Insights into Phosphate Cooperativity and Influence of Substrate Modifications on Binding and Catalysis of Hexameric Purine Nucleoside Phosphorylases

The hexameric purine nucleoside phosphorylase from Bacillus subtilis (BsPNP233) displays great potential to produce nucleoside analogues in industry and can be exploited in the development of new anti-tumor gene therapies. In order to provide structural basis for enzyme and substrates rational optimization, aiming at those applications, the present work shows a thorough and detailed structural description of the binding mode of substrates and nucleoside analogues to the active site of the hexameric BsPNP233. Here we report the crystal structure of BsPNP233 in the apo form and in complex with 11 ligands, including clinically relevant compounds. The crystal structure of six ligands (adenine, 2′deoxyguanosine, aciclovir, ganciclovir, 8-bromoguanosine, 6-chloroguanosine) in complex with a hexameric PNP are presented for the first time. Our data showed that free bases adopt alternative conformations in the BsPNP233 active site and indicated that binding of the co-substrate (2′deoxy)ribose 1-phosphate might contribute for stabilizing the bases in a favorable orientation for catalysis. The BsPNP233-adenosine complex revealed that a hydrogen bond between the 5′ hydroxyl group of adenosine and Arg43* side chain contributes for the ribosyl radical to adopt an unusual C3’-endo conformation. The structures with 6-chloroguanosine and 8-bromoguanosine pointed out that the Cl6 and Br8 substrate modifications seem to be detrimental for catalysis and can be explored in the design of inhibitors for hexameric PNPs from pathogens. Our data also corroborated the competitive inhibition mechanism of hexameric PNPs by tubercidin and suggested that the acyclic nucleoside ganciclovir is a better inhibitor for hexameric PNPs than aciclovir. Furthermore, comparative structural analyses indicated that the replacement of Ser90 by a threonine in the B. cereus hexameric adenosine phosphorylase (Thr91) is responsible for the lack of negative cooperativity of phosphate binding in this enzyme.

NAs can be used in the treatment of a range of human viral infections, such as those caused by HIV, herpesvirus and hepatitis B/C virus [15][16][17][18][19]. They are among the first cytotoxic molecules to be used in the treatment of cancer [20] and have been studied as potential drugs against tuberculosis [21,22], malaria [7,23], trichomoniasis [24] and schistosomiasis [25]. The chemical synthesis of these compounds is generally a costly multistep process that includes several protection and deprotection stages [13,26]. This has encouraged the development of new methods for the synthesis of NAs using PNPs and other enzymes as biocatalysts [13,27,28]. The main advantages of this approach are the higher stereospecificity, regioselectivity and efficiency of enzymes, whose employment usually dispenses group protection and purification steps, optimizing the process [13].
The differences in substrate specificity regarding trimeric and hexameric PNPs have allowed the development of suicide gene therapies strategies against solid tumors [14,29]. Trimeric PNPs are mainly found in mammalian species and are specific for guanine and hypoxanthine (29-deoxy)ribonucleosides whereas hexameric PNPs are prevalent in bacteria and accept adenine as well as guanine and hypoxanthine (29-deoxy)ribonucleosides as substrates [1]. Thus, nontoxic adenosine analogues, which are poor substrates for human PNP, can be cleaved to cytotoxic bases specifically in tumor cells transfected with the bacterial hexameric PNP gene [14]. Main advances in this field have been achieved with the E. coli PNP [30][31][32][33].
In this context, the aim of the present work was to shed light on how a diverse set of substrate modifications affects its binding and catalysis by hexameric PNPs using a structural approach. For this purpose, we choose the hexameric PNP (BsPNP233) from the model specie Bacillus subtilis, which displays great biotechnological potential to produce NAs, including the antiviral drug ribavirin [34]. We have solved the crystal structure of BsPNP233 in the apo form and in complex with 11 ligands comprising sulfate, bases, natural nucleosides and NAs, including clinically relevant compounds. The crystal structure of six ligands (adenine, 29deoxyguanosine, aciclovir, ganciclovir, 8-bromoguanosine, 6-chloroguanosine) in complex with a hexameric PNP are presented for the first time.
Besides providing a broad structural basis for studies aiming at the rational design of BsPNP233 and its homologues for biotechnological applications, this work also bring new insights into the distinct kinetic models for phosphate binding in hexameric PNPs. Furthermore, the structural information showed here may also be instrumental for the development of new inhibitors against hexameric PNPs from pathogens such as Plasmodium falciparum and Trichomonas vaginalis [5,6,8,9] and for the combined design of both hexameric PNPs and prodrugs to improve specificity and efficiency of anti-cancer PNP gene therapies [14].

Expression and Purification of Recombinant BsPNP233
BsPNP233 was expressed in E. coli cells and purified by immobilized metal affinity and size-exclusion chromatographies as described in [35]. The protein concentration was determined by absorption spectroscopy at 280 nm using the theoretical molar extinction coefficient of 16 515 M 21 cm 21 calculated by the program ProtParam [36].

Crystallization
BsPNP233 at 11 mg/ml in 20 mM Tris-HCl pH 7.0, 50 mM NaCl and 1 mM DTT was crystallized by sitting-drop vapordiffusion technique according to conditions previously described [35]. The crystals belong to the space groups P32 1 , P6 3 22, P2 1 2 1 2 1 and H32 with one, two or six monomers per asymmetric unit depending on symmetry and cell dimensions.

Preparation of BsPNP233-ligand Complexes
The protein-ligand complexes were prepared by adding 0.1 ml of 50 mM ligand, dissolved in DMSO, to 1 ml crystallization drops at least 12 h prior to data collection. The ligands used were nucleosides, purine bases and NAs (Table S1). This procedure was performed in drops containing BsPNP233 crystals grown in 0.1 M sodium acetate pH 4.6, 3.2 M sodium chloride, 5% (v/v) glycerol at 291 K.

X-ray Data Collection and Processing
X-ray diffraction experiments were performed on the W01B-MX2 beamline at the Brazilian Synchrotron Light Laboratory (Campinas, Brazil). The data collection was carried out using crystals soaked in a cryoprotectant solution composed by the mother liquor and 20% (v/v) glycerol and flash-cooled in a nitrogen-gas stream at 100 K. The radiation wavelength was set to 1.458 Å and a MAR Mosaic 225 mm CCD detector was used to record the X-ray diffraction data. Data were indexed, integrated and scaled using the HKL-2000 suite [37] or the programs MOSFLM [38] and SCALA [39] from the CCP4 package [40]. Data processing statistics are summarized in Table S1.

Structure Determination and Refinement
The structures were solved by molecular replacement using the programs MOLREP [41] or PHASER [42], both from the CCP4 suite [40]. The first BsPNP233 structure was determined using the atomic coordinates of B. anthracis PNP (PDB code 1XE3) [43] as a search model. The subsequent BsPNP233 structures were solved using the atomic coordinates of BsPNP233 solved at 1.7 Å resolution (BsPNP233-GCV dataset, Table S1) as template. Refinement was carried out using the programs REFMAC5 [44] and COOT [45]. After 20 cycles of rigid body refinement in REFMAC5 [44], the models were refined alternating cycles of restrained isotropic refinement in REFMAC5 [44] and manual rebuilding and real space refinement in COOT [45]. Water molecules were added after refinement of the protein model at chemically reasonable places using COOT [45]. Subsequently, the ligands were added to the model and refined as described above using library descriptions generated by the program SKETCHER from the CCP4 suite [40]. The intensity based twin refinement of REFMAC5 was applied to refine the structures of BsPNP233 in complex with adenosine, 2-fluoradenosine and adenine. The majority of models for the BsPNP233 protein included all but the first and last residues (1 and 233). In the electron density map of the crystal structure solved in the space group P2 1 2 1 2 1 the residue 1 and additional eight residues from the N-terminal his-tag were clearly defined and added to the model. Ramachandran analysis carried out by Molprobity [46] showed that all residues from all models are found in allowed regions (except Gly 121 of the BsPNP233-Ade structure, chain B). Refinement statistics are detailed in Table S1. Weighted 2Fo-Fc maps (2mFo-DFcalc) of ligands as well as a table of interactions between ligands and protein residues are presented in the supplementary material ( Figure S1, Table S2

Figure Preparation
The figures of structures were prepared using PyMOL [47].

Structural Alignment
All structural comparisons were performed using the SSM algorithm [48] available at the program COOT [45] or at the PDBeFold server [49].

Results and Discussion
BsPNP233 Conserves the Quaternary Structure and Topology of Hexameric PNPs in four distinct space groups (P32 1 , P2 1 2 1 2 1 , P6 3 22 and H32). The crystal contacts are similar in the crystal structures solved in P32 1 , P2 1 2 1 2 1 and P6 3 22 but differ in the H32 space group. In the later, we observed additional crystallographic interfaces, resulting from a more compact crystal packing with a lower solvent content (41%) than crystals belonging to other space groups (,56%) ( Figure S2) [35].
The BsPNP233 subunits surround a central axis and alternate in an up/down fashion forming a disc-shaped structure with six active sites: three located at the top face and other three at the bottom face. Analogously to other hexameric PNPs, BsPNP233 is a trimer of dimers where each subunit interacts with the adjacent subunits forming two interfaces: the catalytic, which contains the active site, and the inter-dimeric, involved in hexamer stabilization ( Figure 1A). The inter-dimeric interface is larger than the catalytic interface and both are mainly maintained by hydrophobic interactions. In the ligand-free crystal structure (form II), the inter-dimeric and the catalytic interface areas are 1711 and 1554 Å 2 , respectively.
The BsPNP233 subunit conserves the E. coli hexameric PNP (EcPNP) subunit topology with few exceptions. Its central mixed bsheet lacks the short b10 strand observed in EcPNP [50] and is surrounded by eight (instead of seven) a-helices ( Figure 1B). The extra 5-residues a-helix connects the strands b2 and b3 and is not labeled to preserves the Mao and colleagues notation [50]. BsPNP233 and EcPNP subunits share sequence identity of 58% (PDB code 1ECP, [50]) and superpose with a r.m.s.d of 0.93 Å for 231 Ca atoms aligned ( Figure S3). Structural alignment of BsPNP233 subunit with hexameric PNPs subunits from other Bacillus species resulted in a r.m.s.d of 0.80 Å -0.94 Å for 231 Ca atoms aligned and an average sequence identity of 71% ( Figure  S3).
Analysis of the B-factor distribution in the apo BsPNP233 crystal structure shows that the loop connecting b9 and a7 as well as the N-terminal portion of a7 present the highest B-factor values, highlighting its intrinsic flexibility. As this region surround the active site, its flexibility may be important for catalysis ( Figure 1C).

Free Purine Bases Adopt Alternative Conformations in the Active Site
The crystal structures of BsPNP233 in complex with hypoxanthine (Hyp) and adenine (Ade) showed that the purine-binding site consists of residues Cys 91 , Gly 92 , Phe 159 , Val 177 and Met 179 . Hydrophobic interactions are predominant in the stabilization of both ligands (Figure 2A).
The BsPNP233-Ade binary complex was solved with (BsPNP233-Ade-SO 4 ) or without (BsPNP233-Ade) sulfate ion and represent the first of their kind to be reported for hexameric PNPs. Two subunits were observed in the asymmetric unit of both crystal structures and all of them exhibited clear density for the ligand in the active site ( Figure S1).
Superposition of BsPNP233-Ade and BsPNP233-Ade-SO 4 complexes showed a preferential orientation of Ade in the basebinding site, except in one case where it is rotated by 49u around an axis perpendicular to the base plane ( Figure 2B). This alternative orientation is not followed by significant conformational changes in the active-site residues ( Figure 2B); however, it alters the solvation of the active-site pocket. In the alternative orientation, a crystallographic water molecule in the ribosebinding site is absent. This solvent molecule mediates a hydrogen bond between the AdeN 9 atom and the carbonyl group of Ser 90 in the presence of sulfate ion ( Figure 2B).
Interestingly, the Hyp adopts an orientation similar to the alternative conformation of Ade ( Figure 2C). In this case, a glycerol molecule is located in the ribose-binding site and seems to induce the displacement of Hyp, avoiding a steric clash with the HypN 9 atom. This observation, along with those described above, suggests that binding of the co-substrate ribose-1-phoshate might contribute for stabilizing the base in the favorable orientation for catalysis.
The Hydrogen Bond between the 59 Hydroxyl Group of Ado and Arg 43* Side Chain Contributes for a Ribosyl C39endo Conformation The base moiety of adenosine (Ado) binds to the BsPNP233 active site in a very similar fashion to that seen in homologous PNPs ( Figure 3). However, the ribosyl group adopts a C39-endo form instead of the C49-endo or O49-exo conformations usually observed in Ado complexes with hexameric PNPs (PDB codes Figure 1. Overall structure of BsPNP233. A. Cartoon representation of the hexamer BsPNP233 with adenine (red spheres) bound in the active site. Solid and dashed grey arrows indicate the inter-dimeric and catalytic interfaces, respectively. B. Cartoon representation of BsPNP233 protomer in complex with adenosine (green stick). Loops, a-helices and b-strands are shown in yellow, blue and pink. The a-helices and bstrands were numbered according to the Mao and colleagues notation [50]. C. BsPNP233 protomer colored by B-factors from dark blue (lowest) to red (highest). Adenosine is represented by a black stick. doi:10.1371/journal.pone.0044282.g001 3UAW, [52]; 1ODI, [53]; 1PK7, [54]; 3U40, [55]; 1Z37, [51]; 1VHW, [56]) ( Figure 3). This unusual conformation may be explained by a hydrogen bond between the 59 hydroxyl group of Ado and Arg 43* side chain (residues from the adjacent subunit are designated by an asterisk) not observed in other Ado complexes ( Figure 3). Typically, the 59-OH group of Ado is found interacting with one or two water molecules not observed in BsPNP233-Ado complex, suggesting that the hydration of the active site may influence the ribosyl conformation.
In sulfate/phosphate free complexes of hexameric PNPs with Ado, an O49-exo conformation is usually found. However, in all complexes where the phosphate-binding site is occupied by a sulfate or phosphate, the ribosyl group shows a C49-endo conformation, except for the Thermus thermophilus (TtPNP)-Ado complex (PDB code 1ODI, [53]). Since the side chain of Arg 43* participates in phosphate binding, the presence of this co-substrate and the hydration of the active site probably prevent the interaction between Ado 59-OH group and Arg 43* side chain observed in BsPNP233-Ado complex favoring the ribose to adopt a C49-endo conformation.

29-deoxyguanosine Binding Mode Resembles to that Observed for Adenosine
In the BsPNP233-(29-deoxyguanosine) complex structure, the base of 29-deoxyguanosine (dGuo) binds to the active site in a similar manner to that observed for Ado ( Figure 4). Neither the extra amino group at position 2 nor the carbonyl group at position 6 was observed making hydrogen bonds with the protein residues. A hydrogen bond between dGuoN 7 and Ser 202 O c atoms slightly rotates the base and brings the residue Ser 202 closer to the substrate ( Figure 4B). The lack of the 29-OH group in dGuo is counterbalanced by extra hydrophobic interactions between dGuoC 29 and Glu 178 carbon atoms ( Figure 4A). Comparisons between BsPNP233-dGuo and T. vaginalis PNP (TvPNP)-(29deoxyinosine) complexes showed that the deoxyribosyl group of both ligands conserves the binding mode, whereas the base assumes a little different orientation induced by the dGuoN 7 -Ser 202 O c hydrogen bond exclusively observed in BsPNP233-dGuo complex ( Figure 4C).

BsPNP233 can be Explored as an Alternative in Gene Therapy Approaches using 2-Fluoradenosine as Prodrug
The compound 2-fluoradenosine (F-Ado) is an adenosine analogue which liberates the toxic metabolite 2-fluoradenine when cleaved. Its deoxy form has been studied as a prodrug in an anti-tumor gene therapy approach based on a modified human PNP [57]. The crystal structure of BsPNP233-(F-Ado) complex had two BsPNP233 subunits per asymmetric unit and both presented clear electronic density for the ligand ( Figure S1). In the two independent active sites, F-Ado was found in the same orientation, similar to that of Ado ( Figure 4E). The extra fluorine atom at position 2 is allocated in a hydrophobic micro-   Figure 4D). This motif is fully conserved in EcPNP, which has been tested in anti-tumor gene therapy by activating produgs like F-ado [30].
Two hydrogen bonds (N 6 -Asp 203 O d and N 7 -Ser 202 O c ) observed in BsPNP233-F-Ado complex, but not in BsPNP233-Ado complex, contribute for subtle changes in nucleoside position and base orientation ( Figure 4E). The F-Ado ribosyl moiety adopts the catalytically favorable C49-endo conformation supported by tight hydrogen bonds of nucleoside sugar hydroxyl groups with His 4* , Arg 87 and Glu 180 side chains ( Figure 4E). Structural comparisons among BsPNP233-F-Ado, TvPNP-F-Ado (PDB code 1Z35) and EcPNP-F-Ado (PDB code 1PK9) complexes showed a similar binding mode. However, the presence of phosphate in the EcPNP-(F-Ado) complex displaces by about 0.5 Å the ribosyl moiety, disrupts the N 7 -Ser 202 O c hydrogen bond and leads the N 7 atom closer to the Asp 203 side chain, favoring the catalysis ( Figure 4F).
Since F-Ado binds to the BsPNP233 active site in a manner similar to that of the natural substrate Ado, placing the F 2 atom in a hydrophobic pocket conserved in EcPNP, our structural data indicate that, as well as EcPNP [54], BsPNP233 is able to convert 2-fluoradenosine in the cytotoxic 2-fluoradenine. Thus, we concluded that BsPNP233 can be explored as an alternative in the development of anti-tumors gene therapy approaches using this prodrug or the less toxic 2-fluoro-29-deoxyadenosine [57].
The molecule Cl-Guo displays a similar binding mode to that observed for dGuo ( Figure 5A). However, as the chlorine van der Waals radius is larger than that of oxygen, the Cl 6 substituent pushes the base in the direction of the ribosyl moiety to avoid steric clashes with Gly 92 C a , Val 205 C c2 , and Asp 203 O d1 atoms. This base displacement induces the ribosyl group to adopt an unusual C39exo conformation.
The C39-exo pucker was already observed in the nucleoside 9-b-D-xylofuranosyladenine bound to EcPNP (PDB code 1PR6, [54]) and it is considered incompatible with the sugar conformation required for PNP catalysis [54]. Moreover, structural comparisons with the EcPNP-Ado-PO 4 complex (PDB code 1PK7, [54]) showed that the chlorine atom may prevent the Asp 203 side chain to approach to the N 7 atom to donate a proton during catalysis ( Figure 5B). Thus, these findings suggest that Cl-Guo as well as other NAs with 6-substituents heavier than chlorine cannot be cleaved by BsPNP233 and other hexameric PNPs. The 8-bromoguanosine (Br-Guo) is a ''poor substrate'' of the trimeric PNP from calf spleen [60]. Its first crystallographic portrayal in complex with a protein is described here. The addition of a bromine radical at the C 8 atom of guanosine results in the formation of a halogen bond between Br 8 and Ser 90 O c atoms, which implicates in both positional and rotational displacement of the base by 0.3 Å and 7u, respectively ( Figure 5C). The ribosyl moiety of Br-Guo presents a typical C49-endo conformation and binds to the active site in a very similar fashion to that seen for 29deoxyguanosine ( Figure 5C).
In BsPNP233-(Br-Guo) complex, the side chain of the catalytic residue Asp 203 is facing the O 6 and N 7 atoms of Br-Guo and the Ser 90 side chain is pushed away from the active site in order to accommodate the bromine atom ( Figure 5C). In hexameric PNPs, the hydroxyl group of Ser 90 participates in the coordination of phosphate [61]. The position that it assumes in BsPNP233-sulfate and EcPNP-Ado-PO 4 complexes (PDB code 1PK7 [54]) is incompatible with the presence of Br 8 atom because of steric hindrance ( Figure 5C). Thus, the bromine radical probably prevents the phosphate-Ser 90 interaction being detrimental for binding and correct orientation of phosphate.
Site-directed mutations of human PNP phosphate binding site leads to a decrease in catalytic efficiency ranging from 25-to 185fold [62]. Likewise, impairment of any phosphate interaction in the hexameric PNP active site may reduce the catalytic activity. As Br-Guo probably prevents the phosphate-Ser 90 interaction it might be a ''poor substrate'' or even an inhibitor of BsPNP233 and other hexameric PNPs as well.
This interpretation cannot be applied for trimeric PNPs because Ser 90 is not structurally conserved in trimeric PNPs. However, structural comparisons between BsPNP233-(Br-Guo) and human PNP in complex with guanosine and sulfate (PDB code 1RFG, [63]) indicate that Br-Guo is probably a ''poor substrate'' for trimeric PNPs because of steric hindrance involving the bromine and the side chain of Thr 242 , which would hinder the displacement of the N 7 atom towards Asn 243 side chain for stabilization of the transition state [64] (Figure 5D).
TBN presented an interaction mode very similar to that seen for the natural substrate adenosine ( Figure 6). Slightly differences were observed in its ribosyl moiety that assumed an O49-exo pucker instead of the C39-exo conformation of Ado in complex with BsPNP233 ( Figure 6). The C 7 substituent in TBN makes hydrophobic and van der Waals interactions with residues Cys 91 and Ser 202 . The ribosyl moiety is stabilized by a conserved network of hydrogen bonds involving His 4* , Glu 180 and Arg 87 side chains and by hydrophobic interactions with Glu 178 (C a and C b atoms) and Met 179 C c atom ( Figure 6). Our structural data corroborate the competitive inhibition mechanism of hexameric PNP by TBN defined by in vitro studies [69]. The substitution of N 7 by a carbon prevents the protonation step of the N 7 atom required for catalysis [70], making TBN a non-cleavable adenosine analogue by EcPNP and probably by other PNPs.  Ganciclovir Inhibits Both Trimeric and Hexameric PNPs Ganciclovir (GCV) is an acyclic NA used to treat cytomegalovirus infections [17]. It is also used together with herpes simplex virus thymidine kinase in a suicide gene therapy system that has been studied for the treatment of hepatocellular carcinoma [71]. GCV is an inhibitor of the human PNP (trimeric) [72] and probably has inhibitory effects on hexameric PNPs as well. Our structural data support this hypothesis revealing that GCV binds to the nucleoside binding site of BsPNP233 (Figure 7).
The guanine moiety of GCV conserves the position observed for the 29-deoxyguanosine base but it is rotated by about 10u to accommodate the acyclic chain in the ribose-binding site ( Figure 7A-B). A water molecule mediates hydrogen bonds between the ligand O 6 atom and the side chains of Ser 202 and Asp 203 . The N 7 atom interacts with Ser 202 O c and Gly 92 N atoms, and the base is stabilized by hydrophobic contacts with Ser 90 , Cys 91 , Ser 202 and Phe 159 ( Figure 7A).
Interestingly, the three oxygens of the acyclic radical occupy similar positions to those observed for the three oxygens of dGuo ribosyl group, mimetizing its binding mode ( Figure 7B). From the three hydrogen bonds observed for dGuo ribosyl moiety, the GCV acyclic radical conserves two, involving the His 4* and Glu 180 side chains. Moreover, the C 49 atom of the GCV acyclic moiety preserves the hydrophobic interactions with Met 179 C b and Met 179 C c atoms performed by the dGuo C 39 atom ( Figure 7B). Therefore our data indicate that GCV is also a competitive inhibitor for hexameric PNPs.

Aciclovir Acyclic Chain Adopts Two Conformations in the BsPNP233 Ribosyl Binding Site
Aciclovir (ACV) is an antiviral drug used to treat herpes virus infections [73] and has modest inhibitory effects on human PNP [74]. Here, we present for the first time the crystal structure of a hexameric PNP with ACV. This structure revealed differences in the aciclovir binding mode, which can be explored for drug design targeting hexameric PNPs from pathogens such as P. falciparum [6] and T. vaginalis [8].
Aciclovir binds to the BsPNP233 nucleoside binding site and is stabilized by hydrophobic interactions and a hydrogen-bonding network mediated by solvent molecules (Figure 7C-D). Interestingly, the acyclic tail assumes two alternative conformations that, seen simultaneously, resemble the conformation observed for the ganciclovir acyclic radical ( Figure 7C-E). In one of these conformations, the 39 hydroxyl group of ACV is attached to the carboxyl group of Glu 180 side chain while the carbon atoms make hydrophobic contacts with the main chain of Glu 178 and with the Met 179 C b and Met 179 C c atoms ( Figure 7C). A phosphate ion, modeled with half occupancy based on difference maps, also makes a hydrogen bond with the ligand 39 hydroxyl group ( Figure 7C). The other conformation is stabilized by a hydrogen bond between the 39-OH group of ACV and the His 4* side chain ( Figure 7D).
The ACV guanine moiety assumes a different position and orientation from that observed for GCV ( Figure 7E), getting closer to the Phe 159 side chain. The main chain of Cys 91 and the side chain of Val 177 also contribute with hydrophobic interactions to the base ( Figure 7C-D). The O 6 atom makes water mediated hydrogen bonds with the Asp 203 side chain and with the Phe 159 carbonyl oxygen ( Figure 7C and D). The same is observed for N 1 and N 2 atoms, which interact through a water molecule with the Gln 158 carbonyl oxygen; for N 7 atom, which makes water mediated hydrogen bonds with Asp 203 side chain, and; for N 9 atom, whose interaction with both Ser 90 and Ser 202 hydroxyl groups is also mediated by a solvent molecule ( Figure 7C-D).
Structural comparison between BsPNP233-ACV and human PNP (HsPNP)-ACV (PDB code 1PWY, [74]) complexes showed differences in the binding mode. In the HsPNP-ACV complex, the base N 1 , N 2 , N 7 and O 6 atoms interact directly with active-site residues through hydrogen bonds. In addition, the acyclic chain adopts a different conformation, which is stabilized by hydrophobic interactions with Phe 200 side chain and Ala 116 /Ala 117 main chains ( Figure 7F). To investigate if differences in the interaction mode of aciclovir with BsPNP233 and HsPNP may result in different binding affinities, we estimated the strength of proteinligand interactions using the rerank score function of MOLEGRO [75]. According to this analysis, ACV presented similar predicted binding affinities in both complexes, which was slightly higher (lower rerank score value) for the BsPNP233 complex (Table S3). The same analysis was performed for GCV whose predicted binding affinity was considerable higher than that observed for ACV (Table S3). This result indicates that GCV is a better inhibitor for hexameric PNPs than ACV.

Structural Basis of Distinct Kinetic Models for Phosphate Binding in Hexameric PNPs
The asymmetric unit of the BsPNP233 crystal structure belonging to the H32 space group presented a catalytic dimer whose protomers adopt an open and a closed conformation, respectively ( Figure 8A). The electron density map clearly showed a tetrahedral molecule in the active site of both subunits ( Figure  S1). As the crystallization condition was phosphate free and contained high concentrations of ammonium sulfate, we modeled sulfate ions in both sites.
The open and closed conformations of BsPNP233-sulfate complex were already observed in EcPNP-sulfate/phosphate structures and have been associated with two dissociation constants that characterize phosphate binding to EcPNP [61,76]. The closed conformation is defined by a disruption of helix a7 and subsequent displacement of its N-terminal portion and the precedent loop towards the active site ( Figure 8A). This conformation seems to be triggered by the interaction of Arg 24 side chain with phosphate and results in an approximation of Arg 216 to the catalytic residue Asp 203 ( Figure 8A) [61]. As BsPNP233 protomers are able to adopt open and closed conformations like EcPNP subunits, this suggests that the negative cooperativity of phosphate binding demonstrated for EcPNP [61] is also applied for BsPNP233.
The apparent inability of BcAdoP to adopt the closed conformation seems to be caused by a steric hindrance imposed by Thr 91 to the conformational change that Phe 221 (BsPNP233-Phe 220 ) undergoes for the closed conformation being achieved ( Figure 8B). In BsPNP233 and EcPNP this threonine residue is replaced by a serine, which allows Phe 220 side chain to adopt the rotamer observed in the closed conformation ( Figure 8C). These analyses suggest that the negative cooperativity model of phosphate binding displayed by EcPNP cannot be applied for BcAdoP, as BcAdoP apparently presents only one conformational state. This hypothesis is supported by functional studies, which showed that BcAdoP obeys Michaelis-Menten kinetics [77]. A previous work reported that BsPNP233 is specific for 6aminopurine nucleosides [78]. However, Xie and coworkers [34] recently showed that BsPNP233 (named PNP 702 ) exhibits a broad substrate specificity and present comparable activity towards both guanosine (6-oxopurine nucleoside) and adenosine (6-aminopurine nucleoside). Our structural data is in agreement with Xie and coworkers data indicating that BsPNP233 conserves the same catalytic mechanism proposed for EcPNP [76], where catalysis occurs in the closed conformation ( Figure 8A).

Conclusion
This report provided a broad description of how the hexameric PNP from B. subtilis interacts with natural substrates and the impact of modifications in such substrates on binding and catalysis. The structural analysis reported here can be instrumental for studies aiming to optimize BsPNP233 or other hexameric PNPs for biotechnological applications such as industrial synthesis of nucleoside analogues or gene therapy against solid tumors. An initiative of this sort has been taken for E. coli PNP to optimize the cleavage of the prodrug Me(talo)-MeP-R with great success [29].
The crystal structure of six ligands (adenine, 29deoxyguanosine, aciclovir, ganciclovir, 8-bromoguanosine and 6-chloroguanosine) in complex with a hexameric PNP are presented for the first time. The information extracted from these structures can be extended to homologous hexameric PNPs to help the development of new inhibitors against pathogens such as T. vaginalis [8] and P. falciparum [6] as well as new prodrugs for gene therapies against tumors [30,79].
In addition, our results and comparative analyses shed light on distinct kinetic models for phosphate binding in hexameric PNPs. According to our model the substitution of the conserved residue Ser 90 by a threonine disrupts the open/close mechanism of hexameric PNPs subunits, which results in the loss of the negative cooperativity of phosphate binding.  Table S2 Distances (Å ) between the ligand atoms and interacting BsPNP233 atoms. Potential hydrogen bonds are highlighted by grey boxes. In the case of crystal structures containing more than one complex per asymmetric unit, only one of them is shown in the table. For the ligand adenine a representative structure of the preferential 1 (BsPNP233-Ade complex, chain A) and alternative ¥ (form I, chain A) conformations are presented. (DOC)

Table S3
In silico prediction of ligand binding affinity using the rerank score function of MOLEGRO [75]. ¥ The two values of BsPNP233-ACV complex correspond to the ACV 1 Figure 8. Structural basis of distinct kinetic models for phosphate binding in hexameric PNPs. A. Structural superposition of BsPNP233sulfate open (green) and closed (pink) conformations with the BcAdoP-Ado complex (yellow, PDB code 3UAW, [52]). The cartoon representation highlights the conformational differences observed in the main chain of the b9-a7 loop and the N-terminal portion of helix a7 in the three structures. Dashed lines represent hydrogen bonds and follow the color code of their respective structures. B. The surface representation of BsPNP233 Phe 220 in the closed conformation (pink) and of the BcAdoP Thr 91 evidence the steric hindrance imposed by the Thr 91 C c2 atom to that Phe 220 rotamer. C. The surface representation of BsPNP233 Phe 220 and Ser 90 in the closed conformation shows that the Ser 90 side chain allows the Phe 220 side chain to perform the conformational change needed for the closed conformation takes place. doi:10.1371/journal.pone.0044282.g008 and ACV 2 alternative conformations, respectively. 1 HsPNP-ACV (PDB CODE: 1PWY). (DOC)