Genomic characterization of MDR/XDR-TB in Kazakhstan by a combination of high-throughput methods predominantly shows the ongoing transmission of L2/Beijing 94–32 central Asian/Russian clusters

Isolates from a total of 630 patients living in Kazakhstan (435 men, 195 female) were included in the study. Patient data were only available for patients recruited in the Kazakh National Tuberculosis Reference Laboratory. New cases of tuberculosis accounted for 35.6% of the cases (224 of 630) and 64.4% (n = 406) were previously treated patients. Among 630 patients with pulmonary tuberculosis 69% (n = 435) were men and 23% (n = 145) were women, 50 (8%) were unknown. The age of TB patients varied from 16 to 80 years, with an average age of 37 years.

Public health action taken as a result of notification and surveillance is one of the Public Health National Centre for Tuberculosis of Kazakhstan key roles as stated by the Ministry of Health Act and subsequent Government directives which provide a mandate and legislative basis to undertake necessary follow-up. Part of this follow-up is identification of epidemiological and molecular links between cases, helped by the National Center for Biotechnology that provides expertise and links to the Ministry of Research. This study is part of service development carried out under this framework, and as such explicit ethical approval was not mandatory. Moreover patient data are not traceable by any other third part than by the National Centre for Tuberculosis.

Clinical isolates of MTB were collected from ten regional TB dispensaries (Atyrau, Mangistau, Aktobe, Kostanay, Pavlodar, Semey, Kyzyl-Orda, Zhambyl, Taldykorgan and Almaty) and two prisons run by the State-established committee for the management of the penitentiary system of the Karaganda region and the committee for the management of the penitentiary system of the Akmola region. Isolates were collected from all major regions of Kazakhstan, representing the territory of Kazakhstan. Accordingly, our samples characterize the lineage prevalence of tuberculosis in Kazakhstan as a whole. The high level of sampling from retreated cases however makes it inappropriate to evaluate drug-resistance prevalence in the general population. Clinical isolates of M. tuberculosis were also deposited in the Reference Laboratory of the National Center of Phtisiopulmonology for surveillance studies of anti-TB drug resistance in Kazakhstan. After primary isolation on Löwenstein-Jensen (LJ), mycobacteria were subcultured on the same medium in regional laboratories. Species identification as M. tuberculosis and drug susceptibility testing were performed at the Reference Laboratory by standard procedures [21]. Rifampicin and isoniazid susceptibility tests were carried out on LJ medium containing 40 μg/L rifampicin, 0.2 mg/L isoniazid, or 1 mg/L isoniazid using the absolute concentration method according to the World Health Organization (WHO) recommendations.

DNA was extracted by thermolysis. Briefly, colonies were pelleted, culture medium was discarded, and colonies resuspended in TE (10 mM Tris-HCl, 1 mM EDTA, pH 7.0). Samples were heated to 95 °C for 45 min. The suspension was centrifuged at 15,000 g for 1 min to pellet the cell debris. The supernatant containing the DNA was harvested and transfer into a new tube. The stocks were stored at − 20 °C until further use, or diluted at 1:50 into sterile water in a new tube. Seven hundred DNAs were shipped to France and to the Netherlands for further studies in two shipments (June 2015, n = 93, August 2015, n = 607).

A PCR-based agarose gel method was used to analyze the presence/absence of 0–2 copies of IS6110 in the NTF locus of the L2/Beijing isolates DNAs to discriminate between modern and ancient L2/Beijing [24]. Moreover, a microsphere-based method of this original protocol was also developed and confirmed the gel-based results (results not shown, method to be described elsewhere).

A PCR-RFLP analysis specifically targeting the sigE polymorphism that characterize a mutation specific of the 94–32 L2/Beijing cluster (Central Asian/Russian, including the K1 strain from Uzbekistan) was also implemented [18, 25, 26]. The targeted mutation, when present, creates a double band of 38 and 37 bp by creation of a new AluI site in codon 38 through a CTG - > CTA mutation.

We applied the latest whole-genome sequencing-based (WGS) taxonomical assignation in 7 lineages (L1-L7) complemented by sublineages (i.e. L4.3 for LAM) when a correspondence between spoligotyping-based naming and NGS-based naming was found [27‐30]. VNTR clones are designated according to [18, 25, 26] and L2/Beijing sublineages designation was done according to [17]. Of note, in this regard, TB-SNPID performs better than spoligotyping by sub-classifying L2/Beijing isolates in six subgroups based on SNPs and regions of deletions (RD105, mutT2–58, fbpB-238, pckA-1119, acs-1551, RD131). Conversely spoligotyping allows a finer classification of L4 isolates: at least ten accepted sublineages whereas TB-SNPID assigns only four L4 sublineages (L4.1.1/X, L4.3/LAM, L4.1.2/Haarlem, L4/Euro-American others). Interestingly, TB-SNPID also correctly identifies the previously spoligotyping-misclassified T signatures, now designated as “LAM-RUS” [31, 32].

QGIS v3.0 (www.​qgis.​org, Girona version) was installed on Mas OSX (Apple, Cupertino, CA); Free Kazakhstan shapefiles were downloaded from http://​www.​diva-gis.​org (www.​gadm.​org, version 2.5, July 2015). Available published results as well as our results were recorded into the required csv file format [33, 34]. Geographical coordinates of main cities in Kazakhstan were downloaded by Google® search and and Latitude and Longitude were added to the csv files. Apple pies charts were built following QGIS user’s manual.

Chi-squared test or Fisher’s exact test were used for all statistical results presented in the study; they were either run using Excel® formulas or using an on-line available calculator (https://​biostatgv.​sentiweb.​fr).

TB-SPRINT and TB-SNPID were respectively performed in France and in the Netherlands; each location was blinded to the results generated in the other location. Definitive results from both centers were then compared and cross-checked. One hundred percent identification concordance was obtained on the first studied set of 93 samples (cf. Additional file 1: Table S4, columns E and Q) and 99.3% of concordant subspecies identification was found at the end of the study with only two unexplained discrepancies. Intra-species typability was better by spoligotyping 96% typability (23/700 failed) than by TB-SNPID 86.6 typability (73/543 failed). The L2/Beijing represented 79% (536/677 = 79.1% by TB-SPRINT; SIT1 n = 529, SIT190 n = 5, SIT260 n = 1, SIT1674, n = 1; and 374/470 = 79,5% using TB-SNPID on an partly independent sample set). The total number of isolates identified as non-L2 by the two methods was around 22% (96/470 = 20.4% by TB-SNPID and 24.2% =164/677 by TB-SPRINT).

The most prevalent L4 sublineage was L4.3/LAM (n = 51), that includes the so-called LAM-RUS sublineage (previously T1 or T5-RUS2 in spoligotyping-classification) [31, 32, 35]; the second non-L2 most prevalent lineage was the L4.2.1/URAL genotype [36] (n = 39, labeled as “Euro-American Other” by TB-SNPID). Five isolates were labeled as “Haarlem” (L4.1.2) by TB-SNPID. By spoligotyping, twelve spoligotypes not reported in the SITVIT database were found, among which 4 were also identified as bona fide L4 isolates by TB-SNPID (others: ND). At least three of these non reported spoligotyping patterns were likely linked to represent mixed infection.

No L2.1 (proto-Beijing) nor L2.2.2 (Asia Ancestral 1) were found based on fbpB-238 results in TB-SNPID. As expected, no L2.2.1.1 (Pacific RD150) was found in Kazakhstan based on acs-1551 SNPs. All L2/Beijing identified isolates were further shown to belong to the L2/Modern subgroup (Fig. 1) by the gel-based IS6110-NTF method, confirmed by an NTF-IS6110 microsphere-based assay (results not shown). L2 was further split by TB-SNPID into: Beijing SA+ (n = 163 or 34%), Beijing V+/CHIN+/K2 (n = 16 or 3.4%), Beijing (V-) (n = 73 or 15.5%), Beijing K1 (n = 22 or 4.6%). The correspondence between this designation and the current accepted L2 sublineage nomenclature is shown in Fig. 1, for instance K1 isolates belong to the L2 sublineage 8 from Shitikov et al. [17]. Other L2 epidemic clonal complex are found within these two sublineages [19].

Simultaneous precise identification and predictive drug-resistance typing is of great clinical and epidemiological interest. Typeability on drug-resistance conferring mutations was 76% for TB-SPRINT, and 86.3% by TB-SNPID. Among the set of samples run by the two methods (n = 391), we detected 257 MDR by TB-SPRINT and 247 by TB-SNPID and 75 fully susceptible isolates by TB-SPRINT and 67 by TB-SNPID. All other isolates showed various resistance genotypes (see Additional file 1: Tables S1, S2, S5). MDR prevalence of the isolates in this study lies at 60% (n = 273/470 i.e. 58.1% by TB-SNP-ID; n = 344/515 i.e. 66.8% by TB-SPRINT); full susceptibility around 20% (93/240 = 19.9% and 100/515 = 19.4% respectively); monoresistance to INH around 15% (86/470 and 67/515 = 13%) and RIF monoresistance around 2% (18/470 and 4/515). These figures do however not represent actual resistance prevalence as our sampling was strongly biased in favor of retreated cases (64.4% of the total isolates).

Based on TB-SPRINT, 62.3% L2/Beijing isolates (320/421) were MDR whereas only 4.6% non-L2/Beijing isolates were (24/94). The statistical significance of this result is very high (odds ratio = 9.19 IC₉₅ [5.39, 16.13], p < 2.2 10^− 16), showing that L2 lineage and MDR-TB are strongly associated in Kazakhstan. Table 2 gives a detailed distribution of resistance patterns according to Lineage and sublineage: Beijing SA+ and V+/CHIN+/K2 were more likely to be MDR than Beijing K1 and Beijing −/−/−/− (p-value = 1.38 .10^− 7, Fisher’s exact Test, Additional file 1: Table S8). L4.3 showed a non-significant trend to be more likely MDR as compared to L4.1.2/Haarlem and L4 -EA-other.

RIF resistance associated mutations occurred mostly in rpoB codon 531 (60%; 273/470 and 320/515). Other RIF resistance associated mutations concerned positions 522 (around 3%; 13/470 by TB-SNPID; not assessed by TB-SPRINT) and 526 (around 2.7%; 11/470 and 14/515) and 516-GTC (around 1%; 4/515 by TB-SPRINT, not assessed by TB-SNPID). Some double mutations were detected (14/515 samples analyzed by TB-SPRINT; Additional file 1: Table S1). No mutations were detected in rpoB at codon position 176.

Regarding INH resistance associated mutations, the katG 315 S > T mutation was found in more than 70% of the isolates (348/470 and 385/515), and inh − 15 mutation in around 5% of the samples (34/470 and 5/515). Some double mutations were detected (15/515 samples analyzed by TB-SPRINT; Additional file 1: Table S1).

EMB resistance as associated mutations in embB 306 were found in half of the samples (228/470). Second-line resistance associated mutations in relation to SLID and FQ resistance were detected in decreasing order: the eisG-10A mutation (72/470), rrs1401 (34/470), gyrA94 mutations (30/470), gyrA90 mutations (22/470), eisG-14 T (7/470; Additional file 1: Table S2).

To get a deeper insight on L2 clones circulating in Kazakhstan, we further undertook complementary genomic analysis on a total of n = 356 DNAs (seven L4 as controls, three “failed” by TB-SNPID, and 346 L2/Beijing isolates). We used the recently described SigE SNP characterization method which was shown to be specific of the 94–32 Central Asian/Russian L2/Beijing cluster [18, 25, 26]. We obtained 314 mutants out of 346 L2-isolates characterized (90%), which demonstrates the presence of a limited number of epidemic clones belonging to the Beijing 94–32 cluster. We show that at least two different L2/Beijing epidemiologically linked clusters were circulating in the prison in Stepnogorsk, with at least respectively 20 cases (94–32 cluster) and 8 cases (L2/Beijing second cluster), whereas in Aktobe, a single drug-sensitive L2 clone was circulating during the studied time-frame. This clonal complex is encompassed in the TB-SNPID L2/Beijing/−/−/−/ label (Additional file 1: Table S7).

We then performed a thorough georeferencing using a geographic information system (GIS), synthesizing TB outbreak information of a 15-year period (2001 to 2015). Results are shown in Fig. 2. In the 2005 study published by T. Kubica et al. on 150 patients recruited in 2001 in 9 provinces of Kazakhstan, only 91 isolates could be georeferenced (Fig. 2a). L4 looks quite prevalent at that time compared to L2/Beijing (Fig. 2, left, top), especially in the south-east of the country (Almaty). In the 2015 study published by Y. Skiba et al. on a total of 152 patients recruited in 2008 in 9 provinces of Kazakhstan, a trend towards an increase of the L2/Beijing prevalence can be noticed, with a still high prevalence of L4 in Almaty (Fig. 2a, central). However in our study, performed on a sample of 632 patients in 13 regions, the prevalence of the blue lineage (L2/Beijing) outnumbers the L4, at least in some regions (Fig. 2a, bottom), including Almaty. A statistical analysis demonstrates that in Almaty, the proportion between L2 and L4 is not homogeneous (Chi2 = 52,5; p < 0.001; DF = 2); there is a recent higher prevalence of L2 compared to L4, the same is true in Kostanay (however in that setting, only two time points were available, in 2001 and in 2010–2015). No other statistically significant results were obtained regarding tuberculosis lineage distribution evolution. Of note in the Kyzyl-Orda region where L4 seems to increase between the 2008 and 2010–2015 period, the trend was not statistically significant (see Additional file 1: Table S7).

In Fig. 2b, we retrospectively analyzed the evolution of the prevalence of the L2/Beijing epidemic clones in Kazakhstan, as they had been characterized by different methods: by IS6110-RFLP in the first study, by 24 VNTR typing in the second study, and by combination of various markers (including RD131 and SigE SNP polymorphism) in our study. Figure 2b shows that a limited number of clones (likely IS6110-RFLP clusters 4 and 6 in the 2005 article and Beijing 94–32 cluster in the 2015 article) were already circulating in Kazakhstan between 2001 and 2008 (Fig. 2b, top and central). A further comparison of the IS6110-RFLP K1-K2 patterns obtained in a study run in Uzbekistan (isolates obtained during 2001–2004) showed that some IS6110-RFLP clusters found in Kazkakhstan were highly similar if not identical to the IS6110-RFLP patterns described simultaneously in the Uzbekistan study (Additional file 2: Figure S3) [37]. We performed preliminary typing on 8 VNTR on a limited set of L2/Beijing DNA and obtained results that were similar to those obtained by Skiba et al. 2015 confirming the high prevalence of the 94–32 cluster (results not shown). Finally, the last map of Fig. 2b (bottom), provides the SigE polymorphic results, and shows a clear geographic picture of L2/Beijing 94–32 MDR and XDR transmission that suggests an increase of the transmission of the 94–32 L2/Beijing Central Asian/Russian clusters in Kazakhstan.

In this study, TB-SPRINT and TB-SNPID showed some discrepancies regarding the presence of mutations conferring RIF or INH resistance. Part were intrinsically linked to differences in assay design as shown in Table 1. A thorough discrepancy analysis was undertaken on the first set of 93 isolates (Additional file 1: Table S4). This set of samples showed that only one true and unexplained discordance (red color in Additional file 1: Table S4) was found between the two NAATs. The second discrepancy analysis on all 391 results detected 34 discrepancices between the two NAATs (91.3% of congruence). We did not re-run discrepant results because of lack of budget. Full results are available in Additional file 1: Tables S5 and S6.

Phenotypic DST MDR status was almost always correlated with the NAAT results (n = 9 discrepancies). For six cases, there was a partial agreement on INH resistance, for three others on RIF resistance. Our assays showed altogether limited discrepancies between phenotypic and genotypic DST, in relation to their simple design compared to the complexity of drug resistance emergence and adaptation mechanisms [43]. Discrepancies observed in this study between phenotypic and genotypic DST may be due to errors in phenotypic DST or in genotypic DST since budget was limited to rerun some samples. Infections with mixed resistance genotypes in the patient population may account for another proportion of these discrepancies [44]. Even if phenotypic DST is often still considered as the reference standard, there are inherent difficulties linked to this method, that require quality assurance procedures to track potential errors [45]. In United Kingdom, the switch from phenotypic to genotypic DST is a new reality [46].

Both TB-SPRINT and TB-SNPID methods have advantages and inconveniences: one advantage of TB-SNPID is the “one tube/one run” concept on the MagPix® whereas TB-SPRINT is more suited to Luminex 200® or FlexMap 3D® users. Compared to WGS, TB-SPRINT and TB-SNPID are targeted assays that detect what they are designed to, i.e. a remaining number of samples will always need to be sequenced in particular to discover new mutation responsible of drug resistance. This provides the possibility to adapt these assays to detect specific genotypes. In contrast, WGS is a universal genotyping approach [43]. We recently reported second-line drug resistance assays in Kazakhstan using the 18-Plex “TB-EFI” that targets mutations in eis, gyrA, rrs, embB [23]. Such an assay can be run for a very limited cost in settings where MDR and preXDR-TB is present. Other assays such as the Deeplex® from Genoscreen (Lille, France) are other alternative solution to NGS [47]. Predictive WGS-based bioinformatical pipelines represent an important step forward for TB drug-resistance containment and these methods will progressively spread even if they remain quite expensive (200 US$ per isolate) [48] and considering that bioinformatical pipelines are in evolution [49]. Reagent costs are 20 Euros/sample for TB-SNPID and 12 euros/sample for TB-SPRINT. A raw economic analysis suggest a doubling cost to produce final results using european standard labor costs; these costs are lower in emerging economies. Hence, our assays are cheaper and easier to implement on large numbers of samples than NGS studies for any national, population-based, drug-resistance evaluation study.

Using both general markers provided by the two NAATs and more specialized markers (RD131 in TB-SNPID, SigE run independently) we identified that main circulating clones belong to the L2-Beijing Central Asian/Russian 94–32 sublineage. Further characterization could be performed using the 24 specific SNPs identified by the recent in-depth WGS-based analysis of L2/Beijing 94–32 cluster [18].

We also showed that some but not all of these 94–32 clones were associated with an MDR genotype: the K1 clone was in contrast linked to susceptibility [37]. We now have a better picture of drug-resistance mutations, and of L2 and non L2 sublineages circulating in Kazakhstan.

The L2 taxonomy still needs reconciliation between “ancient” knowledge on IS6110-RFLP, VNTR, RDs, and “modern” WGS acquired results to achieve a unified, deeper, and consensus taxonomical definition of L2 epidemic clones [17, 26]. In addition to purely phylogenetic markers, other polymorphic markers should be investigated to better understand the epidemic success of various clones. Markers of particular interest are undoubtedly adaptive compensatory mutations such as those acting on rpoA and rpoC genes that lower or eliminate the fitness-cost of the drug resistance mutation [19]. An improved global and local knowledge of drug-use history is also of paramount importance to understand the chronology of drug mutation emergence [2, 43]. Looking for epistasis between SNPs, could also be of added value to understand why the 94–32 cluster has been so successful [50].

Achieving improved global tuberculosis control and patient treatment simultaneously is now feasible thanks to WGS or high-throughput SNPs assays. Technology spreading by training should create favorable conditions for sustained laboratory and bioinformatics capacity building or strengthening, and could further be applied to other infectious threats [51, 52]. Genomic assays allow clinical microbiology to change from targeted to nontargeted investigations [53]. However one important issue is the trend towards closed and expensive diagnostics systems development by industrials, that can if not handled carefully contribute to de-skilling and less democratization in health systems. This can generate inequalities due to unaffordable and unsustainable costs as well as creating new local perceptions in populations concerning global health issues if simultaneously linked to austerity policies [54, 55]. Given the strongly regionalized phylogeographical structure of the TB outbreaks, a “one-size fits all” strategy for MDR-and XDR-TB control may not be the optimal solution; as demonstrated in Swaziland, the prevalence of rare drug resistance mutations such as the rpoB I491F, may result in missed MDR-TB cases and inadequate TB treatment and infection control [56]. Global economic issues also remain with respect to sustainability and long-term routine work for WGS-based assays in emerging economies, even if lot of resources are spent to train bioinformaticians [57]. That is the reason why, in the post-genomic era, other locally developed less sophisticated assays, based on and complementary with NGS results, may become important in resource-limited countries [18, 58‐60]. Furthermore, the availability of large amounts of WGS data will allow the precise election of genotypes to be screened. They would in turn help to develop second-generation assays that could rapidly be applied to very large number of samples.