pH-tuneable binding of 2 0-phospho-ADP-ribose to ketopantoate reductase : a structural and calorimetric study

# 2007 International Union of Crystallography Printed in Denmark – all rights reserved The crystal structure of Escherichia coli ketopantoate reductase in complex with 20-monophosphoadenosine 50-diphosphoribose, a fragment of NADP that lacks the nicotinamide ring, is reported. The ligand is bound at the enzyme active site in the opposite orientation to that observed for NADP, with the adenine ring occupying the lipophilic nicotinamide pocket. Isothermal titration calorimetry with R31A and N98A mutants of the enzyme is used to show that the unusual ‘reversed binding mode’ observed in the crystal is triggered by changes in the protonation of binding groups at low pH. This research has important implications for fragment-based approaches to drug design, namely that the crystallization conditions and the chemical modification of ligands can have unexpected effects on the binding modes. Received 5 October 2006 Accepted 25 October 2006

The crystal structure of Escherichia coli ketopantoate reductase in complex with 2 0 -monophosphoadenosine 5 0 -diphosphoribose, a fragment of NADP + that lacks the nicotinamide ring, is reported. The ligand is bound at the enzyme active site in the opposite orientation to that observed for NADP + , with the adenine ring occupying the lipophilic nicotinamide pocket. Isothermal titration calorimetry with R31A and N98A mutants of the enzyme is used to show that the unusual 'reversed binding mode' observed in the crystal is triggered by changes in the protonation of binding groups at low pH. This research has important implications for fragment-based approaches to drug design, namely that the crystallization conditions and the chemical modification of ligands can have unexpected effects on the binding modes.

Introduction
Ketopantoate reductase (KPR; EC 1.1.1.169), an enzyme involved in the biosynthesis of pantothenate (vitamin B 5 ), catalyses the reduction of ketopantoate to form pantoate using NADPH as a cofactor . The enzyme from Escherichia coli, encoded by the panE gene, has been purified and extensively characterized both biochemically (Zheng & Blanchard, 2000a and structurally (Matak-Vinkovic et al., 2001;Lobley et al., 2005). The crystal structure of the apoenzyme was solved at 1.70 Å resolution using the selenomethionine MAD method (Matak-Vinkovic et al., 2001). KPR belongs to the 6-phosphogluconate dehydrogenase superfamily in the SCOP database (Murzin et al., 1995) and has properties of both class A and class B secondary alcohol dehydrogenases (Zheng & Blanchard, 2000a). The secondary structure comprises 15 -helices and 11 -strands. The enzyme is monomeric and has two domains separated by the active-site cleft. The N-terminal domain has the Rossmann-type fold featured in many nucleotide-binding proteins, with a glycine-rich region ( 7 GCGALG 12 ) for nucleotide recognition, whilst the C-terminus is an -helical domain. We have also solved the crystal structure of KPR in complex with NADP + (Lobley et al., 2005). This structure showed the cofactor bound in an extended conformation in the cleft between the two domains and identified the interactions formed with key residues at the enzyme active site (Lobley et al., 2005). More recently, a systematic biophysical analysis was conducted to probe hot spots at the cofactorbinding site using a series of NADP + fragments and analogues (Ciulli et al., 2006). There is a conservation of the NADP-like binding mode across the series of fragments. The 2 0 -phosphate and the reduced nicotinamide are key binding groups for cofactor recognition (Ciulli et al., 2006).
In this paper, we report the three-dimensional structure of KPR in complex with 2 0 -monophosphoadenosine 5 0 -diphosphoribose (2 0 P-ADP-ribose), a fragment of NADPH that lacks the nicotinamide ring, solved to 1.95 Å resolution. The structure surprisingly reveals the ligand to be bound in the opposite orientation to that observed for NADP + , with the adenine ring occupying the lipophilic nicotinamide pocket. Detailed analysis of this unexpected crystallographic result using isothermal titration calorimetry (ITC) and site-directed mutagenesis revealed that this unusual 'reversed binding mode' is triggered by changes in the protonation of binding groups at low pH.

Protein expression, purification and site-directed mutagenesis
For ITC studies, wild-type (WT) KPR and site-directed mutants of KPR were expressed with an N-terminal His 6 tag, whereas WT KPR for crystallization studies had no tag. Both constructs were expressed and purified as described previously (Lobley et al., 2005;Ciulli et al., 2006). Purified His 6 -tagged protein was dialyzed into the appropriate buffer using Slide-A-Lyzer dialysis cassettes (Pierce). The protein concentration was determined from the A 280 , using an absorption coefficient for His 6 -KPR of 62 650 M À1 cm À1 determined from aminoacid analysis [Protein and Nucleic Acid Chemistry Facility (PNAC), Cambridge]. An absorption coefficient of 47 870 M À1 cm À1 estimated from the sequence was used for untagged KPR. The genes for the singly mutated forms of KPR, R31A and N98A were made using the QuickChange Site-Directed Mutagenesis Kit (Stratagene) as described previously (Ciulli et al., 2006).

Crystallization of KPR
Crystallization trials were carried out using the hangingdrop vapour-diffusion method as described previously (Lobley et al., 2005). Prior to crystallization, KPR samples (10-15 mg ml À1 ) were incubated at 277 K with NADPH and pantoate at a final ligand:protein ratio of 2:1 and 5:1, respectively. Needle-like crystals of KPR were obtained in 10% 2-methyl-2,4-pentanediol (MPD) buffered with 0.1 M sodium acetate pH 4.0-5.0. Crystals were cryoprotected in 30% MPD buffered with 0.1 M sodium acetate at the same pH as the well solution. Attempts to crystallize KPR at a pH close to 7 were unsuccessful.
2.3. X-ray data collection, crystal structure determination and refinement X-ray data were collected to 1.95 Å resolution at station 14.4, European Synchrotron Radiation Facility. The principal parameters and data statistics for this data collection are shown in Table 1. DENZO and SCALEPACK from the HKL-2000 suite were used to process and scale the data (Otwinowski & Minor, 1997). The structure was solved using molecular replacement, with apo KPR as the probe structure. AMoRe from the CCP4 suite (Collaborative Computational Project, Number 4, 1994;Navaza, 1994Navaza, , 2001 was used for molecular replacement. After molecular replacement, the initial model underwent simulated annealing using CNS (Brü nger et al., 1998) and initial electron-density maps were generated. Refinement involved an iterative cycle of manual rebuilding in XTALVIEW (McRee, 1999) and maximumlikelihood refinement using REFMAC5 from the CCP4 suite (Murshudov et al., 1997;Collaborative Computational Project, Number 4, 1994). The completed structure is of residues 1-292 of KPR in complex with one molecule of 2 0 P-ADP-ribose. Nonprotein molecules were obtained from the HIC-Up database (Kleywegt & Jones, 1998) and refinement libraries were taken from the PRODRG server (Schü ttelkopf & van Aalten, 2004).

Stability analysis of NADP(H)
The stability of NADPH and NADP + were investigated at pH 5. 1.5 mg NADPH and 1.5 mg NADP + were dissolved in 700 ml 0.1 M sodium acetate pH 5. NMR spectra were recorded at 0, 2, 4, 24 and 48 h time intervals at 298 K on a Bruker DRX 500 MHz spectrometer equipped with a 5 mm triple-resonance inverse (TXI) cryoprobe with Z-gradients using a presaturation pulse sequence. Mass spectrometry was where F obs and F calc are the observed and calculated structure-factor amplitudes. ‡ R free as for R cryst using a random subset of data excluded from the refinement. § Estimated coordinate error based on the free R value as calculated by REFMAC (Collaborative Computational Project, Number 4, 1994;Murshudov et al., 1997). } Calculated using RAMPAGE (Lovell et al., 2003). carried out using a Micromass Quadrupole-Time of Flight (Q-TOF) spectrometer. The degradation product of NADPH was identified by mass spectrometry to be 2 0 P-ADP-ribose: m/z (C 15 H 25Àn N 5 Na n O 17 P 3 ) = 727.96 (n = 4, MH + , 90%), 705.98 (n = 3, 100%), 684.00 (n = 2, 45%).

Isothermal titration calorimetry
ITC experiments were performed on an OMEGA isothermal titration calorimeter from Microcal Inc. at 300 K. Purified protein was exhaustively dialyzed with 0.1 M HEPES pH 7.7 and loaded into the sample cell at a final concentration of 30-80 mM. Sufficient ligand was used to obtain at least 80% saturation of the enzyme at the end of the titration, as estimated using the equation given by Turnbull & Daranas (2003). Typically, 35 injections of 7-8 ml were made at 3-4 min intervals from a 300 ml syringe rotating at 300 rev min À1 . Alternatively, the initial 15 injections were of 3-4 ml and the final 30 injections were of 7-8 ml, all at 3-4 min intervals. The heat change accompanying the titration was recorded as differential power by the instrument and determined by integration of the peak obtained. Titrations of ligands to buffer only were performed to allow baseline corrections. The corrected heat change was then fitted using nonlinear least-squares minimization to obtain the dissociation constants K d , the enthalpy of binding ÁH and the stoichiometry n (Wiseman et al., 1989). A stoichiometry of 1, determined from titrations with NADP + under a high c value, was fixed during curve fitting of data obtained with 2 0 P-ADP-ribose under low c-value conditions (Ciulli et al., 2006). At least two experiments were performed at different ligand concentrations to exclude the occurrence of any concentration-dependent phenomena.

Results and discussion
3.1. Crystallization, structure solution and refinement In an attempt to crystallize the ternary complex between KPR, NADPH and pantoate, we explored conditions similar to those used for the KPR-NADP + crystals (Lobley et al., 2005)  Crystal structure of the KPR-2 0 P-ADP-ribose binary complex and comparison with the KPR-NADP + complex. (a) 2 0 P-ADP-ribose electron density. Project, Number 4, 1994;Navaza, 1994Navaza, , 2001 with apo KPR (Matak-Vinkovic et al., 2001) as the probe structure gave a solution with an R cryst of 21.2% and a correlation coefficient of 55.7%. Initial 2F o À F c and F o À F c electron-density maps clearly showed the presence of a nonprotein molecule in the active site. However, no electron density corresponding to pantoate was identified; instead, it was possible to fit the density corresponding to part of NADPH. Ultimately, one molecule of 2 0 P-ADP-ribose was built into this density (Fig.  1a). Although electron density is incomplete for the terminal ribose, the presence of final 2F o À F c electron density at 3-5 around the phosphate groups gave a clear indication of the position of the ligand and enabled its unambiguous modelling. Maximum-likelihood refinement implemented in REFMAC5 (Murshudov et al., 1997) was used to improve the quality of the electron-density maps and facilitate further rebuilding and improvement of the molecular model until no unexplained electron density remained and the R cryst and R free values converged at 16.6 and 19.4%, respectively (Table 1). As in the apo (Matak-Vinkovic et al., 2001) and holo (Lobley et al., 2005) structures, the C-terminal residues were disordered and nine residues were not rebuilt.
3.2. Identification of 2 0 0 0 P-ADP-ribose as the degradation product of NADPH The most likely source of 2 0 P-ADP-ribose in the structure was as a result of degradation of the NADPH used in the cocrystallization experiments. NADPH is known to be an inherently unstable compound (Burton & Kaplan, 1963;Lowry et al., 1961). The hydrolysis of NADPH to 2 0 P-ADPribose has previously been observed in crystals of acetohydroxyacid isomeroreductase grown in the presence of NADPH (Thomazeau et al., 2000).
We monitored the stability of NADPH and NADP + at pH 5 using NMR spectroscopy to confirm the degradation of NADPH to 2 0 P-ADP-ribose under the crystallization conditions. Several NMR spectra were acquired over 48 h. At pH 5, NADPH degraded completely within 4 h, whereas NADP + did not show any degradation. The degradation of NADPH could be monitored by following the disappearance of the resonances at 2.7 (2H, H4, nicotinamide ring), 6.0 (1H, H6, nicotinamide ring) and 7.0 (1H, H2, nicotinamide ring). The disappearance of these peaks corresponded to the appearance of a doublet at 5.5, which is indicative of the formation of 2 0 P-ADP-ribose.
Mass-spectrometry experiments were used to confirm the formation of 2 0 P-ADP-ribose upon incubation of NADPH at pH 5. The peak observed at 728 Da corresponds to the calculated weight of the protonated 2 0 P-ADP-ribose tetrasodium salt. These measurements confirmed the hydrolysis of NADPH to 2 0 P-ADP-ribose.
3.3. The structure of the binary complex of KPR with 2 0 0 0 P-ADP-ribose 2 0 P-ADP-ribose binds to KPR in the active-site cleft between the N-terminal Rossmann-fold domain and the C-terminal -helical domain. The adenine ring of 2 0 P-ADPribose hydrogen bonds to the main-chain amide and carboxyl of Thr118, the 2 0 -phosphate makes hydrogen bonds to Glu256 and Ser244, Asn98 binds the ribose O2 and O3 atoms and provides part of the adenine pocket, the remaining part being formed by the hydrophobic side chains of Leu11 and Ile96, and the pyrophosphate makes hydrogen bonds with backbone amide NH groups of Ala10 and Leu11 and with a structurally conserved water molecule at the 7 GCGALG 12 glycine-rich loop (Bottoms et al., 2002). There is relatively poor electron density in the region corresponding to the terminal ribose (see Fig. 1a), suggesting that the binding of the ribose moiety may be disordered.
The adenine recognition in the structure presented here can be related to the adenine-binding motif described by Denessiouk et al. (2001). In this recognition motif, which is common to many enzymes that bind adenine cofactors, an amino acid forms hydrogen bonds to the N1 nitrogen and the NH 2 group of adenine with its backbone amide NH and carbonyl groups, respectively. Here this residue is Thr118, which forms two hydrogen bonds of 3.2 and 2.6 Å , respectively. The hydrophobic side chains of Leu11 and Ile96 are positioned below the plane of the adenine ring, which is another common feature of the adenine-recognition motif.

Comparison with the KPR-NADP + binary complex structure
A comparison of the KPR-NADP + and KPR-2 0 P-ADPribose structures is shown in Figs. 1(b), 1(c) and 1(d). Previous biophysical studies identified the adenosyl-2 0 -phosphate pocket and the nicotinamide pocket for NADPH as two distinct hot spots for ligand binding (Ciulli et al., 2006). In the KPR-NADP + binary structure, both pockets are occupied by the cofactor molecule (Fig. 1c). However, in the KPR-2 0 P-ADP-ribose binary structure only one pocket is occupied. Here, the adenine group of 2 0 P-ADP-ribose occupies the apolar nicotinamide pocket (Fig. 1b), whilst no electron density was found for binding in the adenosyl-2 0 -phosphate pocket. The ligand is consequently bound in the opposite orientation to NADP + and we henceforth refer to this as the 'reversed binding mode'. Closer inspection shows that many of the interacting residues, e.g. Lys72, Asn98 and Glu256, are common to both modes of binding (Figs. 1b and 1c). Superposition of the two crystal structures reveals that the adenine ring of 2 0 P-ADP-ribose is almost coplanar with the nicotinamide ring of NADP + and the pyrophosphate groups from each structure overlap well (Fig. 1d).
There are two notable differences in the active-site cleft of KPR-2 0 P-ADP-ribose when compared with KPR-NADP + . Firstly, the side chain of Arg31 moves towards Trp74 and Gln75 in order to close the gap left by the absence of cofactor in the adenine-binding pocket. Secondly, the side chain of Lys176 points in the direction of the adenine ring (Fig. 1b) rather than being involved in a hydrogen bond to Thr117 as in the KPR-NADP + complex. Interestingly, in the KPR-2 0 P-ADP-ribose structure Lys176 adopts the conformation research papers proposed for binding ketopantoate in the KPR-NADPHketopantoate ternary complex model (Lobley et al., 2005).

Comparison with acetohydroxyacid isomeroreductase crystal structures
Acetohydroxyacid isomeroreductase (AHIR; EC 1.1.1.86) catalyses the conversion of acetohydroxy acid into dihydroxyvalerate using NADPH as cofactor, which is the second step in the biosynthetic pathway to the branched side-chain amino acids valine, leucine and isoleucine (Biou et al., 1997). It is known that AHIR can catalyze the same reaction as KPR (Primerano & Burns, 1983;Matak-Vinkovic et al., 2001). In a recently published crystal structure of AHIR (PDB code 1qmg), crystallization was carried out in the presence of NADPH and the substrate 2-aceto-2-hydroxybutyrate at pH 7.2 (Thomazeau et al., 2000). Instead of NADPH, 2 0 P-ADPribose was bound in the active site, which was also shown to be a result of cofactor degradation (Thomazeau et al., 2000).
However, in this structure 2 0 P-ADP-ribose was bound in the same conformation as NADPH, as had been previously seen in the crystal structure of the AHIR-NADPH complex (PDB code 1yve; Biou et al., 1997), which was also obtained at pH 7.2.
The observation of the two different binding orientations for 2 0 P-ADP-ribose in AHIR and KPR was unexpected. However, the conditions under which the two proteins were crystallized were not identical and specifically the pH was quite different. The crystallization conditions for KPR were at a pH between 4.0 and 5.0, whereas those for AHIR were at pH 7.2. Furthermore, recent biophysical studies of KPR with cofactor fragments have shown that 2 0 P-ADP-ribose adopts the same binding mode as NADP + at physiological pH (Ciulli et al., 2006). Therefore, it was considered possible that the difference in pH was giving rise to the reversed binding mode observed in the KPR complex. There is precedent for protein crystallization conditions, particularly pH, having a pronounced effect on the binding mode of a ligand . To resolve this issue, the binding interaction of KPR ITC analysis of 2 0 P-ADP-ribose binding to KPR at different pH values. ITC experiments are shown for His 6 -KPR titrated with 2 0 P-ADP-ribose at 300 K in (a) 0.1 M HEPES-HCl pH 7.7 and (b) 90 mM sodium acetate, 5 mM HEPES pH 4.5. Top panels: raw data for titrations of 1-2 mM ligand into 30-60 mM protein. Each peak corresponds to one injection. An initial 1 ml injection was followed by 35 Â 8 ml (a) or 28 Â 10 ml injections (b). Bottom panel: integration of the data, corrected for the heat of dilution. The line represents the least-squares fit to the single-site binding model by the Origin program.
with NADP + and 2 0 P-ADP-ribose were investigated under different pH conditions using ITC.

Isothermal titration calorimetry
ITC studies have been previously conducted with substrates (Lobley et al., 2005), cofactors and cofactor analogues (Ciulli et al., 2006; binding to KPR at pH 7.7. Titrations of His 6 -KPR with 2 0 P-ADP-ribose were repeated at pH 4.5 under conditions similar to those of crystallization. Typical ITC traces are shown in Fig. 2. Similar experiments were also conducted with NADP + as reference. The thermodynamic parameters of the interactions were characterized ( Table 2) and compared with those determined at pH 7.7. The dissociation constant of NADP + at pH 4.5 was 100 mM, an increase of 17-fold relative to pH 7.7 (Table 2). In contrast, 2 0 P-ADP-ribose was found to bind only four times more weakly at pH 4.5 (K d = 250 mM) than at pH 7.7. Moreover, the relative enthalpic and entropic contributions to affinity changed significantly in the case of the fragment, with a large negative ÁH of À51.1 kJ mol À1 at pH 4.5 compared with a ÁH of À12.6 kJ mol À1 observed at pH 7.7. Large changes in ÁH often indicate a change in binding mode (Ward & Holdgate, 2001;Holdgate & Ward, 2005).

Calorimetric analysis of the effect of R31A and N98A mutations
ITC binding studies suggested a change in the binding mode of 2 0 P-ADP-ribose with pH. To provide additional structural evidence for the proposed changes, single point mutations of Arg31 and Asn98 to Ala were constructed. Mutants R31A and N98A of His 6 -KPR had been previously used to probe fragment binding and locate the site of interaction (Ciulli et al., 2006).
Titrations of 2 0 P-ADP-ribose and NADP + against the R31A and N98A mutants of KPR were conducted at pH 4.5 using ITC under the same conditions used for the WT enzyme and were compared with those obtained at pH 7.7 (Ciulli et al., 2006). Thermodynamic data for the KPR mutants relative to WT are reported in Table 3 as ÁÁG (ÁG mutant À ÁG WT ) together with the relative contributions from ÁÁH (ÁH mutant À ÁH WT ) and ÀTÁÁS [ÀT(ÁS mutant À ÁS WT )]. At pH 7.7, the effects of both mutations on 2 0 P-ADPribose binding are consistent with those on NADP + binding, suggesting that the fragment adopts the same binding mode of NADP + . In particular, the side chain of Arg31 appears to be crucial for ligand recognition by binding Changes in protonation states tune the binding mode of 2 0 P-ADP-ribose. (a) At pH 7.7 both the 2 0phosphate group of the ligand and Glu256 will be negatively charged, carrying a À2 and a À1 charge, respectively. Their interaction would be highly unfavourable. The ligand binds in the same orientation observed for NADP + , anchoring the 2 0 -phosphate close to Arg31 and hydrogen bonding to Glu256 via the terminal ribose. (b) Binding mode observed in the KPR-2 0 P-ADP-ribose crystal structure reported here. At pH 4.5 the 2 0 -phosphate becomes protonated (pK = 6.5) carrying a single negative charge. Glu256 is likely to protonate, allowing a favourable hydrogen bond to the 2 0phosphate group. with the 2 0 -phosphate group (Ciulli et al., 2006). In contrast, the results at pH 4.5 for 2 0 P-ADP-ribose binding differ from those of NADP + under the same conditions, which is consistent with the change of binding mode observed in the crystal structure relative to the cofactor. At pH 4.5, the Arg to Ala mutation does not affect the affinity of 2 0 P-ADP-ribose (ÁÁG < 0.4 kJ mol À1 ; Table 3) and thus no longer discriminates against binding the ligand. These results suggest that Arg31 is not involved in the recognition of 2 0 P-ADP-ribose at pH 4.5.
In the reversed binding mode found in the crystal this side chain is at a distance of more than 8 Å from the ligand. Although there are changes in ÁH and TÁS of 8.0 kJ mol À1 for this interaction on mutation of Arg31, these changes are compensatory and can be explained as being a consequence of a change in protein structure or dynamics, rather than evidence of a direct interaction with Arg31. Conversely, the N98A mutation affected the thermodynamic parameters of 2 0 P-ADP-ribose binding at pH 4.5 (Table 3) in a manner that is consistent with the side chain interacting with the adenosyl group of the ligand. There were large changes in ÁÁH and ÀTÁÁS, although these were compensatory and thus resulted in only a small change in ÁÁG. More importantly, these changes were remarkably different from those observed with NADP + at the same pH, further supporting the change in binding mode.

Effect of pH: rationalization of the reversed binding mode
We interpret these data as showing that the binding mode of 2 0 P-ADP-ribose is reversed at pH 4.5, presumably as a consequence of changes in protonation states (Fig. 3). The second pK of the 2 0 -phosphate group will be around 6.4 (Mas & Colman, 1984). At pH 7.7 the 2 0 -phosphate will be doubly charged and will interact strongly with Arg31, whereas interaction with the carboxylate side chain of Glu256, which is also negatively charged, would be unfavourable (Fig. 3a). In contrast, at pH 4.5 the 2 0 -phosphate would be present as the monoanion and Glu256 will be partially protonated. Interaction with the 2 0 -phosphate of the ligand will then be more favourable (Fig. 3b) and is observed in the crystal structure at a hydrogen-bonding distance of 2.4 Å . These changes in protonation states have triggered an otherwise unfavourable contact between a carboxylate and phosphate group and contributed significantly to the reversed binding mode observed in the crystal.

Conclusions
In a combined crystallographic, calorimetric and mutagenic approach, we have shown how changes in pH tune the binding of a ligand, leading to two distinct binding modes of 2 0 P-ADPribose. The crystal structure of the KPR-2 0 P-ADP-ribose complex is to our knowledge the first example of a reversed binding mode at the active site of a dehydrogenase. Such a novel binding mode may be possible for homologous enzymes from the same family. In the case of AHIR, a side-chain carboxylate (Asp202) binds the ribose hydroxyl groups of the cofactor in a similar way as Glu256 in KPR and could accommodate the reversed binding mode at low pH.
Targeting the cofactor-binding site of dehydrogenases such as KPR is a potential drug-design strategy for the development of novel antimicrobial compounds. This result serves as a cautionary tale for rational design using fragment-based approaches, where structural information, primarily from X-ray crystallography, is used for lead optimization. Furthermore, it stresses the importance of using complementary solution techniques such as ITC to supplement crystallography. High-resolution crystal structures of protein-ligand complexes remain the 'gold standard' for structure-based drug discovery, but due regard must be given to the conditions under which they are determined. Table 3 Effect of point mutations on the thermodynamic parameters of nucleotide binding.