Mucosal Microbiota Associated With Eosinophilic Esophagitis and Eosinophilic Gastritis

Abstract

Objective:

The aim of the study was to determine the mucosal microbiota associated with eosinophilic esophagitis (EoE) and eosinophilic gastritis (EoG) in a geographically diverse cohort of patients compared to controls.

Methods:

We conducted a prospective study of individuals with eosinophilic gastrointestinal disease (EGID) in the Consortium of Eosinophilic Gastrointestinal Disease Researchers, including pediatric and adult tertiary care centers. Eligible individuals had clinical data, mucosal biopsies, and stool collected. Total bacterial load was determined from mucosal biopsy samples by quantitative polymerase chain reaction (PCR). Community composition was determined by small subunit rRNA gene amplicons.

Results:

One hundred thirty-nine mucosal biopsies were evaluated corresponding to 93 EoE, 17 EoG, and 29 control specimens (18 esophageal) from 10 sites across the United States. Dominant community members across disease activity differed significantly. When comparing EoE and EoG with controls, the dominant taxa in individuals with EGIDs was increased (Streptococcus in esophagus; Prevotella in stomach). Specific taxa were associated with active disease for both EoE (Streptococcus, Gemella) and EoG (Leptotrichia), although highly individualized communities likely impacted statistical testing. Alpha diversity metrics were similar across groups, but with high variability among individuals. Stool analyses did not correlate with bacterial communities found in mucosal biopsy samples and was similar in patients and controls.

Conclusions:

Dominant community members (Streptococcus for EoE, Prevotella for EoG) were different in the mucosal biopsies but not stool of individuals with EGIDs compared to controls; taxa associated with EGIDs were highly variable across individuals. Further study is needed to determine if therapeutic interventions contribute to the observed community differences.

Graphical Abstract

An increasing body of literature emphasizes the importance of the microbiome in the pathogenesis of gastrointestinal and allergic diseases. Jensen et al (1) examined early life exposures associated with increased risk of developing eosinophilic esophagitis (EoE). Antibiotic use in infancy was identified as a major risk factor with an odds ratio (OR) of 6. This suggests a potential role for the microbiota in development of EoE. Cesarean delivery and preterm birth remain 2 elements that alter children’s microbiota and were 2 factors that add support to the potential involvement of the microbiome in eosinophilic diseases of the upper digestive tract with an OR of 1.77 and 2.18, respectively (2). A recent antibiotic study in a mouse model of EoE supported the link with early antibiotics and alteration of immune response in the esophagus (3).

The microbiome in EoE has been examined previously (4,5). Prior studies have noted the impact of treatment (6) and identified specific taxa with differential abundance by disease activity and compared to controls. The most consistent finding is an increase in Haemophilus (5–7) that was also observed in a culture-based study (8), though another study in adults with EoE did not corroborate this (9). The microbiome associated with EoE has been recently reviewed (10).

A number of factors contribute to our rudimentary knowledge of the microbiome of the upper gastrointestinal tract affected by eosinophilic gastrointestinal diseases (EGIDs). EGIDs are rare diseases making patient identification challenging, and studies to date have focused on analyses from single sites. The direct assessment of the esophagus and stomach remains poorly defined due to the requirement for endoscopy to obtain samples. Real life influences of diet and treatments can be difficult to control but clearly impact the microbiome patterns.

To address some of these issues, we performed a prospective study of children and adults with EGIDs who were recruited from 10 sites across the United States. We compared the mucosal associated bacteria from biopsies and stool samples obtained from individuals with EoE and EoG to that obtained from control volunteers. We hypothesized that individuals with EGID would have similar microbiome patterns across sites and these would be different from controls. Further, we investigated whether the degree of inflammation associated with the microbiome.

METHODS

Study Design and Participants

We performed a prospective study of children and adults with EoE and EoG as a part of the Consortium of Eosinophilic Gastrointestinal Disease Researchers (CEGIR) which is part of the Rare Diseases Clinical Research Network (RDCRN), an initiative of the Office of Rare Diseases Research, National Center for Advancing Translational Sciences. For the disease cohort, we enrolled children and adults with EoE and EoG (aged 3 years or older) from 10 hospital sites in the United States. Data were entered and managed by the Data Management and Coordinating Center associated with the RDCRN. We defined individuals with active EoE as having symptoms of esophageal dysfunction and a peak count of 15 or more esophageal eosinophils per high-powered field (HPF) and other causes ruled out. Individuals with active EoG were defined as having symptoms of gastric dysfunction and a peak of 30 or more eosinophils per HPF in at least 5 HPFs and other causes ruled out. EoE and EoG individuals who were inactive had less than the requisite numbers of eosinophils. For the control cohort, we analyzed esophageal and gastric biopsies from individuals with no EGID or other identifiable gastrointestinal inflammatory disease. This study was approved by the institutional review boards of the participating institutions via a central institutional review board at Cincinnati Children’s Hospital Medical Center. Participants provided written informed consent.

Procedures

Distal esophageal or gastric biopsy specimens were obtained at the time of endoscopy from either visually abnormal areas if present or random sites. Specimens were flash frozen and maintained at −80°C prior to extraction. We obtained data for individual’s clinical features across sites with the CEGIR questionnaire, which gathers self-reported demographics, ethnic origin, and patient symptoms. Stool was collected in a subset of individuals within a week of the endoscopy and stored at −80°C until analyzed.

DNA Extraction and Bacterial Load

DNA was extracted using the Qiagen Tissue kit per manufacturer’s instructions. Extracted DNA was diluted 1:40 and assayed in triplicate (4 μL template, dilution factor of 10) to determine the total bacterial load (TBL) using the assay described by Nadkarni et al (11).

16S Amplicon Library Construction

Bacterial profiles were determined by broad-range amplification and sequence analysis of 16S rRNA genes following previously described methods (12–14). In brief, amplicons were generated using primers that target approximately 300 base pairs of the V1V2 variable region of the 16S rRNA gene. PCR products were normalized using agarose gel densitometry and pooled at approximately equimolar amounts. The pooled amplicons were purified and concentrated using a DNA Clean and Concentrator Kit (Zymo, Irvine, CA). Pooled amplicons were quantified using Qubit Fluorometer 2.0 (Invitrogen, Carlsbad, CA). The pool was diluted to 4 nM and denatured with 0.2 N NaOH at room temperature. The denatured DNA was diluted to 20 pM and spiked with 10% of the Illumina PhiX control DNA prior to loading the sequencer. Illumina paired-end sequencing was performed on the MiSeq platform using a 500-cycle version 2 reagent kit.

Analysis of Illumina Paired-End Reads

Illumina MiSeq paired-end reads were aligned to human reference genome Hg19 with bowtie2 and matching sequences discarded (15,16). As previously described, the remaining nonhuman paired-end sequences were sorted by sample via barcodes in the paired reads with a python script (13). Sorted paired end sequence data were deposited in the National Center for Biotechnology Information (NCBI) Short Read Archive under accession number SRP156484. The sorted paired reads were assembled using phrap (17,18). Pairs that did not assemble were discarded. Assembled sequence ends were trimmed over a moving window of 5 nucleotides until average quality met or exceeded 20. Trimmed sequences with more than 1 ambiguity or shorter than 200 nt were discarded. Potential chimeras identified with Uchime (usearch6.0.203_i86linux32) (19) using the Schloss (20) Silva reference sequences were removed from subsequent analyses. Assembled sequences were aligned and classified with SINA (1.3.0-r23838) (21) using the 418,497 bacterial sequences in Silva 115NR99 (22) as reference configured to yield the Silva taxonomy. Identical taxonomic assignments were clustered to produce Operational Taxonomic Units. This process generated 31,022,761 sequences for 303 samples (average sequence length: 315 nt; average sample size: 102,385 sequences/sample; minimum sample size: 24,698; maximum sample size: 298,738). The median Goods coverage score was ≥99.86% at the rarefaction point of 24,698. The software package Explicet (v2.10.5, www.explicet.org) (23) was used for data display and analysis (rarefied values for median Good’s coverage and Shannon diversity).

Statistical Analyses

Alpha Diversity

Explicet was used to bootstrap at a rarefaction point of 24,698 to obtain median values of Shannon H Diversity Index measures for each sample. Shannon H diversity and load calculated for individuals with EoE and EoG within disease status (active, inactive) and for control individuals. Active disease assignment was based on published cutoffs for histological assessment of biopsy material (≥30 eosinophils per HPF in EoG and ≥15 in EoE; average of five HPF). Log normal linear regression was used to assess differences in disease status for each diagnosis type (EoE and EoG) for both load and Shannon H diversity.

Beta Diversity

Permutational multivariate analysis of variance (PERMANOVA) was used to assess differences in community composition between therapeutic interventions. Morisita-Horn values were calculated to assess overall community composition in EoE and EoG samples. MH values range from 0 (no similarity) to 1 (completely similar). Principal coordinate analysis was used to assess community differences between samples using 1 – MH for dissimilarity. Ordination plots were color coded base on disease status. Relative abundance (RA) was calculated by dividing taxa count by total library size. Stacked bar charts were used to demonstrate the community composition of each diagnosis and disease status for the most abundant taxa. Taxa whose RAs were less than 1% for every sample were excluded from formal statistical tests. Negative binomials with a log link were used to determine differences in RA between disease status. Statistical analyses were carried out using R version 4.0.2 (www.R-project.org) (2020–06-22).

RESULTS

Study Populations and Samples

Study Population

We studied esophageal biopsies from 111 individuals (93 EoE; 18 controls) and gastric biopsies from 28 individuals (17 EoG; 11 controls) across CEGIR sites. Individual characteristics are provided in Table 1. Concurrent stool samples were collected from 96 individuals (69 EoE, 13 esophageal controls, 11 EoG, and 3 gastric controls, Table 1). Treatment information including swallowed/topical steroids, proton pump inhibitors, and food elimination for eosinophilic disease is summarized in Table 1.

TABLE 1.

Summary of demographic and clinical information

	EoE (n = 93)	HCE* (n = 18)	EoG (n = 17)	HCG* (n = 11)
F:M	26:67	10:8	7:10	4:7
White:Other	85:8	15:3	14:3	9:2
Act.:Inact.†	30:63	–	6:11	–
Age	10.6 (5.7, 16.9)	14.2 (6.9, 35.1)	15.6 (11.5, 17.6)	10.0 (8.9, 16.7)
Eos. count
Active	27 (21, 46)	–	114 (71, 124)	–
Inactive	0 (0, 6)	–	9 (0, 16)	–
Medications
None	6	18	3	11
ppi	63	–	21	–
Steroids	61	–	1	–
Food	55	–	2	–
Stool	69	13	11	3

Disease activity was defined as an average of ≥15 eosinophils/high powered field for EoE, and ≥30 eosinophils/high powered field.

^*HCE = control esophageal participant; HCG = control gastric participant.

^†Act. = active disease; Inact. = inactive disease.

Bacterial Load and Diversity in EoE and EoG

We first examined TBL and Shannon diversity of children and adults with EoE and EoG compared to controls and each other. No significant differences in bacterial load or diversity were found when comparing individuals with either inactive or active EoE to controls (Fig. 1A and C). When comparing results from individuals with inactive or active EoG to controls the load trended to be higher and the diversity lower in controls compared to EoG regardless of disease activity (Fig. 1B and D). Comparing TBL from individuals with EoE to EoG, loads were generally higher in those with EoE whereas Shannon diversity was generally higher in those with EoG. Age did not impact alpha diversity with similar values observed in pediatric and adult subjects.

FIGURE 1.

Total bacterial load and Shannon diversity by group. (A) Total bacterial load for esophageal biopsy from controls and EoE subjects. EoE subjects are divided relative to disease activity [active disease ≥15 eosinophils (eos)/high-powered field (HPF)]. (B) Total bacterial load from gastric biopsy grouped by controls and disease status (active disease ≥30 eos/HPF). (C) Shannon diversity from the same samples as (A). (D) Shannon diversity for gastric biopsies. Groups are the same as (B). (E) Stacked bar chart of community composition for taxa with greater than 1 percent relative abundance in the esophageal biopsies. Relative abundance is shown for the 3 groups; controls (n = 18), inactive EoE (n = 63), and active EoE (n = 30). (F) Distribution of relative abundance for the 4 taxa identified as statistically significant across the 3 groups. Boxes indicate 25–75 percentile and the line is the median. Whiskers represent the highest value within [Q3,Q3+1.5 × IQR] or the lowest value [Q1 − 1.5 × IQR,Q1], and outliers are shown as points. (G) Stacked bar chart of community composition for taxa with greater than 1 percent relative abundance in the gastric biopsies. Relative abundance is shown for the 3 groups; controls (n = 11), inactive EoG (n = 11), and active EoG (n = 6). (H) Distribution of relative abundance for the 2 taxa identified as statistically significant across the 3 groups (as in F). EoE = eosinophilic esophagitis; EoG = eosinophilic gastritis.

Prominent Taxa Associated with EoE and EoG

We next examined differences between specific taxa in individuals with EoE and EoG. Compared to controls (n = 18), individuals with active EoE (n = 30) had more altered community composition than inactive EoE (n = 63) (Fig. 1E). Prevotella and Neisseria accounted for the bulk of RA gain with the decrease in Streptococcus RA in active EoE although this was not statistically significant (Figure 1, Supplemental Digital Content 1, http://links.lww.com/MPG/D38). When comparing the distribution of RA using a negative binomial distribution across the active and inactive EoE compared to controls, 4 taxa were identified as significantly different (Streptococcaceae, Streptococcus, Gemella, Actinobacillus; P < 0.01 for all; Fig. 1F). The primary difference for Streptococcus is reduction in active disease. Streptococcaceae was expanded primarily in inactive EoE. Gemella and Actinobacillus were expanded in EoE irrespective of disease status. The RA for Streptococcaceae, Gemella, and Actinobacillus was generally low with a subset of individuals having high RA.

Analysis of EoG mucosal taxa revealed similar findings as EoE (Fig. 1G). Prevotella was the predominant taxon observed in controls (n = 11). In EoG individuals, Streptococcus RA was increased irrespective of disease activity (inactive n = 11, active n = 6). Two taxa were significantly different between groups; Propionibacterium (P < 0.03) was reduced in EoG and Leptotrichia (P < 0.01, Fig. 1H), which increased consistently in active EoG. The limited numbers for EoG, particularly active disease, decreased our power to identify specific changes. The distribution of RA for prominent taxa are provided in Figure 2, Supplemental Digital Content 2, http://links.lww.com/MPG/D39.

Taxonomic composition was consistent between adult and pediatric individuals (Figure 3, Supplemental Digital Content 3, http://links.lww.com/MPG/D40). The expansion of Streptococcus RA in the adult control esophageal biopsy may be due to geographical differences given the high RA Streptococcus was observed at a single site.

Community Comparison

Community ordination in esophageal biopsies was driven by a small number of taxa (Fig. 2A). Streptococcus was a primary determinant, and exerted the most influence on the first component in the ordination plot (r² = 0.92). The second component was influenced by a collection of taxa with Prevotella (r² = 0.95) and Veillonella (r² = 0.58) inversely related to Neisseria (r² = 0.82) and Fusobacterium (r² = 0.62). Controls were intermixed with the EoE individuals suggesting that the esophageal microbiome is similar across individuals with EoE and controls at the community level. However, the EoE individuals with active disease have greater values in component 1, which suggests a subtle differentiation between communities. The primary difference observed was an increase in taxa with Gram-negative cell wall structure in EoE samples, which could influence inflammatory processes.

FIGURE 2.

Ordination plots for communities associated with esophageal (A) and gastric (B) biopsies. Points represent individual samples and are color coded by group; red = control, green = inactive disease, and blue = active disease. Taxa important for vector loading are shown.

Community composition in gastric biopsy (Fig. 2B) identified a similar pattern, but with more consistent communities observed in the control samples. Prevotella had the most influence on the first component (r² = 0.94). A collection of organisms were most associated with component 2. Neisseria (r² = 0.85) and Fusobacterium (r² = 0.56) were inversely related to Streptococcus (r² = 0.94) and Veillonella (r² = 62). For gastric biopsies, there was more differentiation between the controls and EoG individuals, regardless of disease activity status. The majority of individuals with EoG segregated from the controls, which may suggest a more prominent role at the community level in EoG.

Impact of Therapeutic Interventions on Bacterial Communities

Since past works have identified an impact of therapeutic interventions on the local host microbiome, we next analyzed the bacterial communities derived from individuals according to their treatment state (Fig. 3 and Table 1). This analysis was limited to EoE due to smaller numbers available for EoG. Although proton pump inhibitors appeared to shift community composition (lower Streptococcus) that was mitigated when steroids were prescribed. When examined using PERMANOVA, this observation (P = 0.48), as well as all other comparisons between treatments and no treatments were not statistically significant.

FIGURE 3.

Impact of therapeutic interventions. Summary of community composition for EoE subjects by treatment group for the 3 primary management strategies [proton pump inhibitors (PPI), steroids, and food elimination]. The number of subjects included in each group is indicated. Active and inactive disease subjects were merged for this presentation due to low numbers for some groups. EoE = eosinophilic esophagitis.

Fecal Microbiome

When we determined the microbiome from stool for a subset of individuals in the study (Table 1) and compared them to their associated mucosal biopsy, we found that populations present in the biopsy did not correlate with those measured in paired stool samples. In contrast, principal coordinate analysis demonstrated a highly similar fecal community in EGIDs and controls with the dominant taxa (Faecalibacterium, Lachnospiraceae, and Bacteroides) being typical of human stool samples (Figure 4, Supplemental Digital Content 4, http://links.lww.com/MPG/D41). Mucosal samples were more distributed based on the RA of Prevotella and Streptococcus, which did not appear to associate with any of the subject groups.

DISCUSSION

This study is the first multi-center examination of microbiota associated with mucosal communities from patients with either EoE or EoG. These data provide the most comprehensive examination of these 2 groups to date and expand the list of taxa potentially associated with these diseases. In both the esophagus and stomach, a prominent taxon (Streptococcus for EoE; Prevotella for EoG) represented the primary community member. While community composition was similar at each anatomical site between controls and individuals with EGID there were subtle differences that may have important clinical implications. The taxa that replaced the dominant community member observed in controls varied across the EGID individuals, which likely contributes to the lack of strong statistical signals in the study. This is a general issue for microbiome studies where high individuality increases variance in observations. The impact of treatment on community composition is also a critical observation of this study, which was not feasible in early single center studies due to more similar clinical practice and size. The complexity of clinical interventions will require prospective studies that better control for the heterogeneity of clinical intervention, or specifically test longitudinal analysis of specific therapeutic interventions within individuals for community impacts.

The expansion of groups with Gram-negative cell structure may represent an important feature in EoE. These taxa are generally representative of organisms considered commensal, but would still increase cell wall components associated with inflammation (eg, lipopolysaccharide), which might further exacerbate inflammatory signals in the context of disease. The expansion of Proteobacteria (specifically Neisseria and Haemophilus) has been reported previously (4,5). However, in the current study neither of these taxa was significantly different from controls, which has also been noted in one study of adults with EoE (9). There was a trend consistent with earlier findings with higher median RA observed for both taxa in individuals with EoE. The prior studies were single center, which may suggest local factors (eg, clinical practice) influence the altered communities observed.

Many microbiome studies take advantage of stool sampling to assess more proximal mucosal sites in the gastrointestinal (GI) tract. Here, we took advantage of having analysis of both a mucosal sample and fecal matter at the same time to compare bacterial populations. The availability of fecal samples from a subset of individuals allowed us to assess the utility of a noninvasive sampling strategy to investigate EoE and gastritis. While the ease of collection is a potential benefit, the community structure was more similar between groups than observed for the mucosal samples. In addition, the fecal communities were very distinct from the biopsy sample collected from the site of inflammation. The distinct communities limit the ability to understand specific microbial determinants that might impact disease of the esophagus or stomach. Saliva has shown promise as a noninvasive sample type that may identify cases of active EoE (24).

The primary limitation for this study was the inability to examine therapeutic interventions systematically. The heterogeneity of clinical intervention makes this challenging, and longitudinal designs are probably the most relevant. However, specimen acquisition is problematic for longitudinal designs using standard approaches. Alternatives that are less invasive than traditional endoscopy are under development that should address this issue. In addition, sample size was small for EoG limiting some analyses. Paired stool samples were not available from all individuals. The strengths of this study are the large sample size drawn from multiple centers. The standardized evaluation of clinical samples and participant information is an additional strength.

CONCLUSION

This study of EoE and EoG mucosal microbiota represents a substantial increase in number of individuals examined. There are specific changes in the taxa associated with active disease in both EoE and EoG, which appear modified by therapeutic interventions. The complexity of intervention likely impacts the overall study, which did not select biopsies based on therapeutic intervention. These data provide a strong foundation for prospective studies to disentangle the role of disease activity and therapeutic interventions on the EGID microbiome.

What Is Known

• Microbial communities are associated with allergic conditions and may impact disease course.

• Single center studies have identified specific taxa associated with eosinophilic esophagitis (EoE).

What Is New

• Large multicenter cohort of EoE and eosinophilic gastritis individuals.

• First description of bacterial communities associated with gastric biopsy using microbial ecology approaches.

• Specific taxa were associated with eosinophilic gastrointestinal disease.

• Biopsy communities do not correlate with stool communities.