A new family of periplasmic-binding proteins that sense arsenic oxyanions

Arsenic contamination of drinking water affects more than 140 million people worldwide. While toxic to humans, inorganic forms of arsenic (arsenite and arsenate), can be used as energy sources for microbial respiration. AioX and its orthologues (ArxX and ArrX) represent the first members of a new sub-family of periplasmic-binding proteins that serve as the first component of a signal transduction system, that’s role is to positively regulate expression of arsenic metabolism enzymes. As determined by X-ray crystallography for AioX, arsenite binding only requires subtle conformational changes in protein structure, providing insights into protein-ligand interactions. The binding pocket of all orthologues is conserved but this alone is not sufficient for oxyanion selectivity, with proteins selectively binding either arsenite or arsenate. Phylogenetic evidence, clearly demonstrates that the regulatory proteins evolved together early in prokaryotic evolution and had a separate origin from the metabolic enzymes whose expression they regulate.

. Gene arrangements for the arsenite oxidase (aio), alternative arsenite oxidase (arx) and respiratory arsenate reductase (arr) gene clusters. The aio represented is from Rhizobium sp. str. NT-26, arx from Alkalilimnicola ehrlichii str. MLHE-1 and arr from Chrysiogenes arsenatis. A: molybdenum-containing subunit; B: iron-sulphur protein; C: integral membrane subunit; R: response regulator; S: sensor histidine kinase; X: oxyanion-binding subunit. Putative promoter sites for RpoN are indicated, we could not detect a promoter upstream of arrAB. The genes that encode the alternative arsenite oxidase (arxB'ABC) together with the regulatory cluster (arxXSR) are conserved in anaerobic arsenite oxidisers, all of which are members of the Alpha-, Beta-and Gammaproteobacteria. The aioBA genes are more widespread, with homologues in both prokaryotic domains of life and in the following phyla of the Bacteria, Proteobacteria, Chloroflexi and Thermus-Deinococcus. The aioXSR regulatory module, however, is present only in some arsenite oxidisers in the Alpha-and Betaproteobacteria. In the arsenate respirers, the arrAB genes are present in members of the Firmicutes, Proteobacteria (Gamma-and Epsilonproteobacteria), Chrysiogenetes and Deferribacteres phyla while the arrXSR genes are present in all of these except the Gammaproteobacteria and the Deferribacteres.
Phylogeny and evolutionary history of AioX. Based on sequence comparisons, AioX appears to represent a new subfamily of the PBPs 25 where the majority of the proteins bind tetrahedral oxyanions, including phosphate, sulphate, molybdate and tungstate 26 (Fig. 3). The new subfamily is distinct from the other proteins as it functions in sensing and not oxyanion transport [26][27][28] . As expected, the closest homologues include ArxX and ArrX presumed to be involved in sensing arsenite and arsenate, respectively.
Three-dimensional structure of AioX. Structures of the apo, arsenite-bound and phosphate-bound forms of AioX were solved and deposited in PDB with the following IDs: 6ESK, 6EU7 and 6ESV, respectively. X-ray data and refinement statistics are presented in Table 1. AioX exhibits a typical bilobed structure of the PBP type II class 25 . The 304 amino acid residue polypeptide folds into two similar α/β domains. Each domain contains mixed beta-sheets, one with 6 strands and the other with 5; one strand is antiparallel to the rest in each domain (Fig. 4). The two domains are linked by two crossing-over protein strands, such that one links the beta strand 4 of the first domain to the beta strand 1 of the second domain, while the second crossover links the beta strand 5 of the second domain back to the beta strand 5 of the first domain. In addition, there is a C-terminal extension to the canonical fold, which is the sixth beta strand in the first domain and a C-terminal helix.
The AioX topology closely resembles that of the E. coli phosphonate-binding protein PhnD (Fig. 4). The NT-26 AioX and the E. coli PhnD proteins share 20% sequence identity over 245 aligned residues while the two structures superpose with a core rmsd of 1.84 Å, indicating close structural homology. In this class of proteins, the ligand-binding pocket is located in the space between the two domains.
In the initial single-wavelength anomalous diffraction (SAD) electron density maps generated from the crystals of Se-Met labelled protein there was an unaccounted electron density in the putative ligand-binding region near residue Cys106. The electron density was consistent with the presence of a phosphate ion, from the crystallisation conditions, and its location would suggest that Cys106 might be involved in binding arsenite.
Arsenite-bound AioX displays significant conformational changes. The crystal structure of the phosphate-bound AioX was used to solve the structures of the apo and arsenite-bound forms of the protein by molecular replacement. Figure 5 shows the AioX ligand-binding site with arsenite forming a link to the sulphur atom of Cys106, confirming the role of this residue in arsenite binding. Figure 5 also shows an overlay of the apo . Maximum-likelihood phylogenetic tree of AioX and its homologues. ABP, antigen-binding protein; AioX (red) and ArxX (blue), arsenite-binding proteins; ArrX, (green), arsenate-binding proteins; MBP, molybdate-binding proteins (a: archaeal; b: bacterial); PBP, phosphate-binding proteins (*involved in sensing not transport); PhnD, phosphonate-binding proteins; SBP: sulphate-binding proteins; TBP: tungstate-binding protein.
and the arsenite-bound forms of AioX accentuating the most pronounced differences between the two structures. While the majority of both structures overlay very well, there is a significant difference in the rotamer conformation of Tyr88. The change in the side chain conformation is further associated with the backbone shift for residues 83-88 and a significant conformational change in the loop region comprising residues 53-61, demarked by the black circle in Fig. 5A. This conformational change upon arsenite binding involves a hydrophobic collapse Effective resolution of this data set based on CC 1/2 and Mean((I)/sd(I)) is1.85 Å and 1.97 Å, respectively. R merege at 1.97 Å is 0.267. All data was used in the refinement. . Structure of AioX superimposed with that of PhnD. The bilobal structure of AioX is shown in ribbon representation (yellow) and compared to that of the E. coli protein PhnD in complex with its ligand 2-aminoethyl phosphonate (blue ribbons and blue sticks, respectively). The AioX structure shown is that of the SAD method solved structure, which contains phosphate in its binding pocket. Two conserved tyrosine residues (Y88 and Y131), AioX-specific residue C106, and a phosphate ion are shown as sticks and coloured based on the type of the atom (phosphorous-orange, oxygen-red, nitrogen-blue, and carbon-grey).
SCIentIfIC REPORTS | (2018) 8:6282 | DOI:10.1038/s41598-018-24591-w of the loop, with residues Val56, Phe57 and Leu58 reaching towards the core of the protein and closing of the ligand-binding site. Importantly, the conformation of Tyr88, seen in the apo form of AioX and the conformation of the 53-61 loop, observed in the arsenite-bound AioX, are mutually exclusive, as the conformations of these two parts of the structure would have caused a steric clash. Furthermore, conformational rearrangement upon arsenite binding would inevitably result in changes to the molecular surface of AioX, including its electrostatic potential (Fig. 6). Interestingly, while the structures of apo-AioX and the phosphate-bound AioX contained a monomer in the asymmetric unit, the arsenite-bound molecule crystallised with six molecules in the asymmetric unit of the unit cell. Although the six molecules were arranged as the three repeats of a loosely associated dimer none of the interfaces is sufficiently large to be indicative of a biologically significant assembly. It should be noted that while surface residues 59-61 in the arsenite-bound structure are near the neighbouring molecule in the asymmetric unit, they are too distant to be directly involved in the interactions, while in the apo-AioX Asp61 interacts with the side chain of Arg281 from the crystallographic symmetry related molecule. These small differences in molecular packing support the idea that 53-61 loop conformation change might result in the generation of a new protein interaction surface.
Phosphate-bound and arsenite-bound forms of AioX are not identical. Superposition of the phosphonate-bound structure of PhnD and the phosphate-bound structure of AioX emphasizes the structural similarity between the two proteins, while close inspection of the ligand-binding sites identifies a set of conserved residues that are engaged in interactions with two similar ligands (Figs 4 and 5A). However, the two binding sites are not identical and while one side of the binding pocket contains identical residues, Ser161, Ser163, His192, Tyr88 and Tyr131 specifically (AioX numbering), the other residues in the binding pocket appear to provide substrate specificity. In PhnD, the phosphonate molecule forms an additional hydrogen bond with the side chain of Thr128 (Asn162 in AioX) and a water-mediated interaction with Ser68, while the carboxylate groups of Glu177 and Asp205 interact with the positively charged end of the 2-aminoethyl phosphonate (Tyr212 and Pro240 are present in the corresponding locations in AioX). On the other hand, a phosphate ion found in the binding site of AioX formed AioX-specific hydrogen bonds with Asp210 and Cys106, which in PhnD correspond to Asn175 and Gly65, respectively.
The presence of a phosphate ion in the binding site of AioX leads to the hypothesis that the phosphate ion mimics arsenite binding. A comparison of the X-ray structures however, indicates that while AioX can bind phosphate, the resulting structure does not stabilize the 53-61 loop conformation seen in the arsenite-bound form. Furthermore, even though the structure is similar to the arsenite-bound form, with the main conformation of Tyr88 akin to that of the arsenite-AioX (Fig. 7), electron density for residues 53-61 are ambiguous and due to the disorder, we were unable to confidently build a model for residues 55-57. Within region 55-57 the structure had more similarity to the apo rather than the arsenite-bound form of AioX.
Conservation of AioX-binding pocket in close homologues. The residues that constitute the binding pocket of the arsenite-and phosphate-bound structures of AioX, are Tyr88, Cys106, Tyr131, Ser161, Ser163, His192 and Asp210. Alignment of AioX with ArrX, ArxX and PhnD sequences (Fig. 8), shows that residues are conserved in the AioX subfamily of PBPs, indicating similar substrate-binding capabilities. The alignment also highlights the absence of Cys106 in PhnD. The hydrophobic residues which reach towards the core of the protein in arsenite-bound AioX are also present in Arx except Phe57 is an Ile (also hydropobic) in Arx (Fig. 8). However, the corresponding loop regions in ArrX and PhnD are shorter and more polar with AioX-like residues Val56 and Phe57 missing from that part of the sequence (Fig. 8).
Oxyanion-binding studies demonstrate substrate specificity. Although the residues involved in binding arsenite and phosphate are conserved in the three orthologues, AioX, ArxX and ArrX (Fig. 8), isothermal titration calorimetry (ITC) experiments indicated distinct substrate specificity of these proteins (Fig. 9). We tested a range of oxyanions including antimonite, arsenate, arsenite, molybdate, phosphate, phosphonate, selenate, sulphate, tetrathionate and thiosulphate as potential ligands for each protein. AioX and ArxX bound arsenite with a K d of 177 and 311 nM, respectively ( Fig. 9) and no interaction was detected with any of the other substrates tested (data not shown). It was surprising not to detect phosphate binding to AioX or ArxX given, (1) we solved a structure of AioX with a phosphate ion in the binding pocket (Fig. 4) and (2) that amino acids involved in phosphate binding in the structure of AioX are conserved in ArxX. As we were unable to detect any binding interaction, based on the conditions used in the ITC we estimate that the K d for phosphate would have to be greater than 2 mM (data not shown). ArrX bound arsenate and also phosphate with a K d of 1 and 150 μM, respectively (Fig. 9) while no interaction was detected with any of the other ligands tested (data not shown). Binding of phosphate to ArrX is not surprising given that arsenate and phosphate have analogous chemical structures. Interestingly, ligand-binding affinities for this protein are in micromolar compared to the nanomolar range observed for AioX and ArxX indicating complex tuning of the ion-binding mechanisms, probably involving other structural features beyond direct interactions with the protein side-chains. We used a one-site binding model to fit the data for ArrX with arsenate, as its homology with AioX indicates a single binding pocket, even though the data exhibits a more complex behaviour. A two-site model does fit the data, however the K d values are not replicable between titrations. At pH 8 used in the ITC experiments, the arsenate acquires two forms, H 2 AsO 4 − and HAsO 4 2− , which could account for different K d values in the same titration. Alternatively, the arsenate titration thermogram might include a component resulting from the ligand-induced protein-protein interactions. In summary, at physiological concentrations, AioX, ArxX and ArrX bind the substrate that their cognate molybdenum-containing enzyme metabolises.

Discussion
Structure and potential signalling mechanism of AioX. AioX is a class II PBP 25 and a member of the D-cluster of substrate-binding proteins 26 . Proteins of this type predominantly bind tetrahedral oxyanions 26 and while the binding of arsenate and phosphate by ArrX conforms to this trend, the binding of arsenite by AioX and ArxX is different as arsenite has a trigonal pyramidal geometric structure. The AioX orthologues are the only members of this family that function as signalling proteins. The other members of the family, including the structural homologue PhnD, are part of ABC transporters. Other PBPs shown to be involved in signal transduction include LuxP and TorT, which bind autoinducer-2 29 and trimethylamine 27 and regulate the lux 30 and tor 28 operons, respectively. LuxP and TorT however are structurally and phylogenetically distinct to AioX, indicating independent evolutionary origins.
The arsenite-bound crystal structure of AioX showed significant conformational changes compared with the apo and phosphate-bound structures, specifically in the loop of residues 53-61 and the shape, as well as electrostatic potential, of the protein surface. It is possible that these modifications allow its interaction with the sensor histidine kinase, AioS, leading to its activation. The fact that the arsenite-bound and apo-AioX crystals displayed differences in molecular packing (forming weak local and crystallographic symmetry related dimers, respectively) supports the idea that loop conformation change might result in the generation of a new protein interaction surface.

Substrate specificity. Binding experiments performed by ITC showed that AioX and ArxX could only bind
arsenite from the potential ligands selected, while Arr bound arsenate and phosphate (although the latter with a 150 times higher K d ). Phosphate concentrations that organisms encounter will depend on the environment. A. ehrlichii MLHE-1 is from an environment extremely rich in phosphate (up to 1 mM) 31 , while gold mine tailings often don't contain enough phosphate to support microbial growth 32 , and eutrophied lake sediments may contain  Residues that bind phosphate in AioX are in bold; residues that bind arsenite in AioX are highlighted in yellow; residues that bind phosphonate in PhnD are denoted *; putative signaling loop in AioX is underlined; AioX numbering. Note: there is only one residue that binds phosphate but not arsenite, and one residue that binds phosphonate in PhnD that is not conserved in AioX/ArrX/ArxX. 7 µM phosphate 33 (both Rhizobium sp. str. NT-26 and C. arsenatis were isolated from mining environments). A phosphate ion was identified in the crystal structure of AioX, however, its binding does not stabilize the putative signalling loop as is the case with arsenite binding. As AioX and ArxX did not bind phosphate in conditions tested in ITC, we conclude that AioX and ArxX do not bind phosphate under biologically relevant conditions.
The binding residues involved in arsenite binding in AioX (and presumably ArxX) are also conserved in ArrX, which did not bind arsenite but did bind arsenate and phosphate albeit the latter with about 150-fold less affinity. This finding indicates that binding residues alone do not confer ligand specificity but other factors, such as overall conformation of the protein, may play a role in selectivity. The K d of ArrX for phosphate indicates that in some high phosphate environments expression of the arsenate reductase genes may be inhibited.

Evolution of the XSR regulatory cluster.
Based on phylogenetic studies, some molybdenum-containing enzymes were present in the last universal common ancestor 34 and it is hypothesised that Aio is the common ancestor to all other members of the DMSO reductase family of molybdoenzymes 34,35 . Studies have also shown that while respiratory molybdoenzymes are principally inherited by vertical gene transfer, they are also subject to duplication events and divergence, particularly in the Firmicutes and Proteobacteria phyla 36 . Such a scenario is apparent for divergence of Arx and Arr, where the 'older' enzyme Arr, which uses menaquinone to reduce arsenate, has diverged to form Arx which uses ubiquinone to oxidise arsenite 37 .
The congruency of the AioX, AioS and AioR phylogenies to the 16S rRNA gene indicates that the regulatory genes emerged early in prokaryotic evolution and their functions adapted to that of the molybdoenzymes, whose gene expression they regulate. This is supported by the similar GC content of the regulatory genes and the molybdoenzyme genes (data not shown). Adaptation to conform to the function of a cognate molybdoenzyme is also observed in the NarJ family of proteins, which function as chaperones and are essential for cofactor insertion in several families of molybdoenzymes 38 .
The aioXSR genes are present in some arsenite oxidisers of the Alpha-and Betaproteobacteria. In Polaromonas sp. str. GM1, the Aio is constitutively expressed and a putative phage integrase is present immediately upstream of aioB suggesting that this insertion event may have led to the deletion of the aioXSR genes 39 34,35 which is the case for B. cenocepacia but not the other Burkholderia species. In the other Burkholderia species, the AioA subunit clusters with six Pseudomonas (Gammaproteobacteria, which lack aioXSR) species. Transposases and integrases are located within 10 kb of the aioBA genes in these Burkholderia species, which may indicate loss and horizontal gene transfer of the aioBA genes from Pseudomonas to Burkholderia accounting for the absence of aioXSR in those species.
The XSR gene cluster is conserved in all arx-containing organisms but it appears that it has been lost in some that contain aio. Phylogenetic analyses of the arrAB genes suggests both vertical and horizontal transfer of these genes 35 and it's possible that the regulatory genes were lost after such a horizontal gene transfer event.
In conclusion, we propose that the protein AioX represents a new subfamily of PBPs that bind arsenic oxyanions.

Methods
Cloning of aioX and its orthologues. The aioX gene from Rhizobium sp. str. NT-26 and its orthologues, arxX from A. ehrlichii str. MLHE-1 and arrX from C. arsenatis were cloned without their signal sequences into pPROEX-HTb+ (Invitrogen) for cytoplasmic protein expression in E. coli str. JM109. The aioX and arrX genes were amplified from genomic DNA using PCR with the following primers: aioXF, aioXR, arrXF and arrXR (Supplementary Table S1) whereas a codon-optimised version of the arxX gene was synthesised ( Supplementary  Fig. S1, GeneArt, ThermoFisher Scientific) and then cloned into pPROEX-HTb+. Expression constructs were confirmed by sequencing (GATC Biotech).

Heterologous expression and purification of AioX and its orthologues. AioX, ArxX and ArrX
were heterologously expressed in 1 l lysogeny broth (LB) (in a 5 l flask) containing 100 µg/ml ampicillin and 40 µM isopropyl β-D-1-thiogalactopyranoside (IPTG) at 21 °C with shaking at 180 rpm for 20-24 h. Recombinant proteins were purified from JM109λpir using affinity and size-exclusion chromatography. Cells were harvested by centrifugation at 5000 × g for 10 min (4 °C) and the pellets suspended in 40 ml binding buffer (500 mM NaCl, 20 mM potassium phosphate, 20 mM imidazole, pH 7.3). Cells were centrifuged at 16,000 × g for 10 min (4 °C), suspended in 10 ml/gm (wet weight) binding buffer and broken at 14,000 PSI in a pressure cell homogenizer (Stansted Fluid Dynamics). Debris and unbroken cells were removed by centrifugation at 30,000 × g for 30 min (4 °C). Total cell extract was applied to a 1 ml His Gravitrap (GE Healthcare) Ni-charged column which was then washed with 100 ml binding buffer. Bound proteins were eluted from the column in 5 ml elution buffer (500 mM NaCl, 20 mM potassium phosphate, 200-500 mM imidazole, pH 7.3). The eluate was filtered through a 0.22 µm filter (Millipore), concentrated using a 10,000 MWCO centrifugal filter (Millipore) and loaded onto a Superdex 75 gel filtration column (GE healthcare) pre-equilibrated with either 20 mM Tricine, pH 7.5 (AioX and ArxX) or 50 mM Tris-HCl, pH 8 (ArrX). Chromatography was carried out at 0.8 ml/min and fractions corresponding to the protein peaks were pooled and concentrated with the 10,000 MWCO centrifugal filter (Millipore), snap frozen in liquid nitrogen and stored at −80 °C until required. Protein concentrations were determined with the absorbance at 280 nm, measured with a NanoDrop 2000 spectrophotometer (Thermo), and a predicted molar absorbance coefficient (ExPASy, Swiss Institute of Biofinformatics) 41,42 ; ε (M −1 cm −1 ) for AioX, 42860, ArxX 44350, ArrX 25330. Protein purity was confirmed by sodium dodecyl sulphate-polyacrylamide gel electrophoresis.
For crystallisation, the His-tag was removed from AioX using AcTEV protease (Invitrogen) according to the manufacturer's instruction. The cleaved AioX was then separated from uncleaved protein and the protease using affinity chromatography as described above.
SCIentIfIC REPORTS | (2018) 8:6282 | DOI:10.1038/s41598-018-24591-w Selenomethionine labelling of the recombinant NT-26 AioX. The sulphur atoms of methionine residues in AioX were replaced with selenium by providing selenomethionine in the growth medium during protein expression using a modified protocol 43 . JM109 harbouring pPROEX-HTb + -aioX was grown at 37 °C overnight in 50 ml of M9 medium 44 containing ampicillin (100 μg/ml). The 50 ml was inoculated into 1 l M9 medium and incubated at 37 °C with shaking (170 rpm). Cells were grown until OD 600 nm ~0.5 followed by the addition of an amino acid mix consisting of 100 mg each of lysine, threonine, phenylalanine; 50 mg each of leucine, isoleucine, valine and 60 mg of L-selenomethionine and the cells grown for 15 minutes prior to induction with IPTG. The SeMet-AioX protein was purified as described above with one modification, all buffers contained 5 mM of dithiothreitol.
Crystallisation of NT-26 AioX, data collection and structure determination. For crystallisation of the SeMet-AioX, apo-AioX and AioX-arsenite, the recombinant AioX protein was concentrated to 20 mg/ml. Crystallization screening was performed at 16 °C, using a mosquito (TTP Labtech) and the crystallisation screens from Hampton Research (Index) and Molecular Dimensions (Structure 1&2, PACT and JCSG). Conditions that yielded crystals were as follows: (1) SeMet-AioX crystals grew in 0.49 M NaH 2 PO 4 and 0.91 M K 2 HPO 4 pH 7.2, (2) apo-AioX crystals grew in 20% PEG 3350, 0.2 M sodium malonate and 0.1 M Bis-Tris propane, pH 8.5 and, (3) AioX-arsenite crystals grew in 25% PEG 3350 and 0.1 M Tris-HCl, pH 7.6. Single crystals were collected and mounted in a nylon cryo-loop using 25% glycerol as a cryo-protectant and flash cooled directly in liquid nitrogen prior to data collection.
Crystals were screened and X-ray diffraction data collected at the following synchrotron radiation sources: Diamond Light Source, beamlines I04 and I02 and Soleil synchrotron, beamline Proxima 1. The diffraction images were integrated and scaled using the XDS software package and merged using Aimless in the CCP4 program suite 45 . SAD data were collected from one SeMet-AioX crystal and processed with CCP4 software. The MTZ reflection file obtained was used to determine the position of five selenium atoms through SHELX C/D/E (CCP4). BUCCANEER was used to generate the AioX model. The final AioX model was generated through interactive cycles of model building in Coot 46 and a restrained refinement routine in REFMAC5 47 . The structures of apo-AioX and arsenite-bound AioX were solved using molecular replacement using the phosphate-bound structure as the model, but with the phosphate ion omitted. Phases were obtained with PhaserMR 48 and model building was performed in Coot 46 . Refinement of the new structure was carried out in REFMAC5 47 . The data collection, processing and refinement statistics are listed in Table 1. All figures of crystal structures and molecular surface calculations were prepared using PyMOL 49 . The following PDB entries were used for structural comparisons: 3N5L (a binding protein component of an ABC phosphonate transporter from Pseudomonas aeruginosa) and 3P7I (a PhnD in complex with 2-aminoethyl phosphonate from E. coli) 50 . Three structures and the associated structure factors were submitted to PDB with the following IDs: 6ESK, 6EU7 and 6ESV for apo, arsenite-bound and a phosphate-bound AioX.
Substrate-binding thermodynamics determined by isothermal titration calorimetry. Binding of potential substrates by AioX, ArxX and ArrX was determined by ITC using a MicroCal200 ITC instrument (Malvern) at 25 °C with constant stirring of 750 rpm. Binding parameters were calculated using Microcal Origin software with a one-site binding model to give estimated values of N, K d , ΔH, and ΔS. Antimonite (antimonyl tatrate; C 8 H 10 K 2 O 15 Sb 2 ), arsenate (Na 2 HAsO4), arsenite (NaAsO 2 ), molybdate (Na 2 MoO 4 ), phosphate (KH 2 PO 4 ), phosphonate (2-aminoethyl phosphonic acid; PO 3 H 2 (CH 2 ) 2 NH 2 ), selenate (Na 2 SeO 4 ), sulphate (Na 2 SO 4 ), tetrathionate (K 2 S 4 O 6 ) , thiosulphate (Na 2 S 2 0 3 ) were tested as potential substrates by titrating 1-10 mM substrate into 100 µM protein in 1.5-2 µl injections. Protein-substrate combinations that tested positive for binding were repeated with three independent protein preparations. AioX and ArxX titrations were performed in 20 mM Tricine, pH 7.5 and ArrX in 50 mM Tris-HCl, pH 8. Substrates were prepared in the same buffer as the protein into which they were titrated. Phylogenetic analysis. Sequences from confirmed arsenic metabolisers and other representatives were retrieved from GenBank using BLAST 51 with NT-26 sequences as queries. Phylogenetic analyses were performed using MEGA 52 . Multiple sequence alignments were generated using MUSCLE 53 . Phylogenetic trees were constructed with the maximum-likelihood method, using the Jones-Taylor-Thornton model, and tested by 200 bootstrap resamples. Accession numbers used to construct trees are in Supplementary Table S2.
Data availability statement. All data generated or analysed during this study are included in this published article (and its Supplementary Information file).