Salmonella enterica Serovar Typhi in Bangladesh: Exploration of Genomic Diversity and Antimicrobial Resistance

Typhoid fever, caused by Salmonella enterica serovar Typhi, is responsible for an estimated burden of approximately 17 million new episodes per year worldwide. Adequate and timely antimicrobial treatment invariably cures typhoid fever. The increasing antimicrobial resistance (AMR) of S. Typhi severely limits the treatment options. We studied whole-genome sequences (WGS) of 536 S. Typhi isolates collected in Bangladesh between 1999 and 2013 and compared those sequences with data from a recent outbreak in Pakistan and a laboratory surveillance in Nepal. The analysis suggests that multiple ancestral origins of resistance against ciprofloxacin and ceftriaxone are present in three countries. Such independent genetic events and subsequent dissemination could enhance the risk of a rapid global spread of these highly resistant clones. Given the current treatment challenges, vaccination seems to be the most appropriate short-term intervention to reduce the disease burden of typhoid fever at a time of increasing AMR.

Phylogenetic relationships and new lineages. Genotypes 3.2.2 and 3.3 clustered together in the same classical MLST type (ST2) but formed two distinct subclades in the unweighted pair group method using average linkages (UPGMA) tree based on cgMLST analyses of 3,002 core loci ( Fig. 1a and b). Like genotype 2.3.3, genotype 4.3.1 also generated its own subclade in the tree, with the notable presence of multiple subgroups (Fig. 1b). The same UPGMA tree presented a distinct population structure and a similar genotypic differentiation as observed in the maximum likelihood tree (MLT), which was based on 2,328 SNPs from our WGS data compared to the S. Typhi CT18 reference genome (Fig. 1c) (30). Genotype 1.2.1 mapped closest to the root of the MLT; genotype 4.3.1 was the most remote ( Fig. 1c). As the dominant genotype, 4.3.1 formed a large subclade, with genotype 4.1 in its primary clade. Another primary clade divided into two major subclades, genotypes 3.3 and 3.2.2, which comprised the second and third most prevalent genotypes in Bangladesh. Two small subclades of genotype 2.3.3 and genotype 2.0 were also present in the MLT, rooting with genotype 2.2 and 2.1.7, respectively.
The comparative MLT created using our data and the strains from neighboring countries showed country-specific clusters inside the dominant 4.3.1 genotype and also in genotypes 3.3 and 3.2.2 ( Fig. 2 and 3). This suggests different points of origin for the various lineages of S. Typhi in each country. All XDR Pakistani isolates extended into a single branch of the MLT, showing H58 lineage Ia and a minimally divergent pattern ( Fig. 3; see also Fig. S1 in the supplemental material), whereas the Nepali strains were dominant in lineage II. The isolates from Bangladesh included only four isolates from lineage II but showed two distinct clusters inside genotype 4.3.1 (H58): lineage Ia (n ϭ 223) and a previously nondescribed lineage (n ϭ 108; Fig. 3). The latter lineage did not match the SNP definition of H58 lineage I or lineage II, suggesting a previously undetected H58 lineage (genotype 4.3.1). On the basis of the MLT, this nondescribed lineage could have had the same point of origin as lineage I but then clearly followed a different pattern of divergence and formed its own subclade inside genotype 4.3.1 (Fig. 3). This new H58 lineage can be distinguished by the SNPs at nucleotide position 561056 (C¡A) and 2849843 (A¡C) of the CT18 reference genome (this previously undescribed lineage is referred to as "lineage Bd" in the remainder of the article). Genomic Diversity and AMR of S. Typhi in Bangladesh ® Resistance phenotypes and genotypes. On the basis of analyses performed with five different antibiotics-ampicillin (amp), chloramphenicol (chl), co-trimoxazole (sxt), ciprofloxacin (cip), and ceftriaxone (cro)-the 536 S. Typhi isolates from Bangladesh were found to harbor 12 different phenotypic resistance profiles (phenotypes; Table 2). Isolates with the "MDR, cip-R" profile (n ϭ 202) were most prevalent in our library, followed by "cip-R only" (n ϭ 169) and "Susceptible to all" (n ϭ 62). A comparison of MDR and ciprofloxacin-resistant isolates with different genotypes is presented in Table 3. The single ceftriaxone-resistant (cro-R) strain (MIC Ͼ 32 g/ml) (susceptible to sxt, chl, and cip) in our library, isolated in 2000, displayed genotype 3.3 (haplotype H1) and contained the bla CTX-M-15 gene (ceftriaxone resistance). The other resistance genes  Table 4, including bla TEM-1B (ampicillin resistance); catA1 (chloramphenicol resistance); dfrA7, sul1, and sul2 (co-trimoxazole resistance); and qnrS1 (ciprofloxacin resistance).
Comparison of phenotypic and WGS-derived resistance profiles. On the basis of the resistance genes identified for the five antimicrobial agents, a WGS resistance (WGS-res) profile was assigned and compared with the phenotypic profile of each isolate to evaluate the ability of the WGS approach to predict the resistance phenotype (Table 5). For all antimicrobial agents except ciprofloxacin, the two profiles corresponded at a level of 99% for the resistant isolates. In contrast, some susceptible isolates (n ϭ 33) harbored resistance genes, which reduced the specificity of the method (Ն91%). Three of them had truncated genes (considered inactive genes; Table 5), but the other 30 isolates had the complete coding sequences without any phenotypic resistance. This might suggest impairments (e.g., transcriptomic, translational, protein modification, etc.) in downstream steps of the resistance pathway or the presence of counteracting genes.
On the other hand, for isolates with resistant phenotypes but susceptible WGS res-profiles, we screened for mutations in genes with efflux pump or membrane permeability functions (see Table S1 in the supplemental material). However, no relevant patterns were detected for AMR.
Ciprofloxacin resistance, background mutations, and genotypes. The resistance gene analysis identified the qnrS1 gene in 55 isolates (Table 4) and detected a number of different mutations (n ϭ 24) in the gyrA and gyrB genes encoding DNA gyrase and in the topoisomerase IV enzyme parC and parE genes ( Table 6, columns 1 to 3). The most prevalent mutation was gyrA D538N (n ϭ 352), followed by gyrA S83F (n ϭ 299), gyrA S83Y (n ϭ 125), and parE A364V (n ϭ 69). On the basis of mutations in gyrA/B and parC/E genes, 34 cip-mutation profiles were generated and compared with the ciprofloxacin MIC of each isolate ( Table 6, columns 4 and 5) ( Fig. 4 and 5). All of the profiles, apart from gyrA D538N and parE A364V, were associated with resistance (MIC Ͼ 0.06 g/ml). Two different profiles with triple mutations (gyrA D87G gyrA S83F parC E84K for  Table 6, columns 4 and 5). Profiles with qnr genes also had MICs of Ն1.0 g/ml ( Fig. 4 and 5). No mutations were present in 18 isolates, but 3 of them showed resistance to ciprofloxacin (MIC of 0.25 to 0.5 g/ml; Fig. 4).
Mutations in codon 83 of gyrA (S83F and S83Y) were the most prevalent among our ciprofloxacin-resistant isolates (411/467; 88%) ( Table 6, columns 1 to 3). The S83Y mutation (122/411) was closely associated with genotype 4.3.1 in Bangladesh (123/125; 98%) and was present in 96% of our H58 lineage Bd isolates (104/108; Fig. 3). In contrast, 89% of our lineage Ia isolates had the S83F mutation (198/223) and exhibited lower mean (0.74 versus 1.71 g/ml) and median (0.25 versus 0.5 g/ml) ciprofloxacin MIC values than the lineage Bd isolates. The latter lineage also displayed more divergence than lineage I in Bangladesh (  a Five different antibiotics were considered: ampicillin (amp), co-trimoxazole (sxt), chloramphenicol (chl), ciprofloxacin (cip), and ceftriaxone (cro). MDR (multidrug resistance) refers to co-occurring resistance to amp, sxt, and chl. "S" and "R" refer to susceptible and resistant phenotypes, respectively (interpretations according to EUCAST-2018).   a Four antimicrobials were considered (ampicillin, co-trimoxazole, chloramphenicol, and ceftriaxone); resistance to these agents is caused mainly by acquisition of resistance genes. b Sensitivity data represent proportions of isolates identified as phenotypically resistant by the WGS-res profile. c Specificity data represent proportions of isolates identified as phenotypically susceptible by the WGS-res profile. d For co-trimoxazole (sxt), we considered the presence of dfrA7, plus sul1 and/or sul2 genes to exert the resistance (R) phenotype. e A total of 206 detected sul2 genes matched three different GenBank IDs: FJ197818 (n ϭ 74), GQ421466 (n ϭ 1), and HQ840942 (n ϭ 131). f Of the four sul2 genes, two matched FJ197818 and two HQ840942. One sul1 gene had unreliable bases (N) in its sequence; that result was considered a sequencing error, and the complete sequence was used in calculations. g One sul1 gene had unreliable bases (N) in its sequence; that result was considered a sequencing error, and the complete sequence was used in calculations. h A total of 54 genes matched GQ421466 and one HQ840942. i Only sul2 genes that matched HQ840942 had complete sequences. Genes that matched FJ197818 and GQ421466 were either truncated or mutated. j All catA1 gene sequence had one silent mutation in amino acid 195 (lysine) (CTG¡TTG). In total, 11 ciprofloxacin-resistant isolates (11/467; 2%) did not have any other mutation in DNA gyrase and topoisomerase IV genes, leading to an estimated sensi- 17 tivity of 98% for the WGS method in correctly predicting ciprofloxacin resistance. In contrast, 36 of 69 ciprofloxacin-susceptible isolates had at least one mutation (not linked to a specific genotype) in one of these four genes (specificity ϭ 52%). Comparison with neighboring countries. All genotype 4.3.1 isolates from Bangladesh (this study), Nepal (surveillance in Kathmandu), and Pakistan (outbreak in Sindh) had the same gyrA D538N mutation (Fig. S5). Likewise, the parE A364V mutation was present in all genotype 3.3 isolates from Bangladesh (70/70) and Nepal (17/19) and in genotype 3.3.1 (3/3) isolates from Nepal. Genotype 2.0 isolates from Bangladesh also had the gyrA N529S mutation present (94%; 17/18). However, none of these mutations seemed to have any association with AMR (Fig. 4).
Comparisons performed with the bla CTX-M-15 gene sequence of our ceftriaxoneresistant isolate revealed 92% coverage and 99% identity with the XDR isolate (GenBank accession no. LT906492.1) from the Pakistani outbreak (Table 7). In contrast, the bla CTX-M-15 gene from Bangladesh shared complete homology with the Klebsiella pneumoniae bla CTX-M-15 gene (FJ815436.1). A detailed comparison of our sequence data with the sequences of the isolates from Pakistan and Nepal is presented in Table 7.

DISCUSSION
S. Typhi multilocus sequence types and other genotypes in Bangladesh. Genotyping and the phylogenetic inferences agreed with the genotyping framework interpretation (27) and showed genotype 4.3.1 (haplotype 58, H58) to be dominant among the isolates from Bangladesh (Table 1). This was no surprise, as this genotype possibly emerged from South Asia in the early 1990s and now dominates in regions of typhoid  (26,31). The same genotype was also dominant among the isolates from Nepal and Pakistan (Table 7) (12,29). On the other hand, classical MLST revealed only three sequence types (ST), with dominance of ST1 and ST2, which accords with global MLST report (32). There are 46 complete MLST types available for S. Typhi (33). Interestingly, the third most common MLST type in our data, ST2209, had a complete match with genotype 2.3.3 (100%; 18/18) (Fig. 1a). Isolates of this genotype from 2013 (n ϭ 8) had the same mutation (gyrB S464Y). Five of 8 had a cip-resistant phenotype, which could indicate the beginning of new clonal dissemination (see Fig. S2 in the supplemental material).
Moreover, 99% (349/351) of all ST1 isolates from Bangladesh belonged to genotype 4.3.1 (Fig. 1a). This association was previously described in a small study involving 32 isolates (34). A phylogeographical report of S. Typhi included an estimate that divergence for genotype 4.3.1 commenced in the very late 1980s (26). However, the presence of ST1 could be detected before the 1980s (33,35), as is likely the case for H58.
Presence of genotype-specific mutations. Isolates with genotype 4.3.1 from all three countries shared a common but as-yet-unreported mutation, gyrA D538N (nucleotide position 2332398 of the CT18 genome; Fig. S5). This mutation is not linked to ciprofloxacin resistance ( Fig. 4 and 5) but could be crucial to the structure of the DNA gyrase enzyme, considering the associated change in the isoelectronic point (pI; D¡N: 2.77 ¡ 5.41) of the amino acid due to this mutation (36). Similar associations were also observed between genotype 2.0 and gyrA N529S (hydrophobicity, 3.47 ¡ 1.83), and  Fig. S5) (37). These mutations could have potential as markers to trace genotypes, especially the more prevalent genotypes such as 4.3.1 and 3.3. New H58 lineages with high-level ciprofloxacin resistance. According to the published scheme that defines the different lineages of genotype 4.3.1 (H58) (26,38), lineage Ia was dominant among the isolates from the recent Pakistan XDR outbreak, while most isolates from the Nepal surveillance belong to lineage II (Table 7). An undefined cluster within lineage II was also noticed among the Nepali isolates (Fig. 3), as has been described previously (12,29). Among our isolates from Bangladesh, we found a new lineage of genotype 4.3.1 (H58), Bd (n ϭ 108), which represented the second most dominant lineage after Ia (n ϭ 223). This new lineage has decreased susceptibility to ciprofloxacin compared to lineage Ia (mean MIC, 1.71 versus 0.74 g/ ml). Ciprofloxacin MICs of Ͼ0.06 g/ml are classified as resistant following the EUCAST guidelines. However, as the resistance breakpoint specified by the Clinical and Laboratory Standards Institute (CLSI) is 1 g/ml, some strains could be classified as susceptible in countries that use the CLSI guidelines (12,17,29). Moreover, lineage Bd is probably of local origin, as it was absent in both neighboring countries (Fig. 3) and does not match the published SNP definition of lineage I or II (38). This local variant also had a higher pairwise distance in the SNP matrix (mean, 12.8 versus 11.2) than lineage Ia, suggesting a different pattern of divergence.
Remarkably, a sublineage of lineage Bd (Bdq; n ϭ 55) showed increased resistance compared to other isolates from the same lineage, with median ciprofloxacin MICs of 4.0 g/ml (mean MIC, 3.4 versus 0.4 g/ml; Fig. S4). Sublineage Bdq predominantly carried qnr genes, in addition to gyrA mutations ( Fig. 3 and 5), and showed more clonality than other lineages (mean pairwise distance, 8.4 versus 11.2 for Ia). Moreover, sublineage Bdq emerged recently, as all isolates were from 2006 onward, but became more prevalent after 2007 (Fig. S3). Therefore, antimicrobial treatment with fluoroquinolones of infections caused by sublineage Bdq may lead to failure.
A similar highly resistant lineage with triple mutations (gyrA S83F gyrA D87G parC E84G ) but with no qnr genes was previously reported to cause failure of treatment with gatifloxacin in Nepal (28,29). Our MLT also showed a small subclade (n ϭ 8) inside lineage Ia for Bangladesh, with a triple mutation (gyrA S83F gyrA D87G parC E84K ) and median ciprofloxacin MICs of 8.0 g/ml ( Fig. 4 and 5).
Notably, the number of lineage II isolates (n ϭ 4) in Bangladesh was extremely low (Fig. 3), despite the dominance of this lineage in Nepal and India (26,27,29). The surveillance data from Nepal, which mostly describes the isolates from Kathmandu valley, showed a shifting pattern of H58 lineages (from lineage I to lineage II) over the years (29). Such a changing pattern is not observed in Bangladesh, probably because of relatively high prevalence and dominance of local lineages, such as the previously unreported lineage Bd. The Nepal surveillance also reported association of MDR with lineage I and of cip resistance with lineage II (29). However, no such association has been found for lineage I or lineage Bd in Bangladesh (see Data Set S1 in the supplemental material).
wgSNP analysis suggests regional clonality of S. Typhi in Bangladesh. The wgSNP analyses of our isolates generated 2,328 SNPs, revealing that the S. Typhi population in Bangladesh is highly clonal. However, addition of the isolates from Nepal (n ϭ 198) and Pakistan (n ϭ 100) increased the number of SNPs to 3,251 but decreased the number to 627 for genotype 4.3.1 isolates only (n ϭ 603). As the filtering criteria remain the same, the number of SNPs for all Bangladesh isolates is relatively low compared to the global or multicountry context (25)(26)(27) but is similar to countryspecific data. For example, 1,850 SNPs were detected in isolates from Thailand (n ϭ 44) and 2,187 SNPs in isolates from Nepal (n ϭ 198) (29,39). The wgSNP-MLT data showed distinct differentiation of all genotypes, much like the data from the cgMLST-UPGMA tree, except the latter lacked clear inferences for different H58 lineages (Fig. 1b and c  and Fig. S6). Genotype 1.2.1 mapped close to the root of the MLT, suggesting that this genotype is one of the oldest circulating types. Likewise, being the most distantly related, genotype 4.3.1 could be one of the more recent genotypes circulating in Bangladesh (Fig. 1c) and neighboring countries (Fig. 2).
WGS predicts AMR phenotypes with high sensitivity. The WGS-based resistance profiles showed Ͼ99% sensitivity and Ͼ91% specificity in describing the phenotypes (for amp, sxt, chl, and cro) of AMR isolates (Table 5). Remarkably, the dfrA7 genes (involved in trimethoprim resistance) were always detected in the presence of the sul1 gene (sulfonamide resistance) and never alone. Table 5). Similarly, sul1 was never detected in the absence of dfrA7. Two isolates had discordant results, as we did not detect the concordant resistance genes in WGS analyses (Table 5). Repeating the antimicrobial susceptibility tests (ASTs) reconfirmed the resistant phenotype. Other resistance mechanisms, e.g., efflux pumps or membrane permeability changes, may be involved (40).
Ciprofloxacin resistance in 11 isolates with no mutation in DNA gyrase or topoisomerase IV genes (and no qnr genes) can suggest the presence of other mechanisms. Indeed, MDR bacteria can increase the expression of efflux pump genes, including acrAB, acrEF and tolC (through overexpression of ramA or repression of acrR genes). This enables the bacteria to expel fluoroquinolone molecules, resulting in ciprofloxacin resistance (40)(41)(42)(43), as well as ampicillin or chloramphenicol resistance, even in the absence of bla or catA genes (44)(45)(46). On the other hand, isolates carrying a bla gene without the resistance phenotype could be the result of mutations in the promoter regions of outer membrane protein genes, such as the ompC gene, which facilitates penetration of beta-lactams through the outer membrane (47,48). This could be the scenario for several susceptible isolates (n ϭ 30) in our library that have the full-length resistance gene. However, transcriptomic or proteomic approaches may be required to further explore these possibilities.
Different genotypic backgrounds of ceftriaxone resistance in Bangladesh and Pakistan. The ceftriaxone-resistant (cro-R) strain from our library was isolated in 2000. The first report of a cro-R strain was published in 1999 (16). Interestingly, this isolate harbored the same extended-spectrum-beta-lactamase (ESBL) gene, bla CTX-M-15 (17,49,50), that caused the ceftriaxone-resistant phenotype in an ongoing typhoid outbreak in Pakistan (12). Other ESBL genes, including bla CMY-2 and bla CTX-M-14 , have also been reported in relation with ceftriaxone resistance in other Salmonella species (51, 52) but never in S. Typhi. The sequence identity of bla CTX-M-15 between our isolate and the Pakistani isolates was 99%, with 92% coverage ( Table 7). The resistance phenotype and genotype were also different from those of our isolate ( Table 7). The Pakistani outbreak isolates formed a distinct cluster in the H58-specific MLT and showed high-level clonality ( Fig. 3 and Fig. S1). In contrast, our ceftriaxone-resistant isolate had genotype 3.3, which suggests a different source and geographical origin. Moreover, no other ceftriaxone-resistant strains of genotype 3.3 have been reported from Bangladesh. We hypothesize that acquisition of the bla CTX-M15 gene might compromise the fitness of S. Typhi although as of now no data have been published in support of this. Also, no association with fitness has been found for ciprofloxacin resistance mutations in DNA gyrase genes (53). Therefore, the possibility of a global dissemination of these recently emerging variants cannot be excluded given the successful multicontinent spreading of its H58 ancestor (genotype 4.3.1).
This study had some limitations. The isolates from Pakistan are from a still-ongoing outbreak in Hyderabad and Karachi that started in 2016. The Nepal isolates are from a prospective surveillance in the area of Kathmandu valley and cover a period of 9 years (2008 to 2016). The collection of strains from Bangladesh was selected from a biobank of Ͼ3,000 strains recovered over a period of 15 years (1999 to 2013) from two different hospital settings in Dhaka. The majority (97%) of the isolates from Bangladesh are from children (Ͻ18 years old). Therefore, none of the collections cover the whole population in their respective countries. Also, there is no overlap of the isolate collection periods between Bangladesh and Pakistan. Country-to-country comparisons of the observed data may therefore be biased.
Conclusion. Our study demonstrated that WGS has high sensitivity and specificity for prediction of S. Typhi resistance phenotypes. However, this genomic method still lacks sensitivity and needs fine-tuning for the detection of ciprofloxacin resistance. We detected three different mutations associated with specific genotypes that could be used to develop genotype-specific tracking tools. We report a new, local variant of genotype 4.3.1, lineage Bd, which contains a recently emerged sublineage, Bdq, that exhibits a high level of ciprofloxacin resistance. A triple mutant variant (gyrA S83F gyrA D87G parC E84K ) of lineage Ia with high ciprofloxacin resistance was also detected. A similar triple mutant variant of lineage II (gyrA S83F gyrA D87G parC E84G ) has been reported from Nepal and possesses the same phenotype (28,29). Our ceftriaxone-resistant isolate contains the bla CTX-M-15 gene but has a genotype and gene sequence different from those of the same gene of XDR S. Typhi strains from the Pakistan outbreak, defining a different ancestral origin. Thus, dissemination of this isolate throughout the region from a single point is therefore less likely. However, multiple independent genetic events in neighboring countries and possible subsequent dissemination enhance the risk of the global spread of these highly resistant clones.
The data presented in this study will add to the accumulating information, from Pakistan and Nepal in particular, concerning the increasing drug resistance of S. Typhi. The emergence of XDR S. Typhi is strongly compromising effective treatment of typhoid fever. The spread of these resistant lineages and their occurrence in various Asian countries emphasize the need to inform public health professionals and sensitize the global community. Measures to implement a two-pronged approach for typhoid control need to be accelerated (54,55). Both short-term vaccine interventions for high-risk populations and long-term water and sanitation interventions will undoubtedly be the cornerstones of a global prevention plan to address control of typhoid fever.

MATERIALS AND METHODS
Isolate collection and antimicrobial susceptibility profiles. All S. Typhi isolates used in this study were collected from the Child Health Research Foundation (CHRF) at the Department of Microbiology, Dhaka Shishu (Children) Hospital, in Dhaka, Bangladesh. The CHRF team has been preserving invasive Salmonella isolates since 1999 and maintained a biobank of Ͼ3,500 S. Typhi isolates, largely from children (Ͻ18 years of age). All strains were isolated from the blood of patients diagnosed with typhoid fever in two different settings: hospital inpatients (hospitalized), and out-patients attending the consultation facility (56). Clinical and epidemiological data were collected for all isolates collected from hospital inpatients.
We selected 539 S. Typhi isolates for this study; data were available for those isolates with respect to the date of isolation (1999 to 2013), hospital setting, and phenotypic resistance for five different antibiotics (ampicillin, chloramphenicol, co-trimoxazole, ciprofloxacin, and ceftriaxone). Age data were available for 85% (456/536) cases; among those cases, 97% (443/456) patients were Ͻ18 years of age, while 76% (345/456) were Ͻ5 years of age. We checked the identity of the isolates by the use of standard biochemical tests and Salmonella agglutinating antisera (Thermo Scientific, MA, USA). Antimicrobial susceptibility for ampicillin (amp), co-trimoxazole (sxt), and chloramphenicol (chl) was determined using the disk diffusion method (Oxoid, Thermo Scientific, MA, USA). Broth microdilution was used to determine the MIC values for ciprofloxacin (cip) and ceftriaxone (cro; Sigma-Aldrich, MO, USA). All zone diameter and MIC data were interpreted according to EUCAST v8.0 clinical breakpoints (57). Fig. S7 in the supplemental material shows the complete workflow. All sequence data have been submitted to the European Nucleotide Archive (ENA). Data Set S1 in the supplemental material summarizes relevant details of our isolates.
DNA extraction and whole-genome sequencing. Isolates were grown on MacConkey agar (Oxoid) overnight, and the colonies were suspended in water. The QIAamp DNA minikit (Qiagen, Hilden, Germany) was used to extract DNA from the suspension on the same day. WGS was performed using an Illumina HiSeq 4000 platform (The Oxford Genomics Centre at the Wellcome Trust Centre for Human Genetics, Oxford, United Kingdom). One Salmonella Paratyphi isolate was also sequenced so that it could be included in comparative phylogenetic analysis (as an outgroup).
Data quality check. Sequence data quality was checked using FastQC v0.11.15 (58). We summarized all quality indicators using MultiQC v3 (59). If the summary revealed the presence of adapter sequences, they were removed using Trimmomatic v0.36 (60). KmerFinder was used to confirm the species of the strains (61, 62). Another tool, SeqSero, was used for WGS-based serotyping, to determine the Salmonella serovar of the isolates and confirm the wet-lab serotyping results (63).

WGS data analyses with BioNumerics.
Adaptor-free fastq files were imported into BioNumerics version 7.6.2 (Applied Maths NV, Sint-Martens-Latem, Belgium) and analyzed via the use of the integrated Calculation Engine. For the comparison with isolates from neighboring countries, we used recently published WGS data on 100 S. Typhi isolates from Pakistan (12) and 198 S. Typhi isolates from Nepal (29). The Pakistan isolates were mostly from an ongoing outbreak of XDR S. Typhi, in Hyderabad and Karachi, Sindh, Pakistan, between November 2016 and March 2017 (12). In contrast, the Nepal isolates were part of a hospital-based enteric fever surveillance performed during 2008 to 2016, based on one of the large referral hospitals in Kathmandu Valley, namely, Patan Academy of Health Sciences (PAHS). (29).
Details of the quality control of the WGS data, mapping against the reference genome, filtering the SNPs, allele calling for cgMLST, detecting the presence of acquired resistance genes, and SNP-based genotyping are described in Text S1.
Phylogenetic analyses. We used RaxML v8.2.10 to build maximum likelihood phylogenetic trees (MLT) (65) on the basis of the alignment of 2,328 SNPs from 536 S. Typhi isolates in our study, 3,251 SNPs from 834 isolates in the comparisons with neighboring countries, and 627 SNPs from all 603 H58 isolates. Lineages for all H58 isolates were determined as previously described (38). We employed the generalized time-reversible model and a Gamma distribution to model site-specific rate variation (the GTRGAMMA in RaxML). Support for the MLT phylogeny was assessed via 100 bootstrap pseudoanalyses. The S. Paratyphi A strain from Bangladesh (Sample: 311189_229186) was included as an outgroup for tree rooting. All MLT and UPGMA trees were displayed and annotated using the iTOL6 online version (66). To compute the genetic distances between different groups (e.g., countries, H58 lineages, etc.), a pairwise SNP distance matrix was generated between isolates by computing the number of SNP loci at which pairs of isolates had discordant alleles. Median distances within or between groups were computed from this distance matrix.
Statistical analyses were performed using R v3.5 (64); the same application was used to generate the line graphs and box plots.
Data availability. All sequence data determined in work have been submitted to the European Nucleotide Archive (ENA) (study identifier [ID]: ERP109468).

ACKNOWLEDGMENTS
We thank the High-Throughput Genomics Group at the Wellcome Trust Centre for Human Genetics (funded by Wellcome Trust grant reference 090532/Z/09/Z) for the generation of the sequencing data. We acknowledge the guidance from Maksuda Islam and technical assistance from Hafizur Rahman during the antimicrobial susceptibility tests.
This study received funding from the EU Horizon 2020 research and innovation program under grant agreement no. 643476. The funders had no role in data collection and analysis, decision to publish, or preparation of the manuscript.
A.M.T. received a "Allocations de Recherche pour une Thèse au Sud (ARTS)" PhD scholarship from Institut de Recherche pour le Développement (IRD) and from Fondation Mérieux in France.
We declare no conflicts of interest.