Detecting horizontal co-transfer of antimicrobial resistance genes in bacteria: a network approach

Abstract

Antimicrobials have been widely using as a major resource to treat bacterial infections for almost a century. However, it is not unusual to see antimicrobial resistance emerges in a bacterial species due to natural selection under the usage of antimicrobials. Moreover, numerous studies show that bacteria can accumulate genes encoding resistance to different classes of antimicrobials and share them with other bacteria regardless of ancestry via a biological process called horizontal gene transfer, causing emergence and fast transmission of multidrug resistance. As such, antimicrobial resistance becomes an urgent and global threat to public health, pushing us backwards to the pre-antimicrobial era.

In this thesis, I focus on horizontal co-transfer of resistance genes between bacteria of the same species, which is usually caused by co-localisation of resistance genes in mobile genetic elements, also known as physical linkage between these genes. This kind of linkage plays a pivotal role in the evolution of multidrug resistance, because the mobile elements can translocate, recombine and aggregate, rapidly rendering their host bacteria resistant to a wide spectrum of antimicrobials. By far there is nonetheless not an approach identifying horizontally co-transferred genes in a single bacterial species. Yet most authors of literature reported a few co-mobilised resistance genes each time following biological experiments, and some researchers only applied simple association analysis to representative bacterial isolates of distinct species so as to minimise the possibility that a specific combination of genes is inherited from their most-recent common ancestor. In contrast, intra-species association analysis is severely confounded by strong sample relatedness because of bacterial clonal reproduction. This obstacle leaves a gap between the known high frequency of intra-species horizontal gene transfer and our understandings of this process.

This thesis presents a scalable computational approach that uses whole-genome sequencing data to identify co-transferred antimicrobial resistance genes in bacteria collected in a few decades from the same species. Moreover, it demonstrates applications of the approach to three clinically important pathogens and reports key players, patterns and dynamics underlying the horizontal co-transfer of resistance genes within each species.

In the first chapter, I provide a background to antimicrobial resistance, horizontal gene transfer, whole-genome sequencing and contemporary bioinformatic techniques. I also summarise outcomes of horizontal gene co-transfer for characteristics that we can utilise for inference of physical linkage. Finally, I compare several statistical approaches determining pairwise association between presence-absence status of genes or alleles in bacteria to justify the necessity of controlling for sample relatedness in association analysis.

For the second chapter, I derived a methodology inferring co-transferred genes by integrating gene detection, de novo genome assembly, core-genome and phylogenetic analysis, linear mixed models, hypothesis tests for effects of sample relatedness and evaluation of consistency in pairwise physical distances between resistance alleles in bacterial genomes. This methodology is designed to overcome limitations of existing methods summarised in the first chapter. Moreover, I show interpretations of expected outcomes and discuss constraints of this approach.

The next three chapters present an implementation of my methodology and its applications on antimicrobial resistance genes in three clinically important species of Enterobacteriaceae. First, I conducted an empirical study following a simulation-and-validation strategy on finished-grade full genomes of six strains of Klebsiella pneumoniae to find out an optimal method that measures the pairwise physical distances between alleles in de novo genome assemblies. I found that for each assembly graph, the most accurate measurements are obtained via setting up constraints for both the number of nodes in the graph and the maximum of distance measurements. Second, I developed GeneMates, a computational and integrative software package that implements my methods proposed in the second chapter for the identification of physically linked resistance alleles or for analysing associations between a large number of resistance alleles when controlling for individual relatedness. In particular, GeneMates leverages network topology to identify potential physical linkage between the alleles. For validation, I applied this tool to whole-genome sequencing data of Escherichia coli and Salmonella Typhimurium, whose acquired resistance genes and relevant mobile genetic elements have been well characterised in publications. In result sections, I illustrate clusters of physically linked resistance alleles and discoveries of their vectors. For the last result chapter, I applied GeneMates to genomes of a large global collection of K.pneumoniae strains, which are adept to uptake DNA from various environments. Furthermore, I compared structure and contents of co-localised allele clusters across time and geography and discovered patterns underlying the evolution of multidrug resistance in this species.

To conclude, I have developed and implemented a network approach that performs association tests on presence-absence of resistance alleles in a large collection of bacterial isolates of the same species and infers potential horizontally co-transferred alleles. I have validated this approach using known co-mobilisable resistance genes and the approach showed higher statistical power than existing methods. The GeneMates package will become a powerful tool contributing to routine surveillance of antimicrobial resistance and identifications of known and novel mobile genetic elements. In addition, applications of this package to other kinds of bacterial genes is also feasible and convenient.