Short-Term Longitudinal Analysis of Gut Microbiota Dynamics During Anti-CD19 CAR-T Cell Therapy in Diffuse Large B-Cell Lymphoma Patients

Introduction

Diffuse large B-cell lymphoma (DLBCL) is the most common subtype of non-Hodgkin lymphoma, characterized by biological and clinical heterogeneity.¹ While frontline immunochemotherapy regimens have significantly improved survival outcomes, a substantial proportion of patients ultimately relapse or develop refractory disease.^2,3 Chimeric antigen receptor T (CAR-T) cell therapy has emerged as a promising treatment option for relapsed or refractory DLBCL, offering durable responses in a subset of patients.⁴ However, the treatment is often accompanied by serious complications such as cytokine release syndrome, the mechanisms of which remain incompletely understood.^5,6

Recent evidence suggests that the gut microbiota plays a crucial role in modulating host immune responses and influencing the efficacy and toxicity of cancer immunotherapies, particularly immune checkpoint inhibitors.^7,8 The gut microbiome may serve as a key mediator linking systemic immunity with therapeutic outcomes. Diefenbach et al reported that pretreatment gut microbial composition and diversity differed significantly between indolent lymphomas and aggressive subtypes such as DLBCL.⁹ A study of untreated DLBCL patients revealed distinct gut microbial alterations compared to healthy controls, including reduced beta-diversity and imbalances in key bacterial phyla.¹⁰ These findings highlight the intrinsic connection between gut microbiota and lymphoma biology.

While previous studies have primarily focused on the predictive value of pretreatment gut microbiota for CAR-T therapy outcomes,^5,11 our study targeted patients who achieved confirmed clinical responses. CAR-T cell therapy is known to induce robust immune activation and systemic inflammation; however, it remains unclear whether these responses are accompanied by dynamic alterations in the gut microbiota. In this study, we conducted a longitudinal analysis of 12 DLBCL patients who responded to anti-CD19 CAR-T therapy (complete or partial response) and had fecal samples available at three clinically relevant time points: pre-infusion, and at weeks 1 and 2 post-infusion. This design enabled us to capture microbiota dynamics during the acute post-infusion phase and explore their potential associations with systemic inflammatory and immune parameters, providing new insights into microbiota-immune crosstalk in the setting of effective CAR-T treatment.

Materials and Methods

Study Design and Patients

In this study, all patients with DLBCL were recruited from the Affiliated Hospital of Xuzhou Medical University between September 2023 and May 2024. Patients with a history of gastrointestinal surgery or active gastrointestinal diseases, recent use of oral antibiotics, probiotics, or bowel-cleansing agents (within 4 weeks prior to sampling), coexisting active malignancies, or serious immunodeficiency or autoimmune disorders were excluded. Clinical data, including age, gender, lymphocyte count, serum albumin, neutrophil count, white blood cell count, C-reactive protein (CRP) level, and Eastern Cooperative Oncology Group performance status (ECOG PS), were collected.

All patients received the same anti-CD19 CAR-T product, a second-generation CAR-T therapy incorporating a 4–1BB costimulatory domain, manufactured by Aikangde Biotechnology Co., Ltd. (Suzhou, China). The infused dose ranged from 1.8 to 4.6×10⁶ CAR-T cells/kg [median dose: 2.1×10⁶ CAR-T cells/kg]. Regarding treatment-related toxicities, grade 1 cytokine release syndrome (CRS) occurred in 5 patients and grade 2 CRS occurred in 2 patients. No immune effector cell-associated neurotoxicity syndrome was observed.

Stool Sample Collection and Time Points

Stool samples were collected at three time points: prior to CAR-T cell infusion (baseline), one week after infusion, and two weeks after infusion. All samples were collected using sterile collection kits containing a preservation solution. Participants were instructed to collect stool directly onto clean, dry toilet paper, avoiding contact with other bodily fluids. Approximately 200 mg of stool was sampled from the middle portion using a sterile swab and immediately immersed in the preservation solution. The swab was discarded after ensuring adequate mixing of the sample with the solution by shaking for 30–60 s. Samples were sealed in individual, light-proof bags, labeled with unique identifiers, stored at 4°C, and transported to the laboratory within 24 h for downstream 16S rRNA gene sequencing analysis to systematically evaluate gut microbiome dynamics across different stages of CAR-T therapy.

DNA Extraction and Sequencing

Upon completion of sequencing, demultiplexing of the obtained reads was performed using an in-house bioinformatics script. All downstream analyses were conducted using the Quantitative Insights Into Microbial Ecology 2 (QIIME2, version 2021.11) platform.¹² The DADA2 pipeline¹³ was applied for paired-end read merging, quality filtering, chimera removal, and denoising, followed by clustering of high-quality sequences into amplicon sequence variants (ASVs). Taxonomic assignment of these features was performed against the SILVA database (release 132) using QIIME2’s classify-sklearn classifier, and a feature table was generated.

Bioinformatics and Data Analysis

Alpha diversity was assessed using multiple indices, including ACE, Chao1, observed species, PD_whole_tree, Shannon, Simpson, and Goods_coverage. Beta diversity was calculated based on Bray-Curtis distances and visualized by principal coordinate analysis (PCoA) and sparse partial least squares discriminant analysis (sPLS-DA). Permutational multivariate analysis of variance (PERMANOVA) was performed using the adonis function in the vegan R package to evaluate differences in microbial community structure across time points. Stage-dependent features were identified using linear discriminant analysis effect size (LEfSe) with a logarithmic LDA score threshold of 2.5.¹⁴

Microbial functional prediction was conducted using PICRUSt2¹⁵ based on ASVs clustered from the 16S rRNA sequencing data, and functional annotations were mapped to the Kyoto Encyclopedia of Genes and Genomes (KEGG) database.¹⁶ Differences in predicted pathways between time points were analyzed using Welch’s t-test, with multiple testing correction performed using the Benjamini-Hochberg false discovery rate (FDR) method in STAMP (version 2.1.3).¹⁷

Statistical Analysis

As this was an exploratory longitudinal microbiome study, no formal a priori sample size calculation was performed. The sample size was determined by the number of eligible patients with DLBCL who underwent anti-CD19 CAR-T therapy and had longitudinal fecal samples available during the study period. Given the limited prior effect size estimates for short-term gut microbiota changes in this clinical setting, the present study was intended to generate preliminary data for future validation in larger cohorts.

Longitudinal comparisons of the Chao1 index and selected microbial taxa were performed using linear mixed-effects models, with time point included as a fixed effect and subject ID included as a random effect to account for within-subject correlation due to repeated measurements. Models were fitted using restricted maximum likelihood (REML), and a two-sided P value < 0.05 was considered statistically significant.

Pearson’s correlation analysis was used as an exploratory analysis to assess associations between microbial taxa and inflammatory markers, including lymphocyte count, serum albumin, neutrophil count, white blood cell count, and CRP. A two-sided P value < 0.05 was considered statistically significant unless otherwise specified.

Results

Baseline Characteristics of DLBCL Patients

All patients achieved a clinical response and had longitudinally available fecal samples for microbiome profiling. Among them, 8 (66.7%) achieved a complete response and 4 achieved a partial response following CAR-T cell infusion. The median age of the 12 patients was 54 years, and 58.3% (n = 7) were male. The median LDH level was 202 U/L. According to ECOG PS, 41.7% (n = 5) of patients had a score < 2, while 58.3% (n = 7) had a score ≥ 2. At a median follow-up of 20.4 months, the estimated 1-year overall survival (OS) rate was 83.3%.

α- and β-Diversity Analysis Across Time Points

α-diversity analysis showed that overall microbial richness and diversity exhibited a slight upward trend over time (Figure 1A–G). Richness-related indices, including Chao1, ACE, and observed OTUs, increased slightly at week 2 after treatment, although the differences were not statistically significant. To account for within-subject correlations arising from repeated sampling, linear mixed-effects modeling was further performed and confirmed that the Chao1 index did not differ significantly across time points. The PD_whole_tree index also showed a mild increase at week 2, suggesting a slight enhancement in phylogenetic diversity. The Shannon index increased slightly, whereas the Simpson index decreased marginally at week 2, indicating a trend toward higher diversity and evenness. Good’s coverage values were consistently close to 1 (~99.99%) across all time points, indicating adequate sequencing depth to capture the majority of taxa present. The Venn diagram (Figure 1H) showed that 76 OTUs were detected at all three time points. Notably, week 1 after treatment had the highest number of unique OTUs (n=24), suggesting that CAR-T therapy may cause short-term perturbations in gut microbiota composition, with partial recovery observed by week 2.

Figure 1 Alpha diversity analysis of gut microbiota. (A–G) Various α-diversity indices were used to assess microbial diversity across treatment groups. (H) Venn diagram illustrating the shared and unique differential microbiota among the different treatment groups.

β-diversity was evaluated using PCoA based on Bray-Curtis distances (Figure 2A). The three groups showed a partial separation along the PC1 and PC2 axes; however, PERMANOVA analysis indicated no statistically significant differences (R² = 0.0681, P = 0.122). Pairwise comparisons of Bray-Curtis distances revealed that the most pronounced difference occurred between baseline and week 2 (P < 0.05).

Figure 2 Beta diversity analysis of gut microbiota. (A) Principal Coordinates Analysis (PCoA) based on gut microbiota composition. (B) Sparse partial least squares discriminant analysis (sPLS-DA) illustrating group separation based on microbial community structure. *P < 0.05; **P < 0.01. Abbreviation: NS, not significant.

The sPLS-DA analysis further demonstrated distinct clustering and separation among the three groups (Figure 2B). Notably, samples collected at week 1 exhibited the tightest clustering and were most clearly separated from both baseline and week 2, suggesting a marked shift in microbial composition at this time point. By week 2, the distribution of samples appeared more similar to baseline, indicating a partial recovery following the initial perturbation.

Temporal Changes in Microbiome Composition Following CAR-T Therapy

At the phylum level (Figure 3A), Firmicutes, Bacteroidota, and Proteobacteria were the dominant taxa across all time points. Linear mixed-effects modeling showed that the relative abundance of Bacteroidota was significantly lower at week 1 than at week 2 (P = 0.002), whereas the difference between baseline and week 2 did not reach statistical significance (P = 0.113). Firmicutes showed a decreasing trend over time (Figure 3D), resulting in a reduced Firmicutes/Bacteroidota (F/B) ratio (Figure 3F). The heatmap (Figure 3C) revealed distinct clustering patterns among the three time points. Compared with baseline, samples from weeks 1 and 2 post-treatment showed a reduced relative abundance of certain opportunistic pathogens, such as Escherichia-Shigella and Enterococcus, along with an increased relative abundance of genera commonly associated with gut health, including Faecalibacterium, Lachnospira, and Bacteroides. By week 2, the overall community composition appeared more similar to baseline, although some characteristic shifts persisted, suggesting that the impact of CAR-T therapy on gut microbiota is short-term and partially reversible.

Figure 3 Comparison of gut microbiota abundance across different time points. (A) Relative abundance of dominant gut microbiota at the phylum level. (B) Relative abundance of dominant gut microbiota at the genus level. (C) Heatmap showing hierarchical clustering of gut microbiota at both phylum and genus levels. (D) Relative abundance of Firmicutes across treatment groups. (E) Relative abundance of Bacteroidota across treatment groups. (F) Ratio of Firmicutes to Bacteroidota across treatment groups. *P < 0.05; **P < 0.01. Abbreviation: NS, not significant.

Further analysis showed that Proteobacteria abundance was significantly elevated at week 1 post-treatment (P = 0.048, Supplementary Figure 1A), whereas Enterococcus showed a slight increase at week 1 followed by a decrease at week 2 (Supplementary Figure 1B).

Differentially Enriched Taxa Across Treatment Time Points

LEfSe analysis identified taxa that were significantly enriched at different time points, as shown in the LDA score bar plot (Figure 4A). At baseline, enriched taxa included Coriobacteriales, Fusobacterium mortiferum, and Clostridium leptum. At week 1 post-treatment, the dominant enriched taxa were Enterococcus, Klebsiella, and Lactobacillus, along with multiple members of the families Veillonellaceae, Desulfovibrionaceae, and Peptostreptococcaceae. Several potentially opportunistic taxa, including Enterobacteriaceae within the phylum Proteobacteria, were also enriched at this time point. By week 2 post-treatment, significantly enriched taxa included Prevotellaceae, Parabacteroides, Prevotella, Faecalibacterium, and several health-associated taxa such as Tannerellaceae and Barnesiellaceae. The cladogram illustrated that these differential taxa clustered across multiple taxonomic levels from phylum to genus, indicating a time-dependent restructuring of the microbial community following CAR-T cell therapy (Figure 4B).

Figure 4 Linear discriminant analysis effect size (LEfSe) of gut microbiota. (A) Bar plot of significantly different taxa identified by LEfSe among treatment groups, based on LDA scores. (B) Cladogram illustrating the phylogenetic distribution of differentially abundant taxa identified by LEfSe.

Altered Microbial Functions and Correlations with Inflammatory Indicators

Functional prediction analysis (Figure 5A) showed that the relative abundance of the lipoic acid metabolism pathway was significantly higher at both week 1 and week 2 post-treatment compared with baseline, with P values <0.02 and <0.05, respectively.

Figure 5 Functional prediction and correlation analysis of gut microbiota. (A) Functional prediction of gut microbiota. (B) Correlation analysis of microbial taxa, illustrating relationships among dominant genera.

Correlation analysis (Figure 5B) revealed significant associations between multiple clinical parameters and differential bacterial genera. CRP was positively correlated with Enterococcus, Escherichia-Shigella, and Streptococcus, and negatively correlated with Faecalibacterium and Bifidobacterium. WBC and neutrophil count were positively associated with potentially pathogenic taxa such as Enterococcus and Escherichia-Shigella, whereas lymphocyte count was positively correlated with Faecalibacterium. Albumin levels showed positive correlations with health-associated genera, including Bacteroides and Faecalibacterium, and negative correlations with certain potentially harmful bacteria.

Discussion

In this longitudinal study, we focused on a subset of DLBCL patients who achieved clear clinical responses to CAR-T therapy and profiled their gut microbiota at three early post-infusion time points. We observed subtle and transient changes in microbial community composition, including a trend toward enrichment of health-associated taxa and a reduction in opportunistic pathogens, as well as functional changes inferred from predictive analysis, notably involving lipoic acid metabolism. These findings provide preliminary insight into short-term microbiome dynamics during the early phase after CAR-T infusion and suggest that microbial profiles, combined with host inflammatory markers, may be related to host immune status and treatment-associated inflammation.

In this study, although α diversity metrics did not change significantly, the transient separation of week 1 samples from baseline in sPLS-DA suggested limited temporal variation in intestinal microbial composition during the peak immune activation phase following CAR-T infusion. Given that the overall beta diversity separation was modest, this finding should be interpreted cautiously as a preliminary compositional fluctuation rather than distinct microbiome restructuring. Similar phenomena have been reported in patients receiving immune checkpoint inhibitor therapy. A previous study demonstrated that in patients with hematologic malignancies treated with anti-CD19 CAR-T cells, reduced gut microbiota diversity was associated with immune-related toxicities and was accompanied by significant enrichment of Enterococcus and Actinomyces.¹⁸ In a longitudinal cohort of patients with multiple myeloma, greater post-treatment disruption of the gut microbiota was also linked to increased toxicity.¹⁹ Consistent with these findings, we observed partial reversion of the microbial community toward baseline by week 2, suggesting that such alterations may be transient and partially reversible. Potential mechanisms include inflammation induced by CAR-T-associated cytokine release, alterations in intestinal mucosal permeability, and changes in antimicrobial peptide secretion patterns, all of which have been reported in experimental models of acute inflammation.^20–23

At the phylum level, the increase in the relative abundance of Bacteroidota at week 2, accompanied by a reduction in the F/B ratio, may be of interest, as similar microbial patterns have been associated with improved metabolic status and reduced inflammatory levels in other immunotherapy settings.^24,25 In a study of melanoma patients treated with immune checkpoint inhibitors, responders exhibited significantly higher Bacteroidota abundance and a lower F/B ratio compared with non-responders.⁸ This microbial profile was associated with higher levels of short-chain fatty acids production, improved intestinal barrier function, and lower levels of pro-inflammatory cytokines, which are thought to enhance antitumor immune responses and mitigate treatment-related inflammation.²⁶ Importantly, the F/B ratio has long been regarded as an ecological marker of intestinal and systemic homeostasis. A lower F/B ratio has been linked to reduced endotoxemia, improved metabolic flexibility, and favorable immunotherapy responses in both solid and hematologic malignancies.^27,28 At the genus level, the enrichment of Faecalibacterium, Lachnospira, and Bacteroides is consistent with their known anti-inflammatory properties, including short-chain fatty acid production and maintenance of epithelial barrier integrity. In contrast, the enrichment at week 1 of Enterococcus, Klebsiella, and members of the family Veillonellaceae, many of which have been linked to bloodstream infections and adverse immunotherapy outcomes, may reflect transient dysbiosis during the early phase after CAR-T infusion.⁵

LEfSe analysis identified taxa that were differentially enriched across time points following CAR-T infusion, with several opportunistic and potentially pro-inflammatory taxa enriched at week 1 and several commensal and potentially beneficial taxa enriched at week 2. This pattern may be consistent with a transient perturbation followed by partial recovery.^29,30 Similar transient post-treatment microbial shifts have been reported in other immunotherapy settings. Subsequent recovery of beneficial taxa, particularly butyrate producers, has been linked to restoration of immune homeostasis and attenuation of inflammation.^8,31 The enrichment of Prevotellaceae, Parabacteroides, and Faecalibacterium at week 2 may indicate partial microbial recovery. These taxa are well-recognized producers of short-chain fatty acids, which can enhance mucosal barrier integrity and downregulate pro-inflammatory signaling pathways,³² thereby facilitating immune reconstitution and inflammation control.

Functional prediction analysis based on 16S rRNA sequencing suggested potential enrichment of lipoic acid metabolism following CAR-T cell infusion. Lipoic acid is a critical mitochondrial cofactor involved in energy metabolism and redox regulation, with potent antioxidant properties that enable scavenging of reactive oxygen species, regeneration of other antioxidants such as glutathione, and preservation of mitochondrial function.^33–35 However, this pathway-level enrichment was inferred from predictive bioinformatic analysis rather than measured directly, and should therefore be interpreted cautiously. Moreover, our correlation analysis showed that health-associated taxa enriched in short-chain fatty acid production, such as Faecalibacterium and Bacteroides, were positively correlated with lymphocyte counts and serum albumin levels, whereas opportunistic taxa including Enterococcus and Escherichia-Shigella were positively correlated with CRP, WBC counts, and neutrophil counts.

This study characterized short-term compositional changes and predicted functional shifts in the gut microbiota of patients with DLBCL undergoing CAR-T therapy through repeated sampling and correlation analyses with inflammatory and immune parameters. Although the observed differences were relatively subtle, they provide preliminary evidence that the gut microbiota may undergo early alterations during the acute post-infusion phase. Several limitations should be acknowledged. First, the small cohort size (n = 12) and single-center design limit the statistical power, robustness, and generalizability of the findings. Second, although the longitudinal repeated-sampling design partially reduced inter-individual heterogeneity, gut microbiota composition during CAR-T treatment may still have been influenced by concurrent factors such as antibiotic exposure, dietary changes, hospitalization-related conditions, and lymphodepletion chemotherapy. Third, the use of 16S rRNA sequencing limits taxonomic resolution and the reliability of functional inference, and the predicted functional findings should therefore be interpreted with caution. Accordingly, the present results should be considered exploratory and preliminary. Overall, the observed microbial changes may indicate a potential link between gut microbiota dynamics and host immune status during CAR-T treatment. Larger multicenter studies with longer follow-up and multi-omics approaches are needed to validate these findings and further clarify the interaction between the gut microbiota and host immunity during CAR-T therapy.