A new non-classical fold of varroa odorant-binding proteins reveals a wide open internal cavity

Odorant-binding proteins (OBPs), as they occur in insects, form a distinct class of proteins that apparently has no closely related representatives in other animals. However, ticks, mites, spiders and millipedes contain genes encoding proteins with sequence similarity to insect OBPs. In this work, we have explored the structure and function of such non-insect OBPs in the mite Varroa destructor, a major pest of honey bee. Varroa OBPs present six cysteines paired into three disulphide bridges, but with positions in the sequence and connections different from those of their insect counterparts. VdesOBP1 structure was determined in two closely related crystal forms and appears to be a monomer. Its structure assembles five α-helices linked by three disulphide bridges, one of them exhibiting a different connection as compared to their insect counterparts. Comparison with classical OBPs reveals that the second of the six α-helices is lacking in VdesOBP1. Ligand-binding experiments revealed molecules able to bind only specific OBPs with a moderate affinity, suggesting that either optimal ligands have still to be identified, or post-translational modifications present in the native proteins may be essential for modulating binding activity, or else these OBPs might represent a failed attempt in evolution and are not used by the mites.


Results
Protein expression and purification. First we confirmed the five OBP sequences reported in the NCBI database by cloning the five genes from RNA extracted from whole individuals. Sequences were identical to the published ones at the amino acid level, while few single base substitutions were likely due to differences between the population of mites used in our study with respect to those used for the sequences stored in the NCBI database.
The mature proteins, containing an additional methionine present at the N-terminus, were expressed in bacteria, following standard protocols, as reported in the Methods section. We have successfully obtained VdesOBP1, VdesOBP2, VdesOBP4 and VdesOBP5 in medium to good yields, but we were unable to express VdesOBP3 in amounts adequate for purification and consequently for its characterization. The proteins were purified by two or three chromatographic steps on anion-exchange columns, such as DE-52 and HiPrep-Q. Figure 2 reports a summary of expression and purification for the four OBPs.
Cysteine pairing assessment. VdesOBP1-5 contains six cysteines occurring at conserved positions ( Fig. 1), although in a pattern different from the one typical of insect OBPs. This fact suggested possible differences in the cysteine pairing with respect to insects [10][11][12] . To assess this issue, VdesOBP1, VdesOBP2 and  www.nature.com/scientificreports/ VdesOBP5 were subjected to extensive alkylation with iodoacetamide under non-reducing denaturing conditions; after purification, they were then digested with trypsin followed by a second treatment with chymotrypsin. Resulting digests were analyzed by nanoLC-ESI-Q-Orbitrap-MS/MS, and S-S-crosslinked peptides were assigned using dedicated bioinformatics procedures (Supplementary Table S2 online). Examples of mass spectra of disulphide-bridged peptides from VdesOBP1 and VdesOBP5 digests are reported in Fig. 3 and Supplementary Figure S1 online, respectively. These results and the absence of carboxyamidomethylated peptides in the protein digests demonstrated that all cysteines present in VdesOBPs are involved in disulphide bridges. Surprisingly, the ascertained, conserved disulphide pattern of VdesOBPs was unlike that of insect OBPs. In fact, a C1-C6, C2-C3 and C4-C5 arrangement was identified in VdesOBPs, which was completely different form the C1-C3, C2-C5, C4-C6 typical of insect OBPs 10-12 . This phenomenon brings two major consequences in the structure of VdesOBPs: (1) the pairing of adjacent cysteines should make V. destructor proteins more flexible and less compact than insect OBPs; (2) linking C1-C6 brings together the protein N-and C-termini, while these two ends point generally in opposite directions in insect OBPs.
Protein characterization in solution. The  Overall structure of VdesOBP1. The three-dimensional structure of VdesOBP1 was solved by sulphur single-wavelength anomalous dispersion (SSAD) with a wavelength λ = 1.512 Å on beamline proxima 2A at SOLEIL synchrotron (Saint-Aubin, France). Data were collected at 1.968 Å resolution on a P3 2 21 crystal form and a second dataset was collected at 1.81 Å resolution with wavelength λ = 0.98011 Å (Table 1). Another crystal form was collected, with space group P2 1 at 1.2 Å resolution, and its structure was determined by molecular replacement using the P3 2 21 crystal form. VdesOBP1 crystal forms P3 2 21 and P2 1 were crystallized at pH values of 7.5 and 9.5, respectively. They contain one and two molecules in the asymmetric unit, respectively. After refinement, the final R/Rfree values were 18.6/22.6% and 18.0/20.1% for the two crystal forms, respectively ( Table 1).
The cloned protein is 147 residues long and possesses 6 cysteine residues (Fig. 4). The chain was built in the electron density map of the P2 1 crystal form with the exception of residues 1-6 at the N-terminus, while the P3 2 21 crystal form starts at residue 11 and lacks loop 50-53. We checked the integrity of the protein in the crystal and our analysis showed that the protein was intact, confirming that the N-terminal segment is present but not ordered in the crystal. The r.m.s.d. value between the structures of the two VdesOBP1 crystal forms is 1.55 Å (Fig. 5a).
It has been reported that OBPs may form functional dimers [25][26][27] . However, it cannot be concluded firmly for or against dimerization of OBPs in their biological context. In the crystal, dimers have been observed in the case of AgamOBP1, but this observation may not apply to solutions, due to an insufficient contact area 26 . In the case of VdesOBP1, we observed a monomer using multi-angle static light scattering (MALS)/dynamic light scattering and RI coupled on-line with a size-exclusion chromatography (SEC) column (MALS/UV/RI/SEC) at three pH   Fig. S2 online). In contrast, the P2 1 form contains two molecules in the asymmetric unit forming a crystallographic dimer (Fig. 4b). A crystallographic dimer can be reconstituted in the P3 2 21 form, using a symmetrical molecule in the crystal packing (Fig. 5b). However, the second monomer of one form does not display the same orientation relative to that of the second form when the first monomers are superimposed (Fig. 5b,c). Finally, the PISA server reported an ambiguous response for the P2 1 form (score = 0.49) but discarded the possibility of a biological dimer in form P3 2 21 (score = 0). All in all, it seems highly probable that the biologically relevant form is a monomer.
The binding crevice. The binding crevice, as detected by CavityPlus 28 , is large and wide open in the monomer, with dimensions of the floor of ~ 18 × 18 Å (Fig. 6a). This large surface implies that numerous amino acid side-chains form the walls of the binding site: hydrophobic side-chains (Pro14, Ile16, Trp18, Leu24, Pro 26, Val37, Val38, Leu70, Trp75, Tyr82, Ala87, Ile91, Tyr129, Met133, Ile137), polar non-charged residues (Thr66, Gln130) and charged residues (Glu30, Arg31, Lys34, Glu73, Glu90, Asp134). Hence, although most side chains are from hydrophobic/aromatic residues (16/24), two basic and four acidic residues complete the wall. The active site is empty, except in form P2 1 where a buffer molecule, 2-(N-cyclohexylamino)-ethane sulfonic acid (NCES), is docked against a wall of the binding site (Fig. 6a). Examination of the binding site in the crystallographic dimer revealed that the N-terminus of each monomer protrudes deeply in the binding site of the other monomer, drastically reducing its size. This effect explains the serendipitous binding of NCES as this ligand interacts with the cell wall of one monomer and the C-terminal extension of the other one (Fig. 6b). The sulfonic acid moiety of NCES forms an ionic bond with Arg31A side chain (atom NH 2 ) and with the main chain N-H of Lys12B.

Comparison of VdesOBP1 with other OBPs/PBPs structures.
A search on the PDB with the DALI server reported hits with good statistics, taking into account the Z-score and the r.m.s.d. 29 . The best hit is that of entry 5dic (Z = 7; r.m.s.d. 2.9 Å on 96/115 residues), the fatty acid binding protein OBP56a from the oral disk of the blowfly Phormia regina (PDB ID: 5dic, by Ishida, Y., Leal, W.S., Wilson, D.K., unpublished) with a non-  www.nature.com/scientificreports/ classical OBP fold. This FBP/OBP has been shown to bind the C16 palmitic acid (but not the acetate), as well as the C18 stearic, oleic and linoleic acids, but not decanoic acid 30 . The structure is reported in complex with two molecules of palmitic acid in its wide cavity. Superposition of VdesOBP1 and PregOBP56a shows that VdesOBP1 helices 1, 2, 3, 4 and 5 superimpose well with PregOBP56a helices 1, 3, 4, 5 and 6. Helix 2 of PregOBP56a closes its binding cavity, while a similar effect is not observed in VdesOBP1. This PregOBP56a helix 2 is replaced in VdesOBP1 by a loop between helices 1 and 2, projected out of the protein core (Fig. 7a). However, the Pre-gOBP56 disulphide bridges obey the C1-C3, C2-C5 and C4-C6 connectivity also present in classical OBPs, in contrast with VdesOBP1 (C1-C6, C2-C3, C4-C5). Another good fit (1ow4; Z = 5.4; r.m.s.d 3.3 Å on 8-90/120 residues) is observed with the PBP of the cockroach Leucopheae maderae (LmadPBP) (1ORG). In this case again, LmadPBP exhibits an extra helix instead of a loop between helices 1 and 2 in VdedOBP1 (Fig. 7b). Two disulphide bonds are conserved between the two proteins, the first bridging helices 1 and 2, the second bridging helices 4 and 5 of VdesOBP1 (Figs. 4a, 7b), although as result of different (C1-C6, C2-C3, C4-C5 in VdesOBP1 and C1-C3, C2-C5, C4-C6 in LmadPBP) cysteine linkages. The third disulphide bond that bridges helices 3 and 6 in LmadPBP is not observed in VdesOBP1. Instead, the third disulphide bridge is observed between helix 5 (corresponding to helix 6 in LmadPBP) and the extended N-terminus in VdesOBP1 (Fig. 7b).
Ligand-binding assays. The purified VdesOBP1 did not show any affinity for either of the most common probes used in ligand-binding assays with OBPs, namely N-phenyl-1-naphthylamine (1-NPN) and 1-aminoanthracene (1-AMA). Extraction with dichloromethane to remove possible endogenous ligands did not prove effective. We also subjected the protein to a denaturing treatment with 8 M urea and 1 mM DTT, followed by slow renaturation by extensive dialysis, but we still could not observe any binding with either probes. We also tried 8-phenylamino-naphthalenesulfonic acid (ANS) as well as larger fluorescent reporters 31 , such as N-naphthyl-1-naphthylamine (1-NNN) and N-phenyl-1-aminoanthracene (PAA) with no effect. It is still possible that the fluorescent probe binds the protein, but the event is not accompanied by a change in the emission spectrum. In fact, when adding 1-NPN to the protein, we observed a dose-dependent quenching of the intrinsic tryptophan fluorescence, suggestive of some affinity of this probe for VdesOBP1 (data not shown). In any case, competitive binding using a fluorescent reporter could not be used with VdesOBP1 or with VdesOBP2. Therefore, we attempted to find ligands for these two proteins monitoring the quenching of intrinsic tryptophan fluorescence. On the other hand, this method is limited to chemicals able to accept the energy transferred by the tryptophan (therefore structures with a region of conjugated π-electrons), and is not suitable to evaluate binding affinities. Instead, with VdesOBP4 and VdesOBP5 we obtained clear binding curves using 1-NPN as the fluorescent reporter (Fig. 8a). Therefore, we performed competitive binding experiments only with these two OBPs. Some of the displacement curves obtained with VdesOBP4 and VdesOBP5, using the ligands shown in Fig. 8b, are reported in Fig. 8c,d respectively. We can notice that only coniferyl aldehyde is able to decrease the 1-NPN fluorescence to 50% at the maximum concentration used, qualifying for a modest ligand. None of the other chemicals used in the experiments (listed in Supplementary Table S3 online) with the four proteins proved www.nature.com/scientificreports/ to be a better ligand. Representative curves relative to quenching of the intrinsic fluorescence of tryptophan in VdesOBP1 and VdesOBP2 are reported in Supplementary Fig. S3 online. Summarising these data, we can first observe that some of the tested chemicals proved to be ligands for Vdes-OBP4 and VdesOBP5 in 1-NPN displacement assays, while they determined significant quenching of intrinsic tryptophan fluorescence in VdesOBP1 and VdesOBP2. The same few compounds (coniferyl aldehyde, safranal and vanillin) were the best ligands with all four proteins, although with poor dissociation constants between 10 and 20 μM. These ligands are associated with plants, in particular with leaf damage, and do not bear any relationship to the varroa habits. In particular, none of the honey bee pheromones, such as 9-ketodecanoic acid, homovanillyl alcohol or fatty acid esters were able to bind any of the tested proteins.

Discussion
This work reports the original structural and functional characterization of four out of the five OBPs encoded in the genome of the mite Varroa destructor, a major pest for honey bee. Mass spectrometric analysis demonstrated that the six cysteines VdesOBPs, whose sequence position do not match the pattern in insect OBPs, are paired into three disulphide bridges (C1-C6, C2-C3 and C4-C5) different from that (C1-C3, C2-C5, C4-C6) of their insect counterparts. The overall folding, determined for VdesOBP1, is also novel and different from the conserved structure of insect OBPs, although some similar aspects can be recognised.
VdesOBP1 structure was determined in two crystal forms closely related. Its structure assembles 5 α-helices linked by three disulphide bridges. Comparison with classical OBPs reveals that the second of the six α-helices found in them is lacking in VdesOBP1. Although dimer formation of VdesOBP1 is observed in the crystal, www.nature.com/scientificreports/ it most probably does not represent the functional form in vivo. This dimerization is probably driven by the presence of a wide and fully open cavity, able to accommodate the extended N-and C-terminal stretches of the opposite monomer. The large cavity opening can be ascribed to the absence of helix 2 in OBPs classical fold, which restrains and closes the internal binding site. In VdesOBP1, this helix is absent and replaced by a loop projected outside the protein core. Two disulphide bridges are conserved as compared to classical OBPs, although associated with different sequential cysteine linkages. These observations suggest that VdesOBP1 possesses an original fold remotely related to classical OBPs. Comparison with PregOBP56a suggests that two to three fatty acids could bind inside the VdesOBP1 cavity. Such an arrangement of three ligands was observed in Mamestra brassicae CSP (CSPMbraA6) 30 . The unique structure of VdesOBP1 and the lack of strong ligands pose questions that, at this stage, can only be addressed by formulating hypotheses and models to be verified by further experimental work. A first idea, based on the limited binding activity of these proteins and on the small number of genes (4-5) expressed in V. destructor, as well as in other mites and ticks, might suggest that these genes represent by-products of evolution, which are not used any more. In fact, at least two more families of carrier proteins could be active in Chelicerata, namely NPC2 proteins and lipocalins, apart from a third family (CCPs: candidate carrier proteins) that so far seems to be specific of spiders 21 . NPCs and lipocalins are represented in varroa by 6 and 9 members, respectively, while in other Chelicerata often 10-20 members of NPC2s have been identified, with exceptions of nearly 50 genes, as in the spider mite Tetranychus urticae. Although making sense for the reason reported above, the idea that OBPs in Chelicerata might represent a dead end in evolution and a failed attempt to make efficient carrier proteins contrasts with proteomic findings showing significant expression of such proteins 22 . In fact, four (Vdes-OBP1-4) of the five VdesOBPs were identified in different parts of V. destructor, with one of them (VdesOBP4) particularly abundant in sensory organs, namely the first pair of legs and the mouthparts.  www.nature.com/scientificreports/ If then the expression of OBPs at the protein level indicates that they are functional, then, along with a second hypothesis, their poor affinity might be related to the absence of post-translational modifications. It has been recently shown that the pig OBP1 can bind fatty acids or steroids depending on the phosphorylation and glycosylation status of the protein 32 . Our varroa OBPs, being expressed in bacteria, are devoid of any posttranslational modification. On the other hand, we do not have any information on possible phosphorylation and/ or glycosylation of these proteins in their native forms. This is certainly an interesting aspect, although not easy to investigate, not only for the tiny size of sensory organs, but also because phosphorylation, in particular, may be present only in certain time windows related to the physiology and the behaviour of the mite. Indeed, protein phosphorylation is highly related to signal transduction events regulating essential physiological process, and it has been reported to drastically modify the binding properties of OBPs 32 . Thus, we searched for putative phosphorylation sites in the five OBP sequences, and we found a very large number of serine and threonine residues that could be potentially phosphorylated ( Supplementary Fig. S4 online). In addition, we found few tyrosine residues as potential phosphorylation sites (2 in VdesOBP1, 1 in VdesOBP2 and VdesOBP3, 4 in VdesOBP5).
In the three-dimensional structure of VdesOBP1, no serine or threonine residues are found inside the binding pocket. However, phosphorylation of serine and threonine residues outside the binding pocket (as well as glycosylation of the protein) can affect the folding of the protein and influence its binding affinities. Further speculation would be inappropriate based on the available experimental data. Certainly, a structural study of the native proteins, regarding their post-translational modifications would be essential before proposing alternative modes of action.

Materials and methods
Biological material. V. destructor mites were collected from the cells of an infected honeybee hive and used immediately for RNA extraction.
Reagents. Chemical compounds for buffers and solutions, as well as ligands for binding assays were of analytical grade from Merck, Austria. Ligands were dissolved in spectroscopic grade (Uvasol) methanol. Oligo- www.nature.com/scientificreports/ nucleotides and synthetic genes were custom synthesised at Eurofins Genomics, Germany. Enzymes and kits for DNA extraction and purification were from New England Biolabs, USA.
RNA extraction and gene cloning. Total RNA was extracted using Tri-reagent (Sigma-Aldrich) according to the protocol provided. cDNA was prepared from 2 μL of total RNA using the kit qScript cDNA SuperMix (Quanta Bio) and the manufacturer's procedure. For sequence verification, PCR was performed on cDNA using specific primers at the 5' end and at the 3' end. The primers, which were also used for cloning into the expression vector, also contained the restriction sites NdeI and EcoRI at the 5' and the 3' ends, respectively. After digestion, the PCR products were inserted into pET30 vector previously linearized with the same enzymes. This procedure was adopted for all the genes of V. destructor, except for VdesOBP1 used for structural studies, which was produced from a synthetic gene (Eurofins Genomics), whose nucleotide sequence was optimized for E. coli expression.
Gene cloning. Synthetic  www.nature.com/scientificreports/ the additional settings reported above for Proteome Discoverer and Byonic examinations. Disulphide-bridged peptide identification was considered reliable when BioPharma Finder and pLink results showed a "confidence score" > 95 and/or an "E-value" < 1.0 -10 , respectively. Manual interpretation and verification of the candidate spectra were always performed.
Crystallization, data collection and structure determination of the P3 2  The crystal was cryo-cooled at 100 °K within this crystallization liquor, with the addition of 10% glycerol as cryoprotectant. Structure determination used SAD-phasing method at remote wavelengths (λ = 1.51200 Å) with sulphur atom anomalous dispersion on native crystals. A strong theoretical Bijvoet ratio of 1.12% was expected from the 10 sulphur atoms of the protein that contribute to the anomalous signal at wavelength of 1.512 Å. The sulphur-SAD data set was collected on the PROXIMA-2 beamline (SOLEIL) equipped with a high performance micro-diffractometer with a mini-kappa (MD2), an X-ray fluorescence detector (KETEK), and a fast, low-noise, photon-counting, area detector EIGER X 9 M (238 fps in 9 M mode, 750 fps in 4 M mode). A redundant data set was collected at 1.97 Å resolution. including 3600 images of 0.1° rotation collected with four K orientations of 0°, 30°, 45° and 60°. The data set was processed with XDS 37 . The data sets were indexed and integrated with Pointless and scaled with CCP4 SCALA 38 . VdesOBP1 crystallized in space group P3 2 21 with cell dimensions a = 53.84 Å, b = 53.84 Å, and c = 91.74 Å and and γ = 120 with one molecule in the asymmetric unit, corresponding to a solvent content of 50.87%. Six sulfur atom sites were found by using PHENIX Autosol 39 . The results from PHENIX include a PDB containing the anomalous scatterers, a MTZ file containing the experimentally determined phases, and the optimized phases from density modification. Structure initial construction was performed with PHENIX Autosol 39 and was not further refined. A second data set was collected at 1.20 Å resolution at a wavelength of 0.97856 Å. The ab-initio determined structure was used as a starting model for model refinement. Refinement was performed with REFMAC5 40 and AutoBUSTER 41 alternated with rebuilding using COOT 42 . Crystallization, data collection and structure determination of the P2 1 form. The protein at 20 mg/mL was subjected to screening experiments that were performed using several commercial kits. Vdes-OBP1 was set in 2/3 of the initial crystallization conditions (0.1 M CHES pH 9.5, , 30% w/v PEG3000) and 1/3 of ammonium sulfate screen (0.2 M sodium nitrate, 2.2 M ammonium sulfate). Crystals of VdesOBP1 were briefly soaked in crystallization solution supplemented with 10% (v/v) polyethylenglycol. Diffraction data were collected at 1.2 Å resolution on beamline Proxima-1 at SOLEIL, Paris, France. The data sets were integrated with XDS 37 and were scaled with CCP4 SCALA 38 . VdesOBP1 crystallized in space group P2 1 with cell dimensions a = 31.6 Å, b = 112.96 Å, c = 41.82 Å and β = 98.95, with two molecules in the asymmetric unit, corresponding to a solvent content of 44.48%. The structure was solved by molecular replacement with MOLREP 43 using the structure of VdesOBP1 form P3 2 21 as starting model. Refinement was performed with REFMAC5 40  Multi-angle light scattering. 100 μL of purified VdesOBP1 were analyzed by SEC-MALS using a HPLC system (Thermo Scientific) equipped with a Superdex 75 column coupled with a multi-angle light scattering device (Dawn 8, Wyatt Technology) and an Optilab TrEX differential refractive index detector (Optilab TrEX, Wyatt Technology). At first, the column was first equilibrated with a buffer containing 0.02% w/v NaN 3 for several hours, until the baseline of the Optilab TrEX signal was stable. Then, light scattering and concentration data were analysed in ASTRA to determine molecular weight and size distribution. Inter-detector band broadening and alignment were applied using the algorithms in ASTRA 7 (Wyatt Technology, USA) using BSA protein standard according to manufacturer's recommended methods. Three different buffers were used to understand the behaviour of the protein at different pH values, namely 50 mM CHES pH 10, 50 mM Hepes pH 7 and 50 mM sodium acetate pH 5. The temperature-controlled autosampler was kept at 10 °C. www.nature.com/scientificreports/ 6500 spectrofluorometer (PerkinElmer); slits were set at 5 nm for both excitation and emission, and 1 cm path quartz cuvettes were used. Dissociation constants for 1-NPN were evaluated using Prism software. Affinities of other ligands were calculated from the corresponding [IC] 50 values (the concentration of each ligand halving the initial value of fluorescence), from the equation: where  is the concentration of free 1-NPN and K NPN is the dissociation constant of the complex OBP/1-NPN.