Microbiome and metabolome patterns centered on cancer cachexia link skeletal muscle and adipose tissue depletion to clinical outcomes in locally advanced rectal cancer Authors: Zhixing Kuang, Qing Zhan, Jiannan Tu, Anchuan Li, Wei Guo, Rong Zheng, Hao Jiang, Benhua Xu Published: 2026 ABSTRACT AbstractBackgroundCancer cachexia is a multifactorial syndrome characterized by involuntary weight loss and tissue wasting; it was associated with adverse survival outcomes in patients with locally advanced rectal cancer (LARC). However, its biological basis and links to the gut microbiome, circulating metabolites, and body composition in LARC remain unclear.MethodsWe conducted a retrospective cohort study of 198 LARC patients, including 30 with cachexia. Gut microbial profiles were characterized using 16S rRNA sequencing with predictive functional inference. Serum metabolites were analyzed via orthogonal partial least squares–discriminant analysis (OPLS-DA) and validated by ROC analysis. Associations among microbial taxa, metabolites, body composition, and clinical indices were assessed using correlation analyses. Survival outcomes were evaluated by Kaplan-Meier curves and multivariate Cox regression models, with subgroup and sensitivity analyses performed to validate the robustness of findings.ResultsCachectic patients exhibited marked reductions in BMI, circulating triglycerides and cholesterol, as well as visceral and subcutaneous adipose tissue (VAT and SAT) and skeletal muscle area (SMA) (all P < 0.05). Microbial α- and β-diversity were comparable between groups, yet cachexia was associated with selective enrichment of Peptostreptococcaceae, Erysipelotrichaceae, Veillonella, and Fusobacterium, alongside depletion of Faecalibacterium, Roseburia, and Prevotella. Notably, UBA1819 (Ruminococcaceae) was enriched in cachectic patients and inversely correlated with Faecalibacterium, hemoglobin, and albumin, suggesting intra-family functional divergence. Specific cachexia-enriched taxa, including Veillonella and Erysipelotrichaceae, were inversely associated with SMA and VAT, whereas butyrate-producing genera showed positive correlations with muscle preservation. Functional predictions highlighted enrichment of N-acetylneuraminate degradation in cachexia versus TCA cycle pathways in non-cachectic patients. High abundance of Peptostreptococcaceae and Erysipelotrichaceae was associated with poorer overall survival. Metabolomic profiling identified lithocholic acid, cortisone, 3,5,9-trihydroxyergost-7-en-6-one, and 5,6-DHET as cachexia-associated metabolites (AUC 0.65–0.78), which formed cross-domain correlations with microbial taxa and host tissue measures.ConclusionOur findings reveal a coordinated linkage between the gut microbiome, metabolome, and host body composition in LARC cachexia. Specifically, selective microbial remodeling and metabolite perturbations are closely coupled with skeletal muscle and adipose tissue depletion and adverse survival. These results provide a conceptual foundation for future mechanistic and translational studies exploring the host–microbiome–metabolic interactions in cancer cachexia.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12866-026-05171-9.Keywords: Cancer cachexia, Locally advanced rectal cancer, Gut microbiota, Metabolomics, Skeletal muscle, Adipose tissue Introduction Cancer cachexia is a debilitating wasting syndrome that contributes substantially to cancer-related mortality. It is characterized by involuntary weight loss, systemic inflammation, and metabolic dysfunction arising from dysregulated host–tumor crosstalk, ultimately impairing treatment tolerance and clinical outcomes [1, 2]. Although extensive research has focused on terminal and metastatic disease, metabolic dysregulation in non-metastatic, potentially curable settings, particularly in LARC, remains insufficiently explored [3]. This gap is clinically relevant, as the window of tumor resectability and standard chemoradiotherapy in LARC offers a unique opportunity to investigate early cachexia-associated metabolic alterations before irreversible systemic deterioration ensues. Building upon our prior whole-body [¹⁸F]FDG PET/CT study, we identified a cachexia-associated metabolic phenotype in LARC, characterized by increased glucose uptake in adipose tissue and skeletal muscle despite relatively lower uptake in major visceral organs [4]. These compartment-specific metabolic alterations persisted and became more pronounced after normalization to the aortic blood pool, suggesting that peripheral tissue wasting in cancer cachexia may reflect regulated metabolic reprogramming rather than passive energy deprivation alone. However, whether this hypermetabolic phenotype is quantitatively linked to computed tomography–defined loss of skeletal muscle and adipose tissue, and which upstream biological factors contribute to its development, remain unclear. Emerging evidence indicates that the gut microbiome and circulating metabolome play critical roles in regulating host energy homeostasis and inflammatory signaling, and may therefore contribute to cancer-associated metabolic imbalance and tissue wasting [5–7]. We hypothesized that alterations in the gut microbiota and plasma metabolite profiles are associated with the cachexia-related metabolic phenotype in LARC and are linked to changes in body composition and clinical outcomes. In this study, we integrated gut microbiome profiling, plasma metabolomics, quantitative CT-based body composition analysis, and longitudinal clinical follow-up in patients with LARC. Our objectives were to characterize cachexia-associated microbiome and metabolome alterations, to examine their relationships with tissue-specific metabolic activity and structural depletion of skeletal muscle and adipose tissue, and to evaluate their associations with overall survival. By positioning cancer cachexia as a unifying phenotype connecting metabolic reprogramming, body composition loss, and prognosis, this study aims to provide an integrated framework for understanding cachexia in potentially curable rectal cancer. Materials and methods Study design and participants This retrospective analysis included patients with locally advanced rectal adenocarcinoma (cStage II-III) who underwent neoadjuvant therapy followed by surgical resection at Fujian Medical University Union Hospital between July 2018 and December 2021. Patients were identified through the institutional electronic medical record system. The study protocol was approved by the hospital’s Institutional Review Board. All patients had undergone standardized pretreatment staging, comprising digital rectal examination, biopsy-confirmed colonoscopy, pelvic magnetic resonance imaging (MRI) for T (primary tumor) and N (regional lymph node) staging, and chest/abdominal computed tomography (CT) or MRI to exclude metastatic disease. Following neoadjuvant treatment, response assessment was performed via pelvic MRI. Patients who achieved a clinical complete response (cCR) or near-cCR were offered organ-preserving local excision to pathologically confirm the response and potentially avoid radical resection. Those with suboptimal response proceeded directly to total mesorectal excision (TME). Cancer cachexia was defined according to the international consensus criteria proposed by Fearon et al. (2011) [8] and the Asian Working Group for Sarcopenia (AWGS, 2019) [9], which require: (1) weight loss > 5% over the past 6 months, or (2) weight loss > 2% in individuals with BMI < 18.5 kg/m². To ensure reproducibility, two independent physicians reviewed all weight records and cachexia classifications. Inclusion and exclusion criteria Eligible patients met the following criteria: (1) histological diagnosis of rectal adenocarcinoma; (2) clinical stage II–III (cT3-4N0 or cTanyN1-2) with the tumor located within 10 cm from the anal verge; (3) completion of neoadjuvant chemoradiotherapy and subsequent surgery; (4) Eastern Cooperative Oncology Group (ECOG) performance status 0–1 or Karnofsky score ≥ 70; (5) availability of complete clinical, radiological, pathological, and follow-up records. Key exclusion criteria were: (1) stage I or IV (M1) disease at diagnosis; (2) history of synchronous or metachronous colorectal malignancies; (3) presence of other active cancers within the past 5 years; (4) pre-existing metabolic or endocrine disorders, including diabetes mellitus or hyperthyroidism; (5) significant comorbidities (e.g., severe cardiac, pulmonary, or neurological diseases) that could substantially affect treatment delivery or outcomes. Neoadjuvant therapy and surgical intervention Radiotherapy was planned using intensity-modulated radiotherapy (IMRT) or volumetric modulated arc therapy (VMAT). Target volumes were defined as follows: the gross tumor volume (GTV) included primary tumor and involved lymph nodes based on imaging and clinical examination; the clinical target volume (CTV) encompassed the GTV with a 2-cm craniocaudal margin along the mesorectum, plus elective nodal regions (mesorectal, presacral, internal iliac, and obturator). Planning target volumes (PTV-GTV and PTV-CTV) were created by adding a 10-mm isotropic margin to the GTV and CTV, respectively. Patients received either long-course radiotherapy (LCRT: 50 Gy in 25 fractions to PTV-GTV and 45 Gy to PTV-CTV) or short-course radiotherapy (SCRT: 25 Gy in 5 fractions to PTV-CTV). For those undergoing LCRT, concurrent chemotherapy consisted of two cycles of XELOX (capecitabine 1000 mg/m² twice daily on days 1–14, oxaliplatin 130 mg/m² on day 1, every 3 weeks). This was followed by consolidation chemotherapy with additional XELOX cycles, the number of which was individualized based on multidisciplinary team assessment and patient tolerance. Surgical resection was performed 4–8 weeks after completion of neoadjuvant chemotherapy phase. Postoperative adjuvant chemotherapy with XELOX or FOLFOX was administered to selected patients based on pathological findings such as lymphovascular invasion or persistent nodal involvement. Sample collection and cohort overview Clinical and pathological data were retrieved from the electronic medical record system. Variables collected included: demographic characteristics (sex, age), comorbidities (hypertension, coronary heart disease), personal history (smoking, alcohol consumption), pretreatment laboratory indicators (carcinoembryonic antigen(CEA), carbohydrate antigen 19 − 9(CA-199), neutrophil count, lymphocyte count, platelet count, red blood cell count, hemoglobin, albumin, aspartate aminotransferase, alanine aminotransferase, total bilirubin, triglycerides, total cholesterol, creatinine, urea nitrogen), baseline tumor characteristics (clinical T/N stage, distance from the anal verge, mesorectal fascia status, extramural vascular invasion, lymphovascular invasion, perineural invasion), treatment parameters (radiotherapy type, number of chemotherapy cycles, sphincter preservation), and pathological outcomes (tumor regression grade, pathological complete response, ypTNM stage). The primary endpoint was overall survival (OS), defined as the time from treatment initiation to death from any cause or the last follow-up. Follow-up was conducted every 3 months for the initial 2 years, every 6 months up to year 5, and annually thereafter. Assessments included clinical examination, serum tumor markers (CEA, CA19-9), pelvic MRI, and chest/abdominal CT. Patient survival status was actively verified through direct telephone contact and cross-referenced with the hospital’s electronic medical record system. The data cutoff for the final follow-up analysis was June 19, 2025. Fecal samples were collected from 155 patients, of which 128 met inclusion criteria for microbiome profiling. Blood samples for untargeted metabolomics were collected from 154 patients, with 134 retained after exclusions. Among these, 64 patients had paired fecal and plasma samples for integrative microbiome–metabolome analyses, while the remaining samples were analyzed independently (detailed flowchart in Supplementary Figure S1). Body composition was assessed using contrast-enhanced radiotherapy planning CT images acquired under standardized protocols prior to the initiation of any neoadjuvant treatment. All CT scans were performed within one week of hospital admission. To ensure that all baseline measurements accurately reflect the pre-treatment biological status and to minimize confounding effects from hospital-associated dietary shifts, fecal and blood samples were strictly collected on the second day after initial hospital admission. This standardized timing ensured that all multi-omics and imaging data were captured before the commencement of neoadjuvant chemotherapy or other anti-tumor interventions. Microbiome sequencing and analysis Microbial total DNA was extracted from approximately 0.5 g of each fecal sample using the TGuide S96 Magnetic Soil/Stool DNA Kit (Tiangen Biotech, Beijing, China) according to the manufacturer’s protocol. Briefly, samples underwent mechanical homogenization with grinding beads and chemical lysis at 70 °C for 15 min to maximize DNA recovery. Automated purification was performed on the TGuide S96 platform utilizing magnetic-bead-based technology. DNA quality and concentration were assessed by 1.8% agarose gel electrophoresis and quantified using the Qubit™ 1X dsDNA HS Assay Kit (Invitrogen, USA). Full-length 16S rRNA gene amplification was performed using commercially synthesized universal primers 27 F (5’-AGRGTTTGATYNTGGCTCAG-3’) and 1492R (5’-TASGGHTACCTTGTTASGACTT-3’) (BIOTREE, Shanghai, China), with unique asymmetric barcode sequences incorporated for sample differentiation. PCR reactions (20 µL) contained 4 µL 5× FastPfu Buffer, 2.5 mM dNTPs, 0.4 µL FastPfu Polymerase, 0.8 µL each primer (5 µM), and 10 ng template DNA. The thermal cycling conditions were: 95 °C for 2 min; 25 cycles of 95 °C for 30 s, 55 °C for 30 s, 72 °C for 1 min; followed by 72 °C for 5 min. PCR products were purified using the AMPure PB Beads (Pacific Biosciences, USA). Library construction was performed using the SMRTbell™ Template Prep Kit 1.0 (Pacific Biosciences, USA), with the SMRTbell Binding Kit (Pacific Biosciences, USA) utilized for polymerase binding. Sequencing was conducted on the PacBio Sequel II platform in Circular Consensus Sequencing (CCS) mode. Bioinformatic processing was performed using QIIME2 (v2023.5). Raw sequencing data were processed using SMRT Link software to generate Circular Consensus Sequences (CCS) with minPasses ≥ 5 and minPredictedAccuracy ≥ 0.9. After barcode demultiplexing via lima and primer trimming via cutadapt, sequences were length-filtered (1200 bp to 1650 bp) to obtain Valid CCS. The DADA2 plugin was employed for denoising, dereplication, and chimera removal to generate the Amplicon Sequence Variant (ASV) feature table, with singletons (ASVs appearing only once in the entire dataset) excluded to ensure robustness. Taxonomic annotation was performed against the SILVA database (Release 138) using a combined approach: classify-sklearn (confidence = 0.7) and classify-consensus-vsearch (identity = 0.9, coverage = 0.9). To ensure comparability across samples, the ASV feature table was rarefied to a consistent sampling depth of 4,195 reads per sample (the minimum depth observed) (Figure S2), after confirming that rarefaction curves had reached saturation. Subsequently, Hellinger transformation was applied to the rarefied table using the decostand(method = “hellinger”) function in the vegan R package (v2.6-10) [10]. This Hellinger-transformed dataset was used for beta diversity analyses, including calculating Bray–Curtis dissimilarities for PCoA and NMDS ordination, as well as ANOSIM testing, to reduce the influence of highly abundant taxa and mitigate double-zero issues in compositional data [11]. Alpha diversity metrics (Chao1, Shannon, Simpson, and Faith’s PD) were calculated directly from the rarefied table, with intergroup comparisons performed using the Wilcoxon rank-sum test (P < 0.05 was considered statistically significant) [12]. Differences in community structure between groups were further tested using PERMANOVA (Adonis) with 999 permutations. Metabolomics analysis Untargeted metabolomics analysis was performed on the samples. A 100 µL aliquot of each sample was mixed with 300 µL of methanol extraction solution containing an isotopically-labeled internal standard mixture. After vortexing for 30 s, the mixture was sonicated for 10 min in an ice-water bath and incubated at -40 °C for 1 h, followed by centrifugation at 12,000 rpm for 15 min at 4 °C. The supernatant was collected for LC-MS/MS analysis. A quality control (QC) sample was prepared by pooling equal volumes of supernatant from all samples. Chromatographic separation was carried out on a Thermo Vanquish UHPLC system equipped with a Waters ACQUITY UPLC HSS T3 column (2.1 mm × 100 mm, 1.8 μm). The mobile phase consisted of (A) water containing 5 mmol/L ammonium acetate and 5 mmol/L acetic acid, and (B) acetonitrile. The column temperature was maintained at 4 °C, and the injection volume was 2 µL. Mass spectrometric detection was performed using an Orbitrap Exploris 120 high-resolution mass spectrometer with an electrospray ionization source. The source parameters were set as follows: sheath gas flow rate, 50 Arb; auxiliary gas flow rate, 15 Arb; capillary temperature, 320 °C. Full-scan MS spectra were acquired at a resolution of 60,000, and data-dependent MS/MS spectra were obtained at a resolution of 15,000 with collision energies of 10, 30, and 60 NCE. The spray voltages were + 3.8 kV in positive mode and − 3.4 kV in negative mode. Raw data were converted to mzXML format using ProteoWizard and processed with an in-house R package based on XCMS for peak detection, extraction, alignment, and integration. Metabolite annotation was performed by matching against an in-house MS/MS database (BiotreeDB V2.1) with a cutoff score of 0.3. The confidence of metabolite annotation was categorized according to the recommended levels described in the Nature Methods best-practice guidelines for mass spectrometry–based metabolomics, ranging from Level A (confirmed by authentic reference standards or NMR) to Level D (putative annotation based on MS data only) [13]. Body composition measurement from CT images Body composition was assessed using contrast-enhanced radiotherapy planning CT images acquired under standardized protocols prior to the initiation of neoadjuvant treatment. At the level of the third lumbar vertebra (L3), a single axial CT slice was selected for tissue segmentation, which was performed by an experienced physician (Observer A) and independently verified by a second experienced physician (Observer B) using ImageJ software (version 1.53; National Institutes of Health, Bethesda, MD, USA). Semi-automated segmentation with predefined Hounsfield Unit (HU) thresholds was applied to delineate subcutaneous adipose tissue (SAT), skeletal muscle area (SMA)—encompassing the rectus abdominis, abdominal wall muscles, psoas, quadratus lumborum, and erector spinae—and visceral adipose tissue (VAT). The threshold ranges were set at -190 to -30 HU for adipose tissue and − 29 to 150 HU for SMA. Following initial segmentation, manual refinement was performed using ITK-SNAP, with all segmentations preserved in NIfTI format (.nii.gz). To assess inter-observer reliability, 40 patients were randomly selected, and Observer B independently re-analyzed these images blinded to the original measurements; intraclass correlation coefficients (ICC) demonstrated excellent reproducibility, with values of 0.998 (95% CI 0.939–0.999) for SAT, 0.998 (95% CI 0.979–0.999) for SMA, and 0.997 (95% CI 0.908–0.999) for VAT. Quality assurance was additionally conducted by an experienced radiologist (JH) who reviewed all segmentations, including visual inspection of HU heatmaps, HU distribution histograms, and direct comparison with source images (Fig. 1A-D). Quantification of L3-level body composition on axial CT in a representative patient with LARC. A Axial CT image at the L3 vertebral level (window width 400, window level 60). B HU heatmap of segmented tissue masks demonstrating the three compartments of interest: SMA, SAT and VAT. Color scale indicates Hounsfield unit (HU) values. C Overlay of the HU heatmap on the original CT image, showing spatial localization of HU ranges within each tissue compartment. D Normalized histograms of HU distributions for SMA, SAT and VAT. For each tissue the mean HU, standard deviation and voxel count (N) are reported in the inset Statistical analysis Continuous variables are presented as median with interquartile range (IQR) and were compared between cachectic and non-cachectic patients using the Wilcoxon rank-sum test, while categorical variables were compared using the χ² test or Fisher’s exact test, as appropriate. Gut microbial α-diversity indices were compared using nonparametric tests, and β-diversity was assessed based on Bray–Curtis dissimilarities and visualized by non-metric multidimensional scaling (NMDS), with group differences evaluated using permutational multivariate analysis of variance (PERMANOVA). Differentially abundant microbial taxa were identified using linear discriminant analysis effect size (LEfSe), with a logarithmic LDA score threshold > 2.5 and P < 0.05. For metabolomic analysis, OPLS-DA was applied to assess global metabolic differences between cachectic and non-cachectic patients. Model robustness was evaluated using R² and Q² statistics, along with permutation testing to exclude overfitting. Metabolites contributing to group discrimination were identified based on variable importance in projection (VIP) scores combined with univariate statistical testing. Receiver operating characteristic (ROC) curve analysis was used to assess the discriminative performance of selected cachexia-associated metabolites. To contextualize the identified metabolomic signatures, pathway mapping was performed using MetaboAnalyst 6.0. Differential metabolites were mapped against the KEGG database to identify enriched metabolic frameworks. For survival analysis, cachexia-associated taxa identified by LEfSe were dichotomized into high- and low-abundance groups based on optimal cut-off values determined by maximally selected rank statistics. Intergroup differences in overall survival were assessed using Kaplan–Meier curves with log-rank tests. To evaluate the stability of the selected cut-off values and reduce the risk of Type I error, internal validation was conducted via bootstrap resampling (1,000 iterations). In each bootstrap sample, the optimal cut-off was recalculated and the survival association re-evaluated; the proportion of significant bootstrap replicates (P < 0.05) and the variability of cut-off values were examined. Multivariable Cox proportional hazards regression was performed to evaluate the independent prognostic value of cachexia-associated microbial taxa. Candidate variables were selected based on univariate analysis (P < 0.05). To address multicollinearity among correlated pathological features, sensitivity analyses were performed to determine the optimal combination of covariates. Correlations between microbial taxa, metabolites, body composition parameters, and clinical indices were evaluated using Spearman’s rank correlation coefficient. All statistical tests were two-sided, and P < 0.05 was considered statistically significant. All analyses were conducted using R software (version 4.2.3). Results Patient characteristics and prevalence of cachexia A total of 198 patients with LARC were included, of whom 30 (15.2%) met the diagnostic criteria for cancer cachexia. The prevalence of cachexia was 14.6% (18/123) in male patients and 16.0% (12/75) in female patients. Baseline demographic and clinical characteristics are summarized in Table 1. The two groups were comparable with respect to age, sex, comorbidities (coronary heart disease and hypertension), lifestyle factors (smoking and alcohol consumption), clinical tumor stage (cT, cN, cM), pretreatment hematologic parameters, liver and renal function indices, as well as surgical and systemic treatment strategies (all P > 0.05). Regarding treatment intensity, surgical and systemic treatment strategies, including the number of neoadjuvant, adjuvant, and total chemotherapy cycles, were well-balanced between the two groups (all P > 0.05; Table S1). As expected, cachectic patients exhibited a significantly lower body mass index (BMI) than non-cachectic patients (20.4 ± 3.1 vs. 22.9 ± 3.0 kg/m², P < 0.001). They also showed higher pretreatment levels of carcinoembryonic antigen (CEA) and carbohydrate antigen 19–9 (CA19-9), together with lower serum triglyceride and cholesterol concentrations (all P < 0.05). Notably, cachexia was associated with a higher prevalence of positive circumferential resection margin (CRM) status (80.0% vs. 51.2%, P = 0.006) and perineural invasion (16.7% vs. 4.8%, P = 0.030) (Table S1). Table 1Basic clinical characteristics of patients with LARC stratified by cachexia statusVariableCachexia (n = 30)Non-cachexia (n = 168)P-valueAge group0.723<60 years17 (56.7%)86 (51.2%)≥60 years13 (43.3%)82 (48.8%)Sex0.956Male18 (60.0%)105 (62.5%)Female12 (40.0%)63 (37.5%)Coronary heart disease1.000No30 (100.0%)167 (99.4%)Yes0 (0.0%)1 (0.6%)Hypertension0.957No25 (83.3%)136 (81.0%)Yes5 (16.7%)32 (19.0%)Smoking0.216No24 (80.0%)112 (66.7%)Yes6 (20.0%)56 (33.3%)Alcohol consumption0.981No24 (80.0%)138 (82.1%)Yes6 (20.0%)30 (17.9%)Body mass index, kg/m²20.4 ± 3.122.9 ± 3.0< 0.001CEA, ng/mL6.0 (3.0, 24.2)3.7 (2.2, 7.6)0.005CA19-9, U/mL18.1 (10.5, 39.9)12.1 (7.9, 22.3)0.033Neutrophils, ×10⁹/L3.5 (2.7, 4.7)3.5 (2.8, 4.6)0.975Lymphocytes, ×10⁹/L1.8 (1.5, 2.3)1.7 (1.3, 2.1)0.193Platelets, ×10⁹/L260.5 (213.2, 302.8)249.5 (212.8, 302.5)0.687RBC, ×10¹²/L4.4 (4.1, 4.7)4.5 (4.2, 4.8)0.152Hemoglobin, g/L126.5 (122.0, 136.0)136.0 (122.8, 144.0)0.087Albumin, g/L42.6 (40.9, 44.1)43.2 (40.8, 45.2)0.324AST, U/L14.0 (10.2, 16.8)14.5 (11.8, 20.0)0.273ALT, U/L17.5 (13.2, 20.0)17.0 (13.0, 20.0)0.938Total bilirubin, µmol/L9.8 (7.5, 12.5)10.1 (7.6, 13.2)0.728Triglycerides, mmol/L1.0 (0.8, 1.3)1.2 (1.0, 1.6)0.031Cholesterol, mmol/L4.2 (3.8, 4.9)4.7 (4.2, 5.3)0.027Creatinine, µmol/L67.0 (58.0, 74.0)69.0 (58.0, 83.0)0.293Urea nitrogen, mmol/L4.5 (3.8, 5.8)4.7 (3.8, 5.8)0.736Distance from anal verge, cm5.2 (3.6, 7.1)5.1 (3.7, 6.9)0.872Clinical stage1.000II1 (3.3%)6 (3.6%)III29 (96.7%)162 (96.4%)cT stage0.365T1-20 (0.0%)10 (6.0%)T3-430 (100.0%)158 (94.0%)cN stage1.000N01 (3.3%)6 (3.6%)N1-229 (96.7%)162 (96.4%)CRM status0.006Negative6 (20.0%)82 (48.8%)Positive24 (80.0%)86 (51.2%)EMVI status0.623Negative11 (36.7%)73 (43.5%)Positive19 (63.3%)95 (56.5%)Data are presented as median (interquartile range, IQR) for continuous variables and number (percentage) for categorical variables. Between-group comparisons were performed using the Wilcoxon rank-sum test for continuous variables, and the Chi-square test or Fisher’s exact test for categorical variables, as appropriateAbbreviations: BMI body mass index, CEA carcinoembryonic antigen, CA19-9 carbohydrate antigen 19 − 9, RBC red blood cell count, AST aspartate aminotransferase, ALT alanine aminotransferase, CRM circumferential resection margin, EMVI extramural venous invasion Basic clinical characteristics of patients with LARC stratified by cachexia status Data are presented as median (interquartile range, IQR) for continuous variables and number (percentage) for categorical variables. Between-group comparisons were performed using the Wilcoxon rank-sum test for continuous variables, and the Chi-square test or Fisher’s exact test for categorical variables, as appropriate Abbreviations: BMI body mass index, CEA carcinoembryonic antigen, CA19-9 carbohydrate antigen 19 − 9, RBC red blood cell count, AST aspartate aminotransferase, ALT alanine aminotransferase, CRM circumferential resection margin, EMVI extramural venous invasion Gut microbiota composition and functional profiles differ between cachectic and non-cachectic patients Gut microbial α-diversity did not differ significantly between cachexia and non-cachexia patients, as assessed by observed features, Chao1 richness, Shannon index, Simpson index, and Faith’s phylogenetic diversity (all P > 0.05) (Fig. 2A). Similarly, β-diversity analysis based on Bray–Curtis dissimilarities, validated by PCoA, NMDS, and ANOSIM, revealed no significant global differences in microbial community structure between groups (all P > 0.05; Figure S3). Analysis at the family level revealed higher abundances of f_Fusobacteriaceae, f_Peptostreptococcaceae, and f_Veillonellaceae in cachectic patients compared with non-cachectic controls (Fig. 2B). At the genus level, g_Peptostreptococcus and g_Fusobacterium were similarly enriched, highlighting group-specific microbial composition patterns (Fig. 2C). Gut microbiota characteristics in cachexia and non-cachexia patients with LARC. A Comparison of gut microbial α-diversity between groups. B–C Relative abundance of gut microbiota at the family (B) and genus (C) levels. D–E Differentially abundant taxa identified by LEfSe analysis, shown as LDA scores (D) and a cladogram (E) Linear discriminant analysis effect size (LEfSe) further identified multiple taxa that were differentially enriched between the two groups (Fig. 2D). Taxa enriched in cachexia patients included members of Peptostreptococcaceae, Veillonella, Erysipelotrichacelae, and Fusobacterium whereas non-cachexia patients showed enrichment of several commensal taxa, including Faecalibacterium, Roseburia, and Prevotella. The phylogenetic relationships of these discriminative taxa were illustrated in the LEfSe cladogram (Fig. 2E). Predicted metagenomic functional profiling revealed significant differences in gut microbial metabolic potential between cachectic and non-cachectic patients. Differential abundance analysis at the pathway level demonstrated distinct enrichment patterns. The microbiome of cachectic patients was significantly enriched in pathways related to amino sugar and sialic acid metabolism, including N-acetylglucosamine and N-acetylneuraminate metabolism and degradation, as well as lysine biosynthesis II. In contrast, the non-cachectic group showed higher abundance of pathways involved in core energy metabolism (TCA cycle), biosynthesis of respiratory cofactors (menaquinol-8), nucleotide biosynthesis (pyrimidine nucleotides), and amino acid anabolism (all P < 0.05) (Fig. 3). Predicted functional differences in gut microbial metabolic pathways between cachexia and non-cachexia groups Cachexia-associated gut microbiota are linked to survival outcomes and body composition in LARC To assess the prognostic impact of the specific bacterial taxa identified by LEfSe, patients were stratified into high- and low-abundance groups based on the relative abundance of each identified taxon, using optimal cut-off values determined by maximally selected rank statistics. Kaplan–Meier analysis demonstrated that high abundance of Peptostreptococcaceae (HR = 3.66, 95% CI 1.16–11.54, P = 0.018) and Erysipelotrichaceae (HR = 4.93, 95% CI 1.48–16.37, P = 0.004) were significantly associated with worse overall survival (OS) (Fig. 4A). Internal validation using 1,000 bootstrap replicates confirmed the robustness of these cut-off values, with significant associations observed in 69.0% and 79.2% of replicates for Peptostreptococcaceae and Erysipelotrichaceae, respectively. Univariate analysis identified multiple factors associated with OS, including CA199, LVI, PNI, TRG, and both microbial taxa (all P < 0.05; Table S2). Given the significant correlation between PNI and LVI (Cramer’s V = 0.37, P < 0.001), sensitivity analyses were performed to determine the optimal multivariable model. When adjusted for either PNI or LVI alone, both Peptostreptococcaceae and Both Peptostreptococcaceae and Erysipelotrichaceae remained significantly associated with poorer OS (Table S3). However, including both simultaneously attenuated these associations (both P > 0.05), confirming multicollinearity due to shared prognostic information. Based on these findings, final multivariable models were constructed adjusting for CA199, PNI, and TRG. After adjustment, both Peptostreptococcaceae (HR = 3.92, 95% CI 1.20–12.85, P = 0.024) and Erysipelotrichaceae (HR = 4.34, 95% CI 1.29–14.60, P = 0.018) remained independently associated with poorer OS (Table S4). Gut microbiota links body composition and survival outcomes in patients with LARC. A Kaplan–Meier survival curves stratified by optimal abundance cut-off values of cachexia-associated gut microbial taxa. B Associations between gut microbial abundance (stratified by survival-based cut-offs) and tumor regression grade following neoadjuvant therapy. C Comparison of VAT, SAT, and SMA between cachexia and non-cachexia groups. D Spearman correlation network illustrating associations among gut microbial taxa, body composition parameters, and clinical indicators Postoperative tumor regression grade (TRG) distributions were comparable between high- and low-abundance groups for both taxa, with no significant differences observed (Fisher’s exact test, P = 0.8517 for Peptostreptococcaceae and P = 0.3206 for Erysipelotrichaceae) (Fig. 4B). At the L3 level, cachectic patients exhibited significantly lower VAT area compared with non-cachectic patients (P = 0.0165), along with reduced SAT (P = 0.0347) and SMA (P = 0.0214) (Fig. 4C). Spearman correlation analyses revealed significant associations between specific gut microbial taxa and host physiological parameters. The abundance of f__Erysipelotrichaceae was inversely correlated with VAT area (r = − 0.183, P = 0.041) and positively associated with f__Peptostreptococcaceae (r = 0.427, P < 0.001). The genus g__Veillonella showed negative correlations with skeletal muscle area (SMA; r = − 0.208, P = 0.020), hemoglobin (r = − 0.196, P = 0.028), and albumin (r = − 0.191, P = 0.032). In contrast, SMA was positively correlated with o__Burkholderiales (r = 0.207, P = 0.021), which was itself positively associated with g__Faecalibacterium (r = 0.413, P < 0.001) and inversely correlated with f__Erysipelotrichaceae (r = − 0.202, P = 0.024) and s__Roseburia_inulinivorans (r = − 0.202, P = 0.024). Collectively, SMA showed positive associations with g__Faecalibacterium and s__Roseburia_inulinivorans, but an inverse association with g__Veillonella, whereas VAT depletion was linked to enrichment of f__Erysipelotrichaceae, a taxon also associated with adverse survival outcomes. Additionally, VAT area was strongly positively correlated with triglyceride levels (r = 0.382, P < 0.001), red blood cell count was positively correlated with albumin (r = 0.411, P < 0.001), and g__Faecalibacterium abundance was positively associated with albumin levels (r = 0.177, P = 0.048) (Fig. 4D). Identification and validation of cachexia-related metabolites and cross-domain correlations In the paired cachexia–non-cachexia cohort with matched plasma metabolomics data (n = 64), OPLS-DA provided evidence for global metabolic differences between the two groups, as reflected by a partial separation in the score plot (Figure S4A). The model explained a substantial proportion of class variance (R²Y = 0.683) and showed moderate predictive performance (Q² = 0.269). Permutation testing indicated that the observed separation was unlikely to result from random overfitting (Figure S4B). Based on OPLS-DA–guided feature selection, several metabolites were identified as differentially abundant. Subsequent receiver operating characteristic (ROC) analyses of individual metabolites demonstrated moderate discriminatory performance for cachexia, with AUC values ranging from 0.65 to 0.78. Among these, 3,5,9-trihydroxyergost-7-en-6-one, 5,6-DHET, lithocholic acid, and cortisone showed the highest AUC values in the training set (n = 64, including 13 cachectic patients) (Fig. 5A). When evaluated in an independent, unmatched validation cohort (n = 70, including 11 cachectic patients), these metabolites retained discriminatory ability, with AUC values consistently exceeding 0.60 (Fig. 5B). To further interpret the biological relevance of these signatures, KEGG enrichment analysis was performed. Notably, despite the limited number of hits, arachidonic acid metabolism (P = 0.056, Enrichment Ratio = 17.27) and steroid hormone biosynthesis (P = 0.11, Enrichment Ratio = 8.85) emerged as the top nominally enriched pathways (Figure S5). Metabolite-based discrimination and integrative correlation analysis in cachexia and non-cachexia patients with LARC. A Diagnostic performance of metabolites selected by OPLS-DA in the microbiota-matched cohort. B Receiver operating characteristic (ROC) curves of four metabolites with AUC > 0.6 validated in the unmatched cohort. C Spearman correlation network illustrating associations among gut microbiota, metabolites, and clinical parameters in the matched cohort To explore cross-domain associations, a Spearman correlation network was constructed using the paired dataset, integrating microbial taxa, circulating metabolites, clinical laboratory indices, and body composition parameters. Significant correlations (P < 0.05) formed a connected network linking metabolic markers with host phenotypes and gut microbiota. Within the network, several metabolite-centered associations were particularly prominent. 3,5,9-trihydroxyergost-7-en-6-one showed a negative correlation with s__un_g_Veillonella (r = − 0.277, P = 0.0309) and with SAT_Area (r = − 0.256, P = 0.046), while exhibiting a positive association with o__Peptostreptococcales–Tissierellales (r = 0.258, P = 0.045). In addition, 5,6-DHET was strongly and positively correlated with 3,5,9-trihydroxyergost-7-en-6-one (r = 0.442, P < 0.001), indicating coordinated variation between these two cachexia-associated metabolites. Beyond metabolite–microbiota interactions, significant links between clinical biochemical indices and specific bacterial taxa were also observed. Serum cholesterol was positively correlated with s__Fusobacterium_nucleatum (r = 0.294, P = 0.0213), further integrating lipid metabolism into the multi-domain network. Importantly, the observed associations between gut microbiota, skeletal muscle and adipose tissue measures, and hematological indicators were largely consistent with findings derived from the larger cohort lacking matched metabolomic data, supporting the robustness and stability of these microbiota–host phenotype relationships across different sample sets (Fig. 5C). Discussion Previous microbiome-related studies in LARC have predominantly focused on treatment responsiveness, including pathological tumor regression, radiosensitivity, and neoadjuvant therapy outcomes, with limited attention to host systemic phenotypes beyond the tumor itself. In contrast, cancer cachexia represents a critical yet underexplored dimension of LARC, characterized by progressive skeletal muscle and adipose tissue loss and closely linked to long-term survival. To our knowledge, this is the first study to comprehensively characterize the gut microbiome–metabolome–body composition axis in LARC from the perspective of cancer cachexia. By shifting the analytical focus from tumor response alone to host metabolic vulnerability, we demonstrate that cachexia-associated microbial signatures are tightly coupled to skeletal muscle and adipose tissue depletion and are associated with significantly worse overall survival, independent of postoperative tumor regression (TRG 0–1: 60.0% in cachectic vs. 69.6% in non-cachectic patients; P = 0.406). This integrated approach provides a novel biological context for understanding inter-individual heterogeneity in clinical outcomes among LARC patients beyond conventional tumor-centric metrics. The absence of significant differences in gut microbial α- or β-diversity between cachectic and non-cachectic patients suggests that the development of cachexia in LARC is not characterized by a global disruption of microbial diversity, but rather by selective compositional remodeling of specific taxa. In line with this notion, we observed a marked enrichment of Peptostreptococcaceae, Erysipelotrichaceae, Veillonella, and Fusobacterium in cachectic patients. Notably, these taxa-level alterations appeared consistent across biological sexes, as the sex distribution was well-balanced between our study groups. Importantly, the adverse prognostic impact associated with higher abundances of Peptostreptococcaceae and Erysipelotrichaceae was independent of postoperative tumor regression grade, indicating that their clinical relevance is unlikely to be mediated through tumor radiosensitivity or local treatment response. Instead, these taxa were closely associated with hallmarks of systemic deterioration, including depletion of adipose tissue and skeletal muscle, anemia, and hypoalbuminemia. Collectively, these findings suggest an association between cachexia-associated microbial signatures and a host-centered process of systemic metabolic and nutritional impairment, which may be linked to unfavorable survival. The depletion of key butyrate-producing genera, most notably Faecalibacterium, Roseburia and Prevotella in cachectic patients points to a functional collapse of a critical gut-muscle signaling axis [14, 15]. These taxa are not merely biomarkers of health; their decline directly translates to a local and systemic butyrate deficit [16, 17]. Butyrate serves as the primary energy source for colonocytes, reinforcing gut barrier integrity and suppressing pro-inflammatory pathways. Its systemic anti-inflammatory effects, including the inhibition of cytokines like IL-6, are well-documented [18, 19]. IL-6 is a key pro-inflammatory cytokine that plays a central role in the pathogenesis of cancer cachexia by promoting muscle protein breakdown and inhibiting anabolic signaling [20, 21].Therefore, their loss in cachexia likely creates a permissive environment for systemic inflammation, a known driver of muscle proteolysis via the ubiquitin-proteasome and autophagy-lysosome systems [22, 23].Concomitantly, UBA1819, an uncultured genus within the family Ruminococcaceae, was enriched in cachectic patients and exhibited inverse correlations with Faecalibacterium, hemoglobin, and albumin, reflecting distinct ecological niches within the short-chain fatty acids (SCFA)-producing microbial network and highlighting intra-family heterogeneity and divergent functional roles [24, 25].Previous research linked higher Desulfovibrio piger abundance with increased sarcopenia severity [26]. In contrast, in our study D. piger was enriched in non-cachectic patients and did not correlate significantly with SMA. Mechanistically, D. piger has been shown to engage in cross-feeding with Faecalibacterium prausnitzii, collectively enhancing butyrate production [27]. Consistently, prior literature has demonstrated positive correlations between D. piger and several beneficial taxa, including Prevotella, Roseburia, and Faecalibacterium, which aligns with our observations and underscores the importance of cooperative microbial networks rather than isolated taxa in shaping host metabolic outcomes [28]. Together, these findings suggest that the functional impact of D. piger may depend on the integrity of the surrounding butyrate-producing ecosystem. This mechanistic link is reinforced by our functional predictions: the enrichment of the TCA cycle in non-cachectic patients suggests a microbiome supportive of efficient host energy harvesting, while the N-acetylneuraminate degradation pathway enriched in cachexia may indicate enhanced sialic acid scavenging—a potential microbial adaptation to a nutrient-starved, inflamed gut environment that further depletes resources from the host [29–31]. Taken together, our compositional, functional, and correlative data suggest that the cachexia-associated gut microbiome may transition toward a composition characterized by reduced mutualistic taxa (e.g., Faecalibacterium) and increased abundance of taxa associated with inflammation and metabolic dysregulation (e.g., Peptostreptococcaceae, Erysipelotrichaceae). This shift is correlated with host metabolic vulnerability and tissue wasting, although the cross-sectional design precludes determination of causality or directionality. The reduction in serum cortisone observed in cachectic patients may reflect enhanced peripheral conversion to active cortisol under chronic inflammatory and metabolic stress, functionally amplifying glucocorticoid signaling in skeletal muscle and favoring catabolic programs involved in muscle protein degradation [32, 33].This finding is consistent with the nominally enriched steroid hormone biosynthesis pathway identified in our KEGG pathway analysis (Raw p = 0.11, 1 hit). Additionally, this hormonal alteration might also reflect a tumor-driven “steroid clearance” mechanism, as proposed in recent experimental models where tumors actively deplete circulating steroids to promote host catabolism [34]. In parallel, nominal enrichment was observed in arachidonic acid metabolism (Raw p = 0.0564, 1 hit), driven by decreased 5,6-DHET, a metabolite of the cytochrome P450 epoxygenase pathway. While the lipoxygenase branch of arachidonic acid metabolism is typically pro-inflammatory in cachexia, the epoxygenase pathway generates anti-inflammatory mediators [35, 36]. Reduced 5,6-DHET could be consistent with impaired inflammation resolution, which may play a role in sustained systemic inflammation and tissue wasting. Furthermore, lithocholic acid (LCA), a microbiota-derived secondary bile acid, was markedly reduced in cachectic patients and emerged as a key metabolic node linking gut microbial dysbiosis to host metabolic alterations. Experimental studies have demonstrated that LCA activates TGR5-mediated AKT/mTOR signaling, suppresses FoxO3-driven proteolysis, and promotes skeletal muscle regeneration, underscoring its protective role against muscle wasting [37–39]. Beyond skeletal muscle, LCA-mediated TGR5 activation also supports hepatic metabolic function and bile flow, consistent with the negative correlation observed between LCA and circulating bilirubin levels in our cohort [40]. The depletion of LCA in cachexia may therefore reflect impaired microbiota-driven bile acid signaling and disruption of this regulatory axis. Notably, total bilirubin levels were not significantly elevated in cachectic patients, suggesting that bilirubin homeostasis in advanced cancer is shaped by additional factors, including hepatic synthetic capacity, nutritional status, and systemic inflammation, rather than bile acid signaling alone. Although a limited number of Peptostreptococcaceae strains have the capacity to convert primary to secondary bile acids via 7α-dehydroxylation, this function is not uniform across the order Peptostreptococcales–Tissierellales [41]. Their expansion in cachectic patients may therefore reflect a broader shift toward a proteolytic and pro-inflammatory microbial ecosystem, rather than a direct increase in classical lithocholic acid–producing bacteria. Such ecological remodeling could indirectly impair secondary bile acid production, contributing to the observed depletion of lithocholic acid. Importantly, this inverse association does not imply causality but highlights coordinated alterations of microbial composition and secondary bile acid metabolism in cancer cachexia, warranting further functional investigation. In the present study, SAT and VAT were positively correlated with circulating triglyceride levels, and all three parameters were significantly reduced in cachectic patients. The directionality and mechanisms underlying this association remain unclear, due to a paucity of studies jointly evaluating adipose compartments and lipid profiles in cachexia [42]. Evidence from existing literature remains heterogeneous. A large cohort study including 1,393 patients with gastric cancer reported significantly lower triglyceride and cholesterol levels in cachectic individuals compared with non-cachectic controls, whereas another study of 770 gastric cancer patients, applying comparable diagnostic criteria for cachexia, observed similar trends without reaching statistical significance [43, 44]. Along the same lines, our previous analysis in a prior LARC cohort, as well as a smaller study in patients with gastrointestinal malignancies, demonstrated lower triglyceride and cholesterol levels in cachectic patients, although these differences did not reach statistical significance [4, 45]. Collectively, these findings suggest that reduced circulating lipid levels in cancer cachexia may reflect accelerated lipid turnover and systemic energy exhaustion rather than impaired lipid mobilization. In this setting, concomitant declines in SAT/VAT and serum triglycerides likely represent parallel manifestations of cachexia-associated metabolic remodeling, rather than a direct causal relationship. The heterogeneity across studies further indicates that hypolipidemia is a context-dependent metabolic phenotype rather than a universal hallmark of cancer cachexia. Taken together, our multi-omics data reveal a coordinated axis linking gut microbiome alterations, metabolic reprogramming, and body composition changes in LARC-associated cachexia. The depletion of butyrate-producing taxa (e.g., Faecalibacterium, Roseburia) and secondary bile acid LCA, coupled with reduced 5,6-DHET and cortisone, is consistent with a shift toward a pro-inflammatory and less metabolically supportive microbial ecosystem. These microbial and metabolic signatures were significantly associated with reduced SMI and VAT/SAT depletion, as well as systemic markers of nutritional deterioration (anemia and hypoalbuminemia). Although the cross-sectional design precludes causal inference, these findings suggest that cachexia-associated microbial dysbiosis may contribute to host metabolic vulnerability through impaired short-chain fatty acid production, disrupted bile acid signaling, and attenuated inflammation resolution. In turn, such metabolic shifts appear to parallel and potentially exacerbate progressive tissue wasting. This integrated microbiome–metabolome–body composition pattern provides a host-centered framework that complements conventional tumor-centric metrics and may help explain inter-individual heterogeneity in clinical outcomes and survival among LARC patients. Several limitations merit consideration. First, the retrospective, single-center design precludes causal inference, and the cross-sectional nature of baseline microbiome and metabolomic profiling does not allow assessment of temporal dynamics in cachexia development. Second, while the metabolomic signatures were derived from a discovery cohort and validated in an independent cohort, the number of cachectic patients in both cohorts was relatively limited. This highlights the need for future validation in larger, well-powered cohorts to confirm the robustness of these candidate metabolites. Third, gut microbiota characterization was based on 16 S rRNA sequencing, which limits species-level resolution and functional inference. Although predictive functional analyses provided mechanistic insights, validation using shotgun metagenomics or targeted metabolomic flux analyses is warranted. Fourth, residual confounding by unmeasured factors such as diet, physical activity, and medication use (e.g., antibiotics, proton pump inhibitors) cannot be fully excluded. Finally, the homogeneous LARC cohort enhances internal validity but may limit generalizability to other cancer types or disease stages. Conclusions In summary, cancer cachexia in LARC is characterized by selective gut microbiome and metabolome remodeling rather than global microbial diversity loss. These cachexia-associated alterations are closely linked to progressive skeletal muscle and adipose tissue depletion and are associated with significantly worse overall survival. Collectively, our findings underscore a host-centered metabolic vulnerability in cancer cachexia, characterized by the coordinated shifts across microbial, metabolic, and body composition profiles identified in this study. While these microbial signatures show potential as clinical biomarkers for risk stratification, these findings primarily serve as a conceptual framework for future therapeutic target discovery. Further longitudinal and mechanistic studies are warranted before any therapeutic implications can be considered for managing cancer cachexia. Supplementary Information Supplementary Material 1. Acknowledgements Not applicable. Clinical trial number Not applicable. Abbreviations Body Mass Index Carbohydrate Antigen 19 − 9 Carcinoembryonic Antigen Confidence Interval Circumferential Resection Margin Computed Tomography Extramural Vascular Invasion Hazard Ratio Hounsfield Unit Intraclass Correlation Coefficient Interquartile Range Locally Advanced Rectal Cancer Long-Course Radiotherapy Liquid Chromatography-Tandem Mass Spectrometry Linear Discriminant Analysis Effect Size Lymphovascular Invasion Magnetic Resonance Imaging Non-metric Multidimensional Scaling Overall Survival Permutational Multivariate Analysis of Variance Perineural Invasion Receiver Operating Characteristic Subcutaneous Adipose Tissue Short-Course Radiotherapy Structural Equation Modeling Skeletal Muscle Area Total Mesorectal Excision Tumor Regression Grade Visceral Adipose Tissue Variable Importance in Projection Pathological Tumor Stage after Neoadjuvant Therapy Pathological Nodal Stage after Neoadjuvant Therapy Authors’ contributions Conceptualization: Benhua Xu, Zhixing Kuang; Data curation: Zhixing Kuang, Jiannan Tu, Anchuan Li; Formal analysis: Zhixing Kuang, Qing Zhan, Jiannan Tu; Investigation: Zhixing Kuang, Jiannan Tu, Qing Zhan; Methodology: Zhixing Kuang, Jiannan Tu, Qing Zhan, Wei Guo; Project administration: Benhua Xu; Supervision: Benhua Xu, Hao Jiang, Rong Zheng; Visualization: Zhixing Kuang, Jiannan Tu, Qing Zhan; Writing – Original Draft: Zhixing Kuang, Jiannan Tu; Writing – Review & Editing: Benhua Xu, Hao Jiang; All authors have read and approved the final manuscript. No eligible author has been excluded. Funding This work was supported by the Fujian Provincial Natural Science Foundation Program (Grant No. 2026J001573). Data availability The raw 16S rRNA gene sequencing data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1434023 and are publicly available at https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1434023/. The serum metabolomics data have been deposited in the MetaboLights database under accession number MTBLS13787 and are publicly available at https://www.ebi.ac.uk/metabolights/MTBLS13787. Declarations Ethics approval and consent to participate This study adhered to the Declaration of Helsinki and Ethical Guidelines for Clinical Research. It was approved by the Institutional Review Board of Fujian Medical University Union Hospital (No. 2025KY316). All patients provided written informed consent for treatment prior to therapy. Due to the retrospective design and anonymized data, the requirement for study-specific informed consent was waived. Data were handled in compliance with privacy regulations. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Conflict of interest The authors declare no conflicts of interest. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Zhixing Kuang, Qing Zhan, Jiannan Tu and Anchuan Li contributed equally to this work. Contributor Information Rong Zheng, Email: zhengrrong@outlook.com. Hao Jiang, Email: hydeyjh@163.com. Benhua Xu, Email: benhuaxu@163.com. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The raw 16S rRNA gene sequencing data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1434023 and are publicly available at https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1434023/. The serum metabolomics data have been deposited in the MetaboLights database under accession number MTBLS13787 and are publicly available at https://www.ebi.ac.uk/metabolights/MTBLS13787.