Each contributing study or surveillance programme obtained local ethical and governance approvals, as reported in the primary publication for each dataset. For this study, inclusion of data that were not yet in the public domain by August 2021 was approved by the Observational/Interventions Research Ethics Committee of the London School of Hygiene and Tropical Medicine (ref #26408), on the basis of details provided on the local ethical approvals for sample and data collection (Supplementary file 1).

Attempts were made to include all Typhi sequence data generated in the 20 years since the first genome was sequenced, through August 2021. Genome data and the corresponding data owners were identified from literature searches and sequence database searches (European Nucleotide Archive [ENA]; NCBI Short Read Archive [SRA], and GenBank; Enterobase). Unpublished data, including those from ongoing surveillance studies and routine public health laboratory sequencing, were identified through professional networks, published study protocols (Carey et al., 2020), and an open call for participation in the GTGC. All data generators thus identified were invited to join the GTGC and to provide or verify corresponding source information, with year and location isolated being required fields (‘metadata’, see below). Nearly all those contacted responded and are included as consortium authors on this study. The exceptions, where authors did not respond to email inquiries, were: (i) one genome reported from Malaysia (Ahmad et al., 2017) and n=133 draft genomes reported from India (Katiyar et al., 2020), which were excluded as sequence reads were not available in NCBI; and (ii) n=39 genomes reported in studies of travel-associated or local outbreaks (Burnsed et al., 2018; Hao et al., 2020; Shin et al., 2021), which were included as raw sequence data and sufficient metadata were publicly available. A further n=850 genomes sequenced by US Centers for Disease Control and Prevention and available in NCBI were excluded from analysis because travel history was unknown and most US cases are travel-associated. Table 1 summarises all studies and unpublished public health laboratory datasets from which sequence data were sourced.

Whole genome sequence data, in the form of Illumina fastq files, were sourced from the ENA or SRA or were provided directly by the data contributors in the case of data that was unpublished in August 2021. Run, BioSample, and BioProject accessions are provided in Supplementary file 2, together with contributed metadata and PubMed or preprint identifiers.

Primary sequence analysis was conducted on the Wellcome Sanger Institute compute cluster. Genotypes, as defined under the GenoTyphi scheme (Dyson and Holt, 2021; Wong et al., 2016a), were called directly from Illumina reads using Mykrobe v0.12.1 with Typhi typing panel v20221207, and collated using the Python code available at https://github.com/typhoidgenomics/genotyphi (v2.0, doi: 10.5281/zenodo.7430538; Ingle et al., 2022).

Illumina reads were assembled using the Centre for Genomic Pathogen Surveillance (CGPS) assembly pipeline v2.1.0 (https://gitlab.com/cgps/ghru/pipelines/dsl2/pipelines/assembly/) (Underwood, 2020), which utilises the SPAdes assembler (v3.12.0) (Bankevich et al., 2012; Demin et al., 2020). One readset failed assembly and was excluded. Assemblies were uploaded to Pathogenwatch to confirm species and serovar, and to identify AMR determinants and plasmid replicons (Argimón et al., 2021b). Eight assemblies were excluded as they were identified as non-Typhi: either other serovars of S. enterica (2 Paratyphi B, 2 Enteritidis, 1 Montevideo, 1 Newport, 1 Durban) or other species (1 Klebsiella pneumoniae). Assemblies >5.5 Mbp or <4.5 Mbp in size were also excluded from further analysis (n=35 excluded, see size distributions in Figure 1—figure supplement 1). The resulting 13,000 whole genome assemblies are available in Figshare, doi: 10.26180/21431883 (https://doi.org/10.26180/21431883).

Phylogenetic trees were generated using Pathogenwatch, which estimates pairwise genetic distances between genomes (based on counting SNVs across 3284 core genes) and infers a neighbour-joining tree from the resulting distance matrix (Argimón et al., 2021b). The Pathogenwatch collections used to generate the tree files are available at https://bit.ly/Typhi4311P1 (tree showing position of Rwp1-PK1, in context with other genomes from Pakistan) and https://bit.ly/Typhi232 (tree for genotype 2.3.2 genomes).

Owners of the contributing studies were asked to provide or update source information relating to their genome data, using a standardised template (http://bit.ly/typhiMeta). Repeat isolates were defined as those that represent the same occurrence of typhoid infection (acute disease or asymptomatic carriage) as one that is already included in the dataset. In such instances, data owners were asked to indicate the ‘primary’ isolate (either the first, or the best quality, genome for each unique case) to use in the analysis. Repeat isolates were then excluded from the dataset entirely (excluded from Supplementary file 2).

Data provided on the source of isolates (specimen type and patient health status) are shown in Supplementary file 3. This information was used to identify isolates that were associated with acute typhoid fever. In total, n=6462 genomes were recorded as isolated from symptomatic individuals. A further n=119 were recorded as isolated from asymptomatic carriers. The remaining genomes had no health status recorded (i.e. symptomatic vs asymptomatic carrier); of these, the majority were isolated from blood (n=3365) or the specimen type was not recorded (n=2522). Since most studies and surveillance programmes are set up to capture acute infections rather than asymptomatic carriers, we defined ‘assumed acute illness’ genomes as those not recorded explicitly as asymptomatic carriers (n=119) or coming from gallbladder (n=1) or environmental (n=14) samples; this resulted in a total of 12,831 genomes that were assumed to represent acute illness.

We defined ‘country of origin’ as the country of isolation; or for travel-associated infections, the country recorded as the presumed country of infection based on travel history (Centers for Disease Control and Prevention, 2011; Ingle et al., 2021; Ingle et al., 2019; Matono et al., 2017). Countries were assigned to geographical regions using the United Nations Statistics Division standard M49 (see https://unstats.un.org/unsd/methodology/m49/overview/); we used the intermediate region label where assigned, and subregion otherwise. To identify isolate collections that were suitably representative of local pathogen populations, for the purpose of calculating genotype and AMR prevalences for a given setting, data owners were asked to indicate the purpose of sampling for each study or dataset. Options available were either ‘Non Targeted’ (surveillance study, routine diagnostics, reference lab, other; n=11,086), ‘Targeted’ (cluster investigation, AMR focused, other; n=1862), or ‘Not Provided’ (n=17). Only samples from ‘Non Targeted’ sampling frames with known year of isolation and country of origin were included in national prevalence estimates.

AMR determinants identified in the genome assemblies using Pathogenwatch were used to define AMR genotype as follows. MDR: resistance determinants for chloramphenicol (catA1 or cmlA), ampicillin (bla_TEM-1D, bla_OXA-7), and co-trimoxazole (at least one dfrA gene and at least one sul gene). Ciprofloxacin non-susceptible (CipNS): one or more of the QRDR mutations at GyrA-83, GyrA-87, ParC-80, ParC-84, GyrB-464 or presence of a plasmid-mediated quinolone resistance (PMQR) gene (qnrB, qnrD, qnrS); note, this typically corresponds to MIC ≥0.06 mg/L (Day et al., 2018). CipR: QRDR triple mutant (GyrA-83 and GyrA-87, together with either ParC-80 or ParC-84), or PMQR gene together with GyrA-83, GyrA-87, and/or GyrB-464. This typically corresponds to MIC ≥1 mg/L, and CipR is a subset of CipNS. Ceftriaxone resistant (CefR): presence of an ESBL (blaCTX-M-12, blaCTX-M-15, blaCTX-M-23, blaCTX-M-55, blaSHV-12). XDR: MDR plus CipR plus CefR. AziR: mutation at AcrB-717. The above lists all those AMR determinants that were found here in ≥1 genome and used to define AMR profiles and prevalences; additional AMR genes sought by Typhi Pathogenwatch but not detected are listed in Supplementary Table 2 of (Argimón et al., 2021b).

All statistical analyses were conducted in R v4.1.2 (R Development Core Team, 2021), code is available in R markdown format at https://github.com/typhoidgenomics/TyphoidGenomicsConsortiumWG1 (v1.0, doi:10.5281/zenodo.7487862; Holt, 2022). Genotype and AMR frequencies were calculated at the level of country and UN world region (based on ‘country of origin’) as defined above. Inclusion criteria for these estimates were: known ‘country of origin’, known year of isolation, non-targeted sampling, assumed acute illness (see definitions of these variables above). A total of 10,726 genomes met these criteria; the subset of 9478 isolated from 2010 onwards were the focus of the majority of analyses and visualisations, including all prevalence estimates. The prevalence estimates reported in text and figures are simple proportions; 95% confidence intervals (CIs) for proportions are given in text and supplementary tables where relevant. Annual prevalence rates were estimated for countries that had N≥50 representative genomes and ≥3 years with ≥10 representative genomes. Association between MDR prevalence and prevalence of IncHI1 plasmids amongst MDR genomes was assessed for countries with ≥5% MDR prevalence between 2000 and 2020. The significance of increases or decreases in prevalence was assessed using a Chi-squared test for trend in proportions (using the proportion.trend.test function in R). There are no established thresholds for the prevalence of resistance that should trigger changes in empirical therapy recommendations for enteric fever; hence, we defined our own categories of resistance prevalence for visualisation purposes, to reflect escalating levels of concern for empirical antimicrobial use: (i) 0, no resistance detected; (ii) >0 and≤2%, resistance present but rare; (iii) 2–10%, emerging resistance; (iv) 10–50%, resistance common; (v) >50%, established resistance. Robustness of prevalence estimates was assessed informally, by comparing overlap of 95% CIs computed for different laboratories from the same country (for genomes isolated 2010–2020, and laboratories with N≥20 genomes [Southern Asia] or N≥10 [Nigeria] meeting the inclusion criteria during this period).

All analyses and plots were generated using R v4.1.2, code is available in R markdown format at https://github.com/typhoidgenomics/TyphoidGenomicsConsortiumWG1 (v1.0, doi:10.5281/zenodo.7487862; Holt, 2022). Data processing was done using the R packages tidyverse v1.3.1, dplyr v1.0.7, reshape2 v1.4.4, and janitor v2.1.0; figures were generated using packages ggplot2 v3.3.5, ggExtra v0.9, patchwork v1.1.1, RColorBrewer v1.1-2, and pals v1.7; maps were generated using packages sf v1.0-5, rvest v1.0.2, maps v3.4.0, scatterpie v0.1.7, ggnewscale v0.4.5; trees were plotted using ggtreeio v1.18.1 and ggtree v3.2.1.

A total of 13,000 confirmed Typhi genomes were collated from 65 studies and 5 unpublished public health laboratory datasets (see Table 1, Supplementary file 2). N=35 genomes had assembly sizes outside of the plausible range (4.5–5.5 Mbp, see Figure 1—figure supplement 1), leaving n=12,965 high-quality genomes originating from 110 countries. The distribution of samples by world region (as defined by WHO statistics division M49) is shown in Table 2, with country breakdown in Supplementary file 4. The majority originated from Southern Asia (n=8231), specifically India (n=2705), Bangladesh (n=2268), Pakistan (n=1810), and Nepal (n=1436). A total of n=1140 originated from South-eastern Asia, with >100 each from Cambodia (n=279), Vietnam (n=224), the Philippines (n=209), Indonesia (n=145), and Laos (n=139). Overall, 1106 genomes originated from Eastern Africa, including >100 each from Malawi (n=569), Kenya (n=254), Zimbabwe (n=110). Other regions of Africa were less well represented, with n=384 from Western Africa, n=317 from Southern Africa, n=59 from Middle Africa (so-named in the M49 region definitions, although more commonly referred to as Central Africa), and n=41 from Northern Africa (see Table 2 and Supplementary file 4 for details).

Overall, there were 36 countries with ≥20 genomes (total n=12,409 genomes, 95.7%) and 21 countries with ≥100 genomes (n=11,761 genomes, 90.7%) (see Supplementary file 4). Countries with the most genomes available (n≥100 each) were mainly those where local surveillance studies have utilised WGS for isolate characterisation: India (Britto et al., 2020; da Silva et al., 2022), Bangladesh (Rahman et al., 2020; da Silva et al., 2022), Nepal (Britto et al., 2018; da Silva et al., 2022; Pham Thanh et al., 2016a), Pakistan (da Silva et al., 2022), Cambodia (Kuijpers et al., 2017; Pham Thanh et al., 2016b), Laos (Wong et al., 2015), Vietnam (Holt et al., 2011a), Kenya (Kariuki et al., 2021; Kariuki et al., 2010), Malawi (Feasey et al., 2015), Zimbabwe (Mashe et al., 2021; Thilliez et al., 2022), Ghana (Park et al., 2018), Nigeria (Ikhimiukor et al., 2022a; Wong et al., 2016b), Chile (Maes et al., 2022), Samoa Sikorski et al., 2022; plus South Africa (Smith et al., 2023), the Philippines (Lagrada et al., 2022), United Kingdom, and United States, where Typhi isolates are sequenced as part of national surveillance programmes.

The genome collection included n=3381 isolates that were recorded as travel-associated (see Table 2 and Supplementary file 4), contributed mainly by public health reference laboratories in England (n=1740), USA (n=749), Australia (n=490), New Zealand (n=144), France (n=116), and Japan (n=104). The most common countries of origin for travel-associated isolates were India (n=1241), Pakistan (n=783), Bangladesh (n=264), Fiji (n=102), Samoa (n=87), Mexico (n=60), Chile (n=49), Papua New Guinea (n=45), Nigeria (n=42), and Nepal (n=39). For some typhoid-endemic countries, the majority of genome data originated from travel-associated infections captured in other countries; those in this category with total n≥10 genomes are Guatemala (n=22/22), El Salvador (n=19/19), Mexico (n=60/61), Peru (n=14/14), Haiti (n=12/12), Morocco (n=12/13), Iraq (n=19/19), Malaysia (n=35/35), Fiji (n=102/144), and Papua New Guinea (n=45/86) (full data in Supplementary file 4).

In total, n=10,726 genomes were assumed to represent acute typhoid fever and recorded as derived from ‘non-targeted’ sampling frames, meaning local population-based surveillance studies or reference laboratory-based national surveillance programmes that could be considered representative of a given time (year of isolation) and geography (country and region of origin) (see Methods for definitions). The majority of these isolates (n=9478, 88.4%) originate from 2010 onwards; hence, we focus our reporting of genotype and AMR prevalences on this period. Most come from local typhoid surveillance studies (n=5574) or routine diagnostics/reference laboratory referrals capturing locally acquired (n=1543) or travel-associated (n=2284) cases. All prevalence estimates reported in this study derive from this data subset, unless otherwise stated.

The breakdown of genotype prevalence by world region, for genomes isolated from non-targeted sampling frames from 2010 onwards, is shown in Figure 1a (denominators in Table 2, full data in Supplementary file 5). Annual breakdown of regional genotype prevalence rates is given in Figure 1—figure supplement 2 (raw data, proportions, and 95% CIs in Supplementary file 5). Notably, while our data confirm that H58 genotypes (4.3.1 and derived) dominate in Asia, Eastern Africa, and Southern Africa, they were virtually absent from other parts of Africa, from South and Central America, as well as from Polynesia and Melanesia (Figure 1). Instead, each of these regions was dominated by their own local genotypes. Typhoid fever is no longer endemic in Northern America, Europe, or Australia/New Zealand. The genotype distributions shown for these regions were estimated from Typhi that were isolated locally but not recorded as being travel-associated; nevertheless, these genomes can be assumed to result from limited local transmission of travel-associated infections, and thus to reflect the diversity of travel destinations for individuals living in those regions. Annual national genotype prevalences for well-sampled countries with endemic typhoid are shown in Figure 1b (full data in Supplementary file 6 and Figure 1—figure supplement 3). Below, we summarise notable features of the global genotype distribution, by world region (as defined by WHO statistics division, see Methods).

We estimated the regional and national prevalence of clinically relevant AMR profiles in Typhi for the period 2010–2020, inferred from WGS data from non-targeted sampling frames for which country of origin could be determined (as per genotype prevalences, see Methods). In order to understand the potential implications of these AMR prevalences for local empirical therapy, we categorised them according to a traffic light-style system (see Methods), whereby amber colours signal emerging resistance of potential concern (<10%), and red colours signal prevalence rates of AMR that may warrant reconsideration of empirical antimicrobial use (>10%; see Figure 2 and Figure 2—figure supplement 1). The regional view (Figure 2—figure supplement 1, Supplementary file 7) highlights that CipNS is widespread, whereas CipR, AziR, and XDR have been mostly restricted to Southern Asia. MDR was most prevalent in African regions, and to a lesser degree in Asia. Full country-level data is mapped in Figure 2—figure supplement 2 and detailed in Supplementary file 8. National estimates for countries with sufficient data where typhoid is endemic (≥50 representative genomes available for the period 2010–2020, see Figure 2) indicate that MDR remains common across all well-sampled African countries (39% in Nigeria, 61% in South Africa, 66% in Ghana, 78% in Kenya, 93% in Malawi), but is much more variable in Asia (3% in India [n=67/2267] and Nepal [n=36/1275], 25% in Bangladesh [n=393/1591], 68% in Pakistan [n=1004/1484], 76% in Cambodia [n=167/221]) and essentially absent from Indonesia (n=0), the Philippines (n=0), Samoa (n=0), Mexico (n=1, 1.7%), and Chile (n=0). The underlying genotypes are shown in Figure 2—figure supplement 3, and highlight that MDR in Asia, Eastern Africa, and Southern Africa has been mostly associated with H58 (i.e. 4.3.1 and derived genotypes) but in Western Africa is associated with the dominant genotype in that region, 3.1.1. In contrast, CipNS was associated with more diverse Typhi genotypes in each country, including essentially all common genotypes in Southern Asian countries (Figure 2—figure supplement 3). National annual prevalence data suggest that AMR profiles were mostly quite stable over the last decade (with the notable exception of the emergence and rapid spread of XDR Typhi in Pakistan) but reveal some interesting differences between settings in terms of AMR trends and the underlying genotypes (see Figure 3, Figure 2—figure supplement 3, Figure 2—figure supplement 4, Figure 3—figure supplement 1).

The estimates of genotype and AMR prevalence represented here reflect post hoc analyses of data that were generated for a variety of different primary purposes in different settings, by different groups using varied criteria for sample collection, including in-country surveillance and travel-associated cases recorded in other countries. Whilst datasets known to be biased towards sequencing of AMR strains including outbreak investigations were excluded from prevalence estimates, there is still substantial heterogeneity across data sources. To explore the robustness of these national-level estimates, we compared prevalence estimates for the same country from different studies/sources, where sufficient data existed to do so.

Southern Asian countries were each represented by multiple in-country data sources plus travel-associated data collected in three or four other countries. Figure 6—figure supplement 1 shows genotype prevalence estimates derived from these different sources (for laboratories contributing ≥20 isolates each) and Figure 6a shows the annual genotype frequency distributions (for years with ≥20 isolates). In most cases (67% of genotype-source combinations), genotype prevalence rates estimated from individual source laboratories yielded 95% CIs that overlapped with those of the pooled national estimates (see Figure 6—figure supplement 1). The main exception was for genotype 4.3.1.2 in India; for most source laboratories (many contributing via the Surveillance for Enteric Fever in India [SEFI] network; Carey et al., 2020; da Silva et al., 2022), this was the most prevalent genotype, but the point estimates ranged from 16% to 82%, compared with the pooled estimate of 53.4% (95% CI, 51.4–55.5%), and 95% CIs were frequently non-overlapping (see Figure 6—figure supplement 1). High prevalence of 4.3.1.2 was estimated from contributing laboratories in urban Vellore (82% [95% CI, 78–87%]), Chennai (67% [56–77%]), Bengaluru (70% [62–78%]), and Mumbai (two laboratories, estimates 74% [65–83%] and 63% [46–79%]); with lower prevalence in northern India, New Delhi (three laboratories, estimates 48% [28–68%], 40% [31–49%], 39% [22–56%]) and Chandigarh (39% [33–45%]). Two Indian laboratories were clear outliers, with little or no 4.3.1.2 but very high prevalence of a different genotype: 4.3.1.1 in rural Bathalapalli (81% [67–95%]) and 2.5 in the northern city of Ludhiana (77% [66–88%]). The relative prevalence of 4.3.1.1.P1 (XDR lineage) in Pakistan versus its parent lineage 4.3.1.1 also varied between sources, which could be explained by differences in the sampling periods and locations relative to the emergence of 4.3.1.1.P1 (see Figure 6a); notably, the highest estimate of XDR prevalence (n=27/27, 100%) came from a hospital-based study (Rasheed et al., 2020), which may select for more severe cases that were unresponsive to antibiotics received in the community setting. AMR prevalence estimates were also highly concordant across data sources (see Figure 6—figure supplement 2), and showed strikingly similar temporal trends (Figure 6b).

The only other country represented by ≥10 sequenced isolates each from multiple laboratories was Nigeria; these were located in Abuja (Zankli Medical Center, n=105, 2010–2013) and Ibadan (University of Ibadan, n=14, 2017–2018), and reference laboratories in England (n=15, 2015–2019) and the USA (n=10, 2016–2019) (see Figure 7). Genotype prevalence estimates were concordant across different sources, with single-laboratory 95% CIs overlapping with one another and with the pooled point estimate, for all five common genotypes (see Figure 7). The exception was that genotype 3.1.1 accounted for all n=14/14 isolates sequenced from Ibadan, but ranged from 53% to 70% prevalence at other laboratories and yielded a pooled national prevalence estimate of 67% [95% CI, 60–75%] (see Figure 7a and c). AMR prevalence estimates for Nigeria were more variable across laboratories (see Figure 7b), but this could be explained by their non-overlapping sampling times: Abuja data from earlier years (2010–2013) showed high MDR (49%) and low CipNS (4%); whereas Ibadan data from later years (2017–2018) showed comparatively lower MDR (21%) and higher CipNS (79%), in agreement with contemporaneous travel data (12% MDR, 60% CipNS, from total n=25 isolated 2015–2019).

This study presents the most comprehensive genomic snapshot of Typhi to date, with 12,965 high-quality genomes originating from 110 countries in 21 world regions. The consortium model provides improved consistency and completeness of source data aggregated from 77 laboratories and 66 unique studies. Our dataset also includes 1290 novel genomes sequenced by public health laboratories that would not otherwise have been published, including travel data from countries not previously represented in published Typhi genomics studies (e.g. El Salvador, Guatemala, Haiti, Mexico, and Peru). However, it is a post hoc analysis of isolates that were cultured in different contexts (including routine diagnostics, as well as study settings where culture would not normally be undertaken) and sequenced for different reasons (including retrospective studies, outbreak investigations, and routine surveillance). The study therefore has important limitations, most notably the scarcity of genomic data from many countries and world regions where typhoid is believed to be endemic (GBD 2019 Antimicrobial Resistance Collaborators, 2022), including Northern and Middle Africa, Western Asia, as well as Central and South America (Figures 1—3, Figure 1—figure supplements 2–4, Figure 2—figure supplements 1–3, Figure 3—figure supplement 1). These genomic data gaps reflect an underlying lack of routine blood culture or sustained blood-culture surveillance, and limited resources and expertise in many settings (Ikhimiukor et al., 2022b; Iskandar et al., 2021). In addition, public health authorities may be disincentivized to generate, analyse, and publish genomic data; we hope that this analysis strengthens the case for data generation and sharing for public good. Substantial investments have been made in recent years to improve and expand microbiological surveillance capacity in some low- and middle-income countries, but major regional surveillance gaps remain. It is therefore important to maximise information recovery from available data sources, especially WGS, which provides data on the emergence and spread of AMR variants. While the inference of AMR phenotype from WGS is currently highly reproducible and accurate for Typhi (Argimón et al., 2021b; Chattaway et al., 2021), continued phenotypic antimicrobial susceptibility testing remains crucial to monitor for emerging mechanisms and to guide changes in empiric therapy.

For now, routine sequencing of travel-associated Typhi infections diagnosed in high-income countries helps to fill some molecular surveillance gaps for some regions, assuming that accurate travel history is available and the sequence and metadata (including country of origin) are shared (Ingle et al., 2019). For example, our study included >3000 genomes shared by public health reference laboratories in England, Australia, New Zealand, France, Japan, and the USA. These infections mostly originate in other countries, and can in principle provide informative, if informal, sentinel surveillance for pathogen populations in countries with strong travel and/or immigration links to those with routine sequencing (Ingle et al., 2019). Indeed, for some countries and regions, travel data represented most or all of the available genome data (see Table 2, Supplementary file 4). In this study, where multiple data sources were available for the same country, we found that national genotype and AMR prevalence estimates for the period 2010–2020 were largely concordant between local surveillance studies and travel-associated cases captured elsewhere (Figures 6–7, Figure 6—figure supplements 1–2), particularly when comparing contemporaneous annual prevalence estimates (Figures 6 and 7c). This clearly shows that travel-associated Typhi isolated in low burden countries can be informative for surveillance of some high burden countries, which should serve as incentive for public health reference laboratories to share their data to the fullest extent they are able to under local regulations and encourage culture-based diagnostics in those countries that rely primarily on clinical diagnosis of typhoid fever in local populations, and development of molecular diagnostic tests for local use as travel-associated infections may provide information on predominant genes encoding resistance.

Another key limitation stemming from the post hoc nature of this study is that it is hard to assess how representative the prevalence estimates are for a given region/country and timeframe. The GTGC has developed new source/metadata standards for Typhi (see Methods), that include information on the purpose of sampling, which were completed by the original owners of each dataset (data available in Supplementary file 2). Such ‘purpose-of-sampling’ fields are currently lacking from metadata templates used for submission of bacterial genomes to the public sequencing archives (e.g. NCBI, ENA), and our approach was modelled on that established for sharing of SARS-CoV-2 sequence data, designed by the PHA4GE consortium (Griffiths et al., 2022). In this study, the purpose-of-sampling information was used to identify the subset of genome data that could be reasonably considered to be representative of national annual trends in genotype and AMR prevalence for public health surveillance purposes (n=9478 genomes post 2010; Figures 1—3). These originate mainly from local typhoid surveillance studies (59%), or routine diagnostics/surveillance capturing locally acquired (19%) and travel-associated (24%) infections. The comparisons of estimates for a given country based on different sources of genomes (Figures 6–7, Figure 6—figure supplements 1–2) are reassuring that the general scale and trends of AMR prevalence are reliable. The genome-based estimates are also in broad agreement with available phenotypic prevalence data on AMR in Typhi (Browne et al., 2020; Kariuki et al., 2015), although systematic aggregation of susceptibility data is limited. Both phenotypic and genomic analyses necessarily reflect blood-culture-confirmed cases, which may be biased towards more resistant infections resulting in overestimation of AMR prevalence. Notably, the genome data adds an additional layer of information on resistance mechanisms and the emergence and spread of lineages or variants. Importantly, our study clearly shows that, whilst much attention has been given to the emergence and spread of drug-resistant H58 Typhi, other clones predominate outside of Southern Asia and Eastern Africa (Figure 1) and can be associated with CipNS (Figure 2—figure supplements 3–4), azithromycin (Figure 6), or ceftriaxone (Table 3), the drugs currently recommended by the World Health Organization as first choice treatment for enteric fever (World Health Organization, 2022).

Our data demonstrate that CipNS is emerging or established in all regions except Melanesia (here represented by n=35 genomes from Fiji and Papua New Guinea, mainly from 2010, although more recent reports support a lack of CipNS in Fiji [Davies et al., 2022; Getahun Strobel et al., 2019]; see Figure 2—figure supplement 1). For countries with sufficient data to assess (≥50 genomes), CipNS was emerging or established in all countries except Ghana (Figure 2, Figure 3—figure supplement 1), with no evidence of declining prevalence (Figure 3, Figure 3—figure supplement 1). A diverse range of genotypes and QRDR mutations are involved (Figure 2—figure supplements 3–4), likely reflecting the lack of fitness cost associated with these mutations (Baker et al., 2013). That QRDR mutations are so widespread is highly concerning, as infections with CipNS strains can take longer to resolve, and full clinical resistance can emerge relatively easily against this background, through acquisition of either a mobile qnr gene (as occurred in 4.3.1.1.P1 in Pakistan) or additional QRDR mutations (as occurred in 4.3.1.2.1 in India). Notably, the data suggest that CipR typhoid is now a well-established problem across Southern Asia and is emergent in Chile, Mexico, and South Africa (Figures 2 and 3, Figure 2—figure supplement 1, Figure 3—figure supplement 1). A recent study estimating national annual antibiotic consumption highlighted differences in rates of fluoroquinolone usage between regions and countries, which could potentially drive these differences in resistance prevalence (Browne et al., 2021). The highest rates of fluoroquinolone consumption were estimated in South Asian countries, rising from 1.67 defined daily doses (DDD) per 1000 per day in 2000 to 2.81 DDD/1000/day in 2010 and 2.94 DDD/1000/day in 2018 (see https://www.tropicalmedicine.ox.ac.uk/research/oxford/microbe/gram-project/antibiotic-usage-and-consumption). Fluoroquinolone consumption was also estimated to increase substantially in Latin America, rising from 0.64 DDD/1000/day in 2000 to 1.85 DDD/1000/day in 2010 and 2.26 DDD/1000/day in 2018. Our data show that the highest incidence of CipR burden is associated with four main variants (Figure 2—figure supplement 5). In Pakistan, India, and Bangladesh, it is associated with locally emerged variants; however, the relatively high burden in Nepal is associated with variants acquired from India (Britto et al., 2018; Pham Thanh et al., 2016a). In other regions, CipR burden is low and so far linked mainly to the spread of 4.3.1.2.1 (Britto et al., 2020; da Silva et al., 2022) out of India (Britto et al., 2020; da Silva et al., 2022), plus occasional de novo emergence of resistant variants, which show no evidence of geographical spread (Figure 2—figure supplement 5). However, the high rates of CipNS in Kenya (53%) and Nigeria (40%) are concerning, especially given the increasing usage of fluoroquinolones in these countries (estimated 2.1 DDD/1000/day in 2018 in Kenya and 2.76 DD/1000/day in Nigeria) (Browne et al., 2021), which could potentially drive local emergence and spread of CipR.

While resistance to azithromycin and ceftriaxone have been detected (Table 3, Figures 4—6, Figure 2—figure supplements 1–2, Figure 3—figure supplement 1), their prevalence remains low and, with the exception of XDR 4.3.1.1.P1, clonal expansion of resistant variants has not been observed. To our knowledge, there are no data reported on the fitness cost of acrB mutations or CefR plasmids in Typhi; however, the genomic evidence suggests a higher fitness cost compared with QRDR mutations, providing further support for the use of ceftriaxone or azithromycin over ciprofloxacin as we work to introduce preventative measures. Most instances of ESBL-gene carriage in Typhi (conferring CefR phenotype) have been short-lived (Table 3), suggesting selection against the acquisition of new ESBL genes or plasmids. The expansion and dominance of the XDR 4.3.1.1.P1 genotype in Pakistan is obviously concerning (Figures 4 and 6, Figure 2—figure supplement 3a, Figure 3—figure supplement 1); however, despite circulating at high prevalence in Pakistan for more than 5 years, the strain remains azithromycin-susceptible. There is also limited evidence of local transmission of 4.3.1.1.P1 in other countries; however, most countries near Pakistan have limited data available. A short local outbreak of XDR 4.3.1.1.P1 was reported in China, linked to contamination of an apartment block’s water (Wang et al., 2022) and non-travel-associated cases have been reported in the USA (Hughes et al., 2021). Notably, a CefR+CipR lineage of 4.3.1.2.1 that appears to be well established in Mumbai, India, has been isolated only occasionally since 2015 (Argimón et al., 2021b; Chattaway et al., 2021; Ingle et al., 2021; Jacob et al., 2021; Table 3); however, this is the only example of persistence of a CefR strain besides 4.3.1.1.P1, and there is no evidence it has yet spread outside Mumbai. We hypothesise that the lack of widespread dissemination of 4.3.1.1.P1 and ESBL-positive 4.3.1.2.1 so far may be due to the fitness cost imposed by the associated plasmids (~85 Kbp IncY plasmid in 4.3.1.1.P1 [Klemm et al., 2018]; ~43 Kbp IncX3 plasmid in 4.3.1.2.1 [Argimón et al., 2021b]). The temporal trend data on MDR prevalence and IncHI1 plasmids (Figure 3—figure supplement 2) suggest that migration of the MDR locus from the plasmid to the chromosome may have mitigated the fitness cost associated with plasmid-borne MDR. The same may be true for ESBL genes, that is, the movement of the ESBL locus from the plasmid to the chromosome (as has recently been reported in 4.3.1.1.P1; Nair et al., 2021) may result in a fitter CefR or XDR variant that can spread more easily. Our data show that acrB mutations are occurring spontaneously and independently in multiple locations across a variety of genetic backgrounds (Figure 5). While they are still not prevalent, increased use of azithromycin through public health programmes (e.g. trachoma elimination) as well as widespread misuse of azithromycin to treat SARS-CoV-2 infections and use of azithromycin as first-line therapy for typhoid-like illness may lead to increased selection pressure. It will therefore be important to maintain and expand genomic surveillance, particularly in typhoid endemic countries where azithromycin is used widely. It is also notable that, while they are rare overall, acrB mutations have already arisen in two of the most common CipR lineages (4.3.1.2 and 4.3.1.3.Bdq); this relatively frequent co-occurrence warrants continued monitoring and investigation. While we did not detect the mobile AziR gene mphA, it is circulating in other S. enterica serovars (Nair et al., 2016; Tack et al., 2022) and other enteric bacteria that share plasmids with Typhi (including the human-specific Shigella; Baker et al., 2018), providing another potential mechanism for emergence of AziR in Typhi.

We are at a pivotal stage in the history of typhoid control. Wider access to clean water and improved sanitation have led to a major reduction in global incidence of typhoid fever, which has also been reflected in declining incidence of other enteric diseases (Steele et al., 2016). This should continue but will require sustained investment from national and local governments and thus remains a long-term objective. In the short to medium term, widespread use of TCVs can help to further reduce global incidence of typhoid fever. The WHO has prequalified two TCVs and recommended their use in endemic countries, as well as settings where a high prevalence of AMR Typhi has been reported (World Health Organization, 2018). Gavi, the Vaccine Alliance, has committed funds to support the procurement and distribution of TCVs in typhoid endemic countries (Gavi: The Vaccine Alliance, 2023a; Gavi: The Vaccine Alliance, 2023b). Five countries have undertaken Gavi-supported national introductions (Pakistan, Liberia, Zimbabwe, Nepal, Malawi) and one country has self-financed a national introduction (Samoa) (Neuzil, 2020; Sikorski, 2020). In Pakistan and Zimbabwe, TCV introduction was stimulated by the occurrence of AMR Typhi outbreaks in major urban centres, highlighting that the case for prevention can be stronger when curative therapy is less available. Additional support is likely required to inform TCV decision-making in other typhoid endemic countries, particularly where burden and AMR data are scarce.

With increasingly limited treatment options, vaccines are an even more important tool to mitigate the public health burden of AMR Typhi, both through the prevention of drug-resistant infections and through broader, indirect effects, like reduction of empiric antimicrobial use leading to reduced selection pressure. While TCVs have been shown to be highly effective against drug-resistant Typhi (Batool et al., 2021; Yousafzai et al., 2021), public health policymakers have to weigh the value of TCVs against other competing immunisation priorities. While TCV introduction is scaled up globally, antimicrobial stewardship should also be prioritised. Aggregated, representative data showing distribution and temporal trends in AMR can inform local treatment guidelines to extend the useful lifespan of antimicrobials licensed to treat typhoid fever, potentially including reverting to former last-line drugs in some settings. The traffic light system presented in this analysis (see Figure 2 and Figure 2—figure supplement 1) provides a framework for monitoring trends in AMR and adjusting empiric therapy guidelines accordingly. The WHO recently released its AWaRe (Access, Watch, Reserve) treatment guidelines (World Health Organization, 2022), which indicate that choice of empiric therapy should be guided by severity of presentation and local risk of fluoroquinolone resistance; if low risk, oral ciprofloxacin is recommended for both mild and severe cases and if there is a high risk of fluoroquinolone resistance, oral azithromycin is recommended for mild cases and intravenous ceftriaxone is recommended for severe cases. However, the guidelines do not indicate which prevalence rate of resistance should warrant avoidance of treatment with ciprofloxacin, nor do they indicate where high prevalence rates of resistance might be expected, although it is noted that drug resistance is most prevalent in Asia. There is an opportunity to further refine these recommendations with additional, local information about AMR prevalence and trends over time. Additional data are required from resource-limited settings, where typhoid fever diagnosis is often based on clinical presentation, to optimise these recommendations.

Genomic surveillance has a particularly important role to play in monitoring for changes in clinically important resistances in Typhi, as a shift in resistance mechanism or early evidence of clonal spread, which can only be identified definitively using WGS, could provide early warning of a likely increase in prevalence. This study provides an analytical framework for Typhi genomic analysis, based on an open, robust, reproducible data flow and analysis framework leveraging open-access online data analysis platforms (Typhi Mykrobe for read-based genotyping [Ingle et al., 2022]; the GHRU pipeline for genome assembly [Underwood, 2020], and Typhi Pathogenwatch for assembly-based genotyping and tree-building [Argimón et al., 2021b]). We have made available all data processing and statistical analysis code, and underlying sequence and metadata, via GitHub and FigShare (see Methods). Together, these provide (i) a comprehensive data and code resource for the research and public health communities interested in typhoid surveillance data; (ii) a model for the inclusion of WGS in project-based or routine surveillance studies of typhoid that can be readily replicated and adapted; and (iii) a sustainable model for aggregated analysis of typhoid genomic surveillance data that can readily incorporate new data and extract features (genotypes, AMR determinants, plasmid replicons) of importance to clinical and public health audiences. Notably, this consortium-driven effort shows that new insights can be gained from aggregated analysis of published data, which were not evident from the individual contributing studies, for example (i) the XDR strain 4.3.1.1.P1 existed in Pakistan in 2015, a year earlier than previously reported (Figure 4); (ii) the CefR+CipR strain reported in Mumbai (Argimón et al., 2021b; Jacob et al., 2021) has persisted between at least 2015 and 2020 and is now more easily identified as 4.3.1.2.1 with bla_SHV-12; (iii) persistence of MDR in certain settings is correlated with migration of MDR from plasmid to chromosome (Figure 3—figure supplement 2), which has implications for the future persistence and potentially spread of ESBL strains.

This dataset provides clear, actionable information about the distribution and temporal trends in AMR across multiple countries and regions. Where data gaps exist, the potential of travel-associated data to serve as ‘sentinel’ surveillance has been demonstrated previously by Ingle et al., 2019, and supported by additional data included in this analysis. These data can and should inform prioritisation of TCV introduction and improvements to WASH infrastructure. Sustaining and expanding genomic surveillance can also facilitate measuring the impact of TCV introduction on local bacterial populations, as has been done for previous vaccines like pneumococcal conjugate vaccines. In addition, monitoring for potential ‘strain replacement’ with other Salmonella serovars following TCV introduction can and should inform the prioritisation of the development and deployment of future combination Salmonella vaccines.

The SARS-CoV-2 pandemic illustrated the power of open, continuous data sharing and crowdsourced analysis, and the importance of ensuring that genomic surveillance leads to local benefits. The scale of this analysis, which was made possible through the efforts of an extensive network of collaborators, enables the extraction of key insights of public health relevance. The authors hope that this consortium effort serves as a starting point for continued data generation and sharing and collective analysis, with additional participation from an expanded group of stakeholders. In particular, we hope that researchers and public health authorities from areas with little publicly available data see the value of reporting and sharing genomic data for collective public health benefit. In addition, we hope that the current momentum for donor and government support of molecular surveillance is sustained, so that additional groups are able to generate their own data and fill regional data gaps to inform local public health action.