Functional single-cell omics in gynecological cancers immunotherapy: from heterogeneity dissection to clinical translation

Abstract

Ovarian, endometrial, and cervical cancers represent major threats to women’s health worldwide. The profound heterogeneity of their immune microenvironments has rendered immune checkpoint blockade (ICB) therapies, particularly PD-1/PD-L1 inhibitors, far less effective than anticipated in gynecological malignancies. Conventional research approaches, constrained by coarse analytical resolution, are insufficient to unravel the underlying mechanisms of this therapeutic shortfall. Functional single-cell omics technologies, including single-cell RNA sequencing, single-cell chromatin accessibility profiling, spatial transcriptomics, and multi-omics integration, have provided a new framework for systematically studying the tumor immune microenvironment across three progressive dimensions: cellular identity, functional state, and spatial localization. This review systematically synthesizes the applications of this technological toolkit in the three major gynecological cancers. It reveals distinct bottlenecks across the three dimensions. In ovarian cancer, the core obstacle lies in the spatial dimension, where cancer-associated fibroblast-mediated physical barriers and pro-angiogenic macrophage subsets together construct a “spatial exclusion” microenvironment. In cervical cancer, the principal bottleneck resides in the functional dimension, where HPV status and histological subtype drive a “functional disability” phenotype, in which T cells infiltrate but fail to achieve effective activation, with therapy resistance-associated immune evasion clones further exacerbating functional impairment. In endometrial cancer, however, the immune fate is governed by the identity dimension, where molecular subtypes dictate the baseline nature of immune infiltration, and autophagy-related cellular subpopulations may offer new insights into intratumoral heterogeneity. Although single-cell omics has identified immunologically actionable targets with translational potential, including TREM2⁺ macrophages and the LAG-3/TIM-3 dual-target axis, clinical translation remains constrained by technical and biological bottlenecks, including stringent requirements for tissue sample quality, high assay costs, lack of standardized analytical pipelines, and the discordance between transcriptomic profiles and protein-level functionality.

1. Introduction

Ovarian, endometrial, and cervical cancers are the three major gynecological malignancies and account for a substantial portion of the global cancer burden among women. Cervical cancer ranks as the fourth most common malignancy in women worldwide, with over 600,000 new cases diagnosed annually (1). Although ovarian cancer has a lower incidence rate than cervical cancer, it remains the most lethal gynecological malignancy due to its insidious onset and profound heterogeneity, accounting for approximately 210,000 female deaths each year. Tumor heterogeneity profoundly influences growth kinetics, invasive potential, and drug sensitivity, thereby critically shaping patient prognosis (2, 3). Meanwhile, the incidence of endometrial cancer has been steadily rising in parallel with the global epidemic of metabolic syndrome (4).

Despite the substantial clinical benefits achieved with immune checkpoint blockade (ICB) therapies, particularly PD-1/PD-L1 inhibitors, in multiple solid tumor types, their efficacy in gynecological malignancies has fallen considerably short of expectations. In high-grade serous ovarian cancer (HGSOC) in particular, the objective response rate to PD-1/PD-L1 monotherapy is less than 20% (5–7). This raises a fundamental question: why does the immune system exhibit divergent response states across different gynecological cancers, despite their shared classification as “malignancies”? The three major gynecological cancers display strikingly distinct immune response patterns. Ovarian cancer is characterized predominantly by a “spatial exclusion” phenotype, wherein tumor-infiltrating lymphocytes (TILs) either fail to penetrate the tumor parenchyma or remain confined to the stroma and peripheral margins (6, 8), suggesting the presence of physical or chemotactic barriers within the tumor microenvironment that impede effective spatial access of effector immune cells. Cervical cancer, driven by persistent high-risk HPV infection, is abundant in TILs; however, immune functionality is constrained by pronounced enrichment of immunosuppressive signals, notably transforming growth factor-beta (TGF-β), resulting in a “functional disability” phenotype—T cells are present but “hot yet inert”, with their effector functions restrained by the microenvironmental signaling network (9). Endometrial cancer, in contrast, exhibits a typical “subtype-driven” phenotype, wherein immune features are largely dictated by molecular subtypes: the DNA polymerase epsilon (POLE) ultramutated subtype is distinguished by high tumor mutational burden and dense immune infiltration, whereas the p53-mutated subtype presents as a near-immune desert, suggesting that molecular classification fundamentally determines the immune fate and, consequently, the response to immunotherapy (10).

The therapeutic predicament described above does not arise because of the fundamental cellular nature of these tumors. It does not arise because tumors have been misclassified as ‘hot’, ‘cold’, or ‘immune-desert’. Rather, it arises because conventional bulk analytical approaches lack the resolution to dissect the mechanistic heterogeneity that operates within a given immune phenotype. The advent of single-cell technologies has not challenged the classical immune phenotype framework; instead, it has substantially enriched our understanding of the cellular origins, functional states, and spatial constraints that produce these phenotypes. In other words, single-cell omics does not redefine whether a tumor is hot or cold; it explains why a tumor that appears ‘hot’ by conventional criteria may nonetheless fail to respond to immunotherapy. This review synthesizes the research progress in translating functional single-cell omics discoveries into precision immunotherapy for gynecological malignancies, and examines, from the three dimensions of cellular identity, functional state, and spatial localization, the major challenges that tumor heterogeneity poses for clinical translation.

2. Three-dimensional analytical framework of functional single-cell omics and cellular heterogeneity

In this review, we introduce the term “functional single-cell omics” to describe an integrative analytical framework that goes beyond cellular identity to also capture functional state and spatial context, that is, not only what a cell is, but also what it is doing, where it resides, and which cells it interacts with. This framework differs from conventional ‘single-cell omics’, which is primarily identity-focused, and from ‘single-cell multi-omics’, which refers to the simultaneous measurement of multiple molecular modalities from the same cell. Conventional bulk RNA sequencing (bulk RNA-seq) captures averaged transcriptional signals from tissue samples, wherein the heterogeneity across distinct cell types, functional states, and spatial localizations is masked within population-level means, thereby precluding the resolution of subtle but critical differences among cellular subpopulations (11). Single-cell RNA sequencing (scRNA-seq) has overcome this limitation. Since Tang et al. (12) first achieved whole-transcriptome (mRNA) sequencing of individual mouse blastomere cells in 2009, scRNA-seq has undergone rapid development, ushering molecular biology research into a new era of cellular-resolution profiling. By enabling independent transcript capture and sequencing of individual cells, scRNA-seq allows precise delineation of the identity characteristics, functional states, and molecular phenotypes of each cell (13, 14).

Functional single-cell omics constitutes a continuously evolving methodological framework, the core inquiry of which is to determine the precise functional role of each individual cell within the tumor microenvironment. This interrogation must be advanced progressively across three dimensions: identity, state, and space. Identity confirmation relies primarily on scRNA-seq, which enables unbiased clustering based on whole-genome transcriptional profiles to discriminate cellular lineage origins (e.g., T cells, macrophages, dendritic cells, or fibroblasts) (15, 16). However, transcriptome-based identity assignment ultimately represents a “static label” that fails to elucidate the functional propensity of a given cell within its current microenvironmental context. To address this, state assessment employs single-cell chromatin accessibility sequencing (scATAC-seq) to reveal alterations in chromatin openness at the epigenetic regulatory level (17, 18), or utilizes functional gene set scoring to quantitatively dimension the functional attributes of single-cell transcriptomic data, thereby determining whether a cell resides in an activated, exhausted, quiescent, pro-inflammatory, or anti-inflammatory functional state (19, 20). Nevertheless, both scRNA-seq and scATAC-seq rely on information derived from dissociated single-cell suspensions following tissue digestion, a process during which the spatial localization information of cells within their native tissue architecture is entirely lost (21).

This limitation has catalyzed a technological breakthrough in the spatial localization dimension. Spatial transcriptomics has emerged as a complementary approach, enabling high-throughput gene expression profiling while preserving the spatial architecture of tissues, thereby offering novel perspectives for dissecting cellular spatial organization and biological processes. This technology has become a focal point in the life sciences and was named the Method of the Year 2020 by Nature Methods (22). Spatial transcriptomics assigns spatial coordinates to single-cell information, concurrently delineating cell localization and neighboring intercellular interactions (23–25). The transition from population-averaged measurements to single-cell resolution, further augmented by the spatial dimension, represents a technological leap that constitutes the cornerstone of single-cell omics-driven advances in tumor immune microenvironment research (23, 26).

It is noteworthy that the three dimensions of identity, state, and space do not constitute parallel and independent streams of information, but rather form a cognitive continuum with inherent logical progression (27, 28). Identity confirmation delineates the framework of cell types; state assessment transforms static labels into dynamic functional descriptions; and spatial localization integrates molecular information with tissue architecture. These three dimensions are progressively nested, culminating in a tripartite “identity–state–space” analytical framework (29, 30). Only through systematic integration of information across all three dimensions can the interpretation of the tumor immune microenvironment transcend the conventional static inventory of “which cell types are more or less abundant”. It can then truly rise to a multidimensional landscape that integrates molecular resolution, functional dynamics, and spatial architecture, thereby providing a more precise coordinate system for subsequent mechanistic exploration and targeted intervention. In the following sections, we apply this framework to re-examine the distinct immune microenvironment profiles of the three major gynecological cancers: spatial exclusion in ovarian cancer, functional disability in cervical cancer, and subtype-driven features in endometrial cancer. We further elucidate how single-cell omics addresses each of these clinical challenges through its corresponding analytical dimension.

3. Four core single-cell technologies and gynecological cancers

Single-cell sequencing technologies have evolved into powerful tools for dissecting intercellular interactions and resolving the heterogeneity of the tumor microenvironment, offering novel perspectives for clinical tumor management. The four major technological systems, namely scRNA-seq, scATAC-seq, spatial transcriptomics, and multi-omics integration, progressively dissect the cellular composition and functional networks of the tumor immune microenvironment at the levels of transcriptional profiling, epigenetic regulation, spatial localization, and multidimensional information integration, respectively (20, 31, 32). Their functional positioning and integrated application are summarized in Figure 1. In 2017, scRNA-seq was first applied to the study of immune cells in ovarian cancer ascites, marking the beginning of the single-cell research era in gynecological oncology (33). This approach was later extended by Izar et al., who profiled ~11,000 cells from 22 ascites specimens and provided a comprehensive single-cell landscape of the HGSOC ascites ecosystem (34).

Figure 1.

Functional positioning and integrated application of single-cell and spatial multi-omics technologies. Single-cell RNA sequencing (scRNA-seq) is primarily employed for cell type annotation and classification. Single-cell epigenomic sequencing (scATAC-seq, scBS-seq, scRRBS, EM-seq) and scCUT&Tag are utilized for gene regulatory network (GRN) inference and assessment of cellular functional competence, respectively. Spatial transcriptomics integrates cellular identity with spatial location information, enabling the construction of cell–cell communication networks, identification of fine-scale subpopulations, and spatial dissection of tumor heterogeneity. Each individual omics modality offers a restricted view of biological complexity, and thus cannot comprehensively recapitulate intricate biological processes. In contrast, integrative analysis of multi-omics data enables the establishment of an “identity–state–space” tripartite analytical framework within the native tissue context, thereby facilitating systematic elucidation of the molecular mechanisms and spatial architecture underlying tumor heterogeneity.

3.1. Single-cell RNA sequencing

Within the technological matrix of functional single-cell omics, scRNA-seq is one of the most widely applied technologies. This technique involves the capture, amplification, and sequencing of the transcriptome of individual cells, enabling the clustering of cells based on the similarity of gene expression profiles, and the inference of cellular differentiation trajectories through methods such as pseudo-time analysis (14). In gynecological cancer research, the core value of scRNA-seq lies in its capacity to uncover novel immune cell subsets that have been masked by bulk sequencing. For instance, in ovarian cancer research, investigators have successfully identified a macrophage subpopulation with pro-angiogenic functions (Angio-Mac) using this technology, and confirmed that its function is closely associated with aberrant activation of the NF-κB signaling pathway (35). This finding provides single-cell resolution molecular evidence for dissecting tumor heterogeneity from the cellular identity dimension. Beyond ovarian cancer, scRNA-seq has been extensively applied to cervical and endometrial cancers. In cervical cancer, scRNA-seq has been used to characterize the immune landscape of HPV-positive versus HPV-negative tumors, revealing that HPV-positive samples are enriched in CD4⁺ T cells and cDC2s, whereas HPV-negative tumors are predominantly infiltrated by CD8⁺ T cells (36). In endometrial cancer, scRNA-seq has resolved the immune microenvironment of TCGA molecular subtypes at high resolution, demonstrating that the POLE ultramutated subtype exhibits robust CD8⁺ effector T cell activation, whereas the NSMP subtype frequently presents an immune desert phenotype (37–40).

3.2. Single-cell epigenomic sequencing

If scRNA-seq addresses the question of “what a cell is and what it might do”, then scATAC-seq seeks to answer “why a cell does what it does” (41). This technique identifies active regulatory elements by detecting chromatin openness across the genome, thereby revealing the underlying logic of gene expression regulation at the epigenetic level. The work of Riegel et al. systematically demonstrated the unique value of single-cell epigenomic sequencing in tumor immunology research. Using scATAC-seq, they uncovered cell-state-specific chromatin accessibility heterogeneity that cannot be resolved by conventional bulk sequencing, and identified two distinct states of tumor-infiltrating CD8⁺ T cells, namely precursor exhausted T cells and terminally exhausted T cells, with significant differences in non-coding regulatory elements between them (42). Furthermore, scATAC-seq enables unbiased discovery of novel immune regulatory targets. In a single-cell multi-omics study of melanoma, scATAC-seq localized an open chromatin region upstream of the CD2 gene in T cells, which contains binding motifs for multiple transcription factors including CREB1 and JUN, providing direct physical evidence that CD2 expression is controlled by specific epigenetic regulatory elements (43). In other words, scATAC-seq reveals not the “phenomenon” of functional loss in T cells, but the “mechanism” by which function is epigenetically “locked”. The former represents transcriptional fluctuation, whereas the latter represents homeostatic remodeling at the chromatin level. It is precisely this capacity to dissociate and localize causality at the epigenetic dimension that establishes scATAC-seq as a key technological approach for dissecting the mechanisms governing cellular functional states within the tumor immune microenvironment.

3.3. Spatial transcriptomics

Although the aforementioned two technologies are capable of providing abundant molecular information, they share a common limitation, namely the loss of original spatial localization information of cells during tissue dissociation. Spatial transcriptomics (ST) emerged precisely to address this issue (44, 45). This technology employs capture spots with spatial barcodes on tissue sections, enabling the acquisition of transcriptomic data from each capture spot while preserving the spatial architecture of the tissue (46). In HGSOC research, spatial transcriptomics combined with single-cell sequencing has directly confirmed at the spatial level a long-speculated phenomenon: cancer-associated fibroblasts (CAFs), through TGF-β/SMAD signaling-driven remodeling of the extracellular matrix (ECM), construct a dense fibrotic barrier at the tumor margin, composed predominantly of collagen and fibronectin, which excludes CD8⁺ T cells from the tumor core, thereby constituting the structural basis of the “immune-excluded” microenvironment (6, 47, 48). This finding not only explains the spatial reasons for the poor response of ovarian cancer to immune checkpoint inhibitors, but also provides a theoretical rationale for targeting CAFs to relieve the spatial barrier, elucidating from the spatial localization dimension the microenvironmental causes of restricted T cell spatial accessibility. In cervical cancer, spatial transcriptomics has further demonstrated a strong spatial interaction between CD8⁺ T cells and epithelial subtypes in squamous cell carcinoma, which is markedly diminished in adenocarcinoma (49). In endometrial cancer, high-plex spatial transcriptomic profiling has revealed distinct immune components in mismatch repair-deficient tumors and highlighted a HLA class I/DNMT3A/CD8 modulatory axis that may shape CD8⁺ T cell infiltration (50).

3.4. Single-cell multi-omics integration

3.4.1. Single-cell proteomic technologies (CyTOF)

A single technology often provides only “snapshot” information. For instance, although scRNA-seq can identify a novel macrophage subpopulation, it cannot determine whether it actually suppresses T cells (which requires protein-level validation), nor whether it is located in the tumor core (which requires spatial localization) (51–53). Single-cell proteomic technologies, represented by mass cytometry (CyTOF), extend the analytical dimension to the protein level (54, 55). This technology utilizes metal isotope-labeled antibodies to label surface or intracellular proteins of single cells, followed by quantitative detection via time-of-flight mass spectrometry, enabling simultaneous measurement of dozens of functional proteins at the single-cell level (54, 56). CyTOF enables simultaneous quantification of dozens of immune markers at the single-cell level using metal isotope-labeled antibodies, and has been increasingly applied to dissect the immune microenvironment in gynecological tumors (57).

3.4.2. Multi-omics integration in gynecological cancers

Multi-omics integration can be broadly categorized into vertical integration, which combines multiple molecular modalities measured from the same cell (e.g., scRNA-seq plus scATAC-seq), and horizontal integration, which aligns data from different platforms or samples (e.g., scRNA-seq plus spatial transcriptomics). Commonly used computational strategies include early integration, in which data matrices are concatenated prior to dimensionality reduction; intermediate integration, in which shared latent spaces are learned across modalities (e.g., Seurat, Harmony, MOFA); and late integration, in which modality-specific predictions are combined at the downstream analytical level. A comprehensive study of HGSOC integrating scRNA-seq and spatial transcriptomics revealed significant differences in the immune microenvironment between primary tumors and omental metastases: the primary site exhibited a more active immune status, whereas metastatic lesions were enriched with immunosuppressive signals, suggesting spatial and temporal heterogeneity of immune states across different tumor regions (58). Beyond transcriptome-spatial integration, matched scRNA-seq and scATAC-seq profiling of ovarian and endometrial tumors has linked chromatin accessibility variation to gene expression, revealing previously unannotated regulatory elements that drive hallmark cancer pathways and highlighting the importance of intratumoral heterogeneity (59). In cervical cancer, multi-omics profiling combining proteomics and metabolomics has uncovered sphingolipid metabolism reprogramming in myeloid-derived suppressor cells, identifying the Kng1-sphingosine 1-phosphate axis as a central protein-metabolite regulatory node (60). The aforementioned cross-technology, cross-dimensional integrative analysis synergistically addresses the core demands of dissecting immune microenvironment heterogeneity in gynecological malignancies from the three dimensions of cellular identity, functional state, and spatial localization. Collectively, multi-technology integration has become an increasingly important strategy in current research, although many single-cell studies still rely on a single modality due to cost, sample availability, and technical constraints.

4. Single-cell characteristics of the immune microenvironment in three major gynecological cancers

4.1. Ovarian cancer: the “cold immune” dilemma under high heterogeneity

4.1.1. Clinical background and the necessity of single-cell research

A clinical genomics study in Japan demonstrated that among the most common HGSOC subtype, over 90% harbor TP53 mutations, which are associated with poor prognosis (61). The molecular heterogeneity of HGSOC primarily stems from impaired DNA repair, particularly homologous recombination deficiency (HRD), with approximately half of HGSOC cases being HRD-positive, most commonly attributable to BRCA mutations (62–65). Although BRCA-mutant tumors exhibit higher TIL infiltration due to a higher neoantigen burden, their CD8⁺ T cells are more prone to exhaustion and show diminished responsiveness to PD-1/PD-L1 inhibitor reinvigoration (66, 67). Why do tumors that are “T cell-rich” still fail to respond? Single-cell technologies offer a novel explanation.

Previous studies have broadly classified TAMs into M1 and M2 subtypes; however, this binary framework is insufficient to encompass the functional spectrum of TAMs within the actual tumor microenvironment (68). A large-scale integrative study, encompassing 111 samples with 272,389 CD45⁺ immune cells, constructed a single-cell atlas of the immune microenvironment across these three major gynecological cancers for the first time, identifying 11 macrophage subpopulations, among which two extremes were particularly critical: NFKB1-driven Angio-Mac (characterized by high expression of VCAN/THBS1/VEGFA, activating NF-κB to promote angiogenesis and immunosuppression, and associated with poor prognosis), and PITX1-regulated IFN-Mac_CXCL9 (secreting CXCL9/10/11 to recruit T cells via the CXCR3 axis, and correlated with improved survival) (35). This provides mechanistic clues to explain the clinical paradox of “abundant T cell infiltration but limited immunotherapy response” in BRCA-mutant tumors: even if T cells are physically present in the tumor, their function may be constrained by the immunosuppressive microenvironment driven by TAMs (such as the Angio-Mac subpopulation). Related studies have also confirmed the enrichment of an immunosuppressive network dominated by Trem2/ApoE-high TAMs in recurrent tumors (69). Collectively, these findings argue against indiscriminate TAM depletion and instead prioritize the selective elimination of pro-tumorigenic subsets.

4.1.2. Molecular subtypes of ovarian cancer and single-cell multi-omics insights into clinical outcomes

Epithelial ovarian cancer (EOC) comprises distinct histological subtypes with unique molecular features and clinical behaviors: HGSOC accounts for approximately 75%, whereas low-grade serous (LGSOC), clear cell (OCCC), endometrioid (ENOC), and mucinous (MOC) carcinomas each account for approximately 5–10% (70). The genomic alteration landscape differs markedly across subtypes: HGSOC is characterized by TP53 mutations in 96% of tumors and homologous recombination repair gene (BRCA1/2) defects in approximately half of cases; OCCC frequently harbors ARID1A and PIK3CA mutations; ENOC is dominated by PIK3CA and KRAS alterations; and MOC and LGSOC are characterized by CDKN2A/KRAS and MAPK pathway gene (BRAF/KRAS) alterations, respectively (70–72). Prognosis differs significantly among these histological subtypes, with LGSOC showing the most favorable outcome and HGSOC and ovarian carcinosarcoma being the most aggressive (72–74). Within HGSOC, integrated multi-omics analyses have progressively refined molecular subtyping. The TCGA classified HGSOC into mesenchymal, immunoreactive, differentiated, and proliferative subtypes based on gene expression (75); the CPTAC subsequently performed comprehensive proteomic, phosphoproteomic, and glycoproteomic characterization of 174 ovarian tumors (169 HGSOC), revealing how copy-number alterations influence the proteome and identifying protein acetylations associated with homologous recombination deficiency, though without establishing prognostic proteomic subtypes (76). More recently, Tu et al. performed proteomic analysis of 116 primary HGSOC tumors and identified three reproducible proteomic subtypes with distinct clinical outcomes: S-I (immune-active), S-II (metabolic), and S-III (extracellular matrix dynamic), among which S-II exhibited the best prognosis and S-III the worst, with S-III significantly associated with advanced-stage disease and immune escape (77). In parallel, Tan et al. performed an integrated proteomic and scRNA-seq study of 154 EOC tumor samples and 29 normal fallopian tube tissues, together with scRNA-seq of 8 additional EOC tumors, and identified four proteomic subtypes with distinct clinical significance: malignant proliferative (C1), immune-infiltrating (C2), fallopian-tube-like (C3), and differentiated (C4) (78, 79). The C2 subtype was characterized by enrichment of GZMK⁺ CD8⁺ T cells and phagocytic MRC1⁺ macrophages. CD40 was identified as a subtype-specific prognostic factor for C2, with interactions between CD40⁺ phagocytic macrophages and CD40L⁺IL17R⁻CD4⁺ T cells significantly correlating with favorable prognosis. In patients with non-immune subtypes, TYMP inhibition was validated as a promising therapeutic strategy, with TAS102 (targeting TYMP) combined with 5-FU suggested as a potential regimen (78). Collectively, these findings illustrate how single-cell multi-omics can link molecular subtypes to cellular networks and clinical outcomes, providing a rational basis for subtype-guided therapeutic strategies.

4.1.3. Cancer-associated fibroblast triple barriers and spatial exclusion

Cancer-associated fibroblasts (CAFs) are critical components of the tumor microenvironment (TME) and play pivotal roles in cancer progression and therapeutic response (79, 80). According to the tumor immune phenotype (TIPs) classification framework, the ovarian cancer microenvironment can be categorized into three functional states: immune-inflamed, immune-excluded, and immune-desert (81–85). Among these, the immune-excluded phenotype is the most typical, characterized by the presence of T cells in the stromal region but their inability to penetrate the tumor parenchyma. However, TIPs classification is based on morphological observations of tissue sections and cannot reveal the cellular origins or molecular composition of the exclusion barriers. Single-cell RNA sequencing has provided a refined analytical tool for this purpose. It should be noted that the iCAF/myCAF/apCAF nomenclature was initially established in pancreatic ductal adenocarcinoma and colorectal cancer (86), and direct validation of all three populations in ovarian cancer remains incomplete. In ovarian cancer specifically, myCAF-mediated collagen remodeling and spatial T cell exclusion have been directly demonstrated by spatial transcriptomics and single-cell analyses (87), whereas the immunomodulatory functions of apCAFs and the chemokine-driven MDSC recruitment by iCAFs are primarily extrapolated from other tumor types and have not been definitively established in ovarian cancer. With this caveat, the three subtypes are described as follows: inflammatory CAFs (iCAFs) secrete inflammatory chemokines such as IL-6 and CXCL1, recruiting myeloid-derived suppressor cells (MDSCs) and indirectly suppressing T cell responses (88); myofibroblastic CAFs (myCAFs) secrete dense collagen and remodel the extracellular matrix, forming physical barriers that spatially impede T cell infiltration (89, 90); and antigen-presenting CAFs (apCAFs), although expressing MHC-II molecules, lack co-stimulatory molecules such as CD80/CD86, potentially inducing Treg differentiation or T cell anergy (86, 91, 92). Among these three subtypes, myCAFs provide the structural barrier, while iCAFs and apCAFs synergistically amplify the exclusion effect through inflammatory chemotactic and immunoregulatory pathways. Spatial transcriptomics has further confirmed that in the immune-excluded microenvironment of ovarian cancer, myCAF-enriched regions are accompanied by significant collagen deposition and restricted T cell spatial localization (47, 93), directly validating from the spatial dimension the structural basis of CAF subpopulation-mediated immune exclusion.

4.1.4. Immune differences between primary and metastatic sites

The unique intraperitoneal dissemination pattern of ovarian cancer results in significant differences in the immune microenvironment between primary tumors (ovary/fallopian tube) and metastatic lesions (omentum). Integrated analysis of single-cell and spatial transcriptomics revealed that primary lesions exhibit higher immune activity than omental metastases, and in untreated primary lesions, terminally differentiated CD4⁺ and CD8⁺ T cells predominate, with their dysfunction associated with altered expression of interferon alpha/gamma pathways in epithelial cells, suggesting that epithelial-immune crosstalk constitutes an important mechanism of immunosuppression in HGSOC (58). Further studies have demonstrated that neoadjuvant chemotherapy reshapes the microenvironmental ecology: post-chemotherapy, HLA diversity decreases, PDCD1/CD274 expression is upregulated in specific cell populations such as CD8_ANXA1 and LAMP3⁺ dendritic cells, while certain poor-prognosis-associated subpopulations, including cancer.cell.2 and CAF_C3, show enrichment trends in metastatic lesions and post-chemotherapy samples (94). Collectively, these findings indicate that the immune microenvironment of HGSOC is not only regulated by anatomical site (primary versus metastatic) (95), but is also dynamically shaped by therapeutic pressure, with both factors constituting important dimensions of tumor spatiotemporal heterogeneity.

4.2. HPV-driven immune microenvironment and resistance heterogeneity in cervical cancer

4.2.1. HPV status reshapes the immune microenvironment

Fundamental differences exist in the immune microenvironment between HPV-positive and HPV-negative cervical cancers. Integrated single-cell and spatial transcriptomic analyses revealed that HPV-positive samples are enriched in CD4⁺ T cells and cDC2s, whereas HPV-negative tumors are predominantly infiltrated by CD8⁺ T cells. Mechanistically, in HPV-positive tumors, epithelial cells regulate cDC2s via the ANXA1-FPR1/3 axis, which in turn influences CD4⁺ T cells and interferon-related CD8⁺ T cell subsets; in HPV-negative tumors, epithelial cells primarily affect the monocyte-macrophage lineage, which subsequently interacts with CD8⁺ T cells (36). This finding reveals, from the cellular identity dimension, the fundamental remodeling effect of HPV status on immune cell composition, suggesting that immunotherapeutic strategies may require differential design according to distinct HPV statuses.

4.2.2. Immune differences between histological subtypes

The two major histological subtypes of cervical cancer, squamous cell carcinoma (SCC) and adenocarcinoma (AC), exhibit distinct immune microenvironment profiles. Integrated single-cell and spatial transcriptomic analyses revealed that AC is enriched in epithelial cells and displays a “stroma-rich, immune-cold” phenotype, whereas SCC is more immunogenic and enriched in plasma cells and NK/T cells. Spatial transcriptomics further demonstrated a strong spatial interaction between CD8⁺ T cells and epithelial subtypes in SCC, which was markedly diminished in AC. The study also identified subtype-specific immune and stromal features, including ICOS⁺ Tregs, IDO1⁺, CAFs, and PLVAP⁺ endothelial cells, which may respectively drive immune evasion, angiogenesis, and metastasis (49). These findings reveal, from the spatial dimension, how histological subtypes shape differential microenvironments by altering the spatial distribution and interaction patterns of immune cells, providing a theoretical basis for combining immune checkpoint inhibitors with stromal-targeted agents.

4.2.3. Three-dimensional dissection of the “functional disability” microenvironment in cervical cancer

Synthesizing the single-cell findings from the above three levels, the essence of the “functional disability” immune barrier in cervical cancer can be understood within a three-dimensional framework: HPV status dictates the identity basis of immune cell composition (HPV⁺ tumors enriched in CD4⁺ T cells and cDC2s, HPV⁻ tumors enriched in CD8⁺ T cells); histological subtypes shape the spatial landscape of immune-tumor interactions by modulating the intensity of spatial crosstalk between CD8⁺ T cells and epithelial subtypes; and CD47⁺ drug-resistant clones persisting after chemoradiotherapy represent an extreme phenotype of functional disability, wherein even when T cells are present within the microenvironment, their effector functions remain constrained by immune checkpoint signals. Together, these three dimensions elucidate the nature of “TILs present but inert” in cervical cancer, suggesting that clinical strategies should shift from sole immune checkpoint blockade toward combination regimens that relieve functional suppression, such as anti-TGF-β or anti-CD47 therapies.

4.3. Single-cell dissection of molecular subtypes and immune status in endometrial cancer

4.3.1. Immunological essence of molecular subtypes

The TCGA molecular classification, encompassing POLE ultramutated, microsatellite instability (MSI), p53-abnormal (p53abn), and no specific molecular profile (NSMP) subtypes, has become an important stratifying basis in the clinical management of endometrial cancer (96). Single-cell RNA sequencing has resolved the immune microenvironment of each subtype at high resolution. The POLE ultramutated subtype exhibits an extremely high tumor mutational burden (TMB) and abundant neoantigens, with significant enrichment and robust activation of CD8⁺ effector T cells in the microenvironment, along with upregulated expression of immune checkpoint molecules such as PD-1/PD-L1, constituting the “hot tumor” phenotype most sensitive to immunotherapy (37–39). The p53abn subtype, predominantly represented by serous carcinoma, displays genomic instability; however, a large-cohort study by Martin et al. involving 256 cases revealed intratumoral immune heterogeneity within this subtype: over 50% of cases belong to TIL-enriched subpopulations, which are associated with improved overall survival and disease-specific survival, yet the overall microenvironment of this subtype remains predominantly immunosuppressive (10, 97, 98). The MSI subtype exhibits an intermediate level of immune infiltration between the POLE ultramutated and p53abn subtypes, characterized by the coexistence of CD8⁺ cytotoxic T lymphocytes and regulatory T cells (Tregs), with dual regulatory signals of both immune activation and immunosuppression (99). In contrast, the NSMP subtype, defined as “no specific molecular profile”, represents the most heterogeneous subgroup within the endometrial cancer molecular classification, accounting for approximately 35% to 45% of endometrioid carcinomas (100). Single-cell RNA sequencing has revealed that the NSMP subtype displays the most dispersed distribution of CAF subtypes, and frequently presents an “immune desert” phenotype with low levels of intratumoral immune cell infiltration (40). Furthermore, approximately 16% of NSMP cases exhibit loss of HLA class I molecule expression, which is significantly associated with immune evasion and poor prognosis (101) Why do patients with the same molecular subtype show divergent responses to immunotherapy? Single-cell data have unveiled the existence of “intra-subtype subtypes”. Even among POLE-mutant patients, two subpopulations, namely “immune-normalized” and “immune-exhausted”, have been identified, distinguished by the ratio of Tregs to effector T cells and the balance of co-stimulatory versus co-inhibitory receptor expression (102, 103). Collectively, these findings suggest that molecular subtypes fundamentally address the question of cellular identity composition within the tumor immune microenvironment—namely, which immune cell subpopulations are enriched and in what proportions—yet fine functional differences persist within the same identity label, representing the deeper layer of heterogeneity that can only be discerned at single-cell resolution.

4.3.2. Emerging role of aggrephagy in the endometrial cancer TME

A recent single-cell sequencing-based study in endometrial cancer introduced the perspective of aggrephagy, analyzing 36,227 single cells from 5 samples and subclassifying fibroblasts, macrophages, CD8⁺ T cells, and lymphatic endothelial cells into 2 to 5 functional subsets based on the expression patterns of 44 aggrephagy-related genes. Aggrephagy is now described as a selective form of autophagy that targets misfolded or aggregated proteins for lysosomal degradation, representing a critical component of the cellular proteostasis network (104). Yuan et al. applied this concept to endometrial cancer by subclassifying TME cell populations based on the expression of aggrephagy-related genes, thereby linking a fundamental cellular homeostatic pathway to tumor microenvironment heterogeneity and clinical outcomes (105). CellChat analysis revealed extensive ligand-receptor interactions between these subsets and tumor epithelial cells, with macrophage communication being particularly active. The TUBA1A⁺ macrophage subtype was significantly associated with fewer lymph node metastases and longer survival, with immunohistochemical validation consistent with bioinformatic findings. These aggrephagy-related TME cell subsets demonstrated significant predictive value for both survival and response to immune checkpoint blockade therapy in endometrial cancer patients (105). Importantly, however, this study did not directly establish a causal or functional link between the TUBA1A+ macrophage phenotype and aggrephagy. Collectively, these findings indicate that aggrephagy-related gene signatures can inform TME cell subtyping and are associated with clinical outcomes in endometrial cancer, but whether aggrephagy itself functionally contributes to these phenotypes warrants further experimental validation.

4.3.3. Three-dimensional dissection of the “subtype-driven” microenvironment in endometrial cancer

Synthesizing the above findings, the essence of the “subtype-driven” immune barrier in endometrial cancer can be understood within a three-dimensional framework: the TCGA molecular subtypes determine, at the identity dimension, the fundamental composition and abundance of tumor-infiltrating immune cells—the POLE and MSI subtypes are characterized by CD8⁺ T cell enrichment, the p53abn subtype is dominated by immunosuppression, and the NSMP subtype presents an immune desert phenotype driven by CAF heterogeneity and HLA-I loss. However, fine functional differences persist within the same molecular subtype, as exemplified by the “immune-normalized” and “immune-exhausted” subpopulations within the POLE subtype, which explains the variable therapeutic responses observed among patients with the same subtype. Furthermore, the discovery of aggrephagy-associated cell subsets suggests that functional heterogeneity at the state dimension may partially originate from differences in intracellular homeostatic regulatory pathways such as aggrephagy/autophagy. Collectively, these three dimensions indicate that immunotherapeutic strategies for endometrial cancer should precisely screen beneficiary populations based on molecular subtypes, with priority given to ICB in POLE/MSI subtypes, while simultaneously paying attention to functional heterogeneity within the same subtype and exploring individualized regimens combining targeted therapies against pathways such as aggrephagy. To systematically present the differences in immune characteristics among the three major gynecological cancers within the three-dimensional framework, we have summarized Table 1 and Figure 2.

Table 1.

Single-cell analysis of immune phenotypes and three-dimensional spatial characteristics in three gynecological cancers.

Comparison dimension	Ovarian cancer (HGSOC) (6)	Cervical cancer (9)	Endometrial cancer (10)
Predominant immune phenotype	Predominant immune phenotype: Immune-excluded*	Functionally disabled (TILs “present but inert”)	Subtype-dependent
Phenotypic characteristics	T cells are present in the stromal region but fail to penetrate the tumor parenchyma	Presence of immunosuppressive signals such as TGF-β, with T cell effector functions constrained by the microenvironmental signaling network	POLE ultramutated subtype characterized by high tumor mutational burden and dense immune infiltration, whereas p53-abn subtype characterized by low immune infiltration
Core obstacle in the spatial dimension	CAF/TGF-β construct physical barriers, excluding T cells from the tumor core	Spatial interactions exist but are overridden by immunosuppressive signals	Spatial distribution varies by subtype (T cells widespread in POLE; sparse in p53)
Core obstacle in the state dimension	CD8+ T cell exhaustion; TAMs skewed toward pro-angiogenic phenotype (Angio-Mac)	TILs are abundant but functionally constrained; HPV drives suppressive signals such as TGF-β, impairing T cell function	T cell activation in POLE and MSI; immunosuppression predominates in p53 (97, 98)
Core obstacle in the identity dimension	CAF heterogeneity (myCAF physical barrier + apCAF immunomodulation) (91, 92)	HPV+ enriched in CD4+ T cells/cDC2s; HPV⁻ enriched in CD8+ T cells (36)	High TMB in POLE drives neoantigen generation (38, 39); HLA-I loss in NSMP subtype (101)
Implications for ICB combination strategies	Anti-CAF/anti-TGF-β + ICB (relieve spatial barrier first)	Anti-TGF-β/anti-CD47 + ICB (relieve functional suppression)	Subtype-guided selection (POLE/MSI prioritized; p53 requires combination with other strategies)

*The immune-excluded phenotype is particularly characteristic of HRD-positive tumors with stromal restriction, which exhibit T cell infiltration but impaired intratumoral penetration.

Figure 2.

Schematic of three immune microenvironment dysfunction paradigms in major gynecological cancers. (A) Ovarian cancer – spatial exclusion type: Cancer-associated fibroblasts (CAFs) remodel the extracellular matrix to form a physical barrier at the tumor margin, excluding T cells from the tumor core, while pro-angiogenic macrophage subsets (e.g., Angio-Mac) further reinforce immunosuppression. (B) Endometrial cancer – subtype-driven type: TCGA molecular subtypes determine the baseline immune landscape. The POLE ultramutated subtype displays dense TIL infiltration (a “hot” phenotype), while the MSI subtype shows intermediate infiltration with mixed activation/suppression signals. Over 50% of p53abn cases are TIL-enriched, yet the overall microenvironment remains predominantly immunosuppressive; The NSMP subtype, featuring CAF heterogeneity and HLA-I loss, presents an ‘immune desert’ with sparse immune infiltration. (C) Cervical cancer – functional disability type: HPV⁺ tumors are enriched in CD4⁺ T cells and cDC2s, whereas HPV⁻ tumors are dominated by CD8⁺ T cell infiltration. Squamous cell carcinoma (SCC) is more immunogenic than adenocarcinoma (AC), with spatial crosstalk between CD8⁺ T cells and epithelial subtypes. Despite abundant TILs driven by persistent high-risk HPV infection, immune function is constrained by robust enrichment of immunosuppressive signals, notably TGF-β.

5. From single-cell atlas to clinical decision-making

Single-cell sequencing is driving the transition of tumor target discovery from population averages to single-cell precision. Conventional bulk RNA-seq only reflects population-level average expression and is insufficient to identify rare subpopulations and their specific molecular features. In contrast, single-cell technologies enable the profiling of thousands of cells at single-cell resolution, precisely identifying functional marker-positive subpopulations and significantly enhancing the resolution and specificity of target discovery. On this basis, several immunologically actionable targets with translational potential have been identified in gynecological cancers. However, immune evasion mechanisms differ across tumor types, and accordingly, the corresponding strategies to counteract them should be tailored accordingly.

In ovarian cancer, a tumor classically characterized by spatial exclusion, TREM2⁺ macrophages are enriched in ascites, and their expression is negatively correlated with CD8⁺ T cell activation, suggesting that they suppress immune responses through the TREM2 signaling axis. Myeloid-targeting monoclonal antibodies, including PY314 (anti-TREM2) and PY159 (anti-TREM1), have entered early-phase clinical testing; PY314 has completed a phase I trial in platinum-resistant ovarian cancer, demonstrating acceptable safety and stable disease as monotherapy and in combination with pembrolizumab (106, 107). Concurrently, single-cell analyses have identified a pro-angiogenic macrophage subpopulation (Angio-Mac) associated with poor prognosis, in which the NF-κB signaling pathway is highly activated, suggesting that intervening in NF-κB signaling may suppress tumor angiogenesis by remodeling the functional state of macrophages (35). The common rationale underlying these targets is to dismantle the spatial barrier constructed by immunosuppressive myeloid cells and aberrant angiogenesis within the ovarian cancer microenvironment. While preclinical studies in HGSOC models demonstrate that coordinated blockade of TGF-β and PD-L1 can provoke durable antitumor immunity (108), and TGF-β-driven stromal barriers are increasingly recognized as a proximal mechanism of ICB resistance in ovarian cancer (47), clinical strategies targeting this axis remain predominantly in early-phase testing. These include TGF-β receptor inhibitors such as galunisertib, bifunctional TGF-β/PD-L1 inhibitors such as bintrafusp alfa, and CAF-modulating agents, with no phase III data yet available (47, 109). Notably, the limited clinical success targeting CAFs in ovarian cancer has been explicitly acknowledged in the literature, with calls for well-designed studies with adequate sample sizes before clinical benefit can be expected (110). These strategies therefore remain mechanistically compelling and preclinically promising, but not yet clinically validated in ovarian cancer.

In cervical cancer, a “hot but inert” tumor characterized by functional disability, single-cell sequencing combined with the inferCNV algorithm has identified five tumor epithelial cell subtypes, among which the TSPAN1⁺ EPC subtype accounts for the highest proportion in both high-grade squamous intraepithelial lesions (HSIL) and cervical cancer, and exhibits a significantly elevated CNV score, suggesting greater malignant potential. This subtype has been proposed as a prognostic biomarker and a potential therapeutic target (111). Precision intervention targeting tumor epithelial subtypes may relieve the functional constraints imposed on T cells within the HPV-driven microenvironment. Furthermore, the strategy of anti-CD47 monoclonal antibody combined with radiotherapy has also received biological evidence at the single-cell level (112).

In endometrial cancer, a tumor in which immune fate is dictated by molecular subtypes, single-cell immunoprofiling has revealed the presence of T cell subpopulations co-expressing LAG-3 and TIM-3, which exhibit a deeply exhausted phenotype (113). Combination strategies simultaneously targeting LAG-3 and TIM-3 are currently under clinical investigation in endometrial cancer, including the POD1UM-204 trial evaluating retifanlimab (anti-PD-1) plus INCAGN02385 (anti-LAG-3) and INCAGN02390 (anti-TIM-3) in advanced/metastatic disease (NCT04463771) (114). Separately, PD-1/LAG-3 and PD-1/TIM-3 bispecific antibodies (tobemstomig and lomvastomig) have entered first-in-human testing in other solid tumors, though their evaluation in gynecological malignancies remains pending (115). In addition, integrated single-cell and spatial transcriptomic analyses have identified the TUBA1A⁺ macrophage subtype as being significantly associated with fewer lymph node metastases and longer survival, suggesting that the aggrephagy pathway may participate in the regulation of tumor progression (105). Collectively, these findings indicate that the translational priority in endometrial cancer lies in precisely screening ICB-responsive populations based on molecular subtypes and microenvironmental features, while concurrently exploring individualized strategies that combine targeting of novel pathways such as aggrephagy.

Despite the identification of numerous immunologically actionable targets with translational potential through single-cell omics—including anti-TREM2 monoclonal antibodies that deplete or modulate TREM2⁺ tumor-associated macrophages, which have demonstrated efficacy in preclinical ovarian cancer models (106) and have entered early-phase clinical testing in platinum-resistant ovarian cancer (107), and bispecific antibodies against LAG-3/TIM-3 in endometrial cancer that are currently under clinical investigation (113), only a limited number have advanced to large-scale clinical trials. The roots of this translational bottleneck can be examined from both technical and biological perspectives. At the technical level, single-cell sequencing has traditionally imposed stringent requirements on the quality of fresh tissue specimens, and conventionally obtained formalin-fixed or frozen samples were long considered unable to meet these standards. This is now an active area of methodological development: fixed-RNA and random-primer single-nucleus approaches recover cell types, pathway activities and clinically relevant cell states from FFPE and archived frozen tissue with fidelity comparable to fresh material, as shown by matched-sample benchmarking (116–118). Remaining limitations include reduced transcript capture, systematic reweighting of detection across RNA classes because nuclei rather than whole cells are profiled, and residual variability across tissue types and preservation conditions (117–119). The relatively high cost of detection limits its applicability to large-scale clinical cohorts (23, 120); and the analytical pipelines are complex and lack standardization, rendering cross-study comparisons of results difficult (121, 122). At the biological level, single-cell transcriptomics provides only a static snapshot of gene expression, which cannot fully represent the dynamic processes of protein function and cellular interactions—the discordance between transcript abundance and protein activity remains an inherent limitation that is difficult to circumvent when extrapolating clinical conclusions from single-cell transcriptomic data (123–125). Therefore, most of the aforementioned target discoveries currently remain at the stage of target identification and early validation, and their clinical translational value awaits confirmation through larger prospective cohort studies, multi-omics integrative analyses, and functional experiments.

6. Conclusion

Functional single-cell omics have fundamentally revolutionized our understanding of immune microenvironment heterogeneity in the three major gynecological cancers across the three dimensions of identity, state, and space, elevating traditional coarse population-level descriptions into a networked landscape that integrates molecular resolution, functional dynamics, and spatial architecture. This technological armamentarium provides precise molecular and spatial coordinates for addressing the respective spatial exclusion, functional disability, and subtype-driven barriers in these three gynecological cancers, offering a cell-level evidence-based systematic answer to the clinical dilemma of “why gynecological cancers, despite sharing the same organ system, exhibit profoundly divergent responses to immunotherapy”. However, the translational pathway from single-cell atlas to clinical decision-making remains confronted with multiple challenges, including technical hurdles, cost-effectiveness considerations, and biological complexity. Future efforts should focus on advancing large-scale prospective cohort validation, establishing standardized multi-omics analytical pipelines, and developing detection technologies compatible with routine clinical specimens. Concurrently, single-cell discoveries must be tightly integrated with functional validation and clinical trial design to ultimately translate the cognitive advantages afforded by single-cell resolution into precision therapeutic strategies that benefit patients.

Funding Statement

The author(s) declared that financial support was received for this work and/or its publication. This work was supported by the China Foundation for Youth Entrepreneurship and Employment (Grant No. 25H0209), the Sichuan Cancer Society (Grant No. 25H0764) and Sichuan Provincial Natural Science Foundation Project (Youth Science Fund, Category B, No. 2026NSFSC1669).

Author contributions

D-mL: Data curation, Formal analysis, Investigation, Methodology, Software, Writing – original draft. K-gP: Investigation, Methodology, Writing – original draft. B-wY: Investigation, Methodology, Validation, Writing – original draft. M-rQ: Conceptualization, Methodology, Writing – original draft. J-fX: Methodology, Writing – original draft. LZ: Conceptualization, Methodology, Writing – review & editing. LS: Conceptualization, Data curation, Investigation, Methodology, Writing – review & editing.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.