The primary design of the study is a prospective, observational birth cohort study to describe the acquisition, patterns, and burden of gut bacterial resistome in the first 24 months of life. This study protocol adheres to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for reporting observational studies [22], adapted to capture metagenomic-specific methodological details (STROBE-metagenomics) [23]. The STROBE checklist is provided in Checklist 1.

The study is conducted in the Dzimauli community in the Limpopo Province, northeastern South Africa. The geography and climatic conditions of Dzimauli have been previously described [20,24]. Briefly, Dzimauli is rural and approximately 50 km from Thohoyandou, the main city of the Vhembe district, Limpopo Province. The population predominantly practices subsistence agriculture and petty trading. The area experiences heavy rainfall and flooding due to variations in climatic conditions [24].

A sample size of 200 mother-child pairs was selected based on adequacy for the longitudinal study objectives. The cohort size is sufficient to estimate resistome carriage and burden, and to support mixed-effects and compositional analyses of resistome trajectories and associated exposures across repeated measurements at the community level. It also accommodates anticipated attrition (15%‐20%) over the 24-month period, while maintaining analytical robustness and feasibility. Although conducted in a small rural community, recruiting up to 200 participant pairs with extensive follow-up would enable population-level interpretation of findings or application of the study approach within a similar setting.

Pregnant women in their third trimester are identified through household and outpatient antenatal clinic visits. Upon consent, they are screened and enrolled in the study at the birth of the child. Inclusion criteria for newborns include (1) participant should be available in the study community for at least 6 months, (2) birth weight ≥1500 g, (3) absence of immediate postnatal hospitalization or treatment for severe illness, and (4) mothers should be aged at least 16 years. Infants are excluded if (1) the child is from a multiple pregnancy, (2) the mother has another child enrolled in the AMR study, (3) the mother plans to relocate from the community within 6 months of enrollment, or (4) the mother is unable to consent or declines participation.

Field workers were recruited from the study community and trained to screen and enroll participants and conduct surveillance activities. Data collection spans a 24-month period and includes structured surveillance questionnaires, biological samples, and clinical assessments to capture longitudinal information on maternal and child health, feeding practices, growth, socioeconomic status, hygiene, and environmental exposures [25]. A comprehensive summary of data collection domains, time points, and laboratory analyses is provided in Table 1. Briefly, at enrollment (≤17 d), maternal and child demographic, perinatal, and anthropometric data, as well as treatment, hospitalization, and breastfeeding practices, are collected. Child anthropometrics are subsequently collected monthly for 12 months and quarterly for another 12 months. The study adopted instruments developed and used in the same community by the MAL-ED (The Etiology, Risk Factors and Interactions of Enteric Infections and Malnutrition and the Consequences for Child Health and Development) study [25]. However, for the purpose of this study, household visits were conducted once a week for 24 months, during which time data on illness, antibiotic use, and other treatments received, feeding patterns, and vaccines received were captured by a 7-day recall questionnaire.

Surveillance data are collected with the mother or primary caregiver at various timepoints as detailed in Table 1. For infants aged 1‐8 months, a qualitative 24-hour dietary recall is adapted from the Demographic and Health Survey, which was developed for the same community as part of the MAL-ED study [25]. The qualitative dietary recall complements the once-weekly dietary reports by capturing a broader variety of foods consumed, including specific fruits and vegetables, grains, roots, and animal source foods. A quantitative 24-hour recall, used in the same community from the MAL-ED study [26] was also used from 9 to 24 months, to estimate energy and nutrient intake from nonbreast milk foods during the weaning process. In addition, to assess the household food access insecurity status. To assess food access insecurity, our survey included the nine-question household food insecurity access scale adapted in 2006 by the Food and Nutrition Technical Assistance project for use in low-resource settings [27] and used in the same population in the eight-countryside MAL-ED study by Psaki et al [28]. On a monthly basis, we assess water, sanitation, and hygiene practices of mothers or caregivers, incorporating the 5 key principles and the practical steps to achieve universal access to quality care WHO standards [29], as well as indicators related to antibiotic use and potential drivers of antibiotic resistance. At baseline, 7, 15, and 24 months, we also administer a Self-Reporting Questionnaire screening instrument that effectively measures depression in postpartum women, an instrument previously developed and used in the same population [30]. The Self-Reporting Questionnaire comprises 20 items that assess psychological disturbances related to depressive symptoms occurring within the previous 4 weeks. Questions are answered with a simple “yes” or “no,” and trained field assistants administer the instrument.

Mothers are trained to collect stool and breast milk samples as required. A stool sample is collected from the child and mother at enrollment, and then monthly, and whenever there is diarrhea, for 12 months. Thereafter, stool is collected quarterly, and whenever there is diarrhea, for a further 12 months. Breast milk is collected at enrollment and monthly, and for as long as the mother breastfeeds, for a total duration of 12 months.

Environmental samples, including household drinking water and fecal matter from in-household livestock and pets, are collected once during the study to assess zoonotic E coli resistome. Hydrometeorological exposure data such as precipitation, temperature, humidity, and soil moisture will be extracted from version 2.1 of the Global Land Data Assimilation System.

The main goal is to generate metagenomic sequence data using next-generation sequencing, bioinformatic, and correlational analyses to understand infant microbial composition and resistance dynamics. Total genomic DNA (gDNA) is extracted from infant stool samples using a QIAamp Fast DNA stool min Kit (Qiagen) following the manufacturer’s protocol, with the inclusion of a bead-beating step, to the protocol. Extracted gDNA is quantified with a Qubit dsDNA High Sensitivity Assay, and library preparation is conducted using the Nextera XT DNA Library Prep Kit (Illumina). Sequencing will be performed on the Illumina NextSeq 2000 platform with paired-end reads (2×150 bp) using the P4 flow cell.

The study will first characterize the total gut resistome and then focus on E coli resistome, where the organism will be used as a proxy organism for monitoring AMR in the gut microbiota due to its ubiquity, early colonization in the infant gut, and well-characterized genomic features. In addition, the ESKAPE pathogens’ resistome will be included to broaden our understanding of clinically relevant bacteria. By leveraging E coli and ESKAPE pathogens as sentinel organisms, this study aims to provide a focused yet robust characterization of the early-life enteric resistome, enabling deeper insights into AMR acquisition in low-resource community settings.

To generate data that will contribute to the development of relevant educational packages, a community survey on Knowledge, Attitude, and Practice (KAP) on the use of antimicrobials is underway. A previous study in the Dzimauli area (in the Thulamela municipality in the Vhembe District of South Africa) from 2009 to 2012 reported an estimated population of 9000 [19]. Assuming an annual growth rate of 2.23% based on the population growth rate calculated from the Thulamela municipality population from 2011 to 2022 [38], the population of Dzimauli was projected to be 12,256 by 2023. The sample size for the community KAP survey was estimated as 466 targeted participants using the single-proportion formula with finite population correction (n=12,256), assuming P=0.5, d=0.05, and 95% CI and assuming 80% data completeness. A structured questionnaire is used to collect information on participant demographics, knowledge of antibiotics and antibiotic resistance, as well as attitudes and practices related to antibiotic use. The survey is conducted door-to-door and is administered in English or Tshivenda (the local language), depending on the participant’s preference. Inclusion criteria are those aged 16 years or older. The KAP component consists of 29 closed-ended questions (14 knowledge, 7 attitude, and 8 practice-related items). There is also a question to provide sources of information on antibiotics and 2 open-ended questions on general views. A 4-point Likert scale ranging from “Strongly disagree” to “strongly agree” was adopted to capture participant responses to closed-ended questions. Responses will be analyzed only for participants with complete data sets. The responses will be collapsed into a dichotomous set based on correct or incorrect responses. Total scores will be used to classify “high” (≥80%), “moderate” (60%‐79%), and “low” (<60%) KAPs on antibiotic use [39,40]. Frequency distribution for each KAP will be determined to understand the level of KAP in the community regarding antibiotics. Logistic regression analysis will be conducted to identify factors associated with KAP regarding antibiotic use, and odds ratios with 95% CIs will be reported. Statistical significance will be set at a P value <.05.

Open-ended responses from KAP surveys will be analyzed using thematic analysis. Responses will be transcribed where necessary, coded inductively, and grouped into thematic themes related to antibiotic knowledge, perceptions, practices, and contextual drivers of inappropriate antibiotic use. Qualitative findings will be used to contextualize and interpret quantitative KAP. Integration of qualitative and quantitative data will be conducted informally during interpretation, to support knowledge translation and the co-production of ethnographically relevant AMR stewardship interventions.

In addition to the community KAP survey, an end-assessment KAP survey is administered at 24 months, or when a participant exits the study. This latter survey aims at evaluating whether participation in a longitudinal AMR study has influenced maternal KAP related to antimicrobial use. Table 2 summarizes the different KAP assessments, the target population, and timing.

This activity draws from the KAP and enteric bacteria resistome datasets. During conversation events, participants will share their first-hand experiences on AMR and collectively brainstorm ideas and co-create solutions to tackle AMR challenges. Engagement with stakeholders, such as policy and decision makers, will take place during symposiums, which began during the conception of the project. Policy and decision makers are drawn from district, provincial, and national government health agencies charged with infection prevention and control.

We describe here how the study guarantees data quality, ensures reproducibility, and the ethics of doing research by tracking research integrity, misconduct, and misuse of research findings. The principal investigator reviews informed consent forms on the day they are signed, for correctness and completeness, and stores them securely. The research assistants are trained on the study protocol, data collection tools, and ethical conduct of research prior to commencement of fieldwork and through ongoing refreshers as needed. Case report forms are reviewed within 24 hours and signed by a supervisor. Corrected data on case report forms remains legible, signed, and dated. Repeated data capture errors trigger re-training for the worker or the group. Field and laboratory supervisors hold weekly meetings with staff to review data quality plans, challenges, and progress. Monthly, a supervisor does 5% of the monthly assessments for quality assurance and performs a dry-run training. Approved data are captured in REDCap (Research Electronic Data Capture; Vanderbilt University) system. Devices for anthropometric measurements are maintained and calibrated as recommended. Biorepositories are fitted with temperature monitoring devices and are monitored daily.

A protocol deviation form is completed for every out-of-schedule field assessment, laboratory test, or missing data. Case report forms are scanned and backed up electronically. In total, 85% of completeness and accuracy is the minimum targeted threshold for all data points. Data management is governed by policies and ethics on data creation, storage, sharing, and access. Training on ethics, integrity, and responsible conduct of research emphasizes data falsification, fabrication, plagiarism, and the reputation of the study community, investigators, institutions, and funders.

Potential tangible risks related to research misuse include the risk of misuse of (1) personal data and identifiers of study participants, and (2) analysis of DNA sequence data for reasons other than those of the current research. Soon after consent to participate is obtained, all data collection, tests, and assessments will include a participant identifier on the data collection tool. Samples also have sample identifiers. Through this, participants are delinked from samples, assessments, and datasets. In addition, access to data is password-protected. Study investigators, field, and laboratory workers are trained on the responsible conduct of research involving human participants.

The study generates next-generation sequencing data potentially containing human genes. It is worthwhile to note that it is difficult to deidentify a genome, because a genome by itself is in many ways an identifier. However, in this study, investigators avoid examining parts of the generated sequences that are not of interest to the study in order to avoid incidental findings, such as stumbling on variants that are not of interest to the study but could be of interest to a study participant. Incidental findings pose a dilemma. If this sort of event arises, the matter will be referred to the institutional research ethics committee for guidance on how to proceed. Generally, misuse of data is governed and managed according to legislation and guidelines regulating access to personal information and protection of human dignity.

The study protocol has approval from the Human and Clinical Trials Research Ethics Committee of the University of Venda, South Africa (FSEA/22/MBY/04/0302). Permission was granted by the Limpopo Provincial Department of Health, South Africa, to access local health care facilities in the study community for identification of potential participants (LP_2023-03-014). A Memorandum of Understanding regulates the relationship with the Dzimauli Tribal Authority. Several meetings were held with the community leadership to obtain community approval prior to individual consent. A community advisory committee was established to guide the values of respect, honesty, fairness, and justice vis-à-vis community participation, protection, and expectations from the study. Individually signed informed consent is obtained for all forms of participation in the study. To protect participants’ privacy and confidentiality, specimens and surveillance data are stripped of personal identifiers and coded. Ethical principles according to the Helsinki Declaration are adhered to [41]. The study participants did not receive compensation.

The study was funded in June 2022. Ethical approval was obtained in February 2023 from the Human and Clinical Trials Research Ethics Committee of the University of Venda, South Africa (FSEA/22/MBY/04/0302). Recruitment of mother-child dyads commenced in August 2023, with an expected end date in June 2026. By October 2025, 167 mother-child dyads had been enrolled, and baseline demographic data collection had been completed for the enrolled participants. Stool sample collection continues in accordance with the study follow-up schedule. Laboratory DNA extraction commenced in January 2024, while sequencing of the baseline samples for all participants began in August 2025, with initial data quality control and descriptive analyses expected by May 2026. Comprehensive integrative resistome-microbiome and qualitative analyses are planned following completion of data collection in December 2027. Dissemination of the principal study findings is anticipated through peer-reviewed publications and structured community feedback activities between June 2026 and December 2028.

Among the 167 newborns enrolled by October 2025, 20 participants have completed 24 months of follow-up. Baseline characteristics of these participants are summarized by total numbers with percentages and mean (SDs) to give an overview of participant characteristics. Proportions for each variable are based on the total number of participants with complete data on the assessment or question on the instrument. An attrition rate of 9.6% has been recorded. Figure 1 shows the enrolled strategy, while Tables 3 and 4 show selected baseline descriptive characteristics of the 167 participant-dyads. The majority of the infants (129/167, 77.2%) were enrolled within 14 days after birth. There were no significant differences in sex (χ²₁=0.15; P=.70) among enrolled babies, indicating an approximately equal representation of males and females. The mean weight of the babies at enrollment was 3.4 (SD 0.6) kg, compared to 3.1 (SD 0.5) kg recorded at birth. The mean length of babies at enrollment was 49.5 (SD 4.4) cm. None of the babies were hospitalized prior to enrollment, and 95.8% (160/167) received colostrum at birth. The length-for-age z scores indicated that 3.6% (6/165) were severely stunted, and weight-for-length z scores revealed a 5% (8/161) in the category of severe wasting. At enrollment, the mean age of the mothers was 27.7 (SD 7.1) years. Overall, 29.9 % (50/167) of the mothers had completed at least 12 years of schooling. Of those below 25 years, 30.6% (26/85) were attending at least high school. At enrollment, a baseline stool sample was collected from 96.4% (158/167) of the infant participants. For a robust statistical power calculation, a minimum threshold of 85% of complete baseline data collection for each of: sociodemographics, anthropometrics, treatment, hospitalization, and breastfeeding is being met.

This study protocol describes a prospective observational birth cohort designed to investigate the acquisition, composition, and transmission of ARGs during the first 24 months of life in a South African setting of low socioeconomic status. Early life exposures, particularly within the first 24 months of life, are critical for microbiome establishment and resistome development and shape long-term immunity, metabolism, and resistance profiles [31,32]. By integrating longitudinal metagenomic surveillance with external exposures, including hydrometeorological and contextual data, the study aims to address key gaps in understanding the temporal dynamics of microbiome-resistome and AMR emergence and dissemination outside health care settings. In parallel, the study incorporates KAP assessments on antibiotic use [33] and the co-production of intervention materials, with the goal of translating biological findings into contextually relevant AMR mitigation strategies [34].

Accordingly, a multidimensional longitudinal birth cohort in Dzimauli, a rural South African setting, as a model for community-level AMR stewardship is being established. The cohort is designed to integrate biological, social, and contextual data to provide comprehensive interventional data for community-level AMR. Integrative data generated from this cohort is expected to provide several important contributions to AMR evidence-based interventions. First, the study is expected to generate critical prospective datasets describing the enteric resistome in a low-socioeconomic community, addressing a major gap in community-level AMR surveillance. Longitudinal metagenomic enteric profiles will reveal critical data, including acquisition, composition, persistence, and evolution of ARGs; community-level intervention hotspots; emerging community-level critical pathogens of clinical relevance and potential transmission pathways, including vertical and horizontal transmissions within households and the broader environment. We capture a wide range of exposures, contextual factors, and potential confounders to clearly identify drivers of AMR. Integrative metagenomic and exposure data will provide intervention insights. This approach ensures that intervention strategies are grounded in local realities and informed by community perspectives. Finally, the study will ensure enhancement of the development of leadership in integrative AMR epidemiology in low-income communities for local and global relevance.

Current evidence on AMR in low-resource settings is dominated by cross-sectional studies, which provide only snapshots of resistance patterns at single time points [35,36]. Although such studies are valuable in terms of estimating prevalence, they are limited in their ability to capture proper AMR baseline and temporal dynamics data, including when ARGs are first acquired, which bacteria greatly contribute the most ARGs, how they persist or change over time, and how their transmission pathways evolve. The emergence of AMR itself is complex, shaped by the interaction of biological, environmental, behavioral, and social factors [37,38]. However, a plethora of studies focus on a limited subset of potential drivers, such as antibiotic use or selected environmental variables, without accounting for broader contextual influences [39,40,42]. This narrow focus may bias interpretations of AMR drivers and underestimates the contribution of diverse structural and social determinants. Furthermore, while many AMR studies generate detailed microbiological data, limited studies explicitly link these findings to community knowledge, behaviors, lived experiences, and the production of setting or regional-specific interventions [34,43,44]. These limitations are particularly pronounced in sub-Saharan Africa, where longitudinal AMR data from community settings, especially among infants and young children, remain scarce [45]. As a result, the establishment of an interdisciplinary birth study cohort is key to addressing these limitations and generating the longitudinal, contextualized data necessary for effective AMR surveillance and stewardship.

It is worthwhile to indicate challenges faced so far in the cohort establishment, biospecimen and surveillance data collection, and the strategies being applied to mitigate the challenges. First, due to socioeconomic factors, there is a fair number of relocations of community residents out of the study area. Therefore, effort is required to identify potential eligible participants who aim to stay long enough. This approach contributes to lower attrition and meaningful contributions to the datasets envisaged. Second, the topology of the study community makes the mobility of the field workers to access households difficult. To address this, efforts were made to increase the number of field workers such that the number of households that each field worker is responsible for is reduced. The added resources to address the challenge of mobility pay off by enhancing timely data collection, completeness, and accuracy. Third, the display of fatigue in providing samples prospectively. This behavior was diagnosed as an inadequate understanding of why a specimen, for example, stool, is collected repeatedly. This is addressed through conversations with relevant participants for the purpose of repeated collection and through continuous consenting.

The cohort study has major strengths, including its (1) prospective design and community-based nature, which address a critical evidence gap in community engagement and co-production of scientific evidence in the mitigation of AMR, (2) its broad longitudinal exposure data, including the integration of hydrometeorological and nutritional variables, would enable a holistic assessment of potential drivers of resistome acquisition beyond antibiotic exposure alone, and (3) the use of shotgun metagenomic sequencing to determine the resistome makes it a strong methodological approach to obtain large datasets necessary for the discovery of novel niches and relevant priority organisms driving AMR. Overall, it brings to the fore the role of enteric resistome in child health and livelihoods for global health imperatives.

Despite these strengths, the cohort study should be considered with certain potential limitations. These include the introduction of recall bias, increased risk of missing or delayed data due to the prospective nature and the intensity of data collection; factors such as undocumented exposures, such as over-the-counter antibiotic use, may be missed; the management of attrition rate in some communities, for example, pastoral communities, may be difficult. Furthermore, although metagenomic data provides a high level of granularity, it may not be feasible in resource-constrained research environments.

Large prospective datasets on AMR are necessary for decision-making and intervention in sub-Saharan African settings, particularly in low-socioeconomic settings. Therefore, a comprehensive assessment of the multiple and interacting determinants that drive AMR dynamics remains an essential aspect in AMR stewardship. This protocol is a proposal for AMR research at the community level to identify drivers important for AMR mitigation involving local communities in data and evidence generation.