Phylodynamic modelling of bacterial outbreaks using nanopore sequencing

Abstract

Nanopore sequencing and phylodynamic modelling have been used to reconstruct the transmission dynamics of viral epidemics, but their application to bacterial pathogens has remained challenging. Here, we implement Random Forest models for single nucleotide polymorphism (SNP) polishing to estimate divergence and effective reproduction numbers (Re) of two community-associated, methicillin-resistant Staphylococcus aureus (MRSA) outbreaks in remote Far North Queensland and Papua New Guinea (n = 159). Successive bar-coded panels of S. aureus isolates (2 × 12 per MinION) sequenced at low-coverage (> 5x - 10x) provided sufficient data to accurately infer assembly genotypes with high recall when compared with Illumina references. De novo SNP calling with Clair was followed by SNP polishing using intra- and inter-species models trained on Snippy reference calls. Models achieved sufficient resolution on ST93 outbreak sequence types (> 70 - 90% accuracy and precision) for phylodynamic modelling from lineage-wide hybrid alignments and birth-death skyline models in BEAST2. Our method reproduced phylogenetic topology, geographical source of the outbreaks, and indications of sustained transmission (Re > 1). We provide Nextflow pipelines that implement SNP polisher training, evaluation, and outbreak alignments, enabling reconstruction of within-lineage transmission dynamics for infection control of bacterial disease outbreaks using nanopore sequencing.

Genomic neighbor typing for bacterial outbreak surveillance

“…We next evaluated Sketchy genotyping on two S. aureus outbreaks from remote communities in Papua New Guinea and Far North Queensland (n = 160), which had been sequenced at low-coverage using a dual-library protocol with interspersed nuclease washes (24 strains per MinION flow cell, on a total of 8 flow cells) (4) (Fig. 3, Methods).…”

“…We then assessed ’s performance on an outbreak dataset of S. aureus community infections in Papua New Guinea and Far North Queensland (4, 17, 18), for which we had previously generated matching Illumina reference data (n = 160). In this context, the outbreaks constituted an independent validation dataset, as no S. aureus genomes from these regions had ever been sequenced before (including the very first S. aureus genomes from Papua New Guinea).…”

“…We then selected a series of read cut-offs for predictions (10, 50, 100, 200, 300, 500 and 1000 reads). Ultimately, we selected to report results at the 1000 read threshold for several reasons: first, the threshold marks around 1-3x coverage of the S. aureus genome (depending on read length), after which it becomes feasible to do assembly based genotyping with high recall from nanopore data alone, as demonstrated previously for these outbreak data (20); second, our primary aim was to infer genotypes from as few reads as possible and initial simulations indicated stabilisation of predictions below 1000 reads (Fig. S1); third, reporting by time (as in RASE) is highly volatile due to differences in throughput between libraries (e.g.…”

“…However, whole genomes often cannot be assembled easily from complex metagenomic samples, including blood and lower respiratory infections where the causative pathogen may be at low abundance. Nanopore sequencing is particularly suited for rapid characterisation of pathogen genomes, with potential to be conducted on-site rather than sent to a reference lab, as reads can be streamed from the device and analysed on mobile computing platforms (1)(2)(3)(4). Several methods for bacterial pathogens characterisation on nanopore platforms has been developed 16 over the past few years, including pipelines for batch assembly 17 and marker detection (5,6), novel algorithms for streaming 18 assembly and genotyping (7)(8)(9), and sensitive approaches to 19 antimicrobial resistance prediction, as well as taxonomic iden-20 tification (10)(11)(12).…”

“…In addition to the outbreak in the remote highland provinces of Papua New Guinea, we had become aware of escalating S. aureus infections in remote communities of Far North Queensland, which borders Papua New Guinea through the Torres Strait Islands (18, 19). We collected a snap-shot of strains from Far North Queensland communities (Cape York Peninsula, Cairns and Hinterland, Torres Strait Island) in 2019 (20). This presents a realistic scenario, where surveying cross-border bacterial outbreaks using a comprehensive genomic neighbor typing approach on nanopore devices could provide lineage and genotype data important for deciphering geographical provenance (lineage attribution) and antibiotic susceptibilities of dominant outbreak lineages, potentially informing treatment options.…”