The methods and main findings from GEMS have previously been described [22–24]. Briefly, GEMS participants were recruited from censused populations during 2007–2010 in The Gambia, Mali, Mozambique, Kenya, Bangladesh, India, and Pakistan. Study participants included children aged 0–59 months of age with MSD who presented to a sentinel health facility (see Supplementary Methods for additional details). Children were recruited into 0–11-, 12–23-, and 24–59-month age groups. For each child with MSD (case) enrolled, 1–3 children without diarrhea during the previous week (controls) were recruited. Scientific and ethics committees and institutional review boards of participating institutions in each country as well as the coordinating institution, University of Maryland, Baltimore, approved the study protocol prior to implementation. Informed consent was obtained in the local dialect from all participating caretakers before recruitment of their children into the study.

A panel of enteropathogens was identified from stool specimens, collected at the clinic from MSD cases or obtained at home by caregivers of children in the control group, as previously described [24]. Salmonella spp. were shipped to the Center for Vaccine Development and Global Health (CVD) at the University of Maryland School of Medicine for additional characterization.

At CVD, Salmonella spp. were agglutinated using polyvalent O and O1 antisera followed by serogroups O:2 (A), O:4 (B), O:6,7 and O:7 (C₁), O:6,8 and O:8 (C₂-C₃), O:9 (D₁), O:9,46 (D₂), O:3,10 (E₁), O:11 (F), and O:13 (G) antisera (Denka Seiken, Tokyo, Japan). Serovars Typhimurium, Typhi, Enteritidis, and Paratyphi B were fully serotyped (using O and H typing antisera) and additionally confirmed by polymerase chain reaction (PCR) [25, 26].

Sequence types were determined for all 87 Typhimurium isolates using multi-locus sequence typing (MLST) by PCR and sequencing and/or by examining whole-genome sequences. Sequence typing by MLST followed methodology described previously [15].

The majority of the Salmonella isolates (355 out of 370) were subjected to whole-genome sequencing. Following sequencing, 120 isolates were excluded from subsequent analyses as they did not meet the quality-control criteria. Details of sequencing and phylogenetic analyses are described in the Supplementary Methods.

The susceptibility of the 370 Salmonella isolates to chloramphenicol, ampicillin, ciprofloxacin, trimethoprim/sulfamethoxazole (TMP/SMX), gentamicin, and ceftriaxone was determined using the Kirby-Bauer disk-diffusion method and interpreted according to Clinical and Laboratory Standards Institute guidelines. Multidrug resistance was defined as resistance to ampicillin, chloramphenicol, and TMP/SMX. To assess whether the high resistance of NTS to antimicrobials was associated with antibiotic prescription rate, we determined the percentage of children with MSD (and Salmonella isolated in stools) who had been prescribed (but may or may not have been given) antimicrobial agents after visiting any of the sentinel health facilities that participated in GEMS.

To determine which individual Salmonella serovars were driving the association between Salmonella and MSD that was found in the original GEMS analyses [22], we used the same conditional logistic regression model as previously described [27]. Instead of including Salmonella species in the model we included variables for each Salmonella serogroup/serovar. The association of each serovar with MSD was adjusted for other co-pathogens. The rationale for this approach, generally, and in the unique context of GEMS, has been discussed previously [27]. Analyses were conducted using R version 3.3.2 (R Foundation for Statistical Computing). P values less than .05 were considered statistically significant.

Of the cases with Salmonella identified, 86 (44%) were in the 0–11-month age group, 55 (29%) were in the 12–23-month age group, and 49 (26%) were in the 24–59-month age group (Table 1). Cases experienced severe signs of MSD. Approximately 20% of infants with Salmonella spp. detected had bloody diarrhea, while 100% of children aged 12–59 months old with Salmonella spp. produced watery diarrhea. Of note, a lower proportion of children with MSD who had Salmonella isolated tended to be female in all age groups.

The serovar distribution of the 370 Salmonella isolates (190 from cases and 180 from controls) collected from stools of study participants is shown in Figure 1. Of these, 361 were NTS. Additionally, we recovered 8 Typhi from Asia and 1 Paratyphi A isolate from Bangladesh. The most frequent NTS serovars identified were Typhimurium, serogroup O:8 (C₂-C₃), serogroup O:6,7 (C₁), Paratyphi B Java, and serogroup O:4 (other than Typhimurium or Paratyphi B). Serovar Typhimurium predominated in Africa, whereas serogroup O:6,7 (C₁) and O:8 (C₂-C₃) serovars were the most common in Asia.

The prevalence of the most abundant serovars isolated in GEMS, as well as serovar Enteritidis (due to its importance in iNTS disease in Africa), is shown in Table 2. The individual serovars of isolates are listed in Supplementary Table 1. In general, we found that the rates of NTS isolation were low (≤5.3%) in both cases and controls regardless of age groups, although some site-to-site variation was apparent. At the Kenya site, serovar Typhimurium, the most prevalent serovar, was recovered in stools of MSD cases at a rate of 3.3% for infants, 3.7% for toddlers, and 4.3% for young children. In Bangladesh, Paratyphi B Java, the most prevalent serovar there, was recovered from 2.0% of infants with MSD. Serogroup O:8 (C₂-C₃) organisms were most prevalent in stools at the Pakistan site. The prevalence of NTS in cases and controls in The Gambia, Mali, Mozambique, and India was less than 1.5%.

Previously, 3.2% and 3.7% of MSD episodes in toddlers (12–23 months) and children (24–59 months) at the Kenya site, respectively, and 4.6% of MSD episodes in infants at the Bangladesh site were shown to be attributable to Salmonella [22]. We determined the serovars driving the associations by using a conditional logistic regression model (Supplementary Table 2). In Bangladesh, serogroup O:6,7 (C₁) (odds ratio [OR], 6.4; 95% confidence interval [CI]: 1.84–22.58), O:8 (C₂-C₃) (OR, 6.0; 95% CI: 1.28–28.33), and serovar Paratyphi B Java (OR, 4.8; 95% CI: 1.87–12.29) were significantly associated with MSD. In Kenya, the association was driven by serovar Typhimurium among children aged 12–23 months (OR, 4.3; 95% CI: 1.86–9.93) and 24–59 months (OR, 4.9; 95% CI: 2.09–11.64). All other serovars occurred in too few cases and controls to produce significant results.

Salmonella isolates from the African and Asian sites differed in terms of their antimicrobial susceptibility (Figure 2). Isolates from Africa were susceptible to ciprofloxacin and ceftriaxone, whereas resistance to these antibiotics was observed among Asian NTS isolates. We observed 65.4% of NTS from MSD cases in Kenya to be multidrug resistant (MDR) to ampicillin, TMP/SMX, and chloramphenicol. However, isolation of nonsusceptible NTS was less frequent at the site in The Gambia; only 6.7% of NTS from cases showed an MDR phenotype (Figure 2A). In Asia, Indian isolates showed more resistance to the antibiotics tested than isolates from the other 2 Asian sites (Figure 2B). We observed antimicrobial susceptibility profiles among NTS from controls that were similar to cases at each site except for Kenya and India.

Serovars Typhimurium and Enteritidis from Africa and serogroup O:6,7 (C₁) and serogroup O:13 (G) isolates from Asia showed the highest percentage of antimicrobial resistance (Figure 3). All Enteritidis and Paratyphi B Java isolates recovered from GEMS stools from Asia were pan-susceptible to antimicrobial agents, while the 6 serogroup O:13 (G) isolates from Africa were pan-susceptible. Five of the 8 (62.5%) serovar Typhi from Asia (Pakistan and India) were MDR.

In general, most MSD cases with Salmonella had been prescribed or given an antibiotic (except in Pakistan) (Table 3). Trimethoprim/sulfamethoxazole was the most commonly prescribed antibiotic in Africa, whereas ciprofloxacin was the most common in Asia.

Since Typhimurium was the most important cause of iNTS disease at several GEMS sites and was the most frequent serovar isolated from stools, a phylogenetic analysis was performed. Of 87 Typhimurium isolates, 74 (85.0%) were from Africa (Kenya), while 13 (14.9%) were from Asia (Pakistan, India, and Bangladesh). Table 4 shows the sequence types (ST) of these Typhimurium isolates listed by site of origin.

A phylogeny was constructed using whole-genome sequences to determine the relationship between the Typhimurium ST313 isolates from MSD cases and controls (Figure 4). The African ST313 sequence type has been divided into the older lineage 1 isolates and the more recent lineage 2 [13, 28]. Here, 50 of 55 study isolates analyzed (90.9%) clustered with the ST313 lineage 2 reference genome D23580, 49 of 55 (89%) of which showed the typical MDR phenotype associated with lineage 2, namely resistance to chloramphenicol, co-trimoxazole, and ampicillin. The lineage 2 isolates from the MSD cases and diarrhea-free controls were closely related and could not be distinguished phylogenetically. A group of 5 isolates in stools of cases and controls, regardless of age, formed a small lineage 2 subcluster associated with susceptibility to chloramphenicol.

Four of the 55 isolates (7%) clustered with the ST313 lineage 1 reference genome A130, all of which were isolated from MSD cases in Kenya and were sensitive to chloramphenicol. Of note, the 1 study isolate (specimen 700477) that failed to cluster with lineage 1 or lineage 2 and demonstrated pan-susceptibility to antibiotics was isolated from a case of MSD from Pakistan.