Second phase of the clinical trial of FS-1 against MDR-TB was carried out in Kazakhstan in 2015. Mtb strains have been isolated on a regular basis from patients’ sputum samples before the treatment beginning and during the therapy with FS-1 supplementing the standard regiment of antibiotic treatment of MDR-TB. The complex of anti-tuberculosis antibiotics included pyrazinamide (per os 40 mg/kg), cycloserine (per os 20.0 mg/kg), prothionamide (per os 20.0 mg/kg), capreomycin (intramuscular 20.0 mg/kg) and amikacin (intramuscular 28.0 mg/kg). FS-1 was administrated per os in 2.5 mg/kg in 30 min prior to antibiotics. Sputum samples were inoculated into liquid Löwenstein-Jensen medium (HiMedia Laboratories, India) and cultivated at 37°C for 8 weeks. Culture growth was controlled visually, by microscopy of Ziehl–Neelsen stained smears and by standard diagnostic biochemical tests including positive catalase, niacin and nicotinamidase activities, negative Tween-80 hydrolysis and susceptibility to sodium salicylate [20Bloch, H.; Segal, W. Biochemical differentiation of Mycobacterium tuberculosis grown in vivo and in vitro. J. Bacteriol., 1956, 72(2), 132-141.
[PMID: 13366889] ].

Susceptibility to antibiotics was tested by growing the cultures for 4 weeks at 37°C in test-tubes on solid Löwenstein-Jensen medium (HiMedia Laboratories, India; pH 6.8) supplemented with antibiotics in concentrations recommended by WHO [21World Health Organization. Interim Policy Guidance on Drug Susceptibility Testing (DST) of Second-Line Anti-Tuberculosis Drugs World Health Organization Document, 2008. WHO/HTM/TB/2008.392] and in the literature [22Krüüner, A.; Yates, M.D.; Drobniewski, F.A. Evaluation of MGIT 960-based antimicrobial testing and determination of critical concentrations of first- and second-line antimicrobial drugs with drug-resistant clinical strains of Mycobacterium tuberculosis. J. Clin. Microbiol., 2006, 44(3), 811-818.
[http://dx.doi.org/10.1128/JCM.44.3.811-818.2006] [PMID: 16517859] ]. The following antibiotics were used: isoniazid (0.2 μg/ml), rifampicin (40.0 μg/ml), streptomycin (4.0 μg/ml), ethambutol (2.0 μg/ml), amikacin (30.0 μg/ml), kanamycin (30.0 μg/ml), capreomycin (40.0 μg/ml), ofloxacin (2.0 μg/ml), cycloserine (40.0 μg/ml), ethionamide (40.0 μg/ml) and pyrazinamide (200.0 μg/ml on FAST-3L medium, BIOK, Russia; pH 5.0).

DNA samples were extracted from the Mtb isolates by the Cetyltrimethylammonium bromide (CTAB) method [23Murray, M.G.; Thompson, W.F. Rapid isolation of high molecular weight plant DNA. Nucleic Acids Res., 1980, 8(19), 4321-4325.
[http://dx.doi.org/10.1093/nar/8.19.4321] [PMID: 7433111] ]. The eluted DNA was quantified using the Qubit dsDNA BR Assay Kit (Life Technologies, USA). The samples were sequenced by Macrogen (South Korea) using the Illumina HiSeq 2000 paired-end sequencing technology. Produced read coverage was in the range of 800-1000. Raw DNA reads were deposited at EMBL-ENA under the reference numbers ERR1559736-43. The genome sequence of three clinical MDR-TB isolates M. tuberculosis SCAID 187.0 [24Ilin, A.I.; Kulmanov, M.E.; Korotetskiy, I.S.; Akhmetova, G.K.; Lankina, M.V.; Shvidko, S.V.; Reva, O.N. Complete genome sequence of multi-drug resistant clinical isolate Mycobacterium tuberculosis 187.0 used to study an effect of drug susceptibility reversion by a new medicinal drug FS-1. Genome Announc., 2015, 3(6), e01272-e15.
[http://dx.doi.org/10.1128/genomeA.01272-15] [PMID: 26543112] ], 252.0 and 320.0 were deposited in NCBI with accession numbers CP012506, CP016888 and CP016794, respectively. CLC Genomics Workbench 7.0.3 was used for sequence assembly and variant calling by the quality-based variant detection algorithm.

In the current work, 58,025 amino acid substitutions in 1,623 Mtb strains recorded in GMTV database [14Chernyaeva, E.N.; Shulgina, M.V.; Rotkevich, M.S.; Dobrynin, P.V.; Simonov, S.A.; Shitikov, E.A.; Ischenko, D.S.; Karpova, I.Y.; Kostryukova, E.S.; Ilina, E.N.; Govorun, V.M.; Zhuravlev, V.Y.; Manicheva, O.A.; Yablonsky, P.K.; Isaeva, Y.D.; Nosova, E.Y.; Mokrousov, I.V.; Vyazovaya, A.A.; Narvskaya, O.V.; Lapidus, A.L.; OBrien, S.J. Genome-wide Mycobacterium tuberculosis variation (GMTV) database: a new tool for integrating sequence variations and epidemiology. BMC Genomics, 2014, 15, 308.
[http://dx.doi.org/10.1186/1471-2164-15-308] [PMID: 24767249] ] were analyzed for their distribution in the global Mtb population. Information about drug associated mutations was obtained from the TB Drug Resistance Mutation Database [8Sandgren, A.; Strong, M.; Muthukrishnan, P.; Weiner, B.K.; Church, G.M.; Murray, M.B. Tuberculosis drug resistance mutation database. PLoS Med., 2009, 6(2), e2.
[http://dx.doi.org/10.1371/journal.pmed.1000002] [PMID: 19209951] ].

For every pair of SNP the multi-allelic linkage disequilibrium LD (1) and R² (2) were calculated [25Yang, J.; Shikano, T.; Li, M-H.; Merilä, J. Genome-wide linkage disequilibrium in nine-spined stickleback populations. G3 (Bethesda), 2014, 4(10), 1919-1929.
[http://dx.doi.org/10.1534/g3.114.013334] [PMID: 25122668] ]:

(1)

where k and l are the numbers of alleles for loci A and B,

(2)

Graphviz 2.24 and in-house Python scripts were used for grouping and visualization of functionally linked alleles. Complete genome sequences were aligned by Mauve [26Darling, A.C. Mau. B.; Blattner, F.D.; Perna, N.T. Mauve: multiple alignment of conserved genomic sequences with rearrangements. Genet. Res., 2004, 14(7), 1394-1403.
[http://dx.doi.org/10.1101/gr.2289704] ].

Linkage disequilibrium (LD) indicates the level of dependence between allelic states in different polymorphic loci on genomes. LD varies from 1 (non-random association of two mutations) to -1 (displacement of one mutation by another due to a functional or evolutionary incompatibility). Strong LD may be explained by a functional interplay between two genes, but also it may result from a functionally neutral inheritance of independent mutations from one common ancestor known as the genetic drift.

Many drug resistance mutations showed a strong negative LD in relation to each other. In Fig. (1), several drug resistance mutations with LD ≤ -0.9 were plotted on the circular chromosome of the Mtb type strain H37Rv. The most likely explanation for the negative LD is that the cumulated fitness cost associated with these mutations drastically decreased viability of the mutant strains. Top 20 drug resistance mutations showing significant incompatibility with other mutations are shown in Table 1.

Fig. (1) Location of drug resistance mutations on the chromosome of Mtb H37Rv are depicted by black cycles. Pairs of alleles showing strong negative LD are linked by lines. Open cycles depict several negatively displaced mutations in genes, which were not associated with drug resistance.

Table 1
Top 20 drug resistance alleles incompatible with other drug resistance mutations.

The top 20 polymorphic alleles negatively displacing other drug resistance mutations were localized in six genes encoding embAB arabinosyl-indolyl-acetylinositol synthase, pncA pyrazinamidase-nicotinamidase, fabG1 3-oxoacyl-[acyl]-carrier reductase, kasA 3-oxoacyl-[acyl]-carrier protein synthase, ethA monooxygenase, katG catalase-peroxidase-peroxynitritase and ndh NADH dehydrogenase. They showed strong negative LD links with other drug resistance mutations in inhA NADH-dependent enoyl-[acyl]-carrier-protein reductase; fbpC mycolyl transferase; fabD malonyl CoA-acyl carrier protein transacylase; efpA integral membrane efflux pump; Rv1772 and Rv2242 hypothetical proteins (isoniazid resistance); manB D-alpha-D-mannose-1-phosphate guanylyltransferase; rmlD dTDP-6-deoxy-L-lyxo-4-hexulose reductase, embC membrane indolylacetylinositol arabinosyltransferase and embR transcriptional regulator (ethambutol resistance); gyrA DNA topoisomerase subunit (resistance to fluoroquinolones); thyA thymidylate synthase (para-aminosalisylic acid resistance); rpsL 30S ribosomal protein and gid glucose-inhibited division protein (streptomycin resistance).

Positive LD may indicate synergetic relations between drug resistance mutations leading to an increased tolerance to one or multiple antibiotics. In general, the positive links between drug resistance mutations were not as strong as the negative displacements. Loci linked by LD ≥ 0.75 are shown in Fig. (2).

Fig. (2) Location of drug resistance mutations on the chromosome of Mtb H37Rv are depicted by black cycles. Pairs of alleles showing strong positive LD are linked by lines.

Mutations in four genes: embB, embC, embR and rmlD, all rendering the ethambutol resistance, showed strong positive LD links. These mutations formed the basis of multidrug resistance in Mtb due to their compatibility with many other drug resistance mutations. Other patterns of multidrug resistance were grouped around mutations in katG, kasA, fbpC and efpA.

Relations between drug resistance mutations and other polymorphic alleles, which could be possible compensatory mutations, were identified by R² linkage coefficients. The biggest cluster of co-evolved polymorphic alleles in Fig. (3) is centered round four drug resistance mutations in the 92^nd codon of gid (streptomycin resistance) and in the 229^th codon of accD6, 463^rd codon and 315^th codon of katG, all of them rendering the isoniazid resistance.

Black bars in Fig. (3) depict numerous polymorphic alleles in Mtb genomes showing positive correlations with the drug resistance mutations. Despite they not have been reported in the literature as drug resistance mutations, the strong R² relations of these polymorphisms with the canonical antibiotic resistance sites suggested that at least some of them may be of importance for rendering the drug resistance in Mtb and/or reducing the associated fitness costs.

This analysis showed that out of more than 1,000 known drug resistance mutations only a limited number of them are compatible with each other and even the compatible mutations depend on the allelic states of polymorphisms in many other genes.

In total, 236 MDR-TB isolates were obtained during the clinical trial of FS-1. Mtb cultures were isolated prior to the treatment and then on a weekly basis during the combined treatment course by the antibiotics and FS-1. A gradual reversion of drug resistance to a more sensitive phenotype was observed in series of isolates despite the selective pressure of antibiotics (Table 2). The treatment with FS-1 caused the reversion of drug resistance to aminoglycoside antibiotics (kanamycin, amikacin and capreomycin); ethambutol and cycloserine that was not the case in the control group of patients treated by the standard regiment. Selected Mtb isolates were sequenced to study genomic changes, which could explain the observed drug resistance reversion (Table 2). Complete genome sequences of the initial Mtb isolates from the patients #187, #252 and #320 were deposited in NCBI with the respective accession numbers CP012506, CP016888 and CP016794. Then these sequences were used as references for variant calling. Comparative analysis of the sequenced genomes (marked SCAID #) by Mauve alignment against several known Mtb genomes from NCBI revealed their belonging to the Beijing clade (Fig. 4).

Fig. (3) R2 co-evolutionary links between four drug resistance mutations conferring resistance to streptomycin (light gray oval) and isoniazid (dark gray ovals) and multiple other genomic polymorphic alleles (black terminal bars). Program Graphviz was used for this visualization.

Fig. (4) Phylogenetic relations between the clinical SCAID isolates used in this study and several selected reference Mtb strains.

Table 2
Drug resistance reversion in Mtb isolates from three patients with MDR-TB during the treatment with FS-1.

Beijing clade is characterized by an innate antibiotic resistance [27Merker, M.; Blin, C.; Mona, S.; Duforet-Frebourg, N.; Lecher, S.; Willery, E.; Blum, M.G.; Rüsch-Gerdes, S.; Mokrousov, I.; Aleksic, E.; Allix-Béguec, C.; Antierens, A.; Augustynowicz-Kopeć, E.; Ballif, M.; Barletta, F.; Beck, H.P.; Barry, C.E., III; Bonnet, M.; Borroni, E.; Campos-Herrero, I.; Cirillo, D.; Cox, H.; Crowe, S.; Crudu, V.; Diel, R.; Drobniewski, F.; Fauville-Dufaux, M.; Gagneux, S.; Ghebremichael, S.; Hanekom, M.; Hoffner, S.; Jiao, W.W.; Kalon, S.; Kohl, T.A.; Kontsevaya, I.; Lillebæk, T.; Maeda, S.; Nikolayevskyy, V.; Rasmussen, M.; Rastogi, N.; Samper, S.; Sanchez-Padilla, E.; Savic, B.; Shamputa, I.C.; Shen, A.; Sng, L.H.; Stakenas, P.; Toit, K.; Varaine, F.; Vukovic, D.; Wahl, C.; Warren, R.; Supply, P.; Niemann, S.; Wirth, T. Evolutionary history and global spread of the Mycobacterium tuberculosis Beijing lineage. Nat. Genet., 2015, 47(3), 242-249.
[http://dx.doi.org/10.1038/ng.3195] [PMID: 25599400] ]. Eight canonical drug resistance mutations were found in the three sequenced Mtb strains [8Sandgren, A.; Strong, M.; Muthukrishnan, P.; Weiner, B.K.; Church, G.M.; Murray, M.B. Tuberculosis drug resistance mutation database. PLoS Med., 2009, 6(2), e2.
[http://dx.doi.org/10.1371/journal.pmed.1000002] [PMID: 19209951] ]: GyrA_{Asp:94→Asn/Gly} and GyrA_Ser:95→Thr (FLQ); RpsL_Lys:43→Arg (SM); KatG_{Ser:315→Thr} and KatG_{Arg:463→Leu} (INH); AccD6_{Gln:141→Pro} (PZA); PncA_{Asp:229→Gly} (INH); EmbB_{Met:306→Ile} (EMB); and gid_Glu:92→Asp (SM). Despite of the classical drug resistance mutations in KatG, all three strains were catalase positive. All these strains were resistant to ethionamide (Table 2). In Mtb this resistance usually is conferred by mutations in ethA [8Sandgren, A.; Strong, M.; Muthukrishnan, P.; Weiner, B.K.; Church, G.M.; Murray, M.B. Tuberculosis drug resistance mutation database. PLoS Med., 2009, 6(2), e2.
[http://dx.doi.org/10.1371/journal.pmed.1000002] [PMID: 19209951] ]. In the strain SCAID 320.0 this gene was truncated in the mid by a frame shift mutation that probably made this gene completely nonfunctional in this strain. The strain SCAID 187.0 has a Leu:82→Pro substitution in ethA. It was not reported as an ETH resistance mutation, but it is located very close to other known mutations of this drug resistance. The strain SCAID 252.0 possesses an Ala:248→Asp substitution in the area of other drug resistance mutations within this gene; however, rendering ethionamide resistance by this mutation has not been reported either. Another mutation EthA_{Leu:190→Phe} became abundant in the population SCAID 252 on the 9^th week of the treatment with antibiotics and FS-1, but it disappeared in the later isolates. Canonical mutations in ethA are incompatible with many other drug resistance mutations (Fig. 1) that may explain their absence in these MDR-TB strains. Therapy with FS-1 prevented the accumulation of new potential drug resistance mutations in the SCAID 252 isolates. Accumulation of other mutations, which potentially could lead to drug resistance, was observed also in the isolate SCAID 187.9. Substitutions Phe:95→Leu in cyclohexanone monooxygenase Rv0565c and Arg:264→Gly in CydC cytochrome component transporter were found in 60% of corresponding DNA reads. Both these genes belong to potential prodrug activators [28Fraaije, M.W.; Kamerbeek, N.M.; Heidekamp, A.J.; Fortin, R.; Janssen, D.B. The prodrug activator EtaA from Mycobacterium tuberculosis is a Baeyer-Villiger monooxygenase. J. Biol. Chem., 2004, 279(5), 3354-3360.
[http://dx.doi.org/10.1074/jbc.M307770200] [PMID: 14610090] ]; however, no drug resistance mutations have been reported in these genes in the literature until now. No additional drug resistance mutations were observed in the SCAID 320 isolates.

An increased frequency of mutations was observed in subunits ppsACD of phenolphthiocerol polyketide synthase in all series of isolates. Frame shift indels were found in the genes ppsC and ppsD in the isolates 187.7 and 187.9 with the frequencies up to 60-100%. Substitutions His→Pro in proline rich linker regions between the acyltransferase and dehydratase domains were observed in the isolates 252.9 (30% of reads only in ppsC) and 252.12 (up to 60% of reads in ppsA and ppsC). In the isolate 320.8, this mutation was found in 50% of reads aligned against ppsA. The initial cultures 187.0, 252.0 and 320.0 were free from these mutations. Phthiocerol dimycocerosate is a characteristic cell wall compound of Mycobacteria and an important virulence factor [29Yu, J.; Tran, V.; Li, M.; Huang, X.; Niu, C.; Wang, D.; Zhu, J.; Wang, J.; Gao, Q.; Liu, J. Both phthiocerol dimycocerosates and phenolic glycolipids are required for virulence of Mycobacterium marinum. Infect. Immun., 2012, 80(4), 1381-1389.
[http://dx.doi.org/10.1128/IAI.06370-11] [PMID: 22290144] ]. However, it was recently published that impaired phthiocerol biosynthesis was associated with an enhanced ability to persist in host organisms [30Torrey, H.L.; Keren, I.; Via, L.E.; Lee, J.S.; Lewis, K. High persister mutants in Mycobacterium tuberculosis. PLoS One, 2016, 11(5) e0155127.
[http://dx.doi.org/10.1371/journal.pone.0155127] [PMID: 27176494] ].

Proline rich linkers are important for an appropriate spatial structuring of multidomain proteins [31Williamson, M.P.; Keren, I.; Via, L.E. The structure and function of proline-rich regions in proteins. Biochem. J., 1994, 297(Pt 2), 249-260.
[http://dx.doi.org/10.1042/bj2970249] [PMID: 8297327] ]. Introducing of extra proline residues may alter the efficacy of the phthiocerol biosynthesis in the same way as shift mutations; however, it has not been proved experimentally for these genes yet.

Another commonality of the studied series of isolates was an increase of genetic heterogeneity of the populations during the treatment course (Fig. 5). Total numbers of polymorphic loci (Y axis on the graphs) and frequencies of alternative alleles were counted. The polymorphic loci were grouped into 3 categories: rare mutations (0-20% of reads with alternative alleles), frequent mutations (20-35%) and common polymorphisms (35-50%). While the differences in numbers of polymorphic loci were not so great, it should be accounted that the used sequencing technique had summarized mutations in billions of sampled individual chromosomes, i.e. the graphs in Fig. (5) showed the general trends in accumulation of the most common polymorphisms leaving alone multiple individual mutations. Majority of these mutations were found in hypervariable PE-PGRS genes and in non-coding sequences except for those discussed above.

Fig. (5) Numbers of polymorphic loci in Mtb isolates. Black bars: rare mutations, Grey bars: frequent mutations, White bars: common polymorphisms.