2-Acetamido-2-deoxy-d-glucono-1,5-lactone Sulfonylhydrazones: Synthesis and Evaluation as Inhibitors of Human OGA and HexB Enzymes

Inhibition of the human O-linked β-N-acetylglucosaminidase (hOGA, GH84) enzyme is pharmacologically relevant in several diseases such as neurodegenerative and cardiovascular disorders, type 2 diabetes, and cancer. Human lysosomal hexosaminidases (hHexA and hHexB, GH20) are mechanistically related enzymes; therefore, selective inhibition of these enzymes is crucial in terms of potential applications. In order to extend the structure–activity relationships of OGA inhibitors, a series of 2-acetamido-2-deoxy-d-glucono-1,5-lactone sulfonylhydrazones was prepared from d-glucosamine. The synthetic sequence involved condensation of N-acetyl-3,4,6-tri-O-acetyl-d-glucosamine with arenesulfonylhydrazines, followed by MnO2 oxidation to the corresponding glucono-1,5-lactone sulfonylhydrazones. Removal of the O-acetyl protecting groups by NH3/MeOH furnished the test compounds. Evaluation of these compounds by enzyme kinetic methods against hOGA and hHexB revealed potent nanomolar competitive inhibition of both enzymes, with no significant selectivity towards either. The most efficient inhibitor of hOGA was 2-acetamido-2-deoxy-d-glucono-1,5-lactone 1-naphthalenesulfonylhydrazone (5f, Ki = 27 nM). This compound had a Ki of 6.8 nM towards hHexB. To assess the binding mode of these inhibitors to hOGA, computational studies (Prime protein–ligand refinement and QM/MM optimizations) were performed, which suggested the binding preference of the glucono-1,5-lactone sulfonylhydrazones in an s-cis conformation for all test compounds.

In our previous work [12], 2-acetamido-2-deoxy-D-glucono-1,5-lactone 4-arylsemicarbazones (Table 1, II) were found to be nanomolar inhibitors of both hOGA and HexA/HexB enzymes. Nevertheless, these inhibitors showed only moderate selectivity (K i(Hex) /K i(OGA) = 1-2). This was the first study in which modifications were made to the urethane linking moiety between the glycone and the aromatic residue related to PUGNAc. In the present work, our aim was to make a structure-affinity relationship study of the said linker to determine how respective modifications affect the inhibition of OGA and the selectivity towards HexA/HexB. For this purpose, a library of 2-acetamido-2-deoxy-D-glucono-1,5-lactone arenesulfonylhydrazones (Table 1, target compounds) was prepared and studied as inhibitors of the above enzymes using kinetic and computational methods.

Ar
Products and yields (%) To determine the configuration of the C=N bond in compounds 5, a nuclear Overhauser effect (NOE)-based NMR approach was applied to 5d. A 2D 1 H-1 H EASY ROESY experiment [19] was performed, which is a robust variant of NOE methods for small-and medium-sized molecules. The maximum spatial distance that can provide an NOE signal is about 5 Å. Thus, the E-Z isomeric forms of compound 5d are expected to be distinguished based on informative NOE cross-peaks between the N(2)H and characteristic protons of the carbohydrate unit. In the 1 H-1 H ROESY spectrum ( Figure S1) obtained for compound 5d, NOE cross-peaks between the N(2)H resonance and H(5 ), H(6 a,b) resonances can be observed (Figure 1), which indicate the vicinity in space of these protons and so the existence of the Z isomeric form. Since both the 1 H and 13 C NMR spectra for the other compounds 5 closely resemble those of 5d, one can safely conclude the presence of the Z isomer also in those derivatives.

Inhibition of hOGA and hHexB by 2-Acetamido-2-deoxy-D-glucono-1,5-lactone Arenesulfonylhydrazones
The inhibitory potency of compounds 5a-f was evaluated against recombinant hOGA and hHexB enzymes. The enzymes were prepared and purified as described previously [12,20]. For inhibition of hOGA, fluorescent 4-MU-GlcNAc was used as the substrate. Inhibition constants (K i ) were determined by linear regression of data from Dixon plots using the competitive model as this mode of inhibition was proved by the Cornish-Bowden plots. In the case of inhibition of hHexB, Michaelis-Menten kinetics were evaluated using pNP-GlcNAc as a colorimetric substrate. Inhibition constants (K i ) were determined by nonlinear regression analyses by employing the competitive inhibition model (cf. the Experimental section and Supplementary Materials).
The inhibition constants (K i ) for most inhibitors 5 against hOGA were in the low nanomolar range (Table 3). The most potent inhibitors of hOGA were sulfonylhydrazones 5e and 5f, while 5b proved to be the weakest one. On the other hand, lower K i values against hHexB were obtained for all but one of the compounds, indicating that compounds 5 are better inhibitors of hHexB than of hOGA. Sulfonylhydrazone 5f is the strongest inhibitor of both enzymes and the most selective towards hHexB after 5b (5b is also the least potent inhibitor for both enzymes). Analogously to the parent structure PUGNAc, the newly prepared derivatives 5a-f suffer from virtually no selectivity towards any of the tested enzymes, which suggests that the affinity of the aglycone part of the inhibitors might be similar for both glycosidases.

Computationally Predicted Enzyme-Inhibitor Binding in hOGA
To predict and analyze the hOGA binding modes of inhibitors 5, Prime protein-ligand refinements [21] were performed based on their successful application to the previously reported 2-acetamido-2-deoxy-D-glucono-1,5-lactone semicarbazone series of inhibitors [12]. In this previous study, it was found that Glide native ligand redocking calculations (PDB code: 5UHO) did not reproduce the PUGNAc crystallographic conformation, specifically the orientation of the phenylurethane moiety, so that the Prime refinement and a QM/MM approach was applied [12]. In the present case, however, due to the shortened length of the linker and the greater variance in the structures of the target 2-acetamido-2-deoxy-D-glucono-1,5-lactone arenesulfonylhydrazones compared to PUGNAc in the starting model hOGA-PUGNAc complex (PDB code: 5UHO), refinements were performed in the hierarchical optimization mode (as opposed to the local optimization mode), which involved systematic sampling of inhibitor positions, orientations and conformations, along with enzyme binding site residues. The five output models for each inhibitor were then further refined in the local optimization mode (releasing hydrogen bond constraints, cf. Experimental details), but also used in QM/MM optimizations that should, in theory, better describe the energetic features of the predicted binding modes. More specifically, QM/MM has important applications in areas where standard force field-based methods may not be entirely accurate. A QM description of this novel set of compounds would address any potential shortcomings in force field parameters, particularly in the description of key dihedrals such as the rotations around the N1-N2 bond (cf. atom numbering scheme in Table 4) and the surrounding dihedrals. The binding of 5a-f to hOGA was predicted using this approach; the PUGNAc inhibitor was also included for initial validation by geometrically comparing its predicted complex with its solved crystallographic structure (PDB code: 5UHO). Table 3. Binding affinities of compounds 5 toward hOGA and hHexB enzymes compared to those of some previously known inhibitors.
Compound  The energy results from the Prime enzyme-inhibitor refinements and QM/MM calculations are shown in Table 4. In some cases, close to equivalent poses were obtained and coincided with similar absolute/relative energies. As expected, all the energies are lower following the second Prime enzyme-inhibitor refinement in the local optimization mode; the lowest energy enzyme-inhibitor complex pose was also subject to change in some cases. As mentioned, QM/MM approaches can provide a more accurate picture of the preferred binding mode [22,23]. For PUGNAc, the lowest energy QM/MM optimized poses 4 and 5 were equivalent based on energy and RMSD (heavy atoms) comparisons. Comparing these predicted enzyme-inhibitor models with the crystal structure complex following backbone superimposition, the RMSDs (heavy atoms) for ligand and flexible binding site residues were just 0.762 Å and 0.452 Å, respectively, an initial validation of the refinement protocol. For 5a-f, with the inhibitors described by QM, the key conformations through rotations around the N1-N2 bond (and associated dihedrals) can be more accurately described. Therefore, it was not overly surprising that there was a more considerable change in preferred binding mode following the QM/MM optimizations compared to the Prime enzyme-inhibitor refinements, both in terms of geometry and associated energetics. Most importantly, all inhibitors were observed to have a preference for the s-cis binding conformation around the N1-N2 bond (cf. Table 4), conformations that were originally ranked lower in the initial Prime refinements (where the preference was generally s-trans). The QM/MM optimized lowest energy complexes had inhibitor C1 =N1-N2-H dihedral angles in a similar range −168.0 • to −172.2 • for 5a-e, and slightly less planar (−154.4 • ) for 5f. It is noted that solvation effects are not accounted for in the QM/MM optimizations.
The predicted QM/MM optimized binding geometries of the phenyl-substituted inhibitors 5a-d with hOGA are similar and shown in Figure 2. There are key interactions similar to those reported previously for the 2-acetamido-2-deoxy-D-glucono-1,5-lactone semicarbazone series of compounds and PUGNAc [12]. Notably, the N-acetyl-glucosamine moiety hydrogen bond interactions are conserved as follows (cf. atom numbering in Table 4): inhibitor O3 -hydroxyl with both Gly67 backbone O and Lys98 sidechain; O4 -hydroxyl O with Asn313 sidechain amide and Asp285 sidechain carboxylate; O6 hydroxyl with Asp285 sidechain carboxylate; and the N-acetyl group has hydrogen bonds involving N2 H with the Asp174 sidechain carboxylate and acetyl carbonyl O with the Asn280 sidechain amide. In terms of the sulfonylhydrazone linker between the N-acetyl-glucosamine and phenyl groups, N2H hydrogen bonds with the Tyr219 sidechain hydroxyl O atom and there is also potential for hydrogen bonding of this hydroxyl with ligand N1 atoms. The differences in 5a-d are their phenyl para-substituents. All ligand phenyls have favorable interactions with the Val254 sidechain. There are interactions between the para-substituents (-CH 3 , -CF 3 , -F, and -Cl) and hydrophobic sidechains of Tyr286 and Val255. The CF 3 group (5b) is shifted away from Tyr286, an interaction which may be related to its lower hOGA inhibition potency (K i = 230 nM); alternatively, the methyl-substituted 5a (K i = 78 nM) can form good CH-π contacts with the Tyr286 ring. There are also NH-halogen (enzyme-inhibitor, respectively) interactions for 5b-d involving the Val255 backbone NH. Notably, in the case of the most potent phenyl-substituted compound 5d (Cl substituent, K i = 70 nM), the halogen-hydrogen bond donor (HBD) interactions are consistent with a survey of crystal structure data; the predicted C-Cl-HBD angle (85.4 • ) and the HBD-Cl (4.4 Å) distance are close to the observed most prominent values (HBD defined by the NH nitrogen atom) [24].
The most potent hOGA inhibitors were the naphthyl analogs, 5e (2-naphthyl) and 5f (1-naphthyl), with these ligands close to equipotent (K i -s~30 nM). The predicted binding of these inhibitors from the QM/MM optimizations are shown in Figure 3A,B, respectively. Both ligands are able to exploit favorable interactions with the Val254 sidechain and have the potential for NH-π contacts from the backbone NH of the flexible loop residue Val255, all of which could be the source of their observed superior potencies. The 2-naphthyl substituent of 5e is also aligned with the Tyr286 ring in a T-shaped configuration for π-π stacking interactions with a ring centroid-centroid distance of 5.1 Å. This can be considered consistent with geometric preferences of protein-ligand T-shaped π-π interactions from Protein Data Bank analysis, with T-shaped binding conformations also found to be the most predominant [25]. In the case of 5f, a 180 • flip of the 1-naphthalene ring around the S-C(Ar) bond would lead to alternate π-π interactions with Phe223.

Protein Expression and Purification
The detailed protocol for the expression of hOGA and the subsequent purification procedure of the recombinant protein have been described in our previous work [12]. The gene construct encoding the full-length human OGA fused to N-terminal His 6 -tag was kindly provided by Prof. D. Vocadlo (SFU, Burnaby, BC, Canada). Human β-Nacetylhexosaminidase B (hHexB) was expressed in Pichia pastoris, and the extracellularly produced enzyme was isolated according to the procedure described by Krejzová et al. [20].

Inhibition Studies
The inhibitory potency of compounds 5a-f was evaluated against recombinant enzymes hOGA and hHexB. In the case of hOGA inhibition, 4-MU-GlcNAc was used as a substrate at three concentrations (7 µM, 12 µM, and 25 µM) in the reaction mixtures containing 50 mM potassium phosphate buffer (pH 7.5), an inhibitor 5a-f in a final concentration of 0 to 750 nM, and 3 nM hOGA. In all cases, the enzyme and the inhibitors were preincubated at 37 • C, then the reaction was initiated by the addition of the substrate. All reactions were carried out in a fluorescence cuvette (Hellma Analytics, Müllheim, Germany), held in the thermostat cell holder of a Jasco FP-8200 fluorescence spectrophotometer (Easton, MD, USA) equipped with a Xe lamp light source. The excitation of the reactions was performed at 360 nM and emission was detected at 450 nM every 10 s for 10 min. All measurements were performed in triplicates. Initial velocities were determined and the reciprocal of initial velocities was plotted against inhibitor concentrations. The inhibitory mode of the compounds on hOGA was assessed by the Cornish-Bowden method, which involves the plotting of [S]/v 0 vs. inhibitor concentrations [27]. The linear fitting to the data points resulted in parallel lines referring to competitive inhibition. Inhibition constants (K i ) were determined according to the competitive model of the Dixon method by plotting the reciprocal of initial velocities vs. inhibitor concentrations [28]. The intersection point of the linearly fitted lines gave K i in nM.
In the case of hHexB inhibition, Michaelis-Menten kinetics were evaluated using pNP-GlcNAc as a substrate. The kinetic reactions took place in Eppendorf tubes in a final volume of 400 µL containing 24 nM HexB, pNP-GlcNAc substrate in the range of 0.1-3 mM, and various inhibitor concentrations ranging from 0 to 300 nM in 50 mM citrate-phosphate buffer, pH 5.0. The reactions were incubated at 35 • C and then 50 µL samples were taken every 60 s. Samples were mixed with a quenching solution of 150 µL 0.1 M Na 2 CO 3 and the absorbance of the solutions was measured at 420 nm by a Tecan Sunrise plate reader (Männedorf, Switzerland). All measurements were performed in triplicates. Initial velocities were determined from the data points by linear fit, and then nonlinear regression analyses were performed using the competitive inhibition model to calculate K i values. GraphPad Prism software was used for the calculations.

Protein Preparation
hOGA was prepared for Prime protein-ligand refinement calculations using its cocrystallized complex with PUGNAc (PDB code 5UHO; 3.21 Å resolution) and Schrödinger's Protein Preparation Wizard [21]. Water molecules within 5 Å of the native ligand were initially retained (deleted for subsequent calculations), bond orders were assigned and hydrogen atoms added, with the protonation states for basic/acidic residues based on PROPKA calculated pK a 's at pH 7 [29]. Subsequent optimization of hydroxyl groups, histidine sidechain C/N atom flips and protonation states, and any sidechain O/N atom flips of Gln and Asn residues was based on optimizing hydrogen-bonding networks. The system was finally minimized using OPLS-AA (2005) force field [30] under the constraint of RMSD (heavy atoms) to be maintained within 0.3 Å of the crystallographic atomic positions.

Enzyme-Inhibitor Complex Predictions
Initial models of enzyme-inhibitor complexes for calculations were prepared by mutation of PUGNAc in chain A of the prepared crystallographic complex into the inhibitors 5a-f. Prime v5.4 enzyme-inhibitor refinements [21] were then performed in the hierarchical optimization mode for each new complex. The default OPLSe force field [31] and the VSGB model of solvation [32] were employed. Residues within 5 Å of PUGNAc in 5UHO were free in all refinement calculations (same 360 atoms), but with hydrogen bond constraints (force constant 100 kcal mol −1 Å −2 ) to the N-acetyl-glucosamine moiety to reflect the solved hOGA-PUGNAc crystallographic data; specifically (cf. Table 4), Asp285 carboxylate-inhibitor O4 and O6 hydrogens, Gly67 backbone O-ligand O3 hydrogen, Asp174 and Asp175 sidechain carboxylates-ligand N2 hydrogen, and Asn280 sidechain -NH 2 with inhibitor acetyl carbonyl O atom. The rest of the protein atoms (beyond 5 Å) were constrained. The number of structures (enzyme-inhibitor models) to return in each case was set to 5. Initial validation of the protocol involved the application of this protocol to the cognate PUGNAc ligand for the reproduction of its solved crystallographic binding mode.
The five output complexes for each inhibitor were then further refined using two methods for comparison: additional Prime v5.4 refinements and QM/MM optimizations. For the Prime refinements, in this case, the local optimization mode was used, releasing the hydrogen bond constraints used in the initial hierarchical optimization mode calculations. The same protein residues (360 atoms) were free, with the rest of the protein constrained, as before. For the QM/MM optimizations, again the same protein residues were free and constrained as per the Prime calculations; no hydrogen bond constraints were used. In this case, the respective inhibitor was described using QM (QM region) at the M06-2X/6-31+G** level of theory [33][34][35], and the protein (MM region) was modeled using the OPLS-AA(2005) force field [30]. No cut-offs were used for non-bonded interactions. All QM/MM calculations were performed using QSite (Jaguar v10.2; Impact v81012) [21].

Conclusions
A library of 2-acetamido-2-deoxy-D-glucono-1,5-lactone arenesulfonylhydrazones was prepared from D-glucosamine in a five-step synthetic sequence in 26-44% overall yields. These compounds were assayed for their inhibitory activity against human OGA and HexB enzymes by using fluorimetric and spectrophotometric detection, respectively. Both enzymes were inhibited by the sulfonylhydrazones in the nanomolar range, and hence the compounds were considerably potent. However, the selectivity (K i(hHexB) /K i(hOGA) ) varied from 1.3 to 0.21 to show no significant bias to any of the enzymes. The best inhibitor of hOGA was 2-acetamido-2-deoxy-D-glucono-1,5-lactone 1-naphthalenesulfonylhydrazone with a K i of 27 nM and this compound had a K i of 6.8 nM towards hHexB. The possible binding modes of the inhibitors to the hOGA enzyme were analyzed by computational methods (Prime protein-ligand refinement and QM/MM optimizations) to reveal the predicted interactions responsible for the strong binding of the compounds in a preferred s-cis conformation along the (C=)N-N(-H;-SO 2 ) rotatable bond. This study has extended the relatively few structure-activity relationships of OGA inhibitors. However, the nonselective nature of the inhibition towards hHexB necessitates further structural modifications of the glyconolactone sulfonylhydrazone-type inhibitors.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/ijms23031037/s1. Author Contributions: M.K. synthesized the compounds, recorded routine NMR spectra, carried out the enzyme kinetic measurements, and wrote the paper; I.T. performed the NOE measurements; T.B. designed and supervised the enzymatic measurements; Z.M. and K.S. prepared the recombinant enzymes and participated in the enzyme assays; P.B. supervised and evaluated enzyme kinetics and edited the paper; V.K. supervised the enzyme experiments, raised part of the funding, and edited the paper; J.M.H. performed the computational study and wrote the paper; L.S. conceptualized the research, raised funding, and wrote the paper. All authors have read and agreed to the published version of the manuscript. Data Availability Statement: Not applicable.