Whole-Genome Sequencing of Mycobacterium tuberculosis Provides Insight into the Evolution and Genetic Composition of Drug-Resistant Tuberculosis in Belarus

Known resistance-conferring genotypes explained most phenotypic resistance to first-line drugs in MDR and XDR isolates.To better understand the genetic basis of drug resistance in MDR- and XDR-TB in Belarus, we sequenced 97 isolates that each came from a unique patient. This set included 32 MDR, 40 XDR, and 25 susceptible isolates (see Table 1 for phenotypic drug-resistant profiles). We called variants for each isolate using the H37Rv genome as a reference (Materials and Methods), searched variants against a list of known resistance-conferring mutations (given in Table S1), and compared results to each strain's antimicrobial susceptibility testing (AST) profile for first-line drugs (rifampin, isoniazid, ethambutol, and pyrazinamide). Additionally, we identified all mutations present in the genes searched described above, regardless of whether they were known to confer resistance or not (Data Set S2). Mutations known to confer resistance to rifampin and isoniazid were highly predictive of resistance to these two drugs (Table 2). While 97% of isoniazid resistance could be explained by a single mutation, katG S315T, six different rpoB mutations, including the common S450L, combined to explain 99% of the rifampin resistance in this data set.

The sensitivity for predicting ethambutol resistance was slightly lower (90%), with the primary resistance-conferring mutations in embB being M306I, M306V, and Q497R (Table 2). Of the seven isolates for which known ethambutol resistance-conferring mutations could not explain phenotypic resistance, two isolates carried other polymorphisms in embB that were found only within ethambutol-resistant strains (N296H and D354A; Data Set S2). Although these mutations were not included in our list, both mutations were recently implicated in ethambutol resistance in a study from France (6). Mutations known to confer resistance to pyrazinamide had the lowest sensitivity for predicting phenotypic resistance to this drug; only 16% of resistance could be explained when pncA loss-of-function mutations were included. However, when all 10 nonsynonymous mutations in pncA present in pyrazinamide-resistant isolates were considered, the sensitivity and specificity for detecting resistance rose to 100% (Table 3 and Data Set S2). All of these mutations have been previously implicated in pyrazinamide resistance (http://www.moleculartb.org/gb/pdf/doc/Moltb-tableau-pncA-17_12_14.pdf ), with the exception of pncA L151W, which was present in only a single isolate.

While most strains had resistance-conferring mutations that matched their first-line AST profiles, isolate XTB13-299 was phenotypically MDR but was genetically susceptible based on our list of known mutations. Examination of sequencing reads for XTB13-299 revealed low-frequency polymorphisms at known resistance-conferring sites within rpoB (5 of 237 aligned reads contained the rpoB S450W variant), the inhA promoter (2 of 266 aligned reads contained the inhA c-15t variant), and embB (1 of 229 aligned reads contained the embB M306I variant). These low-frequency mutations suggest that this patient might have been infected with a mixed population of M. tuberculosis, with the minority strain leading to the MDR phenotype.

We also evaluated strains for mutations in rpoA and rpoC, mutant forms that are known to compensate for reductions in fitness caused by rpoB mutations (7, 8). We identified 11 unique single nucleotide polymorphisms (SNPs) in rpoC and 2 SNPs in rpoA present only in rifampin-resistant isolates (Data Set S2). rpoC I491T was the most prevalent SNP cooccurring with the rpoB S450L mutation and was previously predicted to compensate for the rpoB S450L (Escherichia coli S531L) mutation (8). Similarly, most other potential compensatory mutations seen here (Data Set S2) were previously reported (7–11). The exceptions were rpoC E518D, E757G, and A1213E and rpoA G31S variants, which may represent novel compensatory sites in these genes.

Known resistance-conferring genotypes explained most phenotypic resistance to second-line fluoroquinolones and injectables in MDR and XDR isolates.Second-line treatment of TB for the patient cohort included an aminoglycoside/cyclic peptide, a fluoroquinolone, ethionamide, cycloserine, and para-aminosalicylic acid. Of all second-line drugs, fluoroquinolone (ofloxacin) resistance was most often (91%; Table 2) successfully predicted by known resistance-conferring mutations from our list. Seven different resistance-conferring polymorphisms were identified in gyrA, and two were identified in gyrB, the most common being gyrA A90V, D94A, and D94G. Known resistance mutations had a lower sensitivity for predicting resistance to injectables (aminoglycosides and cyclic peptides), ranging from 73% for kanamycin to 83% for capreomycin and 86% for amikacin. The most common mutation was rrs a1400g, which provides cross-resistance to all three aforementioned aminoglycosides and cyclic peptides. Some strains also had kanamycin-specific (eis promoter) or capreomycin-specific (tlyA loss of function) mutations.

Ethionamide resistance was poorly predicted in this data set; while the sensitivity was 100%, the specificity was only 42%. Loss-of-function mutations in ethA had a particularly low specificity of 59%. This effect was not driven by any single mutation but rather poor specificity across a wide range of loss-of-function mutations. No cycloserine resistance was initially predicted, as the only mutation on our list, loss of function of ald, was absent in all samples. However, 15 cycloserine-resistant strains harbored an alr L89R, a mutation that was previously identified in cycloserine-resistant clinical isolates from patients treated in South Africa (12). Although not yet experimentally validated, the addition of this mutation to our list increased sensitivity in this data set to 33%, with 100% specificity (Table 3).

Phylogenetic analysis suggests uncontrolled transmission of MDR-TB in Belarus.To better understand whether the spread of MDR- and XDR-TB in Belarus is driven by person-to-person transmission or de novo evolution of resistance within individual patients, we estimated a Bayesian phylogeny from SNPs predicted in the 97 isolates. The phylogeny revealed that all isolates fell into either the Beijing lineage (lineage 2) or the Euro-American lineage (lineage 4; Fig. 1). MDR isolates were significantly more likely to belong to the Beijing lineage than the Euro-American lineage (P < 0.03, Fisher's exact test), although XDR isolates did not show such a bias (P < 0.68; Fisher's exact test). The phylogenetic analysis also demonstrated that many MDR and XDR isolates clustered into four shallow well-supported (posterior probability >0.95) clades containing closely related strains (mean pairwise distance, 11 to 15 SNPs; Fig. 1, clades I, II, III, and IV). This close similarity among drug-resistant M. tuberculosis strains was comparable to the level of variation recently observed in a single patient in which drug resistance evolved over the course of treatment (13), suggesting an important role for the transmission of MDR- and XDR-TB among patients included in this study.

• Open in new tab
• Download powerpoint

FIG 1

Phylogeny and drug resistance diversity among M. tuberculosis isolated from patients in Belarus from 2010 to 2013. The phylogenetic tree was created using MrBayes and rooted between the Beijing and Euro-American lineages. Thicker branches indicate a posterior probability support of >0.95, and strains phenotyped as MDR, XDR, or resistant to all tested drugs are indicated by color at the terminal branches of the tree. Four closely related clades of MDR and XDR isolates are labeled (I, II, III, and IV). The results of computational spoligotyping (see Materials and Methods) are also shown along representative branches and nodes. A heat map is shown that illustrates, for each strain, the presence of known drug resistance-conferring variants detected in isolate genomes. Clear cells in this heatmap indicate that no known drug resistance variant was detected; black cells indicate that a high-confidence drug resistance variant was detected, and gray cells indicate that there was ambiguity in the variant call made at those positions. A red background indicates the isolate was phenotypically resistant, a blue background indicates the isolate was phenotypically sensitive, and a white background indicates the isolate did not undergo AST for that drug.

To quantify the relative importance of person-to-person transmission and de novo evolution of resistance within individual patients for each drug, we used parsimony and the data presented in Fig. 1 to estimate the number of times genotypic resistance evolved in isolates with AST results. For each drug, we divided this estimate by the total number of phenotypically resistant isolates to determine the fraction of resistance acquired through de novo evolution (Table 4). Our analysis revealed that most of the resistance to isoniazid and rifampin was acquired by patients through transmission (90% and 87%, respectively). This was not the case for second-line drugs. While we predicted that the majority (74%) of amikacin-resistant TB was transmitted to these patients, the majority (54%) of ofloxacin resistance was predicted to have arisen de novo within this population, suggesting a role for both transmission and intrahost evolution in the rising incidence of XDR-TB in Belarus.

While streptomycin was not given to patients as either a first- or second-line therapy during the course of treatment for this study, all MDR and XDR isolates harbored resistance to streptomycin through a number of known resistance-conferring mutations in rpsL, rrs, and gidB (Table 2). While we estimated that most streptomycin resistance was acquired through person-to-person transmission (Table 4), parsimony still predicted 10 independent instances of streptomycin resistance, suggesting that strains acquired this resistance through exposure to streptomycin through treatment prior to this study. As streptomycin was widely used for the treatment of TB in Belarus up to 2012, exposure of circulating M. tuberculosis strains to this drug could have frequently occurred.

Longitudinal samples highlight intrahost evolution and multiple infections.In addition to understanding how MDR- and XDR-TB are evolving and spreading among patients within Belarus, it is also critical to understand the genomic context of a TB infection when treatment fails. We, therefore, sequenced an additional 41 isolates from the original 97 isolates among 34 patients who maintained positive sputum cultures over time, i.e., their infections persisted despite treatment. These included 29 patients with TB that persisted but did not gain additional drug resistances over time and 5 patients with TB that became more resistant over time (Fig. 2).

• Open in new tab
• Download powerpoint

FIG 2

Genetic diversity and drug resistance in serial M. tuberculosis isolates. Individual samples from the same patient are shown as nodes connected by lines. Spoligotypes are indicated by node shape, and same-patient samples from the same or different spoligotype are connected by solid and dashed lines, respectively. Node colors correspond to phenotypic drug resistance. Samples are ordered by spoligotype and by patient number within each spoligotype, and mixed-spoligotype samples are presented last in the figure. The number of SNPs that differ between samples is indicated on the connecting lines. SNPs involving changes that are predicted to impact drug resistance are listed; patient 57 was initially fluoroquinolone resistant yet acquired a second fluoroquinolone resistance-conferring SNP. Percentages indicate prevalence. Asterisks indicate cases where no SNPs were identified between samples, but evaluation of read data indicated the presence of low-frequency variants conferring drug resistance. HIV+ patients are indicated by italics.

When we identified SNPs that differed across same-patient isolates, we observed that most patients (29 of 34) maintained strains that either remained genetically unchanged (0 SNPs) or showed a pattern consistent with intrahost evolution (1 to 4 SNPs) (Fig. 2). In three of these cases, a new phenotypic resistance was acquired, and in all of those cases, the resistance gained was to ofloxacin. However, only one case of phenotypic ofloxacin could be explained using our list of known resistance-conferring mutations (patient 84; Fig. 2). However, examination of sequencing reads revealed resistant subpopulations in the other two patients (61 and 89). In the sample from patient 89, the gyrA D94G mutation was detected at 5% frequency in the sequencing reads. In the sample from patient 61, known resistant alleles, gyrA A90V and D94A, comprised 36% and 42% of the sequencing reads, respectively, aligning across these sites. A third sample taken at least 6 months later from patient 61 showed approximately the same level of heterogeneity at these two positions (Fig. 2) indicating that ofloxacin-resistant and -susceptible subpopulations can coexist in a patient for many months. Daily fluoroquinolone was part of each MDR patient's treatment over the period when these samples were collected (Data Set S1).

Five patients showed a clear pattern of multiple infections rather than evidence for intrahost evolution of the same strain, having >150 SNP differences between isolates sampled at different time points (patients 26, 27, 32, 46, and 74; Fig. 2). In four patients, longitudinal isolates differed by over 800 SNPs and were part of distinct lineages, while the remaining patient isolates belonged to the Beijing lineage and differed by 189 SNPs. In three of the five cases, the earlier isolates were more drug susceptible than the later isolates, which were genotypically predicted to survive the patient's first-line treatment (treatment details are given in Data Set S1). In one of the remaining two cases with three time points, an MDR strain was present at the first and third time points, while the second sample was unrelated but also MDR, suggesting this patient may have been simultaneously harboring multiple MDR strains. In the final remaining case, the patient harbored two XDR strains at different time points. Strikingly, of the five patients showing a pattern consistent with multiple infections, four patients were HIV+ (of 5 total HIV+ patients in this study), representing a significantly greater likelihood of multiple infections in HIV+ patients than in patients not infected with HIV (P < 0.001, Fisher's exact test).