https://orcid.org/0000-0002-1577-0118
Abstract
Exclusive enteral nutrition (EEN) supplementation of the standard of care (SOC) augments steroid responsiveness in patients with acute severe ulcerative colitis (ASUC). EEN is known to alter gut microbial composition. The present study investigates EEN-driven gut microbial alterations in patients with ASUC and examines their correlations with clinical parameters.
Stool samples from patients with ASUC (n = 44) who received either EEN-supplemented SOC (EEN group; n = 20) or SOC alone (SOC group; n = 24) for 7 days were collected at baseline (day 0) and postintervention (day 7). Microbiome analysis was carried out using 16S ribosomal RNA gene sequencing followed by data processing using QIIME2 and R packages.
Seven-day EEN-conjugated corticosteroid therapy in patients with ASUC enhanced the abundances of beneficial bacterial genera Faecalibacterium and Veillonella and reduced the abundance of Sphingomonas (generalized linear model fitted with Lasso regularization with robustness of 100%), while no such improvements in gut microbiota were observed in the SOC group. The EEN-associated taxa correlated with the patient’s clinical parameters (serum albumin and C-reactive protein levels). Unlike the SOC group, which retained its preintervention core microbiota, EEN contributed Faecalibacterium prausnitzii, a beneficial gut bacterial taxon, to the gut microbial core. EEN responders showed enhancement of Ligilactobacillus and Veillonella and reduction in Prevotella and Granulicatella. Analysis of baseline gut microbiota showed relative enhancement of certain microbial genera being associated with corticosteroid response and baseline clinical parameters and that this signature could conceivably be used as a predictive tool.
Augmentation of clinical response by EEN-conjugated corticosteroid therapy is accompanied by beneficial gut microbial changes in patients with ASUC.
Lay Summary
Exclusive enteral nutrition–supplemented corticosteroid therapy in acute severe ulcerative colitis (ASUC) is accompanied by the enrichment of beneficial gut microbial genera, which correlate negatively with the disease activity scores and objective inflammatory markers in ASUC. The baseline gut microbiota in ASUC associates with and may predict corticosteroid response.
The present study shows that 7-day exclusive enteral nutrition (EEN) supplementation of corticosteroid therapy in patients with acute severe ulcerative colitis (ASUC) enriches beneficial gut microbial members Faecalibacterium, and Veillonella and reduces detrimental genera, Sphingomonas, Ruminococcus gnavus, Klebsiella, and Bifidobacterium.
EEN refurbishes the obliterated core microbiota in patients with ASUC by contributing “good” bacteria such as Faecalibacterium prausnitzii and Lactobacillus sp.
While EEN responders harbor beneficial gut microbial genera Veillonella and Ligilactobacillus, nonresponders were found to have enhanced abundances of Prevotella and Granulicatella.
An 8-member baseline gut microbial signature associates with and may predict corticosteroid response in ASUC.
Introduction
Acute severe ulcerative colitis (ASUC) is a life-threatening emergency requiring hospitalization and prompt clinical management.1,2 It affects almost a quarter of patients with ulcerative colitis and has a high morbidity rate.1,3 Intravenous corticosteroids are the first line of therapy for ASUC, but 30% to 40% of patients are nonresponsive and need salvage medical therapy with ciclosporin or anti-tumor necrosis factor therapy, or colectomy.4,5 The efficacy of salvage medical therapy depends on timely decision-making by clinicians, yet it is unaffordable for a substantial proportion of patients in resource-limited regions. Nevertheless, even salvage therapy can be ineffective, leading to colectomy.6
Consequently, there is a need to augment the response to steroids, and a randomized controlled trial7 has shown that this is possible. Carefully tailored dietary programs can act as a conjugate therapeutic modality to steroids. Indeed, there have been numerous studies and clinical trials demonstrating the outstanding performance of exclusive enteral nutrition (EEN) in the induction of remission in patients with Crohn’s disease.8-10 The induction rates of EEN independently are at par with corticosteroid therapy, justifying the adoption of the former as the first line of therapy for Crohn’s disease in clinics.10 The same, however, cannot be claimed in the case of ulcerative colitis or ASUC.11 There have been scant reports studying the efficacy of EEN in mucosal healing and clinical improvement in ulcerative colitis or mice models of ulcerative colitis, but no such studies were reported about ASUC until recently. We employed EEN to augment the response to steroid therapy in patients with ASUC and demonstrated that EEN reduced steroid nonresponse from 43% to 25%, with a significant reduction in hospitalization and colectomy over the subsequent 6 months.7 The benefits of EEN are attributable to local and systemic immune modulatory effects and its ability to mold the gut microbial composition.12,13 Indeed, gut dysbiosis is a pathognomonic feature of inflammatory bowel diseases, and microbial manipulation therapies have the potential for induction and maintenance of remission. Our group has shown that the degree of gut dysbiosis is greater in ASUC than in mild-moderate ulcerative colitis.14 Specific changes in gut microbial composition through dietary therapy might hold the key to ameliorating disease activity.
Here, we decipher the implications of a 7-day EEN regime in conjunction with standard-of-care (SOC) corticosteroid therapy on the gut microbial composition of patients with ASUC. We examine whether the compositional changes correlate with the observed improvements in clinical parameters. This is the first time that changes in gut microbiota in response to EEN in ASUC have been investigated.
Methods
Study Design and Patient Recruitment
Patients hospitalized with ASUC, defined by Truelove and Witt’s criteria,15 were recruited at the Department of Gastroenterology, All India Institute of Medical Sciences, New Delhi between August 2018 and May 2020 after written informed consent (Ref. no. IEC-261/04.05.2018). The cohort and methods have previously been described.7 A total of 51 patients were randomized to EEN for 7 days together with the SOC (EEN group, n = 25) or a normal calorie-monitored diet together with SOC (SOC group, n = 26) (Figures 1 and 2A). Baseline microbiota analysis and its association with the potential response to EEN and corticosteroids were carried out using the day 0 samples from these patients. Patients on EEN received a protein-based semi-elemental feed (Peptamen; Nestlé) according to their caloric requirements for 7 days along with standard medical therapy, while patients on SOC continued a calorie-monitored normal diet with standard medical therapy (intravenous hydrocortisone [400 mg/d], with mesalazine, topical therapy with hydrocortisone enemas, and venous thromboprophylaxis). Patients at high risk of colectomy were identified using the Oxford criteria,16 and nonresponders were rescued either through salvage medical therapy (infliximab/tacrolimus/ciclosporin) or surgery, advised by day 5 of the corticosteroid therapy.
The CONSORT flow diagram describing patient recruitment and sample counts utilized for each analysis type. EEN, exclusive enteral nutrition.
Analysis of alterations in gut microbiota in response to a 7-day exclusive enteral nutrition (EEN)–conjugated corticosteroid therapy in patients with acute severe ulcerative colitis (ASUC) and their correlation with the clinical parameters. A, The study design involved the randomization of an ASUC patient cohort into EEN and standard-of-care (SOC) groups, wherein patients on EEN received a 7-day protein-based semi-elemental feed along with standard corticosteroid therapy, while the SOC group continued a normal diet with standard corticosteroid therapy. Fecal samples were collected at baseline (day 0) (ie, prior to the beginning of the dietary intervention) and on day 7 of intervention for 16S sequencing–based microbiome analysis. The primary outcome was the proportion of patients with corticosteroid failure in each group, while secondary outcomes included changes in inflammatory parameters such as day 5 serum C-reactive protein (CRP) and day 7 serum albumin levels. B, Comparison of α-diversity indices of Shannon index along with the observed richness and abundance-based coverage estimator (ACE) richness index across the paired baseline and day 7 samples of responders in the EEN group, assessed by the Wilcoxon rank sum test. C, Comparison of core microbiota feature between the pre- and postintervention samples in the EEN responders (prevalence cutoff of 0.7). The core microbiota analysis was carried out using the R package microbiomeutilities. D, Differentially abundant bacterial taxa between the baseline and day 7 samples of the patients with ASUC who responded positively to the EEN, carried out by constructing a generalized linear model fitted with Lasso regularization (P < .05), performed using the SIAMCAT R package. E, Scatter plots showing the results of Spearman correlations between the bacterial genera enhanced (Faecalibacterium, Veillonella, Streptococcus, and Schaalia) and reduced (Sphingomonas) after 7 days of EEN and the clinical parameters (day 7 serum albumin and day 5 CRP levels), along with the correlation coefficient (R) and P values. The correlation analysis and heatmap generation was carried out by using the psych and corrplot R packages. To remove noise from the sequencing datasets, all analyses pertaining to microbiota diversity, core identification, differential abundance, and correlation testing were performed by filtering microbial taxa to include only those that occur in more than 30% of the samples. IV, intravenous; ns, non-significant p-value for Wilcoxon rank sum test.
For longitudinal analysis of the microbiota between pre- and postintervention samples, patients with only paired samples were considered for analysis (Figure 1). Due to poor sequence read quality, 7 postintervention samples (day 7) were discarded from the analysis, and 1 preintervention sample (day 0) was removed from the analysis, leaving a total of 88 paired samples (44 baseline samples with their corresponding 44 postintervention samples for the gut microbiota analysis; 20 in the EEN group and 24 in the SOC group).
Analysis of Fecal Microbiota
Fecal samples were collected from all patients at enrollment (day 0) and after completion of the 7-day intervention (day 7). Samples were immediately aliquoted and stored at −80 °C until processed for DNA extraction. DNA extraction was carried out using the Translational Health Science and Technology Institute protocol, with minor modifications.17 Briefly, frozen samples were thawed, precisely weighed to 200 mg, and homogenized using 2- to 3-mm glass beads (Biospec). This was followed by enzymatic cell lysis using lysozyme (10 mg/mL) (Sigma-Aldrich), mutanolysin (25 kU/mL) (Sigma-Aldrich), and lysostaphin (4 kU/mL) (Sigma-Aldrich) at 37 °C for 1 hour. Postincubation samples were subjected to treatment with 4 M guanidine thiocyanate (Sigma-Aldrich) and 10% N-lauryl sarcosine (Sigma-Aldrich), before incubation at 37 °C for 10 minutes and 70 °C for 1 hour. This was followed by mechanical lysis of cells by zirconia beads before supernatants were transferred to fresh tubes and subjected to protein removal. Nucleic acids were pelleted using ice-cold ethanol (96%) and centrifugation at 14 000 rcf for 10 minutes at 4 °C. The final precipitation of DNA was achieved by adding 3 M sodium phosphate and 1 mL of 96% ethanol. The 16s ribosomal RNA (rRNA) gene (V3-V4) amplified from the extracted DNA was sequenced using Illumina MiSeq platform, following which high-quality reads were obtained by using Trimmomatic v0.38 to remove adapter sequences, ambiguous reads, and low-quality reads (reads with more than 10% quality threshold [Quality Value (QV)]; <20 Phred score).
Sequence Analysis
Raw paired-end reads were subjected to demultiplexing, denoising, and chimera removal using the DADA2 pipeline of QIIME2-2022.2,18 which generated an amplicon sequence variant feature table and representative sequences. The bacterial amplicon sequence variant feature table was then annotated via a taxonomy classifier, built using reference sequence annotation and curation pipeline (RESCRIPt), based on the National Center for Biotechnology Information Bio projects 33175 and 33317 (National Center for Biotechnology Information 16s rRNA gene RefSeq database).19 For subsequent analysis through R packages, the QIIME outputs and associated metadata were imported into R as a phyloseq object using QIIME2R (version 0.99.6) (https://github.com/jbisanz/qiime2R.git). The diversity matrices and core microbial features were computed using the R package MicrobiomeUtilities (version 1.00.15).20 A microbial core was assigned on the basis of detection and a minimum prevalence cutoff of 80%. For differential abundance analysis of the microbial data, a generalized linear model fitted with Lasso regularization was performed using the SIAMCAT R package with robustness >0.8. The R package selbal was used for analyzing the baseline microbial balance capable of predicting the response to the therapy. Selbal is a greedy stepwise algorithm for the identification of microbial signatures consisting of 2 groups of taxa whose relative abundances, or balance, are predictive of the response to therapy. The data were collapsed at the genus level and selbal was performed with 5-fold cross-validation, with 5 iterations.21
To consolidate the relationship between clinical parameters and the gut microbial abundance matrices, we performed co-inertia analysis (CIA), using the R package made4 (multivariate analysis of microarray data using ade4).22,23 CIA was performed on the microbal abundance dataset and the clinical (positive clinical response, fecal calprotectin [FCal] and C-reactive protein [CRP] levels, Ulcerative Colitis Endoscopic Index of Severity (UCEIS), and prior ASUC incidence[s]) parameters of baseline patient samples (n = 51). CIA is a multivariate model that identifies successive axes of covariance between 2 datasets involving the same subjects. The global similarity between the datasets is denoted by a correlation coefficient called the Random Variable (RV) coefficient. The RV coefficient is the coefficient of correlation between the two datasets. This coefficient varies between 0 and 1: the closer the coefficient to 1, the stronger the correlation between the tables. The permutation test to determine the significance of RV scores was performed using the called RV.rtest function, which performs a Monte Carlo–based estimation on the sum of eigenvalues from the CIA. Acknowledging the compositional nature of the microbiome datasets, selbal considers microbial signatures given by the geometric means of data from 2 groups of taxa whose relative abundances or balance are associated with the response variable of interest. X+ and X– represents groups of taxa that define the global balance or microbial signature for the response variable. Analyzing the microbial abundance dataset as a contingency table (bearing relative abundances of microbial taxa) and the clinical dataset as quantitative information, co-inertia was computed between the correspondence analysis and performed on the gut microbiota dataset, and principal component analysis was performed on the clinical dataset. The XY scatter plot, representing individual patient samples positioned by arrows, whose origin is described by the clinical parameters and for which the end of the arrow is described by the abundance dataset, shows clustering of the samples on the XY plane, guided by both the clinical and abundance tables.
Software and Tools
All subsequent analysis of the raw microbiome data was performed using R (v.4.1.0; R Foundation for Statistical Computing), RStudio (v.1.4.1717), and associated packages. All microbiome data wrangling was performed using R packages—qiime2R, phyloseq, microbiomeutilities, microbiome, microbiomeSeq, and tidyverse. Differential abundance testing was carried out by using DESeq2, ashr, and scran R packages, and the results were plotted using cowplot and ggplot2 packages.
Results
Of 51 patients with ASUC recruited at the Department of Gastroenterology, All India Institute of Medical Sciences, New Delhi between August 2018 and May 2020,7 the mean age was 35.3 ± 12.7 years, 50% of patients were female, 33% of patients had E3 disease, and 26% had a previous episode of ASUC. Of the 20 patients in the EEN group, 14 responded to the intervention, while 6 were nonresponders. Out of 24 patients in the SOC group, 13 responded and 11 were nonresponders (Figure 1). The median UCEIS at enrollment was 6 (interquartile range, 5-7). Patient demographics and clinical parameters at baseline and postintervention (day 7 serum albumin and day 5 CRP) are shown in Table 1.
Subject demographics and clinical parameters.
| Parameter | EEN group (n = 25) | SOC group (n = 26) |
|---|---|---|
| Age, y | 32.5 ± 11 | 37.9 ± 13.7 |
| Sex | ||
| Female | 15 (60) | 11 (42.3) |
| Male | 10 (40) | 15 (57.6) |
| Extent | ||
| E2 | 6 (24) | 8 (30.7) |
| E3 | 19 (76) | 16 (61.5) |
| Prior ASUC | 7 (28) | 7 (26.9) |
| Steroid use in first year of diagnosis | 22 (88) | 21 (80.8) |
| Prior use of azathioprine | 16 (64) | 15 (57.7%) |
| UCEIS on admission | 5 (5-7) | 6 (5-7) |
| FCal on admission, μg/g | 1176.4 (810.9-1387.9) | 1119.5 (683.7-1491.9) |
| CRP on admission, mg/L | 23.8 (7.5-40.5) | 18.5 (5.5-13.5) |
| Serum albumin levels on admission, g/dL | 3.2 (2.55-3.65) | 3.2 (2.7-3.5) |
| Day 5 CRP, mg/La | 6.4 (1.6-11.75) | 11.4 (4.3-19.5) |
| Day 7 serum albumin, g/dL | 3.4 (2.9-3.8) | 2.7 (2.6-3.2) |
| Parameter | EEN group (n = 25) | SOC group (n = 26) |
|---|---|---|
| Age, y | 32.5 ± 11 | 37.9 ± 13.7 |
| Sex | ||
| Female | 15 (60) | 11 (42.3) |
| Male | 10 (40) | 15 (57.6) |
| Extent | ||
| E2 | 6 (24) | 8 (30.7) |
| E3 | 19 (76) | 16 (61.5) |
| Prior ASUC | 7 (28) | 7 (26.9) |
| Steroid use in first year of diagnosis | 22 (88) | 21 (80.8) |
| Prior use of azathioprine | 16 (64) | 15 (57.7%) |
| UCEIS on admission | 5 (5-7) | 6 (5-7) |
| FCal on admission, μg/g | 1176.4 (810.9-1387.9) | 1119.5 (683.7-1491.9) |
| CRP on admission, mg/L | 23.8 (7.5-40.5) | 18.5 (5.5-13.5) |
| Serum albumin levels on admission, g/dL | 3.2 (2.55-3.65) | 3.2 (2.7-3.5) |
| Day 5 CRP, mg/La | 6.4 (1.6-11.75) | 11.4 (4.3-19.5) |
| Day 7 serum albumin, g/dL | 3.4 (2.9-3.8) | 2.7 (2.6-3.2) |
Values are mean ± SD, n (%), or median (interquartile range).
Abbreviations: ASUC, acute severe ulcerative colitis; CRP, C-reactive protein; EEN, exclusive enteral nutrition; FCal, fecal calprotectin; SOC, standard of care; UCEIS, Ulcerative Colitis Endoscopic Index of Severity.
Subject demographics and clinical parameters.
| Parameter | EEN group (n = 25) | SOC group (n = 26) |
|---|---|---|
| Age, y | 32.5 ± 11 | 37.9 ± 13.7 |
| Sex | ||
| Female | 15 (60) | 11 (42.3) |
| Male | 10 (40) | 15 (57.6) |
| Extent | ||
| E2 | 6 (24) | 8 (30.7) |
| E3 | 19 (76) | 16 (61.5) |
| Prior ASUC | 7 (28) | 7 (26.9) |
| Steroid use in first year of diagnosis | 22 (88) | 21 (80.8) |
| Prior use of azathioprine | 16 (64) | 15 (57.7%) |
| UCEIS on admission | 5 (5-7) | 6 (5-7) |
| FCal on admission, μg/g | 1176.4 (810.9-1387.9) | 1119.5 (683.7-1491.9) |
| CRP on admission, mg/L | 23.8 (7.5-40.5) | 18.5 (5.5-13.5) |
| Serum albumin levels on admission, g/dL | 3.2 (2.55-3.65) | 3.2 (2.7-3.5) |
| Day 5 CRP, mg/La | 6.4 (1.6-11.75) | 11.4 (4.3-19.5) |
| Day 7 serum albumin, g/dL | 3.4 (2.9-3.8) | 2.7 (2.6-3.2) |
| Parameter | EEN group (n = 25) | SOC group (n = 26) |
|---|---|---|
| Age, y | 32.5 ± 11 | 37.9 ± 13.7 |
| Sex | ||
| Female | 15 (60) | 11 (42.3) |
| Male | 10 (40) | 15 (57.6) |
| Extent | ||
| E2 | 6 (24) | 8 (30.7) |
| E3 | 19 (76) | 16 (61.5) |
| Prior ASUC | 7 (28) | 7 (26.9) |
| Steroid use in first year of diagnosis | 22 (88) | 21 (80.8) |
| Prior use of azathioprine | 16 (64) | 15 (57.7%) |
| UCEIS on admission | 5 (5-7) | 6 (5-7) |
| FCal on admission, μg/g | 1176.4 (810.9-1387.9) | 1119.5 (683.7-1491.9) |
| CRP on admission, mg/L | 23.8 (7.5-40.5) | 18.5 (5.5-13.5) |
| Serum albumin levels on admission, g/dL | 3.2 (2.55-3.65) | 3.2 (2.7-3.5) |
| Day 5 CRP, mg/La | 6.4 (1.6-11.75) | 11.4 (4.3-19.5) |
| Day 7 serum albumin, g/dL | 3.4 (2.9-3.8) | 2.7 (2.6-3.2) |
Values are mean ± SD, n (%), or median (interquartile range).
Abbreviations: ASUC, acute severe ulcerative colitis; CRP, C-reactive protein; EEN, exclusive enteral nutrition; FCal, fecal calprotectin; SOC, standard of care; UCEIS, Ulcerative Colitis Endoscopic Index of Severity.
Effect of EEN on Gut Microbiota Profile and its Correlation with Clinical Parameters: Analysis of Paired Pre- and Postintervention Samples
Microbial alterations associated with 7-day EEN supplementation were analyzed by comparing baseline (preintervention) samples with the day 7 (postintervention) samples of patients who responded to the intervention (n = 14). No significant EEN-associated alterations were found in the α diversity indices between the baseline and postintervention samples (Figure 2B). Likewise, no significant alterations were found in the α diversity indices between the baseline and postintervention samples in the SOC group (Supplementary Figure 1A). Core microbiota analysis revealed an obliterated signature in the baseline samples comprising only 4 genera, namely Bifidobacterium, Escherichia, Enterococcus, and Streptococcus in the patients with ASUC (Figure 2C andSupplementary Figure 1C). Core microbiota analysis showed addition of Faecalibacterium, as a core member in the postintervention samples, in the EEN group (Figure 2C), while no core members were added to the postintervention samples in the SOC group (Supplementary Figure 1B).
Analysis of differentially abundant bacterial genera in the postintervention samples, when compared with the preintervention group, was carried out by constructing a generalized linear model fitted with Lasso regularization, performed using the SIAMCAT R package. Postintervention samples showed EEN-associated enhancement in Faecalibacterium, Streptococcus, Schaalia, and Veillonella and reduction in Sphingomonas, Bifidobacterium, Klebsiella, and Mediterraneibacter (Ruminococcus gnavus) (Figure 2D). Differential abundance testing in the SOC group showed enhancement of Klebsiella and Thermus and reduction in Methylobacterium, Prevotella, and Streptococcus postintervention (Supplementary Figure 1C).
Regarding clinical parameters, our randomized controlled trial7 showed that EEN supplementation of SOC shortened hospital stay, reduced serum CRP and FCal levels (both P = .04), and increased serum albumin levels (3.4 ± 0.4 g/dL vs 2.9 ± 0.3 g/dL; P < .01), when compared with SOC alone. The present study analyzing the microbiota in a subset of samples (Figure 1) from the trial also demonstrated a trend of EEN-mediated reduction in the CRP (P = .1) and increased serum albumin levels (median 3.4 [interquartile range, 2.9-3.8] g/dL vs 2.7 [interquartile range, 2.6-3.2] g/dL; P < .01) when compared with SOC alone (Table 1). To test whether these improvements in clinical parameters in patients with ASUC were associated with the microbial alterations in the EEN group, we correlated the differentially abundant bacterial taxa with the patient’s day 5 CRP and day 7 serum albumin levels as markers of inflammation and health, respectively. Significant Spearman correlations (with P < .05) were detected between serum albumin levels and the abundance of Faecalibacterium (R = 0.43, P = .02) and Veillonella (R = 0.41, P = .03). Serum albumin levels showed a trend of positive correlation with the abundances of Schaalia (R = 0.1, P = .57) and Streptococcus (R = 0.35, P = .06), while a trend of correlations was observed between the abundances of Sphingomonas and CRP levels (R = 0.3, P = .12) (Figure 2E).
Effect of EEN vs SOC on the Microbial Diversity, Composition, and Core Features: Analysis of the Postintervention Samples
To elucidate the effects of EEN on the gut microbiota composition, we compared the profiles between the postintervention samples (day 7) of patients with ASUC in the EEN and SOC groups (n = 27; EEN responders = 14, SOC responders = 13). Comparison of α diversity indices between the postintervention samples of the EEN and SOC groups showed a trend toward an EEN-mediated reduction in Shannon index (P = .38) when compared with the SOC group; however, they failed to reach statistical significance (Figure 3A). Differential abundance testing between the postintervention (day 7) samples of the EEN and SOC groups showed EEN-mediated enhancement of Faecalibacterium and Ligilactobacillus, while the SOC group showed enhancement of Blautia and Weissella when compared with the EEN samples (Figure 3B).
Comparison of posttherapy (day 7) gut microbiota diversity and structure between acute severe ulcerative colitis (ASUC) patients in the exclusive enteral nutrition (EEN) vs standard-of-care (SOC) groups. A, Comparison of α-diversity indices of Shannon index along with the observed richness and abundance-based coverage estimator (ACE) richness index in the EEN vs SOC samples of the patients with ASUC who responded positively to either of the therapy, assessed by Wilcoxon rank sum test. B, Differentially abundant bacterial taxa between the EEN and SOC samples who responded positively to their respective therapeutic regimens carried out by constructing a generalized linear model fitted with Lasso regularization (P < .05), performed using the SIAMCAT R package. C, Spearman correlation heatmap between the differentially abundant bacterial genera and the day 7 serum albumin and day 5 C-reactive protein (CRP) levels. The correlation analysis and heatmap generation was carried out by using the psych and corrplot R packages. To remove noise from the sequencing datasets, all microbiota diversity analysis, differential abundance, and correlation testing were performed by filtering microbial taxa to include only those which occur in more than 30% of the samples. ns, non-significant p-value for Wilcoxon rank sum test.
The finding of potentially beneficial taxa in the EEN group compelled us to draw associations between postintervention clinical parameters and the differential gut microbial signature. To test if these gut microbial signatures identified in the EEN vs SOC analysis are associated with postintervention clinical parameters, we correlated the differentially abundant bacterial taxa with the patient’s CRP and serum albumin levels. Spearman correlation analysis revealed a trend of correlation between serum albumin levels and EEN-associated bacterial taxa, including Faecalibacterium (R = 0.36, P = .06) and Ligilactobacillus (R = 0.13, P = .5). In contrast, the SOC group–associated Weissella showed significant positive correlation with the serum CRP levels (R = 0.44, P = .02) (Figure 3C).
Analysis of the Gut Microbial Signature in Responders vs Nonresponders to EEN
With the aim of characterizing the differences in gut microbial profiles between the EEN responders (n = 14) and nonresponders (n = 6), we performed differential abundance testing between their respective postintervention samples. When compared with EEN nonresponders, the responder group showed enhancement of bacterial genera such as Veillonella, Ligilactobacillus, and Klebsiella and a reduction in abundance of Granulicatella and Prevotella (Figure 4A).
![Comparison of gut microbiota diversity and structure between the exclusive enteral nutrition (EEN) responders and nonresponders. A, Differentially abundant bacterial taxa between the EEN responders vs nonresponders, identified at the genus level and determined using a generalized linear model fitted with Lasso regularization (P < .05), performed using the SIAMCAT R package. B, Scatter plots showing the results of Spearman correlations between the bacterial genera enhanced in responders (Ligilactobacillus and Veillonella) and nonresponders (Granulicatella) and the clinical parameters (serum albumin and C-reactive protein [CRP] levels), along with the correlation coefficient (R) and P values. The correlation analysis and heatmap generation was carried out by using the psych and corrplot R packages. C, Principal component analysis (PCA) biplot ordination depicting distribution of samples on the basis of corticosteroid response, overlayed with gut microbial abundance data and clinical parameters (corticosteroid response, day 5 CRP, and day 7 serum albumin). Constructed using ade4 and factoextra R package. D, Table summarizing the point-biserial correlation analysis (enlisting the correlation coefficients, P values, and 95% confidence intervals [CIs]) between the differentially abundant bacterial taxa identified in panel A and the corticosteroid response. To remove noise from the sequencing datasets, all microbiota diversity analysis, differential abundance, and correlation testing were performed by filtering microbial taxa to include only those that occur in more than 30% of the samples.](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/ibdjournal/30/4/10.1093_ibd_izad232/1/m_izad232_fig4.jpeg?Expires=1747861411&Signature=S4ZATLfp5EqVky49bMqHb0bGgIR3Z~kO7mnhW9P3EGhhG4hog6LyioURk0iaXo10X4xX~y~zeIcg8Q75QucBSqpQIh~EU7I7HpyGKkDBoQ6sh5o60hGCyI-oA0joFtUEyTrCpept4wrpzqHJN7dltbwnbR7LtYwVA8ZOOk03UlHO0zL9AF-wQnvlAsLflIDRlKZsJUrO26kBUP0lP3HTJYNgeoLZiAFswDmmKZ81wVccOp2cuGyReyUTejDAqvG347WMWP7EOPdOL5wCxeGO3LDGTZS6LewCDYq3uwB~WB02beuB3WUm5rO5vkDty5cDJI9OZ8rz8y3RB0XhGiQiCA__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Comparison of gut microbiota diversity and structure between the exclusive enteral nutrition (EEN) responders and nonresponders. A, Differentially abundant bacterial taxa between the EEN responders vs nonresponders, identified at the genus level and determined using a generalized linear model fitted with Lasso regularization (P < .05), performed using the SIAMCAT R package. B, Scatter plots showing the results of Spearman correlations between the bacterial genera enhanced in responders (Ligilactobacillus and Veillonella) and nonresponders (Granulicatella) and the clinical parameters (serum albumin and C-reactive protein [CRP] levels), along with the correlation coefficient (R) and P values. The correlation analysis and heatmap generation was carried out by using the psych and corrplot R packages. C, Principal component analysis (PCA) biplot ordination depicting distribution of samples on the basis of corticosteroid response, overlayed with gut microbial abundance data and clinical parameters (corticosteroid response, day 5 CRP, and day 7 serum albumin). Constructed using ade4 and factoextra R package. D, Table summarizing the point-biserial correlation analysis (enlisting the correlation coefficients, P values, and 95% confidence intervals [CIs]) between the differentially abundant bacterial taxa identified in panel A and the corticosteroid response. To remove noise from the sequencing datasets, all microbiota diversity analysis, differential abundance, and correlation testing were performed by filtering microbial taxa to include only those that occur in more than 30% of the samples.
Correlation of these differentially abundant bacterial genera identified in the EEN responder vs nonresponder analysis with clinical parameters showed a trend of correlation between the serum albumin levels and abundances of Ligilactobacillus (R = 0.38, P = .09) and Veillonella (R = 0.22, P = .35). Granulicatella, found to be enhanced in EEN nonresponders, showed a trend of positive correlation with the CRP levels (R = 0.3, P = .2) and a trend of negative correlation with the serum albumin levels (R = −0.14, P = .56) (Figure 4B). A principal component analysis biplot of the gut microbiota taxa and the levels of CRP and serum albumin, built on the postintervention samples of the EEN group, showed localization of beneficial bacterial genera such as Faecalibacterium, Limosilactobacillus, Ligilactobacillus, Veillonella, and Holdemanella with the cluster of samples signifying the responders and having higher serum albumin levels, while genera such as Granulicatella, Phocaeicola, and Mediterraneibacter (R. gnavus) coincided with samples clustering as nonresponders (Figure 4C) Point-biserial correlation of the steroid response showed significant negative correlation with the abundance of Prevotella (R = −0.46, P = .039), along with a trend of positive correlation with abundances of Veillonella (R = 0.23, P = .3), Ligilactobacillus (R = 0.3, P = .2), and Klebsiella (R = 0.19, P = .4) (Figure 4D).
Baseline Gut Microbial Signature and Response to Corticosteroid Therapy in Patients With ASUC
Once we identified the EEN-associated gut microbial modulations, we aimed to determine whether baseline gut microbial signatures were associated with response to 7-day corticosteroid therapy, irrespective of whether they received EEN or SOC alone (n = 51).
Selbal identified 2 groups of taxa that defined the microbial signature for corticosteroid response in the baseline samples, irrespective of EEN supplementation: X+ = {Escherichia, Faecalibacterium, Klebsiella} and X– = {Phocaeicola and Bifidobacterium}. Figure 5 presents the distribution of microbial taxa for responders and nonresponders. Patients with ASUC who did not show response to the 7-day corticosteroid therapy had lower balance scores than responders, implying that there are lower relative abundances of taxa in group X+ than in group X–. The discrimination value of the balance showed an area under the curve value of 0.83 (Figure 5A). The association of these bacterial genera to the corticosteroid response was further validated by generalized linear modeling using Lasso regularization. The model showed significant enrichment of Faecalibacterium and Klebsiella in the baseline gut microbiota of responders, while the baseline microbiota of nonresponders showed enhanced abundances of Bifidobacterium, Phocaeicola, and Collinsella (area under the curve = 0.70) (Figure 5B).
![Characterization of baseline gut microbial signature that correlates (or may predict) the response to corticosteroids in all patient samples (irrespective of dietary intervention) along with their respective receiver-operating characteristic (ROC) curves depicting the predictive ability of the balance and the density plots of balance scores, (A) performed using selbal to construct a 5-fold cross-validated prediction model used to select the best rank-discriminating microbial and (B) using Lasso-regularized generalized linear modeling (P < .05), performed using the glmnet and SIAMCAT R packages. C, Correlation of abundances of negative weights identified in the selbal analysis with the baseline disease activity score (Ulcerative Colitis Endoscopic Index of Severity [UCEIS]) and objective inflammatory parameters (day 0 fecal calprotectin [FCal] and day 0 C-reactive protein [CRP] levels). D, Correlation of abundances of positive weights identified in the selbal analysis with the baseline disease activity score (UCEIS) and objective inflammatory parameters (day 0 FCal and day 0 CRP levels). E, Co-inertia analysis (CIA) of the relationship between the baseline microbial genera and clinical parameters (day 0 FCal, day 0 CRP, day 0 UCEIS, prior acute severe ulcerative colitis [ASUC], and day 7 clinical response to corticosteroids) in patients with ASUC. CIA was carried out between the PCA of the clinical dataset and the correspondence analysis of the gut microbiota abundance dataset. The first scatter plot depicts the samples’ distribution according to the CIA. The start of each arrow is the position of the samples as described by the clinical dataset, while the end of each arrow is the position of the samples as described by the gut microbial abundances. The X and Y loading scatter plots represent the coefficients of the combination of the variables for each of the 2 datasets to define the co-inertia axes. CIA was performed with R package ade4. To remove noise from the sequencing datasets, all microbiota diversity analysis, differential abundance, and correlation testing were performed by filtering microbial taxa to include only those that occur in more than 30% of the samples.](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/ibdjournal/30/4/10.1093_ibd_izad232/1/m_izad232_fig5.jpeg?Expires=1747861411&Signature=5J60~GC6zcDznhANekKrwxaC6zmRvtdPHh2cYfVNvrfGgmBYNyEbEEmqmgCr3EtCcdxWB2JkuLYXc98F~UVDn8PqIyLsZU6tOZQZMvfK6YwPl3mzdxD8mW~Qyx1JDFQfkTybwETI8iWeX5xFAW5Bz4nuD8-fReiHc9bQQXMFiUcDv7XR09~99Dbs9~mNDGtnUHXCxSt0ThJuXARRKq89GRgJSuCLsg2hdRv9WQQVkh-w3Qe8bLhyo3-AKZfyaaoAA~Bysx~d-loiYjZ5NT9Y-uvc97F9wkaL0jbYcUdB-EMtztY3t4qL1pDBNQLzyrHgwlVu8bqePC1Lrv5GyovK0w__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Characterization of baseline gut microbial signature that correlates (or may predict) the response to corticosteroids in all patient samples (irrespective of dietary intervention) along with their respective receiver-operating characteristic (ROC) curves depicting the predictive ability of the balance and the density plots of balance scores, (A) performed using selbal to construct a 5-fold cross-validated prediction model used to select the best rank-discriminating microbial and (B) using Lasso-regularized generalized linear modeling (P < .05), performed using the glmnet and SIAMCAT R packages. C, Correlation of abundances of negative weights identified in the selbal analysis with the baseline disease activity score (Ulcerative Colitis Endoscopic Index of Severity [UCEIS]) and objective inflammatory parameters (day 0 fecal calprotectin [FCal] and day 0 C-reactive protein [CRP] levels). D, Correlation of abundances of positive weights identified in the selbal analysis with the baseline disease activity score (UCEIS) and objective inflammatory parameters (day 0 FCal and day 0 CRP levels). E, Co-inertia analysis (CIA) of the relationship between the baseline microbial genera and clinical parameters (day 0 FCal, day 0 CRP, day 0 UCEIS, prior acute severe ulcerative colitis [ASUC], and day 7 clinical response to corticosteroids) in patients with ASUC. CIA was carried out between the PCA of the clinical dataset and the correspondence analysis of the gut microbiota abundance dataset. The first scatter plot depicts the samples’ distribution according to the CIA. The start of each arrow is the position of the samples as described by the clinical dataset, while the end of each arrow is the position of the samples as described by the gut microbial abundances. The X and Y loading scatter plots represent the coefficients of the combination of the variables for each of the 2 datasets to define the co-inertia axes. CIA was performed with R package ade4. To remove noise from the sequencing datasets, all microbiota diversity analysis, differential abundance, and correlation testing were performed by filtering microbial taxa to include only those that occur in more than 30% of the samples.
The relative abundances of taxa at baseline (day 0), associated with response to 7-day corticosteroid therapy in the ASUC cohort, were then correlated with baseline clinical parameters. Spearman correlation analysis revealed a trend of positive correlations between the abundance of Bifidobacterium sp. and CRP level (R = 0.25, P = .07), UCEIS (R = 0.14, P = .32), and FCal level (R = 0.14, P = .34) (Figure 5C). A significant negative correlation between the abundances of Faecalibacterium and CRP levels (R = −0.3, P = .03) and a trend of negative correlations were noted between the abundances of Klebsiella and UCEIS (R = −0.13, P = .37) and FCal levels (R = −0.04, P = .7) (Figure 5D).
CIA of the relationship between the microbial genera and baseline clinical parameters (day 0 FCal, day 0 UCEIS, day 0 CRP levels, prior ASUC episodes, and positive response to corticosteroids) was performed on the baseline samples (n = 51), followed by a permutation test to analyze the significance of the global similarity between the 2 datasets denoted by the RV coefficient (Figure 5E). CIA showed an RV coefficient of 0.27 (P = .002), along with an overlapping clustering of bacterial genera Faecalibacterium, Veillonella, Klebsiella, Lactobacillus, and Ligilactobacillus with the positive clinical response to corticosteroids (quadrant Q1). Bacterial genera Methylobacterium, Thermus, and Granulicatella coincided with elevated baseline FCal, CRP, and UCEIS (quadrants Q3 and Q4), while Phocaeicola, Collinsella, and Escherichia coincided with prior ASUC exposure (quadrant Q2) (Figure 5E).
Discussion
The study highlights that EEN-augmented corticosteroid therapy in patients with ASUC accompanies beneficial gut microbial alterations, which correlate with clinical parameters. The present data, consistent with our previous report, show that ASUC witnesses severe gut dysbiosis and a decimated core, with Bacteroides, Bifidobacterium, Enterococcus, Streptococcus, and Escherichia as the most abundant genera.14
Analysis of baseline microbial composition in patients with ASUC showed that relative enhancement of baseline microbial genera such as Faecalibacterium, Escherichia, and Klebsiella is associated with corticosteroid response, and this signature could conceivably be used as a predictive tool. Additionally, these baseline genera correlate with the patient’s clinical parameters and thus are indicative of the disease activity.
Differential gut microbiota analysis between baseline and postintervention samples of patients with ASUC in the EEN group identified the EEN-associated enhancement of beneficial genera, Faecalibacterium, Veillonella, and Streptococcus, along with an oral Actinobacteria Schaalia. These members have been highlighted to possess significant immunomodulatory and short-chain fatty acid production activities in the gut. A correlation between these EEN-associated microbial genera and the patient’s serum albumin levels associates these microbial alterations with EEN-mediated improvement in clinical parameters. EEN also mediated a reduction in the abundances of Sphingomonas, Klebsiella, and Bifidobacterium. Apart from these, the abundance of Mediterraneibacter (R. gnavus) was also found to be reduced after EEN supplementation of corticosteroids. R. gnavus is a mucolytic gut bacterium, known to be involved in a plethora of inflammatory gut disorders via mechanisms such as the production of a polysaccharide called glucorhamnan, which stimulates tumor necrosis factor α release from gut dendritic cells.24-26 Analysis of gut microbiota alterations in the SOC group showed absence of any intervention-mediated addition of health-associated gut microbial members. The analysis showed that 7 day corticosteroid therapy alone was incapable of contributing any microbiota-associated benefits in ASUC. Adding to this, these results consolidated the claim that the beneficial gut microbial alterations observed in the EEN group were actually mediated by EEN supplementation and were not a mere artifact apparent due to 7-day corticosteroid therapy.
Differential abundance analysis between the responders and nonresponders in the EEN group further highlighted the relevance of Ligilactobacillus and Veillonella (enhanced in responders) and Prevotella and Granulicatella (enhanced in nonresponders) as the significant differential members associated with the effects of EEN in ASUC. The response-associated lactobacilli have been found to increase the level of anti-inflammatory cytokines while reducing the production of inflammatory cytokines such as interleukin-6, interleukin-1β, and tumor necrosis factor α.27-29
The comparison of gut microbiota diversity and core and differentially abundant features between the posttherapy samples of patients undergoing corticosteroid therapy with and without EEN (SOC group) supplementation revealed interesting insights into the role of dietary augmentation of therapy. The trend toward a reduction in α-diversity indices in the patients who underwent EEN when compared with the patients on SOC showed microbial-modulatory effects of EEN. This EEN-mediated reduction in α-diversity indices has been documented in previous reports, but the cause behind this paradoxical phenomenon and its clinical relevance remains elusive.30,31 Apart from changing the bacterial diversity, EEN was also found to shape core bacterial members, by the addition of Faecalibacterium prausnitzii and Lactobacillus, in the existing ASUC-associated core of Bifidobacterium, Enterococcus, Streptococcus, and Escherichia. It is noteworthy that SOC without EEN supplementation failed to contribute any such beneficial members to the gut bacterial core. Differential abundance analysis highlighted the EEN-mediated addition of beneficial microbes involved in short-chain fatty acid production such as Faecalibacterium and Ligilactobacillus.
Analysis of baseline samples to test whether a gut microbial signature can predict or associate with a potential response to corticosteroids revealed that a higher relative abundance of Faecalibacterium, Klebsiella, and Veillonella and a lower abundance of Phocaeicola, Collinsella, and Bifidobacterium were associated with a positive response to corticosteroids (area under the curve = 0.83). The higher relative abundances of these beneficial gut microbial genera in the preintervention samples showed the crucial role of baseline gut microbiota signature in determining the corticosteroid response in patients with ASUC. These baseline microbial members showed coherence with baseline clinical parameters, which further strengthens their relevance as predictors of response to therapy.
The strength of the present study is that, to the best of our knowledge, it is the first report describing gut microbial alterations associated with EEN augmentation of corticosteroid therapy in patients with ASUC. The gut microbial signatures deciphered in the study and their correlation with clinical parameters establish microbiome manipulation as one of the key mechanisms by which EEN exerts its anti-inflammatory effects in the gut. There are, of course, limitations. We believe that in addition to 16S rRNA gene sequencing data, shotgun metagenomics data would have provided greater depth to the compositional analysis and relevant functional information. Additionally, the baseline predictive signature was built with only 50 day 0 ASUC samples and should be supplemented with a larger sample size and other machine learning algorithms to be validated.
Conclusions
Our study shows that augmentation of clinical response by EEN-conjugated corticosteroid therapy is accompanied by specific gut microbial changes in patients with ASUC. This has implications for both predictions of therapeutic response and pathogenesis.
Funding
This work was supported by the Indian Council of Medical Research—Centre for Advanced Research and Excellence in Intestinal Diseases (grant number 55/4/11/CARE-ID/2018-NCD).
Conflicts of Interest
All authors certify that they have no affiliations with or involvement in any organization or entity with any financial interest or nonfinancial interest in the subject matter or materials discussed in this manuscript.
Data Availability
The data underlying this article will be shared on reasonable request to the corresponding author.
Ethical Considerations
The study was approved by the Institution Ethics Committee of the All India Institute of Medical Sciences, New Delhi. All study participants provided written informed consent.
