Anti-TNFα treatment in Crohn’s disease: Impact on hepatic steatosis, gut-derived hormones and metabolic status

2.1 Sample collection

Patients were prospectively recruited in the Department of Gastroenterology and Hepatology at Essen University Hospital from May 2015 until November 2017. All subjects provided informed written consent, and this study was approved by the Essen University Hospital Ethics committee (Institutional Review Board; 14-6044-BO). This study protocol follows the ethical guidelines of the Declaration of Helsinki.

Subjects were divided into 3 groups (Crohn's disease without anti-TNFα treatment: CD; Crohn's disease with anti-TNFα treatment: CD-TNF and healthy controls: control). Treatment with anti-TNFα-agents included medication with Infliximab (n = 6) or Adalimumab (n = 12). We performed liver stiffness measurement (LSM) to quantify potential liver fibrosis, steatosis was diagnosed by ultrasound and controlled attenuation parameter (CAP) or MRI spectroscopy, if available. Serum samples were collected in a fasted state and stored in aliquots at −80℃ until use for analysis. Standard laboratory parameters were evaluated via the central laboratory of the Essen University Hospital.

2.2 Faecal samples and microbiome analysis

Faecal samples were collected from every patient in sterile tubes and stored at −80℃ until DNA isolation. Individuals were excluded if they received antibiotic treatment during the past 4 weeks. DNA was isolated using the QIamp-DNA isolation kit following manufacturer's instructions (Qiagen, Hilden, Germany), including a mechanical lysis step using dry bead tubes (MoBio Laboratories Inc, Carlsbad, CA, USA) and the Fast Prep^™-24 instrument (MP Biomedicals, Solon, OH, USA) at 6.0 m/s for 45 seconds (2 times). Amplicon libraries were generated as previously described¹⁴ and sequenced on a MiSeq (2 × 250 bp, Illumina, Hayward, CA, USA). All FastaQ files were analysed using the dada2 package in R (www.r-project.org), and as a result, a unique table containing all samples with the sequence reads and abundances was generated. All samples were resampled to equal the smallest library size of 10 366 reads using the phyloseq package and returning 4611 phylotypes. Sequence reads were assigned to a taxonomic affiliation based on the naïve Bayesian classification with a pseudo-bootstrap threshold of 80%. Relative abundances in percentage of Phylotypes, Genus, Family, Order, Class and Phylum were used for downstream analyses. The vegan package (http://CRAN.R-project.org/package=vegan) was used to generate rarefaction curves, the EcoIndR for calculating the richness, relative rarity, Shannon, Simpson and taxonomy diversity indices. For microbiome analysis, both packages from R were used (version 3.4.2. 2017). Multivariate analyses were performed with Past3 (https://folk.uio.no/ohammer/past/), and univariate analyses were performed with GraphPad Prism 7 (GraphPad Software Inc, San Diego, CA, USA). The data comprising the 6 taxonomy ranks were used to construct sample-similarity matrices using the Bray-Curtis algorithm, where samples were clustered with 1000 bootstraps. Significant differences between a priori predefined groups of specimens were evaluated using analysis of similarity (ANOSIM with 9.999 permutations) and permutational multivariate analysis of variance (PERMANOVA with 9.999 permutations), and groups were considered significantly different if the P-value was ≤.05. The abundances of the 6 taxonomy ranks of those phylotypes with a mean >1% were compared by the Mann-Whitney test using the package exactRankTests (CIs at 95%) from R.

Bacterial diversity in faeces was compared between groups to identify individual bacteria as well as disease-associated bacterial signatures. Detected phylotypes were taxonomically assigned to 11 phyla, 22 classes, 35 orders, 67 families, and 168 genera. The global bacterial profiles at all phylogeny ranks (Phylotypes, Genus, Family, Order, Class, and Phylum) were compared using group-average agglomerative hierarchical clustering following sample-pairwise comparisons.

2.3 Serum and faecal sample metabolomics profiling

Faecal samples were extracted by sonification in 1:3 diluted extraction buffer (ethanol and phosphate buffer; Sigma-Aldrich, Steinheim, Germany) from approximately 300 mg of faeces; supernatants were used for quantification. Quantification of metabolomics analysis in serum or faecal sample extracts was performed by liquid chromatography coupled to Electrospray Ionization Tandem Mass Spectrometry (LC-ESI-MS/MS) using the Biocrates^® MxP Quant 500 Kit (BIOCRATES Life Sciences AG, Innsbruck, Austria), which covers up to 630 metabolites and lipids from 26 analyte classes. Data analysis was completed using the Biocrates MetIDQ software. Based on the ANOVA, significantly changed metabolites in each sample (either stool or serum) were used to perform metabolite enrichment analysis using Metaboanalyst (https://www.metaboanalyst.ca/). Lipid enrichment analysis was performed separately using http://lipidontology.com tool.¹⁵

2.4 Bioinformatics and statistical analysis

Statistical significance was determined using an unpaired, two-tailed t test or one-way ANOVA and nonparametric test (Kruskal-Wallis and Dunn's multiple comparison test correction for individual experimental conditions). Correlation analysis was performed using linear regression analysis; all analyses were performed with GraphPad Prism 9. If not stated otherwise, all data are presented as mean ± SEM. Significance was assumed at P ≤ .05.

2.5 Magnetic resonance imaging (MRI) for liver fat detection and bowel movement analysis

MRI of the bowel and liver was performed on a 3 Tesla magnet (Magnetom Avanto, Siemens Health Care, Erlangen, Germany). Conventional axial in and opposed phase (IOP) imaging of the liver was performed and average values for fat and water images were measured using regions of interest (ROI). Thereafter, fat signal fractions can be calculated as shown by Reeder et al¹⁶. Coronal real time true-Fisp sequences of the bowel were acquired for automatic generation of parametric maps facilitating quantification of bowel motility as previously reported.¹⁷

3.1 Characteristics of the study cohort

Forty-nine subjects, including healthy controls (n = 10), CD without anti-TNFα treatment (CD; n = 21) and CD patients receiving anti-TNFα agents (CD-TNF; n = 18), were recruited. Anti-TNF medication was based on intravenous medication with Infliximab (n = 6) or subcutaneous medication with Adalimumab (n = 12). Controls were younger, and CD patients without anti-TNF treatment were older. All CD patients were in clinical remission or had moderate symptoms. No patient had severe symptoms, according to standardized clinical documentation. While Crohn's Disease Activity Index (CDAI) was only available in a subset of patients, there was no significant difference between groups (data not shown). Faecal calprotectin levels were not significantly different between CD and CD-TNF. Table 1 provides an overview of the demographic data of the individual groups.

3.2 Attenuation of hepatic steatosis in CD is associated with anti-TNF treatment

We performed LSM to quantify potential liver fibrosis as well as CAP and fat-fraction measurement by MRI spectroscopy to assess hepatic fat accumulation. CAP values and MRI-fat fraction were higher in all patients with CD (with and without anti-TNF treatment) than in controls (Figure 1A-B). Interestingly, CAP values and MRI-fat fractions were higher in CD patients without anti-TNF therapy than those receiving anti-TNFα agents or controls (Figure 1C-D). In contrast, there was no difference in LSM between any of the groups (Figure 1E). LFTs seem to be lower in the control group; however, those changes failed to show significance (Table 1). The inflammatory status of CD was controlled by the assessment of calprotectin in stool samples and serum-CRP. Here, we did not find any pronounced differences between the 2 CD groups (Table 1). To investigate whether additional medication or conditions impact steatosis, we compared 2 groups (no or mild steatosis: S0/S1; CAP <270 db/m vs severe steatosis S2/S3; CAP>270 db/m). Patients with severe steatosis were less likely to be on anti-TNF treatment but more likely to receive Mesalazine (Table S1). However, there was no difference in CAP and fat fraction between Mesalazine users and nonusers (data are not shown). It should also be noted that neither the use of steroids (n = 6 patients) nor the type of anti-TNF therapy (whether adalimumab or infliximab) had any influence on the CAP value in CD-TNF patients (Tables S2 and S3). Information, if patients received various medications including immune modulatory drugs and CD-specific drugs (including Mesalazine) are summarized in Table S4. Of note, in our cohort, the rate of ileocecal (valve) resection was higher in patients without anti-TNFα treatment (Table 1). Those patients with resection did also present with higher CAP value (Table S1) and lower FGF19 levels (Figure S1A).

3.3 Alterations in gut-liver hormone levels are associated with hepatic steatosis in CD patients

Fibroblast growth factor 19 (FGF19) acts as a gut-derived hormone, suppressing hepatic de novo BA synthesis and regulating lipid and glucose metabolism.⁷ The Farnesoid X receptor (FXR) regulates intestinal FGF19 expression and its release into the bloodstream.¹⁸ CD patients without anti-TNF treatment had increased serum FGF19 compared to control. Serum FGF19 levels were even higher in CD-TNF compared to CD patients (Figure 2A). However, it should be noted that the whole CD group also included a larger number of patients with ileocecal resection (Table 1). In the resected CD patients group, FGF19 levels were lower than in the group without resection (Figure S1A). Also, in CD patients, FGF19 levels correlated inversely with steatosis grade assessed by CAP (Figure 2B) and MRI (Figure S1B). Intriguingly, there was no association between CAP and FGF19 for the total cohort, including healthy controls (Figure S1C). There was also no evidence that differences in FGF19 levels were related to the age of the cohort (Figures S1D).

Glucagon-like peptide 1 (GLP1) is another gut-derived hormone that is released from enteroendocrine cells. FXR activation inhibits GLP1 synthesis and release.¹⁹ In our cohort, CD patients had lower serum levels of GLP1 compared to controls (Figure 2C). There was a direct relationship between GLP1 and CAP (Figure 2D) and GLP1 and MRI (Figure S1E). Again, the association between GLP1 and CAP was present in the CD cohort only (Figure S1F). There was also a positive correlation between GLP1 and liver injury marker ALT (Figure 2E). ALT trended towards a negative correlation with FGF19 but was not significant (Figure 2F).

3.4 Alterations in metabolite and lipid profiles are associated with anti-TNF treatment

We performed metabolomic and lipidomic profiling of serum and stool samples using mass spectrometry. We performed partial least squares discrimination analysis (PLS-DA) in order to highlight group differences which were rather found between the stool's metabolite profile than the serum samples (Figure 3A).

One-way ANOVA was performed on each of the datasets (serum and stool) to identify metabolite concentration and identity changes. Figure 3B shows volcano plots with dysregulated metabolites in each comparison. In serum, we detected an increase in CD patients’ triglycerides compared to controls, which was not observed in serum from CD-TNF patients. Stool samples had increased ceramide levels and cholic acid (CA) in CD samples compared to controls. An increase in ceramides was less pronounced in stool from CD patients with anti-TNF treatment.

Based on the ANOVA, significantly changed metabolites in each sample (either stool or serum) were used to perform metabolite enrichment analysis. Based on this analysis, there was enrichment in serum of ammonia recycling, methylhistidine metabolism, methionine metabolism and other pathways (Figure 3C).

Serum contained substantial enrichment in sphingolipids, saturated fatty acids, hexosylceramides, dihexosylceramides and other lipid categories. An overview of the Lipid Ontology enrichment analysis of serum samples is provided in (Figure 3D).

Levels of neutral lipids such as triglycerides were reduced in the serum of patients on anti-TNF treatment while there are only 2 metabolites common in both CD and CD-TNF patients (Figure 3E). Similar interesting observations were seen in the stool samples. Two BA (CA and glycol-ursodeoxycholic acid; GUDCA) and several ceramides and glycosylceramides were increased in the stool of CD samples while their level was reduced anti-TNF treated cohort (Figure 3E).

3.5 The abundance of individual phylae in CD is associated with anti-TNF treatment

Although there were no differences in phylotype richness and Shannon index between groups (Figure S2A-B), we observed obvious differences between HC and CD in evenness phylotypes represented by Pielous's Evenness index and Simpson index (Figure S2C-D). The results demonstrate that a few bacterial groups comprise the central part of the entire community in CD patients, whereas, in HC, a higher community complexity was found.

The most abundant phylotypes belonged to the phyla Bacteroidetes, Actinobacteria, Firmicutes, Proteobacteria and Fusobacteria, which comprised approximately 99% of the total bacterial community (Figure 4A, B). At the bacterial phylum level in NAFLD, Bacteroidetes and Firmicutes are reported to be decreased, while proteobacteria levels are increased.²⁰ In our cohort, Firmicutes were gradually reduced in abundance from control to CD-TNF to CD (Figure 4C). In comparison, Proteobacteria were increased in control and CD-TNF compared to CD (Figure 4D). Fusobacteria, Bacteriodetes and Actinobacteria's abundances did not show relevant alterations (Figure 4E-G).

3.6 The abundance of specific families in CD

The most abundant families were Ruminococcaceae, Bacteroidaceae, Enterobacteriaceae, Veillonellaceae, Acidaminococcaceae, Lachnospiraceae, Rickenellacea, and Prevotellaceae, Porphyromonadaceae (Figure 5A, B). At the family level in NAFLD, Rikenellaceae and Ruminococcaceae were decreased, and Enterobacteriaceae increased.²⁰ In our cohort, Enterobacteriaceae were gradually increased in abundance from control to CD-TNF to CD (Figure 5C). Contrary Ruminococcaceae were decreased when comparing control to CD-TNF and CD (Figure 5D) while the abundance of between groups Rikenellaceae was not significantly changed (E).

3.7 Association of serum and stool metabolites with the abundance of Firmicutes and Ruminococcaceae

We sought to correlate metabolites and lipids in the metabolomic analysis. We linked those with the abundance of individual phylae and families, which differed between groups in the previous analysis. Analyses show a negative correlation of 83 different triglycerides (TG) in serum with the abundance of Firmicutes (Figure 6A) and Ruminococcaceae (Figure 6B) respectively. As fatty acids (FA) play a crucial role in steatosis pathogenesis, we also correlated serum FA from the metabolomic analysis with the bacterial abundance. In contrast to TG, we failed to show any association in this case (data not shown).

Further correlations with Proteobacteria and Enterobacteriaceae did not provide any differences. Correlation analysis in stool shows a negative association of CA with Firmicutes (Figure 6C) and Ruminococcaceae (Figure 6D). Despite showing differences in metabolomic stool analysis, no correlation was found between ceramides and the abundance of any of the bacteria tested (data not shown).

3.8 Bowel-movement frequency correlates with therapy-induced changes in gut hormones

We used MRI of the bowel in 15 patients to determine if metabolomic and lipid changes correlated with activity (CD n=9; CD-TNF n=6). We assessed voxel velocity as a surrogate marker for bowel activity as previously described.¹⁷ CD patients on anti-TNF treatment showed enhanced bowel-movement activity compared to individuals without anti-TNF treatment (Figure 7A). Interestingly, bowel-movement activity is positively associated with serum FGF19 levels and negatively correlated with GLP1 levels (Figure 7B, C). There was no difference in bowel-movement frequency between patients who received ileocecal resection and those who did not (data not shown).