The impact of Mycobacterium tuberculosis gyrB point mutations on 6-fluoroquinolones resistance profile: in silico mutagenesis and structure-based assessment

Abstract

Despite the efficiency of 6-fluoroquinolone (6-FQ) antibacterials in fighting tuberculosis (TB), the daily reports related to different forms of quinolone-caused “acquired resistance” in Mycobacterium tuberculosis are becoming rather frequent. Alongside the extensively reported mutations targeting predominantly the quinolone resistance-determining region (QRDR) of the gyrA subunit, some recent studies are pointing out the emergence of gyrB point mutations contributing to the M. tuberculosis resistance as well. To clarify the impact of gyrB alterations on 6-FQs resistance, in silico mutagenesis and structure-based methodology were proficiently employed. Three M. tuberculosis single point gyrB mutants (N473T_mod, T474P_mod, and E475V_mod) based on the recently available structural information were developed. The constructed mutant models were utilized as a starting point for performing molecular docking calculations on a set of 145 6-FQs with determined biological activity values, while their resistance profiles (identification of active/inactive 6-FQs) were evaluated relative to that of the wild-type model. This profiling methodology suggested the following order of resistance degree for our models (N473T_mod > T474P_mod > E475V_mod > 3K9F_mod) which was additionally confirmed by molecular docking of a set of pre-selected 48 combinatorially-generated 6-FQ hits. Furthermore, we identified several attractive substructural fragments that could aid the development of novel 6-FQ antibacterials with possible enhanced anti-mycobacterial activity against diverse M. tuberculosis gyrB mutant strains.

---

1. Introduction

Tuberculosis (TB) is one of the global inauspicious health problems. Promoted mainly by the pathogen microorganism Mycobacterium tuberculosis, it is still one of the dominant causes of death cases worldwide together with malaria and human immunodeficiency virus (HIV) infections.^1,2 The latest global TB reports by the World Health Organization (WHO, 2013) indicate a continuous growth of the incidence of this infectious disease with an estimated 8.6 new million TB cases and around 1.3 million deaths registered just in 2012.³

A well established therapeutic target in Mycobacteria is the DNA gyrase enzyme – an omnipresent, superior molecular nanomachine involved in the maintenance of the bacterial cell life through an outstanding control of the mycobacterial DNA topology. It belongs to the type II topoisomerase family of enzymes together with its paralogous form topoisomerase IV.^4,5 However, despite their high level of structural resemblance, these enzymes are involved in distinct intracellular functions – DNA unwinding during the replication process (uniquely controlled by the DNA gyrase) and DNA decatenation (distraction of identical units within the double helical DNA molecule – a process regulated by topoisomerase IV).⁶ While the bacterial genome in majority of the bacterial species usually encodes both type II topoisomerases (DNA gyrase and topoisomerase IV), M. tuberculosis is unusual in its exclusiveness of possessing only one type II topoisomerase enzyme – DNA gyrase.⁷ This molecular nanomachine supercoils the DNA molecule similarly as other gyrases (ATP-dependent catalysis), but simultaneously expresses an augmented relaxation, DNA cleavage, and decatenation activities (ATP-independent catalysis).^8,9 It constitutes of two cardinal subunits, gyrA and gyrB (parC and parE in topoisomerase IV) that together assemble a functional heart-shaped heterotetrameric complex A₂B₂ (C₂E₂ in topoisomerase IV). Both subunits constituting DNA gyrase are composed of two mutually coupled domains oriented one to another – the gyrA breakage-reunion domain (BRD) and gyrB Toprim domain that together form the DNA gyrase catalytic core which is directly involved in the breakage/reunion catalytic step (DNA replication and elongation) facilitated by gyrA-BRD as well as maintaining the helical topology of the double-stranded DNA molecule controlled by the gyrB-Toprim domain (Fig. 1).^10–12

Fig. 1 Structural organization of the M. tuberculosis DNA gyrase in complex with the DNA molecule and an intercalated 6-FQ ligand. (a) Schematic depiction of the entire DNA gyrase enzyme. (b) Surface representation of the DNA gyrase complex (face view (left), top view (middle), and side view (right)) colored by domains (gyrA-BRD in light blue and gyrB-Toprim in light pink), i.e., regions (QRDR-A in olive green and QRDR-B in violet) together with the gyrA (orange)/gyrB (green) hot spots implicated in the 6-FQs resistance. The amino acid residues forming the QBP are coming from both subunits, i.e., gyrA (colored in dark blue) and gyrB (colored in hot pink). (c) Close view of the QBP with an intercalated levofloxacin (LFX) ligand between the DNA base pairs (framed in (b)), representing the most frequently occurred quinolone-caused gyrA (orange) and gyrB (light green) mutations (numbered according to the CAB02426.1 numbering system).³⁶

Quinolone antibacterials and particularly their fluoro derivatives 6-fluoroquinolones (6-FQs) have proved their in vitro/in vivo activity as DNA gyrase inhibitors against M. tuberculosis for more than 40 years.^13,14 They fleetly inhibit the mycobacterial DNA gyrase through establishing a covalent complex between DNA molecule and the enzyme, resulting in a substantial collapse of the nascent DNA topology followed by replication/transcription breakdown, and finally bacterial cell destruction.^15–17 These drugs are still one of the most effective cures for treating TB in general^18,19 and nowadays probably a drug class of first choice in the treatment of multidrug-resistant tuberculosis (MDR-TB; a pathological condition commonly defined as a disease expressed with a bacterial resistance to at least isoniazide and rifampicin).^3,20 However, functional genetic and biochemical studies performed so far revealed several new forms of drug-induced “acquired resistance” in M. tuberculosis known as extensively drug-resistant tuberculosis (XDR-TB; caused by MDR mycobacterial strains that are additionally resistant to any known 6-FQ),^3,21 and even a form of “totally” drug-resistant tuberculosis (TDR-TB; defined as a TB form resistant to 4 first-line and 6 second-line drugs, i.e., a strain characterized as nearly impossible to treat) unfortunately still not recognized by WHO.²²

Resistance to 6-FQs in M. tuberculosis is substantially interceded by amino acid alterations constituting the so called quinolone-resistance determining region (QRDR) of both DNA gyrase subunits (QRDR-A and QRDR-B; Fig. 1) covering the quinolone-binding pocket (QBP).^12,23,24 In contrast to QRDR region of the gyrA subunit (QRDR-A; Fig. 1) that harbors the majority of the mutations known to be implicated in the 6-FQs resistance in M. tuberculosis (T80A, A90V, S91P, D94A, D94G, D94H, and D94Y; see Fig. 1 and Table 1),^12,25–32 currently little knowledge exists regarding the quinolone-caused amino acid substitutions situated in the QRDR region of the gyrB subunit (QRDR-B; Fig. 1).

Table 1 Summary of the most typical M. tuberculosis DNA gyrase (gyrA/gyrB) point mutations extensively elaborated in the last six years

ID	gyrA Mutations	Reference	ID	gyrB Mutations	Reference
a gyrA point mutations known to be strongly implicated in 6-FQs resistance.b gyrB point mutations known to be strongly implicated in 6-FQs resistance.c Mutated residues forming the so called gyrB hot spot region.d gyrB point mutations not implicated in 6-FQs resistance.
1	P8A	27	1	D472H	12 and 25
2	A74S	29 and 32	2	D473N^d	33
3	T80A^a	12, 25 and 27	3	P478A^d	33
4	G88C	29 and 30	4	R485H^d	33
5	A90G	25 and 27	5	S486F^d	33
6	A90V^a	12 and 25–32	6	D500A^b	12 and 31
7	S91P^a	28, 30 and 31	7	D500N	31 and 32
8	D94A^a	12, 25 and 27–31	8	D500H	32
9	D94F	28	9	A506G^d	33
10	D94G^a	12 and 25–32	10	N510D	25
11	D94H^a	25, 27 and 29–31	11	N533T	27
12	D94N	25–29 and 31	12	N538D	12 and 31–33
13	D94Y^a	12 and 25–31	13	N538T^b^,^c	12 and 31
			14	T539N	31
			15	T539P^b^,^c	12 and 31
			16	E540D	31 and 32
			17	E540V^b^,^c	12, 31, 32 and 34
			18	A547V^d	33
			19	G549D	28
			20	G551R^d	28 and 33
			21	G559A^d	33

---

Among the total 21 gyrB point mutations known so far, only four could be distinguished as potentially dangerous (D500A, N538T, T539P, and E540V; see Fig. 1 and Table 1)^31–35 and therefore deserve a special attention. According to a recently performed three-dimensional structural investigation,¹² these mutations are settled in the core of the QBP (Fig. 1); two of them are located within the QRDR-B (D500A and N538T), while the other two are outside QRDR-B region (T539P and E540V).³¹ Except D500 residue that is spatially bit away and also associated with low resistance levels when substituted to alanine, the rest three residues (N538, T539, and E540) assemble a so called gyrB hot spot region (according to the CAB02426.1 numbering system)³⁶ strongly involved in the 6-FQs “acquired resistance” (Table 1).^12,31,35 Therefore, the efforts today are mainly directed toward an early clarification of their role in 6-FQs resistance profile, and consequently a rational optimization of the existing 6-FQs or even design of novel 6-FQ antibacterials through the utilization of advanced structure-based drug design methodologies.^37–39

The present study introduces a rational structure-based approach for proficient identification of novel 6-FQ hits as potential DNA gyrase inhibitors against three virtually-generated M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) utilizing the latest gyrB mutation data (N538T, T539P, and E540V).^31–36 Our recently established and validated M. tuberculosis-DNA gyrase protein homology model in complex with DNA and intercalated 6-FQ (3K9F_mod)³⁹ firmly based on the lately disclosed experimental findings,¹² which QBP emulates the one expressed in the wild-type H37Rv M. tuberculosis strain was employed as a template structure for performing three single-point in silico mutagenesis experiments within the recently determined gyrB hot spot region.³¹ The constructed gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) were initially used to perform molecular docking calculations on a set of 145 6-FQs with experimentally-determined biological activity values, while their resistance profiles (identification of active/inactive 6-FQs) were assessed relative to that of the wild-type model 3K9F_mod³⁹ utilizing two consecutively-coupled virtual screening (VS) validation experiments. The obtained resistance profiles for our models were additionally confirmed in a molecular docking and VS assay using a combinatorial set of 48 pre-selected 6-FQ hits.³⁹ Moreover, this study propounds some tempting, synthetically feasible 6-FQ substructural fragments which might promote the development of novel 6-FQ derivatives as potential DNA gyrase inhibitors against diverse M. tuberculosis gyrB mutant strains.

2. Methods

2.1 6-FQs screening libraries

Two screening libraries of structurally-similar 6-FQ compounds (termed as ExpLib and CombiLibHits, respectively, available as ESI†) were employed to establish and investigate the 6-FQs resistance profile for our constructed M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod). Both libraries (their origin, assembly, and previous exploitation) are extensively elaborated in our previous studies.^38–40 The ExpLib library of total 145 6-FQs with known biological activity values – minimal inhibitory concentrations (MIC_exp [μg mL⁻¹]; of which 114 active 6-FQs with MIC_exp ≤ 1.00 μg mL⁻¹, and 31 inactive molecules with MIC_exp > 1.00 μg mL⁻¹)⁴¹ was used for validation purposes of the constructed mutant models – assessment of their discriminatory performances (evaluation of the models resistance profile through examination of their capability for identification of active/inactive 6-FQs) as previously described.^38,39 CombiLibHits library constructed of total 48 combinatorially-generated 6-FQ analogs previously selected as promising M. tuberculosis DNA gyrase inhibitors,³⁹ which biological activity values (MIC_pred-combi [μg mL⁻¹]) were estimated by implementation of an artificial neural networks (ANNs) predictive model trained and validated on experimental inhibitory data (ExpLib),⁴⁰ was employed to re-confirm the validation results as well as to expose new possible 6-FQs structure–activity relationship (SAR) recommendations.

2.2 In silico mutagenesis and construction of the M. tuberculosis gyrB mutant models

Although both M. tuberculosis DNA gyrase subunits (gyrA and gyrB) have been solved separately by some recent crystallographic studies,^12,42,43 unfortunately a high resolution crystal structure of the entire M. tuberculosis DNA gyrase holoenzyme (A₂B₂ heterotetramer) in complex with the DNA molecule and intercalated 6-FQ ligand still remains an indecipherable issue. To span this problem as well as to approach it as much as possible to the real situation, homology modeling experiments proved to be particularly important.^38,39 Within this scope and most importantly to decipher the impact of gyrB point mutations on 6-FQs resistant profile, we opted to construct three single point M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) taking into account the latterly confirmed gyrB point mutations (N538T, T539P, and E540V) implicated in the 6-FQs resistance.^12,31,35,36

Our recently established and validated M. tuberculosis-DNA gyrase protein homology model³⁹ previously named as 3K9F_mod – a protein model based on the topoisomerase IV crystal structure in complex with DNA and levofloxacin (LFX) ligand (PDB ID: 3K9F)⁴⁴ as well as the recently solved M. tuberculosis gyrA-BRD (PDB ID: 3IFZ)¹² and gyrB-Toprim domain (PDB ID: 3M4I)¹² emulating the DNA gyrase enzyme present in the wild-type H37Rv M. tuberculosis strain was exploited as a template structure for in silico mutagenesis within the aforementioned gyrB hot spot region.³¹ PyMol's⁴⁵ integrated mutagenesis engine was used for in silico amino acids alteration (N473T, T474P, and E475V in our model,³⁹ which are relevant to N499T, T500P, and E501V in,¹² i.e., N538T, T539P, and E540V in ref. 31, 34 and 36; see Table 2 and ESI Fig. S1†), while Swiss-Pdb Viewer⁴⁶ was utilized for in vacuo energy minimization (using GROMOS96 43B1 parameter set)⁴⁷ of each gyrB mutant model.

Table 2 M. tuberculosis gyrB hot spot region – point mutations and their different numbering

ID	Affected residue	Code	Mutation/numbering
			This study^a	Ref. 12	Ref. 31, 34 and 36
a The interchanged amino acid residues constituting the gyrB hot spot region in our previously published M. tuberculosis-DNA gyrase protein homology model 3K9Fmod.^39
1	Asparagine	N	N473T	N499T	N538T
2	Threonine	T	T474P	T500P	T539P
3	Glutamic acid	E	E475V	E501V	E540V

---

The constructed M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod; see ESI Fig. S1†) were subsequently used as a starting point for performing molecular docking calculations.

2.3 Molecular docking calculations

The molecular docking calculations on both 6-FQ compound libraries (ExpLib and CombiLibHits, respectively) within QBPs of the constructed gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) were performed by using GOLD docking environment.⁴⁸ The entire docking procedure (including all settings and required technical parameters) for each mutant model was accomplished in exactly the same manner as for the wild-type 3K9F_mod model (population size = 100, selection pressure = 1.1, number of operations = 100 000, number of islands = 5, niche size = 2, migrate = 10, mutate = 95, cross-over = 95) by running the genetic algorithm (GA) in 10-fold iterative mode per ligand molecule, while experimental coordinates of the co-crystallized LFX ligand and its amino acids coverage were used to define 6-FQs binding pocket (cavity radius of 12.5 Å) as previously described.³⁹

The quality of the constructed mutant models was initially confirmed by comparing the calculated dock poses for the co-crystallized LFX ligand with its natively present spatial conformation^39,44 for each model separately (Table 3).⁴⁹ Moreover, the water molecule present within the QBP which was previously described as an essential co-factor for additional stabilization of the protein–FQ–DNA complex was again introduced in the docking calculations as previously proposed (see Fig. 2).^39,44 For the purpose of comparison of the obtained results with that previously reported,³⁹ the GOLDScore Fitness (GSF) function was used as a scoring function for evaluation of the 6-FQs binding affinity.⁴⁸

Table 3 RMSD values in Angstrom units (Å) obtained by alignment (heavy atoms) of each LFX calculated dock pose and its natively present co-crystallized conformation³⁹ within the QBP of each M. tuberculosis gyrB mutant model (Fig. 2). Dock poses with minimum calculated RMSD values are shown in bold

Model	3K9Fmod^a			N473Tmod			T474Pmod			E475Vmod
a Ligand reproduction results for our previously published M. tuberculosis-DNA gyrase protein homology model 3K9Fmod.^39
Dock pose	Dock 1	Dock 2	Dock 3	Dock 1	Dock 2	Dock 3	Dock 1	Dock 2	Dock 3	Dock 1	Dock 2	Dock 3
RMSD (Å)	1.0568	1.1217	1.0277	1.3005	1.0873	1.0983	1.2689	1.3138	1.0896	1.2689	1.1007	1.2468

---

Fig. 2 Spatial comparison between natively present LFX conformation and its calculated dock poses within the QBP of: (a) wild-type 3K9F_mod model (our previous study),³⁹ (b) N473T_mod model, (c) T474P_mod model, and (d) E475V_mod model. The co-crystallized LFX conformation is represented in dark green, while its reproduced dock poses (for each model separately; see Table 3) are depicted in solid red. M. tuberculosis gyrB point mutations affecting α2-loop (T473), i.e., α2-helix region (P474 and V475) are colored in light green (b–d), while the unaffected residues in hot pink (a–d).

2.4 Validation of the constructed M. tuberculosis gyrB mutant models

Analogously to our previous study,³⁹ the resistance profile for the constructed M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) could be defined as a model's capability for correct discrimination of active 6-FQ compounds among a pool of inactive molecules.⁵⁰ To achieve this, two consecutively-coupled VS validation experiments utilizing the calculated docking data for the set of 145 6-FQ compounds (ExpLib) with experimentally-determined biological activity values (MIC_exp [μg mL⁻¹]) were used. All details concerning the validation methodologies used (settings, theoretical, and mathematical background) are broadly elaborated in our previous study,³⁹ and therefore we give here just a short abridgement.

2.4.1 Receiver operating characteristic (ROC) methodology. The Receiver Operating Characteristic (ROC)⁵¹ as a recommended methodology for evaluation of the VS performances for established protein homology models was used as a primal validation experiment (determination of the mutant models resistance profile) as previously described.³⁹ vROCS tool^52,53 was employed for automated similarity screening of a set of 114 active 6-FQs (constituting the ExpLib library) and an in-house developed set of 13.990 artificial decoys³⁹ against the LFX calculated dock poses (for each mutant model separately) with minimum RMSD (Å) as template structures (queries; see Table 3, the dock poses shown in bold). The generated ROC curves for each gyrB mutant model were used for calculation of the area under the ROC curve (ROC AUC) and the corresponding early enrichment parameters (at 0.5, 1.0, and 2.0% using a framework of ±95% confidence intervals), while the profiling of their resistance levels was performed relative to the wild-type 3K9F_mod model profile.³⁹

2.4.2 Boolean-based clustering. Similarly to our previous study, the thorough assessment of the binding affinity for the investigated 6-FQs within QBP of our developed M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod), and consequently the determination of their resistance profile was additionally performed through utilization of our novel Boolean-based [T/F (true/false)] clustering algorithm.³⁹ The algorithm previously explained in details³⁹ is organized in three consecutive VS levels:

• Level 1: geometry properties assessment (spatial examination of each 6-FQ calculated dock pose relative to the natively present co-crystallized LFX conformation and building a cluster of (T)-signed dock poses).

• Level 2: score-based clustering ((T)-signing of the obtained Level 1 (T)-signed poses with GSF ≥ 80 and building a new cluster of highly scored (T)-signed poses).

• Level 3: activity-based clustering ((T)-signing of the previously extracted Level 2 (T)-signed highly scored poses with MIC ≤ 0.05 μg mL⁻¹, and finally building a cluster of (T)-signed most “active” 6-FQ hits).

As previously described,^38,39 the Level 1 was exclusively employed for validation purposes using the ExpLib library. However, contrary to our previous investigation,³⁹ here we want to stress that CombiLibHits library was previously obtained at the end of our Boolean-based clustering algorithm, and therefore here it was used to re-confirm all validations (Level 1 and Level 2) and previous findings as well as to adduce some novel SAR guidelines.

3. Results and discussion

3.1 Pre-validation of the constructed M. tuberculosis gyrB mutant models

Structure-based VS success using an in silico assembled protein homology model is usually dependent on the quality of the model itself; therefore its proper validation is mandatory. Taking this into account, a pre-validation protocol based on reproduction of the experimentally-determined LFX conformation^39,44 present within the QBP of each constructed M. tuberculosis gyrB mutant model (N473T_mod, T474P_mod, and E475V_mod, respectively) was performed as described in Methods section. Table 3 summarizes the initial validation results for our gyrB mutant models based on the RMSD (Å) values calculated between each computed dock pose and its corresponding experimental LFX conformation (Fig. 2).

As demonstrated in Table 3, all three gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) showed satisfactory performances with calculated RMSD values below 2.0 Å (Fig. 2).⁵⁴ Comparing to the wild-type 3K9F_mod model which calculated dock poses are almost neatly aligned with the co-crystallized LFX conformation (Fig. 2a),³⁹ the predicted LFX dock poses for the mutant models are slightly displaced to the left (Fig. 2b–d). This effect, probably caused as a consequence of the substituted amino acid residues is also congruent with the calculated RMSD values (Table 3); as expected, all three mutant models (N473T_mod, T474P_mod, and E475V_mod) reflected somewhat higher RMSD values for the predicted LFX binding poses relative to the wild-type 3K9F_mod model – a useful information pertaining the resistance nature of our mutant models. However, to thoroughly assess their resistance profile, robust VS validation protocols were implemented.

3.2 In silico profiling of M. tuberculosis gyrB mutant models resistance degree

The in silico profiling of the resistance degree for our constructed M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) was performed through utilization of two consecutively-coupled VS validation protocols as described in Methods section. At the beginning, each gyrB mutant model was subjected to a quantitative evaluation (ROC assessment;⁵¹ see Methods section) of the discriminatory performances (identification of active/inactive 6-FQs from the ExpLib library).⁴¹ Table 4 outlines the ROC validation results for each gyrB mutant model (N473T_mod, T474P_mod, and E475V_mod) obtained from the generated ROC curves (Fig. 3).

Table 4 Statistical parameters describing the resistance profile for our examined M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod): ROC AUC (area under the ROC curve), EF – enrichment factor (early enrichment parameters at 0.5%, 1.0%, and 2.0% by using ±95% confidence interval), and their corresponding p-values (probability to obtain better VS performances for a given model, assuming that the null hypothesis is true). All the comparisons are performed relative to the wild-type 3K9F_mod model³⁹

Model	3K9Fmod	N473Tmod	p	T474Pmod	p	E475Vmod	p
ROC AUC	0.831 [0.78, 0.88]	0.792 [0.74, 0.84]	0.137	0.796 [0.75, 0.84]	0.161	0.799 [0.75, 0.85]	0.184
EF (0.5%)	81.38 [59.67, 103.1]	58.72 [40.35, 77.69]	0.055	54.92 [35.39, 75.86]	0.038	45.43 [28.57, 64.44]	0.007
EF (1.0%)	47.53 [38.32, 57.14]	32.14 [23.15, 41.28]	0.008	32.72 [23.33, 41.44]	0.011	29.20 [20.72, 37.81]	0.002
EF (2.0%)	26.58 [21.74, 31.25]	20.72 [15.66, 25.66]	0.042	19.09 [14.29, 23.81]	0.013	17.33 [12.73, 21.85]	0.002

---

Fig. 3 ROC plot displaying the resistance profile (Table 4) expressed as the fraction of active/decoy found molecules for our models: wild-type 3K9F_mod (solid blue curve), N473T_mod (solid red curve), T474P_mod (solid orange curve), and E475V_mod (solid green curve). The black diagonal line (a line of no discrimination) is depicting the randomly distributed data (RDD [0.5]).

As displayed in Table 4, the unaltered wild-type 3K9F_mod model has again the best discriminatory performances (identification of active/inactive 6-FQs) comparing to the 3K9F_mod-generated gyrB mutant models as supported by the values for the calculated area under the ROC curves (ROC AUC_[3K9Fmod] = 0.831; see Fig. 3); between the mutant models, N473T_mod could be distinguished as a poorest one as denoted by the lowest ROC AUC value (ROC AUC_[N473Tmod] = 0.792), T474P_mod has somehow moderate discriminatory performances (ROC AUC_[T474Pmod] = 0.796), whereas the highest discriminatory performances could be ascribed for the E475V_mod (ROC AUC_[E475Vmod] = 0.799). Inversely, N473T_mod can be described as a model with a highest resistance degree (a model with a lowest capability for identification of active ExpLib 6-FQs), while E475V_mod model has apparently the lowest resistance degree (a model with a highest capability for identification of active ExpLib 6-FQs), which results are also grounded by the p-values for the models selective probability as well as the early enrichment factors calculated relative to the wild-type 3K9F_mod model (Table 4).

As showed in Table 4, all calculated p-values for our gyrB mutant models are below 0.5 (p-value_[N473Tmod] = 0.137, p-value_[T474Pmod] = 0.161, and p-value_[E475Vmod] = 0.184, respectively); consequently, the probability for our mutant models to be more selective than the wild-type 3K9F_mod model is very low. Although, no directly comparable (mainly as a consequence of the significant differences between ROC/RMSD similarity metrics employed), these results are strongly corroborated with the overall outcome of our previously performed RMSD-based initial validation assessment (Table 3 and Fig. 2), and therefore one could suggest the following order of resistance degree for our models (N473T_mod > T474P_mod > E475V_mod > 3K9F_mod).

The resistance profile for the M. tuberculosis gyrB mutant models obtained so far, was additionally confirmed by in-depth visual inspection (see Methods section; Level 1: geometry properties assessment) of the 145 docking-calculated ExpLib 6-FQ binding poses relative to the experimental LFX conformation (for each model separately), through utilization of our Boolean-based [T/F (true/false)] clustering algorithm (see ESI Table S1†).³⁹ As demonstrated there, the Boolean-based (T/F) assessment of the ExpLib compounds docked into the wild-type 3K9F_mod model identified 84 (T)-signed compounds out of total 145 investigated 6-FQs as correctly positioned (see ESI Table S1a†) while our mutant models (N473T_mod, T474P_mod, and E475V_mod) selected almost twice as low number of compounds (40, 42, and 44, respectively) as correctly positioned (see ESI Tables S1b–d†). This result clearly indicates the resistance nature of the M. tuberculosis gyrB mutants in general, but also corroborated our previously obtained validation results.

Moreover, the examination of the (T/F)-signed ExpLib compounds selected as correctly/incorrectly positioned within the QBP of each investigated model (3K9F_mod: 84/61, N473T_mod: 40/105, T474P_mod: 42/103, and E475V_mod: 44/101, respectively), identified a total of 27 6-FQ compounds with biological activity values in the range (0.0016 ≤ MIC_exp [μg mL⁻¹] ≤ 2.000) which are known to be widely used in TB therapy (Table 5). Interestingly, among them only 7 ExpLib compounds (difloxacin, sarafloxacin, clinafloxacin, temafloxacin, moxifloxacin, N-propylciprofloxacin, and balofloxacin) were identified as not just correctly positioned compared to the experimental LFX conformation (see ESI Table S1†), but also active as determined by their biological activity values (MIC_exp ≤ 1.000 μg mL⁻¹) – a result which additionally allowed the estimation of the 6-FQs resistance, i.e., susceptibility profile for our investigated models (Table 5).

Table 5 In silico estimation of the susceptibility (S)/resistance (R) profile for our investigated M. tuberculosis gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) vs. the wild-type 3K9F_mod model,³⁹ for the subset of 27 known ExpLib 6-FQs broadly used in TB therapy. The susceptibility (S) and resistance (R) classification of the models was based on the Boolean-based signing of the ExpLib compounds on (T)-susceptible and (F)-resistant model, respectively (see ESI Table S1†). The ExpLib compounds identified as active against all four investigated models are shown in bold

ID	ExpLib 6-FQ	Name	MICexp [μg mL^−1]	Models
				3K9Fmod	N473Tmod	T474Pmod	E475Vmod
1	Structure002	Ofloxacin	0.0115	R	R	R	S
2	Structure003	Norfloxacin	0.3000	S	R	S	S
3	Structure004	Ciprofloxacin	0.0100	S	R	S	S
4	Structure005	Enoxacin	0.3000	R	R	R	R
5	Structure006	Levofloxacin	0.1200	R	R	R	R
6	Structure012	Difloxacin	0.5000	S	S	S	S
7	Structure013	Ibafloxacin	1.6000	R	R	R	R
8	Structure014	Merafloxacin	0.5000	R	R	R	R
9	Structure015	Fleroxacin	0.5000	R	R	S	R
10	Structure016	Pirfloxacin	2.0000	R	R	R	R
11	Structure017	Sarafloxacin	0.5000	S	S	S	S
12	Structure018	Lomefloxacin	0.5000	S	R	R	R
13	Structure020	Clinafloxacin	0.0100	S	S	S	S
14	Structure021	Temafloxacin	0.0310	S	S	S	S
15	Structure025	Enrofloxacin	0.1250	R	S	S	R
16	Structure026	Pefloxacin	0.3000	R	R	S	R
17	Structure027	Grepafloxacin	0.1200	R	R	R	R
18	Structure037	Gemifloxacin	0.1250	S	S	R	S
19	Structure046	Gatifloxacin	0.0300	R	S	S	S
20	Structure047	N-Isopropylciprofloxacin	0.1250	R	R	S	R
21	Structure048	N-Methylciprofloxacin	0.1250	R	S	S	S
22	Structure050	Ciprofloxacin DR	0.0016	S	R	R	R
23	Structure060	Moxifloxacin	0.0300	S	S	S	S
24	Structure062	N-Propylciprofloxacin	0.1250	S	S	S	S
25	Structure063	N-Benzylciprofloxacin	0.5000	S	R	R	R
26	Structure076	Balofloxacin	0.1250	S	S	S	S
27	Structure081	Difloxacin	0.3900	S	R	S	R

---

The analysis of the substructural fragments attached at R₇ and R₈-position (known to establish direct hydrogen-bonding (HB) interaction with the amino acid residues constituting the gyrB hot spot region)³⁴ of the selected 6-FQ compounds from the ExpLib library (see Table 5; the compounds shown in bold), pointed out fragments with at least one HB acceptor atom (e.g., piperazinyl, 3-methylpiperazinyl, 4-methylpiperazinyl, 4-propylpiperazinyl, 3-aminopyrrolidinyl, 3-methaminopiperidinyl, etc.). These findings are in a strong agreement with the established SAR guidelines for 6-FQ antibacterials.⁵⁵ As the majority of these compounds belong to the class of ciprofloxacin (CIP)- and moxifloxacin (MOX)-like 6-FQ derivatives, these results once again underline their effectiveness for future TB treatments.³¹

3.3 CIP and MOX derivatives as potential novel inhibitors against M. tuberculosis gyrB mutants

To re-confirm our validation results as well as to establish novel possible SAR guidelines, our gyrB mutant models were further screened against a mixed set of 48 combinatorially-generated CIP and MOX compounds (CombiLibHits) previously selected as promising novel DNA gyrase inhibitors against wild-type M. tuberculosis strains.³⁹ For these purposes, the Boolean-based (T/F) clustering algorithm³⁹ (see Methods section) was again used as a VS platform for thorough visual assessment of the CombiLibHits binding poses obtained by docking calculations within the QBP of each constructed gyrB mutant model.

Similarly to the ExpLib compounds, the Boolean-based (T/F) spatial examination of the CombiLibHits compounds (see Methods section; Level 1: geometry properties assessment) identified 23, 31, and 32 (T)-signed compounds, respectively, as spatially well positioned relative to the experimental LFX conformation present within the QBP of each gyrB mutant model (see ESI Tables S2a–c†).

The structural analysis of the first level (T)-signed combinatorial compounds showed that all three gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod) identified a mix of both 6-FQ classes (CIP and MOX structural analogs) constituting the CombiLibHits library. Analogously to our previous study,³⁹ all (T)-signed combinatorial compounds identified during the first level of the Boolean-based (T/F) assessment, were isolated as separate clusters and used in the second level of the CombiLibHits VS assessment (see Methods section; Level 2: score-based clustering). In this routine, a pre-defined GSF limit of (GSF ≥ 80) was used to identify all top-scored clusters, while only highly-scored poses (combinatorial hits) within each top-scored cluster were extracted; 23, 30, and 32 combinatorial compounds for each gyrB mutant model, respectively, were identified as highly-scored hits (Fig. 4) with calculated GSF values above 80 (see ESI Tables S3a–c;† CombiLibHits dock poses highlighted in dark yellow).

Fig. 4 Structure-based predicted binding conformations of the combinatorially-generated 6-FQs (CombiLibHits) within the QBP of each constructed gyrB mutant model: (a) N473T_mod (23 hits), (b) T474P_mod (30 hits), and (c) E475V_mod (32 hits). The mutated gyrB residues (T473, P474, and V475; stick representation) are represented in light green, while the native (unaffected) residues (N473, T474, and E475; stick representation) in hot pink. The water molecule used in the docking calculations is depicted as solid red sphere, the DNA molecule in solid orange color, while the selected CombiLibHits binding poses in native atomic colors.

Moreover, all extracted combinatorial compounds have estimated biological activity values (MIC_pred-combi ≤ 1.00 μg mL⁻¹),^38,39 demonstrating actively predicted 6-FQ structural analogs. Strongly correlated with our previously performed validations, this result is once again highlighting the proposed order of resistance degree for our investigated M. tuberculosis gyrB mutant models (N473T_mod > T474P_mod > E475V_mod > 3K9F_mod).

The substructural examination of the fragments attached at R₇-position of the extracted combinatorial hits (23, 30, and 32 compounds, respectively), revealed the most frequently occurring fragments (Table 6): 6-methyl-4H-furo[3,2-c]pyran-4-one (005), 3-methyl-3,4-dihydro-2H-benzo[b][1,4]oxazine (033), 6-chloro-3-methyl-[1,2,4]triazolo[4,3-b] pyridazine (057), 7-methylquinolin-8-amine (073), 1-methylpiperidine-2,3-dione (096), 1,3-dimethyl-1H-pyrazol-5(4H)-one (102), 3-methyl-1H-pyrazolo[3,4-b]pyridine (116), 2-(chloromethyl)-6-methylpyrimidine-4-ol (137), and 1-methyl-1H-pyrazole (176).

Table 6 The most representative CIP and MOX combinatorial analogs with best predicted biological activity values (MIC_pred-combi [μg mL⁻¹]) selected by VS of CombiLibHits library within the QBP of the constructed gyrB mutant models. The proposed R₇-substructural fragments are represented in bold bonds. The combinatorial compounds identified as well positioned by a model are marked with (+), those rejected by a model with (−), while those 6-FQs identified as well positioned by all three mutant models are depicted with bold IDs

ID	Chemical structure	R1	R7	MICpred-combi [μg mL^−1]	Model
					N473Tmod	T474Pmod	E475Vmod
1		095	057	0.0021	+	+	+
2		100	073	0.0329	+	+	+
3		024	176	0.0250	−	−	+
4		036	096	0.0250	−	+	+
5		028	102	0.0130	−	+	+
6		039	005	0.0499	+	+	+
7		060	033	0.0401	+	+	+
8		028	116	0.0130	+	+	−
9		016	137	0.0300	+	+	+

---

As demonstrated in Table 6, the majority of these fragments (the R₇ building-blocks represented with bold bonds) are small aromatic N-heterocycles with molecular weight between 80 and 170 g mol⁻¹, containing more than two HB acceptor atoms on average – a result that not only supports the existing SAR knowledge for 6-FQ antibacterials,⁵⁵ but also increase the probability for establishing more than one HB interaction with the surrounding gyrB amino acid residues.

4. Conclusions

The present study aimed to investigate the impact of M. tuberculosis gyrB point mutations on the resistance profile of 6-FQ antibacterials broadly used for TB therapy. For these purposes, the recently confirmed gyrB mutation data (N538T, T539P, and E540V, known to be strongly implicated in the 6-FQs resistance)^31–35 assembling the so-called gyrB hot spot region³⁶ situated within the QBP of the M. tuberculosis DNA gyrase enzyme were employed. In silico mutagenesis experiments were performed for construction of three gyrB mutant models (N473T_mod, T474P_mod, and E475V_mod, relevant to N538T, T539P, and E540V according to the CAB02426.1 numbering system) based on our recently established and validated DNA gyrase protein homology model (3K9F_mod) which QBP emulates the one present in the wild-type M. tuberculosis strains.³⁹ The constructed gyrB mutant models were subsequently used as a starting point for performing structure-based and VS assessments, while their outcomes were analyzed relative to that of the wild-type 3K9F_mod model.³⁹

The validation of the constructed gyrB mutant models (determination of their resistance profile) was accomplished through implementation of two consecutively-coupled VS validation protocols (ROC methodology and Boolean-based visual assessment) using the data obtained by structure-based calculation of a set of 145 6-FQs with experimentally-determined biological activity values (ExpLib, MIC_exp [μg mL⁻¹]).⁴¹ The ROC methodology suggested the following order of resistance degree (identification of known active/inactive 6-FQs) for our models (N473T_mod > T474P_mod > E475V_mod > 3K9F_mod), which result was also confirmed by thorough visual assessment (Boolean-based [T/F] clustering) of the calculated ExpLib binding poses within the QBP of each investigated mutant model (see ESI Table S1†). The analysis of the substructural fragments attached at R₇-position (which are known to establish direct HB interaction with the amino acid residues forming the gyrB hot spot region) of the spatially well positioned ExpLib conformations, identified several small fragments (e.g. piperazinyl, pyrrolidinyl, and piperidinyl building-blocks) containing at least one HB acceptor atom – an information undoubtedly congruent with the existing SAR knowledge for 6-FQ antibacterials.⁵⁵ Moreover, the majority of the correctly predicted ExpLib 6-FQs belong to the class of CIP- and MOX-like derivatives; this outcome once again underlines the usefulness of these 6-FQs for future improvements of the existing TB therapy.³¹

Finally, the established resistance profile for the investigated mutant models (N473T_mod > T474P_mod > E475V_mod > 3K9F_mod) was additionally confirmed by in-depth Boolean-based visual assessment of the calculated binding poses for a mixed set of 48 combinatorially-generated CIP and MOX structural analogs (CombiLibHits) previously selected as promising M. tuberculosis DNA gyrase inhibitors.³⁹ Again, the substructural inspection of the R₇-attached fragments for the correctly predicted combinatorial compounds (see ESI Table S3†), revealed several novel building-blocks (see Table 6; fragments depicted in bold bonds), mainly aromatic N-heterocyclic systems with molecular weight in the range between 80 and 170 g mol⁻¹ that contain approximately two or more HB acceptor atoms. These substructural fragments not only support the existing 6-FQs SAR,⁵⁵ but also increase the probability for the ligand to establish more HB interactions with the surrounding gyrB residues and provide a good nesting of the entire ligand conformation within QBP (Fig. 4).

Acknowledgements

Authors thank Agency of Research of R. Slovenia (ARRS) for the financial support through Grants P1-0017 (to M.N. and N.M.) and P1-0012 (to T.S.).

Notes and references

1. C. Dye, S. Scheele, P. Dolin, V. Pathania and M. C. Raviglione, J. Am. Med. Assoc., 1999, 282, 677–686.
2. T. R. Frieden, T. R. Sterling, S. S. Munsiff, C. J. Watt and C. Dye, Lancet, 2003, 362, 887–889.
3. WHO, Global tuberculosis report 2013, 2013, http://www.who.int/iris/bitstream/10665/91355/1/9789241564656_eng.pdf?ua=1, accessed November 2014.
4. J. J. Champoux, Annu. Rev. Biochem., 2001, 70, 369–413.
5. P. Forterre, S. Gribaldo, D. Gadelle and M. C. Serre, Biochimie, 2007, 89(4), 427–446.
6. C. Levine, H. Hiasa and K. J. Marians, Biochim. Biophys. Acta, 1998, 1400(1–3), 29–43.
7. S. T. Cole, R. Brosch and J. Parkhill, et al., Nature, 1998, 393(6685), 537–544.
8. A. Aubry, L. M. Fisher, V. Jarlier and E. Cambau, Biochem. Biophys. Res. Commun., 2006, 348(1), 158–165.
9. F. Collin, S. Karkare and A. Maxwell, Appl. Microbiol. Biotechnol., 2011, 92(3), 479–497.
10. A. J. Schoeffer and J. M. Berger, Q. Rev. Biophys., 2008, 41(1), 41–101.
11. J. Piton, S. Matrat and S. Petrella, et al., Acta Crystallogr., Sect. F: Struct. Biol. Cryst. Commun., 2009, 65, 1182–1186.
12. J. Piton, S. Petrella and M. Delarue, et al., PLoS One, 2010, 5(8), e12245.
13. G. Y. Lesher, E. J. Froelich and M. D. Gruett, et al., J. Med. Chem., 1962, 5(5), 1063–1065.
14. S. Heeb, M. P. Fletcher and S. R. Chhabra, et al., FEMS Microbiol. Rev., 2011, 35(2), 247–274.
15. M. Gellert, K. Mizuuchi and M. H. O'Dea, et al., Proc. Natl. Acad. Sci. U. S. A., 1977, 74(11), 4772–4776.
16. A. Sugino, C. L. Peebles, K. N. Kreuzer and N. R. Cozzarelli, Proc. Natl. Acad. Sci. U. S. A., 1977, 74(11), 4767–4771.
17. K. Drlica and M. Malik, Curr. Top. Med. Chem., 2003, 3(3), 249–282.
18. K. Drlica, Curr. Opin. Microbiol., 1999, 2(5), 504–508.
19. Y. L. Janin, Bioorg. Med. Chem., 2007, 15(7), 2479–2513.
20. S. K. Sharma and A. Mohan, Indian J. Med. Res., 2004, 120(4), 354–376.
21. A. Jain and P. Dixit, J. Biosci., 2008, 33(4), 605–616.
22. A. Slomski, JAMA, 2013, 309(11), 1097–1098.
23. H. R. van Doorn, D. D. An and M. D. de Jong, et al., Int. J. Tuberc. Lung Dis., 2008, 12(7), 736–742.
24. D. A. Duong, T. H. Nguyen and T. N. Nguyen, et al., Antimicrob. Agents Chemother., 2009, 53(11), 4835–4839.
25. A. Aubry, N. Veziris and E. Cambau, et al., Antimicrob. Agents Chemother., 2006, 50(1), 104–112.
26. R. Johnson, E. M. Streicher and G. E. Louw, et al., Curr. Issues Mol. Biol., 2006, 8(2), 97–112.
27. A. V. Groll, A. Martin and P. Jureen, et al., Antimicrob. Agents Chemother., 2009, 53(10), 4498–4500.
28. Z. Cui, J. Wang and J. Lu, et al., BMC Infect. Dis., 2011, 11(78), 1–5.
29. R. W. T. Lau, P.-L. Ho and R. Y. T. Kao, et al., Antimicrob. Agents Chemother., 2011, 55(2), 608–614.
30. J.-I. Sekiguchi, A. Disratthakit, S. Maeda and N. Doi, Antimicrob. Agents Chemother., 2011, 55(8), 3958–3960.
31. A. Pantel, S. Petrella and N. Veziris, et al., Antimicrob. Agents Chemother., 2012, 56(4), 1990–1996.
32. R. Devasia, A. Blackman and S. Eden, et al., Antimicrob. Agents Chemother., 2012, 50(4), 1390–1396.
33. A. Pantel, S. Petrella and S. Matrat, et al., Antimicrob. Agents Chemother., 2011, 55(10), 4524–4529.
34. H. Kim, C. Nakajima and K. Yokoyama, et al., Antimicrob. Agents Chemother., 2011, 55(8), 3661–3667.
35. S. Malik, M. Willby and D. Sikes, et al., PLoS One, 2012, 7(6), e39754.
36. H. Yoshida, M. Bogaki and M. Nakamura, et al., Antimicrob. Agents Chemother., 1991, 35(8), 1647–1650.
37. P. Ghosh and M. C. Bagchi, J. Mol. Model., 2011, 17(7), 1607–1620.
38. N. Minovski, A. Perdih and T. Solmajer, J. Mol. Model., 2011, 18(5), 1735–1753.
39. N. Minovski, A. Perdih, M. Novic and T. Solmajer, J. Comput. Chem., 2013, 34(9), 790–801.
40. N. Minovski, M. Vračko and T. Šolmajer, Mol. Diversity, 2011, 15(2), 417–426.
41. NIAID-Division of AIDS, Anti-HIV/OI/TB Therapeutics Database, http://chemdb.niaid.nih.gov, accessed November 2014.
42. G. Fu, J. Wu and D. Zhu, et al., Nucleic Acids Res., 2009, 37(17), 5908–5916.
43. E. M. Tretter, A. J. Schoeffler, S. R. Weisfield and J. M. Berger, Proteins: Struct., Funct., Bioinf., 2010, 78(2), 492–495.
44. I. Laponogov, X.-S. Pan and D. A. Veselkov, et al., PLoS One, 2010, 5(6), e11338.
45. W. L. DeLano, The PyMol Molecular Graphics System 2008, http://www.pymol.org, accessed November 2014.
46. N. Guex and M. C. Peitsch, Electrophoresis, 1997, 18(15), 2714–2723.
47. W. F. van Gunsteren, S. R. Billeter and A. A. Eising, et al., Biomolecular simulation: the GROMOS96 manual and user guide, http://www.gromos.net, Vdf Hochschulverlag ETHZ, 1996.
48. G. Jones, P. Willet and R. C. Glen, et al., J. Mol. Biol., 1997, 267(3), 727–748.
49. J. Kirchmair, P. Markt and S. Distinto, et al., J. Comput.-Aided Mol. Des., 2008, 22(3–4), 213–228.
50. K. E. Hevener, W. Zhao and D. M. Ball, et al., J. Chem. Inf. Model., 2009, 49(2), 444–460.
51. N. Triballeau, F. Acher and I. Brabet, et al., J. Med. Chem., 2005, 48(7), 2534–2547.
52. ROCS 3.2.0.4, OpenEye Scientific Software, Santa Fe, NM.
53. P. C. D. Hawkins, A. G. Skillman and A. Nicholls, J. Med. Chem., 2007, 50(1), 74–82.
54. M. L. Verdonk, J. C. Cole and M. J. Hartshorn, et al., Proteins: Struct., Funct., Genet., 2003, 52(4), 609–623.
55. G. S. Tillotson, J. Med. Microbiol., 1996, 44(5), 320–324.

---

Footnote

† Electronic supplementary information (ESI) available: Close view of the gyrB hot spot region within our previously published M. tuberculosis-DNA gyrase protein homology model 3K9F_mod³⁹ and the corresponding mutant models produced by in silico mutagenesis (ESI Fig. S1); the results obtained after the Level 1 Boolean-based (T/F (true/false)) clustering (geometry assessment) of the ExpLib dock poses within the QBP of each M. tuberculosis-DNA gyrase protein model (ESI Table S1); the results obtained after the Level 1 Boolean-based (T/F (true/false)) clustering (geometry assessment) of the combinatorially-generated drug-like 6-FQ binding poses (CombiLibHits) for the constructed gyrB mutant models (ESI Table S2); the results obtained after the Level 2 Boolean-based (T/F (true/false)) clustering (score-based clustering) of the (T)-signed combinatorially-generated drug-like 6-FQ binding poses from Level 1 for the constructed gyrB mutant models (ESI Table S3); the screening chemical libraries ExpLib and CombiLibHits available in table format. See DOI: 10.1039/c4ra16031b

---