Unraveling the role of gut microbiome in predicting adverse events in neoadjuvant therapy for rectal cancer Authors: Jingxin Ma, Shengbo Sun, Xin Cheng, Cong Meng, Hanzheng Zhao, Wentao Fu, Yan Gao, Liyan Ma, Zhengyang Yang, Hongwei Yao, Jianrong Su Published: 2024 ABSTRACT ABSTRACTSome patients may develop adverse events during neoadjuvant chemoradiotherapy combined with immunotherapy, influencing response rates. The roles of intestinal microbiome and its metabolites in therapeutic adverse events remain unclear. We collected baseline fecal samples from 21 patients with adverse events (AE group) and 11 patients without adverse events (Non-AE group). Their microbiota and metabolome were characterized using metagenomic shotgun sequencing and untargeted metabolomics. At the species level, the gut microbiota in the Non-AE group exhibits significantly higher abundance of Clostridium sp. Alistipes sp. and lower abundance of Lachnoclostridium sp. Weissella cibaria, Weissella confusa, compared to the AE group (p < .05). A total of 58 discriminative metabolites were identified between groups. Beta-alanine metabolism was scattered. Boc-beta-cyano-L-alanine and CoQ9 were significantly increased in patients without adverse events, while linoleic acid increased in patients with adverse events. The increased Alistipes sp. in the Non-AE group was positively correlated with Boc-beta-cyano-L-alanine and negatively correlated with linoleic acid (p < .05). We constructed a combined microbiome-metabolite model to distinguish Non-AE and AE patients with an AUC of 0.963 via the random forest algorithm. Our findings provided a novel insight into the interplay of multispecies microbial cluster and metabolites of rectal patients with adverse events in neoadjuvant chemoradiotherapy combined with immunotherapy. These microbiota and metabolites deserve further investigations to reveal their roles in adverse events, providing clues for better treatment scenarios.Trial registration number: ClinicalTrials.gov identifier: NCT05368051 Introduction According to the American Cancer Society report, colorectal cancer accounts for the majority of new cases among digestive tract tumors. A total of 46,050 patients out of the predicted 153,020 colorectal patients have rectal cancer, accounting for approximately 30.09% in 2023.1 Therefore, the urgent quest for a good treatment for rectal cancer has become critical. The European Society of Medical Oncology (ESMO) recommends neoadjuvant therapy in the three cases of locally advanced rectal cancer, in any lymph node stage, and in cases where complete resection cannot be achieved.2 Traditional neoadjuvant therapy includes radiotherapy and chemotherapy. The pathological complete response of the therapy was lower than 15%.3,4 Previous studies did not focus on immune checkpoint inhibitors plus concomitant and sequential chemoradiotherapy as the neoadjuvant scheme. Additionally, both subtypes of MSI-H/dMMR and MSS/pMMR locally advanced rectal cancer patients were involved. Taking into consideration the above, our center performed a study enrolled LARC patients receiving immune checkpoint inhibitors and chemoradiotherapy recently. After the neoadjuvant therapy, a total of 32 MSS/pMMR patients were included, with a pathological complete response rate of 43.8% (14/32), which was much higher than traditional neoadjuvant therapies.5 Although the method brought encouraging therapeutic effects, some patients experienced adverse events of 1–2 grade, which impacted the efficacy of treatment. However, the underlying mechanisms involved in the occurrence and development of adverse events are far from being explored. Intestinal microbiota are connected to cancer treatment through metabolites.6,7 Metabolites could exert either genotoxic or conducive functions based on complex mechanisms, such as by providing supportive metabolic flux to promote anabolic metabolism, suppressing inflammation, or recruiting immune cells to tumor.8 For example, researchers found that microbial metabolite butyrate could promote the anti-tumor effects of oxaliplatin by enhancing CD8+T cell response during chemotherapy.9 In a mouse model, Bifidobacterium enhanced the effect of anti-PD-1 antibody by secreting hippurate and inhibiting PD-1 expression to mediate tumor destruction.10 Therefore, the microbiome has come to the forefront in tumor therapy, which may be robust noninvasive targets for the precision medicine.7 However, studies between gut microbiota and adverse events mainly focused on monotherapy by 16S rRNA sequencing which can only reach the genus level of the gut microbiome.11,12 It is crucial to explore the relationship between gut microbiome and adverse events during combined neoadjuvant therapy. Hence, we collected samples of 32 patients with locally advanced rectal cancer (LARC) for metagenomic sequencing. Then, we systematically investigated differences of gut microbiota between no adverse event and adverse event at baseline. Differential microbiota enriched into metabolic pathways. Based on metabolomics, differential metabolites were identified. We then linked the microbiome to metabolites, which revealed possible mechanisms for adverse events. A predicted combined model using a random forest machine was constructed, effectively guiding clinical treatment and paving the way for future exploration. Materials and methods Enrollment of patients and clinical assessment This prospective, single-arm phase 2 clinical trial was approved by the ethics committee of Beijing Friendship Hospital, Capital Medical University (2021-P2-318-01). Since our aim was to predict the occurrence of adverse effects in patients undergoing neoadjuvant chemoradiotherapy combined with immunotherapy based on intestinal microbiome, a normal control was not considered in the study. All patients provided written, informed consent before collection of fecal samples. The study was conducted in compliance with the Helsinki Declaration. The trial registration was at ClinicalTrials.gov Identifier (NCT05368051). pMMR/MSS locally advanced rectal patients aged ≥18 years (cT3-4aN0M0 and cT1-4aN1-2M0) with a tumor distal location ≤10 cm from the anal verge (determined by MRI scan) were eligible for inclusion in this study (Table S8). Thirty-two pMMR/MSS locally advanced rectal patients (cT1-4aN0-2M0) (15 females; 17 males) from October 2021 to August 2022 were enrolled receiving long-course radiotherapy (50 Gy/25 f, 2 Gy/f, 5 days/week), three 21-day cycles of capecitabine (850–1000 mg/m2, bid, po, Days 1–14), and three 21-day cycles of tislelizumab (200 mg, iv.gtt, Day 8) as neoadjuvant therapy (Figure 1). The first two cycles of chemotherapy were provided during XRT treatment. In order to ensure sufficient treatment process, postoperative adjuvant therapies were chosen as FOLFOX or CAPEOX (12–16 weeks) whether with or without pCR according to the NCCN Guidelines. Cases of treatment-related adverse events (TrAEs) were graded according to the Common Terminology Criteria for Adverse Events (CTCAE version 4.0), and the TrAE scores were reviewed by at least two oncologists every 2 weeks during neoadjuvant therapy. Grade refers to the severity of the AE. The CTCAE displays Grades 1 through 5 with unique clinical descriptions of severity for each AE based on this general guideline. Based on the patients’ ratings, we categorized them as non-adverse events (Non-AE group) or adverse events (AE group). Supportive medications were administered to patients to mitigate toxicities. The management of adverse events followed the guidelines of the Society for Immunotherapy of Cancer (SITC) toxicity management working group.13 For instance, medications such as Imodium were used to manage diarrhea, while Ibuprofen and Loxoprofen Sodium were prescribed for abdominal pain. Loratadine and Halometasonum Cream were often used to treat rashes. If adverse events became life-threatening, neoadjuvant therapy was immediately discontinued. Patients without adverse events were confirmed at the last clinical visit. Laboratory specimens were collected at baseline and after neoadjuvant therapy for measurements, including CEA, WBC, LYM, and AST. These indices were detected in the Sysmex XN-Series, which were fully automated blood cell counter/WBC differential systems that can be connected to Sysmex SP-10, an auto mated slide making/staining device, and further to the automated digital cell image analyzer DI-60 that contains the acquisition and classification software.14 Patient baseline clinical information, including age, sex, and BMI, was collected and summarized in Table 1. All patients were on a regular diet without liquid food or extra nutritional supplementation. Probiotics were not allowed in the study. Figure 1.Study design. Locally advanced rectal cancer patients (n = 32) received long-course radiotherapy (50 Gy/25 f, 2 Gy/f, 5 days/week), three 21-day cycles capecitabine (850-1000 mg/m2, bid, po, day1-14), and three 21-day cycles tislelizumab (200 mg, iv.Gtt, day8) as neoadjuvant therapy. Baseline feces were collected prior to treatment. Adverse events during treatment were recorded and rated according to the common terminology criteria for adverse events (version 4.0) to categorize them as no adverse events (non-AE group) or adverse events (AE group).Table 1.Clinical information of enrolled patients.CharacteristicsAE (n = 21)Non-AE (n = 11)PAge (years) 1.000≤599 (42.9%)4 (36.4%) >5912 (57.1%)7 (63.6%) Sex .712male12 (57.1%)5 (45.5%) female9 (42.9%)6 (54.5%) Body mass index (kg/m2) .47218.5–23.911 (52.4%)4 (36.4%) >23.910 (47.6%)7 (63.6%) Hypertension (mmHg) 1.000Yes9 (42.9%)4 (36.4%) No12 (57.1%)7 (63.6%) Family history of tumor .374Yes6 (28.6%)1 (9.1%) No15 (71.4%)10 (90.9%) Smoking .703Yes8 (38.1%)3 (27.3%) No13 (61.9%)8 (72.7%) Drinking .637Yes5 (23.8%)1 (9.1%) No16 (76.2%)10 (90.9%) MRF .721+8 (38.1%)5 (45.5%) -13 (61.9%)6 (54.5%) EMVI .450+10 (47.6%)3 (27.3%) -11 (52.4%)8 (72.7%) Distance from distal border of tumor to anal verge (cm)5.57 ± 2.704.95 ± 2.61.537Length of tumor lesion (cm)4.91 ± 1.954.54 ± 1.46.592CEA (ng/ml)3.86 (1.69–6.05)3.64 (2.27–5.47).876WBC (10^9/L)6.34 ± 1.77.14 ± 1.68.213NEU (%)67.8 (55.2–75.85)59.6 (55.2–69.9).208LYM (10^9/L)1.58 (0.94–2.31)2.23 (2.07–2.42).056PLT (10^9/L)229.71 ± 55.82249.18 ± 60.06.368Hb (g/L)131.67 ± 15.00139.73 ± 12.30.137CRP (mg/L)1.79 (0.96–2.87)1.65 (0.81–9.95).815TBIL (μmol/L)12.97 (10.82–14.78)11.5 (10.16–13.3).194ALB (g/L)43.1 (38.9–46.65)42.6 (39.5–43.3).696GLB (g/L)31 (24.5–37.8)27.5 (26.5–33.4).347ALT (U/L)15 (13–17)16 (12–19).506AST (U/L)19 (15.05–23)17.3 (17–19).327CRE (μmol/L)67.7 (62.1–77.81)73.2 (62–78).755UA (μmol/L)326.9 (277.5–432.7)293 (279.5–366).238Urea (mmol/L)5.16 (4.57–5.9)5.17 (4.37–5.21).584 Study design. Locally advanced rectal cancer patients (n = 32) received long-course radiotherapy (50 Gy/25 f, 2 Gy/f, 5 days/week), three 21-day cycles capecitabine (850-1000 mg/m2, bid, po, day1-14), and three 21-day cycles tislelizumab (200 mg, iv.Gtt, day8) as neoadjuvant therapy. Baseline feces were collected prior to treatment. Adverse events during treatment were recorded and rated according to the common terminology criteria for adverse events (version 4.0) to categorize them as no adverse events (non-AE group) or adverse events (AE group). Table 1.Clinical information of enrolled patients.CharacteristicsAE (n = 21)Non-AE (n = 11)PAge (years) 1.000≤599 (42.9%)4 (36.4%) >5912 (57.1%)7 (63.6%) Sex .712male12 (57.1%)5 (45.5%) female9 (42.9%)6 (54.5%) Body mass index (kg/m2) .47218.5–23.911 (52.4%)4 (36.4%) >23.910 (47.6%)7 (63.6%) Hypertension (mmHg) 1.000Yes9 (42.9%)4 (36.4%) No12 (57.1%)7 (63.6%) Family history of tumor .374Yes6 (28.6%)1 (9.1%) No15 (71.4%)10 (90.9%) Smoking .703Yes8 (38.1%)3 (27.3%) No13 (61.9%)8 (72.7%) Drinking .637Yes5 (23.8%)1 (9.1%) No16 (76.2%)10 (90.9%) MRF .721+8 (38.1%)5 (45.5%) -13 (61.9%)6 (54.5%) EMVI .450+10 (47.6%)3 (27.3%) -11 (52.4%)8 (72.7%) Distance from distal border of tumor to anal verge (cm)5.57 ± 2.704.95 ± 2.61.537Length of tumor lesion (cm)4.91 ± 1.954.54 ± 1.46.592CEA (ng/ml)3.86 (1.69–6.05)3.64 (2.27–5.47).876WBC (10^9/L)6.34 ± 1.77.14 ± 1.68.213NEU (%)67.8 (55.2–75.85)59.6 (55.2–69.9).208LYM (10^9/L)1.58 (0.94–2.31)2.23 (2.07–2.42).056PLT (10^9/L)229.71 ± 55.82249.18 ± 60.06.368Hb (g/L)131.67 ± 15.00139.73 ± 12.30.137CRP (mg/L)1.79 (0.96–2.87)1.65 (0.81–9.95).815TBIL (μmol/L)12.97 (10.82–14.78)11.5 (10.16–13.3).194ALB (g/L)43.1 (38.9–46.65)42.6 (39.5–43.3).696GLB (g/L)31 (24.5–37.8)27.5 (26.5–33.4).347ALT (U/L)15 (13–17)16 (12–19).506AST (U/L)19 (15.05–23)17.3 (17–19).327CRE (μmol/L)67.7 (62.1–77.81)73.2 (62–78).755UA (μmol/L)326.9 (277.5–432.7)293 (279.5–366).238Urea (mmol/L)5.16 (4.57–5.9)5.17 (4.37–5.21).584 Clinical information of enrolled patients. Fecal DNA extraction and library preparation A total of 32 baseline fecal samples were collected after colonoscopy and before neoadjuvant therapy. All patients discontinued anticoagulants, antioxidant herbs, and antidepressants 10 days before the colonoscopy. One day before the examination, patients had three meals consisting of light, low-residue, semi-fluid foods (e.g., white rice porridge, white steamed bread, lotus root powder). They avoided greasy foods, vegetables, beans, and watery fruits. At 19:00, patients dissolved one bag of PEG-electrolyte powder in 1,000 ml of warm water, ensuring it was completely dissolved, and drank it within 60 min while walking around to promote digestion. On the day of the examination, patients drank 2,000 ml of PEG-electrolyte solution, consuming 1,000 ml within 1 h and then drinking 250 ml every 15 min until finished. Walking around during this process helped promote intestinal peristalsis, aiding in effective bowel cleansing. Eligible pMMR/MSS patients were selected based on the pathology report, and samples were collected 7 days after the colonoscopy. No antibiotics or medications were taken close to the sampling time. All fresh fecal samples were placed in sterile containers and frozen at −80°C. Genomic DNA from feces was extracted using the Tiangen Magnetic Soil and Stool DNA kit (DP712). The amount of extracted DNA was measured using the Qubit® dsDNA Assay Kit in the Qubit® 2.0 Fluorometer (Life Technologies, CA, USA), and its purity and integrity were assessed using 1% agarose gel electrophoresis. To build the library, 1 μg of genomic DNA was extracted from the sample. Then, 350 bp pieces of DNA were obtained by ultrasonically fragmenting the material using an E220 Focused Ultrasonicator (Covaris, UK). The NEBNext® Ultra DNA Library Prep Kit for Illumina was used to create individual libraries (NEB, USA). A single-strand circular DNA library was produced by purification, end repair, A-tailing, and PCR amplification of the products. Following the library construction, Qubit2.0 was used for the first quantification, and the library was diluted to 2 ng/µl. Metagenomic shotgun sequencing and taxonomic profiling The pooled library was sequenced by the Illumina NovaSeq 6000 platform according to the manufacturer’s protocol on 150 bp paired-end reads (Novo Gene, China). To obtain high-quality clean reads, raw data were filtered with Readfq (V8). We then used Bowtie2 (version 2.2.4) to map the filtered reads to the human genome (hg19) to remove host contaminants.15 For assembly analysis of clean data, MEGAHIT software (v1.0.4-beta) was utilized. To perform ORF prediction and eliminate redundant genes, we used MetaGeneMark (V2.10) and CD-HIT (V4.5.8).16 Then, we used Bowtie2 to compare the clean data of each sample to the initial gene catalog, calculate the number of reads on each sample, and filter out the genes with reads ≤2 in each sample to obtain the unigenes.17 The unigenes were loaded into DIAMOND software (V0.9.9.110) for taxonomy profiling based on the NR database.18 The LCA algorithm is used to identify the species annotation information of the sequence. Functional annotation of metagenomes Using the DIAMOND software, the metagenomic DNA sequences were mapped to the UniRef90 database.18 The abundance of KEGG Orthologous (KO) groups was then measured against the Kyoto Encyclopedia of Genes and Genomes (KEGG) database using the HUMAnN3 tiered search strategy.19,20 Untargeted metabolomics analyses Thirty-two (Non-AE group: n = 11, AE group: n = 21) samples were tested for fecal metabolomes for each individual using liquid chromatography-mass spectrometry (LC-MS). For each sample, a precooled 80% methanol buffer was added. The mixture was then incubated on ice for 5 min and centrifuged at 15,000 g for 20 min at 4°C. Some of the supernatant was diluted to a final concentration containing 53% methanol by LC-MS grade water. Samples were subsequently transferred to a fresh Eppendorf tube and then were centrifuged at 15,000 g, 4°C for 20 min. Finally, the supernatant was injected into the LC-MS system analysis. Samples were injected onto a Hypersil Gold column (100 × 2.1 mm, 1.9 μm) using a 12-min linear gradient at a flow rate of 0.2 mL/min. The eluents for the positive and negative polarity modes were eluent A (0.1% FA in Water) and eluent B (Methanol). The solvent gradient was set as follows: 2% B, 1.5 min; 2–85% B, 3 min; 85–100% B, 10 min; 100–2% B, 10.1 min; 2% B, 12 min. The MS parameters were as follows: spray voltage 3500 V, capillary temperature 320°C, sheath gas flow rate 35 psi, aux gas flow rate 10 L/min, S-lens RF level 60, Aux gas heater temperature 350°C. In addition, supervised partial least squares discriminant analysis (PLS-DA) was conducted through metaX to discriminate the different variables between groups. The metabolites with P-value <.05 and fold change ≥2 were considered to be differential metabolites. The functions of these metabolites and metabolic pathways were studied using the KEGG database. Markers for predicting adverse events based on microbiome and metabolites To identify markers that distinguish patients from AE and Non-AE, classification models were performed based on differential microbiome, differential metabolites, and associated microbiome-metabolites using random-forest models. The models were validated by leave-one-out cross-validation testing. In each test, the accuracy of the model was examined by receiver operating characteristics. We discarded the variables that were expressed in less than 50% of the patients based on the relative abundances of species. Firstly, model was constructed using the 10 differential floras. Then, 17 metabolites were used to build independent diagnosis model. Finally, microbiome and metabolites with strong relevance were put into the model. All the features were computed the “feature importance” (Mean Decrease Gini). We used Python software (version 3.11.7) and scikit-learn (version 1.4.2) package to perform all analyses. Statistical analysis The alpha diversity was computed to evaluate the richness and evenness of each sample and then compared with the Wilcoxon test. Principal coordinate analysis (PCoA) was performed based on pairwise Bray ‒ Curtis dissimilarity to compare the distance based on R “vegan” package.21,22 Significantly differential taxa between groups were identified with the Wilcoxon test and are shown by a heatmap and linear discriminant analysis (LDA) effect size (LEfSe). Stamp was further applied to identify significantly differentially enriched species based on absolute abundance. Network analysis was used to explore the interplay between species and visualized by Gephi. Normally distributed numerical variables were described by the mean ± standard deviation (X ± s), and a two-tailed Student’s t-test was used to compare the differences between groups. Nonnormally distributed data were described by quartiles, and the Mann-Whitney U-test was used to analyze the intergroup differences. Categorical variables were described as percentages, and Fisher’s test was used for the comparison. Mantel’s test was used to analyze the correlation between differential strains and phylum levels identified in three ways with routine blood, biochemical, and cancer tumor markers. Pearson correlation analysis was used to analyze the associations between microbiome and metabolites. All results were two tailed and were considered statistically significant if the p value was less than 0.05. All statistical analyses were conducted using R software V.4.3.2 and the statistical package for the social sciences (SPSS, version 26, USA). Results No significant difference in clinical characteristics Our study included 32 locally advanced rectal cancer patients (15 females, 17 males) who underwent neoadjuvant chemoradiotherapy (nCRT) combined with immunotherapy. According to the Common Terminology Criteria for Adverse Events (version 4.0), 32 patients were classified as having no adverse events (n = 11) or adverse events (n = 21). There were no statistically significant differences between the Non-AE group and AE group according to demographic/clinical information, such as age, sex, BMI, and each inspection index (Table 1), which excludes the effect of confounding variables on group discrimination before the research. Based on our statistics, 32 patients are currently alive and relapse-free, with a median follow-up period of 940 days. We collected a total of 76 samples at three key time points: at baseline, before the third 21-day cycle of neoadjuvant immunotherapy, and before surgery. A total of 32 baseline fecal samples were collected after colonoscopy and before neoadjuvant therapy. The composition of the gut microbiota was evaluated and predicted by metagenomic sequencing and metabolomics. We were able to obtain annotated genes that satisfied the criteria for subsequent species annotation using Metagenemark and CD-HIT (Figure S1 a and b). We discovered that the correlation coefficient indicated a reasonably high degree of similarity on the sample species abundance patterns, which allowed us to confirm the test’s reliability and the sample’s rationality (Figure S1 c). Differences in Alistipes sp., Weissella confusa and so on between groups at species level We measured the bacterial abundance at the phylum level to better understand the differences in fecal microbiota between groups (Figure 2a). Firmicutes and Bacteroidetes occupied the major positions, accounting for an average of 40.99% and 14.28% of the relative abundance of all groups, respectively. Fluctuations in microbial composition across different groups at genus level were observed (Figure 2b). However, a highly diverse gut microbiome has been associated with less toxicity and better clinical response based on indirect or direct regulation.23–27 For example, direct microbial metabolism mediated chemoresistance (cytidine deaminase degradation of gemcitabine) in pancreatic cancer, while another study demonstrated that Fusobacterium nucleatum targeted TLR4 and MYD88 innate immune signaling and specific microRNAs to activate the autophagy pathway and alter colorectal cancer therapeutic response.28,29 However, we found no significant difference in alpha diversity between groups in our study (Figure 2c, Shannon’s p = .97, reverse Simpson’s p = .73). Beta diversity between groups was also the same (PCoA’s p = .95) (Figure 2d), indicating that there might be individual bacterial taxa related to treatment-related adverse events rather than an overall community shift.30 The abundance at different phylogenetic levels was compared between groups through a linear discriminant analysis (LDA) effect size (LEfSe) and an analysis of variance based on the STAMP. Thirty items of seven levels (kingdom, phylum, class, order, family, genus, and species) were observed (Table S1). Based on the top three outstanding performances, the abundances of Megasphaera micronuciformis, Streptococcus gallolyticus, and Lactobacillus rogosae were found to be higher in the Non-AE group (Figure 3a,b). Additionally, Alistipes sp. and Clostridium sp. were enriched in the Non-AE group. By analyzing the relative abundance of species, 17 differential species were identified (Figure 3c). Paraliobacillus ryukyuensis, Veillonella sp., and Rothia sp. were the top three species among them (Table S2). Weissella cibaria, Weissella confusa, and Weissella paramesenteroides affiliated with Firmicutes, were significantly higher in the AE group (Figure 3d). Spearman’s correlation analysis was used to explore the relationship between differential strains and clinical indices (Figure 4a). Mantel test was also used to deeply analyze the correlation between them. Weissella confusa exhibited the strongest correlation with lymphocyte (LYM) and aspartate aminotransferase (AST) (p < .05) (Table S3). Proteobacteria showed the strongest correlation with LYM (p < .05) (Figure 4b, Table S4). The divergence in the microbial community was also explored, and a slightly stronger correlation network was observed in the Non-AE group compared with the AE group (Figure S2). Bacteroides coprocola, Bacteroides plebeius, and Bacteroides dorei, which were identified as potential driver factors in adverse events based on Netshift analysis (Figure S3). Figure 2.Distribution and phylogenetic diversity of the gut microbiome. (a) The relative abundances of baseline fecal microbes at the phylum level across cancer samples of patients without adverse events (non-AE group) and patients with adverse events (AE group) based on annotated species. The 12 most abundant phyla of each group are shown above. (b) Differences in genus composition between AE and Non-AE groups. (c) Diversity of the gut microbiota across different groups. Alpha diversity was calculated by the Shannon and inverse Simpson indices. (d) Principal coordinates analysis (PCoA) of all baseline fecal samples from locally advanced rectal cancer patients (LARC) using Bray-Curtis distance. Figure 3.Gut microbiome associated with treatment-related adverse events. (a) Linear discriminant analysis (LDA) scores for different groups. The LEfSe classification results and classification chart show the differences in fecal taxa. The size of the charts is proportional to the abundance of the taxon. LDA scores of the different taxa in the stool microbiome of AE group (red) and non-AE group (blue); the length indicates the size of the effect related to the taxonomy, LDA score > 2 is significant. (b) Taxonomic cladogram from LEfSe showed different taxa enriched in the AE group and non-AE group (LDA >2, p < .05). (AE: patients with adverse events; non-AE: patients without adverse events; LEfSe: liner discriminant analysis effect size; LDA: linear discriminant analysis). (c) Differential species in AE and Non-AE groups screened by Wilcoxon rank sum test based on the relative abundance. (d) Differential abundant taxa in baseline fecal samples between AE group and non-AE group based on STAMP according to the absolute abundance of species. Figure 4.Correlations between bacterial genera, species and clinical indices. (a) The layouts of bipartite circular and circle show the relationships among five differential species, routine blood tests, biochemical tests and cancer tumor markers based on Spearman’s correlation analysis. Blue and orange edges indicate negative and positive relationships, respectively. The size of the node is proportional to the degree. Yellow labels represent results of routine blood tests, purple and pink stand for biochemical tests and cancer tumor markers, separately. (b) Correlations between microbial diversity and various inspection indicators. Pairwise comparisons of inspection indicators are shown in the triangle, with a color gradient denoting Spearman’s correlation coefficient. p < .05, p < .01, and p < .001 are denoted as *, **, and ***, respectively. The correlation between five screened differential strains and each clinical inspection indicators is displayed in the lower left corner, while correlation between the seven differential phyla and factors is shown in the upper right corner, which are determined by Mantel’s correlation tests. Edge width corresponds to the Mantel’s r statistic for the corresponding distance correlations, and edge color indicates the statistical significance. Distribution and phylogenetic diversity of the gut microbiome. (a) The relative abundances of baseline fecal microbes at the phylum level across cancer samples of patients without adverse events (non-AE group) and patients with adverse events (AE group) based on annotated species. The 12 most abundant phyla of each group are shown above. (b) Differences in genus composition between AE and Non-AE groups. (c) Diversity of the gut microbiota across different groups. Alpha diversity was calculated by the Shannon and inverse Simpson indices. (d) Principal coordinates analysis (PCoA) of all baseline fecal samples from locally advanced rectal cancer patients (LARC) using Bray-Curtis distance. Gut microbiome associated with treatment-related adverse events. (a) Linear discriminant analysis (LDA) scores for different groups. The LEfSe classification results and classification chart show the differences in fecal taxa. The size of the charts is proportional to the abundance of the taxon. LDA scores of the different taxa in the stool microbiome of AE group (red) and non-AE group (blue); the length indicates the size of the effect related to the taxonomy, LDA score > 2 is significant. (b) Taxonomic cladogram from LEfSe showed different taxa enriched in the AE group and non-AE group (LDA >2, p < .05). (AE: patients with adverse events; non-AE: patients without adverse events; LEfSe: liner discriminant analysis effect size; LDA: linear discriminant analysis). (c) Differential species in AE and Non-AE groups screened by Wilcoxon rank sum test based on the relative abundance. (d) Differential abundant taxa in baseline fecal samples between AE group and non-AE group based on STAMP according to the absolute abundance of species. Correlations between bacterial genera, species and clinical indices. (a) The layouts of bipartite circular and circle show the relationships among five differential species, routine blood tests, biochemical tests and cancer tumor markers based on Spearman’s correlation analysis. Blue and orange edges indicate negative and positive relationships, respectively. The size of the node is proportional to the degree. Yellow labels represent results of routine blood tests, purple and pink stand for biochemical tests and cancer tumor markers, separately. (b) Correlations between microbial diversity and various inspection indicators. Pairwise comparisons of inspection indicators are shown in the triangle, with a color gradient denoting Spearman’s correlation coefficient. p < .05, p < .01, and p < .001 are denoted as *, **, and ***, respectively. The correlation between five screened differential strains and each clinical inspection indicators is displayed in the lower left corner, while correlation between the seven differential phyla and factors is shown in the upper right corner, which are determined by Mantel’s correlation tests. Edge width corresponds to the Mantel’s r statistic for the corresponding distance correlations, and edge color indicates the statistical significance. Metabolic pathway distinctions based on differential orthologous genes HUMAnN3 was used to quantitatively identify the metabolic and biological pathways in our metagenomic shotgun sequencing. We found a total of 60 downregulated orthologous genes and 22 upregulated orthologous genes in the AE group (Table S5). Then, 11 differential microbial pathways were observed on the basis of the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (Table S6). Pantothenate and CoA biosynthesis, beta-alanine metabolism, and amino sugar and nucleotide sugar metabolism were the top three metabolic differential pathways (Figure 5a,b). Figure 5.Difference KEGG functions between AE and Non-AE groups. (a) The classification of the KEGG functions is shown. (b) Representative pathways. PANK: pantothenate Kinase; PPCS: phosphopantothenoylcysteine Synthetase; PPCDC: phosphopantothenoylcysteine Decarboxylase; PPAT: phosphopantetheine adenylyl transferase; DPCK: dephosphocoenzyme a Kinase; CoA: coenzyme A; Cys: cysteine. Difference KEGG functions between AE and Non-AE groups. (a) The classification of the KEGG functions is shown. (b) Representative pathways. PANK: pantothenate Kinase; PPCS: phosphopantothenoylcysteine Synthetase; PPCDC: phosphopantothenoylcysteine Decarboxylase; PPAT: phosphopantetheine adenylyl transferase; DPCK: dephosphocoenzyme a Kinase; CoA: coenzyme A; Cys: cysteine. Boc-beta-cyano-L-alanine, linoleic acid and so on between groups through fecal metabolomic analysis An untargeted LC-MS method was performed to detect 2168 metabolites, including 1356 positive and 812 negative ions of metabolites. Additionally, the metabonomic distributions between groups were distinguished through PLS-DA analysis (Figure 6a). As shown in Figure 6b, we found 28 up-regulated and 30 down-regulated metabolites in the Non-AE group. To more clearly distinguish the two groups, we pulled out differential metabolites at p < .01, as shown in Figure 6c. There were 17 differential metabolites, such as Boc-beta-cyano-L-alanine, CoQ9, and Linoleic acid. Pyrimidine metabolism, tyrosine metabolism, and beta-alanine metabolism were differential pathways as shown in Figure S4. Figure 6.Differential metabolites and complex correlations. (a) Supervised partial least squares discriminant analysis showing discrimination of the Non-AE and AE groups via the first two principal components. (b) Volcano plot revealing the differentially accumulated and significantly altered metabolites between two groups. (c) Heatmap of differential metabolites between Non-AE group and AE group (p < .01, Log2FC >1/Log2FC < −1). (d) A clustered heatmap of the correlations between the abundances of gut microbiota components and differential metabolites. * shows p < .05, ** shows p < .01. Differential metabolites and complex correlations. (a) Supervised partial least squares discriminant analysis showing discrimination of the Non-AE and AE groups via the first two principal components. (b) Volcano plot revealing the differentially accumulated and significantly altered metabolites between two groups. (c) Heatmap of differential metabolites between Non-AE group and AE group (p < .01, Log2FC >1/Log2FC < −1). (d) A clustered heatmap of the correlations between the abundances of gut microbiota components and differential metabolites. * shows p < .05, ** shows p < .01. Alistipes sp. was correlated with Boc-beta-cyano-L-alanine and linoleic acid We evaluated the correlations between 17 selected differential floras and 17 metabolites (Figure 6d and Table S7). As for correlations between metabolites and floras, we found that Alistipes sp. was positively correlated with Boc-beta-cyano-L-alanine and negatively correlated with linoleic acid. Weissella confusa was positively correlated with DL-Norvaline. Excellent predictive efficacy of combined model To identify the adverse events before neoadjuvant therapy, unsupervised random forest classification analysis using the leave-one-out cross-validation method was performed on differential microbiome, differential metabolites, and combined microbiome-metabolites with interplay. At the microbiome and metabolite levels, the random forest classifiers achieved area under the curve (AUC) of 0.872 and 0.926, respectively, while the combined microbiome-metabolite classifier achieved an area of 0.963 (Figure 7a). The random forest classifier also identified the top features of AE and Non-AE patients. For the microbiome classifier, Lactobacillus rogosae, Alistipes sp., and Streptococcus gallolyticus were the top three characteristic features. For the metabolite classifier, linoleic acid, DL-Norvaline, and CoQ9 were the top three characteristic metabolites. For the combined microbiome-metabolite classifier, Alistipes sp. and its associated L-alanine, linoleic acid belonged to the important features (Figure 7b). Figure 7.The diagnostic potential of microbiome and metabolites for adverse events during neoadjuvant therapy. (a) ROC curves show the discriminative ability between non-AE (n = 11) and AE (n = 21) patients using features of differential gut microbiome, differential metabolites and combined microbiome-metabolites. (b) The top eight important features based on Gini importance according to random-forest classifiers based on the microbiome, metabolites and combined microbiome-metabolites. ROC: receiver operating characteristic. The diagnostic potential of microbiome and metabolites for adverse events during neoadjuvant therapy. (a) ROC curves show the discriminative ability between non-AE (n = 11) and AE (n = 21) patients using features of differential gut microbiome, differential metabolites and combined microbiome-metabolites. (b) The top eight important features based on Gini importance according to random-forest classifiers based on the microbiome, metabolites and combined microbiome-metabolites. ROC: receiver operating characteristic. Discussion Although neoadjuvant chemoradiotherapy (CRT) combined with immunotherapy brought higher response rate than monotherapy, some patients experienced adverse events of 1–2 grade, which may impact the efficacy of treatment. Studies have shown that gut microbes could predict the efficacy of immunotherapy or chemoradiotherapy.31 However, few studies were implemented to explore the relationship between gut microbiome and adverse events from neoadjuvant chemoradiotherapy combined with immunotherapy. Hence, we collected 32 baseline fecal samples from LARC patients in our center. Significant differences in the gut microbiome were observed between patients with adverse events and patients without adverse events. In addition, metabolic pathways were enriched based on the differential microbiome. We identified discriminatory metabolites and associated them with floras to explore the possible mechanisms. Finally, we applied random forest machine to construct a predictive model based on the intestinal microbiome and metabolite characteristics to evaluate adverse events of patients to combined therapy. Our results could assist clinical physicians in choosing individualized management for LARC patients in clinical practice. Both intestinal microbiomes and metabolites have recently been identified as important biomarkers for efficacy and toxicity. Epidemiologically, cancer associated with microbiota could account for approximately 15% of cases worldwide.32 In our research, diversity of microorganisms showed no significant difference between AE and Non-AE groups, suggesting that specific microbiota might play a more important role.22,33,34 Five differential strains were identified by differential analysis, including Weissella cibaria, Weissella confusa, Lachnoclostridium sp., Alistipes sp., and Clostridium sp. Specifically, m3 gene expression in Lachnoclostridium_sp. could be used as a biomarker for the noninvasive diagnosis of colorectal cancer, suggesting an important association between Lachnoclostridium_sp. and colorectal cancer progression.35 Similarly, the abundance of Lachnoclostridium_sp. in our study was significantly higher in the AE group. Weissella species are gram-positive, asporogenous, and non-motile bacteria, possessing facultatively anaerobic chemoorganotrophs with an obligately fermentative metabolism. Recent research has indicated that individuals did not get infected by Weissella in a healthy state, while cancer, chemotherapy, or radiotherapy may cause probable immunosuppression leading to opportunistic infections.36 Therefore, we deduced that the immunity of our patients experiencing chemoradiotherapy combined with immunotherapy decreased, leading to a higher risk of therapeutic adverse events. We also found that pathways enriched for differentially orthologous groups and for metabolites intersected, namely the beta-alanine metabolism. Therefore, we supposed that it might be a critical pathway between groups. Alistipes sp. was negatively correlated with inflammatory factors.37,38 Our study found that Alistipes sp. was positively associated with Boc-beta-cyano-L-alanine. Amino acids could produce anti-inflammatory, antioxidant effects, and attenuated intestinal barrier disruption.39 For example, alanine can be transferred into pantothenic acid (Figure 5b), which is a key precursor of coenzyme A (CoA) synthesis and is considered a vascular protector that increases ATP content and reduces the major product of lipid peroxidation.40–43 Therefore, we deduced that Alistipes sp. may facilitate beta-alanine metabolism, increase the synthesis of pantothenic acid, and reduce the occurrence of adverse events. Additionally, Alistipes sp. was negatively correlated with linoleic acid. It is a major constituent of low-density lipoproteins, which was reported to cause more oxidative damage than other free fatty acids and mediate loss of CD4+T lymphocytes, impairing anti-tumor surveillance.44,45 Collectively, the interactions between Alistipes sp., metabolites, and their impact on adverse events are worth studying in the future, which may also guide cancer treatments. In addition, we built and compared the microbiome and metabolite as prognostic biomarkers for adverse events using a random forest model. The results of our study indicated that the metabolite model showed slightly better performance as a predictor than the gut microbiome. While a combined microbiome-metabolite classifier showed better performance for adverse events than that of the microbiome or metabolite, predicting the toxicity of neoadjuvant chemoradiotherapy combined with immunotherapy and further helping clinicians to screen the most appropriate treatment for patients. In short, the AE group was more clearly separated from the Non-AE group at higher levels of classification based on the metagenomic shotgun sequencing technique.46 We identified differential metabolites that might be associated with the occurrence of treatment adverse events and built the first prediction for adverse events of combined therapy. However, there are several limitations in our study. Firstly, given the heterogeneity of human lifestyles and the variation in gut microbiota around the world, the global applicability of our study on the prediction of adverse events needs to be established in larger cohorts. Meanwhile, our center is initiating a multicenter, prospective, randomized controlled study (NCT06312982) to evaluate the effectiveness and safety of neoadjuvant chemoradiotherapy combined with PD-1 blockade (tislelizumab) for treating pMMR/MSS locally advanced rectal cancer (LARC). A total of 375 LARC patients whose distance from the distal border of the tumor to the anal verge was ≤10 cm will be consecutively enrolled. Secondly, we have to acknowledge that different cancer types and modalities of tumor treatment all lead to diversity of the microbiota. For this reason, an independent validation cohort is needed, preferably with the same cancer type.21 Further experiments are essential for validating the model and ensuring that the interactions between bacteria and metabolites we observed are not merely coincidental but reflect true biological mechanisms. Based on our follow-up data, we have not observed any cases of relapse, so we are unable to establish a link between the microbiome and the risk of recurrence. However, existing literature suggests a potential relationship between the microbiome and relapse, and we plan to explore this further in future studies. Additionally, bacteria-based therapy is considered to be a promising solution for malignant tumors because engineered bacteria can improve the therapeutic efficacy of tumor treatment with the rapid development of bioengineering technology and nanotechnology.47 Conclusions Significant distinctions were identified in the composition of the gut microbiome and its metabolites between groups. Moreover, Alistipes sp. and alanine metabolism may alleviate the occurrence of adverse events. Additionally, a microbiome-metabolite model that can noninvasively predict adverse events prior to treatment was built, which allowed us to identify patients at high risk, providing a basis for subsequent individualized treatment regimen. Supplementary Material Supplemental Material.docx