Dynamics of tissue and fecal microbiota in active Crohn’s disease and their ability to predict disease state

Abstract

BACKGROUND

Simultaneous assessment of gut microbiota in stool and tissue samples is crucial for a better understanding of their role in Crohn’s disease (CD), yet most reports have focused on fecal microbiota alone. Additionally, gut microbiota may serve as a clinically useful diagnostic biomarker of CD although data on this is limited.

AIM

To evaluate gut microbiota in tissue and stool samples in patients with active CD to understand the structure and function compared to healthy controls (HC). We also assessed their utility as a diagnostic biomarker of CD.

METHODS

Adult patients with active CD and HC were prospectively recruited for this study. The clinical and investigation details were recorded. Rectal mucosal biopsy and stool samples were obtained to assess the bacterial population. DNA was extracted, the V3-V4 region of the 16S rRNA gene was amplified, and library preparation was done and sequenced on the Illumina MiSeq platform. The bacterial diversity, composition, dysbiosis, predicted function, and predictors of disease state were estimated using the QIIME 2 pipeline and R packages.

RESULTS

We recruited 66 patients with CD (age 39.7 ± 11.1 years, 65.2% males) and 69 HC. Comparison of tissue with fecal microbiota in active CD showed significant differences in composition and predicted function. Both tissue and fecal microbiota from active CD showed reduced microbial diversity and compositional differences compared to HC, and disease state was a key determinant of bacterial population. Differences (CD vs HC) were noted in the abundance of several predicted synthetic and degradation pathways in both tissue and stool bacteria. Tissue microbiota was a superior predictor of active CD than stool (area under receiver operating characteristic curve 0.8 vs 0.62).

CONCLUSION

Gut microbial characteristics revealed significant structural and functional differences between CD and HC in both tissue and stool. Tissue bacteria performed well as a microbial biomarker for clinical diagnosis of CD.

Key Words: Fecal microbiota; Mucosal microbiota; Crohn’s disease; Dysbiosis; Biomarker; Functional difference

Core Tip: Gut microbiota is emerging as a key player in the pathogenesis of Crohn’s disease (CD). The mucosal gut microbiota differs in structure and function compared to fecal microbiota, yet most studies have focused on fecal microbiota alone. We evaluated both mucosal and fecal microbiota in adults with active CD and compared this with the healthy controls. Our results show striking differences in bacterial structural and functional characteristics between mucosal and fecal samples in Crohn's disease. Additionally, the mucosal microbiota proved to be a good predictor of disease state and can serve as a novel potential biomarker for diagnosis of CD.

INTRODUCTION

Gut microbiota is currently recognized as a key player in the pathogenesis of inflammatory bowel disease (IBD) - both Crohn’s disease (CD) and ulcerative colitis, through their influence on host immune response and metabolic pathways in the gut[1,2]. Despite the substantial number of research publications, our understanding of the exact role of the gut microbial community in the development and progression of IBD is still limited[3]. The prevalence of CD is increasing in several developing nations. At the same time, microbial data from these regions is relatively limited but essential to provide further insights into the emergence of this disease[4,5].

The microbial populations in stool and on mucosal surfaces are not similar[3,6-8]. The mucosal bacteria adherent to the epithelial surface are perhaps more likely to influence immune function and inflammatory activity than fecal bacteria. Hence, studying fecal microbiota alone would not provide adequate information on microbial perturbations in disease. There is relatively sparse information on tissue microbiota compared to fecal microbiota in CD, especially in adults[3,9]. Mucosal samples require endoscopy, which is an invasive procedure and hence not easy to obtain compared to the relative ease of getting stool samples. Several studies on mucosal microbiota have included fewer subjects, contain a mix of active and inactive cases, and lack simultaneous evaluation of stool samples[3,6]. Multiple reports suggest mucosal microbiota in the rectum are similar to other parts of the colon[10-14]. As colonoscopy is more invasive, performing only rectal mucosal examination for microbiota study is an attractive and practically feasible option to make management-based decisions. In this study, we evaluated the composition and predicted function of rectal mucosal and fecal microbiota of adult cases with active CD and compared the observations with healthy control subjects.

MATERIALS AND METHODS

We conducted a prospective study from 2017 to 2022. Adult patients were diagnosed with CD based on a combination of clinical, endoscopic, histological, and radiological investigations and where appropriate, exclusion of infectious etiology[15]. As there is no single gold standard for CD, the above strategy for diagnosis has been suggested by the Asia Pacific Consensus. Clinical features include abdominal pain, diarrhoea, weight loss and nutritional deficiencies. Endoscopy shows ulcerations and strictures with intervening normal mucosa in the intestine and histopathological features include chronic inflammation with architectural distortion. Granulomas may be noted in 50%-60% of cases of CD. Imaging features include intestinal wall thickening and enhancement, strictures, fistula, engorged blood vessels, and creeping fat. CD activity index (CDAI) was used to determine disease activity. Those with active CD (CDAI > 150) were included after obtaining written informed consent[16,17]. Patients who used antibiotics or steroids in the past two months and those who were on biological therapy were excluded. The demographic and clinical details and reports of investigations and treatment were recorded on a patient information sheet. Patients who underwent colonoscopy to assess for colonic pathology and were found to have normal mucosal surface and no chronic illness (including irritable bowel syndrome) were recruited as healthy controls (HC). The sample size estimation method for microbial comparison studies is not well established, and we aimed to recruit at least fifty cases and controls each. The study was approved by the institute review board and ethics committee.

DNA extraction and library preparation

Freshly passed stool sample was collected in a sterile container, transported on ice to the laboratory within 1 hour, and stored at -70 °C. The sample was collected preferably before or at least five days after the colonoscopy to avoid the effect of bowel preparation. For colonoscopy, the bowel preparation was done with polyethylene glycol 137.15 g in 2 liters of water on the day before colonoscopy the same was repeated on the day of the colonoscopy. About four bits of rectal mucosal biopsy sample were collected using biopsy forceps and were stored at -70 °C until further processing. The tissue samples were used for microbiota estimation and histopathology was not done. DNA extraction from stool was carried out using the QIAamp Fast DNA stool mini kit (Qiagen) and from tissue using the QIAamp DNA mini kit (Qiagen), according to the manufacturer’s instructions[18]. Amplicons spanning 16S rRNA gene V3-V4 (variable regions 3 and 4) using primers 341F 5’-overhang-CCTACGGGNGGCWGCAG-3’ and 785R 5’-overhang-GACTACHVGGGTATCTAATCC-3’ were produced following the established Illumina protocol (Illumina, 16S Metagenomic Sequencing Library Preparation Protocol Part # 15044223 Rev. B)[19]. Libraries were sequenced on the Illumina MiSeq platform (next generation sequencing). The details of stool and tissue DNA extraction and library preparation are provided in the Supplementary material.

Bioinformatics and statistical analysis

Demultiplexed fastq files were imported into the quantitative insight into microbial ecology pipeline for further downstream processing, which included quality filtering and denoising with the divisive amplicon denoising algorithm 2 plugin, followed by merging of paired reads and chimera removal[20,21]. Taxonomy assignment was performed using the trained Silva classifier 138, which incorporated the specific primer sequence that was used for amplification[22]. The operational taxonomic units (OTUs) that were taxonomically unclassified at phylum rank or were not assigned to bacterial lineages were excluded from further analysis after confirmation. Shannon and Faith’s phylogenetic diversity metrics were used to estimate alpha diversity. Bray-Curtis and Unifrac metrics assessed the dissimilarity between sample groups. Principal coordinates analyses were carried out and plotted to explore the diversity between samples visually. Differences between groups were estimated by non-parametric statistical tests with Benjamini-Hochberg false discovery rate correction for P values. To study the proportion of variance in microbiome composition stratified by grouping variables, Permutational Multivariate Analysis of Variance was done based on the Bray-Curtis dissimilarity distance matrix. Functional prediction of pathways was performed using the phylogenetic investigation of communities by reconstruction of unobserved states (PICRUSt) plugin in quantitative insight into microbial ecology version 2.5.1, based on the MetaCyc pathway database[23,24]. The microbial sequence data would be made available for research purposes on request. All statistical analyses were performed in R version 4.2.0 with appropriate data reduction, analysis, and visualization packages[25-31].

Random forest prediction models were utilized to analyze the differences in the microbiome composition and identify predictors of disease state[32]. Error optimization was done by performing 5 iterations of the models with 10-fold cross-validation using centralized log-ratio transformed genus-level relative abundances. Based on the average error rates obtained from 10-fold cross-validation runs, the minimum error rate (with standard deviation) was calculated, and the number of features corresponding to this model was chosen as the optimal dataset to proceed for each of the groups (active disease and HC) studied. The receiver operating characteristic (ROC) curve was constructed to evaluate the model’s performance, and the area under the ROC curve was used to designate the ROC effect[32]. The mean cross-validation variable importance plot was constructed based on the mean decrease in accuracy values from random permutation.

Microbial dysbiosis was assessed using a cloud-based LOcally linear unbiased dysbiosis (CLOUD) test and microbial dysbiosis index (MDI)[13,33]. CLOUD test was carried out using the dysbiosis R packages[34,35]. This approach assesses the ecological similarity of a set of patient’s microbiome to that of healthy study participants. This allows the characterization of an individual microbiome based on its ecological similarity and stability to healthy participants; a cut-off was derived based on the median value and standard error of the estimate to characterize the microbiome into sets of “dysbiotic” or “healthy-like”. MDI was calculated as described by Gevers et al[13] [log of (total abundance in organisms increased in CD) over (total abundance of organisms decreased in CD) for all samples] to assess dysbiosis[13].

RESULTS

We recruited 66 patients with active CD during the study period. Their mean age was 39.7 ± 11.1 years, and 65.2% were males. Disease onset was noted in most individuals in the 17-40 years of age and, the ileocolonic location was most frequent. Aggressive disease behavior was noted in about 44% of cases. Perianal disease was present in 16.7% cases. The baseline characteristics of the study subjects and controls are shown in Table 1. Sixty-nine adult healthy control subjects were recruited. Their mean age was 39.4 ± 9.7 years, and 68.1% were males. There was no significant difference in age, gender, or dietary pattern between cases and controls. 16SrRNA sequencing was performed on rectal mucosal tissue samples of 66 cases with active CD and 69 HC. Stool samples were provided by 59 cases with active CD and 47 HC. The median number of sequences per tissue sample was 14395. Based on the alpha rarefaction analysis, a sequencing depth of 4000 was chosen for further diversity analysis, and 420000 features were retained in 105 (active CD: 53, HC: 52) samples. The median sequence per stool sample was 25146 and was rarefied to a sequencing depth of 8000 with 848000 features retained in 106 (100%) samples.

Table 1 Baseline characteristics of the study subjects.

Characteristics	Active CD (n = 66)	Healthy (n = 69)
Age (year), mean ± SD	39.7 ± 11.1	39.4 ± 9.7
Gender (males)	65.2%	68.1%
CDAI score	206.3 ± 48.8	-
Fecal calprotectin (microgram/gram)	611.8 ± 798	-
C reactive protein (mg/L)1	18.7 (8.6-49.8)	3.1 (1-5.1)
Age of onset of CD	≤ 16 years: 1.5%	-
	17-40 years: 59.1%
	> 40 years: 39.4%
Location of CD	Ileal: 42.4%	-
	Colonic: 10.6%
	Ileocolonic: 47%
Behaviour of CD	Non-stricturing and non-penetrating (B1): 56.1%	-
	Stricturing (B2): 37.9%
	Penetrating (B3): 6%
Perianal disease	16.7%	-
Smoking	6.1%	5.8%
Residence	Urban: 53%	Urban: 68.1%
	Village: 47%	Village: 31.9%
Diet	Vegetarian: 4.5%	Vegetarian: 7.3%
	Mixed: 95.5%	Mixed: 92.7%

¹Median with interquartile range.

CD: Crohn’s disease; CDAI: Crohn’s disease activity index.

Comparison of tissue microbiota and fecal microbiota in patients with active CD

Fifty-five patients with active CD provided both tissue and stool samples, allowing for comparison of microbiota at the two sites in them. There was no significant difference in the median alpha (within-sample) diversity (Figure 1A and B). The beta diversity estimates showed differential clustering of microbiota from the two sites on principal coordinates analyses plots (Figure 1C and D). The site of the sample (tissue or stool) explained a considerable proportion of variance in the microbial composition (Figure 1E). The relative abundance of a number of bacterial genera was significantly different between tissue and stool samples (Figure 1F). Organisms more common in tissue were Staphylococcus, Sphingomonas, Pelomonas, Flavobacterium, Corynebacterium, etc. while Lactobacillus, Lachnospira, Intestinibacter, Faecalibacterium, Bacteroides, etc. were commoner in stool samples. On PICRUSt analysis at the L2 level, differences were noted in several predicted functional pathways. Most of these pathways were more abundant in stool, possibly due to a higher number of reads from the stool sample than tissue (Figure 1G). However, amino-acid biosynthesis, glycan pathways/cell structure biosynthesis, certain amino acid biosynthesis, and cofactor biosynthesis/polyprenyl biosynthesis pathways were more abundant in tissue bacteria (Figure 1G).

Figure 1 Comparison between tissue and fecal bacteria in patients with active Crohn’s disease. A and B: Violin plot showing Shannon’s alpha diversity and Faith phylogenetic alpha diversity; C and D: Principal coordinates analyses plots based on weighted Unifrac distances and Bray-Curtis dissimilarity estimates; E: Factors affecting bacterial composition; F: Bacterial genera with significant differences in relative abundance between tissue and stool samples; G: Differentially abundant functional pathways between tissue and stool sample.

Comparison of tissue microbiota between active CD and HC

The alpha (Faith’s) diversity was lower in CD compared to controls (Figure 2A and B). The tissue samples seemed to cluster based on disease status when measured by Bray-Curtis distance but not by weighted Unifrac distances (Figure 2C and D). Disease status explained a significant proportion of variance in microbial composition between the two groups (Figure 2E). The relative abundance of Ruminococcus, Blautia, Agathobacter, Marvinbryantia, and subgroups of Lachnospiraceae/Clostridia/Christensenallaceae was lower in active CD compared to controls (Supplementary Figure 1). CLOUD test showed significant dysbiosis in only 7 active cases (Figure 2F) and MDI in only 3 active cases (Supplementary Figure 2). On PICRUSt analysis, methylthioadenosine cycle and methanogenesis from acetic acid pathway were less abundant in active CD than HC (Figure 2G).

Figure 2 Comparison between tissue bacteria in patients with active Crohn’s disease and healthy controls. A and B: Violin plot showing Shannon alpha diversity and Faith phylogenetic alpha diversity; C and D: Principal coordinates analyses plots based on weighted Unifrac distances and Bray-Curtis dissimilarity estimates; E: Factors affecting bacterial composition; F: Presence of dysbiosis based on cloud-based LOcally linear unbiased dysbiosis test; G: Differentially abundant functional pathways between Crohn’s disease and healthy controls.

Comparison of fecal microbiota between active CD and HC

Alpha diversity was lower in cases compared to controls (Figure 3A and B). The tissue samples appeared to cluster with better resolution based on disease status (Figure 2D) than stool samples (Figure 3C and D) on Bray-Curtis dissimilarity distances. The disease status (active disease vs healthy state) explained a significant proportion of microbial composition variance (Figure 3E). At the genus level, the relative abundance of Faecalibacterium, Prevotella, Romboutsia, and Terrisporobacter was lower in active CD than in HC (Supplementary Figure 3). Dysbiosis was noted in 32.2% CD cases based on the CLOUD test (Figure 3F) and 33.9% cases on MDI test (Supplementary Figure 4). There was lower abundance of methanogenesis from acetic acid, glycerol degradation to butanol, glycogen biosynthesis I, L glutamate and glutamine biosynthesis, phenylacetate degradation, etc. pathways in active CD compared to HC (PICRUSt level 6, Figure 3G).

Figure 3 Comparison between fecal bacteria in patients with active Crohn’s disease and healthy controls. A and B: Violin plot showing Shannon alpha diversity and Faith phylogenetic alpha diversity; C and D: Principal coordinates analyses plots based on weighted Unifrac distances and Bray-Curtis dissimilarity estimates; E: Factors affecting bacterial composition; F: Presence of dysbiosis based on cloud-based LOcally linear unbiased dysbiosis test; G: Differentially abundant functional pathways between Crohn’s disease and healthy controls.

Similarity and differences in directional changes of tissue and fecal microbiota in CD compared to controls

The log2 fold change was calculated for the median relative abundance of taxa in stool and tissue. The results showed that Bacteroides, Fusobacterium, and Streptococcus were increased in both stool and tissue in CD, while Prevotella, Enterobacter, Lachnospiraceae, and Roseburia were decreased in both compared to controls (Figure 4). Additionally, Christensenellaceae, Ruminococcus, Rothia, and Clostridium were increased in stool but decreased in tissue.

Figure 4 Bacterial genus increased and decreased in stool and tissue samples. A: Bacterial genus increased and decreased in stool samples of patients with active Crohn’s disease compared to healthy controls; B: Bacterial genus increased and decreased in tissue samples of patients with active Crohn’s disease compared to healthy controls.

Microbial population as a marker of disease state: On random forest analysis, Streptococcus, Phocaeicola and Stenotrophomonas predicted active CD while Pseudomonas, Veillonella, Clostridium innocuum group, Rothia and Ruminococcus gnavus group predicted healthy state in tissue samples with high variable importance (Figure 5A). The predictors of disease in stool samples are shown in Figure 5B. Based on random forest analysis, 40 OTUs were selected from tissue samples to distinguish disease from HC. The ROC had an area under curve (AUC) of 80.69% (72.11-89.28) for the training set and 78.95% (62.09-95.8%) for the validation set (Figure 5C). For the stool samples, 65 OTUs were selected, and the AUC was 61.75% (49.23-74.28) for the training set and 65% (42.76-82.74) for the validation set (Figure 5D). The above analysis showed tissue microbiota to be a good diagnostic biomarker for active CD, while the performance of fecal microbiota was suboptimal.

Figure 5 Bacterial predictors of disease state (active Crohn’s disease and healthy state) in tissue and stool samples. A and B: Random forest analysis showing bacteria predicting active Crohn’s disease and healthy state in tissue sample and stool sample; C: Receiver operating characteristic curve showing ability of top 40 operational taxonomic units in tissue sample selected based on random forest analysis to discriminate between disease and healthy state; D: Receiver operating characteristic curve showing ability of top 65 operational taxonomic units in stool sample to discriminate between disease and healthy state. AUC: Area under curve; CI: Confidence interval.

Disease behavior and gut microbiota

We compared the gut microbial characteristics of patients with aggressive disease (B2, B3) vs non-stricturing/non-penetrating disease (B1). The fecal microbiota showed no significant difference in alpha diversity between the two groups (Shannon P 0.53; Faith’s P 0.62). On random forest analysis, Enterobacter, Roseburia, Lachnospira, Clostridia_UCG-14, Dialister, Howardella, and Sutterella were among the important organisms predicting an aggressive disease state. In contrast, organisms such as Enterococcus, Klebsiella, and Shigella-Escherichia and Mitsuokella predicted B1 disease (Supplementary Figure 5). Similar to stool, tissue microbiota did not show significant difference in alpha diversity (Shannon P 0.1; Faith’s P 0.4) between the two groups. Random forest analysis showed Haemophilus, and Alloprevotella as predictors of B2/B3 disease behavior and Sutterella, Shigella-Escherichia, Prevotella, Megamonas, Streptococcus, Pseudomonas, and Ruminococcus_gnavus_group predicted B1 disease behavior (Supplementary Figure 6).

DISCUSSION

The current study comprehensively assessed the structure and predicted function of tissue and fecal microbiota in adult patients with active CD from a region where this disease is in the rising phase of its epidemiological curve[36,37]. The differences in bacterial composition were quite pronounced between the tissue and stool samples of active CD, and functional differences were also noted. Additionally, tissue microbiota had good accuracy in predicting disease state, while the performance of stool tests was poor. These findings highlight the inadequate information obtained from stool samples and the need to evaluate tissue samples in gut microbiota studies. Bacterial population characteristics at both sites differed significantly between CD and HC. We only included active cases with CD to capture the bacterial population associated with active ongoing inflammation and more likely to be associated with the disease process[38,39].

Several factors, such as geographical location, diet, genetic factors, etc., affect microbial composition in the gut[40,41]. This makes it challenging to identify disease-specific changes in the microbiome. By selecting patients from a defined geographical location and having a control population with similar age and gender distribution and dietary patterns, we have attempted to balance out the effects of these factors in our study. Our results showed disease state as a key predictor of microbial population in both stool and tissue samples, further supporting the role of microbiota in disease pathogenesis. This, in turn, may also have potential implications for microbiota-based therapy for IBD[42].

Dysbiosis in CD has been characterized by a reduction in symbionts like F prausnitzii, Bifidobacterium, Clostridium, etc., and an increase in pathobionts like Pseudomonas, Enterobacteria, etc.[43]. The state of immune tolerance is perturbed and shifts towards a pro-inflammatory state in the presence of microbial dysbiosis and genetic susceptibility[1,44]. Dysbiosis is present in both stool and mucosal samples, but the extent of dysbiosis may be greater in mucosal samples and may provide a superior footprint of bacteria discriminating disease and healthy states[13,45]. An interesting observation in several studies has been the similarity of microbial structure between different parts of the colon[10-14]. Hence, assessing rectal mucosal microbiota may be sufficient and a less invasive option than sampling colonic mucosa at different locations[13]. Therefore, we studied rectal mucosal microbiota among our cases.

As the 16S rRNA sequencing assesses relative rather than absolute abundance, the information on increased or decreased relative abundance may not necessarily reflect an association with disease, and the compositional nature of this data prevents us from employing standard statistical tests. A better method is to transform data like in the MaAslin pipeline or use machine learning methods like random forest analysis to identify organisms predicting the presence of disease or healthy state. Our data showed organisms such as Pseudomonas, Streptococcus, Ruminococcus, Prevotella, Veillonella, Clostridium subgroup, Flavobacterium, Klebsiella, etc. as significant discriminants between diseased and healthy states. Several species of genus Streptococcus, Klebsiella, Escherichia, Shigella, etc. are pathogenic and promote gut inflammation[46-48]. Prevotella, which are anaerobic Gram negative bacteria, have been found to activate toll like receptor 2[49]. On the other hand, Ruminococcus, which belongs to the phylum Firmicutes is an important producer of short chain fatty acid which have anti-inflammatory effects[50]. Christensenellaceae, a commensal bacteria with immunomodulatory properties, has also been noted to be reduced in IBD, as was seen in our cases[51]. As with mucosal microbiota, bacterial populations in stool have also consistently shown reduced diversity (alpha and beta) in patients with CD[3,52-58]. We observed a reduced bacterial alpha diversity (Shannon) among active CD cases compared to HC, and disease status was the most important factor contributing to explained variance in microbial composition. During random forest analysis, many of the organisms identified in our patients’ stool samples as significant predictors of disease were also noted in earlier studies[9,55,59-64].

Comparison of tissue with stool samples has shown both structural and functional differences between active CD and HC. Morgan et al[65] found that seventy clades were over or underrepresented in stool compared to biopsy[65]. The Firmicutes, which are reduced in CD, had an even lower abundance in mucosa compared to the stool sample[65]. Another study from Italy showed Bacteroides and Peptostreptococcaceae to be more abundant in biopsies, while Lactobacillus reuteri, Granulicatella, and Streptococcus were commoner in stool[6]. While we did not note any difference in the bacterial diversity between tissue and stool samples, the type of sample (tissue or stool) contributed the maximum to the explained variance in microbial composition, as has been observed in earlier reports[45]. This reaffirms the limitation of stool samples as a surrogate for overall gut microbial composition. Apart from the contrast between stool and tissue samples, we identified organisms that were increased in both tissue and stool of active CD compared to controls and those showing opposing trends. Whether the opposing trend signifies a shift of organism from one site to another and a similar trend signifies a more direct role in pathogenesis needs to be determined.

As many bacterial functions are shared across species, the functional impact of altered composition provides better information than the type of bacteria. The functional potential of the bacterial community is reflected in the abundance of the metagenome and the metabolic pathways[65]. In fact, the shift in metabolic profile has been found to be more prominent than changes in microbial composition in IBD[65]. The assessment of metagenome requires whole genome sequencing, but the large abundance of human DNA in tissue samples makes the isolated study of bacterial whole genome extremely challenging[41,66]. Hence, for tissue samples, prediction of function by PICRUST from the 16SrRNA sequences is one of the alternatives that provides close to 80% accuracy in assessing the metabolic pathway of the organisms[66]. Reduction in short chain fatty acid metabolic pathways, tryptophan metabolism, and increase in primary bile acid pathways and carbohydrate transport/metabolic pathways have been observed in CD[2,54,63,64,67-69]. On comparing cases with controls, we found differences in the metabolic pathway between the healthy and disease groups in the tissue and stool samples. The differences in the proportion of genes associated with various functions were more in the stool sample (cases vs controls). They were noted for both synthetic (glycogen, glutamine, nucleotide) and degradation (glycerol, phenylacetate, phenylethylamine) pathways. Many functional pathways significantly differed in abundance between the stool and tissue samples of patients with active CD.

In addition to their role in disease pathogenesis, gut microbiota has also shown promise as a biomarker of disease state and predictor of clinical course[13,50,57]. Efforts have been made to identify a microbial signature of CD to help as a diagnostic tool[62]. We used a machine learning classifier method (random forest) to distinguish active CD from healthy subjects[70]. Based on this model, we identified a pre-defined set of bacteria in the mucosal sample to have a good predictive ability for identifying cases with active CD. However, the performance of the fecal microbiota was poor. A previous study has also shown the superior performance of tissue microbiota, including from the rectum (AUC = 0.78), and poor performance of stool samples as disease classifiers (AUC = 0.66)[13].

The imbalance in microbial composition and function (dysbiosis) is considered the driver of inflammation. Several measures of dysbiosis have been used, but since marker gene studies generally assess relative rather than absolute abundance of organisms, it is difficult to get a reliable estimate[71]. Also, they ignore the functional implications of changes in microbial population and assume genomic pathways of all organisms are equally active. Hence, such simplistic estimates of dysbiosis have their own limitations. We used a non-parametric estimate of dysbiosis (CLOUD test), which does not make assumptions about sample distribution, and yet only a minority of samples appeared dysbiotic[33]. MDI, which is a frequently done test for dysbiosis, also showed similar results. There is a need for better or alternative methods to measure dysbiosis, especially for tissue samples.

We have assessed the functional aspects of microbiota using prediction tools rather than whole genome sequencing. This is a limitation of our study. However, for tissue samples, whole genome sequencing of microbes is hampered by the presence of a large amount of human DNA[41]. Our recruitment was based on CDAI score for activity rather than histological activity, which is another limitation. As CD affects different parts of the intestine, and sometimes they may be difficult to access (e.g. small bowel CD), we did not use histological activity.

CONCLUSION

In conclusion, we demonstrate significant alterations in the gut bacterial population in a cohort of patients with active CD compared to HC. We also highlighted the differences between fecal and tissue microbiota and the limitations of studying stool samples alone.