Integrated single-cell and bulk RNA sequencing unravels neutrophil heterogeneity and validates SPP1 as a prognostic biomarker in cervical cancer

Abstract

Background

Cervical cancer (CC) remains a major global health burden, with tumor microenvironment (TME) plasticity and immune evasion driving its progression. The intricate heterogeneity of neutrophils and their complex crosstalk within the TME remain poorly understood, thereby limiting the development of targeted therapeutic strategies. Single-cell RNA sequencing (scRNA-seq) provides an unprecedented level of resolution for dissecting neutrophil subpopulations and elucidating their roles in CC pathogenesis.

Methods

We integrated scRNA-seq data from CC tissues (GSE208653, n = 5) with bulk RNA-seq cohorts from The Cancer Genome Atlas - Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma (TCGA-CESC) and Genotype-Tissue Expression (GTEx) database. Using Seurat-based clustering, pseudotime trajectory analysis and CellChat, we mapped neutrophil dynamics and intercellular communication networks. Differentially expressed genes (DEGs) were analyzed using the limma package, followed by Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) enrichment analyses. A prognostic model was subsequently constructed via LASSO-Cox regression. Key targets were further validated through both in vitro functional assays, immunohistochemical (IHC) and immunofluorescence (IF) analyses of human pathological specimens.

Results

Four functionally distinct neutrophil subtypes, including immature, mature, antitumor, and interferon-stimulated populations, were characterized in CC, demonstrating dynamic heterogeneity during tumorigenesis. Secreted Phosphoprotein 1 (SPP1) was identified as a critical differentially expressed gene, with the SPP1-CD44 axis serving as a key mediator of neutrophil-tumor cell crosstalk. Downregulation of SPP1 markedly suppressed CC cell migration, invasion, and proliferation. Furthermore, a prognostic signature based on risk stratification efficiently categorized patients into high- and low-risk cohorts, with validated clinical utility in predicting survival outcomes.

Conclusions

Our study comprehensively characterized the TME in CC through single-cell transcriptomics. Integrated analysis with bulk RNA-seq established and validated a robust prognostic signature, identifying SPP1 as a key oncogenic driver. The SPP1-centric model demonstrates significant clinical utility for risk stratification. These findings provide new insights into neutrophil heterogeneity and establish a mechanistic foundation for precision therapeutics in CC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-15636-9.

Background

Cervical cancer (CC), the fourth most common female cancer globally, often progresses asymptomatically, leading to over 70% of diagnoses occurring at intermediate or advanced stages [1]. While platinum-based chemotherapy remains a primary therapeutic option [2], its efficacy is severely limited by rapid chemoresistance, which contributes to frequent treatment failure and dismal prognoses. Although novel targeted and immunotherapies offer hope, effective treatment strategies remain insufficient [3, 4]. Therefore, further exploration of new therapeutic targets is vital to enhance the prognosis of CC.

The tumor microenvironment (TME), which comprises a diverse array of cell types and molecules such as immune cells, endothelial cells, and fibroblasts, interacts with cancer cells to influence tumor progression [5]. Among the immune cells infiltrating the TME, neutrophils are particularly abundant and exhibit marked functional heterogeneity, dynamically adapting to the local microenvironment to perform context-dependent roles [6]. Peng et al. identified neutrophil infiltration as a hallmark of CC [7], with Ji et al. further delineating their pro-tumorigenic function through the C-X-C motif chemokine ligands (CXCLs)-C-X-C motif chemokine receptor 2 (CXCR2) axis, which drives metastatic potential [8]. Although the functional heterogeneity of neutrophils has attracted significant interest in CC, its mechanisms and therapeutic targets remain unclear, thereby requiring further investigation.

While bulk RNA sequencing has advanced transcriptome studies, its averaging effect obscures cellular heterogeneity. By establishing cellular phylogenies at single-cell resolution, single-cell RNA sequencing (scRNA-seq) provides unprecedented capacity to reconstruct dynamic cellular evolution, particularly in identifying phenotypically distinct subclusters and pseudotemporally ordered cell states mediating pathological progression [9, 10]. Using scRNA-seq, research on CC has classified tumor-promoting cancer-associated fibroblasts (CAFs) into inflammatory CAFs and myofibroblastic CAFs subtypes with divergent oncogenic pathways [11], while concurrently resolving lipid-associated macrophages as a malignancy-driving subset that may predict disease progression through stromal crosstalk [12]. Nevertheless, the functional heterogeneity of neutrophils and their distinct subsets in CC remains understudied, with limited evidence elucidating their tumor-specific regulatory mechanisms.

This study constructed a single-cell transcriptome atlas of CC via scRNA-seq, revealing heterogeneous cellular compositions. Neutrophil subtypes were investigated using multidimensional profiling approaches, including CellChat and trajectory analysis. The integration of bulk RNA-seq data from the Cancer Genome Atlas - Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma (TCGA-CESC) and Genotype-Tissue Expression (GTEx) datasets enabled the identification of clinically relevant gene signatures for prognostic modeling.

Methods

Data collection and processing

The scRNA-seq data for CC were accessed from the GSE208653 dataset within the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) in November 2024. A total of 5 samples comprising tumor tissues (n = 3) and healthy tissues (n = 2) were selected as raw data. The Seurat R package (4.3.2) was used to filter and obtain 49,775 cells for scRNA-seq analysis. The cell selection criteria were as follows: each sample contained no less than 300 and no more than 7000 cells; each cell expressed more than 300 genes; each gene was expressed in at least 3 cells; the mitochondrial RNA content in each cell was less than 20% [13]. RNA sequencing data and clinical information for CC were retrieved from TCGA-CESC (https://portal.gdc.cancer.gov/, 309 samples: 3 normal, 306 tumor) and the GTEx (http://gtexportal.org/home/, 10 normal samples) in November 2024. When merging the TCGA-CESC and GTEx datasets, the “normalizeBetweenArrays” function from the limma R package (4.1.3) was applied to correct batch effects across platforms or batches. Differential gene expression analysis on the TCGA-CESC and GTEx cohort was performed using the limma package, with |log2FoldChange| > 1 and p-value < 0.05 as significance thresholds for selecting differentially expressed genes (DEGs).

For single-cell RNA-seq analysis, raw counts from GSE208653 were normalized using Seurat’s “NormalizeData” function, which included three sequential steps: (1) scaling gene expression by total cellular counts followed by multiplication with 10,000, (2) natural log-transformation after adding a pseudo-count (count + 1) to avoid undefined values, and (3) identification of 3,000 highly variable genes via the “FindVariableFeatures” algorithm. These genes were centered using the “ScaleData” function. To harmonize datasets and mitigate batch effects, we employed an anchor-based integration strategy using the “FindIntegrationAnchors” function, which established 2,000 conserved biological anchor points that mapped homologous cell populations across datasets. The integrated matrix generated through this approach was used for downstream analyses [14].

Data clustering and dimensionality reduction

In this study, clustering analysis and Uniform Manifold Approximation and Projection (UMAP)-based dimensionality reduction visualization of CC scRNA-seq data were performed using the R package Seurat. The “RunPCA” function was employed to reduce the dimensionality of the top 2,000 highly variable genes via Principal Component Analysis (PCA) [15]. Based on the cumulative standard deviation of principal components, the first 19 principal components were selected for clustering. Cell-cell distances were calculated using the “FindNeighbors” function with default parameters, and a nearest-neighbor graph was constructed based on the overlap between cell neighbors. Subsequently, the “FindClusters” function (resolution = 0.4) partitioned single-cell populations by optimizing cluster identification from the nearest-neighbor graph. Finally, the “RunUMAP” function was applied to perform UMAP-based nonlinear dimensionality reduction and visualize the clustering results. Cell identity markers were curated from published studies and established cell marker databases [16].

Pseudotime trajectory analysis and cell communication network construction

Pseudotemporal trajectories of neutrophils were reconstructed using the “monocle2” R package. A CellDataSet object was generated via the newCellDataSet function, and trajectories were visualized using plot_cell_trajectory. Branch point analysis was performed using plot_genes_branched_pseudotime, with results displayed in heatmaps. Cell-cell communication networks between distinct cell populations in normal and CC tissues were constructed using the R package CellChat (Version = 1.6.1) [17]. Normalized Seurat data were imported as CellChat objects, and ligand-receptor interactions were identified using the CellChatDB.human database.

Differential gene enrichment analysis

Cell subpopulations were reclustered using the “FindClusters” function in Seurat, and DEGs were identified via the “FindAllMarkers” function. Functional enrichment analysis of DEGs was conducted using the Database for Annotation, Visualization and Integrated Discovery (DAVID) (https://david.ncifcrf.gov/) and the Metascape platform (http://metascape.org) for Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. Gene Set Enrichment Analysis (GSEA) was further performed using the SangerBox platform (http://sangerbox.com/index.html), with cross-platform enrichment comparisons. Results were visualized using the ggplot2 package.

Prognostic model construction and validation

To establish a robust prognostic model for CC, DEGs associated with neutrophils from single-cell sequencing data were intersected with DEGs from the TCGA-CESC and GTEx cohorts. The overlapping genes were analyzed using Least Absolute Shrinkage and Selection Operator (LASSO) regression via the “glmnet” package to identify prognostic signatures and calculate risk scores. Patients with CC in TCGA-CESC were randomly partitioned into a training set (70%) and a validation set (30%). Furthermore, the GSE44001 dataset was utilized as an independent external validation cohort to verify the robustness and generalizability of the model. Based on median risk scores derived from the prognostic model, patients in both the internal and external cohorts were subsequently stratified into low-risk and high-risk groups. Survival outcomes were evaluated using the “Survival” and “ggrisk” packages to generate Kaplan-Meier curves and risk distribution plots. Time-dependent ROC curves for 3-, 4-, and 5-year overall survival were plotted with the “timeROC” package. To assess whether the identified signature was an independent prognostic factor, both univariate and multivariate Cox regression analyses were performed using the “survival” package, adjusting for clinical covariates (e.g., age, stage, histology, and race). Results were summarized in forest plots (“foreplot” package) to highlight significant risk factors. A nomogram integrating clinical variables and risk scores was developed using the “rms” package, providing a quantitative tool for individualized survival prediction. This two-step analytical framework ensures both model reliability and clinical interpretability in CC prognosis.

Patient samples

The study was approved by the Ethics Committee of The Third Affiliated Hospital, Sun Yat-sen University (Ethical Approval No. II2023-175) and conducted in accordance with recognized ethical guidelines of the Declaration of Helsinki. Hematoxylin and eosin (H&E), immunohistochemistry (IHC) and immunofluorescence (IF) analyses were performed in tumor tissues from 6 patients with CC. The tissue samples were collected after surgical resection and fixed in 10% neutral buffered formalin for 24–48 h before being processed and embedded in paraffin.

H&E, IHC and Dural-IF staining

For H&E staining, sections were deparaffinized in xylene and rehydrated through a graded ethanol series. They were then stained with hematoxylin for 3–5 min and differentiated under running tap water, followed by eosin staining for 1–2 min. Sections were rinsed, dehydrated, cleared in xylene, and mounted with a coverslip using neutral balsam. For IHC, sections were deparaffinized and rehydrated in a similar manner. Antigen retrieval was done in Tris-EDTA buffer (pH 9.0) using a microwave. After blocking endogenous peroxidase activity with 3% H₂O₂ and non-specific binding with 3% bovine serum albumin (BSA) in TBS, sections were incubated with the primary antibody (anti-SPP1, 1:500, Proteintech, China) for 1.5 h at room temperature. Following rinsing, sections were incubated with a polymer secondary antibody labeled with horseradish peroxidase for 30 min. Color development was carried out with DAB solution until a brown color appeared. Sections were rinsed, counterstained with hematoxylin for 3 min, dehydrated, cleared, and mounted as described above. All stained sections were examined under a microscope to evaluate histological features and antigen expression. For quantitative analysis of IHC staining, ImageJ software was employed to calculate the average optical density (AOD).

To validate the spatial co-localization of SPP1 with neutrophils at the protein level, dual-IF staining was performed on human cervical cancer paraffin-embedded tissue sections. Following deparaffinization, rehydration, and antigen retrieval, sections were permeabilized with PBS containing 0.1% Triton X-100. Non-specific binding was blocked with 5% BSA for 1 h at room temperature. Sections were then incubated overnight at 4 °C with the following primary antibodies: rabbit anti-human SPP1 monoclonal antibody (1:200, Immunoway, USA) and mouse anti-human Myeloperoxidase (MPO) monoclonal antibody (1:150, Immunoway, USA), with MPO serving as a specific marker for neutrophils. The next day, sections were incubated with corresponding fluorescent secondary antibodies for 1 h at room temperature in the dark: Alexa Fluor 488-conjugated goat anti-rabbit IgG (1:400, Abcam, UK) and Alexa Fluor 594-conjugated goat anti-mouse IgG (1:400, Abcam, UK). Fluorescent whole-slide imaging was performed using a 3DHISTECH Pannoramic MIDI scanner.

Cell culture and cell transfection

Human SiHa cells were obtained from Procell Company (Wuhan, China). Cells were cultured in Minimum Essential Medium (MEM, Procell) supplemented with 10% fetal bovine serum (FBS, Procell), 100 U/ml penicillin, and 100 mg/ml streptomycin. They were maintained in a humidified atmosphere at 37°C with 5% CO₂. When the confluence of the cell line reached 50–60%, cells were transfected with 50 pmol/mL small interfering RNA (siRNA) using RNAi transfection reagent (D-Nano Therapeutics, Beijing, China) for 24 hours. After seeding for 48 hours, cells were digested for protein analysis and functional experiments. The siRNA for SPP1 was purchased from Genepharma (Shanghai, China), and the sequences of siSPP1 were as follows: sense: 5’-CCGAUGUGAUUGAUAGUCATT-3’; anti-sense: 5’-UGACUAUCAAUCACAUCGGTT-3’.

Western blot

Protein extraction from cell samples was performed using Radioimmunoprecipitation Assay Buffer (RIPA) buffer (Thermo Fisher Scientific, Waltham, MA, USA), and protein concentration was quantified via the Bicinchoninic Acid (BCA) assay (Keygen Biotech, China). Equal amounts of protein (20 µg per sample) were loaded and separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). After transferring to polyvinylidene fluoride (PVDF) membranes (Gene Molecular Biotech, Inc., Shanghai, China), the membranes were blocked with 5% non-fat milk at room temperature for 1 h. Subsequently, membranes were incubated overnight at 4 °C with the following primary antibodies: anti-SPP1 (1:4000, Proteintech, China) and anti-GAPDH (1:2000, Huabio, China). The membranes were then incubated with a horseradish peroxidase (HRP)-conjugated rabbit IgG secondary antibody (1:4000, Proteintech, China) at room temperature for 1 h. Protein expression levels were detected using an Enhanced Chemiluminescence (ECL) kit (NeoBioscience) and visualized with a Western blot imaging system.

Cell proliferation assay

Cell proliferation was measured using the Cell Counting Kit-8 (CCK-8) assay. Cells were seeded in 96-well plates at a density of 2 × 10³ cells per well in triplicate. After incubation for 0, 24, 48 and 72 h at 37 °C with 5% CO₂, 10 µL of CCK-8 solution was added to each well and incubated for 1 h. The absorbance was measured at 450 nm using a microplate reader.

Transwell migration and invasion assays

Transwell assays were conducted using a 24-well plate equipped with a Transwell chamber system (Corning Inc., USA). For the invasion assay, 100 µL of Matrigel was added to the upper chamber and incubated at 37 °C for 12 h. Transfected cells (1 × 10⁵ cells/well) were seeded in the upper chamber with 200 µL of serum-free medium, while 600 µL of medium containing 20% FBS was added to the lower chamber. After incubation at 37 °C with 5% CO₂ for 48 h (24 h for the migration assay), non-migrated/invaded cells in the upper chamber were removed by gentle washing. The membrane was then stained with 0.3% crystal violet for 30 min. Migrated and invaded cells were imaged under a microscope, and five random fields per well were captured for statistical quantification.

Statistical analyses

Statistical analyses for in vitro experiments, including Western blot (WB), IHC and IF staining, cell proliferation assays, and Transwell migration and invasion assays, were performed using GraphPad Prism 9 (GraphPad Software, USA). For the quantitative analysis of IF co-localization, ImageJ software (National Institutes of Health) was used. The “Plot Profile” function was applied to exhibit fluorescence intensity. Comparisons between two groups were conducted using an unpaired t-test. For multiple group comparisons, a two-way analysis of variance (ANOVA) followed by Tukey’s multiple comparisons test was employed. Data were presented as the mean ± standard deviation (SD). A p-value less than 0.05 was considered statistically significant. Additional data analyses were conducted using R Studio (4.1.3) with appropriate software packages.

Results

Single-cell transcriptomic atlas defines immune-stromal shifts in CC

We analyzed a total of 5 samples from the GSE208653 dataset, including 3 tumor and 2 normal tissue samples. A total of 49,775 cells were obtained after quality control processing using Seurat. All cells in the scRNA-seq dataset were partitioned into 17 distinct clusters through PCA-based clustering with a resolution of 1. As shown in Fig. 1A, the two-dimensional visualization generated by UMAP demonstrated distinct spatial segregation of cellular subpopulations. Cell type annotation was performed using established markers from the Cell Marker database and published research [15]. Our integrated analysis identified ten major cell lineages: NK/T cells, neutrophils, epithelial cells, myeloid cells, fibroblasts, B cells, endothelial cells, mast cells, plasma cells, smooth muscle cells (Fig. 1B). The single-cell transcriptome landscape across different samples is illustrated in Fig. 1C, while Fig. 1D displays the single-cell transcriptome atlas for normal and tumor samples. Cell lineage-specific markers were comprehensively detailed in Fig. 1E, demonstrating NK/T cells expressing NKG7, CCL5, GZMA and CD3G/E/D, fibroblasts exhibiting high expression of DCN, COL1A1, COL3A1, and CD19, B cells characterized by CD19, BANK1 and MS4A1 expression, and neutrophils labeled by NCF1 and SORL1 (Fig. 1F-G). This systematic annotation approach validated the specificity of selected biomarkers for precise cellular identification. Quantitative analysis revealed significantly increased proportions of neutrophils, epithelial cells, myeloid cells, and B cells in tumor specimens compared to normal counterparts (Fig. 1H).

Fig. 1.

Single-cell atlas of normal and tumor tissues in cervical cancer patients. A UMAP plot showing 17 identified cell clusters; each dot represents a single cell. B UMAP plot depicting 10 major cell types in cervical cancer tissue based on characteristic gene signatures. C-D Distribution of distinct cell types in normal versus tumor tissues from cervical cancer patients. E Dot plot displaying expression levels of marker genes used to annotate the 10 cell types. F Heatmap illustrating differential expression of key genes across cell clusters. G UMAP plot highlighting marker gene expression patterns in neutrophil populations. H Bar plot comparing proportions of different cell types in normal and tumor tissues

Four functionally distinct neutrophil subtypes exhibit tumor-specific expansion

Single-cell transcriptomic analysis of the CC atlas revealed a substantial neutrophil population across all samples. Following computational isolation, neutrophils underwent high-resolution re-clustering using our curated database to characterize their functional diversity in CC pathogenesis. Unsupervised clustering analysis revealed seven transcriptionally distinct neutrophil subsets (Fig. 2A), classified into four functional categories: mature, anti-tumor, immature, and interferon-stimulated subtypes (Fig. 2B). UMAP visualization (Fig. 2C-D) demonstrated neutrophil heterogeneity across tissue types, revealing distinct spatial distributions in normal versus tumor microenvironments. In this respect, the distribution of neutrophil subtypes exhibited significant discrepancies between normal and tumor samples. Definitive marker signatures were established through differential expression analysis: CXCR2 was identified as a marker of mature neutrophils, CXCL8 and CXCR4 as markers of antitumor neutrophils, CD84 as a marker of immature neutrophils, and IFIT1, IRF7 and RSAD2 as markers of interferon-stimulated neutrophils (Fig. 2E-F). As shown in Fig. 2G, the marker genes of antitumor and interferon-stimulated neutrophils were significantly overexpressed in tumor tissues. Specifically, the proportions of anti-tumor neutrophils and interferon-stimulated neutrophils were markedly elevated in tumor samples compared to normal samples, whereas the proportion of mature neutrophils was reduced (Fig. 2H). Collectively, our multimodal analysis suggests that specialized neutrophil subsets may act as important modulators of CC progression through distinct functional programs.

Fig. 2.

Transcriptomic landscape of neutrophils in single-cell sequencing. A UMAP plot showing clustering of neutrophils into 7 distinct subsets via scRNA-seq. B UMAP plot of 4 neutrophil subpopulations identified by specific marker genes. C-D Distribution of neutrophil subpopulations in normal versus tumor tissues. E Dot plot exhibiting marker genes for 4 neutrophil subtypes. F Heatmap displaying major gene expression differences among subtypes. G UMAP plot featuring marker gene expression in antitumor neutrophils and IFN-stimulated neutrophils. H Bar plot comparing proportions of neutrophil subtypes in normal and tumor tissues

Pseudotemporal trajectory maps neutrophil differentiation

Cell trajectory analysis, also known as Pseudotime analysis, simulates the developmental trajectory of different cells based on the expression patterns of temporal genes in single-cell samples. To investigate neutrophil dynamics in the CC microenvironment, we performed the cell trajectory analysis by extracting neutrophils, which revealed three branches of neutrophils subtypes (Fig. 3A). Pseudotemporal ordering resolved seven differentiation states, with color gradients reflecting maturation progression from immature (dark) to terminally differentiated states (light) (Fig. 3B-C). The trajectory mapped a progressive differentiation continuum: immature subtypes (Branch 3) transitioned through intermediate states (Branch 2) before giving rise to anti-tumor effector populations (Branch 1). Branch expression analysis modeling (BEAM) function identified 10,505 differentially expressed genes, which clustered into four co-expression modules through a heatmap (Fig. 3D). Module 1 exhibited initial high expression followed by rapid decline, involving cytoplasmic translation and ribosome biogenesis. Module 2 demonstrated decreased expression during the early to mid-stages, followed by a significant upregulation in the late stages, encompassing TNF production and response. Module 3 revealed elevated expression in mid-late stages, covering leukocyte chemotaxis and response to lipopolysaccharide. Module 4 displayed a bell-shaped expression pattern, containing pathways for neutrophil chemotaxis and migration. Pseudotemporal trajectory analysis of eight neutrophil-state marker genes revealed distinct developmental expression patterns (Fig. 3E). Late-stage antitumor neutrophils predominantly expressed C15orf48 and CCL3 (L1), while CD83 showed a progressive increase in expression that correlated with effector differentiation. In addition, IFITM2, MNDA, and S100A8 (A9) exhibited transient expression peaks during intermediate maturation stages.

Fig. 3.

Pseudotime analysis of neutrophils in single-cell sequencing. A Distribution of neutrophil subtypes along the pseudotime axis. B Developmental trajectory of neutrophils inferred using the Monocle2 algorithm. C Trajectory plot showing neutrophil differentiation stages over pseudotime. D Hierarchical clustering of differentially expressed genes (DEGs) along pseudotime and Gene Ontology (GO) enrichment analysis for four DEG clusters. E Pseudotemporal dynamics of the top 8 genes across neutrophil subtypes. Dots represent cells, colors denote clusters, and the y-axis indicates gene expression levels

Cell-cell communication networks in the normal and CC tissue

Systematic profiling of ligand-receptor co-expression networks revealed enhanced cellular crosstalk in CC microenvironments, demonstrating increased interaction complexity within the TME (Fig. 4A). Comparative analysis of communication probabilities in the information network showed distinct signaling patterns between normal and tumor tissues. The results revealed that SPP1, FN1 and GALECTIN signaling pathways exhibited significant enrichment in tumor tissue (green), while CXCL, ANNEXIN, and VISFATIN signaling pathways exhibited higher activity in normal tissue (red) (Fig. 4B). Neutrophil-centric interaction analyses identified tumor-specific SPP1-CD44 axis activation (Fig. 4C), suggesting its potential role in mediating the TME of CC. Multidimensional visualization confirmed widespread SPP1 signaling activation across tumor-infiltrating myeloid cells and neutrophils (Fig. 4D-E), consistent with established roles of SPP1-CD44 interactions in tumor progression through immune modulation [18, 19]. Our analysis further identified specific SPP1-CD44-mediated communication between myeloid cells and neutrophils (Fig. 4F to G). The TME exhibited polarized SPP1-CD44 signaling patterns, with immune-derived (neutrophils and B cells) signals targeting stromal elements, while endothelial-neutrophil crosstalk predominantly engaged myeloid compartments (Fig. 4H).

Fig. 4.

Cell-cell communication analysis. A Comparison of communication quantity and interaction strength between normal and tumor tissues. B Differences in pathway-associated communication number and intensity between tissues. C Dot plot revealing potential ligand-receptor pairs mediating interactions between neutrophils and other cell types. D Heatmap depicting differential ligand-receptor interaction patterns in the SPP1 signaling pathway across tissues. E SPP1 signaling networks in normal and tumor tissues; thicker edges indicate stronger interactions. F Contribution differences of representative ligand-receptor pairs in SPP1 signaling. G Expression profiles of representative SPP1 signaling genes across cell types. H Chord diagrams comparing SPP1-CD44 ligand-receptor expression patterns between tissues

Functional enrichment analysis of neutrophil-derived genes

Functional enrichment analysis was next conducted to uncover the biological roles of the target genes. KEGG analysis revealed that the gene set was significantly enriched in multiple immune-related and cellular signaling pathways, including the Toll-like receptor, TNF, C-type lectin receptor, and chemokine signaling pathways. It also revealed enrichment in pathways linked to cell differentiation and disease, including osteoclast differentiation and tuberculosis (Fig. 5B). GO analysis indicated significant enrichment in biological processes (BPs) such as immune response regulation, innate immune response activation, and positive regulation of cytokine production. For cellular components (CCs), the focus was on secretory granule membrane, specific granule, vacuolar membrane, and lysosomal membrane. In terms of molecular functions (MFs), the gene set showed significant enrichment in protein serine/threonine kinase activity, ubiquitin-like protein ligase activity, and pattern recognition receptor activity (Fig. 5C). Compared with normal tissues, target genes were up-regulated in different tumor cell clusters, especially in neutrophils (Fig. 5A, D). Meanwhile, upregulated expression of target genes was observed in pathways related to immune defense responses, antigen processing and presentation, and metabolic processes, while downregulated expression was observed in humoral immune responses and negative regulation of immune responses (Fig. 5E), which implicate target genes in modulating the immune response in the TME of CC.

Fig. 5.

Identification and functional enrichment of differentially expressed genes (DEGs). A Differential gene expression distribution across cell types. B Bubble plot of KEGG pathway enrichment showing 12 significant pathways associated with DEGs. C Gene Ontology enrichment analysis displaying top 6 representative Biological Processes (BPs), Cellular Components (CCs), and Molecular Functions (MFs). D Volcano plot of DEGs. Genes with |log2FoldChange| > 1 and p < 0.05 were considered significant (red: upregulated; green: downregulated). E Bar plot of enriched up- (red) and down-regulated (blue) pathways

Prediction and validation of prognostic gene signatures

Prognostic gene identification was performed by intersecting DEGs from the TCGA-CESC and GTEx datasets (n = 162 DEGs) with neutrophil-specific DEGs (n = 733) from scRNA-seq, yielding 5 genes: SPP1, ANKRD22, DAPP1, IDO1 and C15orf48 (Fig. 6A). The expression matrix of these genes was standardized, and the LASSO-Cox algorithm was employed to calculate their risk scores (Fig. 6B-C). TCGA-CESC cervical cancer patients were randomly allocated to training (n = 214, 70%) and internal validation (n = 92, 30%) cohorts. Heatmap analysis revealed distinct modeling gene expression patterns, showing significantly higher SPP1 expression in high-risk versus low-risk groups (Fig. 6D-E). Kaplan-Meier survival curves demonstrated poorer prognosis in high-risk groups across both internal cohorts (p < 0.05, Fig. 6G-H). Time-dependent ROC curve analysis validated the signature’s predictive efficacy for overall survival, with training cohort AUCs of 0.62 (3-year), 0.62 (4-year) and 0.66 (5-year), and internal validation cohort AUCs of 0.64 (3-year), 0.73 (4-year) and 0.68 (5-year) (Fig. 6J-K). To further confirm the robustness and generalizability of the signature, the GSE44001 dataset was utilized as an independent external validation cohort. Heatmap analysis revealed that the expression patterns of the signature genes in the external cohort were consistent with those in the internal datasets, showing elevated expression of risk genes (e.g., SPP1) in the high-risk group (Fig. 6F). Kaplan-Meier survival analysis confirmed a significant divergence in prognosis between the two groups, with the high-risk group exhibiting markedly poorer overall survival compared to the low-risk group (p < 0.05, Fig. 6I). The AUCs for 3-, 4-, and 5-year survival in the external validation cohort were 0.71, 0.68, and 0.63, respectively (Fig. 6L).

Fig. 6.

Construction and validation of the risk prediction model. A Venn plot showing overlap between neutrophil-associated DEGs from scRNA-seq and TCGA-CESC samples. B LASSO regression model built using the TCGA-CESC training set. C Cross-validation for parameter optimization in the model. D-F Distribution of risk scores versus survival status in the training set, internal and external validation set. G-I Kaplan-Meier survival curves stratified by risk score in the training set, internal and external validation set. J-L Time-dependent ROC curves predicting 3-, 4-, and 5-year survival in the training set, internal and external validation set

Establishment and validation of clinical prediction models

To evaluate the independent prognostic value of the risk score, both univariate and multivariate Cox regression analyses were performed. The results demonstrated that the risk score remained a significant independent predictor of overall survival, even after adjusting for other clinical covariates, including age, tumor stage, histology, and race (Fig. 7A-B and Fig. S1). To facilitate clinical application and enhance predictive accuracy, we developed a quantitative nomogram incorporating these independent prognostic factors (age, tumor stage, histology, race, and risk score) to predict 3-, 4-, and 5-year survival probabilities (Fig. 7C). Furthermore, calibration curves exhibited excellent agreement between the nomogram-predicted probabilities and the actual observed outcomes, confirming the model’s robust performance and reliability (Fig. 7D-F).

Fig. 7.

Analysis of clinical risk indicators. A Univariate Cox regression analysis of age, stage, histology, race, risk score and Tumor, Node, and Metastasis stage. B Multivariate Cox regression analysis of age, tumor stage, histology, race and risk score. C Construction of the prognostic nomogram. D-E Calibration curves for the nomogram predicting 3-, 4-, and 5-year disease-free survival (DFS) probability

SPP1 co-localizes with tumor neutrophils and its knockdown suppresses oncogenic phenotypes

Next, functional validation of the biological role of SPP1 in CC was conducted through comprehensive experimental analyses. IHC analysis revealed a significantly higher positive signal for SPP1 in CC tissues compared to adjacent normal tissues (Fig. 8A). Quantitative analysis further confirmed a marked increase in SPP1 expression in tumor tissues (p < 0.0001, Fig. 8B). Prompted by the scRNA-seq analysis suggesting high expression of SPP1 in neutrophils, we sought to validate this finding at the protein level using dual-immunofluorescence staining. As shown in Fig. 8C, in cervical cancer tissues, the fluorescent signals of SPP1 (green) and the neutrophil-specific marker MPO (red) exhibited significant spatial overlap, appearing yellow in the merged image. This co-localization pattern strongly indicates the expression of SPP1 protein within tumor-infiltrating neutrophils. Representative line profiles drawn across double-positive cells revealed significantly overlapping peaks of SPP1 (green) and MPO (red) fluorescence intensities. The high degree of synchrony between the two intensity curves along the scanned line provides robust quantitative evidence of spatial co-localization at the subcellular level (Fig. 8D). This result directly validates our prior bioinformatic discovery, positioning SPP1 as a key functional molecule expressed in neutrophils within the TME. To investigate SPP1 function, SPP1 expression in CC cells was knocked down via siRNA transfection. WB revealed a reduction of approximately 50% in SPP1 protein levels in the siSPP1 group compared to the control (p < 0.0001, Fig. 8E, G and Fig. S2A, B). Subsequently, Transwell assays indicated that SPP1 knockdown significantly suppressed the migratory and invasive capacities of CC cells, with a 51% decrease in migrated cells (p < 0.0001) and a 46% reduction in invaded cells (p < 0.0001, Fig. 8F). Furthermore, cell proliferation assays showed that SPP1 knockdown inhibited CC cell proliferation at 48 h post-transfection (p < 0.05), with a significant inhibition rate of approximately 49% observed at 72 h (p < 0.0001, Fig. 8H). These results collectively demonstrate that SPP1 exerts pro-tumorigenic effects in CC, and its overexpression is closely associated with enhanced tumor cell migration, invasion, and proliferation.

Fig. 8.

Expression of SPP1 in tissues and its impact on cellular behavior. A Histological features (H&E staining) and SPP1 protein expression (IHC) in normal and tumor tissue sections. B Quantitative analysis of SPP1 IHC intensity (p < 0.0001). C Immunofluorescence (IF) images showing co-localization of SPP1 and the neutrophil marker MPO in tumor tissue. (From left to right) SPP1 protein signal (green); MPO protein signal (red, marking neutrophils); merged image showing co-localization (yellow). Scan lines (white dashed lines) used for analysis in the double immunofluorescence image. D Fluorescence intensity distribution curves for SPP1 (green) and MPO (red) along the scan line. The overlap of the peaks indicates strong spatial colocalization between SPP1 and the neutrophil marker MPO. E, G Western blot analysis and grayscale quantification of SPP1 expression (p < 0.0001). F Transwell migration and invasion assays showing reduced capabilities in siSPP1-treated CC cells (****p < 0.0001). H CCK-8 assay demonstrating decreased proliferation in siSPP1-treated CC cells (*p < 0.05, ****p < 0.0001)

Discussion

CC, a prevalent malignancy in women, is characterized by a highly inflammatory TME containing abundant immune cell infiltrates, particularly neutrophils, which critically link inflammation to carcinogenesis [20, 21]. To address the limitations of bulk sequencing approaches in resolving cellular heterogeneity and interaction networks within the TME, scRNA-seq was applied, revealing key intercellular communications and identifying SPP1 as a pivotal mediator. Our single-cell atlas revealed decreased proportions of endothelial cells and canonical fibroblasts in tumor samples. This reduction in endothelial cells may reflect vascular dysfunction and potential endothelial-to-mesenchymal transition, leading to their reclassification [22, 23]. Similarly, the decrease in fibroblasts likely signifies their activation and conversion into CAFs [24, 25], illustrating the complex stromal reorganization alongside immune infiltration in CC. Leveraging TCGA and GEO data, we constructed and validated a prognostic model, confirming SPP1’s potential as a biomarker. Most importantly, comprehensive validation using human pathological specimens demonstrated significantly higher expression of SPP1 in CC tissues, while further in vitro functional assays further confirmed its tumor-promoting activities.

Neutrophil functional plasticity and the existence of distinct subsets with opposing roles in tumor progression have been documented in various malignancies [26–28]. Through scRNA-seq profiling, we systematically delineated the differentiation dynamics and functional heterogeneity of neutrophils within the CC microenvironment. Four transcriptionally distinct neutrophil subsets were identified: immature neutrophils, mature neutrophils, antitumor neutrophils, and interferon-stimulated neutrophils. These subsets exhibited tumor-specific spatial distributions and functional states. Besides, upregulated expression of S100A8/S100A9, a calcium-binding protein associated with inflammatory responses and tumorigenic signaling pathways [29], was observed during early maturation stages. Notably, S100A8⁺ neutrophil recruitment via the CXCL5-CXCR2 axis has been linked to lung adenocarcinoma progression [30]. This suggests that in CC, mature neutrophils expressing S100A8/S100A9 may exert a similar pro-tumorigenic role. Our data also demonstrated that antitumor neutrophils exhibited significant CD83 activation, an immune marker associated with T-cell activation via antigen presentation [31, 32], thereby corroborating their functional antigen-presenting capacity [33]. Concurrent upregulation of CCL3/CCL3L1 chemokines in this subset suggests monocyte-mediated immunosurveillance mechanisms, aligning with Zhai et al.’s findings that neutrophil-derived CCL3 inhibits metastatic colonization through monocyte recruitment [34]. In our research, terminal neutrophil differentiation into antitumor states was correlated with the upregulation of C15orf48. While this gene exhibits cancer-specific heterogeneity and neutrophil association across malignancies, its mechanistic role requires further elucidation [35]. This study provides the first comprehensive map of neutrophil developmental trajectories in CC, revealing subset-specific functional heterogeneity and identifying differentiation-associated molecular signatures, thereby establishing a theoretical framework for developing therapeutic strategies targeting neutrophil differentiation states.

Based on our findings, cell-cell communication analysis revealed that the SPP1-CD44 ligand-receptor pair mediates extensive interactions between neutrophils and stromal components, including myeloid cells and fibroblasts, with heightened signaling activity in CC tumors compared to normal tissues. Secreted phosphoprotein 1 (SPP1), a glycosylated secreted protein, regulates cellular adhesion, migration, inflammatory responses, and tumor invasion via CD44 and integrins αvβ3/5 [36, 37]. In hepatocellular carcinoma (HCC), tumor-derived SPP1 activates the PI3K/AKT pathway through CD44, driving hepatic stellate cell differentiation into cancer-associated fibroblasts to promote tumor progression [38]. Xie et al. demonstrated that SPP1-CD44 binding on alveolar epithelial cells induces CXCL1 production, recruiting neutrophils to form pre-metastatic niches that capture circulating tumor cells and facilitate pulmonary metastasis [39]. Our study identified enriched SPP1-CD44 axis interactions between neutrophils and other cells (myeloid cells, B cells and fibroblasts) in CC, suggesting tumor-neutrophil coordination via the axis to drive pro-tumorigenic inflammation and immune remodeling. While SPP1 overexpression correlates with neutrophil, T cell, and macrophage infiltration across malignancies, its functional relevance in CC and neutrophil subpopulations remains underexplored [40, 41]. Notably, functional enrichment analysis was conducted in the present study to confirm SPP1’s strong association with immune response pathways, and cross-validation across bulk and single-cell datasets robustly supported its diagnostic potential as a TME-specific biomarker. These results establish SPP1 and its neutrophil-associated networks as central coordinators of immune-inflammatory cross-talk in CC.

Functional enrichment analyses of target genes revealed significant enrichment in chemokine and TNF signaling pathways, which facilitate neutrophil recruitment, activation, and direct tumor interactions within the TME [42, 43]. Concurrently, enriched immune-regulatory pathways suggest broader modulation of anti-tumor immunity [44, 45]. Given the critical role of immune responses in CC progression and the limitations of current prognostic tools [46], integrating molecular features into novel models is imperative. Our study integrated scRNA-seq insights with bulk transcriptomics (TCGA-CESC and GTEx) to establish a prognostic model. Elevated SPP1 expression correlated with poor prognosis, consistent with its pro-tumorigenic roles in immunosuppression and metastasis. Previous research has demonstrated that increased expression of SPP1 in CC patients exhibit a correlation with poor prognosis [47]. Functional validation via siRNA-mediated SPP1 knockdown revealed significant suppression of CC cell migration, invasion, and proliferation. These experimental findings, coupled with IHC and IF findings, not only confirm SPP1’s established pro-tumorigenic roles in CC but also substantiate its prognostic significance in our model, providing a rationale for its therapeutic targeting.

This study has several limitations that should be acknowledged. First, the prognostic genes were identified through analysis of public datasets with relatively modest sample sizes, necessitating validation in larger-scale clinical cohorts. Second, the inferences from pseudotime trajectory analysis, while informative, are constrained by the limited sample size and the lack of stratification based on clinical progression, which may affect the robustness of the inferred differentiation paths. Third, our experimental validation was limited to preliminary in vitro exploration of SPP1 functionality and lacked in vivo model confirmation. Future investigations should incorporate expanded sample sizes, functional genomics approaches, and clinical intervention trials to fully elucidate the therapeutic implications of neutrophil heterogeneity in CC.

Conclusions

This study utilized scRNA-seq to characterize the CC TME landscape, revealing neutrophil heterogeneity and their roles in trajectory, transcriptional regulation, and intercellular crosstalk. Integration with bulk transcriptomics enabled the establishment and validation of a prognostic model. Most significantly, we identified and validated SPP1 as a key oncogenic driver. SPP1’s elevated expression in tumor tissues and tumor-promoting functions were confirmed through human pathological specimens and in vitro assays. These findings provide deeper biological insights and advance clinical strategies for CC.

Abbreviations

CC

Cervical cancer

TME

Tumor microenvironment

SPP1

Secreted Phosphoprotein 1

CAFs

Cancer-associated fibroblasts

CXCL

C-X-C motif chemokine ligand

CXCR

C-X-C motif chemokine receptor

scRNA-seq

Single-cell RNA sequencing

TCGA-CESC

The Cancer Genome Atlas - Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma

GTEx

Genotype-Tissue Expression

DEGs

Differentially expressed genes

KEGG

Kyoto Encyclopedia of Genes and Genomes

GO

Gene Ontology

GSEA

Gene Set Enrichment Analysis

UMP

Uniform Manifold Approximation and Projection

PCA

Principal Component Analysis

LASSO

Least Absolute Shrinkage and Selection Operator

H&E

Hematoxylin and eosin

IHC

Immunohistochemistry

IF

Immunofluorescence

BPs

Biological Processes

CCs

Cellular Components

MFs

Molecular Functions

AOD

Average optical density

PCC

Pearson's Correlation Coefficient

FBS

Fetal bovine serum

BSA

Bovine serum albumin

siRNA

Small interfering RNA

Authors’ contributions

Siting Lin, Junlin Zhong and Shengjin Yuan designed most of the investigation, performed bioinformatics analyses, wrote original draft and revised the manuscript. Manting Su assisted with the analysis and helped to revise the manuscript. Yiwang Zhang helped to provide clinical sample collection and perform the histological analyses. Xinling Zhang reviewed the entire manuscript and provided significant comments. All authors have read and approved the final manuscript.

Funding

This study was supported by grants from Guangdong Basic and Applied Basic Research Foundation (2023A1515220008), the Municipal and University (Hospital) Joint Funding Project of Guangzhou Municipal Science and Technology Bureau (2023A03J0217), the Science and Technology Projects of Guangzhou (SL2022A03J00358), National Natural Science Foundation of China (No. 82402296), Guangzhou Basic and Applied Basic Research Scheme (2025A03J3213) and 2023 Guangdong Province Hospital Association Ultrasound Medicine Research Special Fund (funded by the Guangdong Province Yi Yang Health Charity Foundation, 2023CSM004).

Data availability

The cervical cancer (CC) scRNA-seq dataset file of GSE208653 was downloaded from the Gene Expression Omnibus (GEO) database ( [https://www.ncbi.nlm.nih.gov/geo/] ). The bulk RNA-seq cohorts were obtained from the TCGA database ([https://portal.gdc.cancer.gov/] ) and the Genotype-Tissue Expression (GTEx) database ( [http://gtexportal.org/home/] ). All data supporting the conclusions are available from the authors on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the ethics committee of the Third Affiliated Hospital of Sun Yat-sen University (No. II2023-175) and written informed consent was obtained from each patient.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.