Differential contributions of the gut microbiota and metabolome to pathomechanisms in ulcerative colitis: an in vitro analysis

ABSTRACT

The gut microbiota has been implicated in onset and progression of ulcerative colitis (UC). Here, we assess potential causal involvement of the microbiota and -associated fecal water (FW) metabolome in altering key functional parameters of the colonic epithelium. Fecal samples were collected from N = 51 healthy controls (HC), N = 36 patients with active UC (UC-A), and N = 41 subjects in remission N = 41 (UC-R). Using in vitro incubation experiments, the FW metabolome’s impact on butyrate oxidation rates/gene expression and cell death (cytotoxicity) of HT-29 cells, cytokine production by PBMC, and barrier integrity of Caco2 monolayers was evaluated. The FW metabolome from patients and individuals hosting the Bacteroides 2 (Bact2) enterotype (69% of UC-A, 31% of UC-R, 3% of HC), characterized by lower levels of median- and short-chain fatty acids and furan compounds, left butyrate oxidation rates unaltered but affected associated gene expression profiles. UC patients/Bact2-carriers’ FW lowered PBMC IL-8 production and increased IL-1β production. Patients’ FW increased cytotoxicity, associated with sulfide compound levels. Bact2 carriers’ FW, displaying higher levels of bile acids, lowered barrier function upon incubation of monolayers. The FW metabolome of patients and individuals hosting a dysbiotic microbiota could contribute to the disruption of functional processes of the colonic epithelium as observed in UC.

KEY MESSAGES

What is already known on this topic.

• The altered gut microbiota is implicated in the pathology of ulcerative colitis, potentially due to a mediating role of the associated fecal water metabolome.

• Colonocytes of patients with ulcerative colitis display altered butyrate oxidation rates, gut barrier function, and immune function in vivo.

What this study adds.

• The gut microbiota in ulcerative colitis, characterized by an increased prevalence of the Bacteroides 2 (Bact2) enterotype, is associated with an altered metabolome comprising lower SCFA, MCFA, and furan concentrations and higher levels of bile acids.

• The UC-associated fecal metabolome alters butyrate oxidation gene expression and cytokine production in colonocytes in vitro. It reduces gut barrier function and results in higher cytotoxicity of fecal water.

• Changes in butyrate oxidation gene expression and immune function are associated both with disease and Bact2 carrier status. Gut barrier function mainly depends on Bact2 status, while fecal water cytotoxicity reflects disease status.

How this study might affect research, practice, or policy.

• This study demonstrates the impact of the altered fecal water metabolome in ulcerative colitis on functional parameters of the colonic epithelium.

• By disentangling effects of Bact2 and patient disease status, the study identifies modulation of the gut microbiota away from enterotype-defied dysbiosis as a potential strategy to increase barrier function.

1. Introduction

Ulcerative colitis (UC) is a chronic-relapsing disease characterized by inflammation of the lining of the colon and rectum. Although UC pathogenesis is complex and remains to be fully elucidated,¹ there is a growing consensus that a disruption of the intestinal microbiota plays a key role in disease onset and/or progression.² Experimental studies revealed that germ-free Tg(epsilon26) mice engrafted with bone marrow from wild-type animals (causing colitis in the specific pathogen-free model) do not develop intestinal inflammation,^3,4 a finding also seen in IL-10 deficient mice.⁵ Furthermore, once established, the inflammatory phenotype has been shown to be transferable to recipient mice by transplantation of the disease-associated microbiota.⁴ In patients, the clinical improvement obtained with therapeutic strategies targeting microbial imbalance such as antibiotics, prebiotics, probiotics, and fecal microbiota transplantations supports the hypothesis of an imbalanced microbiota as a contributing factor in UC pathogenesis.⁶

The microbiota of UC patients is generally characterized by a reduced diversity, comprising low proportions of health-associated bacteria such as Faecalibacterium prausnitzii, Bifidobacterium spp., Clostridium coccoides, and Clostridium leptum.⁷ The presence of such microbiota has been linked to impaired barrier function, altered epithelial cell metabolism, and pro-inflammatory immune responses.^8,9 However, whether the aberrant gut microbiota causes these disruptions of host homeostasis remains to be established. Indeed, while a dysbiotic gut microbiota could potentially induce, sustain, or exacerbate some host responses, changes in the gut ecosystem could also occur because of microbial adaptation to other expressions of a pro-inflammatory environment. Alternatively, microbiome alterations and specific host disease manifestations could coexist without causal relationship. Here, we assess the effect of gut microbiota composition and metabolism on specific aspects of epithelial functioning in an in vitro setting, disentangling the impact of the dysbiotic communities from the impact of genetic predisposition, environmental triggers, and broader individual variability. First, we characterize fecal samples of healthy controls and UC patients in terms of microbiota composition, metabolite profiles, and fecal water cytokine concentrations. To enable ecosystem-level analyses of microbiota variation, we stratify microbiome profiles into a limited number of community types.¹⁰ These enterotypes have previously been linked with a number of host variables including inflammation levels¹¹ and use of medication.^12,13 With respect to host health, the potentially dysbiotic Bacteroides 2 (Bact2) enterotype has been shown to be overrepresented in a plethora of diseases, including UC.¹¹ Next, we evaluate the effect of fecal water obtained from these samples, tightly linked to their associated enterotypes and emerging metabolome, on butyrate oxidation rates, cytokine production, cell survival, and barrier integrity throughout a series of incubation experiments with human cell lines.

2. Results

2.1. Cohort description

One hundred twenty-eight baseline fecal samples were obtained from 51 healthy controls (HC; age = 37.6 ± 15.6, BMI = 23.9 ± 3.3), 36 patients with active ulcerative colitis (UC-A; age = 43.8 ± 15.1, BMI = 24.3 ± 3.6), and 41 patients in remission (UC-R; age = 48.6 ± 15.0, BMI = 25.0 ± 2.9). Thirty-two patients provided an additional follow-up sample taken when their disease activity shifted from active to remission (N = 31) or vice versa (N = 1), based on the evolution of their partial Mayo score (pMayo). Analyses of follow-up samples were performed for validation purposes only and are explicitly identified as such in the sections below. Overall, UC-R patients were older than HCs (Kruskall-Wallis [KW], N = 128, effect size [ES] = 0.083, p_adj = 2.07e⁻³; post-hoc Dunn [phD] test in Supplementary Table S1). More UC-A patients used corticosteroids at the time of sampling compared to UC-R participants (Fisher exact test, N = 77, odds ratio [OR] = 0.032, p_adj = 6.02e⁻⁵; Supplementary Table S1). Samples from UC-A participants displayed lower fecal dry weights (FDW) than HC and UC-R aliquots (KW test, N = 128, ES = 0.12, p_adj = 7.13e⁻⁵; phD in Supplementary Table S2). Fecal water (FW) cytokine profiles were markedly different between patient groups and HCs, with all cytokines quantified displaying distinct concentrations in at least one sub-cohort. IL-1β and IL-8 concentrations both were significantly higher in UC-A FW, with IL-1β concentrations additionally being higher in UC-R compared to HC samples (IL-1β, KW-phD test, N = 128, ES = 0.44, p_adj = 5.32e⁻¹³; IL-8, ES = 0.23, p_adj = 2,35e⁻⁷). TNF-α (ES = 0.19, p_adj = 8.91e⁻⁶) and IL-6 (ES = 0.060, p_adj = 1.09e⁻²) concentrations were lower in UC-R and UC-A FW, respectively. Finally, IL-12p70 levels were reduced in both UC sub-cohorts (ES = 0.060, p_adj = 6.17e⁻⁴; Supplementary Table S2).

2.2. Faecal enterotype prevalence is associated with disease activity

For all fecal samples, microbiota composition was determined using 16S rRNA gene amplicon sequencing. The contributions of age, gender, disease activity, use of medication, FDW, and FW cytokine profiles to baseline microbiome variation were estimated using a distance-based redundancy analysis with forward stepwise regression (dbRDA). Relative genus-level community composition was significantly associated with disease activity, FDW, FW IL-1β concentrations, age, gender, and FW TNF-α levels, jointly explaining 9.9% of microbiome variation (dbRDA on Aitchison distances, N = 128, p = 1.00e⁻⁶; Figure 1a,b; Supplementary Table S3). After combining fecal microbiomes from the present study with N = 1048 Flemish Gut Flora Project (FGFP) profiles,¹⁰ dataset stratification on genus level using a Dirichlet multinomial mixtures-based approach¹⁴ confirmed community variation to revolve around four enterotypes labeled Prevotella (Prev), Ruminococcaceae (Rum), Bacteroides1 (Bact1), and Bacteroides2 (Bact2) based on proportions of dominating taxa. Enterotype prevalence differed markedly between study groups (χ² test, N = 128, χ² = 44.0, p_adj = 7.30e⁻⁸; Figure 1c). While 3.9% of HC stools clustered as Bact2, this proportion increased to 31.0 and 68.6% for UC-R and UC-A samples, respectively (pairwise Fisher exact test, HC vs. UC-R, N = 92, OR = 0.114, p_adj = 2.21e⁻³; HC vs. UC-A, N = 87, OR = 0.022, p_adj = 4.74e⁻¹⁰; UC-A vs. UC-R, N = 77, OR = 5.32, p_adj = 8.94e⁻⁴). This observation confirmed the previously reported association of the putatively dysbiotic Bact2 enterotype with inflammatory phenotypes.¹¹ Bact2 prevalence did not differ significantly between paired samples from patients switching between active disease and remission (paired χ² test, N = 64, χ² = 2.15, p = 0.54; Figure 1d), despite a decrease in 13% points upon transition from UC-A to UC-R. Across enterotypes, Rum samples displayed the highest FDW levels (KW test, N = 128, ES = 0.265, p_adj = 5.54e⁻⁷; phD in Supplementary Table S2), while Bact2 aliquots exhibited higher FW IL-1β concentration compared to Bact1 and Rum samples (ES = 0.17, p_adj = 7.35e⁻⁵).

Figure 1.

Microbiome variation in the UC-A/UC-R/HC cohort a. Independent (black) and cumulative, non-redundant (gray) effect sizes (adjusted R² [%] of the cohort descriptors on the genus-level microbiome community variation; dbRDA on the Aitchison distance matrix, n = 128). (DA = disease activity, FDW = fecal dry weight). b. Principal component analysis (PCA) of genus-level Aitchison distance matrix UC-A/UC-R/HC cohort (n = 128). The percentage of variation explained by the two first PCoA dimensions is reported on the axes. c. High prevalence of the Bact2 enterotype in patient groups compared with healthy controls (pairwise Fisher exact test, n = 128, ***p_adj <0.001, **p_adj <0.01; supplementary table S2). d. Enterotype shifts patients switching between active disease and remission (paired χ2 test, N = 64).

2.3. Enterotype stratification contributes to semi-quantitative metabolome profile variation

Semi-quantitative fecal metabolome profiles were constructed based on the concentrations of volatile organic compounds (VOC; mainly short-to-medium-chain fatty acids, ketones, and alcohols) and bile acids (BA), as well as proportions of nonvolatile organic compounds (NVOC; including amino-acids, sugars, and derivatives as well as medium-to-long-chain fatty acids) detected using a combination of mass spectrometry coupled chromatographic techniques. Aligning with the analysis of gut microbiota composition, the explanatory power of the cohort metadata in metabolome variation was evaluated using a dbRDA, including enterotype as a potential covariate to explore the contribution of the microbiome. Different combinations of metadata variables explained 6.1, 7.1, and 9.8% of VOC (Euclidian distances, p_adj = 1.00e⁻⁶), NVOC (Aitchison distances, p_adj = 1.00e⁻⁶), and BA (Euclidian distances, p_adj = 1.82e⁻²) profile variation, respectively (Figure 2a-c (right); Supplementary Tables S4-S6). Enterotype stratification was the only variable with a significant, non-redundant explanatory power in all three metabolome subsets, explaining 1.2%, 0.8%, and 4.2% of VOC, NVOC, and BA profile variations. Disease activity contributed to metabolite variation in the VOC and NVOC spaces (1.0% and 4.5%). Additional subset-specific contributions to metabolite variation were established for FDW, fecal osmolality, and the use of 5-aminosalicylates to VOC, FW concentrations of IL-12p70 to NVOC, and age, concomitant chronical diseases (y/n), and use of corticosteroids to BA profiles. To assess potential associations between metabolome profiles and genus-level microbiome variation rather than enterotype strata, we performed a Procrustes analysis on the paired datasets. VOC (Procrustes, N = 128, R² = 0.29, p_adj = 3.00e⁻³) and NVOC (R² = 0.27, p_adj = 2.55e⁻²) matrices but not BA profiles (R² = 0.16, p_adj = 0.57) were significantly associated with microbiome composition.

Figure 2.

Metabolome variation in the UC-A/UC-R/HC cohort a. Variation in the volatile organic acid profile (ratio compared to the internal standard; dbRDA on the Euclidian distance matrix; n = 128) b. Variation in the nonvolatile organic acid metabolite profile (clr-value; dbRDA on the Aitchison distance matrix; n = 128). c. Variation in the bile acid metabolite profile (µM; dbRDA on the Euclidian distance matrix; n = 128). (left), Kruskal – Wallis with post hoc Dunn test, ***p_adj <0.001, **p_adj <0.01, and *p_adj <0.05; the body of the box plot represents the first and third quartiles of the distribution and the median line. The whiskers extend from the quartiles to the last data point within 1.5× the interquartile range, with outliers beyond. (right) Independent (black) and cumulative, non-redundant (gray) effect sizes (adjusted R² [%] of the cohort descriptors on the metabolome variation (DA = disease activity, oChrD = other chronic diseases, FDW = fecal dry weight).

Abundances of 4-di-tert-butylphenol (KW test, N = 128, ES = 0.21, p_adj = 6.00e⁻⁵; phD in Supplementary Table S7) and propyl alcohol (ES = 0.12, p_adj = 2.83e⁻³) were higher in UC-A and – R patients compared to HC. Acetophenone, caproic acid and valeric acid (all higher in HC compared to UC-A and -R; ES = 0.18, p_adj = 1.83e⁻⁴; ES = 0.13, p_adj = 1.19e⁻³, and ES = 0.14, p_adj = 2.22e⁻³; Figure 2a) completed the top five of differentially distributed VOCs. Regarding NVOCs, UC-A and – R samples contained lower proportions of d-ribose (KW test, N = 128, ES = 0.35, p_adj = 1.96e⁻⁸; phD in Supplementary Table S8), xylulose (ES = 0.24, p_adj = 4.52e⁻⁶), and maleic acid (ES = 0.17, p_adj = 4.14e⁻⁴) compared to HC and higher proportions of l-histidine (ES = 0.33, p_adj = 3.03e⁻⁸) and taurine (ES = 0.29, p_adj = 3.14e⁻⁷; Figure 2b). Although disease activity did not contribute significantly to BA profile variation, taurodeoxycholic (KW test, N = 128, ES = 0.16, p_adj = 1.04e⁻⁴; phD in Supplementary Table S9), taurochenodoxycholic (ES = 0.16, p_adj = 1.04e⁻⁴), and lithocholic acid (ES = 0.15, p_adj = 1.04e⁻⁴) concentrations were higher in UC-A than in HC samples (Figure 2c). The Bact2 enterotype displayed higher total BA concentrations (Wilcoxon test, N = 128, ES = 0.40, p_adj = 1.23e⁻²; Supplementary Table S9), with notably higher levels of the taurine-conjugated bile acids taurochenodoxycholic acid (ES = 0.40, p_adj = 5.37e⁻⁵) and taurodeoxycholic acid (ES = 0.32, p_adj = 1.86e⁻³). Interestingly, the total sum concentration of bile acids correlated negatively with the FDW (Spearman, ρ= −0.28, p = 1.5e⁻³).

To identify specific metabolites associated with the Bact2 enterotype, we applied the XGBoost machine-learning algorithm (a gradient boost extension of the Random Forest approach). A model built on 70% (N = 89) of the metabolome profiles correctly discriminated between Bact2 and non-Bact2 samples for 89.3% of the holdout test set (N = 39; balanced cross-validation AUC of 0.90), indicating that metabolites and enterotypes are highly interrelated (Supplementary Figure S2). Features contributing most to the predictions were identified using a Shapley Additive Explanations (SHAP) approach. High proportions of the amino acids tryptophan, l-glutamine, phenylalanine, and homoserine, as well as low proportions of the fatty acids maleic, butyric, and valeric acid were predictive for the Bact2 enterotype. Interestingly, there are also low pantothenic acid (vitamin B5) proportions pointed at the Bact2 enterotype (SHAP, N = 89, supplementary figure S1).

2.4. The impact of faecal water on butyrate oxidation gene expression in HT-29 cells depends on donor health status and enterotype

The impact of microbial metabolites on butyrate oxidation was evaluated using human colonic adenocarcinoma cells. HT-29 cells were incubated with a 1/20 dilution of filter-sterilized FW for 24 h. Next, cells were harvested and re-suspended in a Krebs-Henseleit buffer (4e⁶ cells/mL) containing 37kBq of Na-[1-¹⁴C]-labeled butyrate. After incubation for 2 h, released¹⁴CO₂ was captured with hyaminehydroxide and the radioactivity was measured by liquid scintillation counting. Additionally, the impact of exposure to FW on the expression of genes involved in the butyrate oxidation pathway (MCT1 [mono-carboxylate transporter], ACSM3 [acyl-CoA synthetase, medium chain], ACADS [acyl-CoA dehydrogenase], ECHS1 [enoyl-CoA hydratase, short chain 1], HSD17B10 [hydroxysteroid 17-beta dehydrogenase 10], and ACAT2 [acetyl-CoA acetyltransferase]) was assessed after the primary incubation of HT-29 cells in diluted FW. Butyrate oxidation rates did not differ significantly between cell cultures incubated with FW from HC and UC patients (KW test, N = 128, ES = 5.13e⁻⁴, p_adj = 0.59; phD in Supplementary Table S10; Figure 3a). However, FW from UC-A and -R patients reduced ACSM3 expression compared to HC (ES = 0.16, p_adj = 1.28e⁻⁴), while expression of ACAT2 was upregulated after incubation with UC-R and UC-A FW (ES = 0.08, p_adj = 7.28e⁻³). Similarly, butyrate oxidation rates did not differ between Bact2 and the other enterotypes (Wilcoxon test, N = 128, ES = 0.13, p_adj = 0.23; Supplementary Table S10), but ACSM3 expression was lower in cells exposed to Bact2 FW compared to non-Bact2 (ES = 0.25, p_adj = 0.03; Figure 3b). Overall, gene expression profiles are correlated with the concentration of butyric acid in FW (Spearman, N = 128, ACSM3, ρ = 0.49, p_adj = 2.14e⁻⁹; ACAT2, ρ=-0.637, p_adj = 5.62e⁻¹⁶; Supplementary Table S10). Similar correlations have been observed for acetate (ACSM3, ρ = 0.47, p_adj = 2.12e⁻⁸; ACAT2, ρ=-0.46, p_adj = 4.65e⁻⁸) and propionate (ACSM3, ρ = 0.41, p_adj = 1.90e⁻⁶; ACAT2, ρ=-0.42, p_adj = 7.51e⁻⁷). Neither gene expression profiles nor butyrate oxidation rates differed between paired samples from patients switching disease activity (Paired Wilcoxon rank sum test, N = 64, all p_adj > 0.05).

Figure 3.

Variation in butyrate oxidation rate & gene-expression profiles after fecal water incubation a. Butyrate oxidation pathway, abbreviation of the genes with measured expression levels are indicated in bold. Created with www.biorender.com.b Variation in gene expression levels normalized compared to β-actin levels compared between disease strata and according to Bact2 carrier status (Kruskal – Wallis with post hoc Dunn test, n  = 128, ***p_adj <0.001, **p_adj <0.01, and *p_adj <0.05; the body of the box plot represents the first and third quartiles of the distribution and the median line. The whiskers extend from the quartiles to the last data point within 1.5× the interquartile range, with outliers beyond.). c. Highlighted correlations between metabolites and the gene expression profiles (Spearman correlation, n = 128, for visualization purposes a linear fit [black] is added with a 95% confidence interval [gray]).

2.5. Cytokine production profiles of peripheral blood mononuclear cells change upon faecal water incubation in function of donor health status and enterotype

To investigate whether the fecal metabolome contributes to a pro-inflammatory immune response in UC, we incubated peripheral blood mononuclear cells (PMBC) with FW aliquots obtained from the cohort. PBMCs were isolated from the blood of three healthy subjects through density gradient centrifugation. Of these PBMCS, 10⁶ cells/well were incubated with 50 µL of a 1/20 FW dilution for 24 h, after which the cytokine profiles were measured. Incubation of PBMC with FW from UC-A and – R patients induced less IL-8 compared to cells exposed to HC FW (KW test, N = 128, ES = 0.11, p_adj = 4.94e⁻⁴; phD in Supplementary Table S12, Figure 4a). In contrast, UC-A FW induced production of more IL-1β (ES = 0.080, p_adj = 6.13e⁻³), while incubation of PBMC with UC-R resulted in higher concentrations of a TNF-α (ES = 0.049; p_adj = 2.80e⁻²). Higher production of IL-1β after incubation with UC-A compared to UC-R FW was confirmed in the paired samples from patients that switched disease activity (Paired Wilcoxon rank sum-test, N = 64, W-statistic = 487, p_adj = 1.02e⁻²). Exposure to Bact2 FW reduced IL-8 production (Wilcoxon test, N = 128, ES = 0.21, p_adj = 3.68e⁻²; Supplementary Table S12). Concentrations of IL-8 in FW pre-incubation correlated with production of IL-12p70 and IL-1β by PBMC (Spearman, N = 128, ρ = 0.21, p_adj = 1.29e⁻² and ρ = 0.17, p_adj = 4.90e⁻², respectively; Supplementary Table S12, Figure 4b), while FW IL-1β levels auto-correlated with IL-1β production (ρ = 0.21, p_adj = 1.53e⁻²).

Figure 4.

Variation in PBMC cytokine production profiles after fecal water incubation a. Variation in cytokine production profiles in PBMC between disease strata and according to Bact2 carrier status (Kruskal – Wallis with post hoc Dunn test, n  = 128, ***p_adj <0.001, **p_adj <0.01, and *p_adj <0.05; the body of the box plot represents the first and third quartiles of the distribution and the median line. The whiskers extend from the quartiles to the last data point within 1.5× the interquartile range, with outliers beyond.). b. Highlighted correlations between metabolites and the cytokine profiles (Spearman correlation, n = 128, for visualization purposes a linear fit [black] is added with a 95% confidence interval [gray]).

The potential impact of metabolites in FW aliquots on PBMC cytokine production profiles was evaluated through correlation analyses. Most notably, production of IL-8 and IL-1β was negatively correlated with dimethyl trisulfide and dimethyl disulfide FW concentrations (IL-8, Spearman, N = 128, ρ=-0.41, p_adj = 1.03e⁻⁶ and ρ=-0.31, p_adj = 4.00e⁻⁴; IL-1β, ρ=-0.38, p_adj = 4.00e⁻⁴ and ρ=-0.27 p_adj = 2.00e⁻³, respectively; Supplementary Table S11). Additionally, IL-8 measurements were associated with FW butyric acid concentrations (ρ = 0.37, p_adj = 1.48e⁻⁵), while IL-1β production negatively co-varied with a number of furan compounds, among which furan (ρ= −0.42, p_adj = 1.10e⁻⁶), 2-methyl-furan (ρ=-0.39, p_adj = 4.62e⁻⁶), and 2,5-dimethylfuran (ρ=-0.32, p_adj = 2.08e⁻⁴).

2.6. Faecal water obtained from patients is more cytotoxic compared to that obtained from healthy controls

Microbial metabolites may influence host physiology, including cell proliferation, cell differentiation, and apoptosis.¹⁵ To assess the impact on colonocyte survival, HT-29 cells were exposed for 72 h to serial dilutions (1/4–1/1024) of FW obtained from cohort participants. After exposure, cell survival was estimated from the reduction of a tetrazolium salt by the mitochondria of living cells to a colored formazan and expressed as the fold dilution at which 50% of the cells survived. FW from UC-R patients was more cytotoxic than that of HC (KW test, N = 128, ES = 0.040, p = 0.03; Figure 5a; phD in Supplementary Table S13). Bact2 status was not associated with cell survival (Wilcoxon test, N = 128, ES = 0.124, p = 0.16). FW dimethyl trisulfide and dimethyl disulfide concentrations were linked to cytotoxicity (Spearman, N = 128, ρ = 0.42, p_adj = 6.17e⁻⁷ and ρ = 0.37, p_adj = 1.07e⁻⁵, respectively; Supplementary Table S11), whereas pelargonic (ρ=-0.37, p_adj = 1.39e⁻⁵) and myristic acid (ρ=-0.37, p_adj = 1.73e⁻⁵) correlated positively with cell survival. Cytotoxicity of the FW did not change significantly with a switch of disease activity (Paired Wilcoxon rank sum-test, N = 64, W-statistic = 266, p = 0.98).

Figure 5.

Variation in cell death after fecal water incubation a. Variation in fecal water fold dilution rates required to achieve 50% cell death of HT-29 cells between disease strata and according to Bact2 carrier status (Kruskal – Wallis with post hoc Dunn test, n  = 128, ***p_adj <0.001, **p_adj <0.01, and *p_adj <0.05; the body of the box plot represents the first and third quartiles of the distribution and the median line. The whiskers extend from the quartiles to the last data point within 1.5× the interquartile range, with outliers beyond.). b. Highlighted correlations between metabolites and cell toxicity measurements (Spearman correlation, n = 128, for visualization purposes a linear fit [black] is added with a 95% confidence interval [gray]).

2.7. Faecal water from Bact2 samples affects epithelial barrier function of caco-2 colonocyte monolayers

To quantify the influence of FW of HC, UC-A, and UC-R samples on epithelial barrier function, we measured the evolution of the trans-epithelial resistance (TEER) of Caco-2 monolayers upon apical side exposure. Monolayers were incubated in diluted (1/5) FW for 24 h; TEER was quantified at 0 (100%), 2, 4, 8, and 24 h. Given the heterogeneity of individual TEER evolution over time, we identified clusters of similar trajectories using latent class growth analysis (LCGA). Applying third degree polynomials to fit the TEER data (LCGA, N = 128, BIC_1stDegree = 5767, BIC_2ndDegree = 5568, and BIC_3rdDegree = 5290), a three-class model was identified as optimal for trajectory classification (BIC_1class = 5290, BIC_2classes = 5160, BIC_3classes = 5142, and BIC_4classes = 5166). Based on TEER profiles, classes were characterized as constant, gradually decreasing, and sharply decreasing (Figure 6a). While the distribution of these three classes differed only marginally between disease groups (Chi-squared test, HC vs. UC-A vs. UC-R; χ² test, N = 128, χ² = 9.30, p_adj = 4.72e⁻²), the difference between Bact2 and non-Bact2 enterotyped samples was more pronounced (χ² = 16.01, p_adj = 3.33e⁻⁴). Indeed, TEER-trajectories of participants with the Bact2 enterotype remained stable in only 51.4% of cases, compared to 84.1% of those without Bact2 (Figure 6a). TEER profiles from patients switching between active disease and remission showed a similar distribution of classes as profiles from UC-A and UC-R samples (Fisher's exact test, N = 64, OR = 0.53, p_adj = 0.404). Also, in this subset, only 56% of incubations with Bact2 FW resulted in a constant TEER trajectory, compared to 86% for non-Bact2 samples OR = 0.23, p_adj = 2.76e⁻²; Figure 6b).

Figure 6.

Variation in TEER trajectories after fecal water incubation a. (left) Visualisation of the TEER trajectories for the cohort (n = 128). Each thin line represents an individual trajectory, whilst the bold lines represent the mean (+ standard error in grey) per latent growth class as obtained by LCGA. (right) Growth class prevalence associated with Bact2 carrier status (χ² = 16.01, p_adj = 3.33e⁻⁴). b. Variation in metabolite concentrations in faecal water associated with different latent growth classes (Kruskal – Wallis with post hoc Dunn test, n  = 128, ***p_adj <0.001, **p_adj <0.01, and *p_adj <0.05; the body of the box plot represents the first and third quartiles of the distribution and the median line. The whiskers extend from the quartiles to the last data point within 1.5× the interquartile range, with outliers beyond.).

When comparing the fecal water metabolite profiles between TEER-trajectories, azelaic acid (KW test, N = 128, ES = 0.16, p_adj = 3.01e⁻³), 3,5-dihydroxycinnamic acid (ES = 0.12, p_adj = 1.02e⁻²), and palmitic acid (ES = 0.12, p_adj = 1.09e⁻²) were overrepresented in the stable class, while higher concentrations of w-muricholic acid (ES = 0.19, p_adj = 5.01e⁻⁴) and l-glutamine (ES = 0.12, p_adj = 1.02e⁻²) were associated with both gradually and sharply decreasing profiles (Figure 6b; phD in Supplementary Table S14). Additionally, higher concentrations of taurochenodoxycholic acid levels were observed in FW dilutions inducing a gradual decline of TEER (ES = 0.14, p_adj = 4.87e⁻³).

3. Discussion

The present study aimed at evaluating a potential link between fecal microbiota metabolism and epithelial manifestations involved in the pathogenesis of UC, notably reduced butyrate oxidation rates, impaired barrier function, and an exaggerated immune response. Our findings confirmed reports of a disturbed gut microbial ecosystem associated with UC, with a higher prevalence of the Bact2 enterotype, previously linked to intestinal inflammation and faster transit.^10–12 In terms of microbial composition, the Bact2 enterotype has been characterized by an increase in aerotolerant opportunists such as Escherichia and Veilonella, along with a decrease in oxygen-sensitive taxa such as Faecalibacterium .^11,16 Although not statistically significant, a 13-percentage points difference in proportions of the Bact2 enterotype was noted between the active and remission condition among patients that provided both a baseline and follow-up sample. This lack of significance probably resulted from statistical power limitations, although it could also reflect desynchronization of remission/relapse and associated recovery/collapse of a eubiotic microbial community.

In addition, fecal metabolites in UC samples differed from controls. Concentrations of short (SCFA) and medium-chain fatty acids (MCFA) as well as intermediate components such as maleic acid were lower in Bact2-enterotyped stools and samples obtained from UC patients. With respect to UC-associated hyper-activation of the immune system, the lower butyrate concentrations are of key importance. Direct exposure of macrophages to butyrate resulted in down-regulation of lipopolysaccharide (LPS)-induced pro-inflammatory cytokine production through the inhibition of histone deacetylases (HDAC), which could attenuate immune response in UC.¹⁷ Additionally, several amino acids were more abundant in UC and Bact2 samples. Among these, tryptophan is of particular interest, because its downstream metabolites interact with the aryl hydrocarbon receptor (AHR). AHR pathway activation has been shown to promote immune tolerance.^18,19 Consistent with previous reports, bile acid concentrations were higher in samples from UC patients, possibly due to lower BA deconjugation, transformation, and desulfation, hindering colonic reabsorption.²⁰ The negative association between BA concentrations and fecal dry weights suggest a link with transit time, with fast passage rates reducing metabolism and uptake time window while associated lower pH levels additionally affect enzymatic activity.²¹ Reduced bile acid absorption limits farnesoid-X receptor (FXR) activation, decreasing the FXR-driven antagonism of inflammatory pathways.²² Finally, sugars and derivatives such as ribose, xylulose, and maltitol were depleted in both UC and Bact2 samples. Potentially, this could be driven by increased gut permeability, facilitating paracellular passage of sugars, lowering the eventual fecal concentrations.²³ However, as no information on dietary intake by participants was available, depletion caused by a lower intake of dietary fibers cannot be excluded.

Decreased colonic butyrate oxidation rates have been reported for UC patients based both on in vitro studies using isolated colonocytes²⁴ or mucosal biopsies²⁵ and in vivo studies using rectal instillation of labeled butyrate.⁸ In the present study, we observed altered expression of the butyrate oxidation genes ACSM3 and ACAT2 in HT-29 cells after incubation with FW of UC patients/Bact2 carriers when compared to HC/non-Bact2. ACSM3 (lower after incubation with patient/Bact2 FW) couples C2-C6 fatty acids to CoA, in preparation of actual β-oxidation, whereas ACAT2 (higher after incubation with patient FW) catalyses the reversible last step of the fatty acid β-oxidation producing acetyl-CoA out of acetoacetyl-CoA.²⁵ While both associations displayed opposing directions, correlations of gene expression levels with FW butyrate concentrations suggested dependency of substrate availability. Of note, butyrate oxidation rates in HT-29 cells were not significantly altered by incubation with UC/Bact2 FW. The latter implies that while HT-29 cells respond to butyrate concentrations, their potential to metabolize the substrate remained unaffected by other, unspecified constituents of the patient FW metabolome. Hence, our findings suggest that the lower butyrate oxidation rates observed in vitro using patient cells/tissue as well as in vivo can be considered a host manifestation of UC and are not a consequence of microbial metabolism.

The impact of FW on PBMC cytokine profiles reflects pre-incubation FW concentrations, for TNF-α and IL-1β. While these findings might be related to TNF-α mediated induction,²⁶ a potential role of the FW metabolome cannot be excluded. Notably, IL-1β production correlated negatively with a number of furan-containing compounds, a moiety which is under investigation for potential anti-inflammatory effects.^27,28 In contrast, induction of IL-8 in PBMC was higher after incubation with HC/non-Bact2 FW, whereas highest pre-incubation concentrations were measured in FW obtained from UC-A patients. In line with our findings, in vitro incubation of PBMC with butyrate and acetate, which have higher concentrations in HC/non-Bact2 FW, resulted in a higher production of IL-8.²⁹ Higher pre-incubation FW concentrations of IL-8 appear to reflect an altered immune response in UC rather than a response to deviant microbial metabolites, as supported by the higher production of IL-8 by macrophages isolated from IBD patients upon ex vivo stimulation by LPS when compared to those obtained from HC.³⁰

Development of colorectal cancer (CRC) is a serious complication of UC and the risk is highest for those with extensive disease of prolonged duration.³¹ Despite the fact that the risk of CRC-related deaths for UC patients has declined over time due to intensive surveillance programmes,³² when matching for tumor stage, UC patients with CRC still have a poorer 5-year survival than the general population.³³ FW cytotoxicity has been suggested to be implicated in CRC development, as increased cell death could induce compensatory epithelium hyper-proliferation, resulting in higher disease risk.³⁴ In the present study, bile acids, previously characterized as cytotoxic compounds because of their amphiphilic nature,³⁵ did not correlate with FW cytotoxicity. Instead, we observed a link between dimethyl disulfide and dimethyl trisulfide. While to the best of our knowledge not previously identified as cytotoxic, increased concentrations of both components might enhance toxic H₂S production.³⁶ FW cytotoxicity was not different between Bact2 and not-Bact2 donors. Potentially, this could be due to the relatively small sample sizes.

The impact of the fecal metabolome on TEER trajectories of Caco-2 cell monolayers was measured as an indication of overall permeability. Intestinal permeability of UC patients, in active and to a lesser extent in remission, is higher than in HC²³ and may result in LPS translocation, promoting inflammatory cascades.^18,23 FW application to the apical side of Caco-2 monolayers elicited three distinct responses. The distribution of these response types over patient groups and HC revealed only minor differences yet appeared to be linked more firmly to sample enterotype. A first response type resulting in a hardly affected TEER was mainly observed after incubation with non-Bact2 FW. In contrast, FW of Bact2 carriers more often induced a gradual (type 2) or sharp (type 3) decrease of TEER to < 50% of initial values. These observations may be linked to the lower (SC)FA concentrations measured in Bact2 FW, since SCFA alleviated LPS-induced intestinal barrier dysfunction, either by HDAC inhibition or by serving as energy substrate.³⁷ In addition, also BA may decrease epithelial barrier function partially explaining the increased permeability after incubation with high concentration of BA containing FW of Bact2 carriers.^38,39 The striking effect of Bact2 on overall intestinal barrier function in terms of transcellular transport should be investigated further by studying tight junction protein expression or measurement of molecule flux using fluoro- or radio labeled molecules to discriminate between leak or pore pathways of increased intestinal permeability.²³

The present study relies in part on the application of Caco-2 and HT29 immortalized cell lines as screening tools to investigate host-microbiota interactions. Both lines present enterocyte-like structural and functional features, combined with the additional advantages of their relative ease of culturing and lack of host-specific variability.⁴⁰ However, the conclusions of our study could be further explored in a series of follow-up experiments using physiologically more relevant human primary enterocytes as a model.^40,41 Notwithstanding the limitations, our results expand current insights in the role of the gut microbiota and the associated metabolome in the pathomechanisms of UCfor example, regarding the decrease of intestinal barrier function.⁴² The findings presented do suggest a potential role for microbiota modulation efforts as adjuvant therapies in UC clinical care, whether by means of dietary interventions or through the administration of live biotherapeutic products.^43–45

4. Conclusions

Overall, our analyses revealed that the FW of patients with UC may contribute to the pathogenesis of the disease. About 69% of UC-A patients, 31% of UC-R, and 3% of HC hosted the aberrant Bact2 microbiota, resulting in a significant overlap in the effects of disease activity and dysbiosis. Both samples from UC-A and UC-R patients as well as those enterotyped as Bact2 were characterized by low concentrations of butyrate and MCFA, key metabolites for maintaining colonic epithelium integrity. Our in vitro incubation experiments revealed a similar impact of FW metabolites from patients and Bact2-carriers on butyrate oxidation gene expression profiles and cytokine production. However, barrier function of cell monolayers was mostly affected by the Bact2 metabolome, while incubation with patient FW revealed enhanced cytotoxicity, associated with dimethyl disulfide and dimethyl trisulfide concentrations. Overall, our findings support the potential for gut microbiota modulation to revert epithelial dysfunction in IBD, targeting for example the restoration of barrier function by resolving enterotype-defined dysbiosis. Structured in vivo studies remain however required to mechanistically characterize the effect of correlating metabolites on the cellular responses observed.

4.1. Patients and methods

Detailed methods are available in the supplementary methods under the corresponding section.

4.2. Study population

We collected stool samples from HC, UC-A and UC-R. Disease activity of the patients was determined using the partial Mayo score. Quiescent disease was defined as pMayo ≤ 2 and active disease as pMayo ≥ 3.⁴⁶ Patients were recruited via the Gastroenterology Department of the University Hospital Leuven (Belgium) from 2011 to 2014. Exclusion criteria were a history of abdominal surgery (excluding appendectomy), antibiotic use or colonoscopy in the month preceding sampling, a disease extent limited to the rectum for UC patients, and kidney or liver disease, chronic GI disease and the use of antibiotics or GI medication in the month preceding sample collection for HC. The study was approved by the Ethics Committee of the University of Leuven (Belgium, S54170). All participants gave their written informed consent prior to sample collection.

4.3. Sample handling

Fresh fecal samples stored at 4°C were delivered to the laboratory within 24 h after collection. Upon delivery, fecal aliquots were stored at −80°C for microbial community profiling. The remaining sample was centrifuged at 50,000 × g at 4°C for 2 h to prepare FW. The FW was stored at −80°C and filtered through a 0.22 µm filter before further use.

4.4. Volatile organic compounds

VOCs were analyzed using a GC-MS system (Trace GC Ultra and DSQ II, Thermo Electron Corporation, Waltham, MA, USA), coupled on-line to a purge-and-trap system (SOLATek 72 and Velocity XPT, Teledyne Tekmar, Mason, OH, USA), as previously described⁴⁷

4.5. Non-volatile organic compounds

NVOCs were measured using GC-MS after extraction and derivatization, as previously described by Fiehn et al.⁴⁸ These analyses were performed by the West Coast Metabolomics Center (Davis, CA, USA).

4.6. Bile acids

Twenty-three BAs were absolutely quantified by LC-MS/MS using modifications of a previously published method.⁴⁹ Chromatograms were processed by Analyst Software (AB Sciex, Framingham, MA, USA) and MultiQuant Software (AB Sciex). Using internal standard methodologies, BA were absolutely quantified versus 7-point curves of external standards purchased from Steraloid (Newport, RI, USA), Sigma-Aldrich (St. Louis, MO, USA) and Medical Isotopes (Pelham, NH, USA). These analyses were performed by the West Coast Metabolomics Center (Davis, CA, USA).

4.7. Microbiota composition

The fecal microbiota profile of the FGFP cohort was described previously.¹⁰ Fecal DNA extraction and microbiota profiling for the new cohort followed the same protocols.¹⁰ Briefly, nucleic acids were extracted from frozen fecal aliquots using the PowerMicrobiome™ RNA Isolation Kit (MoBio). The manufacturer’s protocol was modified by the addition of a heating step at 90°C for 10 min after vortexing and by the exclusion of the steps where DNA is removed. For bacterial and archaeal characterization, the extracted DNA (dilution 1:10) was further amplified in triplicate using 16S rRNA primers 515F (5’- GTGYCAGCMGCCGCGGTAA-3’) and 806 R (5’- GGACTACNVGGGTWTCTAAT-3’) targeting the V4 region, modified to contain a barcode sequence between the forward primer and the Illumina adaptor sequences to produce single-barcoded libraries.¹⁰ Deep sequencing was performed on a MiSeq platform (2×250 PE reads, Illumina). All samples were randomized, and negative controls (PCR and extraction controls) were taken along and sequenced. After demultiplexing with sdm as part of the LotuS pipeline (v. 1.60)⁵⁰ without allowing for mismatches, fastq sequences were further analyzed per sample using DADA2 pipeline (v. 1.6).⁵¹ Briefly, we removed the primer sequences and the first 10 nucleotides after the primer. After merging paired sequences and removing chimeras, taxonomy was assigned using formatted RDP training set 16.

4.8. Osmolality measurements

FW osmolality (mOsm/kg) was determined using a cryoscopic osmometer (Osmomat 3000; Gonotec GmbH, Berlin, Germany).

4.9. Butyrate oxidation

Human colonic adenocarcinoma HT-29 cells were obtained from ECACC (European Collection of Cell Cultures) and grown in RPMI-1640 (Lonza Group Ltd, Switzerland) with 10% fecal calf serum (Lonza) and antibiotics (gentamicin sulfate; Lonza) at 37°C and 5% CO₂. Butyrate oxidation rate was measured in duplicate as described previously.⁵²

4.10. Relative quantification of the gene expression

The gene expression of the genes involved in butyrate metabolism (MCT1, ACADS2, ACSM3, ECHS1, HSD17B10, ACATS2) and the housekeeping gene β-actin was evaluated using qPCR, as previously described.⁵³

4.11. Faecal water cytotoxicity

FW cytotoxicity was measured in HT-29 cells using a tetrazolium salt-based colorimetric cell viability assay based as previously described.⁵³

4.12. Impact of faecal water on trans-epithelial electrical resistance

Human colonic adenocarcinoma Caco2 cells (European Collection of Cell Cultures) were cultured at 37°C and 5% CO₂ in DMEM with glucose and L-glutamine (Lonza Group Ltd, Switzerland) supplemented with 10% fecal calf serum (Lonza), 1% non-essential amino acids (Lonza), 10 mm 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES; Lonza), 4 µg/mL transferrin (Sigma-Aldrich, Steinheim, Germany) and gentamicin; (5 µg/mL; Lonza).

Caco2 cells (3 × 10⁴) were seeded on Transwell polycarbonate membrane inserts of 6.5 mm diameter and 0.4 µm pore size (Corning Incorporated, Corning, NY, USA). Caco2 monolayers were grown until TEER reached a plateau of at least 1000 Ω x cm² (World Precision Instruments, Sarasota, FL, USA). FW samples in a 1/5 dilution in fresh culture medium were incubated in duplicate on the apical side for 24 h. TEER was measured before incubation (0 h) and after 2 h, 4 h, 8 h and 24 h. results were expressed as % of the initial TEER.

4.13. Production of cytokines in faecal water and after incubation with isolated peripheral blood mononuclear cells

PBMCs were isolated from blood (200 mL) collected from healthy subjects (N = 3).⁵³ Cytokine secretions were measured using a fluorescence based cytometric bead array (CBA) analysis (BD Biosciences, San Diego, CA, USA), according to the manufacturer’s protocol. Interleukin (IL)1β, IL6, IL8, IL12p70 and tumor necrosis factor (TNF)α were measured in the undiluted FW samples and, after FW incubation, in the PBMC culture medium. For each FW sample, the average PBMC supernatant cytokine levels were calculated for the 3 PBMC donors. Cytokine concentrations were calculated using the FCAP Array software (BD Biosciences).

4.14. Statistical analyses

All statistical analyses were performed using R version 4.1.3. Missing values were imputed using half of the minimum obtained value per variable. α-Diversity (inverse Simpson) was calculated using the vegan package.⁵⁴ To estimate non-redundant variation stepwise distance based redundancy analysis (dbRDA) was used, determining p-values with 1000 permutations. For comparisons between multiple groups, Kruskal-Wallis with post-hoc Dunn tests were used, otherwise the Wilcoxon test was used (using the Rstatix package).⁵⁵ Correlations between metabolites, cytokine production, butyrate oxidation rate, and -gene expression and cytotoxicity were assessed using Spearman’s rank order correlation. All p-values were corrected for multiple testing when appropriate using the Benjamini – Hochberg method (p_adj), only p_adj < 0.05 were reported as significant. Visualizations were enhanced using ggplot2.⁵⁶ The microbiota datasets were clustered into enterotypes using the Dirichlet multinomial mixtures (DMM) method implemented in the R package DirichletMultinomial.¹⁴ The XGBoost and LCGA analysis parameters are available in the supplementary methods.

Abbreviations

ACADS

Acyl-CoA dehydrogenase gene

ACAT2

Acetyl-CoA acetyltransferase gene

ACSM3

Acyl-CoA synthetase, medium chain 3 gene

AHR

Aryl hydrocarbon receptor

BA

Bile acids

Bact1

Bacteroides1 enterotype

Bact2

Bacteroides2

BIC

Bayesian information criterion

CRC

Colorectal cancer

DA

Disease activity

dbRDA

Distance-based redundancy analysis

ES

Effect size

FDW

Faecal dry weight

FGFP

Flemish Gut Flora Project

FW

Faecal water

FXR

Farnesoid-X receptor

HC

Healthy controls

HSD17B10

Hydroxysteroid 17-beta dehydrogenase 10 gene

KW

Kruskall-Wallis test

LCGA

Latent class growth analysis

LPS

Lipopolysaccharide

MCT1

Mono-carboxylate transporter 1 gene

MCFA

Medium chain fatty acids

HDAC

Histone deacetylases

NVOC

Non-volatile organic compounds

OR

Odds ratio

pMayo

Partial mayo score

PMBC

Peripheral blood mononuclear cells

Prev

Prevotella enterotype

phD

Post-hoc Dunn test

Rum

Ruminococcaceae enterotype

SCFA

Short chain fatty acids

SHAP

Shapley Additive Explanations

TEER

Trans-epithelial resistance

UC

Ulcerative colitis

UC-A

Active ulcerative colitis

UC-R

Ulcerative colitis in remission

VOC

Volatile organic compounds

Funding Statement

JP is funded by a doctoral fellowship from Flanders Innovation & Entrepreneurship (VLAIO) [HBC.2017.0596]. The Raes lab is funded by KU Leuven, the Rega institute and VIB. GF was funded by the ReALity Innovation Fund, a Research Initiative of the State of Rhineland-Palatinate, Germany. The funders had no role in study design, interpretation, and writing of this work.

Disclosure statement

JR, GF, SV, SVS are inventors on the patent EP2018/084920: A new inflammation associated, low cell count enterotype issued to VIB VZW, Katholieke Universiteit Leuven, KU Leuven and Vrije Universiteit Brussel. JR, DV are inventors on patent EP14184535.4 - 12/09/2014: Biological sampling and storage container, International patent application on 14/09/2015: PCT/EP2015/070977 issued to VIB VZW, Vrije Universiteit Brussel and LRD. No potential conflict of interest was reported by the other author(s).

Author contributions

JP and LB contributed equally. LB, LD and GV acquired data. GV, VDP and LB performed the experimental work. JP, RT and LB prepared the data. JP, LB, DV SVS and GF performed statistical analyses. JP, LB, SVS, JR, GF and KV performed the interpretation. LB, VDP, SV and KV designed the study. JP, LB, GF and KV drafted the manuscript. All authors provided critical input.

Data availability statement

Raw 16S rRNA amplicon sequencing and selected host metadata will be available immediately following publication at the European Nucleotide Archive (ENA) via https://www.ebi.ac.uk/ena/browser/home under accession number PRJEB74446. Other individual de-identified participant data underlying the results reported in this article (text, tables, figures and appendix) will be shared upon reasonable request. Researchers who provide a methodologically sound proposal can request the data by contacting the corresponding authors.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19490976.2024.2424913