Integration of Published Information Into a Resistance-Associated Mutation Database for Mycobacterium tuberculosis (original) (raw)

Abstract

Tuberculosis remains a major global public health challenge. Although incidence is decreasing, the proportion of drug-resistant cases is increasing. Technical and operational complexities prevent Mycobacterium tuberculosis drug susceptibility phenotyping in the vast majority of new and retreatment cases. The advent of molecular technologies provides an opportunity to obtain results rapidly as compared to phenotypic culture. However, correlations between genetic mutations and resistance to multiple drugs have not been systematically evaluated. Molecular testing of M. tuberculosis sampled from a typical patient continues to provide a partial picture of drug resistance. A database of phenotypic and genotypic testing results, especially where prospectively collected, could document statistically significant associations and may reveal new, predictive molecular patterns. We examine the feasibility of integrating existing molecular and phenotypic drug susceptibility data to identify associations observed across multiple studies and demonstrate potential for well-integrated M. tuberculosis mutation data to reveal actionable findings.

Keywords: tuberculosis, drug resistance, resistance-associated mutations, genomic sequencing, drug susceptibility testing, database

Since 2002 there has been a gradual 1.3% annual decrease in the incidence of tuberculosis worldwide. Although this trend is encouraging, it is too weak to lead to elimination of tuberculosis as a public health problem by 2050, which is the goal of the World Health Organization (WHO) [http://www.who.int/tb/strategy/stop_tb_strategy/en/]. The challenge to stop tuberculosis is severely complicated by the increasing incidence of drug-resistant tuberculosis. Although rapid and accurate detection of tuberculosis will be a key factor in conquering tuberculosis, both treatment of the disease and better success preventing its transmission are significantly boosted by accurate information on drug susceptibility [1]. The WHO defines multidrug-resistant tuberculosis (MDR-TB) as resistance to isoniazid and rifampicin, with or without resistance to other first-line drugs [http://www.who.int/tb/challenges/mdr/tdrfaqs/en/]

With the introduction of new drug combinations and regimens, and patients with potentially more complex resistance profiles, it is imperative to be able to provide a comprehensive profile of drug susceptibility in order to select the correct therapies. Bacterial culture-based drug susceptibility testing (DST) is current “gold standard,” but is technically difficult and time-consuming. DST methods are not standardized and results may vary depending on culture techniques employed [2–4], which is especially true for isolates with low-level resistance or when testing resistance to certain second-line drugs. Phenotypic DST methods can also expose laboratory workers to potential infection. Thus new approaches to determining drug resistance are needed.

Detection of resistance-conferring mutations with methods such as polymerase chain reaction and hybridization, targeted sequencing of specific genes, or whole genome sequencing are attractive and promising alternatives to phenotypic DST methods. The data from sequencing especially have been encouraging, with the identification of genes and intergenic noncoding regions associated with drug resistance. However, no systematic study has been performed to correlate genotypic output of either a targeted or whole genome sequencing approach with phenotypic drug response in culture and clinical outcome. In addition, there remains a need to inform logistical decisions on developing simple, rapid, affordable molecular tuberculosis drug-resistance diagnostics, particularly in light of challenges faced by clinical laboratories in low- to middle-resource settings. Currently available DNA sequencers all have reproducibility and performance issues, technical biases, and provide data that require informatics-intensive activities. While welcomed in a research setting, sequencing protocols provide data that need to be reduced to actionable knowledge and may contain false calls (errors) that require base-by-base review. Before sequencing technologies can impact tuberculosis on a global scale, we need to learn how to translate data into statements on drug resistance in a well-supported and reproducible fashion.

To affect guidance in the treatment of patients with tuberculosis, all current analysis and modelling point to the necessity for a solution based on sequence-level results generated as near to the patient as possible [1, 5]. This need for technology proximal to the point of care must be balanced with requirements for data quality. There are multiple sequencing instruments, multiple sample-processing approaches, multiple options for user interfaces, and as-yet incomplete global data associating specific mutations with degrees of resistance to different drugs. Both a protocol to measure mutations and algorithms to interpret the drug resistance implied by mutation data will need to be developed.

Although new sequencing methods are providing an avalanche of genomic information at continually lower cost [6–8], this wealth of information is currently underexploited for diagnostics. Although there are regional and research activities using raw sequencing results from various platforms, there is no generally accepted sequence data-handling approach that is vetted, quality controlled, and readily accessible to the scientific community, let alone usable globally in a clinically constructive way. Efforts to exploit sequencing technologies for tuberculosis treatment and control will require development of solutions tailored to translate this information into easy-to-use and intuitive diagnostic devices.

Data on mutations identified in drug-resistant isolates alone is not enough to determine the association of the mutation with resistance or to demonstrate causality. Therefore, we should focus on data incorporating both susceptible and resistant samples.

Currently, mutation and drug-resistance data for M. tuberculosis are scattered across multiple independent databases, journal articles, and their Supplementary Materials. We sought to identify data on isolates appropriate to integrate into a single database to enable querying data across study sources. The purpose of this work was to determine (i) the challenges to integrating mutation and DST data from multiple sources, and (ii) if data, once integrated, allow for systematic analysis that could inform diagnostics development.

METHODS

Published articles and online repositories were identified and reviewed for data on drug resistance-associated single nucleotide polymorphisms. The following criteria were used to prioritize the data sets for inclusion in this integration project:

(i) DST and mutation data on individual isolates preferred over summary statistics (desired, not required), (ii) number of isolates reported (desired more than 50 isolates in a publication), (iii) mutation data on susceptible isolates (required), (iv) easily understandable methods and results (required), (v) DST data on wild-type isolates (required), (vi) drug-level DST results rather than isolate classification solely by MDR or XDR criteria (required), (vii) publications post-2009 (required since data from pre-2010 publications were included via integration of the tuberculosis drug resistance mutation database (TBDReaMDB) [9]), (viii) understandable, well documented, and readily available data tables in articles, Supplementary Tables, or website portals (required), and (ix) publication not on hold or retracted (required).

Table 1 documents the article sources that were included [10–15]. Table 2 documents the online database resources.

Table 1.

Articles Used as Data Sources

Table 2.

Online Sources of Data

Data Integration

The tuberculosis drug resistance database (TBDR) was established to integrate drug resistance mutation data from the different sources and to enable querying for those mutations that have supporting evidence from multiple studies. The intent was to capture as much information as possible from each study yet also allow analysis across diverse studies. During the development of the database specific issues were identified, and flexibility was built into the database. For instance, some studies reported mutations only at the amino acid level while others reported codon changes. Trivial conversions were implemented, such as translating codons to amino acids or nucleotides to the negative strand if appropriate. The database structure enabled on-the-fly summarization of data at the nucleotide level (eg, “S315T (AGC/ACC)”), the most granular level we store, or at the amino acid level (eg, “S315T”). Some source information, such as the genomic coordinates reported by the Broad/GTBDR database, was captured to permit future efforts to bring identical mutations together.

Some researchers reported mutation observations at the isolate level, while most authors reported just tabulated findings. Sometimes these tabulated findings are co-occurring mutations, which could give some insight into isolate-level observations. TBDR captured as much structure from each study as possible. For isolate-level reports TBDR stored the information on individual isolates then automatically summarized the tabulated results for each drug. For co-occurring mutations, the structure of the data was preserved to enable future analyses that rely on isolate-level information.

Researchers reported rpoB numbering using either Escherichia coli or M. tuberculosis numbering. TBDR uses M. tuberculosis numbering and reports the conversion from E. coli if and when it is performed. Researchers reported promoter mutations using a variety of genes in the same operon. TBDR normalizes these names only for the _fabG1_-hemZ operon promoter, mutations of which appear in TBDR as the inhA promoter. For gyrB, numbering systems have varied [16], and TBDR numbers 714 amino acids (NCBI protein accession number WP_003901763.1).

TBDR was built to provide reproducible results through automated processes. To minimize the manual manipulation of source material, computer programs were written to parse the 1523 data files in the case of the Broad/GTBDR database, an Excel file for TBDReaMDB, and various primary and Supplementary Tables in other publications. The main exception to the automation was PDF document table extraction, which typically required some manual cleanup after a copy and paste.

For published work TBDR stores PubMed identifiers (PMIDs). Since the TBDReaMDB source material did not provide PMIDs, these were identified manually. TBDR connects to PubMed to load reference details via the PMID.

RESULTS

Database Summary

The TBDR database currently contains 39 756 mutations across 29 genes and DST results for 23 drugs and one unspecified fluoroquinolone category. Because some studies performed DST for multiple drugs, the data from 80 studies, including the 73 found in the TBDReaMDB, comprised 148 investigations into drug resistance.

Across the 29 genes, mutations consisted of 1417 distinct amino acid substitutions, 89 distinct regulatory code changes, 105 insertions, and 106 deletions. Table 3 shows the mutations observed in the context of each drug. For example, there were 21 398 mutations observed in isolates subjected to DST for amikacin, and 5556 of these were observed in amikacin-resistant isolates. The numbers in Table 3 are purely descriptive, include mutations measured for genes not expected to be associated with resistance to the particular drug, and do not by themselves inform us about mutation-resistance associations. Table 4 summarizes the number of isolates subjected to DST for each of the drugs in the database. The database content described in Tables 3 and 4 serve as the basis for the calculations and queries described below.

Table 3.

Numbers of Mutations Integrated into TBDR

Table 4.

The Number of Isolates Found Resistant or Susceptible to 23 Drugs

Web Portal to Database

A web portal was established to enable simple access to the database. The portal both facilitated integration efforts and allowed sharing of results among coauthors. There are four main types of tables provided by the portal: (i) a list of drugs with data contained within the database, (ii) a summary of all mutations for a given drug, (iii) a list of mutations provided by a reference source, and (iv) a summary for each drug of all canonical mutations for that drug and an array of resistance-mutation statistics. Sensitivity, specificity, positive predictive value, and negative predictive value were calculated on the pooled counts across studies for the following: the number of (i) clinical isolates with both phenotypic and genotypic results for the mutation, (ii) isolates resistant to the drug, (iii) isolates susceptible to the drug, (iv) isolates with the specific mutation, (v) mutant isolates resistant to the drug, and (vi) mutant isolates susceptible to the drug.

Two modes by which the database reports mutations across studies were deemed potentially useful to inform diagnostics research. The first mode merges all mutations with the same amino acid substitution or promoter position. The second mode merges mutations if the nucleotide-level information is also identical. Because some references did not report codon changes and others reported resistance mutations with the nucleotide change but without the codon, the nucleotide-level reports contain multiple rows for identical mutations. The interface allows selecting studies to exclude from the results report.

Results on Drug Resistance-associated SNPs

For each drug category in the TBDR database, mutations were queried to determine which were observed in at least 3 studies, exhibited a nominal specificity greater than 95% and were found at a higher rate in resistant isolates than in susceptible isolates. Many associations identified were noncanonical since resistance mutations for one drug carry information about resistance to another drug tested in the same study. This phenomenon is to be expected, as resistance to multiple drugs is, unfortunately, not uncommon. Table 5 lists the canonical gene-drug associations we used to limit our presentation on drug resistance-associated mutations. Eleven drugs yielded mutations in canonically associated genes that met the above criteria (Table 5). Supplementary Table 1 lists 106 amino acid substitutions and 11 regulatory resistance mutations that were defined by the query. The table is sorted sequentially on three columns: drug, specificity (highest first), and sensitivity (highest first). Table 6 lists the substitutions and regulatory mutations for 2 first-line drugs, isoniazid and rifampicin, sorted as in Supplementary Table 1. This relatively simple query represents a first attempt at using the integrated data. Further refinement by a panel of experts would surely improve the utility of the resistance mutations list for each drug.

Table 5.

Drug-gene Associations

Table 6.

Drug Resistance-Associated Mutations for Isoniazid and Rifampicina

There exist mutations in canonically drug-resistance-associated genes that do not reliably predict DST results. Data integration can help us identify departures from wild type that do not confer drug resistance. This is a specific strength of TBDR, as it brings together the results of multiple studies in a manner that can be queried to address such concerns. For example, TBDR contains pncA mutations observed only in pyrazinamide-susceptible isolates and in more than one study, including C14G, S59F, F81S, H82Y, Y103C, and A143T. Similarly, 4 studies found rrs 1402 (C->N) mutations in a total of 5 isolates tested for amikacin susceptibility, and all were susceptible, indicating that this mutation may indeed be a poor marker of resistance to this particular drug [17].

Caveats

Table 6 and Supplementary Table 1 identify mutations that are supported by multiple studies and therefore have a higher confidence in their association with drug resistance. These results are not a comprehensive investigation into each drug and mutation, or prediction of resistance to the drug.

The sensitivities for mutations for a given drug in Table 6 and Supplementary Table 1 are not cumulative for predicting drug resistance. First of all, isolates may have multiple mutations and thus contribute to multiple rows in the table. Second, we do not have all data at the isolate level, as summary tables generally do not preserve this important information. If we had all the data at the isolate level (ie, all mutations reported for each isolate), we could indeed ask what sensitivity (and specificity) combinations of mutations would provide for drug resistance across this idiosyncratic collection of samples.

The predictive value of mutations for drug resistance and diagnostics depends strongly on the proportion of mutation-typed samples that are a priori phenotypically resistant. Because the data from many studies are highly biased toward analysis of resistant isolates, the statistics reported are likely quite unreliable as predictions in any particular population. Researchers often avoided typing phenotypically wild-type isolates at their true rates in the population.

DISCUSSION

Integrating tuberculosis drug-resistance data into TBDR required addressing a number of data-handling issues. A straightforward query enabled by the database revealed 96 mutations informative of drug resistance and observed in multiple independent studies.

For sources where complete DST and mutation results were available, TBDR cataloged the association of all resistance mutations with drug sensitivity, including noncanonical associations. Significant noncanonical associations likely arise in part from the evolution of drug resistance to multiple drugs and in part because of ascertainment bias since many studies target populations with MDR and XDR isolates.

The quantitative science required for diagnostics development cannot be addressed fully in an analysis of data such as prepared here. Nevertheless, analyses of existing data should help prioritize mutations. A large impact on diagnostics and patient treatment could be made by using properly integrated data to help the community reach a data-driven consensus regarding tuberculosis drug-resistance predictive mutations.

The Grading of Recommendations Assessment, Development and Evaluation (http://www.gradeworkinggroup.org/) criteria need to be kept in mind when proposing mutations for diagnosis of drug-resistant tuberculosis. A key consideration is how well a mutation actually predicts drug resistance. Because there are associations of particular mutations with phylogeny and of phylogeny with geographic region, these data are important to capture for future drug resistance mutation studies. By including complete or at least expanded sequence results, it should be possible to better understand which mutations are likely to be causal and which are simply markers of resistance in specific populations. It may be important to determine when phylogenetic information has no appreciable impact on mutation-based prediction of drug resistance. The quality of evidence for predicting resistance to drugs will necessarily be better for some mutations than others. In short, an effort to include as many isolate-level records (as opposed to summary table information), and to gather enough information to address phylogeny, should allow for analyses that will boost confidence in drug resistance prediction.

Factors other than phylogeny are confusing to evaluate when associating mutations with drug resistance. Importantly, for some mutations, even causal mutations, imperfect correlation with phenotypic DST likely results from varying phenotypic testing methods or drug concentration thresholds across studies. TBDR records information on DST methods reported by different investigators and further analysis could support expert interpretation. Additionally, although some published studies are clearly annotated by geographic region, others are less clearly annotated and also may include isolates from multiple regions.

The integrated data provide a platform for addressing the multivariate properties of the resistance mutations, although we have not demonstrated such an analysis in this article. A multivariate analysis, using statistical, machine-learning, or other mathematical methods, could determine which of the resistance mutations are most informative, and specifically which do not offer additional information when other resistance mutations have been measured. The results of such a study could inform the selection of a panel of resistance mutations that most efficiently uses resources by reducing redundancy.

The current database and web interface are proof-of-principle tools that enabled the generation of the main results as presented herein. The tools demonstrate that data integration is an important component in development of analysis algorithms to identify drug resistance-predictive mutations.

Much of the data we encountered was presented only in summary tables. For data to be most useful to other investigators and inform diagnostics development, information on isolates should always be reported as complete reports on the isolates, including all DST and mutation typing. While most useful would be entry of the data into an appropriate database, at the very least these complete reports should be released as Supplementary Data. For analysis to best demonstrate the diagnostic potential of molecular patterns, it would be most useful to have data from studies that record treatment regimens and outcome data together with mutation and drug susceptibility phenotypes. The completeness of data gathered and other aspects of data quality control should be carefully targeted in future efforts to collect and analyze tuberculosis drug resistance data.

This demonstration of data integration for M. tuberculosis drug resistance-associated mutations provides two important lessons. First, the knowledge in the community is currently larger than perhaps has been understood by many researchers and diagnostics developers, and could better inform diagnostic development decisions in the near future. Second, we can anticipate many important issues for gathering and analyzing data with more modern tools, such as bacterial whole genome sequencing, which could better inform microbial profiling efforts in the near future.

A database such as TBDR could be expanded or incorporated into another database to address the growing needs for knowledge sharing with respect to sequence data and markers for tuberculosis drug resistance. To develop a relevant data repository, the database will need to clearly address objectives from the community, namely development of tests for detection of drug resistance and clinical impact. At the same time, to develop a sustainable data repository, appropriate partnerships among researchers, clinical trial groups, reference labs, and commercial entities will need to drive the technology development, as different parties have distinct needs. For example, clinical trial groups are required to anonymize data, and commercial parties may need to compare in-house results with database contents in a confidential manner. As an example, the Critical Path to TB Drug Regimens (CPTR) initiative (http://cptrinitiative.org) has developed strong partnerships with clinical trial groups and commercial entities to tackle the challenges facing tuberculosis drug development. As part of its ongoing work, CPTR has established data management practices and technology that can be adapted to assist needs for knowledge sharing with respect to sequence data and markers for tuberculosis resistance. The power of TBDR and subsequent databases that incorporate existing and prospectively gathered genotypes, phenotypes, and metadata lies in the ability to compose and execute queries. These queries will need to be designed by a collaborative effort among data scientists, stakeholders in diagnostics development, expert committees on tuberculosis drug resistance, and computational biologists. In this way the complexities of different drugs and mutation interactions can be addressed, and a consensus for predicting resistance to each drug should be reached.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online (http://jid.oxfordjournals.org). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.

Supplementary Data

Notes

Disclaimer. The views and opinions expressed in this article are those of the authors and do not necessarily represent an official position of the US Centers for Disease Control and Prevention.

Financial support. This work was supported by the Bill and Melinda Gates Foundation. M. S. is funded with Federal funds from the National Institute of Allergy and Infectious Diseases, National Institutes of Health, Department of Health and Human Services, under Contract HHSN272200800014C. The funders had no role in the analysis of data and decision to publish.

Potential conflicts of interest. D. L. D. and M. D. P. are employed by FIND, a nonprofit organization that collaborates with industry partners, including Cepheid and Hain diagnostics among others, for the development, evaluation and demonstration of new diagnostic tests for poverty-related diseases. H. S. and K. D. Y. are the beneficial owners of Knowledge Synthesis Inc. D. A. reports grants from Cepheid, and royalties from a molecular beacons patent pool. D. H. and E. A. report funding from the Bill and Melinda Gates Foundation outside of this work. There are no patents or products with respect to this article. All other authors report no potential conflicts.

All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

References

• 1.Jain A, Mondal R. Extensively drug-resistant tuberculosis: current challenges and threats. FEMS Immunol Med Microbiol. 2008;53:145–50. doi: 10.1111/j.1574-695X.2008.00400.x. [DOI] [PubMed] [Google Scholar]
• 2.Van Deun A, Wright A, Zignol M, Weyer K, Rieder HL. Drug susceptibility testing proficiency in the network of supranational tuberculosis reference laboratories. Int J Tuberc Lung Dis. 2011;15:116–24. [PubMed] [Google Scholar]
• 3.Angra PK, Taylor TH, Iademarco MF, Metchock B, Astles JR, Ridderhof JC. Performance of tuberculosis drug susceptibility testing in U.S. laboratories from 1994 to 2008. J Clin Microbiol. 2012;50:1233–9. doi: 10.1128/JCM.06479-11. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 4.Jiang G-L, Chen X, Song Y, Zhao Y, Huang H, Kam KM. First proficiency testing of second-line anti-tuberculosis drug susceptibility testing in 12 provinces of China. Int J Tuberc Lung Dis. 2013;17:1491–4. doi: 10.5588/ijtld.13.0275. [DOI] [PubMed] [Google Scholar]
• 5.Boehme CC, Nabeta P, Hillemann D, et al. Rapid molecular detection of tuberculosis and rifampin resistance. N Engl J Med. 2010;363:1005–15. doi: 10.1056/NEJMoa0907847. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 6.Catanho M, Mascarenhas D, Degrave W, de Miranda AB. GenoMycDB: a database for comparative analysis of mycobacterial genes and genomes. Genet Mol Res GMR. 2006;5:115–26. [PubMed] [Google Scholar]
• 7.Stucki D, Gagneux S. Single nucleotide polymorphisms in Mycobacterium tuberculosis and the need for a curated database. Tuberc Edinb Scotl. 2013;93:30–9. doi: 10.1016/j.tube.2012.11.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 8.Lu JT, Campeau PM, Lee BH. Genotype-phenotype correlation—promiscuity in the era of next-generation sequencing. N Engl J Med. 2014;371:593–6. doi: 10.1056/NEJMp1400788. [DOI] [PubMed] [Google Scholar]
• 9.Sandgren A, Strong M, Muthukrishnan P, Weiner BK, Church GM, Murray MB. Tuberculosis drug resistance mutation database. PLoS Med. 2009;6:e2. doi: 10.1371/journal.pmed.1000002. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 10.Campbell PJ, Morlock GP, Sikes RD, et al. Molecular detection of mutations associated with first- and second-line drug resistance compared with conventional drug susceptibility testing of Mycobacterium tuberculosis. Antimicrob Agents Chemother. 2011;55:2032–41. doi: 10.1128/AAC.01550-10. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 11.Casali N, Nikolayevskyy V, Balabanova Y, et al. Microevolution of extensively drug-resistant tuberculosis in Russia. Genome Res. 2012;22:735–45. doi: 10.1101/gr.128678.111. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 12.Nosova EY, Bukatina AA, Isaeva YD, Makarova MV, Galkina KY, Moroz AM. Analysis of mutations in the gyrA and gyrB genes and their association with the resistance of Mycobacterium tuberculosis to levofloxacin, moxifloxacin and gatifloxacin. J Med Microbiol. 2013;62(Pt 1):108–13. doi: 10.1099/jmm.0.046821-0. [DOI] [PubMed] [Google Scholar]
• 13.Rodwell TC, Valafar F, Douglas J, et al. Predicting extensively drug-resistant Mycobacterium tuberculosis phenotypes with genetic mutations. J Clin Microbiol. 2014;52:781–9. doi: 10.1128/JCM.02701-13. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 14.Lin S-YG, Rodwell TC, Victor TC, et al. Pyrosequencing for rapid detection of extensively drug-resistant Mycobacterium tuberculosis in clinical isolates and clinical specimens. J Clin Microbiol. 2014;52:475–82. doi: 10.1128/JCM.01821-13. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 15.Miotto P, Cabibbe AM, Feuerriegel S, et al. Mycobacterium tuberculosis Pyrazinamide Resistance Determinants: a Multicenter Study. MBio. 2014;5:e01819–14. doi: 10.1128/mBio.01819-14. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 16.Maruri F, Sterling TR, Kaiga AW, et al. A systematic review of gyrase mutations associated with fluoroquinolone-resistant Mycobacterium tuberculosis and a proposed gyrase numbering system. J Antimicrob Chemother. 2012;67:819–31. doi: 10.1093/jac/dkr566. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 17.Georghiou SB, Magana M, Garfein RS, Catanzaro DG, Catanzaro A, Rodwell TC. Evaluation of genetic mutations associated with Mycobacterium tuberculosis resistance to amikacin, kanamycin and capreomycin: a systematic review. PloS One. 2012;7:e33275. doi: 10.1371/journal.pone.0033275. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Data