The effect of sequencing and assembly on the inference of horizontal gene transfer on chromosomal and plasmid phylogenies

The spread of antibiotic resistance genes on plasmids is a threat to human and animal health. Phylogenies of bacteria and their plasmids contain clues regarding the frequency of plasmid transfer events, as well as the co-evolution of plasmids and their hosts. However, whole genome sequencing data from diverse ecological or clinical bacterial samples is rarely used to study plasmid phylogenies and resistance gene transfer. This is partially due to the difficulty to extract plasmids from short-read sequencing data. Here, we use both short- and long-read sequencing data of 24 clinical extended-spectrum β-lactamase producing Escherichia coli to estimate chromosomal and plasmid phylogenies. We compare the impact of different sequencing and assembly methodologies on these phylogenies and on the inference of horizontal gene transfer. We find chromosomal phylogenies can be estimated robustly with all methods, whereas plasmid phylogenies have more variable topology and branch lengths across the methods used. Specifically, hybrid methods that use long reads to resolve short-read assemblies (HybridSPAdes and Unicycler) perform better than those that started from long-reads during assembly graph generation (Canu). In contrast, the inference of plasmid and antibiotic resistance gene transfer using a parsimony-based criterion is mostly robust to the choice of sequencing and assembly method.