Whole genome sequencing of Mycobacterium tuberculosis isolates and clinical outcomes of patients treated for multidrug-resistant tuberculosis in Tanzania

Background Tuberculosis (TB), particularly multi- and or extensive drug resistant TB, is still a global medical emergency. Whole genome sequencing (WGS) is a current alternative to the WHO-approved probe-based methods for TB diagnosis and detection of drug resistance, genetic diversity and transmission dynamics of Mycobacterium tuberculosis complex (MTBC). This study compared WGS and clinical data in participants with TB. Results This cohort study performed WGS on 87 from MTBC DNA isolates, 57 (66%) and 30 (34%) patients with drug resistant and susceptible TB, respectively. Drug resistance was determined by Xpert® MTB/RIF assay and phenotypic culture-based drug-susceptibility-testing (DST). WGS and bioinformatics data that predict phenotypic resistance to anti-TB drugs were compared with participant’s clinical outcomes. They were 47 female participants (54%) and the median age was 35 years (IQR): 29–44). Twenty (23%) and 26 (30%) of participants had TB/HIV co-infection BMI < 18 kg/m2 respectively. MDR-TB participants had MTBC with multiple mutant genes, compared to those with mono or polyresistant TB, and the majority belonged to lineage 3 Central Asian Strain (CAS). Also, MDR-TB was associated with delayed culture-conversion (median: IQR (83: 60–180 vs. 51:30–66) days). WGS had high concordance with both culture-based DST and Xpert® MTB/RIF assay in detecting drug resistance (kappa = 1.00). Conclusion This study offers comparison of mutations detected by Xpert and WGS with phenotypic DST of M. tuberculosis isolates in Tanzania. The high concordance between the different methods and further insights provided by WGS such as PZA-DST, which is not routinely performed in most resource-limited-settings, provides an avenue for inclusion of WGS into diagnostic matrix of TB including drug-resistant TB.


Background
Tuberculosis (TB), caused by the Mycobacterium tuberculosis complex (MTBC), is the leading cause of mortality, killing about 2 million people each year globally [1]. The highest TB mortality and morbidity occurs in low and middle income countries like Tanzania [1,2]. HIV infection is the most important single predictor of TB incidence across the African continent [3]. This is critical because HIV/AIDS is likely to increase in the risk of progression TB infection by 30 times, due to impairment on the immune system [3]. Sub-Saharan Africa carries a disproportionate burden of HIV, accounting for more than 70% of the global burden of infection [4]. For instance, in 2017, nearly 2.5 million people who contracted TB lived in sub-Saharan Africa, and 665,000 of them died from the disease [5]. The recent introduction of the Xpert®MTB/Rif assay (Cepheid, USA) has shown an increase in detection of drug resistance TB patients and the detection of multidrug-resistant TB (MDRTB) has increased three-to eight-fold compared to conventional testing [6]. Despite the gaps in documentation of MDR-TB cases in several Sub-Saharan African countries (SSA), pooled analysis that involved several studies reported a prevalence of 2.1% of MDR-TB in new cases, signifying a low prevalence of MDR-TB cases in SSA [7]. The relatively low prevalence of MDRTB in SSA might be attributed by the recent introduction of rifampicin in Africa, by the use of rifampicin-free treatment regimens in the continuation phase (during months three to eight), by the growing use of directly observed treatment as recommended under the directly observed treatment, short course (DOTS) strategy, and by the use of fixeddose combination tablets in a few countries [8,9].
Tanzania is among African countries south of Sahara with the highest burdens of TB, with an estimated 295 TB cases per 100,000 adults. The National antituberculosis drug resistance survey conducted in 2010 in Tanzania found the resistance to any of the four firstline drugs in new patients to be 8.3%, while the prevalence of MDR-TB was 1.1%. However, the crude prevalence for any resistance and for MDR-TB in retreated cases increased to 20.6 and 3.9% respectively [10].
The emergence of resistant strains of MTBC to antituberculosis drugs like rifampicin (RIF), isoniazid (INH), pyrazinamide (PZA), ethambutol (EMB), streptomycin (SM) among others; and the emergence of HIV in the 1980s led to resurgence of TB [11]. Multidrug resistant TB (MDR-TB), defined as resistance to at least RIF and INH, is usually caused by MTBC strains that harbour mutations in rpoB, katG and inhA genes associated with RIF and INH resistance. Globally, MDR-TB claims over 580,000people's lives annually [12]. About 10% of MDR-TB isolates exhibit resistance to fluoroquinolones (FQs) and second-line-injectable drugs (SLIDs) (aminoglycosides/ cyclic-polypeptides), referred to as extensive drug resistant TB (XDR-TB), which challenges clinical management [2,13,14]. Patients with XDR-TB are infected with MTBC that have mutations in rrs, tlyA or eis promoter region (SLI related) and gyrA or gyrB genes (FQ related). Clinical management of both MDR-TB and XDR-TB is very complex and there are different treatment regimens [15]. About 50 and 70% of patients treated for MDR-TB and XDR-TB have unfavourable treatment outcomes, respectively. Although XDR-TB is rare in Tanzania, cohort review reports of patients treated for MDR-TB show that treatment success is only 75%. However, control and prevention measures should be strengthened in the country to reduce MDRTB related morbidity and mortality. To improve treatment outcomes and control escalation of MDR/XDR-TB; early, rapid and accurate diagnostic methods are required to detect and decipher susceptibility profile of the MTBC to anti-TB drugs in TB endemic settings such as Tanzania.
As in other TB endemic settings, Tanzania has deployed rapid molecular methods for dual detection of MTBC and susceptibility to either RIF alone or along with INH and second-line injectable and fluoroquinolones. For example, while the Xpert®MTB/Rif assay (Cepheid, USA) detects MTBC and provides information about susceptibility to RIF only [16], the genotype MTBDRplus and MTBDRsl assays (Hain Life sciences, Germany) detect mutations that are strongly associated with multiple types of MDR-TB and also XDR-TB respectively [17,18]. These assays have potential to guide implementation of the new WHO shorter regimen for treating MDR-TB [15,19]. Unlike Whole genome sequencing (WGS), these PCR-probe based assays have limited anti-TB drug-susceptibility value especially if mutations occur outside the target region [20]. In recent years, WGS platforms have become an alternative diagnostic in addressing probebased assay limitations. WGS can identify genotypes predictive of drug-resistance phenotype within the entire region of microbial genome and has potential to determine genetic relatedness and identify transmission dynamics necessary in guiding clinical decisions. There is a need to sequence as many MTBC strains as possible, build libraries of single nucleotide polymorphisms (SNPs) and other variants, compare the relatedness of MTBC strains, and correlate variation with clinical progress or outcomes as a whole. Based on this, we performed WGS for the first time on isolates sourced from patients treated for MDR-TB at Kibong'oto Infectious Diseases Hospital (KIDH) in Tanzania. These data will provide a baseline set of the types and variations in TB in Tanzania. Clinical and laboratory information such as treatment regimen, phenotypic susceptibility profile, prior history of treatment with first line anti-TB drugs, HIV status, culture conversion rate and treatment outcomes were available for analysis.

Baseline demographic and clinical characteristics of study participants
A total of87study participants with positive MTBC culture results were included in the genetic study of whom, 47 (54%) were females. The median age was 35 (Interquartile range = 29-44) years (Table 1). Twenty (23%) were co-infected with HIV and had a median CD4 counts of 246 (IQR =119-388) cells/mm 3 . Table 2 summarizes demographic and clinical characteristics of TB patients. We found association between previous history of TB treatment and phenotypic drug resistance (OR, 0.01 95% CI: 0.0003-0.0995, p = < 0.001) ( Table 2).

Analysis of mutations associated with phenotypic drug resistance to anti-TB drugs
Of the 87 MTBC isolates sequenced, 57 (65.5%) had at least one mutation in a gene that was predictive for 7 anti-drugs ( Table 3). The concordance between phenotypic DST and WGS was 97% for RIF (DST: WGS; 26:28 isolates), 81% INH (DST: WGS; 40:51 isolates) and 95% for SM (DST: WGS; 8:11 isolates) (kappa = 1.00). Moreover, the concordance between Xpert® MTB/RIF (Cepheid, USA) and WGS for detection of RIF resistance was 95% (8:11 isolates) (kappa = 1.00). Substitution of serine to threonine at codon 315 (S315T) of the katG gene accounted for 50 (94.3%) of mutations ( Table 3). The common mutation associated with phenotypic resistant to RIF was S450L (substitution of serine to leucine) accounted for 96% (26/27) of the detected RIF resistance. Of the 57 isolates, 29 (33%) had mutations in the embB gene and Q497R was the common mutation (8/29; 27.6%) ( Table 3). Nineteen isolates had mutations, in pncA genes of which V128G was the common accounting for 7 (36.8%) ( Table 3). In addition, our analysis revealed pncA deletion involving Rv2044c in two drug resistance isolates (MDRTB & polyresistance isolates) that contributed 5.3% (1 isolates) of MDRTB and other resistance pattern (Table 3). Patients with MDR-TB had highly diverse mutations as compared to monoresistance (Table 3). We found 2 pre patients harbouring XDRTB isolates. The drug resistance mutations involving either rrs or gyrB genes against AMK and FQs respectively were detected in low proportions among DR resistance isolates (Table 3). These isolates had mutation either at the rrs gene involving substitution of C to T at codon 51 (C517T) or gyrB gene which involved replacement of R by C at codon 446 (R446C). MTBC isolates from drug susceptible patients had no mutations in genes encoding for INH, RFP, PZE, STM and EMB.
Treatment outcome and culture conversion rates among the drug resistance TB isolates Of the 57 study participants with mutation (s) to at least one of the anti-TB drugs, 47 (82.5%) had successful treatment, 3 (5.3%) died, and 7 (12.3%) defaulted (Table 4). Three patients (2MDR-TB and 1 with polyresistant) died 9 months after enrolment, during which they were in a continuous phase of anti-mycobacterial medications. The median (IQR) culture conversion rates were 51 (30-66) days and 83 (60-180) days for participants with mono/polyresistant and MDR/XDR-TB, respectively (Table 4).

Discussion
In this study, we found high concordance between WGS and conventional culture-based DST in predicting phenotypic drug resistance to anti-TB drugs, ranging from 81% for INH to 97% for RIF, and 95% concordance with Xpert® MTB/RIF (Cepheid, USA) for RIF. These findings are in agreements with those previously reported in different TB endemic settings [21][22][23][24]. The ability of Xpert® MTB/RIF and phenotypic detection of TB drug resistance cannot be underscored as it can pave a way to complementary and confirmatory WGS in early detection of resistance. Studies have shown phenotypic susceptibility tests to serve as reference standards [25], and that both Xpert assay and DST can provide information on polyresistance in pre-XDR-TB [26] despite varying sensitivities [27,28]. The current practice in diagnosing drug resistant tuberculosis is through the WHO-approved probe-based assays like Xpert® MTB/RIF (Cepheid, USA) for RIF [16] and either genotype MTBDRplus for RIF and INH or genotype MTBDRsl (Hain Lifesciences, Germany) for aminoglycosides/capreomycin and fluoroquinolones that must be confirmed with culture-based DST [29]. Unlike WGS, these assays not only have limited DST range to identify hot-spot resistant determining regions but they also cannot inform transmission dynamics [30,31]. The high concordance supports adoption of WGS as an alternative diagnostic tool to complement results from DST of all previously treated presumptive or confirmed rifampicin resistance cases in clinical settings. Although the facilities are not yet available the approach can be advocated in future clinical settings to enable reduction in unnecessary laboratory diagnostic time delays between identification of patients suspected of MDR-TB and initiation of treatment [29]. The potential for WGS to provide a rapid and comprehensive view of the genotype and reliable prediction of the drug susceptibility phenotype has been reported elsewhere [23,32,33]. This is an important platform to inform about susceptibility profile drugs like pyrazinamide (PZA), which is prescribed empirically in Tanzania and other resource limited countries. PZA kills semi-dormant bacilli and has synergetic activity with bedaquiline (BDQ), a key core drug in the currently recommended all-oral injectable free MDR-TB regimen(s). This synergism supports design of a shortened treatment regimens, which favours healthy patient outcomes [34].
In our study, the WGS uncovered 2 pre patient harbouring XDR-TB isolates, which were also detected by phenotypic DST at our hospital and later cured after MDR-TB treatment. This finding adds value to WGS as a potential molecular tool to complement phenotypic-DST and other molecular assay such Gene Xpert® MTB/ RIF (Cepheid, USA) MDR-TB results for informed decision-making prior to anti-tuberculosis therapy. For effective use particularly in low-and middle-income countries, WGS must be robust, easy to use, and affordable.
We also found that the majority of the MTBC isolates from DR-TB participants had mutations in genes predictive of phenotypic resistance to all 7 anti-TB drugs tested including INH (katG), RIF (rpoB), EMB (embB, embC-embA), FQs (gyrB) and second line injectable drugs (rrs) ( Table 3). The commonest mutations found in this study was S450L for rpoB (26/27), S315T on katG (51/56), Q497R (8/29) and G406S (5/29) on embB, C16T (5/29) on embC-embA, V128G (7/19) on pncA and K43R on rpsL genes are in keeping with findings reported elsewhere [35,36]. As expected, patients with MDR-TB had multiple mutations in these genes, confirming the previous concept that drug resistance in mycobacteria spp., and other bacteria evolves as they accumulate mutations either de novo or after multiple and longer exposure to antibiotics [37,38]. This high diversity of mutations among DR-TB patients suggests on-going transmission of MDR-TB strains rather than acquisition through random mutation and selection of drug resistance strains [37,39]. In addition, our analysis revealed a large deletion that involved Rv2044c in   [41]. We found 2 pre-XDR-TB patients possessing R446C and C517Tmutations in the gyrB and rrs genes, in addition S315T, S450L, G406A and C517T mutations in the katG, rpoB, embB and rrs genes respectively. At this juncture, it is worthwhile to translate these  mutations into clinical practice and investigate their association with participant's health outcomes. In this study, participants with MDR-TB were cured using the recommended treatment regimen and reverted late to culture negativity as compared to those with mono-or polyresistance (Table 4). Time to culture conversion was highly prolonged to more than 6 months in participants with pre-XDR-TB, who had multiple mutations. Delayed time to culture conversion in our study is similar to findings by Sangita et al., [42] and Shibabaw et al., [43] in India and Ethiopia who documented reversion time to culture negativity after 125 and 77 days of treatment respectively. In addition, the spectrum of mutations to genes that predict phenotypic resistance to anti-TB drugs determines treatment outcomes. Fortunately, 82.5% (47/57) of study participants achieved treatment success, similar to findings by Meressa et al., [44] in Ethiopia. This success rate is 30% higher than the global MDR-TB treatment success rate [15], positioning the WGS method along with culture based DST as a useful testing algorithm for rigorous microbiological monitoring that will accelerate the 2035 strategic END-TB vision to create a world population free of TB [45]. Over 17% of study participants had unfavourable treatment outcomes (5% mortality and 12% lost-to-follow-up), similar to that reported by Meressa et al. [44],. However, this is lower than that reported by Dheda et al., [46] and Milanov et al., [47] in South Africa and Bulgaria who both documented over 50% unfavourable treatment outcomes. Previous findings have argued that severe malnutrition (BMI < 16 kg/m2), HIV coinfection, previous history of anti-TB therapy and drug adverse events can predict poor treatment outcomes [47,48]. In this study, 29.9% of the study participants had low BMI and 23% were infected with HIV but had no effects on MDR-TB treatment outcomes. In our study, the phylogenetic analysis showed distinct DR-TB spoligotype lineages similar to a previous finding by Kidenya et al. [49] in North-western Tanzania who reported strain variation across different spoligotypedefined M. tuberculosis lineages. We found that, the Central Asian strains (CAS) genotype (lineage 3) were the predominant lineages (Table 5), with no evidence of any change in terms of the CAS dominance over the past few years in Tanzania [50]. The CAS is among the prevalent lineage in the Indian subcontinent, South-East Asia, the Middle-East and East-Africa [51], showing a North-South divide along the Tropic of Cancer in the Eastern hemispheremainly in Asia, and partly prolonged along the horn of Africa [51,52]. The dominance of CAS lineages in our local settings might be attributed to early contact due to migration and trade between people in Asian and East Africa countries. The seemingly prolonged stable dominance of CAS lineage reflects limited movement in of new infections from other sources/origins to influence the constantly circulating strains.
The main strength of our study is that we have correlated and linked WGS with treatment response and outcomes in a resource limited clinical TB endemic setting like Tanzania, which was not available previously. WGS provides high resolution when investigating diseases transmission in outbreaks, provides results for 2nd line drugs which is currently not performed in our settings. In addition to that, the WGS enables earlier use of the most appropriate drug regimen, thus improving patient outcomes and reducing overall healthcare costs [33]. This is the first step to start thinking how this technology could be included and add value in the future DR-TB diagnosis cascade, in high burden areas with limited resources where the need for rapid and accurate tools for assisting clinical decision making for optimal patient care and in predicting the treatment outcomes is very high.
However, our study has some limitations. Phenotypic culture-based DST for PZA, and cycloserine, which were part of the standardized treatment regimen for all participants was missing. In addition, concordance between line probe assays (LPA) genotype MDBTDRplus and MTBDRsl and WGS could not be done, which would have given a full comparison of the robustness of the WGS since the former had not been in use at the time of the study. In addition, WGS was performed retrospectively on clinical isolates, genetic data was not applied in the real time in the clinical setting, and hence no treatment regimen was modified to favour outcomes. Moreover, the use of WGS on frozen isolates of M. tuberculosis does not allow us to make considerations and comparisons between Xpert®MTB/Rif assay, phenotypic DST and WGS. Finally, WGS provides rapid and comprehensive diagnosis of DR-TB and the WHO has recognized its great potential for rapidly diagnosing drugresistant TB in diverse clinical settings. However, there are challenges that need to be overcome prior to the implementation of whole-genome sequencing in resourcelimited countries. These include high cost of equipment, the requirement for technical training, the need for expert guidance on the clinical interpretation of WGS data as the kind of information provided from the previous typing methods, the lack of simple solutions to obtain genome sequencing information directly from sputum samples and how it can be accommodated into our preexisting diagnostic frameworks. Nevertheless, the useful information presented in this study outweighs these limitations and can be used to advocate for a policy change in the adoption of molecular TB diagnostic algorithms.

Conclusion
WGS uncovered mutations in genes that predict phenotypic resistance to anti-TB drugs, signifying its importance in informed decision making prior to anti tuberculosis therapy. Expectedly, WGS was robust in ruling out DR-TB especially in absence of detectable mutations, which was the case in drug susceptible M. tuberculosis clinical isolates. The WGS-based DST best correlated with culture conversion rates and treatment outcome, indicating its potential application in designing an optimally individualized treatment regimen for favourable treatment outcomes in TB endemic settings.

Study settings
This study was conducted at Kibong'oto Infectious diseases hospital (KIDH) in Siha District, Kilimanjaro region, Tanzania. KIDH is the national centre of excellence for clinical management of drug resistant TB in the country and has a bed capacity of 320. The hospital provides TB services to more than 150 and 500 patients with drug resistant and susceptible TB per year, respectively. Recruitment and sputa collections from study participants were done at KIDH. Xpert® MTB/Rif assay, smear microscopy for acid-fast-bacilli (AFB) and isolation of MTBC on Lowenstein-Jensen (LJ) solid medium was performed at KIDH Mycobacteriology laboratory. MTBC isolates were transported to the Central Tuberculosis Reference Laboratory (CTRL) located at Muhimbili National Hospital (MNH) in Dar-es-salaam, Tanzania, for phenotypic susceptibility testing (DST) to first-and second-line anti-TB drugs like RIF, INH, EMB, SM, ofloxacin (OFX) and Kanamycin/Amikacin (KAN/ AMK) and DNA extraction. Due to its complexity, PZA-DST is not routinely performed in Tanzania. MTBC DNA were shipped to London School of Hygiene and Tropical Medicine (LSHTM) for WGS and bioinformatics analysis.

Study design, population and recruitment of participants
This study involved 57 and 30 participants who harboured M. tuberculosis that were resistant and susceptible to anti-TB drugs respectively, presenting at KIDH during the study period in 2014. Participants with resistant isolates were recruited if they had laboratory results by either GeneXpert MTB/RIF or phenotypic culture and DST [16]. The drug resistance was either mono/ polyresistance-resistant or MDR-TB. The rest were recruited if they were infected with MTB that was susceptible to RIF by Xpert® MTB/Rif assay. Eligibility included age ≥18 years, and willingness to sign a written informed consent and provide sputum samples for laboratory analysis. Very sick participants were excluded from the study.

Data collection
A standardized semi-structured questionnaire containing a set of study variables was used to collect data from study participants and medical charts were used to monitor treatment. Pre-treatment data that was collected included clinical information such as symptoms and signs suggestive of pulmonary tuberculosis (PTB), any previous history of TB treatment, HIV status, absolute CD4 + T cell count, the body mass index as the ratio of weight (kg) and (height) 2 (m 2 ) and socio-demographic characteristics such as age and gender. MDR-TB was treated for at least 20 months and monitored monthly with culture and smear microscopy for AFB while DS TB was monitored with smear microscopy only to determine microbiological treatment response and programmatic outcomes. WGS was performed on baseline MTBC isolates.

Laboratory procedures
Samples collection, culture of MTBC and phenotypic drugsusceptibility testing Each study participant provided approximately 4 mL of one spot sputum samples for culture. The samples were processed using the modified Petroff's method [31]. Briefly, 4 mLs of sputum was added to 4 mLs of 4% sodium hydroxide (NaOH). The mixture was vortexed and left to stand at room temperature for 15 min. Thereafter, the volume of the mixture was adjusted to 50 mLs in a falcon tube and concentrated by centrifugation at 3000 g for 15 min. Supernatants were discarded into a container with 25% phenol. Sediments were suspended in Phosphate buffer saline solution (PBS) before being inoculated on LJ culture media as recommended by the Clinical and Laboratory Standard Institute (CLSI) for TB culture [22]. In summary, 200 μL of sputum sediments were inoculated on two slopes of LJ medium containing either pyruvate or glycerol. Culture and identification of MTBC colonies were performed according to locally existing and CLSI standard operating procedures. Phenotypic drug susceptibility testing of the MTBC isolates was performed using a standard proportion method on LJ media [53,54].

DNA extraction and whole genome sequencing
Genomic DNA was extracted at Muhimbili University of Health and Allied Sciences (MUHAS) using a modified protocol based on Restriction Fragment Length Polymorphism (RFLP) developed at the National Institute for Public Health and the Environment (RIVM), Bilthoven, Netherlands [55]. The concentration and quality of the DNA were measured using Qubit™ 4 Fluorometer (Invi-trogen®). DNA samples were shipped to London School of Hygiene and Tropical Medicine (LSHTM), United Kingdom for whole genome sequencing. The DNA were shipped in screw capped cryo-vials and sealed with parafilm materials. During shipment the DNA were packed in double box and transported in iced packed with insulators materials in a thermo-stable shipping box where the Styrofoam was at least 1.5 in. thick. Library preparation, cluster amplification and sequencing were performed according to the manufacturer's instructions (QIAGEN). The QIAseq FX DNA Library Kit (QIAGEN) was used for library preparation according to the manufacturer's protocol. The QIAseq FX DNA Library Kit covers DNA fragmentation for 15 min, adapter ligation and a final purification of the samples using AMPure XP beads (Beckman-Coulter). Library DNA content was analysed using a Qubit 3.0 fluorometer. The sizes of DNA fragments were quantified using the Agilent High Sensitivity DNA Kit (Agilent Technologies) on an Agilent 2100 Bioanalyser (Agilent Genomics) according to the manufacturer's protocol. Samples were pooled to create pools of libraries at a concentration between 10 and 12pM, and loaded into the MiSeq® using MiSeq® v2 Reagent Kit. WGS was performed at the LSHTM using MiseqTM Sequencing 172 System MiSeqV2-500 cycles (Illumina), producing paired-end sequence reads of 151 bp.
Bioinformatics and genomic data analysis Quality control and mapping of reads Initial quality control (QC) reads were characterized using Kraken version v1.1 to exclude contamination and low-quality reads bases of less than 20 Qscore (defined as -10log10 (P), where P is the probability of an error as determined by the sequencing platform). Quality reads were mapped to the H37Rv reference genome (AL123456) using BWA mem [56] with default settings. The median depth and the percentage of the reference covered by a minimum of 10 reads were calculated using SAMtools depth [56] and custom scripts. Samples with sufficient amounts of MTBC data were used for downstream analysis.

Variant calling and in-silico resistance and lineage typing
Variants in drug resistance gene candidates were called using LoFreq version v2.1.2 [57]. These were annotated and compared to the TBProfiler database [58] (http:// tbdr.lshtm.ac.uk/) as it does not perform its own variant calling but annotates existing calls as drug resistance mutations using a database of mutations. Additionally, large indels were identified by looking at depth in candidate genes. The coverage across lineage specific markers was inspected using htsbox (https://github.com/lh3/ htsbox) and custom scripts to assign specific lineages to the isolates.

Phylogeny reconstruction
Variants throughout the whole genome were called using SAMtools/BCFtools version v1.9 [59]. The resulting variant call format (VCF) files were collated together filtering was performed. In short, sites in samples were marked as missing if less than 5 reads aligned to a position and monomorphic sites or those with greater than 10% missing data were removed. Using all SNPs, a maximum likelihood phylogeny was constructed using ExaML version v3.0.21, a state of the art tool for phylogenomic analyses [60]. The phylogenetic reconstruction was used in conjunction with lineage predictions output by TBProfiler. These predictions were based on the universally adopted lineage nomenclature and SNP barcode reported by Coll et al. [61]. This was visualized with drug resistance phenotypes and lineages using iTOL [62]. The lineages were assigned based on the SNP barcode available at doi:https://doi.org/10.1038/ ncomms5812 and has been adopted as the standard for Mtb WGS data.

Analysis of clinical data
Descriptive statistics was performed for demographic characteristics, frequency and distribution of mutation patterns in MDR-TB and other resistance patterns in MTBC. A Pearson chi-square test was used to determine the association between age, gender, HIV status and CD4 counts if applicable, Body mass index (BMI), history of previous TB treatment and genotype drug resistance as the main outcome using STATA version 14 (StataCorp LP, College Station, TX, USA). A p-value of < 0.05 considered statistically significant. Genotype drug resistance was defined as any mutations that are known to confer decreased anti tuberculosis drug susceptibility. Skewed data, Student's -test and Wilcoxon rank-sum tests were used for normally distributed data. The body weight based on BMI values for both adults was categorized into two groups namely; thinness (≤ 18.5) and nor-mal> 18.5 as recommended by the World Health Organization (WHO). Kappa statistics (K) were used to determine agreement between drug susceptibility testing (DST) and genetic drug resistance using four levels of agreement for kappa: < 0.40 (poor), 0.40-0.59 (fair), 0.60-0.80 (good), and > 0.80 (excellent) [63]. Availability of data and materials Raw sequence data are available from the European Nucleotide Archive (ENA) under the project accession number PRJEB29435 (https://www.ebi.ac. uk/ena/data/view/PRJEB29435).

Ethics approval and consent to participate
The Senate Research and Publications Committee of the Muhimbili University of Health and Allied Sciences (MUHAS) in Tanzania approved the study (Ref. No. MU/PGS/PhD/Vol.VI/154). Permission to conduct the study was obtained from KIDH administration. Prior to any study procedure, we obtained written informed consent that was signed by all patients. All study participants received MDR-TB treatment according to existing National and global MDR-TB treatment guidelines in Tanzania [23,27].

Consent for publication
Not applicable.