Methods and Software

The partitioning of pathogenic strains isolated in environmental or human cases to their sources is challenging. The pathogens usually colonize multiple animal hosts, including livestock, which contaminate the food-production chain and the environment (e.g. soil and water), posing an additional public-health burden and major challenges in the identification of the source. Genomic data opens up new opportunities for the development of statistical models aiming to indicate the likely source of pathogen contamination. Here, we propose a computationally fast and efficient multinomial logistic regression source-attribution classifier to predict the animal source of bacterial isolates based on ‘source-enriched’ loci extracted from the accessory-genome profiles of a pangenomic dataset. Depending on the accuracy of the model’s self-attribution step, the modeller selects the number of candidate accessory genes that best fit the model for calculating the likelihood of (source) category membership. The Accessory genes-Based Source Attribution (AB_SA) method was applied to a dataset of strains of Salmonella enterica Typhimurium and its monophasic variant ( S . enterica 1,4,[5],12:i:-). The model was trained on 69 strains with known animal-source categories (i.e. poultry, ruminant and pig). The AB_SA method helped to identify 8 genes as predictors among the 2802 accessory genes. The self-attribution accuracy was 80 %. The AB_SA model was then able to classify 25 of the 29 S . enterica Typhimurium and S . enterica 1,4,[5],12:i:- isolates collected from the environment (considered to be of unknown source) into a specific category (i.e. animal source), with more than 85 % of probability. The AB_SA method herein described provides a user-friendly and valuable tool for performing source-attribution studies in only a few steps. AB_SA is written in R and freely available at https://github.com/lguillier/AB_SA.

The standard workhorse for genomic analysis of the evolution of bacterial populations is phylogenetic modelling of mutations in the core genome. However, a notable amount of information about evolutionary and transmission processes in diverse populations can be lost unless the accessory genome is also taken into consideration. Here, we introduce panini (Pangenome Neighbour Identification for Bacterial Populations), a computationally scalable method for identifying the neighbours for each isolate in a data set using unsupervised machine learning with stochastic neighbour embedding based on the t-SNE (t-distributed stochastic neighbour embedding) algorithm. panini is browser-based and integrates with the Microreact platform for rapid online visualization and exploration of both core and accessory genome evolutionary signals, together with relevant epidemiological, geographical, temporal and other metadata. Several case studies with single- and multi-clone pneumococcal populations are presented to demonstrate the ability to identify biologically important signals from gene content data. panini is available at http://panini.pathogen.watch and code at http://gitlab.com/cgps/panini.

Hypermutable simple sequence repeats (SSRs) are drivers of phase variation (PV) whose stochastic, high-frequency, reversible switches in gene expression are a common feature of several pathogenic bacterial species, including the human pathogen Campylobacter jejuni. Here we examine the distribution and conservation of known and putative SSR-driven phase variable genes – the phasome – in the genus Campylobacter. PhasomeIt, a new program, was specifically designed for rapid identification of SSR-mediated PV. This program detects the location, type and repeat number of every SSR. Each SSR is linked to a specific gene and its putative expression state. Other outputs include conservation of SSR-driven phase-variable genes and the ‘core phasome’ – the minimal set of PV genes in a phylogenetic grouping. Analysis of 77 complete Campylobacter genome sequences detected a ‘core phasome’ of conserved PV genes in each species and a large number of rare PV genes with few, or no, homologues in other genome sequences. Analysis of a set of partial genome sequences, with food-chain-associated metadata, detected evidence of a weak link between phasome and source host for disease-causing isolates of sequence type (ST)-828 but not the ST-21 or ST-45 complexes. Investigation of the phasomes in the genus Campylobacter provided evidence of overlapping but distinctive mechanisms of PV-mediated adaptation to specific niches. This suggests that the phasome could be involved in host adaptation and spread of campylobacters. Finally, this tool is malleable and will have utility for studying the distribution and genic effects of other repetitive elements in diverse bacterial species.

Assembly of bacterial short-read whole-genome sequencing data frequently results in hundreds of contigs for which the origin, plasmid or chromosome, is unclear. Complete genomes resolved by long-read sequencing can be used to generate and label short-read contigs. These were used to train several popular machine learning methods to classify the origin of contigs from Enterococcus faecium, Klebsiella pneumoniae and Escherichia coli using pentamer frequencies. We selected support-vector machine (SVM) models as the best classifier for all three bacterial species (F1-score E. faecium=0.92, F1-score K. pneumoniae=0.90, F1-score E. coli=0.76), which outperformed other existing plasmid prediction tools using a benchmarking set of isolates. We demonstrated the scalability of our models by accurately predicting the plasmidome of a large collection of 1644 E. faecium isolates and illustrate its applicability by predicting the location of antibiotic-resistance genes in all three species. The SVM classifiers are publicly available as an R package and graphical-user interface called ‘mlplasmids’. We anticipate that this tool may significantly facilitate research on the dissemination of plasmids encoding antibiotic resistance and/or contributing to host adaptation.

Plasmid prediction may be of great interest when studying bacteria of medical importance such as Enterobacteriaceae as well as Staphylococcus aureus or Enterococcus. Indeed, many resistance and virulence genes are located on such replicons with major impact in terms of pathogenicity and spreading capacities. Beyond strain outbreak, plasmid outbreaks have been reported in particular for some extended-spectrum beta-lactamase- or carbapenemase-producing Enterobacteriaceae. Several tools are now available to explore the ‘plasmidome’ from whole-genome sequences with various approaches, but none of them are able to combine high sensitivity and specificity. With this in mind, we developed PlaScope, a targeted approach to recover plasmidic sequences in genome assemblies at the species or genus level. Based on Centrifuge, a metagenomic classifier, and a custom database containing complete sequences of chromosomes and plasmids from various curated databases, PlaScope classifies contigs from an assembly according to their predicted location. Compared to other plasmid classifiers, PlasFlow and cBar, it achieves better recall (0.87), specificity (0.99), precision (0.96) and accuracy (0.98) on a dataset of 70 genomes of Escherichia coli containing plasmids. In a second part, we identified 20 of the 21 chromosomal integrations of the extended-spectrum beta-lactamase coding gene in a clinical dataset of E. coli strains. In addition, we predicted virulence gene and operon locations in agreement with the literature. We also built a database for Klebsiella and correctly assigned the location for the majority of resistance genes from a collection of 12 Klebsiella pneumoniae strains. Similar approaches could also be developed for other well-characterized bacteria.

The genus Escherichia is composed of Escherichia albertii, E. fergusonii, five cryptic Escherichia clades and E. coli sensu stricto. Furthermore, the E. coli species can be divided into seven main phylogroups termed A, B1, B2, C, D, E and F. As specific lifestyles and/or hosts can be attributed to these species/phylogroups, their identification is meaningful for epidemiological studies. Classical phenotypic tests fail to identify non-sensu stricto E. coli as well as phylogroups. Clermont and colleagues have developed PCR assays that allow the identification of most of these species/phylogroups, the triplex/quadruplex PCR for E. coli phylogroup determination being the most popular. With the growing availability of whole genome sequences, we have developed the ClermonTyping method and its associated web-interface, the ClermonTyper, that allows a given strain sequence to be assigned to E. albertii, E. fergusonii, Escherichia clades I–V, E. coli sensu stricto as well as to the seven main E. coli phylogroups. The ClermonTyping is based on the concept of in vitro PCR assays and maintains the principles of ease of use and speed that prevailed during the development of the in vitro assays. This in silico approach shows 99.4 % concordance with the in vitro PCR assays and 98.8 % with the Mash genome-clustering tool. The very few discrepancies result from various errors occurring mainly from horizontal gene transfers or SNPs in the primers. We propose the ClermonTyper as a freely available resource to the scientific community at: http://clermontyping.iame-research.center/.

The potential for genome-wide modelling of epistasis has recently surfaced given the possibility of sequencing densely sampled populations and the emerging families of statistical interaction models. Direct coupling analysis (DCA) has previously been shown to yield valuable predictions for single protein structures, and has recently been extended to genome-wide analysis of bacteria, identifying novel interactions in the co-evolution between resistance, virulence and core genome elements. However, earlier computational DCA methods have not been scalable to enable model fitting simultaneously to 104–105 polymorphisms, representing the amount of core genomic variation observed in analyses of many bacterial species. Here, we introduce a novel inference method (SuperDCA) that employs a new scoring principle, efficient parallelization, optimization and filtering on phylogenetic information to achieve scalability for up to 105 polymorphisms. Using two large population samples of Streptococcus pneumoniae, we demonstrate the ability of SuperDCA to make additional significant biological findings about this major human pathogen. We also show that our method can uncover signals of selection that are not detectable by genome-wide association analysis, even though our analysis does not require phenotypic measurements. SuperDCA, thus, holds considerable potential in building understanding about numerous organisms at a systems biological level.

Increasingly rich metadata are now being linked to samples that have been whole-genome sequenced. However, much of this information is ignored. This is because linking this metadata to genes, or regions of the genome, usually relies on knowing the gene sequence(s) responsible for the particular trait being measured and looking for its presence or absence in that genome. Examples of this would be the spread of antimicrobial resistance genes carried on mobile genetic elements (MGEs). However, although it is possible to routinely identify the resistance gene, identifying the unknown MGE upon which it is carried can be much more difficult if the starting point is short-read whole-genome sequence data. The reason for this is that MGEs are often full of repeats and so assemble poorly, leading to fragmented consensus sequences. Since mobile DNA, which can carry many clinically and ecologically important genes, has a different evolutionary history from the host, its distribution across the host population will, by definition, be independent of the host phylogeny. It is possible to use this phenomenon in a genome-wide association study to identify both the genes associated with the specific trait and also the DNA linked to that gene, for example the flanking sequence of the plasmid vector on which it is encoded, which follows the same patterns of distribution as the marker gene/sequence itself. We present PlasmidTron, which utilizes the phenotypic data normally available in bacterial population studies, such as antibiograms, virulence factors, or geographical information, to identify traits that are likely to be present on DNA that can randomly reassort across defined bacterial populations. It is also possible to use this methodology to associate unknown genes/sequences (e.g. plasmid backbones) with a specific molecular signature or marker (e.g. resistance gene presence or absence) using PlasmidTron. PlasmidTron uses a k-mer-based approach to identify reads associated with a phylogenetically unlinked phenotype. These reads are then assembled de novo to produce contigs in a fast and scalable-to-large manner. PlasmidTron is written in Python 3 and is available under the open source licence GNU GPL3 from https://github.com/sanger-pathogens/plasmidtron.

Public health and food safety institutions around the world are adopting whole genome sequencing (WGS) to replace conventional methods for characterizing Salmonella for use in surveillance and outbreak response. Falling costs and increased throughput of WGS have resulted in an explosion of data, but questions remain as to the reliability and robustness of the data. Due to the critical importance of serovar information to public health, it is essential to have reliable serovar assignments available for all of the Salmonella records. The current study used a systematic assessment and curation of all Salmonella in the sequence read archive (SRA) to assess the state of the data and their utility. A total of 67 758 genomes were assembled de novo and quality-assessed for their assembly metrics as well as species and serovar assignments. A total of 42 400 genomes passed all of the quality criteria but 30.16 % of genomes were deposited without serotype information. These data were used to compare the concordance of reported and predicted serovars for two in silico prediction tools, multi-locus sequence typing (MLST) and the Salmonella in silico Typing Resource (SISTR), which produced predictions that were fully concordant with 87.51 and 91.91 % of the tested isolates, respectively. Concordance of in silico predictions increased when serovar variants were grouped together, 89.25 % for MLST and 94.98 % for SISTR. This study represents the first large-scale validation of serovar information in public genomes and provides a large validated set of genomes, which can be used to benchmark new bioinformatics tools.