Structural analogues of reactive intermediates as inhibitors of glucosamine-6-phosphate synthase and phosphoglucose isomerase

ABB Archives of Biochemistry and Biophysics 450 (2006) 39–49 www.elsevier.com/locate/yabbi

Structural analogues of reactive intermediates as inhibitors of glucosamine-6-phosphate synthase and phosphoglucose isomerase Sławomir Milewski *, Agnieszka Janiak, Marek Wojciechowski Department of Pharmaceutical Technology and Biochemistry, Gdan´sk University of Technology, 11/12 Narutowicza St., 80-952 Gdan´sk, Poland Received 27 January 2006, and in revised form 15 March 2006 Available online 5 April 2006

Abstract The active centers of phosphoglucose isomerase (PGI) and the hexose phosphate isomerase domain (HPI) of glucosamine-6-P (GlcN6-P) synthase demonstrate apparent similarity in spatial arrangement of critical amino acid residues, except Arg272 of the former and Lys603 and Lys485 of the latter. Ten derivatives of D-hexitol-6-P, 5-phosphoarabinoate, or 6-phosphogluconate, structural analogues of putative cis-enolamine or cis-enolate intermediates, were tested as inhibitors of fungal GlcN-6-P synthase and PGI. None of the investigated compounds demonstrated equally high inhibitory potential against both enzymes. 2-Amino-2-deoxy-D-mannitol 6-P was found to be the strongest GlcN-6-P synthase inhibitor in the series, with an inhibition constant equal to 9.0 (±1.0) · 106 M. On the contrary, 5-phosphoarabinoate (5PA) exhibited specificity for PGI, with Ki = 2.2 (±0.1) · 106 M. N-acetylation substantially lowered the GlcN6-P synthase inhibitory potential of 2-amino-2-deoxy-D-glucitol-6-P but strongly enhanced inhibitory potential of this compound towards PGI. Molecular modeling studies revealed that interactions of the C1–C2 part of transition state analogue inhibitors with the respective areas demonstrating different distribution of molecular electrostatic potential (MEP) inside HPI and PGI active centers determined enzyme:ligand affinity. In Escherichia coli HPI, a patch of the negative potential created by Glu488 aided by Val399, supposed to stabilize a putative positively charged intermediate, especially attracts ligands containing 2-amino function. The Arg272, Lys210, and Gly271 peptide bond nitrogen system, present in the corresponding space of rabbit PGI, creates an area of positive MEP, stabilizing cis-enolate intermediate and attracting its structural mimics, such as 5PA.  2006 Elsevier Inc. All rights reserved. Keywords: Glucosamine-6-phosphate synthase; Phosphoglucose isomerase; Transition state intermediate; Molecular modeling

L-Glutamine:D-Fructose-6-phosphate amidotransferase (hexose isomerizing), EC 2.6.1.16, known as glucosamine6-phosphate synthase (GlcN-6-P1 synthase) is an enzyme

*

Corresponding author. Fax: +48 58 3471144. E-mail address: [email protected] (S. Milewski). 1 Abbreviations used: DTT, dithothreitol; GAT, glutamine amide hydrolase domain; GlcNAc, N-acetyl-D-glucosamine; GlcN-6-P, D-glucosamine6-phosphate; HPI, hexose phosphate isomerase domain; MEP, molecular electrostatic potential; PGI, phosphoglucose isomerase; ADGP, 2-amino2-deoxy-D-glucitol-6-phosphate; AADGP, N-acetyl-2-amino-2-deoxyD-glucitol-6-phosphate; ADGPE, 2-amino-2-deoxy-D-glucitol-6-phosphate dimethyl ester; ADMP, 2-amino-2-deoxy-D-mannitol-6-phosphate; DGP, 2-deoxy-D-glucitol-6-phosphate; GP, D-glucitol-6-phosphate; 5PAO, Darabinose-5-phosphate oxime; 5PA, 5-phospho-D-arabinoate; 5PAA, 5phospho-D-arabinoamide; 5PAH, 5-phospho-D-arabinohydroxamate; 6PG, 6-phospho-D-gluconate; 6PGA, 6-phospho-D-gluconoamide. 0003-9861/$ - see front matter  2006 Elsevier Inc. All rights reserved. doi:10.1016/j.abb.2006.03.019

that catalyzes the first committed step in the biochemical pathway leading to the formation of an activated form of N-acetyl-D-glucosamine, namely uridine 5 0 -diphosphoN-acetyl-D-glucosamine. This sugar nucleoside provides N-acetyl-D-glucosamine for biosynthesis of lipopolysaccharides, peptidoglycan and teichoic acids in bacteria; chitin in fungi, insects and crustaceans; as well as glycoproteins, glycosaminoglycans and mucopolysaccharides in mammals. The aminosugar-containing biomacromolecules play an important role in both prokaryotic and eukaryotic cells. In fungi and bacteria, deletion of the respective gene encoding the enzyme is lethal [1,2]. In mammals, a temporary depletion of enzyme activity is acceptable, due to the slow turnover of aminosugar-containing macromolecules and a rapid turnover of the mammalian gene encoding the enzyme, known as GFAT [3,4]. Features mentioned above

40

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

make GlcN-6-P synthase a potential target for antimicrobial chemotherapy, especially its antifungal branch [5]. This issue is of crucial importance due to the emerging challenge of disseminated fungal infections and multi-drug resistance. A very limited repertoire of antifungal chemotherapeutics makes the situation even worse. Exploitation of novel targets, including GlcN-6-P synthase, seems therefore an attractive option. Intensive efforts are thus being continued to design GlcN-6-P synthase inhibitors that may become effective antifungals. Prokaryotic GlcN-6-P synthase is a homodimer, while its eukaryotic counterpart supposedly has a tetrameric structure [6]. A 3D structure of the prokaryotic enzyme [7] but not of the eukaryotic one is known. Each subunit of the Escherichia coli enzyme is composed of two domains: glutamine amide hydrolase (GAT) and hexose phosphate isomerase (HPI) [8]. This arrangement seems to be a common feature of all GlcN-6-P synthases. The catalyzed reaction is complex and involves hydrolysis of glutamine, ammonia transfer to fructose-6-phosphate and isomerization of the formed fructosimine-6-phosphate to glucosamine-6-phosphate. The reaction is initiated upon D-fructose-6-phosphate (Fru-6-P) binding to HPI, followed by binding of L-glutamine to GAT and hydrolysis of glutamine amide. The released ammonia is transferred to HPI, where the resulting iminosugar phosphate is isomerized to GlcN-6-P [9]. The mechanism of fructosimine-6-phosphate isomerization postulated by Teplyakov et al. [10] assumes formation of cis-enolamine as a transition state intermediate. This is analogous to the general mechanism of other ketose/aldose isomerizations, where a cis-endiol is formed, for example, as known for phosphoglucose isomerase, EC 5.3.1.9/PGI/ [11]. In this respect it is worth mentioning that both a native GlcN-6-P synthase and its separately expressed HPI domain exhibit PGI-like activity [12].

A few compounds that mimic putative intermediates of the GlcN-6-P synthase-catalyzed sugar isomerization have been described. Structures of some of them are presented in Fig. 1. D-Arabinose-5-phosphate oxime (5PAO), its methylenephosphate analogue and 2-amino-2-deoxyD-glucitol-6-phosphate (ADGP) are considered structural mimics of cis-enolamine [13,14], while 5-methylphosphono-D-arabinohydroximolactone is an analogue of an fructosimine-6-P intermediate [15]. In their extensive studies Bearne and Blouin [16] suggested a crucial role of the 2-amino function for enzyme inhibitory potency of cisenolamine mimics. In addition, 5-phospho-D-arabinoate (5PA), 6-phospho-D-gluconate (6PG), 5-phospho-D-arabinoamide (5PAA) and 5-phospho-D-arabinohydroxamate (5PAH), considered structural mimics of cis-endiol, have been reported as strong inhibitors of PGI [17–20]. In our present studies, we have determined enzyme inhibitory potential of 10 structural mimics of cis-enolamine or cis-endiol intermediates against Candida albicans GlcN-6-P synthase and yeast phosphoglucose isomerase. Comparative analysis of the obtained data, supported by the results of the molecular modeling, has led to the identification of the structural factors determining inhibitory potency of reactive intermediate analogues in regard to both enzymes. Materials and methods Reagents and analytic methods 2-Amino-2-deoxy-D-glucose-6-phosphate, D-gluconic acid 6-phosphate (6PG), 2-amino-2-deoxy-D-mannose, D-glucitol 6-phosphate (GP), 2-deoxy-D-glucose 6-phosphate, D-glucosamine-6-phosphate and sodium borohydride were purchased from Sigma Aldrich, St. Louis, MO. All other chemicals were of the highest purity

Fig. 1. Structures of putative transition state intermediates in HPI- and PGI-catalyzed reactions and their structural analogues.

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

commercially available. 1H and 31P NMR spectra were obtained with Varian Unity Plus 500 at 200 and 500 MHz, respectively. The deuterated solvents were used as an internal lock for 1H NMR and H3PO4 (85% w/v in D2O) as an external lock for 31P NMR. Specific rotations were measured on a Rudolph Autopol II digital polarimeter. Synthesis of inhibitors 2-Amino-2-deoxy-D-glucitol-6-phosphate (ADGP) and 2-deoxy-D-glucitol-6-phosphate (DGP) were synthesized from D-glucosamine-6-phosphate and 2-deoxy-D-glucose6-phosphate, respectively, according to the methods described by Bearne and Blouin [16]. N-Acetyl-2-amino2-deoxy-D-glucitol-6-phosphate (AADGP) and ADGP dimethyl ester (ADGPE) were prepared from ADGP, as described previously [21]. 5-Phospho-D-arabinoate (5PA) was prepared from arabinose-5-phosphate by bromine oxidation, following the protocol of Horecker [22]. 5-Phospho-D-arabinoamide (5PAA) and 6-phospho-D-gluconoamide (6PGA) were synthesized by direct amonolysis of respective lactones, according to the method of Hardre and Salmon [18]. 2-Amino-2-deoxy-D-mannitol-6-phosphate (ADMP) was prepared by a two-step procedure. In the first step, 2-amino-2-deoxy-D-mannose was phosphorylated enzymatically to give 2-amino-2-deoxy-D-mannose-6-phosphate, following a procedure described previously for preparation of kanosamine-6-phosphate from kanosamine [23]. Yield 0.34 g, 70%. 1H NMR (D2O) d: 3.5 (m, 1H), 3.6 (m, 1H), 3.9 (m, 2H), 4.0 (m, 2H), 5.2 (d, 1H, J = 7.5 Hz); 31 P NMR (D2O) d: 1.6 (s, 1P). In the second step, 2-amino2-deoxy-D-mannose-6-phosphate was reduced with sodium borohydride, following a procedure of Bearne and Blouin [16]. The substance was obtained as a thick oil (0.22 g, 76% yield). 1H NMR (D2O) d: 3.55 (m, 1H), 3.7 (m, 2H), 3.9 (m, 2H), 4.0 (m, 2H), 4.15 (m, 1H); 31P NMR (D2O)  d: 2.56 (s, 1P); ½a20 D  3:3 (c = 0.4, H2O). All of the synthesized compounds were at least 99% pure, as judged by HPLC, analysis, conducted as described previously [21].

41

assay, as described by Noltmann [26]. Incubation mixtures contained 0–0.5 mM D-Fru-6-P, 0.4 mM NADP+, 1 mM EDTA, Glc-6-P dehydrogenase, 2.6 units/mL, phosphoglucose isomerase, 0.3 units/mL and a given inhibitor in appropriate concentration, in 50 mM Tris–HCl buffer, pH 8.0. The same conditions were used for determination of the PGI-like activity of GlcN-6-P synthase. Data were plotted as Michaelis–Menten graphs. Kinetic data were subjected to the nonlinear regression analysis using the Enzyme Kinetics Module 1.10 for Sigma Plot software. The inhibitory constants (Ki) were determined in triplicate, to give the mean values ± SD. Molecular modeling Models of the complexes built for the modeling studies were based on the geometry of the crystal structure of E. coli GlcN-6-P synthase HPI complexed with ADGP (1mos) and the rabbit phosphoglucose isomerase complexed with 5-phospho-D-arabinohydroxamate (1koj). The structures of ligands studied were built by means of InsightII molecular modeling package from Accelrys [27]. To verify an applicability of the force fields, the geometries of relevant molecular fragments obtained by molecular modeling were compared with those available in various crystallographic databases. Conformational analysis of ligands planed for the synthesis was performed by the series of stochastic dynamics simulations with the local elevation algorithm, using the GROMOS96 software [28]. The putative geometries of the ligand–enzyme complexes were built by means of the AutoDock program [29]. Lamarckian Genetic Algorithm (LGA) was used as the search method. Ligands were flexible during the calculations with all single bonds marked for rotation. Among the 50 top scoring complexes, the one with a geometry best resembling that of the crystal geometry of ADGP and 5PAH, with respect to the phosphate and the three adjacent ACH(OH)A units, was chosen for further analysis. A distribution of the molecular electrostatic potential (MEP) inside the ligand binding sites of HPI and PGI were calculated by means of the University of Houston Brownian Dynamics program (UHBD) [30] and visualized on the molecular surfaces with the GRASP software.

Enzyme purification and activity assays Results and discussion C. albicans GlcN-6-P synthase, overproduced by Saccharomyces cerevisiae strain YRS C-65, was purified to near homogeneity, as described previously [24]. Enzyme activity was determined by the modified Elson–Morgan method [25]. Enzymatic reactions were carried out in 25 mM sodium phosphate buffer, pH 6.9, containing 1 mM EDTA, 1 mM DTT, D-Fru-6-P (0.5–7.5 mM), L-glutamine (0.625–10 mM), GlcN-6-P synthase (0.1–0.2 lM) and a given inhibitor in appropriate concentration. Phosphoglucose isomerase preparations (S. cerevisiae and rabbit) were purchased from Sigma. Enzyme activity was determined using the Glc-6-P dehydrogenase-coupled

Structural comparison of HPI and PGI Phosphoglucose isomerase (PGI) and GlcN-6-P synthase, or, more precisely, its isomerase domain (HPI) act on the same substrate, D-fructose-6-P, and catalyze the same type of reaction, i.e., ketose/aldose isomerization involving enolic intermediate. Moreover, overexpressed recombinant HPI exhibits Fru-6-P isomerizing activity [12]. Comparison of the ribbon representations of the E. coli HPI/ADGP (1mos) and the rabbit PGI/5PAH (1koj) shown in Fig. 2 reveals some extent of similarity

42

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

Fig. 2. Comparison of the overall fold of the HPI/ADGP complex/1mos/ (left) with that of the rabbit PGI/5PAH complex/1koj/ (right).

between both folds, although substantial differences may be clearly seen. HPI consists of two topologically identical subdomains of equal size. Each domain has ab structure composed of a five-stranded parallel b-sheet flanked on either side by a-helices, thus giving a three-layer aba-sandwich. On the contrary, the PGI subdomains are dissimilar. The small subdomain contains a five-stranded parallel bsheet and the large one—the six-stranded b-sheet core— composed of four parallel and two antiparallel b-strands. PGI is generally larger than HPI and contains some obvious ‘‘additional’’ peripheral a-helices. Despite these differences, the overall shape of both proteins follows the common fold of a flavodoxin type [31]. Both proteins are catalytically functional as homodimers, with a histidyl residue of one monomer serving the active center of the other [9,32]. It may be of some interest that any apparent similarity of the 3D structures cannot be directly deduced from comparison of the respective polypeptide chain sequences, as any attempts to align them reveal no more than a few percent of sequence identity (data not shown). To compare relative positioning and orientation of functional active site amino acid residues of PGI and HPI, we superimposed the two images. Analysis of the resulting model, shown in Fig. 3 reveals that a phosphate-binding loop of PGI, composed of Ser209, Thr211 and Thr214, aided by Ser 159, is well matched by the Ser347–Ser349–Thr352–Ser303 system of HPI. The His388 of PGI, known to be involved in sugar ring opening and the main chain hydroxyls binding [32], nicely overlaps its HPI counterpart, i.e., His504. Finally, little difference has been found between relative locations and orientations of Glu488 of HPI and Glu357 of PGI—both identified as key residues participating in proton transfer between C1 and C2 of respective intermediates [9,33]. Significantly, the key residues: Lys603 of HPI and Arg272 of PGI do not have any obvious counterparts of the same chemical character. Therefore, it is worth mentioning that Arg272 is likely to be responsible for the stabilization of the negatively charged cis-enolate intermediate at the PGI active center [33],whereas Lys603 was found to form

a Schiff base upon reaction with the C2 carbonyl of Fru6-P at the HPI active site [34]. At the PGI active site there is also no counterpart of Lys485 in HPI. e-Amino group of this residue, H-bonded to Glu481 carboxylate aided by Thr302 hydroxyl, was suggested to act as a proton acceptor in deprotonation of the O1 hydroxyl of cis-enolamine intermediate [10]. In the mechanism postulated for PGI, proton exchange between O1 and O2 is likely to be mediated by a water molecule, without participation of any active site residue [35]. In view of the above findings, it is not surprising, that location and conformation of the phosphate residues and consecutive three main chain ACH(OH)A units of ADGP and 5PAH are practically the same, as well as a pattern of their interactions with the respective active center residues.. Differences are observed only for the C1–C2 region of ADGP and its 5PAH counterpart, i.e., the hydroxamate group. In the HPI/ADGP complex, the C2-bonded amino group of the inhibitor interacts directly with the main chain carbonyl oxygen of Val399 and is supposed to interact with e-amino group of Lys603 through the water molecule. The C1 hydroxyl is H-bonded to e-amino group of Lys485 and hydroxyl oxygen of Thr302. In the PGI/5PAH complex, the hydroxamate carbonyl oxygen atom, located at the position corresponding to the C2-bonded amino group of ADGP, is not directly H-bonded to any active center residue. The only possible interactions are likely to be mediated by water with Lys210. The N–OH hydroxyl, corresponding in turn to C1–OH of ADGP, makes hydrogen bonds to Arg272. Relative positioning of both inhibitors in respect to glutamates—Glu488 of HPI and Glu357 of PGI—is obvious in light of the well evidenced participation of these catalytic residues in hydrogen transfer between C1 and C2 of the substrate [10,35]. Detailed analysis of the HPI/ADGP interaction pattern led us to the preliminary assumption that the C2-bonded amino group of a possible ADGP epimer at C2 should be perfectly positioned to establish more favorable interactions with Glu488.

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

43

Fig. 3. (A) A stereo-view stick model of the active site of the HPI/ADGP complex (light green) overlayed onto the active site of the PGI/5PAH complex (red). The active sites contain the His504* and His388* residues taken from the neighboring subunit of the functional homodimers. The inhibitor molecules are tilted with respect to each other. (B) Essential ligand:enzyme interactions in the PGI/5PAH complex (left) and the HPI/ADGP complex (right).

Fig. 6. Comparison of the molecular electrostatic potential distribution inside the active sites of E. coli HPI (left) and rabbit PGI (right). Based on the structures of the HPI/ADGP and PGI/5PAH complexes. Ligands are visualized as wire models.

44

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

Table 1 Competitive inhibition of GlcN-6-P synthase and phosphoglucose isomerase by structural analogues of the reactive intermediates Compound

5PA (1) 5PAA (2) 6PG (3) 6PGA (4) ADGPc (5) AADGPc (6) ADMP (7) ADGPE (8) GPc (9) DGPc (10) 5PAH a b c

Kinetic parameters and free energy of enzyme inhibition C. albicans GlcN-6-P synthase (GlcN-6-P synthesizing activity)

S. cerevisiae PGI

Ki(F6P) (lM)

KM/Ki

DG (kcal/mol)

Ki(F6P) (lM)

KM/Ki

DG (kcal/mol)

>20,000 620 ± 50 >20,000 3400 ± 200 35 ± 4 840 ± 30 9.0 ± 1.0 680 ± 40 1950 ± 250 14,300 ± 800 NTa

— 2.27 — 0.41 40.3 1.68 157 2.07 0.72 0.098 —

— 4.45 — 3.46 6.18 4.27 7.0 4.39 3.76 2.56 —

2.2 ± 0.2 4.6 ± 0.1 148 ± 11 560 ± 16 12,300 ± 500 76 ± 8 17,200 ± 700 >20,000 40 ± 2 680 ± 40 0.23b

31.4 15.2 0.47 0.12 0.006 0.92 0.004 — 1.75 0.103 350

7.85 7.40 5.31 4.51 2.65 5.71 2.45 — 6.10 4.39 9.29

NT, not tested. Inhibition constant from Hardre et al. [19]. Inhibition constants for E. coli GlcN-6-P synthase: ADGP, 25 lM [14] and 19 lM [16]; AADGP, 28 lM [36]; DGP, 15 mM [16]; GP, 2.4 mM [12].

Inhibition of GlcN-6-P synthase and phosphoglucose isomerase by transition state analogues A series of compounds 1–10 (Fig. 1), putative structural mimics of the respective cis-enolamine or cis-endiol intermediates, were tested for their GlcN-6-P synthase and PGI inhibitory activity, to identify the structural factors that determine an effective and selective binding at the HPI center of GlcN-6-P synthase. Compounds studied by us belong to two structural groups: oxidized forms of arabinose-5-P and glucose-6-P and their amides: 5PA 1, 6PG 3, 5PAA 2, and 6PGA 4 and reduced forms of hexose phosphates and their derivatives: ADGP 5, AADGP 6, ADMP 7, ADGPE 8, GP 9 and DPG 10. Some of these compounds are commercially available or methods of their synthesis have been already published. Two of them, namely ADMP 7 and ADGP diethyl ester 8, are completely new compounds. Inhibitory potential of compounds 1–10 in regard to C. albicans GlcN-6-P synthase and yeast and (or) rabbit phosphoglucose isomerases has been quantified. The determined inhibitory constants, KM/Ki ratios and calculated free energy of binding (DG = RT ln Ki) values are summarized in Table 1. All compounds demonstrating measurable inhibitory activity against GlcN-6-P synthase behaved as competitive inhibitors in regard to D-Fru-6-P and non-competitive in regard to L-Gln. This finding confirms binding of the investigated inhibitors to the enzyme at its HPI active site. Four of ten studied compounds were previously tested by other authors as inhibitors of the E. coli enzyme. For three of them we found the Ki values against fungal enzyme almost identical with those reported for the bacterial one (see the footnote in Table 1), thus indirectly confirming close similarity or even apparent identity of HPI active sites of both enzymes versions. The only substantial difference was found in the case of AADGP 6, for which Badet-Denisot et al. reported Ki = 28 lM [36]. In accordance with the

theoretical assumptions, the ADGP epimer at C2, i.e., ADMP 7, was found to be the strongest enzyme inhibitor in the series. Actually, this compound seems to be the most efficient transition state analogue inhibitor of GlcN-6-P synthase known so far, superior to ADGP and arabinose oxime 5-phosphate. Other examined compounds demonstrated lower enzyme inhibitory potency than ADGP. Diminished activity of AADGP 6 was actually expected in light of the results of previous studies of Bearne and Blouin [16], although this finding is not consistent with the earlier results obtained by Badet-Denisot et al. [36]. One may assume that the observed 24-fold increase of the Ki value upon N-acetylation of ADGP is due to the effect of charge loss and/or steric hindrance provided by the N-acetyl function. The Ki value similar to that for AADGP was found for ADGP dimethyl ester 8. It is therefore worth mentioning that the difference in free energy of binding between ADGP and its N-acetyl and ester derivatives is close to 2 kcal/mol, i.e., lower than a generally accepted value for the single hydrogen bond contribution (3–6 kcal/mol). On the other hand, values approaching this lower limit were found in the case of two ADGP analogues lacking the 2-amino function, i.e., D-glucitol-6-P 9 and 2-deoxy-D-glucitol-6-P 10. 5PAA 2 demonstrated moderate inhibitory potential, as Ki of this compound against C. albicans GlcN-6-P synthase was comparable to that of AADGP 6. The respective value for its homologue, 6PGA 4, was about 5-fold higher. No inhibitory effect was found in the case of carboxylate derivatives, 5PA 1 and 6PG 3, at 20 mM, so that the respective Kis could not be actually determined. Compounds 1–10 were found to be competitive inhibitors of S. cerevisiae and rabbit PGI and seven of them demonstrated medium or relatively high activity. No significant differences were found in the case of Ki values determined for the yeast and rabbit version of the enzyme, and only the data obtained for the former one are presented in Table 1.

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

On the contrary to the results obtained for GlcN-6-P synthase, ADGP 5, its dimethyl ester 8 and ADMP 7 demonstrated the lowest inhibitory potential, while 5PA 1 and 5PAA 2 were found to be the strongest PGI inhibitors in the series. The difference in calculated DG values between data obtained for ADGP and 5PAA is well inside the range expected for a single hydrogen bond contribution. Data obtained for AADGP 6 indicate that N-acetylation of the 2-amino function enhances the PGI-inhibitory potential by nearly 3 kcal/mol. This effect should be attributed to the charge loss. It is noteworthy that the Ki value for AADGP 6 is nearly identical with that found for GP 9 but more than an order of magnitude lower than that of DGP 10. Table 1 contains also the literature data for 5phospho-D-arabinohydroxamate (5PAH), the strongest PGI inhibitor known so far [19], never tested as a potential inhibitor of GlcN-6-P synthase. Any direct comparison of the Kis of a given inhibitor towards both enzymes studied in this work is of a little value, since the Michaelis constant of GlcN-6-P synthase for Fru-6-P is much higher than that of PGI (1.41 and 0.07 mM, respectively). The relative enzyme inhibitor affinities are therefore best reflected by the KM/Ki ratios. Analysis of the respective calculated data shown in Table 1 reveals that from this point of view, the tested compounds could be divided into three groups: the first one containing compounds of KM/Ki ratio (GlcN-6-P synthase)  KM/Ki ratio (PGI); the second of KM/Ki ratio (GlcN-6-P synthase)  KM/Ki ratio (PGI); and the third one of compounds demonstrating comparable KM/Ki ratios towards both enzymes. The first group comprises 5PA and 6PG, the second one involves ADGP, ADGPE and ADMP. All the other inhibitors fall into the third group. It is noteworthy that although all compounds 1–10 and 5PAH show structural analogy to respective reactive intermediates, a comparative analysis of the KM/Ki values indicates that only ADGP 5 and ADMP 7 in regard to GlcN-6P synthase, as well as 5PA 1, 5PAA 2 and 5PAH in regard to PGI, demonstrated sufficiently high relative inhibitory potential to be considered true transition state analogue inhibitors of the respective enzyme. Molecular modeling and docking of inhibitors To find out a molecular basis for the observed structure– enzyme inhibitory potency relationships we have performed a two-step in silico studies of the tested inhibitors. For the purpose of the first step, a known crystallographic structure of the E. coli HPI complex with ADGP 5 (1mos) was taken as a starting point. Such an approach was fully justified by the fact that although the respective structure of the fungal enzyme is not known, the amino acid sequences of C. albicans and E. coli HPIs exhibit 44% identity and 80% similarity [6]. Particularly, counterparts of all residues identified in the E. coli enzyme as crucial for substrate binding and catalysis, are present in the C. albicans GlcN-6-P synthase as parts of the highly conserved frag-

45

Fig. 4. Energy-minimized conformer of ADGP resulting from the conformational analysis, overlayed onto a real structure of the inhibitor taken from the HPI/ADGP crystal complex (1mos).

ments. Thus we assumed that conclusions drawn from the modeling studies employing HPI of bacterial GlcN-6P synthase should be applicable for interpretation of inhibition data found for C. albicans GlcN-6-P synthase. ADGP 5 is the only transition state analogue inhibitor of GlcN-6-P synthase for which the actual conformation adopted at the enzyme active site is known. We compared this conformation with that resulting from the conformational analysis, performed following the local elevation algorithm, using the GROMOS96 software. Both structures were nearly identical, as shown in Fig. 4, thus confirming an applicability of the force field used in our calculations. At the next stage we performed a conformational analysis of all 10 compounds that have been tested by us. The conformational space of each compound thus obtained was then analyzed and the optimized conformers were docked into the active center of E. coli HPI. The similar procedure was also applied for docking ligands into the PGI active site, with the PGI/5PAH complex structure (1koj) taken as the starting point. Structures of some of the obtained complexes are shown in Fig. 5. Inspection of the ligand:enzyme interaction pattern found for the C1–C2 (or N1–C2) part of compounds 1– 10 docked into the PGI active site, supplemented by the known crystallographic data available for PGI complexes with 5PA 1, 5PAH and 6PG 3 [33,35,37], provide partial explanation for differences in enzyme inhibitory potentials of these compounds. The N–OH oxygen of 5PAH is locat˚ ) from Arg272 ed at the hydrogen bonding distance (2.9 A side chain guanidynyl moiety nitrogen atom, whereas in ˚ the PGI/5PA complex, a water molecule located at 2.8 A

46

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

Fig. 5. Geometries of the selected HPI/ligand and PGI/ligand complexes. (A) PGI/ADGP; (B) PGI/5PAA; (C) PGI/AADGP; (D) HPI/ADMP; (E) HPI/ 5PAA; (F) HPI/AADGP. All the structures result from the docking experiments.

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

from Arg272 guanidynyl, mediates interaction of this residue with the carboxylate function of the ligand. It seems that thus created favorable interactions prevail over possible repulsive effects that may result from interactions with Glu357. Replacement of the carboxylate function of 5PA 1 with amide in 5PAA 2 makes little difference to the enzyme:inhibitor interaction pattern, as the amide carbonyl ˚ from Arg272 (Fig. 5). On the oxygen is located at 2.8 A other hand, elongation of the inhibitor main chain moves the carboxylate or amide function slightly closer to Glu357 but further away from Arg272 (data not shown). Possibly, the stabilizing effect of Arg272 is thus reduced and the non-favorable interaction with Glu357 prevails. This is disadvantageous, as reflected by the lower PGI inhibitory potency of 6PG 3 and 6PGA 4 (Table 1). A different situation was found in the case of HPI/5PA and HPI/6PG complexes. The negatively charged carboxylate functions of the ligands are there forced to face the Glu488 residue of HPI, thus creating a possibility of nonfavorable, repulsive interactions. Since there is no counterpart of the Arg272 of PGI at the HPI binding site, that could attract the carboxylate function of the ligand and thus stabilize the enzyme–ligand complex, the repulsive effect should dominate. This situation nicely explains the observed lack of any measurable GlcN-6-P synthase inhibitory activity of 5PA 1 and 6PG 3 (Table 1). Obviously, the repulsive interactions between the negatively charged functional groups of the ligands and the Glu488 residue must be reduced or even abolished when the carboxylate function of the ligand is replaced with an amide. The amide moiety of 5PAA 2 docked into the HPI active site was located at the hydrogen bonding distance from Glu488 side chain carboxylate, Val399 main chain carbonyl oxygen and Lys485 e-amino group. All the resulting enzyme:ligand interactions are likely to favorably contribute to the stability of the enzyme:ligand complex. Elongation of the inhibitor main chain by a single ACH(OH)A unit enforces a shift of the amide function towards Val399 at the expense of expanding its distance to Glu488. This in turn, seems to be slightly disadvantageous, as reflected by the difference of the Ki values found for 5PAA 2 and 6PGA 4 (Table 1). The pattern of enzyme:ligand interactions found in the HPI/ADGP complex, shown in Fig. 3, well explains a high GlcN-6-P synthase inhibitory potency of ADGP 5. The N-acetylated derivative of ADGP, namely AADGP 6, when docked into the HPI active site, demonstrated enzyme:inhibitor interaction pattern similar to that found for the parent compound. The acetyl residue introduced at the 2-amino function of ADGP 5 does not seem to pose any substantial steric hindrance to its close neighborhood, since the N-acetyl residue protrudes out of the binding pocket. In the complete enzyme molecule, composed of the GAT and HPI domains, a methyl moiety of the N-acetyl function of AADGP 6 is supposed to point towards the entrance to the hydrophobic channel connecting the two active sites of GlcN-6-P synthase. In the HPI/ADGP complex, the protonated amino group of the inhibitor is

47

likely to establish very favorable interactions with the side chain carboxylate of Glu488 and with the main chain oxygen atom of Val399. These interactions are supposed to be significantly weakened in the HPI/AADGP complex, most probably due to the charge loss at the acetylated nitrogen atom. Replacement of the 2-amino function of ADGP 5 with hydroxyl does not affect the general pattern of the enzyme:inhibitor interaction, as none significant qualitative or quantitative difference were observed between the HPI/ GP and HPI/ADGP complexes (detailed data not shown). We therefore suggest that the lower affinity of GP 9, reflected by its higher Ki value should be entirely attributed to the weakening of the interactions with Glu488 and Val399, expected upon substitution of the positively charged group with the neutral one. Obviously no favorable interactions with Lys603, Val399 or Glu488 are possible in the case of DGP 10, thus giving rise to the substantial reduction of its affinity to the enzyme active center. As expected, advantageous changes in the enzyme:ligand interaction pattern were created upon inversion of C2 configuration in the ADGP molecule. Analysis of the HPI/ADMP complex reveals that a possibility of interaction with Val399 was not lost, but most notably, the distance between the amino function of ADMP 7 and carboxylate of Glu488 became significantly shorter ˚ ). (4.3 fi 3.1 A In the PGI/ADGP complex, the 2-amino function of the ˚ disdocked ligand faces the Arg272 residue at the 3.2 A ˚ . A similar situation was also tance and Glu357 at 3.9 A found for ADGPE 8 and ADMP 7 (data not shown), so that interactions of all these ligands with both crucial active center residues could be possible and a molecular basis for their extremely low PGI-inhibitory potential could not be deduced from the results of the docking experiments. The conclusive explanation was found from calculation of a molecular electrostatic potential distribution inside HPI and PGI active centers. Results are shown in Fig. 6. It turned out, that the centers, although structurally similar, generated quite different MEP distribution pattern. The parts of both active sites responsible for binding the phosphate and adjacent three or four ACH(OH)A units of the ligand similarly generate areas of a positive potential, suitable for binding of negatively charged regions of molecules. However, in the area facing the C1–C2 part of the ligand, HPI demonstrates a large patch of a negative potential, generated mostly by the carboxylate of Glu488 and the amide carbonyl oxygen atom of Val399. Such a pattern is suitable for an efficient attraction of the positively charged parts of the ligand and disfavors ligands bearing a negative charge in this area. The MEP distribution in the respective area of PGI is different. Due to the particular arrangement of the positively charged Arg272 and Lys210 side chains, as well as the Gly271 and Arg272 main chain amide nitrogen atoms, the area of a positive potential extends to almost entire PGI binding pocket and thus compensates for an influence of Glu357. Thus, ligands of the opposite ionic character in the C1–C2 area are supposed to exhibit

48

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

extremely different affinities to both sites. It is therefore not surprising that ligands with negatively charged C1–C2 ‘‘heads,’’ like 5PA 1 and 6PG 3, are relatively strong inhibitors of PGI, but not of GlcN-6-P synthase, while the positively charged ADGP 5 and ADMP 7 exhibit weak, if any affinity to PGI but strong to HPI. Conclusions and implications for the mechanism of sugar phosphate isomerization catalyzed by HPI It is noteworthy that although the putative cis-enediol intermediate of the PGI-catalyzed reaction is structurally similar to the cis-enolamine intermediate of GlcN-6-P synthase, HPI exhibits a phosphoglucose isomerase-like activity and the active sites of PGI and HPI demonstrate a substantial extent of similarity, none of the tested analogues was found to be a strong inhibitor of both enzymes. Comparison of the KM/Ki ratios and results of the molecular modeling and docking studies allows formulation of some general conclusions. Compound containing appropriately positioned carboxylate functions, namely 5PA 1 exhibits good affinity to the PGI active site and very poor if any to that of HPI. Quite the opposite was found for 2-amino-hexitol phosphates, ADGP 5 and ADMP 7. This regularity may be well explained taking into account the relative positioning of the critical residues at the active centers of HPI and PGI and resulting distribution of the molecular electrostatic potential inside both ligand-binding sites. Inspection of a crystal structure of the rabbit muscle PGI complex with 5PA [33], clearly shows strong, favorable interactions between the negatively charged carboxylate function of the inhibitor and a positively charged side chain guanidinyl moiety of Arg272 of the enzyme, supported by Lys210 and Gly271. This network of residues is responsible for stabilization of the high-energy intermediate, cis-enolate, bearing a negative charge at the position corresponding to the carboxylate oxygen atoms of compounds 1 and 2. Such a complex of the positively charged residues is absent from the HPI active site. Instead, the positively charged 2-amino groups of compounds 5 or 7 may establish favorable interactions with the negatively charged side chain carboxylate of Glu488. In light of the above observation, it is not surprising that N-acetylation of the 2-amino group of ADGP affords a derivative that exhibits weaker than the parent compound affinity to the active site of HPI but much stronger to that of PGI. In light of these present findings, an inhibitory potential of D-arabinose-5-phosphate oxime (5PAO) towards GlcN-6-P synthase seems to be of a special interest. Two substantially different inhibitory constants for this compound in respect to E. coli GlcN-6-P synthase have been reported; Ki = 14.3 lM was determined for this compound by Le Camus et al. [13], while Bearne and Blouin [16] reported the value of 1.2 mM. Unfortunately, we did not have 5PAO in our hands, so that were not able to determine the Ki value against the C. albicans enzyme. Nevertheless, some suggestions resulting from our present data can be given. While the N–OH oxygen of 5PAO is supposed to be perfect-

ly positioned to establish favorable interactions with Lys485, there are not any substituents at C2 that could interact with Glu488, Val399 and Lys603. One may therefore expect rather poor enzyme inhibitory properties of 5PAO, closer to that reported by Bearne and Blouin [16] than to that suggested by Le Camus et al. [13]. In our opinion, the previously postulated mechanism of sugar phosphate isomerization at the HPI active site and a role of the Lys603 residue should be re-evaluated in light of the novel findings. Teplyakov et al. [10], in their studies employing bacterial GlcN-6-P synthase, suggested that the Lys603:Fru-6-P Schiff base, transiminated with ammonia, gives rise to the nonprotonated fructosimine-6-P, which upon proton abstraction from C1 affords a high-energy intermediate bearing a negative charge at the C2linked nitrogen atom. The authors presumed that such an unusual intermediate could be stabilized by the dipole of helix 301–317 and by the positively charged Lys603. In our opinion such a possibility is rather unlikely. The actual charge status of Lys603 is not known, but at least two arguments are in favor of its nonprotonated form: (i) only a non-protonated e-amino group may act as a nucleophile attacking carbonyl carbon atom of the substrate to form a Schiff base; (ii) the e-amino group of Lys603 is located at a ˚ ) from Glu396, which in turn is H-bonding distance (2.69 A H-bonded to the Thr402, Lys403 and Ala404 main chain ANHA groups. It seems very likely that such a system stabilizes an unprotonated form of the side chain amino group of Lys603, in a manner similar to that suggested previously for Lys485 [10]. It is noteworthy that the residues 396–404 in the E. coli enzyme and their counterparts in other GlcN-6-P synthases constitute a highly conserved part of the amino acid sequence [6]. There is little doubt that the unprotonated form of e-amino group of Lys603 would be much less efficient in stabilization of the putative negatively charged fructosimine-6-P intermediate. Instead, this group aided by Val399 main chain carbonyl oxygen, should efficiently stabilize a protonated form of fructosimine-6-P, as well as an uncharged cis-enolamine formed upon proton abstraction from C1 by Glu488. Obviously, the protonated fructosimine-6-P intermediate would not be rather stabilized by the protonated Lys603. Therefore, in our opinion, a mechanism of Fru-6-P isomerization catalyzed by GlcN6-P synthase is supposed to follow the lower of the two alternative pathways shown in Fig. 7. Results of our present studies provide further evidence confirming this option. Importance of the 2-amino function for the efficient binding of transition state analogue inhibitors of GlcN-6-P synthase was already demonstrated by Bearne and Blouin [16] but our results, especially the lowest Ki value of ADMP among the tested compounds and diminishment of enzyme inhibitory activity of ADGP upon N-acetylation, suggest that the 2-amino group should exist in a protonated state to ensure stronger ligand binding at the HPI active site. This finding, as well as a very poor inhibitory potency of 5PA and 6PG, support possibility of the existence of a protonated transition intermediate in the GlcN-6-P

S. Milewski et al. / Archives of Biochemistry and Biophysics 450 (2006) 39–49

49

Fig. 7. Schematic representation of two alternative pathways of sugar phosphate isomerization catalyzed by the HPI domain of GlcN-6-P synthase.

synthase-catalyzed isomerization pathway and disfavor presence of any transition species negatively charged at the C2-linked nitrogen atom. Acknowledgments We thank Dr. Ryszard Andruszkiewicz for the synthesis of 5PA and Mrs. Julia Nowak and Sylwia Murawska for the synthesis of ADMP. This work was supported by the grant 3 P05F 012 25 from the Ministry for Science and Information Society Technologies and in part by the grant from the Faculty of Chemistry, Gdan´sk University of Technology. References [1] W.L. Whelan, C.E. Ballou, J. Bacteriol. 124 (1975) 1545–1557. [2] M. Sarvas, J. Bacteriol. 105 (1971) 467–471. [3] C.J. Bates, W.R. Adams, R.E. Handschumacher, J. Biol. Chem. 241 (1966) 1705–1712. [4] S. Marshall, V. Bacote, R.R. Traxinger, J. Biol. Chem. 266 (1991) 10155–10161. [5] E. Borowski, Farmaco 55 (2000) 206–208. [6] S. Milewski, Biochim. Biophys. Acta 1597 (2002) 173–192. [7] A. Teplyakov, G. Obmolova, M.-A. Badet-Denisot, B. Badet, J. Mol. Biol. 313 (2001) 1093–1102. [8] M.-A. Denisot, F. Le Goffic, B. Badet, Arch. Biochem. Biophys. 288 (1991) 225–230. [9] A. Teplyakov, G. Obmolova, M.-A. Badet-Denisot, B. Badet, B.I. Polikarpov, Structure 6 (1998) 1047–1055. [10] A. Teplyakov, G. Obmolova, M.-A. Badet-Denisot, B. Badet, Protein Sci. 8 (1999) 596–602. [11] I.A. Rose, Philos. Trans. R. Soc. Lond–Biol. Sci. 293 (1981) 131–143. [12] C. Leriche, M.-A. Badet-Denisot, B. Badet, J. Am. Chem. Soc. 118 (1996) 1797–1798. [13] C. Le Camus, M.-A. Badet-Denisot, B. Badet, Tetrahedron Lett. 39 (1998) 2571–2572. [14] M.-A. Badet-Denisot, C. Leriche, F. Masserie, B. Badet, Bioorg. Med. Chem. Lett. 5 (1995) 815–820.

[15] C. Le Camus, A. Chassagne, M.-A. Badet-Denisot, B. Badet, Tetrahedron Lett. 39 (1998) 287–288. [16] S.L. Bearne, C. Blouin, J. Biol. Chem. 275 (2000) 135–140. [17] J.M. Chirgwin, E.A. Noltmann, J. Biol. Chem. 250 (1975) 7272–7276. [18] R. Hardre, L. Salmon, Carbohydr. Res. 318 (1999) 110–115. [19] R. Harde, C. Bonnette, L. Salmon, A. Gaudemer, Bioorg. Med. Chem. Lett. 8 (1998) 3435–3438. [20] R. Hardre, L. Salmon, F.R. Opperdoes, J. Enzyme, Inhibition 15 (2000) 509–515. [21] A.M. Janiak, M. Hoffmann, M.J. Milewska, S. Milewski, Bioorg. Med. Chem. 11 (2003) 1653–1662. [22] B.L. Horecker, Meth. Enzymol. 3 (1957) 172–174. [23] A.M. Janiak, S. Milewski, Med. Mycol. 39 (2001) 401–408. [24] S. Milewski, D. Kuszczak, R. Je˛drzejczak, R.J. Smith, A.J.P. Brown, G. Gooday, J. Biol. Chem. 274 (1999) 4000–4008. [25] M. Kenig, E. Vandamme, E.P. Abraham, J. Gen. Microbiol. 94 (1976) 46–54. [26] E.A. Noltmann, J. Biol. Chem. 239 (1963) 1545–1550. [27] InsightII, Accelrys, San Diego, CA, 1997. Available from: . [28] W.R.P. Scott, P.H. Hunenberg, I.G. Tironi, A.E. Mark, S.R. Biller, J. Fennen, A.E. Torda, T. Huber, W.F. van Gunsteren, J. Phys. Chem. 103 (1999) 3596–3607. [29] G.M. Morris, D.S. Goodsell, R.S. Halliday, R. Huey, W.E. Hart, E.K. Belew, A.J. Olson, J. Comput. Chem. 19 (1998) 1639–1662. [30] J.D. Madura, J.M. Briggs, R.C. Wade, M.E. Davis, B.A. Luty, A. Ilin, J. Antosiewicz, M.K. Gilson, B. Bagheri, L.R. Scott, J.A. McCammon, Comput. Phys. Commun. 91 (1995) 57–95. [31] C.I. Branden, J. Tooze, Introduction to Protein Structure, Garland Publishing, London, 1991, pp. 49–51. [32] J.H. Lee, K.Z. Chang, V. Patel, C.J. Jeffery, Biochemistry 40 (2001) 7799–7805. [33] C.J. Jeffery, R. Hardre, L. Salmon, Biochemistry 40 (2001) 1560– 1566. [34] B. Golinelli-Pimpaneau, B. Badet, Eur. J. Biochem. 201 (1991) 175– 182. [35] D. Arsenieva, R. Hardre, L. Salmon, C.J. Jeffery, Proc. Natl. Acad. Sci. USA 99 (2002) 5872–5877. [36] M.-A. Badet-Denisot, C. Leriche, B. Badet, in: R.A.A. Muzzarelli (Ed.), Chitin Enzymology, European Chitin Society, Ancona, pp. 161–166. [37] C.J. Jeffery, B.J. Bahnson, W. Chien, D. Ringe, G.A. Petsko, Biochemistry 39 (2000) 955–964.