These bacteria were isolated from the blood of febrile patients attending Queen Elizabeth Central Hospital in Blantyre, Malawi, as part of the quality assured routine diagnostic microbiology service supported by the Malawi Liverpool Wellcome Programme as described previously. ¹¹ ERS327391 and ERS207185 were previously sequenced by Feasey et al and ERS1509723 and ERS1509734 were sequenced by Gauld et al. ^{11 , 13} using short-read sequencing only, with ERS1509723 and ERS1509734 predicted resistance gene prediction and IncHI1 replicon, unusual for this region. Isolates were recovered from the MLW sample archive and then plated on nutrient agar (Thermo Scientific Oxoid, United Kingdom) and incubated at 37 °C for 18 hours. A single colony was then transferred into 5 ml of nutrient broth (Thermo Scientific Oxoid, United Kingdom) and incubated for 18 hours at 37 °C. The bacteria cells were then concentrated by centrifugation at 3500 rpm for 30 minutes. Total nucleic acids were extracted using the MasterPure complete DNA and RNA Purification Kit (Bioresearch Technologies, United Kingdom) according to the manufacturer’s instructions.

The double-stranded DNA was quantified using the Qubit 4.0 (ThermoFisher Scientific Inc.) fluorometer and normalized to 400ng in 7.5 ul using UltraPure™ Distilled Water (Invitrogen, Life Technologies Limited). The normalised DNA was used for library preparation using the rapid barcoding kit (SBQ-RBK004, Oxford Nanopore Technologies plc) following the manufacturer’s instructions. The prepared library was sequenced on an Oxford Nanopore r9.4.1 flow cell on the MinION Mk1C sequencer (Oxford Nanopore Technologies plc). Data acquisition was done by the MinKNOW ^TM software (version 20.10.6, Oxford Nanopore Technologies plc). Base calling and demultiplexing using Guppy Basecalling Software, Oxford Nanopore Technologies plc. (version 6.0.7+c7819bc). We used Guppy with the default settings and the “dna_r9.4.1_450bps_sup.cfg” configuration file for the guppy_barcoder. We used the command line argument “--trim_barcodes” for the debarcoding step using guppy_barcoder. For guppy_barcoder we passed “SQK-RBK004” to the “--barcode_kits” command line argument.

We retrieved the short reads from ENA and performed trimming and filtering using fastp (version 0.23.1--9f2e255. ^{17 , 18} The reads derived from nanopore sequencing were trimmed and filtered using Filtlong (version 0.2.0--0c4cbe3) with options “--min_length 1000” and “--keep_percent 90” to discard all reads less than 1kb and remove the worst 10% reads. ¹⁹ We used FastQC (v0.11.9_cv7) and NanoPlot (version 1.38.1--e303519) with the default settings to check the quality of the filtered short and long reads respectively. ^{20 , 21}

We performed long-read-first de novo assembly using the Trycycler (v0.5.4) assembly pipeline. ²² Briefly, we created 16 read subsamples using Trycycler subsample. We created draft assemblies on each of the simulated reads using either Flye (version 2.9.1-b1780), Minipolish (version 0.1.2) and Raven (version 1.5.0) assemblers. ^{23 – 25} We used the “--nano-hq” command line argument when generating the Flye assemblies. We run the rest of the assemblers using default settings. We used Any2fasta (version 0.4.2) to convert the gfa formatted output from Minipolish to a fasta file format. ²⁶ We used the generated assemblies to create groups of per-replicon clusters using trycycler cluster. ²² We then performed manual curation steps using trycycler reconcile, trycycler msa, trycycler partition and trycycler consensus following instructions on ( https://github.com/rrwick/Trycycler/wiki/How-to-run-Trycycler ). ²²

We polished the assemblies using medaka (1.8.1), Polypolish (version 0.5.0) and POLCA (MaSuRCA version 4.1.0). ^{27 – 29} The assembly was first polished with the long reads using medaka with “r941_min_sup_g507” passed to the -m command line argument. ²⁷ This polished assembly was further polished using short read data using Polypolish and finally using POLCA. Both Polypolish and POLCA were run using default settings. ^{28 , 29}

The three additional plasmids of early H58 isolates (ERR1764576 (1990, India), ERR1764585 (1992, India), ERR1764588 (1993, India); ⁹ were assembled from Illumina HiSeq 2500 paired end reads using Spades (Ref. 30; version 4.0.0) with the “--plasmid” flag which launches the plasmidSPAdes pipeline that assembles plasmids from whole genome sequencing data. We used the default K-mer values for the assembly.

Bacterial genotyping and plasmid detection was done using the GenoTyphi bacterial typing framework. ^{6 , 31} We implemented GenoTyphi using Mykrobe v0.12.1 with the default settings and the Typhi panel (version 20221208 ^{31 , 32} ). The json output files were parsed into a csv document using the script “parse_typhi_mykrobe.py” provided at the GenoTyphi repository. ³²

We then used AMRFinderPlus (software version 3.11.26 and database version 2023-11-15.1) to assess the presence of AMR genes and plasmids in the assembled contigs. ³³ AMRFinder Plus was run by passing the”–nucleotide –protein –gff –organism –plus” flags. We passed “Salmonella” to the –organism flag. AMRFinderPlus outputs a file with AMR genes and their location on the contigs. The AMR genes are presented in Figure 2. We used ariba (version 2.14.6) using the ARG-ANNOT database to detect AMR genes in the raw reads to annotate the phylogenetic trees. ^{33 , 34}

To detect bacteriophages, we used the PHASTER web tool ( https://phaster.ca/). ^{35 , 36} We loaded the polished assemblies into the web tool and exported the results as a summary.txt file containing the position of phages in the CT18 chromosome, shown in Figure 1 and Figure 2A.

We performed annotation of the hybrid assemblies using Prokka (version 1.14.6) with the default settings and performed a genus-specific annotation with “Salmonella” passed as an argument to the “-–genus” command line option. ³⁷

We used the Artemis Comparison Tool (ACT release 18.2.0) to perform a comparison of assemblies for the isolates sequenced on the MinION platform. ³⁸ We visualised the comparison in the Blast Ring Image Generator (BRIG version 0.95 ³⁹ ). Both of these programs used the Basic Local Alignment Tool (BLAST; Nucleotide-Nucleotide BLAST 2.15.0+) to perform nucleotide sequence comparisons. ⁴⁰ For all sequence comparisons, we used the “blastn” option with the default setting for comparing 2 sequences. We used the CT18 reference genome (AL513382.1) as a reference for the BRIG comparison of the chromosome and used the IncHI1 plasmid from the same isolate (AL513383.1) as a reference for the plasmid comparison. ⁴¹ To assess the presence of AMR genes on the plasmids and parts of the chromosome previously known to contain the insertion of AMR genes, we used the GenoPlotR v 0.8.11 R package which gives a visual comparison of the coding sequences in these regions. ⁴⁰

We constructed a core single-nucleotide polymorphism (SNP) maximum likelihood phylogenetic tree to put the newly assembled isolates in the context of other isolates from Malawi, which uses all conserved sites (i.e. the core genome) when mapping reads against the reference genome. The isolates used in the phylogenetic analysis are from the paper by Gaud et al., where the assembled isolates were first described (7). We used Snippy (version 4.6.0) to perform variant calling and IQ-TREE to construct the phylogenetic tree. We used AL513382.1 as the reference genome for all variant calling steps, snippy-core to create a multiple sequence alignment file with all the variants from the previous step and snp-sites (version 2.5.1) with “-c” argument to extract sites containing exclusively ACGT from the alignment. We also used snp-sites with “-C” to calculate the number of the constant sites which we then used for the “-fconst” argument in IQ-TREE. We performed the IQ-TREE calculation on the alignment with ACGT only generated as described, with the command line parameters “-m GTR+G -bb 1000”.

We performed long-read whole genome sequencing of 4 isolates (ERS327391, ERS207185, ERS1509723, ERS1509734) to better understand unusual antimicrobial resistance gene patterns in 3 isolates from Malawi as observed previously. ¹³ The years of collection for the isolates were between 2010 and 2016. ¹³ Isolates ERS1509723 (BKQT8S) and ERS1509734 (BKQU3X) were selected for their resistance gene pattern, distinct from the rest of the isolates from Malawi from the same period. ¹³ These two isolates are two of only three isolates with IncHI1 plasmids from the same study. ¹³ Isolates ERS327391 (A58390) and ERS207185 (1017142) were selected because they represented isolates at the beginning of the S. Typhi outbreak in Blantyre, Malawi. ¹¹ Strain A58390 represents one of the first H58 isolates causing illnesses during the S. Typhi outbreak. Strain 1017142 represents non-H58 isolates from the same period. ¹¹

The nanopore reads for all samples had a mean Phred Quality score of 13.5 with a mean of 29K (21K - 34K) reads per isolate ( Figure 3). We generated a mean 275 Mb reads (158 Mb-338 Mb) with a mean N50 of 12K (9K - 13K) ( Figure 3).

We generated hybrid assemblies of the four isolates, accession numbers and their metadata are presented in Table 1. Two isolates (A58390 and 1017142) assembled to 1 contig each, and the other two (BKQT8S and BKQU3X) assembled to 2 contigs each. All isolates had a 4.5 Mb contig which represented the chromosome of the isolate. The two isolates with 2 contigs, each had a 185 kb contig that was identified as an IncHI1 plasmid.

The assembled genomes had a mean GC% of 52.0 ( Table 2). Three of the four isolates with long reads belong to the lineage 4.3.1 (H58 haplotype); the two isolates with IncHI1 plasmid were assigned sublineage 4.3.1.1EA1 ⁴² ; the remaining isolate belonging to the lineage 4.1.1 ( Table 2). A pairwise comparison of the assemblies from this study to the Salmonella Typhi reference strain CT18 (AL513382.1) shows a high level of sequence conservation as seen in Figure 1.

We assessed the known chromosomal insertion sites for AMR genes ( Table 3). The insertion site at position 3472059 of the CT18 strain was conserved only in isolate 1017142. The other three isolates had insertion of IS1 insertion sequences ( Figure 2A), whilst the insertion sites at 3815480 and 1690327 of the CT18 strain were conserved across the isolates ( Figure 2B and C).

The plasmids in the isolates BKQT8S and BKQU3X have an average GC content of 45.94% for each isolate. Both isolates had 3 plasmid replicons of the IncHI1 type (IncFIAHI1, IncHI1A and IncHI1BR27; Table 3) which were all on one plasmid. These multi-replicon plasmids are typed as IncHI1 PST2 using pubMLST. ⁴³ A blastn comparison using Megablast shows high levels of similarity between the plasmids ( Figure 4A). Comparing the plasmids to the pHCM1 plasmid isolated from the CT18 strain (AL513383.1) and two of the earliest H58 isolates that acquired IncHI1, ERR1764576 (1990, India), ERR1764585 ( 1992, India), ERR1764588 (1993, India), indicates a loss of coding sequences compared to the early isolates and CT18 in an area with antimicrobial resistance determinants ( Figure 4A); small gaps in the early isolates are mainly mobile element sequences, these gaps are potentially derived from the differences in sequencing methods (short-read-only assemblies of the early isolates). The two plasmid sets have a notable difference to pHCM1 in coding sequences in two regions, HCM1.149 - HCM1.175 (region 1) and HCM1.194 - HCM1.252 (region 2). The first region lost 17 coding sequences, five of these are associated with a mercury resistance operon and one AMR gene ( dhrA14) has relocated to a different area of the plasmids from our study. The second region has been inverted in the Malawian plasmids. The inverted region lost 13 coding sequences, reducing it in size from ~40kb to ~20kb. The missing genes include mercury resistance genes and importantly the two AMR genes catA1 and aph−Id. ⁸ These two AMR genes are responsible for resistance to chloramphenicol and aminoglycosides respectively. Our plasmid maintained aminoglycosides resistance genotype by maintaining a copy of aph(3”)−Ib gene . The AMR genes are highlighted in the ring showing the coding sequences of the reference plasmid ( Figure 4A). Figure 4B further highlights the region in comparison to the pHCM1 plasmid.

Neither lineage 4.1.1 isolate 1017142 nor the 4.3.1 isolate A58390 carried any acquired AMR genes, however there was a gyrA_S83F point mutation in A58390 which is associated with reduced susceptibility to fluoroquinolones. This was the only isolate with mutations in the DNA gyrase. The other two isolates each carried acquired resistance genes ( bla _TEM-1d, sul2, aph(3”)-lb, dfrA14 and tet(B)) on the plasmid which are known to cause resistance to beta-lactams, sulfonamides, aminoglycoside, trimethoprim and tetracyclines.

Using PHASTER to define phage elements on the assemblies reveals at least four intact bacteriophages, one putative phage call and several incomplete bacteriophages per isolate Figure 5. Isolates BKQU3X and BKQT8S show similarity in the number of bacteriophage species per region and the completeness of the bacteriophages in these regions ( Figure 5). Both isolates contain phages similar in composition and size ( Figure 5B).

Performing a phylogenetic analysis highlights the 4.3.1.1.EA1 plasmid and AMR profiles ( Figure 6). Most of the 4.3.1.1.EA1 isolates do not carry plasmids. Those which carry plasmids seem to be in two plasmid combinations, with isolates from this study carrying the plasmid replicon Inc-FIAHI1 which is missing in the isolates from Kenya, while those from Kenya carry the plasmid replicon Inc-HI1-ST6 which is not present in the isolates from Malawi, indicating this was not a direct import event of the lineage from Kenya via traveller. Isolates with plasmids in the Kenyan collection also have additional AMR genes compared to the rest of the 4.3.1.1.EA1 lineage. Isolates which are not 4.3.1.1.EA1 are clustered in distinct clades and uniform AMR and plasmid profiles. There are distinct differences in acquired AMR genes between isolates of the lineage 4.3.1.1.EA1 and the other H58 lineages. The acquired AMR genes present in the lineage 4.3.1.1.EA1 are absent in the other H58 lineages and vice versa.