The primary data for this study were obtained from The Cancer Genome Atlas (TCGA) datasets (v36.0, Jan 2024), systematically downloaded from the University of California Santa Cruz (UCSC) Xena platform (https://xena.ucsc.edu/). These datasets encompassed comprehensive genomic, transcriptomic, and clinical information for HCC patients. Independent validation datasets were acquired from the Gene Expression Omnibus (GEO). Additionally, to establish a baseline for normal tissue gene expression, we incorporated data from the Genotype-Tissue Expression (GTEx) Portal, which offers a comprehensive resource of gene expression data from various healthy human tissues. To ensure data quality and comparability, a rigorous preprocessing pipeline was implemented. Initially, the ‘sva’ package in R was employed to normalize the data and mitigate batch effects and technical variabilities across different experimental runs. This step reduced non-biological sources of variation that could confound downstream analyses. Following normalization, missing data points were addressed using the ‘mice’ (Multivariate Imputation by Chained Equations) package in R. This multiple imputation approach ensures the robust handling of missing data. The resulting preprocessed and complete datasets formed the foundation for subsequent machine learning analyses and experimental validations. All data processing and statistical analyses were performed using R (version 4.1.0) and relevant Bioconductor packages.

To identify robust prognostic candidates, we initially screened 22 MYB-family related genes and their direct downstream targets. A univariate Cox proportional hazards regression was applied as a pre-selection filter (threshold p < 0.05). To address multicollinearity among the remaining significant candidates, we employed the Least Absolute Shrinkage and Selection Operator (LASSO) regression using the ‘glmnet’ R package (v4.1-2). This method effectively selects the most informative features by shrinking coefficients of correlated predictors to zero. The resulting model, and specifically the prognostic value of MYBL2, was subsequently validated externally using independent cohorts from the ICGC (LIRI-JP) and GEO (GSE144269). This methodology ensured that our model remained both powerful and generalizable. The discriminative accuracy of the identified markers was assessed using the area under the receiver operating characteristic (ROC) curve (AUC), calculated with the pROC R package. To evaluate the prognostic value of the identified genes, we conducted Kaplan-Meier survival analysis using an automatic best cutoff algorithm. This analysis was performed using the online tool KM Plotter [22], which integrates gene expression data with clinical outcomes. We investigated the expression pattern of MYBL2 in both normal tissue and tumor tissue using HCCDB, an integrative molecular database specifically designed for HCC research [23]. To ensure the robustness of our findings, we conducted a comprehensive validation of MYBL2 expression across multiple databases. This validation process utilized web-based applications, including BEST (Biomarker Enrichment and Selection Tool) [24] and TNMplot [25], which integrate data from various sources such as The Cancer Genome Atlas (TCGA), and Genotype-Tissue Expression (GTEx).

To develop a comprehensive prognostic scoring model that integrates MYBL2 expression levels and AJCC stage, we employed a multistep approach using various statistical methods and R packages. A nomogram-based prognostic scoring model integrating MYBL2 expression levels and AJCC stage was developed to predict survival probability in hepatocellular carcinoma patients. The model development and validation process involved several steps and utilized various R packages for statistical analysis. Initially, a Cox proportional hazards model was constructed using the ‘survival’ package in R, with MYBL2 expression and AJCC stage as predictors and overall survival as the outcome. Based on this model, a nomogram was formulated using the ‘rms’ package, providing a visual representation of the prognostic model and allowing for the prediction of 1-year, 3-year, and 5-year survival probabilities.

To evaluate the performance of the scoring model, ROC curve, risk score analysis, and Kaplan-Meier survival analysis were conducted using R packages. ROC curves were generated using the ‘pROC’ package and the area under the curve (AUC) to assess the model’s discriminative ability. The nomogram calculated risk scores for each patient and stratified them into high-risk and low-risk groups using the median risk score as the cut-off point. The ‘survminer’ package was used for risk score visualization. Kaplan-Meier survival analysis was utilized to compare survival outcomes between the high-risk and low-risk groups, with the log-rank test used to assess the statistical significance of the difference between the survival curves. This analysis was conducted using both the ‘survival’ and ‘survminer’ packages.

To validate the prognostic significance of MYBL2 in HCC, a comprehensive correlation analysis was performed using three distinct datasets: TCGA, International Cancer Genome Consortium (ICGC), and the GEO dataset GSE144269. The analysis was conducted using the Integrative HCC Gene Analysis (IHGA) web application, a specialized tool designed for HCC research [26]. MYBL2 expression levels were correlated with a comprehensive panel of established prognostic biomarkers for HCC. This panel included alpha-fetoprotein (AFP), glypican-3 (GPC3), cyclin-dependent kinase 4 (CDK4), proliferating cell nuclear antigen (PCNA), baculoviral IAP repeat-containing protein 5 (BIRC5), aurora kinase A (AURKA), discs large-associated protein 5 (DLGAP5), cyclin-dependent kinase 1 (CDK1), marker of proliferation Ki-67 (MKI67), forkhead box protein M1 (FOXM1), DNA topoisomerase II alpha (TOP2A), centromere protein F (CENPF), cyclin-dependent kinase inhibitor 3 (CDKN3), abnormal spindle microtubule assembly (ASPM), NIMA-related kinase 2 (NEK2), minichromosome maintenance complex component 2 (MCM2), cell division cycle 20 (CDC20), and cyclin B1 (CCNB1). Pearson’s correlation coefficient was calculated to quantify the strength and direction of the relationship between MYBL2 and each biomarker.

The biological role of MYBL2 in hepatocellular carcinoma (HCC) cell viability was examined using CRISPR knockout screen datasets, which are publicly available through the Cancer Dependency Map (DepMap) portal [27]. These datasets were originally published by the Broad Institute’s Project Achilles and the Sanger Institute’s SCORE (Sanger’s Catalogue of Somatic Mutations in Cancer) project [28]. The DepMap integrates these large-scale genomic datasets to provide comprehensive insights into cancer cell vulnerabilities. Gene set enrichment analysis (GSEA) for MYBL2 expression in TCGA-LIHC cohort was performed using GENI (Gene ENrichment Identifier), a web-based tool for gene expression analysis [29]. The causative interplay between MYBL2 gene expression levels and cell states at a single cell transcriptomics level that contribute to cellular development trajectory and fate was analyzed with CellTracer [30].

Drug sensitivity assessment was conducted using bulk sample data from three major pharmacogenomic databases: the Genomics of Drug Sensitivity in Cancer (GDSC), the Cancer Therapeutics Response Portal (CTRP), and the Profiling Relative Inhibition Simultaneously in Mixtures (PRISM) [24]. Drug response data and gene expression profiles of cancer cell lines were analyzed to evaluate the association between MYBL2 expression and drug sensitivity.

To investigate the potential role of MYBL2 in immunotherapy response, we examined the correlation of MYBL2 expression with the microsatellite instability (MSI), a predictive biomarker for immune checkpoint inhibitors. These analyses were performed using the ‘TCGAplot’ package in R, which facilitates the exploration and visualization of TCGA data [31].

The correlation between MYBL2 expression and various components of the tumor-immune microenvironment was comprehensively analyzed using the ‘TCGAplot’ package in R, encompassing immune inhibitors, immune checkpoints, immune cell infiltrates, and immune scores. Visualization of the results was achieved through heatmaps and scatter plots generated using the ggplot2 package in R.

To obtain high-resolution insights into MYBL2 expression across immune cell subtypes, single-cell RNA sequencing (scRNA-seq) data analysis was conducted using the HCCDB web application, in which cell type identification, MYBL2 expression profiling, and differential expression analysis between various immune cell populations were conducted [23]. Visualization of scRNA-seq results was performed using UMAP (Uniform Manifold Approximation and Projection) plots to represent cell clusters and MYBL2 expression patterns.

Human HCC cell line HepG2 was purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA; Cat. No. ATCC HB-8065). To ensure cell line identity and prevent contamination, the cell line was authenticated via Short Tandem Repeat (STR) profiling and confirmed to be mycoplasma-free using standard detection methods. The cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM; Thermo Fisher Scientific; Cat. No. 11965-092; Waltham, MA, USA) medium supplemented with 10% heat-inactivated fetal bovine serum (FBS; GIBCO, Thermo Fisher Scientific, Cat. No. 16140-071, Waltham, MA, USA), glutamax, and antibiotic–antimycotic at 37°C in a humidified incubator with 5% CO₂. Cells were seeded at a density of 1.5 × 10⁶ cells per 6 cm culture dish. Twenty-four hours after seeding, we transfected the HepG2 cells with a concentration of 25 nM miR-29a mimic (Cat. No. C-310521-07-0002) (Dharmacon, GE Healthcare, CO, USA) or miR control (Cat. No. CN-001000-01) (Dharmacon, GE Healthcare (now Horizon Discovery), CO, USA) for 24 h with Lipofectamine™ RNAiMAX Transfection Reagent (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA; Cat. No. 13778150) according to the manufacturer’s instructions.

Cells or fresh frozen livers were homogenized using MagNA Lyser (Roche, Mannheim, Germany; Cat. No. 03 358 968 001) in PRO-PREP^™ Protein Extraction Solution (iNtRON Biotechnology, Seongnam, Republic of Korea; Cat. No. 17081). Thirty micrograms of proteins were analyzed through sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and immunoblotting. The primary antibodies anti-MYBL2 (1:40,000, PROTEINTECH, IL, USA; Cat. No. 18896-1-AP) and anti-glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (1:80,000, PROTEINTECH, IL, USA; Cat. No.10494-1-AP). For protein detection, blots were incubated with HRP-conjugated anti-rabbit immunoglobulin-G secondary antibodies (1:10,000; PerkinElmer, MA, USA; Cat. No. NEF812001EA). Chemiluminescent signals were developed using the Western Lightning Plus-ECL system (PerkinElmer, MA, USA; Cat. No. NEF812001EA) and subsequently quantified using ImageJ software.

The oligonucleotides that contained the MYBL2 3^′UTR target sequence (5^′-UGGUGCU-3^′) or MYBL2-3^′UTR-MUT (mutant) sequence (5^′-UCCUCGU-3^′) were annealed and cloned into the pMIR-REPORTTM miRNA Expression Reporter Vector (Ambion^™, Thermo Fisher Scientific, Waltham, MA, USA; Cat. No. AM5795) to generate the pMIR-MYBL2-29a vector or pMIR-MYBL2-MUT vector (negative control). The plasmids were purified using the EasyPrep EndoFree Maxi Plasmid Extraction Kit (BIOTOOLS, Taipei, Taiwan; Cat. No. DPT-BA17). HepG2 cells were cultured in a 10 cm dish and were transfected with 6 μg DNA plasmids of pMIR-MYBL2-29a or pMIR-MYBL2-MUT using TurboFect Transfection reagent (Thermo Fisher Scientific, Waltham, MA, USA). After 24 h of transfection, cells were seeded on a 6 cm dish with a 1.0 × 10⁶ cells/dish density and were grown overnight. The cells were transfected with miR-29a mimic or miR control for 24 h using the RNAiMAX transfection reagent according to the manufacturer’s instructions. After 48 h of transfection, the luciferase activity was measured using Luciferase Reporter Assay Kits (neolite, PerkinElmer, Waltham, MA, USA; Cat. No. R0531).

All animal protocols were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Chang Gung Memorial Hospital (application number: #2020121109) and were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. All animals were housed in a conventional facility at 22°C with a relative humidity of 55% in a 12 h light/12 h dark cycle with food and sterile tap water available ad libitum. Mice were purchased from BioLASCO Taiwan Co., Ltd. (Taipei, Taiwan). Nine-week-old male wild-type (WT) and miR-29a transgenic mice, with a C57BL/6 background, were utilized to investigate the role of miR-29a in HCC. To induce HCC, mice were subjected to a western diet (WD; Teklad diets, TD.120528, Envigo, Madison, WI, USA; Cat. No. TD.120528) and CCl₄ (Sigma-Aldrich, 289116-100ML, St. Louis, MO, USA; Cat. No.: 289116-100ML) treatment for 25 weeks as described previously [32]. The WD consisted of 21.1% fat, 41% sucrose, and 1.25% cholesterol. Mice were intraperitoneally injected with CCl₄ at a dose of 0.32 µg/gram body weight, once a week for 25 weeks [33]. The control mice were fed a normal diet (ND). The study included a total of 16 mice, equally distributed into four groups (N = 4 per group): WT-ND/CCl₄, WT-WD/CCl₄, miR-29aTg-ND/CCl₄, and miR-29aTg-WD/CCl₄. Sample size was determined using the Resource Equation Method for exploratory studies, adhering to the 3Rs principle of Reduction. Despite the small cohort size, the robust phenotypic differences observed provided sufficient statistical power to detect significant effects. The miR-29a transgenic mice, which overexpress miR-29a driven by the phosphoglycerate kinase 1 (PGK1) promoter, were generated and bred as previously described [34]. At the experimental endpoint, mice were deeply anesthetized with 40 mg/kg zoletil and 10 mg/kg xylazine. Maximal blood volume was obtained by cardiac puncture. Immediately after exsanguination, cervical dislocation was carried out to ensure euthanasia, in accordance with the recommendations of the National Institutes of Health Guide for the Care and Use of Laboratory Animals guidelines.

After transfection with the miR-29a mimic or miR control for 24 h, the cell pellets were lysed in protein extraction reagent (iNtRON Biotechnology, Seongnam, Republic of Korea; Cat. No. 17081), homogenized, and centrifugated. Protein extracts (100 μg) were prepared for a relative and absolute quantitation (iTRAQ) gel-free proteomics assay. The samples were first subjected to high-abundance protein depletion with the Pierce Top 12 Abundant Protein Depletion Spin Columns (Thermo Fisher Scientific, Waltham, MA, USA; Cat. No. 85165) and then sent to sample preparation with the iTRAQ Reagents Multiplex Kit (Sciex, Framingham, MA, USA; Cat. No. 4352135). After passing the standard quality control (QC) check, the labeled proteins were analyzed with LC/Q-Exactive Orbitrap MS (Thermo Fisher Scientific, Waltham, MA, USA) for 24 h, and the generated raw data were analyzed with Proteome Discoverer v2.4 (Thermo Fisher Scientific, Waltham, MA, USA) by referring to the MASCOT 2.5 database (Matrix science, London, UK).

Data are expressed as the mean ± standard deviation of at least three independent experiments. p < 0.05 was considered to indicate a statistically significant difference. For high-dimensional comparisons, including differential gene expression and pathway enrichment analyses, p-values were adjusted using the Benjamini-Hochberg False Discovery Rate (FDR) method, with a significance threshold of q < 0.05. Correlations between MYBL2 expression and drug sensitivity (GDSC/CTRP databases) or immune infiltration estimates were considered exploratory hypothesis-generating analyses; nominal p-values are reported for these associations, though key findings (e.g., Sorafenib) remained significant after FDR adjustment. All statistical analyses were performed using R software (version 4.1.0). Key packages included ‘survival’ (v3.2-13) for Cox regression, ‘pROC’ (v1.18.0) for ROC analysis, and ‘Seurat’ (v4.0) for single-cell analysis. Nomogram construction was performed by using Nomogram COX results visualization tool in Hiplot Pro (https://hiplot.com.cn/), a comprehensive web service for biomedical data analysis and visualization. A complete list of software versions and database URLs is provided in Supplementary Table S1.

The flowchart for this study is illustrated in Fig. 1. First, we utilized the Cox Hazard Model to examine the effect of AJCC stage and MYB oncogenes and their downstream targets on the survival rate of the TCGA-LIHC cohort, revealing that stage III, stage IV and 10 genes including MYBL2, BIRC5, CCNA2, CCNB1, CD34, CDK1, CDK2, NCAPH, PLK1 and TAL1 showed significant association with survival (Fig. 2A). We utilized the machine learning algorithm LASSO to identify the most relevant genes for survival prediction. This approach determined the optimal regularization level (Fig. 2B), revealed genes that remained significant at this regularization level (Fig. 2C), and yielded coefficients for seven relevant genes (Fig. 2D): CD34, CCNB1, KIT, PLK1, ATR, IGF1R and MYBL2. Subsequently, we assessed these seven LASSO-selected genes for their ability to discriminate between normal and tumor tissue (Fig. 2F–K). Further investigation into the ability of these genes to stratify patients into high and low relapse-free survival groups (Fig. 2L–P) identified MYBL2 as the most predictive biomarker.

Comprehensive analysis of gene expression indicated basal expression levels of hepatic MYBL2 across normal tissue (Fig. 3A). Elevated MYBL2 expression levels were noted across TCGA pan-cancer types, including HCC (Fig. 3B,C). The expression patterns of GSE144269 (Fig. 3D), GSE14520 (Fig. 3E), and GSE54236 (Fig. 3F) were in line with the TCGA cohort. TNMplot shows a similar pattern (Fig. 3G). Furthermore, high MYBL2 levels correlated significantly with conditions indicative of poor prognosis, including high tumor grade (Fig. 3H), low body mass index (BMI) (Fig. 3I), presence of microvascular invasion (MVI) (Fig. 3J), hepatitis B virus (HBV) positivity (Fig. 3K), and non-response to TACE (Fig. 3L).

A Cox regression analysis summarized the impact of MYBL2 expression on HCC cohorts, revealing that elevated levels are indicative of unfavorable outcomes (Fig. 4A), including in the TCGA cohort (Fig. 4B–E), the ICGC cohort (Fig. 4F) and the GSE144269 cohort (Fig. 4G). To develop a more comprehensive prognostic model, we created a nomogram integrating MYBL2 expression with AJCC stage, generating a prognostic score for each patient (Fig. 4H). Compared to predictions based on stage alone (AUC values of 0.65, 0.69, and 0.67 for 1-, 3-, and 5-year survival, respectively) (Fig. 4I), our prognostic model incorporating both stage and MYBL2 expression demonstrated superior predictive power (AUC values of 0.73, 0.77, and 0.72 for 1-, 3-, and 5-year survival, respectively) (Fig. 4J). Risk score analysis revealed that higher scores were associated with increased mortality and shorter survival times (Fig. 4K). Moreover, the prognostic score effectively stratified HCC patients into groups with significantly different survival probabilities (Fig. 4L). Collectively, these results demonstrate that combining MYBL2 expression with AJCC stage significantly enhances clinical prognostication.

To assess clinical applicability, we examined the correlation between MYBL2 and a panel of established HCC biomarkers. MYBL2 exhibited a positive correlation with most biomarkers (Fig. 5A–C), including AFP (Fig. 5D–F), MKI67 (Fig. 5G–I), PCNA (Fig. 5J–L), and BIRC (Fig. 5M–O). The causative interplay between MYBL2 gene expression levels and cell states that contribute to cellular development trajectory and fate was analyzed with CellTracer at a single cell transcriptomics level (Fig. 5P–R), revealing that MYBL2 positively correlated with the functional score of the Cell Cycle (Fig. 5S) while negatively correlated with that of Inflammation (Fig. 5T). It is worth noting that this “Inflammation” score likely reflects acute inflammatory signaling, which is often suppressed in established tumors, contrasting with the chronic, immunosuppressive immune infiltration observed in our TME analysis. Finally, we explored the role of MYBL2 in HCC cell viability by analyzing CRISPR screens from the DepMap database. Among 23 HCC cell lines, 21 demonstrated growth inhibition following MYBL2 knockout (Fig. 5U), indicating that MYBL2 exerts a pro-survival effect in HCC. Collectively, these findings demonstrate the potential involvement of MYBL2 in promoting cancer cell proliferation and in inhibiting tumor immunoreactivity.

In a comprehensive multi-database pharmacogenetic survey (Fig. 6A), we observed that MYBL2 expression levels were negatively correlated with half-maximal inhibitory concentration (IC50) value of sorafenib in a variety of datasets (Fig. 6B–I).

To clarify the impact of MYBL2 on tumor immune microenvironment, we examined the relationship between MYBL2 and immune signature. Bulk RNA-Seq analysis of the TCGA-LIHC cohort revealed that MYBL2 expression levels positively correlated with the majority of the immune inhibitors (Fig. 7A), as well as with genes of immune checkpoints (HAVCR2, CD274/PD-L1, PDCD1LG2/PD-L2, PDCD1/PD-1, CTLA4 and TIGIT) (Fig. 7B). While MYBL2 showed associations with some immunoreactive infiltrates (decreased macrophage M2 and monocyte; increased T follicular helper cells, memory B cells, and activated memory CD4 T cells), it also correlated with several potentially immunosuppressive infiltrates (macrophages M0, Tregs, resting dendritic cells, and neutrophils) (Fig. 7C). In addition, MYBL2 expression levels had a negative correlation with mucosal-associated invariant T (MAIT) cells (Fig. 7D), and a positive correlation with exhausted T cells (Fig. 7E), and natural regulatory T cells (nTregs) (Fig. 7F). Notably, MYBL2 expression levels positively correlated with the MSI (Fig. 7G), a predictive biomarker for immune checkpoint inhibitors. Furthermore, single-cell transcriptomic analysis (Fig. 7H) identified reduced MYBL2 expression levels in plasmacytoid dendritic cells (pDCs) of HCC compared to those of normal tissue and adjacent tissue (Fig. 7I,J). Taken together, these findings highlight MYBL2’s complex role in modulating the tumor immune microenvironment and its potential relevance to immune checkpoint therapy responses.

To elucidate the regulatory mechanisms of MYBL2 and explore potential therapeutic approaches, we conducted MIRDB-based GSEA, which identified miR-29a as a putative negative regulator of MYBL2 (Fig. 8A,B). We subsequently confirmed a significant negative correlation between miR-29a and MYBL2 expression levels in the TCGA cohort (Fig. 8C). Notably, elevated miR-29a expression was associated with favorable overall survival outcomes in HCC patients (Fig. 8D).

Our in vitro study verified that miR-29a overexpression inhibited MYBL2 expression (Fig. 8E) through direct binding to the MYBL2 3^′-UTR (Fig. 8F,G). To investigate the role of miR-29a overexpression in tumorigenesis in vivo, we utilized wild type (WT) mice and transgenic miR-29a (Tg) mice subjected to a WD/CCl₄-induced HCC model (Fig. 8H). Compared to WT-WD/CCl₄ mice exhibiting hepatic tumor formation, Tg-WD/CCl₄ mice showed no tumor development, decreased expression levels of KI67 (a marker for cancer proliferation), reduced MYBL2 expression, and diminished nuclear percentage of MYBL2 (Fig. 8I–M). Furthermore, Tg-WD/CCl₄ mice demonstrated reduced Masson’s Trichrome staining compared to WT-WD/CCl₄ (Fig. 8N), suggesting that miR-29a overexpression confers a protective effect against WD/CCl₄-induced liver fibrosis.

To profile the functional effect of miR-29a overexpression in HCC cells, we performed iTRAQ-based proteomics analysis (Fig. 8O). Among the downregulated genes, 35 genes were bioinformatically recognized as miR-29a targets (Fig. 8P), with MYBL2 emerging as the most significantly downregulated gene (Fig. 8Q). Pathway enrichment analysis of this downregulated cluster revealed a significant enrichment in “One-carbon metabolism” and “Cysteine” and methionine metabolism” pathways (Supplementary Fig. S1), reinforcing the functional consequence of the miR-29a-MYBL2 axis. Collectively, these findings establish MYBL2 as a prognostic biomarker for HCC and demonstrate its potential as an effective target of miR-29a in suppressing HCC development (Fig. 8R).