Conformational Heterogeneity in Antibody-Protein Antigen Recognition

Background: Antibodies are essential components of the immune system which recognize specific antigens with high affinity. Results: Protein antigen binding sites on antibodies show conformational exchange on a millisecond to second timescale. Conclusion: Conformational heterogeneity at high affinity protein-protein interaction sites may be common and facilitate efficient protein complex formation. Significance: High affinity protein-protein interactions are critical for many biological processes. Specific, high affinity protein-protein interactions lie at the heart of many essential biological processes, including the recognition of an apparently limitless range of foreign proteins by natural antibodies, which has been exploited to develop therapeutic antibodies. To mediate biological processes, high affinity protein complexes need to form on appropriate, relatively rapid timescales, which presents a challenge for the productive engagement of complexes with large and complex contact surfaces (∼600–1800 Å2). We have obtained comprehensive backbone NMR assignments for two distinct, high affinity antibody fragments (single chain variable and antigen-binding (Fab) fragments), which recognize the structurally diverse cytokines interleukin-1β (IL-1β, β-sheet) and interleukin-6 (IL-6, α-helical). NMR studies have revealed that the hearts of the antigen binding sites in both free anti-IL-1β Fab and anti-IL-6 single chain variable exist in multiple conformations, which interconvert on a timescale comparable with the rates of antibody-antigen complex formation. In addition, we have identified a conserved antigen binding-induced change in the orientation of the two variable domains. The observed conformational heterogeneity and slow dynamics at protein antigen binding sites appears to be a conserved feature of many high affinity protein-protein interfaces structurally characterized by NMR, suggesting an essential role in protein complex formation. We propose that this behavior may reflect a soft capture, protein-protein docking mechanism, facilitating formation of high affinity protein complexes on a timescale consistent with biological processes.

Specific, high affinity protein-protein interactions lie at the heart of many essential biological processes, including the recognition of an apparently limitless range of foreign proteins by natural antibodies, which has been exploited to develop therapeutic antibodies. To mediate biological processes, high affinity protein complexes need to form on appropriate, relatively rapid timescales, which presents a challenge for the productive engagement of complexes with large and complex contact surfaces (ϳ600 -1800 Å 2 ). We have obtained comprehensive backbone NMR assignments for two distinct, high affinity antibody fragments (single chain variable and antigen-binding (Fab) fragments), which recognize the structurally diverse cytokines interleukin-1␤ (IL-1␤, ␤-sheet) and interleukin-6 (IL-6, ␣-helical). NMR studies have revealed that the hearts of the antigen binding sites in both free anti-IL-1␤ Fab and anti-IL-6 single chain variable exist in multiple conformations, which interconvert on a timescale comparable with the rates of antibody-antigen complex formation. In addition, we have identified a conserved antigen binding-induced change in the orientation of the two variable domains. The observed conformational heterogeneity and slow dynamics at protein antigen binding sites appears to be a conserved feature of many high affinity protein-protein interfaces structurally characterized by NMR, suggesting an essential role in protein complex formation. We propose that this behavior may reflect a soft capture, protein-protein docking mechanism, facilitating formation of high affinity protein complexes on a timescale consistent with biological processes.
Antibodies are both an integral component of the adaptive immune system and a highly important class of protein therapeutic. Their highly specific and modular nature, coupled with the potential to bind to a huge range of target molecules, makes them one of the most important types of therapeutic available today. Since monoclonal antibodies were first produced in the 1970s (1) there have been significant advances in the ability to engineer antibodies. For example, the successful humanization of mouse antibodies, together with the ability to now produce completely human monoclonal antibodies, has allowed the development of highly specific and potent therapeutics (2). Due to the modular nature of these proteins a number of antigenbinding derivatives have been developed as potential therapeutics, including the fragment antigen-binding (Fab) 6 and single chain variable fragment (scFv). In addition, a variety of more elaborate antibody-based therapeutics is under evaluation, which provide further functionality required for specific applications (3).
To date, a number of Fab fragments have been licensed as therapeutics, with several other antibody fragments in various stages of clinical development. There are currently Ͼ30 United States Food and Drug Administration approved, antibody-derived therapeutics available, with at least nine of these generating revenues of Ͼ1 billion dollars per year (4). Antibody-based therapeutics now constitute a large proportion of global therapeutic sales, and with development times comparable with small molecule drugs and higher success rates, will remain a major part of the pharmaceutical industry.
The high specificity and affinity of antigen recognition lie at the heart of both therapeutic antibody development and the adaptive immune response. This is primarily mediated by the hypervariable loops formed by the complementarity determining regions (CDRs), however, the conformational properties of the CDR loops, together with potential conformational changes induced by antigen binding, remain poorly characterized. The lack of information relating to the antigen binding site also includes very limited knowledge of potential conformational heterogeneity and dynamics on a slow or fast timescale (seconds to picoseconds).
Almost all of the high resolution structures currently available for antibodies have been determined by x-ray crystallography, which limits the ability to gain insights into functionally important conformational dynamics. In addition, there are a relatively limited number of cases where structures have been obtained for both the free antibody and the complex formed with its target protein (5). Consequently, no clear picture has emerged concerning the potential conformational changes induced by antigen binding, and the structural nature of the CDR loops in both the presence and absence of antigen. NMR spectroscopy-based structural biology has now advanced to the point at which detailed structural information can be obtained for proteins and complexes of at least 100 kDa in size (6). This enables the use of NMR spectroscopy to probe the conformational features and properties of both free and antigen-bound antibody fragments, such as scFvs and Fabs.
In this paper, we report the acquisition and analysis of high quality three-dimensional 15 N/ 1 H and 15 N/ 13 C/ 1 H NMR spectra for both free and antigen-bound scFvs and Fabs. This has allowed the determination of comprehensive sequence-specific backbone resonance assignments for the antibody fragments, which have revealed conserved structural changes within distinct antibody fragments upon binding to structurally diverse protein antigens. Furthermore, the NMR data show that the core of the antigen binding sites, in particular the CDR3 loops, exist in multiple conformational states that interconvert on a relatively slow timescale (milliseconds to seconds). Interestingly, this behavior appears to be a conserved feature of many high affinity protein-protein interaction sites (7)(8)(9)(10)(11)(12)(13), suggesting a key role in the formation of tight protein complexes.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The humanized anti-IL-6 scFv and anti-IL-1␤ Fab antibody fragments, human IL-1␤ and human IL-6 were all produced as soluble proteins in Escherichia coli-based expression systems. Tandem pTTod-based expression vectors encoding the heavy and light chain of the anti-IL-6 Fab and anti-IL-1␤ Fab were provided by UCB. To produce the anti-IL-6 scFv, the parent Fab vector was modified by restriction digestion and PCR to produce a single expressible gene encoding the variable light domain (V L ), (Gly 4 Ser) 4 linker, variable heavy domain (V H ) and His 6 tag, as described previously for an anti-IL-1␤ scFv (14). pET-21-based vectors encoding IL-1␤ or IL-6 with an N-terminal His 6 tag containing a tobacco etch virus protease site in the linker were also provided by UCB (14,15).
The Fab and scFv expression vectors were transformed into W3110 E. coli-competent cells for the production of 15 N, 15 N/ 2 H, 15 N/ 13 C/ 2 H, or unlabeled protein. Cells containing the anti-IL-1␤ Fab vector were grown in either Luria-Bertani (LB)-rich medium, or a modified Spizizen's minimal medium (16,17) and expression induced at 37°C (14). Uniformly 15 N labeled and unlabeled anti-IL-6 scFv were similarly expressed; however, 15 N/ 2 H-and 15 N/ 13 C/ 2 H-labeled proteins were produced using a small volume, high cell density expression procedure (18) to obtain high yields of deuterated protein from 50-ml cultures.
IL-1␤ was expressed in Tuner (DE3) pLysS E. coli cells at 25°C, as described previously (14). For the production of IL-6, the expression vector was transformed into Origami B DE3 pLysS E. coli cells, which were grown in either LB, modified Spizizen's minimal medium, or high cell density minimal medium (18), containing 100 g/ml carbenicillin, 12.5 g/ml tetracycline, 15 g/ml kanamycin, and 34 g/ml chloramphenicol. Cultures were grown to an A 600 of 0.3-0.6 at 37°C, cooled to 17°C, IL-6 expression induced by the addition of 100 M IPTG, and cells harvested after approximately a 16-h induction.
The anti-IL-6 scFv, IL-1␤, and IL-6 were purified as described previously (14,15) using a combination of nickel affinity (Ni-nitrilotriacetic acid) and size exclusion chromatography (Superdex 75). The anti-IL-1␤ Fab was also purified as reported previously (14) using a combination of protein G affinity (protein G-Sepharose) and size exclusion chromatography (Superdex 75). During the purification of deuterated samples of the anti-IL-6 scFv, the protein was fully denatured and refolded after nickel affinity chromatography to ensure complete exchange of any buried backbone amide groups. The scFv was denatured by the addition of 5 M guanidine hydrochloride and refolded by dialysis into a 100 mM sodium chloride and 25 mM Tris buffer at pH 7.5. Purified samples of the deuterated anti-IL-1␤ Fab were similarly denatured (6 M guanidine hydrochloride) and refolded by dialysis (100 mM sodium chloride and 25 mM sodium phosphate buffer at pH 6.5) to obtain complete exchange of backbone amide groups.
Surface Plasmon Resonance-On and off rate constants for antigen binding to immobilized antibody fragments were determined by surface plasmon resonance experiments carried out essentially as described previously (15).
A series of two-and three-dimensional TROSY-based spectra (19) were acquired to obtain sequence specific backbone resonance assignments for the anti-IL-6 scFv, anti-IL-1␤ Fab, and antigen-bound complexes, including 15 N/ 1 H TROSY (19), 15 N/ 13 C/ 1 H HNCO, HNCA, HN(CO)CA (20), HNCACB (21) and HN(CO)CACB (22). Typical acquisition times for the twodimensional experiments were 30 ms in F 1 ( 15 N) and 60 ms in F 2 ( 1 H), with a total experimental time of 30 to 90 min. Typical acquisition times for the triple-resonance three-dimensional experiments were 6 -9 ms in indirect 13 C apart from HNCO which were 18 -22 ms, 20 -25 ms in indirect 15 N and 60 -90 ms in direct 1 H, with an overall experimental time of 60 to 96 h. Long range backbone amide NOE data were also obtained from 15 N/ 1 H NOESY-TROSY and NOESY-HSQC spectra (23), which were recorded with acquisition times of 12-15 ms in indirect 1 H, 9 -12 ms in indirect 15 N, and 60 -90 ms in direct 1 H, with an NOE mixing time of 500 ms for the anti-IL-1␤ Fab, 450 ms for the anti-IL-6 scFv-IL6 complex, and 700 ms for the free scFv. Total acquisition times for the NOE experiments were 87-92 h. A series of double-and triple-resonance spectra were also recorded to obtain essentially complete sequencespecific backbone resonance assignments for IL-6 and IL-1␤, as described previously (14,15).
Partially aligned samples of the free and IL-1␤-bound anti-IL-1␤ Fab were produced by the addition of Pf1 phage (Asla Biotech) at 3-4 mg/ml. Backbone amide residual dipolar coupling values were derived from measured differences in the 15 N/ 1 H scalar couplings seen in spectra acquired from isotropic and partially aligned samples, with the separation between equivalent peaks in 15 N/ 1 H HSQC and temperature-compensated TROSY spectra used to determine the couplings (24). Individual two-dimensional spectra were collected for 5-12 h, with acquisition times of 60 ms in direct 1 H and 50 ms in indirect 15 N.
All NMR data were processed using Topspin 2.1 (Bruker Biospin) and analyzed using the SPARKY package (Goddard and Kneller, SPARKY 3, University of California, San Francisco).
Chemical Shift-based Mapping of Interaction Sites-Backbone chemical shift ( 15 N, 13 C, and 1 H) comparisons between free and bound antibody fragments and antigens were made using the minimal shift method (7-9) and by direct comparison of assigned backbone signals for the free and bound proteins where available. For the anti-IL-1␤ Fab, peaks from the comprehensively assigned HNCO spectrum of the free protein were compared with the unassigned HNCO spectrum of the Fab⅐IL-1␤ complex. In the case of the anti-IL-6 scFv, this approach was extended to combine an assigned free versus bound comparison with minimal shift analysis for residues that were not assigned in both states, which provides the fullest overall picture of the backbone chemical shift changes induced by complex formation. Minimal shift analysis was used to map the affects of antibody binding on IL-1␤ and IL-6.
Modeling of the scFv and Fab Structures-Homology models of the anti-IL-1␤ Fab and anti-IL-6 scFv were produced as described previously, with the residual dipolar coupling refined model of the anti-IL-1␤ scFv (PDB accession code 2KH2) used as a template for the anti-IL-6 scFv (14). The homology model obtained for the anti-IL-1␤ Fab was further refined using HADDOCK (25), with a combination of backbone amide residual dipolar coupling data, chemical shift-derived backbone dihedral angles, and backbone H N -H N NOEs included as experimental restraints (14). Analysis of the scFv and Fab structural models produced, including the mapping of antigen bindinginduced chemical shift changes, was carried out using the PyMOL molecular graphics package.

RESULTS AND DISCUSSION
NMR Spectroscopy of Antigen-binding Antibody Fragments-NMR samples of the anti-IL-6 scFv and anti-IL-1␤ Fab were found to be stable for many days at 40 and 45°C, respectively, which allowed the acquisition of a range of high quality twodimensional and three-dimensional NMR spectra, as illustrated in Figs. 1 and 2. The spectra obtained show excellent dispersion of backbone signals ( 15 N, 13 C, and 1 H) and good signal to noise ratios. The line widths observed for resonances in highly deuterated samples of the antibody fragments allowed the recording of an extensive set of triple resonance spectra for both the free and antigen-bound anti-IL-6 scFv and for the free anti-IL-1␤ Fab. The correlations detected between backbone signals in these spectra, together with backbone amide NOEs identified in three-dimensional 15 N-edited NOESY experiments, enabled the determination of comprehensive sequencespecific backbone resonance assignments (H N , N, C ␣ , C ␤ , and CЈ) for both the anti-IL-6 scFv (free and bound to IL-6) and the free anti-IL-1␤ Fab using well established procedures (7,8,14,26,27).
For the free anti-IL-1␤ Fab, 15 N and 1 H resonances from 338 of the 419 assignable backbone amide groups were identified (81% excluding prolines (23) and the two N-terminal residues). Backbone amide signals were not assigned for Ile 2 , Gln 6 -Ser 7 , Ser 9 , Asn 28 , Trp 35 (Fig. 3). More than 94% of the backbone amide-associated spin systems detected for the Fab were assigned to specific residues, with the assignment of only approximately 20 backbone amide signals remaining unclear. We have previously reported essentially complete backbone resonance assignments for the equivalent scFv bound to IL-1␤ (14). Spectra acquired for the scFv and Fab bound to IL-1␤ show strikingly similar shifts for backbone signals (H N , N, and CЈ) arising from the shared antigen binding variable domains (V H and V L ), which clearly indicates an essentially identical structure for the two variable domains and the interface formed with IL-1␤. This is also reflected in the near perfect overlay of 15 N/ 1 H HSQC spectra acquired for IL-1␤ bound to the scFv and Fab (14).
Analysis of HNCO spectra collected for the free and antigenbound anti-IL-1␤ Fab revealed the appearance of 17 new peaks upon binding to IL-1␤. Comparison with the fully assigned spectrum available for the equivalent scFv bound to IL-1␤ (14) clearly showed that these new peaks arose from the backbone amides of residues in the antigen binding CDR loops, in particular, Asn 31 (Figs. 3 and 4). Overall, by reference to the assigned spectra of the free anti-IL-1␤ Fab and equivalent scFv bound to IL-1␤, 355 of the 419 (85%) assignable backbone amide signals ( 15 N and 1 H) were identified for the IL-1␤-bound Fab.
The completeness of the backbone assignments obtained for the antigen binding variable domains of the anti-IL-6 scFv and anti-IL-1␤ Fab are shown in Fig. 3, which highlights many similarities. The overall assignment level for both proteins is high; however, the assignments are slightly more complete for the anti-IL-6 scFv than for the anti-IL-1␤ Fab, with better coverage for both the CDR loops and the framework residues. Interestingly, for both the IL-6-and IL-1␤-targeted antibodies the assignment of the CDRs is incomplete in the absence of bound  antigen. This lack of assignments for the CDR loops reflects the absence of backbone amide signals from many residues present in these regions, in particular, for CDR3 of both the V L and V H domains, which indicates the presence of a number of discreet structural states interconverting on a relatively slow timescale (milliseconds to seconds) (28). The striking similarity in the behavior of the CDRs in both the anti-IL-1␤ Fab and anti-IL-6 scFv suggests that conformational heterogeneity is a common feature of antibody CDR loops, particularly toward the core of the antigen binding surface.
Comparison of the extent of the backbone assignments obtained for the antigen-bound forms of the anti-IL-6 scFv and anti-IL-1␤ Fab reveals a marked difference between the two antibody/antigen systems (Figs. 3 and 4). In both cases, a number of backbone amide signals were recovered upon antigen binding, which were assigned to CDR loop residues and are highlighted on the structures shown in Fig. 4. This appearance of missing backbone amide NMR signals suggests a stabilization of the CDR loops through interactions with the target proteins. However, whereas the majority of the CDR signals are visible for the anti-IL-1␤ Fab bound to IL-1␤, only a relatively small number of new signals are seen for the anti-IL-6 scFv bound to IL-6, with a large portion of CDR3 in both V H and V L remaining unobservable. This difference is most likely due to the distinct structural nature of the two antigens, with IL-1␤ being structurally quite stable compared with a more dynamic IL-6 (14, 15). Interestingly, in spectra of the anti-IL-6 scFv

. Comparison of the extent of the backbone NMR assignments obtained for the variable domains (V L and V H ) of the free and antigen-bound forms of the anti-IL-6 scFv (A) and anti IL-1␤ Fab (B).
Residues for which backbone resonance assignments were made are highlighted in gray for both the free and antigen-bound antibody fragments. The positions of elements of regular secondary structure and CDR loops are also indicated, with ␣-helices shown as red bars, ␤-sheets as green arrows, and CDRs as green boxes. The overall assignment level for the free scFv is slightly higher than for the Fab, but for both antibodies backbone signals from residues in CDR3 of V L and V H remain largely unassigned due to the absence of detectable backbone amide resonances.

90˚90Å nti-IL-1β Fab
Anti-IL-6 scFv . Localization of antibody residues with nonobservable backbone amide NMR signals in the free state and appearance upon antigen binding. Antibody residues with backbone amide resonances that were observable only after antigen binding (green) or were missing in both the free and bound states (yellow) are highlighted on the structural models of the anti-IL-1␤ Fab and anti-IL-6 scFv. Backbone amide signals were not observed for the heart of the antigen binding site in both free antibody fragments. The majority of these signals appeared for the anti-IL-1␤ Fab on binding to IL-1␤, but remained unobservable for the anti-IL-6 scFv bound to IL-6.
bound to IL-6 a number of residues in V H CDR2 (Ser 179 , Thr 181 , Ser 183 , Gly 184 , Thr 187 , Tyr 188 , Arg 190 , Ser 192 , and Val 193 ) and either in, or adjacent to V H CDR3 (Leu 232 , Phe 235 , Tyr 237 , Gln 240 , and Gly 241 ), show significantly broader amide proton signals (Ͼ30 Hz) than observed for the majority of the protein (overall average of 26.3 Ϯ 4.2 Hz), as shown in Fig. 5. This is consistent with some remaining slow conformational exchange behavior at the antigen binding site of the V H domain, which is presumably possible due to the relatively flexible nature of the IL-6 structure (15).
Mapping of Antigen Binding Sites-The effects of antigen binding on the backbone signals (H N , N, and CЈ) of the anti-IL-6 scFv and anti-IL-1␤ Fab were determined by a combination of minimal and actual shift analyses of HNCO spectra obtained for the free and bound antibody fragments. The results from the chemical shift analyses were mapped on to a homology model of the anti-IL-6 scFv (Fig. 6) and a refined homology model of the anti-IL-1␤ Fab (Fig. 7). The locations of residues affected by antigen binding further highlight the absence of NMR signals from a significant number of CDR residues, particularly in the CDR3s of both the V H and V L domains (Figs. 6 & 7). The binding-induced chemical shift changes also clearly show that the visible CDR signals are highly perturbed by antigen binding. An antigen binding site can be clearly identified on both the anti-IL-6 scFv and on the anti-IL-1␤ Fab, which is further confirmed by NMR data obtained for the corresponding IL-1␤ scFv (14).
Interestingly, for both antibody fragments further antigeninduced structural changes were revealed by chemical shift changes that lie beyond the CDR loops. These include changes in the framework residues supporting the CDR loops and in residues at the interface between the V L and V H domains. There are striking similarities in the regions affected for both antibody fragments despite binding to structurally diverse antigens, which strongly suggests that conserved conforma-  The combined H N , N, and CЈ chemical shift changes for residues between free and IL-6-bound anti-IL-6 scFv were assigned a color depending on their magnitude, with shifts of Ͻ0.002 ppm colored white, Ͼ0.03 ppm colored red, and those in between colored on a linear gradient from white to red. Similarly, for residues with backbone signals that remained unassigned in the free or bound state antigen-induced changes were determined by the minimal shift approach, with shifts of Ͻ0.002 ppm colored white, Ͼ0.03 ppm colored blue, and those in between colored on a linear gradient from white to blue. Residues for which no chemical shift data could be obtained are colored yellow. The chemical shift changes observed are mapped onto both a ribbon representation of the scFv and a surface view of the antigen binding site. The most substantial changes are seen in the CDR loops and the interface between the two variable domains.
tional changes beyond the CDR loops are important for effective antigen binding. In contrast to the variable domains, the constant domains present in the anti-IL-1␤ Fab show almost no signs of chemical shift perturbation except at the V L /C L interface. This most likely reflects movements of the V L domain relative to the rest of the protein rather than structural changes within the constant domain.
Mapping of Antibody Binding Sites-Nearly complete backbone and side chain assignments have been reported previously for IL-1␤ and IL-6 (29, 30), but under solution conditions different from those required for this study. It proved relatively straightforward to obtain essentially complete backbone resonance assignments for both cytokines under the new conditions, with backbone amide signals identified for all nonproline residues apart from Glu 51 , Lys 54 , and Glu 69 in IL-6 (98%), and Ala 1 , Arg 4 , Asn 53 , and Glu 64 -Asn 66 in IL-1␤ (96%). The backbone assignments obtained for IL-6 and IL-1␤ allowed further assessment of the nature of the interaction between the antibody fragments and their target antigens (14,15), with minimal shift analysis providing information on the location of the antibody binding surfaces and of any structural changes induced in the bound proteins.
IL-1␤ shows a discreet antibody binding site, with substantial and highly localized chemical shift perturbations (Fig. 8A), con-sistent with a stable structure that is not significantly affected by antibody binding. In stark contrast, IL-6 shows significant perturbation of NMR signals throughout the protein as a result of scFv binding (Fig. 8B). This probably reflects the somewhat plastic and dynamic nature of the IL-6 structure, which has been described previously (15,29), and may be related to the biological function of IL-6 and conformational changes associated with the formation of the IL-6⅐IL-6R␣⅐gp130 signaling complex (15,31). This in turn may explain the continued lack of observable backbone amide signals for residues in both CDR3 loops of the anti-IL-6 scFv bound to IL-6, with the inherent flexibility of IL-6 allowing continued exchange between distinct structural states for the contacting CDR loops. This property of the IL-6⅐scFv complex is in contrast to the IL-1␤⅐Fab, or equivalent IL-1␤⅐scFv complex, in which the majority of the backbone amide CDR signals are seen ( Fig. 4 and Ref. 14).
The exchange between multiple conformational states on a relatively slow timescale, particularly for the CDR3s, is a conserved feature of the two distinct antibodies that have been characterized. The lack of backbone amide signals for these regions implies that a number of discreet structural states are being sampled on a millisecond to second timescale. This behavior is conserved in two antibodies with very different antigen binding sites and has recently been seen in another heavy  chain antibody (32), which strongly suggests that conformational heterogeneity at the heart of the antigen binding site may be a critical feature to ensure effective binding of target proteins. This may simply reflect the need to accommodate flexible surfaces on protein antigens, but also has the potential to play a key role in facilitating productive binding to an extensive contact surface (typically 1700 Ϯ 260 Å 2 (33)).
The slow dynamics of CDR loops at the heart of the antigen binding site and associated sampling of multiple conformations may allow antibodies to recognize and bind protein antigens when approaching at a less than optimal orientation. If this conformational heterogeneity in the CDRs allows the initiation of binding at some significant variance from the ideal orientation for interaction, it could greatly improve the chances of binding and therefore the on rate (k a ) for antibody-antigen complex formation (Fig. 9, anti-IL-1␤ k a ϭ 1.4 ϫ 10 6 M Ϫ1 s Ϫ1 , K D ϭ 80 pM and anti-IL-6 k a ϭ 2.36 ϫ 10 6 M Ϫ1 s Ϫ1 , K D 41 pM). This soft capture type mechanism would also allow initial intermolecular interactions at nonoptimal orientations, facilitating final docking of the proteins. The slow exchange between multiple conformations and relatively high affinity of interactions involving CDR3s appears well suited to the soft capture of protein antigens, perhaps assisted by movement of the variable domains relative to each other. This process could also serve to guide the target protein into the optimal orientation for interactions with residues in the apparently less dynamic CDR1 and CDR2 loops, which make important contributions to overall affinity and specificity (Fig. 10).
Conformational heterogeneity and associated slow motions appear to be a feature of many high affinity protein-protein interaction sites, involved in a diverse range of processes from receptor-mediated signaling to control of gene expression (7)(8)(9)(10)(11)(12)(13). This perhaps suggests that a soft capture type mechanism may be a key feature of many high affinity protein-protein interactions, facilitating a rate of complex formation consistent with that required by a range of biological processes.
Changes in the conformations and dynamics of the CDR3 loops induced by antigen binding appear to be accompanied by reorientations of the antibody variable domains relative to each other, with significant chemical changes seen for residues at the V H /V L interface in both the anti-IL-1␤ Fab and anti IL-6 scFv (Figs. 6 and 7). A number of previous studies have reported antigen-induced changes in the orientation of the variable domains; however, no consistent picture has emerged (34 -37). This may reflect the difficulty in detecting significant but small structural changes in proteins, which can now be reliably identified for relatively large proteins and complexes through the high sensitivity of backbone and side chain NMR chemical shifts to changes in conformation (6 -9, 14, 15, 38).

CONCLUSIONS
The work reported here describes the results of an NMRbased study of the structural characteristics of the antigen binding regions of antibodies selected for their ability to recognize target proteins with very high specificity and affinity. This has revealed conformational heterogeneity and slow dynamics at Chemical Shift Change (ppm) Residue FIGURE 8. Comparison of the effects of antibody binding to IL-1␤ and IL-6. Combined backbone minimal shift data were obtained for the assigned free versus antibody-bound proteins and mapped on to reported structures for the two cytokines (PDB ID 2I1B and 2IL6 for IL-1␤ and IL-6, respectively). A, two-dimensional minimal shift data (H N and N) were used for IL-1␤, with signals from residues shifted by Ͻ0.1 ppm shown in white, Ͼ0.4 ppm shown in red, and those between 0.1 and 0.4 ppm on a linear gradient from white to red. B, triple resonance minimal shift data (H N , N, and CЈ) were used for IL-6, with signals from residues shifted by Ͻ0.01 ppm shown in white, Ͼ0.05 ppm in red, and those between 0.01 and 0.05 ppm shown on a linear gradient from white to red. Residues for which no minimal shift data could be obtained are colored yellow for both proteins. A discreet antibody binding surface is clearly observed for IL-1␤. In contrast, for IL-6 the majority of residues show significant chemical shift changes on antibody binding, which is consistent with induced conformational changes throughout the protein.
the heart of the antigen interaction sites, together with structural changes beyond the contact surfaces induced by antigen binding. The characterization of antibody fragments against two structurally distinct proteins, with IL-6 typical of a relatively flexible ␣-helical structure and IL-1␤ representative of a fairly rigid ␤-sheet topology, has allowed the identification of antibody features and properties associated with antigen recognition that are independent of the structure of the target protein.
A number of structural features and properties have been identified that are common to both the anti-IL-1␤ Fab and anti-IL-6 scFv. Perhaps the most interesting is the finding that the two distinct binding surfaces demonstrate similar conformational heterogeneity and slow dynamics for residues within the CDR3 regions of both variable domains prior to antigen binding. This sampling of a number of distinct conformations, leading to missing backbone NMR signals, appears to be an important feature of the heart of the antigen binding site, which is likely to play a key role in the recognition and/or binding of protein antigens. This inherent flexibility, sampling conforma-tions on a millisecond to second timescale, may mediate an initial soft capture of the antigen over a range of suboptimal orientations of the two proteins, facilitating correct engagement of the antigen with other less flexible and potentially lower affinity CDR residues. This type of mechanism may help to guide initial antibody-antigen contacts down a productive binding pathway, resulting in an increased ability to bind at less than perfect orientations, which may significantly improve the rate of complex formation compared with rigid protein interaction surfaces.
Further antibody structural changes also take place following antigen binding, with an apparent change in the orientation of the two variable domains in both the anti-IL-1␤ Fab and anti-IL-6 scFv. Flexibility at the V H /V L interface may further facilitate the soft capture mechanism by allowing an additional level of adaptability. However, changes in V H /V L orientation could also be linked to processes beyond antigen recognition, such as the initiation of B cell receptor signaling, which remains poorly understood. The conformational heterogeneity and slow dynamics reported here appears to be shared by many high Response Units FIGURE 9. Binding of protein antigens to captured antibody fragments. Typical surface plasmon resonance sensorgrams are shown for the binding of IL-6 and IL-1␤ to captured anti-IL-6 scFv (A) and anti-IL-1␤ Fab (B), respectively. The responses obtained indicate very high affinity antigen binding for both antibody fragments, with a K D of 41 pM determined for the anti-IL-6 scFv and 80 pM for the anti-IL-1␤ Fab.