Phosphite binding by the HtxB periplasmic binding protein depends on the protonation state of the ligand

Phosphorus acquisition is critical for life. In low phosphate conditions, some species of bacteria have evolved mechanisms to import reduced phosphorus compounds, such as phosphite and hypophosphite, as alternative phosphorus sources. Uptake is facilitated by high-affinity periplasmic binding proteins (PBPs) that bind cargo in the periplasm and shuttle it to an ATP-binding cassette (ABC)-transporter in the bacterial inner membrane. PtxB and HtxB are the PBPs responsible for binding phosphite and hypophosphite, respectively. They recognize the P-H bond of phosphite/hypophosphite via a conserved P-H...π interaction, which confers nanomolar dissociation constants for their respective ligands. PtxB also has a low-level binding affinity for phosphate and hypophosphite, whilst HtxB can facilitate phosphite uptake in vivo. However, HtxB does not bind phosphate, thus the HtxBCDE transporter has recently been successfully exploited for biocontainment of genetically modified organisms by phosphite-dependent growth. Here we use a combination of X-ray crystallography, NMR and Microscale Thermophoresis to show that phosphite binding to HtxB depends on the protonation state of the ligand, suggesting that pH may effect the efficiency of phosphite uptake by HtxB in biotechnology applications.


HtxB can form a Complex with Mono-Anionic phosphite at Low pH
In a previous study we determined the structure of P. stutzeri HtxB in complex with hypophosphite (PDB:5ME4), but under the crystallisation conditions used, we were unable to obtain the complex with phosphite 24 . We have now obtained a 1.25 Å structure of the HtxB/phosphite complex (PDB:6EMN) (see Table 1 for data collection and refinement statistics), which was crystallised at pH 5, showing how HtxB accommodates phosphite in its binding pocket.
HtxB is a typical type-II PBP, with a binding pocket positioned in a deep groove between two domains separated by a hinge region. In the HtxB complex with phosphite, the phosphorus centre of the ligand and the four tetrahedral atoms surrounding it (3 oxygen and 1 hydrogen) occupy approximately the same positions as the equivalent atoms in the hypophosphite complex (2 oxygen and 2 hydrogen). The two common oxygen atom positions between the phosphite and hypophosphite make equivalent interactions with residues in the binding pocket, namely the sidechain and mainchain amide of T129, S130, the sidechain hydroxyl of Y97 and the structural water molecule (HOH) ( Fig. 2A-C). The walls of the binding pocket are formed by packing interactions between P71, N128, F158, M176, R178 and D206, with the latter two resides forming a buried salt bridge. The pocket is capped by a tryptophan sidechain (W52) that makes a P-H…π interaction with the R1 hydrogen atom of the phosphite (distance of 2.7 Å and angle of approximately 140°), forming a similar interaction to that present in the hypophosphite complex. The R2 hydrogen in hypophosphite is not involved in any hydrogen bonding interactions within the pocket as the closest residues are outside of van der Waals radius. However, the additional oxygen atom of the phosphite, to which it is equivalent, points towards the D206/R178 salt bridge to one side of the HtxB binding pocket. This phosphite-oxygen atom is 2.8 Å away from the oxygen atom of D206 and 3.2 Å away from the nitrogen atom of R178, suggesting that both residues interact with this phosphite-oxygen, albeit with a rather long hydrogen bond to the arginine (Fig. 2D). Although we could not directly observed the position of the hydrogen atoms in the structure, at pH 5 it is likely that the majority of the phosphite in solution is mono-anionic (pK a 1.3, 6.7, Fig. 1), carrying a single proton on one of the three oxygen atoms. The 2.8 Å distance between the phosphite-oxygen and D206 implies, therefore, that the phosphite is protonated at this position and acts as a hydrogen bond donor to D206. These additional hydrogen bonds appear to provide enough stabilising energy to trap a fully closed complex in the crystal structure, which is virtually identical to the conformation observed in the complex with hypophosphite ( Fig. 2C) (RMSD Cα: 0.25 Å). Above pH 6.7, phosphite is predominantly found as the di-anionic form, lacking an available proton to hydrogen bond with the aspartate, which is consistent with the very weak binding (K d > 1 × 10 4 μM) observed between phosphite and HtxB in our previous MST assays at pH 7.4 24 .

Figure 1.
Chemical structures and pK a s of phosphite. *In this study, pK a1 was determined to be 6.3 by NMR.

Binding of phosphite to HtxB in solution shows a Distinct pH Dependence
In order to characterize the pH dependence of the binding event between HtxB and phosphite, we used hydrogen and phosphorus NMR to directly observe the interaction in solution. The covalent bond between P and H atoms in phosphite gives rise to a 1 J PH coupling which appears as a doublet in the 31 P NMR spectra. This 1 J PH coupling can be used in conjunction with chemical shifts to characterize the protonation state of the phosphite, (for general NMR method see 25,26 ) and can be removed by application of 1 H decoupling during acquisition of 31 P spectra, resulting in a single peak. The linebroadening of these single peaks is dominated by transverse relaxation of the nuclei in question, therefore, binding of phosphite to HtxB would increase the linewidth of observable NMR resonances as a protein-phosphite complex will tumble slower than free phosphite in solution, increasing the transverse relaxation of the phosphorous nucleus and enabling us to detect ligand binding.
In the absence of protein, we performed a pH titration of 10 mM sodium phosphite in 50 mM Tris-acetate buffer across the pH range 4-9 and followed it using both 1D 31 P (Fig. 3) and 1D 1 H NMR ( Supplementary Fig. S1). We also followed 31 P chemical shift changes and 1 J PH couplings across the pH titration, which resulted in a pK a1 value of 6.3 ± 0.1 (Fig. 3D,E and Table 2) compared to previously published values of pK a1 6.2, 6.5 and 6.7 [27][28][29] . Negligible 31 P or 1 H line-width changes were observed for the phosphite peaks across the pH titration ( Fig. 3A and Supplementary Fig. S1). We repeated the pH titration in the same conditions but with the addition of 1 mM HtxB, recording both 1D 1 H and 31 P NMR spectra ( Fig. 3B and Supplementary Fig. S1). The change in chemical shift of the non-decoupled 31 P and 1 H phosphite peaks indicated a pK a of 6.2 ± 0.1, which closely matched the chemical shift change for free phosphite (Supplementary Fig. S1 and Table 2). An upfield shift of the decoupled 31 P phosphite peak was observed as the pH of the buffer was decreased, which is in agreement with the control experiment in the absence of HtxB (Fig. 3B). Minimal chemical shift difference was observed for the phosphite peak on addition of HtxB, however significant linebroadening of the 31 P peak was observed on reducing the pH www.nature.com/scientificreports www.nature.com/scientificreports/ from 9 to 4 (Fig. 3B). This observation indicates that phosphite is binding to HtxB in fast, approaching intermediate, exchange across this pH range, thus has weak binding affinity. The same linebroadening effect at lower pH values is a result of increased transverse relaxation rates that occur only when HtxB is present (Fig. 3E). Maximal line broadening is observed at pH 6, indicating the longest average residence time of phosphite in complex with HtxB is at a pH close to the pK a of phosphite in solution (Fig. 3D, Table 2). Together, this shows that in solution, HtxB is capable of binding phosphite at pH values lower that 6, in agreement with the crystal structure, but that the binding affinity for phosphite is weak.
Having confirmed that HtxB is able to bind phosphite at pH values lower than 6 by NMR, we used Microscale Thermophoresis (MST) to determine a K d for this interaction ( Fig. 4 and Table 3). We labeled the protein with the amine reactive dye, NT-647-NHS, rather than the His-tag fluorescent label that we used in our previous study 24 , which cannot make a stable interaction with the protein at pH 5. A 1:1 ratio of protein molecule to dye was achieved via attachment to a surface exposed lysine group. We determined the K d for HtxB binding to hypophosphite as 650 ± 130 nM (in 40 mM citrate-phosphate buffer, pH 7.4), compared to 560 ± 110 nM in 50 mM pH 7.4 HEPES buffer using the His-tag label (recorded previously 24 ). Under the same citrate buffer conditions, no binding of phosphite to HtxB was detected up to 100 mM phosphite, but on adjusting the pH of the buffer to 5, HtxB binds phosphite weakly, with a K d of 1.74 mM. The reduction of pH also results in an approximately two-fold reduction in binding affinity for hypophosphite, with a recorded K d value of 1.46 μM. Thus, even though HtxB can be crystallised in a closed conformation with a large excess of phosphite at pH 5, the MST results corroborate the NMR experiments, demonstrating that the binding affinity for the ligand in solution is weak. It also acts as a hydrogen bond acceptor to R178, forming a rather long hydrogen bonding interaction (3.2 Å). (B) Superposition of the HtxB complex with phosphite (green) and hypophosphite (orange) showing the conserved position of residues surrounding the binding site and the conserved P-H…π bond with W52. In (A,B) the protein is drawn as a cartoon with the ligands and selected side chains drawn as sticks. The single water in the binding pocket is drawn as a small red sphere. The P-H…π interaction is drawn as a blue dashed line and hydrogen bonds are drawn as orange dashed lines. (C) The same superposition shows that the fold and conformation of the protein is practically identical in complex with phosphite (green) or hypophosphite (orange). The protein backbone is represented as a cartoon and the ligands are drawn as spheres. (D) A schematic showing that R1 is positionally conserved in both phosphite (green) and hypophosphite (orange), with the hydrogen at the R2 position in hypophosphite being replaced by the oxygen atom of the hydroxyl that hydrogen bonds to D206 and R178 in the HtxB-phosphite complex. Note that the hydrogen position on the R2 oxygen of the phosphite is assumed based in the hydrogen bonding network, it cannot be observed in the electron density map. The ligands are shown as sticks and the relative position and interactions with W52, D206 and R178 are indicated schematically with circles and dashed lines, respectively. (2019) 9:10231 | https://doi.org/10.1038/s41598-019-46557-2 www.nature.com/scientificreports www.nature.com/scientificreports/

The effect of D206A and D206N Mutations on Ligand Binding in HtxB
To probe the role of the D206 in ligand binding we generated D206A and D206N mutants of HtxB. Both proteins were produced and purified using the same methods as for WT HtxB. The mutations produced no gross changes in the secondary structure content of the protein, as observed by CD spectroscopy (Supplementary Fig. S2). MST assays show that in the presence of 20 mM acetate the D206A mutant binds hypophosphite at both pH 7.4 and pH  Table 2. (E) pH dependence of line broadening of 10 mM 31 P in the absence of protein (red) and presence of WT HtxB (black) and D206A HtxB (blue). Full width at half-maximum (FWHM) estimated from fitting peaks to a Lorentzian (Eq. 3). Lines in (E) are theoretical cubic splines for guidance only. www.nature.com/scientificreports www.nature.com/scientificreports/ 5, but with a ca. 5000 × reduction in binding affinity compared to the WT protein in the same conditions ( Fig. 4 and Table 3). When we assayed D206A HtxB in acetate-free conditions at pH 7.4, no binding of hypophosphite could be detected up to 400 mM of ligand (data not shown). No evidence of D206A HtxB binding to phosphite at either pH 7.4 or 5 could be detected and 31 P NMR spectra of D206A HtxB in the presence of 10 mM phosphite did not show significant linebroadening of the phosphite peak across the pH range 4-9 (Fig. 3C). The D206N mutation renders D206N HtxB incapable of binding phosphite or hypophosphite at pH 5 or 7.4, as determined by MST with up to 100 mM of ligand (Table 3).
D206A HtxB was crystallised in complex with hypophosphite (PDB:6GHT) in a closed conformation that was equivalent to the WT complex (see Table 1 for data collection and refinement statistics). In the structure, the extra space generated by the alanine mutation allows for a single molecule of acetate to fulfill the role of D206 in the WT protein, forming two hydrogen bonds to R178, and a further hydrogen bond to the indole ring-nitrogen of W68 (Fig. 5A). We cannot account for the origin of the acetate, but this explains why we only detect binding of hypophosphite to D206A HtxB via MST in acetate containing buffers. All of the atoms of the acetate molecule   www.nature.com/scientificreports www.nature.com/scientificreports/ lie in plane of the guanidinium group of the arginine, forming a classic end-on salt bridge interaction. This differs from the direction of approach of the aspartate to the arginine in the WT complex, where the two oxygen atoms of the aspartate lie 1 Å below the plane of the guanidinium group, with the Cα-Cβ bond of D206 offset by approximately 75° from the same plane (Fig. 5B). It is not possible to say from the electron density map or interactions with the surrounding residues if the acetate is protonated at the oxygen that points towards the hypophosphite, but the crystallisation pH is less than the pK a of acetate (4.76) 30 . The mode of binding of the hypophosphite and the degree of domain closure is the same as the WT protein (RMSD Cα: 0.19 Å) (Fig. 5C), but the binding of the acetate slightly disrupts the local packing around the binding pocket, causing a change in rotamer conformation of M18 (Fig. 5D). Clearly, the high concentration of hypophosphite in the crystallisation experiment along with the presence of acetate favours the formation of a closed, ligand-bound complex of D206A HtxB, that would not readily form in solution, given the weak binding affinity for ligand. However, the observation of acetate binding to facilitate the formation of the buried salt bridge in the closed complex of D206A HtxB, suggests that this intramolecular interaction is a prerequisite for capture of the ligand.
D206N HtxB crystallised in a partially open conformation (PDB:6GHQ, crystallised in the presence of hypophosphite) (see Table 1 for data collection and refinement statistics) in complex with a sulphate molecule that binds at the N-terminal end of helix 3, presumably being partially stabilised by the helix dipole at this location (Fig. 6A). Both sulphate and phosphate could be refined into the electron density map at this position, but the crystallisation conditions contained 0.2 M sulphate and we determined that phosphate binding to D206N HtxB was incredibly weak, therefore it was unlikely to co-purify ( Supplementary Fig. S3). The sulphate binding site is approximately 6 Å (S-P distance) away from the native hypophosphite binding site in the WT HtxB structure (Fig. 6B), forming hydrogen bonds with a number of surrounding water molecules (Fig. 6A), including the www.nature.com/scientificreports www.nature.com/scientificreports/ mainchain amides of P71, W72, G73, and the sidechain of T129, which is one of the conserved binding pocket residues required for hypophosphite binding. A comparison of the D206N HtxB structure with the closed structure of WT HtxB (PDB:5ME4) and the open structure of apo P. stuzeri PtxB (PDB:5O2K) 24 , using DynDom 31 , shows that one domain of D206N HtxB is rotated by 35° relative to the other around an axis that lies between the two domains, which is approximately half of the rotation observed between the closed and open complexes (60°) of PtxB (Fig. 6E). This shows that D206N HtxB crystallises in a partially open conformation. Considering the inability of D206N HtxB to bind phosphorus ligands in solution, along with the observation of this partially open confirmation, this suggests that the D206N mutant is unable to achieve a closed, ligand-bound state.
Superposition of the ligand binding domain (residues 95-204) from D206N HtxB with WT HtxB (Fig. 6B) shows that N206 points away from the ligand binding site in the mutant protein, rotated ca. 180° away from its position in the closed WT complex, and therefore, not in a position where it can interact with R178. We note that residue 206 sits on one of the two hinge regions of the protein (F96-K98 and P204-N206), suggesting it may have some flexibility. Closer comparisons of the D206N, D206A and WT HtxB structures (Fig. 6C,D), focusing on the conformation of the larger domain (superposition of the N-terminus to F96 and P204 to the C-terminus) shows key differences in the interactions that residue 206 and its neighbours make with the surrounding polypeptide. In the WT and D206A HtxB structures the carbonyl of residue 206 hydrogen bonds with the mainchain amide of V89 (3.0 Å), as part of the hydrogen bonding network across the anti-parallel beta-sheet in this domain. In the D206N HtxB structure, the sidechain oxygen of N206 sits within hydrogen bonding distance of the V89 amide (2.9 Å), altering the mainchain bond angles and rotating the N206 carbonyl by 50° away from its position in the WT/D206A HtxB structures. This change in mainchain conformation results in a 180° flip in the N205 carbonyl, requiring N205 to adopt a different rotamer in the D206N structure and leading to a new hydrogen bond formed with R178. R178 itself, also forms a hydrogen bond with the mainchain carbonyl of F96 (2.9 Å), which also sits in the hinge region. It seems likely that this rearrangement of the mainchain very close to the hinge region, together with the concomitant change in the hydrogen bond network across the domain interface might result in the D206N HtxB mutant being unable to form a closed conformation in complex with phosphorus ligands, explaining why no ligand binding could be detected for this mutant in our solution studies.

Discussion
PBPs exist in a state of dynamic equilibrium between open and closed conformations, facilitated by a highly flexible hinge region 32,33 . When a specific ligand is bound to the protein, a closed state is favoured and stabilised to allow for cargo to be shuttled to the transporter in the inner membrane. We have shown that HtxB is capable of binding phosphite at pH 5 due to a change in the protonation state of the ligand that facilitates the formation of extra hydrogen bonds within the binding pocket of the protein, enabling the formation of a closed complex. Although this complex can be captured by crystallisation with a large excess of phosphite (5 mM), the K d of HtxB for phosphite at pH 5 is only 1.74 mM, which is ca. 1000 × less than hypophosphite at the same pH. At neutral pH we could detect no binding of phosphite to HtxB, suggesting that HtxB is not a good phosphite transporter in conditions equivalent to those found in nature. In our structure at pH 5, the protonated phosphite forms a hydrogen bond with both D206 and R178. In complex with hypophosphite, neither of these residues form a direct interaction with the ligand, but instead form an intramolecular salt bridge across the domain interface. This generates one wall of the ligand binding pocket, delimiting a space adjacent to the ligand. Based on the much weaker binding of phosphite to HtxB at neutral pH, we therefore predict that binding of di-anionic phosphite would generate a sub-van der Waals distance between the third oxygen position and the D206/R178 salt bridge, producing a steric clash that would prevent closure of the domains around the ligand and lead to subsequent ligand release. It is only by prototonating the oxygen at this position by reducing the pH to below 6 that it can act as a hydrogen bond donor, forming a stabilising interaction with the residues of the salt bridge. However, given the weak binding, it is likely that the off-rate of the ligand is very rapid. Perturbation of this salt bridge by introducing a D206A or D206N mutations completely disrupts hypophosphite or phosphite binding to HtxB at pH 7.4 or 5. It is only by the addition of acetate that partial binding of hypophosphite can be restored in the D206A mutant, but with a 5000 × reduction in binding affinity compared to WT. This further suggests that the formation of the intramolecular salt bridge between D206 and R178 is critical for HtxB to bind ligands. Comparisons of the D206N structure, which crystallised in a partially open conformation, with open and closed complexes from closely related proteins suggests that D206 and R178 are positioned on flexible regions of the protein that are brought into close proximity during domain closure. This suggests that the residues surrounding the binding pocket are arranged in such a way that binding of the correct ligand is a requirement for domain closure; this is in addition to the selection of a P-H bond by the P-H…π interaction with the capping aromatic residue.
PBPs that have evolved to bind phosphite, hypophosphite and other reduced phosphorus compounds are structurally and sequentially closely related. The mode of binding of phosphite in complex with HtxB can therefore be directly compared with that of phosphite in the phosphite-binding PBP, P. stutzeri PtxB (PDB: 5O2J). This shows how an alternative positioning of the π system, which confers ligand specificity in both proteins (W52 in HtxB compared to Y203 in PtxB), to opposite sides of the binding pocket dictates that the R1 hydrogen of the phosphite must point in different directions (Fig. 7A,B). Thus, two of the phosphite-oxygen atoms have a conserved position in each protein, with the phosphite rotated ca. 180° around an axis that bisects the phosphorus center and lies equidistant between the two common oxygen atoms. The effect of this is that the ligands are bound as mirror-images of each other (Fig. 7D) and is an example of how subtle changes of sequence on a similar molecular scaffold can influence specificity dependent on the relative geometric positions of interacting atoms in the ligand.
Binding affinities for phosphite and hypophosphite in the nanomolar range are required in this family of transporter because ligand availability in natural environments is very low. We could not detect binding of phosphite to HtxB at neutral pH, which suggests that phosphite uptake is unlikely to be sustained solely by the HtxBCDE (2019) 9:10231 | https://doi.org/10.1038/s41598-019-46557-2 www.nature.com/scientificreports www.nature.com/scientificreports/ The superposition is carried out on the smaller of the two domains (residues 99-200 in HtxB), containing the important loop that recognises the oxygen atoms of the cargo (HtxB; N128-S130, PtxB; S124-S126).
www.nature.com/scientificreports www.nature.com/scientificreports/ transporter in typical aqueous conditions. However, recent work by Motomura et al. 15 , building on previous work from the same group in E. coli 13 , shows how the HtxBCDE transporter is capable of sustaining growth in supplemented phosphite conditions of a strain of the cyanobacteria, Synechococcus elongatus, that had been genetically modified to lack any alternative phosphorous import systems. The modified strain could not survive in simulated pond or fresh water media containing no phosphite, but could out-compete the WT strain in conditions containing 0.2 mM phosphite, whilst in the presence of 0.2 mM phosphate, the WT strain dominated the cultures and the modified strain perished. This conferred phosphite-dependency has therefore been proposed as a potential mechanism for biological containment of genetically engineered organisms. Based on the binding data and structures reported here, we suspect that the observed phosphite uptake by HtxB could be due to a slight acidification of the growth media under the experimental conditions used, enabling uptake of mono-anionic phosphite at www.nature.com/scientificreports www.nature.com/scientificreports/ concentrations consistent with the K d of 1.74 mM that we recorded. We propose that uptake of phosphite is likely to be better at a pH less than 6, which might allow growth on lower phosphite concentrations.
More broadly, we have shown that subtle changes to the binding site of HtxB (the difference of one or two hydrogen bonds) decrease binding affinity of ligands by 2-3 orders of magnitude, thus it is likely that specific point mutations around the third oxygen site could, with very subtle structural changes, increase the binding affinity of HtxB for phosphite. In an attempt to explore this further we have modelled the binding pocket from a sub-family of "HtxB-like" proteins from the soil-dwelling bacteria Bradyrhizobium sp. BTAi1, Methylopila sp. 73B and Lutibaculum baratangense AMV1 (ca. 50% sequence identity to HtxB), which also share some sequence similarities with PtxB around the ligand binding pocket. We have identified three positions around the ligand binding site that vary between HtxB and the HtxB-like proteins, which are N128 to serine (also serine in PtxB), F158 to histidine (also histidine in PtxB) and M176 to tyrosine or phenylalnine (serine in PtxB) (Fig. 7E). Each of these residue differences generate extra space in the binding pocket of the HtxB-like protein model, resulting in an extra 20 Å 3 of volume, compared to HtxB (model, 88 Å 3 ; HtxB, 68 Å 3 , calculated with CASTp 34 ), plus, the extra histidine moiety provides a hydrogen bond donor/acceptor within proximity to the ligand, suggesting that in the HtxB-like proteins, this histidine is likely to carry out a similar role as in PtxB, forming an interaction with bound ligand. As the pocket is both bigger, with extra capacity for hydrogen bonding, and still possesses the tryptophan-mediated P-H…π cap, it is likely, based in the understanding gained from this study, that it can bind phosphite. With further investigation, there could be scope to optimise binding of phosphite, while maintaining the exclusion of phosphate, to HtxB or this family of HtxB-like proteins, to optimise these transporters for bioengineering and biocontainment strategies using phosphite.
X-ray crystallography. 5 mM sodium phosphite or sodium hypophosphite was added to HtxB or each of the HtxB D206N/A mutants (10 mg/ml) and crystallised by sitting-drop vapor diffusion at 290 K (50 μl drop and 50 μl reservoir) using a Mosquito LCP crystallisation robot equipped with a humidity controlled stage (TTP Labtech) and commercial screens (Qiagen Nextal PACT and JCSG+). WT HtxB with phosphite crystallised in conditions containing 0.2 M magnesium chloride, 0.1 M Na acetate pH 5 and 20% (w/v) PEG 6000. D206N HtxB was crystallised in 0.2 M ammonium sulphate, 0.1 M Bis-Tris pH 5.5 and 25% (w/v) PEG 3350. D206A HtxB was crystallised in complex with hypophosphite in 0.1 M Bis-Tris pH 5.5 and 25% (w/v) PEG 3350. A single crystal of each was harvested and cryoprotected in its mother-liquor with an additonal 25% ethylene glycol prior to plunge-cooling with liquid nitrogen. Crystals were mounted (100 K) on beamlines i04-1, i04 and i03, at the Diamond Light Source, respectively, and data were collected at wavelengths of 0.92819 Å, 0.97951 Å or 0.9718 Å, respectively. Data were processed using the Xia2 pipeline 35 , which showed that the crystals belonged to either the spacegroup P212121 (WT and D206A HtxB) or P21212 (D206N HtxB) (See Table 1 for data collection statistics and cell dimensions). The structures were determined by molecular replacement with Phaser 36 using a monomer of HtxB (PDB:5ME4) as a search model. Model building was carried out in Coot 37 , refinement in Refmac5 (ccp4i) 38,39 and superposition with Superpose 40 (See Table 1 for refinement and validation statistics). Structural validation was carried out with Molprobity 41 . Coordinates and structure factors are deposited in the Protein Data Bank under accession codes 6EMN, 6GHT and 6GHQ. NMR. One dimensional 1 H and 31 P spectra to characterize the pH dependent binding between HtxB and phosphite were acquired using a Bruker 500 MHz Avance spectrometer equipped with a 5 mm broadband probe tuned to 202.45631 MHz for phosphorous, and 500.130 MHz for proton. A spectral width of 50 ppm centered at 0 ppm enabled the observation of the relevant phosphorous signals, while a spectral width of 25 ppm centered at 4.7 ppm for 1 H enabled the observation of all 1 H resonances. 1 H spectra were acquired using 1 H presaturation and a Hahn-Echo refocusing pulse prior to acquisition. Spectra were typically accumulations of 256 transients with an acquisition time of 0.65 s. Both decoupled and un-decoupled 31 P spectra were acquired using zgig and zg pulse programs respectively from the Bruker library. Spectra were typically accumulations of 512-1024 transients with a 10 s inter-scan delay and a 0.81 s acquisition time. Both 1 H and 31 P NMR spectra were processed using an exponential window function and 5 Hz linebroadening. A 100% D 2 O capillary with 200 mM phosphate in HEPES(K + ) buffer at pH 7.2 was included in the 5 mm NMR tube for the 31 P experiments recorded for HtxB, to act both as a reference, and as a quality control measure for the shimming of the probe.
Microscale thermophoresis. 20 μM protein was labeled with a two-fold excess of NT-647-NHS dye (NanoTemper Technologies, Munich Germany), following the manufacturer's instructions. Protein had a degree of labeling of 1 dye molecule per protein molecule. Binding affinity of proteins for ligands was determined using Microscale Thermophoresis (MST) with a Monolith NT.115 instrument (NanoTemper Technologies, Germany). 10 μL of 20 nM labelled protein was mixed with 10 μL of ligand in 40 mM citrate-phosphate, 250 mM NaCl, 0.05% Tween-20 pH 7.4 or 5.4 μL of protein-ligand mixture was loaded into "Premium Grade Capillaries" (NanoTemper Technologies) and thermophoresis was measured at 22 °C for 22 s with 40% LED power and medium thermophoresis power. Data from three independent measurements were combined and analysed using the MO.Affinity (2) MRW NMR figures. NMR spectra were output as ASCII files and represented in Igor. Chemical shifts and line broadening was estimated by fitting peaks to a Lorentizian function (Eq. 3).
where y 0 is the baseline, A is the peak amplitude, x 0 is the center of the peak and B is the Full Width Half Maximum. pK a values were calculated by fitting Eq. 4 to either chemical shift data from 31 P decoupled data, or coupling constants from 1 H and 31 P 1D spectra of phosphite peaks.

Data Availability
The atomic coordinates and structure factors are available via the PDB under accession codes PDB: 6EMN, 6GHT and 6GHQ.