WO2016046734A2

WO2016046734A2 - Compounds and compositions for treatment of tuberculosis

Info

Publication number: WO2016046734A2
Application number: PCT/IB2015/057280
Authority: WO
Inventors: Alan Gilbert CHRISTOFFELS; Ekow OPPON; Ruben Earl Ashley CLOETE
Original assignee: University Of The Western Cape
Priority date: 2014-09-22
Filing date: 2015-09-22
Publication date: 2016-03-31
Also published as: WO2016046734A3

Abstract

The present invention relates to compounds which have a higher binding affinity than the natural substrate or natural ligand of three protein targets involved in Mycobacterium tuberculosis metabolic pathways. The protein targets being cytidylate kinase (cmk), nicotinate mononucleotide adenylyltransferase (nadD) and ATP synthase epsilon chain (AtpC). The invention further relates to pharmaceutical compositions containing the compounds, uses of the pharmaceutical compositions and methods of treatment of tuberculosis.

Description

COMPOUNDS AND COMPOSITIONS FOR TREATMENT OF TUBERCULOSIS

BACKGROUND OF THE INVENTION

Mycobacterium tuberculosis is the causative agent of tuberculosis (TB). Worldwide, this pathogen accounts for approximately two million deaths and nine million new TB cases annually. Of the 22 countries with the highest burden of pulmonary TB in the world. South Africa (SA) ranks 5* with 10000 deaths yearly. One region of South Africa in particular, the KwaZulu Natal (KZN) province, is severely affected by TB and was the hot spot for an HIV-associated extensively drug-resistant TB (XDR-TB) outbreak in 2005.

Three M. tuberculosis strains from the KZN region were isolated from sputum samples obtained from patients co-infected with HIV and sequenced by the Broad Institute. These three strains represent three levels of varying drug resistance phenotypes namely: susceptible, multiple drug resistant (MDR) and extensively drug resistant (XDR). MDR-TB, is defined as TB infection resulting from strains of bacteria resistant to first-line drugs namely Isonaizid (INH) and Rifampicin (RIF). Of these three strains, the XDR TB strain is the most difficult to treat because it is resistant to first-line anti-TB drugs, any one of the second line injectable drugs (Capreomycin, Kanamycin, or Amikacin) as well as at least one fluoroquinolone drug. Given this growing resistance to anti-TB drugs, identification of new putative targets coupled with the design of novel, potent drugs is urgently required.

A widely used general method of identifying putative drug targets entails the comparison of host and pathogen metabolic pathways. This approach allows the identification of pathways unique to the pathogen and the consequent selection of genes from these pathways as possible drug targets. Tools such as the TDR target database and AssessDrugTarget allow for the prioritization and ranking of identified putative drug targets. However, most targets identified using these tools do not map to metabolic pathways and are not involved in dormancy growth conditions. This present invention relates to identifying targets of drug resistance pathways. The rationale behind this is that despite the emergence of drug resistant mutations in a druggable target in a given metabolic pathway, the same pathway can be disrupted via a different target. This approach is attractive when it is considered that current anti-TB drugs target enzymes (DNA and RNA polymerase, DNA gyrase) in the cascading information-processing pathways instead of metabolic enzymes. Specifically, the inventors aimed at identifying of metabolic pathways for genes known to be associated with anti-TB drug resistance and the subsequent investigation of products within these pathways as putative drug targets. Identified putative targets were filtered to exclude non-viable candidates. We consequently identified three putative targets, Rv1712, Rv2421 c and RV1311 for docking and molecular dynamics studies using potential inhibitors.

Cytidylate kinase (Rv1712/CMK) is a key enzyme involved in the phosphorylation of ATP to form nucleoside diphosphates in the pyrimidine metabolic pathway. The known TB drug target rpoB (Rv0667) resides in the purine and pyrimidine pathway and the gene coding for this target has been shown to contain several mutations associated with RIF resistance, making Rv1712 an attractive putative target. In the present study a 3D structure for Rv1712 is presented and potential inhibitors identified based on molecular docking studies.

The enzyme NadD catalyses the transfer of an adenyl group from ATP to NaMN to form NaAD which is then converted to the ubiquitous intermediate Nicotinamide adenine dinucleotide (NAD). Metabolic pathway and BLAST analysis indicated that Rv2421c is involved in nicotinate and nicotinamide metabolism in M tuberculosis and has no equivalent human ortholog. To the best of our knowledge, Rv2421 c has not been reported in the literature as a potential M. tuberculosis drug target and is not under investigation by the Tuberculosis Structural Genome Consortium (TBSGC) for structure determination. It is up-regulated during dormancy conditions of M. tuberculosis survival and lacks a close homolog in mice making it an attractive drug target. We predicted a 3D structure for the putative drug target Rv2421c and performed molecular dynamics and docking studies to identify novel potential inhibitors. We present for the first time a structural description of M. tuberculosis Rv2421c and potential inhibitors identified after performing docking studies.

A further target, ATP synthase epsilon chain (Rv131 1/atpC) protein catalyzes the production of ATP from ADP in the presence of a proton gradient across the membrane generated by electron transport complexes of the respiratory chain. Rv1311 belongs to the ATPase epsilon chain family and is involved in reaction steps of the Oxidative phosphorylation pathway. A previous study by Yagi and colleagues (2007), which solved the crystal structure of the ATP bound 8 subunit from a thermophillic Bacillus, a close homolog of Rv1311 and two solution structures (apo and holo forms) showed that the C-terminal domain does undergo conformational changes in the presence and absence of ATP binding (Yagi et al., 2007). Rv1311 has also been shown to be essential for growth in Mycobacterium tuberculosis and is conserved between several Mycobacterium species. To the best of our knowledge, Rv1311 has not been reported in the literature as a potential M. tuberculosis drug target and is not under investigation by the Tuberculosis Structural Genome Consortium (TBSGC) for structure determination. We predicted the 3D structure for the putative drug target Rv1311 , performed molecular dynamics simulation studies in the presence and absence of substrate ATP to identify features responsible for substrate binding, and identified potential inhibitors of Rv1311.

SUMMARY OF THE INVENTION

In one aspect of the invention there is provided for a compound selected from the group consisting of:

5-(4-amino-2-oxopyrimidin-1 (2H)-yl)-4-hydroxy-2-(hydroxymethyl)

tetrahydrofuran-3-yl dihydrogen phosphate,

2'-deoxycytidine 5'-monophosphate,

cytidine 5'-triphosphate disodium salt,

cytidine 5'-monophosphate disodium salt,

nicotinic acid mononucleotide,

flavin adenine dinucleotide disodium salt hydrate,

adenosine 5'-triphosphate, disodium salt hydrate, and

adenosine 5'-diphosphate monopotassium salt dehydrate,

or pharmaceutically acceptable salts or derivatives thereof for use in the treatment of tuberculosis.

In one embodiment the tuberculosis is multiple drug resistant tuberculosis.

In another embodiment of the invention the compound targets a Mycobacterium tuberculosis protein by forming a protein-compound complex. It will be appreciated that the protein may be an enzyme, preferably an enzyme involved in a metabolic pathway. More preferably, the enzyme is selected from the group consisting of cytidylate kinase (cmk), nicotinate mononucleotide adenylyltransferase (nadD) and ATP synthase epsilon chain (AtpC). It will be appreciated that the compounds of the invention outcompete the natural substrates of the respective enzymes thus interrupting a metabolic pathway in Mycobacterium tuberculosis.

In yet another embodiment of the invention there is provided for the modification of chemical groups on the compound in order to improve its efficacy. Those of skill in the art will appreciate that the incorporation of for instance lipophilic cages may assist in transporting the compound to the cytoplasm of a cell in order to deliver it to the target protein.

In another aspect of the invention there is provided for pharmaceutical compositions comprising the compounds of the invention. It will be appreciated that the pharmaceutical compositions may comprise one or more of the compounds of the invention. Further, those of skill in the art will appreciate that the pharmaceutical composition may further include pharmaceutical excipients, diluents or carriers. Other suitable additives, including stabilizers and such other ingredients as may be desired may be added.

The invention further extends to methods of treatment of a subject for tuberculosis infection, wherein the method comprises administering an effective amount of one or more of the compounds described above to a subject. It will further be appreciated that the subject may be a mammal, preferably a human. BRIEF DESCRIPTION OF THE FIGURES

Non-limiting embodiments of the invention will now be described by way of example only and with reference to the following figures:

Figure 1: KEGG metabolic pathway map for Nucleotide metabolism (pyrimidine metabolism) in M. tuberculosis H37rV strain. Rv1712 or cmk drug target candidate is shown in the highlighted box for step 2.7.4.14 of this specific pathway. Known drug resistance gene Rv0667 or rpoB is shown in the highlighted box for step 2.7.7.6 of this pathway. M. tuberculosis specific genes are coloured in green.

Figure 2: KEGG metabolic pathway map for Energy metabolism (oxidative phosphorylation) in M. tuberculosis H37rV strain. Rv2984 or ppk drug target candidate is shown in the highlighted box for step 2.7.4.1 of this specific pathway. Rv2194 or qcrC drug target candidate is shown in the highlighted box. Both drug target candidates Rv1305 (atpE) and Rv1311 (atpC) are shown in the highlighted box for step 3.6.3.14 of this specific pathway. The proposed drug target Rv2195 or qcrA is shown in the highlighted box and involved in cytochrome C reductase. Drug candidates Rv1456c or COX15 is shown in the highlighted box involved in cytochrome C oxidase and Rv1622c or CydB is shown in the highlighted box involved in ubiquinol-cytochrome oxidase. Known drug resistance gene Rv1854c or ndh is shown in the highlighted box for step 1.6.99.3 of this pathway.

Figure 3: KEGG metabolic pathway map for Nicotinate and Nicotinamide metabolism in M. tuberculosis H37rV strain. Rv2421c or nadD drug target candidate is shown in the highlighted box for step 2.7.7.18 of this specific pathway. Known drug resistance gene Rv2043c or pncA is shown in the highlighted box for step 3.5.1.19 of this pathway.

Figure 4: Ribbon diagram of the structural model of M. tuberculosis cytidilate monokinase (Rv1712). The CORE domain, NMP-binding domain and the LID domain are labelled as well as ligands CMP and S04 shown as sticks. Figure generated using PyMOL.

Figure 5: DOPE score energy profiles graph of the structural model for Rv1712 and templates 1Q3T, 1CKE, 1 KDO and 2H92. Generated using Gnuplotv4.2.

Figure 6: RMSD of the backbone atoms of model for Rv1712 (green) and substrate C5P (red) during the 30000ps simulation. Generated using Gnuplotv4.2. Figure 7: The variation in total and potential energy for the Rv1712-C5P complex during the 30000ps simulation. Generated using Gnuplotv4.2.

Figure 8: RMS fluctuations of all Ca residues for the Rv1712 during the last 25000ps simulation.

Figure 9: Radius of gyration of all Ca residues for the Rv1712 during the 30000ps simulation.

Figure 10: Superimposition of lowest DOPE score models for the initial and newly generated structures for Rv1712. The model represents the initial model without 3R20 used as a template and 3R20 used as a template for model construction. Ligands S04 and CMP are shown as sticks. RMSD = 0.387A.

Figure 11: Interactions between the top ten compounds and active site residues of Rv1712. Panel A shows that the compound ZINC_A is displaying six hydrogen bond interactions. Panel B shows that the compound ZINCJ3 is displaying five hydrogen bond interactions. Panel C shows that the compound ZINC_C is displaying nine hydrogen bond interactions. Panel D shows that the compound ZINC_D is displaying four hydrogen bond interactions. Panel E shows that the compound ZINC_E is displaying two hydrogen bond interactions. Panel F shows that the compound ZINC_F is displaying six hydrogen bond interactions. Panel G shows that the compound ZINC_G is displaying four hydrogen bond interactions. Panel H shows that the compound ZINC_H is displaying five hydrogen bond interactions. Panel I shows that the compound ZINCJ is displaying four hydrogen bond interactions. Panel J shows that the compound ZINC_J is displaying four hydrogen bond interactions. The dashed lines represent hydrogen bonds. Figures generated using PoseView.

Figure 12: Multiple sequence alignment between Rv2421c (probable nicotinate-nucleotide adenylyl transferase (nadD), 212AA in size) and four homologous templates 2QTR, 1YUM, 2H29 and 1 KAM. The alignment was performed using modellerv9.7 script align2d_mult.py and was visually represented using Jalview v2.3. The highly conserved signature sequence motif [(H T)XGH], the conserved active site residues which recognizes DND for binding and the conserved SXXXX(R K) residues are shown in boxes.

Figure 13: Ribbon diagram of the 3D structure of M. tuberculosis probable nicotinate-nucleotide adenylyl transferase (Rv2421c). The two structural domains: N- terminal composed of residues 1-154 and the C- terminal domain composed of residues 155-191 are labelled as well as substrate DND. Figure was generated using Py OL.

Figure 14: DOPE score energy profiles graph of model Rv2421c and templates 2QTR, 1YUM, 2H29 and 1KAM. Gapped regions in the plot correspond to gaps in the alignment. Figure generated using Gnuplotv4.2 .

Figure 15: Ramachandran plot for the LDSM of Rv2421c. Generated using Procheck.

Figure 16: Molecular dynamic simulation analysis of Rv2421c in complex with substrate DND. Panel A shows RMSD of the backbone atoms of system 2 for Rv2421c and ligand DND during the 30000ps simulation, Panel B shows the variation in total and potential energy for the Rv2421c-DND complex during the 30000ps simulation. Panel C shows the RMS fluctuations of all Ca residues for the Rv2421c during the 30000ps simulation (three regions of high fluctuations, L1-L3). Panel D shows the Radius of gyration of all Ca residues for the Rv2421c during the 30000ps simulation. Plots generated using Gnuplotv4.2.

Figure 17: Interactions between the top four compounds and the DND binding site residues of Rv2421c. Panel A shows that the compound ZINC94S03216 is displaying 7 hydrogen bond interactions and two hydrophobic contacts. Panel B shows that the compound ZINC08551088 is displaying 11 hydrogen bond interactions and two hydrophobic contacts. Panel C shows that the compound ZINC85629877 is displaying 10 hydrogen bond interactions and two hydrophobic contacts. Panel D shows that the compound ZINC77285165 is displaying 6 hydrogen bond interactions and two hydrophobic contacts. The dashed lines represent hydrogen bonds and residues involved in hydrophobic contacts to compounds are shown. Figures generated using PoseView.

Figure 18: Superimposition of LDSM and crystal structure 3MLB.The blue model represents the LDSM of Rv2421c and 3MLB solved in complex with inhibitor ZINC58655383. Substrate DND and ZINC58655383 are shown as sticks. RMSD = 0.776A.

Figure 19: Multiple structural sequence alignment between Rv1311 (probable ATP synthase epsilon chain (atpC), 144AA in size) and four homologous templates. The alignment was performed using modellerv9.7 script align2d_mult.py and is visually represented using Jalview v2.3. Structurally conserved ATP-binding site residues are shown in boxes. ATP-binding site sequence motif [(l/L)DXXRA]. Figure 20: The 3D structure of the LDSM of M. tuberculosis Rv1311. The two structural domains are labelled as; NTD and CTD each composed of residues 1- 84 and 90-120, while the ligand ATP is shown as a stick representation. Figure was generated using PyMOL.

Figure 21 : DOPE score energy profiles graph of model Rv1311 and templates 2E5Y, 2QE7, 2RQ6 and 2RQ7. Gapped regions in the plot correspond to gaps in the alignment. Figure generated using Gnuplotv4.2 .

Figure 22: A) RMSD of the backbone atoms of model for Rv1311 , system 1 , system 2 and ligand ATP during the 30000ps simulation. B) RMS fluctuations of all Ca residues for the Rv1311 , system 1 and system 2 during the 30000ps simulation (regions of high fluctuations are labelled L1-L5). C) Radius of gyration of all Ca residues for the Rv1311 , system 1 and system 2 during the 30000ps simulation. Generated using Gnuplotv4.2.

Figure 23: Superimpositions of snapshot structures for Rv1311 at 15, 25 and 30ns between systems 1 and 2. Panel A shows impositions at 15ns with system 1 and system 2 and distance measurement (11.5A) between residue Pro103 of both systems. Panel B shows snapshot structures at 25ns with system 1 and system 2 and distance measurement (24.3A) between residue Pro103 of both systems. Panel C shows snapshot structures at 30ns with system 1 and system 2 and distance measurement (8.3A) between residue Pro103 of both systems. ATP is shown as a stick structure. Figures generated using PyMol.

Figure 24: Superimpositions of the range of movements of the two MD systems of the first principal component. System 1 and system 2 with the corresponding loop regions labelled. Figures generated using PyMol.

DETAILED DESCRIPTION OF THE INVENTION

The present invention will now be described more fully hereinafter with reference to the accompanying drawings, in which some, but not all embodiments of the invention are shown.

The invention as described should not be limited to the specific embodiments disclosed and modifications and other embodiments are intended to be included within the scope of the invention. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation. As used throughout this specification and in the claims which follow, the singular forms "a", "an" and "the" include the plural form, unless the context clearly indicates otherwise.

The terminology and phraseology used herein is for the purpose of description and should not be regarded as limiting. The use of the terms "comprising", "containing", "having" and "including" and variations thereof used herein, are meant to encompass the items listed thereafter and equivalents thereof as well as additional items.

By "tuberculosis infection" is meant any condition, disease or disorder that has been correlated with the presence of an existing Mycobacterium infection, preferably the Mycobacterium infection is caused by the presence of one or more of the following causative organisms Mycobacterium tuberculosis, Mycobacterium bovis, Mycobacterium marinum, Mycobacterium intracellulare, Mycobacterium avium (including sub sp. paratuberculosis and avium), Mycobacterium leprae, Mycobacterium ulcerans, Mycobacterium kansasii, Mycobacterium vanbaalenii, Mycobacterium gilvum, Mycobacterium sp. MCS, Mycobacterium sp. KMS, Mycobacterium sp. JLS, Mycobacterium smegmatis and Mycobacterium abscessus. Most preferably the tuberculosis infection is caused by Mycobacterium tuberculosis.

The pharmaceutical compositions and compounds of the invention can be provided either alone or in combination with other compounds (for example, nucleic acid molecules, small molecules, peptides, or peptide analogues), in the presence of a liposome, an adjuvant, or any carrier, such as a pharmaceutically acceptable carrier and in a form suitable for administration to mammals, for example, humans, cattle, pigs, etc.

As used herein a "pharmaceutically acceptable carrier" or "excipient" includes any and all antibacterial and antifungal agents, coatings, dispersion media, solvents, isotonic and absorption delaying agents, and the like that are physiologically compatible. A "pharmaceutically acceptable carrier" may include a solid or liquid filler, diluent or encapsulating substance which may be safely used for the administration of the pharmaceutical compositions or compounds to a subject. The pharmaceutically acceptable carrier can be suitable for intramuscular, intradermal, intravenous, intraperitoneal, subcutaneous, oral or sublingual administration. Pharmaceutically acceptable carriers include sterile aqueous solutions, dispersions and sterile powders for the preparation of sterile solutions. The use of media and agents for the preparation of pharmaceutically active substances is well known in the art. Where any conventional media or agent is incompatible with the active compound, use thereof in the pharmaceutical compositions of the invention is not contemplated. Supplementary active compounds can also be incorporated into the compositions.

Suitable formulations or compositions to administer the pharmaceutical compositions and compounds of the present invention to subjects with a tuberculosis infection or subjects which are presymptomatic for a condition associated with tuberculosis infection fall within the scope of the invention. Any appropriate route of administration may be employed, such as, parenteral, intravenous, subcutaneous, intramuscular, intracranial, intraorbital, ophthalmic, intraventricular, intracapsular, intraspinal, intrathecal, intracistemal, intraperitoneal, intranasal, aerosol, topical, or oral administration.

As used herein the term "subject" includes mammals, preferably human or animal subjects, but most preferably the subjects are human subjects. .

For pharmaceutical compositions, an effective amount of the compounds of the present invention can be provided, either alone or in combination with other compounds, or they may be linked with suitable carriers and/or other molecules, such as lipophilic cages. The presence of lipophilic cages may assist in enhancing transport of the compounds to the cytoplasm of the cell in order to deliver them to the target proteins.

In some embodiments, the pharmaceutical compositions or compounds according to the invention may be provided in a kit, optionally with a carrier and/or an adjuvant, together with instructions for use.

An "effective amount" of a compound or pharmaceutical composition according to the invention includes a therapeutically effective amount. A "therapeutically effective amount" refers to an amount effective, at dosages and for periods of time necessary, to achieve the desired therapeutic result, such as treatment of an infection or a condition associated with such infection. The outcome of the treatment may for example be measured by a decrease in bacterial count, inhibition of target metabolic pathways, delay in development of a pathology associated with tuberculosis infection, or any other method of determining a therapeutic benefit. A therapeutically effective amount of a compound may vary according to factors such as the disease state, age, sex, and weight of the individual, the ability of the compound to elicit a desired response in the individual, the nature and severity of the disorder to be treated or prevented, the route of administration, and the form of the composition. Dosage regimens may be adjusted to provide the optimum therapeutic response. A therapeutically effective amount is also one in which any toxic or detrimental effects of the compound are outweighed by the therapeutically beneficial effects.

Any of the compositions of the invention may be administered in a single dose or in multiple doses. The dosages of the compositions of the invention may be readily determined by techniques known to those of skill in the art or as taught herein.

Dosage values may vary and be adjusted over time according to the individual need and the judgment of the person administering or supervising the administration of the pharmaceutical compositions or compounds of the invention. It may be advantageous to formulate the compositions in dosage unit forms for ease of administration and uniformity of dosage.

The term "preventing", when used in relation to an infectious disease, or other medical disease or condition, is well understood in the art, and includes administration of a composition which reduces the frequency of or delays the onset of symptoms of a condition in a subject relative to a subject which does not receive the composition.

The term "therapeutic" treatment is well known to those of skill in the art and includes administration to a subject of one or more of the pharmaceutical compositions or compounds of the invention. If the composition is administered after manifestation of the unwanted condition, the treatment is therapeutic (i.e., it is intended to diminish, ameliorate, or stabilise the existing unwanted condition or side effects thereof).

Certain compounds of the present invention possess asymmetric carbon atoms (optical centers) or double bonds; the racemates, diastereomers, geometric isomers and individual isomers are all intended to be encompassed within the scope of the present invention.

The present invention relates the identification of eight compounds that have inhibition activity against an attenuated form of Mycobacterium tuberculosis. The compounds of the present invention out compete the natural substrate for binding to Mycobacterium tuberculosis protein target.

The compounds were identified in silico and showed higher binding affinity to protein targets cytidylate kinase (cmk) represented by Rv1712, nicotinate mononucleotide adenylyltransferase (nadD) represented by Rv2421c and ATP synthase epsilon chain (AtpC) represented by Rv1311 , when compared to their natural substrates CMP, DND and ATP, respectively. The inventors predicted the three dimensional structures of these proteins in complex with their natural substrate. Thereafter molecular dynamics were performed to determine a stable protein- substrate complex. Docking studies were performed to the stable protein structure first using the substrate and then using compounds identified from the ZINC database that shared 60-90% structural similarity to the substrate. Interactions between the protein residues and compounds were predicted and compared to those observed within the experimental structure. Eight compounds showed higher binding affinity and number of interactions. These compounds were experimentally validated by performing inhibition growth assays. Eight compounds were found to have an effect on the growth of an attenuated Mycobacterium tuberculosis strain.

Using computational docking methods, the inventors identified compounds with higher binding affinity than the substrate or natural ligand to three protein targets cytidylate kinase (cmk), nicotinate mononucleotide adenylyltransferase (nadD) and ATP synthase epsilon chain (AtpC). Inhibition assays conducted with the compounds identified using the computational methods revealed eight compounds with activity against an attenuated strain of Mycobacterium tuberculosis. For cytidylate kinase (cmk) the compounds with activity against M. tuberculosis are:

5-(4-amino-2-oxopyrimidin-1 (2H)-yl)-4-hydroxy-2-(hydroxymethyl)

tetrahydrofuran-3-yl dihydrogen phosphate,

2'-deoxycytidine ^-monophosphate,

cytidine 5'-triphosphate disodium salt; and

cytidine 5'-monophosphate disodium salt.

For ATP synthase epsilon chain (AtpC) the compounds with activity against M. tuberculosis are:

adenosine 5'-triphosphate, disodium salt hydrate; and

adenosine 5'-diphosphate monopotassium salt dihydrate.

For nicotinate mononucleotide adenylyltransferase (nadD) the compounds with activity against M. tuberculosis are:

nicotinic acid mononucleotide; and

flavin adenine dinucleotide disodium salt hydrate.

These compounds could be developed into new anti-tuberculosis drugs for treatment of drug resistant strains of Mycobacterium tuberculosis.

The following examples are offered by way of illustration and not by way of limitation. EXAMPLE 1

Identification of potential M. tuberculosis targets

The BROAD institute (http://www.broadinstitute.org/) has sequenced the whole genomes of three KwaZulu Natal TB strains (KZN 4207, KZN 1435 and KZN 605) and performed comparative analysis among these datasets. Ten mutations were verified in eight genes that account for drug resistance and 26 unique non- coding and coding polymorphisms in 20 other genes between the MDR and XDR strains. We mapped the 20 genes comprising 26 novel mutations identified by the BROAD institute to metabolic pathways. However, most of these genes lacked pathway data (75%) and 40% of these genes were described as hypothetical proteins. Nevertheless, two genes mapped to pathways that are involved in protein transport and transcription. Literature based searches were performed and ten additional genes known to be associated with first and second-line TB drug resistance were identified.

The inventors focussed on the 18 known genes that are associated with drug resistance because of additional information on the essentiality of these genes for M. tuberculosis survival.

Metabolic pathway analysis and target prioritization

The 18 known drug resistance genes were used to query the KEGG M. tuberculosis H37Rv strain database (http://www.genome.jp/kegg- bin/show organism?org=mtu). The mapping of these genes to metabolic pathways provided an opportunity to investigate other genes (excluding known genes) as drug candidates. Six of the 18 drug resistance genes have not been annotated with any pathway data and were assigned to what are called "pathway holes". We therefore focused on targets that map to pathways containing first-line drug resistance genes as these pathways have been extensively characterized. Three metabolic pathways were selected because of experimental evidence for gene products within these pathways associated with importance for bacterial growth during active and latent states and their potential for developing antimicrobial agents for other genes within these pathways. These three pathways comprise 105 genes of which 14 are known TB genes that were excluded from our search for drug targets. Thirty-eight of the remaining 91 H37rV genes were not currently under investigation by the TB Structural Genome Consortium (TBSGC) as reported on the TBSGC web portal. Only one target currently being investigated by the TBSGC was included in our candidate list because no structure has yet been elucidated for this protein. The TubercuList World Wide Web Server

(http://genolist.pasteur.frl/TubercuList) was used to query the combined 39 genes to find relevant information regarding essentiality of the gene, if its function is known (meaning it has a biological role within a reaction step of the pathway) and if a 3D structure is known. TubercuList provided valuable experimental evidence for growth based on transposon site hybridisation assays and other mutagenic studies done on all M. tuberculosis genes confirming their essentiality or non-essentiality. Seventeen of the 39 potential drug targets can be classified as essential genes with known biological function and were used for BLASTp searches against the H. sapiens database at NCBI (http://www.ncbi.nlm.nih.gov/Blast.cgi) (build GRCh37/hg19). Among these 17 proteins, nine showed no similarity to human proteins and were chosen as potential drug targets, eliminating possible human host protein-drug interactions (expectation score < 0.0005). The amino acid sequence of each of the nine selected drug candidates were used to perform a BLASTp search against 24 Mycobacterium species to determine inter-species sequence conservation. A high degree of sequence conservation suggests that mutations in these proteins are not tolerated thereby prohibiting the spontaneous occurrence of drug resistance. The M. tuberculosis targets that were retained were investigated for their possible role in M. tuberculosis survival during latency conditions. Genes that are essential for dormancy survival are up-regulated during latent conditions of M. tuberculosis and should be targeted in an attempt to combat TB persistence. The M. tuberculosis protein accession numbers were used to query the Tuberculosis Database (http://www.tbdb.org/) to obtain gene expression data collated for each target. These include samples, conditions and the statistically significant measurements (p-value) for each experiment. Conditions most often associated with persistence have been reported as hypoxia, starvation and change in pH. Additionally, the lack of close homologs to human gut flora would facilitate the design of M. tuberculosis specific drugs and inform in vivo mouse infection studies. To this end, protein BLAST searches were carried out against three species of human gut flora bacteria Staphylococcus aereus (taxid: 1280), Enterococcus faecalis (taxid: 1351) and Escherichia coli (taxid: 562) and mouse proteins (Mus musculus).

Homology modelling

Homology models were constructed for M. tuberculosis protein Rv1712 (cytidylate kinase) using MODELLER version 9.7. Homologous template structures were identified by scanning the target amino-acid sequence against known structures deposited in the Protein Data Bank (PDB) using the profile. build module of MODELLER, while multiple sequence alignments were obtained using the salign routine. The alignment was optimized manually and inputted into MODELLER'S model-mult-hetero module for model construction. Structural templates for Rv1712 were cytidine monophosphate kinase from Streptococcus pneumoniae (1Q3T), cytidylate kinase from Staphylococcus aureus binding cytidine-5-monophosphate (2H92), as well as CMP from E. coli both as free enzyme (1CKE) and in complex with 2 cytidine monophosphate (1 KDO). Fifty models were generated for Rv1712 using MODELLER and all models were constructed in the presence of ligand C5P to improve the final model conformation as we intend exploiting the C5P binding site for future docking studies.

Model evaluation

The LDSM was selected for qualitative analysis as it is considered to be the model closest to the proteins native state. The DOPE score energy profiles were calculated for the selected model and the four templates and compared graphically using Gnuplot (version 4.2) to locate regions of high energy. The accuracy of the LDSM was assessed by calculating the normalised DOPE Z-scores. Usually models with a DOPE Z-score of -1 and lower are considered as good quality structures. PROSAII was used to assess the reliability of regions within the model that are important for substrate binding while PROCHECK was used to determine if the model satisfied stereochemical constraints. ERRAT, that allows checking of crystal structures based on non-bonded atom-atom interactions, was used to assess the overall quality of the LDSM relative to reliable, high-resolution structures. Finally, the 3D structural similarity between model and the four templates were assessed using the root mean square deviation (RMSD) value as calculated in PyMol (http://www.pymol.org). The lower the RMSD value, the more similar the structures are with respect to the backbone conformation.

Energy refinement and molecular dynamics

The LDSM was further refined using energy minimization steps. The energy minimization included 1000 steps of steepest descent and 5000 steps of conjugated gradients using GROMACS to remove bad van der Waals force contacts. The energy minimized structure was then subjected to 30ns molecular dynamics simulation to determine stability of the structure in complex with ligand C5P using GROMACS. The total and potential energy terms were calculated using GROMACS tools. Also, the RMSD and R SF analysis was calculated using GROMACS for the equilibrated structure over the whole trajectory as well as the radius of gyration.

Parameter optimisation

The ligand structure C5P was docked to the active site pocket residues Asp31 , Tyr36, Arg37, Arg106, Arg128, Asp129, Arg178, Asp182 and Arg185 of model Rv1712, respectively. The top ten binding modes produced by the docking run of the ligand C5P were compared to the most stable ligand trajectory using PYMOL. This was done to determine if AutoDock-vina correctly predicted the top binding pose as the most energy favourable and if it can replicate the most stable ligand trajectory in silico. If AutoDock-vina correctly predicted the top binding mode as the most energetically favourable, this would suggest successful parameter optimisation.

Molecular docking

Docking C5P to the equilibrated structure

The equilibrated protein structure Rv1712.gro was converted to pdb structure using editconf tool and SOL, sodium and C5P molecules were deleted. The final pdb structure of Rv1712 and the ligand structure C5P were converted to a set of united atom aliphatic carbons, aromatic carbons, polar hydrogens, hydrogen bonding nitrogens and oxygens with charges called pdbqt using AutoDock Tools. This was done to correct for errors such as missing atoms, added H₂0, more than one molecule chain breaks and alternate locations. After receptor and ligand preparation, the center of mass is calculated for protein Rv1712 with the ligand present. A python script center of mass.py is executed and returns X, Y and Z coordinates for the grid space. This defines the search space before the docking simulation commences. The various parameters for the docking process are stored in a configuration file named conf.txt that contains input parameters for the docking simulation. These parameters include: center of mass coordinates, grid space and exhaustiveness of the search algorithm. The nine modes produced by the docking run are compared to the experimental pose using PYMOL (http://www.pymol.org).

Compound screening and interaction analysis

The ZINC database was searched using the simple molecular input line entry (SMILE) format of ligand C5P to identify structurally similar compounds. After screening, 48 compounds were identified at 95% structural similarity and downloaded in sdf format from the ZINC database. The compounds were converted to pdbqt format using Open Babel in preparation for docking simulation. The processed compounds were docked to the overall simulated protein structure of Rv1712 using AutoDock Vina software. Rigid body docking was performed to obtain several possible conformations and orientations for the compounds docked at the receptor's active site. The Grid box dimensions of previously parameter optimised dockings were implemented. Docking energies calculated during the run were extracted and ranked using a python script called vina_screen_get_top.py with lowest energy values at the top. The binding energies reported represent the sum of the intermolecular energy, total internal energy and torsional free energy minus the energy of the unbound system. The top ten lowest energy binding modes for each compound were visually inspected in PyMol and were further analysed using PoseView and BINANA programs. PoseView determines four types of interactions namely; i) hydrogen bonds, ii) hydrophobic, iii) metal interactions and iv) π interactions, while BINANA was used to calculate all the aforementioned interactions as well as salt bridges because this functionality was not present in PoseView.

Genome comparisons and pathway analysis

Comparative data generated by the BROAD institute (http://www.broadinstitute.org/) confirmed ten mutations in eight known genes that account for first and second-line drug resistance while 26 mutations were identified in 20 novel genes among the MDR and XDR KZN strains. Further literature searches yielded ten additional genes implicated in first and second-line anti-TB drug resistance. This study only focussed on known drug resistance genes as TB drugs have successfully targeted these. The combined 18 known drug resistant genes were then mapped onto pathways in the KEGG database. Twelve genes mapped to 15 metabolic pathways, while six could not be assigned to any KEGG pathway (Table 1). Three of the fifteen metabolic pathways namely. Pyrimidine (total of 42 genes). Oxidative phosphorylation (total of 47 genes) and, Nicotinate and Nicotinamide metabolism (total of 16 genes) were selected for investigation because of their functional importance during active bacterial growth as well as in the latent state.

Furthermore, previous studies highlighted these pathways as promising targets for developing antimicrobial agents against slow growing bacteria. Table 1 : Metabolic pathways of known M. tuberculosis genes involved in drug resistance

Selection and prioritization of candidate genes

Of the 105 genes identified in the three selected pathways, 14 are known TB drug targets and were excluded from further analyses. The accession numbers for the remaining 91 genes were cross-checked with the open reading frames accessions of genes under investigation by the TBSGC to avoid duplication of research efforts. Thirty-eight putative targets were not under investigation by the TBSGC. An additional putative target, Rv2984, currently under investigation by the TBSGC, lacks a solved 3D structure and was therefore included in our analysis (Table 2). Out of the 39 combined putative targets, only 17 were found to be essential for M tuberculosis survival, have known biological function and have no solved 3D structure (Table 2).

Table 2: M. tuberculosis putative drug target genes

Gene TBSGC Essential PDB Known Human Conserved Up or down for growth structure biological homolog among 24 regulated during

(8, 28) available function <P < mycobacterial dormancy

0.0005) species

Rv1712 NUI yes no yes no yes Up regulated

Rv2984 Ul yes no yes no yes Up regulated

Rv2194 NUI yes no yes no yes Down regulated

Rv1311 NUI yes no yes no yes Down regulated

Rv1305 NUI yes no yes no yes Down regulated

Rv2195 NUI yes no yes no yes Down regulated

Rv1622c NUI yes no yes no yes Up regulated

Rv1456c NUI yes no yes no yes Down regulated

Rv2421c NUI yes no yes no yes Up regulated

Rv3199c NUI no - - - - -

Rv1595 NUI yes no yes yes - -

Rv3393 NUI no - - - - -

Rv0156 NUI no - - - - -

Rv0157 NUI no - - - - -

Rv3158 NUI no - - - - -

Rv3157 NUI no - - - - -

Rv3156 NUI no - - - - -

Rv3154 NUI no - - - - -

Rv3152 NUI no - - - - -

Rv3151 NUI no - - - - -

Rv3150 NUI no - - - - -

Rv3149 NUI no - - - - -

Rv3148 NUI no - - - - -

Rv3147 NUI no - - - - -

Rv3145 NUI no - - - - -

Rv0392c NUI no - - - - -

Rv1812c NUI no - - - - -

Rv3043c NUI yes no yes yes - - Rv2193 NUI yes no yes yes - -

Rv0248c NUI no - - - - -

Rv3319 NUI no - - - - -

Rv1307 NUI yes no yes yes - -

Rv1308 NUI yes no yes yes - -

Rv1309 NUI yes no yes yes - -

Rv1310 NUI yes no yes yes - -

Rv2196 NUI yes no yes yes - -

Rv1623c NUI no - - - - -

Rv1629 NUI yes no yes yes - -

Rv0321 NUI no - - - - -

Rv3644c NUI no

essentiality

data

Out of the 17 M tuberculosis putative targets, the three dimensional structure of 9 targets showed no homology to human proteins (p > 0.0001) and shared approximately 67-100% sequence similarity and 82-99% sequence coverage with orthologs in 24 other Mycobacterium species. Microarray data analysis further revealed that four (Rv1712, Rv2984, Rv1622c and Rv2421c) of the nine genes were up-regulated during dormancy conditions (starvation, hypoxia and oleic acid) with p- values ranging between 0.02 and le-09. The remaining five genes (Rv2194, Rv1305, Rv1456c, Rv2195 and Rv1311) were found to be down- regulated but with weak statistical support (p > 0.01) under similar conditions. Further analysis performed to validate the suitability of selected targets included BLASTp searches against three species of human gut flora bacteria {Staphylococcus aereus, Enterococcus faecalis and Escherichia coli) and mouse proteins {Mus musculus). The nine proteins showed no homology to any of the mouse proteins (p > 1X10^"4). However, six proteins showed varying degrees of sequence similarity and coverage to at most two intestinal bacterial species. These were Rv1712, Rv1311 , Rv2421c, Rv2984, Rv1622c, and Rv1305. Of these, Rv1712, Rv1311 , and Rv2421c show approximately 40% sequence similarity and 90% sequence coverage to orthologs in the three host intestinal bacterial species. Rv2984 and Rv1622c showed 30-40% sequence identity to orthologs in two bacterial species while Rv1305 showed 55% sequence similarity and 87% sequence coverage to an ortholog in host gut bacteria. Three proteins namely Rv2194, Rv2195 and Rv1456c showed no homology to any proteins in the three human gut bacteria. KEGG pathway maps for each of the nine genes were reconstructed using the M. tuberculosis H37rV strain database to visually illustrate the location of each gene within their respective pathways (Figures 1 , 2 and 3). Interestingly the Rv1622c and Rv1456c are involved in more than one metabolic pathway which share a common link (Table 3).

Table 3: Metabolic pathways of M. Tuberculosis putative drug targets

Template selection and model building

Searching the protein data bank (PDB), four homologous templates were identified for Rv1712. These were 1Q3T, 2H92, 1CKE and 1 KDO. The multiple sequence alignment between Rv1712 and the four templates showed sequence identities of 43% (1 Q3T), 40% (2H92), 39% (1CKE) and 40% for 1 KDO. A structure- based alignment for Rv1712 and the four templates indicates that P-loop residues (10-16) in the N-terminal domain constituting the ATP-binding site and the CMP- binding residues of the nucleoside monophosphate (NMP) to be highly conserved. Perfectly conserved residues include D33, T34, Y38, R39, R105, R126, D127, R176, D180 and R183. The conserved nature of the active site residues between Rv1712 and the four templates potentially allows these residues to be used in identifying novel inhibitors for enzymes of Rv1712 in docking studies.

Fifty structural models were constructed for Rv1712 out of which the model with the lowest DOPE score was selected for further analysis. The GA341 score was equal to 1.0. The Rv1712 three-dimensional model is made up of nine a-helices and eight β- strands. It additionally contains three domains: the CORE domain (residues 10-16), the NMP-binding domain (33-100) and the LID (155-168) domain (Figure 4). The NMP-binding domain is free to rotate during substrate binding and recognizes both CMP or dCMP. Model Quality Validation

The DOPE score profile of the selected Rv1712 model is similar to that of templates 1Q3T, ICKE, IKDO and 2H92 (Figure 5). The absence of high energy regions imply that the model is close to the native structure of the protein. The normalised DOPE Z-score for the lowest DOPE score model (LDSM) was found to be -0.77 indicative of model reliability. Moreover, the model conforms to permitted stereochemical restraints as per PROCHECK [14] with 91.6% of residues in most favoured and none in disallowed regions of the Ramachandran plot. The ERRAT overall quality factor indicates that 85% of residues in the 3D structure have a low error rate thus making the model comparable to high resolution 3D structures (0.5- 1A). The Prosa Z-score for the LDSM of Rv1712 (-7.67) was comparable to that of the templates (-7.31 to -7.85). The RMSD values suggest that the LDSM of Rv1712 is more similar to 2H92 (0.53 A) and 208R (0.62 A) than 1Q3T (3.25 A) and 1CKE (2.33 A). This indicates very little deviation from the main chain carbon atoms between target and template(s) suggesting homology and similarity between the structures.

Molecular Dynamics

Analysis of the structural model in complex with ligand C5P indicates a rapid increase in RMSD during the first 2500ps which then gradually decreases after 5000ps for both the protein's backbone and ligand C5P (Figure 6). An equilibrium phase was reached within 5000ps suggesting that 30ns was sufficient for stabilizing the structure. The average total energy and the potential energy reaches convergence at -1.11074e+06 and -1.35662e+06 KJ/Mole, respectively (Figure 7). The RMS fluctuation for residues averaged ~0.1 (nm) and the radius of gyration for the molecule became constant after 5000ps fluctuating between 1.75 and 1.8 (nm) (Figure 8 and 9).

Virtual compound screening and interaction analysis

The lowest energy conformation obtained for docking C5P to the equilibrated structure was found at a binding affinity value of -7.3 kcal/mol. Screening of the 48 compounds obtained from the ZINC database against the equilibrated structure yielded 19 compounds with higher binding affinity values than (-7.3kcal/mol). We selected the top ten minimum energy binding modes for interaction analysis. According to PoseView, the natural substrate C5P forms one hydrogen bond with residue Glnll3, and one n-stacking interaction with Tyr36, while, salt bridge interaction analysis confirmed one positively charged residue Arg37 with the predicted structure of Rv1712. The number of hydrogen bonds formed between ZINC A, ZINC B, ZINC C, ZINC D, ZINC E, ZINC F, ZINC G, ZINC H, ZINCJ and ZINC_J and Rv1712 were 6, 5, 9, 4, 2, 6, 4, 5, 4 and 4, respectively (Figure 11 A, B, C, D, E, F, G, H, I and J). BINANA interaction analysis indicated salt bridge formation between all the compounds and positively charged residues of Rv1712 ranging between two and three residues namely: Lys14, Asp129 and Arg185. Nine out of the ten inhibitors confirmed the n-stacking interaction with Tyr36 of Rv1712 except for compound ZINC C.

Target selection protocols

This study focussed only on genes in metabolic pathways that contain known drug resistance genes. This prerequisite ensured that we targeted only functionally characterised enzymes thereby increasing the likelihood of selecting effective putative target candidates given that anti-TB drugs have successfully targeted these pathways. Target validation included five filtering steps and two subsequent quality tests during prioritization. Previous studies used freely available resources such as AssessDrugTarget and the TDR target database to identify potential drug targets in pathogenic bacteria. For example, the transcription factor, devR, and the enzymes glnE and fbpD have been proposed as potential drug targets in M tuberculosis (Hasan S, et al 2006). However, neither glnE nor fbpD are part of a metabolic pathway despite ranking high on the metabolic list in the referenced study. Moreover, the target fbpD was ranked on the M tuberculosis species-specific list of the referenced study although not being an essential enzyme. Furthermore, the enzymes (glnE and fbpD) do not have an assigned biological role within an organism, making them harder to study with biochemical and biophysical methods.

Predicted sub-cellular localization of putative targets

Of the identified putative targets, six (Rv2984, Rv2194, Rv1305, Rv1622c, Rv2195, Rv1456c) are membrane proteins while three (Rv1712, Rv1311 and Rv2421 c) are cytoplasmic proteins. Membrane proteins are considered easier drug targets because drugs targeting these proteins bind to them on the external side of the cell which eliminates the need for substantial modification of the drug molecule. The structures of membrane proteins are however difficult to solve using biophysical methods and as such experimental assessment of a target's druggability potential is also difficult. Strategies for the delivery of drugs to intracellular targets have dramatically improved over the years and include the use of cell penetrating peptides, pH responsive carriers and endosome-disrupting agents. Cytidylate kinase and Polyphosphate kinase as drug targets

Out of the nine putative targets obtained using our approach, six (Rv1712, Rv2984, Rv2194, Rv1311 , Rv1305 and Rv1622c) have been previously identified as potential drug targets, which further supports the robustness of our method. Of these, only Rv1712 (cytidylate kinase) and Rv2984 (polyphosphate kinase) have been investigated both computationally and experimentally. However, the 3D structure of cytidylate kinase has not been made publicly available for comparison and no inhibitors have been identified for this protein. In the case of polyphosphate kinase, experimental evidence has been previously provided that indicates it is a promising drug target. These indications coupled to our findings support the hypothesis that cmk and ppk genes could represent important M tuberculosis drug targets. To our knowledge, the putative targets arising from this study, including Rv2195, Rv1456c and Rv2421 c, have not been proposed as potential drug targets. We propose them as additional putative targets that ought to be further investigated for rational drug design.

Pathway and drug target validation

The nine genes reported in this study reside in three metabolic pathways (Pyrimidine, Oxidative phosphorylation, Nicotinate and Nicotinamide metabolism). Of these genes, two (Rv1622c and Rv1456c) map to an additional pathway the 'two component system' that is unique to bacteria, fungi and plants. Drugs targeting these targets would therefore be specific to bacteria with minimal human toxicity. The Rv1712 putative target is a key enzyme involved in the phosphorylation of ATP to form nucleoside diphosphates in the pyrimidine pathway. The known TB drug target rpoB (Rv0667) resides in the purine and pyrimidine pathway and the gene coding for this target has been shown to contain several mutations associated with RIF resistance. The identification of Rv1712 in this pathway points to an attractive alternative target whose inhibition would still disrupt this pathway. Five (Rv2984, Rv2194, Rv1311 , Rv1305 and Rv2195) putative candidates map to the oxidative phosphorylation pathway. The known gene ndh, coding for Rv1854c, resides in this pathway and is the target for INH. Studies have shown several mutations in this gene that accounts for INH resistant cases. If these five putative targets are targeted for inhibition, they could disrupt this pathway leading to the killing of M tuberculosis which has limited ATP available to it in its dormant phase. Rv2421c plays an important role in the transfer of phosphorous groups in both nicotinate and nicotinamide salvage and de novo pathways. The pncA gene coding for Rv2043c occurs in this pathway and is targeted by the PZA drug that is highly effective at killing persistent bacilli in the initial phase of TB therapy. Previous studies have indicated that pncA mutations conferred resistance to PZA. Successful inhibition of Rv2421c may therefore eradicate slowly growing persistent bacilli in TB infection.

Although the three selected pathways are also found in the human host, the specific putative targets are unique to bacteria and show no sequence similarity to any human proteins. Potential drugs are therefore not likely to affect these pathways in the human host. Furthermore, the absence of close homologs in mice for the nine genes makes mouse infection studies feasible. All nine putative targets shared a significant degree of sequence similarity and coverage in several mycobacterial species ( . leprae, M. abscessus, and M. ulcerans) pathogenic to humans. Inhibitors with a broad spectrum can therefore be designed. Given that the identified putative targets are conserved in E. coli, H. pylori and N. meningitidis, TB drugs designed to bind to these targets may be used for infections caused by these bacteria. The conservation of these targets probably points to their essentiality in these organisms. From a drug design perspective, targeting of conserved targets is appealing because such targets are not easily prone to mutations. Comparative sequence analysis showed that three genes (Rv2194, Rv2195 and Rv1456c) lacked close homologs to several human gut bacteria and as such targeting these candidates is unlikely to disrupt the host microbiome balance. Although Rv1712, Rv2984, Rv1311 , Rv1305, Rv1622c and Rv2421c showed some degree of sequence similarity and coverage to several gut bacteria proteins, further investigations are warranted. The target for the widely used TB drug Rifampicin, DNA directed RNA polymerase, is highly conserved across all species. However, five residues in three regions show no sequence conservation between eukaryotic RNA polymerase and bacterial RNA polymerase which explains the selectivity of RIF. Although target Rv1305 satisfied all our prioritization criteria and is proposed as a potential target, its expression in humans makes us not rank it top of our putative targets list. However, further investigation to decode the subtle differences between the bacterial and human forms is warranted. Given that our results are in agreement with experimental findings that Rv1712 is essential for cellular processes in M tuberculosis as a highly important drug target, we selected it for further analysis.

Template search, model building and quality assessments

The amino acid sequence identity for Rv1712 and several solved structures in the PDB ranged between 39-43%. As such these structures could be used as modeling templates. The LDSM of Rv1712 satisfied all quality checks i.e. normalised DOPE Z- score, stereochemical restraints, PROSA Z-score and ERRAT overall quality factor score. Moreover, inclusion of the recently solved crystal structure of the M smegmatis cytidylate kinase as a template in our modelling did not significantly alter the model structure obtained for Rv1712 (Figure 10). We therefore believe that the model for Rv1712 approximates the native protein structure (Figure 2).

Molecular dynamics

The simulation of Rv1712 in complex with C5P indicated that 30ns was sufficient for stabilizing the protein structure. This was verified by the fluctuating of the backbone RMSD value between 0.3 and 0.4 (nm) and C5P between 0.1 and 0.15 (nm). Furthermore, energy analysis showed that the molecule had reached convergence throughout the simulation. The RMSF analysis indicated that very few regions had large fluctuation values that corresponded to active site amino acid residues. The radius of gyration for the molecule became constant after 5000ps suggesting that the Rv1712-C5P complex has a stable surface structure suitable for virtual screening and drug design.

Virtual screening and interaction analysis

A total of 19 compounds showed higher binding affinity values compared to C5P docked to the equilibrated structure of M tuberculosis protein model Rv1712. Selecting the top ten compounds and performing interaction analysis provided support for the higher binding affinity values. The higher binding affinity values could be due to the larger number of hydrogen bonds and salt bridge interactions observed between the ten docked compounds and protein model Rv1712. The larger number of interactions suggests that these compounds may have a stronger affinity for the cytidylate kinase enzyme. Additionally, nine of the compounds reproduce a similar ligand-binding orientation of C5P confirming that these compounds occupy the correct ligand-binding site. Also, these compounds make interactions with a range of newly identified residues within the binding pocket of C5P as well as known binding site residues. The residues identified are hydrophilic in nature and crucial for substrate recognition and structural stability of the protein. We propose the use of the ten compounds in experimental testing to determine if they are non-toxic to human cell lines and able to inhibit M tuberculosis growth. Summary

In this study, known drug resistance genes were mapped onto pathways in the KEGG pathway database and the component proteins of three essential pathways obtained.

Putative drug targets were obtained from this list by the application of several stringent filtering criteria such as gene essentiality for M. tuberculosis growth, known gene function, absence of the gene in the human host and conservation of gene across Mycobacterium species. From this pipeline, nine putative drug target candidates (Rv1712, Rv2194, Rv2984, Rv1311 , Rv2195, Rv2421c, Rv1305, Rv1622c and Rv1456c) were identified. To our knowledge, this study is the first to report three (Rv2195, Rv2421c and Rv1456c) of the nine genes as potential drug targets with supporting evidence. Among the three newly identified targets, Rv1456c is of particular interest, because it maps to more than one metabolic pathway, indicating that a drug targeting several choke points could be designed. As a case study we selected one target Rv1712, predicted its 3D structure using homology modelling and proceeded to dock several putative inhibitors to the target. This yielded 19 compounds that bound to Rv1712 with high binding affinity. Additional interaction analysis of the top ten compounds suggested that these compounds could be potential inhibitors of Rv1712.

EXAMPLE 2

Identification of the potential M. tuberculosis drug target Rv2421c

M. tuberculosis genes known to be involved in first-line and second-line drug resistance were used to query KEGG M. tuberculosis (H37Rv) strain database for metabolic pathway annotations (Table 4). The KEGG database produce metabolic pathway maps for genes queried, which is a molecular interaction/reaction network diagram generated computationally. The novel target Rv2421c was identified from among other targets located within the nicotinate nicotinamide metabolic pathway. The protein sequence of Rv2421c was downloaded from the TubercuList World Wide Web Server (http://genolist.pasteur.frl/TubercuList) and subjected to a BLASTp search against H. sapiens database (build GRCh37/hgl9) in order to determine if any sequence similarity was detectable between Rv2421c and human proteins. Using the protein sequence of Rv2421c a BLASTp search was performed against 24 Mycobacterium strain species to determine the degree of conservation of Rv2421c. A high degree of conservation suggest that mutations in this protein are not tolerated thereby prohibiting the spontaneous occurrence of drug resistance. Furthermore, the lack of homologs to host gut bacteria and mice proteins would facilitate the design of M. tuberculosis specific drugs and inform in vivo mouse infection studies. To this end, BLAST searches against three species of human gut flora bacteria (Staphylococcus aereus, Enterococcus faecalis and Escherichia Co!i) and mouse proteins (Mus musculus) were performed. The Tuberculist website was used to find relevant information regarding essentiality of Rv2421c and if its biological function is known. Additionally, the protein's accession number (Rv2421c) was used to query the Tuberculosis Database (Reddy et al., 2009) to obtain gene expression data collated from previous experiments. The rationale for this analysis was to determine if Rv2421c is up-regulated during latent conditions of M. tuberculosis infection suggesting that it would be a promising target to pursue as 90% of the world population is latently infected with M. tuberculosis.

Table 4: KEGG pathways for genes known to be associated with 1st and 2nd line drug resistance.

Drug Gene(s) Mutation Pathway

Rv1484 (inhA) -8 T/A and mycolic acid biosynthesis

Ser94Ala

Rv1908c (katG) Ser315Thr tryptophan metabolism and methane metabolism

INH (Isoniazid) 1st Rv3795 (embB) Met306lleu arabinosyl transferase

Rv2428 (ahpC) Ser315Thr glutathione metabolism

Rv2245 (kasA) Asp66Asn Not available

Rv1854c (ndh) Thr110Ala and oxidative phosphorylation

Arg268His

PZA (Pyrazinamide) 1st Rv2043c (pncA) Ala132Gly, nicotinate and nicotinamide

Thr114Met metabolism

RIF (Rifampicin) 1st Rv0667 (rpoB) Asp435Gly,

Leu452Pro,

He1106Thr,

Ser531Leu or His526Tyr purine metabolism,

pyrimidine

metabolism, R A

polymerase

Rv3795 (embB) Met306Val arabinosyl transferase

Rv3795 (embB) Met306lleu arabinosyl transferase

EMB (Ethambutol) 1st

Rv3793 (embC) Met306Val arabinosyl transferase

Rv3794 (embA) Met306lleu arabinosyl transferase Rv3266c (rmID) Not available nucleotide sugars metabolism, streptomycin biosynthesis, polyketide sugar unit biosynthesis

STR (Streptomycin) 1st Rv3919c (gidB) Indel Not available

Rv1483 (mapA, Ser94Ala mycolic acid biosynthesis

ETH (Ethionamide) 2nd fabG1)

Rv3854c (ethA, ethR) Not available aminobenzoate degradation

Rv0005 (gyrB) Not available Not available

FLQ (flouroquinolones) 2nd Rv0006 (gyrA) Ala90Val, Not available

Asp94Gly

The gene highlighted in bold is known to be involved in first-line drug resistance and resides in the nicotinate and nicotinamide metabolic pathway.

Abbreviations used: A-Adenine, Ala-Alanine, Arg-Arginine, Asn-Glutamine, Asp-Aspartate, Gly-Glycine, His-Histidine, indel-insertion deletion, lleu-lsoleucine, Leu-Leucine, Met- Methionine, N.A-Not available, Pro-Proline, Ser-Serine, T-Thymidine, Thr-Threonine, Tyr- Tyrosine, Val-Valine.

Homology modelling

A 3D model was built for Rv2421c as no crystallographic structure was available for the M. tuberculosis protein in the Protein Data Bank (PDB). The protein sequence of Rv2421c was used to search the PDB to identify suitable templates for homology modelling. The best possible templates were selected for model construction based on overall structural similarity and sequence identity over other similar structures. The software program MODELLER9V12 (Eswar et al., 2008) was used for 3D model construction. MODELLER9V12 generates 3D structures for proteins based on satisfaction of spatial restraints derived from the sequence alignment between target and template. MODELLER9V12 also allows for the calculation of quality assessment scores such as the lowest discrete optimized potential energy (DOPE) score and GA341 score. Usually models with the lowest DOPE score are considered more accurate models and were therefore used for further analysis.

Quality assessment of lowest DOPE score model

The lowest DOPE score model (LDSM) was selected to construct a residue- based pseudo-energy profile. The energy profiles generated for the model and templates can be used in a plotting program called GNUPLOTv4.2 to visually represent the energy functions. This is useful to identify regions of high pseudo- energy that correspond to errors in the model. To obtain a more reasonable estimate of the accuracy of the model we calculated the normalised DOPE Z-scores. Usually models with a DOPE Z-score of -1 and lower are considered as good quality structures. Stereochemical analysis was done using PROCHECK to determine if residues of the model were located in allowed regions of the Ramachandran plot according to the method of Laskowski et al 1993. The PROSAII online web-tool was used to assess the reliability of regions within the modelled target. This is important for accurate prediction of substrate binding to receptor regions. Finally, the 3D structural similarity between model and the four templates were assessed using the root mean square deviation (RMSD) value as calculated in PyMol (http://www.pymol.org). The lower the RMSD value, the more similar the structures are with respect to the backbone conformation.

Energy refinement and Molecular dynamics

The LDSM in complex with ligand nicotinic acid adenine dinucleotide (DND) was further refined using energy minimization steps. The energy minimization included 1000 steps of steepest descent and 5000 steps of conjugated gradients using GROMACS to remove close van der Waals force contacts. The molecular dynamic simulation of Rv2421 c was carried out using GROMACS 4.6.1 employing GROMOS 96 forcefield. The topology for DND was generated by uploading the atomic coordinates in PDB file format to the PRODRG server. Briefly, the system consisting of Rv2421c and ligand DND was solvated with SPC water molecules in a cubic box (1.2nm thick) and eight sodium ions was added to neutralize the negative charge of the system. The energy minimized system was then subjected to a short position restrained dynamics of 20ps and then a full 30ns molecular dynamics simulation without restraints. For the 20ps simulation the Berendsen temperature coupling method was used according to the method of Berendsen et al 1984, with constant coupling of 0.1 ps at 300K under conditions of position restraints (all-bonds). Electrostatic forces were calculated using Particle Mesh Ewald method. For the full 30ns simulation the same conditions was applied as used in the 20ps simulation except the v-rescale temperature coupling was implemented. The energy terms (Total and potential energy), root mean square deviation (RMSD), root mean square fluctuation (RMSF) analysis and radius of gyration was calculated to determine if the system reached convergence over the 30ns simulation.

Optimising docking parameters

The ligand DND was docked to the 30ns snapshot structure of Rv2421c using AutoDock Vina. The conserved active site residues were manually verified using multiple sequence alignments to other homologous proteins. The receptor and ligand structure DND was prepared using AutoDock tools which adds polar hydrogens, gasteiger charges and saves the output file in pdbqt format. This was done to correct for errors such as missing atoms, added H₂0, more than one molecule chain break and alternate locations. After receptor and ligand preparation the center of mass was calculated for the receptor structure with the ligand DND present. A python script center of mass.py was executed and returns X, Y and Z coordinates for the grid space. The various parameters for the docking process was stored in a configuration file. The configuration file contained input parameters for the docking simulation such as: center of mass coordinates, grid space and exhaustiveness of the search algorithm. AutoDock Vina produces nine binding modes for the ligand and ranks them according to energy values.

Virtual compound screening and Interaction analysis

The SMILE format entry of the ligand DND was used to search the ZINC database to identify structurally similar compounds. The search was done using 50, 80 and 90% identity settings for ZINC database compounds using DND. The structures identified with similar chemical and structural properties were downloaded from the ZINC database in sdf format and converted to pdbqt using Open Babel. A total of 669 compounds were identified for DND and each were docked to the snapshot structure of Rv2421c using previously defined docking parameters. We also screened a known inhibitor ZINC58655383 to determine if any of the compounds showed higher binding affinity than substrate DND and the known drug. The compounds that showed higher binding affinities than ligand DND to the snapshot structure were visually inspected in PyMol and were further analysed using Pose View and BINANA programs. PoseView determines four types of interactions namely; i) hydrogen bonds, ii) hydrophobic, iii) metal interactions and iv) π interactions, while BINANA was used to calculate all the afore mentioned interactions as well as salt bridges because this functionality was not present in PoseView.

Target Validation

The candidate Rv2421c was selected as a potential drug target because it satisfied the stringent selective criteria applied in this study. The BLASTp search of Rv2421c protein sequence against several Mycobacterium species indicated that 19 out of the 24 strain species had sequence identities above 70% and sequence coverage values of approximately 75-99% with expectation values ranging between (3X10-83 and 3X10-122). Several of these mycobacterial species are pathogenic to humans for example, (M. leprae, M. abscessus and M. ulcerans) allowing broad range drug design. Microarray data analysis revealed that Rv2421c is up-regulated when treated with arachidonic acid having a statistically significance p- value of 1X10^"6 making it an attractive drug target, especially to treat cases of latently infected tuberculosis. Arachidonic acid is one of the carbon sources which provide . tuberculosis with enough energy to survive during dormant conditions and after dormancy. To further validate the suitability of Rv1421c as a dmg target, a BLASTp search against three species of human gut flora bacteria (Staphylococcus aereus, Enterococcus faecalis and Escherichia coli) and mouse proteins (Mus musculus) was conducted. Rv2421c showed approximately 40% sequence similarity and 72 to 97% sequence coverage to all three host intestinal bacterial species, while no sequence homology was found to mice proteins. Although sequence similarity was detected to host gut bacteria there are conformational differences between the active site pocket that might allow for the design of drugs with a higher specificity for M. tuberculosis. The lack of homology to mouse proteins facilitates the use of mouse infection studies.

Template selection and model building

Searching PDB, with the protein sequence of M. tuberculosis Rv2421c produced four crystal structures, (PDB ID: 2QTR, 1YUM, 2H29 and 1 KAM). The sequence identity between Rv2421c and the homologous templates were 40% (2QTR), 35% (1YUM), 36% (2H29) and 40% (1 KAM), all having highly significant p- values (p < 10X10^"10). The multiple sequence alignment between Rv2421c and 2QTR, 1YUM, 2H29 and 1 KAM show that active site residues important for the recognition of the DND substrate is well conserved (Figure 12). The nucleotidyl transferase consensus sequence motif is highly conserved between the four templates and Rv2421c, residues H10, G12 and H13 crucial for dimerization of the enzyme. Fifty 3D models were constructed for Rv2421c using ODELLER9V12. The lowest DOPE score model (model 20) was selected for qualitative analysis and is visually represented using PyMOL (Figure 13). The secondary structural arrangement of Rv2421c includes 9 alpha helices and 5 Beta strands and 14 turns or coils (Figure 13). The overall structure of Rv2421c represents a compact folded structure with substrate DND located within the core of the protein suggesting a stable protein-ligand complex. Qualitative analysis of LDSM

The DOPE score profile of the Rv2421c LDSM corresponds to that of templates 2QTR, 1YUM, 2H29 and 1KAM (Figure 14), which indicates that the generated model is a reasonable approximation. The LDSM shows no regions of high energy and the trace profile (green line) corresponds to that of the templates indicating that the 3D model predicted for Rv2421c is close to its native conformation. The normalised DOPE Z-score for the LDSM was found to be -0.603 indicating reliability of the model. The LDSM for Rv2421c satisfied stereochemical restraints and passed criteria subjected to in PROCHECK located at SWISS-MODEL. The phi/psi angles distribution of residues within the LDSM of Rv2421c were 88.6% in most favoured regions and 1.8% in disallowed regions according to the Ramachandran plot (Figure 15). The Prosa Z-score of -6.49 for the LDSM of Rv2421c falls within range of experimental structures with similar lengths. The RMSD values were found to be less than 2A between the LDSM of Rv2421c and 1YUM (1.346A) and 1KAM (0.842A) than 2QTR (0.731 A) and 2H29 (0.765A). This indicates very little deviation from the main chain carbon atoms between Rv2421c and the four templates (1YUM, 1KAM, 2QTR and 2H29) suggesting homology and similarity between the structures.

Molecular Dynamics

Analysis of the Rv2421c in complex with substrate DND indicates a rapid increase in RMSD during the first 2500ps which then gradually decreases after 5000ps for both the protein's backbone and DND (Figure 16A). An equilibrium phase was reached within 5000ps suggesting that 30ns was sufficient for stabilizing the structure. The average total energy and the potential energy reaches convergence at at -1.47474e+06 and -1.80487e+06 KJ/Mole, respectively (Figure 16B). The RMS fluctuation analysis indicates three regions that are highly flexible within the protein these correspond to LI (72-75), L2 (127-130) and L3 (167-171), respectively (Figure 16C). The radius of gyration for the molecule became constant after 10000ps fluctuating around 1.56 (nm) suggesting that the protein structure have a stable surface structure suitable for docking studies (Figure 16D).

Virtual compound screening and interaction analysis

Screening the 669 compounds obtained from the ZINC database against the snapshot structure of Rv2421c yielded four compounds with higher binding affinities compared to DND and ZINC58655383 (5). These four compounds (ZINC94303216, ZINC08551088, ZINC85629877 and ZINC77285165) also showed a higher number of interactions compared to DND and ZINC58655383. The number of hydrogen bonds formed between docked molecules (ZINC94303216, ZINC08551088, ZINC85629877 and ZINC77285165) and residues of Rv2421c protein model were 7, 11 , 10 and 6 for each compound compared to the 8 of substrate DND and zero of known compound ZINC58655383 (Figure 17A, B, C and D, Table 5). The number of hydrophobic bonds formed between (ZINC94303216, ZINC08551088, ZINC85629877 and ZINC77285165) and Rv2421c were 2, 2, 2 and 3, respectively (Figure 17 A, B, C and D). However, only ZINC94303216 showed a higher number of salt bridges 2 compared to 1 of DND and zero of ZINC58655383 (Table 5).

Table 5: Residues forming interactions between Rv2421c and the top four compounds compared to DND and the known inhibitor (ZINC58655383).

residues.

() indicate the number of interactions formed between residue and the compound atom. Rv2421c as a potential drug target

The novel drug target Rv2421c is unique to bacteria, lacks a human homolog and is upregulated during dormancy conditions of Mycobacterium persistence, making it an attractive target for drug design. Rv2421c plays an important role in the transfer of phosphorous groups in both nicotinate and nicotinamide salvage and de novo pathways. The pncA gene coding for Rv2043c occurs in this pathway and is targeted by the Pyrazinamide (PZA) drug that is highly effective at killing persistent bacilli in the initial phase of TB therapy. Previous studies have indicated that pncA mutations conferred resistance to PZA. Successful inhibition of Rv2421c may therefore eradicate slowly growing persistent bacilli in TB infection. Furthermore, Rv2421c is a cytoplasmic protein which might expedite structure elucidation using experimental methods to assess the target's druggability.

Template search, model building and guality assessments

The amino acid sequence identity between Rv2421c and the four solved structures in the PDB ranged between 35-40%. As such these four structures could be used as modelling templates. The LDSM of Rv2421c satisfied all quality checks i.e. normalised DOPE Z-score, stereochemical restraints and PROSA Z-score. Moreover, superimposing the LDSM to the recently solved crystal structure (PDB ID: 3MLB) of the B. anthracis nicotinate mononucleotide adenylyltransferase did not indicate any structural differences with respect to the backbone conformation (Figure 18). We therefore believe that the LDSM for Rv2421c approximates the native protein structure of M. tuberculosis (Figure 13).

Molecular dynamics

The simulation of Rv2421c in complex with DND indicated that 30ns was sufficient for stabilizing the protein structure. This was supported by the fluctuating of the backbone RMSD values between 0.3 and 0.45 (nm) and DND between 0.2 and 0.25 (nm). Furthermore, energy analysis showed that the molecule had reached convergence throughout the simulation. The RMSF analysis indicated that three regions had large fluctuation values; however none of these flexible regions corresponded to DND binding site residues suggesting conserved importance of these active site residues. The radius of gyration for the molecule became constant after 10000ps suggesting that the Rv2421c-DND complex has a stable surface structure suitable for virtual screening and drug design. Virtual screening and interaction analysis

A total of four compounds showed higher binding affinity values compared to DND and ZINC58655383 docked to the 30ns snapshot structure of M. tuberculosis protein model Rv2421c. Additionally, performing interaction analysis provided support for the higher binding affinity values. The higher number of interactions observed namely; hydrogen bonds, hydrophobic and electrostatic interactions might contribute to the stronger affinity of these compounds for Rv2421c and in doing so potentially prohibit substrate DND from binding. Also, these compounds make interactions with a range of newly identified residues within the binding pocket of DND as well as known conserved binding site residues. Interestingly, the known inhibitor makes only interactions with three hydrophobic residues, while ZINC77285165 showed similar interactions and an additional five hydrogen bonds with Rv2421c residues. This makes ZINC77285165 a strong candidate for inhibition studies. We propose the use of the four compounds in experimental testing to determine if they are non-toxic to human cell lines and able to inhibit M tuberculosis growth.

Summary

We selected Rv2421c as a potential drug target because it has been shown to be essential for M. tuberculosis growth, has a known biological role, shares no homology to the human host, conserved between various Mycobacterium species and up- regulated during latent conditions of M. tuberculosis survival. Rv2421c maps to Nicotinate and Nicotinamide metabolic pathway, an essential pathway for M. tuberculosis survival during active and latent conditions. Furthermore, the 3D structure was predicted for Rv2421c and structurally verified using molecular dynamics simulations in complex with substrate DND and subsequently used for docking studies to identify potential lead compounds. We identified four compounds ZINC94303216, ZINC08551088, ZINC85629877 and ZINC77285165 that showed not only higher binding affinity but a higher number of interactions compared to DND and known drug ZINC58655383. We propose these four compounds as potential lead molecules that should be investigated experimentally for their inhibitory effects on M. tuberculosis growth and toxicity to human cells. EXAMPLE 3

Validation of Rv1311 as a potential M. tuberculosis drug target

The protein sequence of Rv1311 was downloaded from the Tuberculist website and subjected to a BLASTp search against H. sapiens database (build GRCh37/hgl9) in order to determine if any human homologs was present. Using the protein sequence of Rv1311 a BLASTp search was performed against 24 Mycobacterium strain species to determine the degree of conservation of Rv1311. A high degree of conservation suggests that mutations in this protein are not tolerated thereby prohibiting the spontaneous occurrence of drug resistance. Lack of homologs in mice proteins facilitate in vivo mouse infection studies and therefore BLAST searches were performed against mouse proteins (Mus musculus). The TubercuList was used to find relevant information regarding essentiality of Rv1311 for the growth of M. tuberculosis and if its biological function is known.

Homology Modeling

Ten (10) 3D models were built for Rv1311 as no crystallographic structure was available for the M. tuberculosis protein in the Protein Data Bank (PDB). The protein sequence of Rv1311 was used to search the PDB to identify suitable templates for homology modeling. The best possible templates were selected for model construction based on overall structural similarity and sequence identity over other similar structures. The software program MODELLER9V12 was used for 3D model construction. MODELLER9V12 generates 3D structures for proteins based on satisfaction of spatial restraints derived from the sequence alignment between target and template. MODELLER9V12 also allows for the calculation of quality assessment scores such as the lowest discrete optimized potential energy (DOPE) score and the statistical potential Z-score (GA341 score). Usually models with the lowest DOPE score are considered more accurate models.

Quality assessment of lowest DOPE score model

The lowest DOPE score model (LDSM) was used construct a residue-based pseudo-energy profile. The energy profiles generated for the model and templates can be used in a plotting program GNUPLOTv4.2 to visually represent the energy functions. This is useful to identify regions of high pseudo-energy that correspond to errors in the model. To obtain a more reasonable estimate of the accuracy of the model, the normalised DOPE Z-scores was computed. Usually models with a DOPE Z-score of -1 and lower are considered as good quality structures. Stereochemical analysis was done using PROCHECK to determine if residues of the model were located in allowed regions of the Ramachandran plot according to the method of Laskowski et al 1993). The PROS All online web-tool was also used to assess the reliability of regions within the modelled target. This is important for accurate prediction of substrate binding to receptor regions. Finally, the 3D structural similarity between model and the four templates were assessed using the root mean square deviation (RMSD) value as calculated in PyMol (http://www.pymol.org). The lower the RMSD value, the more similar the structures are with respect to the backbone conformation.

Energy refinement and molecular dynamics

The LDSM was refined using energy minimization steps. The energy minimization included 1000 steps of steepest descent and 5000 steps of conjugated gradients using GROMACS to remove close van der Waals force contacts. The molecular dynamic simulation of Rv1311 was carried out using GROMACS 4.6.1 employing GROMOS 96 forcefield. The topology for adenosine triphosphate (ATP) was generated by uploading the atomic coordinates of ATP in PDB file format to the PRODRG server. Briefly, two systems were generated consisting of apoenzyme Rv1311 (system 1) and the second Rv1311 in complex with ligand ATP (system 2) both were solvated with SPC water molecules in a cubic box of at least 1A in length. Ten sodium ions was added to system 1 and twelve to system 2 to neutralize the negative charge of the systems. Each system was then subjected to a short position restrained dynamics of 20ps and then a full 30ns molecular dynamics simulation without restraints. For the 20ps simulation the Berendsen temperature coupling method was used according to the method of Berendsen et al 1984, with constant coupling of 0.1 ps at 300K under conditions of position restraints (all-bonds).

Electrostatic forces were calculated using Particle Mesh Ewald method. For the full 30ns simulation the same conditions was applied except the v-rescale temperature coupling was implemented. The root mean square deviation (RMSD), root mean square fluctuation (RMSF) analysis and radius of gyration were calculated to determine if the system reached convergence over the 30ns simulation. Principle component analysis (PCA) was performed using GROMACS implementing the g_covar and g_anaeig functions. VMD was performed according to the method of Humphrey et al 1996, to visually inspect motions along the trajectory.

Interaction analysis

Interactions between the ligand (ATP) and the protein were analysed using the PoseView program. We took four snapshots of our system 2 over (15, 20, 25 and 30ns) time frames and analysed interactions made between ligand ATP and Rv1311 and compared it to the homologous template (2E5Y and ATP). PoseView determines four types of interactions namely; i) hydrogen bonds, ii) hydrophobic bonds between protein atom and ligand iii) metal interactions between ligand atom and metal ion and iv) TT interactions between aromatic rings of a protein and a ligand.

Target Validation

No homology was detectable between Rv1311 and human proteins (p > 0.0005), while the BLASTp search of Rv1311 protein sequence against several Mycobacterium species indicated that 19 out of the 19 strain species had sequence identities above 70% and sequence coverage values of approximately 99% with expectation values ranging between (2X10^"36 and 8X10^"54). Several of these mycobacterial species are pathogenic to humans for example, (M. leprae, M. abscessus and M. ulcerans) allowing for the design of drugs against these mycobacterial species. To further validate the suitability of Rv1311 as a drug target, a BLASTp search against mouse proteins (Mus musculus) was conducted. Rv1311 showed no sequence homology to mice proteins. The lack of homology to mouse proteins facilitates the use of mouse infection studies by M. tuberculosis to validate potential lead compounds.

Template selection and model building

A search of PDB with the protein sequence of Rv1311 identified four crystal structures, (PDB codes: 2E5Y from Thermophilic Bacillus, 2QE7 from Thermoalkaliphilic bacterium bacillus sp. ta2.al, 2RQ6 from Thermosynechococcus Elongatus BP-1 and 2RQ7 from T. Elongatus bp-1 f1). The sequence identities range from 34% (2RQ6, 2RQ7 and 2E5Y) to 44% (2QE7), expectation values range between 2x10^"8 - 1.1x10^"10. The MSSA between Rv1311 and 2E5Y (chain A), 2QE7 (chain H), 2RQ6 (chain A), and 2RQ7 (chain A) shows that the ATP-binding site motif [(l/D)DXXRA] within the C-domain is well conserved (Figure 19). The structurally conserved residues between the four templates and Rv1311 are I89, D90 and A94. These are crucial for catalytic activity of the enzyme (Figure 19). A total of ten 3D models were constructed for Rv1311 because sequence identity and overall sequence coverage between the target and template were high given the scores. The LDSM is visually represented using PyMOL (Figure 20). The secondary structural arrangement of LDSM for Rv1311 includes four alpha helices, 9 beta strands and 10 turns (Figure 20). The overall structure of Rv1311 is composed of two domains: an N- terminal β sandwich domain (NTD) and a C-terminal a-helical domain (CTD), well conserved between several bacterial species (Figure 20). The CTD consist of four a-helices and are folded into a hairpin (Yagi et al., 2010). The CTD domain contains the ATP-binding motif that is important for catalytic activity of the enzyme.

Qualitative analysis of LDSM

The DOPE score profile of the LDSM (model 6) for Rv1311 is comparable to that of the homologous templates 2E5Y, 2QE7, 2RQ6 and 2RQ7, with no distinct regions of high energy (Figure 21), indicating that the generated model is close to its native conformation. The normalised DOPE Z-score calculated for model 6 was - 0.32; an indication that the model was accurately predicted. The LDSM for Rv1311 satisfied stereo-chemical restraints subjected to in PROCHECK. The Ramachandran plot indicated that 88.7% of residues are in most favoured regions and 0.0% are located in disallowed regions. For the LDSM of Rv1311 the GA341 score was equal to 1 and the Prosa Z-score -3.52 was close to the template values ranging between - 4.42 and -5.75, providing support that, the 3D model for Rv1311 is close to the protein's native conformation. The RMSD values for the superimposition of the LDSM to the three templates were 2E5Y (0.137A), 2RQ6 (1.446A) and 2RQ7 (1.746A). All values were <2A, suggesting stereo-chemical similarity between target and template structures. Only template 2QE7 could not be aligned to the Rv1311 LDSM due to a considerable amount of mismatch errors.

Molecular dynamics

Analysis of the RMSD values for system 1 (Figure 22A), indicates that the protein's backbone atoms fluctuates considerably. It also showed rapid increases in RMSD values during the first 1500ps and then after 20000ps and thereafter remained constant after 26800ps for the remainder of the simulation (Figure 22A). On the other hand, the RMSD values for system 2 showed that, the C alpha backbone atoms reached a plateau after 2000ps (0.3-0.4nm) while the ligand ATP reaches equilibrium after approximately 6600ps fluctuating ~0.3nm (Figure 22A). The RMSD values of both systems suggest that system 1 is less stable than system 2; consistent with higher RMSD values and that 30ns was sufficient for stabilizing the protein structure of system 2. The RMS fluctuation for the total residues indicates that regions L1 (87-88), L2 (91-92), L3 (95-96), L4 (103-104), and L5 (105) represent higher RMSF values (0.3-0.6nm). The results also supports the RMSD values that these regions are the most flexible regions in the protein structure for system 1 (Figure 22B). While for system 2 RMS fluctuation analysis indicates three regions that are highly flexible within the protein, these correspond to L1 (91-93), L2 (95-96) and L3 (101-103), respectively (Figure 22B). These regions correspond to a loop region within the protein which is highly flexibly in the absence of ATP. The radius of gyration for system 1 and 2 are mirror images of each other. Both systems became constant after approximately 1000ps fluctuating between 1.35 and 1.4nm suggesting that the protein structure of both systems have a stable surface structure (Figure 22C).

ATP binding domain

In an attempt to elucidate any conformational changes of the Mycobacterium tuberculosis Rv1311 upon ATP binding, we superimposed snapshots (15, 25 and 30) of system 1 with system 2. Superimposing the two structure systems at different time frames provided RMSD values ranging between 1 and 2A suggesting very little deviation between structures (Figure 23A, B and C). However, visual inspection of the two systems using 15-30ns indicated large mobility of the loop region (residues 101-105) of the two systems suggested. At 15 and 25ns, system 1 had a more open conformation demonstrated by the large differences in the loop region (11.5A and 24.3A), while system 2 adopted a more closed conformation (Figure 23A and B). For 30ns, both systems showed to some extent a similarity in the loop region conformation thereby confirming high flexibility of the loop region for system 1 in the absence of ATP (Figure 23C). This analysis demonstrates the ability of ATP to induce major stability of this region upon binding potentially allowing enzyme catalytic activity. Our results are in agreement with those of presented in the study of Yagi ef al 2007, demonstrating that the CTD takes on two conformations depending on ATP- binding. Principal component analysis was performed to describe predominant movements of the two MD simulations. The first principal component described 54.1% and 36% of the motion for system 1 and 2, respectively (Table 6). The collective movement described by the first three principal components were 72% for system 1 and 60% for system 2 (Table 6). The results from the PCA analysis were similar to that of the superimposed snapshots where the loop region of system 2 takes on a narrow shape fluctuating between elongated and bent conformations (Figure 24). This suggests that ATP might contribute to stability of the bent topology of the protein structure. Table 6: The percentage motion in each of the first three, and total sum of the first three, principal components (PrCs).

Interaction analysis

To account for the dynamic nature of Rv1311 we performed interaction analysis between ATP and snapshots 15, 20, 25 and 30ns of system 2. Interaction analysis revealed that ATP makes interactions with the two known binding site residues Asp90 and Glu84 at snapshots 20 and 30ns (Table 7), while at 15ns ATP makes interactions with no known binding site residue interactions (Table 7). Although ATP makes only one known interaction with Glu84 at 25ns, it made in total seven hydrogen bond interactions with ATP more than 2E5Y. We propose the use of structures at snapshots 20, 25 and 30ns in docking studies as these structures are closer to the proteins active state and reproduces in some cases interactions with known ATP binding site residues.

Table 7: Interactions formed between snapshots of Rv1311 and 2E5Y with ATP.

Summary

We selected Rv1311 as a potential drug target because it has been shown to be essential for M. tuberculosis growth, has a known biological role, shares no homology to the human host and mice proteins and is conserved between various Mycobacterium species. Rv1311 maps to a known targeted metabolic pathway in M. tuberculosis namely; Oxidative phosphorylation, an essential pathway for M. tuberculosis survival. Its 3D structure has been predicted for Rv1311 and molecular dynamics studies were performed to identify features responsible for the stability of the structure. The structure in complex with ATP proved to be stable. We also identified the loop region as being highly flexible in the absence of ATP adopting an open conformation. Furthermore interaction analysis suggest the use of structural snapshots 20, 25 and 30ns of system2 for docking studies to identify inhibitors that would bind to the known binding site residues to avoid ATP binding and henceforth prohibit enzyme catalytic activity.

REFERENCES

1. Berendsen HJ, The Journal of chemical physics 1984, 81 :3684.

2. Hasan S, et al. (2006). PLoS Computational Biology, 2:e61.

3. Laskowski RA, et al. Journal of applied crystallography 1993, 26:283-291.

4. Yagi H, et al. (2007) Proceedings of the National Academy of Sciences 104:11233-11238.

Claims

1. A pharmaceutical composition comprising a compound selected from the group consisting of:

5-(4-amino-2-oxopyrimidin-1 (2H)-yl)-4-hydroxy-2-(hydroxymethyl)

tetrahydrofuran-3-yl dihydrogen phosphate;

2'-deoxycytidine 5'-monophosphate;

cytidine 5'-triphosphate disodium salt;

cytidine 5'-monophosphate disodium salt;

nicotinic acid mononucleotide;

flavin adenine dinucleotide disodium salt hydrate;

adenosine 5'-triphosphate, disodium salt hydrate; and

adenosine 5'-diphosphate monopotassium salt dehydrate,

or a pharmaceutically acceptable salt thereof, together with a pharmaceutically acceptable carrier.

2. The pharmaceutical composition of claim 1 , for use in the treatment of a tuberculosis infection.

3. The pharmaceutical composition for use of claim 2, wherein the tuberculosis infection is caused by Mycobacterium tuberculosis.

4. The pharmaceutical composition for use of claim 2 or 3, wherein the tuberculosis infection is selected from the group consisting of susceptible tuberculosis, multiple drug resistant tuberculosis and extensively drug-resistant tuberculosis.

5. The pharmaceutical composition for use of claim 3 or 4, wherein the compound selectively binds to an enzyme involved in a Mycobacterium tuberculosis metabolic pathway to form an enzyme-compound complex.

6. The pharmaceutical composition for use of claim 5, wherein the enzyme is selected from the group consisting of cytidyiate kinase (cmk), nicotinate mononucleotide adenylyltransferase (nadD) and ATP synthase epsilon chain (AtpC).

7. The pharmaceutical composition for use of claim 5 or 6, wherein the compound outcompetes a natural substrate of the enzyme, thereby interrupting the metabolic pathway.

8. The pharmaceutical composition for use of claim 2, wherein one or more chemical groups of the compound are modified to improve its efficacy.

9. The pharmaceutical composition for use of claim 8, wherein the compound is modified by incorporating a lipophilic cage to assist in transporting the compound to the cytoplasm of a cell.

10. The pharmaceutical composition for use of any one of claims 1 to 9 further comprising pharmaceutical excipients, diluents, carriers or other suitable additives.

11. The pharmaceutical composition for use of claim 11 , wherein the pharmaceutical composition is delivered to the cell by a delivery vehicle selected from the group consisting of: a cell penetrating peptide, a pH responsive carrier or an endosome-disrupting agent.

12. A method of treating tuberculosis infection in a subject, the method comprising administering an effective amount of one or more of the pharmaceutical compositions of any one of claims 1 to 11 to the subject.

13. The method of claim 12, wherein the subject is a mammal.

14. The method of claim 12 or 13, wherein the subject is a human.

15. Use of a compound selected from the group consisting of:

5-(4-amino-2-oxopyrimidin-1(2H)-yl)-4-hydroxy-2-(hydroxymethyl)

tetrahydrofuran-3-yl dihydrogen phosphate;

2'-deoxycytidine 5'-monophosphate;

cytidine 5'-triphosphate disodium salt; cytidine 5'-monophosphate disodium salt;

nicotinic acid mononucleotide;

flavin adenine dinucleotide disodium salt hydrate;

adenosine 5'-triphosphate, disodium salt hydrate; and

adenosine 5'-diphosphate monopotassium salt dehydrate, in the manufacture of a medicament for use in a method of treating tuberculosis infection in a subject.

16. The use of claim 15, wherein the tuberculosis infection is caused by Mycobacterium tuberculosis.

17. The use of claim 15 or 16, wherein the tuberculosis infection is selected from the group consisting of susceptible tuberculosis, multiple drug-resistant tuberculosis and extensively drug-resistant tuberculosis.

18. The use of any one of claims 15 to 17, wherein the compound selectively binds to an enzyme involved in a Mycobacterium tuberculosis metabolic pathway to form an enzyme-compound complex.

19. The use of claim 18, wherein the enzyme is selected from the group consisting of cytidylate kinase, nicotinate mononucleotide adenylyltransferase and ATP synthase epsilon chain.

20. The use of claim 18 or 19, wherein the compound outcompetes a natural substrate of the enzyme, thereby interrupting the metabolic pathway.

21. The use of claim 15, wherein one or more chemical groups of the compound are modified to improve its efficacy.

22. The use of claim 21 , wherein the compound is modified by incorporating a lipophilic cage to assist in transporting the compound to the cytoplasm of a cell.