This study included 41 patients under the medical follow-up by a gastroenterologist at Soon Chun Hyang University Bucheon Hospital between 2021 and 2022, consisting of 19 UC patients and 22 CD patients. The inclusion criteria were South Korean adults. They were restricted from taking antibiotics, probiotics, and fermented dairy products for at least 3 weeks before stool sampling. Pregnant women and those diagnosed with other diseases were excluded. Disease severity was assessed based on the Mayo score and Truelove–Witts severity score for UC patients (Truelove and Witts, 1955; Rubin et al., 2019) and the Crohn’s disease activity index for CD patients (Di Palma and Farraye, 2011). All participants provided their stool samples under their informed and signed consent. Additionally, the clinical database contained the blood test results obtained from the hospital for diagnostic purposes, along with patient information (Supplementary Data S1). This protocol was approved by the Institutional Review Board of Soon Chun Hyang University Bucheon Hospital (SCHBC 2021–01–028-002). To compare the differences in gut microbiota between each IBD group and healthy individuals, we applied the 16S rRNA gene amplicon sequencing results (n = 11) and culturomics findings (n = 8) from healthy subjects analyzed in our previous study to this study (Park et al., 2024). Healthy controls (HC) were defined as individuals with no medical history of gastrointestinal or metabolic diseases, recent antibiotic use, or consumption of fermented milk or probiotics.

To minimize changes in bacterial composition and viability in stool, we followed the Human Microbiome Project stool collection protocol (Wu et al., 2019) with some modifications (Supplementary Figure S1). Clinicians provided patients with sufficient information and education on how to store stools immediately after defecation at home. The samples were kept at a low temperature (< 4°C) and under anaerobic conditions with a GasPak EZ anaerobe container system (Becton Dickinson, MD, United States) until arrival at our laboratory. All samples were processed within 24 h of collection in an anaerobic chamber with a gas mixture of 5% CO₂, 10% H₂, and 85% N₂ to eliminate oxygen. The shape and condition of stool samples were evaluated based on the Bristol stool scale and disease activity index, which included an assessment of stool consistency and visible blood. Stool consistency was scored as normal (0), loose stool (1), or diarrhea (2). Blood in the stool was scored as no visible blood (0), streaks of blood (1), apparent blood (2), or mostly blood (3) (Ogata et al., 2017).

A portion of the sample was placed in a collection tube for later quantification of stool inflammatory biomarkers (fecal calprotectin, FCP; fecal occult blood, FOB) using enzyme-linked immunoassay tests from CalproLab (Oslo, Norway) and Epitope Diagnostics Inc. (CA, United States), respectively. The remaining stool sample was homogenized and resuspended in saline at a concentration of 0.25 g/L. Some portions were stored at −80°C until DNA extraction and the rest were utilized for bacterial culturomics.

We followed a streamlined culturomics protocol proposed in our previous study (Park et al., 2024). Fecal suspensions were immobilized into polysaccharides consisting of xanthan and gellan gums for a long-term incubation (Cinquin et al., 2004; Yeo et al., 2023; Park et al., 2024). The fecal gel beads were inoculated into preculture media at a final concentration of 5 g stool/L. The streamlined culturomics was conducted under aerobic and anaerobic conditions for 30 days at 37°C. These included BACT/ALERT FAN Plus Culture Bottles (BioMérieux, Marcy l’Etoile, France; FA PLUS Ref. 410851 for aerobic cultivation; FN PLUS Ref. 410852 for anaerobic cultivation) and modified Gifu Anaerobic Medium (mGAM; Nissui Pharmaceutical, Tokyo, Japan) (Park et al., 2024). All preincubation media were supplemented with 10% rumen fluid and 10% sheep blood. The cultured solutions were regularly taken (> six times for 30 days), serially diluted into saline, and inoculated on mGAM agar for bacterial isolation. After serial dilution, the fecal suspension without preincubation was also directly plated onto mGAM agar. Based on the colony morphology differences, colonies were randomly picked, inoculated into mGAM broth, and plated onto mGAM agar. Bacterial colonies were identified using MALDI-TOF MS on a MALDI Biotyper Sirius system (Bruker Daltonics, Bremen, Germany). If the signal of the bacterial spectrum was insufficient or did not match the MBT 8,468 MSPs library (score ≤ 1.69), we extracted genomic DNA using Chelex 100 resin (Bio-Rad Laboratories, CA, United States). The 16S rRNA gene was then sequenced using 27F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 1492R (5′-GGTTACCTTGTTACGACTT-3′) primers using an ABI PRISM 3730XL DNA analyzer (Applied Biosystems, CA, USA) by SolGent (Daejeon, South Korea). Identified strains were preserved in 10% glycerol stock solution and stored at −80°C. The taxonomic lineage and names were updated based on the List of Prokaryotic names with Standing in Nomenclature database (LPSN).¹ Species that showed the 16S rRNA gene sequence similarity of less than 98.65% with the closest neighbor were classified as potential novel species. A phylogenetic tree was constructed using PhyloT² based on the 16S rRNA reference genes, which contained 360 species of this study. The tree was visualized using the iTOL tool³ (Letunic and Bork, 2024). We utilized the phenotypic characteristics of type strains registered in BacDive⁴ for oxygen tolerance and Gram characteristics. We also applied the bacterial oxygen tolerance database.⁵

Unlike the next-generation sequencing-based method that provides information on relative abundance based on ASV, the form of data obtained from culturomics is the number of isolated microorganisms under the given culture conditions. We followed the calculation methods for culturomics defined by Tidjani Alou et al. (2017) to compare isolated gut bacterial composition between groups. One of the key measures is the unique/total (U/T) ratio, which is the ratio of the number of unique species to the total species in each group. In this context, unique species were defined as bacteria isolated from only one subject in each group. Another critical measure is the relative frequency, the ratio of subjects in which a species is found to the total number of subjects in each group. For instance, if a species is isolated from all subjects, its relative frequency is 1.0; if it is found in only 10% of the subjects, its relative frequency is 0.1. Using these concepts, we calculated each species’ relative frequency differences (RFDs) by subtracting the HC group frequency from the UC or CD group frequency to identify species enriched or depleted in each IBD subtype. Additionally, we evaluated the differences in the number of isolates within specific bacterial taxa across groups.

Genomic DNA was extracted from fecal samples using a ZymoBIOMICS DNA Miniprep Kit (Zymo Research, CA, United States) according to the manufacturer’s protocol. The sequence libraries were prepared according to Illumina’s instructions (Amplicon et al., 2013) and then sequenced by Macrogen (Seoul, South Korea) on an Illumina MiSeq system. We obtained 3,139,607 demultiplexed sequences with an average of 56,064 reads ranging from 35,562 to 76,011 reads per sample in the QIIME2 environment. The low-quality regions of the reads were denoised using the DADA2 pipeline (Callahan et al., 2016), and a rooted phylogenetic tree was generated. Taxonomic assignment was conducted using weighted naive classifiers trained on Silva 138.1 data specific to human stool samples (Kaehler et al., 2019). The QIIME2 files were further processed using the phyloseq package in R v4.2.2 (McMurdie and Holmes, 2013). The assigned features ranging from 25,024 to 52,733 were rarefied to 25,000-read sampling depth. The linear discriminant analysis (LDA) effect size (LEfSe) was used to select differential taxa in each IBD group compared with the HC group (Segata et al., 2011). The 16S rRNA gene amplicon sequences were deposited in the GenBank SRA database (PRJNA975689). Additionally, the SRA database for healthy controls (PRJNA975692 and PRJNA1094612) was imported and applied for comparison.

To quantify the bacterial cell number in stool samples, ZymoBIOMICS™ Spike-in Control I (Zymo Research, United States) containing equal cell numbers of two bacteria strains, Imtechella halotolerans and Allobacillus halotolerans, was added to each stool sample prior to DNA extraction. The protocol was followed according to the manufacturer’s instructions. The subsequent sequencing and analysis process was carried out in the same manner.

Statistical analyses were conducted in GraphPad Prism (v.10.2.2) and R software (v.4.3.2). Two-tailed Mann–Whitney test, two-sided Fisher’s exact test, and one-way ANOVA with Kruskal-Wallis test were used to analyze the clinical data and stool biomarkers. The two-sided Fisher’s exact test was applied to assess statistical significance in the U/T ratio and RFDs. Multiple Mann–Whitney test was used to analyze the number of cultured species in each phylum. One-way ANOVA with the Kruskal-Wallis test was used to analyze differences between groups in alpha diversity (Chao1 and InvSimpson indices), relative abundances at the phylum level, and bacterial cell counts in stool samples. The significance of the group distribution in beta diversity was determined by permutational multivariate analysis of variance (PERMANOVA) and permutational multivariate analysis of dispersion (PERMDISP). The correlation coefficients and statistical significance between the relative proportions (%) and inflammation biomarkers were calculated using Pearson correlation.

Our analysis included data from 41 IBD patients recruited for this study, along with data from 11 healthy individuals from our previous study (Park et al., 2024). Most of the 41 IBD patients analyzed were diagnosed with mild severity at the time of stool sample collection (58% with UC and 64% with CD) (Table 1). Most of the subjects in the UC group (95%, 18/19) had never been prescribed biologics, while a significant portion of the CD group (68%, 15/22) had a history of biologics treatment. While the ESR levels were similar between the two IBD groups, CRP levels were significantly higher in CD patients (4.6-fold, p = 0.03). Based on the morphological characteristics of the stool samples, we were able to classify samples as showing mushy or watery stool consistency or visible blood in the IBD groups. Notably, significantly higher levels of occult blood were quantified in the UC group compared to the HC group (p = 0.01). Moreover, the number of bacterial cells and FCP levels in the stool showed significant differences between the CD and HC groups, while the UC group exhibited intermediate values. Detailed clinical information is provided in Supplementary Data S1.

We compared the microbiota composition between the HC and IBD groups via 16S rRNA gene amplicon sequencing. In the alpha diversity analysis estimated Chao1 and InvSimpson indices, IBD patients exhibited significantly low microbial diversities compared to the HC group (Figure 1A). This tendency was also observed in the beta diversity based on the phylogenetic distance among communities, where the three groups showed significant group distribution (PERMANOVA p = 0.001 and PERMDISP p = 0.045) (Figure 1B). Notably, the CD group showed a distinct cluster from the HC group. In contrast, the UC group was distributed widely, overlapping with the HC and CD groups. At the phylum level, we observed significantly higher relative abundances of Pseudomonadota in both IBD subtypes and decreased Actinomycetota populations in CD patients (Figure 1C). Additionally, Fusobacteriota was detected in 40.9% (9/22) of CD patients. LEfSe analysis was conducted to compare the relative contributions of different taxa using an adjusted p-value cutoff of 0.05 for the Kruskal–Wallis test and an LDA score cutoff of 3.0. We identified 19 different taxa at the species level between the UC and HC groups, of which seven were differentially enriched in the UC group (Figure 1D; Supplementary Table S1). In contrast, a total of 55 differential taxa were identified in pairwise comparisons between the CD and HC groups (Figure 1E; Supplementary Table 2). Among them, 10 species were enriched in the CD group. The entire taxa list, detailed LDA score, p-values, and adjusted p-values are described in Supplementary Tables S1, S2.

Among the collected stool samples from IBD patients, 53.8% (28/52) of samples were included in the culturomics analysis. Ten samples from each of the UC and CD groups were used for cultivation (Supplementary Table S3). Additionally, culturomics data of eight healthy individuals from our previous study were included (Park et al., 2024). The culturomics cohort was comparable to the entire cohort in terms of clinical characteristics, inflammatory biomarkers, and 16S rRNA gene amplicon sequencing results (Supplementary Table S3; Figure 2). Our cultured strain libraries included a total of 14,095 isolates, consisting of 6,134 isolates from the UC patients and 7,961 isolates from the CD patients. The gut bacterial library of UC patients was classified into 211 species, corresponding to four phyla, 11 classes, 15 orders, 40 families, and 91 genera (Supplementary Data S2). The gut bacterial library of CD patients was classified into 164 species, corresponding to five phyla, 12 classes, 18 orders, 35 families, and 72 genera (Supplementary Data S3). From IBD patients, three species, including Selenobaculum gibii gen. Nov. sp. nov. (Yeo et al., 2023), were classified as potential novel species (Supplementary Table S4).

The combined dataset of 258 species from the IBD groups and 263 species from the HC group yielded a total of 367 species (Figure 2A). The number of subjects from whom each species was isolated, i.e., the isolation frequency, showed diverse patterns at the species level (Outer circle in Figure 2A). The species isolated from over 70% of subjects in each group were Enterococcus faecium, Clostridium innocuum, and Escherichia coli. The human gut bacterial library was comprised of Bacillota (65.7%), Bacteroidota (13.1%), Pseudomonadota (12.0%), Actinomycetota (8.4%), and Fusobacteriota (0.8%) (Figure 2B). The overall phylum composition in the UC group was similar to that of the HC group, except for the absence of Fusobacteriota. In contrast, the CD group showed the relative composition of phylum Bacillota (62.2%), Bacteroidota (11.6%), Pseudomonadota (12.8%), Actinomycetota (12.2%), and Fusobacteriota (1.2%). The number of exclusively isolated species was highest in the HC group (109 species), followed by the UC (55 species) and CD groups (31 species) (Figure 2C). The species diversity of Actinomycetota was significantly lower in both IBD groups compared to the HC group (Figure 2D). Similarly, the species diversities of Bacillota and Bacteroidota were significantly lower in the CD group. Moreover, our U/T ratio comparison revealed that the isolation of unique bacterial species was lower in the IBD groups compared to the HC group, with the CD group, in particular, showing significantly lower unique anaerobes (−18.9%, p = 0.008) and Gram-negative populations (−18.9%, p = 0.04) (Table 2).

Next, we identified bacterial species showing differences between the HC and each IBD group by calculating RFDs based on the isolation frequency of cultured species (Table 3). There were notable differences in the microbial signature patterns between the UC and CD groups. In the pairwise analysis between the UC and HC groups, 12 species were identified as statistically significant, with Bifidobacterium dentium, Enterococcus gallinarum, Flavonifactor plautii, and Peptoniphillus gorbachii being enriched in the UC group. In contrast, the pairwise analysis between the CD and HC groups revealed differences in 16 species, with Mediterraneibacter gnavus, Thomasclavelia ramosa, and Morganella morganii being enriched in the CD group. Consistent with our findings, the previous LEfSe analysis also identified B. dentium, M. gnavus, and T. ramosa as taxa, showing differential abundances in each IBD group, further supporting their potential role in IBD pathogenesis (Figures 1D,E; Supplementary Tables S1, S2). The culturomics analysis revealed 13 species with significantly lower RFDs in the CD group compared to the HC group. Among these, B. bifidum, Collinsella aerofaciens, Phocaeicola coprocola, and Parabacteroides merdae were also identified as depleted taxa in the CD group compared to the HC group through LEfSe analysis (Figure 1E; Supplementary Table S2). Intriguingly, C. aerofaciens was consistently depleted in both UC and CD patients, as revealed by both culturomics and LEfSe analysis (Figures 1D,E).

In this study, culturomics complemented 16S rRNA gene amplicon sequencing by providing insights into culturable species diversity, while sequencing offered information on relative abundances. Results showed a significant reduction in the relative abundances and culturable species diversity of Actinomycetota in the CD group compared to the HC group (Figures 1C, 2D). While Pseudomonadota exhibited an overall increase in relative abundances in the CD group, their cultured species diversity did not significantly differ from that of the HC group, suggesting an expansion of specific Pseudomonadota species (Figures 1C, 2D). Furthermore, the LEfSe analysis revealed that Bifidobacterium and Enterobacterales were the taxa that best differentiated between the HC and CD groups, belonging to the phyla Actinomycetota and Pseudomonadota, respectively (Figure 1E; Supplementary Table S2).

Given the limitations of species-level classification in 16S rRNA gene amplicon analysis and the high sequence similarity within the order Enterobacterales (Gupta et al., 2019; Gupta et al., 2019), it is challenging to obtain additional information on the species-level diversity of Bifidobacterium and Enterobacterales. When estimating bacterial cell numbers in the spiked initial cultured stool samples, the CD group exhibited significantly lower Bifidobacterium levels compared to the HC group (average 1.4 × 10⁹ cells/g stool vs. 1.0 × 10¹⁰ cells/g stool) (Figure 3A). This finding was corroborated by the culture-based analysis, where 1,491 isolates (nine species) were obtained from healthy individuals (n = 8), whereas only 317 isolates (eight species) were obtained from CD patients (n = 10) (Figure 3B). When examined as relative proportions within each group, distinct patterns were observed for each species. Specifically, B. dentium and B. breve exhibited a greater increase in proportion in the CD group compared to the HC group (Figure 3C). Additionally, the remaining Bifidobacterium species represented lower proportions in the CD group compared to the HC group, particularly for B. bifidum (p = 0.0002) and B. longum (p = 0.007).

The spiked initial fecal samples from CD patients exhibited a higher number of Enterobacterales cells compared to healthy individuals (average 2.1 × 10⁹ cells/g stool vs. 3.1 × 10⁸ cells/g stool) (Figure 3D). The culture-based analysis yielded a similar trend, with CD patients showing approximately 2.7 times more isolates (2,593 isolates; 17 species) than the HC group (976 isolates; 23 species) (Figure 3E). For E. coli, which was isolated from all subjects (Figure 2A), the isolation count was similar between the groups (Figure 3E). However, its relative proportion was significantly higher in the HC group (78% of Enterobacterales) (Figure 3F). In contrast, Proteus vulgaris (p = 0.03), Proteus mirabilis (p < 0.0001), and M. morganii (p < 0.0001) showed significant increases in both isolation counts and relative proportions in the CD group, suggesting a potential role in the pathogenesis of CD (Figure 3F).

We further investigated bacterial species correlated with both blood (ESR, CRP) and fecal (FCP, FOB, Stool DAI) inflammatory biomarkers, applying a prevalence cutoff for species present in more than 30% of total samples (Figure 4). In our culturomics analysis, significant correlations were identified with 13 bacterial species (Figure 4A). Among these, B. pseudocatenulatum and B. catenulatum had significant negative correlations with stool DAI scores (p = 0.028 and 0.024, respectively). From the 16S rRNA gene amplicon sequencing results (Figure 4B), unidentified Bifidobacterium populations showed significant negative correlations with the blood inflammation biomarkers (p = 0.002 for ESR and 0.012 for CRP). Additionally, P. merdae and C. aerofaciens, which showed differences in relative frequency across groups in our culturomics results (Table 3), exhibited negative correlations with ESR (p = 0.003) and stool DAI scores (p = 0.03), respectively, when analyzed using the relative abundances (Figure 4B). Among taxa showing positive correlations with the biomarkers, C. innocuum, T. ramosa, and M. gnavus were commonly selected from both approaches (Figure 4). When considering their enrichments in CD patients (Figure 1E; Table 3; Supplementary Table S2), the observed correlations suggest a strong association with increased fecal inflammation as indicated by higher stool DAI scores and FCP values in CD patients (Table 1).