Tissue and stool microbiome in pediatric inflammatory bowel disease patients: diversity differs in patients with relapsing and non-relapsing Crohn’s disease

Inflammatory bowel diseases (IBD) are a group of chronic, relapsing inflammatory conditions of the gastrointestinal tract with increasing prevalence in the population [1]. Two major subtypes, Crohn’s disease (CD) and ulcerative colitis (UC), exhibit distinct clinical and pathological features. The pediatric cases (pIBD) are frequently associated with a more complicated disease course compared to adult-onset IBD [2]. The prevalence of pediatric IBD varies across studies but is generally increasing, with the greatest numbers in Western Asia and Western Europe (up to 21.6 and 17.4 cases per 100,000 person-years, respectively) [3].

The etiology of pIBD remains elusive, but it involves a complex interplay of genetic, environmental, and microbial factors. The disruptions in the gut microbiota composition and function are implicated in the pathogenesis of IBD in both pediatric and adult patients and may involve the loss of certain taxa of microorganisms and a reduction in the overall alpha diversity of the microbial community. However, whether dysbiosis is a cause or a consequence of the inflammatory process remains unclear. Prior research on tissue samples has revealed that pIBD patients consistently exhibit lower gut microbiome diversity compared to healthy individuals. Furthermore, these studies report a characteristic shift in microbial composition, marked by decreased levels of Firmicutes, and the genera Faecalibacterium, and Bifidobacterium, and increased levels of Escherichia or Veillonella in the gut microbiota of pIBD patients [4‐6].

Current diagnostic approaches rely on a combination of clinical, endoscopic, and histological assessments. Stool analysis provides a non-invasive assessment of gut microbiota composition and presence of inflammatory markers (i.e., fecal calprotectin), while tissue biopsies obtained through endoscopy offer valuable information about the extent and severity of intestinal inflammation and histological features characteristic of each IBD subtype. Despite significant progress in the management of pediatric IBD, about one-third of patients develop a relapse within a year of diagnosis [7]. A reliable prognostic marker that would predict the course of the disease at the time of diagnosis would help to set a treatment strategy and reduce the incidence of relapse.

In this study, we characterized the tissue and fecal microbiome of a therapeutically naïve cohort of 33 pCD patients, 23 pUC patients, and 26 non-IBD pediatric controls. The variable outcomes observed within these pIBD groups allowed for the identification of potential prognostic markers associated with pCD relapse.

A biopsy sample stored in RNAlater™ Stabilization Solution (Invitrogen™, Thermo Fisher Scientific, Waltham, MA, USA) was thawed on ice, and 1 × 1 mm section was aseptically separated. The resulting tissue fragment was placed into ZR BashingBead™ Lysis tubes (0.1 & 0.5 mm) containing 750 µL of ZymoBIOMICS™ Lysis Solution, which is part of the ZymoBIOMICS™ DNA Miniprep Kit (Zymo Research, Irvine, USA). The stool sample stored in ESwab tubes with Amies medium (Copan, Ave, Murrieta, USA) was thawed on ice, homogenized by vortexing, and 200 µL was added to the ZR BashingBead™ Lysis tubes (0.1 & 0.5 mm), along with 550 µL of ZymoBIOMICS™ Lysis Solution. In addition, ZymoBIOMICS Fecal Reference with TruMatrix™ Technology and ZymoBIOMICS Gut Microbiome Standard (both Zymo Research, Irvine, USA) were used as positive controls for both DNA isolation and sequencing. The taxonomic composition from the sequencing controls correlated with the expected composition (Spearman’s rank correlation: r_s = 0.799 and r_s = 0.939, respectively; both p < 0.001). Microbial DNA-free water (Sigma-Aldrich, Burlington, Massachusetts, USA) was used as a negative control. To each sample, 20 µL of Proteinase K (20 mg/mL stock concentration; GEXPRK01-I5, Elisabeth Pharmacon, Brno, Czech Republic) was added, and the sample was incubated for 1 h at 55 °C and 600 RPM in the Thermomixer comfort. Homogenization was performed 4 times for 60 s at 9000 RPM, with a 2-minute pause between cycles, using the Precellys Evolution homogenizer (Bertin Technologies SAS, France). The subsequent steps of the extraction followed the manufacturer’s recommendations for the ZymoBIOMICS™ DNA Miniprep Kit (Zymo Research, Irvine, USA). The quality of the isolated DNA was checked using a NanoDrop™ 2000/2000c spectrophotometer. The quantity of extracted DNA was measured using the QuantiFluor^® dsDNA System, and the isolated DNA was stored at − 20 °C until further analysis.

Raw paired-end 16S rRNA gene sequences underwent quality control using FastQC (v0.11.5) [12] and FastQ Screen (v0.15.3) [13]. Adapter trimming and quality filtering (Q ≥ 25, minimum length 40 bp) were performed with Fastp (v0.20.1) [14]. Host reads were subsequently removed by alignment against the GRCh38 reference genome using Bowtie2 (v2.4.2) [15]. Denoising, chimera removal, and amplicon sequence variant (ASV) inference were performed using DADA2 implemented via QIIME 2 (v2024.5). Each sequencing run was processed separately for error correction, read merging, and chimera detection. Taxonomic assignment of representative ASVs obtained by DADA2 was performed using the Naïve Bayes classifier [16] and the SILVA database v138 (99% identity) [17]. Contaminant ASVs were identified and removed using Decontam (v1.12.0), applying the prevalence method with negative controls indicated in the metadata. Taxa with contamination scores exceeding 0.1 were discarded. The ASV table was further filtered by minimum abundance (≥ 10 reads across all samples) and minimum prevalence (≥ 2 samples).

Alpha diversity metrics (Shannon, Simpson, observed ASVs) and beta diversity metrics (Bray–Curtis, Jaccard, unweighted UniFrac, and weighted UniFrac) were calculated on rarefied feature Tables (10,000 reads/sample). Group differences in alpha diversity were assessed using the Kruskal–Wallis test, while beta diversity differences were tested using PERMANOVA. Principal coordinates analysis (PCoA) plots were constructed using beta diversity metrics to visualize sample clustering patterns. Differential abundance analysis was performed using ANCOM-BC, a QIIME 2 plugin (v2024.5), with a significance threshold of 0.05.

The multivariate logistic regression models were applied in order to identify a combination of predictors that hold prognostic value for disease outcome. The best model for each outcome was determined through bidirectional stepwise selection, which iteratively added or removed microbial taxa to minimize the Akaike information criterion (AIC). Predictive performance was evaluated through the sensitivity and specificity of the models, summarized in ROC curves (https://cran.r-project.org/web/packages/boot/index.html). A risk score formula for predicting individual outcomes was developed as a linear combination of microbial abundance levels, weighted by regression coefficients derived from the logistic regression model. Patients were stratified into high-risk and low-risk groups based on a threshold designed as the value maximizing the sum of sensitivity and specificity. Confidence intervals (CIs) for sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) for the threshold were calculated using Wilson’s formula with 2,000 bootstrap replicates of the ROC curve [18, 19]. An AUC greater than 0.7 was considered the threshold for acceptable markers [20].

The diversity of microbial community composition was evaluated in tissue samples from pIBD patients and a control group. Alpha diversity analysis revealed no statistically significant differences in taxon richness between patient and control groups. However, both the pCD and pUC groups showed a trend towards decreased richness (p = 0.083 and 0.086, respectively; Fig. 2A, B) and towards lower Shannon index values compared to controls (p = 0.105 and 0.138, respectively; Fig. 2A, B). Beta diversity analysis showed a clear difference between the microbial communities of controls and patients with pCD (p = 0.004) and pUC (p = 0.001; Fig. 2C). Based on differential abundance analysis, in pCD tissue samples, 15 taxa were enriched and three were depleted. In pUC samples, nine taxa were enriched and 24 were depleted (Fig. 2D). At the genus level, Enterococcus was the most significantly enriched taxon in both pCD and pUC patients. The most depleted taxa were Coprobacter in pCD and Alistipes in pUC (Fig. 2D). A complete list of abundances of significantly different taxa is shown in Supplementary Table S2.

The highly variable clinical course of inflammatory bowel diseases (IBD) presents a major challenge in pediatric patient diagnosis and management [22]. Moreover, the limited correlation observed between clinical activity, biological markers, and endoscopic findings in IBD patients complicates accurate assessment of disease severity [23]. Prediction of pIBD prognosis plays an important role in the clinical decisions and disease management and wPCDAI/PUCAI and FCP are currently used by clinicians as non-invasive markers for pIBD prognosis prediction [24, 25]. In the present study, we have observed that pCD patients who relapsed within one year after diagnosis had a significantly higher wPCDAI value at diagnosis. Moreover, the clinical scoring index wPCDAI demonstrated a moderate discriminatory power (AUC = 0.744) when considering that it offers a non-invasive approach prognosis assessment of pCD patients.

Given the very limited number of available pIBD tissue microbiome studies [26, 27], this research, encompassing 33 pCD patients, 23 pUC patients, and 26 non-IBD controls, makes an important contribution. In addition, the microbiome composition of pIBD patients remains an unexplored area for the identification of prognostic markers. This study found a decreased overall alpha diversity in the tissue microbiome of pIBD patients, consistent with the previous findings in pCD [26] and pUC [28]. Similarly, the determined beta-diversity differences observed here align with other studies [29, 30]. While the specific enriched and depleted taxa differed between pCD and pUC in our cohort, certain taxa, such as Enterococcus, were enriched in both, which has been previously reported [31].

To identify the most effective prognostic markers for pCD in tissue microbiome, we employed the ROC curve analysis. Our results revealed that specific microbial taxa demonstrated significant discriminatory potential between pCD patients with and without relapse. Notably, Barnesiella (AUC = 0.818), Butyricimonas (AUC = 0.790), and Collinsella (AUC = 0.773) emerged as promising markers, with all three taxa showing depletion in pCD patients experiencing relapse. Conversely, none of the taxa significantly enriched in pCD patients with relapse exceeded AUC threshold of 0.7, suggesting that the depletion of certain microbial members might be a stronger indicator of a poorer prognosis in pCD. Previous attempts to identify microbial markers for the prediction of pIBD severity include the work by Meij et al. (2018) [32], which found that reduced Alistipes finegoldii and A. putredinis among pIBD patients could be a good indicator of the disease severity (AUC = 0.87). The study by Wang et al. (2021) [33] proposed a disease severity prediction based on eleven bacterial genera (11-OTU stool model, AUC = 0.84–0.88). In contrast, the overall stability of the microbiome in pUC patients, irrespective of their disease activity status [34], suggests why a predictive marker for relapse was not found in this study. As suggested by our study, the genus Barnesiella is a promising microbial marker for predicting pCD disease prognosis. The depleted abundance of Barnesiella in pCD patients with relapse aligns with the findings reported by Alipour et al. (2016) [28], who showed a decreased abundance of the Barnesiellaceae family in tissue samples from both pCD and pUC patients compared to non-IBD children. Chen et al. (2022) [35] identified the genus Barnesiella as a potential marker, among five others, for predicting positive patient responsiveness to adalimumab, a commonly used IBD treatment that targets and neutralizes TNF-alpha. Moreover, Vestergaard et al. (2024) [36] demonstrated a lower abundance of Barnesiella in individuals with IBD as well as decreased abundance of UCG-003 (family Oscillospiraceae), which they noted as the first such observation in the literature [36]. Consistent with these findings, our study also found UCG-003 to be significantly depleted in both the tissue and fecal microbiomes of pCD patients experiencing relapse. The significance of Barnesiella depletion extends beyond IBD, as it has been identified in a meta-analysis as a more universal microbial marker for additional intestinal diseases, including colorectal cancer [37]. A higher abundance of Barnesiella in healthy individuals compared to those with intestinal diseases suggests its correlation with intestinal health [37]. Moreover, the study of Mancabelli et al. (2017) [37] showed a high abundance of Christensenellaceae R-7 group in healthy individuals, and this taxon was found to be depleted in pCD patients with relapse in our study (with an AUC of 0.733 determined in this study). Although the precise role bacterial genera associated with microbiome of healthy individuals is not clear, Barnesiella may be involved in degradation of complex carbohydrates, facilitating cross-feeding interactions with other beneficial gut microbes often reduced in IBD [38].

In our study, alpha and beta diversity significantly differed between relapsing and non-relapsing pCD patients in tissue samples, however, this distinction was not evident in stool samples. A similar finding was reported by Gevers et al. (2014) [26] that suggested a higher diagnostic power in tissue samples compared to stool samples. However, other authors including Wang et al. (2016) [39] found results that are comparable between tissue and stool samples. A depletion of Barnesiella genus found in this work was observed in both tissue and stool samples from pCD patients experiencing a relapse. Confirming this depletion in both sample types is critical for establishing a reliable prognostic marker. Moreover, this finding is important since the identification of prognosis markers of pCD patients in stool samples could avoid repeated intestinal biopsies. The Barnesiella depletion showed strong standalone predictive performance for relapse within 12 months in tissue (AUC = 0.818), but poor performance in stool (AUC = 0.696). However, this prediction was considerably stronger when combined with wPCDAI (AUC = 0.872 in tissue and 0.852 in stool). This combined performance is substantially higher than the wPCDAI standalone score (AUC = 0.744). Predictors with an AUC >0.8 are generally considered excellent markers [20]. At the individual patient level, Barnesiella in tissue sample microbiome was completely absent in 12 out of 16 relapsing patients and only in 3 out of 11 non-relapsing patients (Fig. 4B), suggesting the presence of both qualitative and quantitative differences between pCD and controls. The abundance of Barnesiella in the human microbiota appears to be a promising prognostic marker, but establishing its clear clinical threshold may be challenging, particularly for tests relying on real-time PCR detection. Setting this threshold remains a goal for future research. In this context, determination of complete 16S rRNA gene microbiome analysis, including determination of microbiome diversity, composition, and taxon richness, may offer a more reliably interpretable alternative, as in this case, Barnesiella abundance thresholds could be defined based on its percentage within the total microbiome and could be determined together with microbiome richness.

Unlike most other studies that primarily analyze the fecal microbiome, a key strength of this study lies in the simultaneous analysis of both tissue and stool samples obtained from the same pIBD patients. This unique approach enabled us to establish microbiome characteristics directly from tissue, which better reflects the patient’s intestinal environment, and then directly correlate these findings with corresponding stool samples. However, the limitations of this study include a relatively small number of patients and non-IBD controls within the cohorts, imperfectly matched biopsy areas between patient and non-IBD control samples, and the exclusion of patients with persistent active disease after induction therapy from relapse group, which may have introduced selection bias and potentially limits the generalizability of our findings to the broader pediatric IBD population.