Aspartame Drives the Continuous Progression from MASLD to HCC: An Integrated Analysis

Aspartame Drives the Continuous Progression from MASLD to HCC: An Integrated Analysis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Aspartame Drives the Continuous Progression from MASLD to HCC: An Integrated Analysis Xusheng Zhang, Rong Tan, Yongxin Ma, Qi Wang, Bnedong Chen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9171872/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 14 You are reading this latest preprint version Abstract Background Aspartame (APM), a widely used sweetener, has been linked to cancers, yet its molecular impact on metabolic dysfunction-associated steatotic liver disease (MASLD) and subsequent hepatocellular carcinoma (HCC) remains undefined. We integrated network toxicology, bulk RNA-seq and docking to map the mechanism. Methods APM targets were retrieved from ChEMBL, STITCH and SwissTargetPrediction. MASLD and HCC RNA-seq data from GEO were used to call DEGs. WGCNA identified disease modules and hub genes. Intersection of APM targets, DEGs and hubs defined core genes for GO/KEGG and PPI analyses. CytoHubba (DMNC, EPC, Degree, MCC), LASSO, RF and SVM-RFE shortlisted key genes, and docking verified APM binding. Results Twelve genes intersected across APM, MASLD and HCC datasets. Enrichment supports a “dual-track” mechanism: APM-MASLD targets suppress bile-acid export, impair lipid clearance and fuel steatosis; MASLD-HCC targets jointly activate TNF/IL-17 and chemical-carcinogenesis pathways, indicating chronic inflammation bridges steatosis to cancer; APM-HCC targets map to p53, nuclear-receptor and xenobiotic-response networks, revealing APM hijacks receptor signalling to impose proliferative stress that, coupled with p53 loss, drives clonal selection. Machine-learning nominated EGR1 and PTGS2 as top diagnostic genes (AUC > 0.7); docking showed high-affinity APM binding (–7.1 and–7.9 kcal mol⁻¹, respectively), identifying them as key relays in APM-induced HCC. Conclusions EGR1 and PTGS2 are central nodes through which APM precipitates MASLD and accelerates progression to HCC. We propose a “dual-track” oncogenic paradigm: Track A follows the canonical MASLD-HCC axis (bile-acid retention - lipid deposition - TNF/IL-17-driven ROS-mutational amplification), whereas Track B allows APM, via PTGS2/EGR1, to usurp gate-keeper proteins governing proliferation and apoptosis, initiating malignant programming before overt steatosis develops. These findings provide mechanistic insight into APM-related hepatocarcinogenesis, nominate tractable diagnostic biomarkers and therapeutic targets, and inform future re-evaluation of APM carcinogenicity classifications. aspartame hepatocellular carcinoma metabolic dysfunction-associated steatotic liver disease network toxicology molecular docking Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Introduction Hepatocellular carcinoma (HCC) accounts for roughly four-fifths of primary liver cancer and ranks third among global cancer deaths[ 1 – 3 ]. GLOBOCAN 2022 recorded 389 000 new cases and 331 000 deaths in China—over forty percent of the world total-rising two to three percent each year[ 4 ]. Curative resection or transplantation lifts five-year survival to seventy percent, yet fewer than thirty percent are operable at diagnosis; combined targeted and immunotherapy extends median overall survival from under one year to about twenty months in advanced disease. The aetiological profile is shifting: hepatitis B-related tumours are declining through vaccination and antiviral therapy[ 5 ], while MASLD, renamed from non-alcoholic fatty liver disease (NAFLD) in 2023, now increases at more than five percent annually, contributing thirty to forty percent of new HCC in Western countries and fifteen to twenty percent in China[ 6 ]. The molecular route from MASLD to HCC remains obscure, with no actionable targets[ 7 , 8 ]. This epidemiological rise parallels four decades of widespread sugar substitution. Non-nutritive sweeteners are promoted to counter obesity and type 2 diabetes[ 9 ]; aspartame, one hundred and eighty times sweeter than sucrose, dominates the market. Approved in 1981[ 10 ], it is now present in roughly six thousand products-soft drinks, chewing gum, yoghurt, vitamin chews-with annual global consumption exceeding twenty-five thousand tonnes[ 11 , 12 ]. China’s current food-additive standard still lists aspartame as permissible without quantitative limit, and one third of packaged foods declare its use; teenagers already ingest thirty to fifty percent of the acceptable daily intake. Since rodent lymphoma findings were reported in 2005, the safety of chronic exposure has been debated[ 13 ]. A 2023 International Agency for Research on Cancer classification as “possibly carcinogenic” and a European Food Safety Authority statement that mechanistic evidence for low-dose effects remains incomplete have intensified demands for high-resolution human data[ 14 ]. Current experimental evidence is dominated by rodent studies at doses fifty to two hundred times the acceptable daily intake, whereas primate or human data are scarce; most reports focus on direct DNA damage or oxidative stress, leaving the indirect oncogenic route-metabolic remodelling, chronic inflammation and micro-environmental imbalance-largely unexplored[ 14 – 17 ]. Within the liver, intestinal esterase cleaves aspartame into methanol and phenylalanine, both requiring hepatic metabolism[ 18 ]. Sustained low-level exposure can deplete folate during methanol oxidation, raise homocysteine and promote lipid deposition[ 18 ]; phenylalanine can inhibit the L-arginine-nitric-oxide pathway and aggravate insulin resistance; the methyl ester moiety can consume glutathione, trigger endoplasmic-reticulum stress and activate nuclear factor kappa B and the NLRP3 inflammasome. These events mirror the two-hit hypothesis of MASLD, suggesting that prolonged aspartame intake could initiate a non-viral pathway from steatosis through fibrosis to malignancy[ 19 ]. Traditional cohort studies lack the resolution to track such effects across one or two decades, and animal protocols that add two to four grams per litre to drinking water deliver eight to fifteen times the human acceptable daily intake, limiting translational relevance. We therefore combined population-based liver transcriptomes, computational toxicology and machine-learning algorithms to chart the aspartame-associated molecular network linking MASLD with HCC and to pinpoint nodes amenable to therapeutic intervention. Figure abstract Methods 2.1 Data download and processing The present study obtained HCC (GSE14323, N:91; GSE14520, N:445; GSE25097, N:268; GSE36376, N:240; GSE76427, N:115) and MASLD (GSE126848, N:75; GSE135251, N:96; GSE89632, N:70) transcriptome datasets from the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/ ). All datasets underwent standardized preprocessing, including background correction, gene ID annotation conversion, and data normalization to ensure comparability and analytical accuracy. GeneCards ( https://www.genecards.org/ ) was used to retrieve MASLD- and liver-cancer-related target genes. In parallel, this study integrated ChEMBL ( https://www.ebi.ac.uk/chembl/ ), STITCH ( https://stitch.embl.de ), and SwissTargetPrediction ( https://www.swisstarget -prediction.ch) to systematically search for potential human target genes of APM for subsequent intersection analysis and functional investigation. Target structures of EGR1(PDB ID:4X9E) and PTGS(PDB ID:5F1A) were retrieved from the RCSB Protein Data Bank ( https://www.rcsb.org/ ), the conformational information of APM was obtained from PubChem ( https://pubchem.ncbi.nlm.nih.gov/ ), and molecular docking was performed using CB-Dock2 ( https://cadd.labshare.cn/cb-dock2/index.php ). 2.2 Identification of APM-induced MASLD and HCC targets For the training datasets downloaded from GEO, differential expression analysis between HCCC and normal controls, and between MASLD and normal controls was performed with the R limma package; |log₂FC| > 0.585 and FDR < 0.05 served as screening criteria to identify differentially expressed genes (DEGs). To identify co-expression gene modules and key hub genes related to MASLD and HCC, weighted gene co-expression network analysis (WGCNA) was applied to the training datasets. After the pickSoftThreshold function determined the optimal soft-threshold power β, an adjacency matrix was constructed and converted into a topological overlap matrix (TOM) to quantify gene co-expression similarity. Hierarchical clustering based on the TOM dissimilarity matrix was performed, and the dynamic tree-cut algorithm was used to identify co-expression modules (minimum module size 90, cut height 0.25). Module eigengenes (ME) were calculated to represent the overall expression pattern of each module; Pearson correlation analysis screened modules significantly associated with MASLD and liver cancer (P 0.6 and |GS| > 0.6 were selected as hub genes. Finally, the intersection of the obtained DEGs, WGCNA hub genes, and APM potential target genes was taken to acquire common target genes of APM-induced MASLD and HCC, serving as the candidate gene set for downstream analyses. 2.3 Functional enrichment and PPI network analysis To explore the potential molecular mechanisms by which APM sequentially induces MASLD and liver cancer, functional enrichment and protein-protein interaction (PPI) network analyses were performed on the candidate target genes. GO and KEGG pathway enrichment analyses were conducted with DAVID ( https://david.ncifcrf.gov )[ 20 ]. GO covers Biological Process (BP), Cellular Component (CC), and Molecular Function (MF). Fisher’s exact test evaluated enrichment significance; P < 0.05 was set as the cutoff to identify significantly enriched biological functions and signalling pathways. Protein interaction relationships among candidate targets were retrieved from STRING ( https://string-db.org/ )[ 21 ], imported into Cytoscape for network visualisation, and CytoHubba algorithms (DMNC, EPC, Degree, MCC) were applied to screen core targets and reveal interaction patterns and potential regulatory mechanisms. 2.4 Screening and validation of key target genes To select core genes that play central roles in APM-induced MASLD and liver-cancer development, a multi-machine-learning integration strategy was adopted. LASSO[ 22 ], random forest (RF)[ 23 ], and support-vector-machine recursive-feature elimination (SVM-RFE)[ 24 ] were each used to screen core intersection targets; the three resulting lists were intersected with the CytoHubba (DMNC, EPC, Degree and MCC) output to obtain the final core intersection targets. 2.5 Pathway exploration of core target genes with the KEGG PATHWAY Database We further employed the KEGG PATHWAY Database ( https://www.kegg.jp/kegg/kegg.html ) to map the signalling pathways of the two core differentially expressed genes in liver disease. Results 3.1 Differential expression analysis for MASLD and HCC, and APM target screening GEO2R was applied separately to each included HCC dataset (GSE14323, GSE14520, GSE25097, GSE36376, and GSE76427) for differential expression analysis with thresholds of |log₂FC| > 0.585 and FDR < 0.05, and DEGs for each dataset were obtained as shown in Fig. 1 A. The DEG lists from the five datasets were then intersected, and genes present in any two datasets were retained for further analysis, yielding 1,716 candidate target genes (Fig. 1 B). Subsequently, batch correction was performed across the five datasets, and principal component analysis results before and after batch correction are shown in Fig. 1 C–D. Differential expression analysis between the HCC and normal control (N) groups was conducted on the batch-corrected data using the limma package with |log₂FC| > 0.585 and FDR < 0.05 as selection criteria, resulting in 701 candidate targets (Fig. 1 E). Next, weighted gene co-expression network analysis (WGCNA) was performed on the batch-corrected HCC data. The pickSoftThreshold function was first used to evaluate a range of soft-threshold powers β, and the optimal β = 8 was selected based on the scale-free topology fit index (Fig. 2 A, B). A weighted adjacency matrix was then constructed and transformed into a Topological Overlap Matrix (TOM) to quantify gene co-expression similarity. Subsequently, hierarchical clustering was performed based on the TOM dissimilarity matrix, combined with the dynamic tree cut algorithm, with parameters set as follows: minimum module size = 90, deepSplit = 3, and module merge cut height = 0.25. Ultimately, seven gene co-expression modules with distinct expression patterns were identified, and the correspondence between modules and the gene dendrogram is shown in Fig. 2 C. Among them, the blue module exhibited a strong positive correlation with the disease, and 461 hub genes highly associated with HCC were identified (Fig. 2 D, E). In addition, 910 HCC-related genes were retrieved from GeneCards (Supplementary Material 1). The intersection of the GEO2R results, Limma package analysis results, WGCNA results, and GeneCards retrieval was then performed, with candidate target genes identified in any two of these methods being included in the next step of analysis, resulting in a total of 854 HCC-related genes. The included MASLD sequencing datasets (GSE126848, GSE135251, and GSE89632) were batch-corrected; PCA before and after correction is shown in Fig. S1 A,B. Limma was then applied to the corrected data to compare the MASLD and normal control groups. With thresholds of |log₂FC| > 0.585 and FDR < 0.05, 1,514 MASLD candidate targets were obtained (Fig. 3 A). Next, weighted gene co-expression network analysis (WGCNA) was performed on the batch-corrected MASLD data. pickSoftThreshold evaluated a series of soft-threshold powers β, and the optimal β = 8 was selected based on the scale-free topology fit index (Fig. 3 B,C). A weighted adjacency matrix was constructed and converted into a Topological Overlap Matrix (TOM) to quantify gene co-expression similarity. Hierarchical clustering based on the TOM dissimilarity matrix, combined with dynamic tree cut (minimum module size = 90, deepSplit = 3, merge cut height = 0.25), identified seven co-expression modules; the correspondence between modules and the gene dendrogram is shown in Fig. 3 D. The blue module exhibited the strongest positive correlation with MASLD, yielding 557 hub genes (Fig. 3 E). Intersection of Limma results, WGCNA results, and GeneCards retrieval (Supplementary Material 2), retaining genes detected by any two approaches, produced 390 MASLD-related genes (Fig. 3 F). Next, ChEMBL, STITCH, and SwissTargetPrediction were systematically queried for potential human targets of APM; the union of the three databases yielded 12 APM targets (Fig. 3 G). The intersection of the obtained HCC target genes, MASLD-related hub genes, and APM targets revealed 16 APM–MASLD overlaps, 18 APM–HCC overlaps, 36 MASLD–HCC overlaps, and 12 genes common to all three sets (Fig. 3 H). (A–B) Soft-threshold selection: left panel shows scale-free topology fit index R², right panel shows mean connectivity, red line marks R² = 0.8; (C) Gene clustering dendrogram: colored bars indicate distinct co-expression modules; (D) Module–trait correlation heatmap: numbers are correlation coefficients with P values in parentheses, color indicates correlation strength; (E) Importance analysis of different modules; (F) Intersection of GEO2R results, Limma results, WGCNA clusters and GeneCards yielded HCC target genes for downstream analysis. (A) Heat-map of differential expression results from the batch-corrected MASLD data analysed with the R “Limma” package; (B–C) Soft-threshold selection: left panel shows scale-free topology fit index R², right panel shows mean connectivity, red line marks R² = 0.8; (D) Gene clustering dendrogram: coloured bars represent distinct co-expression modules; (E) Module–trait correlation heat-map: numbers are correlation coefficients with P values in parentheses; colour indicates correlation strength; (F) Intersection of Limma DEG results, WGCNA clusters and GeneCards yielded MASLD-related targets for downstream analysis; (G) Union of APM targets from ChEMBL, STITCH and SwissTargetPrediction; (H) Intersection among APM targets, MASLD signature genes and HCC signature genes. 3.2 Functional Enrichment Analysis of Candidate Targets The 12 key intersection targets of APM-induced MASLD and HCC are shown in Fig. 4 A. Based on the DAVID database, GO and KEGG functional enrichment analyses were first performed on the 12 intersection targets of APM and MASLD to investigate the specific molecular mechanisms involved in APM-induced MASLD. GO analysis results showed that these genes were primarily involved in biological processes such as protein/vesicle localization, response to stimulus, regulation of biological process, negative regulation of biological process, and cellular process (Fig. 4 B). KEGG enrichment analysis revealed that these genes were mainly enriched in signaling pathways including Bile secretion, Lipid and atherosclerosis, TNF signaling pathway, and AGE-RAGE signaling pathway in diabetic complications (Fig. 4 C). Continuing with enrichment analysis of the 36 intersecting target genes between MASLD and HCC, these genes were found to be primarily involved in biological processes such as lipid localization, response to xenobiotic stimulus, and lipid transport; cellular components including membrane raft, endoplasmic reticulum lumen, and RNA polymerase II transcription regulator complex; and molecular functions such as DNA-binding transcription factor binding, RNA polymerase II-specific DNA-binding transcription factor binding, and modified amino acid binding (Fig. 4 D). KEGG enrichment analysis demonstrated that these genes were mainly enriched in signaling pathways including Chemical carcinogenesis-receptor activation, Transcriptional misregulation in cancer, Apelin signaling pathway, Chemical carcinogenesis-DNA adducts, AMPK signaling pathway, TNF signaling pathway, IL-17 signaling pathway, and PPAR signaling pathway (Fig. 4 E). Furthermore, enrichment analysis of the 18 intersecting target genes between APM and HCC revealed that these genes were primarily involved in functions such as carboxylic acid transport, Nuclear receptors meta-pathway, response to xenobiotic stimulus, regulation of lipid transport, regulation of blood vessel endothelial cell migration, osteoblast differentiation, and cell population proliferation. The signaling pathways involved included the PID NFAT TF pathway and p53 transcriptional gene network (Fig. 4 F). These functions and pathways may represent important mechanisms through which aspartame directly induces HCC. (A) Network diagram of APM–MASLD–HCC intersection targets; (B) GO enrichment of APM–MASLD targets; (C) KEGG pathway enrichment of APM–MASLD targets; (D) GO enrichment of MASLD–HCC targets; (E) KEGG enrichment of MASLD–HCC targets; (F) GO and KEGG enrichment of APM–HCC targets. 3.3 Diagnostic model construction Using the 12 intersection targets, 113 machine-learning algorithms and their ensembles were trained on the batch-corrected compilation of the first four HCC datasets (GSE14323, GSE14520, GSE25097, GSE36376) as the training set; each of the four was also used alone as an internal validation set, and GSE76427 served as the external test set. Prediction performance is shown in Fig. 5 A; the “glmBoost + GBM” combination achieved the highest diagnostic efficacy, maintained in both validation and test sets. ROC-AUC curves for training, validation and test sets under this algorithm are displayed in Fig. 5 B–G. Confusion matrices further confirmed high diagnostic performance (Fig. 5 H–M). Expression profiles of the eight genes retained by “glmBoost + GBM” (PTGS2, SLC1A2, SPP1, EGR1, IGFBP2, ACSL1, THBS1, PPARG) are shown in Fig. 6 A. Diagnostic evaluation indicated that, except for SPP1 and THBS1, the remaining six genes exhibit robust diagnostic capacity for HCC and represent potential biomarkers (Fig. 6 B). (A) Heat-map of predictive performance for training, validation and test sets across different machine-learning algorithms; (B–G) ROC-AUC curves for the training set, validation sets (GSE14323, GSE14520, GSE25097, GSE36376) and test set (GSE76427); (H–M) Confusion matrices for the training set, validation sets (GSE14323, GSE14520, GSE25097, GSE36376) and test set (GSE76427). 3.4 SHAP analysis To further verify the diagnostic fidelity of the model, SHAP analysis was performed on the eight modelling genes. SHAP scores were calculated and ranked by absolute value (Fig. 6 C); EGR1 dominated model explainability with a mean|SHAP| of 0.135, followed by IGFBP2, ACSL1, etc. The eight-gene HCC predictor achieved robust performance across five algorithms (RF, SVM, XGB, GBM, KNN; AUC > 0.9, Fig. 6 D). The waterfall plot in Fig. 6 E illustrates that all eight genes contributed positively to the predicted probability of an individual sample, raising it from a baseline of 0.851 to 0.999, underscoring their role as core switches. This is further corroborated by the SHAP feature-importance bar chart (Fig. 6 F). (A) Differential-expression analysis of the eight model genes between HCC and normal tissues; (B) Diagnostic performance of the eight model genes for HCC; (C) SHAP importance bar plot of the model genes; (D) Predictive performance of the HCC model constructed from the eight genes across five algorithms (RF, SVM, XGB, GBM and KNN); (E) Waterfall plot showing how the eight model genes influence the predicted probability for an individual sample; (F) SHAP feature-importance bar chart quantifying the average magnitude and direction of each gene’s contribution to the final prediction. 3.5 Construction of a risk-prediction model We next re-screened the 12 intersecting targets with additional machine-learning approaches. LASSO regression retained seven feature targets (Fig. S2 A). SVM-RFE achieved peak performance when six variables were kept (Fig. S2 B). Random-forest screening, with Ntree = 142 and minimal error, selected genes whose importance score exceeded 10, yielding three feature targets (Fig. S2 C). Protein–protein interactions (PPI) among the 12 intersecting targets were then analyzed with STRING (Fig. 7 A). The network was imported into Cytoscape and interrogated with four CytoHubba algorithms—DMNC (Fig. 7 B), EPC (Fig. 7 C), Degree (Fig. 7 D) and MCC (Fig. 7 E). The hub proteins identified by each method were intersected, producing six core proteins (Fig. 7 F). (A) PPI network of the 12 intersecting targets; (B) Core proteins identified by the DMNC algorithm; (C) Core proteins identified by the EPC algorithm; (D) Core proteins identified by the Degree algorithm; (E) Core proteins identified by the MCC algorithm; (F) Intersection of the results produced by the four built-in CytoHubba algorithms (DMNC, EPC, Degree and MCC). Next, the core targets retained by machine-learning screening were intersected with the core proteins extracted from the PPI network, yielding two final core target genes (Fig. 8 A). A risk-prediction nomogram for APM-induced HCC was then constructed on the basis of these two genes (Fig. 8 B). Calibration curves demonstrated good agreement between predicted and observed probabilities, indicating reliable model performance (Fig. 8 C). Decision-curve analysis (DCA) showed that the model provides additional net benefit within the clinically relevant threshold range of 5–45%, underscoring its potential for translation into a clinical prediction tool (Fig. 8 D). The clinical impact curve revealed that predicted probabilities between 10% and 40% achieve the highest net clinical benefit—detecting the greatest number of early-stage HCC cases with the fewest additional examinations (Fig. 8 E). (A) Intersection of core proteins selected by CytoHubba algorithms and core target genes retained by machine-learning screening; (B) Nomogram of the risk-prediction model built on the two intersecting core target genes; (C) Calibration curve; (D) Decision curve; (E) Clinical impact curve. 3.6 Molecular-docking prediction of APM binding to core targets To verify whether APM can directly bind the core HCC targets, we performed molecular-docking simulations with EGR1 and PTGS2. APM displayed favorable binding affinities for both proteins, with calculated binding energies of − 7.1 kcal mol⁻¹ (EGR1) and − 7.9 kcal mol⁻¹ (PTGS2); absolute values > 7 kcal mol⁻¹ indicate stable complex formation(Fig. 9 ). These data provide preliminary molecular-level evidence that APM can directly engage the identified core target proteins, detailed results are provided in Table S1 . 3.7 Exploration of core-target signaling pathways Using the KEGG PATHWAY Database, we mapped the signaling cascades in which the two core differentially expressed genes are implicated. The results suggest that APM, by targeting PTGS2/COX-2, triggers sustained lipid accumulation, ROS generation, mutational burden and neutrophil infiltration; together these insults push hepatocytes into compensatory proliferation and ultimately drive hepatocellular carcinoma (Fig. 10 ). Discussion As evidence accumulates, three key rodent studies have shaped the debate of APM expose: the Ramazzini Institute reported increased lymphomas and leukaemias[ 25 ]; the NTP diet-based carcinogenicity bioassay observed a rise in hepatocellular carcinomas in female mice; and human data show a “weak but coherent” signal. In the NutriNet-Santé cohort (n = 102 865), 10-year follow-up yielded an overall cancer HR of 1.15 (95% CI 1.03–1.28), with the strongest hint for liver cancer[ 26 ]. Consequently, IARC re-classified APM as Group 2B (“possibly carcinogenic to humans”) in 2023. Yet APM-centric oncogenic research remains scarce. Since FDA approval in 1981 only ~ 40 years have elapsed[ 27 ]; no study has accrued > 15 years of individual-level follow-up. Meanwhile, ubiquitous low-calorie products have silently driven population exposure upward, but investigative effort has not kept pace, and the cumulative impact of ultra-long-term, low-dose APM is still terra incognita. The liver—central to metabolism and xenobiotic clearance—has received disproportionately little attention despite chronic, escalating APM doses. Here, leveraging multi-centre transcriptomes and a machine-learning framework, we chart a molecular roadmap in which low-dose APM propels the sequential evolution “metabolism to inflammation to fibrosis to MASLD to HCC”. We further propose a second, direct oncogenic route that bypasses the steatotic stage and targets hepatocytes per se. Functional enrichment and protein-interaction networks converge on two intersecting yet partly independent signalling clusters: (i) bile-acid efflux blockade, lipid mis-trafficking and TNF/AGE–RAGE cascades that explain the MASLD–HCC continuum; and (ii) p53-network destabilisation and NFAT-driven hyper-proliferation that reveal direct transformation pressure. GO functional enrichment analysis and KEGG pathway enrichment analysis of APM–MASLD intersecting targets identify “bile-acid–lipid metabolic imbalance” as the first hit. KEGG pathway terms such as “protein/vesicle localisation” and “negative regulation of biological process” imply early membrane-trafficking reprogramming. KEGG pathway places “bile secretion” at the top; retained bile acids cripple the FXR–SHP negative loop, raising intrahepatic free cholesterol and sphingolipids-consistent with simultaneous enrichment of “lipid and atherosclerosis”[ 28 ]. By impairing bile-acid export, APM lowers lipid clearance and supplies substrates for droplet deposition—the classical “first hit”[ 29 ]. Intersection genes of MASLD–HCC are simultaneously enriched for “TNF signalling”, “IL-17 signalling” and “chemical carcinogenesis–receptor activation”, implicating chronic inflammation as the bridge between simple steatosis and malignancy[ 30 ]. Mechanistically, APM-derived methanol depletes folate, raises homocysteine and activates NF-κB; phenylalanine inhibits eNOS, reduces NO bioavailability and synergistically elevates ROS. ROS inflicts mtDNA mutations and, via an IL-17C–neutrophil chemotaxis feed-forward loop, continuously recruits neutrophils that release reactive nitrogen species[ 30 ]. Persistent AGE–RAGE signalling glycation maintains this inflammatory milieu, fuels mutation accumulation and provides clonal advantage for malignant transformation. Genes unique to the APM–HCC axis enrich for “p53 transcriptional network”, “nuclear receptors meta-pathway” and “response to xenobiotic stimulus”, suggesting that APM acts as an exogenous ligand that hijacks nuclear-receptor signalling, forcing quiescent hepatocytes into chronic replicative stress. Coupled with p53 loss, this drives clonal selection for malignancy. Molecular docking shows APM–PTGS2/COX-2 binding energy − 7.9 kcal mol⁻¹, compatible with stable enzyme activation. Sustained COX-2 elevation (i) induces c-Myc and Cyclin D1 via the PGE₂–EP₄ axis, propelling cell-cycle traverse, and (ii) suppresses p53 acetylation, blunting transcription of GADD45 and PUMA. The resulting “hyper-proliferation plus disabled genome surveillance” creates a double-edged environment that can launch transformation before overt steatosis appears. This finding aligns with EFSA’s 2023 conjecture that “long-term, low-dose exposure might directly perturb proliferation control” and offers mechanistic footing for the 2B classification[ 31 ]. EGR1 and PTGS2-short-listed by both machine-learning and CytoHubba-appear in both APM-MASLD and APM-HCC modules and can be viewed as “signal gates” between the two routes. EGR1, an immediate-early transcription factor, is rapidly induced by oxidative stress and up-regulates IL-8, TGF-β1 and COL1A1, bridging inflammation to fibrosis. Within an established steatohepatitic milieu, EGR1 prolongs lipid retention by repressing PPARα, creating a self-reinforcing “metabolism–inflammation” loop. PTGS2, as outlined, directly drives proliferation via COX-2-PGE₂. Their temporal–dose expression gradient may dictate APM outcome: short-term or low-dose exposure favours EGR1-centric simple steatosis, whereas higher cumulative doses shift the balance toward PTGS2-dominant direct transformation or accelerated MASLD-to-HCC progression. We leveraged cross-cohort bulk RNA-seq paired with 113 algorithmic combinations to minimise batch effects and over-fitting; external validation (GSE76427) achieved AUC > 0.9, underscoring generalisability. The “intersection–enrichment–docking” triad embeds affinity data into macro-pathways, closing the loop from omic association to structural feasibility. Limitations remain: (i) absence of blood APM concentrations precludes dose–response modelling; (ii) docking predicts but does not prove binding—surface plasmon resonance (SPR) or isothermal titration calorimetry (ITC) validation confirmation is needed; (iii) human liver transcriptomes cannot capture spatio-temporal nuances of entero-hepatic APM metabolite cycling. Future work should establish low-dose dietary APM murine models coupled with hepatocyte-specific EGR1 or PTGS2 knockout to establish causality, while parallel human studies with accurate exposure assessment and ultra-long-term follow-up are urgently warranted. Integrating enrichment analyses and experimental evidence, we propose a “dual-track” model for APM-driven hepatocarcinogenesis. Track A follows the canonical MASLD-to-HCC axis, driven by bile-acid blockade → lipid deposition → TNF/IL-17 amplification → ROS-mediated mutagenesis. Track B bypasses steatosis altogether: APM directly engages PTGS2/EGR1 to hijack proliferation and apoptosis gatekeepers, accomplishing malignant priming before any histological fat accumulation is detectable. This schema not only accounts for the epidemiologic observation of “non-obese HCC” but also offers actionable molecular nodes for revising APM dietary guidelines. For Track A, FXR agonists or TNF-α monoclonal antibodies could interrupt the inflammatory loop; for Track B, low-dose COX-2-selective inhibitors or EGR1-DNA–binding-domain antagonist peptides may serve as novel chemopreventive strategies. Next, we will quantify serum EGR1/PTGS2 expression in a prospective cohort to evaluate their utility as early-surveillance biomarkers for APM-associated liver cancer and to accelerate the translational pipeline from toxicology to clinic. Demonstrating the causal links of both “APM–metabolism–cancer” and “APM–cancer” chains will provide the evidence base required to upgrade regulatory classification from the current 2B (“possibly carcinogenic to humans”) to “sufficient evidence of carcinogenicity”. Conclusion Using bulk RNA-seq we identified EGR1 and PTGS2 as the core nodes through which APM drives MASLD-to-HCC progression; these same genes also represent putative diagnostic markers for APM-attributable liver cancer. Enrichment analyses support a “dual-track” model of APM hepatocarcinogenesis. Track A follows the canonical MASLD-HCC axis, initiated by bile-acid blockade, lipid deposition, TNF/IL-17 signalling and ROS-mediated mutagenesis. Track B bypasses steatosis: APM directly targets PTGS2/EGR1 to hijack proliferation and apoptosis gatekeepers, accomplishing malignant priming before histological fat accumulation is evident. These findings provide new mechanistic insight into APM-induced HCC, nominate actionable diagnostic/therapeutic targets, and furnish evidence to help refine APM’s carcinogenic classification, although experimental validation remains to be completed. Declarations Acknowledgements First, we would like to thank the editors and reviewers of this journal for their contributions to this study. We also extend our gratitude to the official sources of the following databases and tools for their data and analytical support: C hEMBL, STITCH and SwissTargetPrediction , PubChem, RCSB Protein Data Bank, GEO database, and CB-Dock2. Author contributions ZX and TR contributed equally. TR, ZX, MY, WQ and CB participated in the conception and design of the study. TR, ZX and MY organized the database and statistical analysis. ZX, MY and TR divided the work and participated in the picture drawing. ZX and TR wrote the frst draft of the manuscript. WQ and CB participated in the revision of the manuscript. All authors read and agreed to the fnal manuscript and authorship arrangement. Funding This research is supported by the Ningxia Natural Science Foundation, Project No.:2023AAC02073 and Central Government-Guided Local Science and Technology Development Special Project, Project No.:2024FRD05060. Data availability The sequencing data used in this study were obtained from the GEO (https:// www.ncbi.nlm.nih.gov/geo/) database, including GSE14323 (N=91), GSE14520 (N=445), GSE25097 (N=268), GSE36376 (N=240), GSE76427 (N=115) for HCC, and GSE126848 (N=75), GSE135251 (N=96), GSE89632 (N=70) for MASLD. Simultaneously, the target genes of HCC and MASLD were searched using GeneCards (https://www.genecards.org/), and the targets of APM were searched using ChEMBL(https://www.ebi.ac.uk/chembl/), STITCH(https://stitch.embl.de), and SwissTargetPrediction (https://www.swisstarget -prediction.ch). The target structures of EGR1 (PDB ID: 4X9E) and PTGS (PDB ID: 5F1A) were retrieved from the RCSB Protein Data Bank (https://www.rcsb.org/), the conformational information of APM was obtained from PubChem (https://pubchem.ncbi.nlm.nih.gov/), and molecular docking was performed using CB-Dock2 (https://cadd.labshare.cn/cb-dock2 /index.php). All of which are available in publicly available databases. This study complies with its data use and publication rules. Clinical trial number Not applicable. Ethics, Consent to Participate, and Consent to Publish declarations Not applicable. References Koshy A: Evolving Global Etiology of Hepatocellular Carcinoma (HCC): Insights and Trends for 2024 . Journal of clinical and experimental hepatology 2025, 15 (1):102406. Pol S: Hepatocellular carcinoma (HCC) . Medecine tropicale et sante internationale 2024, 4 (4):mtsi.v4i. Buonaguro L: Human Hepatocellular Carcinoma (HCC) . CANCERS 2020, 12 (12):3739. Filho AM, Laversanne M, Ferlay J, Colombet M, Piñeros M, Znaor A, Parkin DM, Soerjomataram I, Bray F: The GLOBOCAN 2022 cancer estimates: Data sources, methods, and a snapshot of the cancer burden worldwide . INT J CANCER 2025, 156 (7):1336-1346. Lebossé F, Zoulim F: Hepatitis B vaccine and liver cancer . B CANCER 2021, 108 (1):90-101. Banini BA, Sanyal AJ: NAFLD-related HCC . ADV CANCER RES 2021, 149 :143-169. Ray K: NAFLD-HCC: target cholesterol . Nature reviews. Gastroenterology & hepatology 2018, 15 (7):390. Yip TC, Lee HW, Chan WK, Wong GL, Wong VW: Asian perspective on NAFLD-associated HCC . J HEPATOL 2022, 76 (3):726-734. Ahmad SY, Friel JK, Mackay DS: Effect of sucralose and aspartame on glucose metabolism and gut hormones . NUTR REV 2020, 78 (9):725-746. Yellowlees H: Aspartame . British medical journal (Clinical research ed.) 1983, 287 (6396):912-913. Aspartame . In: Drugs and Lactation Database (LactMed®). Bethesda (MD): National Institute of Child Health and Human Development; 2006. Anton SD, Martin CK, Han H, Coulon S, Cefalu WT, Geiselman P, Williamson DA: Effects of stevia, aspartame, and sucrose on food intake, satiety, and postprandial glucose and insulin levels . APPETITE 2010, 55 (1):37-43. Shaher SAA, Mihailescu DF, Amuzescu B: Aspartame Safety as a Food Sweetener and Related Health Hazards . NUTRIENTS 2023, 15 (16):3627. Soffritti M: Understanding the link between aspartame and cancer . EXPERT REV ANTICANC 2024, 24 (9):793-802. Chen D, Hou X: Aspartame carcinogenic potential revealed through network toxicology and molecular docking insights . SCI REP-UK 2024, 14 (1):11492. IARC WGOT: Aspartame, methyleugenol, and isoeugenol . Lyon (FR): International Agency for Research on Cancer; 2024. Iqbal MJ, Kabeer A, Abbas Z, Siddiqui HA, Calina D, Sharifi-Rad J, Cho WC: Interplay of oxidative stress, cellular communication and signaling pathways in cancer . Cell communication and signaling : CCS 2024, 22 (1):7. Jarmakiewicz-Czaja S, Sokal-Dembowska A, Filip R: Effects of Selected Food Additives on the Gut Microbiome and Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD) . Medicina (Kaunas, Lithuania) 2025, 61 (2):192. Finamor IA, Bressan CA, Torres-Cuevas I, Rius-Pérez S, Da Veiga M, Rocha MI, Pavanato MA, Pérez S: Long-Term Aspartame Administration Leads to Fibrosis, Inflammasome Activation, and Gluconeogenesis Impairment in the Liver of Mice . Biology 2021, 10 (2):82. Huang DW, Sherman BT, Tan Q, Kir J, Liu D, Bryant D, Guo Y, Stephens R, Baseler MW, Lane HC et al : DAVID Bioinformatics Resources: expanded annotation database and novel algorithms to better extract biology from large gene lists . NUCLEIC ACIDS RES 2007, 35 (Web Server issue):W169-W175. Szklarczyk D, Kirsch R, Koutrouli M, Nastou K, Mehryary F, Hachilif R, Gable AL, Fang T, Doncheva NT, Pyysalo S et al : The STRING database in 2023: protein-protein association networks and functional enrichment analyses for any sequenced genome of interest . NUCLEIC ACIDS RES 2023, 51 (D1):D638-D646. Xie Y, Shi H, Han B: Bioinformatic analysis of underlying mechanisms of Kawasaki disease via Weighted Gene Correlation Network Analysis (WGCNA) and the Least Absolute Shrinkage and Selection Operator method (LASSO) regression model . BMC PEDIATR 2023, 23 (1):90. Becker T, Rousseau A, Geubbelmans M, Burzykowski T, Valkenborg D: Decision trees and random forests . American journal of orthodontics and dentofacial orthopedics : official publication of the American Association of Orthodontists, its constituent societies, and the American Board of Orthodontics 2023, 164 (6):894-897. Zhang HH, Ahn J, Lin X, Park C: Gene selection using support vector machines with non-convex penalty . Bioinformatics (Oxford, England) 2006, 22 (1):88-95. Gnudi F, Panzacchi S, Tibaldi E, Iuliani M, Sgargi D, Bua L, Mandrioli D: Hemolymphoreticular Neoplasias from the Ramazzini Institute Long-term Mice and Rat Studies on Aspartame . ANN GLOB HEALTH 2023, 89 (1):43. Debras C, Chazelas E, Srour B, Druesne-Pecollo N, Esseddik Y, Szabo De Edelenyi F, Agaësse C, De Sa A, Lutchia R, Gigandet S et al : Artificial sweeteners and cancer risk: Results from the NutriNet-Santé population-based cohort study . PLOS MED 2022, 19 (3):e1003950. Jones SK, McCarthy DM, Vied C, Stanwood GD, Schatschneider C, Bhide PG: Transgenerational transmission of aspartame-induced anxiety and changes in glutamate-GABA signaling and gene expression in the amygdala . P NATL ACAD SCI USA 2022, 119 (49):e2081847177. Lu H, Mao Z, Zheng M, Zhang M, Huang H, Chen Y, Lv L, Chen Z: Identification of hub gene for the pathogenic mechanism and diagnosis of MASLD by enhanced bioinformatics analysis and machine learning . PLOS ONE 2025, 20 (5):e324972. Wang Z, Huang Y, Guo Z, Sun J, Zheng G: Interferon-Linked Lipid and Bile Acid Imbalance Uncovered in Ankylosing Spondylitis in a Sibling-Controlled Multi-Omics Study . INT J MOL SCI 2025, 26 (16):7919. Wang Y, Fleishman JS, Li T, Li Y, Ren Z, Chen J, Ding M: Pharmacological therapy of metabolic dysfunction-associated steatotic liver disease-driven hepatocellular carcinoma . FRONT PHARMACOL 2024, 14 :1336216. Keshavarz-Rahaghi F, Pleasance E, Kolisnik T, Jones SJM: A p53 transcriptional signature in primary and metastatic cancers derived using machine learning . FRONT GENET 2022, 13 :987238. Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial2.csv SupplementaryFiguresS1andS2.docx SupplementaryMaterial1.csv TableS1.docx image1.png Figure abstract Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 07 Apr, 2026 Reviews received at journal 03 Apr, 2026 Reviews received at journal 02 Apr, 2026 Reviewers agreed at journal 30 Mar, 2026 Reviewers agreed at journal 30 Mar, 2026 Reviews received at journal 29 Mar, 2026 Reviewers agreed at journal 29 Mar, 2026 Reviews received at journal 28 Mar, 2026 Reviewers agreed at journal 28 Mar, 2026 Reviewers invited by journal 27 Mar, 2026 Editor invited by journal 27 Mar, 2026 Editor assigned by journal 27 Mar, 2026 Submission checks completed at journal 26 Mar, 2026 First submitted to journal 25 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9171872","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":614711859,"identity":"e8cebde0-755e-4b08-a070-e49b0670150e","order_by":0,"name":"Xusheng Zhang","email":"","orcid":"","institution":"General Hospital of Ningxia Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xusheng","middleName":"","lastName":"Zhang","suffix":""},{"id":614711860,"identity":"55a6ee0d-fc26-418e-ada9-533f4f2222e1","order_by":1,"name":"Rong Tan","email":"","orcid":"","institution":"People's Hospital of Ningxia Hui Autonomous Region","correspondingAuthor":false,"prefix":"","firstName":"Rong","middleName":"","lastName":"Tan","suffix":""},{"id":614711861,"identity":"3efbed33-3968-4e0d-ae6a-ccfed2921cef","order_by":2,"name":"Yongxin Ma","email":"","orcid":"","institution":"General Hospital of Ningxia Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yongxin","middleName":"","lastName":"Ma","suffix":""},{"id":614711862,"identity":"02bce350-00f9-435a-ab9f-93045e66ca52","order_by":3,"name":"Qi Wang","email":"","orcid":"","institution":"General Hospital of Ningxia Medical University","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Wang","suffix":""},{"id":614711863,"identity":"e8e7b24e-dfa4-4e6c-b8a5-07d2cb27bbd7","order_by":4,"name":"Bnedong Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8UlEQVRIiWNgGAWjYDACCQYGZiDF2ADED6BiBkRrYYYpJV4LmwRRWvhnNx97XFBxR7Zfuv1adWHbtsQG9uZtEgw1d3BbcudYuvGMM8+MZ845U3Z7xpnbiQ08x8okGI49w6nFQCLHTJq37XDihhs5abd5KoBagCISjA2H8WjJ/ybN+w+ipZjHAKhF/g0hLTls0rwNIC3px5ghtvDg1yJxI81Mesaxw8YzZ+QwS/OcuW3cxpNWbJFwDLcW/hnJz6QLag7L9kukP/zM23Zbtp/98MYbH2pwa0ECPJDoYAMRCcRoYGBgf0CculEwCkbBKBhxAAAFDVc3prPbwQAAAABJRU5ErkJggg==","orcid":"","institution":"General Hospital of Ningxia Medical University","correspondingAuthor":true,"prefix":"","firstName":"Bnedong","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2026-03-19 17:08:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9171872/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9171872/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105810336,"identity":"9148bd04-96a9-4748-b160-bb4b9e775992","added_by":"auto","created_at":"2026-03-31 11:12:08","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1922291,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDifferential expression analysis of HCC versus normal sequencing data with GEO2R.\u003c/strong\u003e (A) Volcano plots of GEO2R results for the included HCC datasets GSE14323, GSE14520, GSE25097, GSE36376 and GSE76427. (B) Intersection of DEG results from the five HCC datasets. (C) PCA of HCC data before batch correction. (D) PCA after batch correction. (E) Heat-map of limma-based DEG analysis on batch-corrected HCC data.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/5328e923d8f1e333736f2661.png"},{"id":105810338,"identity":"f524bd66-4d78-4c7a-824b-b10fb397db30","added_by":"auto","created_at":"2026-03-31 11:12:08","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2661650,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWGCNA of HCC sequencing data.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A–B) Soft-threshold selection: left panel shows scale-free topology fit index R², right panel shows mean connectivity, red line marks R² = 0.8; (C) Gene clustering dendrogram: colored bars indicate distinct co-expression modules; (D) Module–trait correlation heatmap: numbers are correlation coefficients with P values in parentheses, color indicates correlation strength; (E) Importance analysis of different modules; (F) Intersection of GEO2R results, Limma results, WGCNA clusters and GeneCards yielded HCC target genes for downstream analysis.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/06988858d56ee58f50ccf335.png"},{"id":105904753,"identity":"62ba78d8-2810-4ded-b292-6f0125b24625","added_by":"auto","created_at":"2026-04-01 10:10:24","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":6318639,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eScreening of MASLD signature genes and APM targets.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Heat-map of differential expression results from the batch-corrected MASLD data analysed with the R “Limma” package; (B–C) Soft-threshold selection: left panel shows scale-free topology fit index R², right panel shows mean connectivity, red line marks R² = 0.8; (D) Gene clustering dendrogram: coloured bars represent distinct co-expression modules; (E) Module–trait correlation heat-map: numbers are correlation coefficients with P values in parentheses; colour indicates correlation strength; (F) Intersection of Limma DEG results, WGCNA clusters and GeneCards yielded MASLD-related targets for downstream analysis; (G) Union of APM targets from ChEMBL, STITCH and SwissTargetPrediction; (H) Intersection among APM targets, MASLD signature genes and HCC signature genes.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/f2adf1cd58bf759374a79110.png"},{"id":105904610,"identity":"a80906e5-d57e-4333-8b7d-3357935c109a","added_by":"auto","created_at":"2026-04-01 10:09:51","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1586365,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEnrichment analysis of candidate intersection targets.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Network diagram of APM–MASLD–HCC intersection targets; (B) GO enrichment of APM–MASLD targets; (C) KEGG pathway enrichment of APM–MASLD targets; (D) GO enrichment of MASLD–HCC targets; (E) KEGG enrichment of MASLD–HCC targets; (F) GO and KEGG enrichment of APM–HCC targets.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/0c8edd01dd2b0940d255424f.png"},{"id":105810344,"identity":"d2ab825d-6f87-47c0-82ec-2e339ca795b5","added_by":"auto","created_at":"2026-03-31 11:12:08","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":3892548,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDiagnostic model construction and validation.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Heat-map of predictive performance for training, validation and test sets across different machine-learning algorithms; (B–G) ROC-AUC curves for the training set, validation sets (GSE14323, GSE14520, GSE25097, GSE36376) and test set (GSE76427); (H–M) Confusion matrices for the training set, validation sets (GSE14323, GSE14520, GSE25097, GSE36376) and test set (GSE76427).\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/a1df346183ce16b8f1596488.png"},{"id":105810346,"identity":"0c1b1f8e-9546-4d05-b6c0-5733cdf4d59f","added_by":"auto","created_at":"2026-03-31 11:12:08","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1168197,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP analysis of the model genes.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Differential-expression analysis of the eight model genes between HCC and normal tissues; (B) Diagnostic performance of the eight model genes for HCC; (C) SHAP importance bar plot of the model genes; (D) Predictive performance of the HCC model constructed from the eight genes across five algorithms (RF, SVM, XGB, GBM and KNN); (E) Waterfall plot showing how the eight model genes influence the predicted probability for an individual sample; (F) SHAP feature-importance bar chart quantifying the average magnitude and direction of each gene’s contribution to the final prediction.\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/71a0a51480e16b411f372c39.png"},{"id":105810347,"identity":"c4bf73e9-9711-4bbe-9914-7f48a84827e1","added_by":"auto","created_at":"2026-03-31 11:12:08","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1328731,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePPI network of the 12 intersecting targets and core-protein identification with CytoHubba algorithms.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) PPI network of the 12 intersecting targets; (B) Core proteins identified by the DMNC algorithm; (C) Core proteins identified by the EPC algorithm; (D) Core proteins identified by the Degree algorithm; (E) Core proteins identified by the MCC algorithm; (F) Intersection of the results produced by the four built-in CytoHubba algorithms (DMNC, EPC, Degree and MCC).\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/82f9c58af6fe81b71446e258.png"},{"id":106093182,"identity":"789e318a-f50b-4a97-9063-66f64fd242ee","added_by":"auto","created_at":"2026-04-03 11:35:45","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":3087627,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConstruction and validation of the risk model.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Intersection of core proteins selected by CytoHubba algorithms and core target genes retained by machine-learning screening; (B) Nomogram of the risk-prediction model built on the two intersecting core target genes; (C) Calibration curve; (D) Decision curve; (E) Clinical impact curve.\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/25848115bb67f19e03990c36.png"},{"id":105904533,"identity":"cbfb2782-8fe2-4be7-a224-6ad6f40a69b2","added_by":"auto","created_at":"2026-04-01 10:09:19","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":1625782,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMolecular docking results.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image10.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/e0af84de4870bbb07bfb470b.png"},{"id":105810349,"identity":"60d0fa48-7b2c-4a9c-a4d6-44972fc2c549","added_by":"auto","created_at":"2026-03-31 11:12:08","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":2342044,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eProposed mechanism by which APM promotes HCC via PTGS2/COX-2 targeting, leading to sustained lipid deposition, ROS accumulation, mutational burden, and inflammatory neutrophil infiltration.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image11.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/38b70589ac7f5e6093d6bb96.png"},{"id":106723785,"identity":"93695a2e-0002-4b84-afd0-fb077255b3ba","added_by":"auto","created_at":"2026-04-12 18:14:34","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":26847136,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/0278ff3f-9ac1-478f-942e-d4dc3db31043.pdf"},{"id":105904318,"identity":"5f8c2a5b-f928-437a-b89a-3c549adfbd38","added_by":"auto","created_at":"2026-04-01 10:07:22","extension":"csv","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":497211,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial2.csv","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/74c57d1b2d9eaae02d3c0844.csv"},{"id":105904605,"identity":"ad29827f-4c34-4f2b-a327-3a0c96cad176","added_by":"auto","created_at":"2026-04-01 10:09:50","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":928341,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryFiguresS1andS2.docx","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/04b3f9267eca43894cfd09ee.docx"},{"id":105810341,"identity":"60df99d8-a219-485d-a9fa-79eca0eb5a22","added_by":"auto","created_at":"2026-03-31 11:12:08","extension":"csv","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":1828903,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial1.csv","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/4eee8049208b805a16563e32.csv"},{"id":105904602,"identity":"a1b60d1d-06a7-4366-92d7-bd12cb15e5e5","added_by":"auto","created_at":"2026-04-01 10:09:48","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":11886,"visible":true,"origin":"","legend":"","description":"","filename":"TableS1.docx","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/58c32fb444ea422f4ca9a507.docx"},{"id":105904548,"identity":"4dee7bfe-6618-40ac-b845-6e902118e5ea","added_by":"auto","created_at":"2026-04-01 10:09:28","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":3957323,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure abstract\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-9171872/v1/510decd8bbfe6dc2aafff6c9.png"}],"financialInterests":"No competing interests reported.","formattedTitle":"Aspartame Drives the Continuous Progression from MASLD to HCC: An Integrated Analysis","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHepatocellular carcinoma (HCC) accounts for roughly four-fifths of primary liver cancer and ranks third among global cancer deaths[\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. GLOBOCAN 2022 recorded 389 000 new cases and 331 000 deaths in China\u0026mdash;over forty percent of the world total-rising two to three percent each year[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Curative resection or transplantation lifts five-year survival to seventy percent, yet fewer than thirty percent are operable at diagnosis; combined targeted and immunotherapy extends median overall survival from under one year to about twenty months in advanced disease. The aetiological profile is shifting: hepatitis B-related tumours are declining through vaccination and antiviral therapy[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e], while MASLD, renamed from non-alcoholic fatty liver disease (NAFLD) in 2023, now increases at more than five percent annually, contributing thirty to forty percent of new HCC in Western countries and fifteen to twenty percent in China[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. The molecular route from MASLD to HCC remains obscure, with no actionable targets[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis epidemiological rise parallels four decades of widespread sugar substitution. Non-nutritive sweeteners are promoted to counter obesity and type 2 diabetes[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]; aspartame, one hundred and eighty times sweeter than sucrose, dominates the market. Approved in 1981[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], it is now present in roughly six thousand products-soft drinks, chewing gum, yoghurt, vitamin chews-with annual global consumption exceeding twenty-five thousand tonnes[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. China\u0026rsquo;s current food-additive standard still lists aspartame as permissible without quantitative limit, and one third of packaged foods declare its use; teenagers already ingest thirty to fifty percent of the acceptable daily intake. Since rodent lymphoma findings were reported in 2005, the safety of chronic exposure has been debated[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. A 2023 International Agency for Research on Cancer classification as \u0026ldquo;possibly carcinogenic\u0026rdquo; and a European Food Safety Authority statement that mechanistic evidence for low-dose effects remains incomplete have intensified demands for high-resolution human data[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eCurrent experimental evidence is dominated by rodent studies at doses fifty to two hundred times the acceptable daily intake, whereas primate or human data are scarce; most reports focus on direct DNA damage or oxidative stress, leaving the indirect oncogenic route-metabolic remodelling, chronic inflammation and micro-environmental imbalance-largely unexplored[\u003cspan additionalcitationids=\"CR15 CR16\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Within the liver, intestinal esterase cleaves aspartame into methanol and phenylalanine, both requiring hepatic metabolism[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Sustained low-level exposure can deplete folate during methanol oxidation, raise homocysteine and promote lipid deposition[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]; phenylalanine can inhibit the L-arginine-nitric-oxide pathway and aggravate insulin resistance; the methyl ester moiety can consume glutathione, trigger endoplasmic-reticulum stress and activate nuclear factor kappa B and the NLRP3 inflammasome. These events mirror the two-hit hypothesis of MASLD, suggesting that prolonged aspartame intake could initiate a non-viral pathway from steatosis through fibrosis to malignancy[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTraditional cohort studies lack the resolution to track such effects across one or two decades, and animal protocols that add two to four grams per litre to drinking water deliver eight to fifteen times the human acceptable daily intake, limiting translational relevance. We therefore combined population-based liver transcriptomes, computational toxicology and machine-learning algorithms to chart the aspartame-associated molecular network linking MASLD with HCC and to pinpoint nodes amenable to therapeutic intervention.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eFigure abstract\u003c/b\u003e \u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Data download and processing\u003c/h2\u003e \u003cp\u003eThe present study obtained HCC (GSE14323, N:91; GSE14520, N:445; GSE25097, N:268; GSE36376, N:240; GSE76427, N:115) and MASLD (GSE126848, N:75; GSE135251, N:96; GSE89632, N:70) transcriptome datasets from the Gene Expression Omnibus (GEO, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ncbi.nlm.nih.gov/geo/\u003c/span\u003e\u003cspan address=\"https://www.ncbi.nlm.nih.gov/geo/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). All datasets underwent standardized preprocessing, including background correction, gene ID annotation conversion, and data normalization to ensure comparability and analytical accuracy. GeneCards (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.genecards.org/\u003c/span\u003e\u003cspan address=\"https://www.genecards.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) was used to retrieve MASLD- and liver-cancer-related target genes.\u003c/p\u003e \u003cp\u003eIn parallel, this study integrated ChEMBL (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ebi.ac.uk/chembl/\u003c/span\u003e\u003cspan address=\"https://www.ebi.ac.uk/chembl/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), STITCH (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://stitch.embl.de\u003c/span\u003e\u003cspan address=\"https://stitch.embl.de\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), and SwissTargetPrediction (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.swisstarget\u003c/span\u003e\u003cspan address=\"https://www.swisstarget\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e -prediction.ch) to systematically search for potential human target genes of APM for subsequent intersection analysis and functional investigation.\u003c/p\u003e \u003cp\u003eTarget structures of EGR1(PDB ID:4X9E) and PTGS(PDB ID:5F1A) were retrieved from the RCSB Protein Data Bank (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.rcsb.org/\u003c/span\u003e\u003cspan address=\"https://www.rcsb.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), the conformational information of APM was obtained from PubChem (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pubchem.ncbi.nlm.nih.gov/\u003c/span\u003e\u003cspan address=\"https://pubchem.ncbi.nlm.nih.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), and molecular docking was performed using CB-Dock2 (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://cadd.labshare.cn/cb-dock2/index.php\u003c/span\u003e\u003cspan address=\"https://cadd.labshare.cn/cb-dock2/index.php\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Identification of APM-induced MASLD and HCC targets\u003c/h2\u003e \u003cp\u003eFor the training datasets downloaded from GEO, differential expression analysis between HCCC and normal controls, and between MASLD and normal controls was performed with the R limma package; |log₂FC| \u0026gt; 0.585 and FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05 served as screening criteria to identify differentially expressed genes (DEGs).\u003c/p\u003e \u003cp\u003eTo identify co-expression gene modules and key hub genes related to MASLD and HCC, weighted gene co-expression network analysis (WGCNA) was applied to the training datasets. After the pickSoftThreshold function determined the optimal soft-threshold power β, an adjacency matrix was constructed and converted into a topological overlap matrix (TOM) to quantify gene co-expression similarity. Hierarchical clustering based on the TOM dissimilarity matrix was performed, and the dynamic tree-cut algorithm was used to identify co-expression modules (minimum module size 90, cut height 0.25). Module eigengenes (ME) were calculated to represent the overall expression pattern of each module; Pearson correlation analysis screened modules significantly associated with MASLD and liver cancer (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Genes with |MM| \u0026gt; 0.6 and |GS| \u0026gt; 0.6 were selected as hub genes.\u003c/p\u003e \u003cp\u003eFinally, the intersection of the obtained DEGs, WGCNA hub genes, and APM potential target genes was taken to acquire common target genes of APM-induced MASLD and HCC, serving as the candidate gene set for downstream analyses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Functional enrichment and PPI network analysis\u003c/h2\u003e \u003cp\u003eTo explore the potential molecular mechanisms by which APM sequentially induces MASLD and liver cancer, functional enrichment and protein-protein interaction (PPI) network analyses were performed on the candidate target genes.\u003c/p\u003e \u003cp\u003eGO and KEGG pathway enrichment analyses were conducted with DAVID (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://david.ncifcrf.gov\u003c/span\u003e\u003cspan address=\"https://david.ncifcrf.gov\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. GO covers Biological Process (BP), Cellular Component (CC), and Molecular Function (MF). Fisher\u0026rsquo;s exact test evaluated enrichment significance; P\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was set as the cutoff to identify significantly enriched biological functions and signalling pathways.\u003c/p\u003e \u003cp\u003eProtein interaction relationships among candidate targets were retrieved from STRING (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://string-db.org/\u003c/span\u003e\u003cspan address=\"https://string-db.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], imported into Cytoscape for network visualisation, and CytoHubba algorithms (DMNC, EPC, Degree, MCC) were applied to screen core targets and reveal interaction patterns and potential regulatory mechanisms.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Screening and validation of key target genes\u003c/h2\u003e \u003cp\u003eTo select core genes that play central roles in APM-induced MASLD and liver-cancer development, a multi-machine-learning integration strategy was adopted. LASSO[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], random forest (RF)[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], and support-vector-machine recursive-feature elimination (SVM-RFE)[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] were each used to screen core intersection targets; the three resulting lists were intersected with the CytoHubba (DMNC, EPC, Degree and MCC) output to obtain the final core intersection targets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Pathway exploration of core target genes with the KEGG PATHWAY Database\u003c/h2\u003e \u003cp\u003eWe further employed the KEGG PATHWAY Database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.kegg.jp/kegg/kegg.html\u003c/span\u003e\u003cspan address=\"https://www.kegg.jp/kegg/kegg.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) to map the signalling pathways of the two core differentially expressed genes in liver disease.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Differential expression analysis for MASLD and HCC, and APM target screening\u003c/h2\u003e \u003cp\u003eGEO2R was applied separately to each included HCC dataset (GSE14323, GSE14520, GSE25097, GSE36376, and GSE76427) for differential expression analysis with thresholds of |log₂FC| \u0026gt; 0.585 and FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05, and DEGs for each dataset were obtained as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA. The DEG lists from the five datasets were then intersected, and genes present in any two datasets were retained for further analysis, yielding 1,716 candidate target genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB). Subsequently, batch correction was performed across the five datasets, and principal component analysis results before and after batch correction are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC\u0026ndash;D. Differential expression analysis between the HCC and normal control (N) groups was conducted on the batch-corrected data using the limma package with |log₂FC| \u0026gt; 0.585 and FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05 as selection criteria, resulting in 701 candidate targets (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eE).\u003c/p\u003e \u003cp\u003eNext, weighted gene co-expression network analysis (WGCNA) was performed on the batch-corrected HCC data. The pickSoftThreshold function was first used to evaluate a range of soft-threshold powers β, and the optimal β\u0026thinsp;=\u0026thinsp;8 was selected based on the scale-free topology fit index (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA, B). A weighted adjacency matrix was then constructed and transformed into a Topological Overlap Matrix (TOM) to quantify gene co-expression similarity. Subsequently, hierarchical clustering was performed based on the TOM dissimilarity matrix, combined with the dynamic tree cut algorithm, with parameters set as follows: minimum module size\u0026thinsp;=\u0026thinsp;90, deepSplit\u0026thinsp;=\u0026thinsp;3, and module merge cut height\u0026thinsp;=\u0026thinsp;0.25. Ultimately, seven gene co-expression modules with distinct expression patterns were identified, and the correspondence between modules and the gene dendrogram is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC. Among them, the blue module exhibited a strong positive correlation with the disease, and 461 hub genes highly associated with HCC were identified (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD, E). In addition, 910 HCC-related genes were retrieved from GeneCards (Supplementary Material 1). The intersection of the GEO2R results, Limma package analysis results, WGCNA results, and GeneCards retrieval was then performed, with candidate target genes identified in any two of these methods being included in the next step of analysis, resulting in a total of 854 HCC-related genes.\u003c/p\u003e \u003cp\u003eThe included MASLD sequencing datasets (GSE126848, GSE135251, and GSE89632) were batch-corrected; PCA before and after correction is shown in Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003eA,B. Limma was then applied to the corrected data to compare the MASLD and normal control groups. With thresholds of |log₂FC| \u0026gt; 0.585 and FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05, 1,514 MASLD candidate targets were obtained (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA).\u003c/p\u003e \u003cp\u003eNext, weighted gene co-expression network analysis (WGCNA) was performed on the batch-corrected MASLD data. pickSoftThreshold evaluated a series of soft-threshold powers β, and the optimal β\u0026thinsp;=\u0026thinsp;8 was selected based on the scale-free topology fit index (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB,C). A weighted adjacency matrix was constructed and converted into a Topological Overlap Matrix (TOM) to quantify gene co-expression similarity. Hierarchical clustering based on the TOM dissimilarity matrix, combined with dynamic tree cut (minimum module size\u0026thinsp;=\u0026thinsp;90, deepSplit\u0026thinsp;=\u0026thinsp;3, merge cut height\u0026thinsp;=\u0026thinsp;0.25), identified seven co-expression modules; the correspondence between modules and the gene dendrogram is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD. The blue module exhibited the strongest positive correlation with MASLD, yielding 557 hub genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE). Intersection of Limma results, WGCNA results, and GeneCards retrieval (Supplementary Material 2), retaining genes detected by any two approaches, produced 390 MASLD-related genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF).\u003c/p\u003e \u003cp\u003eNext, ChEMBL, STITCH, and SwissTargetPrediction were systematically queried for potential human targets of APM; the union of the three databases yielded 12 APM targets (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eG). The intersection of the obtained HCC target genes, MASLD-related hub genes, and APM targets revealed 16 APM\u0026ndash;MASLD overlaps, 18 APM\u0026ndash;HCC overlaps, 36 MASLD\u0026ndash;HCC overlaps, and 12 genes common to all three sets (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eH).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(A\u0026ndash;B) Soft-threshold selection: left panel shows scale-free topology fit index R\u0026sup2;, right panel shows mean connectivity, red line marks R\u0026sup2; = 0.8; (C) Gene clustering dendrogram: colored bars indicate distinct co-expression modules; (D) Module\u0026ndash;trait correlation heatmap: numbers are correlation coefficients with P values in parentheses, color indicates correlation strength; (E) Importance analysis of different modules; (F) Intersection of GEO2R results, Limma results, WGCNA clusters and GeneCards yielded HCC target genes for downstream analysis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(A) Heat-map of differential expression results from the batch-corrected MASLD data analysed with the R \u0026ldquo;Limma\u0026rdquo; package; (B\u0026ndash;C) Soft-threshold selection: left panel shows scale-free topology fit index R\u0026sup2;, right panel shows mean connectivity, red line marks R\u0026sup2; = 0.8; (D) Gene clustering dendrogram: coloured bars represent distinct co-expression modules; (E) Module\u0026ndash;trait correlation heat-map: numbers are correlation coefficients with P values in parentheses; colour indicates correlation strength; (F) Intersection of Limma DEG results, WGCNA clusters and GeneCards yielded MASLD-related targets for downstream analysis; (G) Union of APM targets from ChEMBL, STITCH and SwissTargetPrediction; (H) Intersection among APM targets, MASLD signature genes and HCC signature genes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Functional Enrichment Analysis of Candidate Targets\u003c/h2\u003e \u003cp\u003eThe 12 key intersection targets of APM-induced MASLD and HCC are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA. Based on the DAVID database, GO and KEGG functional enrichment analyses were first performed on the 12 intersection targets of APM and MASLD to investigate the specific molecular mechanisms involved in APM-induced MASLD. GO analysis results showed that these genes were primarily involved in biological processes such as protein/vesicle localization, response to stimulus, regulation of biological process, negative regulation of biological process, and cellular process (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB). KEGG enrichment analysis revealed that these genes were mainly enriched in signaling pathways including Bile secretion, Lipid and atherosclerosis, TNF signaling pathway, and AGE-RAGE signaling pathway in diabetic complications (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC).\u003c/p\u003e \u003cp\u003eContinuing with enrichment analysis of the 36 intersecting target genes between MASLD and HCC, these genes were found to be primarily involved in biological processes such as lipid localization, response to xenobiotic stimulus, and lipid transport; cellular components including membrane raft, endoplasmic reticulum lumen, and RNA polymerase II transcription regulator complex; and molecular functions such as DNA-binding transcription factor binding, RNA polymerase II-specific DNA-binding transcription factor binding, and modified amino acid binding (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD). KEGG enrichment analysis demonstrated that these genes were mainly enriched in signaling pathways including Chemical carcinogenesis-receptor activation, Transcriptional misregulation in cancer, Apelin signaling pathway, Chemical carcinogenesis-DNA adducts, AMPK signaling pathway, TNF signaling pathway, IL-17 signaling pathway, and PPAR signaling pathway (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE).\u003c/p\u003e \u003cp\u003eFurthermore, enrichment analysis of the 18 intersecting target genes between APM and HCC revealed that these genes were primarily involved in functions such as carboxylic acid transport, Nuclear receptors meta-pathway, response to xenobiotic stimulus, regulation of lipid transport, regulation of blood vessel endothelial cell migration, osteoblast differentiation, and cell population proliferation. The signaling pathways involved included the PID NFAT TF pathway and p53 transcriptional gene network (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eF). These functions and pathways may represent important mechanisms through which aspartame directly induces HCC.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(A) Network diagram of APM\u0026ndash;MASLD\u0026ndash;HCC intersection targets; (B) GO enrichment of APM\u0026ndash;MASLD targets; (C) KEGG pathway enrichment of APM\u0026ndash;MASLD targets; (D) GO enrichment of MASLD\u0026ndash;HCC targets; (E) KEGG enrichment of MASLD\u0026ndash;HCC targets; (F) GO and KEGG enrichment of APM\u0026ndash;HCC targets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Diagnostic model construction\u003c/h2\u003e \u003cp\u003eUsing the 12 intersection targets, 113 machine-learning algorithms and their ensembles were trained on the batch-corrected compilation of the first four HCC datasets (GSE14323, GSE14520, GSE25097, GSE36376) as the training set; each of the four was also used alone as an internal validation set, and GSE76427 served as the external test set. Prediction performance is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA; the \u0026ldquo;glmBoost\u0026thinsp;+\u0026thinsp;GBM\u0026rdquo; combination achieved the highest diagnostic efficacy, maintained in both validation and test sets. ROC-AUC curves for training, validation and test sets under this algorithm are displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB\u0026ndash;G. Confusion matrices further confirmed high diagnostic performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eH\u0026ndash;M). Expression profiles of the eight genes retained by \u0026ldquo;glmBoost\u0026thinsp;+\u0026thinsp;GBM\u0026rdquo; (PTGS2, SLC1A2, SPP1, EGR1, IGFBP2, ACSL1, THBS1, PPARG) are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA. Diagnostic evaluation indicated that, except for SPP1 and THBS1, the remaining six genes exhibit robust diagnostic capacity for HCC and represent potential biomarkers (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(A) Heat-map of predictive performance for training, validation and test sets across different machine-learning algorithms; (B\u0026ndash;G) ROC-AUC curves for the training set, validation sets (GSE14323, GSE14520, GSE25097, GSE36376) and test set (GSE76427); (H\u0026ndash;M) Confusion matrices for the training set, validation sets (GSE14323, GSE14520, GSE25097, GSE36376) and test set (GSE76427).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.4 SHAP analysis\u003c/h2\u003e \u003cp\u003eTo further verify the diagnostic fidelity of the model, SHAP analysis was performed on the eight modelling genes. SHAP scores were calculated and ranked by absolute value (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eC); EGR1 dominated model explainability with a mean|SHAP| of 0.135, followed by IGFBP2, ACSL1, etc. The eight-gene HCC predictor achieved robust performance across five algorithms (RF, SVM, XGB, GBM, KNN; AUC\u0026thinsp;\u0026gt;\u0026thinsp;0.9, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eD). The waterfall plot in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eE illustrates that all eight genes contributed positively to the predicted probability of an individual sample, raising it from a baseline of 0.851 to 0.999, underscoring their role as core switches. This is further corroborated by the SHAP feature-importance bar chart (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eF).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(A) Differential-expression analysis of the eight model genes between HCC and normal tissues; (B) Diagnostic performance of the eight model genes for HCC; (C) SHAP importance bar plot of the model genes; (D) Predictive performance of the HCC model constructed from the eight genes across five algorithms (RF, SVM, XGB, GBM and KNN); (E) Waterfall plot showing how the eight model genes influence the predicted probability for an individual sample; (F) SHAP feature-importance bar chart quantifying the average magnitude and direction of each gene\u0026rsquo;s contribution to the final prediction.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Construction of a risk-prediction model\u003c/h2\u003e \u003cp\u003eWe next re-screened the 12 intersecting targets with additional machine-learning approaches. LASSO regression retained seven feature targets (Fig. \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003eA). SVM-RFE achieved peak performance when six variables were kept (Fig. \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003eB). Random-forest screening, with Ntree\u0026thinsp;=\u0026thinsp;142 and minimal error, selected genes whose importance score exceeded 10, yielding three feature targets (Fig. \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003eC). Protein\u0026ndash;protein interactions (PPI) among the 12 intersecting targets were then analyzed with STRING (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA). The network was imported into Cytoscape and interrogated with four CytoHubba algorithms\u0026mdash;DMNC (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB), EPC (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eC), Degree (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eD) and MCC (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eE). The hub proteins identified by each method were intersected, producing six core proteins (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eF).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(A) PPI network of the 12 intersecting targets; (B) Core proteins identified by the DMNC algorithm; (C) Core proteins identified by the EPC algorithm; (D) Core proteins identified by the Degree algorithm; (E) Core proteins identified by the MCC algorithm; (F) Intersection of the results produced by the four built-in CytoHubba algorithms (DMNC, EPC, Degree and MCC).\u003c/p\u003e \u003cp\u003eNext, the core targets retained by machine-learning screening were intersected with the core proteins extracted from the PPI network, yielding two final core target genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eA). A risk-prediction nomogram for APM-induced HCC was then constructed on the basis of these two genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eB). Calibration curves demonstrated good agreement between predicted and observed probabilities, indicating reliable model performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eC). Decision-curve analysis (DCA) showed that the model provides additional net benefit within the clinically relevant threshold range of 5\u0026ndash;45%, underscoring its potential for translation into a clinical prediction tool (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eD). The clinical impact curve revealed that predicted probabilities between 10% and 40% achieve the highest net clinical benefit\u0026mdash;detecting the greatest number of early-stage HCC cases with the fewest additional examinations (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eE).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e(A) Intersection of core proteins selected by CytoHubba algorithms and core target genes retained by machine-learning screening; (B) Nomogram of the risk-prediction model built on the two intersecting core target genes; (C) Calibration curve; (D) Decision curve; (E) Clinical impact curve.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Molecular-docking prediction of APM binding to core targets\u003c/h2\u003e \u003cp\u003eTo verify whether APM can directly bind the core HCC targets, we performed molecular-docking simulations with EGR1 and PTGS2. APM displayed favorable binding affinities for both proteins, with calculated binding energies of \u0026minus;\u0026thinsp;7.1 kcal mol⁻\u0026sup1; (EGR1) and \u0026minus;\u0026thinsp;7.9 kcal mol⁻\u0026sup1; (PTGS2); absolute values\u0026thinsp;\u0026gt;\u0026thinsp;7 kcal mol⁻\u0026sup1; indicate stable complex formation(Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e). These data provide preliminary molecular-level evidence that APM can directly engage the identified core target proteins, detailed results are provided in Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.7 Exploration of core-target signaling pathways\u003c/h2\u003e \u003cp\u003eUsing the KEGG PATHWAY Database, we mapped the signaling cascades in which the two core differentially expressed genes are implicated. The results suggest that APM, by targeting PTGS2/COX-2, triggers sustained lipid accumulation, ROS generation, mutational burden and neutrophil infiltration; together these insults push hepatocytes into compensatory proliferation and ultimately drive hepatocellular carcinoma (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eAs evidence accumulates, three key rodent studies have shaped the debate of APM expose: the Ramazzini Institute reported increased lymphomas and leukaemias[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]; the NTP diet-based carcinogenicity bioassay observed a rise in hepatocellular carcinomas in female mice; and human data show a \u0026ldquo;weak but coherent\u0026rdquo; signal. In the NutriNet-Sant\u0026eacute; cohort (n\u0026thinsp;=\u0026thinsp;102 865), 10-year follow-up yielded an overall cancer HR of 1.15 (95% CI 1.03\u0026ndash;1.28), with the strongest hint for liver cancer[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Consequently, IARC re-classified APM as Group 2B (\u0026ldquo;possibly carcinogenic to humans\u0026rdquo;) in 2023. Yet APM-centric oncogenic research remains scarce. Since FDA approval in 1981 only\u0026thinsp;~\u0026thinsp;40 years have elapsed[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]; no study has accrued\u0026thinsp;\u0026gt;\u0026thinsp;15 years of individual-level follow-up. Meanwhile, ubiquitous low-calorie products have silently driven population exposure upward, but investigative effort has not kept pace, and the cumulative impact of ultra-long-term, low-dose APM is still terra incognita. The liver\u0026mdash;central to metabolism and xenobiotic clearance\u0026mdash;has received disproportionately little attention despite chronic, escalating APM doses.\u003c/p\u003e \u003cp\u003eHere, leveraging multi-centre transcriptomes and a machine-learning framework, we chart a molecular roadmap in which low-dose APM propels the sequential evolution \u0026ldquo;metabolism to inflammation to fibrosis to MASLD to HCC\u0026rdquo;. We further propose a second, direct oncogenic route that bypasses the steatotic stage and targets hepatocytes per se. Functional enrichment and protein-interaction networks converge on two intersecting yet partly independent signalling clusters: (i) bile-acid efflux blockade, lipid mis-trafficking and TNF/AGE\u0026ndash;RAGE cascades that explain the MASLD\u0026ndash;HCC continuum; and (ii) p53-network destabilisation and NFAT-driven hyper-proliferation that reveal direct transformation pressure.\u003c/p\u003e \u003cp\u003eGO functional enrichment analysis and KEGG pathway enrichment analysis of APM\u0026ndash;MASLD intersecting targets identify \u0026ldquo;bile-acid\u0026ndash;lipid metabolic imbalance\u0026rdquo; as the first hit. KEGG pathway terms such as \u0026ldquo;protein/vesicle localisation\u0026rdquo; and \u0026ldquo;negative regulation of biological process\u0026rdquo; imply early membrane-trafficking reprogramming. KEGG pathway places \u0026ldquo;bile secretion\u0026rdquo; at the top; retained bile acids cripple the FXR\u0026ndash;SHP negative loop, raising intrahepatic free cholesterol and sphingolipids-consistent with simultaneous enrichment of \u0026ldquo;lipid and atherosclerosis\u0026rdquo;[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. By impairing bile-acid export, APM lowers lipid clearance and supplies substrates for droplet deposition\u0026mdash;the classical \u0026ldquo;first hit\u0026rdquo;[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIntersection genes of MASLD\u0026ndash;HCC are simultaneously enriched for \u0026ldquo;TNF signalling\u0026rdquo;, \u0026ldquo;IL-17 signalling\u0026rdquo; and \u0026ldquo;chemical carcinogenesis\u0026ndash;receptor activation\u0026rdquo;, implicating chronic inflammation as the bridge between simple steatosis and malignancy[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Mechanistically, APM-derived methanol depletes folate, raises homocysteine and activates NF-κB; phenylalanine inhibits eNOS, reduces NO bioavailability and synergistically elevates ROS. ROS inflicts mtDNA mutations and, via an IL-17C\u0026ndash;neutrophil chemotaxis feed-forward loop, continuously recruits neutrophils that release reactive nitrogen species[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Persistent AGE\u0026ndash;RAGE signalling glycation maintains this inflammatory milieu, fuels mutation accumulation and provides clonal advantage for malignant transformation.\u003c/p\u003e \u003cp\u003eGenes unique to the APM\u0026ndash;HCC axis enrich for \u0026ldquo;p53 transcriptional network\u0026rdquo;, \u0026ldquo;nuclear receptors meta-pathway\u0026rdquo; and \u0026ldquo;response to xenobiotic stimulus\u0026rdquo;, suggesting that APM acts as an exogenous ligand that hijacks nuclear-receptor signalling, forcing quiescent hepatocytes into chronic replicative stress. Coupled with p53 loss, this drives clonal selection for malignancy. Molecular docking shows APM\u0026ndash;PTGS2/COX-2 binding energy \u0026minus;\u0026thinsp;7.9 kcal mol⁻\u0026sup1;, compatible with stable enzyme activation. Sustained COX-2 elevation (i) induces c-Myc and Cyclin D1 via the PGE₂\u0026ndash;EP₄ axis, propelling cell-cycle traverse, and (ii) suppresses p53 acetylation, blunting transcription of GADD45 and PUMA. The resulting \u0026ldquo;hyper-proliferation plus disabled genome surveillance\u0026rdquo; creates a double-edged environment that can launch transformation before overt steatosis appears. This finding aligns with EFSA\u0026rsquo;s 2023 conjecture that \u0026ldquo;long-term, low-dose exposure might directly perturb proliferation control\u0026rdquo; and offers mechanistic footing for the 2B classification[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eEGR1 and PTGS2-short-listed by both machine-learning and CytoHubba-appear in both APM-MASLD and APM-HCC modules and can be viewed as \u0026ldquo;signal gates\u0026rdquo; between the two routes. EGR1, an immediate-early transcription factor, is rapidly induced by oxidative stress and up-regulates IL-8, TGF-β1 and COL1A1, bridging inflammation to fibrosis. Within an established steatohepatitic milieu, EGR1 prolongs lipid retention by repressing PPARα, creating a self-reinforcing \u0026ldquo;metabolism\u0026ndash;inflammation\u0026rdquo; loop. PTGS2, as outlined, directly drives proliferation via COX-2-PGE₂. Their temporal\u0026ndash;dose expression gradient may dictate APM outcome: short-term or low-dose exposure favours EGR1-centric simple steatosis, whereas higher cumulative doses shift the balance toward PTGS2-dominant direct transformation or accelerated MASLD-to-HCC progression.\u003c/p\u003e \u003cp\u003eWe leveraged cross-cohort bulk RNA-seq paired with 113 algorithmic combinations to minimise batch effects and over-fitting; external validation (GSE76427) achieved AUC\u0026thinsp;\u0026gt;\u0026thinsp;0.9, underscoring generalisability. The \u0026ldquo;intersection\u0026ndash;enrichment\u0026ndash;docking\u0026rdquo; triad embeds affinity data into macro-pathways, closing the loop from omic association to structural feasibility. Limitations remain: (i) absence of blood APM concentrations precludes dose\u0026ndash;response modelling; (ii) docking predicts but does not prove binding\u0026mdash;surface plasmon resonance (SPR) or isothermal titration calorimetry (ITC) validation confirmation is needed; (iii) human liver transcriptomes cannot capture spatio-temporal nuances of entero-hepatic APM metabolite cycling. Future work should establish low-dose dietary APM murine models coupled with hepatocyte-specific EGR1 or PTGS2 knockout to establish causality, while parallel human studies with accurate exposure assessment and ultra-long-term follow-up are urgently warranted.\u003c/p\u003e \u003cp\u003eIntegrating enrichment analyses and experimental evidence, we propose a \u0026ldquo;dual-track\u0026rdquo; model for APM-driven hepatocarcinogenesis. Track A follows the canonical MASLD-to-HCC axis, driven by bile-acid blockade \u0026rarr; lipid deposition \u0026rarr; TNF/IL-17 amplification \u0026rarr; ROS-mediated mutagenesis. Track B bypasses steatosis altogether: APM directly engages PTGS2/EGR1 to hijack proliferation and apoptosis gatekeepers, accomplishing malignant priming before any histological fat accumulation is detectable. This schema not only accounts for the epidemiologic observation of \u0026ldquo;non-obese HCC\u0026rdquo; but also offers actionable molecular nodes for revising APM dietary guidelines. For Track A, FXR agonists or TNF-α monoclonal antibodies could interrupt the inflammatory loop; for Track B, low-dose COX-2-selective inhibitors or EGR1-DNA\u0026ndash;binding-domain antagonist peptides may serve as novel chemopreventive strategies.\u003c/p\u003e \u003cp\u003eNext, we will quantify serum EGR1/PTGS2 expression in a prospective cohort to evaluate their utility as early-surveillance biomarkers for APM-associated liver cancer and to accelerate the translational pipeline from toxicology to clinic. Demonstrating the causal links of both \u0026ldquo;APM\u0026ndash;metabolism\u0026ndash;cancer\u0026rdquo; and \u0026ldquo;APM\u0026ndash;cancer\u0026rdquo; chains will provide the evidence base required to upgrade regulatory classification from the current 2B (\u0026ldquo;possibly carcinogenic to humans\u0026rdquo;) to \u0026ldquo;sufficient evidence of carcinogenicity\u0026rdquo;.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eUsing bulk RNA-seq we identified EGR1 and PTGS2 as the core nodes through which APM drives MASLD-to-HCC progression; these same genes also represent putative diagnostic markers for APM-attributable liver cancer. Enrichment analyses support a \u0026ldquo;dual-track\u0026rdquo; model of APM hepatocarcinogenesis. Track A follows the canonical MASLD-HCC axis, initiated by bile-acid blockade, lipid deposition, TNF/IL-17 signalling and ROS-mediated mutagenesis. Track B bypasses steatosis: APM directly targets PTGS2/EGR1 to hijack proliferation and apoptosis gatekeepers, accomplishing malignant priming before histological fat accumulation is evident. These findings provide new mechanistic insight into APM-induced HCC, nominate actionable diagnostic/therapeutic targets, and furnish evidence to help refine APM\u0026rsquo;s carcinogenic classification, although experimental validation remains to be completed.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFirst, we would like to thank the editors and reviewers of this journal for their contributions to this study. We also extend our gratitude to the official sources of the following databases and tools for their data and analytical support: C\u003cstrong\u003ehEMBL, STITCH and SwissTargetPrediction\u003c/strong\u003e, PubChem, RCSB Protein Data Bank,\u0026nbsp;GEO database,\u0026nbsp;and CB-Dock2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eZX and TR contributed equally. TR, ZX, MY, WQ and CB participated in the conception and design of the study. TR, ZX and MY organized the database and statistical analysis. ZX, MY and TR divided the work and participated in the picture drawing. ZX and TR wrote the frst draft of the manuscript. WQ and CB participated in the revision of the manuscript. All authors read and agreed to the fnal manuscript and authorship arrangement.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis research is supported by the Ningxia Natural Science Foundation, Project No.:2023AAC02073 and Central Government-Guided Local Science and Technology Development Special Project, Project No.:2024FRD05060.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe sequencing data used in this study were obtained from the GEO (https:// www.ncbi.nlm.nih.gov/geo/) database, including GSE14323 (N=91), GSE14520 (N=445), GSE25097 (N=268), GSE36376 (N=240), GSE76427 (N=115) for HCC, and GSE126848 (N=75), GSE135251 (N=96), GSE89632 (N=70) for MASLD. Simultaneously, the target genes of HCC and MASLD were searched using GeneCards (https://www.genecards.org/), and the targets of APM were searched using ChEMBL(https://www.ebi.ac.uk/chembl/), STITCH(https://stitch.embl.de), and SwissTargetPrediction (https://www.swisstarget -prediction.ch). The target structures of EGR1 (PDB ID: 4X9E) and PTGS (PDB ID: 5F1A) were retrieved from the RCSB Protein Data Bank (https://www.rcsb.org/), the conformational information of APM was obtained from PubChem (https://pubchem.ncbi.nlm.nih.gov/), and molecular docking was performed using CB-Dock2 (https://cadd.labshare.cn/cb-dock2 /index.php).\u003c/p\u003e\n\u003cp\u003eAll of which are available in publicly available databases. This study complies with its data use and publication rules.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical trial number\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics, Consent to Participate, and Consent to Publish declarations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eKoshy A: \u003cstrong\u003eEvolving Global Etiology of Hepatocellular Carcinoma (HCC): Insights and Trends for 2024\u003c/strong\u003e. \u003cem\u003eJournal of clinical and experimental hepatology\u003c/em\u003e 2025, \u003cstrong\u003e15\u003c/strong\u003e(1):102406.\u003c/li\u003e\n\u003cli\u003ePol S: \u003cstrong\u003eHepatocellular carcinoma (HCC)\u003c/strong\u003e. \u003cem\u003eMedecine tropicale et sante internationale\u003c/em\u003e 2024, \u003cstrong\u003e4\u003c/strong\u003e(4):mtsi.v4i.\u003c/li\u003e\n\u003cli\u003eBuonaguro L: \u003cstrong\u003eHuman Hepatocellular Carcinoma (HCC)\u003c/strong\u003e. \u003cem\u003eCANCERS\u003c/em\u003e 2020, \u003cstrong\u003e12\u003c/strong\u003e(12):3739.\u003c/li\u003e\n\u003cli\u003eFilho AM, Laversanne M, Ferlay J, Colombet M, Pi\u0026ntilde;eros M, Znaor A, Parkin DM, Soerjomataram I, Bray F: \u003cstrong\u003eThe GLOBOCAN 2022 cancer estimates: Data sources, methods, and a snapshot of the cancer burden worldwide\u003c/strong\u003e. \u003cem\u003eINT J CANCER\u003c/em\u003e 2025, \u003cstrong\u003e156\u003c/strong\u003e(7):1336-1346.\u003c/li\u003e\n\u003cli\u003eLeboss\u0026eacute; F, Zoulim F: \u003cstrong\u003eHepatitis B vaccine and liver cancer\u003c/strong\u003e. \u003cem\u003eB CANCER\u003c/em\u003e 2021, \u003cstrong\u003e108\u003c/strong\u003e(1):90-101.\u003c/li\u003e\n\u003cli\u003eBanini BA, Sanyal AJ: \u003cstrong\u003eNAFLD-related HCC\u003c/strong\u003e. \u003cem\u003eADV CANCER RES\u003c/em\u003e 2021, \u003cstrong\u003e149\u003c/strong\u003e:143-169.\u003c/li\u003e\n\u003cli\u003eRay K: \u003cstrong\u003eNAFLD-HCC: target cholesterol\u003c/strong\u003e. \u003cem\u003eNature reviews. Gastroenterology \u0026amp; hepatology\u003c/em\u003e 2018, \u003cstrong\u003e15\u003c/strong\u003e(7):390.\u003c/li\u003e\n\u003cli\u003eYip TC, Lee HW, Chan WK, Wong GL, Wong VW: \u003cstrong\u003eAsian perspective on NAFLD-associated HCC\u003c/strong\u003e. \u003cem\u003eJ HEPATOL\u003c/em\u003e 2022, \u003cstrong\u003e76\u003c/strong\u003e(3):726-734.\u003c/li\u003e\n\u003cli\u003eAhmad SY, Friel JK, Mackay DS: \u003cstrong\u003eEffect of sucralose and aspartame on glucose metabolism and gut hormones\u003c/strong\u003e. \u003cem\u003eNUTR REV\u003c/em\u003e 2020, \u003cstrong\u003e78\u003c/strong\u003e(9):725-746.\u003c/li\u003e\n\u003cli\u003eYellowlees H: \u003cstrong\u003eAspartame\u003c/strong\u003e. \u003cem\u003eBritish medical journal (Clinical research ed.)\u003c/em\u003e 1983, \u003cstrong\u003e287\u003c/strong\u003e(6396):912-913.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eAspartame\u003c/strong\u003e. In: \u003cem\u003eDrugs and Lactation Database (LactMed\u0026reg;).\u003c/em\u003e Bethesda (MD): National Institute of Child Health and Human Development; 2006.\u003c/li\u003e\n\u003cli\u003eAnton SD, Martin CK, Han H, Coulon S, Cefalu WT, Geiselman P, Williamson DA: \u003cstrong\u003eEffects of stevia, aspartame, and sucrose on food intake, satiety, and postprandial glucose and insulin levels\u003c/strong\u003e. \u003cem\u003eAPPETITE\u003c/em\u003e 2010, \u003cstrong\u003e55\u003c/strong\u003e(1):37-43.\u003c/li\u003e\n\u003cli\u003eShaher SAA, Mihailescu DF, Amuzescu B: \u003cstrong\u003eAspartame Safety as a Food Sweetener and Related Health Hazards\u003c/strong\u003e. \u003cem\u003eNUTRIENTS\u003c/em\u003e 2023, \u003cstrong\u003e15\u003c/strong\u003e(16):3627.\u003c/li\u003e\n\u003cli\u003eSoffritti M: \u003cstrong\u003eUnderstanding the link between aspartame and cancer\u003c/strong\u003e. \u003cem\u003eEXPERT REV ANTICANC\u003c/em\u003e 2024, \u003cstrong\u003e24\u003c/strong\u003e(9):793-802.\u003c/li\u003e\n\u003cli\u003eChen D, Hou X: \u003cstrong\u003eAspartame carcinogenic potential revealed through network toxicology and molecular docking insights\u003c/strong\u003e. \u003cem\u003eSCI REP-UK\u003c/em\u003e 2024, \u003cstrong\u003e14\u003c/strong\u003e(1):11492.\u003c/li\u003e\n\u003cli\u003eIARC WGOT: \u003cstrong\u003eAspartame, methyleugenol, and isoeugenol\u003c/strong\u003e. Lyon (FR): International Agency for Research on Cancer; 2024.\u003c/li\u003e\n\u003cli\u003eIqbal MJ, Kabeer A, Abbas Z, Siddiqui HA, Calina D, Sharifi-Rad J, Cho WC: \u003cstrong\u003eInterplay of oxidative stress, cellular communication and signaling pathways in cancer\u003c/strong\u003e. \u003cem\u003eCell communication and signaling : CCS\u003c/em\u003e 2024, \u003cstrong\u003e22\u003c/strong\u003e(1):7.\u003c/li\u003e\n\u003cli\u003eJarmakiewicz-Czaja S, Sokal-Dembowska A, Filip R: \u003cstrong\u003eEffects of Selected Food Additives on the Gut Microbiome and Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD)\u003c/strong\u003e. \u003cem\u003eMedicina (Kaunas, Lithuania)\u003c/em\u003e 2025, \u003cstrong\u003e61\u003c/strong\u003e(2):192.\u003c/li\u003e\n\u003cli\u003eFinamor IA, Bressan CA, Torres-Cuevas I, Rius-P\u0026eacute;rez S, Da Veiga M, Rocha MI, Pavanato MA, P\u0026eacute;rez S: \u003cstrong\u003eLong-Term Aspartame Administration Leads to Fibrosis, Inflammasome Activation, and Gluconeogenesis Impairment in the Liver of Mice\u003c/strong\u003e. \u003cem\u003eBiology\u003c/em\u003e 2021, \u003cstrong\u003e10\u003c/strong\u003e(2):82.\u003c/li\u003e\n\u003cli\u003eHuang DW, Sherman BT, Tan Q, Kir J, Liu D, Bryant D, Guo Y, Stephens R, Baseler MW, Lane HC\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eDAVID Bioinformatics Resources: expanded annotation database and novel algorithms to better extract biology from large gene lists\u003c/strong\u003e. \u003cem\u003eNUCLEIC ACIDS RES\u003c/em\u003e 2007, \u003cstrong\u003e35\u003c/strong\u003e(Web Server issue):W169-W175.\u003c/li\u003e\n\u003cli\u003eSzklarczyk D, Kirsch R, Koutrouli M, Nastou K, Mehryary F, Hachilif R, Gable AL, Fang T, Doncheva NT, Pyysalo S\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eThe STRING database in 2023: protein-protein association networks and functional enrichment analyses for any sequenced genome of interest\u003c/strong\u003e. \u003cem\u003eNUCLEIC ACIDS RES\u003c/em\u003e 2023, \u003cstrong\u003e51\u003c/strong\u003e(D1):D638-D646.\u003c/li\u003e\n\u003cli\u003eXie Y, Shi H, Han B: \u003cstrong\u003eBioinformatic analysis of underlying mechanisms of Kawasaki disease via Weighted Gene Correlation Network Analysis (WGCNA) and the Least Absolute Shrinkage and Selection Operator method (LASSO) regression model\u003c/strong\u003e. \u003cem\u003eBMC PEDIATR\u003c/em\u003e 2023, \u003cstrong\u003e23\u003c/strong\u003e(1):90.\u003c/li\u003e\n\u003cli\u003eBecker T, Rousseau A, Geubbelmans M, Burzykowski T, Valkenborg D: \u003cstrong\u003eDecision trees and random forests\u003c/strong\u003e. \u003cem\u003eAmerican journal of orthodontics and dentofacial orthopedics : official publication of the American Association of Orthodontists, its constituent societies, and the American Board of Orthodontics\u003c/em\u003e 2023, \u003cstrong\u003e164\u003c/strong\u003e(6):894-897.\u003c/li\u003e\n\u003cli\u003eZhang HH, Ahn J, Lin X, Park C: \u003cstrong\u003eGene selection using support vector machines with non-convex penalty\u003c/strong\u003e. \u003cem\u003eBioinformatics (Oxford, England)\u003c/em\u003e 2006, \u003cstrong\u003e22\u003c/strong\u003e(1):88-95.\u003c/li\u003e\n\u003cli\u003eGnudi F, Panzacchi S, Tibaldi E, Iuliani M, Sgargi D, Bua L, Mandrioli D: \u003cstrong\u003eHemolymphoreticular Neoplasias from the Ramazzini Institute Long-term Mice and Rat Studies on Aspartame\u003c/strong\u003e. \u003cem\u003eANN GLOB HEALTH\u003c/em\u003e 2023, \u003cstrong\u003e89\u003c/strong\u003e(1):43.\u003c/li\u003e\n\u003cli\u003eDebras C, Chazelas E, Srour B, Druesne-Pecollo N, Esseddik Y, Szabo De Edelenyi F, Aga\u0026euml;sse C, De Sa A, Lutchia R, Gigandet S\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eArtificial sweeteners and cancer risk: Results from the NutriNet-Sant\u0026eacute; population-based cohort study\u003c/strong\u003e. \u003cem\u003ePLOS MED\u003c/em\u003e 2022, \u003cstrong\u003e19\u003c/strong\u003e(3):e1003950.\u003c/li\u003e\n\u003cli\u003eJones SK, McCarthy DM, Vied C, Stanwood GD, Schatschneider C, Bhide PG: \u003cstrong\u003eTransgenerational transmission of aspartame-induced anxiety and changes in glutamate-GABA signaling and gene expression in the amygdala\u003c/strong\u003e. \u003cem\u003eP NATL ACAD SCI USA\u003c/em\u003e 2022, \u003cstrong\u003e119\u003c/strong\u003e(49):e2081847177.\u003c/li\u003e\n\u003cli\u003eLu H, Mao Z, Zheng M, Zhang M, Huang H, Chen Y, Lv L, Chen Z: \u003cstrong\u003eIdentification of hub gene for the pathogenic mechanism and diagnosis of MASLD by enhanced bioinformatics analysis and machine learning\u003c/strong\u003e. \u003cem\u003ePLOS ONE\u003c/em\u003e 2025, \u003cstrong\u003e20\u003c/strong\u003e(5):e324972.\u003c/li\u003e\n\u003cli\u003eWang Z, Huang Y, Guo Z, Sun J, Zheng G: \u003cstrong\u003eInterferon-Linked Lipid and Bile Acid Imbalance Uncovered in Ankylosing Spondylitis in a Sibling-Controlled Multi-Omics Study\u003c/strong\u003e. \u003cem\u003eINT J MOL SCI\u003c/em\u003e 2025, \u003cstrong\u003e26\u003c/strong\u003e(16):7919.\u003c/li\u003e\n\u003cli\u003eWang Y, Fleishman JS, Li T, Li Y, Ren Z, Chen J, Ding M: \u003cstrong\u003ePharmacological therapy of metabolic dysfunction-associated steatotic liver disease-driven hepatocellular carcinoma\u003c/strong\u003e. \u003cem\u003eFRONT PHARMACOL\u003c/em\u003e 2024, \u003cstrong\u003e14\u003c/strong\u003e:1336216.\u003c/li\u003e\n\u003cli\u003eKeshavarz-Rahaghi F, Pleasance E, Kolisnik T, Jones SJM: \u003cstrong\u003eA p53 transcriptional signature in primary and metastatic cancers derived using machine learning\u003c/strong\u003e. \u003cem\u003eFRONT GENET\u003c/em\u003e 2022, \u003cstrong\u003e13\u003c/strong\u003e:987238.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"dion","sideBox":"Learn more about [Discover Oncology](https://www.springer.com/12672)","snPcode":"","submissionUrl":"","title":"Discover Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"aspartame, hepatocellular carcinoma, metabolic dysfunction-associated steatotic liver disease, network toxicology, molecular docking","lastPublishedDoi":"10.21203/rs.3.rs-9171872/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9171872/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eAspartame (APM), a widely used sweetener, has been linked to cancers, yet its molecular impact on metabolic dysfunction-associated steatotic liver disease (MASLD) and subsequent hepatocellular carcinoma (HCC) remains undefined. We integrated network toxicology, bulk RNA-seq and docking to map the mechanism.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eAPM targets were retrieved from ChEMBL, STITCH and SwissTargetPrediction. MASLD and HCC RNA-seq data from GEO were used to call DEGs. WGCNA identified disease modules and hub genes. Intersection of APM targets, DEGs and hubs defined core genes for GO/KEGG and PPI analyses. CytoHubba (DMNC, EPC, Degree, MCC), LASSO, RF and SVM-RFE shortlisted key genes, and docking verified APM binding.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eTwelve genes intersected across APM, MASLD and HCC datasets. Enrichment supports a \u0026ldquo;dual-track\u0026rdquo; mechanism: APM-MASLD targets suppress bile-acid export, impair lipid clearance and fuel steatosis; MASLD-HCC targets jointly activate TNF/IL-17 and chemical-carcinogenesis pathways, indicating chronic inflammation bridges steatosis to cancer; APM-HCC targets map to p53, nuclear-receptor and xenobiotic-response networks, revealing APM hijacks receptor signalling to impose proliferative stress that, coupled with p53 loss, drives clonal selection. Machine-learning nominated EGR1 and PTGS2 as top diagnostic genes (AUC\u0026thinsp;\u0026gt;\u0026thinsp;0.7); docking showed high-affinity APM binding (\u0026ndash;7.1 and\u0026ndash;7.9 kcal mol⁻\u0026sup1;, respectively), identifying them as key relays in APM-induced HCC.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eEGR1 and PTGS2 are central nodes through which APM precipitates MASLD and accelerates progression to HCC. We propose a \u0026ldquo;dual-track\u0026rdquo; oncogenic paradigm: Track A follows the canonical MASLD-HCC axis (bile-acid retention - lipid deposition - TNF/IL-17-driven ROS-mutational amplification), whereas Track B allows APM, via PTGS2/EGR1, to usurp gate-keeper proteins governing proliferation and apoptosis, initiating malignant programming before overt steatosis develops. These findings provide mechanistic insight into APM-related hepatocarcinogenesis, nominate tractable diagnostic biomarkers and therapeutic targets, and inform future re-evaluation of APM carcinogenicity classifications.\u003c/p\u003e","manuscriptTitle":"Aspartame Drives the Continuous Progression from MASLD to HCC: An Integrated Analysis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-31 11:11:59","doi":"10.21203/rs.3.rs-9171872/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-07T05:48:23+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-03T08:21:03+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-02T18:56:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"32445979626889827391274037762449488271","date":"2026-03-30T15:36:07+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"308548153704380107719556204390772693214","date":"2026-03-30T13:18:18+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-29T08:27:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"94530514822065286761436850362187142823","date":"2026-03-29T07:42:15+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-28T12:01:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"168170965339790007365057420880550907093","date":"2026-03-28T11:44:22+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-27T14:59:01+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-27T10:27:45+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-27T05:53:10+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-26T05:09:28+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Oncology","date":"2026-03-26T03:50:21+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"dion","sideBox":"Learn more about [Discover Oncology](https://www.springer.com/12672)","snPcode":"","submissionUrl":"","title":"Discover Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f11ada67-529a-4ace-880b-ac3f6ae8b2c9","owner":[],"postedDate":"March 31st, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-10T11:38:10+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-31 11:11:59","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9171872","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9171872","identity":"rs-9171872","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}