Small Intestine Bacterial Overgrowth is associated with increased Campylobacter and epithelial injury in duodenal biopsies of Bangladeshi children

Small intestine bacterial overgrowth (SIBO) has been associated with enteric inflammation, linear growth stunting, and neurodevelopmental delays in children from low-income countries. Little is known about the histologic changes or epithelial adherent microbiota associated with SIBO. We sought to describe these relationships in a cohort of impoverished Bangladeshi children. Undernourished 12-18-month-old children underwent both glucose hydrogen breath testing for SIBO and duodenoscopy with biopsy. Biopsy samples were subject to both histological scoring and 16s rRNA sequencing. 118 children were enrolled with 16s sequencing data available on 53. Of 11 histological features, we found that SIBO was associated with one, enterocyte injury in the second part of the duodenum (R = 0.21, p = 0.02). SIBO was also associated with a significant increase in Campylobacter by 16s rRNA analysis (Log 2-fold change of 4.43; adjusted p = 1.9 x 10−6). These findings support the growing body of literature showing an association between SIBO and enteric inflammation and enterocyte injury and further delineate the subgroup of children with environmental enteric dysfunction who have SIBO. Further, they show a novel association between SIBO and Campylobacter. Mechanistic work is needed to understand the relationship between SIBO, enterocyte injury, and Campylobacter.


Introduction
Environmental enteric dysfunction (EED) is an asymptomatic syndrome of small bowel dysfunction with adverse impact on childhood growth and development [1,2].It is classically characterized by the alteration of small intestinal architecture including villous atrophy, elongation of crypts, and increased infiltration of inflammatory cells in the lamina propria although recent work to more fully describe the histological changes in EED has also described Paneth and Goblet cell abnormalities in patients with EED [2,3].EED involves enteric inflammation, loss of barrier integrity, and nutrient malabsorption [4,5].It also involves a dysbiosis of the small intestine [6][7][8][9].This syndrome is highly prevalent among the residents of low-and middle-income countries residing in unsanitary living conditions [10,11].Poor water, sanitation, and hygiene practices along with the consumption of fecally contaminated food and water lead to sustained exposure to enteric pathogens [12].This alters the local immune system and is associated with an intestinal dysbiosis [2,7,10].
Small intestine bacterial overgrowth (SIBO) is a particular dysbiosis which has been documented in a subset of children with EED [9,13,14].Glucose-hydrogen-breath-testing (GHBT) is the most commonly used diagnostic test for SIBO as it is non-invasive and well tolerated.SIBO, as measured by GHBT, has been documented in 11% of Bangladeshi 18 week old children and the prevalence increases to 30-40% by 2 years of age [7,14].Recent work in a Bangladeshi cohort showed that total bacterial load, as determined by 16s rRNA sequencing, in duodenal aspirate was associated with enteropathy in children [8].Findings from Madagascar and the Central African Republic have shown an association between increased amounts of culturable total bacteria in the duodenum and stunted growth [15,16].Additionally, key taxa of stunting and enteropathy were identified by both studies [15,16].SIBO, as diagnosed by GHBT has also been associated with fecal markers of inflammation [14].The GHBT has also been associated with linear growth stunting [14,17].All of these studies have been cross-sectional.However, one longitudinal study in Bangladesh demonstrated SIBO in the first two years of life predicted future stunting and language delays [7].
Hydrogen breath testing measures excess hydrogen produced by the upper intestinal flora in response to an oral glucose load.This dysregulated hydrogen economy of the gut is not present in all children with EED and the exact nature of this dysbiosis remains uncertain.Further, it is unclear if the dysbiosis detected by the GHBT is driven by lumenal shifts in bacteria or by shifts in those bacteria adherent to the intestinal wall.These populations are distinct and interact with the host differently [18].
All previous studies demonstrating a link between SIBO and EED have done so by demonstrating elevated biomarkers of inflammation in children with SIBO as compared to those without.However, no description of differences in histology between impoverished children with SIBO and those without has been published.Further, the only descriptions of the small intestinal microbiota associated with SIBO have involved study of duodenal aspirates with no description of the mucosa-adherent microbiota.They have also not used the GHBT for SIBO diagnosis which has been shown to predict future stunting and is associated with enteric inflammation [7][8][9].Therefore, we conducted this study to test our hypotheses that the GHBT would be associated with EED as documented by histology and that the mucosa-adherent duodenal microbiome would differ significantly between impoverished SIBO positive and negative children.

Ethics statement
This project was approved by the Research Review Committee and the Ethics Review Committee at the International Centre for Diarrhoeal Disease Research, Bangladesh and the Institutional Review Board at the University of Virginia.

Study design and settings
This analysis was nested within the Bangladesh Environmental Enteric Dysfunction (BEED) study.BEED was a community-based interventional trial conducted in the slum setting of Mirpur in Dhaka, Bangladesh.This area is densely populated with >38,000 people living per square kilometer [19].We enrolled children aged 12 to 18 months with a length-for-age Z score (LAZ) �-1 SD.This cutoff was chosen for the larger BEED study as indicative of children with chronic malnutrition.Participants were excluded if they were exclusively breastfed, had severe acute malnutrition, severe anemia, tuberculosis, any congenital anomaly or deformity, any severe or chronic disease, an ongoing episode of diarrhea, or a history of persistent diarrhea in the month preceding enrollment.All children received a directly-observed nutritional therapy of an egg and 150 ml milk for 90 days.Subsequently, a subset underwent esophagogastroduodenoscopy (EGD) if their LAZ failed to improve by the end of the 90 day intervention.Enrolled children also had GHBT for detection of SIBO within 7 days of EGD.The detailed procedure of GHBT and the methodology of the BEED study have been published [20,21].EGD was performed on 120 children, of which 118 also had the GHBT.We have included those 118 children in this analysis.However, due to a cold shipment failure effecting 65 samples, we could not analyze the microbiome data from all 118 children.Data from 53 children were included in the microbiome analysis (S1 Fig).
The BEED study was approved by the Institutional Review Board of the International Centre for Diarrhoeal Disease Research, Bangladesh.All methods were carried out in accordance with relevant guidelines and regulations.Written informed consent was obtained from the parents or legal guardians at enrollment as well as prior to EGD.Biopsy samples for 16S rRNA analysis were sent to the UVa under a materials transfer agreement with the icddr,b, Dhaka, Bangladesh.

Data collection
Socioeconomic and household information was obtained from the parents or caregivers of the children in the Bangladeshi cohort at enrollment.Anthropometry data was collected by trained field staff using standard procedures based on the manuals of World Health Organization (WHO) and Centers for Disease Control and Prevention (CDC) guidelines.LAZ was calculated using WHO anthropometry software.EGD was performed by gastroenterologists with expertise in the clinical care of childhood digestive disorders.Two biopsy tissues were obtained from each child, one from the second part of the duodenum (D1) and another from the duodenal bulb (D2).Biopsy tissues were oriented immediately after collection under a microscope.Biopsies were placed in AllProtect (Qiagen, Inc.Hilden, Germany) immediately after collection, and stored at -80 C until bacterial DNA extraction.

SIBO testing
SIBO was assessed by GHBT in the 7 days prior to EGD.Children who were acutely ill or had received antibiotics in the 14 days preceding a scheduled GHBT were rescheduled until after the 14-day period or illness had resolved, whichever was longer.Children were fasted for 2 hours prior to GHBT.We collected a baseline breath sample and then administered a glucose solution of 100g glucose in 500 ml sterile water administered at 5 ml/kg body weight over 10 minutes.Breath was then collected every 20 minutes for 3 hours.Samples were collected using the Quintron (Milwaukee, WI, USA) child breath collection bag and one-way flutter valve which was connected to an appropriately sized pediatric anesthesia mask.Breath samples were immediately analyzed using a Quintron BreathTracker SC (Milwaukee, WI, USA) gas chromatograph.Samples with lower than expected CO2, per the manufacturer's protocol, were considered contaminated with room air, discarded, and immediately recollected.Children were allowed only water during the fasting and testing periods.The GHBT results were dichotomized to positive or negative with a child was labeled as SIBO positive if they had a single post-glucose hydrogen reading �12 ppm over their baseline value.This cutoff was chosen prior to the publication of the American College of Gastroenterology Clinical Guideline for Small Intestinal Bacterial Overgrowth based on data demonstrating a decreased sensitivity when higher cutoffs are used [27][28][29][30].However, given recent analysis showing the trapezoidal area under the hydrogen curve (SIBO AUC) was a better predictor of linear growth in children with EED, SIBO AUC was also utilized to study relationships between the GHBT, duodenal histology, and the mucosa-associated microbiota [7].

Statistical analysis
Statistical analyses were done using R version 3.5.3(https://www.rproject.org,Foundation for Statistical Computing, Vienna, Austria) software.The baseline characteristics of the children were described using number with proportion for categorical variables.Mean ± standard deviation (SD) were used for normally distributed continuous variables.The asymmetric numeric variables are reported with median and inter-quartile range (IQR).The components of histological index were compared between SIBO-positive and SIBO-negative children using Wilcoxon-Mann-Whitney tests as a dichotomous cut-off is the standard interpretation of the glucose hydrogen breath test [29,30].As an exploratory endeavour, based on our previous findings using SIBO AUC, we then determined the correlation between SIBO AUC and components of the histological index using Spearman's rank correlation test [7].
The V4 region of the 16S gene was amplified and sequenced using the primer sequences (S2 Table ) and methodology published by Kozich, et al. and was done in triplicate [31].The library was processed using the DADA2 (v.1.16.0)pipeline and taxonomy was assigned using Silva v.138.Reads were truncated at 190 base pairs for forward reads and 170 base pairs for reverse reads, and a pair-end library was used with expected size of 250 base pairs.Rare taxa that did not have at least 3 copies present in at least 5% of the samples were removed.One sample which only has 3 reads was removed from analysis.
Samples were rarefied using 'rarefy_even_depth' function in the R package 'phyloseq'.Shannon index was used to calculate alpha diversity (the intra-sample diversity).The significance of differences in alpha diversity across groups was tested using the Wilcoxon rank sum test implemented in the R function wilcox.test.A Brays-Curtis dissimilarity Principle Coordinates Analysis (PCoA) on taxa relative abundance was used for visualization of beta diversity (the intra-group diversity).
We assessed for taxa differences in children who were SIBO positive (i.e.>12 ppm increase in exhaled H2 over baseline reading) and negative.We then tested for differences across the 11 histological aspects scored.The significance between taxa relative abundance Bray-Curtis distances and SIBO or histology was evaluated using univariate PERMANOVA tests on Bray-Curtis distance using R function 'adonis()' in package 'vegan' [32].As there was insufficient power to test for differences in SIBO positivity and negativity using PERMANOVA, we divided the children based on the SIBO AUC into the top and bottom 50 th percentiles.These groups, as well has histological features that had a PERMANOVA p value <0.1 were analyzed to identify differentially abundant taxa.Univariate taxa differential abundance in two groups of samples was tested using a negative binomial model for the overdispersion and Poisson process intrinsic to microbiome data, as implemented in DESeq2 package in R, utilizing the Benjamini and Hochberg method for multiple testing correction [33].Initial models were then repeated, correcting for breastfeeding status (some vs. none).

Results
One hundred eighteen Bangladeshi children underwent EGD and GHBT.No adverse events due to EGD occurred.Of them, 53 children had 16s rRNA data available.One child had only 100 reads and was removed from further analysis.Twenty-two (41.5%) children were male, and their mean (±SD) age was 18.7 (±2.1) months.Overall, 15 (12.7%) of the children tested positive for SIBO.Of the 53 children with 16s rRNA data available, 8 (15.1%) tested SIBO positive.The socio-demographic information of the children is presented in Table 1.

Histological differences by GHBT results
No significant difference was observed for the scores of the 11 histological elements between SIBO-positive and SIBO-negative children when GHBT output was dichotomized based on the 12 ppm cutoff.SIBO AUC was positively correlated with the score of enterocyte injury in the second part of the duodenum (R = 0.21, p = 0.02) (Fig 1).The median enterocyte injury score in the second part of the duodenum was 0.30 with an interquartile range of 0.00-0.50.Linear regression was used to determine the linear trend in the association between AUC and enterocyte injury score (Fig 1C).Based on this analysis, for every 1 ppm increase in SIBO AUC, the enterocyte injury score increased by 0.0003 in the second part of the duodenum.The average SIBO AUC in the cohort was 1467 ppm (range 160 to 5960) meaning the average child had an increase of 0.44 points in enterocyte injury score due to SIBO while the most severe case had an increase of 1.7 points.None of the other components had a statistically significant correlation with SIBO AUC (S2 Fig).

16S rRNA analysis
A median of 15,792 raw reads per sample (range 3-184,266) was generated from the biopsy samples.After removing rare taxa and the sample with 3 total reads, a median of 15,812 reads remained (range 2,151-184,266).There were 117 unique amplicon sequence variants (ASVs) identified.ASVs in children in the top and bottom 50 th percentile for SIBO AUC are shown in  ).

Microbiome of SIBO in Bangladeshi children
There was no difference in alpha diversity of the microbiota as measured by Shannon Diversity index between SIBO positive and negative children (median 2.55 vs. 2.44, p = 0.67).A PCoA plot did not demonstrate separation between SIBO negative and positive children (Fig 3A).PERMANOVA p value was 0.74 comparing the top and bottom 50 th percentiles for SIBO AUC.Differential abundance analysis did not select any taxa predictive of SIBO positivity.In models corrected for breastfeeding status, no taxa were significantly associated with SIBO status, SIBO AUC (top vs bottom 50 th percentile), villous atrophy, or crypt hyperplasia.

Discussion
This work is the first to investigate duodenal histology and the epithelial adherent microbiota of the duodenum in relation to glucose hydrogen breath testing in a low-income setting.The main finding was that increased SIBO AUC was associated with enterocyte injury.A growing body of literature has described high rates of GHBT positivity in children from low-incomes countries [7,14,[34][35][36].Recent work shown GHBT positivity to be a risk factor for linear growth stunting and language delay [7].SIBO, as diagnosed by GHBT, has been associated with several markers of intestinal inflammation including fecal Reg1B (an anti-apoptotic, proproliferative protein secreted by damaged epithelial cells) and fecal myeloperoxidase (a marker of neutrophilic infiltrate in the gut) [14].SIBO, as determined by 16s rDNA read count in duodenal aspirate, has also been associated with systemic inflammatory cytokines [8].SIBO, defined by culture of duodenal aspirates, has been associated with increased fecal calprotectin and alpha-1-antitrypsin with the culturable SIBO microbiota leading to decreased lipid absorption when transplanted to murine intestinal cells [9].This work furthers these findings by showing a direct association of GHBT positivity and epithelial damage while showing GHBT positivity is not associated with increases in the classic markers of EED histology (i.e.crypt hyperplasia, villous blunting, and inflammatory infiltrate into the lamina propria).While the observed statistically significant effect size seems small, correlation to clinically significant outcomes such as growth or neurodevelopment is unknown.
Recent work describing the histology of EED has shown eight of the 11 parameters utilized in our analysis to have sufficient variance to be useful in the assessment of EED in children https://doi.org/10.1371/journal.pntd.0012023.g003[22].While villous atrophy, crypt hyperplasia, and lymphocytic infiltrate have been the traditional histologic hallmarks of EED, recent evidence suggests depletion of Paneth and goblet cells is also a predominating feature [3].Transcriptomic analysis has shown dysregulated lipid metabolism to be common in children with EED [37].Further, distinct bile acid signatures have been identified in children with EED [38].This cumulative evidence strongly implies a small intestinal dysbiosis is pathogenic in children with EED [3].
Several studies have used duodenal sampling to diagnose SIBO and have described microbial populations in the luminal aspirate [8,16].These studies have also shown an association between SIBO and stunted growth.This work is the first to investigate the epithelial adherent microbiota in this setting.Campylobacter was increased in patients with higher SIBO AUC although this associated was not shown in a model corrected for breastfeeding status.This likely reflects an interaction between breastfeeding and the microbiota which decreased our power to detect significant associations, especially given our low sample size.A potential link between Campylobacter and SIBO has been described whereby Campylobacter can induce anti-Cytolethal Distending Toxin (CDT) antibodies which cross react with vinculin in the enteric nervous system leading to irritable bowel symptoms.In a rat model, anti-CDT titers were associated with the development of SIBO [28].This suggests that epithelial adherent Campylobacter can precipitate SIBO by an immune mediated mechanism [39,40].The cumulative burden of Campylobacter infection in children from low-income countries has been directly associated with chronic malnutrition and stunting [41,42].Contrary to what we would have expected based on this literature, we detected less Campylobacter in patients with villous atrophy or crypt hyperplasia.Campylobacter's pathogenesis depends on multiple virulence factors responsible for adherence including CadF which binds epithelial fibronectin and is essential for invasion [43].The helical shape of Campylobacter specifically allows the organism to target crypts for adhesion and colonization [44].It may be that an unhealthy epithelium with malformed crypts represents a barrier to Campylobacter adhesion or that epithelial turnover is a defense mechanism against Campylobacter.Thus, while Campylobacter infection may be an inflammatory insult creating epithelial cell damage, the organism may struggle to adhere to a previously damaged small intestine.Campylobacter typically colonizes the distal small intestine and large intestine and thus our findings may be specific to the proximal small bowel thus limiting the conclusions we can draw on Campylobacter pathogenesis [45,46].
Our study had several notable strengths.First, SIBO was assessed by GHBT, a biomarker that has been directly associated with poor linear growth.Our pathological assessment was robust, using a validated scoring system by multiple pathologists.Finally, the collection of specimens was protocolized to preserve nucleic acid.There are also several important limitations that should be considered when interpreting findings from this analysis.Our sample size was relatively small which may limit our ability to detect smaller effect sizes, especially when covariates such as breastfeeding were included in our models.Our measure of breastfeeding (some vs none) was crude where more robust breastfeeding data would likely offer valuable insights into the relationship breastfeeding may play between the enteric microbiota and our outcomes of interest.Biopsy samples also had lower read counts compounding the effects of low sample size.Biopsy sampling involved only two samples from each patient and it is unknown if EED has patchy disease similar to celiac disease or inflammatory bowel disease which could have led to missed pathology.Also, it should be emphasized that our work examined only the mucosa-adherent microbiome obtained from biopsy samples.The luminal microbiome may play a significant role in SIBO pathogenesis and breath hydrogen test positivity which involve taxa not detected in our analysis.Finally, we excluded children with severe acute malnutrition which likely represents the most extreme end of the EED spectrum and thus there is an element of selection bias to this study.

Conclusions
The findings of this work suggest that SIBO may be associated with epithelial cell damage and Campylobacter colonization of the epithelial layer of the small intestine.Further work is needed to better understand the relationship between SIBO, EED, and Campylobacter infection as SIBO may be a modifiable risk factor leading to malnutrition and neurodevelopmental delay.

S3 Table.
The average relative abundances of all detected taxa are displayed for all children for whom 16s rRNA data is available (Column B).Also displayed are the average relative abundances for children in the top 50 th percentile (Column C) and the bottom 50 th percentile for SIBO AUC (Column D).Rare taxa that did not have at least 3 copies present in at least 5% of the samples were removed.(XLSX)

$NFig 1 .
Fig 1. Histologic comparison of SIBO positive and negative children.Histologic assessment of duodenal biopsies from children with and without SIBO demonstrated increases in intramucosal Bruner's Glands (p = 0.07) (panel A) and foveolar cell metaplasia (p = 0.07) (panel B) in the second and first parts of the duodenum, respectively, in children who were SIBO positive.Enterocyte injury directly correlated with SIBO AUC in the first part of the duodenum (p = 0.02) (panel C). https://doi.org/10.1371/journal.pntd.0012023.g001

Fig 2 .
Fig 2. Phylum comparison between SIBO status.Phylum level determinations are depicted for both the top and both 50 th percentile for SIBO AUC.DESeq2 only selected the genus Campylobacter as discriminating these two groups.https://doi.org/10.1371/journal.pntd.0012023.g002

Fig 3 .
Fig 3. Alpha and Beta diversity analysis for children with and without SIBO, villous atrophy, and crypt hyperplasia.Children with and without SIBO, villous atrophy, and crypt hyperplasia were assessed for differences in the Shannon alpha diversity index using Wilcoxon-Mann-Whitney tests.No significant differences (NS) were noted.PCoA plots were constructed for each variable without notable separation of positive and negative children.

Fig 4 .
Fig 4. Association of Campylobacter spp. with SIBO and duodenal histology.Campylobacter reads were higher in children with a SIBO AUC in the top 50 th percentile and those without villous atrophy or crypt hyperplasia.Data are displayed as median and upper limits of the interquartile range (whiskers).https://doi.org/10.1371/journal.pntd.0012023.g004

Table 1 . Baseline Characteristics of the Children with and without 16s rRNA profiling.
# Count (%)Abbreviations: LAZ, length-for-age Z score; WAZ, weight-for-age Z score; WLZ, weight-for-length Z score; SIBO, small intestine bacterial overgrowth; AUC, area under the curve; ppm, parts per million; WAMI, water & sanitation, assets, maternal education and household income.