Royal Society Publishing

Large scale molecular dynamics simulation of native and mutant dihydropteroate synthase–sulphanilamide complexes suggests the molecular basis for dihydropteroate synthase drug resistance

Fabrizio Giordanetto, Philip W Fowler, Mansoor Saqi, Peter V Coveney


Antibiotic resistance is hampering the efficacy of drugs in the treatment of several pathological infections. Dihydropteroate synthase (DHPS) has been targeted by sulphonamide inhibitors for the past 60 years and has developed different amino acid mutations to survive sulpha drug action. We couple homology modelling techniques and massively parallel molecular dynamics simulations to study both the drug-bound and apo forms of native and mutant DHPS. Simulations of the complex between sulphanilamide and Streptomyces pneumoniae, DHPS shows how sulphanilamide is able to position itself close to 6-hydroxymethyl-7, 8-dihydropteridine-phosphate in a suitable position for the enzymatic transformation whereas in the mutant complex the sulpha drug is expelled from the catalytic site. Our simulations, therefore, provide insight into the molecular basis for drug resistance with S. pneumoniae DHPS.


1. Introduction

Paediatric pneumonia is the most widespread acute respiratory infection in developing countries claiming 3.5 million deaths annually (Garenne et al. 1992). Streptomyces pneumoniae, alongside Haemophilus influenzae, is the principal causative agent of severe paediatric pulmonary diseases. On the recommendation of the World Health Organisation, trimethoprim in combination with sulphamethoxazole (co-trimoxazole) has been largely prescribed in order to reduce the high death rate caused by pneumococcal infections (Straus et al. 1998). However, in recent years, high rates of antimicrobial resistance have emerged. Specifically, high rates of co-trimoxazole resistance against S. pneumoniae isolates have been reported (Klugman 1990). The target for sulphamethoxazole, a sulphonamide drug, is dihydropteroate synthase (DHPS), a protein that is essential for micro-organism proliferation. DHPS is a fundamental enzyme for the de novo synthesis of folate cofactors in lower eukaryotes and prokaryotes whereas it is absent in mammals. It catalyses the condensation of 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate with para-aminobenzoic acid to form pyrophosphate and 7, 8-dihydropteroate as shown in figure 1 (Shiota et al. 1969). Dihydropteroate is subsequently transformed to folic acid through the action of dihydrofolate synthase. Since DHPS is only present in micro-organisms, this enzyme has been long exploited as a selective drug target. The sulpha drugs, including sulphonamides and sulphones, were the first class of synthetic antimicrobial agent and have been used widely since 1933 (Wingard 1991). They are synthetic ligands whose structures are very similar to that of the natural DHPS substrate, para-aminobenzoic acid, and act as alternative substrates forming sulphapteroates (Brown 1962; Shiota et al. 1969; figure 1). These compounds are not recognised by dihydrofolate synthase and cannot be transformed further to folate (Swedberg et al. 1998).

Figure 1

Enzymatic reaction catalysed by DHPS. Incorporation of para-aminobenzoic acid leads to the formation of 7, 8-dihydropteroate, which continues the folate biosynthesis pathway. Reaction with sulphanilamide produces sulphapteroate and blocks the following synthetic steps.

In many cases, resistance to sulphonamides is associated with chromosomal mutations within the gene encoding DHPS (Huovinen et al. 1995). Different studies have reported the occurrence of single and/or multiple amino acid mutations on sulphonamide-resistant clinical isolates of S. pneumoniae DHPS (Lopez et al. 1987; Maskell et al. 1997; Padayachee & Klugman 1999). Amino acid mutations that cause drug resistance are likely to play a prominent role in the molecular recognition process between the protein and the drug. Three-dimensional information on the structure of both native and mutant proteins, possibly bound to the corresponding drug, would therefore prove valuable in an attempt to better characterize drug resistance mechanisms. Unfortunately, no experimental three-dimensional structures for S. pneumoniae DHPS are currently available. At the time of our investigation, three crystal structures of native DHPS from Escherichia coli (Achari et al. 1997), Staphylococcus aureus (Hampelle et al. 1997) and from Mycobacterium tuberculosis (Baca et al. 2000) had been reported. Structures of Bacillus anthracis DHPS have been recently released but as yet not reported. In all these experimental structures, DHPS shows a ‘TIM-barrel’ fold (see figure 2). The catalytic and sulpha drug binding sites are located at the carboxyl-terminal pole of the barrel and are surrounded by four different loops: 1, 2, 5 and 7, according to the notation employed by Baca et al. (2000). The crystal structure of E. coli DHPS (Achari et al. 1997) has been solved as a complex between DHPS, 6-hydroxymethyl-7, 8-dihydropteridine and sulphanilamide. The complex displays the sulpha drug embraced between loops 2 and 7 at the edge of the catalytic site. Unfortunately, there is no crystal structure with para-aminobenzoic acid bound which makes it difficult to compare drug- and ligand-bound, native- and mutant-forms of DHPS.

Figure 2

Cartoon representation of native S. pneumoniae DHPS. The positions of the amino acid mutations, under study here (see figure 3), are highlighted as yellow spheres. The heavy atoms of three key amino acids are also displayed in a sphere representation.

Figure 3

Sequence to structure alignment employed to build the native and mutant S. pneumoniae DHPS models. Mtb refers to M. tuberculosis. Black shading indicates amino acid identities while grey shading represents conserved physico-chemical properties. Red shading highlights the mutated amino acids conferring sulpha drug resistance to S. pneumoniae DHPS.

Most of the amino acid mutations that confer drug resistance in S. pneumoniae DHPS are located on loops 2 and 5 (Lopez et al. 1987; Dallas et al. 1992; Fermer & Swedberg 1997; Maskell et al. 1997; Vedantam et al. 1998; Padayachee & Klugman 1999). This can be seen for the three mutants under study (‘A’, ‘B’ and ‘C’; Maskell et al. 1997) in figure 2. The high degree of flexibility of these loops observed in the three different crystal structures does not support the use of static molecular models for drug resistance, even if such an experimental structure was available. Thus we build a starting structure using homology modelling, a reasonable approach when homologues of high sequence identity exist, and then evolve the structure using large-scale molecular dynamics (MD). Since our MD simulations depend upon the accuracy of our homology models, we present a validation of our structural prediction approach. We also briefly discuss some of the issues when using MD in a structural prediction context. To validate the insights produced by computational studies, it is important that they may be further investigated by experiment.

Reducing the elapsed wall-clock time required to perform a simulation is a basic requirement for strengthening this interaction between experimental and computational studies. The comparatively recent availability of highly scalable MD codes and tightly coupled massively parallel computers permits us to turnaround simulations of these large systems in less than a day in contradistinction to the weeks or months required for conventional MD codes and serial or loosely coupled parallel hardware platforms. The deployment of MD codes on a suitable computational Grid allows yet further improvements, for example, permitting the scientist to run more numerous and longer simulations to improve confidence in any result. Alternatively, as discussed elsewhere in this issue, a computational Grid allows distributed calculations not previously possible, for example computing the difference in binding free energy of sulphanilamide between the native and mutant. We believe that the use of high-performance computing and computational Grids in the life sciences will bring benefits to all concerned.

Section 2 details the tools and algorithms we used to build and simulate the homology models. We describe how the homology models were built and discuss their validation in §3. The purpose of this paper is to investigate the structural differences between the native- and mutant-forms of S. pneumoniae DHPS in both apoprotein and complexed states and these differences are examined in §4. It has been shown experimentally that for sulphanilamide or para-aminobenzoic acid to bind, 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate must already be bound (Vinnicombe & Derrick 1999). Hence, we will primarily study the differences in behaviour between native DHPS and a typical sulpha-resistant mutant form (‘A’) complexed with sulphanilamide and 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate. The other two mutants (‘B’ and ‘C’) are considered but not simulated. We also study the differences in loop interactions between the native-and mutant-apo forms of S. pneumoniae DHPS. Studying situations where the enzyme does and does not have the drug-bound helps us to understand how the system may behave during the binding event, the dynamics of which elude present simulation capabilities. Based on our simulations, we propose and discuss a molecular model for drug resistance within S. pneumoniae DHPS.

2. Materials and methods

The amino acid sequences of S. pneumoniae DHPS native- and mutant-forms were retrieved from the EMBL database (Rodriguez-Tomé et al. 1996) (accession numbers: AF245691, AF245698, AF245696 and AF245697 for native DHPS and the ‘A’, ‘B’ and ‘C’ mutants, respectively). Multiple sequence alignments were performed employing ClustalW (Thompson et al. 1994) and T-Coffee (Notredame et al. 2000). Three-dimensional structure coordinate files were retrieved from the Protein Data Bank (PDB; Berman et al. 2000). Search and evaluation of structural neighbours and structural comparison were carried out using Dali (Holm & Sander 1993). Amino acid conservation was determined using phylogenetic relationships (Clamp 1998). Secondary structure predictions were obtained using PsiPred V2.0 (Jones 1999b) and PredictProtein (Rost & Sander 1993, 1994; Rost et al. 1994). Sequence to structure alignments were performed using GenThreader (Jones 1999a) and T-Coffee (Notredame et al. 2000).

The three-dimensional models of native and mutant DHPS were generated by application of restraint-based homology modelling methods as implemented in the program Modeller v6.2 (Šali & Blundell 1993). Loop fragments were generated by loop search algorithms available within Modeller (Fiser et al. 2000) and evaluated according to the best-score values. Structural evaluation of the overall three-dimensional models was accomplished using the package Procheck (Laskowski et al. 1993). The results obtained were consistent with those obtained for the template structure and they did not reveal any unusual features. Alignment of the structures and calculation of root mean square deviation (RMSD) values were accomplished using both VMD scripts (Humphrey et al. 1996) and the McLachlan algorithm (McLachlan 1982) as implemented in the program ProFit.

All the MD used in the investigation into the molecular basis for drug resistance was performed employing the Amber PARM98 force field (Cornell et al. 1995), as previously ported (Grindon et al. 2004) to the large-scale atomistic/molecular massively parallel simulator (LAMMPS; Plimpton & Hendrickson 1996). As part of the validation of the structural prediction method MD was carried out using the Amber 98 forcefield in Namd2 (Kalé et al. 1999).

Atomic charges for 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and sulphanilamide were calculated by fitting the corresponding electrostatic potentials (RESP approach; Bayly et al. 1993; Cieplak et al. 1995) as computed with Gaussian98 Revision A.71 and its 6-31G* basis set at the Hartee-Fock level of theory. Force field (van der Waals, dihedral, bond and angle) parameters were taken from the general amber force field (Wang et al. 2001).

The final homology-built structures were initially subjected to 3000 steps of energy minimisation in vacuo. Subsequently, the systems were neutralized by adding sodium counter-ions and solvated by equilibrated TIP3P water, thus yielding a box of approximate size 75×75×75 Å. This resulted in systems of approximately 37 000 atoms. Solvent and ions were energy-minimized and then evolved using MD for 20 ps holding the protein atoms fixed and both solvent and solute were energy-minimized again. The apoprotein simulations were adjudged to have equilibrated using standard metrics after 100 ps and the complexed DHPS simulations after 400 ps. To collect adequate statistics, the apoprotein and complex DHPS systems were evolved for a further 1 and 2 ns, respectively. A single 2 ns production run in a NVT ensemble required only 30 h wall-clock time on 128 Cray T3E processors thanks to the high scalability of LAMMPS (Plimpton & Hendrickson 1996). Solvent and solute were separately coupled to a Langevin thermostat to maintain the temperature at 300 K. Electrostatic interactions were computed using the particle–particle particle–mesh Ewald summation method (Hockney & Eastwood 1998) while time-stepping made use of the reversible reference system propagation algorithm (r-RESPA) scheme (Tuckerman et al. 1991). In this framework, bonded forces are computed every shorter subtime-step (1 fs) while short-range non-bonded pairwise (van der Waals) and long-range (Coulomb) forces are calculated every longer time-step (2 fs).

3. Results

Building and validating the S. pneumoniae homology models was a significant challenge. We start this section by detailing the homologues used before discussing the validation of our homology modelling approach. The section finishes with a description of the homology modelling and a brief introduction to the MD simulations.

(a) Homologues

The native S. pneumoniae DHPS protein comprises 314 amino acid residues. Fold recognition programs identified the known three-dimensional structures of DHPS from E. coli (PDB code: 1aj0; Achari et al. 1997), S. aureus (PDB code: 1ad4; Hampelle et al. 1997) and M. tuberculosis (PDB code: 1eye; Baca et al. 2000) as suitable templates for a model of unliganded DHPS. However, we employed only the DHPS structure from native E. coli as a template when modelling complexed DHPS since it has been obtained as a complex with sulphanilamide. Native S. pneumoniae DHPS shares sequence identity percentages of 39, 35 and 36% with E. coli, S. aureus and M. tuberculosis DHPS, respectively. Our derived alignment is presented in figure 3 and the additional ‘B’ and ‘C’ mutants are shown. Residue numbering is based on the native S. pneumoniae DHPS shown in this figure. S. pneumoniae proteins align well with the corresponding templates, showing several amino acid identities and good conservation of physico-chemical properties. However, S. pneumoniae DHPS differs significantly from the template structures in terms of the lengths of loop 5 (from 18 to 26 inserted residues) and loop 7 (from 5 to 11 inserted amino acids).

(b) Validation of structural prediction approach

It is important that the S. pneumoniae homology model is of high accuracy (i.e. close to its equilibrium ensemble of structures) for use in MD simulations. If it were not, then while in theory we would eventually approach equilibrium, in practice we would find this would take a very long time and would certainly require more computational resource than we could currently expect to have access to. To validate our approach, we used two of the three S. pneumoniae homologues (S. aureus & M. tuberculosis DHPS) to predict a structure for the third (E. coli DHPS) which we then compared to its known structure (PDB code: 1ajz). Modeller was used in a standard way to generate an ensemble of 100 models from which the structure with the lowest objective function was chosen. No loop modelling was performed. This process is deliberately less sophisticated than that used for the S. pneumoniae homology model to avoid the possibility that knowledge of the target structure influenced the validation process. We expect the homology modelling process used for S. pneumoniae to be more accurate since it has more homologues and uses more information, for example the diversity in homologue loop conformation.

We used metrics derived by the Critical Assessment of techniques for protein Structure Prediction meetings (CASP)2 to assess the accuracy of our predictions (Zemla et al. 2001). The resulting model predicted the secondary structure exceptionally well: 93.6% of residues were correctly predicted (CASP Q3 score) and had a total alpha carbon (Cα) RMSD of 2.52 Å when aligned on the core secondary structure and a core CαRMSD of 1.80 Å. 83% of the Cα enjoy better than a 2 Å similarity which translates into a CASP GDT-TS score of 0.74. To compare this to the state-of-the-art we examined a similar target from the fourth CASP experiment. T0122, a Tryptophan synthase alpha subunit from Pyrococcus furiosus, also has a TIM-barrel topology; its best homologue has identical function (from Salmonella typhimurium—PDB code: 1geq) and shares a sequence identity of 34.9% (Tramontano et al. 2001). Our prediction falls in the top third of the 86 predictions from the CASP results for this protein, having an average CαRMSD of 2.81 Å and an average GDT-TS of 0.72 (predictions with CαRMSD<5 Å). However, for S. pneumoniae we expect to achieve a better model since we used two homologues of higher sequence identity.

As we would expect, it is the loops that are predicted least accurately, with loops 1 and 5 predicted least well. The crystal beta values indicate areas of flexibility in the protein since the E. coli crystal structure was determined at sufficiently high resolution (2 Å) to minimize the error component. We find good correlation between CαRMSD and beta values. That the residues matching least well between the homology model and crystal structure are also those with the largest uncertainty in position in the crystal structure give us further confidence in the technique. As mentioned later, loop 1 displays several conformations despite sharing a high sequence identity across all the homologues. Not taking account of this flexibility results in a poor prediction as illustrated by the CαRMSD of 6.32 Å over this loop.

Of more importance is that the active sites are well predicted: the 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate binding site comprises 10 residues, seven of which are part of the core structure and consequently it is well predicted with a carbon atom RMSD of 1.71 Å. If T62 (a loop 2 residue) is excluded this falls to 1.27 Å. The sulphanilamide active site comprises three loop residues and has a carbon atom RMSD of 1.57 Å, although the orientation of the side chains is such that it is unlikely sulphanilamide could bind to this structure. Although the homology model is of good accuracy in terms of these conventional metrics, we argue that, from a theoretical perspective, the protein structure prediction problem is better restated as assessing whether the microstates represented by the prediction and target both belongs to the same macrostate under physiological conditions. Unfortunately, as we shall see, it is difficult to prove or disprove that this is the case but we find that it is rewarding to examine the problems encountered.

To generate an ensemble of structures under physiological conditions we solvate, neutralize and thermalize both homology model and crystal structure (referred to hereafter as prediction and target, respectively) and then evolve both systems using MD. After equilibration, comparing the structures of both prediction and target at time t yields RMSD(t). The average of these RMSDs is 3.32 Å which is higher than the value we obtained for our homology model (2.52 Å). Efforts to try to retrieve parts of the trajectories where the models were more similar and therefore (we assume) closer to each other in phase space by correlating RMSD with thermodynamics variables, have so far failed. Hence it is not currently possible to recover these ‘improved’ predictions blind, i.e. without knowledge of the target structure.

It is extremely difficult to determine if these trajectories belong to the same macrostate. Indeed, it is far from clear that it is valid to compare the RMSD of 3.32 Å between the two trajectories at 300 K with the RMSD between the prediction and target of 2.52 Å at 100 K. Taken at face value we appear to be confirming the widespread view that MD ‘worsens’ structural predictions in nearly all cases (Schonbrun et al. 2002). Were we to submit a structure from the prediction trajectory to CASP, a similar comparison would be made since our prediction at 300 K would be compared with an X-ray crystallographic structure at 100 K.

In both cases we need to acknowledge that there are two effects at work. First, we are comparing systems at different temperatures and, second, we are comparing a static X-ray crystallographic structure with a MD trajectory. It is manifestly flawed to compare RMSDs across different temperatures and consequently we cannot say that MD has degraded our initial prediction. To test the second effect, we could try and simulate our solvated system at 100 K: unfortunately, the explicit water would freeze and the forcefield may no longer be valid. In conclusion, there are no unambiguous means directly available for assessing the performance of MD when predicting protein structures.

(c) Homology modelling

We built three-dimensional models for the native and mutant, unliganded and complexed DHPS proteins using comparative modelling techniques as described in §2. All the resulting structures fold in a core barrel comprised eight β-strands (V11–N17, M49–G53, L87–D91, L108–D111, V132–M135, I198–D201, I230–L232 and V280–V283) surrounded by eight α-helices (L32–A44, E69–E83, S95–A103, M121–A128, P172–E190, K210–H224, D263–R275 and V286–A302). The central barrel accommodates the catalytic site with a pterine binding pocket comprising D91, N110, I112, V133, M135, D201, G203, F206, F231, G233, K237 and R282. Substrate/ligand access to the catalytic core is modulated at the carboxyl terminal pole of the barrel by four main structural elements: loop 1 (V18–A31), loop 2 (G54–E68), loop 5 (F136–L171) and loop 7 (G233–R262).

All homology models built were highly similar to their homologue(s) as shown by a maximum RMSD of 0.53 Å when using the α-carbon atoms of secondary structural elements for the fit. The main differences were found in the loop conformations. Loop 1 displayed the most diverse spatial arrangement in the templates: it is found in an open conformation in the E. coli structure (Achari et al. 1997) whereas in the M. tuberculosis DHPS (Baca et al. 2000) it closes over the catalytic site. The same loop adopts an intermediate conformation in the S. aureus DHPS structure (Hampelle et al. 1997). In the S. pneumoniae apo DHPS models loop 1 is in an ‘averaged conformation’ similar to the one displayed by S. aureus. Loop 2 has not been observed in the M. tuberculosis DHPS crystal structure (Baca et al. 2000) while the other templates showed different geometries. The resulting unliganded DHPS models possess an averaged conformation for loop 2.

Amino acid mutations conferring sulpha-drug resistance in S. pneumoniae occur within loops 2 and 5. The extra amino acids on loop 2 are accommodated without large conformational changes when compared to the native DHPS model (figure 2). The biggest difference is observed for loop 2 of the ‘A’-mutant, which yields a RMSD value of 1.01 Å when superimposed on the corresponding one in the native protein. The restraint-based comparative modelling approach handled the long insertions found in S. pneumoniae loops 5 and 7 automatically. Both loops pointed away from the core of the TIM-barrel and they did not display any specific interactions with other structural elements. We applied loop optimisation tools within Modeller (Fiser et al. 2000) to identify possible alternative loop geometries. However, after scoring and structural evaluation, the best resulting candidates were highly similar overall.

A different approach was required for constructing the complexed S. pneumoniae DHPS homology model. The S. pneumoniae DHPS ‘A’ mutant contains the two inserted amino acids (an arginine and a proline) in loop 2 conferring sulpha drug resistance (Padayachee & Klugman 1999; figures 2 and 3). The conformation of R58 found in the template structure, necessary for the coordination of both 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and sulphanilamide (Achari et al. 1997), was preserved by keeping the first part of loop 2 (G54–P59) fixed during the model building procedure. This yields final RMSDs of 0.11 and 0.35 Å for native and mutant loop 2, respectively. The mutated amino acids pointed away from the core of the TIM-barrel and they did not display any interaction with 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and sulphanilamide.

The biggest difference between the amino acid sequences of E. coli and S. pneumoniae DHPS lies within loop 5. Here, an insertion of 18 residues needs to be allocated in the latter protein. In the crystal structure of E. coli DHPS, loops 2 and 5 enjoy several molecular contacts. These interactions were maintained in the final S. pneumoniae models. The inserted amino acids are solvent-exposed in the final conformation of loop 5 and, according to distance criteria, they are not involved in any molecular interactions with the rest of the system.

The catalytic site is located at the carboxyl-terminal side of the core barrel. In the E. coli DHPS structure (PDB code: 1aj0), 6-hydroxymethyl-7, 8-dihydropteridine and a sulphate ion have been co-crystallized with sulphanilamide. An additional crystal structure from the same work (PDB code: 1aj2; Achari et al. 1997) shows 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate but not sulphanilamide co-crystallized with DHPS. The two structures are very similar (RMSD 0.31 Å) and, when superimposed, the sulphate ion in 1aj0 occupies the position of the second phosphate group in the 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate molecule. In the homology-built complexes we removed the original sulphate ion and added the pyrophosphate moiety to 6-hydroxymethyl-7, 8-dihydropteridine using the conformation found in 1aj2. Thus, this procedure yielded 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate in the catalytic site which is necessary in order to simulate the molecular behaviour of sulphanilamide.

6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate is involved in an extended hydrogen bond network in the catalytic pocket of native and mutant DHPS as shown in figure 4. The phenyl ring of sulphanilamide lies almost perpendicular to the plane defined by the pteridine ring of 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and is packed between residues R58 on loop 2 and K237 on loop 7 while the sulphonamide function is hydrogen bonded to the carbonyl backbone groups of residues S235 and R236 on loop 7 (figure 4).

Figure 4

Top (a) and side (b) views of the binding site of S. pneumoniae DHPS homology-built three-dimensional model. Key hydrogen bonds, with bond distances in Angstroms, are shown. All distances are from heavy-atom to heavy-atom and therefore exclude hydrogens. SUL, sulphanilamide; DHPPP, 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate.

In conclusion, all the interactions that involve 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and sulphanilamide in the E. coli DHPS template structure (151) are conserved in the final S. pneumoniae DHPS models. This is desired and supports the reliability of the models built using homology-based methods and enables comparative analysis of the MD simulations now to be described.

(d) Molecular dynamics simulations

Starting from the homology-built structures, followed by energy minimization and thermal equilibration, as described in §2, we subjected both native and mutant S. pneumoniae complexed DHPS to production MD runs for 2 ns. Unliganded DHPS was simulated for 1 ns to allow for investigation of loop–loop interactions. RMSD values from the corresponding starting structures for the complexed DHPS simulations are presented in table 1 for complete structures and selected fragments including the loops, 6-hydroxymethyl-7, 8-dihydropteridine, its pyrophosphate group and sulphanilamide. A summary of the observed amino acid interactions between the different loops (and ligands where relevant) for all the DHPS models is presented in table 2. All the molecular interactions presented and discussed here formed during the simulations and remained stable throughout. They are depicted for the native and mutant complexed S. pneumoniae DHPS in figure 5.

Figure 5

Close-up of the binding site, as displayed by the average structure computed from the last 400 ps of the MD runs, for the native (NAT) and mutant (MUT) S. pneumoniae DHPS. Hydrogen bonds are indicated by lines. Key interactions are coloured red: these are described in detail in table 2.

View this table:
Table 1

RMSD comparison matrix for the native and mutant S. pneumoniae DHPS

View this table:
Table 2

Comparison of the average distances (Å) and durations (ns) for key molecular interactions as observed between the native and mutant simulations

4. Discussion

We are dealing with a subtle problem to which MD is well suited: the sulphanilamide-complexed native and mutant S. pneumoniae DHPS proteins under study here only differ by the presence of two residues on loop 2 in the latter (Padayachee & Klugman 1999). Apart from these amino acid insertions in the mutant, the two homology-built models are virtually identical, as shown by the RMSD values in table 1. Moreover, the spatial arrangement of sulphanilamide and 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and their corresponding molecular interactions are the same in the two starting models (figure 4).

Interestingly, the two systems display a completely different structural response to the drug during the MD simulations (figure 5). Native S. pneumoniae DHPS admits spatial rearrangement of sulphanilamide in the binding pocket: the resulting stable configuration is well set-up for the subsequent catalytic transformation of sulphanilamide and 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate into sulphapteroate (figure 5). A similar sulphanilamide configuration did not arise during the sulpha resistant S. pneumoniae DHPS simulation; indeed, the sulphanilamide moiety slowly drifted away from its catalytic partner (figure 5).

The results indicate that this important difference is mainly due to the interactions between loops 1 and 2 which have been shown to play a decisive role in the coordination of sulphanilamide and 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate (Achari et al. 1997; Baca et al. 2000). In the starting models, loops 1 and 2 do not display any molecular interactions (table 2). Loop 2 is extensively involved in the orientation of both the ligand and the catalytic substrate: the side chain hydroxyl group of T57 is hydrogen bonded to O16 (atom numbering as shown in figure 1) on 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and the guanidinium group of R58 has electrostatic interactions with the pyrophosphate moiety and van der Waals contacts with the phenyl ring of sulphanilamide (figure 4). The side chain of N17, the first amino acid before loop 1, is linked to the pyrophosphate via a hydrogen bond.

During the MD simulation of native complexed DHPS, however, loops 1 and 2 interact with each other via D21 and S62, respectively (table 2). This molecular interaction is also observed during MD simulations of unliganded native DHPS (table 2). The hydrogen bond formed between these two residues stabilizes the corresponding loop conformations. Specifically, the conformation of loop 1 enables N17 to maintain its hydrogen bond with the pyrophosphate group throughout the entire MD simulation period (table 2). N17 is a highly conserved amino acid among the different micro-organisms and it is therefore very likely that it plays a decisive role in DHPS function. In fact, the hydrogen bond between N17 and pyrophosphate has a direct impact on the spatial position of the terminal phosphate group of 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate, which slightly rotates towards N17 and establishes an additional hydrogen bond with the side chain of S56 on loop 2 (table 2). Due to the movement of the pyrophosphate group, the electrostatic repulsion between the phosphate and the sulphonamide group oxygen atoms triggers the rearrangement of sulphanilamide at about 750 ps of elapsed simulation time. The phenyl ring of sulphanilamide started the simulation in a perpendicular orientation with respect to the pteridine ring of 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and after 1150 ps moved parallel to its catalytic partner. This orientation was maintained throughout the remainder of the simulation (to 2 ns), indicating a stable state (table 2, figure 5).

We attribute the specific spatial arrangement of sulphanilamide observed during the native MD run to the position of the amino group of sulphanilamide with respect to 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate. Here, the amino group of sulphanilamide is hydrogen bonded to O16 of the first phosphate group in the pteridine molecule (figure 5). This quasi-electrostatic interaction places the nitrogen atom (N7) in van der Waals contact with C11 of 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate at 3.24 Å (table 2).

From our simulation data, we can propose a possible reaction mechanism for the condensation of sulphanilamide with 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate (figure 6). The first step in the recognition process is the establishment of a hydrogen bond between the amino group of sulphanilamide and O16 on the pyrophosphate moiety of 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate. This step brings the amino group of the drug close to the C11 of 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate from where the lone pair of electrons on the nitrogen atom could easily locate the carbon atom (C11), which is highly polarized due to the adjacent double bond and the electronegative oxygen atom (figure 6). The catalytic transformation can then proceed via a concerted mechanism that finally produces sulphapteroate and pyrophosphate. The simulation data suggest that the release of pyrophosphate is primarily assisted by N17, S56, T57 and R58 side chains, which are involved in an extensive hydrogen bond network with the pyrophosphate moiety, as shown in figure 5.

Figure 6

Schematic of a proposed enzymatic reaction mechanism ‘performed’ by S. pneumoniae DHPS, as suggested by the spatial arrangement of sulphanilamide observed during the MD simulation of the native protein.

The simulation of mutant complexed DHPS did not reproduce the interesting binding mode observed for sulphanilamide in the native DHPS MD run. Rather, sulphanilamide moved progressively further away from 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate during the second half of the 2 ns simulation. A possible explanation of this different behaviour can be found by analysing the interplay between loops 1 and 2 observed for the mutant DHPS and comparing it with the situation in native DHPS. The sulpha-resistant mutant possesses two additional amino acid residues on loop 2, an arginine and a proline (figures 2 and 3). During the simulation, the mutated proline alters the conformation of loop 2 and prevents S62 from interacting with D21, as observed in the native MD run (table 2). Instead, the mutated arginine residue establishes a salt bridge with D21 on loop 1 (table 2).

Interestingly, we observe different behaviour during MD simulations of unliganded mutant DHPS (table 2). Rather than the D21–S62 interaction being disrupted during the mutant simulations, it is the D21–R58 interaction which is not preserved. This arginine is invariant throughout the different species and is involved in the complexation of sulpha-drugs (Achari et al. 1997). It is likely that the presence of the sulphanilamide during the complexed simulations stabilizes R58 in its initial conformation and it is the D21–S63 interaction instead which is not preserved across the mutants. We note that in the complexed simulations, R58 interacts with sulphanilamide (table 2) and propose that this difference between the unliganded and complexed simulations is due to the lack of stabilizing interactions formed between the drug and the enzyme. Regardless of whether the drug is present or not, the mutants show different networks of interactions between loops 1 and 2 to that of the native form indicating a change in loop flexibility due to the presence of mutated residues.

The loss of the D21–S63 interaction provides an important structural difference between complexed native and mutant S. pneumoniae DHPS. The modified conformation of loop 2, due to the presence of a less flexible residue (proline) and the ion pair involving the mutated arginine, has a direct influence on loop 1 geometry. The electrostatic interaction between the mutated arginine and D21 restrains the mobility of N17, the residue adjacent to the beginning of loop 1. A direct consequence, as indicated by the formation times, is that the hydrogen bond between N17 and the pyrophosphate group breaks (table 2) while the second phosphate group enjoys enhanced conformational freedom with respect to the native DHPS MD run (table 1). In the latter case, the electrostatic repulsion between the tightly bound pyrophosphate (hydrogen bonded to N17, S56, T57 and R58) and the sulphonamide function triggered the three-dimensional rearrangement of sulphanilamide.

The situation is different in the mutant DHPS: pyrophosphate lacks the hydrogen bond with N17 and the effects of electrostatic repulsion are minimized by a small simultaneous rotation of both pyrophosphate and sulphanilamide. Indeed, the sulphonamide NH group is now involved in a hydrogen bond with O19 on the pyrophosphate function (table 2). This molecular interaction remains stable for 300 ps (table 2). After this point, the modified flexibility of loop 2 alters the conformation of R58 and the corresponding molecular restraint exerted on sulphanilamide (tables 1 and 2). The greater conformational freedom experienced by sulphanilamide leads to the scission of the hydrogen bond between N10 and pyrophosphate (table 1). This missing interaction enables the movement of sulphanilamide away from 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate for the remainder of the simulation. Sulphanilamide establishes hydrogen bonds with the backbone carbonyl groups of T57 on loop 2 and E164 on loop 5 during its separation from 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate (table 2, figure 5). The expulsion of sulphanilamide from the binding site observed in the mutant S. pneumoniae MD simulation places the amino group of sulphanilamide more than 11 Å away from the C11 of 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate, thereby inhibiting rather than encouraging incipient chemical transformation.

5. Conclusion

Understanding the molecular basis of drug resistance is frequently non-trivial. Streptomyces pneumoniae DHPS is a challenging case because of subtle amino acid replacements located on highly flexible structural elements, the loops. We have addressed the problem of loop flexibility and its relevance to drug resistance mechanisms by means of extensive MD simulations starting from homology-built models of native and mutant S. pneumoniae DHPS in both unliganded and complexed forms. The present study provides an explicit molecular mechanism for drug resistance within S. pneumoniae DHPS. Inclusion of the drug in the MD simulations enables us to pinpoint amino acids that may be responsible for the altered response to sulpha drugs. The interplay between loops 1 and 2 observed in the native DHPS simulation is found to confine sulphanilamide in close proximity to its catalytic partner. The resulting molecular arrangement is favourably prepared for the ensuing enzymatic transformation, and we have been able to propose a possible reaction pathway on the basis of our simulations. Such detailed understanding of the molecular mechanism may also allow the design of inhibitors that are themselves less sensitive to the phenomenon of drug resistance.

We have observed different interactions between loops 1 and 2 for native and mutant forms of S. pneumoniae DHPS. Most notably, the interaction which is disrupted by the mutant is different for both apo and complexed forms. The simulations show that this is a direct result of the presence or absence of sulphanilamide.

During the simulation of the sulpha-resistant form of S. pneumoniae DHPS, sulphanilamide is unable to dock close to 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate but rather moves away from the catalytic site. Our simulation data indicate that the modified amino acids on the mutant S. pneumoniae DHPS studied here are directly responsible for the alteration in the proteins response to the drug, by firstly preventing it getting close to 6-hydroxymethyl-7, 8-dihydropteridine-pyrophosphate and then expelling it from the catalytic pocket.

The detailed molecular model for drug resistance within S. pneumoniae DHPS, which we have formulated here on the basis of large scale MD simulations, provides new insights into this complex problem. Our theoretical model could be tested by tailoring specific site-directed mutagenesis experiments and/or by the rational design of new and, hopefully, more successful synthetic inhibitors. We hope that someone will take up this challenge. We believe that a similar approach, combining homology modelling and large scale MD, may have wider applicability in addressing analogous problems of mutant based drug resistance, for example within the HIV proteases.


We thank EPSRC (UK) for providing access to an 816 processor Cray T3E at the Computer Services for Academic Research (CSAR) located in Manchester (UK) under RealityGrid grant GR/R67699, the NSF (US) for providing access to 3000 processor Compaq Alphaserver (LeMieux) located at Pittsburgh Supercomputing Centre (USA) under NSF PACI grant ASC030006P and HEFCE (UK) for funding the purchase of a Silicon Graphics Onyx2 machine under a JREI award. The PhD studentship of F.G. is funded by Queen Mary, University of London and that of P.W.F. by EPSRC.



View Abstract