry (Table 2). The L5MK16 VH-to-VL linked diabody,
346
T84.66 anti-CEA Diabody Crystal Structure
Table 2. Geometric analysis utilizing the Gelfand core coordinate set Measure ˚ )a Distance O –O2 (A ˚ )b Distance VL36–VH91 (Fv1) (A ˚ )b Distance VL36–VH91 (Fv2) (A ˚ )c Linker spans (scFvA) (A ˚ )c Linker spans (scFvB) (A Linker residues ˚ /residue) Linker scFvA (A ˚ /residue) Linker scFvB (A Angle X –O– O2 (deg.)d Angle X2–O2 –O (deg.)d Inter-axial X angle (deg.)e Angle Y– O–O2 (deg.)f Angle O2 –O–Y2 (deg.)f Inter-axial Y angle (deg.)g Torsion angle (deg.)h
L5MK16
T84.66Di
NC102
MFE-232
36.4 14.0 13.8 15.7 16.6 6 2.6 2.8 70.2 71.7 238.1 115.6 116.5 52.1 2139
36.1 14.2 14.1 20.8 15.1 8 2.6(3.0) 1.9(2.2) 71.5 66.8 241.7 72.5 101.5 26.0 283
29.1 13.7 13.7 24.0 24.1 8 3.0 3.0 53.4 53.4 273.3 135.8 135.8 91.6 2137
26.1 14.2 14.2 21.5 21.5 8 2.7 2.7 87.6 87.6 24.8 177.4 177.4 174.8 2172
The analysis performed here is an extension of that used by Perisic23 to describe the differences in domain arrangement between the VH-to-VL linked L5MK16 diabody crystal structure (PDB accesion no. 1LMK) and a theoretical model of a VH-to-VL linked dimer of the D1.3 antibody Fv.59 This comparison uses vectors in a similar way to analyze the differences between the domain arrangements of the VL-to-VH linked T84.66Di and the L5MK16 diabody crystal structures as well as two other potential diabodies based on the back to back arrangement of Fv domains seen in the crystal structures of NC10 and MFE-23. A geometrically invariant core, or set of framework residues for the Fv module of antibodies constructed by Gelfand;43,60 were used as the reference molecule as an attempt to remove bias from the selection of the Fv for least-squares alignment and the coordinate system constructed by the same authors used as the vector system for geometric analysis. The Y-axis of the Gelfand coordinate system extends from the origin, or centroid of the invariant core, out along the pseudo 2-fold axis between the VL and VH domains. The X-axis then aligns with a vector bridging the VL and VH domains, the Z-axis then bisects the two V domains. The Gelfand coordinate axes are thus aligned to features of the variable domain Fvs such as the pseudo-2-fold and the CDR faces, and are generally applicable to the study of variable region and Fv module orientations. For the analysis presented here, one Fv module (Fv1) of a diabody was aligned to the Gelfand coordinates and the transformation required to align the Fv1 to the unaligned Fv (Fv2) for each diabody was determined. This transformation was then used to rotate and translate a copy of the Gelfand coordinate system axes (described by O, X, Y and Z) to an identical position in the second Fv (Fv2), the new axes described by O2, X2, Y2 and Z2. A further axis between the origins of the Fv1 (O) and Fv2 (O2) coordinate axes was then constructed. The VH-to-VL linked diabodies were aligned to the Gelfand coordinates using VL-toVL and VH-to-VH alignment using a subset of core coordinates common to the VH and VL domains. Due to the reversal of the linker connectivity in the T84.66Di (VL-to-VH linked), this molecule was aligned VL-to-VH and VH-to-VL with the Gelfand coordinates, using the same set of common V domain coordinates. The domain arrangements in each diabody could then be quantified. a Center-to-center distance (O– O2 axis length). b V domain separation using highly conserved framework residues as anchor points within each Fv. c Linker span length using the Ca atom of the last residue in the VH or VL (S H112 or K/R L107) and the atom in the first residue of the VL or VH (D/E L1 or E/Q H1) as anchor points within each chain of the diabody. d Orientation of the CDR faces with respect to the center-to-center axis described by angles X– O–O2 and O –O2 –X2. e Orientation of the CDR faces with respect to one another or inter axial X angle ¼ (angle X–O –O2) þ (angle X2–O2–O) –1808. f Orientation of the psuedo 2-folds of the Fvs with respect to the center-to-center axis described by angles Y –O– O2 and O–O2–Y2. g Orientation of the pseudo 2-folds in each diabody to each other described by an inter-axial Y angle ¼ (angle Y –O– O2) þ (angle Y2 –O2–O) –1808. h Torsion angle made by each Y-axis pair to the O– O2 axis (Y –O– O2–Y2) derived from O and used for comparison to the Y elbow angles described by Lawrence.44
however, also shows near symmetry with less than a degree difference for the two angles made between the Y-axis and the Fv center-to-center axis. In contrast, the T84.66Di VL-to-VH linked diabody was highly asymmetric with a difference of 298 for the same angles. The VH-to-VL linked diabodies all exhibit similar, relatively high torsion angles for the Y-axes and center-to-center axes (Y-axes torsion angles), so that the Y-axes are rotated away from each other by 2 139 to 2 1728, whereas the T84.66Di VL-to-VH linked diabody had a much smaller Y-axis torsion angle of 2 838 and the pseudo 2-folds are almost perpendicular to each other when viewed down the Fv center-to-center (O –O2) axis. Comparison of the distances between the Fv centers (O to O2 axis length) of the four diabodies showed that the L5MK16 diabody had the greatest ˚ ) between the two Fv centers separation (36.4 A
but, conversely, had the shortest linker length. The distance between the Fv centers of the T84.66Di ˚ ) was similar to the L5MK16 diabody. (36.1 A The proposed diabody distances between the Fv ˚ for the ˚ for the NC102 and 26.1 A centers, 29.1 A MFE-232, were much shorter. Perhaps indicating the possibility that these proposed diabodies may not be realistic in vivo. The paths of both scFv linkers of the T84.66Di were visible (cf. only one linker resolved clearly in the L5MK16 diabody structure). The distances spanned by the linkers were calculated between the Ca atoms of the designated C-terminal residues (Ser112 and Lys107) and N-terminal residues in the V domains of each scFv. The longest linker spans were observed for the proposed diabodies while the shortest span was shown in the T84.66Di ˚ ) with connected density for the (scFvB, 15.1 A linker path. The T84.66Di scFvA linker spanned
T84.66 anti-CEA Diabody Crystal Structure
347
Figure 2. Stereo diagrams of Fv dimers or diabodies. Stereo renderings of the four different diabodies; shown as a Ca trace with the light chain CDRs rendered pale blue and the heavy chain CDRs in red, the linkers of the known dia˚ . WebLab bodies are rendered in yellow. The selection of axes length in X, Y and Z were chosen arbitrarily as 40 A Viewer Lite (http://molsim.vei.co.uk/weblab/) was used to render the Figure. For further details, see the legend to Table 2.
348 ˚ with clear but disconnected density for 20.8 A the linker path, and a more disordered (higher B-factors) and extended structure that was exposed to solvent. The linker span length values gave a ˚ , 2.6 A ˚, longest per residue linker length of 2.8 A ˚ and 2.7 A ˚ for the L5MK16, T84.66Di, NC102 3.0 A and MFE-232 diabodies, respectively, indicating that NC102 had the most strained linkage. With a ˚ per residue linker length possible theoretical 3.8 A for a fully extended peptide, however, the NC102 structure was still well within the possibility of diabody connectivity. The flexibility inherent in diabody Fv module orientations has been demonstrated elegantly in electron micrographs of an NC10 diabody, containing a five residue linker, complexed to an antiidiotype 3-2G12 Fab fragment.44 In these electron micrographs, a variable angle between the antiidiotype 3-2G12 Fabs, complexed to each NC10 diabody Fv, was correlated to an elbow angle between the two antigen-binding sites of the diabody. The elbow angle ranged from 608 to 1808 with a normal distribution around a mean of 1228. A modelling analysis of the L5MK16 diabody, also with a five residue linker, gave rise to similar conclusions about the allowed flexibility.45 In the present work an inter-axial Y angle was correlated with the Lawrence elbow angle and the 91.68 inter-axial Y angle calculated for the NC102 proposed diabody is within the expected range.44 The inter-axial Y angles of 528 and 174.88 calculated for the other VH-to-VL linked diabodies, L5MK16, and MFE-232 respectively, also fall within or were close to the angle ranges described by Lawrence44 and by Holliger.45 In contrast, the T84.66Di inter-axial Y angle (2 6.08) is very different from these reported values and shows that the Fv modules are angled towards each other in this alternately linked diabody. One conclusion, on the basis of contact residues and asymmetric Fv alignment, is that the T84.66Di VL-to-VH linked diabody exhibits greater flexibility than is allowed for the VH-to-VL linked diabodies.. Supporting this, all the VH-to-VL linked diabodies showed some symmetry in their pseudo 2-fold arrangements (necessarily for the proposed diabodies) and the angles of the pseudo 2-folds to the center-to-center axis are all obtuse. Thus, asymmetry and increased flexibility might be a characteristic of VL-to-VH linking, although the evidence is limited by the small number of structures studied and crystal packing constraints. Also, the prediction that VL-to-VH linked diabodies with a negative inter-axial Y angle and an apparent occlusion of the antigen-binding sites (and subsequent reduction in affinity) are a preferred orientation is unlikely, since analysis of T84.66Di in complex with antigen demonstrates that this diabody is capable of binding two molecules of CEA (monomeric molecular mass, 180 kDa) simultaneously in solution.18 Further evidence supporting the flexibility of diabodies in solution is provided by characteri-
T84.66 anti-CEA Diabody Crystal Structure
sation of a T84.66 diabody scFv with a short linker peptide and containing a Cys residue at the C terminus.46 This Cys-diabody exists exclusively as a disulfide-bonded dimer, indicating that even though the C termini of the scFv modules are over ˚ apart in the T84.66Di crystal structure (42 A ˚ 60 A in the L5MK16 diabody), in solution the VL-to-VH linked T84.66Di could swivel to allow close association of the C termini. Further analysis of domain contacts in diabody crystal structures and models, with a variety of linker lengths, will be important in designing functional avidity, For example, if VL-to-VH linked diabodies are inherently restricted in the movement of the Fv modules, then the mutation of some inter-domain loop residues may be required to prevent this constraint. Alternatively, loop residues could be designed to stabilize the Fv modules in a particular orientation to minimize entropy loss upon bivalent binding. This could include either disulphide stabilization or reversible cross-linkers that could be designed to occlude one binding site until the targeted site has been reached and then the second binding site could be “unleashed”.
Conclusion The T84.66Di is a potentially important therapeutic molecule for the diagnosis and treatment of some cancers.9,18,46 We anticipate that molecular designs based on the T84.66 diabody structure may enhance flexibility and stability in vivo and thereby improve functional affinity. Indeed, it is possible to now design stabilized T84.66 diabodies that have a preferred binding orientation for binding two adjacent, asymmetric antigens on the cell surface. Perhaps more importantly, the knowledge of antigen-binding orientation can determine preferred fusion sites on the T84.66 diabody, since optimal clinical formulation may require conjugation to a range of effector molecules, such as radioisotopes for imaging, enzymes for prodrug therapy, toxins for targeted cell-killing, viruses for gene therapy or lipids for improved systemic delivery. Indeed, molecular designs will direct many efficient therapeutic uses of diabodies, including T84.66. In one important example, the problems of renal retention in diabody-directed radioimmunotherapy have been reduced using novel metabolizable chelate linkers.47 Finally, our comparative analysis of diabodies utilised the Gelfand Fv coordinate system (Table 2 and Figure 2) and we suggest that this be adopted as a standard method for future comparison of other scFv dimers and higher order multimers.
Acknowledgements This work was supported as part of a PhD scholarship (J.A.C) jointly administered by LaTrobe
349
T84.66 anti-CEA Diabody Crystal Structure
University and The Biomolecular Research Institute. The T84.66Di coordinates have been deposited with the RSCB (PDB, accession no. 1MOE). 15.
References 1. Todorovska, A., Dolezal, O., Kortt, A. A., Hoogenboom, H. R. & Hudson, P. J. (2001). Design and application of diabodies, triabodies and tetrabodies for cancer targeting. J. Immunol. Methods, 248, 47 – 66. 2. Tomlinson, I. & Holliger, P. (2000). Methods for generating multivalent and bispecific antibody fragments. Methods Enzymol. 326, 461– 479. 3. Carter, P. (2001). Improving the efficacy of antibodybased cancer therapies. Nature Rev. Cancer. 1, 118 – 129. 4. Hudson, P. J. & Souriau, C. (2001). Recombinant antibodies for cancer diagnosis and therapy. Expert Opin. Biol. Ther. 1, 845– 855. 5. Casey, J. L., Napier, M. P., King, D. J., Pedley, R. B., Chaplin, L. C., Weir, N. et al. (2002). Tumour targeting of humanised cross-linked divalent-Fab0 antibody fragments: a clinical phase I/II study. Br. J. Cancer, 86, 1401 –1410. 6. Willuda, J., Kubetzko, S., Waibel, R., Schubiger, P. A., Zangemeister-Wittke, U. & Pluckthun, A. (2001). Tumor targeting of mono-, di-, and tetravalent antip185(HER-2) miniantibodies multimerized by selfassociating peptides. J. Biol. Chem. 276, 14385– 14392. 7. Plu¨ckthun, A. & Pack, P. (1997). New protein engineering approaches to multivalent and bispecific antibody fragments. Immunotechnology, 3, 83 – 105. 8. Kortt, A. A., Dolezal, O., Power, B. E. & Hudson, P. J. (2001). Dimeric and trimeric antibodies: high avidity scFvs for cancer targeting. Biomol. Eng. 18, 95 – 108. 9. Yazaki, P. J., Wu, A. M., Tsai, S. W., Williams, L. E., Ikler, D. N., Wong, J. Y. et al. (2001). Tumor targeting of radiometal labeled anti-CEA recombinant T84.66 diabody and T84.66 minibody: comparison to radioiodinated fragments. Bioconjug. Chem. 12, 220– 228. 10. Tahtis, K., Lee, F. T., Smyth, F. E., Power, B. E., Renner, C., Brechbiel, M. W. et al. (2001). Biodistribution properties of (111)indium-labeled C-functionalized trans-cyclohexyl diethylenetriaminepentaacetic acid humanized 3S193 diabody and F(ab0 )(2) constructs in a breast carcinoma xenograft model. Clin. Cancer Res. 7, 1061– 1072. 11. Power, B. E., Caine, J. M., Burns, J. E., Shapira, D. R., Hattarki, M. K., Tahtis, K. et al. (2001). Construction, expression and characterisation of a single-chain diabody derived from a humanised anti-Lewis Y cancer targeting antibody using a heat-inducible bacterial secretion vector. Cancer Immunol. Immunother. 50, 241– 250. 12. Nielsen, U. B., Adams, G. P., Weiner, L. M. & Marks, J. D. (2000). Targeting of bivalent anti-ErbB2 diabody antibody fragments to tumor cells is independent of the intrinsic antibody affinity. Cancer Res. 60, 6434–6440. 13. Adams, G. P., Shaller, C. C., Chappell, L. L., Wu, C., Horak, E. M., Simmons, H. H. et al. (2000). Delivery of the alpha-emitting radioisotope bismuth-213 to solid tumors via single-chain Fv and diabody molecules. Nucl. Med. Biol. 27, 339– 346. 14. Dolezal, O., Pearce, L. A., Lawrence, L. J., McCoy, A. J., Hudson, P. J. & Kortt, A. A. (2000). ScFv multi-
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
mers of the anti-neuraminidase antibody NC10: shortening of the linker in single-chain Fv fragment assembled in V(L) to V(H) orientation drives the formation of dimers, trimers, tetramers and higher molecular mass multimers. Protein Eng. 13, 565–574. Neumaier, M., Shively, L., Chen, F. S., Gaida, F. J., Ilgen, C., Paxton, R. J. et al. (1990). Cloning of the genes for T84.66, an antibody that has a high specificity and affinity for carcinoembryonic antigen, and expression of chimeric human/mouse T84.66 genes in myeloma and chinese hamster ovary cells. Cancer Res. 50, 2128– 2134. Wong, J. Y., Williams, L. E., Yamauchi, D. M., OdomMaryon, T., Esteban, J. M., Neumaier, M. et al. (1995). Initial experience evaluating 90 yttrium-radiolabeled anti-carcinoembryonic antigen chimeric T84.66 in a phase I radioimmunotherapy trial. Cancer Res. 55, 5929s – 5934s. Wong, J. Y., Chu, D. Z., Yamauchi, D., OdomMaryon, T. L., Williams, L. E., Liu, A. et al. (1998). Dose escalation trial of indium-111-labeled anticarcinoembryonic antigen chimeric monoclonal antibody (chimeric T84.66) in presurgical colorectal cancer patients. J. Nucl. Med. 39, 2097– 2104. Wu, A. M., Williams, L. E., Zieran, L., Padma, A., Sherman, M., Bebb, G. G. et al. (1999). Anticarcinoembryonic antigen (CEA) diabody for rapid tumor targeting and imaging. Tumor Targeting, 4, 47 – 58. Wu, A. M., Chen, W., Raubitschek, A., Williams, L. E., Neumaier, M., Fischer, R. et al. (1996). Tumor localization of anti-CEA single-chain Fvs: improved targeting by non-covalent dimers. Immunotechnology, 2, 21 – 36. Hu, S. Z., Shively, L., Raubitschek, A. A., Sherman, M., Williams, L. E., Wong, J. Y. et al. (1996). Minibody: a novel engineered anti-carcinoembryonic antigen antibody fragment (single-chain Fv-CH3) which exhibits rapid, high-level targeting of xenografts. Cancer Res. 56, 3055– 3061. Yazaki, P. J., Shively, L., Clark, C., Cheung, C., Le, W., Szpikowska, B. et al. (2001). Mammalian expression and hollow fiber bioreactor production of recombinant anti-CEA diabody and minibody for clinical applications. J. Immunol. Methods, 253, 195– 208. Wu, A. M., Yazaki, P. J., Tsai, S. w., Nguyen, K., Anderson, A. L., McCarthy, D. W. et al. (2000). Highresolution microPET imaging of carcinoembryonic antigen-positive xenografts by using a copper-64labeled engineered antibody fragment. Proc. Natl Acad. Sci. USA, 97, 8495– 8500. Perisic, O., Webb, P. A., Holliger, P., Winter, G. & Williams, R. L. (1994). Crystal structure of a diabody, a bivalent antibody fragment. Structure, 2, 1217 –1226. Malby, R. L., McCoy, A. J., Kortt, A. A., Hudson, P. J. & Colman, P. M. (1998). Three-dimensional structures of single-chain Fv – neuraminidase complexes. J. Mol. Biol. 279, 901– 910. Boehm, M. K., Corper, A. L., Wan, T., Sohi, M. K., Sutton, B. J., Thornton, J. D. et al. (2000). Crystal structure of the anti-(carcinoembryonic antigen) single-chain Fv antibody MFE-23 and a model for antigen binding based on intermolecular contacts. Biochem. J. 346, 519–528. Lee, Y. C., Boehm, M. K., Chester, K. A., Begent, R. H. & Perkins, S. J. (2002). Reversible dimer formation and stability of the anti-tumour single-chain Fv
350
27.
28.
29.
30.
31. 32.
33. 34.
35. 36.
37.
38.
39.
40.
41.
42.
antibody MFE-23 by neutron scattering. Analytical ultracentrifugation, and NMR and FT-IR spectroscopy. J. Mol. Biol. 320, 107– 127. Murakami, M., Kuroki, M., Arakawa, F., Kuwahara, M., Oikawa, S., Nakazato, H. & Matsuoka, Y. (1995). A reference of the GOLD classification of monoclonal antibodies against carcinoembryonic antigen to the domain structure of the carcinoembryonic antigen molecule. Hybridoma, 14, 19 – 28. Ramakrishnan, C. & Ramachandran, G. N. (1965). Stereochemical criteria for polypeptide and protein chain conformations. II. Allowed conformations for a pair of peptide units. Biophys. J. 5, 909– 933. Laskowski, R. A., MacArthur, M. W., Moss, D. S. & Thornton, J. M. (1993). PROCHECK: a program to check the stereochemicai quality of protein structures. J. Appl. Crystallog. 26, 283– 291. Fan, Z. C., Shan, L., Guddat, L. W., He, X. M., Gray, W. R., Raison, R. L. & Edmundson, A. B. (1992). Three-dimensional structure of an Fv from a human IgM immunoglobulin. J. Mol. Biol. 228, 188– 207. Chothia, C. & Lesk, A. M. (1987). Canonical structures for the hypervariable regions of immunoglobulins. J. Mol. Biol. 196, 901–917. Chothia, C., Lesk, A. M., Gherardi, E., Tomlinson, I. M., Walter, G., Marks, J. D. et al. (1992). Structural repertoire of the human VH segments. J. Mol. Biol. 227, 799– 817. Al-Lazikani, B., Lesk, A. M. & Chothia, C. (1997). Standard conformations for the canonical structures of immunoglobulins. J. Mol. Biol. 273, 927– 948. Morea, V., Tramontano, A., Rustici, M., Chothia, C. & Lesk, A. M. (1998). Conformations of the third hypervariable region in the VH domain of immunoglobulins. J. Mol. Biol. 275, 269–294. Shirai, H., Kidera, A. & Nakamura, H. (1996). Structural classification of CDR-H3 in antibodies. FEBS Letters, 399, 1 – 8. Kabat, E. A., Wu, T. T., Perry, H. M., Gottesman, K. S. & Foeller, C. (1991). Sequences of Proteins of Immunological Interest NIH, Publ. no. 91-3242, 5th edit., DHHS, PHS, National Institutes of Health, Bethseda, MD. Chothia, C., Novotny, J., Bruccoleri, R. & Karplus, M. (1985). Domain association in immunoglobulin molecules. The packing of variable domains. J. Mol. Biol. 186, 651– 663. Novotny, J. & Haber, E. (1985). Structural invariants of antigen binding: comparison of immunoglobulin VL-VH and VL-VL domain dimers. Proc. Natl Acad. Sci. USA, 82, 4592– 4596. Kortt, A. A., Malby, R. L., Caldwell, J. B., Gruen, L. C., Ivancic, N., Lawrence, M. C. et al. (1994). Recombinant anti-sialidase single-chain variable fragment antibody. Characterization, formation of dimer and higher-molecular-mass multimers and the solution of the crystal structure of the single-chain variable fragment/sialidase complex. Eur. J. Biochem. 221, 151– 157. Boehm, M. K. & Perkins, S. J. (2000). Structural models for carcinoembryonic antigen and its complex with the single-chain Fv antibody molecule MFE23. FEBS Letters, 475, 11 – 16. Whitlow, M., Filpula, D., Rollence, M. L., Feng, S. L. & Wood, J. F. (1994). Multivalent Fvs: characterization of single-chain Fv oligomers and preparation of a bispecific Fv. Protein Eng. 7, 1017– 1026. Kuttner, G., Keitel, T., Giessmann, E., Wessner, H., Scholz, C. & Hohne, W. (1998). A phage library-
T84.66 anti-CEA Diabody Crystal Structure
43.
44.
45.
46.
47. 48.
49. 50.
51. 52. 53. 54.
55.
56.
57.
58.
derived single-chain Fv fragment in complex with turkey egg-white lysozyme: characterization, crystallization and preliminary X-ray analysis. Mol. Immunol. 35, 189– 194. Gelfand, I. M., Kister, A. E., Kulikowski, C. & Stoyanov, O. (1998). Geometric invariant core for the V(L) and V(H) domains of immunoglobulin molecules. Protein Eng. 11, 1015– 10125. Lawrence, L. J., Kortt, A. A., Iliades, P., Tulloch, P. A. & Hudson, P. J. (1998). Orientation of antigen binding sites in dimeric and trimeric single chain Fv antibody fragments. FEBS Letters, 425, 479– 484. Holliger, P., Brissinck, J., Williams, R. L., Thielemans, K. & Winter, G. (1996). Specific killing of lymphoma cells by cytotoxic T-cells mediated by a bispecific diabody. Protein Eng. 9, 299– 305. Li, L., Olafsen, T., Anderson, A. L., Wu, A., Raubitschek, A. A. & Shively, J. E. (2002). Reduction of kidney uptake in radiometal labeled peptide linkers conjugated to recombinant antibody fragments. Site-specific conjugation of DOTA-peptides to a Cys-diabody. Bioconjug. Chem. 13, 985– 995. Harris, L. J., Skaletsky, E. & McPherson, A. (1998). Crystallographic structure of an intact IgG1 monoclonal antibody. J. Mol. Biol. 275, 861– 872. Esnouf, R. M. (1997). An extensively modified version of MolScript that includes greatly enhanced coloring capabilities. J. Mol. Graph. Model. 15, 132– 134. Merritt, E. A. & Murphy, M. E. P. (1994). Raster3D version 2.0—a program for photorealistic molecular graphics. Acta Crystallog. sect. D, 50, 869– 873. Varghese, J. N., van Donkelaar, A., Hrmova, M., Fincher, G. A., Baliac, D. X. & Barnia, Z. (1999). Structure of an exoglucanase complexed with conduritol B epoxide from a 50m crystal using monocapillary optics. Acta Crystallog. sect. A, 55, 329– 330. Matthews, B. W. (1985). Determination of protein molecular weight, hydration, and packing from crystal density. Methods Enzymol. 114, 176–187. Otwinowski, Z. & Minor, W. (1996). Processing of X-ray diffraction data collected in oscillation mode. Methods Enzymol. 276, 307– 326. Navaza, J. (1994). AMoRe: an automated package for molecular replacement. Acta Crystallog. sect. A, 50, 157– 163. Rini, J. M., Stanfield, R. L., Stura, E. A., Salinas, P. A., Profy, A. T. & Wilson, I. A. (1993). Crystal structure of an HIV-1 neutralizing antibody 50.1 in complex with its V3 loop peptide antigen. Proc. Natl Acad. Sci. USA, 90, 6325– 6329. Jones, T. A., Zou, J. Y., Cowan, S. W. & Kjeldgaard, M. (1991). Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Crystallog. sect. A, 47, 110 – 119. Bru¨nger, A. T., Adams, P. D., Clore, G. M., DeLano, W. L., Gros, P., Grosse-Kunstleve, R. W. et al. (1998). Crystallography and NMR system: a new software suite for macromolecular structure determination. Acta Crystallog. sect. D, 54, 905– 921. Adams, P. D., Pannu, N. S., Read, R. J. & Bru¨nger, A. T. (1997). Cross-validated maximum likelihood enhances crystallographic simulated annealing refinement. Proc. Natl Acad. Sci. USA, 94, 5018– 5023. Hodel, A., Kim, S.-H. & Bru¨nger, A. T. (1992). Model bias in macromolecular crystal structures. Acta Crystallog. sect. A, 48, 851– 858.
T84.66 anti-CEA Diabody Crystal Structure
59. Holliger, P., Prospero, T. & Winter, G. (1993). “Diabodies”: small bivalent and bispecific antibody fragments. Proc. Natl Acad. Sci. USA, 90, 6444– 6448. 60. Gelfand, I. M. & Kister, A. E. (1997). A very limited number of keywords (main patterns) describes all
351
sequences of the human variable heavy (VH) and kappa (Vkappa) domains. Proc. Natl Acad. Sci. USA, 94, 12562– 12567.
Edited by I. Wilson (Received 3 September 2002; received in revised form 3 December 2002; accepted 3 December 2002)