Docking of known UPPS inhibitors
In earlier work (8), we reported the X-ray crystal structures of five bisphosphonate compounds bound to E. coli UPPS. We found up to four distinct binding sites for BPH-629 (PDB 1D code 2E98). However, only Site 1 (Figure 2A) was occupied in all of the five bisphosphonate/UPPS structures reported, with ligand interactions involving ASP-26, ASN-28, ARG-39, HIS-43, ARG-51, ARG-77, PHE-89, ARG-102, and HIS-103 being common to all structures. On average, there were ∼14 bisphosphonate–protein interactions in Site 1, but only ∼9 for Sites 2–4, compared with 15 interactions (in Site 1) for the substrate analog, S-thiolo-farnesyl diphosphate (FSPP). This suggests that Site 1 might represent the strongest binding site for bisphosphonate UPPS inhibitors. To test this hypothesis further, we carried out a computational docking investigation using the Glide programa (28,29).
Figure 2. (A) Superposition of the docked (colored by atom type) and co-crystallized poses (green) of BPH-629 bound to the 2E98 crystal structure. (B) Docked poses of 29 BPH inhibitors into the 2E98 structure. (C) Docked poses of 1i, 1j, 4a, 4g, 4j, 4l, and 4m into the 2E98 structure. Ligand poses showed a very poor alignment with the substrate, farnesyl pyrophosphate (FPP) (shown in green). (D) Docked poses of 1i, 1j, 4a, 4g, 4j, 4l, and 4m into the fourth most populated MD-derived structure. All ligand poses showed good alignment with the substrate, FPP (shown in green).
Download figure to PowerPoint
The co-crystallized structure of UPPS and BPH-629 (PDB ID 2E98), after removal of ligands, was subjected to numerous docking calculations with the program GLIDE at the XP level. Four BPH-629 ligands are present in the active site of chain A of UPPS in the crystal structure, identifying four distinct binding sites. As shown in Figure 2, the docking calculation accurately reproduced the crystallographic poses of BPH-629 in the UPPS-active site. Although four binding sites were identified in the UPPS crystal structure, the poses of all bisphosphonate (BPH) ligands generated by GLIDE were located primarily in the first binding site (Figure 2B). These docking results are also in good agreement with the experimental observation that, in the crystal structure, ligands bind more tightly to the first binding site (8). To further validate our model, the 29 BPH ligands (Figure 2), with known potency against UPPS, were docked into the structure, and the estimated free energy of binding (Glide XP score) of each ligand was compared with its respective pIC50. When these docking results are compared with the enzyme inhibition pIC50 results (pIC50 = −log10 IC50 [M]), we find a correlation coefficient of −0.5 (Figure S3). The correlation coefficient is promising but not strong. This could be in part because of the many approximations inherent in docking software, or artifacts from crystallographic conditions including the high concentration of ligand in the medium (5 mm). The structural changes induced by the three other inhibitors occupying the active site, which were removed for docking, may also affect the correlation.
In recent work, Peukert et al. (11) described a class of potent and selective UPPS inhibitors. The scaffold from a known UPPS inhibitory compound was modified to create a small library of substituted tetramic acid and dihydropyridin-2-one-3-carboxamides. These compounds, inspired by the binding mode of FPP, possess two hydrogen bond acceptors and a hydrophobic group, which are important interaction sites. The proposed inhibitors showed sub-micromolar UPPS inhibition and antibacterial activity against Gram-positive bacteria. To investigate the binding mode of this class of inhibitors, inhibitors 1i, 1j, 4a, 4g, 4j, 4l, and 4m (Figure 2) were docked into the crystal structure 2E98. These inhibitors bear no structural similarity with the BPH compounds, and they were designed to adopt a binding mode similar to the natural substrate, FPP. As displayed in Figure 2C, docking of these inhibitors into 2E98 generated unexpectedly poor results with unreasonable binding poses. The molecules were distributed over all four binding sites with none of the poses showing a good alignment with FPP. This raised the question: is the protein in a different conformation when it binds these non-bisphosphonate inhibitors?
Identifying inhibitor-bound UPPS conformations from MD simulations
To investigate the dynamic behavior of UPPS, we performed clustering analysis of the MD trajectories of the HIP43 and HID43 systems. Each trajectory was fit to the alpha carbons of all UPPS subunit A residues, with the exception of the C and N-termini and the flexible segment containing residues 73–80. Clustering was performed by employing the Gromos method with RMS differences of a selection of active site residues (residues 23–51, 67–93, 96, 110, 141–145, 194, 204, 221–222) within gromacs version 3.3.1.b To choose the appropriate cutoff radius, several cutoffs were investigated, resulting in a final cutoff of 1.8 Å for the HID43 simulation and a cutoff of 2.2 Å for the HIP43 simulation. The first five clusters represent more than 90% of the entire trajectory. Docking of the 29 BPH ligands into the five representative MD cluster structures did not show any improvement over the results from the crystal structure. The clustered structures demonstrate that the crystal conformation is not highly populated in the apo enzyme trajectory. It is worth noting that the crystal structure 2E98 contains four BPH-629 molecules in its active site because of the high inhibitor concentration used in the experiment (8). Thus, in this case, it is expected that induced fit effects are playing an important role in the binding process, promoting large conformational changes in the active site of the enzyme, expanding the volume from the unbound state. The active site of 2E98 is very large when compared with the most populated structures extracted from the MD simulations (1032 Å3 compared to an average of 332 Å3 volume in the MD simulations). Clustering analysis showed that all representative structures from the MD simulation displayed active sites with significantly decreased volumes when compared to the crystal structure. We believed that this highly open conformation could be favoring the binding of BPH-containing ligands. To test this hypothesis, we used the povme software to calculate the volume of the pocket throughout the MD trajectory, and in numerous crystal structures. Figure 3A displays the volume of the HIP UPPS simulation binding pocket calculated for each selected frame along the time-course of the simulation, as well as a time-averaged size along the simulation, and the size of selected crystal structures. As can be seen, conformational states that show pocket volume close to the one observed in the holo crystal structure (2E98) are rarely sampled in the simulation of the apo form of UPPS. The crystal structure 2E99 is bound to another bisphosphonate BPH-608 (11) and also has a large pocket size, 873 Å3. Although protein coordinates from the crystal structure, 2E98, were used to build our UPPS model, when simulated in the absence of ligands, the pocket volume decreases significantly after the initial equilibration steps and never returns to its initial value during the production phase. Interestingly, at approximately 12 ns, the pocket widens to 939 Å3 and reaches within 100 Å3 of the size observed in the crystal structure (Figure 3A). In order to evaluate the influence of the protein pocket size on the docking results, the frame with the largest pocket was selected and used for further calculations. Docking of the BPH compounds to the selected structure showed a very good agreement with the experimental pIC50 values, with a correlation coefficient of −0.8 between the docking score and pIC50 (Figure S3). This represents a significant improvement when compared to the results obtained from the crystal structure 2E98 (correlation coefficient of −0.5). We partially attribute this improvement to the opening of the bisphosphonate binding region in Site 1. In the 2E98 crystal structure, the positively charged residues HIS43 and ARG77 are in close proximity. This region widens in the largest MD-derived structure, minimizing the scoring penalty for ligands with a positively charged N near the positively charged region of the protein, such as BPH-641 and BPH-642 (the two lowest scoring ligands when docked against 2E98). This repulsive force is strong enough that both BPH-641 and BPH-642 take on unexpected poses with poor scores (Figures S1 and S4). The wider bisphosphonate binding region is thus able to accommodate a larger variety of bisphosphonate containing ligands.
Figure 3. Volume distribution of the HIP43 undecaprenyl pyrophosphate synthase binding pocket 2E98 crystal structure and the apo crystal structure. (A) Volume of the binding pocket along the MD trajectory. The black line shows data taken every 10 ps, the overlayed gray line is the average over every 100ps. (B) Frequency at which different volumes of the pocket are sampled. The size of the bisphosphonate-bound crystal structure (2E98), the newly described apo crystal structure (Apo), and cluster 4 that docked tetramic acids well, are represented by labeled dashed lines in both graphs.
Download figure to PowerPoint
To help confirm the nature of this pocket closure on ligand removal (or expansion on ligand binding), we crystallized the E. coli UPPS in the ligand-free form. Full crystallographic data and structure refinement details are given in Table 1, and a comparison of this structure (PDB ID code 3QAS) with that of the bisphosphonate-containing species (PDB ID code 2E98) is shown in Figure 4. Crystal structures of the ligand-bound and apo form of UPPS reveal that a major pocket closure occurs upon ligand removal, from 1032 to 432 Å3 and is in good agreement with the time-averaged pocket volume (332 Å3) of the apo form obtained from the MD simulation.
Figure 4. (A) The apo crystal structure in green and the bisphosphonate-bound crystal structure in blue. (B) The apo crystal structure with 1 Å spheres filling the active site pocket. (C) The bisphosphonate-bound crystal structure with 1 Å spheres filling the active site pocket. Note the significantly larger pocket size in the bisphosphonate-bound structure when compared to the apo crystal structure.
Download figure to PowerPoint
To further investigate the binding mode of tetramic acid and dihydropyridin-2-one-3-carboxamide inhibitors, molecules 1i, 1j, 4a, 4g, 4j, 4l, and 4m (Chart 1) were docked into the five most representative MD structures. Interestingly, only docking into the most representative member of the fourth cluster generated ligand poses similar to the one observed for the FPP natural substrate (Figure 2D). Unlike the results obtained from the crystal structure 2E98, all docked tetramic acid and dihydropyridin-2-one-3-carboxamide inhibitors reproduced the binding mode of the substrate FPP (Figure 2D). Because our bisphosphonates bind to structures with open active sites, we wanted to know whether these compounds, which were designed using a pharmacophore hypothesis of FPP binding, bind to the same size pockets as FPP-bound or apo crystal structures. Four crystal structures are described by Guo et al. (8) which have IPP or FSPP, an FPP analogue, in the active site. These structures had active site volumes ranging from 295 to 330 Å3, which is similar to the calculated volume of the fourth cluster, 377 Å3. This indicates that tetramic acids and dihydropyridine inhibitors, as well as the natural substrates, bind to more closed forms of the enzyme, similar to the apo state, while bisphosphonates bind to an open form.
As the conformational states that bind tetramic acid and dihydropyridine inhibitors are sparsely populated, and those that bind bisphosphonates are rarely sampled in our MD simulations, our results suggest that a population shift mechanism (30) may play an important role in changing the equilibrium towards other conformations upon inhibitor binding. Furthermore, it appears that the active site of the apo structure opens and expands considerably upon binding of bisphosphonate ligands, shifting the population of UPPS enzymes to a markedly different conformation. To further investigate this population shift, we plotted the principal components (PC) of our HIP trajectory (Figure 5C) and highlighted both the BPH binding structure (green) and tetramic acid and dihydropyridine binding structure (blue). Principal component analyses break the complex motions of molecular dynamics simulations into just a few variables. The two eigenvectors shown are the principal components of motion that account for the most motion. In the event of inhibitor binding, the PC results suggest a shift away from the center of the most highly sampled area of the apo MD simulation. Therefore, our results indicate that structurally diverse inhibitors recognize a specific set of conformational states of the receptor, which can vary significantly between families of ligands.
Figure 5. Principal component analyses (PCA). (A) Principal component analysis (PCA) calculated from the HID43 trajectory. (B) The extreme conformations of the flexible loop (residues 72–82) in HID43 are shown in red and blue. (C) PCA calculated from the HIP43 trajectory. The green circle indicates the largest conformation, and the blue circle indicates cluster 4. (D) The extreme conformations of the flexible loop (residues 72–82) in HIP43 are shown in red and blue.
Download figure to PowerPoint
Breathing motion of the catalytic pocket
The most dynamic region of UPPS is the loop comprising residues 72–82. Besides the terminals, the root mean square fluctuation (RMSF) of this loop region is by far the dominant feature in Figure S5 for both the HID43 and HIP43 simulations. The RMSF plots also reveal that the HIP43 simulations show a slightly more flexible protein, especially the helices as defined by residues 80–100 and 120–140. One of the highly populated conformational states of HID43 shows ARG77 deep inside the UPPS active site. Similar conformations were not observed in the HIP43 simulation. This result suggests that differences in electrostatic interactions, originating from different protonation states of active site HIS43, may affect the loop dynamics and the motion of the catalytically important residues on the flexible loop. It has been proposed that ARG77 plays an important role in the catalytic mechanism by helping transfer the pyrophosphate group from one substrate to another in the active site (5). Kinetic studies have shown a 1000-fold decrease in activity when ARG77 is mutated to ALA77 (7). Our MD trajectories revealed that ARG77 more extensively samples regions between the first and second sites in the HID43 than in the HIP43 simulations. This behavior can be attributed to the difference in electrostatic forces originating from the doubly protonated HIP43, which prevents this movement. The flexible loop in the HID43 simulation shows substantially more movement into and out of the pocket. The effect of the different protonation states of HIS43 on the dynamic behavior of UPPS can also be seen in Figure 5. The projection of the trajectories onto the first two eigenvectors calculated from principal component analysis reveals that each system clearly samples different regions of the conformational space. This result may have catalytic implications, as it suggests that ARG77 is only able to successfully transfer the pyrophosphate group between the two sites when HIS43 is singly protonated. Additionally, the imidazole group of HIS43 interacts directly with the pyrophosphate, and experimental results have shown that a 1000-fold decrease in catalytic activity is observed when this residue is mutated to ALA43. These results support the hypothesis proposed by Chang et al. (6) who suggest the initial binding of FPP is encouraged by the protonated imidazole group followed by proton donation from HIS43 to FPP, inducing changes in the dynamic behavior of the flexible loop.