Gut Microbial Signatures Associated With Clinical Remission in Inflammatory Bowel Disease Treated With Biologics: A Comprehensive Multi‐Cohort Analysis

ABSTRACT

Background and Aims

The relationship between gut microbiota and biological treatment response in inflammatory bowel disease (IBD) remains incompletely understood. We sought to characterize microbial signatures associated with clinical remission and develop a prediction model for clinical remission.

Methods

We analyzed 16 S rRNA gene sequencing data from two independent public cohorts (n = 231) treated with biologics (infliximab: n = 23; adalimumab: n = 22; ustekinumab: n = 186). Microbial diversity and taxonomic compositions were compared between the remission and non‐remission groups. Random Forest algorithm was employed to construct a prediction model using differential genera and clinical features, with performance evaluated through cross‐validation. The model was further validated in a local cohort (n = 29).

Results

Significant differences in alpha and beta diversity were observed between the remission and non‐remission groups (p < 0.05). MaAsLin2 analysis identified 25 differentially abundant genera (p < 0.05). Among these, we selected the top 10 genera with highest importance scores (Parabacteroides_B_862066, Agathobaculum, Ruminococcus_E, Sutterella, Clostridium_R_135822, Hominilimicola, Onthenecus, Butyricimonas, Bariatricus, Hominenteromicrobium) to build the Random Forest model, notably all enriched in remission patients. The model demonstrated robust predictive performance for clinical remission (AUC: 0.895), which was further validated in the local cohort (AUC: 0.750).

Conclusion

There is a relationship between gut microbial signatures and biological treatment outcomes in IBD patients. A predictive model based on gut microbiota composition may help stratify patients for treatment response. Further investigation of microbiome modulation strategies may enhance therapeutic efficacy.

1.

Summary.

• Summarize the established knowledge on this subject:

  • ◦Inflammatory bowel disease (IBD) is characterized by chronic, recurrent inflammation of the gastrointestinal tract, with the highest prevalence rates reported in western countries and an increasing incidence in developing countries.
  • ◦Currently, biologics are a crucial treatment approach for IBD; however, 24%–52% of patients do not respond to this therapy.
  • ◦Current evidence suggests that IBD development involves complex interactions among genetic, environmental, immune, and microbial factors, with existing studies showing limitations in consistently explaining the relationship between microbiota and therapeutic response.

• What are the significant and/or new findings of this study:

  • ◦This multi‐cohort analysis revealed significant differences in microbial diversity between IBD patients achieving and not achieving clinical remission after biological therapy.
  • ◦The study identified 25 differentially abundant genera, with 10 key genera used to develop a predictive Random Forest model for clinical remission.
  • ◦The predictive model demonstrated robust performance, with an AUC of 0.895 in cross‐validation and 0.750 in local cohort validation.
  • ◦The research highlights the potential of baseline gut microbial signatures as non‐invasive biomarkers for predicting biological treatment outcomes in IBD patients.

2. Introduction

Inflammatory bowel disease (IBD), which includes ulcerative colitis (UC) and Crohn's disease (CD), is characterized by chronic, recurrent inflammation of the gastrointestinal tract [1, 2]. The highest prevalence rates for IBD have been reported in western countries; however, the incidence is steadily increasing in developing countries, creating a growing economic and healthcare burden globally [3]. Currently, biologics are a crucial treatment approach for IBD; however, 24%–52% of patients do not respond to this therapy [4, 5, 6]. Identifying factors that influence IBD treatment outcomes and developing tools to predict individual responses to biologics could significantly enhance the effectiveness of personalized treatment.

Biological agents have emerged as cornerstone therapies for various inflammatory conditions, including IBD [7, 8], rheumatoid arthritis [9, 10], psoriasis [11, 12], and systemic lupus erythematosus [13, 14]. Current evidence suggests that IBD development involves complex interactions among genetic, environmental, immune, and microbial factors, with gut microbiota potentially playing a crucial role in the efficacy of biological therapies. For instance, Schierova et al. reported significant alterations in the fecal microbiome and specific antibacterial responses during anti‐TNF therapy in IBD patients [15]. Furthermore, Doherty et al. demonstrated that baseline fecal microbial signatures could serve as predictive biomarkers for identifying IBD patients likely to achieve remission with ustekinumab therapy [16]. However, due to limitations such as small sample sizes and inadequate patient stratification, existing studies have not reached consistent conclusions regarding the relationship between microbiota and therapeutic response, warranting further investigation in this field.

To address these limitations, we integrated data from multiple independent cohorts to comprehensively investigate the association between baseline gut microbiota and biological treatment response in IBD patients. Our analysis revealed that patients achieving remission exhibited higher microbial diversity at baseline, albeit with reduced inter‐genera network connectivity. Through differential analysis, we identified several characteristic genera with predictive value. These findings highlight the potential of baseline gut microbial signatures as non‐invasive biomarkers for predicting biological treatment response in IBD patients, a finding validated across multiple cohorts.

3. Materials and Methods

3.1. Screening of Published Studies and Construction of Public Cohort

In this study, we conducted a multi‐cohort analysis to investigate the association between baseline gut microbiota and treatment outcomes in IBD patients receiving biological therapy. To identify microbiological studies analyzing gut microbiota alterations in IBD patients undergoing biological therapy, we performed a comprehensive literature search across the National Center for Biotechnology Information (NCBI), Web of Science (WOS), and Embase databases using keywords including “inflammatory bowel disease,” “Crohn's disease,” “ulcerative colitis,” “biologics,” “metagenomics,” and “gut microbiome.” Raw 16 S rRNA sequencing data and clinical information from two populations (18 UC patients and 213 CD patients) from the United States and Czech Republic were obtained using NCBI BioProject IDs (PRJNA757573 [15] and PRJNA418765 [16]). The raw sequencing data were downloaded from the Sequence Read Archive (SRA) using the NCBI SRA toolkit. Demographic and clinical treatment information was extracted from the published articles.

The inclusion criteria for studies were as follows: (1) utilization of fecal samples, (2) employment of 16 S rRNA amplicon sequencing, and (3) public availability of sequencing data and metadata for IBD patients. Exclusion criteria encompassed: (1) patients who used probiotics, antibiotics, or bowel preparation agents within 1 month prior to sample collection, and (2) patients diagnosed with Clostridioides difficile infection or colorectal cancer. The search period spanned from January 2012 to December 2023. For the assessment of biological therapy response, clinical remission was evaluated after 22 weeks of treatment, which aligns with standard clinical trial endpoints. Clinical remission was defined using the following criteria: partial Mayo score (pMayo) ≤ 1; Crohn's disease activity index (CDAI) < 150 or Harvey‐Bradshaw index (HBI) < 5.

3.2. Patient Recruitment and Sample Collection for Local Cohort

Twenty‐nine IBD patients were recruited from The Sixth Affiliated Hospital of Sun Yat‐sen University between October 2022 and March 2024. Fecal samples were collected at baseline prior to biological therapy initiation with informed consent from all participants. The study protocol was approved by the Ethics Committee of The Sixth Affiliated Hospital of Sun Yat‐sen University (approval number: 2022ZSLYEC‐136) and conducted in accordance with the Declaration of Helsinki.

Inclusion criteria were as follows: (1) age between 18 and 75 years; (2) IBD diagnosis confirmed by colonoscopy; and (3) voluntary participation with signed informed consent.

Exclusion criteria included: (1) use of probiotics, antibiotics, or bowel preparation agents within 1 month prior to sample collection; (2) concurrent active systemic immune or infectious diseases, including severe allergies, rheumatoid arthritis, systemic lupus erythematosus, viral hepatitis, or acquired immune deficiency syndrome; (3) presence of other untreated primary malignancies; (4) severe organ dysfunction or failure; and (5) any other conditions deemed unsuitable for study participation by the investigators.

For eligible patients, fecal samples were collected in sterile containers and immediately transported in a 4°C cooling box. All samples were transferred to −80°C storage within 12 h of collection and maintained at this temperature until DNA extraction.

3.3. 16 S rRNA Gene Sequencing of Local Fecal Samples

Genomic DNA was extracted using a TIANamp Soil DNA kit (TIANGEN Biotech, China). The V3‐V4 region of the bacterial 16 S rRNA gene was amplified using primers 341F (5′‐CCTAYGGGRBGCASCAG‐3′) and 806R (5′‐GGACTACNNGGGTATCTAAT‐3′). Libraries were constructed using the TruSeq DNA PCR‐Free Sample Preparation Kit following manufacturer's protocol. After quality assessment, the qualified libraries were sequenced on an Illumina NovaSeq 6000 platform to generate 250 bp paired‐end reads. Detailed information about DNA extraction, library preparation, and sequencing procedures is provided in the Supporting Information S1.

3.4. Analysis of 16 S rRNA Gene Amplicon Sequencing Data

Raw sequencing data processing and quality control were performed using QIIME2 [17](Version: 2024.5). According to previous studies [18, 19], paired‐end reads were first processed using Cutadapt [20] (qiime2/q2‐cutadapt plugin) to remove adapter sequences and low‐quality reads (quality score < 20), and then merged using VSEARCH [21] (qiime2/q2‐vsearch plugin). Reads shorter than 150 bp and samples with fewer than 10,000 reads were filtered using QIIME2 plugins. The filtered sequences were processed for chimera detection using VSEARCH against the Greengenes2 database (Version: 2024.09), and potential contaminants including mitochondrial and chloroplast sequences were removed. To address PCR biases, ASVs present in less than 10% of all samples were filtered out. Taxonomic classification was performed using the Greengenes2 taxonomy database [22], which utilizes full‐length sequences to minimize potential biases from different 16 S rRNA gene regions. Detailed information about the bioinformatic analysis pipeline is provided in the Supporting Information S1.

3.5. Prediction Models Training by Machine Learning

The machine learning framework integrated patients' baseline clinical and fecal microbiota data (28 features in total) collected prior to biological therapy, utilizing the “mlr3” R package (version 0.22.0) [23] for model construction. Data were split into training (80%) and internal validation (20%) sets using stratified 10‐fold cross‐validation. A series of Random Forest models were constructed by progressively incorporating features (ranging from 1 to 28 features). Model fine‐tuning employed random search in the hyperparameter space with nested 10‐fold cross‐validation, iterating over 100 evaluations. The Area Under the Receiver Operating Characteristic curve (AUC) served as the primary metric for model performance assessment. The best‐performing model was subsequently applied to our local cohort and validated using an independent external cohort to evaluate its predictive capability. To assess the model's performance in different treatment subgroups, we further evaluated the top 10 feature model separately in anti‐TNF and anti‐IL‐12/23 subgroups.

3.6. Statistical Analysis

“MMUPHin” R package [24] was used to mitigate batch effects between datasets. Alpha‐diversity indices (Shannon and InvSimpson indices) and beta‐diversity were calculated using the “MicrobiotaProcess” R package [25]. Beta‐diversity analysis was performed via Principal Coordinate Analysis (PCoA) based on Bray‐Curtis distances using Hellinger‐transformed abundance data. PERMANOVA (adonis2) was employed to assess the effects of multiple variables (including BioProject, Disease, Treatment, Group, Age, Gender, BMI) and their two‐way and three‐way interactions with 9999 permutations. Alpha diversity indices between groups were compared using two‐sided Wilcoxon rank‐sum tests, with statistical significance defined as p < 0.05. Differential abundance analysis of bacterial genera between groups was performed using two complementary approaches. Linear discriminant analysis Effect Size (LEfSe) was conducted using the “microeco” R package [26], with significant differences defined as p < 0.05 and Linear Discriminant Analysis (LDA) > 2. Additionally, Compound Poisson Linear Models (CPLM) from the “Maaslin2” R package [27] were employed to assess the multivariate associations between clinical remission and potential confounding factors, with false discovery rate (FDR) controlled at Q‐value < 0.25 using the Benjamini‐Hochberg procedure. Compound Poisson Linear Models (CPLM) from the “Maaslin2” R package [27] were employed to assess the multivariate associations between clinical remission and potential confounding factors. Microbial co‐occurrence networks were constructed based on Spearman rank correlations calculated using the “WGCNA” R package [28], with significant correlations defined as |r| > 0.7 and p < 0.05. Genera with relative abundance < 0.1% were excluded from the analysis. The networks were visualized using the “igraph” R package [29] and Gephi software (version 0.10.1, http://gephi.org).

4. Results

4.1. Construction of the Study Cohort

Initial screening of PubMed, WOS, and Embase databases yielded over 1700 relevant studies. After systematic evaluation, 10 studies met the inclusion criteria, containing clinical data from IBD patients receiving biological therapy. However, only two studies provided sufficient metadata to determine clinical remission status following biological therapy. The characteristics of both publicly available studies and our local cohort are summarized in Table 1.

TABLE 1.

Characteristics of included studies.

Study	IBD Subtype	Country	Patients (n)	Biological	Assay type	BioProject ID
Doherty, 2018 [16]	CD, UC	United States	186	Ustekinumab	Amplicon	PRJNA418765
Schierova, 2021 [15]	UC	Czech republic	45	Adalimumab, infliximab, ustekinumab	Amplicon	PRJNA757573
Present study	CD	China	29	Ustekinumab	Amplicon	PRJCA036999

Among the two public cohorts (231 patients) and our local cohort (29 patients), patient characteristics and treatment distributions are shown in Figure 1A. The local sequencing data has been deposited in the China National Center for Bioinformation (CNCB) under BioProject ID PRJCA036999, which will be released according to the database's data sharing policy. In the public cohorts, after excluding patients with undefined clinical remission status, 133 patients were included in the final analysis, with 53 achieving clinical remission and 80 without clinical remission as shown in Table 2. These patients received adalimumab (n = 20), infliximab (n = 18), and ustekinumab (n = 95). In our local cohort, all patients were treated with ustekinumab (n = 29), with their baseline characteristics shown in Table 3. The detailed comparison between public and local cohorts is shown in Table S1, and the characteristics between anti‐TNF and anti‐IL‐12/23 subgroups in the public cohort are presented in Table S2. The research workflow of this study is illustrated in Figure 1B. At the phylum level, taxonomic analysis revealed Bacillota and Bacteroidota as the predominant phyla across all cohorts (Figure 1C).

FIGURE 1.

Overview of study design and cohort characteristics in IBD patients receiving biological therapy. (a) Distribution of clinical characteristics in the public cohort, including age, BMI, remission status, disease type, gender, study source, and treatment type. (b) Distribution of the same characteristics in the local cohort. Pie charts show the percentage breakdown of each variable. (c) Schematic overview of the study workflow, including patient groups, sample collection, sequencing method, and data analysis steps. (d) Relative abundance of dominant bacterial taxa in non‐Remission and Remission groups. Stacked bar plots show the compositional differences in intestinal microbiota between the two response groups, with different colors representing distinct bacterial taxa.

TABLE 2.

Baseline characteristics of public IBD patients with and without remission to biologic therapy.

	Remission (n = 53)	Non‐remission (n = 80)	p value
Age, years (mean, SD)	37.06 (12.51)	37.52 (11.84)	0.827
Male, n (%)	20 (37.7)	21 (26.2)	0.225
BMI, kg/m2 (mean, SD)	24.44 (4.89)	25.60 (6.75)	0.285
Ulcerative colitis, n (%)	7 (13.2)	7 (8.8)	0.595
Disease duration, years (mean, SD)	8.87 (7.57)	12.46 (9.49)	0.022
Localization, n (%)			0.933
E1	2 (3.8)	1 (1.3)
E2	1 (1.9)	2 (2.6)
E3	4 (7.5)	4 (5.1)
L1	13 (24.5)	19 (24.4)
L2	13 (24.5)	20 (25.6)
L3	20 (37.7)	32 (41.0)
HBI (mean, SD)	2.95 (2.75)	5.33 (2.08)	0.166
pMayo (mean, SD)	5.29 (2.14)	6.86 (2.04)	0.184
CDAI (mean, SD)	301.24 (60.79)	333.57 (68.00)	0.039
Disease activity, n (%)			< 0.001
Clinical remission	14 (26.4)	1 (1.2)
Mild activity	12 (22.6)	4 (5.0)
Moderate activity	26 (49.1)	72 (90.0)
Severe activity	1 (1.9)	3 (3.8)

Note: CD lesion location: L1 (ileal), L2 (colonic), L3 (ileocolonic), L4 (upper GI); UC extent: E1 (proctitis), E2 (left‐sided), E3 (extensive/total colitis). Disease activity was defined as: clinical remission (CDAI < 150, HBI < 5, pMayo < 2), mild activity (CDAI 150–220, HBI 5‐7, pMayo 2–4), moderate activity (CDAI 221–450, HBI 8–16, pMayo 5–7), and severe activity (CDAI > 450, HBI > 16, pMayo > 7).

TABLE 3.

Baseline characteristics of local IBD patients with and without remission to biologic therapy.

	Remission (n = 16)	Non‐remission (n = 13)	p value
Age, years (mean, SD)	29.69 (8.17)	35.31 (11.48)	0.136
Male, n (%)	12 (75.0)	7 (53.8)	0.424
BMI, kg/m2 (mean, SD)	20.07 (3.41)	18.95 (2.67)	0.344
Crohn's disease, n (%)	16 (100.0)	13 (100.0)	1.000
Disease duration, years (mean, SD)	5.12 (5.14)	5.69 (8.36)	0.824
Localization, n (%)			0.172
L1	2 (12.5)	3 (23.1)
L2	4 (25.0)	0 (0.0)
L3	10 (62.5)	9 (69.2)
L4	0 (0.0)	1 (7.7)
HBI (mean, SD)	5.88 (1.82)	8.85 (3.74)	0.009
CDAI (mean, SD)	212.54 (51.56)	291.38 (113.10)	0.019
Montreal Classification, n (%)			0.492
B1	8 (50.0)	5 (38.5)
B2	3 (18.8)	1 (7.7)
B2, B3	4 (25.0)	4 (30.8)
B3	1 (6.2)	3 (23.1)
Disease activity, n (%)			0.122
Clinical remission	4 (25.0)	1 (7.7)
Mild activity	8 (50.0)	4 (30.8)
Moderate activity	4 (25.0)	8 (61.5)

Note: CD lesion location: L1 (ileal), L2 (colonic), L3 (ileocolonic), L4 (upper GI); CD behavior: B1 (inflammatory), B2 (stricturing), B3 (penetrating); UC extent: E1 (proctitis), E2 (left‐sided), E3 (extensive/total colitis). Disease activity was defined as: clinical remission (HBI < 5), mild activity (HBI 5–7), moderate activity (HBI 8–16), and severe activity (HBI > 16).

4.2. Correlation Between Microbial Diversity and Biologics‐Induced Remission

To understand the alterations in gut bacterial composition between IBD patients who achieved and did not achieve remission following biological therapy, we first evaluated sequencing depth using rarefaction curves, which indicated adequate coverage of microbial diversity across all samples (Figure S1). We then analyzed differences in microbial diversity across these groups. Alpha diversity analysis using Shannon and InvSimpson indices revealed significantly higher diversity in the remission group compared to the non‐remission group in the combined dataset and Doherty's cohort (PRJNA418765) (both p < 0.05) (Figure 2A–B). A similar trend of higher diversity in the remission group was also observed in Schierova's cohort (PRJNA757573), although this difference did not reach statistical significance (Shannon: p = 0.19; InvSimpson: p = 0.23).

FIGURE 2.

Alpha and beta diversity analysis reveals distinct microbiome patterns between remission and non‐remission groups in IBD patients. (a) Shannon index and (b) InvSimpson index comparisons between Remission and non‐Remission groups across combined cohort, Doherty et al., and Schierova et al. datasets. Box plots illustrate the distribution of α‐diversity index values for each group. (c) Bar plot showing R ² values from beta diversity analysis based on various factors. Asterisks indicate statistical significance (***p < 0.001, **p < 0.01). (d) PCoA plot comparing microbiome composition between Remission and non‐Remission groups. The plot shows the first two principal coordinates with the percentage of variation explained by each axis. Ellipses represent 95% confidence intervals for each group. R ² and P for group separation are indicated. Box plots in (a), (b), and (d) show median (central line), interquartile range (box), and 1.5 times the interquartile range (whiskers).

Subgroup analysis revealed that CD patients who achieved remission showed significantly higher microbial diversity compared with those who did not achieve remission (p < 0.05). Similarly, among patients treated with Ustekinumab, those who achieved remission demonstrated higher microbial diversity compared with non‐remission patients (p < 0.05) (Figure S2, Figure S3).

Subsequently, we performed the beta diversity analysis to examine the relationship between various factors and changes in the gut microbiota of individuals with IBD. We identified the study source as the strongest contributor to gut microbiota variation (R ²  = 0.077, p < 0.001), followed by clinical remission outcome (R ²  = 0.018, p < 0.01). Other factors including BMI (R ²  = 0.011), gender (R ²  = 0.010), age (R ²  = 0.009), disease (R ²  = 0.009), and treatment (R ²  = 0.007) showed relatively minor contributions (p > 0.05, Figure 2C). Principal Coordinate Analysis (PCoA) demonstrated significant clustering differences between the remission and non‐remission groups (R ²  = 0.0182, p = 0.0037, Figure 2D).

Collectively, our results indicate substantial disparities in both alpha and beta diversity between clinical remission and non‐remission groups. Nonetheless, given that variability in microbial features is influenced by multiple factors, subsequent analyses should include a comprehensive accounting of these variables when comparing the clinical remission and non‐remission groups.

4.3. Alterations of Gut Bacterial Genera Are Associated With Biologics‐Induced Remission in IBD

We compared the taxonomic relative abundance between remission and non‐remission groups to investigate the association between biological therapy response and baseline gut microbiota composition. Initially, using an LDA > 2 criterion, we identified 78 differential genera, with 37 enriched in the remission group and 41 in the non‐remission group (Figure S4A). When performing differential analysis using MaAsLin2 on the complete genera dataset, we ultimately identified 25 significantly different genera after adjusting for confounding factors including study, disease type, treatment regimen, age, gender, and disease location (Figure S5). Among these, 17 genera were significantly enriched in the remission group, whereas 8 genera were significantly decreased. These distinct microbial signatures between remission and non‐remission groups suggest that baseline gut microbiota composition might serve as a potential predictive biomarker for biological therapy response.

4.4. Machine Learning Model for Predicting Biologics‐Response Using Baseline Stool Samples

Given these significant taxonomic differences and their potential clinical implications, we next sought to develop a machine learning model to predict therapeutic outcomes based on baseline gut microbiota profiles. Previous studies have shown that gut microbiota can predict biological therapy responses in various diseases, including IBD [30, 31]. However, there remains a lack of robust predictive models using baseline gut microbiota features to forecast therapeutic outcomes in IBD patients prior to treatment initiation. Therefore, we developed a machine learning prediction framework based on baseline gut microbial characteristics (Figure 3A) to effectively predict clinical remission before biological therapy initiation. To establish the optimal predictive model, we first retained genera that were consistently present across both public cohorts. We then evaluated the feature importance scores of these 25 differentially abundant genera along with patients' baseline clinical characteristics (age, gender, and BMI) using random forest modeling (Figure 3B). To determine the optimal number of features, we systematically assessed the model performance using different numbers of features (ranging from 1 to 28) through internal cross‐validation. As shown in Figure 3C, the model achieved stable performance with 10 features (AUC = 0.895), with minimal improvement observed when additional features were included. Using the top 10 ranked features based on model importance scores (Parabacteroides_B_862066, Agathobaculum, Ruminococcus_E, Sutterella, Clostridium_R_135822, Hominilimicola, Onthenecus, Butyricimonas, Bariatricus, Hominenteromicrobium), our random forest model demonstrated excellent predictive capability in the cross‐validation of two public cohorts (Training AUC = 0.895, 95% CI: 0.836–0.974). More importantly, when validated on an independent local cohort, the model maintained robust performance (Testing AUC = 0.750, 95% CI: 0.560–0.923), demonstrating its potential for clinical application (Figure 3D). Notably, the 10‐feature model also showed good predictive performance in both treatment subgroups, with AUC values of 0.775 and 0.905 for the anti‐TNF and anti‐IL‐12/23 cohorts, respectively (Figure S6).

FIGURE 3.

Machine learning‐based prediction model for IBD treatment outcome. (a) Schematic diagram showing the integration of clinical demographics, mycobiome data, and machine learning approach for model development. (b) Feature importance scores based on AUC (common features between two cohorts), displaying the relative contribution of different bacterial taxa and clinical factors to the prediction model. Features are ranked by their importance scores on the x‐axis. (c) Model performance curve showing the relationship between the number of features and model accuracy (AUC). The shaded area represents the confidence interval. (d) ROC curves for the top 10 features model, showing performance on both training (AUC = 0.895, 95% CI: 0.836–0.974) and testing (AUC = 0.750, 95% CI: 0.560–0.923) sets. The diagonal dashed line represents random chance (AUC = 0.5).

Notably, all these top 10 predictive genera were enriched in the remission group (Figure 4A–J) and predominantly belonged to class Clostridia within the phylum Bacillota, which has been previously associated with anti‐inflammatory effects and improved intestinal health in IBD patients [32, 33].

FIGURE 4.

Relative abundance comparison of top 10 discriminative bacterial taxa between remission and non‐remission groups. (A‐J) Box plots showing the relative abundance of key bacterial taxa: (a) Parabacteroides_B_862066, (b) Agathobaculum, (c) Ruminococcus_E, (d) Hominilimicola, (e) Onthenecus, (f) Butyricimonas, (g) Sutterella, (h) Clostridium_R_135822, (i) Bariatricus, and (j) Hominenteromicrobium in non‐Remission (green) and Remission (pink) groups. Statistical significance is indicated by asterisks: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Box plots display median values (central line), interquartile ranges (box), and individual data points showing the distribution of relative abundances.

Further in‐depth analysis of the interrelationships among these 10 differential genera revealed distinct characteristics of microbial community interactions. The results showed that multiple bacterial pairs in the non‐remission group exhibited strong correlations (correlation coefficient |r| > 0.7) (Figure 5A), whereas these tight associations were markedly weakened in the remission group (Figure 5B), suggesting that microbial community interaction patterns may be closely related to disease prognosis. In conclusion, our study not only identifies baseline gut microbiota composition, particularly the abundance of members within the phylum Bacillota, as a potential predictor of biological therapy response in IBD patients but also reveals that differences in microbial community interaction networks may play a crucial role in disease prognosis.

FIGURE 5.

Correlation analysis of the top 10 bacterial features in non‐Remission and Remission groups. (A) Correlation heatmap showing relationships between bacterial taxa in the non‐Remission group. (B) Correlation heatmap showing relationships between bacterial taxa in the Remission group. The color scale indicates correlation strength from −1.0 (dark blue) to 1.0 (dark red), with white representing no correlation. The size and darkness of squares represent correlation magnitude and statistical significance, respectively. Red asterisks indicate statistical significance.

4.5. Microbial Co‐Occurrence Networks in Biologics‐Induced Remission

To explore the interactions among bacterial taxa and their correlation with biological therapy efficacy, we further investigated the relationships between bacterial genera using co‐occurrence networks. We calculated the correlation strength between genera in both remission and non‐remission groups using baseline samples (Table S3). The network structure differed significantly between the remission group (162 nodes, 276 edges) and the non‐remission group (199 nodes, 719 edges). After eliminating nodes with fewer connections (degree < 5), the remission group network contained four major clusters (modularity index: 0.73 (Figure 6A)), while the non‐remission group showed 6 clusters (modularity index: 0.73) (Figure 6B). The interactions between genera intensified in the remission group as prognosis worsened. This finding is consistent with the correlation differences between the 10 genera that showed significant differences between the two groups, as mentioned previously. The genera in both network diagrams primarily belonged to the phylum Bacillota. Overall, these results suggest that baseline network complexity and bacterial interconnectivity might serve as indicators of clinical remission, with simpler, more loosely connected networks being associated with better response to biological therapy.

FIGURE 6.

Co‐occurrence networks of bacterial genera in remission and non‐remission IBD patients. (a) Network of bacterial interactions in the remission group, showing a relatively sparse network structure. (b) Network of bacterial interactions in the non‐remission group, displaying a more complex and interconnected community structure. In both networks, nodes represent bacterial genera with size proportional to relative abundance. Different colors indicate distinct bacterial phyla according to the legend. Edges represent significant correlations between genera (Spearman's correlation coefficient |r| > 0.7, p < 0.05). Node positions are determined by the Fruchterman‐Reingold algorithm, with more closely related taxa positioned nearer to each other. The networks reveal distinct microbial community organization patterns between remission and non‐remission states.

5. Discussion

In the present study, we analyzed the role of the gut microbiome in predicting biological therapy outcomes in IBD patients through integration of multiple 16 S rRNA sequencing datasets. By leveraging open‐access data from diverse cohorts, we enhanced statistical power while minimizing cohort‐specific biases [15]. Our analysis focused on the 16 S rRNA gene sequence, providing reliable genus‐level taxonomic classification [34, 35]. Through characterization of gut microbiome profiles and their association with treatment responses, we identified therapy‐specific bacterial patterns and developed a predictive model for biological therapy outcomes.

Building upon these findings, we conducted a detailed analysis of baseline fecal samples, observing significantly higher alpha diversity in patients who achieved remission, although this pattern varied in smaller subgroups. This observation aligns with previous studies [36, 37] and reflects the potential role of gut microbial diversity in IBD treatment outcomes, as lower microbial diversity has been linked to IBD onset [38, 39, 40], while higher pre‐treatment diversity may enhance treatment efficacy [41, 42, 43]. Beta‐diversity analysis revealed significant differences consistent across studies, with larger R ² between studies than between remission and non‐remission groups, indicating robust differences despite sequencing variations.

To address the variability in sample collection, sequencing techniques, and gut microbiota composition across geographical regions [16, 44], we adjusted for cohort heterogeneity and potential confounding factors, including age, gender, disease location, and treatment remission. This adjustment was applied when analyzing genus‐level microbial differences between patient groups. Despite inherent inter‐group variations, our findings revealed a key pattern: patients with divergent prognoses exhibited distinct microbial compositions at baseline. These results support the potential of pre‐treatment gut microbiome profiles as predictive indicators of biological therapy outcomes.

Analysis of individual bacterial taxa revealed 25 marker genera associated with biological therapy outcomes in IBD patients. Based on their importance scores in the random forest model, we focused on the top 10 genera for further analysis. Seven belonged to the phylum Bacillota (Agathobaculum, Ruminococcus_E, Clostridium_R_135822, Hominilimicola, Onthenecus, Bariatricus, and Hominenteromicrobium), while Butyricimonas and Parabacteroides_B_862066 were from Bacteroidota, and Sutterella from Pseudomonadota. Both Bacillota and Bacteroidota are predominant phyla in healthy human gut [45], with Bacillota playing crucial roles in immune regulation through SCFA production [46, 47].

Consistent with these findings, most of our identified genera demonstrate strong associations with SCFA production and immune regulation. Among the phylum Bacillota members, Agathobaculum enhances intestinal barrier function and reduces inflammation through butyrate production, with its decrease linked to multiple sclerosis, Parkinson's disease, and osteoporosis [48, 49, 50]. Ruminococcus_E promotes mucosal repair and modulates immune responses through butyrate production, potentially serving as a biomarker for IBD treatment efficacy [39]. Clostridium_R_135822 may suppress inflammation by inducing Treg differentiation and producing SCFAs [33, 51, 52]. Onthenecus produces butyrate and maintains intestinal barrier integrity [53], while Bariatricus contributes to metabolic health regulation [54]. Both Hominilimicola, a keystone genus in dietary fiber metabolism, and Hominenteromicrobium are essential for gut homeostasis through their production of SCFAs [55, 56, 57]. From other phyla, Parabacteroides_B_862066 shows anti‐inflammatory effects through SCFA production [58]. Sutterella's enrichment in remission patients aligns with its beneficial associations in cancer immunotherapy responses [59, 60].

The genus Butyricimonas demonstrates remarkable duality in disease contexts, warranting special attention in our analysis. Multiple studies have documented its protective effects across various conditions, including cognitive impairment [61], diabetes [62], and irritable bowel syndrome [63]. Contrasting evidence, however, reveals potential pathogenic associations, such as Butyricimonas paravirosa‐related bacteremia in acute terminal ileitis and increased abundance in post‐operative esophageal cancer recurrence [64]. Our findings mirror this complexity: initial LEfSe analysis indicated significant enrichment of Butyricimonas in the non‐remission group, yet subsequent analysis using MaAsLin2 to control for confounding factors revealed its significant enrichment in the remission group. These apparently contradictory results align with the genus's documented dual nature and underscore the importance of rigorous statistical approaches in microbiome analysis. Despite the controlled analysis suggesting a potentially beneficial role of Butyricimonas in IBD treatment outcome, future research must elucidate the specific mechanisms and conditions under which this genus promotes or impairs therapeutic efficacy.

Microbial communities in the human gut form a complex ecosystem, where interactions between microorganisms are crucial to host health. Recent studies suggest that these interactions may influence IBD patients' responses to biological therapies [43]. Our research found that IBD patients who failed to achieve clinical remission after biological treatment exhibited significantly stronger interactions among gut microbial genera. This suggests that in patients with poorer outcomes, beneficial bacteria may be suppressed by other microbial groups, potentially disrupting the gut microecological balance and affecting key metabolite production such as SCFAs. These findings highlight that microbial community interaction patterns could be critical in determining IBD treatment efficacy.

Numerous studies have confirmed the potential of gut microbial markers in IBD diagnosis [65]. Through multi‐cohort analysis, we demonstrate for the first time that the microbiota can serve as reliable predictive markers for biological therapy outcomes in IBD patients. Our research uniquely explores the association between gut microbiota and clinical remission by developing a universal prediction model applicable across biological therapeutics. Using pre‐treatment baseline samples, we constructed a model that effectively forecasts clinical remission and showed excellent performance in external validation cohorts. While measuring patients' microbial abundance before treatment can guide clinical therapy [66, 67, 68], identifying characteristic microbes across populations remains challenging due to gut microbiota heterogeneity.

This study has several limitations. First, the limited number of included cohorts may affect the generalizability of our results. Second, the resolution of 16 S rRNA sequencing only allows for reliable taxonomic annotation at the genus level and is restricted to changes within the bacterial kingdom. Future studies employing metagenomic sequencing are needed to explore changes at more refined taxonomic levels and across other microbial kingdoms. Finally, this study lacks mechanistic experiments to elucidate how specific gut microbiota influences intestinal metabolism and biological therapy efficacy. Further research is required to unravel these complex interactions.

6. Conclusion

In conclusion, this study demonstrates that significant alterations in gut microbiota composition are associated with clinical remission in IBD patients receiving biological therapy. Furthermore, we found that baseline gut microbial signatures can effectively predict clinical remission in these patients. These findings highlight the potential role of the gut microbiome in modulating the efficacy of biological therapy for IBD and suggest that microbial features may serve as valuable biomarkers for treatment outcome prediction.

Author Contributions

Qingyang Zheng: conceptualization (lead), data curation (lead), investigation (lead). Yun Zhong: software (equal), formal analysis (lead). Haifeng Lian: writing – original draft (equal). Jieru Zhuang: writing – original draft (equal). Lichun Wang: methodology (lead). Jianyong Chen: investigation (supporting), project administration (equal). Huaiming Wang: software (equal), data curation (supporting), funding acquisition (equal). Hui Wang: supervision (lead), resources (lead), funding acquisition (equal). Xijie Ye: validation (lead), visualization (lead). Zicheng Huang: Writing – review and editing (equal). Keli Yang: project administration (equal), writing –review and editing (equal). All authors reviewed and approved the final version of the manuscript.

Ethics Statement

This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of The Sixth Affiliated Hospital, Sun Yat‐sen University (approval number: 2022ZSLYEC‐136).

Consent

Written informed consent for fecal sample usage was obtained from all participants in our cohort. Due to the public nature of NCBI databases, no informed consent was required for the use of publicly available data.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information S1

Supporting Information S2

Figure S1

Figure S2

Figure S3

Figure S4

Figure S5

Figure S6

Table S1

Table S2

Table S3

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.