Rearrangement of the Former VL Interface in the Solution Structure of a Camelised, Single Antibody VH Domain

J. Mol. Biol. (1996) 259, 957–969

Rearrangement of the Former VL Interface in the Solution Structure of a Camelised, Single Antibody VH Domain Lutz Riechmann MRC Laboratory of Molecular Biology Hills Road, Cambridge CB2 2QH, UK

The solution structure of the isolated antibody heavy chain variable domain (VH)-P8 was determined by NMR spectroscopy. The VH had previously been modified (camelised) at three positions in its former antibody light chain variable domain (VL) interface to reduce hydrophobicity by mimicking camelid heavy chains naturally devoid of light chains. The architecture of two pleated b-sheets and the conformation of the H1 and H2 loops in VH-P8 are very similar to those in non-camelised, VL-associated VH domains. Major differences concern the H3 loop, which no longer points towards the now absent VL, and three residues in the former VL interface. The side-chains of Val37 and Trp103 are buried and the Arg38 side-chain exposed in VH-P8. In non-camelised, VL-associated VH domains the side-chains of Val37 and Trp103 are in contact with the VL while the Arg38 side-chain is buried within the VH. Reorientation of Trp103 is due to the local structure in the b-bulge of strand G. Reorientation of Val37 and Arg38 is caused by a disruption of regular b-structure in strand C opposite the b-bulge in strand C'. These changes, combined with the more hydrophilic side-chains of the camelised residues, reduce hydrophobicity and prevent non-specific binding of camelised VH domains, which proved critical for their use as small recognition units. The VH-P8 structure also indicates structural reasons for two other mutations specific for light-chain-lacking camel immunoglobins. Absence of the VH-typical Arg94/Asp101 salt bridge at the base of the H3 loop in VH-P8 may explain why a positively charged residue at position 94 is not conserved in camels. Reorientation of Val37 suggests a function of the camel-specific phenylalanine residue at this position in the hydrophobic core of light-chain-lacking camel heavy chains. 7 1996 Academic Press Limited

Keywords: NMR; immunoglobulin; camel; hydrophobicity; antibody engineering

Abbreviations used: 2D, two-dimensional; 3D, three dimensional; CDR, complementarity-determining region; Chaps, 3-[(3-Cholamidopropyl) dimethylammonio]-1-propanesuphonate; COSY, correlated spectroscopy; CT, constant time; HMQC, hetero multiple quantum correlation spectroscopy; HSQC, hetero single quantum correlation spectroscopy; Fv, heterodimer of VH and VL; Fab, antigen-binding antibody fragment; H1, H2 and H3, hypervariable loops 1, 2 and 3; NOE, nuclear Overhauser enhancement; NOESY, NOE spectroscopy; RMSD, root mean square deviation; TOCSY, total correlation spectroscopy; TPPI, time proportional phase incrementation; VH, antibody heavy chain variable domain; VL, antibody light chain variable domain. Residues are numbered according to Kabat et al., 1991. Mutants are denoted by the wild-type amino acid residue (one-letter code) followed by the residue number and the mutant residue. 0022–2836/96/250957–13 $18.00/0

Introduction Immunoglobulins are a powerful tool for the specific targeting of ligands and objects presenting them. The expression of recombinant antibodies and their display on the surface of phage has recently increased their potential even further (Winter et al., 1994). Through phage display, antibody fragments with specificity for virtually any antigen can be isolated in vitro (Griffiths et al., 1994). Selected antibodies can afterwards be produced as soluble proteins in eukaryotic cells or bacteria for biotechnological and pharmaceutical applications. For bacterial expression and phage display it was advantageous to modify the original architectural format of the antibody molecule, which consists of 7 1996 Academic Press Limited

958 two multi-domain heavy and light chains. Bacterial folding of entire immunoglobulin molecules is less efficient than the folding of their Fv fragments, which consist of a single pair of heavy and light chain variable domains, or Fab fragments containing in addition a pair of light and heavy chain constant domains. In some situations, like the targeting of solid tumours, smaller sized antibody fragments can be therapeutically even more effective than complete antibodies due to their faster diffusion. Attempts were therefore made to reduce further the size of the minimum antibody fragment required for antigen binding. These attempts focused on the antibody heavy chain variable (VH) domains as it was observed relatively early that heavy chains can sometimes retain a significant portion of antigen affinity in the absence of light chains (Utsumi & Karush, 1964). Furthermore, in camels (HamersCasterman et al., 1993), antibodies naturally devoid of light chains have been discovered. Similarly, bacterially expressed VH domains with antigen specificity were rescued from immunised mice (Ward et al., 1989). This led to the design of VH-based b-domains (Pessi et al., 1993) and to the modification of a human VH domain for use as a small recognition unit through the mimicking of camel heavy chains (Davies & Riechmann, 1994). Libraries of both designs were expressed on the surface of phage and antigen-specific single-domain recognition units were selected (Martin et al., 1994; Davies & Riechmann, 1995a). VH domains represent therefore arguably the smallest immunoglobulin-based recognition unit. Despite a large amount of structural information about VH domains as part of Fv or Fab fragments (see, for example, the review by Alzari et al., 1988) very little is known about the structure of VH domains when removed from this context. So far only the secondary structure of a single camelid, human VH domain has been determined through NMR analysis (Riechmann & Davies, 1995). More information about their structure, stability and interactions with ligands is needed to optimise VH domains for their use as small recognition units. Towards this goal we report here the 1H and 13C side-chain NMR assignments and the detailed three-dimensional solution structure of a camelised VH domain based on 15N and 13C resolved nuclear Overhauser enhancement (NOE) data.

Results VH domains Several years ago, single VH domains were proposed as a template for the design of small recognition units (Ward et al., 1989). Their exploitation however proved more difficult than anticipated. Purification was often hampered by poor bacterial expression and low stability. Nonspecific binding (or ‘‘stickiness’’) of VH domains, presumably through the exposed, hydrophobic

Solution Structure of a Camelised VH Domain

light chain variable domain (VL) interface, caused loss of protein during work-up procedures and compromised antigen specificity. When we therefore found a VH of human origin that was expressed well (10 to 15 mg of purified VH from one litre of bacterial culture) even in the absence of a VL, its VL interface still had to be modified to reduce stickiness. This was achieved through the introduction of the mutations G44E, L45R and W47G (Davies & Riechmann, 1994) in the human VH domain (VH-P1) to mimic camelised heavy chains occurring naturally without light chains (Hamers-Casterman et al., 1993). Residues 44, 45 and 47 are central to interactions with the VL (Poljak et al., 1975; Chothia et al., 1985). VH-P1, which has no known antigen specificity, was later used as a template for the generation of a repertoire of phage-displayed VH domains through randomisation of its complementarity-determining region (CDR) 3. VH domains specific for hapten, protein and peptide ligands were isolated from this library (Davies & Riechmann, 1995a; L.R., unpublished results). Camelising enabled also the structural analysis of the VH by NMR spectroscopy. A differently camelised VH (VH-P8, which corresponds to VH-P1 except for residue Ile47) gave better expression than VH-P1 and was therefore more amenable to isotopic labelling. Both VH domains did not aggregate in solution and behaved as monomeric molecules in gel-filtration experiments. The larger NMR linewidth of VH-P8 was reduced to that of VH-P1 in the presence of the detergent 3-[(3-Cholamidopropyl)dimethylammonio]-1-propanesulphonate (Chaps) to allow a detailed multi-dimensional NMR analysis (Davies & Riechmann, 1994). The backbone assignments, including those of the 15N and 13Ca nuclei of VH-P8 and its secondary structure, have been reported before (Riechmann & Davies, 1995). At the level of the secondary structure no differences from VH domains as part of Fv or Fab fragments were observed at that stage. Structural analysis of the Ile47-containing VH-P8 is relevant as ligand binding, camelised VH domains selected after phage display (see above) were also expressed as Ile47-containing soluble proteins due to their better expression and higher stability compared to the Gly47-containing VH domains. The G47I mutation does not usually affect antigen binding or specificity (Davies & Riechmann, 1995a, 1996). Furthermore, amino acid residues with branched aliphatic side-chains at position 47 are also found in naturally occurring, light-chain lacking camel heavy chains (Muyldermans et al., 1994). Assignments and NOESY analysis The backbone assignments of VH-P8 were extended to the side-chain 1Hb and 13Cb signals using data from HBHA(CO)NH and CBCA(CO)NH experiments. The remaining 1H and 13C

959

Solution Structure of a Camelised VH Domain

(a)

(c)

(b)

(d)

Figure 1. (a) Inter-residue NOE-derived distance constraints per residue. (b) Ca (continuous trace) and heavy atom (broken trace) rms deviation (RMSD) for the 20 best structures to the mean structure versus residue number. (c) Angular order parameter of F (broken trace) and C (continuous trace) for each residue calculated for the 20 best structures. An identical angle in all structures corresponds to an order parameter of 1.0. A random distribution of angles corresponds to an order parameter of 20−0.5 = 0.22. (d) Distances between the Ca atoms (except those of residues 52a, 98 and 99) of the average VH-P8 structure and the Pot-VH (bold continuous trace), VH-P8 and the D1.3-VH (broken trace) or the Pot-VH and the D1.3-VH (thin continuous trace) when superposed on the Ca atoms of residues 2 to 52, 55 to 96 and 101 to 112. b-Strands A to G are indicated.

side-chain signals were then identified through connectivities to the a and b chemical shifts in HCCH-correlated spectroscopy (COSY) and HCCH-total correlation spectroscopy (TOCSY) experiments. The 13C chemical shifts of the aromatic side-chains were identified in a two-dimensional (2D) constant time hetero single quantum correlation spectroscopy (CT-HSQC) spectrum based on their 1H chemical shifts. Interpretation of the previously reported threedimensional (3D) 15N-1H NOE spectroscopy (NOESY)-hetero multiple quantum correlation spectroscopy (HMQC) spectrum recorded with the 15 N-labeled VH was completed using the additional side-chain assignments. NOE crosspeaks in this and the 3D 13C-1H NOESY-HMQC spectra recorded

with the 13C/15N-labeled VH in 2H2 O were assigned mostly on the basis of chemical shifts alone. Assignment ambiguities in the 13C-1H NOESYHMQC experiments were resolved as much as possible through identification of corresponding NOEs originating from the other partner. Some remaining ambiguities could be resolved later using intermediate results from the structure calculations. Structure calculation A total of 509 sequential, 197 medium-range (=i − j =E4) and 752 long-range (=i − j = > 4) interresidue (Figure 1(a)) and a further 1175 intraresidue distance constraints were derived from the

960

Solution Structure of a Camelised VH Domain

Figure 2. Hydrogen bond distribution in the b-sheets of VH-P8. The b-strand architecture of a typical immunoglobulin VH domain is shown. Arrowed lines indicate hydrogen bonds with the arrow pointing from the (NH) donor residue to the (CO) acceptor group. Continuous arrowed lines indicate hydrogen bonds resulting in protection of the involved NH from 2H2 O exchange, which were used as distance constraints in the X-PLOR calculations. Broken arrowed lines indicate likely hydrogen bonds in VH-P8 according to the calculated structures. Conditions for a possible hydrogen bond were an H-O dis˚ , an N-H-O tance of less than 2.5 A angle of between 120° and 180° and an H-O-C angle of between 90° and 150° in the average of the 20 best VH-P8 structures. Hydrogen bonds not present in VH-P8 are indicated by crossed, arrowed lines. Hydrogen bonds observed in VH-P8 but not present in the Pot-VH are marked with asterisks. b-Strands A to G are indicated.

NOE analysis. A disulphide bond constraint for residues 22 and 92 as well as 33 hydrogen bond constraints (Figure 2) deduced from 2H2 O exchange experiments based on the secondary structure were also used for the structure calculations. Of the 60 calculated structures, 42 converged with no more than ten NOE violations by more ˚ each. Figure 3 illustrates the dependence than 0.3 A of the average backbone (Ca, N and C' atoms) rms deviation on the ensemble size of the structures when ordered according to their total X-PLOR energy. Thus the rms deviation for the backbone of the 40 energetically best structures to their mean ˚ . The structural statistics of the 20 structure is 0.70 A energetically best structures, whose ensemble has ˚ for the an average overall rms deviation of 0.59 A ˚ backbone and 1.04 A for the heavy atoms, are given in Table 1.

well defined in the calculated structures (Figures 1(b) and 4). They have an average rms devi˚ for the backbone and 0.92 A ˚ for the ation of 0.43 A

General structure The principal architecture of the VH consists of nine anti-parallel b-strands forming two pleated sheets (Figures 4 and 5). Residues in these sheets (81 out of 117 residues; 3 to 12 (strand A), 17 to 25 (B), 32 to 40 (C), 44 to 52 (C'), 56 to 60 (C"), 66 to 73 (D), 76 to 82b (E), 87 to 96 (F) and 102 to 112 (G)) are

Figure 3. Rms deviation (RMSD) profile of calculated VH-P8 structures in order of their total X-PLOR energy. The mean backbone (Ca, N and C') rms deviation to the mean structure is plotted against increasing ensemble size of structures.

961

Solution Structure of a Camelised VH Domain

Table 1. Structural statistics of the 20 energetically best VH-P8 structures Mean energy terms: Total X-PLOR energy (kcal/mol) Van der Waals energy (kcal/mol) Distance constraints energy (kcal/mol) Distance constraints: Average rms deviation NOE ˚) distance violation (A ˚ in all Sum of violations >0.3 A 20 structures Rms deviations from the ideal geometry: ˚) Bond lengths (A Bond angles (°) Improper angle (°)

645.2 (233.9) 152.4 (213.9) 133.1 (215.0)

0.032 (20.002) 7 0.0044 (20.0011) 0.74 (20.02) 0.61 (20.02)

˚ ): Atomic rms deviations to mean structure (A Residues Backbone atoms Heavy atoms 1 to 112 0.59 (20.10) 1.04 (20.16) b-strands 0.43 (20.06) 0.92 (20.17) (Residues 3 to 12, 17 to 25, 32 to 40, 44 to 52, 56 to 60, 66 to 73, 76 to 82b, 87 to 96, 102 to 112)

heavy atoms as well as high angular order parameters of F and C angles for the 20 best structures (Figure 1(c)). Beside the terminal residue the least well-defined regions are the H2 and H3 loops and the turn connecting strands E and F (Figures 1(b) and 4). To detect features particular to the isolated, camelised VH domain, its solution structure was compared to the X-ray structures of the VH domains from the human Fv Pot (Fan et al., 1992) and the mouse Fv D1.3 (Bhat et al., 1994) providing high-resolution models of VL-associated VH domains (Figure 5). The Pot-VH originated from the same human VH family (subgroup III) as the

camelised VH-P8. A total of 92 of the 117 VH-P8 residues in the entire sequence are identical in the two VHs. CDR1 and 2 have the same length in both proteins. The H2 loop (residues 52a to 55) is identical in sequence, while CDR1 differs in sequence at positions 30 to 33. The structure of the Pot-VH (except for CDR3) should therefore provide a reasonable approximation for the structure of the VH-P8 wild-type when complexed with a VL. A comparison of the CDR3 structures in VH-P8 and the Pot-VH is not informative, as the Pot-VH contains four additional residues. VH-P8 was therefore compared with the D1.3-VH, which has a CDR3 of identical length. However, only 60 out of 117 are identical in the D1.3-VH and VH-P8. The CDR2 in the D1.3-VH contains one residue less. The principle architecture of the camelised VH-P8 based on the nine anti-parallel b-strands is very similar to that of both the Pot-VH and the D1.3-VH (Figure 5). The hydrogen bonding pattern is mostly identical to that in the Pot-VH (Figure 2). The b-bulges in strands C' and G are also found at identical positions. The most prominent differences concern a different backbone conformation in the b-bulge of strand G resulting in a reorientation of the Trp103 side-chain and a disruption of regular b-structure between residues 37 and 38 in strand C, which causes a kink of the polypeptide chain opposite to the location of the b-bulge in the adjacent strand C' (Figure 2). The effect on the surface properties of the former VL interface are discussed below. An improved, least-squares global superposition using the program O (Jones et al., 1991) yields an Ca rms deviation between VH-P8 and the Pot-VH

Figure 4. Superposition of the 20 energetically best VH-P8 structures to the lowest energy structure on the Ca atoms of residues 2 to 112. The stereo plot shows backbone carbon and nitrogen atoms only.

962

Solution Structure of a Camelised VH Domain

Figure 5. Molscript (Kraulis, 1991) diagrams showing the superposition of the best VH-P8 structure (red) and the X-ray structure of the VH from the human Fv Pot (yellow) or the D1.3-VH (green) to the Ca atoms of residues 2 to 52, 55 to 96 and 101 to 112. Hypervariable loops (Hl, H2, H3) and b-strands (A, B, C, C', C", D, E, F, G) are indicated.

˚ for 82 aligned residues. Non-aligned of 2.01 A regions include the terminal residues and residues 10, 11, 28 to 30 (in CDR1), 43 to 61 (strands C', H2 and strand C"), 84 to 86 and 96 to 99 (in CDR3). The Ca rms deviation between VH-P8 and the D1.3-VH ˚ with 91 aligned residues. Non-aligned is 2.06 A residues in this case are the terminal residues and residues 11, 12, 30 (in CDR1), 43 to 47 (strand C'), 52a to 61 (strand C") and 85, 86, 98 and 99 (in CDR3). The same rms deviation between the ˚ with a Pot-VH and the D1.3-VH is only 1.28 A satisfactory alignment of all residues. Large rms deviations between VH-P8 and the two VH domains from crystal structures of Fv fragments are not restricted to the mutated, former VL interface but are found throughout the whole sequence (Figure 1(d)). However, the high rms deviation values are misleading as individual pairs of b-strands and even some of the loops superpose well. For example, strand C' and C" do not align in the global superposition of VH-P8 with both the Pot-VH and the D1.3-VH. However, the local least-squares superposition of residues 48 to 60, which include these two strands and the H2 loop, ˚ (Pot-VH) and yields Ca rms deviations of 1.12 A ˚ (D1.3-VH). The deviations are therefore not 1.64 A solely caused by changes of structural details but also by small, rigid-body movements of different regions in VH-P8. A reason for this type of discrepancy between the crystal structures and that of VH-P8 might be the smaller number of distance

constraints available for the loops and turns at the top and bottom of the VH-P8 molecule (Figure 1(a)). The differences in strand C and G discussed above and changes in the H3 conformation (see below) however persist even after local least-square superposition and are significant structural changes caused by the camelising mutations in the former VL interface and/or the absence of a VL partner. CDR structures Conformations of the hypervariable loops in CDR1 and 2 of VH-P8 are principally similar to those of the VL-associated VH domains. The H1 loop of VH-P8 folds like that of the Pot-VH into canonical structure 1 following the nomenclature of Chothia et al. (1992). This structure is determined by a sharp turn around Gly26 and the conformation of residues Ala24, Phe27, Phe29, Met34 and Arg94. The turn around Gly26 and the orientation of the Phe27 side-chain are identical in VH-P8 and the Pot-VH (Figure 6). The Phe29 side-chain is buried between the side-chains of residues 24 and 34, albeit in an orientation slightly different from that in the ˚ for the Pot-VH. The mean rms deviation is 1.59 A Ca atoms of residues 26 to 32 from the average of the 20 best VH-P8 structures and from the Pot-VH after superposition on these atoms. The conformation of the H2 loop in VH-P8 is almost identical to that of the Pot-VH. The average

Solution Structure of a Camelised VH Domain

963

Figure 6. H1, H2 and H3 loops from the five best VH-P8 structures (lines only) and from the Pot-VH (ball-and-stick models in H1 and H2) or the D1.3-VH (ball-and-stick model in H3) after superposition onto the Ca atoms of residues 23 to 35 (H1), 52 to 56 (H2) and 91 to 104 (H3) of the best VH-P8 structure. The backbone atoms only are shown except for residues 27, 29, 94 and 101, for which the side-chains are also shown. The salt bridge between the side-chains of residues 94 and 101 in the D1.3-VH is indicated by a broken line.

˚ for the Ca atoms of residues rms deviation is 1.05 A 52 to 56 from the two VHs when superposed onto these atoms. In both proteins the CDR2 loop forms the canonical structure 3 of H2 regions, which is

one of the two structures found for H2 loops with six residues between position 52 and 56 (Figure 6; Chothia et al., 1992). A slight difference is seen for residue 52a, which however is the most poorly

964 defined position in CDR2 of the NMR structure (Figures 1(b) and (c)). The H3 loop (residues 94 to 102) of VH-P8 occupies a conformation very different from the H3 of identical length in the D1.3-VH (Figure 6). Even a local superposition of the Ca atoms of residues 91 to 104 from the D1.3-VH and VH-P8 yields an ˚ for these atoms. In average rms deviation of 2.69 A a more global superposition, as in Figure 5, it becomes evident that the H3 loop in the D1.3-VH is orientated towards the VL while it is orientated more towards the H1 and H2 loops in the VH-P8 structures (Figures 4 and 5). Another change in CDR3 concerns the salt bridge between the side-chains of Arg94 and Asp101 at the base of the H3 hairpin, which is formed in most VL-associated VH domains but not in VH-P8 (Figure 6). The side-chain of Arg94 is relatively disordered in the NMR structures. It is, however, in all of the 20 best structures, buried within the VH. In contrast, the side-chain of residue 101 is orientated towards the solvent in all structures. The relevant atoms of the two side-chains are in none of the structures in hydrogen-bond range. The Asp101 side-chain contributes in VH-P8 to the hydrophilicity of the former and now exposed VL interface. Former VL interface The VL interface of non-camelised VH domains when part of Fv fragments is formed mainly by b-strands C' and G, which both include a b-bulge, and the H3 loop. The differences in the H3 loop have been described above. Within strand C' residues 44, 45 and 47 are those contributing most to the VL interface (Poljak et al., 1975; Chothia et al., 1985). They had been mutated during the camelising of VH-P8. There is no principle difference in the conformation of these residues. The side-chain of Arg45 is solvent exposed in VH-P8, while Leu45 is pointing towards the VL in the Pot-VH. The Glu44 side-chain is also solvent exposed in VH-P8. In most non-camelised VH-domains residue 44 is a glycine (Kabat et al., 1991). Residues 44 and 45 therefore both reduce the hydrophobic character of the former VL interface in VH-P8 (see Figure 8). The Ile47 side-chain folds like the VH-Trp47 side-chain in the Pot-Fv back onto the backbone of residue 49 in strand C'. It is partly exposed to the solvent in VH-P8 but its accessible surface area is smaller than the VL-contacting surface area of Trp47 in an Fv (Figures 7 and 8). Although no changes concerning the VH-typical b-structure are observed in strand C', there are major differences in the neighbouring strand C. Residues Val37 and Arg38 have opposite orientations in VH-P8 and the Pot-VH (Figure 6). This is due to a distortion of the b-structure within this strand between residues 37 and 38 as indicated by, for b-structure, abnormal (Wu¨thrich, 1986) F angles of on average 111° in the 20 best structures causing a kink in the polypeptide chain (Figures 7 and 8). The structure is well defined in this region

Solution Structure of a Camelised VH Domain

(Figure 7) and residues 37 and 38 both have very high order parameters between 0.98 and 1.0 for their torsion angles (Figure 1(c)). The backbone distortion in strand C causes the side-chain of Val37 to be buried in the interior of VH-P8 where it interacts with the backbone of strand C'. In the Pot-VH and other VL-associated VH domains, the Val37 side-chain forms the bottom of a hydrophobic pocket in the VL interface, which accommodates the side-chain of the highly conserved light chain residue Phe98. In contrast, the side-chain of Arg38 is buried within the Pot-VH but solvent exposed in VH-P8 (Figure 8). As far as backbone interactions are concerned, the NH group of residue 39 in VH-P8 does not hydrogen bond with the backbone oxygen atom of residue 89 in strand F as in VL-associated VH domains. It instead forms a hydrogen bond with the CO group of residue 46 from strand C' (Figure 2). There is also no longer a hydrogen bond between the NH of residue 46 and the carbonyl group of residue 38. The backbone conformation within the b-bulge of strand G in VH-P8 also differs from that in VL-associated VH domains (Figure 7). The Trp103 side-chain would be exposed to the solvent in an Fv structure after removal of the VL. In VH-P8 it points towards the interior of the protein where it interacts with the backbone of strands A and F. As a result most of the Trp103 side-chain is no longer exposed to the solvent (Figure 8).

Discussion The solution structure of the single antibody domain VH-P8 has been solved to a medium overall resolution with well-defined regions of b-structure and less well-defined connecting loops and turns (Figures 1 and 4). The general architecture of two pleated b-sheets and the conformation of the hypervariable loops H1 and H2 is very similar to that in crystal structures of VL-associated VH domains. However, the overall Ca rms deviation of more ˚ between VH-P8 and VH domains from the than 2 A crystal structures of the human Fv Pot or the mouse Fv D1.3 signifies important differences in the details of the structures. The same rms deviation between the VH domains from the Pot-Fv and ˚ . Structural changes, especially D1.3-Fv is only 1.3 A in the former VL interface of VH-P8, are most likely due to either the camelising mutations and/or the absence of a VL partner. The presence of its VH partner had very little effect on the structure of the VL monomer of the antibody 26-10, whose solution structure was solved to a resolution comparable to that of VH-P8 (Constantine et al., 1994). The backbone rms deviation of this isolated VL and the same VL in the crystal structure of the complex with its VH is only ˚ . It seems therefore that either this isolated 1.05 A VL domain is less affected than VH-P8 by the presence of its natural immunoglobulin partner or that indeed the camelising mutations are most

Solution Structure of a Camelised VH Domain

965

Figure 7. Ca trace of the five best VH-P8 structures (top) superposed to the Ca trace of the energetically best structure and of the Pot-VH (bottom) in the same orientation looking onto the VL interface. The Ca trace of residues 32 to 52 and 95 to 112 and the side-chains of residues 37, 45, 47, 100 (100d in the Pot-VH) and 103 are shown.

critical for the structural abnormalities observed in VH-P8. The most significant differences between the VH-P8 structure and those of VL-associated VH domains concern the former VL interface and the H3 loop. The H3 loop is orientated more towards the other two hypervariable loops in VH-P8 than it is in VH domains of Fv fragments (Figure 5). It also lacks the salt bridge between residues Arg94 and Asp101, which is typical for VL-associated VH domains (Figure 6). The different H3 orientation in VH-P8 and the D1.3-VH is not surprising as VH residues 94 and 97 to 100 are in direct contact with the VL in the D1.3-Fv. As there are no such interactions possible for the H3 in the isolated VH-P8, a D1.3-VH-like conformation is unlikely. Within the former VL interface local changes in b-strand C and the b-bulge of strand G reversed the orientation of the side-chains from three residues (Val37, Arg38 and Trp103) in VH-P8 (Figure 7). The hydrophobic side-chains of residues 37 and 103 are almost completely buried within VH-P8, while in Fv fragments most of their surface area is in contact

with the VL. In contrast, the charged Arg38 side-chain is buried within the VH of an Fv fragment while it is exposed in VH-P8. The new conformations of these residues and the camelising mutations in the VL interface of VH-P8 (G44E, L45R and W47I) transform a large hydrophobic patch in the surface of a single VH domain into a partly hydrophilic one (Figure 8). This proved essential for the use of single VH domains as small recognition units. Non-specific binding, probably through the former VL interface, was found to interfere with efficient antigen selection of ligandspecific VH domains from phage-displayed libraries (Davies & Riechmann, 1995a). It cannot be excluded that the detergent Chaps present during the NMR analysis had some effect on the VH-P8 structure described here. Neither a bound form of Chaps nor NOEs between free Chaps and the VH were seen in the NMR spectra. Its influence can therefore not be assessed at present. However, the structural differences between VH-P8 and non-camelised VH domains all contribute to the hydrophilicity of the former VL

966

Solution Structure of a Camelised VH Domain

Figure 8. Grasp (Nicholls et al., 1991) built surface plots of the best VH-P8 structure and the Pot-VH in the same orientation as in Figure 7 looking onto the VL interface. Side-chain surfaces of hydrophobic amino acid residues (AVLIMFPYW) are coloured in yellow, side-chain surfaces of charged residues (EDKR) in blue. All other side-chain and backbone surfaces are shown in white.

interface in VH-P8. It seems unlikely that these changes are mainly due to the presence of Chaps, which would probably interact better with a more hydrophobic surface. Based on the structure of VH-P8 alone it is impossible to decide whether the conformational differences in the VL interface between VH-P8 and VL-associated VH domains are due to the absence of the VL, the introduction of the camelising mutations, the presence of the detergent Chaps or a combination of these factors. However, thermostability analysis of some VH mutants showed that the camelising mutations, but not the absence of the VL, are sufficient to induce at least one of the observed structural changes. Although the absence of the VL is likely to affect the structure of the H3 loop in general, the salt bridge between the side-chains of Arg94 and Asp101 at the bottom of the H3 hairpin is most likely present in the isolated, non-camelised wild-type of VH-P8 (i.e. VH-Ox13 without the camelising mutations G44E, L45R and W47I) but not in VH-P8 itself (Figure 6). When Arg94 was mutated to an alanine residue in VH-Ox13 to abolish the salt bridge, the melting point of the resulting VH was reduced by 3.5°C corresponding to a loss of stabilisation energy for folding by about 0.9 kcal (Davies & Riechmann, 1996). This amount is typical for the loss of a surface salt bridge (Horovitz et al., 1990). In the case of VH-P8 the same R94A mutation did not alter the melting point (Davies & Riechmann, 1996). The observed effects on the structure of the single VH domain also have implications for the in vitro

design of immunoglobulin-based recognition units. Notably the altered conformation of the H3 loop makes it unlikely that antigen interactions of a VH in the context of an Fv fragment are unaffected by the removal of the VL partner and the introduction of camelising mutations in the former VL interface. Thus both VH-P8 and its non-camelised wild-type VH-Ox13 no longer bind the phenyloxazolone ligand of the parent Fv (Davies & Riechmann, 1994). Similarly in cases of other isolated VH domains the conservation of antigen specificity exhibited by the parent Fv is usually not observed. The three mutations in the VL interface of VH-P8 were designed to mimic camel VH domains, which occur naturally without a VL partner. It is therefore interesting to speculate if structural features observed in VH-P8 are likely to be present in single camel VH domains. The VH-P8 residues Trp103 and Arg38 are identical and highly conserved in camel VH domains (Muyldermans et al., 1994). To reduce the hydrophobicity of the former VL interface, it would make sense if these residues had also adopted their reversed VH-P8 conformations in camel VH domains. Residue 37 is a highly conserved valine in human VH3 domains (89 of 92 sequences; Kabat et al., 1991). In contrast, 16 out of 17 published camel sequences have a phenylalanine (12) or a tyrosine (4) residue at this position. Based on the orientation of residue 37 in VH structures as part of Fv fragments the significance of this mutation was difficult to interpret (Muyldermans et al., 1994) because an aromatic side-chain in the same

967

Solution Structure of a Camelised VH Domain

orientation would increase the surface hydrophobicity of an isolated VH. However, according to the conformation of Val37 in VH-P8, the Phe37 side-chain would be buried within a camel VH and might even stabilise the global structure as part of its hydrophobic core. This speculation is supported by thermodenaturation experiments of the V37F mutants of VH-P8 and other camelised VHs, which indicate a stabilising effect (Davies & Riechmann, 1995b, 1996). In the H3 loop of VH-P8 the Arg94/Asp101 salt bridge is not formed. Residue 94, which is conserved as either arginine (31 of 95) or lysine (28 of 95) in human VH3 domains (Kabat et al., 1991), is only in two of 17 camel VH domains a positively charged amino acid residue (lysine), while in 14 of the 17, residue 94 is an alanine (Muyldermans et al., 1994). The absence of the Arg94/Asp101 salt bridge in the VH-P8 structure indicates why there is no need for conservation of a positively charged residue 94 in camel VH domains.

Materials and Methods Protein preparation The design and bacterial expression of VH-P8 has been described previously (Davies & Riechmann, 1994). The VH was isotopically labelled, purified and prepared for the NMR analysis as reported before (Riechmann & Davies, 1995). NMR spectroscopy NMR experiments were performed at 303 K on a Bruker AMX-500 spectrometer using samples in 98% H2 O/8% 2H2 O or 99.9% 2H2 O. The HBHA(CO)NH experiment (Grzesiek & Bax, 1993) was acquired with 32 complex points for a spectral width of 1572 Hz in the 15N dimension, 250 complex points for a spectral width of 4500 Hz in the indirect 1H dimension and 1024 complex points for a spectral width of 7042 Hz in the acquisition dimension using 32 scans per increment. The CBCA(CO)NH experiment (Grzesiek & Bax, 1993) was acquired as above except for the recording of 116 complex points in the 13C dimension (carrier frequency 50.7 ppm) in place of the indirect 1H dimension. In the HCCH-COSY experiment (Ikura et al., 1991) 180 real points were recorded for a spectral width of 3500 Hz in the indirect 1H dimension (carrier frequency 3.0 ppm), 60 complex points for a spectral width of 4530 Hz in the 13C dimension (carrier frequency 56.4 ppm) and 1024 complex points for 6430 Hz in the acquisition dimension (32 scans per increment). The HCCH-TOCSY experiment (Bax et al., 1990) was acquired with 180 real points in the indirect 1H dimension (spectral width 3300 Hz, carrier frequency 3.0 ppm), 112 complex points in the 13C dimension (spectral width 4854 Hz, carrier frequency 29.2 ppm) and 1024 complex points for acquisition (spectral width 6024 Hz, 32 scans). Three 13C-1H NOESY-HMQC experiments (Ikura et al., 1990) were recorded with 256 real points in the indirect 1H dimension (spectral width 5500 Hz), 64 complex points (32 in the case of the NOESY for the aromatic region) in the 13C dimension (spectral width 4673 Hz, 4673 Hz or 4000 Hz; carrier frequency 56.4 ppm, 27.9 ppm or

126 ppm, respectively) and 1024 complex points for acquisition (spectral width 6024 Hz, 16 scans). A CT-HSQC 2D experiment (Vuister & Bax, 1992) was acquired with the 15N/13C sample in 2H2 O using a constant time delay of 16 ms for observation of the aromatic side-chain signals ( 13C carrier frequency 120 ppm). Quadrature detection was achieved using time proportional phase incrementation (TPPI; Marion & Wu¨thrich, 1983) in the indirect 1H dimensions and TPPI/States (States et al., 1982) for the heteronuclear dimensions. If not stated otherwise the 1H carrier frequency was set to that of water (4.75 ppm). Other experimental details and data analysis were as described previously (Riechmann & Davies, 1995). Constraints The NOE constraints used for the X-PLOR calculations were derived from a 3D 15N-1H NOESY-HMQC experiment and the three 3D 13C-1H NOESY-HMQC experiments (tm = 100 ms in all cases). NOE crosspeaks were volume integrated and grouped into five groups ˚ (16% of all NOEs), with upper distance bounds of 6.0 A ˚ (52%), 4.0 A ˚ (19%), 3.0 A ˚ (8%) and 2.5 A ˚ (4%). These 5.0 A bounds were chosen according to the distances of backbone-backbone NOEs within the identified b-sheet regions or according to the distances of intra-residue aromatic ring NOEs with established standard distances (Wu¨thrich, 1986). Lower bounds were set to zero. No stereospecific assignments were made. A disulphide bond constraint was introduced for residues 22 and 92. Based on slowly exchanging backbone amide protons within the b-strands, 33 hydrogen bonds had been identified previously (Riechmann & Davies, 1995; Figure 2). The following distance constraints were included ˚ E H-O E 2.3 A ˚ ; 2.5 A ˚ E N-O E 3.3 A ˚. for these; 0.0 A Structure calculation Structures were calculated from initial coordinates with randomised torsion angles using a modified simulated annealing protocol within X-PLOR, version 3.1 (Bru¨nger, 1992; Nilges et al., 1988). In the calculations r−6 distance averaging was employed. For NOEs of methyl groups the upper bound for constraints was increased by 31/6 and for NOEs of degenerate aromatic or CH2 protons the upper bound was increased by 21/6. A total of 60 structures were calculated. The constraints list was corrected during different stages of the structure calculation based on the emerging average structure. Structures were superposed and the atomic mean square deviations from the mean were calculated using the program Clusterpose (Diamond, 1992, 1995). Angular order parameters were calculated as described (Hyberts et al., 1987). Structures were further analysed and visualised using Grasp (Nicholls et al., 1991), InsightII (Biosym), O (Jones et al., 1991) and Molscript (Kraulis, 1991). The coordinates of the 20 energetically best structures and a list of the distance constraints used in the X-PLOR calculations were submitted to the Brookhaven Protein Data Bank (1VHP).

Acknowledgements I thank Fred Allain and David Neuhaus for advice during the structure calculations.

968

References Alzari, P. M., Lascombe, M.-B. & Poljak, R. J. (1988). Three-dimensional structure of antibodies. Annu. Rev. Immunol. 6, 555–580. Bax, A., Clore, G. M. & Gronenborn, A. M. (1990). 1H-1H correlation via isotropic mixing of 13C magnetization, a new three-dimensional approach for assigning 1H and 13C spectra of 13C-enriched proteins. J. Magn. Reson. 88, 425–431. Bhat, T. N., Bentley, G. A., Boulot, G., Green, M. I., Tello, D., Dall’acqua, W., Souchon, H., Schwartz, F. P., Mariuzza, R. A. & Poljak, R. (1994). Bound water molecules and conformational stabilization help mediate an antigen-antibody association. Proc. Natl Acad. Sci. USA, 91, 1089–1093. Bru¨nger, A. (1992). X-PLOR Version 3.1: A System for Crystallography and NMR. Yale University, New Haven. 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. Chothia, C., Lesk, A. M., Gherardi, E., Tomlinson, I. M., Walter, G., Marks, J. D., Llewelyn, M. B. & Winter, G. (1992). Structural repertoires of the human VH segments. J. Mol. Biol. 227, 799–817. Constantine, K. L., Friedrichs, M. S., Metzler, W. J., Hensley, P. & Mueller, L. (1994). Solution structure of an isolated antibody VL domain. J. Mol. Biol. 236, 310–327. Davies, J. & Riechmann, L. (1994). Camelising human antibody fragments: NMR studies on VH domains. FEBS Letters, 339, 285–290. Davies, J. & Riechmhann, L. (1995a). Antibody VH domains as small recognition units. Bio/Technology, 13, 475–479. Davies, J. & Riechmann, L. (1995b). An antibody VH domain with a lox-Cre site integrated into its coding region: bacterial recombination within a single polypeptide chain. FEBS Letters, 377, 92–96. Davies, J. & Riechmann, L. (1996). Single antibody domains as small recognition units: design and in vitro antigen selection of camelised, human VH domains with improved protein stability. Protein Eng. 9. In the press. Diamond, R. (1992). On the multiple simultaneous superposition of molecular structures by rigid body transformation. Protein Sci. 1, 1279–1287. Diamond, R. (1995). Coordinate-based cluster-analysis. Acta Crystallog. sect. D, 51, 127–135. Fan, Z.-C., Shan, L., Guddat, L. W., He, X.-M. & Gray, W. R. (1992). Three-dimensional structure of an Fv from a human IgM immunoglobulin. J. Mol. Biol. 228, 188–207. Griffiths, A. D., Williams, S. C., Hartley, O., Tomlinson, I. M., Waterhouse, P., Crosby, W. L., Konterman, R. E., Jones, P. T., Low, N. M., Allison, T. J., Prospero, T. D., Hoogenboom, H., Nissim, A., Cox, J. P. L., Harrison, J. L., Zaccolo, M., Gheradi, E. & Winter, G. (1994). Isolation of high affinity human antibodies directly from large synthetic repertoires. EMBO J. 13, 3245–3260. Grzesiek, S. & Bax, A. (1993). Amino acid type determination in the sequential assignment procedure of uniformly 13C/15N-enriched proteins. J. Biomol. NMR, 3, 185–204. Hamers-Casterman, C., Atarhouch, T., Muyldermans, S., Robinson, G., Hamers, C., Bajyana Songa, E.,

Solution Structure of a Camelised VH Domain

Bendahman, N. & Hamers, R. (1993). Naturally occurring antibodies devoid of light chains. Nature, 363, 446–448. Horovitz, A., Serrano, L., Avron, B., Bycroft, M. & Fersht, A. R. (1990). Strength and co-operativity of contributions of surface salt bridges to protein stability. J. Mol. Biol. 216, 1031–1044. Hyberts, S. G., Ma¨rki, W. & Wagner, G. (1987). Stereo assignment of side-chain protons and characterisation of torsion angles in Eglin c. Eur. J. Biochem. 164, 625–635. Ikura, M., Kay, L. E., Tschudin, R. & Bax, A. (1990). Three-dimensional NOESY-HMQC spectroscopy of a 13C-labeled protein. J. Magn. Reson. 86, 204–209. Ikura, M., Kay, L. E. & Bax, A. (1991). Improved three-dimensional 1H-13C-1H correlation spectroscopy of a 13C-labeled protein using constant-time evolution. J. Biomol. NMR, 1, 299–304. 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. Kabat, E. A., Wu, T. T., Perry, H. M., Gottesmann, K. S. & Foeller, C. (1991). Sequences of Immunological Interest, 5th edit., US Department of Health and Human Services, Washington, DC. Kraulis, P. J. (1991). Similarity of protein G and ubiquitin. J. Appl. Crystallog. 24, 946–950. Marion, D. & Wu¨thrich, K. (1983). Application of phase-sensitive two dimensional correlated spectroscopy (COSY) for measuring 1H-1H spin-spin coupling constants in proteins. Biochem. Biophys. Res. Commun. 113, 967–974. Martin, F., Toniatti, C., Salvati, A. L., Venturini, S., Ciliberto, G., Cortese, R. & Sollazzo, M. (1994). The affinity-selection of a minibody polypeptide inhibitor of human interleukin-6. EMBO J. 13, 5303–5309. Muyldermans, S., Atarhouch, T., Saldanha, J., Barbosa, J. A. R. G. & Hamers, R. (1994). Sequence and structure of VH domains from naturally occurring camel heavy chain immunoglobulins lacking light chains. Protein Eng. 7, 1129–1135. Nicholls, A., Sharp, K. A. & Honig, B. (1991). Protein folding and association: insights from the interfacial and thermodynamic properties of hydrocarbons. Proteins: Struct. Funct. Genet. 11, 281–296. Nilges, M., Gronenborn, A. M., Bru¨nger, A. T. & Clore, G. M. (1988). Determination of 3-dimensional structures of proteins by simulated annealing with interproton distance restraints—application to crambin, potato carboxypeptidase inhibitor and barley serine proteinase inhibitor-2. Protein Eng. 2, 27–38. Pessi, A., Bianchi, E., Crameri, A., Venturini, S., Tramontano, A. & Sollazzo, M. (1993). A designed metal-binding protein with a novel fold. Nature, 362, 367–369. Poljak, R. J., Amzel, L. M., Chen, B. L., Phizackerley, R. P. & Saul, F. (1975). Structural basis for the association of heavy and light chains and the relation of subgroups to the conformation of the active site of immunoglobulins. Immunogenetics, 2, 393–394. Riechmann, L. & Davies, J. (1995). Backbone assignment, secondary structure and protein A binding of an isolated, human antibody VH domain. J. Biomol. NMR, 6, 141–152. States, D. J., Haberkorn, R. A. & Ruben, D. J. (1982). A two-dimensional nuclear Overhauser experiment

Solution Structure of a Camelised VH Domain

with pure absorption phase in four quadrants. J. Magn. Reson. 48, 286–292. Utsumi, S. & Karush, F. (1964). The subunits of purified rabbit antibody. Biochemistry, 3, 1329–1338. Vuister, G. W. & Bax, A. (1992). Resolution enhancement and spectral editing of uniformly 13C-enriched proteins by homonuclear broadband 13C decoupling. J. Magn. Reson. 98, 428–435. Ward, E. S., Gu¨ssow, D., Griffiths, A. D., Jones, P. T. & Winter, G. (1989). Binding activities of a repertoire of single immunoglobulin variable domains secreted from Escherichia coli. Nature, 341, 544–546. Winter, G., Griffiths, A. D., Hawkins, R. E. & Hoogenboom, H. R. (1994). Making antibodies by phage display technology. Annu. Rev. Immunol. 12, 433–455. Wu¨thrich, K. (1986). NMR of Proteins and Nucleic Acids, John Wiley & Sons, New York.

969 Edited by P. E. Wright (Received 2 February 1996; received in revised form 1 April 1996; accepted 18 April 1996)

Supplementary material for this paper (comprising one table of 1H, 15N and 13C chemical shifts for VH-P8 at 303 K and pH 6.2) is available from JMB On line.