Investigating the mechanisms of PhIP-induced colorectal cancer through network toxicology, machine learning, and molecular dynamics simulation Authors: Yuqi Gao, Qian Ye, Yunhuan Zhen, Yan Wang, Yi Wang, Yuping Zeng, Xiaoqin Liu, Xiaoru Zhou, Weiwei Chen Published: 2026 ABSTRACT AbstractBackgroundOver the past few years, 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP)— a compound from grilled or processed meats—has emerged as a major player in cancer development, especially colorectal cancer (CRC). This work dives into its potential links to CRC and uncovers the key genes that bridge this connection.MethodsWe tapped into various databases to pinpoint target genes tied to PhIP and CRC, then ran protein-protein interaction (PPI) analyses for visualization. Next, we explored underlying mechanisms through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment. To nail down predictions, we tested 107 machine learning pipelines and picked the best one, validating its accuracy and the core genes’ prognostic value across datasets. Next, molecular docking and dynamics simulations probed the interactions between these genes and PhIP. Finally, cell proliferation was assessed using Cell Counting Kit-8 (CCK-8) and 5-ethynyl-2’-deoxyuridine (EdU) assays, and polymerase chain reaction (PCR) was performed to validate the expression levels of the hub genes.ResultsOur analysis identified 39 overlapping genes, from which a machine learning model (glmBoost + Enet) identified six candidate targets: CDK4, CEBPB, COMT, SOX9, TIMP1, and TOP2A. To prioritize these, a hierarchical screening framework was applied. Molecular docking and dynamics simulations identified CDK4, COMT, and TIMP1 as the most stable interactors with PhIP. Functional assays confirmed that PhIP treatment significantly enhanced the proliferation of CRC cells. Crucially, quantitative PCR (qPCR) validation in multiple CRC cell lines identified TIMP1 as the primary target, showing the most consistent and significant upregulation upon PhIP exposure.ConclusionsIn essence, these genes drive PhIP is role in CRC, offering novel insights into its molecular pathways. This could reshape how we tackle food-related pollutants, paving the way for better prevention and targeted therapies.Keywords: Colorectal cancer (CRC), network toxicology, molecular dynamics simulation, machine learning Highlight box Key findings • We constructed a machine-learning-based predictive model to evaluate the potential impact of dietary 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP) exposure on colorectal cancer (CRC). • High tumor expression of CEBPB and TIMP1 was associated with poorer survival outcomes. • Structural analyses showed that PhIP docked tightly to CDK4, CEBPB, COMT, SOX9, TIMP1, and TOP2A; molecular dynamics (MD) further supported more stable complexes with CDK4, COMT, and TIMP1. • Proliferation assays demonstrated that PhIP promotes CRC cell proliferation, and reverse transcription quantitative polymerase chain reaction (RT-qPCR) validation highlighted TIMP1 as a core gene mediating this pro-proliferative effect. What is known and what is new? • PhIP, a heterocyclic amine generated during high-temperature meat cooking, is epidemiologically linked to increased CRC risk, but its key molecular targets and pathways remain insufficiently defined. • By integrating machine learning, network toxicology, docking/MD simulations, and experimental validation, we identified a TIMP1-centered signature connected to prognosis and provided mechanistic evidence supporting direct PhIP-protein interactions. What is the implication, and what should change now? • TIMP1 may serve as candidate biomarkers for risk stratification and prognosis in PhIP-related CRC and as potential therapeutic targets for further validation. • The identified binding partners (CDK4, COMT, TIMP1) offer testable mechanistic hypotheses and support both mechanistic follow-up studies and strengthened strategies to reduce dietary PhIP exposure. Introduction Colorectal cancer (CRC) remains a formidable global health challenge, claiming countless lives and driving up healthcare costs. Its roots often lie in a complex mix of environmental and dietary factors (1). CRC was responsible for roughly 9.4% of cancer-related deaths worldwide and made up nearly one in ten cancer cases, trailing only lung cancer in its prevalence (2). This sobering reality underscores the urgent need to dig deeper into what fuels this disease. One culprit that is drawn significant attention is 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP)—a heterocyclic amine born from the sizzle of high-temperature meat cooking (3,4). Studies have consistently linked higher PhIP consumption to a greater risk of CRC, with the danger cutting across both colon and rectal cancer subtypes. Notably, this connection holds firm regardless of whether someone smokes, pointing to PhIP is potent carcinogenic nature. As global meat consumption climbs, finding ways to curb PhIP exposure—through smarter cooking techniques or dietary shifts—could be a game-changer for CRC prevention (5,6). In this study, we aimed to externally validate the previously published model from (https://github.com/soochow-Feng/DNA-Methylation-regulator) using independent cohorts from Gene Expression Omnibus (GEO). By calculating the model is predicted risks in these public datasets, we evaluated its discrimination and calibration for the overall survival (OS). Simultaneously, we leveraged network toxicology, molecular docking, and molecular dynamics (MD) simulations to dissect how PhIP and its associated proteins may influence CRC pathways, thereby providing mechanistic insight into the model is biological rationale (7). It is a holistic approach, letting us systematically probe how target proteins interact and predict the molecular mischief caused by toxins. Meanwhile, molecular docking helps us visualize how PhIP binds to key proteins, revealing its affinity and preferred docking sites, while MD simulations dive into the nitty-gritty of how these interactions evolve over time and affect protein function (8). Together, these methods aim to shed new light on PhIP is role in CRC, paving the way for better safety assessments and preventive strategies. By exploring patterns seen in combination therapies, we also hope to uncover ways to boost therapeutic outcomes, offering a fresh angle on tackling this pervasive disease. We present this article in accordance with the MDAR and TRIPOD reporting checklists (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0832/rc). Methods Experimental design To explore PhIP is toxic potential, we kicked things off by assessing its toxicity using an online prediction tool. Next, we dove into a network toxicology framework to pinpoint PhIP and CRC targets, pulling data from publicly accessible databases. For a closer look at CRC-specific gene activity, we analyzed the GSE52060 dataset, focusing on genes upregulated in tumor tissues. Using R software (R 4.3.1), we cross-referenced three target sets—PhIP, CRC, and differentially expressed genes—to find their overlap. These shared targets became the foundation for predictive modeling, where we tested 12 algorithms across 107 machine learning pipelines. The model with the highest C-index won out, and its genes were deemed our candidate targets. To wrap up, we ran molecular docking between these candidate targets and PhIP, followed by MD simulations on the tightest-binding complexes to confirm how stably PhIP interacts with CRC-related proteins. In short, this pipeline aims to map out PhIP is carcinogenic risks and its molecular targets (Figure 1). Integrative workflow for this study. CRC, colorectal cancer; GEO, Gene Expression Omnibus; PhIP, 2-amino-1-methyl-6- phenylimidazo[4,5-b]pyridine. Acquisition of PhIP chemical composition We sourced PhIP is molecular formula and its 2D and 3D structures from trusted databases like PubMed (https://pubmed.ncbi.nlm.nih.gov/) and PubChem (https://pubchem.ncbi.nlm.nih.gov/). Toxicity analysis and prediction To gauge PhIP is safety risks, we turned to ProTox 3.0 (https://tox.charite.de/protox3), a handy online platform for predicting small-molecule toxicity. By inputting PhIP is SMILES structure, we got a detailed breakdown of its toxicological profile, covering various endpoints that hint at potential hazards. We then used the Swiss Target Prediction database (http://swisstargetprediction.ch/) to predict PhIP is targets in humans, selecting “Homo sapiens” as the species. To keep things clean, we removed duplicate targets and standardized their names using the UniProt database (https://www.uniprot.org/). Disease-related target screening We hunted for CRC-relevant genes by searching “CRC” in the GeneCards (https://www.genecards.org/) and OMIM (https://www.omim.org/) databases. To ensure we were zeroing in on the most relevant players, we set a relevance score cutoff of 10, cherry-picking genes above this threshold to build our CRC target library. Acquisition of differential genes between CRC and adjacent tissues We downloaded the GSE52060 dataset from the GEO database (https://www.ncbi.nlm.nih.gov/geo/) and based on the available clinical data, classified the samples into 23 tumor tissues and 23 corresponding adjacent normal tissues (46 samples in total). The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Using the Limma package in R, we performed differential expression analysis to identify genes upregulated in tumors (P value <0.05, logFC >1). We then overlapped these with PhIP is targets to pinpoint genes that might drive PhIP-related CRC progression. A Venn diagram helped us visualize this intersection, making the connections crystal clear. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis To understand what our intersected genes (PhIP, CRC, and differentially expressed) are up to, we ran GO and KEGG pathway enrichment analyses using R packages like “GeneTerProfiler”, “Enrichplot”, and “Org.Hs.eg.db” (9). These tools revealed the genes’ roles in biological processes, molecular functions, and cellular components, plus the signaling pathways they influence. To make the results pop, we created bar charts for a more intuitive view of the data. Screening of core targets for PhIP-CRC and construction of the PPI network We explored PPI among the intersected PhIP and CRC genes using the STRING database (https://cn.string-db.org/), setting a confidence threshold of ≥0.4 to capture meaningful connections. The resulting network was visualized with Cytoscape 3.9.1, giving us a clear map of how these proteins talk to each other (10). Development of a machine learning-based prediction model for PhIP-CRC To craft a robust diagnostic model, we used a diagnostic framework published on GitHub that comprises 12 core machine-learning algorithms. By combining these algorithms in multiple configurations and tuning their hyperparameters, we generated 107 distinct pipelines and evaluated them to identify robust configurations. The 12 algorithms include Lasso, Ridge, RandomForest, XGBoost, NaiveBayes, among others. The implementation was used as published (https://github.com/soochow-Feng/DNA-Methylation -regulator). We combined multiple distinct sample datasets for the final model following established methods (11,12). GSE39582 was used as the training dataset, while GSE23878, GSE33113, and GSE52060 served as independent validation cohorts (Table 1). All cohorts were reconstructed using the same 39 selected genes. Inclusion criteria: for the disease groups, we included all samples that had available OS information, regardless of other clinical covariates. Samples lacking OS data were excluded from analyses involving survival outcomes to ensure consistency in calculating prognostic performance metrics such as receiver operating characteristic (ROC) and area under the curve (AUC). Following reconstruction, models that included fewer than three genes were excluded before proceeding with the evaluation, and we calculated model- and gene-level ROC and AUC values for each cohort, visualizing the results via heatmap. Table 1Sample counts by GEO accession, cohort (training/validation) and group (tumor/normal)GEO accessionCohortGroupNumberGSE39582Training setTumor566Normal19GSE23878Validation setTumor35Normal24GSE33113Validation setTumor90Normal6GSE52060Validation setTumor23Normal23GEO, Gene Expression Omnibus. Sample counts by GEO accession, cohort (training/validation) and group (tumor/normal) GEO, Gene Expression Omnibus. Survival analysis of candidate targets For survival analysis, we used the GEPIA2 tool and pulled colorectal adenocarcinoma (COAD) and rectal adenocarcinoma (READ) cases from the The Cancer Genome Atlas (TCGA) database. We split our five candidate targets into high- and low-expression groups based on median expression levels, with OS as the endpoint (13). Molecular docking To investigate how PhIP interacts with CRC candidate targets, we employed molecular docking using AutoDock 1.5.6, with PyMOL for visualization (14,15). We retrieved protein structures from the PDB and AlphaFold databases (https://alphafold.ebi.ac.uk/) and converted PhIP’s SMILES from PubChem into mol2 format, optimizing its energy with Chem3D. Docking was performed with a binding energy cutoff of ≤−5.0 kcal/mol to filter significant interactions (16). MD simulation of candidate targets To get a closer look at how PhIP interacts with CRC-related proteins, we turned to MD simulations—a computational powerhouse that lets us watch molecular systems dance and evolve in real time. This approach helped us unpack the intricate ways PhIP binds to its targets, shedding light on the physical and chemical shifts that might fuel CRC. By zooming in at the atomic level, MD simulations revealed the nitty-gritty of intermolecular interactions: bonds forming and snapping, molecules twisting and sliding, all of which hint at how PhIP might disrupt protein structures and functions critical to CRC (17). We relied on Gromacs, a go-to tool for molecular simulations, to build our setup. This involved minimizing energy, balancing temperature and pressure, and tracking the system is dynamic behavior. Before diving into the simulations, we used AutoDock Vina to predict how PhIP docks with its protein partners, giving us a starting point for the MD runs. With Gromacs 2022, we simulated 100 nanoseconds of molecular action, applying the Amber99sb-ildn force field to model the proteins’ behavior (18), and the GAFF force field, via Sobtop software, to shape PhIP is ligand topology (http://sobereva.com/soft/Sobtop) (19). The protein-PhIP complex was placed in a cubic box with periodic boundaries, surrounded by TIP3P water molecules to mimic a realistic environment. To handle electrostatic interactions, we used the particle-mesh Ewald (PME) method paired with the Verlet algorithm. The system went through 50,000 steps of NPT (constant number, pressure, and temperature) and NVT (constant number, volume, and temperature) equilibration, each with a 2 fs coupling constant over 100 ps, ensuring everything was stable. Van der Waals and Coulombic interactions were calculated with a 1.2 nm cutoff. The simulation itself ran at a steady 300 K and 1 bar, using a 2 fs timestep across 50 million steps to cover the full 100 ns. Cell lines and reagents In this study, we worked with two human CRC cell lines, HT-29 (ATCC HTB-38) and Caco-2 (ATCC HTB-37), which were grown as monolayers in a controlled environment at 37 ℃ with a humid atmosphere and 5% CO2. For the culture medium, we opted for either DMEM or McCoy’s 5A, each enriched with 10% fetal bovine serum, 100 units/mL penicillin, and 100 µg/mL streptomycin to support cell growth. We sourced PHIP from Shanghai Yuanye and prepared a 10 mmol/L stock solution by dissolving it in a small amount of DMSO—just 0.1 of the total volume—before topping it up with 0.9 volumes of the medium to get the right concentration. Cell proliferation assays In our experiments, we seeded HT-29 and Caco-2 cells into 96-well plates, each well containing 100 µL of culture medium. We set up five replicate wells per experimental group and included a negative control with no cells. After letting the cells settle for 24 hours, we introduced PHIP at varying concentrations (0.01, 0.1, 1, 10, and 100 µM) and treated the cells for 48 hours. Following this, we incubated the plates in the dark at 37 ℃ for 2 hours, then used a Bio-Rad microplate reader to measure the optical density (OD) at 450 nm, following the manufacturer is guidelines. To assess cell response, we calculated inhibition and survival rates by comparing the OD values of treated groups to those of the blank control. For the EdU assay, we plated cells in 24-well plates, then fixed and stained them. We treated the cells with 50 µM EdU for 30 minutes, followed by a 10-minute incubation with 1 µg/mL DAPI. Afterward, we examined and photographed the cells under a fluorescence microscope to capture the results. For the colony formation assay, we prepared a single-cell suspension of 1000 HT-29 or Caco-2 cells in complete medium and distributed them into 6-well plates. After three weeks, we stained the visible colonies with crystal violet and took photos to document the outcomes. All the above experiments were performed in triplicate independently. Real-time PCR Total RNA was extracted from TRI GENE Plus cells. Subsequently, complementary DNA (cDNA) was synthesized using a high-capacity cDNA conversion kit, followed by quantitative measurement using SYBR Green qPCR. Relative gene expression levels were calculated using the 2−ΔΔCT method. Complementary RNA levels of CDK4, COMT, and TIMP1 were calibrated against β-actin. All the above experiments were performed in triplicate independently. Statistical analysis OS analysis was performed using the GEPIA2 database. Molecular docking was conducted using AutoDock 1.5.6, and MD simulations were carried out using GROMACS. All other bioinformatics analyses were performed using R software. Statistical analyses were conducted using GraphPad Prism 10.1. Data are presented as the mean ± standard error of the mean (SEM). Unpaired t-tests were used for pairwise comparisons. All statistical charts were generated using GraphPad Prism 10.1. P value <0.05 was considered statistically significant. Results Toxicity assessment of PhIP To assess the toxicity of PhIP, we employed ProTox-III, which categorized the compound as toxicity grade 4. PhIP demonstrated notable toxicity in several domains, including carcinogenicity, organ damage, and respiratory effects, with a striking 94% likelihood of carcinogenic potential standing out as the most significant concern. These findings are illustrated in Figure 2A. Bioinformatics analysis reveals molecular mechanisms linking PhIP exposure to colorectal cancer. (A) Toxicity analysis of PhIP. (B) Differential analysis of GSE52060. (C) Common targets of CRC, PhIP, and differential genes. (D) PPI network constructed by common targets. (E) Gene ontology term analysis plot of common targets. (F) KEGG analysis of potential targets. CRC, colorectal cancer; GEO, Gene Expression Omnibus; KEGG, Kyoto Encyclopedia of Genes and Genomes; PhIP, 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine; PPI, protein-protein interaction. Target selection for PhIP and CRC We used SwissTargetPrediction to predict potential targets for PhIP, resulting in a list of 772 candidates. For CRC, we compiled 1,565 relevant targets from OMIM and GeneCards, carefully merging and removing duplicates. Analysis of the GSE52060 dataset pinpointed 758 targets as potential risk factors for CRC progression (see Figure 2B). By integrating these datasets and eliminating redundancies, we narrowed down to 39 shared targets that likely mediate PhIP is role in CRC development. A Venn diagram in Figure 2C visually captures this overlap. PhIP-CRC protein-protein interaction (PPI) network construction We constructed a PPI network for the 39 shared PhIP-CRC targets using the STRING database, setting a confidence threshold of ≥0.4. After filtering out isolated nodes, all 39 targets remained (Figure 2C). Visualized in Cytoscape 3.9.1, the network ranks targets by maximal clique centrality (MCC), with deeper colors and larger nodes indicating stronger connections (Figure 2D). GO and KEGG enrichment analysis of targets From the intersection of PhIP, CRC, and differentially expressed genes, we identified 39 overlapping genes. GO analysis revealed that these genes primarily regulate processes like mitotic cell cycle transitions, nuclear division, and chromosome organization. Meanwhile, KEGG pathway analysis highlighted enriched pathways, including cell cycle, human T-cell leukemia virus 1 infection, IL-17 signaling, cellular senescence, microRNAs in cancer, and TNF signaling. These results suggest that PhIP may drive CRC onset and progression by disrupting cell cycle dynamics, mitosis, and oncogenic pathways. See Figure 2E,2F for detailed visualizations. Machine learning-based PhIP-CRC predictive model development Using the 39 overlapping genes, we developed a predictive model for PhIP-related CRC. After evaluating 107 combinations of machine learning pipelines across 12 algorithms, the glmBoost+Enet[alpha=0.1] model emerged as the top performer, achieving an impressive average concordance index (C-index) of 0.985—outshining all other models (Figure 3A; https://cdn.amegroups.cn/static/public/tcr-2026-0832-1.xlsx, https://cdn.amegroups.cn/static/public/tcr-2026-0832-2.xlsx). Predictive model based on common genes validated across multiple datasets. (A) Development of a predictive model based on common genes. (B) Boxplot of differential genes of candidate genes in the training set for the optimal model. (C) Boxplot of differential genes of candidate genes in the validation set GSE23878 for the optimal model. (D) Boxplot of differential genes of candidate genes in the validation set GSE33113 for the optimal model. (E) Boxplot of differential genes of candidate genes in the validation set GSE52060 for the optimal model. AUC, area under the curve. Predictive model construction and validation The glmBoost+Enet[alpha=0.1] model was built around six candidate targets: CDK4, CEBPB, COMT, SOX9, TIMP1, and TOP2A. To validate its performance, we conducted differential expression analysis across three test datasets, confirming that all six genes were significantly overexpressed in tumor tissues compared to healthy ones (Figure 3B-3E). ROC curves for each dataset underscored the model is robust diagnostic potential (Figure 4A-4D). Model performance evaluated by ROC curves and survival analysis of key genes. (A) ROC curves of candidate genes in the training set for the optimal model. (B) ROC curves of candidate genes in the validation sets GSE23878 for the optimal model. (C) ROC curves of candidate genes in the validation set GSE33113 for the optimal model. (D) ROC curves of candidate genes in the validation set GSE52060 for the optimal model. (E) Survival analysis of TIMP1 in CRC patients. (F) Survival analysis of CEBPB in CRC patients. AUC, area under the curve; CI, confidence interval; CRC, colorectal cancer; HR, hazard ratio; ROC, receiver operating characteristic. Survival analysis of candidate targets We performed survival analysis using the TCGA COAD and READ datasets via GEPIA2, grouping samples by median gene expression. Elevated levels of CEBPB and TIMP1 were linked to worse survival outcomes (Figure 4E,4F). Molecular docking of PhIP with CRC candidate targets To explore how PhIP interacts with CRC-related proteins, we conducted molecular docking using AutoDock 1.5.6. The six candidate targets—CDK4, CEBPB, COMT, SOX9, TIMP1, and TOP2A—were docked with PhIP, all showing low binding energies (affinity <−5 kcal/mol), indicative of strong interactions. These findings suggest PhIP plays a significant role in CRC is molecular mechanisms. Binding modes and interaction sites were visualized using PyMOL for clarity (Figure 5, Table 2, https://cdn.amegroups.cn/static/public/tcr-2026-0832-3.xlsx). Molecular docking of candidate targets with PhIP. PhIP, 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine. Table 2Top five binding energies (kcal/mol) from molecular docking, RMSD lower bound (rmsd l.b.), and RMSD upper bound (rmsd u.b.)TargetBinding energy (kcal/mol)rmsd l.b.rmsd u.b.CDK4−8.300−8.21.1122.277−7.92.3483.077−7.62.5933.877−7.522.757CEBPB−5.400−5.279.44281.439−5.212.47914.375−5.110.41511.149−4.929.11930.62COMT−8.400−8.22.2495.435−7.79.90910.993−7.72.8244.654−7.73.7395.611SOX9−6.300−6.33.7116.136−6.310.05711.133−6.20.942.281−6.21.9932.89TIMP1−7.400−7.416.83219.26−716.68819.42−6.820.06622.466−6.618.97121.368TOP2A−7.300−767.98970.492−770.1472.147−6.63.8826.052−6.452.754.985RMSD, root mean square deviation. Top five binding energies (kcal/mol) from molecular docking, RMSD lower bound (rmsd l.b.), and RMSD upper bound (rmsd u.b.) RMSD, root mean square deviation. MD simulation of candidate targets sites To further validate PhIP is binding affinity, we performed MD simulations for five candidate targets (CDK4, COMT, SOX9, TIMP1, TOP2A) with binding affinities <−6 kcal/mol. Conformational stability was assessed via root mean square deviation (RMSD), where lower values reflect greater stability. The PhIP complexes with CDK4, COMT, and TIMP1 stabilized after 80 ns, while those with SOX9 and TOP2A showed more structural variability. The PhIP-CDK4 complex was particularly stable, with an RMSD below 0.4 nm. Root Mean Square Fluctuation (RMSF) analysis indicated lower flexibility in the PhIP-CDK4, PhIP-COMT, and PhIP-TIMP1 complexes (most residues <1 nm, except at the C- and N-termini), whereas PhIP-SOX9 and PhIP-TOP2A complexes were more flexible. Radius of gyration (Rg) analysis suggested tighter protein conformations upon PhIP binding to COMT, SOX9, TIMP1, and TOP2A, with PhIP-CDK4 maintaining consistent Rg over the 100 ns simulation, indicating minimal structural shifts (Figure 6). Molecular dynamics analysis of pro tein-ligand complexes using RMSD, RMSF, and Rg. (A-E) RMSD values of the protein-ligand complexes over time. (F-J) RMSF values over time of the amino acid backbone atoms of the protein-ligand complexes. (K-O) Rg values for the protein-ligand complexes over time. PhIP, 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine; Rg, radius of gyration; RMSD, root mean square deviation; RMSF, root mean square fluctuation. The MM-PBSA calculations revealed a clear affinity gradient: CDK4 > TIMP1 > COMT > TOP2A > SOX9. CDK4 and TIMP1 emerged as the most stable candidates, yielding remarkably low binding energies of −17.20 and −16.93 kcal/mol, respectively (https://cdn.amegroups.cn/static/public/tcr-2026-0832-4.xlsx). This stability is further supported by their consistent hydrogen bonding (Figure S1A-S1E) and well-equilibrated SASA (Solvent Accessible Surface Area) profiles (Figure S1F-S1J), suggesting a robustly buried binding state. Their high affinity is predominantly driven by van der Waals interactions (approx. −29 kcal/mol), reflecting a tight hydrophobic fit and high structural complementarity (Figure S1K-S1O). Interestingly, while COMT exhibited massive electrostatic attraction (−154.25 kcal/mol) and rapid SASA stabilization, much of this gain was offset by a substantial polar desolvation penalty. On the other end, the binding of TOP2A was hampered by unfavorable electrostatics (+5.03 kcal/mol) and significant SASA fluctuations. The weakest performance was observed in SOX9 (−3.77 kcal/mol), where sparse hydrogen bonding and a high standard deviation point to a lack of conformational stability, aligning with its inability to reach a steady-state SASA during the simulation (Figure S1, https://cdn.amegroups.cn/static/public/tcr-2026-0832-4.xlsx). PhIP promotes the survival and proliferation of CRC cells in vitro Subsequently, the results were validated through in vitro experiments (Figure 7). When we treated cells with varying PHIP concentrations and compared them to a control group, we noticed a marked uptick in proliferation rates over 48 hours. The effect was particularly striking at a 0.1 µM concentration, where the growth curve showed a clear acceleration (Figure 7A,7B). PhIP inhibits Caco-2 and HT-29 cell proliferation, colony formation, and gene expression. (A,B) The effect of different concentrations of PhIP on the proliferation of Caco-2 and HT-29 cells over 48 hours. (C,D) The 0.1 µM concentration of PhIP in the colony formation assay for Caco-2 and HT-29 Cells. (E,F) The PCR experiment on candidate genes (TIMP1, COMT, CDK4) in Caco-2 and HT-29 cells treated with 0.1 µM concentration of PHIP. (G-J) The EdU cell. Magnification: 10×. *, P<0.05; **, P<0.01; ***, P<0.001. PCR, polymerase chain reaction; PhIP, 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine. To dig deeper, we ran cell cloning assays (Figure 7C,7D), and EdU proliferation assays (Figure 7G-7J) with 0.1 µM PHIP. The results were telling: treated cells formed more colonies and showed a higher count of EdU-positive cells, strongly supporting PHIP is role in boosting proliferation. PhIP promotes the expression of key genes in CRC To further explore PhIP is effects, we extracted RNA from HT-29 and Caco-2 cell lines treated with 0.1 µM PhIP and performed PCR analysis to check the expression of three genes—COMT, CDK4, and TIMP1—which had shown strong correlations in our MD simulations. In the HT-29 cells treated with PhIP, we saw a notable boost in TIMP1 and CDK4 expression, but COMT levels took a significant dip (P<0.05). Meanwhile, in the Caco-2 cells exposed to PhIP, TIMP1 and COMT expression surged, whereas CDK4 levels dropped considerably (P<0.05) (Figure 7E,7F). Discussion This study dives into the intricate molecular mechanisms by which PhIP, a dietary carcinogen, may drive CRC. We employed a multifaceted bioinformatics and computational approach to map out these connections. Our process began with extracting PhIP- and CRC-related target genes from databases like PubChem, GeneCards, OMIM, and the Swiss Target Prediction. We then analyzed CRC datasets from the GEO, pinpointing genes significantly upregulated in CRC pathogenesis. By cross-referencing these with PhIP-related targets, we honed in on a select group of candidate genes central to PhIP is carcinogenic effects. To build predictive models, we tested 12 machine learning algorithms, running each 107 pipelines to ensure robust and reproducible results. The top-performing model is genes were subjected to molecular docking and dynamics simulations to probe their interactions with PhIP metabolites. Our results highlight a core set of genes with remarkable predictive power for PhIP-induced CRC. These genes showed strong binding to PhIP is reactive intermediates and maintained stable interactions under simulated physiological conditions. Pathway enrichment analyses further tied these genes to critical oncogenic processes, painting a detailed picture of how dietary PhIP exposure may fuel colorectal tumorigenesis. This framework offers promising biomarkers for early detection, risk assessment, and targeted therapies in CRC. Our findings resonate with global research on PhIP is genotoxic and carcinogenic properties. Animal studies, for instance, show that PhIP exposure boosts intestinal tumor rates, underscoring its role in tumor initiation and progression (20). Additionally, PhIP also tweaks signaling pathways like Wnt/β-catenin and MAPK, further driving malignant transformation (21,22). PhIP, a common heterocyclic amine, emerges when meat or fish is cooked at high temperatures—think grilling, frying, or roasting above 150 ℃. You’ll find it in concentrations anywhere from 0.56 to 69.2 ng/g, and the longer or hotter the cooking, the more PhIP builds up. It is especially prevalent in grilled or fried meats and their juices, making it the most significant cancer-causing HCA in our diets. Studies show that in a group of 10 healthy volunteers, urinary PhIP excretion ranged from 0.12 to 1.97 ng over 24 hours. Daily intake varies widely by region: in Japan, it is estimated at 0.005–0.3 pg per person, while in the US, it can hit 16.6 ng/kg/day, reflecting differences in how we cook and what we eat. Furthermore, PhIP triggers tumors in organs like the colon, breast, liver, lung, skin, and bladder. Its cancer-causing potential comes from PhIP-DNA adducts, which mess with DNA, often causing guanine deletions in the 5’-GTGGGAT-3’ sequence of the Apc gene. This, along with microsatellite instability, can disrupt DNA repair and pave the way for cancer (23). Epidemiological data suggests that regularly eating meat cooked at high temperatures could raise CRC risk by 2 to 6 times. Interestingly, hair analyses show higher PhIP levels in women, possibly tied to metabolism or diet. Exposure varies globally—16.6 ng/g in Japan, 39.42 ng/g in the US, and 44.8 ng/g in Uganda—driven by distinct culinary habits. PhIP metabolites also form DNA adducts in human tissues, hinting at elevated risks for breast, colorectal, and prostate cancers (24). The structural basis for the regulatory influence of PhIP in CRC was explored by examining its binding characteristics with several core proteins, focusing on hydrogen bonding and site-specific stability. Our docking simulations first identified a high-affinity interaction with CDK4, where PhIP forms a stable hydrogen bond at the VAL-96 residue. Since VAL-96 is located within the hinge region—a crucial segment for adenosine triphosphate (ATP) binding and the kinase is catalytic function—this binding likely compromises CDK4 is ability to drive cell proliferation (25,26). Given that CDK4 overexpression is a hallmark of aggressive tumor growth and metastasis in CRC, particularly in microsatellite-stable tumors, PhIP is targeting of this hinge residue represents a potential mechanism for disrupting oncogenic signaling and the epithelial-mesenchymal transition (27). In contrast, the interaction between PhIP and CEBPB appeared much more marginal. This complex exhibited the weakest affinity among all tested targets, with a binding energy failing to reach the −6 kcal/mol threshold. Furthermore, no conventional hydrogen bonds were identified, suggesting that any association between PhIP and CEBPB is likely transient or mediated by weak non-covalent forces. While CEBPB is known to fuel inflammation and tumor survival in late-stage CRC (28,29), these docking results suggest that CEBPB may not be a primary direct target for PhIP, or perhaps its regulation occurs through more complex, indirect cellular mechanisms rather than high-affinity structural binding. In the case of SOX9, another critical transcription factor, PhIP was found to form a hydrogen bond with the GLY-97 residue. This structural association is particularly relevant as SOX9 is frequently overexpressed in advanced CRC, where it serves as a valuable prognostic marker and a driver of metastasis (30). Beyond promoting tumor growth and invasion, SOX9 is instrumental in sustaining cancer stem cell traits and facilitating immune evasion within the tumor microenvironment. Given its involvement in chemotherapy resistance (31), the interaction between PhIP and the GLY-97 residue may interfere with these oncogenic functions, potentially offering a way to modulate the aggressive phenotype associated with SOX9 elevation. A much stronger structural tethering was observed for TIMP1, where the binding of PhIP is anchored by hydrogen bonds with GLU-51 and TYR-58. These specific interactions are of interest because elevated TIMP1 levels in CRC plasma and tissues are closely linked to lymph node metastasis and reduced survival (32). Functionally, TIMP1 does more than inhibit proteases, it activates the PI3K/AKT signaling pathway to support cancer stem cell-like properties and cell survival (33). By engaging these specific residues, PhIP may influence TIMP1 is ability to remodel the extracellular matrix (ECM) and drive the gene expression profiles required for metastasis (34). The interaction with TOP2A further underscores the multi-targeted potential of PhIP, characterized by hydrogen bonds with residues GLN-542 and ASP-543. TOP2A is indispensable for resolving DNA supercoiling during replication; however, its overexpression in CRC often leads to the uncontrolled division and genomic instability that define malignancy (35). As a marker tied to advanced disease stages and lymph node involvement, TOP2A is role in driving invasive capacity—often through altered ECM and EMT-related gene expression—makes it a significant target (36). The binding of PhIP to these specific residues could potentially disrupt DNA topological homeostasis, thereby blunting the metastatic drive of CRC cells. Lastly, the metabolic enzyme COMT showed a distinct binding pattern involving hydrogen bonds at ILE-139 and SER-169. COMT is essential for catecholamine and estrogen metabolism and is frequently aberrantly expressed in advanced CRC, correlating with poor prognosis (37). By targeting these residues, PhIP may interfere with the methylation of catechol estrogens, thereby impacting estrogen-mediated signaling which modulates proliferation in certain CRC subtypes. This metabolic shift could also influence oxidative stress and inflammation within the tumor microenvironment, potentially disrupting the niche that facilitates tumor progression (38). Collectively, while the affinity for CEBPB remains weak, the robust binding patterns observed with CDK4, SOX9, TIMP1, TOP2A, and COMT provide a structural rationale for how PhIP might exert a broad regulatory effect across diverse oncogenic pathways in CRC. Molecular docking revealed that all six key targets—CDK4, CEBPB, COMT, SOX9, TIMP1, and TOP2A—bind specifically to PhIP. MD simulations further showed that PhIP complexes with CDK4, COMT, and TIMP1 exhibit notable stability. These results support the idea that PhIP directly interacts with these genes, altering their functions and potentially driving CRC development. KEGG pathway analysis highlighted the cell cycle pathway as particularly enriched, emphasizing its role in regulating cell survival and proliferation in PhIP-induced CRC. Survival analysis showed that higher CEBPB and TIMP1 expression correlates with worse outcomes, likely due to their roles in enhancing tumor proliferation, invasion, metastasis, immune evasion, and angiogenesis (39,40). These proteins not only hold prognostic value but also present potential targets for innovative therapies. The integration of network toxicology and experimental validation provides a holistic view of PhIP-induced colorectal carcinogenesis. Our KEGG analysis identified the Cell Cycle and TNF signaling as the most significantly enriched pathways. Notably, these pathways are mechanistically linked to our validated hub genes. For instance, CDK4, a core component of the cell cycle machinery identified through machine learning, showed high binding affinity to PhIP in MD simulations, suggesting a direct impact on mitotic transitions (41,42). Furthermore, the significant upregulation of TIMP1 in PhIP-treated CRC cells, combined with the enrichment of the TNF signaling pathway, indicates that PhIP may facilitate a pro-tumorigenic inflammatory microenvironment (43,44). Together, these findings suggest that PhIP exerts its carcinogenic effects by synergistically disrupting cell cycle homeostasis and activating inflammatory signaling via key nodes such as CDK4 and TIMP1. This study is strengths lie in its integrative approach. By combining network toxicology, machine learning, bioinformatics, and big data analytics, we’ve developed a novel way to quickly uncover complex links between environmental pollutants and diseases. Molecular docking and dynamics simulations offered a close-up view of chemical-biological interactions at the molecular level, shedding light on their mechanisms and dynamics. The comprehensive methodology—spanning network analysis, predictive modeling, docking, and simulations—bolsters the study is scientific rigor and reliability, enabling rapid evaluation of understudied contaminants. Despite the insights provided, several limitations of this study should be acknowledged. First, PhIP is a well-characterized pro-carcinogen that requires metabolic activation, typically mediated by cytochrome P450 enzymes (CYP1A2), to form highly reactive intermediates capable of DNA binding and oncogenic transformation (20,45). Since our target prediction was primarily based on the parent structure of PhIP, it is possible that potential interactions between its active metabolites and specific carcinogenic-related proteins were not fully captured. Second, while we employed a hierarchical screening workflow and validated the mRNA expression of the hub gene TIMP1 via qPCR in two CRC cell lines, the experimental evidence remains preliminary. The expression levels of other predicted targets (CDK4 and COMT) and the precise downstream molecular mechanisms through which PhIP exerts its effects via these genes require further exploration through functional assays and in vivo animal models. Lastly, the reliance on predictive network databases may introduce algorithmic biases. Future studies integrating metabolomics and proteomics are warranted to provide a more comprehensive landscape of PhIP-induced carcinogenesis. Looking ahead, epidemiological studies are needed to solidify the link between PhIP exposure and CRC incidence. Longitudinal research tracking exposure and disease development over time will be crucial for establishing causality. Such efforts will help validate key targets and pathways, paving the way for preventive and therapeutic strategies to counter PhIP is harmful effects. This holistic research approach will deepen our understanding of the health impacts of dietary and environmental pollutants, providing timely, evidence-based support for tackling emerging public health challenges. Conclusions In summary, this study employed an integrated framework—combining network pharmacology, machine learning, and multi-level computational simulations—to systematically investigate the oncogenic mechanisms of PhIP in CRC. Through a hierarchical screening process, TIMP1 was identified and validated as a primary target among six initial candidates (CDK4, CEBPB, COMT, SOX9, TIMP1, and TOP2A). Our experimental findings demonstrate that PhIP promotes CRC cell proliferation and significantly upregulates TIMP1 expression, highlighting its role as a critical mediator of PhIP-induced malignancy. While the other candidate genes showed strong computational potential, future studies are warranted to fully elucidate their individual biological contributions. These findings offer fresh insights into environmental health risks and position TIMP1 as a promising biomarker and therapeutic target for PhIP-associated CRC. Supplementary The article’s supplementary files as Acknowledgments We thank all the researchers who shared public data and the patients who participated in the sequencing efforts. Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Footnotes Reporting Checklist: The authors have completed the MDAR and TRIPOD reporting checklists. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0832/rc Funding: This work was supported by the Guizhou Provincial Basic Research Program (Natural Science) (Nos. [2022]340 and zk[2025]475, to W.C.) and the Cultivation Project of Guizhou Medical University (No. 20NSP040 to W.C.). Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0832/coif). The authors have no conflicts of interest to declare. Data Sharing Statement Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0832/dss References Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0832/prf Supplementary Materials The article’s supplementary files as Data Availability Statement Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0832/dss