Unraveling CD4⁺ T cell heterogeneity and cell death-associated genes in high-grade serous ovarian cancer: a comprehensive analysis of single-cell RNA and spatial transcriptome sequencing

Abstract

Background

CD4⁺ T cells infiltrate the tumor microenvironment, but their role in high-grade serous ovarian cancer (HGSOC) progression remains unclear. Cell death-associated genes critically impact HGSOC advancement.

Methods

We integrated single-cell and spatial transcriptomics to analyze CD4⁺ T cell heterogeneity and identify cell death-associated genes. Key candidates were experimentally validated via qPCR and Western blotting in independent HGSOC cohorts. Furthermore, a T cell–tumor cell co-culture system was employed for functional validation of candidate gene effects on tumor behavior.

Results

To comprehensively characterize the heterogeneity of CD4⁺ T cells in HGSOC, we performed an integrated analysis including inferCNV, pseudotime trajectory, SCENIC transcription regulatory networks, and hdWGCNA. This multi-omics approach defined distinct CD4⁺ T cell subgroups and identified key regulons. Notably, a co-expression network module derived from hdWGCNA was leveraged to construct a prognostic model, which demonstrated significant predictive power for patient survival. Spatial transcriptomic analysis identified co-localized niches of IL7R⁺CD4⁺ T cells and Tregs, exhibiting highly concordant gene expression with scRNA-seq data. Orthogonal validation confirmed significant dysregulation of MAL (p < 0.001) and ANXA1 (p < 0.001) in HGSOC. Critically, in vitro co-culture experiments demonstrated that silencing MAL in T cells significantly suppressed ovarian cancer cell proliferation, migration, and invasion, highlighting its functional role in modulating the tumor microenvironment.

Conclusion

Our findings elucidate mechanisms of cell death-associated genes in CD4⁺ T cell subgroups, with experimentally grounded biomarkers offering therapeutic options for HGSOC.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12094-026-04266-z.

Introduction

High-grade serous ovarian cancer (HGSOC) has limited early symptoms and is often diagnosed at a terminal stage to become one of the most lethal and prognostically unfavorable gynecological cancers [1]. Surgical debulking and platinum chemotherapy are the standard of therapy for HGSOC [2]. The presence of residual cancer cells and the development of resistance to chemotherapy contribute to cancer recurrence, with 5-year survival rates of 41% and 20% for patients in FIGO stages III and IV [3]. Notwithstanding progress in the treatment of HGSOC, the 5-year survival rates for the majority of patients continue to be below the desired level owing to the emergence of resistance and the adverse effects of therapeutic agents.

The molecular diversity of tumor subclones is widely considered a core basis of tumor heterogeneity, driving tumor progression, metastasis, and drug resistance formation through the aberrant activation or inhibition of key oncogenes and signaling pathways [4, 5]. However, increasing evidence suggests that genomic or transcriptomic alterations within tumor cells themselves are insufficient to fully explain the complexity of tumor behavior [6]. Immune cells in the tumor microenvironment (TME) exert continuous selective pressure on the survival, proliferation, and therapeutic sensitivity of tumor cells through cell–cell interactions, cytokine secretion, and immune regulatory networks [7, 8]. These processes, especially those mediated by CD4⁺ T cells, can profoundly shape tumor heterogeneity. Regulatory T (Treg) cells, an important immunosuppressive subtype of CD4⁺ T cells, weaken anti-tumor immune surveillance by suppressing immune responses [9]. Treg cells show significant infiltration in various solid tumors, including HGSOC. Currently, research primarily focuses on changes in Treg cell abundance or their overall immunosuppressive effects, while a systematic analysis of the functional heterogeneity within CD4⁺ T cells and their potential differentiation trajectories is still lacking.

Cell death is an essential biological process that works crucially in preserving tissue balance and controlling many physiological activities. Dysregulation of cell death can lead to diseases marked by uncontrolled proliferation, most notably cancer [10]. Immunogenic cell death (ICD) is a type of cell death that can actively provoke an immunological response and is of great importance in the field of cancer treatment. In anti-PD-1 immunotherapy, the anti-cancer immune response triggered by ICD can augment the therapeutic efficacy of both anti-PD-1 immunotherapy and traditional anticancer chemotherapies [11, 12]. Nevertheless, the intricate molecular mechanisms governing ICD remain incompletely elucidated and necessitate additional research for a comprehensive understanding.

In this study, we analyzed 31,403 cells from HGSOC samples to construct single-cell transcriptome atlases, revealing the roles of two subsets of CD4⁺ T cells in HGSOC. We used the CD4⁺ T cell population to identify genes potentially critical for diverse cell types in the tumor microenvironment. Furthermore, we identified cell death-related genes that could serve as predictive biomarkers for the efficacy of chemotherapy and anti-PD-1 immunotherapy. To functionally validate these T cell-derived candidates, we performed in vitro experiments to assess their roles in ovarian cancer. The results of this study provide new mechanistic clues for understanding the interaction between the immune microenvironment and tumor cell phenotype in HGSOC and lay a theoretical foundation for developing precision treatment strategies.

Methods

Data acquisition and preprocessing

ScRNA-seq data of GSE184880 was gathered from the GEO (https://www.ncbi.nlm.nih.gov/geo/) database, including five samples of normal ovarian tissues and seven HGSOC tissues. For scRNA-seq data, the single-cell expression matrix quality control was tested using Seurat (version 4.3.0.1). The screening criteria were that each single cell expressed at least 2000 genes and a Unique Molecular Identifier (UMI) > 100. Ultimately, 31,403 cells from 8 samples (4 normal ovarian tissue samples and 4 HGSOC tissue samples) were retained.

The dimensionality of the scRNA-Seq dataset was reduced by principal component analysis (PCA). The core cell clusters were determined using the FindClusters function of Seurat, and the resolution was set as default (res = 0.6). The findings were visualized using UMAP plots. The Seurat Findallmarker function was utilized to screen preferentially expressed genes in clusters or DEGs between tumor cells and normal cells. Spatial transcriptomic data of one patient with HGSOC (CytAssist_11mm_FFPE_Human_Ovarian_Carcinoma) were acquired from 10 × genomics (https://www.10xgenomics.com/resources/datasets). Seurat package (version 4.3.0.1) is also utilized for analyzing ST data [13].

Estimation of copy number variations in cancer cells

In the single-cell RNA-seq data preprocessing stage, we annotated cells based on classic T-cell marker genes (CD3D, CD4, CD8A) while excluding cells expressing epithelial/tumor markers (EPCAM, KRT8, KRT18, MUC16). Only cells clearly marked as CD4⁺ T cells and not expressing epithelial/tumor-specific genes were retained for CNV analysis. The InferCNV package (version 1.18.1) was utilized with default parameters to detect CNVs in major cell types and CD4⁺T cell subgroups to recognize real cancer cells. Endothelial cells were deemed the control group.

CD4 + T cell subclustering workflow  

1. Cell subset extraction: CD4⁺ T cells were isolated from the global dataset using pre-annotated cell identities (cell_type = ’CD4T’).

2. Feature selection: the top 3000 highly variable genes were identified using the variance-stabilizing transformation (vst) method implemented in Seurat v4.3.0.1.

3. Data scaling: expression matrices were scaled using Z-score normalization via ScaleData with default parameters.

4. Dimensionality reduction: Principal component analysis (PCA) was performed using the top 3000 variable genes, retaining the first 20 principal components (PCs) capturing > 85% cumulative variance.

5. Neighborhood graph construction: A k-nearest neighbor (KNN) graph was built based on Euclidean distances in PCA space (dims = 1–20).

6. Unsupervised clustering: Cell clusters were identified using the Louvain algorithm at resolution 0.6 via FindClusters.

7. Batch effect correction: Technical variability across sample groups was mitigated using Harmony (v0.1.1) with default parameters (group.by.vars = "group").

Trajectory analysis

Using the R package Slingshot (version 2.10.0) [14], pseudotime analysis was performed to map differentiation in the TME and confirm the dramatic translational relationships among cell types.

Simultaneous gene regulatory network analysis

A computational method, SCENIC (version 1.3.1) [15] is applied for constructing regulatory networks and identifying various cell states from scRNA-seq data. SCENIC was performed on all single cells to compare the difference between cell clusters according to transcription factors or their target genes, and the preferentially expressed regulons were estimated by the Limma (version 3.56.2) package [16].

Cell–cell communication analysis with Cellchat

To obtain an integrated analysis of cell–cell communication molecules, the authors conducted cell–cell communication analysis by CellChat v1 [17] and CellChat v2 (version 1.6.1) [18]. The expression of ligand and receptor genes in each cell was projected into a reference communication network with an artificial selection, and the probability of communication in each pathway was close to gene expression. We validated the communication probability at a statistical level between subclones and other non-malignant cells, even the communication between different subclones.

hdWGCNA analysis

All genes of T cells in GSE184880 were included for hdWGCNA [19] (‘hdWGCNA’ package, version 0.2.20). This was done to identify the gene sets of interest and guarantee the inclusion of all relevant information. WGCNA was constructed based on the most optimal value. A screening process was used to identify and group co-expression modules according to their similarity.

Functional and pathway enrichment analysis

The R’clusterProfiler’ package (version 4.10.0) was used to screen functional annotations and interpretations of DEGs or marker genes of cell types [20], involved in GO and KEGG analyses. Fisher’s exact test is capable of identifying the significant phrases and pathways. A significant enrichment threshold was set at p < 0.05 and FDR < 0.25.

Multimodal intersection analysis (MIA) and SPOTlight analysis

Spatial Transcriptomics data and scRNA-seq data were integrated by MIA [21]. The hypergeometric cumulative distribution confirmed the significance of the overlap between cell-type marker genes and ST genes. We employed the SPOTlight [22] package (version 1.5.2), a robust, accurate, and sensitive tool specifically designed to predict the cell composition of each spot in the spatial transcriptomics (ST) data.

Cell culture

The T cells used in this study were the human Jurkat cell line (derived from acute T-lymphoblastic leukemia, ATCC TIB-152), serving as a CD4⁺ T cell functional model. A2780 (human ovarian cancer cells, ATCC CRL-1572), SKOV3 (human ovarian cancer cells, ATCC HTB-77), and (human ovarian epithelial cells, normal control) were cultured in RPMI-1640 or DMEM medium containing 10% fetal bovine serum (FBS, Gibco) and 1% penicillin–streptomycin (Gibco) at 37 °C in a humidified incubator with 5% CO₂. Transwell cells were used for co-culture in this study. Jurkat cells were seeded in the upper chamber with 1 × 10⁵ cells and A2780 cells in the lower chamber, and cultured in RPMI-1640 medium containing 10% FBS at 37 °C and 5% CO₂ for 48–72 h. Jurkat cells were then subjected to MAL gene knockout (siRNA) or empty vector control treatment, and the knockout efficiency was verified by qPCR and Western blot.

RNA extraction and quantitative real-time PCR

Total RNA was extracted from immortalized ovarian surface epithelial cells (IOSE80, control) and two ovarian cancer cell lines A2780 (epithelial ovarian carcinoma, derived from untreated patients with low metastatic potential) and SKOV3 (serous ovarian carcinoma, ascites-derived with high invasiveness)-using TRIzol reagent (#5,301,100, Xinjing Bio, China) followed by phase separation with Buffer EX (#9,025,100, Xinjing Bio). RNA precipitation was performed with isopropanol (Baisi Chemical, China) and washing with 75% ethanol. RNA integrity was verified by spectrophotometry prior to reverse transcription using the SureScript RT Kit (#QP056, GeneCopoeia, USA) in 20 μL reactions containing 1 μg RNA, 5 × RT Buffer, and 20 × RTase Mix under thermal conditions: 25°C/5 min → 42°C/15 min → 85°C/5 min. Quantitative PCR was performed on the SLAN-96S System (Hongshi Medical, China) using 2 × SYBR Green Master Mix (MPC2203026, Servicebio, China) with gene-specific primers: MAL (F: 5’-AGGAC TACGG CTGCT TCTTG-3’, R: 5’-GAGG TAGTC CTGGT GATGG A-3’) and GAPDH (F: 5’-GGAGC GAGAT CCCTC CAAAA T-3’, R: 5’-GGCTG TTGTC ATACT TCTCA TGG-3’). Each 20 μL reaction contained 10 μL master mix, 0.4 μM primers, and 2 μL cDNA. Cycling parameters: 95 °C/30 s initial denaturation; 40 cycles of 95 °C/15 s → 60 °C/30 s. Relative quantification used the 2^(− ΔΔCt) method where ΔΔCt = [(Ct_target − Ct_GAPDH)_sample—(Ct_target − Ct_GAPDH)_control] with IOSE80 as calibrator, and significance determined by Student’s t test (Prism v9.0).

Western blot analysis

Protein lysates were extracted from immortalized ovarian surface epithelial cells (IOSE80), epithelial ovarian carcinoma cells (A2780), and serous ovarian carcinoma cells (SKOV3) using RIPA lysis buffer (#R002, Solarbio) supplemented with 1 mM PMSF (#P0100, Boster), followed by centrifugation at 12,000 g for 15 min at 4 °C (5415D centrifuge, Eppendorf). Protein concentrations were determined via Bradford assay (#AR0145, Boster) on a 721 spectrophotometer (Shanghai Sunny Hengping). Equal amounts of protein (20 μg) were separated on 10% SDS-PAGE gels (#AR0138, Boster) with Tris–glycine-SDS buffer (#AR0139, Boster) at 120 V for 90 min using a DYCZ-24DN electrophoresis system (Liuyi Instrument), then transferred to PVDF membranes (#IPVH00010, Millipore) at 300 mA for 90 min in Tris–glycine transfer buffer (#AR0141, Boster) with a VE-386 transfer tank (Yuanpinghao). Membranes were blocked with 5% BSA-TBS (#AR0187, Boster) for 1 h, then incubated overnight at 4 °C with primary antibodies: anti-ANXA1 (1:1000; #DF7586, Affinity) or anti-β-actin (1:5000; #ab6276, Abcam) in Western antibody diluent (#AR10117, Boster), followed by 1-h incubation with HRP-conjugated goat anti-rabbit IgG (1:5000; #ab7090, Abcam). Protein bands were visualized using Western Lightning Chemiluminescence Reagent (#NEL10300EA, PerkinElmer) on a FluorChem FC2 imaging system (Alpha Innotech), and quantified by densitometry in ImageJ (v1.53t) with β-actin normalization.

Cell proliferation assay

Cells were seeded into 96-well plates, with 100 μL of cell suspension added to each well, adjusting the density to 1000 cells/well. Cell proliferation was measured at 0 h and 24 h using a cell counting kit-8 (CCK-8, Boster Biotechnology, Wuhan, Shanghai). Absorbance of each well was measured at 450 nm using a microplate reader to reflect cell proliferation levels.

Flow cytometry assay for apoptosis

After digesting cells with 0.25% trypsin, approximately 5 × 10⁵ cells were collected from each group, centrifuged at 1000 rpm for 5 min, and the supernatant was discarded. Resuspend cells in 500 μL 1 × Binding Buffer, add 5 μL Annexin V-FITC (AP101-100-kit, Hangzhou Lianke Biotechnology), and 10 μL PI, mix gently, and incubate in the dark for 10–20 min in an ice bath. During incubation, gently pipette 2–3 times to improve staining. After staining, detect the proportion of apoptotic cells using flow cytometry.

Transwell migration assay

Wash each group of cells twice with PBS, digest with trypsin–EDTA, then stop digestion with complete culture medium and mix thoroughly by pipetting. After counting, take 1 × 10⁴ cells, centrifuge at 1000 rpm for 5 min, discard the supernatant, resuspend the cells in 200 μL serum-free culture medium, and add to the upper chamber of a Transwell apparatus (TCS020024, Guangzhou Jiete Biofiltration). Add 1 mL of complete culture medium containing the drug to the lower chamber and incubate at 37 °C for 24 h. After incubation, wash twice with PBS, fix with 4% paraformaldehyde for 30 min, stain with 1% crystal violet for 15 min, wash again with PBS, and air-dry the membrane. Observe and count the number of migrating cells under a microscope (20 ×) on a glass slide.

Scratch healing assay

Seed cells from each group into 6-well plates and culture until the monolayer is nearly confluent. Use the tip of a 200 μL pipette to make a straight scratch along the fixation reference line, keeping it as vertical as possible and avoiding tilting. Gently wash three times with PBS to remove detached cells, and add serum-free culture medium. During culture, photograph and record the scratch healing process at 0 h and 24 h.

Results

An expression atlas of HGSOC by scRNA-seq

To determine the heterogeneity and diversity of high-grade serous ovarian cancer (HGSOC), we downloaded scRNA-seq data from seven HGSOC samples and five adjacent non-cancerous tissue samples from the GEO database (Fig. 1) [23]. After quality control, we ultimately obtained 31,403 cells from eight samples (4 HGSOC and 4 adjacent non-cancerous tissue samples). The ovarian tissue cell atlas was outlined through UMAP plots (Fig. 2A). Cell types were identified based on cross-referencing cluster-specific genes with previously identified indicators of cell populations. According to the UMAP plots, the signature genes were highly expressed in their corresponding cell types (Fig. 2B). The DotPlot revealed that seven cell types were designated by marker genes, such as CD4⁺ T cells (CD4, CTLA4, FOXP3, IL7R), B cells (CD19, MS4A1, CD79A, CD38, SDC1), CD8⁺ T cells (CD3E, CD3D, CD8A, NKG7, MK167, GZMB, GZMK), endothelial cells (PECAM1, EGFL7, EMCN, CDH5, VWF), fibroblasts (DCN, COL1A1, COL1A2, ACTA2, PDGFRA, POSTN), epithelial cells (EPCAM, KRT8, KRT18, KRT19), and myeloid cells (FCGR3A, CD68, LYZ, CD14, CD83, S100A8, MNDA) (Fig. 2C).

Fig. 1.

Flowchart depicting the design of this research study

Fig. 2.

Identification of high-grade serous ovarian cancer (HGSOC) cells by single-cell RNA-seq. A The UMAP plot of the major cell types of scRNA-seq data. Area total of 31,403 cells were obtained after filtration from eight samples. Cell types were annotated according to their differential genes. B Expression levels of representative marker genes across the cell types in HOSOC. The color key from gray to red indicates relative expression levels from low to high. C Dotmap shows marker gene expression levels in each cell cluster. D, E Histography of the proportion of different tissue types in HGSOC; complete quantitative data are available in Supplementary Table S1. F Heatmap depicting the activation levels of various signaling pathways across seven cell types. The color intensity corresponds to the degree of pathway activation, offering valuable insights into the dynamic signaling landscape within the cellular microenvironment

Next, the composition of the samples was investigated to determine the dominant cell types among the tumor samples, containing fibroblasts, CD8⁺ T cells, myeloid cells, and epithelial cells. As Fig. 2D, there was a remarkable rise in the proportion of fibroblasts and endothelial cells in tumor samples in comparison with normal samples. Tumors exhibit higher percentages of epithelium, CD8⁺ T cells, and myeloid cells than normal samples. Single-cell profiling revealed profound shifts in the ovarian ecosystem upon malignant transformation. All 7 annotated cell types exhibited differential abundance between normal (n = 16,690 cells) and tumor tissues (n = 14,713 cells), with CD4⁺ T cells and CD8⁺ T cells showing dramatic expansion (Fig. 2D–E, Supplementary Table S1). Additionally, a heatmap was used to identify the expression levels of several important signaling pathways for each cell (Fig. 2F). Considering these results, we conducted further analyses on epithelial cells and CD4⁺ T cells to investigate their roles in tumorigenesis and cancer progression.

The aberrant expression of genes correlates with metastasis of HGSOC

The HGSOC cells are highly heterogeneous, which is attributed to the patterns associated with copy number variations (CNVs). According to previous studies, HGSOC originates from epithelial cells, so we conducted inferCNV analysis to examine the CNV of each cell population with reference cells of endothelial cells (Fig. 3A). We identified typical HGSOC genomic alterations, and epithelial cells had a larger amount of chromosome CNVs than other cell types. Across the CNV heatmap, a darker red color indicated a higher proportion of chromosome amplification; the blue color is darker with higher levels of chromosome loss. The heatmap revealed a gain of epithelial cells in chromosome 8,11,19,20, and chromosome 6 gain of CD4⁺T cells, CD8⁺T cells, and myeloid cells. Subsequently, we calculated the CNV scores for these six cell populations, revealing that CD8⁺T cells, epithelial cells, and CD4⁺T cells exhibited the highest scores (Fig. 3B).

Fig. 3.

Heterogeneity and copy-number variation (CNV) analysis of high-grade serous ovarian cancer (HGSOC) cells. A The hierarchical heatmap shows large-scale CNVs in six major cell types. B The boxplot illustrates the CNV scores of the six major cell types. Epithelial cells exhibit the highest CNV scores, followed by CD8⁺T and CD4⁺ T cells. C Thirteen main CD4⁺T subgroups were identified using Uniform Manifold Approximation and Projection analysis. D The heatmap displays the average expression levels of the top ten differentially expressed genes among seven subgroups. The color legend represents the normalized gene expression levels across the subgroups. E Gene Set Expression Analysis DotPlot for ontology gene sets in the MSigDB database among thirteen CD4⁺T subclusters. F The hierarchical heatmap shows large-scale CNVs in thirteen CD4⁺T cell subgroups. G The boxplot illustrates the CNV scores of the thirteen CD4⁺T cell subgroups. Cluster 1 exhibits the highest CNV scores, followed by clusters 2 and 7

CD4⁺ T cells laid a vital impact in defining the TME and regulating the lymphocyte immune response to tumors by producing cytokines and undergoing differentiation into specific subgroups. Reclustering of 3104 CD4⁺ T cells (2,136 tumor vs. 968 normal) identified thirteen subgroups (Fig. 3C). Complete quantitative metrics, including cluster-specific marker genes and inter-group distributions, are detailed in Supplementary Table S2. Based on the heatmap, the top 10 marker genes in each subgroup of 13 subtypes (Fig. 3D) showed that IGKC, GZMB, LAG3, XAF1, CCL3, IGLC2, and GNIY were highly expressed in cluster 1. We found that MALAT1, STK4, HSPA1B, MIAT, and TC2N were highly expressed in cluster 2, while FOSB, JUN, and FOS had high-expressed levels in cluster 7. GSEA indicated that abnormal joint morphology, endoplasmic reticulum lumen, and collagen-containing extracellular matrix signaling pathways were highly activated in CD4⁺T cell subgroups (Fig. 3E). As a chemokine, CCL3 is a key factor influencing the migration and maintenance of relevant immune cells within infected tissues or TME. Nuclear factor (NF)-κB is constitutively activated, facilitating the expression of inflammation-associated chemokines in the TME, including CCL3 and CXCR4. Hence, clusters 0 and 1 exhibiting high expression of CCL3 and CXCR4 may possess the capability to promote inflammation. The DotPlot also showed that ribonucleoprotein complex, cytosolic ribosome, cytoplasmic translation, and structural constituents of ribosome signaling pathways were suppressed in CD4⁺T cell subgroups.

To further validate the occurrence of malignant cells in the thirteen CD4⁺T cell subgroups, we conducted inferCNV analysis (Fig. 3F). In our observations, amplifications were noted in chromosomes 17, 21, 22, and 2 in Cluster 1, while chromosome 3 was amplified in Cluster 2. We conducted calculations of CNV scores for thirteen subgroups, and the highest scores were observed for clusters 1, 2, and 7 (Fig. 3G). Therefore, cells in CD4⁺ T subgroups 1, 2, and 7 show CNV alterations that could be associated with their functional modulation in the tumor microenvironment.

Dynamics of IL7R⁺ CD4⁺T cells and Treg cells in HGSOC

We reclustered CD4⁺ T cells and identified two distinct subtypes, namely IL7R⁺CD4⁺ T cells and Treg cells (Fig. 4A). Through differential gene expression analysis, we screened eight cluster-specific marker genes and could distinguish between IL7R⁺CD4⁺ T cells (CD4, CCR7, GZMK, and IL7R), and Treg cells (FOXP3, CTLA4, TIGIT and GZMB) (Fig. 4B). Trajectory analysis using the Slingshot package in R was performed to investigate the transition of IL7R⁺CD4⁺ T cells and Treg cells toward exhaustion. The investigation revealed that IL7R⁺CD4⁺ T cells existed in medium and late stages, and Treg cells in early stages of HGSOC (Fig. 4C). However, the results of trajectory analysis by using CytoTRACE v0.3.3 did not appear to be entirely consistent between IL7R⁺CD4⁺ T cells and Treg cells [24] (Fig. 4D). Treg cells exhibited enhanced expression of BATF, TNFRSN18, GBP5, STAT3, VIM, and TIGIT (Fig. 4E). Moreover, the trajectory analysis (BATF, TNFRSF18, GBP5, TIGIT) showed that the genes were generated at the early stage of Treg cells differentiation, and their expression decreased over time. Analyses of functional enrichment were performed on the top 100 differentially expressed genes (DEGs) between Treg cells and IL7R⁺CD4⁺T cells (Fig. 4F, G). Treg cells were related to cytoplasmic translation, T cell differentiation, ribosome biogenesis, lymphocyte differentiation, regulation of T cell activation (Fig. 4F). KEGG analysis revealed the enrichment of antigen processing, and presentation, phagosome, PD-L1/PD-1 pathway in cancer, Th17 cell differentiation (Fig. 4G). Moreover, the expression of BATF was significantly altered between tumor and normal tissue (Fig. 4H). Basic leucine zipper ATF-like transcription factor (BATF), as a transcription factor of the AP-1 family that functions in the differentiation of T-helper 17 cells (Th17). In the TME of ovarian cancer, Th17 cells play a pathogenic role and produce several cytokines that can promote angiogenesis and cell proliferation [25, 26].

Fig. 4.

Identification of CD4⁺ T cell subgroups. A A UMAP plot of CD4 + T subgroups, including IL7R⁺CD4⁺ T cells and Treg cells. B Levels of the expression of subgroup marker genes across the cell types. The color key from gray to red indicates relative expression levels from low to high. C Pseudotime analysis of SlingShot showing CD4⁺T subgroups and state dynamics. Red to blue colors represent the pseudotime order. D Pseudotime analysis of cytoTRACE showing CD4 + T subgroups and state dynamics. Red represents less differentiation, and blue represents more differentiation. E Six representative genes with different expression patterns in the process of CD4⁺T cell subgroup differentiation: BATF, VIM, GBP5, TNFRSF18, TIGIT, and STAT3. F Gene Ontology analysis of top 50 state-specific upregulated genes along the pseudotime. G Kyoto Encyclopedia of Genes and Genomes analysis of top 50 state-specific upregulated genes along the pseudotime. H Boxplot of the BATF expression between tumor and normal samples by using GEPIA2. I Violin plot of the SCENIC showing the specificity of transcription factor activation in the five cell types

Through SCENIC analysis, we identified the upstream transcription factors in IL7R⁺CD4⁺ T cells and Treg cells (Fig. 4I). Coincidentally, a high activation of BATF was noted in tumor-infiltrating Treg cells, consistent with the outcomes of Slingshot trajectory analysis. IKZF2, ZNF891, FOXO1, ZNF615, and MXD4 were highly activated in tumor Treg cells. Additionally, MXD4, IRF1, and GATA3 were found to be activated in tumor-infiltrating IL7R⁺CD4⁺ T cells. BATF, an AP-1 family transcription factor, is crucial in controlling the differentiation of lineage-specific cells across the immune system. Interestingly, the depletion of BATF has been demonstrated to improve the antitumor performance of CAR-T cells [27]. BATF has been demonstrated to drive the exhaustion of CAR-T cells by combining and upregulating associated genes. Cancers such as lung cancer, kidney cancer, and melanoma that activate these transcription factors have been associated with poor prognoses [28]. Overall, these analyses have revealed that different developmental subtypes of CD4⁺ T cells in HGSOC demonstrate distinct and heterogeneous expression programs.

Functional and pathway enrichment analysis in hdWGCNA

To investigate the progress of IL7R⁺CD4⁺ T cells and Treg cells in HGSOC and integrate scRNA-seq data of CD4⁺ T cells within a comprehensive analytical framework, we employed the expression profile for hdWGCNA via the R package “hdWGCNA” to develop the gene co-expression networks of HGSOC [19]. All 2271 genes in the gene expression matrix were included in the analysis. In total, five modules associated with HGSOC were identified based on their expression profiles (Fig. 5A and B). We demonstrated a significant correlation between the five co-expression modules and the development of HGSOC, including “M1”, “M2”, “M3”, “M4”, and “M5” (Fig. 5C–D). The genes contained in these five modules are 107, 162, 932, 978, and 92. Subsequently, GO enrichment analysis was utilized to comprehensively acknowledge biological functions, molecular functions, and cellular components involved in the development of tumor cell immune infiltration state based on these genes. Our analysis of the M2 genes showed the significant involvement of these active genes in cytoplasmic translation, ribosome biogenesis, and cellular respiration (Fig. 5E). KEGG pathway analysis of the M4 genes revealed their association with chemical carcinogenesis mediated by diverse signaling pathways in HGSOC, including reactive oxygen species (ROS) related pathways and oxidative phosphorylation (Fig. 5F).

Fig. 5.

Identification of modules associated with the development of high-grade serous ovarian cancer (HGSOC). A Dendrogram clustering of all genes using dissimilarity method and assignment of modules. B The powers for soft-thresholding were chosen; panels on the left show the influence of soft-threshold power on the scale-free topology fit index. Panels on the right show the average network connectivity for various weighting factors. C The UMAP plots of the five co-expression gene modules. D Co-expression plots for modules M1, M2, M3, M4, and M5. E Gene Ontology functional enrichment analysis of modules M1, M2, M3, M4, and M5 in high-dimensional weighted gene co‐expression network analysis (hdWGCNA). F Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis of modules M1, M2, M3, M4, and M5 in hdWGCNA

Construction and verification of the prognostic model based on the M4 module genes of high-dimensional weighted gene co‐expression network analysis

Differential expression analysis was employed using the limma package in R to identify key tumorigenesis-associated genes based on the GSE26712 dataset. It identified 1,436 DEGs (|logFC|≥ 1, p < 0.05), containing 518 upregulated genes and 918 downregulated genes (Fig. 6A). According to the differential expression analysis using the limma package, Cox regression analysis was conducted to find survival-related genes. Through overlap analysis of the upregulated, M4 module genes, we identified that TAGLN2, MAL, TXN, PPP1R14B, CORO1B, LSM4, and PPP1CA were upregulated and belong to M4 module genes (Fig. 6B, D). Subsequently, Cox regression analysis performed on these overlapping genes revealed that MAL (also known as T-cell differentiation protein) exhibited an independent prognostic effect (p < 0.05; Fig. 6D). The expression level of MAL in HGSOC patients is shown in Fig. 6C. The time-dependent ROC curve revealed the AUCs of MAL at 1, 3, and 5 years were 62.00%, 63.58%, and 57.72% respectively (Fig. 6F, left). The Kaplan–Meier (KM) analysis revealed a statistically significant decrease in the overall survival (OS) rate among patients in the high-expression group compared to those in the low-expression group (p = 0.009, Fig. 6F, right). These results thus suggest that MAL elicits moderate performance and holds potential as a prognostic biomarker in HGSOC.

Fig. 6.

Construction and assessment of a prognostic risk model. A Volcano plot depicting differential expression analysis of GSE26712. Blue dots correspond to genes that are downregulated, whereas red dots correspond to genes that are upregulated, logFC = 1. B Overlap analysis between the upregulated genes and module M4 genes. C The difference in the expression of MAL in HGSOC compared to normal samples is shown through Boxplot. D The Forest plot depicting the results of Cox regression analysis using the upregulated genes (TAGLN2, MAL, TXN, PPP1R14B, CORO1B, LSM4, PPP1CA) obtained from the overlap analysis, and MAL was chosen to construct a risk model. E The area under the ROC curves (AUCs) of the risk model for predicting 1, 5, 10, and 12-year survival rates were, respectively, 52.80%, 53.94%, 66.47%, and 69.21% in the TCGA-OV cohort. F The AUCs of the risk model for predicting 1, 3, and 5-year survival rates were 62%, 63.58%, and 57.72% in GSE26712, respectively (left). Kaplan–Meier analysis demonstrated that individuals with a high-risk score had significantly worse overall survival (OS) compared to those with a low-risk score in GSE26712 (p = 0.009) (right). G The AUCs of the risk model for predicting 1, 3, and 5-year survival were, respectively, 57.00%, 54.80%, and 57.38% in GSE32062 (left). Kaplan–Meier analysis demonstrated that individuals with a high-risk score had significantly worse OS compared to those with low-risk scores in GSE32062 (p = 0.036) (right)

For the determination of the robustness of the model, we used the GSE32062 and TCGA-OV cohorts and performed the time-dependent ROC curve, showing that the AUCs of MAL at 1, 3, and 5 years were, respectively, 57.00%, 54.80%, and 57.38% (Fig. 6G, left). In the TCGA-OV cohort, the 1- and 5-year AUCs were lower (52.8% and 53.94%), while the predictive performance improved notably at longer follow-ups, with AUCs reaching 66.47% and 69.12% at 10 and 12 years, respectively (Fig. 6E). The KM curve analysis of MAL in GSE32062 demonstrated a significant difference between risk groups (p = 0.036; Fig. 6G, right).

Spatial transcriptomics analysis resolved heterogeneity of HGSOC

To comprehensively map the cellular architecture of HGSOC, we employed two complementary computational approaches: SPOTlight and multimodal intersection analysis (MIA). MIA prioritized key regions of interest, while SPOTlight revealed dynamic cell state interactions within these regions. This dual-strategy approach ensured both broad spatial annotation and granular cellular resolution, addressing limitations inherent to single-method analyses. Using hematoxylin and eosin (H&E)-stained ovarian cancer slides from the 10 × Genomics database (Fig. 7A, left), we applied SPOTlight [22] to deconvolve the tissue composition, identifying cell types including CD8⁺ T cells, B cells, epithelial cells, myeloid cells, fibroblasts, endothelial cells, IL7R⁺CD4⁺ T cells, and Treg cells (Fig. 7A, right). Subsequently, the expression levels of MAL on the tissue sections were examined (Fig. 7B). We employed the multimodal intersection analysis (MIA) as previously reported to combine the scRNA-seq with ST datasets [21]. This integration provided deeper insight into tissue heterogeneity and spatial organization. MIA delineated distinct tissue regions with unique cellular enrichments: fibroblasts in spatial 3 and 6; myeloid cells, B cells, and Treg cells in spatial 7; and CD8⁺ T cells, endothelial cells, and IL7R⁺CD4⁺ T cells in spatial 6 (Fig. 7C). We found a significant overlap between the ST and scRNA-seq data in the highly expressed genes of IL7R⁺CD4⁺ T cells and Treg cells (Fig. 7D). KEGG pathway analysis of overlapped significant genes revealed enrichment of genes involved in the T cell receptor signaling pathway, Th1/Th2/Th17 cell differentiation, the JAK-STAT signaling pathway, and the regulation of T cell proliferation (Fig. 7E, F). GO enrichment analysis of overlapped genes in region spatial 6/IL7R⁺CD4⁺T cells is shown in Figure S1A. Corresponding analysis for region spatial 7/Treg cells is presented in Figure S1B.

Fig. 7.

Spatial transcriptomics analysis of high-grade serous ovarian cancer (HGSOC) samples. A Representative hematoxylin–eosin-stained (H&E) image of HGSOC (left) and annotated cell types of HGSOC on the ST slide by SPOTlight (right). B H&E images showing the expression of MAL. C The MIA map of all scRNA-seq-identified cell types and ST clusters. D Venn diagram showing the overlaps between ST and scRNA-seq data of specifically expressed genes of Treg cells and IL7R⁺CD4⁺ T cells. E Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of the intersecting genes identified in the overlap analysis of Treg cells. F KEGG pathway enrichment analysis was performed on the intersecting genes identified in the overlap analysis of IL7R⁺CD4⁺ T cells

Cell–cell cross-talk in the HGSOC microenvironment

To explore intercellular communication networks involving ligands, receptors, and cofactors, we employed CellChat [17], which is a computational tool designed to infer and analyze cell-to-cell communication networks via scRNA-seq data. Among the communication hubs, circle plots revealed that fibroblasts, endothelial cells, and myeloid cells were the most prominent in cell–cell interactions and their strengths (Fig. 8A). Treg cells communicated mainly with the myeloid cells (Fig. 8B). These plots further indicated that MIF, TNF, CXCL, and CCL signaling pathways mediated the complex interactions between IL7R⁺CD4⁺ T cells and other cell types (Fig. 8C). Specific receptor-ligand pairs, including MIF-(CD74 + CXCR4) and MDK-NCL, showed strong interactions between epithelial cells and IL7R⁺CD4⁺ T cells (Fig. 8D, E). MIF, whose upregulation is linked to poor cancer prognosis [29], serves as a key ligand for CD74, leading to ERK activation via β-arrestin1 recruitment [30]. MIF can also bind CXCR2/4/7, triggering monocyte activation and recruitment [31].

Fig. 8.

Cellular interactions analysis of single-cell RNA sequencing (scRNA-seq) data and ST-seq data. A Global communications are presented as circle plots showing the weight/strength and number of significant ligand-receptor pairs in eight cell clusters of scRNA-seq data. B An interaction circle plot illustrating the relationship between Treg cells and other cell types. The arrow indicates the specific cell type the receptor is expressed in. A higher number of interactions is indicated by a wider edge or a larger dot. C The four circle plots show the representative critical signaling pathways, namely MIF, TNF, CXCL, and CCL. The arrow indicates the specific cell type the receptor is expressed in. D The bubble plot shows the possible ligand-receptor pairs from major cell types to Treg/IL7R⁺CD4⁺ T cells in the tumor microenvironment (TME). E The bubble plot shows the possible ligand-receptor pairs from Treg/IL7R⁺CD4⁺ T cells to major cell types in the TME. F The spatial transcriptomics (ST) plot shows the annotated cell types of HGSOC. The annotations indicate a region expressing high levels of IL7R⁺CD4⁺ T cells (pink), fibroblasts (yellow), and Treg cells (brown). G The heatmap shows the number and weight/strength of significant ligand-receptor pairs in eight cell clusters of ST-seq data. H–J The heatmap illustrates the functions performed by various cell types in TME within the network of tumor necrosis factor (TNF), NOTCH, and vascular endothelial growth factor (VEGF) signaling pathways (left), respectively. The communications are presented on the ST plot, showing the weight/strength of this signaling pathway in eight cell types

Following spot annotating of the spatial region (Fig. 8F), we applied CellChat V2 to analyze the cell-to-cell communication networks within ovarian cancer spatial transcriptomics. We created and quantified the global signaling cross-talk atlas, providing the amounts and degrees of cell–cell interactions in heatmaps (Fig. 8G). In the global overview, IL7R⁺CD4⁺ T cells, endothelial cells, and Tregs occurred in most intercellular communications (Fig. 8G, left). The outgoing signaling between the 39 most relevant pathways in CellChat was identified using pattern recognition analysis. IL7R⁺CD4⁺ T cells primarily acted as signal senders in the TNF and VEGF pathways (Fig. 8H, J), and as key influencers in the NOTCH and TNF pathways (Fig. 8I). TNF (TNFSF2) regulates immune function by activating NF-κB to enhance immune cell recruitment [32], and can also influence cell proliferation, death, and differentiation [33]. Other pathways, including EGF, ICAM, IFN-I, OXT, and WNT, were also identified as potential drivers of HGSOC progression (Fig. S1C–H). Collectively, these results underscore the significant role of IL7R⁺CD4⁺ T cells and Tregs in shaping the communication landscape of HGSOC.

Analysis of key genes immune infiltration and anti-PD-1 immunotherapy in HGSOC

Considering the potential influence of key genes on immune responses, we performed a comprehensive analysis of immune cell infiltration in HGSOC utilizing the CIBERSORT algorithm (Fig. 9A). The correlation heatmap between key genes and 22 immune cells revealed a strong association between the expression of TNFRSF18, CRLF3, KLRB1, IRF1, and BATF and the infiltration of CD4 memory-activated T cells, gamma delta T cells, and M1 Macrophages in tumor samples (Fig. 9B). A significant positive correlation was observed between MAL and M1/M0 macrophages, as well as gamma delta T cells. Subsequently, we assessed the association between these key genes and response to anti-PD-1 immunotherapy. Analysis of the GSE160751 dataset revealed a notable decline in the expression levels of ANXA1 and MAL following anti-PD-1 immunotherapy (Fig. 9C, D). Using the GSE188249 dataset, we further compared key gene expression between anti-PD-1 responders and non-responders. The data suggested that increased expression levels of ANXA1 and IRF1 were linked to improved responsiveness to anti-PD-1 immunotherapy (Fig. 9E, F). Furthermore, we validated the protein expression and subcellular localization of ANXA1 and IRF1 via immunohistochemistry using data from The Human Protein Atlas. ANXA1 localized primarily to the cytoplasm, membrane, and nucleus, while IRF1 was predominantly nuclear (Fig. 9G, H). Overall, our findings highlight the potential influence of key genes in regulating immune cell infiltration and modulating response to anti-PD-1 immunotherapy in HGSOC.

Fig. 9.

The cell death-related genes immune infiltration in HGSOC. A Immune cell infiltration landscape in HGSOC. B The heatmap illustrated the correlated expression levels of 11 key genes and the infiltration of immune cells in HGSOC. C and D The box plots showed the expression level of ANXA1 and MAL between the post anti-PD-1 immunotherapy group and the untreated group, respectively. E and F The box plots showed the expression level of ANXA1 and IRF1 between the anti-PD-1 immunotherapy response group and the non-response group, respectively. G and H The expression of ANXA1 and IRF1 proteins between tumor tissue and normal tissue from the HPA database

The expression levels of MAL and ANXA1 in ovarian cancer cell lines

To substantiate the bioinformatic predictions, we quantified MAL and ANXA1 expressions across established ovarian cancer cell lines (SKOV3 and A2780) and normal epithelial controls (IOSE80). The results showed a significant upregulation of ANXA1 mRNA in both ovarian cancer cell lines (SKOV3: p < 0.001; A2780: p < 0.001; Fig. 10A). In contrast, MAL mRNA expression was significantly decreased in both cell lines (SKOV3: p < 0.001; A2780: p < 0.001; Fig. 10B). Western blot analysis confirmed elevated ANXA1 protein abundance among the three groups (p < 0.01; Fig. 10C). MAL protein abundance was lowest among the three groups (p < 0.001; Fig. 10D). We utilized the GSE18004 dataset to analyze the survival information of HGSOC patients who underwent combined treatment with carboplatin, paclitaxel chemotherapy, and bevacizumab. The ROC curve was performed to assess the prediction capability of ANXA1 and MAL (Fig. 10E and F). The AUCs for ANXA1 and MAL were found to be 0.59 and 0.56, respectively. Despite downregulation of MAL in tumor tissues, elevated MAL expression correlated with poor overall survival (HR = 1.176, 95% CI 1.054–1.312; p = 0.004), suggesting potential stage-dependent or microenvironment-specific roles.

Fig. 10.

Independent validation of MAL and ANXA1 dysregulation in high-grade serous ovarian carcinoma. A and B qPCR analysis of ANXA1 (A) and MAL (B) mRNA in HGSOC patient samples (n = 3) versus benign controls (n = 3). Data normalized to GAPDH (mean ± SEM; p < 0.001 vs controls, unpaired t test). C and D Western blot quantification: Representative immunoblots (left) and densitometric analysis (right) of ANXA1 (C) and MAL (D) protein. Band intensities normalized to β-actin (p < 0.001 vs controls). E and F ROC curves evaluating diagnostic accuracy of ANXA1 (E) and MAL (F) mRNA in the GSE18004 cohort. The AUC are 0.59, and 0.56, respectively

MAL expression in CD4⁺ T cells and its association with tumor cell behavior in high-grade serous ovarian cancer

To further validate the potential impact of T cell-related gene expression changes on tumor biological behavior and analyze the association between CD4⁺ T cell dysfunction and tumor progression revealed in single-cell analysis, we constructed a T cell–tumor cell co-culture system. Jurkat cells were used as a T cell model and co-cultured with the ovarian cancer cell line A2780 in vitro to investigate the effect of MAL gene expression changes on tumor cell behavior. We examined the expression of MAL and ANXA1 in different cell types. The results showed that compared with primary CD4⁺ T cells, the mRNA and protein expression levels of MAL and ANXA1 were significantly increased in both Jurkat cells and A2780 cells (Fig. 11A–D). The results indicate that these genes are relatively highly expressed in both tumor-associated immune cell models and tumor cells and may be involved in the regulation of immune–tumor interactions in the tumor microenvironment.

Fig. 11.

T cell-derived MAL regulates tumor cell proliferation, apoptosis, migration, and invasion in a co-culture system. A and B qPCR analysis of ANXA1 (A) and MAL (B) mRNA in T cell-tumor cell co-culture system (n = 3; mean ± SEM; p < 0.001, unpaired t test). C and D Western blot quantification: representative immunoblots (left) and densitometric analysis (right) of ANXA1 (C) and MAL (D) protein in T cell-tumor cell co-culture system. Band intensities normalized to β-actin (n = 3; data are mean ± SEM; *p < 0.05, ***p < 0.001 by unpaired t test). E qPCR analysis of MAL mRNA expression transfected with control siRNA (si-NC) or two independent MAL-targeting siRNAs (si-RNA-1, si-RNA-2). si-RNA-2 was used for subsequent experiments. Data are mean ± SEM from three independent experiments (***p < 0.001 vs. si-NC; one-way ANOVA). F Cell proliferation measured by CCK-8 assay after MAL knockdown (si-RNA) compared to si-NC control over time. Data are mean ± SEM (***p < 0.001 at indicated time points; two-way ANOVA). G, H Apoptosis analysis by flow cytometry. Representative dot plots of Annexin V/PI staining (G) and quantification of total apoptotic cells (H) (mean ± SEM; ***p < 0.001 vs. si-NC; unpaired t test). I Representative images and quantification of wound-healing assays at 0 and 24 h post-scratch. Migration distance is reduced upon MAL knockdown (mean ± SEM; ***p < 0.001; unpaired t test). J, K Transwell migration assay. Representative images (J) and quantification of migrated cells (K) showing impaired migration in si-MAL-treated cells (mean ± SEM; ***p < 0.001 vs. si-NC; unpaired t test). L, M Matrigel invasion assay. Representative images (L) and quantification of invaded cells (M) showing reduced invasion upon MAL knockdown (mean ± SEM; ***p < 0.001 vs. si-NC; unpaired t test)

Given the potential role of MAL in TCR signaling and T cell function regulation, we further silenced the MAL gene in Jurkat cells. After assessing the silencing efficiency by transfecting multiple siRNA sequences, siRNA-2 was found to significantly reduce MAL expression levels and was therefore selected for subsequent functional experiments (Fig. 11E). In co-culture functional assays, silencing MAL in Jurkat cells significantly inhibited the proliferation of A2780 cells (Fig. 11F) and promoted their apoptosis (Fig. 11G, H), compared to the control. Furthermore, Transwell assays demonstrated that silencing MAL expression in T cells significantly inhibited the migration and invasion of A2780 cells (Fig. 11I–M). These results suggest that high MAL expression in T cells may indirectly create favorable conditions for tumor cell proliferation and invasion by weakening anti-tumor immune effects or inducing immunosuppression-related states.

Discussion

Cellular heterogeneity and the immune landscape of HGSOC

Ovarian cancer remains one of the most challenging gynecological malignancies to treat, with HGSOC being one of the subtypes with the worst prognosis [1]. Clinical data show that approximately 80% of ovarian cancer-related deaths are attributable to HGSOC, primarily due to its highly aggressive nature, biological complexity, and adaptive resistance to standard treatments [23]. Although debulking surgery combined with platinum-based chemotherapy remains the main treatment strategy in recent years, patients experience high recurrence rates and limited long-term survival benefits [2].

Traditional research often attributes the high heterogeneity of HGSOC to the intrinsic genetic instability of tumor cells, particularly the widespread CNVs and chromosomal structural abnormalities [34]. However, increasing research indicates that immune cell and stromal cell remodeling within the TME is a significant driver of HGSOC progression, immune escape, and treatment resistance [35, 36]. Therefore, relying solely on molecular characteristics at the tumor cell level is insufficient to fully explain the biological behavior of the disease and its heterogeneous response to treatment. In this study, we constructed a single-cell transcriptome atlas of HGSOC using scRNA-seq technology, systematically analyzing the tumor tissue microenvironment. Through unsupervised clustering analysis, we identified seven major cell populations, encompassing key components such as malignant epithelial cells, T cells, B cells, myeloid cells, fibroblasts, and endothelial cells. This multicellular atlas clearly reveals the high complexity of HGSOC at the cellular composition and functional state levels, indicating that its biological heterogeneity does not originate solely from tumor cells themselves, but is jointly shaped by the dynamic interactions between tumor cells and various non-tumor cells.

Although immune cells typically do not exhibit typical malignant CNV characteristics in inferCNV analysis, this does not imply their passive role in tumor progression. On the contrary, our analysis revealed that immune cells, especially CD4⁺ T cells, exhibit significant functional heterogeneity and state differentiation within the HGSOC tumor ecosystem. This heterogeneity is not manifested through genomic instability, but rather through transcriptional regulatory networks, functional program activation, and coupling with tumor cell signaling pathways.

Aberrant CD4⁺ T cell subpopulations exhibit malignant-associated features

Several single-cell studies have identified low to moderate levels of CNV signals in tumor-infiltrating immune cells across various solid tumors, such as hepatocellular carcinoma, lung cancer, and colorectal cancer [37, 38]. These CNVs are associated with chronic inflammatory stimulation, persistent antigen exposure, and immune exhaustion [38], suggesting that somatic copy-number alterations are not exclusive to cancer cells but may reflect profound cellular remodeling within the tumor ecosystem [39]. In this study, we used inferCNV analysis with endothelial cells as a reference and found that the overall CNV level of CD4⁺T cells was second only to epithelial tumor cells. This result is consistent with previous research on genomic instability in tumor-associated immune cells and further expands the applicability of this phenomenon in HGSOC. We also found that CNV abnormalities within CD4⁺T cells exhibit significant subset specificity, a characteristic often overlooked in previous studies analyzing whole immune cells.

Through fine re-clustering, we identified several CD4⁺T cell subsets (clusters 1, 2, and 7) with significantly elevated CNV levels. These subpopulations also exhibit significant tumor-associated features. Cluster 1 was characterized by high expression of CCL3, a chemokine known to recruit regulatory T cells (Tregs) into the TME, thereby suppressing antitumor immunity and fostering tumor progression [40]. Notably, CCL3 has been shown to induce VEGF secretion in tumor cells, activating angiogenesis pathways and promoting tumor invasion and metastasis [41]. Consistently, we observed significant enrichment of the VEGF signaling pathway within this Treg-associated CD4⁺ T cell subset, providing single-cell-level support for a functional CCL3–VEGF axis in HGSOC. Meanwhile, cluster 7 displayed strong activation of the AP-1 transcription factor complex (via FOS/FOSB), a key regulator of chronic inflammation, immune cell plasticity, and tumor-promoting immune responses [42]. Previous studies have found that sustained activation of AP-1 can drive T cells to a dysfunctional or tumor-supportive state and is associated with poor prognosis in various cancers [43]. Therefore, the AP-1 activation state we observed in a specific subset of CD4⁺T cells likely reflects the long-term reprogramming effect of the tumor microenvironment on immune cells.

Functional dynamics of Treg and IL7R⁺CD4⁺ T cells in HGSOC

Regulatory T (Treg) cells, a key immunosuppressive subset of CD4⁺ T cells, play a crucial role in tumorigenesis and progression [9]. We found that Treg cells and IL7R⁺CD4⁺ T cells exhibit different differentiation trajectories and functional states in HGSOC. Through Slingshot and CytoTRACE trajectory analysis, we observed that IL7R⁺CD4⁺ T cells were mainly in the mid-to-late developmental stages, while Treg cells were mainly in the early stages, reflecting the dynamic evolution of the immune response in the tumor microenvironment [44]. This heterogeneity in function and developmental state suggests that relying solely on tumor cell or traditional immune cell classification is insufficient to reveal the temporal dimension and mechanisms of immune regulation [45, 46].

Functional enrichment analysis revealed that Treg cell-related genes were significantly enriched in T cell differentiation, lymphocyte activation regulation, and cytokine signaling pathways, while IL7R⁺CD4⁺T cells were involved in immune signaling and cell death-related pathways. Furthermore, SCENIC analysis showed that transcription factors were highly activated in Treg cells, which may drive immunosuppression and tumor progression in the tumor microenvironment, such as BATF, IKZF2, and FOXO1. These results are further consistent with the findings of the hdWGCNA module analysis in this study. HdWGCNA constructed a co-expression network of CD4⁺T cells and identified several modules significantly associated with HGSOC progression and poor prognosis. Among them, the genes in the M2 and M5 modules were highly enriched in cytokine secretion, inflammatory responses, and inflammation-induced apoptosis pathways.

MAL and T cell dysfunction and HGSOC prognosis

MAL is highly expressed in thymocytes and peripheral CD4⁺ T cells, accounting for 65–90% [47]. This gene is involved in regulating the directional transport of cell membrane proteins, delivering specific proteins to the apical region of epithelial cells, maintaining cell polarity and the integrity of signal transduction [48]. Although previous studies have reported downregulation of MAL expression in both whole tumor tissue and tumor epithelial cells [49, 50], this does not necessarily contradict the association between high expression and poor prognosis. Bulk RNA-seq analysis of overall expression levels cannot distinguish cell types, and its downregulation trend may primarily reflect altered expression regulation within the tumor cells themselves [51]. Single-cell analysis allows us to identify MAL expression signatures in specific immune cell subsets [52]. In this study, we observed significant upregulation of MAL in a CD4⁺ T cell subset within the tumor microenvironment; these cells may be undergoing functional adjustment rather than malignant transformation.

This type of MAL-mediated transport mechanism is not only important in epithelial cell function but also plays a key role in immune cell activation. In T cells, MAL can promote the localization of LCK to the plasma membrane, which is a core step in T cell receptor (TCR) signaling and immune synapse formation [53]. LCK initiates the TCR cascade signaling by interacting with CD3 immune receptor tyrosine activation motifs (ITAMs), and subsequently activates downstream transcription factors, such as BATF [54, 55]. BATF has been shown to be closely associated with immune exhaustion and poor prognosis in various tumors, especially playing an important role in regulating immunosuppressive CD4⁺T cell subsets [56]. In this study, we found that BATF was significantly upregulated in HGSOC CD4⁺T cells, consistent with previous reports on BATF’s involvement in immunosuppression and transcriptional reprogramming in the tumor microenvironment, suggesting that the MAL-LCK-BATF signaling axis may play a central role in HGSOC immunosuppression and tumor progression.

The absence or functional limitation of MAL may lead to impaired T cell activation and differentiation [57], which may partially explain the poor efficacy of CAR-T cells in solid tumor therapy. Previous studies have found that removing the LCK binding site in the CAR structure can reduce IL-2 induction and significantly enhance the anti-tumor activity of CAR-T cells, indicating that its regulation of LCK-related signaling can improve the efficacy of cell therapy [58, 59]. Therefore, targeting MAL or regulating its mediated LCK-targeted transport may be a new strategy for optimizing CAR-T cell therapy for solid tumors. In addition to its function in T cell signaling, MAL also functions as a Toll-like receptor (TLR) adaptor protein in tumor cells. MAL-mediated TLR signaling can activate the NF-κB pathway, and NF-κB plays an important role in several key tumor biological processes, including tumor metastasis, cancer stem cell (CSC) maintenance and immune escape [60, 61]. Sustained activation of the NF-κB pathway has been shown to promote CSC proliferation, survival, migration and treatment resistance [62, 63].

Immune cell death-related genes and response to anti-PD-1 immunotherapy

PD-1 is expressed in some Treg cells, and its ligands PD-L1/PD-L2 play an immunomodulatory role in the tumor microenvironment, participating in maintaining immune tolerance [64]. Previous studies have shown that inhibiting PD-1 in Treg cells may actually enhance their suppressive function and promote tumor progression [65]. This indicates that simply blocking PD-1 does not always restore immune activity; its efficacy is influenced by the immunosuppressive state in the tumor microenvironment and Treg heterogeneity. Therefore, understanding the complexity of the tumor immune microenvironment, especially the functional state of CD4⁺T cell subsets, is crucial for optimizing immunotherapy strategies.

Key regulators of immunogenic cell death (ICD) and T cell function include ANXA1, IRF1, and FOXO1. ANXA1 is a biomarker of ICD that can promote tumor immune escape by inducing PD-1 expression [66, 67]. IRF1 and FOXO1 are involved in immune signaling and apoptosis regulation, modulating the responsiveness of T cells to tumor antigens [68–71]. In our study, both IRF1 and FOXO1 expression were significantly elevated in patients responsive to anti-PD-1 therapy compared to non-responders, and high expression correlated with improved clinical response. This is consistent with previous studies, suggesting that enhanced ICD-related signaling can improve the efficacy of anti-PD-1 immunotherapy and may improve T cell activity and anti-tumor immune response.

Although this study validated the regulatory role of T cell-derived MAL on tumor cells through co-culture experiments, in vitro co-culture cannot fully simulate the complex tumor microenvironment in HGSOC patients, including the multiple influences of immune cells, stromal cells, and cytokine networks. Thus, the in vivo applicability of the results requires further validation. Second, although the prognostic model was built using datasets from TCGA-OV, GSE26712, and GSE32062, it lacks independent validation in a large-scale, prospective clinical cohort. Consequently, its full clinical predictive power remains to be determined and necessitates future prospective validation in HGSOC patients.

Conclusion

This study has revealed a significant role of the genes IRF1 and ANXA1 in the response to anti-PD-1 immunotherapy, suggesting their potential as biomarkers for predicting treatment outcomes. Our study employed various genomic techniques and analysis methods to develop a novel prognostic model to predict overall survival in HGSOC patients. Our findings further revealed previously unknown roles of MAL, indicating its potential as a promising target for treating HGSOC in the future.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

HGSOC

High-grade serous ovarian cancer

MAL

Myelin and lymphocyte proteins

Treg

Regulatory T cells

TME

Tumor microenvironment

scRNA-seq

Single-cell RNA sequencing

ICD

Immunogenic cell death

SCENIC

Single-Cell rEgulatory Network Inference and Clustering

inferCNV

Infer copy number variation

MIA

Multimodal intersection analysis

EMT

Epithelial-mesenchymal transition

TNFR

Tumor necrosis factor receptor

BATF

Basic leucine zipper ATF-like transcription factor

Th17

T-helper 17 cells

hdWGCNA

High dimensional weighted gene co-expression network analysis

GO

Gene ontology

KEGG

Kyoto encyclopedia of genes and genomes

KM

Kaplan–Meier

OS

Overall survival

ST-seq

Spatial transcriptomics sequencing

H&E

Hematoxylin and eosin

MIF

Macrophage migration inhibitory factor

TNF

Tumor necrosis factor

IHC

Immunohistochemistry

CSCs

Cancer stem cells

PD-1

Programmed death 1

Author contributions

Conceptualization: Shaobo Wu. Data curation: Yaoyao Li, Jiayi Yu. Investigation: Yanhong Lv, Yurou Zhou. Methodology: Chunfeng Wu. Resources: Lei Ding, Di zhang. Supervision: Jun Tang. Writing—original draft: Shaobo Wu, Chunfeng Wu. Writing—review and editing: Jun Tang, Shaobo Wu.

Funding

This work was supported by the Scientific Research Project of Shanghai Jinshan District Health Commission (JSKJ–KTQN–2023–06).

Data availability

The datasets utilized in this study are available in the GEO repository (GSE184880, GSE26712, and GSE32062, https://www.ncbi.nlm.nih.gov/geo/) and 10 × genomics (https://www.10xgenomics.com/resources/datasets).

Declarations

Conflict of interest

The authors declare that there are no competing interests.

Ethical approval and consent to participate

All datasets in the present study were downloaded from public databases, including the TCGA and GEO databases. These public databases allowed researchers to download and analyze public datasets for scientific purposes, and thus ethics approval was not required.

Consent for publication

The authors in this study agreed to the publication of the manuscript.