Three Camelid VHH Domains in Complex with Porcine Pancreatic α-Amylase

Camelids produce functional antibodies devoid of light chains and CH1 domains. The antigen-binding fragment of such heavy chain antibodies is therefore comprised in one single domain, the camelid heavy chain antibody VH (VHH). Here we report on the structures of three dromedary VHH domains in complex with porcine pancreatic α-amylase. Two VHHs bound outside the catalytic site and did not inhibit or inhibited only partially the amylase activity. The third one, AMD9, interacted with the active site crevice and was a strong amylase inhibitor (K i = 10 nm). In contrast with complexes of other proteinaceous amylase inhibitors, amylase kept its native structure. The water-accessible surface areas of VHHs covered by amylase ranged between 850 and 1150 Å2, values similar to or even larger than those observed in the complexes between proteins and classical antibodies. These values could certainly be reached because a surprisingly high extent of framework residues are involved in the interactions of VHHs with amylase. The framework residues that participate in the antigen recognition represented 25–40% of the buried surface. The inhibitory interaction of AMD9 involved mainly its complementarity-determining region (CDR) 2 loop, whereas the CDR3 loop was small and certainly did not protrude as it does in cAb-Lys3, a VHH-inhibiting lysozyme. AMD9 inhibited amylase, although it was outside the direct reach of the catalytic residues; therefore it is to be expected that inhibiting VHHs might also be elicited against proteases. These results illustrate the versatility and efficiency of VHH domains as protein binders and enzyme inhibitors and are arguments in favor of their use as drugs against diabetes.

The fundamental molecular recognition molecules of the humoral immune response are remarkably homogeneous throughout the vertebrate phylum. All immunoglobulins are multimers of heterodimeric chains where each heavy (H) 1 chain of four or five domains is linked by disulfide bridges to a light (L) chain of two domains (1). The antigen-binding part of the immunoglobulins is formed invariably by the N-terminal domains of both the H and L chains. These domains display a large sequence variation concentrated in three regions per domain, the CDRs.
However, important deviations of this conserved immunoglobulin organization have been observed. In some immunoglobulin isotypes of camelids from the old world (camels and dromedaries) or from the new world (llamas and vicunas) the L chain is missing (2). Furthermore their H chain is devoid of the CH1 domain (3,4) due to an unconventional splicing event during the mRNA maturation. The antigen-binding fragment of the heavy chain antibodies is therefore comprised in one single domain, the unique N-terminal variable domain referred to as VHH that replaces a four-domain Fab fragment in the immunoglobulin structure (5). This VHH domain is obtained after a DNA recombination between dedicated VHH germline gene segments with D and J minigenes. The dromedary VHH germline genes are quite diverse. They can be grouped into seven subfamilies (6) and contain additional hotspots for mutation that will add to the diversity of the antigen binding repertoire. Moreover the VHH domain often acquires a disulfide bond between its CDR3 and CDR1 or position 45 (5). A considerable interest in the humoral immune system of Camelidae comes from the observation that their heavy chain antibodies and the recombinant VHHs as well contain a much higher proportion of molecules that interact directly with the active site cleft of enzymes (7). From a structural viewpoint, the three-dimensional structures of VHH complexes with lysozyme, RNase, carbonic anhydrase, and two dye haptens as well as an unbound VHH have been determined (8 -14). All three dromedary anti-enzyme VHHs of known structure are derived from the same VHH germline subfamily (subfamily 2a), and only one of these inhibits the enzymatic activity of its antigen. This cAb-Lys3 inhibitor of chicken egg white lysozyme has a remarkable paratope architecture where part of its long CDR3 protrudes from the remaining antigen-binding site and inserts into the active site of the enzyme, mimicking the lysozyme natural substrate (15).
Here we report the three-dimensional structures of three complexes between porcine pancreatic ␣-amylase (PPA, Ref. 16) and camelid VHH fragments. These three binders originated from VHH germline genes of three different subfamilies, two for which no structural information is yet available. Crystal structures of complexes between PPA and carbohydrate or proteinaceous inhibitors are known at atomic resolution (17)(18)(19). We investigated inhibitors that were raised by the humoral immune response in a few weeks time and that possess an enzyme inhibiting potency similar to the natural inhibitors that co-evolved with amylase over many millions of years.

MATERIALS AND METHODS
VHH Preparation and Characterization-Periplasmic expression and immobilized metal affinity chromatography purification of the three PPA binder VHH proteins, in fusion with a C-terminal His 6 tail, was performed according to Lauwereys et al. (7). The VHH proteins were further purified by gel filtration and mixed with PPA in a 2 to 1 molar ratio, and the complexes were separated from free antibody on Superdex 75 column (Amersham Biosciences) in 50 mM Tris (pH 7.5), 100 mM NaCl. The inhibition of the enzymatic activity of PPA by the various VHHs was tested on 2-chloro-4-nitrophenyl maltotrioside or on "blue-starch" (Phadebas, Pharmacia-Upjohn) according to the protocols in Ref. 7 or as recommended by the supplier, respectively.
Crystallographic Procedures-All crystals were obtained using the hanging drop method of vapor diffusion by mixing 1 l of protein solution with 1 l of reservoir solution. A single crystal of the PPA⅐AMB7 complex was obtained in 10 -15% polyethylene glycol 20,000, 0.1-0.2 M imidazole malate (pH 8.0). A unique monoclinic crystal was obtained that could not be further reproduced (Table I). Triclinic crystals of the PPA⅐AMD9 complex were obtained in 0.8 M phosphate buffer (NaH 2 PO 4 and K 2 HPO 4 ) at pH 7.0 (Table I). Triclinic crystals of the PPA⅐AMD10 complex were produced in 32% polyethylene glycol 4000, 0.1 M sodium citrate, 0.2 M ammonium acetate (pH 5.0) ( Table I).
Data were collected at the European Synchrotron Radiation Facility (Grenoble, France) at beamline ID14-EH2 for AMB7 and AMD10 on an ADSC-Q4 detector. AMD9 was collected at beamline BM14 with an imaging plate detector. Data were integrated with DENZO (20) and reduced with SCALA (21). Collection statistics, presented in Table I, indicate a good quality for the AMD9 and AMD10 complexes, whereas the data are incomplete for AMB7 due to spot overlap.
The three structures were solved by molecular replacement with AMoRe (22). The initial solution for the complex with AMB7 was obtained using native amylase (Ref. 16, 1FJH) and the anti-RR6 VHH fragment R2 (Ref. 12, 1QDO) as search models. The three amylases were positioned readily by AMoRE, yielding a correlation coefficient of 0.53 and an R value of 37.6 at 4.0-Å resolution. The VHHs, however, were positioned manually by visual inspection of the difference maps using Turbo-Frodo (23). After rigid body and minimization refinement with CNS (24), R and R free dropped to 28.4 and 33.5%, respectively, at 2.0-Å resolution. For the AMD9 complex, the same search models were used. The amylase molecules were found by AMoRE, resulting in R and R free values of 36.8 and 39.0%, respectively, at 3.5-Å resolution. The packing consists of a dimer related by a pure 2-fold axis and a translated dimer. The complete model of the four complexes was subjected to a rigid body minimization and B-factor refinement, which lowered R and R free to values of 33.4 and 35.2%, respectively, at 1.8-Å resolution. For the PPA⅐AMD10 complex, the VHH search model was cAb-Lys3 (Ref. 8, 1MEL). The two amylase molecules and the two VHH fragments were readily located by AMoRe, leading to a correlation coefficient of 0.56 and an R value of 36.9% at 2.7-Å resolution.
The same refinement procedure was used with the three complexes using CNS (24). Rounds of minimization/B-factor refinement were alternated with model rebuilding at the display with Turbo-Frodo. The complexes with AMD9 and AMD10 exhibit an excellent geometry and good R values (Table I). In contrast, the complex with AMB7 suffers from data incompleteness and hence exhibits a geometry quality closer to a model at medium resolution. The water-accessible surfaces were calculated with DSSP (25) implemented in Turbo-Frodo (23). The radius of the water probe used was 1.5 Å. The coordinates have been deposited in the Protein Data Bank at Research Collaboratory for Structural Bioinformatics (www.rcsb.org/pdb/) as entries 1KXT, 1KXQ, and 1KXV.

Characterization of the VHH Binders
A dromedary was immunized with PPA, and the antigen binding repertoire of the heavy chain antibodies was cloned in a phage display vector. After three rounds of panning with the antigen we identified several binders (7) of which three (AMB7, AMD9, and AMD10 VHH) were selected for structural investigation. The sequences revealed that the three binders are derived from germline genes of different subfamilies (Ref. 6 and Fig. 1A). The VHH germline gene used to generate the AMB7 VHH binder is a member of subfamily 4b since it has a 16amino acid-long CDR2 and a Cys (at position 45). Its CDR3 is 19 amino acids long and contains a Cys that could form a disulfide bridge with Cys-45. The AMD9 VHH binder contains 17 amino acids in its CDR2, meaning that the VHH germline used is either of subfamily 2a or 5a. The presence of Phe-37, Gly-47, Ala-49, and Val-78 instead of Tyr, Leu, Ser, and Leu, respectively, suggests that is most likely derived from 2a, the most frequently found subfamily in dromedary. This binder is special in the sense that the CDR1 is shortened by 3 amino acids possibly due to a deletion around the palindromic nucleotide sequence (codons 29 -33 in clone cvhhp11). Since the deletion in the CDR1 has removed the Cys and since no Cys occurred in the CDR3 region of 14 amino acids in size, this structure is not stabilized by an interloop disulfide bridge. The VHH germline used for the AMD10 binder is of subfamily 3b because it has the characteristic 16-amino acid-long CDR2 and Cys-30, Tyr-37, and Phe-47. Its CDR3 is short for a VHH (13 amino acids) and contains a Cys. The affinity of the three VHHs for PPA ranged from 3.5 (AMD9) to 235 nM (AMB7) with AMD10 having an intermediate affinity of 25 nM as measured on an IAsys biosensor (Ref. 7 and Table II). The gel filtration of a stoichiometric mixture of the binders with PPA proved that binding occurred at a 1:1 molar ratio. These binders were also chosen because they inhibit the PPA to different extents. Only AMD9 VHH had the capacity to inhibit the hydrolysis of the small organic 2-chloro-4-nitrophenyl maltotrioside substrate (Ref. 7 and Table II). The hydrolysis of blue-starch, a large water-insoluble cross-linked starch polymer carrying a blue dye, by ␣-amylase to form water-soluble blue fragments was almost completely blocked by AMD9 VHH, whereas the AMB7 VHH showed a largely retarded solubilization of the chromophore, and AMD10 VHH exhibited only a weak effect (Table II). This suggests that the three VHHs interact at three different epitopes of the PPA and/or utilize different enzyme inhibition modes.

The VHH Structures
The VHH polypeptidic chains are not complete and are visible in density from residues 2 to 25 and 28 to 111 for AMB7 VHH (123 residues), from 1 to 111 for AMD9 VHH (118 residues), and from 2 to 112 for AMD10 VHH (119 residues) (Kabat numbering, Ref. 30). The three VHH structures adopt the classical immunoglobulin fold (Fig. 1B) and do not present impor-tant deviations in their frameworks: the root mean square deviation calculated with the 88 framework C␣ atoms ranged from 0.61 (AMB7/AMD9) to 0.84 Å (AMB7/AMD10). Indeed most deviations were observed at the CDRs level, but some significant divergences also occurred elsewhere: in the loop adjacent to the CDRs (amino acids 71-78), in the loop at the bottom of the VHHs (amino acids 39 -44), and in the segment 102-106 just after the CDR3 in the PPA⅐AMD10 complex.
FIG. 1. Sequences and three-dimensional structures of the three VHHs. The three CDRs (1-3) are colored red, green, and blue, respectively, and the cysteines are purple. A, anti-PPA AMB7, AMD9, and AMD10 VHH amino acid sequences. The numbering is according to Kabat (30). B, from left to right, views of the AMB7, AMD9, and AMD10 VHHs in the same orientation with the CDR3 oriented in front (view made with SPOCK (33)).
In the case of AMD10 VHH, the second disulfide bridge tethers CDR3 Cys-100c and CDR1 Cys-30. In the latter case, the relatively short CDR3 and the presence of this disulfide bridge may explain the deviation in the framework conformation observed between residues 102-106; in addition, the presence of two prolines (residues 100a and 100d) may also contribute to this effect.
The CDRs 1-3 in the VL and 1-2 in the VH domains of immunoglobulins have been shown to adopt a restricted set of conformations depending on their length and amino acid sequence (26). In the VHHs of camelids, the set of conformations of CDR1 and CDR2 has been found to extend beyond that observed in classical VHs (27). The CDRs in the three VHHs are well defined in density except for two residues that were not visible in the CDR1 of AMB7 VHH. The conformation of this CDR1, however, resembles that of canonical type 1. For the two other VHHs (AMD9 and AMD10) the CDR1s did not fit with any CDR1 of known canonical type or with any VHH known structures. In contrast, CDR2 canonical types were readily identified as type 1 for AMB7 and AMD10 VHHs and as type 2 for AMD9 VHH.
The CDR3 loops do not resemble the canonical types even if some conformational preferences and classes have been identified (28). Their lengths of 19, 14, and 13 residues, respectively, in the three VHHs can be considered as average for dromedary VHHs (Fig. 1A). Amazingly, when looking at the VHH side bearing the CDR3 (Fig. 1B), one CDR3 is going to the left (AMB7), another to the right (AMD9), and the last one upwards (AMD10). Clearly the disulfide bridges may play a role in this conformational dispersion. In the AMB7 VHH, the disulfide bridge pulls the CDR3 in a zone never observed in any of the other camelid VHHs. Proline 96 twists the CDR3 chain toward the solvent. It adopts an antiparallel two-stranded ␤-sheet structure (␤-hairpin) between residues 99 and its end at residue 100k (Fig. 1B). A proline at this position redirects the chain close to the protein core with a classical framework conformation. The AMD9 CDR3 covers the VH/VL interface as shown in other VHH studies and does not present any special features. In the AMD10 VHH, the disulfide bridge between CDR1 and the short CDR3 maintains the loops close together and makes it possible for the CDR3 to protrude as observed in the anti-lysozyme VHH (8).

Structure of the Complexes
Overall Structures-In the three complexes, the complete PPA polypeptidic chain was visible in the electron density map. The structures of the PPA complexes with AMB7, AMD9, and AMD10 VHHs contained three, four, and two PPA⅐VHH complexes in their asymmetric unit, respectively. When the PPA⅐VHH complexes contained in the asymmetric unit from the same crystal form were superimposed, very low root mean square deviation on the C␣ atoms were observed (below 0.3 Å in all cases). Superimposing the PPAs belonging to the three different complexes yielded surprisingly low root mean square deviation values below 0.5 Å for all atoms.  bound (pink). The PPA surface is in gray, and its visible catalytic residue Asp-300 is in red. The C␣ trace of the VHHs is orange, the three CDRs (1-3) are colored red, green, and blue, respectively, and each VHH surface is transparent. Inset, close-up view of the PPA active site with the inhibitory AMD9 VHH bound. The saccharidic inhibitor acarbose has been positioned in the active site according to the x-ray structure as a probe of the saccharide position. Two residues of the VHH, Tyr-52 and Arg-52a (CPK yellow and orange), clash with the modeled acarbose (views made with SPOCK (33)). The PPA⅐VHH Associations-Superposition of the three complexes using the PPA coordinates indicates clearly the different location of the VHHs at the PPA surface (Fig. 2). The AMD9 VHH interacts directly with the active site region of PPA and blocks the entrance of the V-shaped crevice. AMB7 VHH binds PPA at one end of the elongated active site crevice but far from the catalytic residues (Fig. 2). In contrast, AMD10 VHH binds PPA far from the catalytic crevice (Fig. 2).
Calculations of water-accessible surfaces have been used to evaluate the contacts of each VHH with its PPA partner. In Table III, the surface areas of the VHH shielded from the solvent by bound PPA have been reported. AMD9 VHH had the largest contact area with a total of 1151 Å 2 (Table III). This belongs to the highest values currently observed in complexes between classical antibodies (Fab or Fv fragments) and proteins. The two other fragments, AMB7 and AMD10 VHHs, interacted with PPA with large contact surface areas of 854 and 882 Å 2 , respectively (Table III).
In all three complexes, several parts of the VHHs interacted with PPA. Dissecting these interactions indicates a surprisingly large contribution of the framework residues to the VHH⅐PPA contacts with 28, 25, and 40% of the total contact surface in the three VHHs, respectively (Table III). Some contribution of framework residues to the interaction with the antigen has already been reported for classical immunoglobulins. However, the extent of these interactions is much smaller than for the present VHHs, the amount of buried surfaces covered by framework residues being comprised of between 1 and 9% of the total interaction (29). Such interactions have also been observed to be important in VHH/hapten interactions (13). The main interactions, however, were provided by the three CDRs, CDR3 being the main contributor in the complexes of AMB7 and AMD10 VHHs and the least for AMD9 VHH. In this latter case, CDR2 dominated the interaction and provided the inhibitory contact (Table III). In all cases, the CDR1 was little involved in the interactions.
The CDR3 plays a minor role in PPA inhibition when considering the AMD9⅐PPA complex. It did not protrude from the paratope and did not penetrate into the active site in the way that was reported for the lysozyme⅐cAb-Lys3 complex (8). This protruding geometry can certainly be established upon formation of a disulfide bridge between the CDR1 and the CDR3. This feature, however, was absent in AMD9 VHH. In contrast, this feature was observed in the AMD10 VHH in which the CDR3 protruded at the top of the VHH body. Despite this favorable conformation, AMD10 VHH did not bind to the catalytic crevice either. The AMD10⅐PPA interactions involved mainly framework residues on a side of the VHH and the CDR3 and other framework residues located close to it. In AMB7 VHH, the long hairpin CDR3 and some framework residues around it provided most of the interactions.
Interaction of AMD9 VHH at the Catalytic Site-The AMD9 VHH covered a large surface of PPA in the complex (1108 Å 2 ) within the active site crevice. This VHH, however, did not occupy a central position in the active site as compared with other proteinaceous inhibitors of PPA (18,19). The CDR2 loop 50 -55 contacted the upper side of the V-shaped extended active site crevice, the side chains of Tyr-52 and Arg-52a filling the amylose path at its non-reducing end. The Tyr-52-accessible surface was reduced from 32 to 9 Å 2 upon complexation and that of Arg-52a from 179 to 46 Å 2 , making this residue the largest contributor in the AMD9 complex formation. The Arg-52a guanidinium group was located 4.8 Å away from the catalytic residue Asp-300 and established a strong stacking contact (d ϭ 3.6 Å) with the Trp-59 indole ring of PPA, which has been shown to interact with the hydrophobic face of saccharides. This stacking has also been observed in other structures of PPA in complex with proteinaceous inhibitors such as tendamistat and ␣-AI (18,19).
Comparison of the PPA Structures in the VHH Complexes with Native Amylase-The three current PPA structures remain very close to the native PPA structure. Native PPA has been shown to exhibit striking differences when inhibited by a small molecule or some proteinaceous inhibitor such as ␣-AI (but not by tendamistat) (18,19). The core A domain was very similar in all PPA structures as well as in the VHH⅐PPA complexes with the exception of some loops near the active site (see below). In the VHH complexes, the C domain of PPA adopted slightly different orientations compared with the native enzyme. The B domain stretch 125-155 was distorted especially in the case of AMD9 VHH, while the effect was less pronounced with AMB7 VHH and non-existent in AMD10 VHH. Three loops deviated as compared with native PPA: loops 220 -225, 237-241, and 348 -354. The largest deviation with regards to the native enzyme involved the 348 -354 loop. The effect was larger in AMD9 VHH (3-Å displacement) and weaker in AMB7 and AMD10 VHHs. Two loop deviations were common between the complexes of PPA with VHH and with proteinaceous inhibitors. In this latter case, deviations were observed at 137-153, 237-241, and 350 -359 and at the active site of PPA.

Correlates of Binding Topology and Inhibition Data
Among the three VHH binders of PPA, AMD9 VHH was the only one to interact directly in the V-shaped active site crevice close to the catalytic residues. Its binding affinity (K d ϭ 3.5 nM, Table II) correlates well with its very large interaction surface area with PPA, the largest found among VHH/protein interactions. The affinity value was of the same order of magnitude as the K i value obtained upon competitive inhibition by AMD9 VHH of the PPA activity on a small pseudo-substrate, 2-chloro-4-nitrophenyl maltotrioside. This inhibition is also a clear indicator of the proximity of AMD9 VHH to the catalytic residues. In contrast, the two other binders were unable to inhibit the hydrolysis of the small substrate by the PPA, their epitope being located outside of the catalytic area. The K d value of AMD10 VHH was 25 nM, while that of AMB7 VHH was 1 order of magnitude larger despite the same interacting surface area. The action of PPA on large substrates, mimicked by the bluestarch Phadebas amylase test, was indeed inhibited by AMD9 VHH (Table II). AMB7 VHH, unable to inhibit the activity of PPA on a small substrate, exhibited, however, a marked inhibitory effect (70%) of the PPA activity on starch. This can be explained by the localization of AMB7 VHH in the starch binding crevice ϳ15 Å away from the catalytic residues, at the reducing end, opposite to the AMD9 VHH location. AMD10 VHH was unable to inhibit PPA activity on small substrates and presented a very reduced effect with large substrates that can be rationalized by its location far from the catalytic residues.

Conclusions
The crystal structures of the three VHHs in complex with PPA revealed water-accessible surface areas similar to or even larger than those observed in the complexes between proteins and classical antibodies (32). For the first time, a strong involvement of framework residues in opportunistic interactions with a proteinaceous antigen was observed. Such interactions have recently been documented in the case of a complex of a VHH with a hapten (12). Indeed the involvement of framework residues could compensate in some cases for the lack of VL and hence provide interaction surfaces of the same magnitude as for classical immunoglobulins. In brief, for a smaller footprint of VHHs compared with Fv fragments, involvement of framework residues makes it possible for VHHs to reach an interaction surface area as large, or even larger, than that observed with Fv fragments. Furthermore it makes it possible for the VHH to interact with its protein antigen with different topologies of orientations: front-wise, side-wise, penetrating, and flat-wise. The participation of the framework residues in the antigen recognition is expected to occur more frequently for VHHs than for VHs because we noticed an overall increased accumulation of mutations (i.e. variability) throughout the entire sequence relative to that of dromedary VHs (6). Hence these mutants introduced by an active somatic hypermutation mechanism can be selected during the affinity maturation. This is confirmed by the anti-PPA VHH fragments AMD9 and AMD7 that exhibited excellent K d and K i values around 10 nM. These values are much better that those obtained with small saccharidic inhibitors such as acarbose (K i ϭ 1 M), a molecule used against diabetes. Indeed anti-PPA natural proteinaceous inhibitors such as tendamistat and ␣-AI, which co-evolved with amylase over millions of years, exhibit smaller K i values (0.01 and 0.1 nM, respectively). These values are not out of reach with VHHs provided that directed evolution techniques are used.