Relationships between the Microbiome and Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer
Article information
Abstract
Purpose
This study aimed to investigate the dynamic changes in the microbiome of patients with locally advanced rectal cancer (LARC) undergoing neoadjuvant chemoradiotherapy (nCRT), focusing on the relationship between the microbiome and response to nCRT.
Materials and Methods
We conducted a longitudinal study involving 103 samples from 26 patients with LARC. Samples were collected from both the tumor and normal rectal tissues before and after nCRT. Diversity, taxonomic, and network analyses were performed to compare the microbiome profiles across different tissue types, pre- and post-nCRT time-points, and nCRT responses.
Results
Between the tumor and normal tissue samples, no differences in microbial diversity and composition were observed. However, when pre- and post-nCRT samples were compared, there was a significant decrease in diversity, along with notable changes in composition. Non-responders exhibited more extensive changes in their microbiome composition during nCRT, characterized by an increase in pathogenic microbes. Meanwhile, responders had relatively stable microbiome communities with more enriched butyrate-producing bacteria. Network analysis revealed distinct patterns of microbial interactions between responders and non-responders, where butyrate-producing bacteria formed strong networks in responders, while opportunistic pathogens formed strong networks in non-responders. A Bayesian network model for predicting the nCRT response was established, with butyrate-producing bacteria playing a major predictive role.
Conclusion
Our study demonstrated a significant association between the microbiome and nCRT response in LARC patients, leading to the development of a microbiome-based response-prediction model. These findings suggest potential applications of microbiome signatures for predicting and optimizing nCRT treatment in LARC patients.
Introduction
Colorectal cancer is one of the leading causes of cancer-associated mortality, with approximately one-third of malignancies occurring in the rectum [1]. The management paradigm for locally advanced rectal cancer (LARC) has evolved significantly in recent years, marked by two key trends. On one hand, there has been a shift toward using aggressive neoadjuvant treatment before surgery [2]. On the other hand, there is a growing trend to avoid surgery and instead opt for a close surveillance strategy known as non-operative management (NOM) [3]. Numerous efforts have been made to identify suitable candidates for the NOM strategy, however, this remains challenging due to the lack of definitive biomarkers [4].
Beyond tumor-intrinsic characteristics, the microbiome has emerged as a promising, non-invasive biomarker in colorectal cancer [5]. Landmark studies have successfully enhanced the efficacy of anticancer therapies by modulating the gut microbiome in both mouse and human models [6,7]. These findings support the potential of the microbiome as a predictor of treatment efficacy, prompting numerous ongoing investigations [5,8]. However, while our understanding of the microbiome’s interaction with systemic therapy is growing, its role in the context of chemoradiotherapy remains largely unexplored.
In this study, we aimed to investigate the relationship between the microbiome and response to neoadjuvant chemoradiotherapy (nCRT) in patients with LARC. We specifically focused on comparative analyses of the microbiome profiles in: (1) tumor tissue versus adjacent normal tissue, (2) pre-nCRT versus post-nCRT samples, and (3) responders versus non-responders to nCRT. Additionally, we sought to develop a microbiome-based predictive model for nCRT response, aiming to evaluate the potential of the microbiome as a biomarker for treatment efficacy.
Materials and Methods
1. Patient selection and treatment
The inclusion criteria were patients diagnosed with LARC, defined as clinical T category 3 or higher or with nodal involvement, who underwent nCRT followed by surgical resection between January 2008 and December 2016. The nCRT regimen consisted of concurrent chemoradiotherapy with either capecitabine or 5-fluorouracil (5-FU) monotherapy, without prior induction chemotherapy. Patients who had been exposed to antibiotics, steroids, or immunosuppressants within 4 weeks prior to sample collection were excluded. Pathologic response was evaluated by two independent colorectal pathologists using the three-point tumor regression grade (TRG) system of the American Joint Commission on Cancer [9]. This system categorizes the TRG as follows: TRG 0, no residual viable tumor; TRG 1, single or small groups of tumor cells; TRG 2, residual viable tumor cells outgrown by fibrosis; and TRG 3, minimal or no tumor cells killed. Patients with a TRG of 0 or 1 were classified as responders, and those with a TRG of 2 or 3 were classified as non-responders.
2. Sample collection and processing
This study utilized tissue samples from an existing repository, originally collected with patient consent for general research purposes. The requirement for additional informed consent was waived by the IRB. For each patient, both tumor and nearby normal tissue samples were collected from prenCRT biopsies and post-nCRT surgical specimens (S1 Fig.). The tumor and normal tissue areas were identified on hematoxylin and eosin-stained slides and manually dissected using a scalpel. Formalin-fixed paraffin-embedded (FFPE) tissues were digested in a cell lysis solution containing proteinase K, and DNA was extracted.
To construct standard exome capture libraries, the Agilent SureSelect target enrichment protocol was constructed for Illumina paired-end sequencing libraries, with 1 µg of input genomic DNA. The SureSelect Human All Exon V6 probe set (Agilent Technologies, Inc.) was used in all cases. FFPE genomic DNA was sheared using a Covaris LE220 focused ultrasonicator (Covaris LLC), and exome capture was performed according to the manufacturer’s protocol. Indexed libraries with an average coverage of 30× were sequenced using the NovaSeq 6000 platform (Illumina, Inc.) at Macrogen, Inc.
3. Bioinformatic and statistical analysis
The GATK PathSeq algorithm was used to perform computational subtraction of human reads, followed by alignment of residual reads to both human and microbial reference genomes [10]. These alignments resulted in taxonomic classification of the reads into bacterial, viral, archaeal, and fungal sequences. Classification was conducted across the following levels: domain, phylum, class, order, family, and genus. At the genus level, the relative abundance of each organism was calculated and normalized per sample so that the total relative abundance for each sample was summed to one. To exclude noisy variations, only the taxa present in more than five samples were included.
Alpha diversity was evaluated using observed operational taxonomic units (OTUs) along with Shannon and inverse Simpson indices. For beta diversity analysis, Bray-Curtis and Euclidean distances were employed in conjunction with principal coordinate analysis (PCoA) and non-metric multidimensional scaling (NMDS) techniques. To identify discriminatory taxa between groups, linear discriminant analysis (LDA) effect size (LEfSe) was conducted, setting a threshold of 1.0 and utilizing the LDA score [11]. LEfSe results were visualized as a cladogram. Microbial abundance in responder and non-responder groups was depicted as a heatmap using the R package “ComplexHeatmap” [12].
For association network analysis and prediction model development, a Bayesian network modeling approach was utilized. This method is well-suited for capturing complex interactions within large datasets such as microbiome data, providing a robust probabilistic framework for prediction, and offering a graphical representation that facilitates interpretation of high-dimensional data. The Bayesian network classified microbial taxa into binary states (presence/absence) based on their relative abundances and analyzed association networks among taxa using the maximum spanning tree algorithm and Pearson’s correlation. In this network, each taxon was represented as a node, with significant interactions depicted by edges connecting the corresponding nodes. A microbiome-based prediction model for nCRT response was constructed using supervised learning with an augmented Markov blanket algorithm, based exclusively on pre-nCRT samples. Internal validation was performed by estimating the area under the receiver operating characteristic curve (AUC) through 5-fold cross-validation, combined with bootstrapping resampling (1,000 iterations).
Statistical comparisons between samples were performed using t tests or one-way analysis of variance (ANOVA). All statistical analyses were conducted using R software ver. 4.3.1 (R Foundation for Statistical Computing), and a twosided p < 0.05 was considered statistically significant. Bayesian network analysis and association network generation were performed using BayesiaLab 9.0 (Bayesian Ltd.).
Results
1. Patient characteristics
Patient characteristics are summarized in Table 1. A total of 103 samples from 26 patients with LARC were included in this study. Of the 26 patients, nine patients (34.6%) were classified as responders and 17 (65.4%) as non-responders. Both groups exhibited similar characteristics. The median age at diagnosis was 61 years (range, 46 to 70 years) for responders and 60 years (range, 41 to 81 years) for non-responders. All patients completed planned nCRT and underwent surgical resection. The total radiation dose ranged from 50.4 Gy to 54.0 Gy, delivered at 1.8 Gy per fraction. Concurrent Capecitabine was administered to 55.6% responders and 58.8% non-responders, whereas 5-FU was administered to 44.4% responders and 41.2% non-responders. No patient achieved a complete pathological response in the surgical specimens.
Patient characteristics
2. Comparison between tumor and normal tissue samples
In the entire sample set, the microbiome communities were predominantly composed of four bacterial phyla: Bacteroidetes, Proteobacteria, Firmicutes, and Actinobacteria. These four phyla collectively represented over 95% of the total microbial composition.
For the comparison between tumor and adjacent normal rectal tissues, we examined pre-nCRT samples consisting of 26 paired tumor and normal tissue samples. We observed a high degree of similarity in microbial composition across all taxonomic levels between the paired samples. Alpha diversity analysis showed no significant differences between two tissue types in terms of observed OTUs (p=0.940), Shannon index (p=0.966), and inverse Simpson index (p=0.849) (Fig. 1A). Additionally, beta diversity analysis using Bray-Curtis and Euclidean distance metrics demonstrated highly consistent patterns between the paired tissue types (Fig. 1B). These findings collectively indicate a similarity in both the composition and diversity of the microbiome between the paired tumor and adjacent normal tissue samples before nCRT.
Comparison of microbial diversity between tumor and adjacent normal rectal tissue samples (n=26 per group; pre–neoadjuvant chemoradiotherapy samples only). (A) Alpha diversity assessment using observed operational taxonomic units (OTUs), Shannon index, and inverse Simpson index. (B) Beta diversity evaluation using principal coordinates analysis (PCoA) and non-metric multidimensional scaling (NMDS) plots based on Bray-Curtis and Euclidean distance metrics.
3. Comparison between pre-nCRT and post-nCRT samples
For the comparison of the microbiome before and after nCRT, we analyzed 26 paired tumor samples collected at both pre-nCRT and post-nCRT time points. The analysis revealed dramatic shifts in microbial composition across all taxonomic levels during nCRT. At the phylum level, the abundances of Firmicutes and Bacteroidetes declined, while the abundance of Proteobacteria increased after nCRT. Alpha diversity analysis showed that the pre-nCRT samples had significantly higher richness and evenness compared to the post-nCRT samples, as indicated by the number of observed OTUs (p < 0.001), Shannon index (p=0.010), and inverse Simpson index (p < 0.001) (Fig. 2A). Beta diversity analysis confirmed these differences, with distinct clustering of the pre-nCRT and post-nCRT groups in both PCoA and NMDS plots using Bray-Curtis and Euclidean distances (Fig. 2B). A detailed taxonomic analysis revealed a decline in commensal bacteria such as Bacteroides, Prevotella, and Bifidobacterium after nCRT. Conversely, there was a significant increase in opportunistic pathogens, including Mycobacterium and Pseudomonas. These findings suggest that nCRT may reshape the microbiome community by reducing its diversity and altering its composition.
Comparison of microbial diversity between pre– and post–neoadjuvant chemoradiotherapy (nCRT) samples (n=26 per group; tumor samples only). (A) Alpha diversity assessment using observed operational taxonomic units (OTUs), Shannon index, and inverse Simpson index. (B) Beta diversity evaluation using principal coordinates analysis (PCoA) and non-metric multidimensional scaling (NMDS) plots based on Bray-Curtis and Euclidean distance metrics. *p < 0.05, ***p < 0.001.
4. Comparison between responder and non-responder samples
We compared the microbiome between responders and non-responders at both pre-nCRT and post-nCRT time points, as well as the changes occurring from pre- to post-nCRT. To gain maximum insights from the available data, the analysis included samples from both tumor and adjacent normal rectal tissue.
Alpha diversity showed no significant difference between responders and non-responders at the pre-nCRT time point. However, both groups exhibited a significant decrease in alpha diversity after nCRT, with this effect being more pronounced in non-responders (S2 Fig.). LEfSe analysis revealed that non-responders experienced more extensive changes in microbiome composition during nCRT (Fig. 3A). These changes were characterized by an increase in the Proteobacteria phylum and specific pathogens such as Mycobacterium and Pseudomonas. Conversely, responders maintained a relatively stable microbiome composition during nCRT.
Comparison of microbial profiles between neoadjuvant chemoradiotherapy (nCRT) responders (n=36) and non-responders (n=67). (A) Cladogram of differentially abundant taxa before and after nCRT: red nodes indicate increased abundance post-nCRT; green nodes show decreased abundance. (B) Heatmap of pre-nCRT microbiome composition at family level. (C) Relative abundance of four major butyrate-producing genera. R, responder; NR, non-responder. (D) Heatmap of post-nCRT microbiome composition at family level. (E) Relative abundance of four opportunistic pathogens.
Within the pre-nCRT samples, we found that the responder group harbored a higher abundance of Lachnospiraceae and Ruminococcaceae families, known for their butyrate-producing bacteria, compared to the non-responder group (Fig. 3B). This observation was further corroborated at the genus level, with Ruminococcus, Roseburia, Lachnospira, and Faecalibacterium significantly enriched in responders (Fig. 3C). In contrast, post-nCRT analysis showed that the non-responder group exhibited a distinct microbial signature. Families known to harbor opportunistic pathogens, such as Corynebacteriaceae, Enterobacteriaceae, Mycobacteriaceae, and Pseudomonadaceae, flourished among the non-responders (Fig. 3D). This trend was also observed at the genus level, with Mycobacterium, Pseudomonas, Corynebacterium, and Enterobacter showing significantly higher abundance in non-responders (Fig. 3E).
5. Response-prediction model
The association network analysis revealed distinct patterns of microbiome interactions between responders and non-responders (Fig. 4A). In the responder group, butyrate-producing bacteria formed a robust network with strong correlations, whereas the non-responder group showed a clustered network of opportunistic pathogens that displayed a strong correlation. Notably, these interactions were not observed in the opposite groups, as butyrate-producing bacteria lacked networks in non-responders and opportunistic pathogens lacked networks in responders.
Bayesian network analysis of microbial interactions and neoadjuvant chemoradiotherapy (nCRT) response prediction. (A) Microbial association networks in responders (n=36) and non-responders (n=67). (B) Bayesian network model for predicting nCRT response showing five selected microbial taxa (n=18 for responders; n=34 for non-responders; pre-nCRT samples only). (C) Effect of selected microbial taxa abundances on nCRT response rate.
Finally, the microbiome-based model predicting nCRT response was developed based on the pre-nCRT samples (Fig. 4B). This model identified five key microbes as the optimal set for predicting nCRT response: Ruminococcus, Lactobacillus, Lachnospira, Thermus, and Sphingomonas. Of these, Ruminococcus, Lactobacillus, and Lachnospira, which are all butyrate-producing bacteria, demonstrated a direct positive correlation with nCRT response rates (Fig. 4C). In contrast, the abundance of Sphingomonas showed an inverse relationship with response rate beyond a certain threshold. This model demonstrated acceptable performance in internal validation, achieving a mean AUC of 0.82 through 5-fold cross-validation and 0.84 using bootstrapping with 1,000 resampling iterations.
Discussion
In this study, microbiome analysis was conducted at longitudinal time points of pre- and post-nCRT in patients with LARC, focusing on the association between the microbiome and the nCRT response. Highly similar microbiome profiles were observed between the paired tumor and normal tissue samples, suggesting that intratumoral bacteria may have originated from adjacent normal tissues. In contrast, throughout nCRT, the microbiome community underwent significant changes in terms of composition and diversity. Notably, the patterns of change differed between responders and non-responders. Non-responders demonstrated extensive changes in microbial composition, particularly an increase in opportunistic pathogens. Meanwhile, responders exhibited higher baseline levels of butyrate-producing bacteria that together formed robust networks and experienced more stable microbiome shifts during nCRT. Furthermore, the response-prediction model identified five key microbes, with butyrate-producing bacteria playing a pivotal role in predicting response to nCRT.
The relationship between the microbiome and cancer treatment is complex and remains at an early stage of exploration [13]. Several preclinical and clinical studies have reported drastic alterations in microbiome composition and diversity during radiotherapy or chemotherapy [14,15]. However, only a few studies have investigated microbiome data in the setting of chemoradiation. Yi et al. [16] conducted the largest study in this area, analyzing 167 fecal samples from 84 LARC patients. They identified two distinct types of microbiome communities among the samples: one characterized by a pathological microbiome and the other by a beneficial microbiome driven by butyrate-producing microbes. Notably, the latter ‘beneficial microbiota’ showed significantly higher response rates for each clustering method. They also found that butyrate-producing bacteria such as Roseburia, Dorea, and Anaerostipes were overrepresented in responders, whereas Coriobacteriaceae and Fusobacterium were overrepresented in non-responders. Sanchez-Alcoholado et al. [17] evaluated fecal samples from 40 patients with colorectal cancer and compared the microbiome profiles between nCRT responders and non-responders. They reported that the responder group was more enriched in butyrate-producing bacteria and had significantly higher levels of butyric acid than the non-responder group. Collectively, these findings are broadly consistent with our results, although some variations may arise from differences in sample types: those studies examined fecal samples, whereas our study utilized tissue samples.
Tissue samples directly reflect the tumor microenvironment, providing site-specific information crucial for elucidating tumor-microbiome interactions [18]. In contrast, fecal samples, containing microbes from the entire colorectum, may less accurately represent the tumor-associated microbiome. Nevertheless, their non-invasive nature and suitability for repeated sampling make fecal samples more practical as biomarkers in the clinical settings [19]. Research on the concordance between tissue and fecal samples has yielded mixed results. Some studies have identified distinctly different microbial communities in these two sample types, while others have found sufficient similarity in their core microbial compositions to support the use as reliable biomarkers [18-21]. Considering the relevance of tissue samples in capturing tumor-specific microbiomes and the clinical utility of the fecal samples, a comprehensive approach integrating both sample types would provide more robust information and potentially serve as an effective clinical tool.
Butyrate, a major component of short-chain fatty acids, plays an essential role in maintaining gut homeostasis and serves as a primary energy source for colonocytes [22]. Interestingly, research suggests an association between butyrate-producing bacteria and an improved response to anticancer treatments. Several studies have demonstrated that patients with higher levels of butyrate-producing bacteria show better outcomes following systemic therapy [23,24]. Regarding radiotherapy, a recent study using intestinal organoids from colorectal cancer patients revealed that butyrate can suppress tumor proliferation and enhance radiation-induced cell death [25]. The mechanisms through which butyrate influences the therapeutic response are not fully understood but may include direct inhibition of histone deacetylases, alteration of cancer cell metabolism, and activation of G protein– coupled receptor signaling pathways [26]. Importantly, the effects of butyrate are not uniform, varying with concentration and host genotype, which adds layers of complexity to our understanding.
This study identified an increase in opportunistic pathogens among non-responders following nCRT. While this could be attributed to chance or confounding factors, we propose a potential link between nCRT and increased opportunistic pathogens. In a state of homeostasis, the microbiome protects the host against colonization by true and opportunistic pathogens via “colonization resistance” mechanism [27]. Conversely, when microbiome dysbiosis occurs, this protective mechanism could be compromised, creating an environment conducive to pathogen proliferation. Multiple studies have consistently reported an association between chemoradiotherapy and microbiome dysbiosis [14,15]. This study also demonstrated substantial changes in the microbiome communities after nCRT, suggesting that nCRT may contribute to the observed increase in opportunistic pathogens through microbiome dysbiosis. Moreover, lymphocyte depletion induced by chemotherapy or radiotherapy has been shown to significantly increase the risk of opportunistic infections, further supporting this hypothesis [28]. However, current study design precludes evaluating this hypothesis or explaining the predominance of this phenomenon in the non-responder group. Future prospective studies with longitudinal immune marker monitoring and larger microbiome datasets are needed to elucidate these potential interactions and clinical implications.
This study has several limitations. First, this study had a relatively small sample size, which may limit the generalizability of our findings. Although previous microbiome studies have demonstrated significant results with comparable cohort sizes [16,17,24], external validation using independent cohorts is necessary to confirm the robustness and broader applicability of our results. Second, our predictive model incorporated both tumor and normal tissue samples, considering that microbiome profiles from both tissue types could hold clinical significance. While this inclusion of paired samples may have introduced potential statistical bias, the implementation of consistent sampling methodology across all patients likely mitigated this risk. Third, our analysis was confined to the genus-level taxonomy, which limited our ability to conduct reliable functional pathway analysis and identify more granular information that species-level data would have provided. Fourth, the use of FFPE specimens for microbiome analysis presents technical challenges, including potential DNA degradation and fragmentation during the preservation process. These specimens may be susceptible to contamination during processing and storage [29]. Though several studies have shown encouraging concordance between FFPE and fresh-frozen tissue microbiome profiles, these methodological constraints necessitate careful interpretation of our results [30].
Despite these limitations, our findings provide valuable insights into the potential utility of microbiome signatures for treatment response prediction. In the current landscape of personalized medicine, where accurate outcome prediction is increasingly crucial, the integration of microbiome signatures with established clinical factors may enhance predictive accuracy. Future large-scale studies with comprehensive microbiome analysis and external validation will be essential to establish robust predictive models for LARC patients.
In conclusion, this study demonstrated a bidirectional relationship between the microbiome and nCRT in LARC patients. The nCRT significantly altered the diversity and composition of the microbiome. Conversely, specific microbial populations, particularly butyrate-producing bacteria, were associated with favorable responses to nCRT. We developed a microbiome-based model for predicting nCRT response, which achieved robust performance in internal validation. Further research is needed to explore the clinical potential of the microbiome as a predictive biomarker and to optimize LARC treatment.
Electronic Supplementary Material
Supplementary materials are available at Cancer Research and Treatment website (https://www.e-crt.org).
Notes
Ethical Statement
This study was approved by the Institutional Review Board (IRB) of Seoul National University Hospital (IRB No. H-2204-119-1317). The requirement for informed consent was waived due to the retrospective nature of the study.
Author Contributions
Conceived and designed the analysis: Lee HI, Jang BS, Chie EK.
Collected the data: Lee HI, Lee HI, Jang BS, Chang JH, Kim E, Lee TH, Park JH, Chie EK.
Contributed data or analysis tools: Lee HI, Jang BS, Chang JH, Kim E, Lee TH, Park JH, Chie EK.
Performed the analysis: Lee HI, Jang BS.
Wrote the paper: Lee HI.
Supervision: Chie EK.
Funding acquisition, resources, writing - review & editing: Chie EK.
Conflicts of Interest
Conflict of interest relevant to this article was not reported.
Funding
This work was funded by Patient-Centered Clinical Research Coordinating Center (PACEN) funded by the Ministry of Health & Welfare, Republic of Korea (RS-2021-KH120306) for Prof. Eui Kyu CHIE, MD, PhD and the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. NRF-2017R1A2B4003535) and Seoul National University Hospital (0320210410) for Prof. Ji Hyun CHANG, MD, PhD.