This was a retrospective laboratory-based study. We used archived samples from children aged less than 5 years collected in 2015 from a cross-sectional study investigating the prevalence of fecal carriage and seroprevalence of non-typhoidal Salmonella at three sites in Kenya ( Supplementary Figure 1 ). Volunteers were recruited from three sites in Mukuru, Nairobi County; Asembo, Siaya County; and Junju, Kilifi County ( Supplementary Figure 1 ). Mukuru (referred to as Nairobi) is an urban informal settlement (slum area) in Nairobi City (15, 16); Asembo (referred to as Siaya), in Western Kenya, is a semirural setting (17) and was a Kenyan site included in the GEMS study (4, 18); while Junju (referred to as Kilifi) is a rural setting on the Kenyan Coast where children have been enrolled in a longitudinal malaria cohort from birth to the age of 15 years separate from the current study (19, 20). Participants were identified evenly across the three sites by age-stratified, simple random sampling into 0–6-month, 7–11-month, 12–23-month, 24–35-month, 36–47-month, and 48–59-month strata, taking equal proportions in each stratum. The sampling frame was based either on the most recent census data (Nairobi), population-based surveillance census data (Siaya), or demographic health surveillance data (Kilifi). Fieldworkers within these areas recruited the participants after information sharing and obtaining informed consent from the parent/guardian of the participating children.

Participants were asked to attend a clinic visit at each trial location, at enrolment where anthropometric measurements such as weight, height (length for those under 2 years of age), and mid-upper arm circumference (MUAC) were collected. In addition, a short questionnaire was administered to the parent/guardian to collect information on sociodemographic and environmental (e.g., drinking water source, washing water source, toilet type) characteristics. Data on diarrhea episodes (defined as >3 loose stools in 24 h) and febrile illness (defined as temperature >38°C) within the last 2 weeks of the enrolment visit were also collected. A point-of-care test was conducted for hemoglobin level as well as a Plasmodium falciparum malaria HRP-2-based rapid diagnostic test (RDT). As part of the main study, the primary sample for analysis was stool, which was prioritized for collection, whereas sera were the secondary samples collected alongside the stools; thus, not all volunteers who were enrolled had a 2-ml blood sample collected for serum separation. A subset of the stool samples from the Kilifi site underwent PCR using the TaqMan array (21), with the results for the Shigella PCR targets available for analysis. Serum samples from children aged less than 5 years were assayed for serotype-specific IgG to Shigella.

After collection, serum samples were stored at −80°C in the KEMRI-Wellcome Trust Biobank until 2021 when an aliquot of each was shipped to GSK Vaccine Institute for Global Health in Siena, Italy, for analysis. Anti-Shigella serum IgG was measured by standardized enzyme-linked immunosorbent assay (ELISA) using S. flexneri 6 and S. sonnei LPS or S. flexneri 1b, S. flexneri 2a, and S. flexneri 3a O-antigen (OAg) as coating antigens (at final concentrations of 15 µg/ml in PBS for S. flexneri 6, 0.5 µg/ml in PBS for S. sonnei, 2 µg/ml in carbonate buffer for S. flexneri 1b, 0.5 µg/mL in carbonate buffer for S. flexneri 2a, and 1 µg/ml in PBS for S. flexneri 3a, respectively) and 1:4500 dilution of goat anti-human IgG-alkaline phosphatase secondary antibody (Wholesaler, UK Sigma-Aldrich) using the methodology previously described (22). ELISA units were expressed relative to a five-parameter human antigen-specific antibody standard serum curve composed of 10 standard points and 2 blank wells (run in duplicate on each plate). One ELISA unit is defined as the reciprocal of the dilution of the standard serum that gives an absorbance value [optical density (OD) measured at 405 nm minus the OD measured at 490] equal to 1 in the assay. Standard serum to assess anti-S. sonnei LPS-specific antibodies was obtained by pooling sera from four individual high responders initially enrolled in a clinical trial who received two vaccinations with S. sonnei monocomponent 1790GAHB (23), whereas the standard sera used to assess anti-S. flexneri-specific OAg antibodies were obtained by pooling sera prevaccination from nine individual subjects enrolled in the clinical trial conducted in Pakistan and India (24), where Shigella is endemic, that were screened for response against each of the abovementioned S. flexneri serotypes. Anti-S. sonnei standard was used to determine anti-S. sonnei-specific antibodies in a phase 2b trial conducted with 1790GAHB (25), and the assay was fully characterized in terms of standard curve accuracy, linearity, specificity, precision, and limits of quantification; the anti-S. sonnei standard sera were calibrated in order to meet the definition of 1 EU as the reciprocal of the dilution giving an OD equal to 1 for this study. The primary anti-S. flexneri standard sera were calibrated against each coating antigen, and the assay was characterized in terms of standard curve accuracy and specificity. The limits of blanks and quantification were established.

Several QC criteria were applied on each run, particularly minimum R ² value ≥0.96 for the 5PL curve fit to standard dilution series, maximum background <0.15 OD for S. sonnei coating and <0.1 for S. flexneri coating, minimum value of OD maximum ≥2.6, range between 0.5 OD and 2 OD for 1 EU/ml, and Control Dev <40% to the expected EU/ml both for the high and low controls. If at least one of the abovementioned criteria was not met, the entire layout was repeated under the same experimental conditions. For each sample, the EU/ml was determined as the average of EU/ml of the triplicate at each specific sera dilution if the control CV was <30%; otherwise, the sample was rerun under the same conditions. The limits of standard curve accuracy (LSCAs) were set at the nominal value of the standard at the last and the first serial dilutions, respectively, where the confidence interval of RE% with 90% probability fell within the acceptance range of [−25%; +25%]. The lower limit of quantification (LLoQ) was set as equivalent to lower LSCA multiplied by 100 (the lowest sera dilution tested in the assay). Specificity was determined by running ELISA on the samples depleted with 50 μg/ml of homologs or heterologous polysaccharides in comparison with the undepleted control. The assay was considered specific if inhibition was >80% in the presence of homologous polysaccharide and <20% with heterologous polysaccharides (either from the same species, thus S. sonnei for S. flexneri and S. flexneri for S. sonnei, and from a different species—Salmonella Typhimurium for all).

Anti-specific LPS/OAg IgG antibodies were summarized by age class and site as GMCs, while seropositivity was expressed as a binary variable (1 = positive; 0 = negative) and defined as an antibody concentration at least four-fold greater than the lower limit of quantification (4 * LLoQ) for each specific serotype. The LLoQ was 9 EU/ml for S. sonnei, 8 EU/ml for S. flexneri 1b, 2 EU/ml for S. flexneri 2a, 11 EU/ml for S. flexneri 3a, and 2 EU/ml S. flexneri 6.

Our outcome variable of interest was IgG antibodies against Shigella measured by ELISA, assessed for each sample collected from each subject, classified as seropositive or seronegative as defined above. More specifically, as a population-level summary, we looked at seroprevalence (based on the percent of samples seropositive) and at the magnitude of IgG antibodies (as reflected by antibody GMC). Our explanatory variables for seropositivity to Shigella serotypes included sociodemographic characteristics, water source, toilet type, drinking water type, and serotypes. Sociodemographic characteristics for this study included age (in months), sex, diarrhea, and fever and were summarized using frequencies with anthropometric measures. Anthropometric measures were defined as follows: underweight [weight-for-age Z-score (WAZ) <−2]; wasting [weight-for-height Z-score (WHZ) <−2], and stunting [height-for-age Z-score (HAZ) <−2] based on the WHO Child Growth Standards (26). Anemia was defined by a hemoglobin level of less than 10 g/dl. In addition, data on anti-Salmonella Enteritidis and Salmonella Typhimurium IgG concentrations from the parent study were made available and classified as seropositive or seronegative to determine their association with Shigella serotype-specific antibodies. TaqMan array PCR data (only available for the Kilifi site), specific for S. sonnei and S. flexneri 1b, 2a, 3a, and 6, were used to further determine their relationship with the antibody concentrations.

The dataset was checked for missing values and completeness and exported to R version 4.0.3 (R Core Team 2019) and Stata version 15 (Stata Corp., College Station, TX, USA) for analysis. We used descriptive statistics for sociodemographic characteristics. We presented the continuous variables using median and interquartile ranges (IQRs), while categorical variables were presented using frequency and percentages. We used the Kruskal–Wallis test to determine the statistically significant difference between the medians of the continuous variables and the independent groups (study sites). We used the chi-square test to assess the association between categorical variables. Participants were grouped and analyzed into six age categories: 0–6 months, 7–11 months, 12–23 months, 24–35 months, 36–47 months, and 48–59 months. Shigella serotype-specific antibodies were presented together with the 95% confidence interval across age groups and by site. The non-parametric test of trend was used to compare the association between sites and the ranks across ordered age groups. We visualized the prevalence of the IgG antibody by age using a bar plot and the associated 95% Agresti–Coull confidence intervals, violin plots for the distribution of IgG antibodies by age, and heatmaps for seroprevalence by age and site. Pearson correlation was used to test the association among the respective serotype-specific antibody responses.

The chi-square test for trend was used to compare the association between seropositivity and age groups. Non-parametric one-way analysis of variance (ANOVA) test was used to determine the association between serotype antibody concentrations and age groups.

We used univariable and multivariable logistic regression models to investigate factors associated with Shigella serotype-specific antibodies. For each antigen, the site was checked as a potential effect modifier for the association between age and seropositivity. The presence of effect modification was determined using a likelihood ratio test (LRT). To determine the explanatory variables to be included in the final models (one for each Shigella serotype), we used LRT. We reported both the crude and adjusted odds ratios with their corresponding 95% confidence intervals (CIs). All tests were considered significant at p <0.05 with no adjustment for multiple comparisons.

A total of 474 participants (169 from Nairobi, 185 from Siaya, and 120 from Kilifi) with a median age of 13.4 months (IQR 7.0–35.6) had their samples analyzed ( Table 1 ). There were no age or sex differences across the sites.

As shown in Table 1 , diarrhea and fever in the preceding 2 weeks were reported among 13.5% (64/474) and 23.6% (112/474) participants, respectively, and diarrhea and fever were reported among 7.2% (34/474) and 19.6% (93/474) family members, respectively. At the time of sampling, 15.2% (72/474) of the participants had anemia, 20.5% (97/474) were stunted, 5.9% (28/474) were wasted, and 9.5% (45/474) were underweight. Several site-specific differences were observed: diarrhea episodes were more commonly reported in Nairobi (p = 0.02); both participants and household members had more frequent fever episodes in Siaya (p < 0.001); and anemia (p < 0.001), wasting (p = 0.050), and being underweight (p = 0.03) were more common in Kilifi.

The observed seroprevalence of antibodies to S. flexneri 1b, S. flexneri 2a, S. flexneri 3a, and S. sonnei was 50.8%, 61.4%, 44.7%, and 43.5%, respectively, showing a relationship with age. However, no age-related difference was observed for S. flexneri 6 (100% across all ages and sites apart from 0 to 6 months for Kilifi and Siaya at 92.9% and 97.5%, respectively) ( Table 2 and Figure 1 ). The overall age-related difference was significant for S. flexneri 2a (p = 0.01) and S. sonnei (p = 0.01), for which a U-shaped relationship was observed where the seroprevalence was 56.1% and 33.6% at 0–6 months and decreased to 39.5% and 18.4% in the 7–12-month group, respectively, for S. flexneri 2a and S. sonnei. Seroprevalence increased in the subsequent age groups with peaks observed above the 36-month and 48-month age groups for S. flexneri 2a and S. sonnei, respectively. Seroprevalence was the lowest among the 7–11-month age groups against S. flexneri 1b, S. flexneri 2a, S. flexneri 3a, and S. sonnei. Prevalence against S. flexneri 1b, S. flexneri 2a, S. flexneri 3a, and S. sonnei was ≥70% by the age of 59 months. A significant prevalence increase was shown in the second year of life compared with the first year of life for S. flexneri 2a (p = 0.006), S. flexneri 3a (p = 0.006), and S. sonnei (p = 0.05). Except for S. flexneri 6, there was no evidence that the site modifies the association between age and antigen seropositivity ( Figure 2 ).

Overall, the magnitude of the antibody response increased with age ( Figure 3 and Supplementary Figure 2 ). The geometric mean concentration of the antibodies in the 12–23-month age groups was significantly higher compared with the 7–11 months for all five serotypes (Kruskal–Wallis, p = 0.01–0.0001). A significant difference was also observed for S. flexneri 2a, S. flexneri 6, and S. sonnei in the antibody concentrations between 0–6-month- and 7–11-month-old children, whereas there was no difference for S. flexneri 1b [24.7 (95% CI 17–36) vs. 17.7 (95% CI 13–25)] and S. flexneri 3a [23 (95% CI 16–32) vs. 18.2 (95% CI 13–25)]. Concerning the site, the patterns were similar across the three sites with no difference in the geometric mean antibodies against S. flexneri 2a between those 0–6 months and 7–11 months or those 7–11 months and 12–23 months old across all the three sites ( Figure 4 ) with low antibody levels detected despite the high seroprevalence ( Figure 1 ). Interestingly, all antibodies against S. flexneri 6 were above the seropositivity cutoff ( Figure 3D ).

The breadth of antibody response by age and site is shown in Figure 5 as well as the correlation of the antibody response among the various Shigella serotypes, Salmonella Typhimurium, and Salmonella Enteritidis ( Supplementary Figure 3 ). Overall, exposure to multiple serotypes (n = 2 to 5) increased from ~50% at 7–11 months to ~85% by the age of 12–23 months ( Figure 5A ). Overall, in the 48–59-month age group, approximately 50% of children had antibodies to all serotypes, and approximately 50%, 45%, and 35% of participants in Nairobi, Siaya, and Kilifi, respectively, had antibodies to all five serotypes ( Figure 5B ). All antibody responses across the five Shigella serotypes were significantly positively correlated. Importantly, a strong association was observed between the presence of S. flexneri 1b and S. flexneri 3a (ρ = 0.864, p < 0.001) ( Supplementary Figure 3 ).

In the multivariable analysis, being seropositive against S. flexneri 3a increased the odds of being S. flexneri 1b positive by 23.85 (95% CI 11–54, p < 0.001) ( Supplementary Table 4 ). We further observed a positive correlation between the responses to the five Shigella serotypes with antibody responses against both Salmonella Enteritidis and Salmonella Typhimurium ( Supplementary Figure 3 ). In a subset of subjects, analysis of stools by PCR did not show any relationship between the presence of Shigella in stool samples and antibodies ( Supplementary Table 3 ). PCR positivity for this analysis was only 9.8%. All the other covariates tested after adjustment had no effect on seropositivity.