Study Population

MTBC isolates were collected during January–June 2015 in the clinical laboratory of Odessa Regional TB Hospital (now Odessa Regional Center for Socially Significant Diseases; Odessa, Ukraine), where most clinical MTBC cultures from the region are routinely isolated and analyzed. We completed subculturing and WGS on a subset of culture isolates obtained at the clinical laboratory during the study period. The sampling strategy aimed to include a similar proportion of MDR/XDR TB isolates and non-MDR (polyresistant and pansusceptible) strains. We included samples on the basis of routine phenotypic rifampin and isoniazid susceptibility testing that categorized isolates as MDR or non-MDR TB strains; otherwise, inclusion was unbiased and isolates were included directly from the clinical pipeline when human and technical resources were available. Isolates from both new and retreatment cases were eligible for inclusion; retreatment cases included relapse, reinfection, failure of treatment, or interrupted treatment episode. In accordance with the clinical protocol in the laboratory, all MTBC cultures that were not included in our study were autoclaved and disposed.

Figure 1

Figure 1. Locations where Mycobacterium tuberculosis complex (MTBC) DNA samples were collected in Odessa and Vinnytsia regions. Yellow dots indicate locations of patients infected with a MTBC lineage 2 Ukraine outbreak strain. Green...

Our study was mainly set in the Odessa region, which had the highest TB burden (3,039 cases) in Ukraine, the second highest MDR TB burden (30.5% for new cases and 46.9% for retreatment), and TB incidence twice as high as the country’s average (127.9 cases/100,000 population) in 2017 (16). In total, we included 186 MTBC isolates, each obtained from 1 patient; most were living in the Odessa region (95.2%). The rest of the samples (4.8%) were from the bordering administrative region of Vinnytsia (Figure 1). Inclusion of these samples supported investigation of epidemiologic links between both regions. Thus, the analysis is focused on the differences between 103 non-MDR and 74 MDR/XDR MTBC strains all isolated from southern Ukraine. Patient characteristics assessed for association with MDR/XDR TB were age, sex, residence, HIV status, and case definition.

Contact Tracing of Patients Infected with a Ukraine Outbreak Strain

We conducted a retrospective review of epidemiologic contact tracing for patients infected with a Ukraine outbreak strain through patient records. Because Ukraine lacks resources for contact tracing and TB is stigmatized, we anticipated that the journal information would yield few contacts. To assist with contact identification, we interviewed physicians providing care for the patients with a Ukraine outbreak strain of TB. We maintained data confidentiality in accordance with regional ethics approvals in Ukraine and Umea, Sweden (N 2012-292-31M). We immediately shared information regarding nosocomial transmission of Ukraine outbreak strains with the administration of Odessa Regional TB Hospital to support infection control measures.

Phenotypic Characterization of Drug Resistance and DNA Isolation

We performed primary isolation and drug-susceptibility testing of MTBC strains using clinical diagnostics protocol for TB in Ukraine. In brief, we treated sputum samples with the N-acetyl-L-cysteine-NaOH (NALC) method for decontamination and further processed them for Ziehl-Neelsen staining using GeneXpert-MTB/RIF (Cepheid, https://www.cepheid.com), followed by inoculation on Löwenstein–Jensen (LJ) medium and BACTEC MGIT960 tubes (Becton Dickinson, https://www.bd.com) for culture confirmation and subsequently for phenotypic drug susceptibility testing. We used the following drug concentrations (mg/L, by medium): isoniazid, LJ 0.2, MGIT960 0.1; rifampin, LJ 40.0, MGIT960 1.0; ethambutol, LJ 2.0, MGIT960 5.0; pyrazinamide, MGIT960 100.0; and streptomycin, LJ 4.0, MGIT960 1.0. Individual reagents were not always available for routine drug-susceptibility testing, and resistance profiles in this manuscript are based on molecular markers (Appendix 1 Table 1, https://wwwnc.cdc.gov/EID/article/26/3/19-0525-App1.xlsx). We extracted mycobacterial DNA from LJ cultures as previously described (17) using cetyltrimethylammonium bromide.

Whole-Genome Sequencing

We completed WGS using Illumina technology (MiSeq, NextSeq 500, and Nextera XT library preparation kit; Illumina, https://www.illumina.com) according to the manufacturer’s instructions. We mapped raw read data (FASTQ files) to the M. tuberculosis H37Rv genome (GenBank accession no. NC_000962.3) using Burrows-Wheeler Aligner maximal exact matches (18) and refined mappings with the Genome Analysis Toolkit software package (19). We detected variants, including single-nucleotide polymorphisms (SNPs) and insertions and deletions (indels), with Samtools mpileup (20). For a concatenated sequence alignment, the basis for the phylogenetic reconstruction, we considered only SNPs with minimum thresholds of 4 reads in both forward and reverse orientation, 4 reads calling the SNP with a Phred score >20, and 75% SNP frequency. We excluded consecutive SNPs for the phylogenetic reconstruction detected within <12 bp, which can occur as artificial variants around indels and which would inflate the genetic distance of individual isolates. We combined all remaining SNP positions that had a clear base call for all strains and matched the threshold levels in >95% of all strains in one FASTA alignment, further excluding repetitive regions and resistance-associated genes.

Phylogenetic Reconstruction

We calculated maximum-likelihood trees with FastTree version 2.1.9 (21) using a general time-reversible nucleotide substitution model and 1,000 resamplings. The consensus tree was rooted with the midpoint root option in FigTree version 1.4 (http://tree.bio.ed.ac.uk/software/figtree). We obtained cladograms of outbreak strains and number of branch-specific mutations by maximum parsimony using BioNumerics version 7.6 (Applied Maths, https://www.applied-maths.com).

We inferred phylogenetic lineages from specific SNPs based on Coll et al. (22) and Merker et al. (8). As proxy for TB cases associated with direct transmission events, we considered a maximum pairwise genetic distance between >2 MTBC isolates of 5 SNPs, proposed by Walker et al. as a strict threshold that identified cases associated with household transmission (23).

Genotypic Drug Resistance Prediction

We extracted polymorphisms from 37 drug resistance– and bacterial fitness–associated genomic targets (Appendix 1 Table 1). We excluded known and newly identified phylogenetic, non–resistance-related variants (8,24,25) from the genotypic drug resistance prediction (Appendix 1 Table 3).

Resistance genotypes were defined as follows: for wild-type alleles (H37Rv reference sequence) for all resistance-associated targets we inferred antimicrobial susceptibility (genotypic wild type, gWT). We classified isolates with unknown mutations as genotypic non–wild-type (non-WT) with no further classification. We considered isolates with known resistance-mediating mutations to be resistant to the respective antimicrobial drug (Appendix 1 Table 1). Using genotypic resistance predictions, we classified the isolates as XDR (known mutations mediating resistance to isoniazid, rifampin, >1 second-line injectable drug, and a fluoroquinolone), MDR (isoniazid and rifampin resistance but not XDR), and non-MDR (either pansusceptible or any resistance but not MDR).

Statistical Analysis

We created an association plot in R version 3.3 (https://www.r-project.org) using the vcd package version 1.4. The underlying assumption is that the proportion of MDR/XDR MTBC strains within 1 group resembles the proportion of this group among all analyzed isolates (i.e., the resistance level is independent from the phylogenetic groupings). We calculated Pearson residuals to measure the departure from independence from each cell; values of +2 represent significant deviation at α = 0.05 and values of +4 deviation at α = 0.001 (26). We prepared a box plot of pairwise genomic distances of MTBC strains from defined strain groups with R version 3.3 using the ggplot2 package version 2.2 (Appendix 2 Figure 1).

We analyzed factors associated with MDR/XDR TB by logistic regression (univariate and multivariate model) using the glm function in R version 3.3.1 (27). We excluded the variable MTBC genotype from multivariate analysis because cell counts included 0. We compared means of pairwise SNP distances with 1-way analysis of variance (ANOVA) and Bonferroni multiple comparison tests.

Study Population

During the study period, TB service identified 1,026 patients with >1 culture–positive specimen in the Odessa region. Of these cultures, 330 isolates were identified as MDR.

Our study included 16.17% (169/1,026) of the identified positive cultures. MDR/XDR TB specimens constituted 21.82% (72/330) of all MDR/XDR TB–positive cultures identified during the study period. The remaining 97 isolates were phenotypically non-MDR MTBC strains (50 pansusceptible and 47 monoresistant or polyresistant strains). In addition, we randomly collected and analyzed 6 pansusceptible and 2 MDR/XDR MTBC isolates from TB patients registered in the bordering region of Vinnytsia (Vinnytsia Regional TB Hospital, Vinnytsia).

After completing WGS, we withdrew 4.8% (9/186) of MTBC isolates from analysis because of discrepancies between phenotypic drug susceptibility testing and the applied genome-based drug resistance prediction. Therefore, we included 177 MTBC clinical isolates in analysis.

MDR/XDR MTBC Beijing Outbreak Clades

Figure 2

Figure 2. Maximum-likelihood phylogeny based on 10,339 SNPs, and employing general time-reversible substitution model for 177 clinical MDR/XDR and non-MDR Mycobacterium tuberculosis complex isolates from southern Ukraine. Branches are color-coded according to the...

We classified a total of 177 clinical isolates based on WGS data from 10,339 variable positions, according to a recently proposed SNP barcode (22). Furthermore, we classified 89 of the isolates (50.3%) as Beijing genotype (lineage 2.2.1) and the remaining 88 (49.7%) as Euro-American lineage 4. We differentiated lineage 4 strains to the following genotypes and sublineages: 8/88 (9.1%) Ghana (lineage 4.1); 32/88 (36.4%) H37Rv-like (lineages 4.7 and 4.8); 16/88 (18.2%) Haarlem (lineages 4.1.2 and 4.1.2.1); 21/88 (23.9%) LAM (lineages 4.3.3, 4.3.4.1, and 4.3.4.2); and 11/88 (12.5%) Ural (lineage 4.2.1) (Figure 2).

Within Beijing lineage 2.2.1, we identified 3 closely related subgroups that exhibited a lower genetic diversity compared with other lineage 2.2.1 strains, based on intragroup pairwise SNP distances between any 2 strains. Two of those subgroups were previously reported in other settings (8); we identified them by clade-specific SNPs, and termed them Central Asia outbreak and Europe/Russia W148 outbreak. The third subgroup, named Ukraine outbreak, had not been previously described. These Beijing clades showed a lower median pairwise SNP distance of 25 (interquartile range [IQR] 20.0–32.5) for Central Asia outbreak, 39 (IQR 25.0–48.0) for Europe/Russia W148 outbreak, and 14 (IQR 11.5–18.0) for Ukraine outbreak, contrasting with a distance of 122 (IQR 170.8–139.0) for other Beijing strains (p<0.05, one-way ANOVA, Bonferroni multiple comparison test) (Appendix 2 Figure 1).

When we considered a strict threshold of 0–5 SNPs, as previously proposed for MTBC strains isolated from household contacts (23), only 9.0% of all patients (16/177) could be linked to 7 putative transmission networks comprising 2–4 patients each. We found these networks or molecular clusters among the 3 Beijing outbreaks (3 transmission events) and other Beijing strains (1 transmission event), and also among Euro-American strains (lineages 4.1 and 4.8; 3 transmission events) (Appendix 1 Table 1).

Contact Tracing Review for Beijing Ukraine Outbreak Cases

Epidemiologic links between clustered patients were partially confirmed by retrospective contact tracing review of patients infected with the Ukraine outbreak strain; of 18 cases, 7 were new and 11 retreatment (Figure 1; Appendix 1 Table 4; Appendix 2 Figure 2). Ukraine outbreak cases were registered in several places (Figure 1); 3 patients had contact through a psychiatric hospital (patient isolate odir-1606, with pairwise SNP distance of 8; odir-1746, SNP distance 10; odir-1747, with distance of 12) and 2 cases (odir-5192 and odir-1636, SNP distance 3) had close contact through their immediate neighborhood (Appendix 1 Table 4; Appendix 2 Figure 2). The neighborhood contact had also been detected by applying the strict SNP threshold. Three other case-patients had a close family MDR TB contact, but bacterial DNA was not available for these cases. No apparent connection could be established among 13 of 18 Ukraine outbreak cases.

Patient Factors Associated with MDR/XDR TB

To pinpoint certain demographic and treatment-related factors associated to MDR/XDR TB in southern Ukraine, we further used logistic regression analysis (Table 1). After multivariate analysis, 3 factors—previous TB treatment (OR 3.5, 95% CI 1.0–12.8; p = 0.04), living in a city (OR 6.0, 95% CI not applicable; p = 0.005), and infection with a Beijing outbreak strain (OR 7.4, 95% CI 1.8–29.9; p<0.001) —were significantly associated with MDR/XDR TB. Multivariate analysis included the additional variables age, gender, HIV status, and MTBC genotype.

Transmitted and Acquired Drug Resistance among Beijing Outbreak Strains

Figure 3

Figure 3. Association plot comparing expected and observed numbers of MDR/XDR and non-MDR strains from different phylogenetic groups. The colors and bar heights reflect the Pearson residuals, the width of the boxes is...

Consistent with results from the phylogenetic and logistic regression analyse, Beijing outbreak strains were the main carriers of MDR/XDR TB. The Europe/Russia W148, Central Asia, and Ukraine outbreak strains accounted for 29.9% of the cohort, but they contributed to more than two thirds (68.9%) of all MDR/XDR TB cases (Figure 3; Appendix 1 Table 2). The 3 Beijing outbreak strains all harbored high rates of first-line drug resistances to isoniazid (93%–100%), rifampin (93%–100%), streptomycin (93%–100%), ethambutol (86%–100%), and pyrazinamide (75%–100%) (Table 2). In contrast, first-line drug resistance rates among all other Beijing and non-Beijing strains were reduced (8%–33% for individual drugs). Similarly, resistance rates to second-line drugs among the 3 outbreak strains were high (ofloxacin 14%–44%, kanamycin 14%–50%, and prothionamide 57%–100%) compared with other Beijing (ofloxacin 11%, kanamycin 17%, and prothionamide 11%) and non-Beijing (ofloxacin 3%, kanamycin 8%, and prothionamide 15%) strains (Table 2).

An association plot confirmed the results from the logistic regression analysis that demonstrated that MDR/XDR TB among patients in southern Ukraine is clearly linked to the 3 Beijing outbreak clades and that drug resistance is not equally distributed among the MTBC strains, as one would expect from random treatment failures (i.e., acquired resistance) (Figure 3). For example, the Ukraine outbreak strain, constituting 10.2% of the total cohort, was associated with 24.3% of all MDR/XDR TB cases. In contrast, non-Beijing strains, constituting 49.7% of the cohort, were associated with only 17.6% of all MDR/XDR cases.

Figure 4

Figure 4. Maximum parsimony phylogenies of 3 Mycobacterium tuberculosis complex Beijing clades in Ukraine: Europe/Russia W148 outbreak (A), Ukraine outbreak (B), and Central Asian outbreak (C). Mutations mediating drug resistance to different antibiotics...

Strains from the 3 outbreaks demonstrated clade-specific drug resistance–related mutations, mainly to first-line drugs (Figure 4). All Ukraine outbreak strains shared identical mutations mediating resistance to isoniazid (katG S315T, inhA −15c/t), prothionamide (inhA −15c/t), rifampin (rpoB S450L), streptomycin (rpsL K88R), and ethambutol (embA −12 c/t and embB Y334H), suggesting that patients have been infected primarily with a strain already resistant to at least these 5 drugs. On the other hand, mutations in the pncA gene conferring pyrazinamide resistance were diverse, indicating individual and more recent drug resistance acquisition under selective pressure (e.g., failing treatment regimens) (Figure 4). Similar patterns can be observed for the other 2 outbreaks: shared first-line resistance mediating mutations; unique mutations mediating resistance to kanamycin (eis −8 del c, −10 g/a, −12 c/t, −14 c/t, −15 c/g, −35 g/a, −37 g/t); cross-resistance to all injectable drugs (rrs 1401 a/g); and resistance to the fluoroquinolones (gyrA A90V, S91P, D94G, D94N, D94Y, D94A) (Figure 4; Appendix 1 Table 1). We did not identify any known resistance marker for cycloserine and linezolid (Appendix 1 Table 2).

Strains of the 3 outbreak clades acquired putative compensatory mutations, e.g., in rpoB or rpoC previously suggested to mitigate the growth deficit of rifampin-resistant strains (10,11). These mutations occur either jointly or after the rifampin resistance mediating mutation in the rpoB rifampin resistance determining region (Figure 4).