Cessation of exclusive breastfeeding and seasonality, but not small intestinal bacterial overgrowth, are associated with environmental enteric dysfunction: A birth cohort study amongst infants in rural Kenya

Background: Environmental Enteric Dysfunction (EED) is a chronic intestinal inflammatory disorder of unclear aetiology prevalent amongst children in low-income settings and associated with stunting. We aimed to characterise development of EED and its putative risk factors amongst rural Kenyan infants. Methods: In a birth cohort study in Junju, rural coastal Kenya, between August 2015 and January 2017, 100 infants were each followed for nine months. Breastfeeding status was recorded weekly and anthropometry monthly. Acute illnesses and antibiotics were captured by active and passive surveillance. Intestinal function and small intestinal bacterial overgrowth (SIBO) were assessed by monthly urinary lactulose mannitol (LM) and breath hydrogen tests. Faecal alpha-1-antitrypsin, myeloperoxidase and neopterin were measured as EED biomarkers, and microbiota composition assessed by 16S sequencing. Findings: Twenty nine of the 88 participants (33%) that underwent length measurement at nine months of age were stunted (length-for-age Z score <-2). During the rainy season, linear growth was slower and LM ratio was higher. In multivariable models, LM ratio, myeloperoxidase and neopterin increased after cessation of continuous-since-birth exclusive breastfeeding. For LM ratio this only occurred during the rainy season. EED markers were not associated with antibiotics, acute illnesses, SIBO, or gut microbiota diversity. Microbiota diversified with age and was not strongly associated with complementary food introduction or linear growth impairment. Interpretation: Our data suggest that intensified promotion of uninterrupted exclusive breastfeeding amongst infants under six months during the rainy season, where rainfall is seasonal, may help prevent EED. Our findings also suggest that therapeutic strategies directed towards SIBO are unlikely to impact on EED in this setting. However, further development of non-invasive diagnostic methods for SIBO is required. Funding: This research was funded in part by the Wellcome Trust (Research Training Fellowship to RJC (103376/Z/13/Z)). EPKP was supported by the MRC/DfID Newton Fund (MR/N006259/1). JAB was supported by the MRC/DFiD/Wellcome Trust Joint Global Health Trials scheme (MR/M007367/1) and the Bill & Melinda Gates Foundation (OPP1131320). HHU was supported by the NIHR Oxford Biomedical Research Centre (IS-BRC-1215-20008). For this prospective observational birth cohort, infants from the 10 closest villages surrounding Junju dispensary, Kilifi County, Kenya (3·85°S 39·73°E), were recruited within 14 days of birth and followed until the tenth month of life. Sample size was based on a statistical significance test comparing a continuous outcome (LM ratio at six months) between two groups: exposed and unexposed to SIBO (at least one positive GBHT). Predicted mean LM ratio ranged from 0.17 to 0.38 with standard deviation 0.15 to 0.30 and SIBO prevalence 10 to 30%.21, 22, 23, 24 74 participants yielded 95% power to detect a two-fold difference in mean LM ratios between SIBO-exposed and SIBO-unexposed groups (alpha 0.05; mean LM ratios 0.2 and 0.4). This number was increased by 15% to allow for non-parametric testing25 and by 15% for loss to follow-up. 100 participants were therefore enrolled. At enrolment, household, maternal, and participant characteristics were elicited. Length, weight, and mid upper arm circumference (MUAC) measurement and GBHT and LM testing were carried out monthly, within 1–3 days of each other. See Supplementary Text for detailed LM and GBHT methods. To assess inter-observer variability, on one occasion, 20 participants underwent repeated measurement. Between 5·5 and 6·5 months of age, participants underwent timed venesection targeting 90 min after ingestion of oral LM solution.26 Complete blood count was also performed and participants with haemoglobin <8 g/dL were treated with 14 days of oral iron. Acute illness events (diarrhoea, breathing difficulty and fever), antibiotics, exclusive breastfeeding status and other foods given were captured by weekly home visits. Attendances at local primary and secondary government facilities were captured through passive and active surveillance. Season was defined by prospective weekly recording of local rainfall: April, May, June, July, November, and December were classified ‘rainy’ and remaining months ‘dry’. Non-diarrhoeal stool was collected from disposable nappies into sterile containers on alternate weeks by participants’ caretakers and fieldworkers notified for collection within one hour. Samples were transported at 2–8 °C to the nearby dispensary where they were frozen in liquid nitrogen. Samples collected ≤10 days before a successful LM test underwent Enzyme-linked Immunosorbent Assay (ELISA) measurement of alpha-1-antitrypsin, neopterin, and myeloperoxidase concentration and Illumina sequencing targeting the V3/V4 region of the bacterial 16S rRNA gene. See Supplementary Text for detailed laboratory methods. Statistical analysis was carried out using Stata version 15·1 and R version 3·6·1. Principal component analysis was used to create a continuous variable that captured variation in participants’ household WASH characteristics (Table S1). GBHT results were classified according to Rome criteria27 (positive: ≥1 breath hydrogen reading ≥12 ppm above baseline). Tests with baseline >5 ppm were deemed ‘indeterminate’.28 To examine factors associated with linear growth impairment, a nested cross-sectional mixed effects linear regression analysis was undertaken where each observation was a two-month period of linear growth (change in LAZ) and its preceding and contemporaneous events. Fixed effects were selected prospectively based on upstream proximity to stunting in the hypothesised causal framework (Figure 1). Clustering of repeated observations by participant was adjusted for by random effects. To determine factors associated with EED, a second nested cross-sectional multivariable mixed effects linear regression analysis was employed where each observation (outcome) was an LM test or stool EED biomarker result. A priori testing for interaction between season and exclusive breastfeeding status was done. 16S analyses were performed at a rarefaction depth of 30,000 sequences. Ribosomal Sequence Variant (RSV) count and Shannon index served as covariates in EED and linear growth impairment models. Variation in beta diversity was explored via permutational multivariate analysis of variance (PERMANOVA) using one sample per infant across three age strata (0–3, 4–6 and 7–9 months), with sequencing run as a permutation constraint. Random Forests were used to determine changes in RSV-level microbiota composition associated with age, linear growth impairment, and EED biomarkers (see Appendix for details). Spearman’s rank correlation coefficients were used to determine associations (false discovery rate (FDR) p < 0.1) between RSV abundance and age. Further detail on 16S analysis methods are given in the Supplementary Text. Approval was granted by Kenya Medical Research Institute (KEMRI) Scientific & Ethics Review Unit (2983) and Oxford Tropical Research Ethics Committee (37-15, 566-15). Local language written informed consent was obtained from parents/guardians for all participants by fieldworkers trained in Good Clinical Practice. The funder played no role in the writing of the manuscript or in the decision to submit for publication. The authors have not been paid to write this article by any agency. All authors had access to the data and accept responsibility to submit for publication.