We obtained clinical data and RNA expression profiles for patients with hepatocellular carcinoma from the Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. There were also single-cell RNA sequencing (scRNA-seq) datasets of HCC samples downloaded from the GEO database (GSE149614). The GSE149614 dataset comprises samples from 10 HCC patients, with a total of 71,915 cells passing quality control filters. Clinical characteristics of these patients are summarized in Supplementary Table S1, including age, sex, tumor stage, etiology, and survival outcomes where available (Chen Y. et al., 2024; Chlahbi et al., 2024).

This study was designed as a bioinformatic discovery and characterization study to establish RNF157 expression patterns in HCC and develop an associated prognostic signature. The selection of FAP and CD11b as experimental targets was based on our single-cell RNA sequencing analysis, which revealed that RNF157 exhibited heterogeneous expression patterns across different cell types within the tumor microenvironment (TME), particularly showing significant associations with cancer-associated fibroblasts (CAFs) and tumor-associated macrophages (TAMs). FAP (Fibroblast Activation Protein) serves as a canonical marker for activated CAFs, while CD11b is a well-established marker for myeloid-derived cells including TAMs. The FAP and CD11b knockdown experiments were designed as proof-of-concept studies to demonstrate that the TME populations identified through our computational analyses as RNF157-associated could be experimentally modulated, rather than to directly test RNF157 function.

Venn diagram visualization was employed to identify overlapping genes implicated in disease processes, subsequently capturing core hub genes through intersection analysis of disease-associated gene sets. This approach facilitated efficient candidate gene selection for subsequent analysis. Utilizing the STRING database, we conducted protein-protein interaction (PPI) network analysis (Chen CM. et al., 2024; Du et al., 2024; Watanabe et al., 2024). This analytical framework enabled examination of interactions and relationships among intersected genes, providing mechanistic insights into potential disease processes.

Differential expression analysis was performed on genes demonstrating prognostic correlations, incorporating patient survival status and duration data. The TCGA-LIHC dataset was randomly divided into training (70%) and internal validation (30%) cohorts using a fixed random seed for reproducibility. We employed 10-fold cross-validation during model training to prevent overfitting. Additionally, we performed bootstrap resampling (1,000 iterations) to assess model stability and reported the confidence intervals for all performance metrics. External validation was performed using the GSE14520 dataset from the GEO database. To enhance analytical depth, unsupervised clustering methodologies were applied to stratify patients into distinct donor groups. Through this technique, we identified natural data clusters potentially representing discrete prognostic groups or disease subtypes. Following clustering procedures, we implemented a scoring framework designed to quantify prognostic implications of identified gene expression profiles. Principal component analysis (PCA) was conducted on the first two components to establish this analytical framework (Watanabe et al., 2024; Effati et al., 2024; Jeong et al., 2024; Rasul et al., 2024).

The “Seurat” R package was used for scRNA-seq data analysis. Quality control of data was conducted after cells with a number of features <200 and a mitochondrial content >20% were removed. These thresholds were selected based on established protocols in the field and the specific characteristics of liver tissue samples. Hepatocytes naturally exhibit higher mitochondrial content due to their metabolic activity, and overly stringent filtering could result in the loss of biologically meaningful hepatocyte populations. Quality control metrics (genes per cell, UMIs per cell, mitochondrial percentage distributions) are presented in Supplementary Figure S1. Integration and batch effect correction of single-cell data across samples. For unsupervised cell phenotyping, the LogNormalization method was used, then PCA and tSNE (t-Distributed Stochastic Neighbor Embedding) were processed. Cell types in each cluster were annotated using package ‘SingleR’ and marker genes with differential expression levels across cell types were identified by the package of “FindAllMarkers” (He et al., 2024; Zhang et al., 2024).

The CAFs were obtained from cancerous liver tissues resected from HCC patients, with the approval of the Institutional Review Board. THP-1 monocytes (ATCC) were differentiated into TAMs, as previously described. Cells were cultured in DMEM (Gibco) medium supplemented with 10% FBS, 100 U/mL penicillin; 100 μg/mL streptomycin at 37 °C and incubated with 5% CO₂. Cell culture medium was refreshed every 2–3 days, and cells were subcultured by trypsin/EDTA (0.25%) treatment when they became 80%–90% confluent.

shRNA sequences targeting FAP (for CAFs) and CD11b (for TAMs) were designed using an online tool (https://rnaidesigner.thermofisher.com), synthesized, and cloned into the lentiviral pLKO vector. 1-puro (Addgene, United States). The following shRNA sequences were used: sh-FAP: 5′-GCC​TCT​GAA​GGT​ATA​TGA​CTA-3′; sh-CD11b: 5′-GCA​GGT​CAT​CAA​GTA​CGA​TGT-3′. Non-targeting scrambled shRNA (5′-CCT​AAG​GTT​AAG​TCG​CCC​TCG​AAA​AAA​AA-3′) was used as negative control (NC). Lentiviral particle generation was achieved by cotransfecting 293T cells with shRNA expression plasmids, psPAX2 packaging plasmid, and pMD2 envelope plasmid. G with Lipofectamine 3000 (Invitrogen, United States) as the manufacturer recommended. The viral supernatants were harvested 48–72 h after transfection, filtered using 0.45 μm filters and applied to CAF and THP-1-derived TAM infection in the presence of 8 μg/mL polybrene. Stable transfectants were selected with 2 μg/mL puromycin over a period of 7 days. Knockdown efficiency was confirmed by qRT-PCR and Western blot.

THP-1 monocytes were induced to differentiate into macrophages with 100 ng/mL phorbol 12-myristate 13-acetate (PMA, Sigma-Aldrich) for 48 h. After differentiation, cells were washed twice with PBS and cultured in new DMEM for another 24 h. For TAM-like polarization, differentiated macrophages were treated for 48 h with conditioned medium obtained from HCC cell cultures (diluted 1:1 with fresh DMEM). TAM phenotype was verified by flow cytometry and RT-qPCR for the expression of markers (CD163, CD206) before shRNA transfection.

Cell harvest for flow cytometry was performed using enzyme-free cell dissociation buffer (Gibco) to maintain the surface antigens. About 1 × 10⁶ cells were washed for two with cold PBS/1% FBS and then staining with fluorochrome-conjugated antibody in the dark at 4 °C for 30 min. For CAF analysis, cells were stained with FITC-conjugated anti-FAP antibody (FITC-CA, Abcam, 1:100 dilution). For TAM analysis, cells were stained with anti-CD11b antibody PE-conjugated (PE-A, BD Biosciences, 1:100) to assess the population of CD45⁺CD11b+. After staining, cells were washed twice in PBS 1% FBS and resuspended in 500 μL of PBS for analysis. Flow cytometry was carried out on BD FACSCanto II flow cytometer (BD Biosciences, United States) and the data analyzed by FlowJo software (version 10.0.7, Tree Star Inc., United States). At least 10,000 events were acquired for each sample. The gating strategy to exclude dead cells was conducted for live cell population based on FSC and SSC characteristics (P1 region) with subsequent fluorescence intensity analysis to detect various cell populations (FAP+/FITC-CA+ in CAFs and CD11b+/PE-A+ for TAMs).

All the experiments were performed using three independent biological replicates (n = 3), with technical duplicates for each biological replicate. Statistical analyses were performed on biological replicates, and data are shown as mean ± SD. The GraphPad Prism 9.0 software (GraphPad Software, United States) was used for statistical analysis. Student's t-test was used for analyses between two groups, and one-way analysis of variance (ANOVA) with Tukey’s post hoc test for comparisons between more than two groups. For correlation analyses, Benjamini–Hochberg false discovery rate (FDR) correction was applied for multiple testing, and correlations were considered significant only when FDR <0.05. P values <0.05 were considered to be statistically significant.

To address our first research question regarding cellular heterogeneity in HCC, we performed comprehensive single-cell analysis. A PCA plot (Figure 1A) shows the distribution of cells from different samples relative to each other where individual cells are plotted as points and color represents sample identity. The figure with the scree plot shows how much of the variance is explained by each principal component which can help us decide how any components to retain for further analysis. Figure 1B displays the t-distributed stochastic neighbor embedding (t-SNE) plot identifying well-defined cellular communities, colour-coded according to different cellular clusters, possibly representing cell states or subtypes present in liver cancer microenvironment. Another t-SNE plot is shown in Figure 1C, where cells are classified by type and the high complexity of TME with immune infiltrates, cancer associated fibroblasts and malignant cells is emphasized. Figure 1D presents a uniform manifold approximation and projection (UMAP) plot, which is an alternative representation of cell type distribution that frequently better maintains global structure than t-SNE. These findings establish the foundation for investigating RNF157 expression patterns across these diverse cellular populations.

Building on the cellular heterogeneity analysis, we next examined RNF157 expression patterns across the identified cell populations. Figures 2A–D show the expression levels of RNF157 gene in various cell types in liver cancer. A dot blot plot of average RNF157 expression and the frequency of cells expressing RNF157 in different trait features is shown in Figure 2A. The dot size indicates percentage of cells expressing RNF157, and colour represents the average expression level. This is further delineated by cellular identity in Figure 2B through showing the expression of RNF157 in different cell types (malignant cells, immune cells and cancer associated fibroblasts). The direct spatial distribution of RNF157 expression across the cellular plane is visually presented in a UMAP plot, Figure 2C whereat we observe distinct colors representing different levels of RNF157 gene expression. A density plot is overlaid on the UMAP in Figure 2D, with darker spots indicating larger fraction of RNF157 expression. These results revealed that RNF157 expression was particularly associated with CAF and TAM populations, providing the rationale for our subsequent experimental validation targeting these cell types.

On the other hand, Figures 3A–D depict the patterns and relationships of different genes in liver cancer. Figure 3A Heatmap shows levels of its expression across different cell types like CAFs, TAM and T-cells. Color strength corresponds to the expression level of a gene, enabling understanding of cellular diversity in the tumor microenvironment. Dot plots for the interaction and pathways of genes are shown in Figures 3B–D. Plots are showing significant interactions (dot size is interaction strength and the color represents expression). Figure 3B highlights some important pathways and the detailed gene interactome in these pathways is shown where Figures 3C,D elaborate on individual gene interactions within them also providing additional insight to their probable contribution to tumor development and immune response.

Figure 4A shows the PCA plot of cellular distribution from multiple samples, where colors represent sample labels. An associated scree plot indicates the variance explained by each principal component, thereby helping to inform decisions in dimensionality reduction. The cell clustering into groups based on gene expression profiles is shown in Figure 4B (UMAP plot), color- and cluster-coded. This visualisation can guide to potential subtypes of cells in the tumour microenvironment. Figure 4C extracts RNF157 expression in a UMAP mapping on the cellular landscape. The heat map shows a gradient that represents different expression leveles of RNF157, the hotter is higher. Figure 4D displays a density plot on top of the UMAP indicating presence of cells with higher concentration expressing RNF157.

In Figures 5A–D, the prognostic factors and survival prediction is examined in liver cancer. Heatmap of the different gene signatures is shown in Figure 5A, which demonstrates the C-index for each cohort by colors. This index quantifies the predictive power of each gene signature; it identifies which gene signatures are statistically most relevant for prognosis. A rose diagram depicting chromosomal locations of genes correlated with the prognosis in liver cancer was shown in Figure 5B. This illustration is useful for suggesting loci of interest that are worth studying further. There is a nomogram in Figure 5C, a predictive model that calculates survival probability using the gene expression of AKR1B10, CDK4 and CXCL1. It should be noted that this is a multi-gene prognostic model developed through systematic screening of RNF157-correlated genes. AKR1B10, CDK4, and CXCL1 were identified as genes significantly co-expressed with RNF157 and independently associated with patient prognosis. The model therefore represents an RNF157-centered gene signature rather than a model based solely on RNF157 expression. Score points are obtained from each gene and integrate into overall points that allot 1-year, 2-year, and 3-year survival probabilities. A calibration plot comparing the predicted survival probabilities of nomogram and actual observations is illustrated in Figure 5D.

Figures 6A–F analyze predictive ability and survival outcome of a prognostic model in liver cancer. Figures 6A–C are receiver operating characteristic (ROC) curves for predicting the survival of patients at 1 year, 3 years and 5 years; The area under the curve (AUC) value reflects model accuracy. Kaplan-Meier survival curves of high-risk versus low-risk groups are shown in Figures 6D–F. The curves are strikingly different for the survival probabilities at a given time, and p-values demonstrate statistical significance of the difference.

The risk stratification and survival of liver cancer patients were further evaluated by Figures 7A–C. Each figure is composed of two plots: a risk score distribution, and a survival status chart. Patients are ranked according to their risk score in the left panels of Figures 7A–C, and the high-risk (red) and low-risk (blue) groups are also shown. The separation of groups suggests that the model may have potential utility in stratifying patients according to their propensity for risk. The plots on the right present survival times for the same patients as dots (red if died and blue if alive). These observations suggest associations between risk scores and survival outcomes that require validation in prospective cohorts.

Figure 8A shows a heat map of the correlations between the crucial genes (including AKR1B1 and CDK4) and different sorts of immune cells. Figures 8B,C further illustrate correlations between RNF157-associated genes and specific immune pathways and immune cell subtypes. Dot color and size correspond to the direction of correlation and strength, respectively; red = positive correlation, green = negative. *Indicates significant correlation (FDR <0.05 after Benjamini–Hochberg correction). It is important to note that these correlations do not establish causal relationships. Alternative explanations for these associations include confounding by shared upstream regulators, cell type composition effects, or the possibility that immune infiltration drives gene expression rather than vice versa. These hypotheses require future experimental validation.

Heatmap for DNA methylation in different genes of various cancer types as demonstrated in Figure 9A. Circle size reflects significance (FDR) and color reflects the direction of methylation change (red = hypermethylation, blue = hypomethylation). Figure 9B classifies genetic variant categories and types in samples. Pie charts in Figure 9C indicate the proportion of copy number variations (CNVs) for each gene involved. Figures 9D,E further show the differences in immune cell abundance between mutant and wild-type groups, as well as changes associated with copy number variation status. These observations are hypothesis-generating, and determining whether methylation alterations or CNVs causally affect RNF157 expression or function requires dedicated experimental studies.

The flow cytometry charts illustrate distribution and characteristics of liver cancer-associated fibroblasts (CAFs) and tumor-associated macrophages (TAMs) under different conditions. Figures 10A,C (CAF-NC vs. CAF-sh) show cell distribution of all events in control group (CAF-NC) and treatment group (CAF-sh). Both groups have distinct cell populations, with similar proportions of cells in region P1 (approximately 61.74%). Figures 10B,D (CAF-NC: P1 vs. CAF-sh: P1) further analyze cells in region P1, revealing that the proportion of cells marked with FAP and FITC-CA is higher in the CAF-NC group (45.54%) compared to the CAF-sh group (22.01%), indicating that CAF-sh treatment successfully reduced a specific subset of CAFs. Figures 10E,G (TAM-NC vs. TAM-sh) display cell distribution of all events in control group (TAM-NC) and treatment group (TAM-sh). Similar to CAF groups, both groups also have distinct cell populations, with similar proportions of cells in region P1 (approximately 63.17% and 62.29%). Figures 10F,H (TAM-NC: P1 vs. TAM-sh: P1) further analyze cells in region P1, showing that the proportion of cells marked with CD11b and PE-A is higher in the TAM-NC group (35.03%) compared to the TAM-sh group (24.18%), suggesting that TAM-sh treatment successfully reduced a specific subset of TAMs. These results demonstrate that the TME populations computationally associated with RNF157 expression can be experimentally manipulated, validating our analytical approach. However, these experiments do not establish a direct regulatory relationship between RNF157 and FAP/CD11b expression.