Rapid antibiotic-resistance predictions from genome sequence data for Staphylococcus aureus and Mycobacterium tuberculosis.
Bradley P, Gordon NC, Walker TM, Dunn L, Heys S, Huang B, Earle S, Pankhurst LJ, Anson L, de Cesare M, Piazza P, Votintseva AA, Golubchik T, Wilson DJ, Wyllie DH, Diel R, Niemann S, Feuerriegel S, Kohl TA, Ismail N, Omar SV, Smith EG, Buck D, McVean G, Walker AS, Peto TE, Crook DW, Iqbal Z. · 2015
Read Rapid antibiotic-resistance predictions from genome sequence data for Staphylococcus aureus and Mycobacterium tuberculosis. by Bradley P, Gordon NC, Walker TM, Dunn L, Heys S, Huang B, Earle S, Pankhurst LJ, Anson L, de Cesare M, Piazza P, Votintseva AA, Golubchik T, Wilson DJ, Wyllie DH, Diel R, Niemann S, Feuerriegel S, Kohl TA, Ismail N, Omar SV, Smith EG, Buck D, McVean G, Walker AS, Peto TE, Crook DW, Iqbal Z. (2015) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4703848.txt.
Introduction
Phelim Bradley, N. Claire Gordon, Timothy M. Walker, and a large team predicted antibiotic resistance from bacterial genomes in Nature Communications in 2015. Standard resistance tests take days for fast-growing bacteria and weeks for tuberculosis. The team's software, Mykrobe predictor, reads raw sequence data and reports species and resistance in about three minutes on a laptop. It works by comparing sequence fragments against a graph built from known bacterial diversity.
For Staphylococcus aureus, its calls across twelve antibiotics matched lab testing closely on an independent set of 470 isolates. For tuberculosis, on 1,609 isolates, it caught most resistance, though gaps in genetic knowledge limited it. Traces of rare variants helped flag the most resistant strains. The team also showed the approach can run on new portable sequencing machines. Faster resistance answers could steer treatment days earlier than culture allows.
Catalog record
| CREATOR | Bradley P, Gordon NC, Walker TM, Dunn L, Heys S, Huang B, Earle S, Pankhurst LJ, Anson L, de Cesare M, Piazza P, Votintseva AA, Golubchik T, Wilson DJ, Wyllie DH, Diel R, Niemann S, Feuerriegel S, Kohl TA, Ismail N, Omar SV, Smith EG, Buck D, McVean G, Walker AS, Peto TE, Crook DW, Iqbal Z. |
|---|---|
| YEAR | 2015 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 73 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Rapid antibiotic-resistance predictions from genome sequence data for Staphylococcus aureus and Mycobacterium tuberculosis By Phelim Bradley, N. Claire Gordon, Timothy M. Walker, Laura Dunn, Simon Heys, Bill Huang, Sarah Earle, Louise J. Pankhurst, Luke Anson, Mariateresa de Cesare, Paolo Piazza, Antonina A. Votintseva Published in: Nature Communications DOI: 10.1038/ncomms10063 ABSTRACT The rise of antibiotic-resistant bacteria has led to an urgent need for rapid detection of drug resistance in clinical samples, and improvements in global surveillance. Here we show how de Bruijn graph representation of bacterial diversity can be used to identify species and resistance profiles of clinical isolates. We implement this method for Staphylococcus aureus and Mycobacterium tuberculosis in a software package (‘Mykrobe predictor') that takes raw sequence data as input, and generates a clinician-friendly report within 3 minutes on a laptop. For S. aureus, the error rates of our method are comparable to gold-standard phenotypic methods, with sensitivity/specificity of 99.1%/99.6% across 12 antibiotics (using an independent validation set, n=470). For M. tuberculosis, our method predicts resistance with sensitivity/specificity of 82.6%/98.5% (independent validation set, n=1,609); sensitivity is lower here, probably because of limited understanding of the underlying genetic mechanisms. We give evidence that minor alleles improve detection of extremely drug-resistant strains, and demonstrate feasibility of the use of emerging single-molecule nanopore sequencing techniques for these purposes. The marked increase in antibiotic use in health care and agriculture since the 1940s has driven a rise in frequency of drug-resistant bacterial strains, which now present a global threat to public health. Clinical isolates resistant to most drugs have now been seen for many species including Mycobacterium tuberculosis, Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Neisseria gonorrhoeae, Acinetobacter baumannii and Pseudomonas aeruginosa1. Antimicrobial susceptibility testing is therefore now central to the treatment of serious bacterial infections diagnosed by culture, and is used to determine the protocols for first-line antibiotic use when culture is not available. At present, phenotyping tests take at least 1–2 days to complete for rapidly growing bacteria such as S. aureus, and can take weeks in slow-growing bacteria such as M. tuberculosis. Microbial genome sequencing has the potential to substantially increase the speed of antibiotic resistance detection for many pathogens2 and in addition provides valuable information on relatedness that could contribute to surveillance. The key biological constraint is the extent of our understanding of the genotype-to-phenotype correspondence—that is, the genotype needs to be sufficiently predictive of resistance. Increasingly, this correspondence is high for many bacterium/drug combinations. For example, Gordon et al.3 recently demonstrated that a curated panel of mutations and genes known to cause drug resistance in S. aureus was sufficient to predict resistance for 12 antibiotics with a sensitivity of 97% and specificity of 99%. Thus, at least for one species, a genotype-based method could deliver results with accuracy suitable for use in a clinical laboratory. We set out to develop methods applicable to standard clinical samples, and solve the multiple computational challenges that act as barrier to routine and rapid deployment of such a system in clinical practice. These challenges include not only the need to determine species and predict resistance, but also developing a framework extensible to many species, and ensuring accessibility of the tool to a user base who may be unskilled in bioinformatics. Furthermore, although clinical samples currently undergo protocols that tend to remove diversity (for example, blood culture for S. aureus and solid/liquid culture for M. tuberculosis), future developments to reduce bed-to-diagnosis time will surely involve reductions in culture time, and potentially increase the levels of diversity in the sample. Thus we set out to build a system robust to mixture, and to establish whether an appreciable proportion of phenotypic resistance was explained by low-frequency alleles. Methods using genome sequence data for species identification range from the specific4 to the sensitive5, but generally performance is measured globally in terms of detection of species presence. However, for clinical use we need considerable flexibility in tuning sensitivity and specificity for different species, potentially weighted to minimize clinical risk. For example, there may be a species associated with high mortality (for example, S. aureus) that can occur in samples mixed with other species (for example, coagulase-negative staphylococci (CoNS), which are common contaminants of blood cultures, being present on the skin through which the blood was taken), and that may even share the same resistance genes and thus confound inference. Various methods have been used for genotyping resistance features: mutations and genes have been detected by whole-genome assembly3, genes by assembly and BLAST6, or single nucleotide polymorphisms (SNPs) and indels by mapping78. These methods have been demonstrated to have adequate performance in some circumstances. However, traditional whole-genome bacterial assembly is fundamentally based on the assumption that all data comes from a single haploid genome9, and so is ill-suited for mixed samples, and mapping to a single-reference results in error rates that depend on genetic distance of the sample from the reference10. There are pre-existing tools for expert users that incorporate resistance prediction61112, none of which handle the issue of contaminating related species in clinical samples, or minor clones—we include comparative data below. Here we aim to move beyond proof-of-concept of how sequencing might work in the clinic, to a general framework for genotype-based antimicrobial-resistance prediction, with concrete implementations for two species where drug resistance is of global concern: S. aureus and M. tuberculosis. The user should be able to obtain a report interpretable by a clinician by simple drag-and-drop of raw sequence data file. We evaluate extensively against clinical gold standards using current (Illumina) sequence data, including the impact of minor population detection. We notice that minor alleles improve ability to distinguish multi-drug resistant (MDR) from extensively drug-resistant (XDR) tuberculosis. We demonstrate that our method also works with an emerging strand-sequencing technology (Oxford Nanopore Technologies (ONT), MinION), giving perfect concordance with phenotype after just 7 h of sequencing, both for SNP-based and gene-based resistance. Finally, we discuss what is needed to apply our framework to other species. RESULTS Using population genome graphs for genotyping We show in Fig. 1a a cartoon of the genetic diversity in a bacterial species and two options for building a reference variation structure. In option (i) we show the standard approach, where we select an arbitrary strain (strain 1) to be the reference genome, along with one copy of each plasmid gene. In this work we have developed an alternate approach, shown in option (ii). We start with a curated knowledge base of resistant/susceptible alleles, and assemble a de Bruijn graph13 of them on different genetic backgrounds, along with many examples of resistance genes. This forms our reference graph. In Fig. 1b we show the corresponding analyses of a mixed sample. The traditional approach is shown in option (i), whereby sequence reads are mapped to the reference genome and genes, requiring the mapping and inference to cope with the divergence between sample strains and the reference. Our approach (option (ii)) directly compares the de Bruijn graph of the sample with the reference graph. This results in statistical tests for the presence of resistance alleles that are unbiased by choice of reference or assumptions of clonality. Moreover, these tests will improve as the catalogue of diversity in the species grows. Our approach is implemented in