Multimodal single-cell network analysis uncovers BSG/CD147 as an early biomarker and signaling hub in hepatocellular carcinoma

Multimodal single-cell network analysis uncovers BSG/CD147 as an early biomarker and signaling hub in hepatocellular carcinoma | 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 Multimodal single-cell network analysis uncovers BSG/CD147 as an early biomarker and signaling hub in hepatocellular carcinoma Thomas DW Wang, Eun-Young K Choi, Shuo Feng, Hui Jiang, Thomas DS Wang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8593693/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Hepatocellular carcinoma (HCC) is a leading cause of cancer-related mortality worldwide, largely due to late-stage diagnosis and the limited sensitivity of current biomarkers such as α-fetoprotein (AFP). Early detection requires molecularly defined targets that capture the initial steps of malignant transformation. Single-cell RNA sequencing (scRNA-seq) offers high-resolution insight into tumor heterogeneity and lineage progression to enable the identification of early biomarkers. This study aimed apply scRNA-seq analysis to detect clinically important molecular patterns that define the early stages of malignant transformation in HCC and facilitate the diagnosis of small or ambiguous lesions. Methods Two independent scRNA-seq datasets (GSE149614 and GSE189903) comprising non-tumor and HCC tissues were analyzed. Following batch correction and clustering, hepatocyte subpopulations were characterized by differential expression, pseudotime, and CytoTRACE analyses to reconstruct the trajectory from normal to malignant states. High-dimensional weighted gene co-expression network analysis (hdWGCNA) was used to identify stage-associated modules, while CellChat and protein-protein interaction analyses delineated intercellular signaling networks. Target expression was validated in paired human liver specimens using quantitative immunofluorescence. Results scRNA-seq revealed a continuum of hepatocyte states characterized by progressive stemness and oncogenic pathway activation (MYC, E2F, G2M). Module Hep-M20 exhibited the strongest correlation with tumor stage and identified BSG/CD147 as a central hub gene with monotonic upregulation along pseudotime and strong correlation with stemness potential. CellChat analysis uncovered a cyclophilin (PPIA/PPIB)-dependent tumor-stroma signaling axis that positions BSG/CD147 as the key mediator for intercellular communication between tumor hepatocytes, fibroblasts and T cells. Ex vivo validation confirmed significantly higher BSG/CD147 protein expression in HCC versus background liver (P = 2.9×10⁻ 11 ) with excellent diagnostic accuracy (AUC = 0.93–0.96; sensitivity 86–87%; specificity 93–97%) including in lesions < 2 cm that are frequently indeterminate on conventional imaging. Conclusions This study establishes BSG/CD147 upregulation as an early molecular event in hepatocarcinogenesis that integrates hepatocyte dedifferentiation, microenvironmental signaling, and tumor progression. Strong and specific expression in small lesions < 2 cm underscores potential as a precision biomarker and imaging target for early HCC detection, risk stratification, and therapeutic development. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Hepatocellular carcinoma (HCC) is a common primary liver malignancy, and a leading cause of cancer-related deaths worldwide [ 1 , 2 ]. The incidence continues to rise, driven by chronic hepatitis B and C infections, metabolic liver disease, and cirrhosis [ 3 ]. Despite advances in imaging and therapy, most patients are diagnosed at an advanced stage when curative interventions, such as resection or transplantation This pathway is likely to foster a microenvironment conducive to invasion, angiogenesis, and immune modulation are no longer useful [ 4 ]. Early-stage HCC is frequently asymptomatic, and current biomarkers lack the sensitivity and specificity needed for reliable detection [ 5 ]. This limitation underscores the need for biologically based biomarkers that can identify early or indeterminate (LI-RADS 3/4) lesions on imaging to improve clinical decision-making [ 6 ]. Many imaging techniques, such as ultrasound (US), computed tomography (CT), and magnetic resonance imaging (MRI), are used in clinical practice [ 7 ] and serum biomarkers, such as α-fetoprotein (AFP) and des-γ-carboxy prothrombin (DCP), are commonly used to assist in detection and monitoring. However, diagnosing HCC at an early stage remains a significant challenge. These techniques often lack sensitivity, especially with small lesions or in the context of liver cirrhosis [ 8 , 9 ]. Furthermore, false positive diagnoses may lead to unnecessary invasive procedures or undue patient anxiety. Therefore, molecular signatures, such as serum or tumor biomarkers that deliver higher sensitivity and specificity for the early and accurate detection of HCC as diagnostic tools, are greatly needed. Tumor heterogeneity is a well-known feature of HCC, and results in high diversity at the cellular, molecular, functional, and lineage levels [ 10 ]. Cellular and molecular variability poses a significant barrier to early diagnosis [ 11 ]. This biological complexity gives rise to distinct molecular profiles even at early disease stages and in small nodules. Thus, conventional imaging phenotypes or bulk biopsy assessments may fail to capture the full molecular lesion complexity. In these contexts, molecular signature, such as gene‑expression panels, pathway‑activation markers, or cell‑type‑specific transcripts, may provide the sensitivity needed to detect small or early HCC lesions by capturing the molecular “footprint” of malignant transformation that precedes overt radiological changes [ 12 ]. Traditional bulk transcriptomic methods suffer from aggregation of signals across all cells within the sample, thereby overlooking both rare populations and interactions among distinct cell types [ 13 ]. In contrast, single‑cell RNA sequencing (scRNA‑seq) has revolutionized this landscape by enabling high‑resolution transcriptomic analysis at the level of individual cells. This approach is capable of revealing rare hepatocyte and progenitor‑like cells, reconstructing lineage trajectories, and delineating signaling networks that underlies cancer transformation. Although scRNA‑seq has been used to profile immune and stromal HCC components, the full potential to identify clinically actionable early biomarkers remains unrealized. High-resolution data generated by scRNAseq allows for the discovery of candidate early biomarkers, such as specific transcripts expressed only in emerging malignant subpopulations or in the supportive microenvironmental niche, which remain undetected in bulk transcriptomic analyses. The aim of the study was to apply scRNA-seq analysis to identify clinically relevant molecular signatures that capture the process for early malignant HCC transition and support the diagnosis of small or indeterminate lesions. Materials and Methods Data acquisition from scRNA-seq datasets Single-cell transcriptomic datasets were obtained from the Gene Expression Omnibus (GEO) repository ( https://www.ncbi.nlm.nih.gov/geo ) to investigate gene expression dynamics during HCC transformation. Two independent datasets were evaluated. The discovery dataset (GSE149614) consisted of liver specimens from n = 10 HCC patients distributed in 4 tissue types, including non-tumor liver, primary tumor, portal vein tumor thrombus, and metastatic lymph nodes. Only cells from non-tumor liver and primary tumor tissues were retained to focus on early cancer transformation. This cohort captured a wide range of tumor-node-metastasis (TNM) stages and etiologies, including hepatitis B virus (HBV), hepatitis C virus (HCV), and metabolic liver disease. A validation dataset (GSE189903) was used for independent module preservation analysis and included n = 4 HCC patients representing diverse tumor sizes. Single cells were prepared from different tumor regions and adjacent normal tissue. This approach enabled assessment of the robustness and reproducibility of gene co-expression network modules derived from the primary dataset across independent patient cohorts and sequencing platforms. Single-cell RNA-seq data processing and quality control All analyses were conducted using R software, v4.4.0 with the Seurat v5 package [ 14 ]. Raw gene expression count matrices were imported and converted into Seurat objects. Quality control procedures were applied to exclude low-quality cells using standard filtering criteria, including the removal of cells with high mitochondrial gene content that indicate apoptotic cells, abnormally low or high gene counts that suggest empty droplets or doublets, and low total RNA expression. GSE149614 was preprocessed, and only batch correction was performed. For GSE189903, cells with nCount_RNA < 800, nFeature_RNA 25% were excluded. These quality control steps ensured that only high-quality, biologically meaningful single-cell profiles were retained. Batch correction was performed to correct technical variations across patients and sequencing runs using the Seurat integration pipeline. Shared biological features were aligned across diverse datasets to minimize the influence of technical artifacts and ensure integrity for downstream analyses, and included clustering, differential gene expression, and pseudotime trajectory modeling. Cell clustering, annotation, and characterization After initial quality filtering, the FindVariableFeatures function was used to identify highly variable genes with a focus on those that contribute most to cellular heterogeneity. The data were then normalized and centered using ScaleData, and dimensionality reduction was performed using RunPCA. The appropriate number of principal components for downstream analysis was determined with the ElbowPlot function. A shared nearest-neighbor graph was constructed using FindNeighbors. Unsupervised clustering was performed with FindClusters to identify transcriptionally distinct cell populations. For visualization in 2D space, RunTSNE and RunUMAP were applied. Cluster-specific gene expression was evaluated using FindAllMarkers for global differential expression gene (DEG) analysis across all clusters, and FindMarkers for pairwise comparisons between selected clusters. Clusters were annotated based on expression of known liver cell-type marker genes. Only clusters identified as hepatocytes were retained to focus on tumor-associated changes. Seurat visualization functions, including dot plots, violin plots, and bar plots, were used to display gene expression patterns, assess cellular composition, and compare distributions across normal and tumor samples. Analysis of stemness using CytoTRACE CytoTRACE (Cellular Trajectory Reconstruction Analysis using gene Counts and Expression) was used to infer stemness and differentiation potential across normal, pro-tumor, and tumor clusters [ 15 ]. The pre-processed Seurat object from dataset GSE149614 was used for the CytoTRACE analysis. The filtered and normalized scRNA-seq data was converted into a gene expression matrix by extracting raw counts using the as.matrix() function (arkov_object[["RNA"]]@counts). This matrix served as input to the CytoTRACE() function from the CytoTRACE R package (v0.3.3) using default parameters unless otherwise specified. CytoTRACE scores, which reflect transcriptional diversity and infer cellular plasticity, were incorporated into the Seurat object metadata. These scores were visualized on UMAP embeddings to assess differentiation gradients across clusters. Higher CytoTRACE scores corresponded to less differentiated, more stem-like states and were used to inform downstream trajectory analyses. Pathway enrichment analysis Single-cell pathway enrichment analysis was performed using the escape R package (v2.2.3) on a Seurat v5 object containing hepatocytes. Enrichment scores were extracted from the "escape.ssGSEA" assay and converted into a matrix with cells in rows and pathways in columns. Average enrichment scores per pathway were computed across normal, pro-tumor, and tumor groups using dplyr. Averaged data was visualized using pheatmap by applying row-wise scaling and hierarchical clustering to highlight differential pathway activity patterns among groups. High-dimensional weighted gene co-expression network analysis (hdWGCNA) hdWGCNA was performed to identify transcriptional programs associated with HCC transition. Hepatocyte subsets were first extracted from the integrated Seurat object. Using the hdWGCNA R package, a gene expression correlation matrix was computed, and a soft-thresholding power of 10 was selected to generate a scale-free weighted adjacency matrix [ 16 ]. This parameter was chosen based on the standard approach in WGCNA, where the power was selected to approximate scale-free topology, a common property of biological networks. Specifically, the scale-free topology fit index and mean connectivity across a range of powers were examined. Power = 10 was the lowest value at which networks began to exhibit scale-free behavior (R² >0.85). The soft-thresholding power controls how strongly co-expression similarities (correlation values) are weighted in the adjacency matrix. Higher powers emphasize stronger correlations and suppressed weaker ones. This improved the reliability of module detection, leading to more biologically coherent gene modules. A topological overlap matrix (TOM) was then constructed to improve network robustness by emphasizing high-confidence gene-gene interactions. Gene modules were identified using average linkage hierarchical clustering and was followed by dynamic tree cutting to define module boundaries. Genes that did not cluster into any module were assigned to the gray module and indicated low or nonspecific connectivity. Module eigengenes, representative expression profiles of each module, were correlated with key clinical features, including tissue origin (normal versus tumor), tumor stage, and viral infection status (HBV/HCV) to assess potential biological relevance. Hub genes were defined as genes with the highest intramodular connectivity using the GetHubGenes function and were used to identify key regulators within each module. Preservation of key gene modules Module preservation analysis was conducted using GSE189903 to assess the reproducibility of co-expression modules identified in the primary dataset (GSE149614). The GetModulePreservation function from the hdWGCNA package was employed to compute Z-summary scores, which integrate multiple metrics of module preservation and quality, including intramodular connectivity and network density. The PlotModulePreservation function was used to visualize module preservation, and display module size against Z-summary scores. Modules with Z-summary values > 10 were considered highly preserved and indicated strong reproducibility across datasets. Identification of stage-associated biomarkers A multi-step integrative analysis was performed using gene co-expression modules and DEG data to identify robust biomarkers associated with HCC progression. First, hepatocyte-specific gene co-expression modules were generated using hdWGCNA. Modules were prioritized based on a strong correlation with clinical variables, such as tumor stage and tissue origin. A set of DEGs was generated by comparing gene expressions between tumor-derived and normal hepatocytes to refine candidate genes with functional relevance. The DEGs were then intersected with the gene members of prioritized modules to identify overlapping genes that were both differentially expressed and co-expressed within tumor stage-associated modules. To identify genes associated with tumor stage, DEG analysis was performed using the Seurat package with tumor stage set as the active identity class (Idents). DEG analysis was performed using the FindAllMarkers() function to restrict the analysis to a predefined list of intersecting genes. This approach identified stage-specific marker genes by comparing expression profiles across tumor stages. Expression dynamics were visualized using violin plots to allow for identification of genes that demonstrated progressive upregulation with increasing tumor stage. TCGA Analysis of Stage-Associated Gene Expression Transcriptomic and clinical data for HCC were obtained from The Cancer Genome Atlas (TCGA) using the TCGAbiolinks R package. Raw gene expression data (HTSeq counts, STAR workflow) were downloaded and processed using the GDCquery, GDCdownload, and GDCprepare functions. Gene-level expression was quantified, and BSG/CD147 expression was extracted using the corresponding Ensembl gene ID. Only primary tumor samples were retained to focus on tumor-specific expression. Samples lacking valid stage annotations or gene expression were excluded from further analysis. A Spearman’s rank correlation was performed to assess the relationship between BSG/CD147 expression and tumor progression. Additionally, boxplots, violin plots, and smoothed regression lines were generated to visualize stage-wise expression differences. Outliers were identified and excluded using the IQR method for sensitivity analyses. Due to the limited number of Stage IV samples, visualizations were focused on Stages I–III to ensure balanced group comparisons. Trajectory analysis of biomarker expression Pseudotime trajectory analysis was performed using the Monocle 2 R package to model the dynamic progression of hepatocyte transformation during HCC development [ 17 ]. Hepatocyte clusters previously identified using Seurat were used to construct the trajectory. These clusters were selected to capture the continuum from early to advanced tumorigenic states. Monocle 2 was applied to order cells along a pseudotime axis and infer a developmental trajectory from transcriptomic changes across individual hepatocytes. Gene expression was projected onto the trajectory to investigate the transcriptional activation patterns of candidate biomarkers. Pseudotime expression analysis was then performed to assess the onset and progression of gene activation along the transformation continuum from normal to malignancy. Cell-cell communication analysis To infer and analyze intercellular communication networks, CellChat R package, v1.6.11 was used [ 18 ]. The normalized expression matrix and metadata (cell type annotations) were used to create a CellChat object via createCellChat() function. The object was then subset to include only signaling-relevant genes using subsetData() function. Analysis used the built-in CellChatDB.human ligand-receptor interaction database. In addition, BSG/CD147-PPIA and BSG/CD147-PPIB interactions were manually incorporated to investigate their role in cell-cell signaling. These interactions were curated based on known protein-protein interaction evidence, and added to the database, including the ligand-target pair: BSG/CD147 (target); PPIA, PPIB (ligands), and signaling pathway name: BSG_PPIA_PPIB. Interactions were appended to the CellChatDB $ interaction and CellChatDB $ complex slots prior to data subsetting. The modified database was then used to subset the expression data with subsetData() and proceed with communication probability inference. Communication probabilities were computed using computeCommunProb(), followed by computeCommunProbPathway() and aggregateNet(), to infer and summarize pathway-specific signaling networks. Only interactions involving cell types with sufficient representation (minimum 10 cells) were retained. Visualizations such as netVisual_circle(), netVisual_bubble(), and netAnalysis_signalingRole_network() were employed to explore outgoing and incoming signaling patterns. The contribution of each ligand-receptor pair to the pathway was quantified using netAnalysis_contribution(). Protein-protein interaction analysis The BSG/CD147 protein-protein interaction (PPI) network was obtained from the STRING database ( https://string-db.org ) using a confidence score cutoff of 0.7. The network was visualized and analyzed in Cytoscape (v3.10.3). To assess context-specific interaction dynamics, expression-weighted PPI scores were calculated for normal, protumor, and tumor groups based on the average expression levels of interacting proteins and visualized using heatmap. Ex vivo validation of target expression in HCC Tissue procurement and use in this study were conducted in accordance with institutional guidelines and approved by the Institutional Review Board (IRB) of University of Michigan (Approval No: HUM00248521). Formalin-fixed, paraffin-embedded (FFPE) human liver sections were obtained from the archived tissue bank in the University of Michigan Department of Pathology. Sections (5 µm thick) were cut, mounted on Superfrost Plus glass slides (Fisher Scientific), and deparaffinized. Antigen retrieval was performed in sodium citrate buffer prior to staining. Slides were blocked with 5% goat serum for 1 hour at room temperature (RT), followed by overnight incubation at 4°C with monoclonal anti-BSG/CD147 antibody (MA529060, Invitrogen) at a 1:500 dilution. After 3X washes with phosphate-buffered saline containing Tween-20 (PBST, 3 min each), sections were incubated with a Cy5.5-conjugated secondary antibody for 1 hour at RT. Slides were then washed 3X with PBST and mounted with 1.5 µm coverslips using ProLong Gold Antifade Reagent with DAPI (8961; Cell Signaling Technologies). Fluorescence images were acquired using a confocal microscope with 20X objective under identical exposure settings for tumor and background liver. Mean fluorescence intensities were quantified by placing 3X 20×20 µm² boxes entirely within liver tissues using custom MATLAB software (MathWorks, Inc), while avoiding regions of saturated signal. Adjacent sections were processed for routine pathology (H&E) and independently evaluated by an expert liver pathologist (EKC). On each slide, the pathologist delineated the tumor region and adjacent background liver, which were then used to quantify the fluorescence signal. Statistical analysis Unless otherwise specified, default statistical tests were used for all R functions. DEG analysis was performed using the Wilcoxon Rank Sum test to assess statistical significance. Image quantification data were analyzed using GraphPad Prism, v10.4.1. Paired t-tests were applied where appropriate to assess differences between matched groups. Receiver operating characteristic (ROC) curve analysis was used to evaluate sensitivity and specificity. Linear regression was used to assess the relationship between tumor size and target-to-background (T/B) ratio, calculated as the fluorescence intensity of the tumor (T) divided by that of the surrounding background (B) liver. Spearman analysis was performed to determine correlation coefficients. Results Data acquisition from scRNA-seq datasets Single-cell transcriptomic analysis was performed to identify candidate biomarkers for early HCC detection. The discovery dataset (GSE149614) consisted of liver specimens from n = 10 HCC patients. Only cells from non-tumor liver and primary tumor were retained to focus on early cancer transformation. A validation cohort (GSE189903) was used for module preservation analysis. This dataset included n = 4 HCC patients with diverse tumor stages, sizes and locations, and was used to evaluate robustness and reproducibility of gene co-expression patterns derived from the primary dataset across independent patient cohorts and technical platforms. Single-cell RNA-seq data processing and quality control Initial analyses revealed substantial batch effects in both datasets with cells clustering primarily by patient origin and individual sample identifiers as shown on tSNE plots, Fig. S1 A , B . Batch-driven clusters were eliminated using the Seurat integration pipeline. Post-integration, cells from different patients and samples exhibited extensive intermixing, and reflected successful alignment of biologically similar cell populations, Fig. S1 C , D . This correction enabled reliable interpretation of cellular heterogeneity by ensuring that clustering and downstream analyses were driven by biological variations rather than technical differences. After stringent quality control and integration, the final datasets comprised 34,414 tumor-derived and 28,687 normal liver cells (GSE149614), and 43,656 tumor-derived and 30,614 normal liver cells (GSE189903). Cell clustering, annotation, and characterization Identification and Annotation of Liver Cell Types Unsupervised clustering identified 32 distinct cell clusters that represented a range of liver cell types. Annotations based on established marker genes revealed a diverse cellular composition, including hepatocytes, Kupffer cells, macrophages, dendritic cells, T cells, NK cells, B cells, fibroblasts, and endothelial cells, and reflects liver microenvironment heterogeneity, Fig. S2 A . Stratification by tissue origin showed differential distribution of these populations between normal and tumor with several clusters enriched in either tissue type, Fig. S2 B . Specific cell populations were preferentially represented in either normal or tumor, Fig. S2 C . DEG analysis revealed clear transcriptional differences between normal and tumor. Tumor hepatocytes showed upregulation of genes associated with malignancy, while normal hepatocytes retained expression of genes linked to physiological liver function, Fig. S2 D . Clusters 3, 4, 5, 8, 9, and 26 were identified as hepatocytes based on marker gene expression. Hepatocyte Subpopulation Analysis tSNE analysis was used to explore hepatocyte heterogeneity between normal and tumor. Hepatocyte clusters (red) were identified based on distinct marker profiles, Fig. 1 A. Cluster composition analysis revealed differential enrichment levels. Cluster 26 was composed predominantly of normal hepatocytes (N), while clusters 4, 5, and 9 were enriched in tumor (T). Clusters 3 and 8 contained cells from both normal and tumor, designated as pro-tumor (PT), and represented transitional cellular states, Fig. 1 B. CytoTRACE was used to evaluate stemness properties and was found to increase in tumor versus normal and pro-tumor clusters, Fig. 1 C. Dot plot analysis highlighted elevated expression of tumor-specific markers in clusters 4, 5 and 9, and expression of normal genes in cluster 26, Fig. 1 D. Clusters 3 and 8 exhibited a hybrid expression pattern by co-expressing both tumor and normal markers to support their role as intermediate states during tumor progression. Pathway enrichment analysis demonstrated significant upregulation of key cancer-related pathways within tumor clusters. Pathways such as MYC, which regulates cellular proliferation and metabolism [ 19 ]; DNA repair, which maintains genomic integrity [ 20 ]; the G2M checkpoint, which ensures proper cell cycle progression and prevents propagation of damaged DNA [ 21 ]; and E2F targets, which drive DNA synthesis and cell cycle regulation [ 22 ], were all elevated, Fig. S3. High-dimensional weighted gene co-expression network analysis (hdWGCNA) A soft-threshold value of 10 was chosen to build a gene network to best reflect natural biological relationships while keeping the connections simple, Fig. S4 A - D . Using this network approach, 22 groups of genes that show similar activity patterns, called modules, and were identified among hepatocytes, Fig. 2 A. Genes that did not fit into any module were placed in a “gray” group and showed weak or inconsistent relationships with others. A dot plot shows how these gene modules change across hepatocyte clusters as they progress from normal to tumor cells, Fig. 2 B. Hep-M18 and Hep-M20 expression increased steadily from normal hepatocytes (cluster 26) to early tumor-like (clusters 3 and 8) and finally tumor-rich clusters (4, 5, 9). When comparing modules with clinical features, several meaningful patterns emerged. One group, Hep-M20, showed the strongest link to tumor stage (ρ = 0.51) to suggest a role in cancer progression. Hep-M12 was most associated with whether the tissue came from a normal or tumor, and Hep-M6 correlated most with viral infection status (HBV or HCV), Fig. 2 C. Preservation of key gene modules Module preservation analysis was performed using hdWGCNA to assess robustness and reproducibility for co-expression modules across diverse conditions. GSE149614 was used as the discovery network, and GSE189903 was used to validate module preservation by applying the modulePreservation() function with 20 permutations. Zsummary.pres and Zsummary.qual statistics reflect preservation of density and connectivity, respectively, Fig. 2 D. Modules with Z-summary scores > 10 were considered highly preserved. Hep-M2 and Hep-M20 both met this threshold, and confirmed robustness and biological relevance across independent patient cohorts. However, the Hep-M2 module did not exhibit a clear progression from normal to tumor, and was not used to identify stage-associated biomarkers. Identification of stage-associated biomarkers DEG analysis between normal and tumor hepatocytes were intersected with those from modules Hep-M20 to identify candidate biomarkers linked to HCC progression. This integrative analysis yielded 205 overlapping genes within Hep-M20, and represented a refined set of candidates likely involved in hepatocyte transformation, Fig. 3 A. Stage specific DEG analysis was performed to identify the genes involved in tumor progression. Violin plots of the top 18 DEGs demonstrated stage-dependent upregulation with increased expression from normal liver to advanced tumor stages, Fig. 3 B. Among these, Basigin (BSG)/CD147 emerged as a promising candidate, and its location on the cell surface makes it an accessible and practical target for molecular imaging. Target expression was significantly elevated in tumor hepatocytes versus normal liver cells, Fig. 3 C. A progressive increase across tumor stages reached peak levels in advanced disease (stage IIIB-IV), Fig. 3 D. TCGA data was analyzed to validate stage-associated BSG/CD147 expression. A progressive increase was observed with advanced pathology, Fig. 3 E. A significant positive correlation was found between BSG/CD147 expression and tumor stage in TCGA-LIHC primary tumor samples. Trajectory analysis of biomarker expression A single cell pseudotime trajectory was constructed using Monocle 2 to model transcriptional dynamics during HCC progression. The trajectory began with cluster 3 (pro-tumor hepatocytes) at the root, and extended through clusters 4, 5, and 9 to represent progressively advanced tumor states, thereby forming a continuum from early to late tumorigenesis, Fig. 4 A. The arrangement of cells along pseudotime reflect their inferred developmental order with early-stage cells positioned near the root and advanced tumor cells occupying the terminal branches, Fig. 4 B. Mapping the tissue origin of cells onto the trajectory further supports biological relevance. Normal cells were located primarily at the early end of the pseudotime axis, while tumor-derived cells clustered at later stages to support a temporal transition from normal to malignant hepatocytes, Fig. 4 C. Gene trajectory plots demonstrated that BSG/CD147 expression increased progressively along the trajectory, Fig. 4 D. This trend was further supported by smoothed scatter plots that revealed sustained and gradual gene upregulation over pseudotime, Fig. 4 E. CytoTRACE scores projected onto the pseudotime trajectory revealed a gradual increase with lower scores observed at early stages and higher scores toward the terminal branches, Fig. 4 F. A significant positive correlation was observed between BSG/CD147 expression and stemness potential (Spearman’s coefficient r = 0.48, P < 2.2×10 − 16 to suggest that higher expression is associated with increased stemness, Fig. 4 G. Cell-cell communication analysis CellChat analysis was used to explore how different cell types communicate with each other. Key signaling pathways were identified using a database of known ligand–receptor pairs, and the strength of communication among cell groups was measured. Cell–cell communication patterns were visualized using CellChat, with interaction networks quantified by both interaction count, S5 A and interaction weight, S5 B. In the network map, thicker connecting lines represent more frequent or stronger interactions among cell types, Fig. S5 A , B . Fibroblasts showed the highest level of communication activity, interacting with tumor cells (59 interactions, weight = 0.66), normal liver cells (53 interactions, weight = 0.70), and pro-tumor cells (40 interactions, weight = 0.41) to suggest a central role in shaping the tumor environment. A heatmap further illustrates the strength of signaling among different cell types, Fig. S5 C . Overall, fibroblasts sent strong signals to immune cells, normal hepatocytes, and tumor cells to highlight their major role as communication “hubs” within the tissue. Analysis of cell-to-cell communications showed that the BSG/CD147 signaling pathway plays a major role in shaping the tumor microenvironment, Fig. 5 A. Among all cell types, dendritic cells sent out signals with greatest strength (0.71), and were identified as the main “messenger” cells, while fibroblasts received the most signals (1.43) to become key “receivers” within the network, Fig. S6 A , B . Strong communications was also observed from fibroblasts to both T cells and tumor cells, Fig. 5 B. Tumor cells showed the highest overall signaling strength to suggest a dominant tumor-to-T cell communication pathway mediated by BSG/CD147 and cyclophilins, Fig. 5 C. Tumor cells had the highest network “hub” score to highlight their central role, while fibroblasts ranked highest as signal receivers, Fig. 5 D. Overall, the analysis revealed distinct cellular roles as tumor cells function as communication hubs, fibroblasts as primary recipients, and immune and endothelial cells as important regulators within the BSG/CD147-cyclophilin signaling network, Fig. 5 E. The strength of cyclophilin A and B signaling (PPIA/PPIB) was further quantified across normal, pre-tumor, and tumor conditions, Fig. 5 F. Protein-protein interaction analysis BSG/CD147 was found to interact with several key proteins, including PPIA, PPIB, MMP1, and SLC16A1, to support its known role in matrix degradation and metabolic adaptation, Fig. 5 G. Network analysis identified PPIA and PPIB as hub interactors. Expression-weighted PPI scores revealed that tumor cells displayed the highest interaction strength, followed by pro-tumor and normal groups, to reflect a progressive activation of BSG/CD147-associated signaling during tumor development, Fig. 5 H. Ex vivo validation of target expression in HCC BSG/CD147 expression in HCC was further validated by immunofluorescence staining of paired human liver specimens. Tumor regions exhibited markedly stronger signal intensity compared with adjacent cirrhosis or background liver, Fig. 6 A,B. ROC analysis demonstrated excellent diagnostic accuracy, AUC = 0.96 with 86% sensitivity and 97% specificity, to distinguish HCC from background cirrhotic liver, Fig. 6 C. A weak positive correlation was observed between tumor size and protein expression, Fig. 6 D. Signal was further analyzed in HCC tumors < 2 cm in diameter to assess its potential utility for early detection, Fig. 6 E. Despite the small size, these lesions showed clearly elevated fluorescence relative to background liver, Fig. 6 E. ROC analysis for nodules < 2 cm demonstrated high diagnostic accuracy, AUC = 0.95 with 87.5% specificity and 93% sensitivity, Fig. 6 F. Although the mean T/B ratio was higher in larger tumors, the ability of BSG/CD147 to generate detectable contrast in < 2 cm nodules underscores promise to identify early or borderline lesions that are otherwise difficult to classify radiologically (LI-RADS 3/4). Target expression also exhibited inter-patient variability, Fig. 6 F, with the coefficient of variation (CV) of T/B ratios greater in large tumors (77.7%) than in small ones (40.2%) suggesting biological heterogeneity increases with tumor size, Fig. 6 E. Importantly, BSG/CD147 signal remained clearly detectable within small HCC foci embedded in cirrhotic tissue to support potential utility for distinguishing indeterminate liver nodules (< 2 cm). In a representative specimen, strong fluorescence was observed in HCC tumor regions with only weak background signal in adjacent cirrhosis, Fig. S7 A - D . A few isolated tumor cells within cirrhotic areas also demonstrated focal strong CD147 expression. Discussion This study integrated single-cell transcriptomics, network analysis, and ex vivo validation, and identified Basigin (BSG/CD147) as a robust early HCC biomarker and signaling hub. Using two independent scRNA-seq datasets, the transcriptional trajectory was reconstructed from normal hepatocytes through pro-tumor intermediates to malignant states, and a continuum of increasing stemness and oncogenic activation was identified. Multiple gene modules associated with tumor progression were found using hdWGCNA. The Hep-M20 module showed the strongest correlation with pathological stage. BSG/CD147 emerged from within this module as a key hub gene and exhibited progressive upregulation along pseudotime. Strong correlation with stemness potential implicates gene activation as an early molecular hallmark of hepatocarcinogenesis. Mechanistically, BSG/CD147 appears to function not only as a marker but also as a driver of early malignant transformation. Expression correlated with CytoTRACE-derived stemness to suggest that CD147 upregulation accompanies hepatocyte dedifferentiation toward progenitor-like phenotypes known to promote tumor initiation and recurrence [ 23 ]. Previous studies have shown that CD147 regulates metabolic reprogramming, extracellular matrix remodeling, and epithelial-mesenchymal transition through interactions with cyclophilins (PPIA/PPIB) and downstream MAPK/ERK, PI3K/AKT, and HIF-1α signaling cascades [ 24 – 26 ]. Our single-cell and network analyses extended these observations by revealing that CD147-based signaling networks mediate cross-talk between tumor hepatocytes and fibroblasts, the 2 dominant cell types of the tumor microenvironment to form a cyclophilin-dependent tumor-stroma communication axis. Additionally, our findings suggest that BSG/CD147 can engage T cells through PPIA and PPIB ligands to establish an immunomodulatory communication channel that shapes the tumor immune microenvironment. Our findings are consistent with the recent report that elevated CD147 expression in tumor cells is associated with an immunosuppressive tumor-immune microenvironment (TIME) in HCC as evidenced by increased infiltration of regulatory T cells [ 27 ]. This pathway likely fosters a microenvironment conducive to invasion, angiogenesis, and immune modulation, thereby integrating intrinsic tumor signaling with extrinsic stromal remodeling [ 28 ]. Ex vivo validation further demonstrated that BSG/CD147 expression was significantly elevated in HCC compared with adjacent background liver, including in lesions < 2 cm that are often radiologically indeterminate (LI-RADS 3/4). Our findings are consistent with the observations of a significantly elevated expression of CD147 in HCC compared to non-tumor [ 29 ]. ROC analyses confirmed high diagnostic accuracy for small tumors to underscore the potential of BSG/CD147 for early detection. Strong membrane localization, high T/B ratio, and minimal expression in surrounding cirrhosis or background liver establishes a favorable foundation for development of targeted contrast agents. Incorporating BSG/CD147 assessment into HCC surveillance frameworks could substantially enhance performance for early diagnosis and enable biological risk stratification of indeterminate nodules. Unlike prior studies that largely evaluated CD147 expression in bulk tumor tissue or serum as a static diagnostic or prognostic marker [ 30 – 32 ], this work uses single-cell transcriptomics and pseudotime network modelling to demonstrate CD147 upregulation as an early event in hepatocyte transformation that is correlated with increasing stemness and malignant progression. Despite the strengths of this integrated approach, several limitations should be noted. The sample size (n = 16) for small (< 2 cm) lesions was modest, which may limit generalizability. In addition, the scRNA-seq datasets analyzed were annotated by pathological stage rather than derived directly from indeterminate nodules to restrict temporal resolution for early transformation events. Finally, while computational modeling identified plausible ligand-receptor and pathway interactions, further functional validation is needed to confirm the biological significance of BSG/CD147-mediated signaling in the tumor microenvironment. Our results define BSG/CD147 upregulation as an early molecular event in HCC transformation to connect stemness, intercellular signaling, and microenvironmental remodeling in a unified model for early hepatocarcinogenesis. Consistent expression in small, radiologically ambiguous lesions (< 2 cm) highlights the translational potential as a precision biomarker for early HCC detection and image-guided diagnosis. Integration of BSG/CD147-targeted strategies into molecular imaging or surveillance workflows could transform the early HCC management by enabling detection, risk stratification, and intervention at the most curable stages of liver cancer. Abbreviations AFP α—fetoprotein AUC area under the curve BSG Basigin CD147 Cluster of Differentiation 147 CT computed tomography CV coefficient of variation CytoTRACE Cellular Trajectory Reconstruction Analysis using gene Counts and Expression DAPI 4′,6—diamidino—2—phenylindole DEG differential gene expression FFPE formalin—fixed paraffin—embedded GEO Gene Expression Omnibus GSEA gene set enrichment analysis H&E hematoxylin and eosin HBV hepatitis B virus HCC hepatocellular carcinoma HCV hepatitis C virus hdWGCNA high—dimensional weighted gene co—expression network analysis HIF 1α—hypoxia—inducible factor 1 alpha LIHC Liver Hepatocellular Carcinoma (TCGA cohort) LI RADS—Liver Imaging Reporting and Data System MAPK/ERK mitogen—activated protein kinase/extracellular signal—regulated kinase MRI magnetic resonance imaging PBST phosphate—buffered saline with Tween—20 PPI protein—protein interaction PPIA peptidylprolyl isomerase A (cyclophilin A) PPIB peptidylprolyl isomerase B (cyclophilin B) ROC receiver operating characteristic RT room temperature scRNA seq—single—cell RNA sequencing TCGA The Cancer Genome Atlas T/B target—to—background TGF β—transforming growth factor beta tSNE t—distributed stochastic neighbor embedding UMAP uniform manifold approximation and projection US ultrasound WGCNA weighted gene co—expression network analysis Declarations Grant support This study was funded in part by the National Institutes of Health (NIH) R01 CA285303 (TDW). Ethics approval Human tissue procurement and use in this study were conducted in accordance with institutional guidelines and approved by the Institutional Review Board (IRB) of University of Michigan (Approval No: HUM00248521). Consent to participate: Not Applicable Consent for publication: Not Applicable Conflict of interest : The authors report no conflicts of interest. Funding: This work was supported in part by the National Institutes of Health (NIH) R01 CA285303 (TDSW). Author Contribution TDWW performed the data analysis and prepared all figures. SJ conducted the experiments, prepared reagents and samples, contributed to data collection, reviewed the results, drafted and revised the manuscript, and finalized the text. SJ also conceived and supervised the study. EKC conducted histopathologic evaluation and provided expert interpretation of liver tissue specimens. SF collected the ex vivo specimen and performed H&E staining. HJ contributed to statistical analysis and computational data processing. TDSW provided scientific guidance, contributed to study design, and assisted with data interpretation. All authors read and approved the final manuscript. Acknowledgements: The authors thank the University of Michigan Department of Pathology for archived human liver specimens. Data Availability The single-cell RNA sequencing datasets analyzed during this study are publicly available from the Gene Expression Omnibus (GEO) under accession numbers GSE149614 and GSE189903. Processed data, analysis scripts, and figure source files generated during this study are available from the corresponding author upon reasonable request. References Llovet JM, Kelley RK, Villanueva A, et al. Hepatocellular carcinoma. Nat Rev Dis Primers. 2021;7:6. https://doi.org/10.1038/s41572-020-00240-3 . Oh JH, Jun DW. The latest global burden of liver cancer: A past and present threat. Clin Mol Hepatol. 2023;29(2):355–7. 10.3350/cmh.2023.0070 . Epub 2023 Mar 9. PMID: 36891606; PMCID: PMC10121295. Alberts CJ, Clifford GM, Georges D, Negro F, Lesi OA, Hutin YJ, de Martel C. Worldwide prevalence of hepatitis B virus and hepatitis C virus among patients with cirrhosis at country, region, and global levels: a systematic review. Lancet Gastroenterol Hepatol. 2022;7(8):724–35. 10.1016/S2468-1253(22)00050-4 . Epub 2022 May 14. PMID: 35576953; PMCID: PMC9259503. Patel N, Yopp AC, Singal AG. Diagnostic delays are common among patients with hepatocellular carcinoma. J Natl Compr Canc Netw. 2015;13(5):543–9. 10.6004/jnccn.2015.0074 . PMID: 25964640; PMCID: PMC5624109. Parikh ND, Mehta AS, Singal AG, Block T, Marrero JA, Lok AS. Biomarkers for the Early Detection of Hepatocellular Carcinoma. Cancer Epidemiol Biomarkers Prev. 2020;29(12):2495–503. 10.1158/1055-9965.EPI-20-0005 . Epub 2020 Apr 1. PMID: 32238405; PMCID: PMC7529652. van der Pol CB, Lim CS, Sirlin CB, McGrath TA, Salameh JP, Bashir MR, Tang A, Singal AG, Costa AF, Fowler K, McInnes MDF. Accuracy of the Liver Imaging Reporting and Data System in Computed Tomography and Magnetic Resonance Image Analysis of Hepatocellular Carcinoma or Overall Malignancy -. Syst Rev Gastroenterol. 2019;156:976–86. Addissouky TA, Sayed IETE, Ali MMA, et al. Latest advances in hepatocellular carcinoma management and prevention through advanced technologies. Egypt Liver J. 2024;14:2. https://doi.org/10.1186/s43066-023-00306-3 . Yan Q, Sun YS, An R, Liu F, Fang Q, Wang Z, Xu T, Chen L, Du J. Application and progress of the detection technologies in hepatocellular carcinoma. Genes Dis. 2022;10(5):1857–69. PMID: 37492708; PMCID: PMC10363596. Wu G, Bajestani N, Pracha N, Chen C, Makary MS. Hepatocellular Carcinoma Surveillance Strategies: Major Guidelines and Screening Advances. Cancers. 2024;16:3933. Safri F, Nguyen R, Zerehpooshnesfchi S, George J, Qiao L. Heterogeneity of hepatocellular carcinoma: from mechanisms to clinical implications. Cancer Gene Ther. 2024;31(8):1105–12. 10.1038/s41417-024-00764-w . Epub 2024 Mar 18. PMID: 38499648; PMCID: PMC11327108. Nguyen PHD, Ma S, Phua CZJ, Kaya NA, Lai HLH, Lim CJ, Lim JQ, Wasser M, Lai L, Tam WL, Lim TKH, Wan WK, Loh T, Leow WQ, Pang YH, Chan CY, Lee SY, Cheow PC, Toh HC, Ginhoux F, Iyer S, Kow AWC, Young Dan Y, Chung A, Bonney GK, Goh BKP, Albani S, Chow PKH, Zhai W, Chew V. Intratumoural immune heterogeneity as a hallmark of tumour evolution and progression in hepatocellular carcinoma. Nat Commun. 2021;12:227. Huang ZL, Xu B, Li TT, Xu YH, Huang XY, Huang XY. Integrative Analysis Identifies Cell-Type-Specific Genes Within Tumor Microenvironment as Prognostic Indicators in Hepatocellular Carcinoma. Front Oncol. 2022;12:878923. Li X, Wang CY. From bulk, single-cell to spatial RNA sequencing. Int J Oral Sci. 2021;13:36. https://doi.org/10.1038/s41368-021-00146-0 . Satija R, Farrell JA, Gennert D, Schier AF, Regev A. Spatial reconstruction of single-cell gene expression data. Nat Biotechnol. 2015;33(5):495–502. 10.1038/nbt.3192 . Epub 2015 Apr 13. PMID: 25867923; PMCID: PMC4430369. Gulati GS, Sikandar SS, Wesche DJ, Manjunath A, Bharadwaj A, Berger MJ, Ilagan F, Kuo AH, Hsieh RW, Cai S, Zabala M, Scheeren FA, Lobo NA, Qian D, Yu FB, Dirbas FM, Clarke MF, Newman AM. Single-cell transcriptional diversity is a hallmark of developmental potential. Science. 2020;367(6476):405–11. 10.1126/science.aax0249 . PMID: 31974247; PMCID: PMC7694873. Morabito S, Reese F, Rahimzadeh N, Miyoshi E, Swarup V. hdWGCNA identifies co-expression networks in high-dimensional transcriptomics data. Cell Rep Methods. 2023;3(6):100498. 10.1016/j.crmeth.2023.100498 . PMID: 37426759; PMCID: PMC10326379. Trapnell C, Cacchiarelli D, Grimsby J, et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat Biotechnol. 2014;32:381–6. https://doi.org/10.1038/nbt.2859 . Jin S, Plikus MV, Nie Q. CellChat for systematic analysis of cell-cell communication from single-cell transcriptomics. Nat Protoc. 2025;20(1):180–219. 10.1038/s41596-024-01045-4 . Epub 2024 Sep 16. PMID: 39289562. Dang CV. MYC on the path to cancer. Cell. 2012;149(1):22–35. 10.1016/j.cell.2012.03.003 . PMID: 22464321; PMCID: PMC3345192. Jackson S, Bartek J. The DNA-damage response in human biology and disease. Nature. 2009;461:1071–8. https://doi.org/10.1038/nature08467 . Malumbres M, Barbacid M. Cell cycle, CDKs and cancer: a changing paradigm. Nat Rev Cancer. 2009;9:153–66. https://doi.org/10.1038/nrc2602 . Dimova D, Dyson N. The E2F transcriptional network: old acquaintances with new faces. Oncogene. 2005;24:2810–26. https://doi.org/10.1038/sj.onc.1208612 . Muramatsu S, Tanaka S, Mogushi K, Adikrisna R, Aihara A, Ban D, Ochiai T, Irie T, Kudo A, Nakamura N, Nakayama K, Tanaka H, Yamaoka S, Arii S. Visualization of stem cell features in human hepatocellular carcinoma reveals in vivo significance of tumor-host interaction and clinical course. Hepatology. 2013. doi/10.1002/hep.26345 . https://onlinelibrary.wiley.com/ . Li M, Zhai Q, Bharadwaj U, Wang H, Li F, Fisher WE, Chen C, Yao Q. Cyclophilin A is overexpressed in human pancreatic cancer cells and stimulates cell proliferation through CD147. Cancer. 2006;106(10):2284-94. 10.1002/cncr.21862 . PMID: 16604531. Nigro P, Pompilio G, Capogrossi MC. Cyclophilin A: a key player for human disease. Cell Death Dis. 2013;4(10):e888. 10.1038/cddis.2013.410 . PMID: 24176846; PMCID: PMC3920964. Sakamoto M, Miyagaki T, Kamijo H, Oka T, Boki H, Takahashi-Shishido N, Suga H, Sugaya M, Sato S. CD147-Cyclophilin a Interactions Promote Proliferation and Survival of Cutaneous T-Cell Lymphoma. Int J Mol Sci. 2021;22(15):7889. 10.3390/ijms22157889 . PMID: 34360654; PMCID: PMC8346093. Shigematsu Y, Kanda H, Takahashi Y, Takeuchi K, Inamura K. Relationships between tumor CD147 expression, tumor-infiltrating lymphocytes, and oncostatin M in hepatocellular carcinoma. Virchows Arch. 2025;487(1):117–25. 10.1007/s00428-024-03939-w . Epub 2024 Oct 12. PMID: 39395054. Nyalali AM, Leonard AU, Xu Y, Li H, Zhou J, Zhang X, Rugambwa TK, Shi X, Li F. CD147: an integral and potential molecule to abrogate hallmarks of cancer. Front Oncol. 2023;13:1238051. Mamori S, Nagatsuma K, Matsuura T, Ohkawa K, Hano H, Fukunaga M, Matsushima M, Masui Y, Fushiya N, Onoda H, Searashi Y, Takagi I, Tagiri H. Useful detection of CD147 (EMMPRIN) for pathological diagnosis of early hepatocellular carcinoma in needle biopsy samples. World J Gastroenterol. 2007;13(21):2913–7. 10.3748/wjg.v13.i21.2913 . PMID: 17589939; PMCID: PMC4171141. Huang DF, Zhang WJ, Chen J, Jiao ZG, Wang XL, Rao DY, Li WS, Hu D, Xie FF, Wang XX, Li ZZ, Yi XM, Wu JY, Jiang Y, Wang Q, Zhong TY. Hepatocellular carcinoma cell-derived small extracellular vesicle-associated CD147 serves as a diagnostic marker and promotes endothelial cell angiogenesis via the PI3K/Akt pathway. Extracell Vesicles Circ Nucl Acids. 2023;4(4):532–47. 10.20517/evcna.2023.30 . PMID: 40357132; PMCID: PMC12066417. Wu J, Hao ZW, Zhao YX, Yang XM, Tang H, Zhang X, Song F, Sun XX, Wang B, Nan G, Chen ZN, Bian H. Full-length soluble CD147 promotes MMP-2 expression and is a potential serological marker in detection of hepatocellular carcinoma. J Transl Med. 2014;12:190. 10.1186/1479-5876-12-190 . PMID: 24996644; PMCID: PMC4227008. Lee A, Rode A, Nicoll A, Maczurek AE, Lim L, Lim S, Angus P, Kronborg I, Arachchi N, Gorelik A, Liew D, Warner FJ, McCaughan GW, McLennan SV, Shackel NA. Circulating CD147 predicts mortality in advanced hepatocellular carcinoma. J Gastroenterol Hepatol. 2016;31(2):459 – 66. 10.1111/jgh.13148 . PMID: 26312403. Additional Declarations No competing interests reported. Supplementary Files Supplementaryfile.docx Cite Share Download PDF Status: Posted Version 1 posted 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-8593693","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":574250349,"identity":"429268e1-e4fc-4a3f-b218-4a356acbec5c","order_by":0,"name":"Thomas DW Wang","email":"","orcid":"","institution":"University of Michigan–Ann Arbor","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"DW","lastName":"Wang","suffix":""},{"id":574250350,"identity":"b5e165d8-6d82-4655-ba7f-570302dd43a3","order_by":1,"name":"Eun-Young K Choi","email":"","orcid":"","institution":"University of Michigan–Ann Arbor","correspondingAuthor":false,"prefix":"","firstName":"Eun-Young","middleName":"K","lastName":"Choi","suffix":""},{"id":574250351,"identity":"64ccfc0b-eae6-4418-b844-5adaa9f7cd18","order_by":2,"name":"Shuo Feng","email":"","orcid":"","institution":"University of Michigan–Ann Arbor","correspondingAuthor":false,"prefix":"","firstName":"Shuo","middleName":"","lastName":"Feng","suffix":""},{"id":574250352,"identity":"da35abbd-af74-4d63-bbc1-733a052385df","order_by":3,"name":"Hui Jiang","email":"","orcid":"","institution":"University of Michigan–Ann Arbor","correspondingAuthor":false,"prefix":"","firstName":"Hui","middleName":"","lastName":"Jiang","suffix":""},{"id":574250353,"identity":"8e774a67-9127-4a85-a7fc-eb3c7ee3a0d7","order_by":4,"name":"Thomas DS Wang","email":"","orcid":"","institution":"University of Michigan–Ann Arbor","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"DS","lastName":"Wang","suffix":""},{"id":574250354,"identity":"d1ef573b-7e84-4c0b-a086-335c63304840","order_by":5,"name":"Sangeeta Jaiswal","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0klEQVRIiWNgGAWjYBACxgbGBiBlwcAPJA+A2QeI0yLBINlArBYokGAwOAA2gAgtzO2Hmz/z1EjYG9/ITjzwcweDHN+NBAIO60lsk+Y5JpG47UbuhoO9ZxiMJQlqaUhsY85hk0gwA2o5zNjGkLiBoJb+h82fc/4BHTYDoqWesJYZiQ3SuW0SjBskIFoSDAhredgm/bdPInHGmbdAv7RJGM488wC/FsP+9McfZ3yzsedvz9384WebjTzfcQK2GDag8iXwKwcBecJKRsEoGAWjYMQDAJeSTMiQaZKVAAAAAElFTkSuQmCC","orcid":"","institution":"University of Michigan–Ann Arbor","correspondingAuthor":true,"prefix":"","firstName":"Sangeeta","middleName":"","lastName":"Jaiswal","suffix":""}],"badges":[],"createdAt":"2026-01-13 15:23:34","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8593693/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8593693/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":101297113,"identity":"d88aae3a-7b8d-41de-bf59-ff69f7d112ba","added_by":"auto","created_at":"2026-01-28 09:25:22","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":648229,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHepatocyte subpopulations\u003c/strong\u003e. \u003cstrong\u003eA\u003c/strong\u003e) Single cell clusters for normal and tumor specimens are shown. \u0026nbsp;Hepatocyte clusters (red) were selected based on distinct gene expression patterns, and demonstrated differential gene enrichment between normal and tumor. \u003cstrong\u003eB\u003c/strong\u003e) Stacked bar plot illustrates proportional composition: 26 normal (N); 3 and 8 pro-tumor (PT); 4, 5, and 9 tumor (T). \u0026nbsp;Cluster 26 was composed predominantly of cells from normal liver (orange). Clusters 4, 5, and 9 showed marked enrichment in tumor (teal). \u0026nbsp;Clusters 3 and 8 contained cells from both normal and tumor to reflect transitional states. \u003cstrong\u003eC\u003c/strong\u003e) CytoTRACE analysis shows increase in stemness property from normal (cluster 26) to tumor (cluster 4, 5, and 9) cells. \u003cstrong\u003eD\u003c/strong\u003e) Dot plot shows expression profiles for key differentially expressed genes across hepatocyte clusters. \u0026nbsp;Dot intensity indicates average gene expression, while dot size represents percentage of cells expressing each gene in cluster.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8593693/v1/9c48601f303b944e401766b9.jpeg"},{"id":101296813,"identity":"5bf7f0da-441f-44cf-a024-a8ee0cf0497e","added_by":"auto","created_at":"2026-01-28 09:21:04","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":871501,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGene modules associated with HCC progression\u003c/strong\u003e. \u003cstrong\u003eA\u003c/strong\u003e) High-dimensional WGCNA (hdWGCNA) was used to generate a dendrogram that illustrates hierarchical gene clustering in distinct hepatocyte co-expression modules (depicted by unique colors). \u0026nbsp;Genes not assigned are included in the gray module. \u003cstrong\u003eB\u003c/strong\u003e) Dot plot shows gene expression for all modules identified (Hep-M1 to Hep-M22) across hepatocyte clusters. \u0026nbsp;Dot size indicates proportion of cells within each cluster expressing genes from the respective module (% Exp) while color intensity reflects the average module expression level (Avg Exp). \u0026nbsp;Modules, including Hep-M2, Hep-M18, and Hep-M20, exhibit progressively elevated expression from normal (26) through pro-tumorigenic (3 and 8) and tumor-specific (4, 5, and 9). \u003cstrong\u003eC\u003c/strong\u003e) Heatmap depicts correlations between module eigengenes (rows) and clinical features (columns), including tissue origin (\"site\"; normal versus tumor), pathological stage (\"stage\"), and viral infection status (\"virus\"). \u0026nbsp;Positive correlations (red) indicate module enrichment with clinical features, and negative correlations (blue) represent inverse relationships. \u0026nbsp;Module Hep-M20 demonstrates the strongest positive correlation (ρ = 0.51) with stage, and highlights potential associations with tumor progression. \u003cstrong\u003eD\u003c/strong\u003e) Module preservation analysis was performed using Z-summary statistics to demonstrate reproducibility and robustness of gene modules from the exploratory dataset (GSE149614) by comparison with the validation dataset (GSE189903). \u0026nbsp;The Z-summary quality scores reflect internal consistency and connectivity, while the Z-summary preservation scores indicate module preservation. \u0026nbsp;Modules with scores \u0026gt;10 (Hep-M2, Hep-M20) demonstrate strong preservation, and reinforce biological validity and significance across validation cohorts.\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8593693/v1/9929612f9d02aa5f405b1951.jpeg"},{"id":101297520,"identity":"ca9bc130-2ea8-43cb-bd2b-7cc6ade81ad7","added_by":"auto","created_at":"2026-01-28 09:27:37","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":709101,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentification of stage-specific HCC biomarkers\u003c/strong\u003e.\u0026nbsp; \u003cstrong\u003eA\u003c/strong\u003e) Venn diagram illustrates intersection of differentially expressed genes between tumor and normal, and genes contained within module Hep-M20.\u0026nbsp; This intersection identified 205 overlapping genes, and represents a group of candidate genes associated with tumor progression.\u0026nbsp; \u003cstrong\u003eB\u003c/strong\u003e) Violin plots show expression profiles of the top 18 overlapping genes from Hep-M20 across different pathological stages (N: normal; I, II, IIIA, IIIB, IV).\u0026nbsp; Distinct stage-dependent expression patterns emerge to highlight genes progressively upregulated with tumor progression.\u0026nbsp; \u003cstrong\u003eC\u003c/strong\u003e) Violin plots compare expression of Basigin (BSG)/CD147 in normal, pro-tumor and\u0026nbsp; tumor hepatocytes.\u0026nbsp; \u003cstrong\u003eD\u003c/strong\u003e) Stage-specific violin plots illustrate progressive increase of this biomarker across tumor stages from normal (N) through early-stage tumors (I-IIIA) to advanced stages (IIIB-IV) to underscore potential for use as an early diagnostic gene for HCC progression.\u0026nbsp; \u003cstrong\u003eE\u003c/strong\u003e) TCGA data analysis confirms positive correlation between BSG/CD147 and stage with Spearman’s coefficient ρ = 0.196, P = 2.8×10\u003csup\u003e-4\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8593693/v1/1772c89e6b40a1a0b908b0c3.jpeg"},{"id":102298529,"identity":"e695b43f-ff8f-462f-971d-9351608e5e9c","added_by":"auto","created_at":"2026-02-10 10:43:13","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":824531,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePseudotime trajectory analysis\u003c/strong\u003e.\u0026nbsp; \u003cstrong\u003eA\u003c/strong\u003e) Single-cell pseudotime trajectories display normal, pro-tumor, and tumor hepatocyte clusters (3, 4, 5, and 9) that capture a continuum of states from normal to early tumor and late stage tumor.\u0026nbsp; \u003cstrong\u003eB\u003c/strong\u003e) Pseudotime progression illustrates the inferred temporal hepatocyte ordering along the differentiation trajectory. \u0026nbsp;Cells at the root (dark blue) represent early or transitional states, and cells at the terminal branches (light blue) correspond to more advanced tumor states.\u0026nbsp; \u003cstrong\u003eC\u003c/strong\u003e) Mapping of tissue origin (normal and tumor) onto the pseudotime trajectory highlights a clear transition from predominantly normal-derived hepatocytes at early pseudotime points to tumor-derived hepatocytes in later stages.\u0026nbsp; \u003cstrong\u003eD\u003c/strong\u003e) Expression profiles of BSG/CD147 is projected on the pseudotime trajectory. \u0026nbsp;Color intensity indicates relative gene expression levels, and reveals a marked increase in expression as cells progress toward advanced tumor stages.\u0026nbsp; \u003cstrong\u003eE\u003c/strong\u003e) Scatter plots demonstrate relative gene expression of BSG/CD147 over pseudotime to further support gene upregulation. \u0026nbsp;Smoothed trend lines show a progressive increase in gene expression along pseudotime to support it role as a potential biomarker for HCC progression.\u0026nbsp; \u003cstrong\u003eF\u003c/strong\u003e) Projection of CytoTRACE score shows cells with low stemness potential situated at the trajectory beginning while cells with high CytoTRACE score were located at the trajectory end. \u0026nbsp;\u003cstrong\u003eG\u003c/strong\u003e) Scatterplot showing correlation between CytoTRACE score and BSG expression with Spearman’s coefficient r = 0.48, P \u0026lt;2.2×10\u003csup\u003e-16\u003c/sup\u003e.\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8593693/v1/d0c04c135041c65a6e0d6132.jpeg"},{"id":101214628,"identity":"c7f764fd-1908-4048-a11b-ef27e5c6082e","added_by":"auto","created_at":"2026-01-27 10:36:12","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":965749,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBSG/CD147 mediated cell-cell communication in tumor microenvironment\u003c/strong\u003e. \u003cstrong\u003eA\u003c/strong\u003e) NetVisual plot of BSG/CD147 signaling illustrates communication networks among cell types. \u0026nbsp;Node size corresponds to cell group abundance. Edge thickness represents relative signaling probability. \u003cstrong\u003eB\u003c/strong\u003e) Circle plot depicts the strength of fibroblast-derived signals to target cells that highlight significant communications with T and tumor cells. \u003cstrong\u003eC\u003c/strong\u003e) Circle plot shows the strongest outgoing signaling from tumor cells toward T cells and indicates a dominant tumor-to-immune cell communication axis via BSG. \u003cstrong\u003eD\u003c/strong\u003e) Authority versus hub scores reveal distinct cellular roles in BSG/CD147 signaling. \u0026nbsp;Each point represents a cell population positioned according to its hub score (outgoing signaling strength) and authority score (incoming signaling strength). \u003cstrong\u003eE\u003c/strong\u003e) Signaling role heatmap depicts cellular functions within the BSG/CD147 signaling network and displays the relative sending (hub) and receiving (authority) roles of each cell population. \u0026nbsp;Tumor cells show strong hub activity to indicate a central role in initiating signaling, while fibroblasts exhibit dominant receiver activity. \u0026nbsp;\u003cstrong\u003eF\u003c/strong\u003e) Bar plot compares interaction strengths across normal, pro-tumor, and tumor conditions to demonstrate that tumor cells exhibit the highest signaling probability, followed by pro-tumor and normal cells. \u003cstrong\u003eG\u003c/strong\u003e) Protein-protein interaction network was derived from STRING (confidence ≥0.7) and visualized in Cytoscape. \u0026nbsp;Nodes represent proteins, and edges display predicted or validated interactions. \u003cstrong\u003eH\u003c/strong\u003e) Heatmap showing the expression-weighted protein-protein interaction (PPI) scores for BSG/CD147 across normal, pro-tumor, and tumor cells. \u0026nbsp;Tumor cells exhibit the highest PPI activity, followed by pro-tumor and normal cells, to indicate enhanced BSG/CD147-mediated signaling during tumor progression.\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8593693/v1/986007daed825177de91b769.jpeg"},{"id":101296774,"identity":"88b3f589-c322-49b8-b10a-061f1f0d4399","added_by":"auto","created_at":"2026-01-28 09:20:30","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1008830,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEx vivo validation of BSG/CD147 expression in HCC\u003c/strong\u003e.\u0026nbsp; \u003cstrong\u003eA\u003c/strong\u003e) Representative immunofluorescence (BSG/CD147-Cy5.5, red) and corresponding histology (H\u0026amp;E) of paired human liver tissues show markedly higher BSG/CD147 signal in HCC compared with adjacent cirrhotic and normal regions.\u0026nbsp; \u003cstrong\u003eB\u003c/strong\u003e) Quantified fluorescence intensities demonstrate significantly elevated T/B ratios in tumor relative to matched background (paired t-test, P = 2.9×10⁻\u003csup\u003e11\u003c/sup\u003e).\u0026nbsp; \u003cstrong\u003eC\u003c/strong\u003e) ROC analysis differentiated HCC from non-tumor with high diagnostic performance (AUC = 0.96, 86% sensitivity, and 97% specificity).\u0026nbsp; \u003cstrong\u003eD\u003c/strong\u003e) Spearman correlation analysis reveals moderate positive association between tumor size and T/B ratio (ρ = 0.32, P = 0.04) to support increased BSG/CD147 expression with tumor progression.\u0026nbsp; \u003cstrong\u003eE\u003c/strong\u003e) T/B ratios in small (\u0026lt;2 cm) and large (\u0026gt;2 cm) HCC tumors shows significantly higher mean signal in larger lesions of 8.1±1.4 versus 3.2±0.32, P = 0.027.\u0026nbsp; Violin plot shows greater signal variability among large tumors (CV = 77.7%) than smaller ones (CV = 40.2%). \u003cstrong\u003eF\u003c/strong\u003e) ROC curve analysis of small lesions demonstrates high diagnostic performance for distinguishing small HCC from background cirrhosis (AUC = 0.95; sensitivity 87%; specificity 93%). \u003cstrong\u003eG\u003c/strong\u003e) Heatmap summarizes inter-patient heterogeneity, and illustrates consistent elevation of BSG signal in HCC (\u0026lt;2 cm) versus background cirrhotic liver across n =16 patients.\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8593693/v1/a1766e5adc29092576d14b07.jpeg"},{"id":102397506,"identity":"08ae701c-952d-4f8f-837d-88f80f9cfde1","added_by":"auto","created_at":"2026-02-11 10:17:33","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6267783,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8593693/v1/9eafe74a-bfe1-4240-af8a-96d7e10ae853.pdf"},{"id":101297105,"identity":"3189314d-1e11-4794-ae25-bc3304b1d3f8","added_by":"auto","created_at":"2026-01-28 09:25:19","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":2220401,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryfile.docx","url":"https://assets-eu.researchsquare.com/files/rs-8593693/v1/cfe4b9bba970ee13bb9d039b.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Multimodal single-cell network analysis uncovers BSG/CD147 as an early biomarker and signaling hub in hepatocellular carcinoma","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHepatocellular carcinoma (HCC) is a common primary liver malignancy, and a leading cause of cancer-related deaths worldwide [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The incidence continues to rise, driven by chronic hepatitis B and C infections, metabolic liver disease, and cirrhosis [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Despite advances in imaging and therapy, most patients are diagnosed at an advanced stage when curative interventions, such as resection or transplantation This pathway is likely to foster a microenvironment conducive to invasion, angiogenesis, and immune modulation are no longer useful [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Early-stage HCC is frequently asymptomatic, and current biomarkers lack the sensitivity and specificity needed for reliable detection [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. This limitation underscores the need for biologically based biomarkers that can identify early or indeterminate (LI-RADS 3/4) lesions on imaging to improve clinical decision-making [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMany imaging techniques, such as ultrasound (US), computed tomography (CT), and magnetic resonance imaging (MRI), are used in clinical practice [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] and serum biomarkers, such as α-fetoprotein (AFP) and des-γ-carboxy prothrombin (DCP), are commonly used to assist in detection and monitoring. However, diagnosing HCC at an early stage remains a significant challenge. These techniques often lack sensitivity, especially with small lesions or in the context of liver cirrhosis [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Furthermore, false positive diagnoses may lead to unnecessary invasive procedures or undue patient anxiety. Therefore, molecular signatures, such as serum or tumor biomarkers that deliver higher sensitivity and specificity for the early and accurate detection of HCC as diagnostic tools, are greatly needed.\u003c/p\u003e \u003cp\u003eTumor heterogeneity is a well-known feature of HCC, and results in high diversity at the cellular, molecular, functional, and lineage levels [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Cellular and molecular variability poses a significant barrier to early diagnosis [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. This biological complexity gives rise to distinct molecular profiles even at early disease stages and in small nodules. Thus, conventional imaging phenotypes or bulk biopsy assessments may fail to capture the full molecular lesion complexity. In these contexts, molecular signature, such as gene‑expression panels, pathway‑activation markers, or cell‑type‑specific transcripts, may provide the sensitivity needed to detect small or early HCC lesions by capturing the molecular \u0026ldquo;footprint\u0026rdquo; of malignant transformation that precedes overt radiological changes [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTraditional bulk transcriptomic methods suffer from aggregation of signals across all cells within the sample, thereby overlooking both rare populations and interactions among distinct cell types [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. In contrast, single‑cell RNA sequencing (scRNA‑seq) has revolutionized this landscape by enabling high‑resolution transcriptomic analysis at the level of individual cells. This approach is capable of revealing rare hepatocyte and progenitor‑like cells, reconstructing lineage trajectories, and delineating signaling networks that underlies cancer transformation. Although scRNA‑seq has been used to profile immune and stromal HCC components, the full potential to identify clinically actionable early biomarkers remains unrealized. High-resolution data generated by scRNAseq allows for the discovery of candidate early biomarkers, such as specific transcripts expressed only in emerging malignant subpopulations or in the supportive microenvironmental niche, which remain undetected in bulk transcriptomic analyses. The aim of the study was to apply scRNA-seq analysis to identify clinically relevant molecular signatures that capture the process for early malignant HCC transition and support the diagnosis of small or indeterminate lesions.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData acquisition from scRNA-seq datasets\u003c/h2\u003e \u003cp\u003eSingle-cell transcriptomic datasets were obtained from the Gene Expression Omnibus (GEO) repository (\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) to investigate gene expression dynamics during HCC transformation. Two independent datasets were evaluated. The discovery dataset (GSE149614) consisted of liver specimens from n\u0026thinsp;=\u0026thinsp;10 HCC patients distributed in 4 tissue types, including non-tumor liver, primary tumor, portal vein tumor thrombus, and metastatic lymph nodes. Only cells from non-tumor liver and primary tumor tissues were retained to focus on early cancer transformation. This cohort captured a wide range of tumor-node-metastasis (TNM) stages and etiologies, including hepatitis B virus (HBV), hepatitis C virus (HCV), and metabolic liver disease. A validation dataset (GSE189903) was used for independent module preservation analysis and included n\u0026thinsp;=\u0026thinsp;4 HCC patients representing diverse tumor sizes. Single cells were prepared from different tumor regions and adjacent normal tissue. This approach enabled assessment of the robustness and reproducibility of gene co-expression network modules derived from the primary dataset across independent patient cohorts and sequencing platforms.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSingle-cell RNA-seq data processing and quality control\u003c/h3\u003e\n\u003cp\u003eAll analyses were conducted using R software, v4.4.0 with the Seurat v5 package [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Raw gene expression count matrices were imported and converted into Seurat objects. Quality control procedures were applied to exclude low-quality cells using standard filtering criteria, including the removal of cells with high mitochondrial gene content that indicate apoptotic cells, abnormally low or high gene counts that suggest empty droplets or doublets, and low total RNA expression. GSE149614 was preprocessed, and only batch correction was performed. For GSE189903, cells with nCount_RNA\u0026thinsp;\u0026lt;\u0026thinsp;800, nFeature_RNA\u0026thinsp;\u0026lt;\u0026thinsp;500, or mitochondrial gene percentage\u0026thinsp;\u0026gt;\u0026thinsp;25% were excluded. These quality control steps ensured that only high-quality, biologically meaningful single-cell profiles were retained. Batch correction was performed to correct technical variations across patients and sequencing runs using the Seurat integration pipeline. Shared biological features were aligned across diverse datasets to minimize the influence of technical artifacts and ensure integrity for downstream analyses, and included clustering, differential gene expression, and pseudotime trajectory modeling.\u003c/p\u003e\n\u003ch3\u003eCell clustering, annotation, and characterization\u003c/h3\u003e\n\u003cp\u003eAfter initial quality filtering, the FindVariableFeatures function was used to identify highly variable genes with a focus on those that contribute most to cellular heterogeneity. The data were then normalized and centered using ScaleData, and dimensionality reduction was performed using RunPCA. The appropriate number of principal components for downstream analysis was determined with the ElbowPlot function. A shared nearest-neighbor graph was constructed using FindNeighbors. Unsupervised clustering was performed with FindClusters to identify transcriptionally distinct cell populations. For visualization in 2D space, RunTSNE and RunUMAP were applied. Cluster-specific gene expression was evaluated using FindAllMarkers for global differential expression gene (DEG) analysis across all clusters, and FindMarkers for pairwise comparisons between selected clusters. Clusters were annotated based on expression of known liver cell-type marker genes. Only clusters identified as hepatocytes were retained to focus on tumor-associated changes. Seurat visualization functions, including dot plots, violin plots, and bar plots, were used to display gene expression patterns, assess cellular composition, and compare distributions across normal and tumor samples.\u003c/p\u003e\n\u003ch3\u003eAnalysis of stemness using CytoTRACE\u003c/h3\u003e\n\u003cp\u003eCytoTRACE (Cellular Trajectory Reconstruction Analysis using gene Counts and Expression) was used to infer stemness and differentiation potential across normal, pro-tumor, and tumor clusters [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. The pre-processed Seurat object from dataset GSE149614 was used for the CytoTRACE analysis. The filtered and normalized scRNA-seq data was converted into a gene expression matrix by extracting raw counts using the as.matrix() function (arkov_object[[\"RNA\"]]@counts). This matrix served as input to the CytoTRACE() function from the CytoTRACE R package (v0.3.3) using default parameters unless otherwise specified. CytoTRACE scores, which reflect transcriptional diversity and infer cellular plasticity, were incorporated into the Seurat object metadata. These scores were visualized on UMAP embeddings to assess differentiation gradients across clusters. Higher CytoTRACE scores corresponded to less differentiated, more stem-like states and were used to inform downstream trajectory analyses.\u003c/p\u003e\n\u003ch3\u003ePathway enrichment analysis\u003c/h3\u003e\n\u003cp\u003eSingle-cell pathway enrichment analysis was performed using the escape R package (v2.2.3) on a Seurat v5 object containing hepatocytes. Enrichment scores were extracted from the \"escape.ssGSEA\" assay and converted into a matrix with cells in rows and pathways in columns. Average enrichment scores per pathway were computed across normal, pro-tumor, and tumor groups using dplyr. Averaged data was visualized using pheatmap by applying row-wise scaling and hierarchical clustering to highlight differential pathway activity patterns among groups.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eHigh-dimensional weighted gene co-expression network analysis (hdWGCNA)\u003c/h2\u003e \u003cp\u003ehdWGCNA was performed to identify transcriptional programs associated with HCC transition. Hepatocyte subsets were first extracted from the integrated Seurat object. Using the hdWGCNA R package, a gene expression correlation matrix was computed, and a soft-thresholding power of 10 was selected to generate a scale-free weighted adjacency matrix [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. This parameter was chosen based on the standard approach in WGCNA, where the power was selected to approximate scale-free topology, a common property of biological networks. Specifically, the scale-free topology fit index and mean connectivity across a range of powers were examined. Power\u0026thinsp;=\u0026thinsp;10 was the lowest value at which networks began to exhibit scale-free behavior (R\u0026sup2; \u0026gt;0.85). The soft-thresholding power controls how strongly co-expression similarities (correlation values) are weighted in the adjacency matrix. Higher powers emphasize stronger correlations and suppressed weaker ones. This improved the reliability of module detection, leading to more biologically coherent gene modules. A topological overlap matrix (TOM) was then constructed to improve network robustness by emphasizing high-confidence gene-gene interactions. Gene modules were identified using average linkage hierarchical clustering and was followed by dynamic tree cutting to define module boundaries. Genes that did not cluster into any module were assigned to the gray module and indicated low or nonspecific connectivity. Module eigengenes, representative expression profiles of each module, were correlated with key clinical features, including tissue origin (normal versus tumor), tumor stage, and viral infection status (HBV/HCV) to assess potential biological relevance. Hub genes were defined as genes with the highest intramodular connectivity using the GetHubGenes function and were used to identify key regulators within each module.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePreservation of key gene modules\u003c/h3\u003e\n\u003cp\u003eModule preservation analysis was conducted using GSE189903 to assess the reproducibility of co-expression modules identified in the primary dataset (GSE149614). The GetModulePreservation function from the hdWGCNA package was employed to compute Z-summary scores, which integrate multiple metrics of module preservation and quality, including intramodular connectivity and network density. The PlotModulePreservation function was used to visualize module preservation, and display module size against Z-summary scores. Modules with Z-summary values\u0026thinsp;\u0026gt;\u0026thinsp;10 were considered highly preserved and indicated strong reproducibility across datasets.\u003c/p\u003e\n\u003ch3\u003eIdentification of stage-associated biomarkers\u003c/h3\u003e\n\u003cp\u003eA multi-step integrative analysis was performed using gene co-expression modules and DEG data to identify robust biomarkers associated with HCC progression. First, hepatocyte-specific gene co-expression modules were generated using hdWGCNA. Modules were prioritized based on a strong correlation with clinical variables, such as tumor stage and tissue origin. A set of DEGs was generated by comparing gene expressions between tumor-derived and normal hepatocytes to refine candidate genes with functional relevance. The DEGs were then intersected with the gene members of prioritized modules to identify overlapping genes that were both differentially expressed and co-expressed within tumor stage-associated modules. To identify genes associated with tumor stage, DEG analysis was performed using the Seurat package with tumor stage set as the active identity class (Idents). DEG analysis was performed using the FindAllMarkers() function to restrict the analysis to a predefined list of intersecting genes. This approach identified stage-specific marker genes by comparing expression profiles across tumor stages. Expression dynamics were visualized using violin plots to allow for identification of genes that demonstrated progressive upregulation with increasing tumor stage.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eTCGA Analysis of Stage-Associated Gene Expression\u003c/h2\u003e \u003cp\u003eTranscriptomic and clinical data for HCC were obtained from The Cancer Genome Atlas (TCGA) using the TCGAbiolinks R package. Raw gene expression data (HTSeq counts, STAR workflow) were downloaded and processed using the GDCquery, GDCdownload, and GDCprepare functions. Gene-level expression was quantified, and BSG/CD147 expression was extracted using the corresponding Ensembl gene ID. Only primary tumor samples were retained to focus on tumor-specific expression. Samples lacking valid stage annotations or gene expression were excluded from further analysis. A Spearman\u0026rsquo;s rank correlation was performed to assess the relationship between BSG/CD147 expression and tumor progression. Additionally, boxplots, violin plots, and smoothed regression lines were generated to visualize stage-wise expression differences. Outliers were identified and excluded using the IQR method for sensitivity analyses. Due to the limited number of Stage IV samples, visualizations were focused on Stages I\u0026ndash;III to ensure balanced group comparisons.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eTrajectory analysis of biomarker expression\u003c/h2\u003e \u003cp\u003ePseudotime trajectory analysis was performed using the Monocle 2 R package to model the dynamic progression of hepatocyte transformation during HCC development [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Hepatocyte clusters previously identified using Seurat were used to construct the trajectory. These clusters were selected to capture the continuum from early to advanced tumorigenic states. Monocle 2 was applied to order cells along a pseudotime axis and infer a developmental trajectory from transcriptomic changes across individual hepatocytes. Gene expression was projected onto the trajectory to investigate the transcriptional activation patterns of candidate biomarkers. Pseudotime expression analysis was then performed to assess the onset and progression of gene activation along the transformation continuum from normal to malignancy.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eCell-cell communication analysis\u003c/h2\u003e \u003cp\u003eTo infer and analyze intercellular communication networks, CellChat R package, v1.6.11 was used [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The normalized expression matrix and metadata (cell type annotations) were used to create a CellChat object via createCellChat() function. The object was then subset to include only signaling-relevant genes using subsetData() function. Analysis used the built-in CellChatDB.human ligand-receptor interaction database. In addition, BSG/CD147-PPIA and BSG/CD147-PPIB interactions were manually incorporated to investigate their role in cell-cell signaling. These interactions were curated based on known protein-protein interaction evidence, and added to the database, including the ligand-target pair: BSG/CD147 (target); PPIA, PPIB (ligands), and signaling pathway name: BSG_PPIA_PPIB. Interactions were appended to the CellChatDB\u003cspan\u003e$\u003c/span\u003einteraction and CellChatDB\u003cspan\u003e$\u003c/span\u003ecomplex slots prior to data subsetting. The modified database was then used to subset the expression data with subsetData() and proceed with communication probability inference.\u003c/p\u003e \u003cp\u003eCommunication probabilities were computed using computeCommunProb(), followed by computeCommunProbPathway() and aggregateNet(), to infer and summarize pathway-specific signaling networks. Only interactions involving cell types with sufficient representation (minimum 10 cells) were retained. Visualizations such as netVisual_circle(), netVisual_bubble(), and netAnalysis_signalingRole_network() were employed to explore outgoing and incoming signaling patterns. The contribution of each ligand-receptor pair to the pathway was quantified using netAnalysis_contribution().\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eProtein-protein interaction analysis\u003c/h2\u003e \u003cp\u003eThe BSG/CD147 protein-protein interaction (PPI) network was obtained from the STRING database (\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) using a confidence score cutoff of 0.7. The network was visualized and analyzed in Cytoscape (v3.10.3). To assess context-specific interaction dynamics, expression-weighted PPI scores were calculated for normal, protumor, and tumor groups based on the average expression levels of interacting proteins and visualized using heatmap.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eEx vivo validation of target expression in HCC\u003c/h2\u003e \u003cp\u003e Tissue procurement and use in this study were conducted in accordance with institutional guidelines and approved by the Institutional Review Board (IRB) of University of Michigan (Approval No: HUM00248521). Formalin-fixed, paraffin-embedded (FFPE) human liver sections were obtained from the archived tissue bank in the University of Michigan Department of Pathology. Sections (5 \u0026micro;m thick) were cut, mounted on Superfrost Plus glass slides (Fisher Scientific), and deparaffinized. Antigen retrieval was performed in sodium citrate buffer prior to staining. Slides were blocked with 5% goat serum for 1 hour at room temperature (RT), followed by overnight incubation at 4\u0026deg;C with monoclonal anti-BSG/CD147 antibody (MA529060, Invitrogen) at a 1:500 dilution. After 3X washes with phosphate-buffered saline containing Tween-20 (PBST, 3 min each), sections were incubated with a Cy5.5-conjugated secondary antibody for 1 hour at RT. Slides were then washed 3X with PBST and mounted with 1.5 \u0026micro;m coverslips using ProLong Gold Antifade Reagent with DAPI (8961; Cell Signaling Technologies). Fluorescence images were acquired using a confocal microscope with 20X objective under identical exposure settings for tumor and background liver. Mean fluorescence intensities were quantified by placing 3X 20\u0026times;20 \u0026micro;m\u0026sup2; boxes entirely within liver tissues using custom MATLAB software (MathWorks, Inc), while avoiding regions of saturated signal. Adjacent sections were processed for routine pathology (H\u0026amp;E) and independently evaluated by an expert liver pathologist (EKC). On each slide, the pathologist delineated the tumor region and adjacent background liver, which were then used to quantify the fluorescence signal.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eUnless otherwise specified, default statistical tests were used for all R functions. DEG analysis was performed using the Wilcoxon Rank Sum test to assess statistical significance. Image quantification data were analyzed using GraphPad Prism, v10.4.1. Paired t-tests were applied where appropriate to assess differences between matched groups. Receiver operating characteristic (ROC) curve analysis was used to evaluate sensitivity and specificity. Linear regression was used to assess the relationship between tumor size and target-to-background (T/B) ratio, calculated as the fluorescence intensity of the tumor (T) divided by that of the surrounding background (B) liver. Spearman analysis was performed to determine correlation coefficients.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eData acquisition from scRNA-seq datasets\u003c/h2\u003e \u003cp\u003eSingle-cell transcriptomic analysis was performed to identify candidate biomarkers for early HCC detection. The discovery dataset (GSE149614) consisted of liver specimens from n\u0026thinsp;=\u0026thinsp;10 HCC patients. Only cells from non-tumor liver and primary tumor were retained to focus on early cancer transformation. A validation cohort (GSE189903) was used for module preservation analysis. This dataset included n\u0026thinsp;=\u0026thinsp;4 HCC patients with diverse tumor stages, sizes and locations, and was used to evaluate robustness and reproducibility of gene co-expression patterns derived from the primary dataset across independent patient cohorts and technical platforms.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eSingle-cell RNA-seq data processing and quality control\u003c/h2\u003e \u003cp\u003eInitial analyses revealed substantial batch effects in both datasets with cells clustering primarily by patient origin and individual sample identifiers as shown on tSNE plots, Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003cb\u003eA\u003c/b\u003e,\u003cb\u003eB\u003c/b\u003e. Batch-driven clusters were eliminated using the Seurat integration pipeline. Post-integration, cells from different patients and samples exhibited extensive intermixing, and reflected successful alignment of biologically similar cell populations, Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003cb\u003eC\u003c/b\u003e,\u003cb\u003eD\u003c/b\u003e. This correction enabled reliable interpretation of cellular heterogeneity by ensuring that clustering and downstream analyses were driven by biological variations rather than technical differences. After stringent quality control and integration, the final datasets comprised 34,414 tumor-derived and 28,687 normal liver cells (GSE149614), and 43,656 tumor-derived and 30,614 normal liver cells (GSE189903).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eCell clustering, annotation, and characterization\u003c/h2\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003eIdentification and Annotation of Liver Cell Types\u003c/h2\u003e \u003cp\u003eUnsupervised clustering identified 32 distinct cell clusters that represented a range of liver cell types. Annotations based on established marker genes revealed a diverse cellular composition, including hepatocytes, Kupffer cells, macrophages, dendritic cells, T cells, NK cells, B cells, fibroblasts, and endothelial cells, and reflects liver microenvironment heterogeneity, Fig. S2\u003cb\u003eA\u003c/b\u003e. Stratification by tissue origin showed differential distribution of these populations between normal and tumor with several clusters enriched in either tissue type, Fig. S2\u003cb\u003eB\u003c/b\u003e. Specific cell populations were preferentially represented in either normal or tumor, Fig. S2\u003cb\u003eC\u003c/b\u003e. DEG analysis revealed clear transcriptional differences between normal and tumor. Tumor hepatocytes showed upregulation of genes associated with malignancy, while normal hepatocytes retained expression of genes linked to physiological liver function, Fig. S2\u003cb\u003eD\u003c/b\u003e. Clusters 3, 4, 5, 8, 9, and 26 were identified as hepatocytes based on marker gene expression.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eHepatocyte Subpopulation Analysis\u003c/h2\u003e \u003cp\u003etSNE analysis was used to explore hepatocyte heterogeneity between normal and tumor. Hepatocyte clusters (red) were identified based on distinct marker profiles, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA. Cluster composition analysis revealed differential enrichment levels. Cluster 26 was composed predominantly of normal hepatocytes (N), while clusters 4, 5, and 9 were enriched in tumor (T). Clusters 3 and 8 contained cells from both normal and tumor, designated as pro-tumor (PT), and represented transitional cellular states, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB. CytoTRACE was used to evaluate stemness properties and was found to increase in tumor versus normal and pro-tumor clusters, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC. Dot plot analysis highlighted elevated expression of tumor-specific markers in clusters 4, 5 and 9, and expression of normal genes in cluster 26, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eD. Clusters 3 and 8 exhibited a hybrid expression pattern by co-expressing both tumor and normal markers to support their role as intermediate states during tumor progression. Pathway enrichment analysis demonstrated significant upregulation of key cancer-related pathways within tumor clusters. Pathways such as MYC, which regulates cellular proliferation and metabolism [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]; DNA repair, which maintains genomic integrity [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]; the G2M checkpoint, which ensures proper cell cycle progression and prevents propagation of damaged DNA [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]; and E2F targets, which drive DNA synthesis and cell cycle regulation [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], were all elevated, Fig. S3.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eHigh-dimensional weighted gene co-expression network analysis (hdWGCNA)\u003c/h2\u003e \u003cp\u003eA soft-threshold value of 10 was chosen to build a gene network to best reflect natural biological relationships while keeping the connections simple, Fig. S4\u003cb\u003eA\u003c/b\u003e-\u003cb\u003eD\u003c/b\u003e. Using this network approach, 22 groups of genes that show similar activity patterns, called modules, and were identified among hepatocytes, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA. Genes that did not fit into any module were placed in a \u0026ldquo;gray\u0026rdquo; group and showed weak or inconsistent relationships with others. A dot plot shows how these gene modules change across hepatocyte clusters as they progress from normal to tumor cells, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB. Hep-M18 and Hep-M20 expression increased steadily from normal hepatocytes (cluster 26) to early tumor-like (clusters 3 and 8) and finally tumor-rich clusters (4, 5, 9). When comparing modules with clinical features, several meaningful patterns emerged. One group, Hep-M20, showed the strongest link to tumor stage (ρ\u0026thinsp;=\u0026thinsp;0.51) to suggest a role in cancer progression. Hep-M12 was most associated with whether the tissue came from a normal or tumor, and Hep-M6 correlated most with viral infection status (HBV or HCV), Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003ePreservation of key gene modules\u003c/h2\u003e \u003cp\u003eModule preservation analysis was performed using hdWGCNA to assess robustness and reproducibility for co-expression modules across diverse conditions. GSE149614 was used as the discovery network, and GSE189903 was used to validate module preservation by applying the modulePreservation() function with 20 permutations. Zsummary.pres and Zsummary.qual statistics reflect preservation of density and connectivity, respectively, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD. Modules with Z-summary scores\u0026thinsp;\u0026gt;\u0026thinsp;10 were considered highly preserved. Hep-M2 and Hep-M20 both met this threshold, and confirmed robustness and biological relevance across independent patient cohorts. However, the Hep-M2 module did not exhibit a clear progression from normal to tumor, and was not used to identify stage-associated biomarkers.\u003c/p\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003eIdentification of stage-associated biomarkers\u003c/h2\u003e \u003cp\u003eDEG analysis between normal and tumor hepatocytes were intersected with those from modules Hep-M20 to identify candidate biomarkers linked to HCC progression. This integrative analysis yielded 205 overlapping genes within Hep-M20, and represented a refined set of candidates likely involved in hepatocyte transformation, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA. Stage specific DEG analysis was performed to identify the genes involved in tumor progression. Violin plots of the top 18 DEGs demonstrated stage-dependent upregulation with increased expression from normal liver to advanced tumor stages, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB. Among these, Basigin (BSG)/CD147 emerged as a promising candidate, and its location on the cell surface makes it an accessible and practical target for molecular imaging. Target expression was significantly elevated in tumor hepatocytes versus normal liver cells, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC. A progressive increase across tumor stages reached peak levels in advanced disease (stage IIIB-IV), Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD. TCGA data was analyzed to validate stage-associated BSG/CD147 expression. A progressive increase was observed with advanced pathology, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE. A significant positive correlation was found between BSG/CD147 expression and tumor stage in TCGA-LIHC primary tumor samples.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section3\"\u003e \u003ch2\u003eTrajectory analysis of biomarker expression\u003c/h2\u003e \u003cp\u003eA single cell pseudotime trajectory was constructed using Monocle 2 to model transcriptional dynamics during HCC progression. The trajectory began with cluster 3 (pro-tumor hepatocytes) at the root, and extended through clusters 4, 5, and 9 to represent progressively advanced tumor states, thereby forming a continuum from early to late tumorigenesis, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA. The arrangement of cells along pseudotime reflect their inferred developmental order with early-stage cells positioned near the root and advanced tumor cells occupying the terminal branches, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB. Mapping the tissue origin of cells onto the trajectory further supports biological relevance. Normal cells were located primarily at the early end of the pseudotime axis, while tumor-derived cells clustered at later stages to support a temporal transition from normal to malignant hepatocytes, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC. Gene trajectory plots demonstrated that BSG/CD147 expression increased progressively along the trajectory, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD. This trend was further supported by smoothed scatter plots that revealed sustained and gradual gene upregulation over pseudotime, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE. CytoTRACE scores projected onto the pseudotime trajectory revealed a gradual increase with lower scores observed at early stages and higher scores toward the terminal branches, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eF. A significant positive correlation was observed between BSG/CD147 expression and stemness potential (Spearman\u0026rsquo;s coefficient r\u0026thinsp;=\u0026thinsp;0.48, P\u0026thinsp;\u0026lt;\u0026thinsp;2.2\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;16\u003c/sup\u003e to suggest that higher expression is associated with increased stemness, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eG.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003ch2\u003eCell-cell communication analysis\u003c/h2\u003e \u003cp\u003eCellChat analysis was used to explore how different cell types communicate with each other. Key signaling pathways were identified using a database of known ligand\u0026ndash;receptor pairs, and the strength of communication among cell groups was measured. Cell\u0026ndash;cell communication patterns were visualized using CellChat, with interaction networks quantified by both interaction count, S5\u003cb\u003eA\u003c/b\u003e and interaction weight, S5\u003cb\u003eB.\u003c/b\u003e In the network map, thicker connecting lines represent more frequent or stronger interactions among cell types, Fig. S5\u003cb\u003eA\u003c/b\u003e,\u003cb\u003eB\u003c/b\u003e. Fibroblasts showed the highest level of communication activity, interacting with tumor cells (59 interactions, weight\u0026thinsp;=\u0026thinsp;0.66), normal liver cells (53 interactions, weight\u0026thinsp;=\u0026thinsp;0.70), and pro-tumor cells (40 interactions, weight\u0026thinsp;=\u0026thinsp;0.41) to suggest a central role in shaping the tumor environment. A heatmap further illustrates the strength of signaling among different cell types, Fig. S5\u003cb\u003eC\u003c/b\u003e. Overall, fibroblasts sent strong signals to immune cells, normal hepatocytes, and tumor cells to highlight their major role as communication \u0026ldquo;hubs\u0026rdquo; within the tissue.\u003c/p\u003e \u003cp\u003eAnalysis of cell-to-cell communications showed that the BSG/CD147 signaling pathway plays a major role in shaping the tumor microenvironment, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA. Among all cell types, dendritic cells sent out signals with greatest strength (0.71), and were identified as the main \u0026ldquo;messenger\u0026rdquo; cells, while fibroblasts received the most signals (1.43) to become key \u0026ldquo;receivers\u0026rdquo; within the network, Fig. S6\u003cb\u003eA\u003c/b\u003e,\u003cb\u003eB\u003c/b\u003e. Strong communications was also observed from fibroblasts to both T cells and tumor cells, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB. Tumor cells showed the highest overall signaling strength to suggest a dominant tumor-to-T cell communication pathway mediated by BSG/CD147 and cyclophilins, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC. Tumor cells had the highest network \u0026ldquo;hub\u0026rdquo; score to highlight their central role, while fibroblasts ranked highest as signal receivers, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD. Overall, the analysis revealed distinct cellular roles as tumor cells function as communication hubs, fibroblasts as primary recipients, and immune and endothelial cells as important regulators within the BSG/CD147-cyclophilin signaling network, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eE. The strength of cyclophilin A and B signaling (PPIA/PPIB) was further quantified across normal, pre-tumor, and tumor conditions, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eF.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section2\"\u003e \u003ch2\u003eProtein-protein interaction analysis\u003c/h2\u003e \u003cp\u003eBSG/CD147 was found to interact with several key proteins, including PPIA, PPIB, MMP1, and SLC16A1, to support its known role in matrix degradation and metabolic adaptation, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eG. Network analysis identified PPIA and PPIB as hub interactors. Expression-weighted PPI scores revealed that tumor cells displayed the highest interaction strength, followed by pro-tumor and normal groups, to reflect a progressive activation of BSG/CD147-associated signaling during tumor development, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eH.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec29\" class=\"Section2\"\u003e \u003ch2\u003eEx vivo validation of target expression in HCC\u003c/h2\u003e \u003cp\u003eBSG/CD147 expression in HCC was further validated by immunofluorescence staining of paired human liver specimens. Tumor regions exhibited markedly stronger signal intensity compared with adjacent cirrhosis or background liver, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA,B. ROC analysis demonstrated excellent diagnostic accuracy, AUC\u0026thinsp;=\u0026thinsp;0.96 with 86% sensitivity and 97% specificity, to distinguish HCC from background cirrhotic liver, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eC. A weak positive correlation was observed between tumor size and protein expression, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eD. Signal was further analyzed in HCC tumors\u0026thinsp;\u0026lt;\u0026thinsp;2 cm in diameter to assess its potential utility for early detection, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eE. Despite the small size, these lesions showed clearly elevated fluorescence relative to background liver, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eE. ROC analysis for nodules\u0026thinsp;\u0026lt;\u0026thinsp;2 cm demonstrated high diagnostic accuracy, AUC\u0026thinsp;=\u0026thinsp;0.95 with 87.5% specificity and 93% sensitivity, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eF. Although the mean T/B ratio was higher in larger tumors, the ability of BSG/CD147 to generate detectable contrast in \u0026lt;\u0026thinsp;2 cm nodules underscores promise to identify early or borderline lesions that are otherwise difficult to classify radiologically (LI-RADS 3/4). Target expression also exhibited inter-patient variability, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eF, with the coefficient of variation (CV) of T/B ratios greater in large tumors (77.7%) than in small ones (40.2%) suggesting biological heterogeneity increases with tumor size, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eE. Importantly, BSG/CD147 signal remained clearly detectable within small HCC foci embedded in cirrhotic tissue to support potential utility for distinguishing indeterminate liver nodules (\u0026lt;\u0026thinsp;2 cm). In a representative specimen, strong fluorescence was observed in HCC tumor regions with only weak background signal in adjacent cirrhosis, Fig. S7\u003cb\u003eA\u003c/b\u003e-\u003cb\u003eD\u003c/b\u003e. A few isolated tumor cells within cirrhotic areas also demonstrated focal strong CD147 expression.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study integrated single-cell transcriptomics, network analysis, and ex vivo validation, and identified Basigin (BSG/CD147) as a robust early HCC biomarker and signaling hub. Using two independent scRNA-seq datasets, the transcriptional trajectory was reconstructed from normal hepatocytes through pro-tumor intermediates to malignant states, and a continuum of increasing stemness and oncogenic activation was identified. Multiple gene modules associated with tumor progression were found using hdWGCNA. The Hep-M20 module showed the strongest correlation with pathological stage. BSG/CD147 emerged from within this module as a key hub gene and exhibited progressive upregulation along pseudotime. Strong correlation with stemness potential implicates gene activation as an early molecular hallmark of hepatocarcinogenesis.\u003c/p\u003e \u003cp\u003eMechanistically, BSG/CD147 appears to function not only as a marker but also as a driver of early malignant transformation. Expression correlated with CytoTRACE-derived stemness to suggest that CD147 upregulation accompanies hepatocyte dedifferentiation toward progenitor-like phenotypes known to promote tumor initiation and recurrence [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Previous studies have shown that CD147 regulates metabolic reprogramming, extracellular matrix remodeling, and epithelial-mesenchymal transition through interactions with cyclophilins (PPIA/PPIB) and downstream MAPK/ERK, PI3K/AKT, and HIF-1α signaling cascades [\u003cspan additionalcitationids=\"CR25\" citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Our single-cell and network analyses extended these observations by revealing that CD147-based signaling networks mediate cross-talk between tumor hepatocytes and fibroblasts, the 2 dominant cell types of the tumor microenvironment to form a cyclophilin-dependent tumor-stroma communication axis. Additionally, our findings suggest that BSG/CD147 can engage T cells through PPIA and PPIB ligands to establish an immunomodulatory communication channel that shapes the tumor immune microenvironment. Our findings are consistent with the recent report that elevated CD147 expression in tumor cells is associated with an immunosuppressive tumor-immune microenvironment (TIME) in HCC as evidenced by increased infiltration of regulatory T cells [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. This pathway likely fosters a microenvironment conducive to invasion, angiogenesis, and immune modulation, thereby integrating intrinsic tumor signaling with extrinsic stromal remodeling [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eEx vivo validation further demonstrated that BSG/CD147 expression was significantly elevated in HCC compared with adjacent background liver, including in lesions\u0026thinsp;\u0026lt;\u0026thinsp;2 cm that are often radiologically indeterminate (LI-RADS 3/4). Our findings are consistent with the observations of a significantly elevated expression of CD147 in HCC compared to non-tumor [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. ROC analyses confirmed high diagnostic accuracy for small tumors to underscore the potential of BSG/CD147 for early detection. Strong membrane localization, high T/B ratio, and minimal expression in surrounding cirrhosis or background liver establishes a favorable foundation for development of targeted contrast agents. Incorporating BSG/CD147 assessment into HCC surveillance frameworks could substantially enhance performance for early diagnosis and enable biological risk stratification of indeterminate nodules. Unlike prior studies that largely evaluated CD147 expression in bulk tumor tissue or serum as a static diagnostic or prognostic marker [\u003cspan additionalcitationids=\"CR31\" citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], this work uses single-cell transcriptomics and pseudotime network modelling to demonstrate CD147 upregulation as an early event in hepatocyte transformation that is correlated with increasing stemness and malignant progression.\u003c/p\u003e \u003cp\u003eDespite the strengths of this integrated approach, several limitations should be noted. The sample size (n\u0026thinsp;=\u0026thinsp;16) for small (\u0026lt;\u0026thinsp;2 cm) lesions was modest, which may limit generalizability. In addition, the scRNA-seq datasets analyzed were annotated by pathological stage rather than derived directly from indeterminate nodules to restrict temporal resolution for early transformation events. Finally, while computational modeling identified plausible ligand-receptor and pathway interactions, further functional validation is needed to confirm the biological significance of BSG/CD147-mediated signaling in the tumor microenvironment.\u003c/p\u003e \u003cp\u003eOur results define BSG/CD147 upregulation as an early molecular event in HCC transformation to connect stemness, intercellular signaling, and microenvironmental remodeling in a unified model for early hepatocarcinogenesis. Consistent expression in small, radiologically ambiguous lesions (\u0026lt;\u0026thinsp;2 cm) highlights the translational potential as a precision biomarker for early HCC detection and image-guided diagnosis. Integration of BSG/CD147-targeted strategies into molecular imaging or surveillance workflows could transform the early HCC management by enabling detection, risk stratification, and intervention at the most curable stages of liver cancer.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAFP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eα\u0026mdash;fetoprotein\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003earea under the curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBSG\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBasigin\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCD147\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCluster of Differentiation 147\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ecomputed tomography\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ecoefficient of variation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCytoTRACE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCellular Trajectory Reconstruction Analysis using gene Counts and Expression\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDAPI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003e4\u0026prime;,6\u0026mdash;diamidino\u0026mdash;2\u0026mdash;phenylindole\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDEG\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003edifferential gene expression\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eFFPE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eformalin\u0026mdash;fixed paraffin\u0026mdash;embedded\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGEO\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGene Expression Omnibus\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGSEA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003egene set enrichment analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eH\u0026amp;E\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ehematoxylin and eosin\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eHBV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ehepatitis B virus\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eHCC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ehepatocellular carcinoma\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eHCV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ehepatitis C virus\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ehdWGCNA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ehigh\u0026mdash;dimensional weighted gene co\u0026mdash;expression network analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eHIF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003e1α\u0026mdash;hypoxia\u0026mdash;inducible factor 1 alpha\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLIHC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLiver Hepatocellular Carcinoma (TCGA cohort)\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRADS\u0026mdash;Liver Imaging Reporting and Data System\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMAPK/ERK\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emitogen\u0026mdash;activated protein kinase/extracellular signal\u0026mdash;regulated kinase\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMRI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emagnetic resonance imaging\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePBST\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ephosphate\u0026mdash;buffered saline with Tween\u0026mdash;20\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePPI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eprotein\u0026mdash;protein interaction\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePPIA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003epeptidylprolyl isomerase A (cyclophilin A)\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePPIB\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003epeptidylprolyl isomerase B (cyclophilin B)\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ereceiver operating characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eroom temperature\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003escRNA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eseq\u0026mdash;single\u0026mdash;cell RNA sequencing\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTCGA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eThe Cancer Genome Atlas\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eT/B\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003etarget\u0026mdash;to\u0026mdash;background\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTGF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eβ\u0026mdash;transforming growth factor beta\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003etSNE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003et\u0026mdash;distributed stochastic neighbor embedding\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eUMAP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003euniform manifold approximation and projection\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eUS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eultrasound\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eWGCNA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eweighted gene co\u0026mdash;expression network analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003ch2\u003eGrant support\u003c/h2\u003e\n\u003cp\u003eThis study was funded in part by the National Institutes of Health (NIH) R01 CA285303 (TDW).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHuman tissue procurement and use in this study were conducted in accordance with institutional guidelines and approved by the Institutional Review Board (IRB) of University of Michigan (Approval No: HUM00248521).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to participate:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot Applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot Applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e \u003cstrong\u003eConflict of interest\u003c/strong\u003e:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors report no conflicts of interest.\u003c/p\u003e\n\u003ch2\u003eFunding:\u003c/h2\u003e\n\u003cp\u003eThis work was supported in part by the National Institutes of Health (NIH) R01 CA285303 (TDSW).\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eTDWW performed the data analysis and prepared all figures. SJ conducted the experiments, prepared reagents and samples, contributed to data collection, reviewed the results, drafted and revised the manuscript, and finalized the text. SJ also conceived and supervised the study. EKC conducted histopathologic evaluation and provided expert interpretation of liver tissue specimens. SF collected the ex vivo specimen and performed H\u0026amp;E staining. HJ contributed to statistical analysis and computational data processing. TDSW provided scientific guidance, contributed to study design, and assisted with data interpretation. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003ch2\u003eAcknowledgements:\u003c/h2\u003e\n\u003cp\u003eThe authors thank the University of Michigan Department of Pathology for archived human liver specimens.\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe single-cell RNA sequencing datasets analyzed during this study are publicly available from the Gene Expression Omnibus (GEO) under accession numbers GSE149614 and GSE189903. Processed data, analysis scripts, and figure source files generated during this study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eLlovet JM, Kelley RK, Villanueva A, et al. Hepatocellular carcinoma. Nat Rev Dis Primers. 2021;7:6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41572-020-00240-3\u003c/span\u003e\u003cspan address=\"10.1038/s41572-020-00240-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOh JH, Jun DW. The latest global burden of liver cancer: A past and present threat. Clin Mol Hepatol. 2023;29(2):355\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3350/cmh.2023.0070\u003c/span\u003e\u003cspan address=\"10.3350/cmh.2023.0070\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2023 Mar 9. PMID: 36891606; PMCID: PMC10121295.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlberts CJ, Clifford GM, Georges D, Negro F, Lesi OA, Hutin YJ, de Martel C. Worldwide prevalence of hepatitis B virus and hepatitis C virus among patients with cirrhosis at country, region, and global levels: a systematic review. Lancet Gastroenterol Hepatol. 2022;7(8):724\u0026ndash;35. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S2468-1253(22)00050-4\u003c/span\u003e\u003cspan address=\"10.1016/S2468-1253(22)00050-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2022 May 14. PMID: 35576953; PMCID: PMC9259503.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatel N, Yopp AC, Singal AG. Diagnostic delays are common among patients with hepatocellular carcinoma. J Natl Compr Canc Netw. 2015;13(5):543\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.6004/jnccn.2015.0074\u003c/span\u003e\u003cspan address=\"10.6004/jnccn.2015.0074\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 25964640; PMCID: PMC5624109.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParikh ND, Mehta AS, Singal AG, Block T, Marrero JA, Lok AS. Biomarkers for the Early Detection of Hepatocellular Carcinoma. Cancer Epidemiol Biomarkers Prev. 2020;29(12):2495\u0026ndash;503. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1158/1055-9965.EPI-20-0005\u003c/span\u003e\u003cspan address=\"10.1158/1055-9965.EPI-20-0005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2020 Apr 1. PMID: 32238405; PMCID: PMC7529652.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan der Pol CB, Lim CS, Sirlin CB, McGrath TA, Salameh JP, Bashir MR, Tang A, Singal AG, Costa AF, Fowler K, McInnes MDF. Accuracy of the Liver Imaging Reporting and Data System in Computed Tomography and Magnetic Resonance Image Analysis of Hepatocellular Carcinoma or Overall Malignancy -. Syst Rev Gastroenterol. 2019;156:976\u0026ndash;86.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAddissouky TA, Sayed IETE, Ali MMA, et al. Latest advances in hepatocellular carcinoma management and prevention through advanced technologies. Egypt Liver J. 2024;14:2. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s43066-023-00306-3\u003c/span\u003e\u003cspan address=\"10.1186/s43066-023-00306-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan Q, Sun YS, An R, Liu F, Fang Q, Wang Z, Xu T, Chen L, Du J. Application and progress of the detection technologies in hepatocellular carcinoma. Genes Dis. 2022;10(5):1857\u0026ndash;69. PMID: 37492708; PMCID: PMC10363596.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu G, Bajestani N, Pracha N, Chen C, Makary MS. Hepatocellular Carcinoma Surveillance Strategies: Major Guidelines and Screening Advances. Cancers. 2024;16:3933.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSafri F, Nguyen R, Zerehpooshnesfchi S, George J, Qiao L. Heterogeneity of hepatocellular carcinoma: from mechanisms to clinical implications. Cancer Gene Ther. 2024;31(8):1105\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41417-024-00764-w\u003c/span\u003e\u003cspan address=\"10.1038/s41417-024-00764-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2024 Mar 18. PMID: 38499648; PMCID: PMC11327108.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNguyen PHD, Ma S, Phua CZJ, Kaya NA, Lai HLH, Lim CJ, Lim JQ, Wasser M, Lai L, Tam WL, Lim TKH, Wan WK, Loh T, Leow WQ, Pang YH, Chan CY, Lee SY, Cheow PC, Toh HC, Ginhoux F, Iyer S, Kow AWC, Young Dan Y, Chung A, Bonney GK, Goh BKP, Albani S, Chow PKH, Zhai W, Chew V. Intratumoural immune heterogeneity as a hallmark of tumour evolution and progression in hepatocellular carcinoma. Nat Commun. 2021;12:227.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang ZL, Xu B, Li TT, Xu YH, Huang XY, Huang XY. Integrative Analysis Identifies Cell-Type-Specific Genes Within Tumor Microenvironment as Prognostic Indicators in Hepatocellular Carcinoma. Front Oncol. 2022;12:878923.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi X, Wang CY. From bulk, single-cell to spatial RNA sequencing. Int J Oral Sci. 2021;13:36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41368-021-00146-0\u003c/span\u003e\u003cspan address=\"10.1038/s41368-021-00146-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSatija R, Farrell JA, Gennert D, Schier AF, Regev A. Spatial reconstruction of single-cell gene expression data. Nat Biotechnol. 2015;33(5):495\u0026ndash;502. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nbt.3192\u003c/span\u003e\u003cspan address=\"10.1038/nbt.3192\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2015 Apr 13. PMID: 25867923; PMCID: PMC4430369.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGulati GS, Sikandar SS, Wesche DJ, Manjunath A, Bharadwaj A, Berger MJ, Ilagan F, Kuo AH, Hsieh RW, Cai S, Zabala M, Scheeren FA, Lobo NA, Qian D, Yu FB, Dirbas FM, Clarke MF, Newman AM. Single-cell transcriptional diversity is a hallmark of developmental potential. Science. 2020;367(6476):405\u0026ndash;11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1126/science.aax0249\u003c/span\u003e\u003cspan address=\"10.1126/science.aax0249\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 31974247; PMCID: PMC7694873.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorabito S, Reese F, Rahimzadeh N, Miyoshi E, Swarup V. hdWGCNA identifies co-expression networks in high-dimensional transcriptomics data. Cell Rep Methods. 2023;3(6):100498. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.crmeth.2023.100498\u003c/span\u003e\u003cspan address=\"10.1016/j.crmeth.2023.100498\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 37426759; PMCID: PMC10326379.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTrapnell C, Cacchiarelli D, Grimsby J, et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat Biotechnol. 2014;32:381\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nbt.2859\u003c/span\u003e\u003cspan address=\"10.1038/nbt.2859\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin S, Plikus MV, Nie Q. CellChat for systematic analysis of cell-cell communication from single-cell transcriptomics. Nat Protoc. 2025;20(1):180\u0026ndash;219. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41596-024-01045-4\u003c/span\u003e\u003cspan address=\"10.1038/s41596-024-01045-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2024 Sep 16. PMID: 39289562.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDang CV. MYC on the path to cancer. Cell. 2012;149(1):22\u0026ndash;35. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cell.2012.03.003\u003c/span\u003e\u003cspan address=\"10.1016/j.cell.2012.03.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 22464321; PMCID: PMC3345192.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJackson S, Bartek J. The DNA-damage response in human biology and disease. Nature. 2009;461:1071\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nature08467\u003c/span\u003e\u003cspan address=\"10.1038/nature08467\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMalumbres M, Barbacid M. Cell cycle, CDKs and cancer: a changing paradigm. Nat Rev Cancer. 2009;9:153\u0026ndash;66. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nrc2602\u003c/span\u003e\u003cspan address=\"10.1038/nrc2602\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDimova D, Dyson N. The E2F transcriptional network: old acquaintances with new faces. Oncogene. 2005;24:2810\u0026ndash;26. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/sj.onc.1208612\u003c/span\u003e\u003cspan address=\"10.1038/sj.onc.1208612\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMuramatsu S, Tanaka S, Mogushi K, Adikrisna R, Aihara A, Ban D, Ochiai T, Irie T, Kudo A, Nakamura N, Nakayama K, Tanaka H, Yamaoka S, Arii S. Visualization of stem cell features in human hepatocellular carcinoma reveals in vivo significance of tumor-host interaction and clinical course. Hepatology. 2013. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003edoi/10.1002/hep.26345\u003c/span\u003e\u003cspan address=\"doi/10.1002/hep.26345\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://onlinelibrary.wiley.com/\u003c/span\u003e\u003cspan address=\"https://onlinelibrary.wiley.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi M, Zhai Q, Bharadwaj U, Wang H, Li F, Fisher WE, Chen C, Yao Q. Cyclophilin A is overexpressed in human pancreatic cancer cells and stimulates cell proliferation through CD147. Cancer. 2006;106(10):2284-94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/cncr.21862\u003c/span\u003e\u003cspan address=\"10.1002/cncr.21862\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 16604531.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNigro P, Pompilio G, Capogrossi MC. Cyclophilin A: a key player for human disease. Cell Death Dis. 2013;4(10):e888. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/cddis.2013.410\u003c/span\u003e\u003cspan address=\"10.1038/cddis.2013.410\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 24176846; PMCID: PMC3920964.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSakamoto M, Miyagaki T, Kamijo H, Oka T, Boki H, Takahashi-Shishido N, Suga H, Sugaya M, Sato S. CD147-Cyclophilin a Interactions Promote Proliferation and Survival of Cutaneous T-Cell Lymphoma. Int J Mol Sci. 2021;22(15):7889. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/ijms22157889\u003c/span\u003e\u003cspan address=\"10.3390/ijms22157889\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 34360654; PMCID: PMC8346093.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShigematsu Y, Kanda H, Takahashi Y, Takeuchi K, Inamura K. Relationships between tumor CD147 expression, tumor-infiltrating lymphocytes, and oncostatin M in hepatocellular carcinoma. Virchows Arch. 2025;487(1):117\u0026ndash;25. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00428-024-03939-w\u003c/span\u003e\u003cspan address=\"10.1007/s00428-024-03939-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2024 Oct 12. PMID: 39395054.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNyalali AM, Leonard AU, Xu Y, Li H, Zhou J, Zhang X, Rugambwa TK, Shi X, Li F. CD147: an integral and potential molecule to abrogate hallmarks of cancer. Front Oncol. 2023;13:1238051.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMamori S, Nagatsuma K, Matsuura T, Ohkawa K, Hano H, Fukunaga M, Matsushima M, Masui Y, Fushiya N, Onoda H, Searashi Y, Takagi I, Tagiri H. Useful detection of CD147 (EMMPRIN) for pathological diagnosis of early hepatocellular carcinoma in needle biopsy samples. World J Gastroenterol. 2007;13(21):2913\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3748/wjg.v13.i21.2913\u003c/span\u003e\u003cspan address=\"10.3748/wjg.v13.i21.2913\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 17589939; PMCID: PMC4171141.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang DF, Zhang WJ, Chen J, Jiao ZG, Wang XL, Rao DY, Li WS, Hu D, Xie FF, Wang XX, Li ZZ, Yi XM, Wu JY, Jiang Y, Wang Q, Zhong TY. Hepatocellular carcinoma cell-derived small extracellular vesicle-associated CD147 serves as a diagnostic marker and promotes endothelial cell angiogenesis via the PI3K/Akt pathway. Extracell Vesicles Circ Nucl Acids. 2023;4(4):532\u0026ndash;47. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.20517/evcna.2023.30\u003c/span\u003e\u003cspan address=\"10.20517/evcna.2023.30\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 40357132; PMCID: PMC12066417.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu J, Hao ZW, Zhao YX, Yang XM, Tang H, Zhang X, Song F, Sun XX, Wang B, Nan G, Chen ZN, Bian H. Full-length soluble CD147 promotes MMP-2 expression and is a potential serological marker in detection of hepatocellular carcinoma. J Transl Med. 2014;12:190. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/1479-5876-12-190\u003c/span\u003e\u003cspan address=\"10.1186/1479-5876-12-190\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 24996644; PMCID: PMC4227008.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee A, Rode A, Nicoll A, Maczurek AE, Lim L, Lim S, Angus P, Kronborg I, Arachchi N, Gorelik A, Liew D, Warner FJ, McCaughan GW, McLennan SV, Shackel NA. Circulating CD147 predicts mortality in advanced hepatocellular carcinoma. J Gastroenterol Hepatol. 2016;31(2):459\u0026thinsp;\u0026ndash;\u0026thinsp;66. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/jgh.13148\u003c/span\u003e\u003cspan address=\"10.1111/jgh.13148\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 26312403.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8593693/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8593693/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eHepatocellular carcinoma (HCC) is a leading cause of cancer-related mortality worldwide, largely due to late-stage diagnosis and the limited sensitivity of current biomarkers such as α-fetoprotein (AFP). Early detection requires molecularly defined targets that capture the initial steps of malignant transformation. Single-cell RNA sequencing (scRNA-seq) offers high-resolution insight into tumor heterogeneity and lineage progression to enable the identification of early biomarkers. This study aimed apply scRNA-seq analysis to detect clinically important molecular patterns that define the early stages of malignant transformation in HCC and facilitate the diagnosis of small or ambiguous lesions.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eTwo independent scRNA-seq datasets (GSE149614 and GSE189903) comprising non-tumor and HCC tissues were analyzed. Following batch correction and clustering, hepatocyte subpopulations were characterized by differential expression, pseudotime, and CytoTRACE analyses to reconstruct the trajectory from normal to malignant states. High-dimensional weighted gene co-expression network analysis (hdWGCNA) was used to identify stage-associated modules, while CellChat and protein-protein interaction analyses delineated intercellular signaling networks. Target expression was validated in paired human liver specimens using quantitative immunofluorescence.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003escRNA-seq revealed a continuum of hepatocyte states characterized by progressive stemness and oncogenic pathway activation (MYC, E2F, G2M). Module Hep-M20 exhibited the strongest correlation with tumor stage and identified BSG/CD147 as a central hub gene with monotonic upregulation along pseudotime and strong correlation with stemness potential. CellChat analysis uncovered a cyclophilin (PPIA/PPIB)-dependent tumor-stroma signaling axis that positions BSG/CD147 as the key mediator for intercellular communication between tumor hepatocytes, fibroblasts and T cells. Ex vivo validation confirmed significantly higher BSG/CD147 protein expression in HCC versus background liver (P\u0026thinsp;=\u0026thinsp;2.9\u0026times;10⁻\u003csup\u003e11\u003c/sup\u003e) with excellent diagnostic accuracy (AUC\u0026thinsp;=\u0026thinsp;0.93\u0026ndash;0.96; sensitivity 86\u0026ndash;87%; specificity 93\u0026ndash;97%) including in lesions\u0026thinsp;\u0026lt;\u0026thinsp;2 cm that are frequently indeterminate on conventional imaging.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThis study establishes BSG/CD147 upregulation as an early molecular event in hepatocarcinogenesis that integrates hepatocyte dedifferentiation, microenvironmental signaling, and tumor progression. Strong and specific expression in small lesions\u0026thinsp;\u0026lt;\u0026thinsp;2 cm underscores potential as a precision biomarker and imaging target for early HCC detection, risk stratification, and therapeutic development.\u003c/p\u003e","manuscriptTitle":"Multimodal single-cell network analysis uncovers BSG/CD147 as an early biomarker and signaling hub in hepatocellular carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-27 10:36:01","doi":"10.21203/rs.3.rs-8593693/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3732e081-36d6-49f1-8b7f-df5c3fe6dcc4","owner":[],"postedDate":"January 27th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-01-28T05:25:46+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-27 10:36:01","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8593693","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8593693","identity":"rs-8593693","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}