Environmental Exposures Associated with Enteropathogen Infection in Six-Month-Old Children Enrolled in the ECoMiD Cohort along a Rural–Urban Gradient in Northern Ecuador

Abstract

Enteropathogens are major contributors to mortality and morbidity, particularly in settings with limited access to water, sanitation, and hygiene infrastructure. To assess transmission pathways associated with enteropathogen infection, we measured household environmental conditions and assayed 22 enteropathogens using TaqMan Array Cards in stool samples from 276 six-month-old children living in communities along a rural–urban gradient in Northern Ecuador. We utilized multivariable models, risk factor importance, and distance-based statistical methods to test factors associated with infection. Most children (89%) carried at least one pathogen, and 72% carried two or more. Bacterial infections (82% of participants) were more common than viruses (58%) or parasites (9.1%). Infants living in the urban site had decreased infection risks compared to those in rural locations. Improved water and sanitation were most predictive of reduced infection risk. Improved water was associated with decreased enterotoxigenic E. coli prevalence, and improved sanitation was associated with lower prevalence of any infection and specifically norovirus. Animal exposure was associated with increased Salmonella prevalence. Children measured during the rainy season had fewer viral and more bacterial infections. Identifying environmental exposures associated with specific pathogen outcomes provides insights into transmission pathways, which contribute critical information for developing effective strategies to improve child health.

Short abstract

Children living in communities in Northern Ecuador experience high rates of enteropathogen infection. Fewer infections were observed in urban compared to rural areas and in households with better water and sanitation conditions.

Introduction

Enteropathogens encompass an array of bacterial, viral, and parasitic agents of infectious diarrhea, which are responsible for 1.6 million deaths annually.¹ Diarrhea disproportionately affects young children in low- and middle-income countries (LMICs), with limited access to water, sanitation, and hygiene (WASH) resources.² Even absent diarrhea, persistent and repeated enteric infections in early life can lead to growth faltering, cognitive deficits, and reduced vaccine efficacy.³⁻⁸

Enteropathogens are transmitted via multiple fecal-oral pathways, as described by the F-diagram, which summarizes environmental reservoirs of fecal contamination (fluids, fields, flies, fingers, fomites, and food).⁹ While enteropathogens can spread person-to-person, they are primarily environmentally mediated,¹⁰ persisting in water and food. Many bacterial and some parasitic pathogens can grow outside the host,^2,11 and interdependent transmission pathways make it difficult to identify specific drivers of infection. Additionally, the relative importance of transmission routes varies by the ecology of the specific pathogen and environmental conditions,¹¹⁻¹³ necessitating pathogen-specific data to understand the role of each pathway in infection.

WASH infrastructure improvements and promotion of WASH-related behaviors (e.g., handwashing, sanitary food preparation) aim to interrupt transmission by reducing enteropathogen concentrations in the environment and human exposures. However, results of WASH intervention studies have been mixed, where several recent trials have found null or negative effects on enteric disease-related health outcomes in young children.^7,14−18 This may be because WASH interventions have not adequately addressed the multiple environmental reservoirs and transmission pathways for enteropathogens.^19,20 There is a need to understand how specific environmental conditions impact enteropathogen infection and related child health outcomes in LMIC settings, using study designs that assess the relative importance of various transmission pathways for specific pathogens.

Many studies have focused on a single or small number of enteropathogens.^e.g.,21,22 Recent molecular technologies allow for the simultaneous detection of multiple enteropathogens in human stool,^23,24 providing a comprehensive characterization of enteropathogen burdens, including asymptomatic infections and coinfections, in various settings.²⁵⁻²⁹ Multi-country data on enteric infections in LMIC settings show that pathogen prevalence and diversity vary by geography, enteropathogen type, and species.^30,31 Our previous work in Ecuador^e.g.,21,32 and studies elsewhere^25,33 indicate that pathogen carriage rates also differ between urban and rural settings within a country. Enteric disease risks in LMIC settings are not homogeneous,³⁴ and differences may be related to variation in environmental exposures between settings.

In this study, we examine environmental exposures associated with pathogen-specific enteric infection in six-month-old infants across a rural-urban gradient in Northern Ecuador. We focus on six-month-olds, because interactions with the environment become more frequent at this age, increasing child exposures to transmission pathways. This is also a critical period for growth and development. We investigate household exposures, including drinking water source and quality, sanitation infrastructure, contamination of child and caregiver hands, building materials, exposure to animals, and household crowding. We also consider the context of urbanicity and season because these factors influence household-level exposures. Examining associations between household exposures and pathogen-specific infection can inform our understanding of the role of the environment in child enteropathogen disease burdens, and aid in the development of evidence-based WASH interventions that address multiple modes of environmental enteropathogen transmission.

Methods

Study Design

We investigated environmental exposures associated with enteric infection in six-month-olds enrolled in the ECoMiD (‘Enteropatógenos, Crecimiento, Microbioma, y Diarrea’) prospective birth cohort study.³⁵ Recruitment and enrollment are described in the Supporting Information (SI) and elsewhere.³⁵ Participants were recruited from ten communities spanning a rural-urban gradient, ensuring a range of environmental conditions. Study communities were grouped into four prespecified³⁵ categories: (i) the city of Esmeraldas, the provincial capital (population ∼162,000; “urban” site), (ii) the town of Borbón, the commercial center of a politically defined rural region (population ∼5,000; “intermediate” site), (iii) rural villages accessible by road near Borbón, including Timbiré, Selva Alegre, Colon Eloy and Maldonado (populations ∼500–1000 each; “rural road” communities), and (iv) rural villages only accessible by river or with limited car accessibility, including Santo Domingo, Zancudo, Colon de Onzole, and San Francisco (populations ∼200–700 each; “rural river” communities) (Figure 1). Detailed information about ECoMiD, including recruitment and study protocols, has been published previously.

Figure 1.

Map of the study area. Participants were recruited from an urban community (city of Esmeraldas), intermediate community (town of Borbón), and rural communities accessible by road (Maldonado, Timbiré, Selva Alegre, Colon Eloy) and by river (Zancudo, Santo Domingo, Colon de Onzole, San Francisco).

Since 2019, 521 participants have been enrolled in ECoMiD. TaqMan Array Card (TAC) analysis of stool samples was planned for first phase enrollees (n = 472). Of these, 81 had incomplete visits due to the COVID-19 shutdown, 50 had a stool sample collected outside the target window (±3 weeks from their six-month birthday), 6 withdrew and 4 died prior to the six-month visit, 20 moved away, 18 were unavailable (e.g., family traveling, could not contact mother), 3 caregivers refused, 5 had field team errors, and 9 had lab errors. We therefore include 276 six-month-olds in the present analysis. Participants included were not significantly different than those excluded across any exposures or covariates considered (Table S1).

The study protocol was approved by the institutional review boards of the University of Washington (UW; STUDY00014270), Emory University (IRB00101202), the University of California, San Francisco (21–33932), the Universidad San Francisco de Quito (USFQ; 2018–022M), and the Ecuadorian Ministry of Health (MSPCURI000253–4).

Exposure Variables and Covariates

Trained fieldworkers surveyed mothers or caregivers about mother and child demographics, household socioeconomic status and WASH practices and infrastructure, food security, animal exposure (based on the FECEZ index;³⁶ details provided in (SI Methods), breastfeeding, and other factors. Fieldworker observations were used to gather information about household hygiene practices and construction materials. Child vaccine cards were observed to determine receipt of four-month vaccines (most children had not yet received their six-month vaccines) and information about delivery mode (vaginal/cesarean). E. coli in drinking water and maternal and child hand rinses were measured using both Petrifilms (3M, St. Paul, Minnesota) and the Colilert presence-absence test (IDEXX laboratories, Westbrook, Maine) (see SI Methods).

Environmental exposures considered encompassed: household crowding (number of people in household/number of rooms), flooring material, community type, exposure to animals, improved/unimproved water and sanitation, E. coli in drinking water and on mother and child hands, household hygiene, and rainy/dry season. Covariates measured were: household asset score (a proxy for socioeconomic status), maternal ethnicity, age, and education, and child sex, birth order, birth mode, breastfeeding in the past 14 days, and receipt of four-month vaccines. Detailed information on exposure and covariate timing, calculations, and imputation of missing data is provided in the SI (Table S2, Table S3, and SI Methods).

Sample Collection and Nucleic Acid Extraction

Child stool samples were collected at age six-months (182 ± 21 days). Caregivers collected stool in sterile containers, which were refrigerated or stored in a study-provided cooler. Stool was collected and processed by fieldworkers within 1 h if unrefrigerated or within 3 h if refrigerated. In the field lab, stool samples were aliquoted and stored at −196 °C in a liquid nitrogen tank as previously described.³⁵ Stool aliquots were transported to USFQ in liquid nitrogen and stored at −80 °C. At USFQ, total nucleic acids (TNA) were extracted from frozen stool using the QIAamp Fast DNA Stool Mini Kit (QIAGEN, Germantown, MD) with modifications (see SI Methods). Frozen TNA samples were shipped from USFQ to UW using Techni-Ice (Frankston, Australia) to maintain cold chain, inspected upon receipt to confirm that they had arrived l frozen, and immediately stored at −80 °C.

TaqMan Array Cards

At UW, TNA samples were analyzed for bacterial, viral, parasite, and antimicrobial resistance gene targets using TAC. Controls were PhHV (viral DNA control seeded during extraction), MS2 (bacteriophage RNA control seeded during extraction), and pan E. coli (uidA; endogenous control expected to be present in stool samples). Primers, probes, and their sources are described in Table S4. Because this study focused on enteropathogens, SARS-CoV-2 and antimicrobial resistance gene target results were not included.

Positive control plasmids (PCPs) containing the primer and probe regions of all targets (Table S5) were ordered from GeneArt Gene Synthesis (ThermoFisher Scientific) and Genewiz PriorityGENE (Azenta Life Sciences, South Plainfield, NJ). PCPs were amplified via transformation and an RNA PCP was created for the RNA targets (see SI Methods).

Samples or PCPs (20 μL) were combined with AgPath-ID One-Step RT-PCR master mix (50 μL), AgPath-ID One-Step RT-PCR enzyme (4 μL), and nuclease-free water (26 μL). After addition of template, 100 μL of reaction mixture was loaded, TAC cards were centrifuged twice at 1200 × g for 1 min, sealed, and cycled using the QuantStudio 7 Flex instrument (ThermoFisher Scientific) with the following conditions: 45 °C 20 min, 95 °C 10 min, 40 cycles of 95 °C 15 s and 60 °C 1 min. A no template control (NTC) was included on each card. Treatment of the NTCs is described in the SI.

The presence or absence of each target was determined using the automatic baseline and manual adjustment of assay thresholds within the QuantStudio Real-Time PCR Software (v.1.7.2, ThermoFisher Scientific). Targets were considered present if they amplified with a CT value <35. Assay performance was confirmed using seven aggregate standard curves with PCP serial dilutions from 10⁷ to 10¹ copies (Table S6). These serial dilution cards were interspersed throughout the testing period to confirm ongoing performance of the assay. Once results were obtained for individual TAC targets, presence or absence of each enteropathogen was determined using the criteria in Table S7.

Statistical Analyses

We adopted multiple statistical approaches to address challenges inherent to environmental epidemiology, and the WASH field in particular, where multiple exposures often correlate. We used traditional multivariable linear regression models for single exposure models alongside statistical methods for variable selection (ridge penalized regression) and variable reduction (nonparametric multidimensional scaling (NMDS)) to assess all exposures of interest simultaneously.

We summarized enteropathogen outcomes as (i) any infection, (ii) coinfection, (iii) number of pathogens, and infection with any (iv) bacteria, (v) virus, or (vi) parasite. We also tested associations between environmental exposures and the presence/absence of individual pathogens. Outcome data were binary for the multivariable models and ridge regression, except for number of pathogens, which was treated as a count outcome. For the NMDS, we used binary presence/absence data for individual pathogens to create distance-based ordinations of infection profiles. Analyses were conducted in R v.4.2.3;³⁶ code is available on Github: https://github.com/kjojess/ECoMiD-6m-TAC-environment-manuscript.

Multivariable Regression Models

Each environmental exposure was assessed as a single exposure variable with each outcome in separate models due to collinearity of environmental variables (Figure S1). Thus, individual exposure estimates are not independent of other exposures and may include effects along shared pathways. Parameters of interest were risk ratios (RR) or prevalence ratios (PR) for binomial outcomes, depending on the timing of exposure collection (Table S2), and count ratio for the pathogen count. We fit modified Poisson models through generalized estimating equations (GEE) with robust standard errors using the R package geepack.^38,39,40 Potential confounders were selected a priori as contextually important covariates that may be associated with both the exposure and outcome. We adjusted for potential confounders: asset score, maternal education, child sex, birth mode, breastfeeding in the past 14 days, and receipt of four-month vaccines (Table S8). Prior to model fitting, we also prescreened a set of variables potentially associated with infection: child age in days, maternal age, birth order, maternal ethnicity, and household food security, and included any variables associated with the outcome (p < 0.2; Table S9). We also adjusted for community type and season, except when community type or season were considered as exposures (Table S8). Finally, we assessed collinearity between exposures and sequentially removed covariates with a variance inflation factor (VIF) > 10 to achieve final model form. Results for models where community type was dropped due to VIF > 10 were not considered. For individual pathogen infection outcomes, we did not estimate models for pathogens with prevalence <5%. Though we focus on raw p-values for the adjusted models in the results and discussion, we also applied a Benjamini-Hochberg (BH) correction to control for false discovery (Table S10).

For outcomes associated with improved toilet or drinking water, we performed a secondary analysis to assess the type of toilet (basic, limited, unimproved, or open defecation) or water source (rainwater, public tap, protected well or tubewell, and piped water connection, or surface water) associated with infection. Models were run as described above for the primary analysis.

Risk Factor Importance

We used ridge penalized regression, which can handle multicollinearity and prioritize variables among a set that are minimally associated with the outcome,⁴¹ to identify the most important exposures for predicting each infection outcome. Following variable standardization, ridge regressions were run on all exposures together using the R package glmnet,⁴² with the penalty (λ) selected through cross-validation to minimize mean-squared error. Variable importance was ascertained using the coefficient values at the optimal λ value.

NMDS Ordinations

We used NMDS, a dimension reduction technique, to visualize differences between children based on the profile of enteropathogens detected. In the R package vegan,⁴³ we used the Jaccard similarity index to calculate distance matrices based on individual pathogen presence/absence. Separate NMDS ordinations were calculated for the outcomes any infection (binary presence/absence of all pathogens), any bacteria (binary presence/absence of bacterial pathogens), and any virus (binary presence/absence of viral pathogens). We did not run NMDS for “any parasite” due to low prevalence of parasite infections. Exposure variables were fit to NMDS ordinations using the vegan function “envfit” with 10,000 permutations. This function is used to fit environmental variables onto an ordination to determine how well environmental factors explain variation in community composition data within an ordination space. Resulting p-values were adjusted using a BH correction.

Results

Household Environmental Conditions

Environmental conditions varied by community type (Figure 2, Table S11). Water: Nearly all households (98%) used an improved water source (tubewells, protected well or spring, public tap, piped water connection directly to the household, or rainwater⁴⁴) for drinking water. Unimproved drinking water sources (unprotected wells/springs or surface water from a river or stream) were reported in the intermediate site and the rural road communities, which had 4.5% and 1.1% unimproved water, respectively. E. coli was detected in household drinking water less often in the urban site (57% of households) than in rural areas (96% of households in the rural river communities). Sanitation: Most households (93%) had improved sanitation, defined as a shared or unshared toilet that discharged to a sewer system, septic tank, or pit latrine, a ventilated pit latrine, or a pit latrine with a slab.⁴⁵ Unimproved sanitation, which included a toilet that discharged to another place, a pit latrine without a slab, a bucket or hanging toilet, and open defecation, was most common in the rural river communities (28%). Hygiene: The prevalence of E. coli in maternal hand rinses was high across all communities (≥80%) and was slightly higher in the rural river communities. Detection of E. coli in child hand rinses was less common in the urban site (45% of households) compared to the rural sites (100% of households in rural river communities). The median hygiene score was 7.0 (IQR 5.0–8.0) across all communities, with a range of scores from 1 to 9 (9 indicating better hygiene).

Figure 2.

Exposures along the rural-urban gradient. Y-axes indicate the percent of households in each community with a given exposure (a-e); or the mean of continuous exposure scores across households with standard error bars (i and j).

Across all communities, 40 households (14%) were considered crowded (≥3 people per room), with slightly more crowding reported in the urban and intermediate sites. Most households had improved flooring types (ceramic tiles, polished wood or parquet, or cement floors; 67%), versus unimproved flooring (palm/bamboo, earth/sand, or wooden boards; 33%). Improved flooring was most common in the urban site (85%) and rural road communities (78%), compared to the intermediate site (49%) and rural river communities (36%). The median animal exposure score was 0.10 (maximum possible score 4.0; IQR: 0–0.45) across all communities. The highest median score (indicating greater exposure to animals and animal feces) was for rural road communities (0.29, IQR: 0–0.68), followed by rural river communities (0.16, IQR: 0–0.42), the intermediate site (0.13, IQR: 0–0.47), and the urban site (0, IQR: 0–0.31).

Enteropathogen Prevalence

Among 276 six-month-old children, most (89%) were positive for at least one pathogen and 72% were positive for multiple pathogens (Table S7). Bacterial pathogens were most common (82%), followed by viruses (58%), and parasites (9.1%) (Table S7 and Figure 3). Children residing in the urban site had the lowest overall prevalence of infection. The most common bacterial infections were enteroaggregative E. coli (EAEC; 63%), diffuse adherent E. coli (DAEC; 31%), and atypical enteropathogenic E. coli (aEPEC; 29%). Shigatoxigenic E. coli (STEC) was detected least frequently (0.7%). The most common viral infections were enterovirus (40%) and adenovirus (16%). Sapovirus was detected least frequently (2.9%). The most common parasite infections were Giardia (5.8%) and Cryptosporidium (2.2%). Cyclospora cayetanensis and Entamoeba histolytica were rarely detected (0.4% each).

Figure 3.

Prevalence of a) bacteria, b) virus, and c) parasite pathogens in n = 276 child stool samples by community type, where n = 67 in urban, n = 87 in intermediate, n = 97 in rural (road), and n = 25 in rural (river) communities.

Of the 22 pathogens analyzed, the median number detected per child was 3.0 (IQR: 1.0–4.0) (Table S7 and Figure 4). Of 10 bacterial pathogens, the median per child was 2.0 (maximum: 6.0). There were fewer bacterial pathogens per child in the urban community (median = 1.0) compared to the intermediate and rural communities (median = 2.0). Of six viral pathogens, the median per child was 1.0 (maximum: 4.0). As with bacteria, median viral pathogens per child was lower in the urban site (0.0) compared to the intermediate the rural communities (1.0). Of six parasites, the median number per child across all communities was 0.0 (maximum: 2.0).

Figure 4.

Average number of a) any, b) bacteria, c) virus, and d) parasite pathogens detected in n = 276 child stool samples by community type, where n = 67 in urban, n = 87 in intermediate, n = 97 in rural (road), and n = 25 in rural (river) communities. The total number of enteropathogens assayed was 22, with 10 bacteria, 6 viruses, and 6 parasites.

Environmental Exposures and Enteric Infection

Multivariable Models

Associations between environmental exposures and infection varied along the rural-urban gradient (Figures 5 and 6, Tables S6–S8). Here, we focus significant associations in the adjusted models. Compared to the urban site, the number of pathogens in child stools was 38–41% higher in the intermediate and rural sites. Children from the intermediate site also had a 28% increased risk of coinfection (95% CI: 2–60%), and children in the rural road communities had a 21% increased risk of any bacterial infection (95% CI: 2–43%) and were more than twice as likely to be infected with DAEC. Living in the rural river communities was associated with 43% increased risk of EAEC infection (95% CI: 1–104%).

Figure 5.

Covariate-adjusted (purple) and unadjusted (green) associations of environmental variables with overall pathogen infection outcomes. All models are for single exposure variables. Results for “Any parasite” are available in the Supporting Information as there were no significant associations with environmental variables. The x-axes are a risk ratio (crowding, improved flooring, improved water, community type), prevalence ratio (animal score, improved toilet, water stored, E. coli in drinking water, E. coli in child handwash, E. coli in adult handwash, hygiene score rainy), or count ratio (all pathogen no. models) depending on the outcome and timing of the exposure. Asterisks indicate significance at p < 0.05.

Figure 6.

Covariate-adjusted (purple) and unadjusted (green) associations of environmental variables with individual pathogen infection outcomes. All models are for single exposure variables. Results for pathogens where there were no significant associations with environmental variables are available in the Supporting Information. The x-axes are a risk ratio (crowding, improved flooring, improved water, community type), prevalence ratio (animal score, improved toilet, water stored, E. coli in drinking water, E. coli in child handwash, E. coli in adult handwash, hygiene score rainy), or count ratio (all pathogen no. models) depending on the outcome and timing of the exposure. Asterisks indicate significance at p < 0.05.

Household sanitation and water were also key determinants of infection. Having an improved toilet was associated with 9% lower prevalence of infection with any pathogen (95% confidence interval (CI): 2–17%) and 72% lower prevalence of norovirus infection (95% CI: 26–90%). The association between toilet type and norovirus infection was driven by households reporting open defecation; compared to households with improved sanitation, open defecation was associated with >6X increased norovirus prevalence (95% CI: 1.5–29), with no association for other sanitation types. Improved water was associated with 78% decreased risk of ETEC (95% CI: 34–93%) and 90% decreased risk of astrovirus (95% CI: 13–99%). Higher animal exposure scores were associated with >2X higher prevalence of Salmonella infection (95% CI: 1.19–8.2). Higher hygiene scores were associated with a 23% increase in norovirus prevalence (95% CI: 1–50%), and the presence of E. coli on child hands was associated with a 12% decrease in the prevalence of bacterial infections (1–23%). There were no significant associations between enteropathogen infection and household crowding, improved flooring, drinking water storage, or E. coli contamination of mother hands or drinking water. Similarly, there were no associations between environmental variables and the following infection outcomes: any parasite, aEPEC, typical enteropathogenic E. coli (tEPEC), enteroinvasive E. coli (EIEC)/Shigella, Campylobacter, enterovirus, rotavirus, and Giardia. The prevalence of any bacterial infection increased 13% in the rainy season (95% CI: 1–26%), as did infection with EAEC (21%; 95% CI: 1–46%). Prevalence of viral infections decreased by 26% (95% CI: 8–40%) in the rainy season. Results for these outcomes are presented in Figure S2.

Risk Factor Importance

Improved water and toilet were among the most important variables for predicting the lower prevalence of infection (Figures 7a and 7b). Exceptions were infection with any parasite, norovirus, and Giardia, where improved water was important but predicted increased infection risk. Improved flooring was an important variable for several outcomes and predicted decreased risk of any infection, coinfection, and any bacterial infection. Living in the intermediate and rural sites predicted an increased risk of infection compared to urban site, though with varying importance by outcome and opposite direction for specific pathogens (decreased risk of infection in the intermediate and/or rural sites for enterovirus, rotavirus, adenovirus, norovirus, and Giardia). Season had moderate to high importance for the presence of any pathogen, any bacteria, and any virus, but was less important than community type or improved water and sanitation.

Figure 7a.

Relative importance of environmental variables extracted from penalized regression for overall enteropathogen infection outcomes. Positive direction of association (yellow) indicates higher predicted prevalence of the outcome and negative direction of association (purple) indicates lower predicted prevalence of the outcome.

Figure 7b.

Relative importance of environmental variables extracted from penalized regression for individual pathogen infection outcomes. Positive direction of association (yellow) indicates higher predicted prevalence of the outcome and negative direction of association (purple) indicates lower predicted prevalence of the outcome.

Risk factor importance varied by outcome for the other environmental exposures assessed. Household crowding was important for coinfection and EAEC, adenovirus, and Giardia infection, and generally predicted decreased risk. Animal exposure was the most important variable for DAEC and predicted decreased infection risk. However, animal exposure was predicted to increase infection risk for several other pathogen-specific outcomes, including EAEC, tEPEC, and enteroinvasive E. coli (EIEC)/Shigella. Hygiene score and E. coli in drinking water and maternal and child handwashes were relatively unimportant in most models, though E. coli detection in child and maternal hand washes were important predictors for decreased risk of Giardia infection. Model results for aEPEC, EIEC/Shigella, Campylobacter, and Salmonella infection outcomes had variable importance scores near zero for all exposures (Table S12).

NMDS Ordinations

Two-dimensional NMDS visualizations (Figure 8) showed substantial overlap by community type and environmental variables for any infection, any bacterial infection, and any viral infection, suggesting that the overall profiles of the 22 pathogens assayed did not substantially differ across the variables tested. Vector fitting revealed that living in an urban community and animal exposure were correlated with infection with any bacterial pathogen (p = 0.03 each; Table S13). However, each of these environmental variables explained only 3% (R²=0.03) of variation.

Figure 8.

Nonparametric multidimensional scaling (NMDS) ordinations summarizing between-sample Jaccard similarity for enteropathogens in child stool. Comparisons were conducted using presence/absence data for any pathogen (a), any bacterial pathogen (b), and any viral pathogen (c). Ellipses indicate 95% CIs around data points for each community type.

Discussion

Understanding how enteric infection burdens vary by pathogen type and household environmental conditions along an urban-rural gradient within a single country contributes important information for developing effective strategies to improve child health. Our findings highlight links between the environment and enteropathogen infection in early childhood and provide a novel contribution by examining comprehensive and robust environmental characterization at multiple socioecological levels analyzed alongside multiple enteropathogen outcomes, across a unique urban-rural gradient. In our analysis of environmental exposures associated with enteric infection in six-month-old children in Ecuador, we found that infections were highly prevalent, environmental conditions varied by urbanicity, and environmental exposures influenced infection. Though results varied by infection outcome, living in the city of Esmeraldas and access to improved water and sanitation were generally associated with decreased infection risk. We assessed multiple etiologic agents of enteric infection, since disease burdens, pathogen ecology, environmental reservoirs, and modes of transmission associated with infection vary by the type of enteropathogen examined. Sanitation access was protective for norovirus infection, animal exposure was a risk for Salmonella infection, living in a rural location was a risk for EAEC and DAEC infection, and improved water was protective for ETEC and Astrovirus infection. Connecting pathogen-specific data with specific environmental exposures provides information for developing more targeted and efficient mitigation strategies by identifying likely transmission pathways for specific pathogens. While prevalence varied by community type and specific household conditions, NMDS analyses suggest that overall infection profiles—or the set of pathogens found in each child—did not vary substantially by location or environmental conditions.

Enteropathogen Prevalence

The overall pathogen detection rate for six-month-old children in our study was 89%, with 72% of children carrying multiple pathogens. These infection rates are higher than previously reported enteric infection prevalence in Ecuador^e.g.,21,32 and South America,^e.g.,46,47 but these estimates of enteric disease burdens have largely relied on culture-based or immunoassay approaches for pathogen detection. These more conventional pathogen detection methods are less sensitive than the culture-independent molecular methods like the TAC technology used in this study.⁴⁸ Infection estimates vary by participant age, study setting, and the pathogens evaluated. The Malnutrition and Enteric Disease (MAL-ED) multicenter longitudinal cohort study⁴⁹ also used TAC to assess enteropathogen carriage and infection prevalence across multiple ages and countries. We used MAL-ED TAC data (accessed via ClinEpiDB⁵⁰) to compare our findings to MAL-ED’s South American sites in Peru and Brazil. Infection prevalence in Peru was very similar to those observed in this study (any infection: 92%; coinfection: 71%), though lower infection rates were observed in Brazil (any infection: 75%; coinfection: 37%).

Across all MAL-ED sites, infection prevalence for six-month-old children was strikingly similar to this study when considering the same pathogens (any infection: 91%; coinfection: 72%). Bacterial and viral infection prevalence in this study and MAL-ED were also similar, though rates of parasite infection were higher in MAL-ED than our study (16% versus 9.1%). Other studies from resource-limited settings that used TAC to detect pathogen infections in young children also reported comparable enteric infection prevalence. For example, the Safe Start trial found 89% enteropathogen and 67% coinfection prevalence in infants aged five months in a peri-urban community in Kenya.⁵¹ Other studies in Kenya²⁶ and Tanzania⁵² have reported 93% and 80% enteropathogen carriage in children with diarrhea.

Key findings in MAL-ED included between-country and pathogen-specific differences in enteric disease burdens (Figure S3). Here, we show that enteric infections can also vary substantially within a single country. However, the relationships between urbanicity and infection were different for MAL-ED than this study, depending on the specific pathogen examined. For example, MAL-ED found a higher prevalence of ETEC in urban compared to rural sites, while we observed the opposite trend. Similarly, we saw a much higher prevalence of aEPEC infection in the rural river communities than in the more populous sites, whereas MAL-ED reported highest incidences of aEPEC in urban areas. Other interesting pathogen-specific differences include higher incidence of Campylobacter, adenovirus, astrovirus, norovirus, and sapovirus infections in MAL-ED vs this study.

Environmental Exposures and Infection

Urbanicity and Improved Water and Sanitation

Children living in the urban city of Esmeraldas had a lower risk of infection for nearly every infection outcome examined. Compared to the town of Borbón and the rural river and road communities, Esmeraldas also had more households with improved flooring and improved sanitation, were less likely to have E. coli contaminated hands and water and had lower animal exposure scores. Only a small number of participants had unimproved water and sanitation, but those who did were at high risk of infection. The same trend emerged when we considered all environmental exposures simultaneously in the variable importance analysis; improved water and sanitation were strongly predictive of decreased infection prevalence for several outcomes. However, these results should be interpreted with caution given the small sample sizes in the unimproved water and sanitation groups. Alongside water and sanitation, community type was important and likely captures unmeasured variation in household exposures across the rural-urban gradient. Unmeasured sources of variability along the gradient might include exposures to enteropathogens from food and community water sources.

The relatively lower infection rates in the city of Esmeraldas compared to the more rural communities differed from previous research in Ecuador that found higher rates of enteric infection in urban compared to rural areas.³² In particular, our previous studies in the same region found that travel to urban areas increased risk of infection with diarrheagenic E. coli (DEC)⁵³ and suggested that DEC strains migrate from urban toward rural populations.⁵⁴ We attributed these findings to increased transmission in urban settings, potentially from crowding and population density. However, our previous studies of enteropathogens in Ecuador have included samples from the capital city of Quito, while this work was limited to the single urban setting of Esmeraldas. We also did not see any association between household crowding and infection.

Cities may have better public health infrastructure and WASH access than rural areas resulting in lower overall enteropathogen prevalence in urban settings. This is likely the case here, where increased coverage of improved water and sanitation facilities in Esmeraldas, in addition to other potentially unmeasured environmental and/or population factors (e.g., nutrition and immune status), were protective for infection. Depending on the specific setting and outcomes examined, results of studies examining the association between urbanicity and/or population density and infection risk have been mixed, with some reporting higher disease burdens in rural or low density areas^55,56 and others reporting the opposite trend.^57,58

While we did observe lower infection prevalence in Esmeraldas, the overall child enteric infection profiles did not vary substantially by community type, according to distance-based NMDS analyses. There was no visual clustering of participants based on enteric infection profiles by community type, or for any other environmental exposure variables examined. Thus, while environmental factors may influence overall magnitude (i.e., prevalence), they may not affect the distribution of pathogens in circulation in the population.

Seasonality

We observed a clear association between season and infection prevalence. Children sampled during the rainy season (January-May) were more likely to have infections with any pathogen or bacterial pathogens but had decreased viral pathogen detection. There is a substantial body of evidence that elevated temperature and rainfall are associated with increased rates of enteric infection and diarrhea,^e.g.,59,60 and our results are consistent with previously observed studies documenting elevated rates of bacterial infections in hotter, wetter seasons compared to cooler, drier seasons.^61,62 The seasonal trends we observed were primarily driven by increased prevalence of EAEC and tEPEC in the rainy season and adenovirus and enterovirus in the dry season. Seasonality of these pathogens varies by study and context. Some studies have found increased prevalence of tEPEC, EAEC, and adenovirus in rainy weather,^e.g.,63,64 while others have reported opposite signals (i.e., viral infection prevalence increasing in dry weather,^65,66 as seen in this study) or a lack of seasonality.⁶⁴ The seasonal patterns reported here warrant further investigation into the relationships between enteric infection, meteorological conditions (i.e., rainfall and temperature), and perhaps behavioral and immune-related factors associated with seasonal variations in enteropathogen burdens in the region.

Animal Exposure

The animal exposure score, an indicator of child exposure to animals and their feces,³⁷ was associated with Salmonella infection in the multivariable analysis. Animals are ubiquitous in and around household environments in the study communities, particularly in the town of Borbón and the rural road and river sites, where poultry (mostly chickens) often roam freely, potentially contributing to Salmonella transmission. Poultry and poultry-related food items are well-characterized reservoirs for Salmonella and other enteropathogens.⁶⁷⁻⁶⁹ Animal exposure was not associated with or predictive of Campylobacter infection in our analyses, despite the fact that, like Salmonella, Campylobacter is frequently associated with poultry.⁶⁹⁻⁷¹ Animal exposure was of high to moderate importance for decreased predicted prevalence of some outcomes, specifically coinfections and infection with DAEC and Giardia. There are several reasons animal exposure might be protective for enteric infection, including improved nutritional status from animal food sources, better household hygiene practices from awareness of zoonotic transmission risk, and immune conditioning from animal contacts.⁷²⁻⁷⁴

E. coli Contamination and Hygiene Practices

E. coli in mother and child hand rinses and drinking water were predictive of decreased infection prevalence for any parasite and specifically for Giardia in the variable importance analysis. E. coli in child handwashes was also associated with decreased prevalence of bacterial infection in the multivariable models. These associations between E. coli detection and decreased disease risk were unexpected because contaminated water and hands are usually associated with increased risk of infection.^12,75 We were similarly surprised to find that higher hygiene scores predicted increased norovirus prevalence. This discrepancy may be related to the enteropathogen’s mode of transmission. For example, consumption of contaminated food may be more important than direct inoculation by contaminated water and hands for some pathogens. A household with an ill child may also more frequently engage in handwashing behaviors, leading to reverse causation between an infection and E. coli contamination, particularly if the exposure and outcome are measured concurrently.^76,77 There were also counterintuitive relationships between Giardia, norovirus, and any parasite infection outcomes and improved water, where these infections increased in households with improved water. This further suggests that the relationships between water, sanitation, and hygiene are complex.

Strengths and Limitations

This study benefited from comprehensive data on household environmental conditions and demographics from multiple communities along a rural-urban gradient in Northern Ecuador, as well as outcome data on 22 different enteropathogens for children at a common age. However, measuring 276 children at a single age limited our power for assessing more rare outcomes. Inclusion of a broader set of ages could capture additional infections and provide insights into what behaviors and developmental stages are associated with exposure pathways for specific pathogens.

Despite excluding a number of six-month-olds from the full ECoMiD cohort, we found that participants not included in the analysis were comparable to those included. We are confident that our methods capture a representative sample of the six-month-old infants in the town of Borbón and rural sites, but our sample group from the urban city of Esmeraldas may not be fully representative of the population due to the much larger population size and difficulties accessing some neighborhoods because of safety concerns.

There was collinearity and a lack of variability in some of the environmental exposure variables; water, sanitation, improved flooring, and detection of E. coli contamination of drinking water and on child and mother hands were colinear with community type along the rural-urban gradient. However, we used multiple analytical tools to overcome this and evaluated results holistically. While community type did covary with these exposures, it is likely that it also captures unmeasured variability in environmental conditions between communities. We included community type as a covariate for all models, which allowed us to control for this potentially unmeasured variability.

APPENDIX

ECoMiD Investigators

Principal Investigators/co-Investigators

Karen Levy, Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, 98195, USA

Joseph N.S. Eisenberg, Department of Epidemiology, University of Michigan, Ann Arbor, Michigan, 48109, USA

Gwenyth O. Lee, Rutgers Global Health Institute, Rutgers University, New Brunswick, New Jersey, 08901, USA

Gabriel Trueba, Instituto de Microbiología Colegio de Ciencias Biológicas y Ambientales, Universidad San Francisco de Quito, Quito, 170901, Ecuador

William Cevallos, Centro de Biomedicina, Universidad Central del Ecuador, Quito, 170136, Ecuador

Benjamin F. Arnold, F.I. Proctor Foundation and Department of Ophthalmology, University of California San Francisco, San Francisco, California, 94158, USA

Kostantinos T. Konstantinidis, School of Civil and Environmental Engineering and School of Biological Sciences, Georgia Institute of Technology, Atlanta, Georgia, 30332, USA

Lab Coordination and Analysis

Gabriela Vasco, Instituto de Microbiología Colegio de Ciencias Biológicas y Ambientales, Universidad San Francisco de Quito, Quito, 170901, Ecuador

Stuart Torres, Instituto de Microbiología Colegio de Ciencias Biológicas y Ambientales, Universidad San Francisco de Quito, Quito, 170901, Ecuador

Lesly Simbaña, Instituto de Microbiología Colegio de Ciencias Biológicas y Ambientales, Universidad San Francisco de Quito, Quito, 170901, Ecuador

Viviana Albán, Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, 98195, USA

Manuel Calvopiña, Escuela de Medicina, Universidad de las Americas, Quito, 170137, Ecuador

Christine S. Fagnant-Sperati, Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, 98195, USA

Nicolette Zhou, Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, 98195, USA

Kelsey J Jesser, Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, 98195, USA

Field Data Collection

Mauricio Ayovi, Instituto de Microbiología Colegio de Ciencias Biológicas y Ambientales, Universidad San Francisco de Quito, Quito, 170901, Ecuador

Adriana Lupero, Instituto de Microbiología Colegio de Ciencias Biológicas y Ambientales, Universidad San Francisco de Quito, Quito, 170901, Ecuador

Molly K. Miller-Petrie, Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, 98195, USA

Data Management

Jesse Contreras, Department of Epidemiology, University of Michigan, Ann Arbor, Michigan, 48109, USA

Jessica Uruchima, Department of Epidemiology, University of Michigan, Ann Arbor, Michigan, 48109, USA

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.4c07753.

• Recruitment and enrollment; environmental exposure variables and covariates; lab methods; comparison, processing, binning, and descriptive statistics of exposure and covariate variables; multivariable model results; variable importance scores for ridge regression/variable importance analyses; hygiene score scale; TaqMan array card assay details; positive control plasmid sequences; pathogen classifications; covariates included in models; prescreening variable results; NMDS statistics; correlation matrix for environmental risk factor variables; associations of variables with pathogen infections; prevalence of pathogens from this study and MAL-ED (PDF)

Author Present Address

¹ Department of Population Health Sciences, Georgia State University, Atlanta, Georgia 30302, USA

Author Contributions

^# Co-first Authors: Kelsey J. Jesser and Nicolette A. Zhou contributed equally to this work.

The authors declare no competing financial interest.

Notes

^† This work was submitted by the authors on behalf of the ECoMiD Investigators.