Prophages

We first evaluated VirSorter results against the manually curated prophages from (Casjens, 2003 (link)). Each genome was processed with VirSorter, PhiSpy (Akhter, Aziz & Edwards, 2012 (link)), Phage_Finder (Fouts, 2006 (link)) and PHAST (Zhou et al., 2011 (link)). For each tool, a prophage was considered as “detected” when a prediction covered more than 75% of the known prophage. For a more detailed example case of prophage detection in a complete bacterial genome including both prophages and genomic islands, the same tools were applied to the manually annotated Pseudomonas aeruginosa LES B58 genome (Winstanley et al., 2009 (link)).
VirSorter was then compared with the same prophage detection tools on the set of simulated SAGs. In that case, a viral sequence was considered as detected if predicted as completely viral or as a prophage. All the additional detections were manually checked to verify if the region was indeed viral (originating from a prophage in one of the microbial genomes rather than from a viral genome) or a false positive. The same approach was used for the simulated microbial and viral metagenomes results.
For each set of predictions, two metrics are computed. First, the Recall value corresponds to the number of viral sequences correctly predicted divided by the total number of known viral sequences in the dataset, and reflects the ability of the tool to find every known viral sequence in the dataset. Second, the Precision value is computed as the total number of viral sequences correctly predicted divided by the total number of viral sequences predicted, and indicates how accurate the tool is in its identification of viral signal.

STs were deduced from the genome assemblies using mlst (v2.16.1; https://github.com/tseemann/mlst) and assigned to pre-defined CCs on PubMLST [40 (link)] (https://pubmlst.org/campylobacter). The same database was used to calculate core-genome MLST (cgMLST) (Oxford scheme) on PubMLST [40 (link)]. The relationship between STs, phenotypic resistance, seasonality and the presence of pTet plasmid was assessed using Fisher’s exact test (GraphPad Prism 9.1.2). Source attribution patterns were assigned based on ST–ecotype associations described in recent publications [41–43 (link)]. Gene annotation was carried out from the draft assemblies using Prokka v.1.13 [44 (link)]. The presence of plasmids or prophages was inferred from the assemblies using MOB-suite v.2.0.1 [45 (link)]. The predicted plasmid sequences were used as queries in blastn with threshold values set to >80 % coverage and >95 % identity. The sequences of accession numbers CP017866 and CP014746 were used to define the predicted sequences as pTet and pVir, respectively. Pan-genome analyses were carried out using Roary v.3.12.0 [46 (link)] with an amino acid identity cut-off of 95 % and splitting homologous groups containing paralogues into groups of true orthologues. A summary of the pan-genome composition and visualization of gene diversity is provided in Fig. S1(a–c), available with the online version of this article.
In parallel, whole-genome SNP (wgSNP)-based alignments were built from trimmed reads using the Snippy v.4.3.6 pipeline (https://github.com/tseemann/snippy). The closed genome of strain C. jejuni subsp. jejuni NCTC 11168 (GenBank assembly accession no. GCA_000009085.1) was used as a reference in read mapping. Areas of putative recombination were removed from the resulting alignment using Gubbins v.2.2.0 [47 (link)] and default settings (five iterations and >3 base substitutions to identify a recombination event). Maximum-likelihood phylogenies were obtained from the recombination-removed alignments using the tree building option FastTree v2.1.4 [48 (link)]. The core-genome phylogeny was visualized using iTOL [49 (link)] and the pan-genome genes calculated in Roary were displayed alongside the recombination-removed phylogenetic tree using Phandango [50 (link)] (https://jameshadfield.github.io/phandango). Virulence gene detection was carried out using ABRicate (version 0.8.10; https://github.com/tseemann/abricate) equipped with VFDB (Virulence Factor Database) [51 (link)]. Hits with less than 80 % identity or coverage were filtered out of the analysis.
The PubMLST C. jejuni database was screened for the major flagellin protein, FlaA (encoded by the flaA gene). All 2058 deposited C. jejuni sequences classified as CC-257 (as of November 4th 2022) were searched for the presence of the flaA sequence by blastn [NCBI, National Institutes of Health (NIH)] analysis using the DNA sequence from C. jejuni strain NCTC 11168 as reference. Presence of the gene was determined by >90 % alignment and identity with the query sequence (E value=0). Finally, the presence of type VI secretion system (T6SS) genes, encoding a total of 13 core components (TagH, TssA–TssG, TssI–TssM), was assessed using previously published reference sequences [52 (link)] and the blastn tool. The presence of genes was defined as DNA identity and coverage of ≥90 %.

The presence of AMR genes was scanned in the genomic assemblies of the outbreak strains clade using AMRFinderPlus v3.10.16 [49 (link)]. To assess penicillin susceptibility, the protein sequences of the penicillin binding proteins (PBPs) were extracted and screened for amino acid substitutions known to correlate with decreased penicillin susceptibility in streptococci [50 (link)]. In brief, the sequence of each PBP was aligned using muscle v3.8.1551 [51 (link)] and visualized in SEAVIEW v5.0.5 [52 (link)]. The prokka annotations of the Thai zoonotic clade were queried to identify acquired resistance genes in the outbreak strain using Panaroov1.2.9 [53 (link)]. In addition, the STC78 complete genome was scanned for integrative and conjugative elements (ICEs) and prophages using ICEFinder [54 (link)] and PHASTER [55 (link)], respectively. Acquired AMR genes and their genomic context were manually inspected using Artemis v18.1.0 [56 (link)]. PubMed was searched for primary research articles describing mobile genetic elements (MGEs) carrying the same AMR genes to identify putative homologous MGEs. The annotated MGEs were aligned using clinker and clustermap.js v.0.021 [57 (link)]. The plasmid acquired by the outbreak strain was visualized using ApE v3.0.8 [58 (link)] and a blastn [59 (link)] search was performed against bacterial reference genomes. To assess the presence of potential genes of interest, ABRicate (https://github.com/tseemann/abricate) was used with a custom database containing the sequences of 52 genes previously found to be putatively associated with zoonotic potential of S. suis strains [46 (link)].