key: cord-0274460-qad9gork
authors: Aiemjoy, K.; Seidman, J. C.; Saha, S.; Munira, S. J.; Sajib, M. S. I.; Al Sium, S. M.; Sarkar, A.; Alam, N.; Nusrat Jahan, F.; Kabir, M. S.; Tamrakar, D.; Vaidya, K.; Shrestha, R.; Shakya, J.; Katuwal, N.; Shrestha, S.; Yousafzai, M. T.; Iqbal, J.; Dehraj, I. F.; Ladak, Y.; Maria, N.; Adnan, M.; Pervaiz, S.; Carter, A. S.; Longley, A.; Fraser, C.; Ryan, E. T.; Nodoushani, A.; Fasano, A.; Leonard, M.; Kenyon, V.; Bogoch, I. I.; Jeon, H. J.; Haselbeck, A.; Park, S. E.; Zellweger, R.; Marks, F.; Owusu-Dabo, E.; Adu-Sarkodie, Y.; Owusu, M.; Teunis, P.; Garrett, D. O.; Luby, S.; Qamar, F. N.; Saha,
title: Estimating the incidence of typhoidal Salmonella from cross-sectional serology using antibody dynamics following infection
date: 2021-10-21
journal: nan
DOI: 10.1101/2021.10.20.21265277
sha: ed73d97bad30a0dee216b126092ccaa10925bace
doc_id: 274460
cord_uid: qad9gork

The incidence of enteric fever, an invasive bacterial infection caused by typhoidal Salmonellae, is largely unknown in regions lacking blood culture surveillance. New serologic markers have proven accurate in diagnosing enteric fever, but whether they could be used to reliably estimate population-level incidence is unknown. We collected longitudinal blood samples from blood culture-confirmed enteric fever cases enrolled from surveillance studies in Bangladesh, Nepal, Pakistan and Ghana and conducted cross-sectional sero-surveys in the catchment areas of each surveillance site. We used ELISAs to measure quantitative IgA and IgG antibody responses to Hemolysin E (HlyE) and S. Typhi lipopolysaccharide (LPS). We used Bayesian hierarchical models to fit two-phase power-function decay models to the longitudinal sero-responses among enteric fever cases and used the joint distributions of the peak antibody titers and decay rate to estimate population-level incidence rates from cross-sectional serosurveys. The longitudinal antibody kinetics for all antigen-isotypes were similar across countries and did not vary by clinical severity. The incidence of typhoidal Salmonella infection ranged between 41.2 per 100 person years (95% CI: 34.0-50.1) in Dhaka, Bangladesh to 5.8 (95% CI: 4.8-7.1) in Kavrepalanchok, Nepal and followed the same rank order as clinical incidence estimates. The approach described here has the potential to expand the geographic scope of typhoidal Salmonella surveillance and generate incidence estimates that are comparable across geographic regions and time.

Typhi and Paratyphi A, B, or C, is a major cause of preventable morbidity and mortality in lowand middle-income countries (LMICs) (1) . The incidence of enteric fever is typically ascertained using clinical surveillance, where blood culture-positive enteric fever cases are tallied and reported relative to a catchment area population. Blood culture requires considerable microbiological diagnostic infrastructure that is not widely accessible in many LMICs where the vast majority of enteric fever cases occur. Another limitation of blood culture surveillance is that only individuals seeking care at a surveillance site are captured, whereas many patients receive treatment from pharmacies, private practitioners, or other healthcare providers outside the reach of traditional surveillance systems. Even when available, the estimated sensitivity of blood culture is only 60% (2) and may vary substantially depending on prior antibiotic use and symptom duration. For these reasons, blood culture surveillance covers only a small proportion of the at-risk populations around the world, underestimates the true burden of disease and yields incidence estimates that are not easily comparable across geographic regions and time. There remains a critical need for more accurate, efficient, and cost-effective approaches to estimate and monitor enteric fever incidence in LMICs. This need is particularly acute in the context of the recent World Health Organization (WHO) recommendation and Gavi funding for typhoid conjugate vaccine (TCV) introduction in settings with a high enteric fever burden. Many at-risk countries lack the requisite data on typhoid incidence needed to inform decisions about TCV introduction and to monitor the impact of vaccine introduction.

Population-based serologic surveillance (sero-surveillance) can be used to estimate pathogen transmission in settings where robust facility-based surveillance is not available. Additionally, it may overcome sources of uncertainty or bias in facility-based surveillance caused by healthcare seeking and prior antibiotic use. Cross-sectional sero-surveillance has been useful for rapidly generating data on the transmission of COVID-19, pertussis, dengue, and other diseases (3) (4) (5) (6) (7) .

However, sero-surveillance for typhoidal Salmonella has been limited by the lack of sensitive and specific serologic markers. Serologic responses to the most widely used antigen to-date, virulence (Vi) capsular polysaccharide, have poor diagnostic performance during acute infection and cannot distinguish between natural infection and immune responses to Vi-based vaccine (8) .

S. Typhi lipopolysaccharide (LPS) have demonstrated promising accuracy in the diagnosis of acute enteric fever (9) (10) (11) but have not yet been evaluated as tools to understand population-level infection incidence. HlyE, a pore-forming toxin, causes cytotoxicity and invades epithelial cells (12) (13) (14) . It is present in S. Typhi and S. Paratyphi A but is rarely found in other Salmonella serovars (15) . LPS is a major component of the outer membrane of Gram-negative organisms and is a potent inducer of the innate immune system. LPS is made of three parts: 1) lipid A; 2) the core oligosaccharide which is common across the Salmonella serovars; and 3) the O antigen polysaccharide (or O antigen), the outermost portion of LPS which is the most variable and consists of oligosaccharide repeats. The O antigen is one factor used to determine the serologic specificity of Salmonella; S. Typhi LPS has O antigens 9 and 12 (16) .

Here, we model the longitudinal immune responses to HlyE and LPS among blood cultureconfirmed enteric fever cases enrolled from two multi-year, hospital-based, enteric fever surveillance studies: The Surveillance for Enteric Fever in Asia Project (SEAP) in Bangladesh, Nepal, and Pakistan, (17) and the Severe Typhoid in Africa (SETA) surveillance study in Ghana (18) . We then used the longitudinal antibody dynamics to generate typhoidal Salmonella incidence estimates from cross-sectional population sero-survey data. We compare the serologic incidence estimates to clinical incidence estimates from the same catchment areas (17, 18) .

In prospective surveillance for acute febrile illness between 2017 and 2020, we enrolled 1420 blood-culture confirmed enteric fever cases (407 from Bangladesh, 543 from Nepal, 399 from Pakistan and 71 from Ghana) from the SEAP and SETA studies (Table 1) . We followed cases for a median of 382 days after fever onset (interquartile range 94 -696 days) and collected and analyzed 4126 longitudinal blood samples ( Figure S1 ).

In parallel to clinical surveillance, we conducted cross-sectional serosurveys in the catchment communities around each health facility study site, collecting and analyzing capillary blood Table 2 ).

In the population-based serosurveys, median antibody responses for all antigen-isotypes increased with age and were highest in Dhaka, Bangladesh and lowest in Kavre, Nepal and Agogo, Ghana ( Figure 1 ). Among culture-confirmed enteric fever cases, median responses at 1 month and 6 months after fever onset were comparable across sites and higher than the median values for the serosurvey participants at each site. ( Figure S3 ). Among enteric fever cases, correlation in responses across antigen-isotypes was higher among younger ages and at earlier time points (closer to infection) ( Figure S4 ).

We used models with two-phase kinetics, including a power-function decay rate, to describe the longitudinal change in antibody levels among culture-confirmed cases, estimating baseline (y0) and peak antibody responses (y1), time to peak (t1), and decay rate (α) and decay shape (r). All four antigen-isotypes reached peak levels within three weeks of fever onset ( Table S1 ). The overall decay rate for HlyE IgA was slightly faster than for HLyE IgG. The shape parameters for all antigenisotypes deviate from 1, indicating non-exponential decay (Table S1 ). Median responses remained elevated above baseline levels for 28.8 months for HlyE IgG, 13.6 months for LPS IgG, 11.4 months for HlyE IgA and 2.9 months for LPS IgA. The model fitted antibody trajectories were similar across all four study sites ( Figure 3 ). The distributions for peak antibody responses were nearly identical across countries and the differences between distributions all centered near 0 for all antigen-isotype combinations. There was some inter-country variation in the decay rate with Bangladesh having the slowest decay rate, followed by Pakistan, Nepal then Ghana ( Figure 3 ). The differences in incidence estimates using both the overall and countryspecific longitudinal parameters (peak, decay rate and decay shape) are presented in supplemental Figure S5 . We also compared sero-responses among enteric fever cases who were hospitalized (32.8% (458/1420) and not hospitalized, and found comparable antibody trajectories for all antigen-isotype combinations ( Figure S6 ).

We evaluated Vi-IgG responses among enteric fever cases and the population sample in Nepal.

After a modest rise, Vi responses plateaued with no subsequent decay (rate = 0) ( Figure 4 , Table   S1 ). Median Vi IgG levels remained elevated above baseline for at least 32 months and did not vary significantly by age ( Figure 4B , p > 0.05). For the Nepal data, we calculated the ratio of 28day responses among culture-confirmed Typhi cases compared with the population-level mean for each antigen-isotype and found that the ratio was highest for HlyE IgG (median 30.9, IQR 16.1-54.6) and lowest for Vi IgG (median 1.5, IQR 1.0 -2.1) ( Figure 4C ).

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 21, 2021.

We defined suspected re-infections as cases with a 3-fold or higher increase for 2 or more antigen-isotypes occurring at least 3 months after their initial infection episode. Thirty-seven participants met this definition, yielding a re-infection incidence rate per 100-person years of 5. Observations including and after the suspected re-infection event were excluded from the longitudinal decay parameter estimation.

We used the joint distributions of the peak antibody titers, decay rate and decay shape estimated from enteric fever cases to estimate population-level incidence rates from cross-sectional data, extending previously published methods (10) . Because of the differences in the LPS sero- Figure 5 ). In Dhaka, Agogo and the KGH and AKU catchment areas, the incidence was higher in the youngest age group (<5 years); whereas in Kathmandu, incidence was highest among individuals 16 years and older.

Incidence estimates using both HlyE and LPS IgA and IgG are presented in the supplemental material (Table S2) .

We compared our population-based sero-incidence estimates to clinical blood culture-based incidence estimates derived from the same catchment area populations. We included both crude incidence (culture-confirmed cases divided by the catchment population and observation time) and adjusted incidence estimates (accounting for blood culture sensitivity and the proportion of . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint patients with typhoid-like illness who had blood culture at a surveillance site), following previously described methods (17, 22) . We focused comparisons among children under 5 years, as the sero-incidence will reflect recent infections, during the same time period as the clinical incidence study. The sero-incidence estimates followed the same rank order as the incidence estimates for each catchment area, with Dhaka, Bangladesh having the highest incidence, followed by KGH in Karachi, Pakistan; AKU in Karachi; Kathmandu, Nepal; Kavre, Nepal;

Agogo, Ghana ( Figure 6 ). Sero-incidence estimates were between 9 and 24 times higher than care-seeking adjusted clinical incidence rates (incidence ratio: 24. 

To estimate the sample size needed to consistently estimate incidence, we re-sampled each catchment area population at various sample sizes ranging from 50 to 600. The percent of resampled incidence estimates falling within the 95% CI of the full estimate (ie coverage) was 88% in Dhaka and 72% in Kavre with a sample size of 100; 100% in Dhaka nd 87% in Kavre with a sample size of 300 and 100% in Dhaka and 95% in Kavre with a sample size of 600 ( Figure S8 , panel A). The average half-width of the confidence interval relative to the incidence estimate fell below 25% at a sample size of 250 for Dhaka and 450 for Kavre ( Figure S8 , panel B).

We describe an approach to generate enteric fever incidence estimates from cross-sectional serosurveys. Leveraging longitudinal cohorts of enteric fever cases in four countries, we found IgA and IgG responses to HlyE and LPS were markedly increased for many months following infection, and these elevations were consistent across populations with varying infection pressures. We estimate the incidence of typhoidal Salmonella infection in the catchment areas of six study sites across four countries in which prospective blood culture-based surveillance was . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ;  undertaken, finding that the rank order of sero-incidence estimates tracked with clinical enteric fever incidence estimates.

We used estimates incorporating heterogeneity in the decay of antibody responses among enteric fever cases to calculate the incidence of infection. Rather than traditional sero-epidemiological methods that use cut-offs to determine whether a person has been recently infected, we leveraged the quantitative, longitudinal measurements of the antibody responses, integrating IgA and IgG isotype data, to generate population-level incidence estimates. A strength of this approach is that we can incorporate several types of uncertainty: measurement error, biologic noise and interindividual variation in sero-responses. While seroresponse curves showed large individual variation among patients, we found the median fitted peak antibody responses were consistent across the four different countries. There was some variation in the decay rates across countries, with Bangladesh having the slowest decay rate and Ghana having the fastest decay rate. We identified two potential explanations for this difference. The first is that reinfections are more common in settings with a higher force of infection. While we attempted to capture reinfections in the longitudinal case data, some reinfections may have not been identified. Second, in settings with higher incidence, individuals with typhoid are more likely to have been previously exposed, and secondary antibody responses may wane more slowly.

Previous typhoid seroepidemiological studies have used IgG responses to the Vi polysaccharide capsule, which is the antigen used in typhoid conjugate vaccines. We found that the ratio of elevated antibody responses among clinical cases at 28 days compared to the population serosurvey participants was lowest for Vi IgG and highest for HlyE IgG. Whereas HlyE IgG responses were a median of 30fold greater at 28 days following fever onset in culture-confirmed typhoid cases compared with the population mean, Vi IgG responses were only a median of 1.5 times greater, with a quarter of typhoid cases having 28-day titers that fell below the population mean.The minimal difference between titers in acute cases relative to the population for Vi IgG implies that this marker has limited utility in distinguishing acute cases from prior exposures.

Furthermore, we found no increase in serologic responses to Vi by age, echoing an earlier study from Kathmandu (23) . A study from Fiji found anti-Vi IgG responses increase with age but identified no differences in age-response patterns between high and low incidence communities . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint (8) . Further, the use of Vi in WHO-recommended vaccines will preclude distinguishing natural infections from vaccine-derived immunity in settings where vaccines are utilized. Additionally, Vi is not present in S. Paratyphi A, limiting its utility for sero-surveillance in contexts where paratyphoid fever is prevalent. Together, these circumstances underscore the importance of using alternative antigens in assessing enteric fever seroincidence.

The seroincidence estimates for all catchment areas were substantially higher than populationbased clinical incidence estimates, even after adjusting for care-seeking and sensitivity of blood culture. A "high" incidence of clinical enteric fever has been previously defined as >100 cases per 100,000 person-years (24), whereas we estimated sero-incidence of > 4,000 per 100,000

person-years in all study areas, with rate ratios of seroincidence to clinical incidence ranging from 9 to 24. This likely implies a substantial incidence of asymptomatic or paucisymptomatic infections that remain subclinical, like those observed with other infectious diseases (6, (25) (26) (27) (28) .

One benefit of sero-incidence estimates is the lack of sensitivity to care-seeking behavior, healthcare access, and cultural differences in reporting illness; this may enable less biased estimates of the force of infection in a community. Given that incidence thresholds for vaccine introduction have been based on clinical incidence (29, 30) , further data relating seroincidence to clinical incidence and case fatality ratio across diverse settings will be useful to translate seroincidence estimates into metrics that can inform policy decisions.

Geostatistical models predict the highest incidence of typhoid occurs in LMICs, where clean water and sanitation infrastructures are poor (1, 31, 32) . However, many of these countries with predicted high incidence lack blood culture-based data on enteric fever. For those in which population-based incidence estimates are available, data are typically geographically and temporally sparse, due to the resources needed to initiate and sustain surveillance. Serosurveys represent a potential scalable complement to blood-culture based surveillance, in that they can be conducted in a shorter time with fewer resources, generating estimates for rural areas where culture-based data are typically lacking. We found that sample sizes of 200 to 400 individuals per age strata were sufficient to consistently estimate incidence depending on the burden of typhoid in the population, with higher burden settings requiring smaller sample sizes. Our studies . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. in South Asia were performed using finger stick dried blood spots, which is more easily implementable than venous phlebotomy in field settings. Alternatively, banked serum and existing sero-surveys, such as Demographic and Health Surveys (33) could be leveraged to estimate the incidence of typhoidal Salmonellosis. Multipathogen sero-surveillance approaches are increasingly emerging as efficient means for population-based surveillance of vaccination coverage and infectious diseases of public health importance.

In South Asia, 10-20% of enteric fever cases are caused by S. Paratyphi A, for which vaccines are not currently available. We found sero-responses for S. Typhi-LPS (containing O antigens 9 and 12) were slightly lower among S. Paratyphi A cases than among S. Typhi cases, which we expected given the S. Paratyphi A only shares O12 with S. Typhi A. Additionally, we observed elevated anti-LPS antibody responses among patients with invasive non-typhoidal Salmonella (iNTS) from Ghana. These cross-reactive responses are likely due to the shared O12 antigen, a trisaccharide repeat backbone that is identical among Salmonella groups A, B, and D (16, 34) .

As such, we used HlyE sero-responses, which did not differ between S. Typhi and S. Paratyphi A, for our primary incidence estimates. Our earlier studies have demonstrated high specificity of anti-HlyE antibody responses for typhoid or paratyphoid diagnosis, compared with other bacterial infections (11, 35) . We are working to identify S. Typhi-or S. Paratyphi-specific immune responses to enable serovar-specific seroincidence estimation.

The results of this study should be interpreted within the context of several limitations. We incorporated age-dependence in antibody responses among enteric fever cases by fitting separate longitudinal age curves to each age strata. Methods to formally account for age-dependence in the longitudinal decay curves require further development. We assumed that the antibody kinetics were similar for all cases, but it is possible that asymptomatic individuals have different antibody kinetics. If the peak antibody responses are lower and the decay more rapid for asymptomatic cases, our method would underestimate the true incidence rate. In regions with high transmission intensity, individuals may be frequently re-exposed to Salmonella Typhi/Paratyphi, and the shape and parameters of antibody kinetics from secondary and tertiary exposures will likely be different from primary infections (36) . With our age-stratified approach, . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint we accounted for some of the differences between primary and secondary exposures, in that our parameter estimates for seroresponse kinetics among older children likely reflect multiple exposures. Longitudinal studies with serial blood culture are needed to accurately describe the kinetics of true re-infections. Long-term carriage of typhoidal Salmonella may alter the longitudinal serologic responses and is an area of future exploration. We found antibody kinetics did not significantly differ between hospitalized and non-hospitalized enteric fever cases, implying that, among culture-confirmed cases who sought care at study facilities, sero-responses were not dependent on clinical disease severity. Whether individuals who have asymptomatic infections or who receive care at lower acuity facilities (e.g. pharmacies) have similar antibody responses could not be determined from this study. While we know typhoid transmission to be seasonal in some environments, our modeling approach assumes a constant force of infection throughout the year and therefore averages the incidence rate across the year. Finally, while serosurveys may provide an efficient means for incidence estimation in resource-constrained settings, they do not obviate the need for blood culture surveillance, which is the only method for monitoring antimicrobial resistance--a serious and growing threat to the effective treatment of typhoid.

The WHO recommends that countries with high typhoid incidence introduce TCVs in national programs, resulting in an urgent need for typhoidal Salmonella incidence data to make informed decisions regarding vaccine introduction. Clinical surveillance for enteric fever is limited to settings with facilities equipped to perform blood cultures; even when available, incidence estimates may be highly sensitive to biases in care-seeking behaviors, antibiotic use and variable diagnostic sensitivity (22) . We describe a sero-surveillance tool that can efficiently generate population-level typhoidal Salmonella incidence estimates. This approach has the potential to expand the geographic scope of typhoid surveillance, generate much-needed subnational data on its burden and yield incidence estimates that are comparable across geographic regions and time.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint

The Sero-epidemiology and Environmental Surveillance (SEES) Study enrolled cohorts of blood culture-confirmed enteric fever cases from healthcare facilities and a population-based sample of participants from household surveys. We enrolled blood culture confirmed enteric fever cases from the Surveillance for Enteric Fever in Asia Project (SEAP) (17) and collected longitudinal blood samples over two years of follow-up. Concurrently, we conducted population-based serosurveys in the catchment areas of each of the clinical surveillance sites. We also obtained blood samples from the Surveillance for Enteric Fever in Africa (SETA) -Ghana Study (18), including longitudinal blood samples from enteric fever patients and cross-sectional samples from neighborhood controls of enteric fever cases. We characterized the dynamics of antibody responses to HlyE and LPS among cases across countries and age groups, and then applied a model using these kinetics to cross-sectional data to estimate the sero-incidence of infection in the population.

Enteric fever cases: All blood culture-positive enteric fever cases enrolled through SEAP were eligible to participate in the ancillary serologic study. Cases were enrolled from five SEAP (17) . For prospective cases, we collected plasma at the time of enrollment into SEAP and capillary blood collected on filter paper (i.e., dried blood spot; DBS) at 28 days, 3 months, 6 months, 12 months and 18 months post-enrollment. In Bangladesh, the majority of cases were followed up to 6 months post-infection and a subset of 170 cases were followed up for longer. We also enrolled cases retrospectively using stored baseline plasma, when available, and collected DBS at the time of enrollment and then for additional scheduled visits following . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

Capillary blood samples were collected on TropBioTM filter papers (Cellabs Pty Ltd., Brookvale, New South Wales, Australia), air-dried for at least two hours at room temperature, then stored with desiccant in individual plastic bags at -20 C until processing. All plasma was stored at ≤ -70 C until processing. SEAP samples were analyzed at the SEAP study laboratories in each country; SETA and North American samples were analyzed at Massachusetts General Hospital. We used kinetic enzyme-linked immunosorbent assays (ELISAs) to quantify antibody . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint levels present in plasma and eluted DBS samples. For the DBS samples, we cut two filled, filter paper protrusions and submerged them in 133 L of 1XPBS 0.05% Tween buffer overnight at 4 C and eluates were recovered after centrifugation. The eluate was used immediately or stored at 4 C for up to 2 weeks. We coated plates with S. Typhi lipopolysaccharide (LPS, containing serogroup antigens O9 and O12) (2.5 µg/mL), purified HlyE (1 µg/mL) or Vi antigen (Sanofi Pasteur; 2 µg/mL) as previously described (11, 38) . Plasma was added to the plate in duplicate at the following dilutions for each antigen and antibody isotype (IgA, IgG, respectively): LPS 

We fit a two-phase model of the sero-response to each antigen and isotype to describe kinetics of the antibody response to typhoidal Salmonella infection (39) (40) (41) . Briefly, we model an exponential rise in antibody level followed by a non-exponential (power function) decay episode.

The power function decay allows for rapid initial decrease followed by a period of sustained low antibody concentration. Additional details of the model have been published previously (39, 40) .

We used a Bayesian hierarchical framework to fit the sero-response model to individual data, defining joint parameter distributions for each antigen-isotype combination and obtaining predictive posterior samples using Markov chain Monte Carlo (MCMC) sampling (40) for baseline (y0) and peak antibody responses (y1), time to peak (t1), and decay rate (α) and decay shape (r). The models were run in JAGS (42) , using the rjags package (43) . We estimated the time course of antibody response according to three age strata at time of infection: <5 years, [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] years, >15 years. The age strata were chosen to compare the sero-incidence estimates to published clinical incidence estimates from the same catchment areas. We report the median and . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint 95% credible interval (CrI) from each posterior distribution. To formally compare differences in the distribution of modeled parameters across countries and age groups we subtracted the mcmc distributions and calculated the proportion of the distribution of differences that was greater or less than zero.

When following individuals over time, re-infections are possible and increase in likelihood with longer duration follow-up. We defined suspected re-infections as the occurrence of a 3-fold or higher increase in quantitative antibody response in individuals between visits at least 3 months from fever onset for 2 or more antigen-isotype combinations, unless the absolute value of the difference between measurements was less than one. The 3-fold threshold was derived by calculating the median fold change from baseline to 28 days among blood culture confirmed cases. For antibody decay estimates, we excluded longitudinal observations from individuals who met the definition of a re-infection from the time of the identified re-infection and all remaining follow-up observations. We calculated the incidence of suspected re-infection by dividing the total number of possible re-infections identified by the sum of the observed person time in each country; we used bootstrap resampling to calculate 95% confidence intervals.

To estimate incidence, we used the method outlined previously (19) . Briefly, we assumed that incident infections in the study sample occurred at random as a Poisson process with rate ë and created a likelihood function for the observed cross-sectional population sample values based on the longitudinal kinetics following infection. We used this model to generate maximum likelihood profiles for ë. We estimated ë with each antigen (HlyE, LPS) and isotype (IgA, IgG) separately and also jointly estimated ë by combining their likelihood functions (44) . We estimated age-stratified incidence rates in the population using the seroresponse parameter distributions for each age strata.

We accounted for two sources of noise in the serologic responses: measurement noise of the assay (described by the coefficient of variation [CV] across replicates) and biologic noise (measured as background response to the antigen-isotype among never-exposed, negative . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint controls) as detailed in (19) . We calculated CVs from a panel of 100 negative and positive controls that were run in triplicate at each site. We estimated biologic noise using the North American control population described above. To calculate lower detection limits we conducted serial dilutions of each antigen isotype and chose a threshold at which the CV was greater than or equal to 30% and added two standard deviations (45) then normalized to the positive control.

Lower detection limits were estimated for each laboratory running the assays. Measurement and biological noise parameters and the limits of detection used for each site are detailed in Table S3 .

We used mixed effect models to adjust the standard errors to clustering by sampling unit for the population-based geographic samples (Bangladesh, Nepal, and Pakistan).

To determine a sufficient sample size to consistently characterize incidence, we resampled the catchment area population data and estimated the incidence using various sample sizes ranging from 50 to 600. We used bootstrapping to select 100 samples with replacement for each sample size and estimated the incidence in each sample. We then calculated the percent of resampled estimates falling within the 95% confidence intervals for the full-sample estimate and the halfwidth of the confidence intervals relative to the incidence estimates.

For SEAP/SEES, we obtained written informed consent from all eligible participants and the parents/guardians of participants aged <18 years prior to collection of blood samples and completion of the questionnaire. We also obtained written assent from children aged 15 S8 . Relationship between serosurvey size and estimated seroincidence for a high (Dhaka) and more moderate (Kavre) incidence community. Table S1 : Model parameter estimates for antibody kinetics by antigen, immunoglobulin isotype and age group. Table S2 : Estimates of typhoidal Salmonella infection from cross-sectional serological data by study community, age group, and antigen/isotype combinations used in model. Table S3 : Measurement error, biologic noise and lower censoring limits for each antigen-isotype . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint Figures   Fig. 1 . Age trends in serologic responses to HlyE and LPS in cross-sectional surveys from six communities. IgA and IgG responses to HlyE and LPS are compared across age strata and catchment area populations. Boxes indicate the median and interquartile range. Antibody responses were measured using kinetic enzyme-linked immunosorbent assays (ELISAs). Asterisks denote levels of statistical significance for differences across age strata using the Kruskal-Wallis test.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) (C) . Panel (C) restricts to cases 5-15 years old because S. Paratyphi A cases were older than S. Typhi cases.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; Fig. 3 . Antibody kinetics, peak response and decay rate among blood culture-confirmed enteric fever cases by study country. (A) Median longitudinal antibody decay profiles fit to ELISA-measured antibody responses in each study country. (B) The distribution of modelpredicted peak antibody responses in each study country. (C) Boxplots and distributions of model-predicted antibody decay rates across study countries. All comparisons are among children 5 to 15 years to account for the different age-distribution of cases across countries.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint Fig. 5 . Estimated sero-incidence of typhoidal Salmonella by study community and age group. Age groups are denoted by point shapes for the median, with lines indicating the 95% credible interval. Boxes reflect the height of the median estimate for the overall population-based serosurvey. Fig. 6 . Comparison of estimates for crude and adjusted clinical enteric fever incidence with typhoidal Salmonella seroincidence. Crude incidence reflects culture-confirmed S. Typhi and S. Paratyphi A cases divided by the catchment population and time, while adjusted incidence accounts for imperfect sensitivity of blood culture and the proportion of acute febrile illnesses captured by the surveillance system. The horizontal axis indicates incidence, and scale differs for type of estimate. Estimates are shown for children under 5 years of age, for the serological estimates to coincide with the period of clinical surveillance.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 21, 2021. ; https://doi.org/10.1101/2021.10.20.21265277 doi: medRxiv preprint

The global burden of typhoid and paratyphoid fevers: a systematic analysis for the Global Burden of Disease Study

The Relationship Between Blood Sample Volume and Diagnostic Sensitivity of Blood Culture for Typhoid and Paratyphoid Fever: A Systematic Review and Meta-Analysis

Seroprevalence of antibodies against Chlamydia trachomatis and enteropathogens and distance to the nearest water source among young children in the Amhara Region of Ethiopia

Enteropathogen antibody dynamics and force of infection among children in low-resource settings

Reconstruction of antibody dynamics and infection histories to evaluate dengue risk

The incidence of Bordetella pertussis infections estimated in the population from a combination of serological surveys

Seroprevalence of SARS-CoV-2 in Bhubaneswar, India: findings from three rounds of community surveys

A cross-sectional seroepidemiological survey of typhoid fever in Fiji

Immunoproteomic Analysis of Antibody in Lymphocyte Supernatant in Patients with Typhoid Fever in Bangladesh

Characterization of Anti-Salmonella enterica Serotype Typhi Antibody Responses in Bacteremic Bangladeshi Patients by an Immunoaffinity Proteomics-Based Technology

Plasma Immunoglobulin A Responses Against 2 Salmonella Typhi Antigens Identify Patients With Typhoid Fever

A novel PhoP-regulated locus encoding the cytolysin ClyA and the secreted invasin TaiA of Salmonella enterica serovar Typhi is involved in virulence

The Salmonella Typhi hlyE gene plays a role in invasion of cultured epithelial cells and its functional transfer to S. Typhimurium promotes deep organ infection in mice

Characterization of a Pore-Forming Cytotoxin Expressed by Salmonella enterica Serovars Typhi and Paratyphi A

Comparison of genome degradation in Paratyphi A and Typhi, human-restricted serovars of Salmonella enterica that cause typhoid

Relationships among the rfb regions of Salmonella serovars A, B, and D

Incidence of Typhoid and Paratyphoid Fever in Bangladesh, Nepal and Pakistan: Results of the Surveillance of Enteric Fever in Asia Project

The Severe Typhoid Fever in Africa Program: Study Design and Methodology to Assess Disease Severity, Host Immunity, and Carriage Associated With Invasive Salmonellosis

Estimation of seroconversion rates for infectious diseases: Effects of age and noise

Development of ELISAs for diagnosis of acute typhoid fever in Nigerian children

Specific and cross-reactive immune response to oral Salmonella Typhi Ty21a and parenteral Vi capsular polysaccharide typhoid vaccines administered concomitantly

Integrating Facility-Based Surveillance With Healthcare Utilization Surveys to Estimate Enteric Fever Incidence: Methods and Challenges

Kinetics of the Natural

The global burden of typhoid fever

Immune response-eliciting exposure to Campylobacter vastly exceeds the incidence of clinically overt campylobacteriosis but is associated with similar risk factors: A nationwide serosurvey in the Netherlands

Estimation of incidences of infectious diseases based on antibody measurements

Comparison of seroprevalence of SARS-CoV-2 infections with cumulative and imputed COVID-19 cases: Systematic review

Global seroprevalence of SARS-CoV-2 antibodies: A systematic review and meta-analysis

Comparison of Strategies and Incidence Thresholds for Vi Conjugate Vaccines Against Typhoid Fever: A Cost-effectiveness Modeling Study

Cost-effectiveness of routine and campaign use of typhoid Vi-conjugate vaccine in Gavi-eligible countries: a modelling study

The burden of typhoid fever in low-and middle-income countries: A meta-regression approach

Burden of typhoid fever in low-income and middle-income countries: a systematic, literature-based update with risk-factor adjustment

The DHS Program -Quality information to plan, monitor and improve population, health, and nutrition programs

Live oral typhoid vaccine Ty21a induces crossreactive humoral immune responses against Salmonella enterica serovar Paratyphi A and S. Paratyphi B in humans

Evaluation of a Rapid Point-of-Care Multiplex Immunochromatographic Assay for the Diagnosis of Enteric Fever

Waning and boosting: on the dynamics of immune status

Celiac Disease Genomic, Environmental, Microbiome, and Metabolomic (CDGEMM) Study Design: Approach to the Future of Personalized Prevention of Celiac Disease

Identification of immunogenic Salmonella enterica serotype Typhi antigens expressed in chronic biliary carriers of S

A two-phase within-host model for immune response and its application to serological profiles of pertussis

Linking the seroresponse to infection to within-host heterogeneity in antibody production

Kinetics of the IgG antibody response to pertussis toxin after infection with B. pertussis

JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling, Working Papers

Package 'rjags

Biomarker dynamics: estimating infection rates from serological data

EP17A2 | Evaluation of Detection Capability for Clinical Laboratory Measurement Procedures

Alexander Yu and the SEAP Nepal and SEAP Pakistan teams. Funding: Bill and Melinda Gates Foundation

Author contributions: Conceptualization

We acknowledge the contributions of Shampa Saha, Sultana Aflatun Rubana,

We declare no competing interests.Data and materials availability: De-identified data and all analysis code will be made available on github.