Our study comprised isolates from countries in Latin America and Europe that were previously serotyped as Salmonella Paratyphi B var. Java. We selected strains from different countries as follows.

We isolated genomic DNA from selected strains from the Netherlands, Colombia, Costa Rica, and Guatemala using the UltraClean Microbial DNA Isolation Kit (QIAGEN, https://www.qiagen.com). We performed WGS on the MiSeq platform using 2 × 250-bp reads and the NextSeq platform using two 2 × 150-bp reads (Illumina, https://www.illumina.com). WGS of historical strains was performed at the Plateforme de microbiologie mutualisée from the Pasteur International Bioresources network (PIBnet, Institut Pasteur, https://www.pasteur.fr) as previously described (36). We assembled the genomes of historical and newly sequenced strains with SPAdes version 3.10.1 (37) and screened for antimicrobial resistance genes and chromosomal mutations using ResFinder 3.1 (38). We subtyped plasmids using PlasmidFinder 2.0 and plasmid MLST (pMLST) 2.0 (39). For newly sequenced genomes, we performed 7-gene MLST at the strain level with MLST 1.8 (40).

For phylogenetic single-nucleotide polymorphism (SNP) analysis of the core genome, we aligned WGS of all Salmonella Paratyphi B. var. Java ST28 isolates using Parsnp version 1.2 (41), excluding non-ST28 strains. We used Gubbins (42) to detect and visualize recombination regions in the core genome alignment and performed time-resolved phylogeny on recombination-filtered SNPs of the Salmonella Paratyphi B. var. Java ST28 isolates that were extracted from the Gubbins results and used for divergence dating in BEAST (43). We included only newly obtained genomes with coverage >30. We used isolation dates as tip dates in the phylogenetic tree, as outlined in the BEAST manual, with the following modifications: 10,000,000× sampling and a general time-reversible model plus γ correction as the distance model. We tested a strict clock, relaxed logarithmic clock, and relaxed exponential clock as the clock model. We used a Bayesian skyline plot with 3 groups as demographic models to adjust for expected population changes and the effective sample size (ESS) method to select the strict clock model because it had the highest ESS values. All ESS values obtained with the strict clock model were >1,000. To generate Bayesian skyline plots for the European and Latin American populations, we repeated time-resolved phylogeny analysis using BEAST on 2 subsets containing the historical isolates and the isolates from Europe and Latin America.

We annotated genomes using Prokka version 1.13 (44), followed by orthology predictions using Roary (45). We differentiated chromosome and plasmid contigs with an in-house built tool. In brief, we scored contigs for the presence of known plasmid genes (https://github.com/aldertzomer/RFPlasmid), single-copy chromosomal marker genes (46), and kmer profiles and inferred their likely origin (plasmid or chromosomal) using a Random Forest model trained on known plasmid and chromosome assemblies (A.L. Zomer, unpub. data).

We conducted principal component analysis (PCA) for all isolates on the gene presence/absence tables from the output of orthology predictions. We made comparisons using the accessory (noncore) genome of chromosome contigs, complete plasmid composition with all plasmid contigs, and plasmid composition with only plasmid contigs >50 kb. We made additional characterizations of prophage sequences using the PHAge Search Tool Enhanced Release (47) and BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi).

We deposited sequences of the historical and newly sequenced strains in the short-read archive of the European Nucleotide Archive (ENA) under project no. PRJEB31547. Accession numbers of all collected genomes are provided (Appendix 1 Table 1).

Most of the strains from poultry that we sequenced from Colombia (48/52), Costa Rica (15/15), Guatemala (4/5), and the Netherlands (17/17) belonged to ST28 (Table). Similarly, nonbroiler strains from a fish product and a turkey in the Netherlands and 1 from a pig carcass in Costa Rica belonged to ST28. Isolates from 3 human and 1 reptile samples from Costa Rica or the Netherlands belonged to STs different from ST28 (Appendix 1 Table 1).

In the phylogenetic trees, the strains collected from the Latin America countries formed a separate cluster. In contrast, the strains collected from the Netherlands clustered with strains originating from other countries in Europe (Figure 1). An additional cluster was formed by the historical strains, which were neither from Europe nor Latin America. The molecular clock was estimated at 3.5 × 10^–7 substitutions/site/year (1.7 SNPs/genome/year, 95% CI 1.44–2.0 SNPs/genome/year). The output of BEAST indicated separation between strains from Europe and Latin America occurred in 1987 (95% CI 1978–1988) (Figure 1; Appendix 2 Figure 1). From the Bayesian Skyline plot, we inferred that the effective population size (N_e) of strains from Europe increased sharply in 1995 (95% CI 1992–1998) and in Latin America in 2005 (95% CI 2001–2007) (Figure 2), 10 years later than in Europe.

We found a chromosomal class 2 integron (with dfrA1-sat1-aadA1 [GenBank accession no. AB188271.1]) in historical strain IP_6155/87, collected in 1987 in Saudi Arabia. However, this integron was not in the other 2 historical strains from Austria (originating from Israel [IP_6395/88]) and Saudi Arabia [IP_7734/92]). In strain IP_6155/87, the alignment of the class 2 integron was divided into multiple contigs. We also found the integron in all strains from Europe and Latin America in contigs of ≈50 kb carrying the complete integron or divided into multiple contigs. Alignments of ≈50 kb contigs carrying the complete integron, revealed 100% identity within and between strains from European and Latin American clades.

Resistance to β-lactams in strains from Europe was mainly mediated by bla_TEM-1B or bla_TEM-1B-like genes. These genes were found in 64% of isolates in the European clade. In strains from the Latin American clade, bla_CMY-2 was most prevalent and was carried in 50% of strains (Figure 1). bla_TEM-1B gene is known to confer resistance to aminopenicillins and bla_CMY-2 gene to extended-spectrum cephalosporins. In the European clade, bla_TEM-1B/1B-like was mainly found on IncI1 plasmids (17/42) and, to a lesser extent, on IncX4 (7/42) and IncHI2 (3/42) plasmids. In strains from the Latin American clade, bla_CMY-2 was mainly found on IncI1/ST12 plasmids (25/33) (Figure 1). The sul2 gene, conferring resistance to sulphonamides, was frequently encountered in strains from Europe (48%) and Latin America (14%). In the European clade, sul2 was mainly found to co-localize with bla_TEM-1B/1B-like on IncI1 plasmids (17/30), whereas in the Latin American clade, sul2 was found on ColRNAI-like plasmids (9/9). β-lactam and sulphonamide resistance genes in the Latin American clade were observed only in strains from Colombia (Figure 1). In addition, >50% of strains from the European clade exhibited known DNA gyrase mutations conferring resistance to fluoroquinolones. The observed mutations were gyrA D87G (16/33), D87Y (10/33), and S83F (7/33) (Figure 1). Although no strains from the Latin American clade carried the chromosomal gyrA mutations, they did carry qnrB19-harboring plasmids in 96% of the cases. qnrB19 is known to confer reduced susceptibility to quinolones. This gene was co-localized with ColRNAI-like plasmids in strains from Latin America and in 1 strain from Europe from recent years (2015) (Figure 1).

We observed a marked distinction for the accessory genes located on chromosome contigs between the European and Latin American clades (Figure 3). The separation in the PCA was associated with the presence of a prophage highly similar to the Salmonella phage SEN34 (National Center for Biotechnology Information reference sequence NC_028699.1) in strains from Latin America. This phage was also found in the genome of a Salmonella serovar Saintpaul strain submitted in Canada (GenBank accession no. CP022491.1). A few strains from Colombia closer to the cluster of strains from Europe in the PCA (UGBOG142, UGVIL373, and SSIII_4_C2) lacked the sequence of this phage. Plasmid composition was similar in strains from Europe and Latin America (Figure 4). The profiles of plasmid composition were characterized by the presence of IncI1 plasmids (cluster I), IncHI2 (cluster II), ColRNAI-like (cluster III), and combinations of IncI1 and IncHI2 plasmids (cluster IV). Although the IncI1 plasmids had different pMLST sequence types, their content appears to be remarkably similar as they are in proximity in the PCA plot (cluster I). When exploring this further using only plasmid contigs >50 kb in strains from both European and Latin American clades, we observed that the IncI1 plasmid contigs clearly have near identical content (cluster I in Appendix 2 Figure 2). In strains with multiple plasmid contigs, in addition to the cluster of IncI1-like plasmids, IncHI2-like plasmid contigs were differentiated in cluster II (Appendix 2 Figure 2).