Exploring the Chemical Space around 8-Mercaptoguanine as a Route to New Inhibitors of the Folate Biosynthesis Enzyme HPPK

As the second essential enzyme of the folate biosynthetic pathway, the potential antimicrobial target, HPPK (6-hydroxymethyl-7,8-dihydropterin pyrophosphokinase), catalyzes the Mg2+-dependant transfer of pyrophosphate from the cofactor (ATP) to the substrate, 6-hydroxymethyl-7,8-dihydropterin. Recently, we showed that 8-mercaptoguanine (8-MG) bound at the substrate site (KD ∼13 µM), inhibited the S. aureus enzyme (SaHPPK) (IC50 ∼ 41 µM), and determined the structure of the SaHPPK/8-MG complex. Here we present the synthesis of a series of guanine derivatives, together with their HPPK binding affinities, as determined by SPR and ITC analysis. The binding mode of the most potent was investigated using 2D NMR spectroscopy and X-ray crystallography. The results indicate, firstly, that the SH group of 8-MG makes a significant contribution to the free energy of binding. Secondly, direct N 9 substitution, or tautomerization arising from N 7 substitution in some cases, leads to a dramatic reduction in affinity due to loss of a critical N 9-H···Val46 hydrogen bond, combined with the limited space available around the N 9 position. The water-filled pocket under the N 7 position is significantly more tolerant of substitution, with a hydroxyl ethyl 8-MG derivative attached to N 7 (compound 21a) exhibiting an affinity for the apo enzyme comparable to the parent compound (KD ∼ 12 µM). In contrast to 8-MG, however, 21a displays competitive binding with the ATP cofactor, as judged by NMR and SPR analysis. The 1.85 Å X-ray structure of the SaHPPK/21a complex confirms that extension from the N 7 position towards the Mg2+-binding site, which affords the only tractable route out from the pterin-binding pocket. Promising strategies for the creation of more potent binders might therefore include the introduction of groups capable of interacting with the Mg2+ centres or Mg2+ -binding residues, as well as the development of bitopic inhibitors featuring 8-MG linked to a moiety targeting the ATP cofactor binding site.


Introduction
Antibiotic resistance is rapidly emerging as one the most significant health challenges of this century [1]. In Europe alone, 25,000 deaths were reported in 2007 as a result of antimicrobial resistance, with an estimated cost of J1.3 billion per year [2]. Compounding this problem is the fact that antibiotic drug discovery is on the decline -a reflection of the considerable challenges associated with identifying both viable new targets as well as drugs to target them, but also a general lack of interest from large pharmaceutical companies. Most alarmingly, Methicillinresistant S. aureus (MRSA) has evolved globally into a range of strains with varying phenotypes [3]. It has become resistant to both oxacillin and erythromycin, and resistance to levofloxacin is reported to be on the rise [4]. Community-acquired MRSA (caMRSA) is a relatively recent threat among patients without conventional risk factors. The epidemic USA300 strain of caMRSA is exceptionally virulent due to high levels of alpha toxin and the phenol-soluble modulins [4]; remarkably, it accounts for over half of all illnesses caused by the entire range of S. aureus species.
Logical targets for antimicrobials are essential enzymes that are unique to microorganisms, of which those of the folate biosynthesis pathway are prime examples. Folate is essential for the growth of all living cells, with the reduced form, tetrahydrofolate, used in the biosynthesis of thymidine, glycine and methionine. However, only bacteria and lower eukaryotes synthesize folate de novo; mammals and higher eukaryotes obtain it from their diet by active transport. The folate pathway enzymes, dihydropteroate synthase (DHPS) and dihydrofolatereductase (DHFR) are the targets for the sulfa drugs and Trimethoprim, respectively, which are used to treat diseases such as malaria, pneumocystis pneumonia (PCP), and, more recently, caMRSA infections.
It is well established that point mutations in pathogenic DHPS and DHFR genes contribute to widespread resistance to the aforementioned drugs. Recently, structure-based investigations have identified new inhibitors of DHPS that bind to the pterin site, remote from the sulpha drug site [5], as well as a new lead candidate for inhibiting the quadruple mutant DHFR enzyme conferring resistance in Plasmoidium falciparum [6]. These studies exemplify the application of modern drug discovery approaches to old targets as a means of generating potential new antibiotics.
An alternative approach to combating resistant isolates is the development of inhibitors for as-yet-to-be-drugged enzymes within the folate pathway. Hydroxymethyl-pterin pyrophosphokinase (HPPK) is one such enzyme, responsible for catalysing the transfer of a pyrophosphate group from the ATP to the pterin substrate, 6hydroxymethyl-7,8-dihydropterin (HMDP) (Fig. 1A). HPPK structures from many microbial sources have been solved (E. coli, H. influenza, S. pneumoniae, S. cerevisiae, Y. pestis, F. tularensis and S. aureus) [7][8][9][10][11][12][13][14]. All have a thioredoxin-like fold containing the binding sites for both the substrate and the ATP cofactor. X-ray structural studies have revealed that major conformational changes, particularly in loop L3, occur throughout the catalytic cycle [15]. Structural and kinetic studies [16][17][18][19][20] have also established that ATP binds (K D = 2.6-4.5 mM) prior to the substrate, which binds with sub-micromolar affinity. The pterin stacks between two highly conserved aromatic residues (Tyr or Phe) and both the substrate and cofactor are fixed in position by a multitude of hydrogen bonds; in total, they interact with 26 separate residues.
While much is known about the structure of HPPK, very few small molecule inhibitors have been developed (Fig. 1B). The gemdimethyl-and 7-phenethyl-substituted pterin analogues, 3 and 4, were reported to be HPPK inhibitors over three decades ago by Woods [21]. They have since been crystallized bound to the E. coli enzyme [11,22], and 3 was utilized in a number of structural studies aimed at understanding the catalytic trajectory of HPPK [23,24]. Recent inhibitor design has included the production of bitopic ligands featuring pterin coupled to adenosine via monothrough to tetra-phosphate linkers (5), with the longest linker providing the best affinity (K D = 0.47 mM) and inhibition (IC 50 = 0.44 mM) [18]. Bitopic ligands featuring a more drug-like piperidine bridge (6) [20], or gem-dimethyl pterin in combination with a piperidine or amide-sulphone linker (7 [20] and 8 [25]), have also been reported, however no gain in potency has been achieved (8 did, however, display a novel binding mode in which the base was flipped).
Very recently, we showed that the simple guanine derivative, 8mercaptoguanine , is able to inhibit HPPK from S. aureus (K D ,11 mM, IC 50 = 41 mM) through interaction with the HMDP pocket [8]. Binding was found to be non-competitive with either the cofactor (ATP) or its non-hydrolyzable analogue, AMPCPP, as judged by both surface plasmon resonance (SPR) and isothermal titration calorimetry (ITC) analysis. A 1.65 Å resolution X-ray crystal structure revealed a high degree of stereo-electronic complementarity between 8-MG and the HMDP-binding pocket, together with an extensive network of hydrogen bonds, accounting for the unusually high binding affinity of the small 8-MG molecule (183 Da) ( Fig. 2A, B). Most intriguingly, NMR analysis on the 8-MG/AMPCPP ternary complex provided compelling evidence that the SH group of 8-MG interacts with the L3 loop of SaHPPK, locking it onto a ''closed'' conformation above the active site [26].
Herein, we report the results of a study interrogating the chemical space available within the active site of SaHPPK and the chemical developability of 8-MG as a HPPK inhibitor. As part of this study, a series of N 7 -and N 9 -substituted 8-MG analogues have been synthesized, and their interaction with SaHPPK examined using a combination of SPR, ITC, NMR spectroscopy and X-ray crystallography, in order to determine which of these positions are amenable to the lead optimization extension strategy. Additionally, a small number of C 8 -sustituted analogues have been studied to allow further assessment of the relative importance of the SH group of 8-MG to the overall binding characteristics of this compound.

Structure-based Hypotheses and Design of 8-MG Analogues
As shown in Figure 2A and B, the pyrimidine heterocycle (ring A) of 8-MG is ''perfectly tailored'' for the pterin-binding pocket of HPPK, as evidenced by full saturation of all hydrogen donors and acceptors, and the sandwiching of the ring between the two phenylalanine residues, Phe54 and Phe123. This ring was therefore considered of limited value as a site for further chemical modification aimed at improving binding affinity and potency. Instead, our efforts focused on exploring the effects of substitution at the N 7 , C 8 and N 9 positions of ring B.
Predicting the likely outcome of substituent changes/additions to 8-MG is complicated by the fact that the L2 and L3 catalytic loops in HPPK can adopt a diverse range of conformations, leading to drastic changes in the microenvironment of ring B (Fig. 2D) (loops L2 and L3 are also inherently dynamic in the apo and cofactor-bound states on the micro to millisecond timescale [23], [8]). The 8-MG/SaHPPK X-ray structure (PDB: 3QBC) itself displays an extended L3 loop conformation [8], and is therefore limited in guiding modelling and structure-based design of 8-MG analogues from the N 7 , C 8 and N 9 positions. In the first instance, we therefore chose to explore the effect of replacing the mercapto group of 8-MG with a variety of other substituents (compounds 10a-10f, Table 1) in order to probe tolerance to substitution at this position. In part, this was performed in order to test our hypothesis (based on earlier 15 N chemical shift and NMR relaxation measurements [8]), that Gly90 or Trp89 at the tip of the L3 loop form a favorable contact to the mercapto group at the C 8 position, which serves to fix L3 into a ''closed'' conformation resembling that observed in the ternary complex of E. coli HPPK, HMDP and AMPCPP (PDB: 1Q0N) (Fig. 2C) [15].
Our substituent choices for the N 9 position were inspired by the structure of the ternary complex of E. coli HPPK with the phenethyl HMDP analogue (2-amino-6-methoxy-7-methyl-7-phenethyl-7,8-dihydropterin) and AMPCPP (PDB:1DY3) [22]. Within this structure, the phenyl ring of the substrate analogue makes two hydrophobic intermolecular interactions; one edge-on to Trp89 in loop L3 and the other to the side-chain of Leu45 (Val46 in S. aureus) in loop L2. From an overlay of 1DY3 and 3QBC (Fig. 2D), it was conjectured that the appendage of a hydrophobic group to the N 9 position of 8-MG could afford similarly favorable interactions with side-chains present in loops L2 and L3. Four hydrophobic substituents of increasing size (CH 3 , C 2 H 5 , CH 2 C 6 H 5 , CH 2 CH 2 C 6 H 5 ) were thus chosen for investigation (compounds 15a-15d, Table 1). In order to deliver a stronger binder, it was recognized that any favorable interaction(s) afforded by these groups would have to more than compensate for the loss of the hydrogen-bond between the N 9 H group and the Val 46 carbonyl in the SaHPPK/8-MG complex ( Fig. 2A, B).
Analysis of the SaHPPK/8-MG crystal structure revealed a water-filled pocket proximal to the N 7 position (Fig. 2B, C). Given the hydrophilic nature of this region, it was postulated that attachment of a suitable polar substituent might enhance binding through provision of additional interactions with the polar sidechains and/or bound waters present, coupled with entropicallyfavorable water displacement. A small series of 8-MG analogues featuring alcohol, amine and guanidinium pendants attached to N 7 were therefore included within our test set (compounds 21a-21e).

Synthesis of 8-MG Analogues
8-(Methylamino)guanosine, 9, synthesized as described in the literature [27], was hydrolyzed using 1 M HCl to afford the first of the test compounds, 8-(methyamino)guanine (10a). All other derivatives with C 8 substitution (10b-10f) were commercially sourced.  The synthetic routes to N 9 -substituted guanines are wellestablished, in part because of the use of the N 9 -substituted drugs, acyclovir and ganciclovir, in the treatment of herpes virus infections [28]. An expedient synthesis of the N 9 -methyl guanine from 2-amino-6-chloropurine, exploiting the N 9 -directing effect of the chloro-substituent, has been previously reported and involved alkylation with methyl iodide followed by hydrolysis to install the oxo group [29]. We found this method could also be employed to provide ethyl, benzyl and 2-phenethyl substituents at the N 9position (Fig. 3). Transformation of 13a to the 8-mercapto derivative 15a had been previously been demonstrated by bromination at C 8 to provide 14a [30], followed by treatment with thiourea [31]. We found this similar transformation could be applied to our other derivatives providing the brominated analogues 14b-d and the SH-containing target compounds, 15b-d.
The N 7 -substituted 8-MG analogues were prepared via alkylation of 8-bromo-N 2 -acetylguanine (18), formed in two steps from guanosine (16) according to a literature method [32,33] (Fig. 4). Benzylation of 18 at N 9 with benzyl bromide has been reported previously under conditions that required no base [32]. We found that alkylation with other reagents proceeded well when the pH of the reaction mixture was adjusted to 3. These reactions generally yielded a ca. 1:1 mixture of N 7 -and N 9 -substituted isomers, from which the desired N 7 -alkylated intermediates were isolated following either silica gel or preparative-scale reverse-phase HPLC. Installation of the 8-mercapto group was then achieved by reaction with sodium thiosulfate in the presence of a catalytic quantity of aluminium-trichloride [34], and the target 8-MG analogues isolated following removal of any protecting groups under the appropriate conditions (Fig. 4). Compound 21e, featuring an ethyl guanidinium group, was prepared from the amino analogue, 21c, through reaction with pyrazole carboxamidine (Fig. 5).
All final compounds were purified by preparative HPLC to .95% purity.

SPR and ITC Analysis of Binding
Initially, the binding of each of the test compounds to SaHPPK was quantitatively assessed using SPR (Fig. S1, S2). Compared with the parent compound (8-MG), SPR data for the synthesized analogues did not appear to be compromised by their solubility in aqueous buffer at the maximum concentration used (126 mM). Moreover, all sensorgrams ( Fig. S1 and S2) were of high quality and consistent with near perfect 1:1 stoichiometric binding of analogues. Table 1 lists the estimated equilibrium dissociation constants (K D ) . It is clear from the data that replacement of the 8mercapto group is highly detrimental to binding; compounds 10a and 10b, featuring -NHCH 3 and -SCH 3 groups at the C 8 position, did not bind SaHPPK at all (although binding of compound 10a was detected (K D = 108 mM) in the presence of saturating amounts of ATP), whilst all other C 8 -substiuted analogues exhibited 15-20fold lower K D values than 8-MG. This supports the hypothesis that the 8-mercapto group of 8-MG aids binding through the formation of one or more favorable interactions with SaHPPK. The precise nature of this/these interaction(s) remain unclear, however the considerably inferior binding affinity of the C 8 -OH analogue (10e) suggest that it is unlikely to be a simple hydrogen bond to a loop L3 residue, as we speculated might be the case earlier [8].
The 8-MG derivatives with simple hydrophobic substituents at the N 9 position (15a-d) exhibited 10-20-fold lower affinities for SaHPPK, indicating that any potential positive contribution to binding afforded by these groups is not sufficient to make up for the loss of the intramolecular N 9 -H?Val46 carbonyl hydrogen bond. Extension of the 8-MG scaffold via the N 9 position, therefore, does not appear to be a promising strategy for lead optimization.
Of the N 7 -substituted 8-MG analogues, compound 21a, with an ethyl alcohol substituent, displayed comparable affinity to 8-MG (K D ,12 mM), whilst the analogues with amine and guanidinum pendants (21c-21e) displayed slightly weaker binding to SaHPPK; the carboxylate pendant-bearing analogue, 21b, did not bind. This indicates that addition of substituents at the N 7 position are tolerated, and that extension from this ring position is likely the most promising avenue for future development of more potent 8-MG analogues. It should be noted, however, that in contrast to 8-MG, the binding of compounds 21a, 21c and 21d was found to be 10-15-fold weaker under saturating ATP conditions, suggesting extensions from ring B into the space towards the Mg 2+ centres and the ATP binding site leads to competitive binding with the ATP cofactor. Any future lead optimization studies will need to bear this in mind.
To corroborate the ligand binding affinities determined by SPR, and to determine the enthalpic and entropic contributions to the free energy of binding, ITC experiments were performed for Table 1. Structures of C 8 , N 9 and N 7 -substituted guanine analogues and their binding affinities to SaHPPK, as determined by SPR.   , showed a lower binding enthalpy than 8-MG, but its binding to SaHPPK was associated with a much lower entropic penalty. Given that 21a has more rotatable bonds than 8-MG, this may suggest that binding of the latter may lead to a greater degree of immobilization of the catalytic loops within SaHPPK. To investigate the factors contributing to the free energy of binding, we solved the structure of SaHPPK in complex with 21a.

X-ray Structure of SaHPPK in Complex with Compound 21a
Attempts were made to co-crystallize each of the strongest binding compounds (21a and 21c-21e) with SaHPPK, however diffraction quality crystals could only be obtained for compound 21a. These provided excellent quality electron density data, and a high-resolution X-ray structure (1.85 Å ) of the SaHPPK/21a binary complex (Fig. 6A) determined via molecular replacement (crystal data and details of the data collection and refinement are provided in Table 3). A head-to-tail protein dimer was found in the asymmetric unit, similar to that observed for the earlier SaHPPK/8-MG structure (PDB: 3QBC) [8], with the ligand bound to the pterin sites of both protein monomers. The ethyl alcohol pendant projects into the space leading towards the Mg 2+ binding site, making two hydrogen bond contacts with a pair of bound water molecules (Fig. 6B). Presumably, these interactions in part compensate for the loss of the hydrogen bond between the N 9 -H of 8-MG and the backbone carbonyl of Val46, which occurs as a consequence of the tautomerization accompanying alkylation at the N 7 position. A water molecule found in the cavity under N 7 in the SaHPPK/8-MG structure has been displaced in the SaHPPK/ 21a structure, and there is a tightly bound water between the hydroxyethyl and Asp97 which orients Asp97 in a similar position to that found of Asp97 of the EcHPPK/AMPCPP/HMDP structure (where Mg 2+ sits). Superposition of the SaHPPK/21a structure with that of EcHPPK/AMPCPP/HMDP (PDB: 1QON) [15] indicates that if Mg 2+ ions and ATP were simultaneously bound, the oxygen of the hydroxyethyl pendant of 21a is displaced by ,1 Å and would lie only 1.5-1.6 Å from one of the metal ions, which is considerably less than the Mg-O bond length observed in the 1QON structure (2.1 Å ) and sterically unfavorable (Fig. 6C). This is the likely reason for the cofactor and metal competition observed for 21a.

Heteronuclear NMR Analysis of Compound 21a Binding to SaHPPK
Titration of 21a into 15 N-labelled apo (data not shown) and magnesium-loaded SaHPPK enzyme led to broadening and disappearance of several, common peaks in the 2D 15 N HSQC NMR spectrum (Fig. 7A, B), which is characteristic of the intermediate exchange timescale, and indicates that binding of 21a is not magnesium-dependent. This is similar to what was observed for the binding of 8-MG to SaHPPK (Fig. 7A) and is consistent with the fact that density characteristic of magnesium was not observed in the X-ray crystal data of the SaHPPK/21a complex.
The observed intermediate exchange regime for the binding of 8-MG and 21a is possibly dictated in part by the slow ms-ms timescale motion of loop L3 [23,24]. While the spectra (Fig. 7A, B) appear to be very similar, however, closer inspection reveals that the sidechain He1-Ne1 peak of Trp89 (in loop L3) is only perturbed in the 8-MG bound spectrum (Fig. 7A). This is mechanistically interesting and may indicate that this region of loop L3, adjacent to the substrate-binding loop L2, is involved in binding of 8-MG but not 21a. Following on from this, the observed larger entropic penalty to the free energy of binding of 8-MG as compared to 21a (Table 2) may derive in part from this increase in loop L3 rigidity in the presence of 8-MG, whilst the more favorable enthalpic contribution likely reflects the formation of the N 9 -H Val46 intermolecular hydrogen bond (as observed in the X-ray structure). Reduced loop L3 involvement in 21a binding, on the other hand, is likely a result of the loss of the N 9 -H Val46 intermolecular hydrogen bond (due to tautomerization from substitution at N 7 ), which would reduce any dampening of the adjacent loop L2 dynamics. Ligand-induced loop L2 and L3 dampening can be detected and investigated directly by NMR, but in order to do this the NMR timescale needs to be shifted out from the intermediate regime. This was previously accomplished by  binding 8-MG to the AMPCPP bound SaHPPK enzyme [8]. The results of our heteronuclear NMR spin relaxation studies reported therein revealed dampening of loop L2 and L3 motion on the fast timescale compared to the apo or the AMPCPP (or ATP)-bound SaHPPK enzyme. Unfortunately, it was not possible to investigate the enzyme dynamics in the same way for the binding of 21a as it was found to be competitive with AMPCPP. From a comparison of the X-ray structures of the 21a/SaHPPK binary complex (this work) with our previous 8-MG/SaHPPK binary structure [8], the expulsion of a bound water underneath N 7 is likely to be thermodynamically favorable for the binding of 21a, and in line with the observed reduced entropic penalty. This may be the reason for the reduced entropic penalty associated with binding of the 21a-c series as a whole; see Table 2.
Repeating the titration in the presence of a saturating amount of the cofactor analogue, AMPCPP (K D = 3 mM), led to broadening of the same pterin site signals and the peak corresponding to the sidechain of Trp89 displayed very little change, in accord with the lack of involvement of this residue in 21a binding, as described previously. In accordance with the SPR data, however, cross peaks in slow exchange, characteristic of formation of a ternary complex (observed in our earlier study of the interaction of 8-MG with SaHPPK in the presence of AMPCPP (Fig. 7C)) were absent (Fig. 7D), indicating that binding of 21a to SaHPPK is competitive with AMPCPP.

Conclusion
8-MG represents a promising scaffold for the potential development of a new antibiotic drug targeting the folate pathway enzyme, HPPK. This study has shown that the 8-mercapto group plays a pivotal role in binding, and ought to be maintained in any future lead optimization studies. Extension from the N 9 position within ring B leads to a dramatic loss of affinity and is therefore not a viable site for chemical modification. Substitution at the N 7 position, however, is tolerable, as exemplified by N 7 -hydroxyethyl-8-MG (21a), which was found to bind SaHPPK with comparable affinity to the parent compound. An important caveat is that extension into the space surrounding the N 7 atom leads to competitive binding with the ATP cofactor. To provide a meaningful enhancement in potency, future studies will therefore need to focus on the development of N 7 pendants that interact strongly with the residues surrounding this pocket. This could include the introduction of groups to bind to the absolutely conserved metal-binding residues, Asp95 and Asp97, within the apo form of the enzyme. An alternate route to an increase in potency could involve changing the nature of ring B of the 8-MG core such that the N 9 -H Val46 H bond is maintained whilst still allowing extension from the N 7 position towards the highly conserved metal-binding residues. We are currently investigating this approach.
Compared to the reported bitopic inhibitors for HPPK [18,20,25], both 8-MG and 21a are less potent, yet they have better ligand efficiencies (K D ,10 mM over 12 and 15 heavy atoms, respectively, compared to K D , 3 mM over 40+ heavy atoms). 8-MG could potentially be linked to adenosine to provide a bitopic ligand with considerably enhanced affinity, though problems associated with linking two subsite binders as a route to higher affinity have been well documented [35,36]. Ultimately, incremental step-wise chemical evolution of the 8-MG scaffold in a more conventional manner may prove the most efficient route to developing an inhibitor with superior pharmacodynamic and pharmacokinetic properties.
Finally, it is worth noting that it has recently been shown that 8-MG can also bind to the pterin pocket in DHPS, the adjacent, downstream enzyme to HPPK [37]. The chemical strategies described herein may therefore prove beneficial for the design of more potent DHPS inhibitors based on the 8-MG scaffold, and perhaps even for the development of agents capable of inhibiting multiple enzymes within the folate biosynthesis pathway.

Chemistry -General Methods
Melting points were determined on a Mettler Toledo MP50 melting point system and are uncorrected. The abbreviation dec. indicates that the compound decomposed at the specified Table 3. X-ray structure data processing and refinement statistics.

Spacegroup
Monoclinic, P21  temperature. 1 H and 13 C NMR spectra were recorded on a Bruker Ultrashield 400 Plus at 400 MHz and 101 MHz, respectively. Analytical HPLC was performed on a Waters Alliance 2690 fitted with a Waters 5996 PDA detector and a Phenomenex Luna C 8 column (5 mm, 100 Å , 150 6 4.60 mm). Analyses were conducted using a gradient of 0 to 64% acetonitrile in water over 10 min with 0.1% trifluoroacetic acid (TFA) throughout. Preparatory HPLC was performed on a Waters Prep LC 4000 system fitted with a Waters 486 Tunable Absorbance Detector and either a Phenomenex Luna C 18 (10 mm, 100 Å , 250 6 30 mm) column or a Phenomenex Luna C 8 (10 mm, 100 Å , 50 6 21.2 mm) column. Low resolution mass spectrometry was performed on an Agilent 6120 single quadrapole LCMS system using electrospray ionization. High resolution mass spectrometry was performed on a Waters Premier XE time-of-flight mass spectrometer using electrospray ionization.

Surface Plasmon Resonance (SPR)
All SPR binding experiments were performed as described previously [8]. The only difference was the use of a sulfhydryl reactive maleimide-activated biotin derivative (Thermo Scientific, 1-biotinamido-4-(49-[maleimidoethylcyclohexane]-carboxamido)butane. The maleimide-activated biotin was attached to the exposed surface cysteine residue of SaHPPK according to manufacturer's instructions. The resulting site-specific biotinylated protein was immobilized onto the sensor chip surface using the Biotin capture kit (GE Healthcare). All analogues were serially diluted (either 2-or 3-fold from 126 mM down to 1.5 mM) in SPR binding buffer (50 mM HEPES, 150 mMNaCl, 1 mM TCEP, 0.05% (v/v) Tween-20, 10 mM MgCl 2 , 5% (v/v) DMSO, pH 8.0) and injected for 30 sec contact time at 60 mL/min and then allowed to dissociate for 60 sec. Binding sensorgrams were processed using the Scrubber (version 2c, BioLogic Software, Campbell, Australia). To determine the binding affinity (equilibrium dissociation constant; K D ), responses at equilibrium for each compound were fit to a 1:1 steady state affinity model available within Scrubber.

Isothermal Titration Calorimetry (ITC)
Experiments were performed using an iTC200 instrument (MicroCal) at 298 K, with the ligands titrated into solutions of SaHPPK using 1862.2 mL injections. Data were fitted using Origin software to yield the thermodynamic parameters, DH, KD and N (the binding stoichiometry), assuming a cell volume of 0.2 mL. These were then used to calculate the Gibb's free energy of binding, DG (-RT.lnK a ), and entropy of binding, DS (using DG = DH -TDS). A stock solution of SaHPPK was dialyzed overnight into 50 mM HEPES, 1 mM TCEP, 10 mM MgCl 2 , pH 8.0 buffer with the addition of 5% DMSO (v/v) prior to running the experiment. For titrations with compounds 21a-e, SaHPPK was typically at 30 mM and the ligand stocks were at 1-1.5 mM dissolved in the above buffer then diluted into more of the same buffer. There was no apparent issue with limited solubility of 21a-c compromising either the stock solutions or the injected concentrations.

X-ray Crystallization and Structure Determination
Crystallization experiments were performed as described previously [9]. Briefly, co-crystallization was set-up in the JCSG+ Suite commercial crystal screens (Qiagen) at 281 K using sittingdrop vapor-diffusion method with droplets consisting of 150 nL protein solution and 150 nL reservoir solution and a reservoir volume of 50 mL. Crystals of the SaHPPK in complex with 7-(2hydroxyethyl)guanine (21a) were observed in conditions containing 240 mM sodium malonate and 20% polyethylene glycol 3350. Data were collected at the MX-2 beamline of the Australian Synchrotron (see Table 3 for statistics) using a one degree oscillation angle, 360 frames were obtained for a complete data set. These data were indexed using XDS [38] and scaled using SCALA [39].
The SaHPPK structure (3QBC) was used to solve the initial phases of the binary complex by molecular replacement using Phaser [40]. Refinement was performed using REFMAC5 [41] and the Fourier maps (2F O -F C and F O -F C ) were visualized in Coot [42]. After several rounds of manual rebuilding, 21a and water molecules were added and the model further refined to a resolution of 1.85 Å (R free (%) = 26.4, R work (%) = 20.9).
The coordinates of SaHPPK in complex with 21a have been deposited at the Protein Data Bank with accession number 4ad6.