By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Glutaminyl cyclases (QCs) catalyze the formation of pyroglutamate-modified amyloid peptides deposited in neurodegenerative disorders such as Alzheimer’s disease. Inhibitors of QC are currently in development as potential therapeutics. The crystal structures of the potent inhibitor PBD150 bound to human and murine QC (hQC, mQC) have been described recently. The binding modes of a dimethoxyphenyl moiety of the inhibitor are significantly different between the structures, which contrasts with a similar Ki value. We show the conformation of PBD150 prone to disturbance by protein–protein interactions within the crystals. Semi-empirical calculations of the enzyme–inhibitor interaction within the crystal suggest significant differences in the dissociation constants between the binding modes. To probe for interactions in solution, a site-directed mutagenesis on hQC was performed. The replacement of F325 and I303 by alanine or asparagine resulted in a 800-fold lower activity of the inhibitor, whereas the exchange of S323 by alanine or valine led to a 20-fold higher activity of PBD150. The results provide an example of deciphering the interaction mode between a target enzyme and lead substance in solution, if co-crystallization does not mirror such interactions properly. Thus, the study might provide implications for rapid screening of binding modes also for other drug targets.
Many neuropeptides and secretory proteins from plants and animals contain a pyroglutamate (pGlu) residue at their N-terminus (1,2). Frequently, the residue mediates receptor interaction, as shown, for example, for the hormones thyrotropin-releasing hormone or gonadotropin-releasing hormone (3,4), and protects against degradation by aminopeptidases. The N-terminal pGlu formation represents a finishing step in the process of peptide and protein post-translational maturation. The reaction is catalyzed by glutaminyl cyclases (QCs), which cyclize N-terminal glutaminyl residues into pGlu following limited proteolysis of the substrate precursors (5–9). Besides pGlu-modified hormones and proteins, QCs have been recently identified as catalysts of pGlu formation at the N-terminus of amyloid peptides such as Aβ, ADan, or ABri (10,11). Based on numerous investigations, it appears that the N-terminal modification increases the aggregation propensity of these amyloid peptides and also enhances resistance to degradation by proteases (12–17). Both characteristics are likely responsible for an increased toxicity of pGlu-Aβ as observed in animal models. On the basis of these observations, QC inhibition is currently in development as a therapeutic strategy in neurodegenerative disorders (18).
Recent co-crystallization and structural analysis of inhibitor PBD150, which has been applied successfully in pharmacological studies (18,19), revealed differences in binding modes of a phenylmethoxy part of the inhibitor. The residue has been shown to significantly increase the potency of the compounds in structure–activity relationship (SAR) assessments (20,21). To characterize the binding in solution and to draw conclusions about the validity of the binding mode in QC crystal structure, we assessed the interaction using a combined approach of in silico analyses and site-directed mutagenesis to gain information of drug–target interactions in solution. The study was intended to analyze the different conformations of the inhibitor in crystal structures from mice and men and to provide clues for similar investigations on other potential drug targets.
Materials and Methods
The Escherichia coli strain DH5α was applied for all cloning procedures. Pichia pastoris strain X33 (AOX1, AOX2) was used for the expression of the different hQC variants. Yeast was grown, transformed, and analyzed according to the manufacturer’s instructions (Invitrogen, Karlsruhe, Germany). The glutaminyl peptides were obtained from Bachem (Bubendorf, Switzerland) or synthesized as described elsewhere (22). Recombinant pyroglutamyl aminopeptidase (pGAP) from Bacillus amyloliqefaciens was purchased from Qiagen (Hilden, Germany) and glutamic acid dehydrogenase (GDH) was from Fluka (Steinheim, Germany). Imidazole, benzimidazole, and cysteamine were purchased from Sigma-Aldrich (Steinheim, Germany). The low-salt LB medium required for propagation of E. coli and the buffered glycerol complex medium (BMGY) or buffered methanol complex medium (BMMY), which are required for propagation of yeast, were prepared according to the Pichia manual and ‘Fermentation process guidelines’ (Invitrogen).
In silico analysis of the binding modes of PBD150 in hQC and mQC
For all calculations, a 22-CPU computer cluster running centos 5.4 (Tikanga, http://www.centOS.org, the CentOS project) as the operation system was used. All preparation and visualization steps were performed with the software package MOE (vers. 2010.10; Chemical Computing Group, Montreal, QC, Canada). For the QMScore and PWD calculations, the DivCon Discovery Suite, which could be integrated in MOE as MOE/DivCon (build 438; QuantumBio Inc., State College, PA, USA), was used. The pdb files (3pbb and 3si2) were obtained from the protein data base (http://www.rcsb.org). The protonate 3D function was used to assign all hydrogen atoms to the corresponding system, in which standard options were kept. An energy minimization step using the mmff94x force field was performed applying a gradient of 0.01. The heavy atoms of the systems were tethered by a force constant of 100, and the stereochemistry was not a subject of alterations by the system. PBD150 was marked, and all water atoms with a distance >7 Å to the ligand were removed in each system. Two databases were created, one with the proteins, including the remaining water molecules and zinc ions, the second database contained the ligands. The QMScore calculation window was used to set up the calculation for the quantumbio software. Thereby, the ‘Many-to-Many’ option in the ‘Mapping Mode’ was used, and the formerly created db files have been chosen in the corresponding fields. Before the ‘Molecule Tests’ were performed, the system was switched to the OPLS-AA force field, and the ‘Create’ function was used to generate the corresponding working database. The ‘Hamiltonian’ was set to ‘PM3’, the ‘Calculation’ was set to ‘Linear Scaling’, and the ‘PWD’ option was switched on. The ‘Shift’ value was set to 9. The calculation was started by using a script, which was calling the moebatch application. Thereby, a sun grid engine was used as the queuing system for parallel execution. For the visualization of the attraction forces between the inhibitor and the amino acids in the active center, the PWD module of the MOE/DivCon-Suite was applied.
Method evaluation: Comparable calculations were made using the crystal structure of human QC, where five different inhibitors were modeled into the QC. These inhibitors share an activity range of four orders of magnitude. They represent key compounds in the quantitative activity relationship of the compound class from PBD150 [compounds 26, 33, 44 and 55 (PBD150) from (20) and compound 9 from (21)]. Thereby an r2 between the pKi value and the QMScore of 0.9 was measured. The same calculations were performed for mQC – co-crystallized with four different inhibitors, sharing an activity range of 1 order of magnitude, but from different chemical classes. Hereby, again an r2 of 0.9 for the correlation of the pKi values and the QMScore was observed. More details of these calculations are given in the supplementary information (Tables S2 and S3).
The hQC cDNA was inserted into the yeast expression vector pPICZαB (Invitrogen) via the PstI and NotI restriction sites. Additionally, an N-terminal His6-tag using the primers Cs (sense) and Cas (antisense) was introduced (Table S5). The mutations in human QC were introduced by PCR-mediated site-directed mutagenesis according to standard PCR techniques using the appropriate primer pairs listed in Table S5. Following whole-plasmid amplification, the parent DNA was digested using DpnI (quik-change II site-directed mutagenesis kit; Stratagene, Santa Clara, CA, USA). The cDNA sequence was verified by sequencing applying the primer Ss, Sas was used in case of the K144-mutation (Table S5).
Transformation of Pichia pastoris and mini-scale expression
Plasmid DNA was amplified in the E. coli DH5α and purified (Plasmid Miniprep Kit; Qiagen). Twenty to thirty micrograms DNA was linearized using PmeI, precipitated, and dissolved in deionized water. One to five micrograms DNA was applied for transformation of competent yeast cells [P. pastoris strain X33 (AOX1, AOX2)] by electroporation according to the manufacturer’s instructions (Bio-Rad, Munich, Germany). Yeast was grown, transformed, and analyzed according to the manufacturer’s instructions (Invitrogen). A selection of transgenics was achieved on YPDS plates containing 150 μg/mL Zeocin. To test yeast clones upon expression, recombinants were grown for 24 h in 10-mL conical tubes containing 2 mL of BMGY. Afterward, cell suspension was centrifuged and cells were resuspended in 2 mL of BMMY containing 0.5% (v/v) methanol. This concentration was maintained by addition of methanol every 24 h. QC activity was determined in the supernatant after 72 h. To finally confirm the insertion of the correct hQC variant, genomic DNA was prepared according to standard molecular biological techniques. Target DNA was amplified by PCR using primer pairs Cs/Sas (K144) and Ss/Cas and sequenced. Clones displaying highest QC activity were chosen for large expression.
Large-scale expression and purification
For large-scale expression, high-cell-density fermentation and expression in shake flasks were applied. Fermentation was carried out in a 5-L reactor (Biostad B; Braun Biotech, Melsungen, Germany), essentially as described elsewhere (23). Briefly, fermentation was initiated in basal salt medium supplemented with trace salts at pH 5.5. Biomass was accumulated in a batch and a fed-batch phase with glycerol as the sole carbon source for about 28 h. Expression of target protein was induced by methanol feeding according to a three-step profile recommended by Invitrogen (‘Pichia fermentation process guidelines’). The large-scale expression in shake flasks was performed in a final volume of 4 L, essentially as described for the mini-scale expression for activity screening. In addition, the OD 600 was adjusted to 1 before the change from BMGY to BMMY. Both expression modes were stopped after 68 h. Cells were separated from the medium by centrifugation at 6000 g and 4 °C for 20 min. The pH was adjusted to 6.8 by the addition of saturated Tris buffer, and the resulting turbid solution was centrifuged at 16 000 × g for 30 min at 4 °C.
Starting purification, histidine was added to a final concentration of 1 mm. The supernatant was applied with a flow of 12 mL/min in upward flow direction onto an expanded bed adsorption column [Streamline column 25 (2.5 × 22 cm – settled) with Streamline Chelating Sepharose; GE Healthcare, Uppsala, Sweden]. Before, the column was saturated with Ni2+ ions and equilibrated with 50 mm sodium phosphate buffer, pH 6.8, and 300 mm sodium chloride. Bound enzyme was washed with 1.5 L of 50 mm sodium phosphate buffer, pH 6.8, and 300 mm sodium chloride. Enzyme was eluted in downward flow direction at a flow rate of 8 mL/min using 50 mm phosphate buffer, pH 6.8, and 300 mm sodium chloride containing 100 mm histidine. QC-containing fractions were pooled, and ammonium sulfate was added to a final concentration of 800 mm. After centrifugation at 100 000 × g (4 °C) for 1 h, the resulting solution was applied onto a Butyl Sepharose Fast Flow column (1.6 × 13 cm; GE Healthcare) at a flow rate of 2 mL/min. Bound enzyme was washed for three column volumes applying 50 mm sodium phosphate buffer, pH 6.8, and 800 mm ammonium sulfate and eluted in reversed flow direction with 5 mm sodium phosphate buffer, pH 6.8. Fractions containing QC activity were pooled and dialyzed overnight at 4 °C against a 100-fold excess (v/v) of 30 mm Bis-Tris, pH 6.8. The solution was centrifuged at 100 000 × g (4 °C) for 1 h and applied onto an Uno Q6 column (12 × 53 mm; Bio-Rad) at a flow rate of 4 mL/min. Bound enzyme was washed for three column volumes applying 30 mm Bis-Tris, pH 6.8. Enzyme was eluted with 30 mm Bis-Tris, pH 6.8, containing 360 mm sodium chloride. QC-containing fractions were pooled, and purity was analyzed by SDS–PAGE (Servagel TG 4-20; Serva, Heidelberg, Germany) and Coomassie Blue staining (Figure S1, supplementary information). The purified enzyme was stored at −20 °C after addition of glycerol [50% (v/v)] or without glycerol at −80 °C.
CD spectroscopic evaluation
hQC WT and mutants were desalted by size exclusion chromatography using a Sephadex G-25 fast desalting column (1.0 × 10 cm; GE Healthcare), which was equilibrated in 10 mm potassium phosphate buffer, pH 6.8. The protein concentration was adjusted to 39 μm (WT), 31 μm (F325A), 36 μm (I303N), 39 μm (W329Y), 39 μm (K144M), 30 μm (Y299A), 28 μm (hQCloop), and 36 μm (S323T), respectively. CD spectra were acquired with a Jasco J-715 polarimeter (Jasco, Groß-Umstadt, Germany) using quartz cuvettes of 0.1 cm pathlength and an external thermostat (Julabo F25; Julabo, Seelbach, Germany). For wavelength scans, the mean of 10 scans between 190 and 260 nm (far-UV/amide region) was calculated and corrected by subtraction of the buffer spectrum (Figure 4).
QC activity was assayed essentially as described (24). The assay reactions consisted of varying concentrations of H-Gln-AMC (7-amido-4-methylcoumarin), H-Gln-βNA (β-naphthylamide), or another glutaminyl peptide in 50 mm Tris–HCl, pH 8.0. For a spectrophotometric assessment, samples additionally contained 30 U/mL GDH, 0.25 mm NADH/H+, and 15 mmα-ketoglutaric acid. Reactions were started by addition of QC, and activity was monitored by recording the decrease in absorbance at 340 nm. For a fluorometric detection of QC activity, reactions contained the substrate (H-Gln-βNA or H-Gln-AMC) and 0.4 U/mL pyroglutamyl aminopeptidase as auxiliary enzyme. The excitation/emission wavelengths were 380/460 nm (H-Gln-AMC) and 320/410 nm (H-Gln-βNA). Reactions were started by addition of QC. QC activity was determined from a standard curve of the fluorophore under assay conditions. All determinations were carried out at 30 °C using the BMG Fluostar reader for microplates (BMG Labtechnologies, Offenburg, Germany). For inhibitor testing, the sample composition was the same as described above, except for the added inhibitory compound. Inhibitory constants were determined using H-Gln-AMC in a concentration range between 0.25 and 4 Km. Evaluation of all kinetic data was performed using GraFit software (version 5.0.4. for windows; Erithacus Software Ltd., Horley, UK) via a non-linear regression by using default settings.
In silico analysis of the binding mode of PBD150 in human and murine QC
The crystal structures of human and murine QC with bound PBD150 have been published recently [hQC: 3pbb (25), mQC: 3si2 (26)]. The primary interaction of the inhibitor is mediated by the imidazole warhead of the inhibitor binding to the fourth coordination site of the active site zinc ion. The imidazole moiety adopts virtually identical orientations in murine and human QC. Differences are observed for the binding of the thiourea and the 3,4-dimethoxyphenyl part of PBD150. Interestingly, the differing binding modes of PBD150 to both QCs do not reflect all aspects of previous SAR analyses (20,21), raising the possibility that crystal packaging may influence the conformation. This assumption is further supported by very similar inhibition constants of PBD150 for both enzymes in solution (Ki[hQC] = 100 ± 10 nm; Ki[mQC] = 114 ± 11 nm) and a very high conservation of the corresponding amino acids in both active sites.
To characterize the enzyme–inhibitor interactions in detail, semi-empirical calculations of the binding energies of PBD150 in the active sites were performed. The calculations clearly support a major contribution of the imidazole part to the overall binding energy. A pair-wise decomposition analysis of the amino acids and water molecules involved in PBD150 binding are shown in Figure 1 and Table S1. The results of the calculations are two different terms. Thereby, the attraction force between the ligand and the enzyme is described by the EAB value. The QMScore itself is a general description of the whole binding energy, including not only the attraction but also the repulsion forces. The QMScore is therefore usually used to judge the binding modes. Summing up the values for EAB, a nearly identical attraction force is obtained in both complexes. In contrast to the QMScore, evaluation of the binding modes clearly suggests a stronger binding of PBD150 in the murine enzyme, resulting in 4–10 times higher affinity (Tables 1, S2 and S3). Thus, the results of the calculation are in strong contrast to the experimental values, where PBD150 was slightly more potent to inhibit the human enzyme.
Table 1. QMScores (calculated on the basis of the crystal structure) and dissociation constants of PBD150, determined using human and murine QC
The difference in the QMScores of 24 should result in 4–10 times higher affinity of PBD150 to murine QC. Detailed results of the calculation are given in the supplementary information (Tables S1–S3).
100 ± 10
114 ± 11
An explanation for this apparent discrepancy between the calculated and the experimental values could be given, if the whole elemental cells of the crystal structures are analyzed (Figures 2 and 3). In murine and human QC crystals, a loop of a neighbor protein chain is interacting directly with the inhibitor and the active site of the enzyme (Figures 2A and 3A). In human QC, an additional H-bond of the N2 of PBD150 is formed with R118 of the neighbor protein chain (Figure 2B). Also, an additional hydrophobic interaction between the phenyl ring of the inhibitor and Y115 of the neighboring protein molecule is likely. In murine QC, the orientation of the phenyl ring of PBD150 is influenced by an additional edge-to-pi interaction with S152 of the neighbor protein chain (Figure 3B).
The calculations and the assessment of the protein–PBD150 interactions suggest that the binding pose of the inhibitor, especially the dimethoxyphenyl part, is highly influenced by the packing in the corresponding crystal structure. Therefore, we aimed at probing the inhibitor–protein interactions in solution by site-directed mutagenesis and assessment of the inhibitory potency of PBD150.
Heterologous expression and characterization of human QC mutants
Based on the different enzyme–ligand interaction appearing in the murine and human crystal structures 3si2 and 3pbb, we chose the following amino acids as mutation sites to probe PBD150 in solution (compared with Figures 1 and 2B, amino acid numbering of human QC): (i) W329; (ii) F325; (iii) I303, building up a hydrophobic area at the entrance of the active site, which is likely to interact with the phenyl ring of PBD150; (iv) S323, located close to the mentioned hydrophobic cleft but obviously not directly involved in the interaction; (v) Y299, either fixing two loop regions in mQC (Y300-T266) or being involved in a hydrogen bond network to the methoxy oxygens of PDB150; (vi) we considered a complete exchange of the variable loop region (N296-Q304)hQC containing Y299 (variant: hQCloop). The replacing sequence was obtained from the isoenzyme of human QC, isoQC (UniProt entry Q9NXS2, P317–E325). While this loop is fixed in the mQC, it might gain flexibility in the human QC and adopt a different orientation in solution. (vii) Finally, we exchanged K144 to probe the specificity of the approach, because it should not be involved in the binding of PBD150.
The hQC variants were isolated by heterologous expression in the methylotrophic yeast P. pastoris (strain X33). Previously, P. pastoris has been successfully used for isolation of animal QCs, for example, human QC (23), murine QC (27), or the secreted isoform of the QC from D. melanogaster (28). Because all proteins showed a very similar behavior in protein chromatography, a gross influence of the mutations on the protein structure was not expected. In average, 11 mg per 2 L of fermentation broth or 4 L of shake flask broth was obtained. The purity of the protein was assessed using SDS–PAGE (sodium dodecyl sulfate–polyacrylamide gel electrophoresis); an exemplary analysis is depicted in Figure S1.
To study whether the protein secondary structure formation was influenced by the amino acid exchanges, far-UV CD spectroscopy was applied. The spectra of several mutants are shown exemplarily in Figure 4. All enzyme variants exhibit virtually identical spectra, supporting that the mutagenesis did not affect the fold of the proteins.
To investigate the influence of the amino acid exchanges on catalysis of the protein, the steady-state kinetic parameters, Michaelis–Menten constant Km, and turnover number kcat for conversion of H-Gln-βNA, H-Gln-Gly-OH, H-Gln-Phe-Ala-NH2, and H-Gln-Lys-Arg-Leu-NH2 was determined. The results of the analysis depicted as ratio of the kinetic parameter of hQC mutant and wild type (WT) are shown in Figures 5 and 6. The ratios of mutant over wild type kcat values are close to 1 for most hQC variants, indicating that the catalytic step is not affected by introduction of the mutations. The only exception is the exchange of W329. Any replacement of the indole ring leads to a dramatic drop of kcat. A role of W329 in stabilizing the enzyme–substrate complex has been suggested by in silico calculations previously (29).
In contrast to kcat, significant differences in the ratio of the Km values appear (Figure 6). The most prominent increase was observed by amino acid exchanges of I303, F325, and W329, which build up a hydrophobic binding site. An increase in the Km value is also observed for the peptide surrogate and the small peptides with K144 variants. However, the effect disappears for the longer peptide substrates, maybe due to involvement of compensatory binding site(s). Interestingly, replacing S323 by threonine or exchanging the variable loop region (N296-Q304)hQC (hQCloop) even provokes a slightly improved conversion of the substrates. An exchange of Y299A results in Km and kcat values indistinguishable from the WT enzyme.
Summarizing, the newly generated enzyme variants showed activity and were thus well suited for the further study of inhibitor binding. The data suggest a significant contribution of K144, S323, I303, and F325 to substrate binding but not to the catalytic step. Thus, these residues are also potential candidates for a secondary enzyme–inhibitor interaction.
Binding of PBD150 to human QC and its variants
Exchange of K144, Y299, and hQCloop
To assess the relevance of secondary interactions for inhibitor binding, the inhibition constant Ki of PBD150, imidazole, benzimidazole, and cysteamine was determined. The binding features of the scaffold structures reflect the primary interaction with the active site zinc ion. To provide the results clearly, again the ratios of the Ki values of hQC mutants and WT were calculated (Table 2, for Ki values refer Table S4). An exchange of K144, Y299, and the variable loop (N296 – Q304)hQC (hQCloop) did not result in alterations of inhibitory binding of the scaffold structures imidazole, benzimidazole, and cysteamine (Tables 2 and S4). Similarly, the Ki of PBD150 was not or only little affected. The results thus suggest that a binding interaction of Y299 or the loop region - which was suspected from the crystal structure - does not occur in solution. Interestingly, the loop exchange also implies that the region is not responsible for the different Ki values of PBD150 for hQC and its isoenzyme [Ki[hQC] = 100 ± 10 nm; (26); Ki[isohQC] = 236 nm (30)].
Table 2. Relative changes in Ki compared with the wild type human QC, caused by site-directed mutagenesis
Ratios ± SD have been calculated from three independent evaluations by dividing the Ki value of the hQC variant by that determined with the WT enzyme. The Ki values are provided in the supplementary material (Table S4). Reactions were carried out in 50 mM Tris, pH 8.0, at 30 °C.
1.30 ± 0.10
1.10 ± 0.10
1.70 ± 0.10
1.21 ± 0.01
0.68 ± 0.03
1.10 ± 0.10
2.60 ± 0.10
3.35 ± 0.02
1.41 ± 0.01
1.70 ± 0.10
1.20 ± 0.10
2.37 ± 0.01
1.01 ± 0.04
1.00 ± 0.04
0.75 ± 0.01
1.22 ± 0.01
0.98 ± 0.04
1.15 ± 0.02
1.05 ± 0.01
2.20 ± 0.20
1.00 ± 0.02
1.40 ± 0.10
0.80 ± 0.10
1.21 ± 0.03
0.68 ± 0.01
0.90 ± 0.10
0.81 ± 0.03
0.20 ± 0.01
0.90 ± 0.10
1.09 ± 0.02
1.40 ± 0.10
0.21 ± 0.01
0.84 ± 0.04
1.00 ± 0.02
1.11 ± 0.04
0.074 ± 0.001
4.20 ± 0.10
11.4 ± 0.1
0.90 ± 0.10
190 ± 2
3.80 ± 0.30
6.70 ± 0.03
0.80 ± 0.10
815 ± 5
2.60 ± 0.10
5.60 ± 0.30
0.76 ± 0.02
381 ± 2
2.30 ± 0.05
1.40 ± 0.10
0.48 ± 0.02
66.3 ± 2.2
1.70 ± 0.10
2.22 ± 0.01
0.92 ± 0.02
6.20 ± 0.10
2.42 ± 0.01
1.80 ± 0.05
0.77 ± 0.05
112 ± 6
0.96 ± 0.04
1.23 ± 0.03
0.85 ± 0.01
1.90 ± 0.10
0.16 ± 0.01
0.26 ± 0.01
1.40 ± 0.05
0.56 ± 0.01
1.00 ± 0.10
4.10 ± 0.10
1.10 ± 0.10
1.00 ± 0.10
Exchange of I303, F325, and W329
The amino acids I303, F325, and W329 define a hydrophobic binding site, which could be described as flat tray at the entrance of the active site. According to the calculation, I303, F325, and W329 should significantly contribute to the binding of PBD150 (Figure 1A, Table S1). A similar influence of I303 and W329 substitution on Ki would suggest an enzyme–inhibitor interaction similar to crystal structure 3pbb (Figure 1A). A more prominent influence of I303 (and S323) would argue for an enzyme–ligand complex as described by murine QC crystals (Figure 1B).
First differences in inhibitory binding become obvious with the scaffold structures. While cysteamine binding stays virtually unaffected by all amino acid exchanges, binding of benzimidazole and imidazole does moderately change by mutation of F325 (Tables 2 and S4). Although the differences in binding are below one order of magnitude for most enzyme variants, an increase in the binding constant appears more eminent for benzimidazole than for imidazole, indicating that an interaction of the hydrophobic binding site might occur with the condensed ring structure. This conclusion is supported by only little changes in Ki with F325Y. The most prominent influence is observed in variant F325A (imidazole: 4.2 Ki[WT]; benzimidazole: 11.4 Ki[WT]). Interestingly, removal of a hydrogen donor close to the primary enzyme–inhibitor interaction site (W329F) reduces the Ki values of the scaffold structures imidazole and benzimidazole.
The binding constants of PBD150 were increased with all enzyme variants at F325 and I303, except I303V. Reducing the size of the residue (I303A) or introduction of a polar amino acid (I303N) resulted in a substantial increase in the Ki value by up to two orders of magnitude. In this regard, a major commitment to binding of PBD150 can be attributed to F325. Any exchange of the amino acid yielded a more than 100-fold increase in Ki. The lowest potency was obtained with F325N, which showed an 815-fold increase in Ki.
In contrast to the already studied amino acids of the hydrophobic binding area I303 and F325, replacement of W329 by Y or F does not affect the Ki of PBD150.
Exchange of hQC S323
With regard to the binding of PBD150, the role of a serine residue in close proximity of the hydrophobic site was not clearly predictable from in silico modeling. Whereas the EAB calculations show a small but nearly identical contribution for both enzymes (Table S1), an exchange of S323 (S323A/T/V) did not alter the binding of the scaffold structures. The inhibitors cysteamine, benzimidazole, and imidazole have not been expected to interact with the targeted region of the enzyme.
Interestingly, a significant decrease in the Ki values of PBD150 was observed. The effect is likely caused by an enlargement of the hydrophobic interaction area. Such a conclusion is supported by the most prominent shift to 0.07 Ki[WT] for the mutant S323V. Deletion of the hydroxy group (S323A) or introduction of an additional methyl group (S323T) shows a similar decrease in the binding constant of PBD150 to 0.2 Ki[WT]. The results emphasize the importance of hydrophobic interactions for the secondary binding site of PBD150.
Discussion and Conclusions
Compelling evidence suggests a role of QC activity in pathological conditions such as Alzheimer’s disease, familial British and Danish dementia (18). QCs generate the N-terminal pGlu modification of amyloid peptides, which causes an increase in stability and amyloidogenicity. N-terminally modified Aβ species are major constituents of deposits, and their occurrence appears to correlate with disease progression of AD. Therefore, our current aim of research is to develop inhibitors of QC as novel therapeutics. The first competitive inhibitors imidazole, benzimidazole, and cysteamine have been identified several years ago (22). On the basis of these scaffold structures, an improvement in the inhibitors was achieved due to modifications by medicinal chemistry methods, and first SARs could be deduced (20,21).
The present study aimed at an investigation of the binding characteristics of one of the potent inhibitors, PBD150, to human QC in solution. The rationale for the investigations was the unexpected finding that the binding modes of the inhibitor to hQC and mQC differed with regard to positioning of a 3,4-dimethoxyphenyl thiourea moiety of the inhibitor, although similar Ki values have been obtained. A closer assessment revealed that the packaging of the protein molecules in the crystals obviously affect the orientation of the inhibitor because of binding to a surface-exposed region. The disturbances of the binding mode – and potentially of surface-exposed loops at the active site – hamper a clear conclusion about the conformation of the inhibitor and the most relevant sites of interaction in the hydrophobic binding area. To obtain a validation of the binding interaction, we aimed at establishing a second algorithm to probe the important interaction site of PBD150, potentially enabling similar assessments with follow-up compounds.
We generated a set of human QC variants, which carry amino acid changes at the putatively relevant sites of inhibitor interaction. Summarizing the observations, we conclude that the hydrophobic funnel at the entrance of the catalytically active site represents the most relevant site (besides the complexation of the metal ion) for secondary interaction of hQC and PBD150. The significant rises in the inhibition constants by mutation of F325 and I303 as well as the enforced potency by increasing the hydrophobicity at S323 support a major contribution to binding of the inhibitor (Figure 7). In contrast, the only minor effects evoked by mutation of W329 disagree with a binding situation as observed in human QC crystals, where PBD150 is positioned within proximity to W329. Likewise, the calculation of the binding energies supported similar contributions of I303, F325, and W329.
The binding mode of the inhibitor obtained by the presented mutagenesis study thus shows greater consent with the binding of PBD150 in the murine QC crystal (compare Figures 1B and 7). In accordance with this conclusion, the calculated energies of interaction already favored such a conformation (Tables 2 and S4). The reasons for the repositioning of PBD150 are either a direct interaction between the neighboring molecule in the crystal package and the inhibitor or a repositioning of the loop at the active site (the exchanged loop) caused by the protein–protein interaction. The data support that the binding conformation of PBD150 to human QC in solution significantly differs from that recently obtained by co-crystallization. The combination of semi-empirical calculations and site-directed mutagenesis provided useful tools to clarify the (seeming) discrepancies observed in the binding modes to the highly homologous murine and human QC.
The design of the present study might have implications for inhibitory characterization of other drug targets. The approach has the advantage that – based on an established procedure of protein expression and purification – different inhibitors can be rapidly screened to validate conclusions drawn from in silico design of compounds. This is especially of interest, because computer-assisted drug design represents a useful tool to predict binding of compounds and to facilitate design of novel structures. The testing of drug efficacy can be thus combined with an in vitro assessment of the tentative binding mode in solution, even in a high-throughput manner. With regard to QC, current investigations concentrate on the modification of the approach to facilitate isolation of isoenzyme-specific drug candidates.
We thank D. Ruiz Carillo, J. Rahfeld, and Prof. M. Stubbs for helpful discussion and support. We are grateful for the support of M. Kossner and G. Kirsten (Chemical Computing Group) in the integration of the QM suite in MOE. We further thank L. Westerhoff (QuantumBio Inc.) for helpful discussions and technical support. The project was supported by grants FK0313185 and FK0315662 of the federal ministry of science and education of Germany (BMBF) to Probiodrug (H.-U. D.).
Conflict of Interest Statement
The authors declare following conflict of interest: BK, MB, MW, UH, and SS are employees of Probiodrug. HUD serves as CSO of Probiodrug and holds stocks of the Probiodrug group.