Prediction of antibiotic resistance by gene expression profiles

ARTICLE Received 10 Aug 2014 | Accepted 7 Nov 2014 | Published 17 Dec 2014 DOI: 10.1038/ncomms6792 OPEN Prediction of antibiotic resistance by gene expression profiles Shingo Suzuki1, Takaaki Horinouchi1 & Chikara Furusawa1 Although many mutations contributing to antibiotic resistance have been identified, the relationship between the mutations and the related phenotypic changes responsible for the resistance has yet to be fully elucidated. To better characterize phenotype–genotype mapping for drug resistance, here we analyse phenotypic and genotypic changes of antibiotic-resistant Escherichia coli strains obtained by laboratory evolution. We demonstrate that the resistances can be quantitatively predicted by the expression changes of a small number of genes. Several candidate mutations contributing to the resistances are identified, while phenotype–genotype mapping is suggested to be complex and includes various mutations that cause similar phenotypic changes. The integration of transcriptome and genome data enables us to extract essential phenotypic changes for drug resistances. 1 Quantitative Biology Center (QBiC), RIKEN, 6-2-3 Furuedai, Suita, Osaka 565-0874, Japan. Correspondence and requests for materials should be addressed to C.F. (email: ). NATURE COMMUNICATIONS | 5:5792 | DOI: 10.1038/ncomms6792 | www.nature.com/naturecommunications & 2014 Macmillan Publishers Limited. All rights reserved. 1 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6792 T he emergence of multi-drug-resistant bacteria is a growing concern for global public health1–3, as doses of antibiotics have conferred a selective advantage for naturally emergedresistant bacteria to cause drug ineffectiveness4,5. A number of mutations have been identified and shed light on how bacterial cells acquire antibiotic resistance6,7. For some of these mutations, scientists can easily extract the causal relationship to drug resistance, such as a resulting modification in a specific drug target8. However, the relationship between a mutation and drug resistance is not always a simple one-to-one correspondence. Multiple mutations are often required to acquire high levels of resistance to a specific drug7,9,10, whereas a single mutation can cause various phenotypic changes that change the resistance and susceptibility to various drugs simultaneously11. Studies using mutant libraries have revealed that a large number of genes influence drug resistance and susceptibility, including many genes not directly involved in known drug-resistant machineries12–14. Furthermore, non-additive (for example, synergistic and antagonistic) responses to combinatorial drug treatments suggest interplay among the mechanisms of drug resistances15,16. Overall, the complex relationship between drug resistance acquisition, genetic alternations and global phenotypic changes remains unclear. Laboratory evolution of bacteria17 is a powerful tool for investigating the acquisition dynamics of drug resistance7,18. In such experiments, bacterial cells are exposed to fixed drug concentrations around which the cell growth is partially or completely inhibited such that a selective advantage for resistant strains is maintained. Although some essential factors in drug resistance evolution, including horizontal gene transfer (HGT)19 and interspecies communication20, are difficult to analyse using laboratory evolution, this experimental system has several advantages in comparison with in vivo experiments when studying de novo acquisition of drug resistance, including a wellcharacterized ancestor strain, a defined environment and parallel evolution experiments that discriminate necessary and unnecessary phenotypic/genetic changes. In general, the genomewide phenotypic and genotypic analysis of emerging resistant strains in laboratory evolution offers to clarify the relationship between phenotype–genotype changes and drug resistances. In this study, we performed laboratory evolution of Escherichia coli under various drug treatment conditions to obtain resistant strains. For each obtained drug-resistant strain, transcriptome and genome re-sequencing analyses were performed to identify fixed mutations and gene expression changes. Furthermore, we analysed how the acquisition of resistance to one drug changes the resistance and susceptibility to other drugs. By integrating these data and using a simple mathematical model, we succeed to quantitatively predict resistances to various drugs based on the gene expression levels of a small number of genes. The phenotype–genotype relationship in resistant strains is analysed to elucidate the contribution of the fixed mutations to the drug resistance. Results Laboratory evolution of antibiotic-resistant E. coli cells. We selected 11 antibiotics that cover a wide range of action mechanisms, including drugs that disrupt cell wall synthesis, protein synthesis, folic acid biosynthesis and DNA replication (Table 1). E. coli MDS42 cells were cultured in M9 synthetic medium with eight different concentrations of drugs and were propagated daily from a well containing the highest drug concentration possible, in which cells were able to sustain growth (see Methods for details). To evaluate the reproducibility of the evolutionary pathways, for each antibiotic, four independent 2 Table 1 | List of antibiotics used for experimental evolution. Antibiotics name Cefoperazone Abbreviation Class CPZ Cefixime CFIX Amikacin AMK Neomycin NM Doxycycline DOXY Chloramphenicol CP Azithromycin AZM Trimethoprim Enoxacin Ciprofloxacin Colistin TP ENX CPFX CL Cellular target Cephalosporin, Cell wall b-lactam Cephalosporin, Cell wall b-lactam Aminoglycoside Protein synthesis, 30S Aminoglycoside Protein synthesis, 30S Tetracycline Protein synthesis, 30S Protein synthesis, 50S Azalide, Protein synthesis, macrolide 50S Folic acid synthesis Quinolone DNA gyrase Quinolone DNA gyrase Peptide Cell membrane culture lines were propagated in parallel. After 90 days propagation, significant increases in minimum inhibitory concentrations (MICs) were observed in the culture series of all 11 antibiotics except colistin (Fig. 1; all time courses are presented in Supplementary Fig. 1). In addition to these cultures, we observed 90 days propagation of two independent culture lines under the antibiotic-free condition, where all other conditions were identical to the other culture lines, as control. For all resistant strains, we confirmed drug resistances after cultivation for at least 30 generations in the absence of the drug, indicating that the phenotypes of drug resistance were stably memorized. Quantification of cross-resistance and hyper-susceptibility. To explore how the resistance acquisition to one drug changes the resistance and susceptibility to other drugs, for each obtained resistant strain, we measured the MICs of the 25 antibiotics shown in Supplementary Table 1. Figure 2a,b show changes in the MICs of various drugs for chloramphenico (...truncated)