A community-based cross-sectional study, including a questionnaire-based surveillance of the households and collection of child growth data, child urine, and fecal samples were performed in the Haramaya woreda (district), East Hararghe Zone, Oromia Region, Ethiopia. Beginning in 2018, Haramaya University (HU) initiated a full household enumeration of 12 kebeles (wards) within Haramaya woreda using methods developed for the Kersa Health and Demographic Surveillance System (26). For this cross-sectional study, five out of the 12 kebeles were selected in a way to capture the spatial heterogeneities in the landscape (e.g., land use, land cover, altitude, agricultural practices) that could influence risk factors contributing to health outcomes.

Eligibility criteria of this study were designed to enroll children receiving traditional maternal care and without underlying health conditions that might affect the primary study outcomes. Previous studies (MAL-ED) have indicated that the prevalence of Campylobacter spp. in children in low- and middle-income countries (LMIC) increases linearly in the first year of life and then reaches a steady state or slow decline with peak levels occurring around 12 months of age (15). A child was included in the study if he/she was 11–13 months of age when consent was given, and if the child's mother was the caretaker. The child was excluded from the study if he/she presented visible congenital abnormality or had serious medical illnesses, or the child or his/her mother required extended stay in hospital after birth. Households with at least three chickens within the homestead (defined as the small collection of households that are physically connected to one another), willing to participate and conform to the requirements of the study were included. Families not residing in Haramaya woreda for at least 3 months, currently participating in another study on animal husbandry, or with a mother who did not live in Haramaya woreda when the child was born were excluded.

The sample size of this study considered estimation of the three primary outcomes: Campylobacter colonization, EED, and stunting. Based on a binomial distribution, a sample size of 100 for the whole study population allowed estimation of a 50% prevalence with precision of 10 at 95% confidence (R function binom:: binom.confint).

While adjusting for population weight of kebeles, 102 children (one child from each household) were randomly selected from the sampling frame of the selected five kebeles. Field work started on October 16, 2018 and continued for 10 weeks until December 21, 2018. The enrolled children and their mothers were invited to come to a local health post or administration office for anthropometric measurements, administration of a sugar solution, and collection of a urine sample for each child for later measurement of lactulose excretion (a marker for EED in urine), as well as completion of interviews and providing a fecal sample for each child.

Questionnaires were developed in collaboration with the Haramaya University (HU) field team using validated indices whenever possible, such as minimum dietary diversity of infant and young children (MDD-IYC). In consultation with social scientists and field workers, the study questionnaire was finalized so that it was culturally appropriate and locally adapted. In the local language (Afan Oromo), bilingual data collectors collected information on demographics, livelihoods, wealth, animal ownership, animal management and disease, water, sanitation and hygiene, health, and nutrition. Study data were collected and managed using REDCap electronic data capture tools hosted at the University of Florida Clinical and Translational Science Institute (27).

During the implementation of household surveillance, stool samples from the selected children were also collected aseptically in a sterile plastic sheet for EED and PCR analyses, and meta-total RNA sequencing (MeTRS, a culture-independent sequencing method that profiles the presence and quantity of RNA transcripts in a sample). Samples were flash-frozen in the field using liquid nitrogen and stored at −80°C until further use. Genomic DNA and RNA were extracted for conventional PCR and MeTRS, respectively (25). Details of laboratory procedures on PCR and MeTRS were described in Terefe et al. (25). For the dual sugar absorption test, an oral solution consisting of 200 mg L-rhamnose and 1000 mg lactulose in 10 mL sterile water was prepared as previously described (28). Study personnel administered the entire solution to each child over a period of 5 min. A urine bag was then placed on the child 30 min after consuming the solution and all urine collected over the subsequent 60 min was returned to the data collector. Analysis of lactulose and rhamnose was conducted according to a previously described procedure (29, 30). In stool samples, myeloperoxidase (MPO) was measured using a commercially available enzyme-linked immunosorbent assay (MPO RUO, Alpco, Salem, NH). One child had severe diarrhea and difficulty in producing feces, we were unable to collect a fecal sample from him; another child produced a small amount of feces that was only sufficient for PCR testing. The samples sizes for PCR, MeTRS, and MPO analyses were therefore 101, 100, and 100, respectively.

Threshold cutoffs for moderate and severe EED based on the dual sugar absorption test and fecal MPO were defined following previously described methods (31, 32). As reported by Ordiz et al. (31), urinary excretion of lactulose and rhamnose were highly correlated as illustrated in Supplementary Figure 1, and following these authors, we adopted the percentage of lactulose excretion as a biomarker for gut permeability associated with EED. Cutoff values for the percentage of lactulose (%L) were as suggested by Agapova et al. (33): 0.2 < %L ≤ 0.45% for moderate EED, and %L > 0.45 for severe EED. Following Kosek et al. (32), we defined a threshold for fecal MPO of 2,000 ng/ml for gut inflammation and the third quartile of the observed data (11,000 ng/ml in our cohort) as a threshold for severe inflammation.

The child's weight, recumbent length, and Mid-Upper Arm Circumference (MUAC) measurements were collected by trained caregivers using Seca 334 baby scales, Seca 417 length boards (Itin Scale Co., Inc., Brooklyn, NY), and MUAC tape precise to 0.01 kg, 0.1, and 0.2 cm, respectively. Length measurements were taken three times. The average length was calculated by taking the arithmetic mean of the data after outlier removal as described in Supplementary Appendix 1. Length-for-age, weight-for-age, and weight-for-length Z scores (LAZ, WAZ, and WLZ, respectively) were calculated according to WHO Child Growth Standards using R package anthro. Severe acute malnutrition was defined according to Ethiopian clinical standards (MUAC <110 mm and length >65 cm or a WLZ ≤ −3) (34).

A total of 12 children with one (or more) of the following medical condition(s) were referred to a healthcare facility, including four cases of dehydration, five cases of diarrhea and fever, and six cases of diarrhea. Two children were referred to a health post for additional follow-up for severe acute malnutrition.

Primary outcome variables of interest were Campylobacter colonization (by PCR), EED, and stunting. Descriptive analysis was undertaken to explore background features of the study population. 95% binomial proportion confidence intervals for prevalence were based on Wilson score. Explanatory variables potentially associated with the outcome variables were selected based on a priori knowledge (1–4, 7, 8, 10, 12, 13, 16). All data analyses were performed in the statistical language R version 3.5.1 (35).

Composite variables were generated to assist in analyzing the sociodemographic data. We created Water, Sanitation, and Hygiene (WASH) ladder variables (i.e., drinking water, sanitation, and hygiene ladders) based on the Joint Monitoring Program (JMP) by WHO and UNICEF (22, 36). Tropical livestock unit (TLU), a composite metric for quantifying all farming animals in a household, was calculated following published literature (37).

Bivariate and multivariate analyses were conducted to explore associations between the three outcome variables and potential explanatory variables using unadjusted and adjusted logistic regression models [controlling for child's age (in days), sex, and kebele of residence]. To strengthen the robustness of the models, continuous [i.e., child's and mother's age, TLU, household wealth (asset and income)] and ordinal explanatory variables were, respectively dichotomized by medians or generally accepted cut-offs (i.e., minimum dietary diversity score of 5). We established multiple exposure models by including the three confounders (child's age group, sex, and kebele of residence) and candidate variables with p-values < 0.20 from the adjusted single exposure models (Supplementary Tables 4–6), and backward stepwise selections were conducted to sort out the significant explanatory variables (38). Given the small sample size, all p-values and 95% confidence intervals (CIs) were estimated by likelihood-ratio (LR) test, and we did not include interaction between the explanatory variables in the models. Observations with missing values were excluded from the analyses.

Campylobacter prevalence in child fecal samples was 50% (95% CI: 41–60%, n = 101) as deduced from PCR-based detection (23), and 12% of the children were reported to have current diarrhea. MeTRS revealed 88% (95% CI: 80–93%, n = 100) prevalence of the genus Campylobacter among the studied samples. Eight Campylobacter species showed the highest prevalence: C. jejuni, C. hyointestinalis, C. coli, Campylobacter sp. RM6137, C. upsaliensis, an uncultured Campylobacter sp. and Campylobacter sp. RM12175 (Table 4). Interestingly, most of the children (82%, n = 100) were observed to be infected with multiple species of Campylobacter.

As expected, absorption of lactulose (%L¯ = 0.27) and rhamnose (%R¯ = 4.03) are highly correlated (Pearson's ρ= 0.44, P-value < 0.01) (Supplementary Table 1 and Supplementary Figure 1). Based on the specified threshold cutoffs for MPO and %L biomarkers, the prevalence of EED in the study children was deduced to be 50% (95% CI: 40–60%, n = 100), among whom 17% (95% CI: 11–26%) had severe EED (Supplementary Table 2).

Comparing with the reference distribution of LAZ and WLZ (both with a mean = 0 and SD = 1), the full distribution of LAZ values was shifted to the left, while the distribution of WLZ had a mode at 0 but was tailed to the left (Figure 1). The means for LAZ, WAZ, WLZ, and MUAC were −1.88 (SD = 1.56), −1.25 (SD = 1.18), −0.42 (SD = 0.91), and 130 mm (SD = 11), respectively (Supplementary Table 3). Stunting was observed in 41% (95% CI: 32–51%) of the study population, while severe stunting occurred in 19% (95% CI: 12–27%) of the children (Table 5). Wasting and severe wasting were observed in 5% (95% CI: 2–11%) and 1% (95% CI: 0–5%) of the population, respectively, and 3% (1–8%) of children were both stunted and wasted (Table 5).

Supplementary Tables 4–6 present crude and adjusted odd ratios, and Table 6 depicts the models with single (with p [probability]-values <0.2) and multiple risk factor(s), i.e., the Campylobacter model considering children's current breastfeeding and animal source food consumption practices as risk factors. The five kebeles could be categorized into three sets by assembling the three kebeles with similar prevalence of Campylobacter (by PCR) into a new group as a confounder for kebele effect in Campylobacter models (Supplementary Table 4). Given this three-level kebele effect did not yield statistically different prevalence of EED and stunting (Supplementary Tables 5, 6), this scheme was kept in the models of these two outcomes (i.e., EED and stunting) to account for spatial heterogeneity. Statistical analysis showed that the group of children who are being breastfed and also currently consuming ASF had significantly higher odds for the presence of Campylobacter spp. when adjusting for confounders and other risk factors at the 5% level (Table 6). EED was treated as a binary variable by combining categories of moderate and severe EED as listed in the Supplementary Table 2. The household's access to a basic or safely managed drinking water source significantly reduced the odds of EED at the 5% level (Table 6). No significant crude or adjusted associations were found between stunting and any of the explanatory variables (Supplementary Table 6), hence multiple exposure models were not constructed. Also, the aforementioned associations (measured through the logistic regressions) between the explanatory factors and the three health endpoints (i.e., colonization with Campylobacter spp., EED, and stunting) and their prevalence estimates are depicted in Figure 2.

According to the 2018 Global Hunger Index, Ethiopia ranks 93rd out of 199 qualifying low/middle income countries and suffers from a level of hunger that is considered serious (39). The prevalence of stunting in children under 5 years of age has decreased from 57.4% in 2000 to 38.4% in 2018. In the study area, we found that the prevalence of stunting among children 12–16 month of age is 41% (95% CI: 32–51%), which is higher than a 2020 study in southern rural Ethiopia (31.4%) (40). Conversely, our study found a lower rate of wasting 5% (2–11%) compared to the southern rural region in this country (14.3%) (40). On the other hand, in rural Ethiopia, average HAZ value has been reported to be −1.7 at approximately 9.6–20 months of age (7), whereas the average LAZ in our setting at that age was −1.9. Hence, chronic malnutrition in the area of this study appears to be more serious than the country's rural average.

The prevalence of Campylobacter spp. among almost 50% of the under-five children in our setting is similar to Tanzania from the MAL-ED study and another recent surveillance study conducted in southern rural Ethiopia (15, 40), when measured by PCR, confirming a high burden of Campylobacter colonization. High exposure to pathogenic microorganisms is also indicated by high prevalence of diarrhea and fever (41, 42). The results of this study aligns with findings from previous studies that indicate a considerable number of diarrheal episodes can be associated with Campylobacter spp., which is one of the most commonly detected species in stool samples in Peru, South Africa, Brazil, and Nepal (41, 42). The metagenomic results confirmed that the thermotolerant C. jejuni and C. coli, which are commonly associated with chickens and livestock, are frequently colonizing at high abundance levels in the gut of young children, that we have also presented in more details (18). Notably, we observed the presence of C. upsaliensis, which is a member of the C. jejuni group but commonly associated with cats and dogs. An important finding of our study is the occurrence of a large number of other, non-thermotolerant Campylobacter species known to be highly diverse. Their reservoirs are poorly characterized, especially in low- and middle-income countries. The frequent presence of these non-thermotolerant species, including C. hyointestinalis, showing higher abundance than C. jejuni/C. coli, indicates a salient impact of various animals on child health in Ethiopia. Previous studies have shown that intestinal colonization of C. upsaliensis, C. fetus, and C. hyointestinalis can induce gastroenteritis in humans (43–45). Apart from human isolates, C. hyointestinalis subsp. hyointestinalis has been mainly isolated from ruminants while C. hyointestinalis subsp. lawsonii has been isolated from pigs (46). The primary reservoirs of C. fetus subsp. fetus include cattle and sheep, but this subspecies has also been isolated from the feces of other animal species. Among the Campylobacter spp. identified by MeTRS in this study, C. fetus subsp. venerealis is a cause of infection in cows (44), and Campylobacter sp. RM 6137 and Campylobacter sp. RM12175 are unnamed novel species in the C. fetus group that have been isolated from feces of wild pigs and cows, respectively (47). Therefore, frequent occurrence of C. jejuni/C. coli in children supports the hypothesis that chickens are a main reservoir of these bacteria. The presence of other species of Campylobacter implies that domestic and farmed animals other than poultry may also significantly contribute to environmental exposure and colonization of children. This inference is further supported by the increased risk of Campylobacter colonization associated with ASF consumption, which in this region mainly includes raw cow's milk (Table 6).

The observed significantly increased risk of Campylobacter colonization associated with current breastfeeding, which is in accordance with a growing body of evidence indicating that exclusive breastfeeding may provide a gut environment, more favorable to growth of (certain species) of Campylobacter (23, 48). It is currently not clear how breastfeeding provides a competitive advantage to Campylobacter species in the infant gut (23). Nonetheless, the significance and importance of breastfeeding for overall child infant nutrition, development, and protection against infectious diseases is well-documented (48). Our findings should not be interpreted as to discourage breastfeeding of infants and young children but underscore the importance of controlling Campylobacter exposure, as well as overall gut health, to protect and ensure proper growth of children.

There is no gold standard for classifying EED based on biomarkers, and different approaches have been used to define this endpoint (28, 32). We adopted the Lactulose: Rhamnose test, which is now considered more appropriate than the Lactulose: Mannitol test as a marker for intestinal permeability, and MPO as a well-established marker of neutrophil activity and gut inflammation (28). A classification based on these two markers was then used to define moderate and severe EED and revealed a high burden of EED in our study area. Fecal MPO results are known to be strongly variant within and between populations geographically. The median MPO concentration in our setting is lower than the median of all sites in the MAL-ED cohort (32). This does not necessarily imply a lower EED burden in our study population, because the variance of MPO may be due to different age of children in the two studies, since MPO is age-dependent (49).

Typically, the diarrheal prevalence has been reported to be 20–25% among the under-five children in the study region (50, 51). However, the mother interviewed in our study have reported a higher prevalence (48%) of recent diarrhea (in the previous 15 days) among the children aged between 12 and 16 months. A high prevalence of recent diarrhea in children, as reported by their mothers, has also reported from Pakistan (63%), Bangladesh (53%), and Peru (53%), whereas lower prevalence was reported in India (21%), Nepal (27%), South Africa (4%), Tanzania (6%), and Brazil (5%) (15). These differences in diarrheal prevalence might be due to variations in selection criteria, geo-climatic conditions, and WASH and feeding practices. For example, in congruence to our study, increased diarrhea has been found in populations consuming more milk in southeast Asia (52). Further investigation of the microbial content of ASF, as well as more careful characterization of symptomatic diarrhea would aid in better understanding this potential driver of stunting among different child cohorts.

Only 9% of children in our study population met the MDD-IYC of five out of eight food groups (53). While intake of diverse ASF can significantly improve child diet quality, no child consumed any animal flesh, only about half of children consumed dairy products, and very few of them consumed eggs. Though seasonality of diet may vary significantly in parts of Ethiopia (54), evidence does not suggest that these data reflect lower than normal rates of MDD-IYC than would otherwise be observed in the study area. Moreover, increased diversity in ASF for younger children may vary seasonally with parents income (55), which was not captured in this study.

Our fieldwork indicated that people in the Haramaya woreda do not consider exposure to animal feces a public health issue, and household members in the community had little knowledge or concern about the risks of zoonotic exposure. In the study region, access to or use of proper sanitation facilities was observed to be very poor and human and/or animal excreta were found scattered in most homesteads. Despite the general suboptimal drinking water supply, we found access to the basic or safely managed source of drinking water was associated with reduced odds of EED. This emphasizes the necessity of improving the drinking water supply in the study region. Moreover, people's perception of the significance and adequate practice of hand-washing was low in general. The above contexts of WASH practices could not only induce microbial exposure (e.g., Campylobacter spp.) to the mothers but also to their children.

Most of the study families had their own livestock and poultry, and most animals were freely roaming around without proper management of animal droppings. As observed, most of the families kept the smaller animals in house overnight due to security concerns; in over half of households, chickens roamed freely within household compounds during nighttime. This kind of animal husbandry, in addition to the poor status of WASH practices, place human populations at high risk of exposure to zoonotic pathogens and may explain the observed high prevalence of Campylobacter and EED in children. Nonetheless, other research underlines the difficulty of unraveling these complex networks: despite multiple observational studies that found household-level WASH to be strongly associated with child linear growth, three landmark randomized clinical trials found WASH interventions had no statistically significant impact on the linear growth, even though effective behavior change and prevalence reduction for some enteric pathogens were achieved (56, 57).

Only a few significant associations between explanatory variables and stunting, EED, and Campylobacter colonization among young children could be ascertained, which might be due to relatively small sample size of this study. Other limitations of this study included the use of the cross-sectional survey data to evaluate variation in health status as a consequence of multiple factors interacting on different time scales, including children's diet and breastfeeding practices, and their health conditions (diarrhea and fever statuses), which are complex interactions particularly susceptible to recall bias (58).