The study cohort used in the present study included patients described in our previous study (20). The present study involved conducting untargeted metabolomics analysis on fecal samples obtained from patients with CPs and healthy individuals at the Shanghai Jiading District Hospital of Chinese Medicine (Shanghai, China) between March and August 2024. All participants were enrolled consecutively upon providing written informed consent. Patients undergoing colonoscopy and diagnosed with CPs and healthy individuals without CPs were included. A total of 30 patients with CPs were classified as the CP group, whereas 30 healthy individuals were classified as the healthy control (HC) group. Participants in each group were matched based on age, sex and body weight during enrollment. The CP group was selected according to the following inclusion criteria: Aged >18 years, and a diagnosis of CP confirmed by colonoscopy and histopathology. The exclusion criteria for the CP group were as follows: i) Use of antibiotics or probiotics within the past 2 months; ii) a previous history of CRC. The inclusion criteria of the HC group was: Aged >18 years without CP confirmed by colonoscopy examination. The exclusion criteria for the HC group were as follows: i) Presence of any underlying diseases, including autoimmune diseases such as ankylosing spondylitis and systemic lupus erythematosus, infectious diseases, cachexia, organ failure, respiratory diseases and cardiovascular diseases; ii) history of surgery or chronic drug use; iii) use of antibiotics or probiotics within the past 2 months.

Fecal samples were collected by the participants after enrolling in the study. Immediately after collection, the samples were placed on dry ice and stored at -80˚C until further use. Subsequently, the samples were sent to Shanghai Bioprofile Biotechnology Co., Ltd. for untargeted metabolomics analysis. These samples were mixed with 400 µl cold (4˚C) methanol acetonitrile (v/v; 4:1) and homogenized with a tissue crusher. The mixture was then sonicated (53 kHz) for 20 min in an ice bath. Following this, the mixture was incubated at -20˚C for 1 h, and centrifuged at 4˚C for 20 min at 16,000 x g. Next, the supernatants were harvested and were vacuum dried.

UHPLC (LC-30AD; Shimadzu Corporation) coupled with Q-Exactive™ Plus (Thermo Fisher Scientific, Inc.) was used for metabolomics profiling analysis.

Samples were placed in a 4˚C autosampler and underwent liquid chromatography separation using an ACQUITY UPLC^® HSS T3 column (2.1x100 mm; 1.8 µm; Waters Corporation). The injection volume was 4 µl and the column temperature was 40˚C. The flow rate was set at 0.3 ml/min, and the mobile phase consisted of two components: i) A (0.1% formic acid in water); and ii) B (0.1% formic acid in acetonitrile). The gradient started at 0% buffer B for 2 min, increased linearly to 48% over 4 min, was further increased to 100% over 4 min and held for 2 min, then decreased to 0% buffer B within 0.1 min, followed by a 3 min re-equilibration period.

Electrospray ionization (ESI) in both positive and negative modes was used for separate acquisition of MS data. The heated-ESI source was configured with the following conditions: i) Spray voltage at 3.8 kV (positive) and 3.2 kV (negative); ii) capillary temperature at 320˚C; iii) sheath gas (nitrogen) flow at 30 arbitrary units (AU) with temperature at 300˚C and with nebulizer pressure at 100 pounds per square inch; iv) aux gas flow at 5 AU; v) probe heater temperature at 350˚C; and vi) S-Lens radio frequency level at 50. The instrument was set to acquire full MS data over the m/z range of 70-1,050 Da. The full MS scans were performed at a resolution of 70,000 at m/z 200 and 17,500 at m/z 200 for MS/MS scan. The maximum injection time was set at 100 msec for MS and 50 msec for MS/MS. The isolation window for MS/MS was set to 2 m/z and the normalized collision energy (stepped) was set to 20, 30 and 40 V for fragmentation.

The raw MS data underwent processing using MS-DIAL (version 4.9) (21) to align peaks, correct retention time and extract peak areas. Metabolites were identified based on accurate mass (with a mass tolerance of <10 parts per million) and MS/MS data (with a mass tolerance of <0.01 Da) through matching against The Human Metabolome Database (https://hmdb.ca/), MassBank (https://massbank.eu/MassBank), Global Natural Product Social Molecular Networking (https://gnps.ucsd.edu) library databases and Baipu Metabolite Standard Library (in-house database of Shanghai Bioprofile Biotechnology Co., Ltd.). Only variables with >50% non-zero measurement values in at least one group were retained from the extracted-ion features.

R programming language (v4.0.3) and R packages were used for all multivariate data analyses and modeling (https://www.r-project.org/). Metabolomics data were mean-centered using Pareto scaling (22). Models were constructed using principal component analysis (PCA), orthogonal partial least squares (OPLS)-discriminant analysis (DA) and partial least squares (PLS)-DA. All models were evaluated and tested for overfitting using permutation tests. The descriptive performance of the models was assessed based on the cumulative R2X (with a perfect model having R2X, 1) and R2Y (with a perfect model having R2Y, 1) values, while their predictive performance was evaluated using cumulative Q2 (with a perfect model having Q2, 1) and a permutation test (n=200). A permuted model should be unable to predict the classes: The R2 and Q2 values at the Y-axis intercept should be lower than those of the non-permuted model.

Discriminating metabolites that differentiate patients with CPs from HCs were identified utilizing a statistically significant threshold of the variable influence on projection (VIP) values derived from the OPLS-DA model, along with two-tailed unpaired Student's t-test (P-value) applied to the normalized raw data at the univariate analysis level. The VIP score represents the contribution of the variable to discriminating between sample classes, calculated as the weighted sum of squares of PLS weights for each variable. The mean VIP value is 1, and VIP values >1 and P<0.05 are considered significant, indicating a strong discriminatory ability and serving as criteria for biomarker selection. Fold change was determined as the logarithm ratio of the average mass response (area) between two arbitrary classes. Additionally, the identified differential metabolites were subjected to cluster analyses using R package (v4.0.3) and a heatmap was generated. Receiver operating characteristic (ROC) analysis was performed and the area under the curve (AUC) was calculated to assess the predictive capability of potential bacterial markers for distinguishing between the CP and HC groups.

To identify the disrupted biological pathways, KEGG pathway analysis was conducted on the differential metabolite data using the KEGG database (http://www.kegg.jp). KEGG enrichment analyses were performed using Fisher's exact test, and multiple testing correction was applied using the false discovery rate method. Pathways enriched at P<0.05 level were considered statistically significant. The correlations between change in metabolites were estimated with Pearson correlation tests using R package (v4.0.3).

No significant differences were found between the CP and HC groups in terms of age, sex, height, weight or BMI, as previously shown (20). The histological types of CPs comprised the following: i) Tubular adenoma (63.3%); ii) tubular adenoma/serrated polyps (3.3%); iii) tubular adenoma/hyperplastic polyps (16.7%); iv) tubular adenoma/hyperplastic polyps/serrated polyps (3.3%); v) serrated adenoma (3.3%); and vi) hyperplastic polyps (10.0%).

After filtering and pre-processing data, a total of 62,209 peaks were normalized and utilized for subsequent analyses. The unsupervised PCA revealed that the quality control study pools formed a distinct cluster at the center of the corresponding study samples (Fig. S1). PLS-DA, a supervised multivariate analysis, effectively distinguished CP and HC samples into well-defined groups using the untargeted data, supported by strong model statistics (R2Y, 0.990; R2X, 0.214; Q2, 0.845; Fig. 1A). Similarly, supervised multivariate analysis using OPLS-DA also discriminated CP and HC groups with good model statistics (R2Y, 0.990; R2X, 0.214; Q2, 0.831; Fig. 1B).

Next, differential abundance analysis of metabolites was performed, and a total of 300 metabolites were identified (Table SI). Fig. 2A displays the top 30 metabolites with statistically significant differences (based on P-value) between the CP and HC groups, with a VIP score >1 and P<0.001 (Table SII). Specifically, the log₂ fold change, VIP and P-value for prostaglandin A1 were -7.453520969, 2.161393135 and 4.88x10^-8, respectively (Table SI). Correlation analysis showed a positive correlation between these identified metabolites (Fig. S2; Table SIII). Specifically, 6-keto-decanoylcarnitine, trimethadione, salsolinol-1-carboxylic acid, histidyl-proline, norethindrone, gibberellin A12 aldehyde, 6-isopentenyladenine-9-β-D-glucopyranoside, prostaglandin E2, yucalexin B5 and 5,6-dihydroxyprostaglandin F1a were significantly less abundant in the CP group compared with in the HC group (Fig. 2A). By contrast, phosphatidylcholine (PC; 16:0/18:2) exhibited a significantly greater abundance in the CP group compared with in the HC group (Fig. 2A). The hierarchical clustering heatmap (Fig. 2B) visually illustrates the distribution of metabolites that significantly differ between the CP (green) and HC (blue) groups.

KEGG pathway analysis identified multiple perturbed metabolic pathways between the CP and HC groups. The perturbed pathways (Fig. 3) encompassed ‘neuroactive ligand-receptor interaction’, ‘aminoacyl-tRNA biosynthesis’, ‘central carbon metabolism in cancer’, and ‘phenylalanine, tyrosine and tryptophan biosynthesis’. These findings indicated that metabolic pathways, as well as individual metabolites, may be altered in patients with CPs.

All 300 metabolites with differential abundance underwent ROC analysis. Fig. 4 summarizes the ROC analysis results for six significantly abundant metabolites. 6-Keto-decanoylcarnitine, trimethadione, salsolinol-1-carboxylic acid and 6-isopentenyladenine-9-β-D-glucopyranoside were ranked in the top 30 by VIP score and with AUC values in the top 5% (≥0.967), indicating accurate performance for the CP status. In addition, PC (16:0/18:2) and prostaglandin A1 achieved an AUC of 1 indicating perfect discrimination.