Serum samples of healthy individuals and patients with inflammatory colorectal polyps and colorectal adenomas were collected from the Affiliated Jiangyin Hospital of Nanjing University of Chinese Medicine. The physicians requested the medically indicated blood collection. All participants were diagnosed using colonoscopy, and individuals with cancer, cardiovascular diseases, immunodeficiency diseases, and other nervous system diseases that may impact metabolism were excluded. Patients with a history of long-term drug use were also excluded. Patients and healthy subjects gave written informed consent before enrollment. All procedures were approved by the medical ethics committee of the Affiliated Jiangyin Hospital of Nanjing University of Chinese Medicine in accordance with the Declaration of Helsinki. Prior to colonoscopy, blood samples were obtained from all patients. All blood samples were promptly centrifuged at 2,000 × g for 15 min, and serum was aliquoted into clean Eppendorf tubes and kept at −80 °C before analysis.

To minimize the inter-individual variation, we performed sample pooling before processing. Three to six individual samples from the same group were pooled to constitute a pool. The sequential precipitation and delipidation method was used for the depletion of high-abundance proteins (Li et al., 2020). A total of 50 μL of water and 250 μL of methyl tert-butyl ether (MTBE) (Sigma, 306975) were added to 50 μL of the pooled serum, followed by vortex mixing. Subsequently, 150 μL of methanol (Merck, 1.06007.4008) was added and the mixture was incubated at 4 °C for 30 min. After incubation, the mixture was centrifuged at 21,000 × g for 30 min at 4 °C, and the supernatant was carefully transferred to a new tube. Then 500 μL of MTBE and 100 μL of water were added. After efficient mixing, the solution separated into two phases. The lower phase was collected and desalted using an HLB SPE column (Waters, 186000383) according to the manufacturer’s instructions.

The resulting samples were resuspended in 8 M urea (Sigma, U5128)/100 mM NH₄HCO₃ (Honeywell, 40867) and reduced with 20 mM dithiothreitol at 37 °C for 1 h followed by alkylation with 40 mM iodoacetamide for 30 min at room temperature in the dark. The urea concentration was diluted to less than 2 M with four volume of 50 mM NH₄HCO₃. The trypsin was added at a trypsin:protein ratio of 1:50 (w/w) for digestion at 37 °C overnight. Samples were acidified with trifluoroacetic acid to a final concentration of 1% for C18 StageTip binding and desalting. The eluted peptides were lyophilized and subsequently reconstituted in 0.1% (v/v) formic acid solution. Following this, equal sample aliquots were subjected to LC-MS/MS analysis.

All samples were analyzed in data independent acquisition mode, and 2 μg peptides were loaded on 20-cm column packed in-house with C18 3 um ReproSil particles (Dr. Maisch GmbH, r13.aq.). Peptide mixtures were injected on an EASY-nLC 1200 system (Thermo Fisher Scientific) coupled to the mass spectrometer (Q Exactive Plus, Thermo Fisher Scientific), and separated by a non-linear 120 min gradient using mobile phase A (100% H2O, 0.1% formic acid) and B (80% acetonitrile, 0.1% formic acid). Peptides were eluted (300 nL/min) by gradient as follows: 2%–5% B, 5 min; 5%–32% B, 90 min; 32%–45%, 12 min; 45%–100% B, 3 min; 100% B, 10 min. Column temperature was maintained at 50 °C. The survey scan changes in the 300–1,500 m/z range with a maximum injection time of 150 ms and a resolution of 70,000. The automated gain control (AGC) target was 3e6. Following the full MS scan, DIA scans were acquired at a resolution of 35,000 and AGC target 2e5 with 20 m/z isolation window. The precursors were fragmented using HCD and normalized collision energy set to 27%, and maximum injection time was automatic. The data were recorded in centroid mode.

Raw files were analyzed with DIA-NN (version 1.8.1). The spectral library was created using the human proteome downloaded from Uniprot (retrieved in March 2023) with the ‘Deep learning-based spectra and RTs prediction’ enabled. The false discovery rate (FDR) of precursor was set as 1%. Other settings were used as default parameters. Statistical analyses and graphical representations were performed in R (https://www.r-project.org/). All bioinformatics analyses were done with R (https://www.r-project.org/), Origin and Metascape (Zhou et al., 2019). In each experimental group, more than half of the proteins with quantitative values are retained. Missing values are imputed using the KNN algorithm. Batch effects were removed with R package sva from the proteins quantified across both cohort 1 and cohort 2. Principal component analysis (PCA) was performed using the R package factoextra, and results were visualized in Origin. We performed sPLS-DA (KA et al., 2011) and LASSO regression, employing the R packages mixOmics and glmnet, respectively. Receiver operating characteristic (ROC) curve analysis was carried out with the pROC package. Protein abundance was compared using ANOVA test with 5% FDR. P values were adjusted by Benjamini & Hochberg. The data presented in the study are deposited in the iProX repository, accession number PXD 063783.

Serum levels of FLNA and APOA4 were quantified using commercial human ELISA kits (FLNA: ELK Biotechnology, Cat# ELK4516; APOA4: ELK Biotechnology, Cat# ELK11097) according to the manufacturers’ protocols. All measured concentrations were normalized to total protein content and are expressed as ng/mL.

For serum proteomic profiling, samples were collected from two independent cohorts (Figure 1; Supplementary Table S1). Cohort 1 comprised pooled samples from 45 patients with inflammatory colorectal polyps, 60 patients with colorectal adenomas, and 33 healthy controls. This pooling strategy was employed to ensure sufficient protein quantity for detection while mitigating inter-individual variability and maintaining biological representativeness. Cohort 2 consisted of individual samples from 15 patients with inflammatory colorectal polyps, 12 patients with colorectal adenomas, and 15 healthy controls, serving as an independent validation set. Blood contains a reservoir of potential diagnostic biomarkers. To exploit this potential, serum samples were subjected to high-abundance protein depletion, followed by reduction, alkylation, and tryptic digestion to generate peptides. Peptide samples were analyzed by nanoflow liquid chromatography-tandem mass spectrometry (LC-MS/MS) to characterize alterations in the serum proteome of colorectal adenoma patients. In two cohorts, colorectal adenoma patients exhibited a significant male predominance (75.0%), consistent with established epidemiological patterns. Additionally, no statistically significant differences were demonstrated in age or polyp number among healthy individuals, patients with inflammatory polyps, and those with adenomas (Supplementary Figure S1).

In cohort 1, a total of 5,568 peptides were identified, with per-sample counts ranging from 4,307 to 4,891. From these, 633 human proteins were quantified in at least one sample, and the per-sample protein count ranged from 337 to 524 (Figure 2A). In cohort 2, proteomic analysis identified 7,716 peptides in total (range: 4,020–5,588 per sample). Among these, 916 human proteins were quantified in at least one sample, with individual sample quantifications ranging from 267 to 524 proteins (Figure 2B). To assess the reliability of the proteomic data, we examined the raw MS/MS data and found that over 95% of peptides were matched by ≥2 spectral counts in cohort 1. The average spectral count for all peptides in this cohort was 28.9, indicating high peptide-level reliability (Supplementary Figure S2A). Furthermore, 521 proteins (82.3%) were supported by ≥2 peptides, with an average of 7.0 peptides per protein (Supplementary Figure S2B), confirming high confidence in protein-level quantification. Similarly, in cohort 2, over 95% of peptides were matched by ≥2 spectral counts, and 73.6% of proteins were supported by ≥2 peptides (Supplementary Figures S2C–D).

We further evaluated quantitative precision by calculating the coefficient of variation (CV) of protein expression across all samples. Samples H, A, and B showed high measurement precision in both cohorts, with CVs of 3.1%, 2.8%, and 2.7% in cohort 1, and 3.5%, 4.4%, and 3.9% in cohort 2, respectively (Figures 2C,E). These low CV values reflect the high reproducibility of the entire analytical workflow, including sample preparation, data acquisition, and bioinformatic processing. Principal component analysis (PCA) was performed to visualize overall proteomic profiles (Figures 2D,F). Notably, the three groups in cohort 2 showed clearer separation than those in cohort 1, suggesting improved group discrimination in cohort 2. The distribution of MS/MS spectral counts of quantified peptides and the distribution of peptide numbers of quantified proteins are shown in Supplementary Figure S2.

Serum proteomic profiling of cohort 1 revealed 49 significant DEPs (one-way ANOVA, adjusted p < 0.05) distinguishing patients with colorectal inflammatory polyps, patients with adenomas and healthy controls. Hierarchical clustering of these proteins demonstrated distinct molecular stratification, as visualized in the accompanying heatmap (Figure 3A). The DEPs were subsequently analyzed using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses. Significant enrichment was observed in biological process terms and KEGG pathways including inflammatory response, fluid homeostasis regulation, actin cytoskeleton reorganization, platelet activation, and focal adhesion (Figure 3B; Supplementary Table S3).

To identify potential adenoma biomarkers, we integrated the proteomic data from cohort 1 and cohort 2 using the subset of proteins quantified in both cohorts. Following batch effect removal, the overall data distribution was consistent across cohorts (Supplementary Figure S3A). Principal component analysis (PCA) further confirmed minimal inter-cohort variation (Supplementary Figure S3B), while clearly separating the three experimental groups (Supplementary Figure S3C). The top five proteins identified by sPLS-DA in cohort 1, based on variable importance in projection (VIP) scores, were transgelin-2 (TAGLN2), FLNA, PDZ and LIM domain protein 1 (PDLIM1), fermitin family homolog 3 (FERMT3), and PFN1 (Figure 4A). The intersection of 49 differentially expressed proteins (DEPs) from cohort 1, 149 DEPs from cohort 2 (one-way ANOVA, adjusted p < 0.05), and the top 20 VIP proteins from sPLS-DA analysis yielded 14 consensus candidate proteins (Figure 4B). Subsequent LASSO regression analysis of these 14 proteins identified four candidate biomarkers (Figures 4C,D). Among these, APOA4, FERMT3, and FLNA exhibited consistent expression changes in the adenoma groups of both cohorts (Figure 4E). In contrast, THBS1 was downregulated in cohort 1 but showed no significant change in cohort 2. Additionally, APOA4 and FLNA levels differed significantly between inflammatory polyps and adenomas in at least one cohort. Based on these findings, we selected APOA4 and FLNA for further validation of their serum levels.

To validate APOA4 and FLNA as potential biomarkers, we conducted an independent validation study in cohort 3 (Figure 5A; Supplementary Table S4), including 37 healthy controls (H), 38 inflammatory polyp patients (A), and 45 adenoma patients (B). Serum samples from these individuals were collected and then subjected to ELISA analyses to measure the serum levels of APOA4 and FLNA. Notably, the serum levels of APOA4 measured by ELISA did not correlate with the prior LC-MS/MS proteomic findings (Supplementary Figure S4). In contrast, ELISA quantification of FLNA revealed a progressive decrease in serum levels along the H-A-B pathological continuum (Figure 5B). Receiver Operating Characteristic (ROC) analysis demonstrated FLNA’s diagnostic performance with AUC values of 0.810 (95% CI: 0.713–0.907) for H vs. B discrimination, 0.734 (95% CI: 0.622–0.85) for A vs. B, and 0.639 (95% CI: 0.51–0.77) for H vs. A, indicating moderate-to-good diagnostic accuracy for adenoma detection (Figure 5C). FLNA reached an AUC value of 0.772 (95% CI: 0.673–0.871) to distinguish patients with colorectal adenoma from both healthy controls and inflammatory colorectal polyp patients (Figure 5D). At the determined cutoff, FLNA demonstrated high specificity but limited sensitivity for distinguishing B from H, B from A, and B from the combined group of H&A. The classifier showed high positive predictive values (PPV: 0.96 for H vs. B; 0.87 for A vs. B) but comparatively low negative predictive values (NPV: 0.68 for H vs. B; 0.67 for A vs. B) (Supplementary Figure S5). Serum levels of FLNA exhibited a negative correlation with the progression from normal tissue to polyp and subsequent development into adenoma (Figure 5E), as well as with the total number of polyps present (Figure 5F). Western blot analysis revealed no significant difference between inflammatory polyp patients and healthy controls, and adenoma patients showed significantly decreased FLNA expression relative to both healthy controls and inflammatory polyp cases (Supplementary Figure S6). Circulating FLNA likely derives from both passive leakage of damaged or remodeling tissues and active inflammatory secretion. The observed serum FLNA reduction may directly result from decreased tissue expression.