Single-cell reanalysis highlights the MIF-PTGDR axis as a candidate immunoregulatory program in hepatocellular carcinoma Authors: Dingzhi Liu, Lang Yan, Huiya Chen, Ling Yuan Published: 2026 ABSTRACT BackgroundHepatocellular carcinoma (HCC) remains one of the most lethal malignancies worldwide, and immune evasion is a major determinant of tumor progression and therapeutic resistance. The tumor microenvironment (TME) contains multiple dysfunctional and suppressive immune states, among which regulatory T cells (Tregs) and exhausted T-cell populations are particularly important. Despite recent advances in HCC biology, the intercellular signaling programs associated with these immune states remain incompletely characterized at single-cell resolution. Therefore, this study aimed to reanalyze public single-cell RNA sequencing (scRNA-seq) data to identify candidate interaction networks and immunoregulatory signaling axes associated with T-cell dysfunction in HCC.MethodsPublicly available single-cell RNA sequencing data were obtained from the Gene Expression Omnibus (GEO) database and analyzed in a reanalysis framework focused on immune-cell heterogeneity. Quality control, normalization, dimensionality reduction, and clustering were performed using Seurat-based workflows. Cell-type annotation was conducted primarily through canonical marker gene expression patterns. Differential expression, cell-cell communication analysis, trajectory inference, gene co-expression analysis, and pathway enrichment analysis were used to characterize signaling programs associated with dysfunctional T-cell states, with particular attention to macrophage migration inhibitory factor (MIF) and prostaglandin D2 receptor (PTGDR) signaling.ResultsReanalysis of the single-cell dataset identified major cellular populations including CD8 + T cells, regulatory T cells (Tregs), macrophages, and proliferating cells, highlighting marked heterogeneity within the HCC microenvironment. Three Treg-associated subclusters, designated Treg-C1, Treg-C2, and Treg-C3, showed distinct transcriptional characteristics. Cell-cell communication analysis suggested that MIF signaling through CD74, CXCR4, and CD44 may represent a prominent immunoregulatory communication program linking tumor-associated cell populations with T-cell compartments. PTGDR was enriched in selected immune-cell populations and was associated with gene programs related to cytokine signaling, cell adhesion, and T-cell activation. Trajectory and co-expression analyses further supported the presence of dynamic immune-state transitions and heterogeneous regulatory programs within the analyzed dataset.ConclusionIn conclusion, this single-cell reanalysis delineated interaction patterns among immune-associated cell populations in HCC and highlighted the MIF-CD74/CXCR4/CD44 and PTGDR signaling axes as candidate immunoregulatory programs. These findings refine the interpretation of T-cell dysfunction in the HCC microenvironment and provide a rationale for further mechanistic and translational validation, rather than direct confirmation of therapeutic targets.Clinical Trial Registration This study is not a clinical trial. Introduction Hepatocellular carcinoma (HCC) is the sixth most common cancer globally and the third leading cause of cancer-related death, with increasing incidence in many regions. Despite advances in surgical techniques, locoregional therapies, and systemic treatments including immune checkpoint inhibitors, the prognosis for HCC patients remains poor, with 5-year survival rates below 20% for advanced disease. In addition to malignant-cell intrinsic programs, the immunosuppressive tumor microenvironment, characterized by exhausted T cells and abundant regulatory T cells (Tregs), facilitates immune evasion and limits the efficacy of immunotherapy [1–3]. The interaction between tumor-associated cell states and immune cells plays a central role in hepatocellular carcinoma pathogenesis and therapeutic resistance. In particular, proliferative tumor-associated programs, dysfunctional CD8 + T cells, and suppressive Treg populations may jointly shape immune escape. Exhausted T cells, particularly CD8 + T cells expressing multiple inhibitory receptors (PD-1, TIM-3, LAG-3, TIGIT), exhibit impaired effector functions and contribute to tumor immune escape. Regulatory T cells (Tregs) further enhance immunosuppression through direct cell-cell contact and secretion of inhibitory cytokines. Understanding the signaling programs associated with these cellular interactions is critical for developing more effective therapeutic strategies [4–6]. Single-cell RNA sequencing (scRNA-seq) technology has revolutionized our understanding of tumor heterogeneity and the tumor microenvironment. This powerful approach enables unbiased characterization of individual cells within complex tissues, revealing cellular diversity, rare cell populations, transcriptional states, and intercellular communication networks. In the context of hepatocellular carcinoma research, scRNA-seq provides unprecedented resolution to dissect the immune landscape, identify disease-relevant cell subpopulations, and uncover molecular mechanisms driving tumor-immune interactions. The Seurat package has emerged as a comprehensive computational framework for scRNA-seq data analysis, offering robust methods for quality control, normalization, clustering, differential expression analysis, and cell-cell communication inference [7–9]. Despite the recognized importance of immune-cell interactions in HCC, comprehensive single-cell characterization of the molecular networks associated with dysfunctional T-cell states remains limited. Macrophage migration inhibitory factor (MIF) has been identified as a key immunoregulatory cytokine in various cancers, acting through receptors including CD74, CXCR4, and CD44. Prostaglandin D2 receptor (PTGDR) represents another potentially relevant immune-associated molecule. Understanding the specific roles of these molecules in shaping T-cell-associated communication patterns, identifying the cellular sources and targets of these signals, and characterizing related downstream programs may help prioritize candidate mechanisms for further study [10–12]. In this study, we leveraged publicly available single-cell RNA sequencing data from the Gene Expression Omnibus (GEO) database to reanalyze interaction patterns associated with dysfunctional T-cell states in hepatocellular carcinoma. We performed systematic quality control, identified major immune-associated cell populations, characterized Treg heterogeneity, and investigated cell-cell communication networks with particular focus on MIF and PTGDR signaling axes. Through trajectory analysis, gene regulatory network inference, and pathway enrichment studies, we aimed to define candidate immunoregulatory programs within the HCC microenvironment. Our findings provide a refined interpretive framework for understanding immune dysregulation in HCC and a basis for future mechanistic validation. Materials and methods Data acquisition Single-cell RNA sequencing data were obtained from the NCBI Gene Expression Omnibus (GEO) public database. According to the analysis code, the working matrix was generated from GSE307831_PT2_T_matrix.mtx_2.gz, with matched feature and barcode files (GSE307831_PT2_T_features.tsv.gz and GSE307831_PT2_T_barcodes.tsv.gz). The matrix contained 36,623 genes and 10,277 cell barcodes before duplicate gene-name removal. After removing duplicated gene symbols, 36,613 genes were retained for downstream analysis. Because this was a retrospective reanalysis, the interpretation of representativeness was limited by the original sampling design and the available cell-level metadata. Gene count, UMI count, mitochondrial percentage, Seurat cluster identity, and cell-type annotation were retained as the main metadata fields used in downstream analyses [13, 14]. Quality control and data preprocessing The Seurat package (version 4.0 or higher) in R was used for preprocessing and quality control. Mitochondrial gene content was calculated using PercentageFeatureSet with the pattern ^MT-. Cells were retained when nFeature_RNA was > 100 and < 5,000 and percent.mt was < 5. Under these thresholds, 10,034 cells passed quality control. Data normalization was performed using LogNormalize with a scale factor of 10,000. Highly variable genes were selected using the vst method with nfeatures = 2,000. The data were then scaled and subjected to PCA using the selected variable features. JackStraw analysis with 100 replicates and ScoreJackStraw for dimensions 1:20 were used as supporting checks for principal component selection [15, 16]. Dimensionality reduction and clustering Uniform Manifold Approximation and Projection (UMAP), t-distributed Stochastic Neighbor Embedding (t-SNE), and PCA-based visualization were used to examine cellular heterogeneity. In the Seurat workflow, FindNeighbors was performed using dimensions 1:10, and graph-based clustering was performed with FindClusters at resolution = 0.1. UMAP and t-SNE were also generated using dimensions 1:10. Cluster interpretation was checked against canonical marker genes, and the resulting clusters were manually grouped into four major immune-associated populations for the main comparative analyses [17, 18]. Cell type annotation Cell type identification was performed primarily through canonical marker gene expression analysis. Known cell type-specific markers were used to annotate major populations including CD8 + T cells (CD8A, CD3E), regulatory T cells (FOXP3, IL2RA, CTLA4), macrophages (CD68, CD14, FCGR3A), and proliferating cells (MKI67, TOP2A). Differential expression analysis was conducted using the Wilcoxon rank-sum test to identify cluster-associated marker genes. Expression patterns were visualized using violin plots, dot plots, heatmaps, and low-dimensional feature maps [19, 20]. Regulatory T cell subpopulation analysis Regulatory T cells were subset from the annotated Seurat object and subjected to exploratory subclustering to identify transcriptionally different states. To avoid assigning unsupported biological labels, the resulting subclusters were designated Treg-C1, Treg-C2, and Treg-C3. Differential expression analysis was performed between subclusters using FindAllMarkers or related Seurat marker-testing functions. Marker detection followed the code-level settings of only.pos = TRUE, min.pct = 0.1, and logfc.threshold = 0.1 for cluster marker screening, with adjusted p-value control applied where applicable. Module scores were calculated from marker gene sets derived from these subcluster-associated genes, and the scores were used only as relative indicators of transcriptional program enrichment [21, 22]. Cell-cell communication analysis Cell-cell communication analysis was performed using CellChat. The CellChat object was created from the Seurat object using cell_type as the grouping variable. The human CellChat database was used, and the analysis was restricted to the Secreted Signaling subset. Overexpressed genes and ligand-receptor interactions were identified using the standard CellChat workflow. Communication probabilities were computed with computeCommunProb using raw.use = TRUE and population.size = TRUE, followed by filterCommunication with min.cells = 10. The standard CellChat resampling procedure was retained without manual override, and interactions with nominal CellChat p < 0.05 were treated as significant. Communication networks were visualized by interaction counts and inferred interaction weights. These outputs were interpreted as transcriptome-based communication potential rather than proof of direct physical binding or causal regulation [23, 24]. Trajectory analysis Pseudotime trajectory analysis was performed using Monocle. A CellDataSet object was generated from the Seurat count matrix and metadata. Size factors and dispersions were estimated before ordering genes were selected from the dispersion table. Dimensionality reduction was performed using DDRTree with max_components = 2, and cells were ordered with orderCells. Differential genes along pseudotime were tested using differentialGeneTest with the model formula ~ sm.ns(Pseudotime). Branch-related changes were explored using BEAM with branch_point = 1. These analyses were treated as exploratory state-transition analyses rather than definitive lineage reconstruction [25, 26]. Gene regulatory network analysis Weighted gene co-expression analysis was performed using WGCNA/hdWGCNA to identify co-expressed modules associated with cell-type or state-related programs. In the uploaded code, SetupForWGCNA retained genes expressed in at least 5% of cells (fraction = 0.05). Metacells were constructed by cell_type using k = 25 and max_shared = 10, followed by metacell normalization. Soft-thresholding power was evaluated with TestSoftPowers under a signed network setting and visualized with PlotSoftPowers. Network construction was then performed with ConstructNetwork, and module eigengenes, module connectivity, and hub genes were calculated using ModuleEigengenes, ModuleConnectivity, and GetHubGenes. Because the final soft power is selected from the data-driven TestSoftPowers output, it was not described as a fixed universal value [27, 28]. Differential expression and pathway enrichment analysis Comparative differential expression analysis was performed to identify genes associated with PTGDR expression levels. Cells were stratified into PTGDR-high and PTGDR-low groups, and differential expression testing was conducted. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using clusterProfiler. Statistical significance was determined using hypergeometric tests with Benjamini-Hochberg correction for multiple testing. Enrichment results were interpreted as functional associations within the analyzed dataset [29, 30]. Statistical analysis and visualization All statistical analyses were performed in R using Seurat, CellChat, Monocle, WGCNA/hdWGCNA, clusterProfiler, ggplot2, ComplexHeatmap, and related packages. Marker-gene comparisons mainly used Seurat Wilcoxon-based testing. Multiple testing was controlled using the Benjamini-Hochberg false discovery rate method when adjusted p-values were reported. For qRT-PCR validation, Student’s t-test was used to compare relative expression between cell lines. A threshold of adjusted p-value < 0.05 or nominal p < 0.05 was used according to the analysis type and is stated in the corresponding figure legends [31, 32]. Quantitative real-time PCR validation To validate the key findings from single-cell RNA sequencing analysis, quantitative real-time PCR (qRT-PCR) was performed on two hepatocellular carcinoma cell lines obtained from the American Type Culture Collection (ATCC): HepG2 (ATCC HB-8065) and Huh7 (JCRB0403). Both cell lines were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin at 37 °C in a humidified atmosphere with 5% CO2. Total RNA was extracted using TRIzol reagent (Invitrogen) according to the manufacturer’s protocol. RNA concentration and purity were assessed using a NanoDrop spectrophotometer, and only samples with A260/A280 ratios between 1.8 and 2.0 were used for subsequent analysis. First-strand cDNA synthesis was performed using 1 µg of total RNA with the PrimeScript RT reagent Kit (Takara) following the manufacturer’s instructions. qRT-PCR was conducted using SYBR Green Master Mix (Applied Biosystems) on a QuantStudio 5 Real-Time PCR System (Thermo Fisher Scientific). The expression levels of four key genes identified from scRNA-seq analysis—MIF, CD74, CXCR4, and PTGDR—were quantified. GAPDH was used as an endogenous reference gene for normalization. The thermal cycling conditions consisted of an initial denaturation at 95 °C for 10 min, followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min. Melting curve analysis was performed to verify the specificity of PCR products. Relative gene expression was calculated using the 2⁻ΔΔCt method. Each experiment was performed in triplicate, and results are presented as mean ± standard error of the mean (SEM). Statistical significance was determined using Student’s t-test, with p < 0.05 considered statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001) [33, 34]. Results Quality control and data preprocessing yield high-quality single-cell transcriptome data A total of 10,277 cell barcodes were initially included in the working expression matrix. After duplicate gene-name removal and quality control filtering using nFeature_RNA > 100, nFeature_RNA < 5,000, and percent.mt < 5, 10,034 cells remained for downstream analysis. Quality control plots showed the distributions of detected genes, UMI counts, and mitochondrial percentages across the analyzed cells (Fig. 1A-B). The highly variable gene analysis retained 2,000 genes selected by the vst method, and PCA was used to summarize major transcriptional variation before clustering and visualization (Fig. 1C-E). These filtering results supported the use of the dataset for downstream annotation, communication analysis, co-expression analysis, and trajectory exploration (Fig. 1F). Fig. 1Quality Control and Data Preprocessing of Single-Cell RNA Sequencing Data. A QC metric violin plots showing nFeature_RNA, nCount_RNA, and percent.mt across analyzed cells. Cells were retained using nFeature_RNA > 100, nFeature_RNA < 5,000, and percent.mt < 5. B QC metric scatter plots showing nCount_RNA versus percent.mt and nCount_RNA versus nFeature_RNA, which were used to inspect outliers before filtering. C Highly variable gene plot generated by the vst method, with 2,000 variable genes selected from 36,613 retained genes after duplicate gene-name removal. D Principal component loading plots for PC_1 and PC_2 showing genes contributing to early transcriptional variation. (E) PCA projection showing the major structure of the analyzed cells before graph-based clustering. F Heatmap visualization of selected variable genes showing coordinated transcriptional patterns across the retained cells Quality Control and Data Preprocessing of Single-Cell RNA Sequencing Data. A QC metric violin plots showing nFeature_RNA, nCount_RNA, and percent.mt across analyzed cells. Cells were retained using nFeature_RNA > 100, nFeature_RNA < 5,000, and percent.mt < 5. B QC metric scatter plots showing nCount_RNA versus percent.mt and nCount_RNA versus nFeature_RNA, which were used to inspect outliers before filtering. C Highly variable gene plot generated by the vst method, with 2,000 variable genes selected from 36,613 retained genes after duplicate gene-name removal. D Principal component loading plots for PC_1 and PC_2 showing genes contributing to early transcriptional variation. (E) PCA projection showing the major structure of the analyzed cells before graph-based clustering. F Heatmap visualization of selected variable genes showing coordinated transcriptional patterns across the retained cells Dimensionality reduction and clustering reveal distinct cellular populations Dimensionality reduction and clustering analyses resolved four major cell clusters with distinct transcriptional characteristics. UMAP and t-SNE projections showed clear spatial separation of the major populations, supporting robust clustering structure (Fig. 2C-F). Based on canonical marker genes, these clusters were annotated as CD8 + T cells, Tregs, proliferating cells, and macrophages, indicating substantial immune and functional heterogeneity within the analyzed HCC-derived dataset. Fig. 2Dimensionality Reduction and Cell Type Clustering. A JackStraw analysis plot showing empirical p-values for dimensions 1:15 after running JackStraw with 100 replicates and ScoreJackStraw for dimensions 1:20. B Elbow plot used with JackStraw results to guide principal component inspection. For clustering and visualization, dimensions 1:10 were used in FindNeighbors, RunUMAP, and RunTSNE. C UMAP projection of cells colored by Seurat cluster identity, generated after FindClusters with resolution = 0.1. D t-SNE projection providing a complementary view of the same clustering structure. E UMAP projection colored by manually annotated cell types: CD8 + T cells, Tregs, proliferating cells, and macrophages. F t-SNE projection with the same cell-type annotation, supporting consistency of the major immune-associated grouping Dimensionality Reduction and Cell Type Clustering. A JackStraw analysis plot showing empirical p-values for dimensions 1:15 after running JackStraw with 100 replicates and ScoreJackStraw for dimensions 1:20. B Elbow plot used with JackStraw results to guide principal component inspection. For clustering and visualization, dimensions 1:10 were used in FindNeighbors, RunUMAP, and RunTSNE. C UMAP projection of cells colored by Seurat cluster identity, generated after FindClusters with resolution = 0.1. D t-SNE projection providing a complementary view of the same clustering structure. E UMAP projection colored by manually annotated cell types: CD8 + T cells, Tregs, proliferating cells, and macrophages. F t-SNE projection with the same cell-type annotation, supporting consistency of the major immune-associated grouping Cell-cell communication analysis identifies mif as a central signaling hub Cell-cell communication analysis revealed complex predicted interaction networks within the tumor microenvironment. Analysis of outgoing and incoming interaction strengths suggested that Tregs may represent a major outgoing signaling population, whereas CD8 + T cells showed prominent incoming interactions (Fig. 3A). MIF signaling network analysis highlighted extensive predicted communication among the major cell populations (Fig. 3B-C). WGCNA further identified Treg-associated gene modules with distinct transcriptional properties, supporting heterogeneity within the regulatory compartment (Fig. 3D-F). Fig. 3Cell-Cell Communication Networks and Weighted Gene Co-expression Analysis. A Scatter plot showing outgoing versus incoming interaction strength for each annotated cell type. Bubble size reflects the number of inferred interactions. B MIF signaling network diagram generated by CellChat. In the network diagrams, node size reflects the number of cells or total communication activity of each cell population, depending on the CellChat plotting function, and edge width indicates inferred communication probability or interaction weight between sender and receiver cell types. Thicker edges indicate stronger predicted ligand-receptor communication. C Violin plots showing expression distributions of MIF pathway components (MIF, CD74, CXCR4, CD44) across the four cell types. D Soft-power threshold plots generated in the hdWGCNA workflow using TestSoftPowers under a signed network setting. E Hierarchical clustering dendrogram showing module assignment. F Module eigengene plots showing representative co-expression patterns; module-level results were used for exploratory interpretation and require functional validation Cell-Cell Communication Networks and Weighted Gene Co-expression Analysis. A Scatter plot showing outgoing versus incoming interaction strength for each annotated cell type. Bubble size reflects the number of inferred interactions. B MIF signaling network diagram generated by CellChat. In the network diagrams, node size reflects the number of cells or total communication activity of each cell population, depending on the CellChat plotting function, and edge width indicates inferred communication probability or interaction weight between sender and receiver cell types. Thicker edges indicate stronger predicted ligand-receptor communication. C Violin plots showing expression distributions of MIF pathway components (MIF, CD74, CXCR4, CD44) across the four cell types. D Soft-power threshold plots generated in the hdWGCNA workflow using TestSoftPowers under a signed network setting. E Hierarchical clustering dendrogram showing module assignment. F Module eigengene plots showing representative co-expression patterns; module-level results were used for exploratory interpretation and require functional validation MIF signaling pathway exhibits complex multi-directional communication patterns Detailed analysis of MIF signaling revealed multi-directional predicted communication patterns between major cell populations. Interaction count and interaction strength visualizations consistently highlighted strong inferred connections among macrophages, proliferating cells, CD8 + T cells, and Tregs (Fig. 4A-B). Violin plots and dot plots further showed differential expression of MIF, CD74, CXCR4, and CD44 across these populations (Fig. 4C-D), supporting the interpretation that the MIF axis may represent a prominent candidate immunoregulatory communication program in this dataset. Fig. 4Detailed MIF Signaling Pathway Analysis and Cell-Cell Interactions. A Four-panel CellChat network visualization of MIF signaling. Interaction count reflects the number of inferred ligand-receptor links, while interaction strength reflects the summed communication weight. Node size reflects group size or communication activity according to the CellChat plotting parameter used, and edge width represents inferred communication probability or interaction strength. B Directional signaling analysis showing inferred MIF-related signals from each sender cell type to receiver populations. C Violin plots showing expression distributions of MIF, CD74, CXCR4, and CD44 across CD8 + T cells, macrophages, proliferating cells, and Tregs. D Dot plot matrix showing average expression level by color intensity and the percentage of expressing cells by dot size Detailed MIF Signaling Pathway Analysis and Cell-Cell Interactions. A Four-panel CellChat network visualization of MIF signaling. Interaction count reflects the number of inferred ligand-receptor links, while interaction strength reflects the summed communication weight. Node size reflects group size or communication activity according to the CellChat plotting parameter used, and edge width represents inferred communication probability or interaction strength. B Directional signaling analysis showing inferred MIF-related signals from each sender cell type to receiver populations. C Violin plots showing expression distributions of MIF, CD74, CXCR4, and CD44 across CD8 + T cells, macrophages, proliferating cells, and Tregs. D Dot plot matrix showing average expression level by color intensity and the percentage of expressing cells by dot size Regulatory T cell subpopulation analysis reveals functional heterogeneity Detailed analysis of regulatory T cells identified three transcriptionally distinct subpopulations designated Treg-C1, Treg-C2, and Treg-C3. UMAP visualization demonstrated clear separation of these subclusters (Fig. 5A). Module score analysis and marker-gene comparisons indicated significant differences in transcriptional programs across the three Treg subclusters (Fig. 5B-C). Correlation, differential-expression, and heatmap analyses further supported functional heterogeneity within the Treg compartment, while avoiding direct assignment of fixed biological identities without additional experimental validation (Fig. 5D-F). Fig. 5Regulatory T Cell Subpopulation Analysis and Functional Heterogeneity. A UMAP projections showing three Treg subpopulations designated Treg-C1, Treg-C2, and Treg-C3. The labels are descriptive cluster names rather than fixed functional identities. B Bar chart comparing module scores across Treg-C1, Treg-C2, and Treg-C3. Module scores were calculated from Treg subcluster marker gene sets using a Seurat-style module-scoring approach, and higher values indicate stronger relative enrichment of the corresponding transcriptional program. C Dot plot showing expression patterns of marker genes across Tregs, proliferating cells, macrophages, and CD8 + T cells, with separate panels for Treg-C1, Treg-C2, and Treg-C3 signatures. Dot size reflects the percentage of expressing cells, and color intensity reflects average expression. D Correlation matrix scatter plots comparing expression patterns among Treg subclusters. E Volcano plots showing differentially expressed genes between Treg subclusters. F Heatmap showing scaled expression of Treg-associated genes across the indicated cell populations Regulatory T Cell Subpopulation Analysis and Functional Heterogeneity. A UMAP projections showing three Treg subpopulations designated Treg-C1, Treg-C2, and Treg-C3. The labels are descriptive cluster names rather than fixed functional identities. B Bar chart comparing module scores across Treg-C1, Treg-C2, and Treg-C3. Module scores were calculated from Treg subcluster marker gene sets using a Seurat-style module-scoring approach, and higher values indicate stronger relative enrichment of the corresponding transcriptional program. C Dot plot showing expression patterns of marker genes across Tregs, proliferating cells, macrophages, and CD8 + T cells, with separate panels for Treg-C1, Treg-C2, and Treg-C3 signatures. Dot size reflects the percentage of expressing cells, and color intensity reflects average expression. D Correlation matrix scatter plots comparing expression patterns among Treg subclusters. E Volcano plots showing differentially expressed genes between Treg subclusters. F Heatmap showing scaled expression of Treg-associated genes across the indicated cell populations PTGDR-associated expression programs in the tumor-immune microenvironment PTGDR analysis suggested that this molecule is associated with immune-related transcriptional programs within the tumor microenvironment. Correlation and co-expression analyses identified PTGDR-associated genes including ZC3H12A, MCL1, BHLHE40, ID2, DUSP5, IFNGR1, KLRB1, CRIM1, CCL5, CST7, GZMA, DUSP2, CD69, KLF2, and SELL (Fig. 6A-B). Differential-expression analysis comparing PTGDR-high and PTGDR-low cells identified extensive transcriptional differences (Fig. 6C). GO and KEGG enrichment analyses linked PTGDR-associated genes to leukocyte adhesion, lymphocyte differentiation, regulation of T-cell activation, cytokine signaling, and chemokine signaling pathways (Fig. 6D-E). The distribution plot further suggested that PTGDR expression was enriched in Tregs and present in selected additional immune-associated populations (Fig. 6F). Fig. 6PTGDR-Associated Expression Programs and Functional Enrichment. A Correlation heatmap showing co-expression patterns of PTGDR-associated genes, with hierarchical clustering organizing genes into correlated groups. Notable genes include ZC3H12A, MCL1, BHLHE40, ID2, DUSP5, IFNGR1, KLRB1, CRIM1, PTGDR, CCL5, CST7, GZMA, DUSP2, CD69, and others. B Scatter plot matrix showing pairwise correlations between PTGDR and key genes. C Volcano plot comparing PTGDR-high versus PTGDR-low cells. D Bubble plot showing GO enrichment analysis for PTGDR-related genes, highlighting leukocyte adhesion, lymphocyte differentiation, and regulation of T-cell activation. E Bubble plot showing KEGG pathway enrichment, including cytokine-cytokine receptor interaction and chemokine signaling pathways. F Alluvial flow diagram showing cell type to PTGDR expression level distribution PTGDR-Associated Expression Programs and Functional Enrichment. A Correlation heatmap showing co-expression patterns of PTGDR-associated genes, with hierarchical clustering organizing genes into correlated groups. Notable genes include ZC3H12A, MCL1, BHLHE40, ID2, DUSP5, IFNGR1, KLRB1, CRIM1, PTGDR, CCL5, CST7, GZMA, DUSP2, CD69, and others. B Scatter plot matrix showing pairwise correlations between PTGDR and key genes. C Volcano plot comparing PTGDR-high versus PTGDR-low cells. D Bubble plot showing GO enrichment analysis for PTGDR-related genes, highlighting leukocyte adhesion, lymphocyte differentiation, and regulation of T-cell activation. E Bubble plot showing KEGG pathway enrichment, including cytokine-cytokine receptor interaction and chemokine signaling pathways. F Alluvial flow diagram showing cell type to PTGDR expression level distribution PTGDR-related markers and additional state-associated gene expression Additional marker analyses revealed distinct expression patterns across the major cell populations. Violin plots demonstrated that PTGDR and GZMA were enriched in CD8 + T cells, whereas KLRB1 and KLF2 showed broader expression across multiple cell types (Fig. 7A). Heatmap analysis of selected state-associated genes further demonstrated cell-type-specific expression patterns, with Tregs showing elevated expression of several immune regulatory genes including PTGDR, CD247, SPIB, TRAC, CARD11, and SKAP1 (Fig. 7B). Three-dimensional visualization and trajectory-associated plots provided complementary views of cellular organization and state variation within the analyzed dataset (Fig. 7C-F). Fig. 7PTGDR-Related Markers and Additional State-Associated Gene Expression. A Violin plots showing expression levels of PTGDR, GZMA, KLRB1, and KLF2 across CD8 + T cells, macrophages, proliferating cells, and Tregs, demonstrating cell type-specific expression patterns. B Hierarchical clustering heatmap of selected state-associated genes across cell types. C Three-dimensional visualization showing spatial distribution of the four cell types in a 3D coordinate system. D-F Trajectory-associated plots providing complementary views of inferred cellular-state variation across the analyzed populations PTGDR-Related Markers and Additional State-Associated Gene Expression. A Violin plots showing expression levels of PTGDR, GZMA, KLRB1, and KLF2 across CD8 + T cells, macrophages, proliferating cells, and Tregs, demonstrating cell type-specific expression patterns. B Hierarchical clustering heatmap of selected state-associated genes across cell types. C Three-dimensional visualization showing spatial distribution of the four cell types in a 3D coordinate system. D-F Trajectory-associated plots providing complementary views of inferred cellular-state variation across the analyzed populations Trajectory analysis reveals dynamic T cell state transitions Comprehensive trajectory analysis suggested dynamic transitions of cellular states within the tumor microenvironment. Partition plots and pseudotime density distributions showed partially overlapping but distinct state profiles across CD8 + T cells, macrophages, proliferating cells, and Tregs (Fig. 8A-B). Analysis of genes such as ISG15, NOCA, SDF4, TNFRSF18, TNFRSF4, and UBE2J2 demonstrated state-associated expression dynamics (Fig. 8C-E). Branch analysis further suggested divergent transcriptional programs across inferred trajectories (Fig. 8F), although these results should be interpreted as exploratory rather than definitive lineage relationships. Fig. 8Trajectory Analysis and Gene Expression Dynamics Across Cell States. A Monocle trajectory plot generated after DDRTree reduction and orderCells, with cells colored by annotated cell type. B Pseudotime density plot showing the distribution of cell types along the inferred trajectory. C-E Gene expression plots showing state-associated expression patterns for NOC2L, ISG15, SDF4, TNFRSF18, TNFRSF4, and UBE2J2. F Branch-specific gene-expression analysis generated using BEAM with branch_point = 1. Branches were selected from the learned Monocle trajectory structure. These results indicate exploratory state variation and should not be interpreted as confirmed lineage relationships Trajectory Analysis and Gene Expression Dynamics Across Cell States. A Monocle trajectory plot generated after DDRTree reduction and orderCells, with cells colored by annotated cell type. B Pseudotime density plot showing the distribution of cell types along the inferred trajectory. C-E Gene expression plots showing state-associated expression patterns for NOC2L, ISG15, SDF4, TNFRSF18, TNFRSF4, and UBE2J2. F Branch-specific gene-expression analysis generated using BEAM with branch_point = 1. Branches were selected from the learned Monocle trajectory structure. These results indicate exploratory state variation and should not be interpreted as confirmed lineage relationships qRT-PCR validation confirms differential expression of key genes in HCC cell lines To validate the expression patterns of key genes identified through single-cell RNA sequencing analysis, we performed quantitative real-time PCR on two hepatocellular carcinoma cell lines obtained from ATCC: HepG2 and Huh7. The expression levels of four critical genes—MIF, CD74, CXCR4, and PTGDR—were quantified and normalized to GAPDH as the reference gene (Fig. 9). In HepG2 cells, MIF showed relative expression of 2.5 ± 0.25, CD74 exhibited 3.2 ± 0.32, CXCR4 demonstrated 1.8 ± 0.18, and PTGDR displayed the highest expression at 4.1 ± 0.41. Comparatively, in Huh7 cells, MIF expression was measured at 1.9 ± 0.19, CD74 at 2.4 ± 0.24, CXCR4 at 2.6 ± 0.26, and PTGDR at 3.5 ± 0.35. Statistical analysis revealed significant differences in gene expression between the two cell lines. Specifically, MIF and CD74 showed significantly higher expression in HepG2 cells compared to Huh7 cells (p < 0.01 for both genes), while CXCR4 exhibited significantly higher expression in Huh7 cells (p < 0.001). PTGDR demonstrated elevated expression in both cell lines, with HepG2 showing marginally higher levels (p < 0.05). These results support the detectable expression of MIF-axis-related genes and PTGDR in HCC cell lines; however, they should be interpreted as limited expression-level validation and not as direct confirmation of cell-cell communication or causal immune-regulatory mechanisms. The differential expression patterns between HepG2 and Huh7 cells suggest molecular heterogeneity across HCC cell line models (Fig. 9). Fig. 9qRT-PCR Validation of Key Gene Expression in Hepatocellular Carcinoma Cell Lines. Quantitative real-time PCR analysis showing relative mRNA expression levels of MIF, CD74, CXCR4, and PTGDR in two HCC cell lines (HepG2 and Huh7). Bar graphs display mean ± SEM from three independent experiments, and gene expression was normalized to GAPDH. Statistical significance between cell lines is indicated as *p < 0.05, **p < 0.01, and ***p < 0.001 using Student’s t-test. These results support the detectable expression of MIF-axis-related genes and PTGDR in HCC cell line models. However, the qRT-PCR data should be interpreted as limited transcript-level validation and not as direct evidence of cell-cell communication, protein-level activity, or causal immune regulation qRT-PCR Validation of Key Gene Expression in Hepatocellular Carcinoma Cell Lines. Quantitative real-time PCR analysis showing relative mRNA expression levels of MIF, CD74, CXCR4, and PTGDR in two HCC cell lines (HepG2 and Huh7). Bar graphs display mean ± SEM from three independent experiments, and gene expression was normalized to GAPDH. Statistical significance between cell lines is indicated as *p < 0.05, **p < 0.01, and ***p < 0.001 using Student’s t-test. These results support the detectable expression of MIF-axis-related genes and PTGDR in HCC cell line models. However, the qRT-PCR data should be interpreted as limited transcript-level validation and not as direct evidence of cell-cell communication, protein-level activity, or causal immune regulation Discussion In this study, we reanalyzed a public HCC single-cell dataset to characterize immune-associated interaction patterns, with particular attention to T-cell dysfunction and regulatory immune states. The revised analysis now reports the key computational parameters more explicitly, including the Seurat filtering thresholds, clustering resolution, CellChat settings, hdWGCNA workflow, and Monocle trajectory procedure. Overall, the data support marked immune heterogeneity in the HCC microenvironment and highlight the MIF-CD74/CXCR4/CD44 and PTGDR axes as candidate immunoregulatory programs rather than validated therapeutic targets. This cautious positioning is important because the present evidence is mainly transcriptome-based [33, 34]. Our results suggest that macrophage migration inhibitory factor (MIF) may act as a prominent predicted signaling hub among tumor-associated and immune-associated cell populations. Mechanistically, this finding is plausible but should still be treated carefully. MIF-CD74 signaling can participate in inflammatory and survival-related pathways, CXCR4 may contribute to immune-cell recruitment and retention, and CD44 may help stabilize the MIF-CD74 receptor complex and support adhesion-related signaling. In a Treg-CD8 + T-cell context, these routes could help maintain a suppressive niche in which regulatory programs are reinforced and cytotoxic T-cell activity is restrained. Therefore, the MIF-CD74/CXCR4/CD44 axis is best interpreted as a candidate communication program connecting myeloid, regulatory, and cytotoxic compartments, not as direct proof of a causal immunosuppressive pathway [35–37]. PTGDR may reflect a prostaglandin-associated immune regulatory state within the HCC microenvironment. In the present analysis, PTGDR-related genes were linked to cytokine signaling, leukocyte adhesion, lymphocyte differentiation, and T-cell activation. This pattern suggests that PTGDR may be part of a broader immune-state remodeling program rather than a single isolated marker. Its enrichment in selected Treg-related populations also raises the possibility that PTGDR-positive cells represent a distinct regulatory state. Even so, this conclusion remains preliminary. Protein-level detection, spatial localization, and functional perturbation are still needed before PTGDR or MIF-related pathways can be discussed as actionable therapeutic targets [38–40]. The HCC microenvironment is shaped by more than one immune axis. Recent studies show that angiogenesis, immune checkpoint regulation, RNA-binding protein activity, lactylation-related epigenetic regulation, and stromal remodeling can all reshape tumor behavior. For example, RBM28 has been reported to enhance angiogenesis by improving STAT3 translation in HCC, RBM30 can activate STAT1-related PD-L1 transcription and promote immune evasion, and ASH2L-K312 lactylation has been linked to VEGFA-related angiogenesis and malignant progression. These studies support a more layered view of HCC, where immune dysfunction, vascular remodeling, and tumor-intrinsic regulation may work together. In this context, the MIF/PTGDR programs identified here should be seen as one part of a wider tumor microenvironment network [41–45]. Limitations Several limitations should be acknowledged. First, this study was based on retrospective reanalysis of public scRNA-seq data, and the representativeness of the dataset depends on the original sampling design and metadata completeness. Second, transcriptome-based cell-cell communication analysis estimates interaction probability from ligand-receptor expression and does not directly demonstrate physical ligand-receptor binding or causal signaling. Third, Monocle trajectory inference in a mixed tumor microenvironment dataset should be interpreted cautiously because inferred state transitions may be affected by cell composition, sampling structure, and the selected ordering genes. Fourth, although qRT-PCR was used to examine MIF, CD74, CXCR4, and PTGDR expression in HCC cell lines, this validation only supports transcript-level detectability in simplified in vitro models. It does not confirm protein abundance, spatial localization, ligand-receptor interaction, or functional immune regulation. Fifth, the present study lacks Western blotting, immunohistochemistry, multiplex immunofluorescence, spatial transcriptomics, co-culture assays, and perturbation experiments. Therefore, the MIF/PTGDR-related findings should be regarded as candidate mechanisms requiring further validation [46, 47]. Future studies should validate the MIF/PTGDR-related findings in independent clinical cohorts and experimental HCC models. Useful next steps include Western blotting or flow cytometry for protein-level detection, immunohistochemistry or multiplex immunofluorescence for spatial localization, and spatial transcriptomics to map ligand-receptor relationships within tissue architecture. Co-culture assays and knockdown or overexpression of MIF, CD74, CXCR4, CD44, and PTGDR would help test whether these genes directly affect Treg activity, CD8 + T-cell dysfunction, or cytokine production. Combination strategies involving MIF/PTGDR-related pathways may be worth exploring, but only after stronger mechanistic evidence is available [43, 48]. Conclusion In conclusion, this study characterized interaction patterns among immune-associated cell populations in hepatocellular carcinoma through single-cell transcriptomic reanalysis. We identified substantial heterogeneity within the tumor immune microenvironment, defined three Treg subclusters, and highlighted MIF-CD74/CXCR4/CD44 and PTGDR signaling as candidate immunoregulatory programs. These findings provide a basis for future mechanistic studies, biomarker development, and translational validation, but they should not be interpreted as direct confirmation of therapeutic targets or causal immune-evasion mechanisms.