C3⁺ cancer-associated fibroblast-derived GAS6 promotes cervical cancer metastasis via activation of AXL signaling pathway in a syngeneic orthotopic mouse model

Abstract

Background

Metastatic cervical cancer remains a leading cause of cancer-related death in women, although several mouse models of metastatic cervical cancer exist, many fail to preserve an immunocompetent primary tumor microenvironment, limiting their ability to adequately model metastatic progression. This study aimed to establish a syngeneic orthotopic cervical cancer model that faithfully reproduces the metastatic cascade and defines cellular interactions associated with metastasis in tumor microenvironment.

Methods

U14 cells derived from mouse cervical cancer were engineered to stably express luciferase (U14-Luc) for the detection of micrometastases via bioluminescence imaging. The syngeneic orthotopic mouse model was established by intra-cervically injecting U14-Luc into immunocompetent C57BL/6 mice under direct visualization. Primary and metastatic tumor burden were detected by in vivo or ex vivo organ bioluminescence imaging. Histological analysis of tumor tissues was performed using Hematoxylin-Eosin staining. Single-cell RNA sequencing was used to identify cellular interactions and molecular pathways in primary tumors (PTs) and paired lymph node metastasis tumors (LNMTs).

Results

In this mouse model, primary tumors consistently developed near the external cervical os and vaginal fornix. By day 14 post-injection, cancer cells had invaded into the pelvic viscera, particularly the rectum, and disseminated to regional lymph nodes. By day 28, distant metastases, predominantly in the lungs, were observed. Single-cell analysis revealed that crosstalk between cancer cells and T cells in LNMT promoted immune escape via the SPP1 pathway. Additionally, a C3⁺ fibroblast subtype enriched in the PTs was correlated with unfavorable prognosis in cervical cancer patients. These C3⁺ fibroblasts enhanced tumor invasiveness through activation of the GAS6/AXL pathway.

Conclusions

This syngeneic orthotopic mouse model effectively replicates the metastatic cascade of cervical cancer. Its ability to closely mimic the tumor microenvironment makes this model a valuable immunocompetent orthotopic platform for investigating the metastatic mechanism and evaluating novel therapeutic strategies for cervical cancer.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08392-8.

Background

Despite global efforts to promote HPV prevention, cervical cancer remains the fourth leading cause of cancer-related deaths in women due to its aggressive nature [1, 2]. Understanding the mechanism of cervical cancer invasion and metastasis is crucial for developing effective therapeutic strategies. However, traditional mouse models of cervical cancer, such as subcutaneous xenografts [3, 4], random peritoneal seeding [5–7] and footpad tumor models [8, 9], fail to replicate the immunocompetent primary tumor (PT) microenvironment and do not effectively reflect the aggressive metastatic behavior of cervical cancer cells. Accordingly, it is calling for a syngeneic orthotopic mouse model which more accurately mimic the metastatic cascade of cervical cancer.

There are two primary methods for generating orthotopic mouse models of cervical cancer: (i) Laparotomy-based orthotopic transplantation or intracervical injection [10, 11]; (ii) transvaginal allograft implantation [12] or chemical carcinogen instillation [13, 14] onto mechanically abraded cervical epithelium without direct visualization. However, each method has limitations. Orthotopic transplantation of tumor tissue dose not model clinical progression through precancerous lesion [15]. Intracervical injection, performed under direct laparotomy, frequently results in abdominal and cervical trauma, leading to unintended tumor cells extravasation and peritoneal seeding. Transvaginal models, generating under non-visualized conditions, can cause inaccurate tumor cell colonization, such as in the submucosal layer of the vaginal wall. Chemical-induction protocols also introduce safety risks to both animals and researchers.

Genetically engineered mouse models, such as K14 transgenic mice that express HPV16 E6/E7, can develop gynecologic malignancies spontaneously after 6 months when treated with high-dose estrogen chronically [16]. While this spontaneous tumorigenesis model aims to recapitulate the continuum from HPV infection to cervical carcinoma, it exhibits prolonged latency of cervical tumorigenesis, and central tolerance to HPV16 E6/E7, markedly attenuating its suitability for modeling metastatic process. Moreover, prior to the development of cervical neoplasia, K14-transgenic animals often develop esophageal squamous cell carcinoma or head-and-neck tumors, and frequently succumb to cachexia, complicating their use in metastatic cervical cancer research [17]. Collectively, the inherent limitations of existing preclinical cervical cancer models significantly hinder their ability to accurately recapitulate cancer metastasis. This underscores the need for improved preclinical models that more faithfully mirror the invasive and metastatic characteristics of cervical cancer.

Here, we report the establishment of a syngeneic orthotopic mouse model using U14 cells, originally derived from a methylcholanthrene-induced murine cervical squamous carcinoma [18]. Although this model lacks high-risk HPV infection, mice transplanted with luciferase-expressing U14 cells (U14-Luc) faithfully mimic clinical disease progression, including local invasion, lymph node metastasis tumors (LNMTs) and distant metastasis. Single-cell RNA sequencing (scRNA-seq) analysis of PTs and paired LNMTs from this model revealed that crosstalk between cancer cells and T cells in LNMT promotes tumor immune evasion through the SPP1 pathway. Moreover, C3⁺ fibroblasts, enriched in the PTs, were identified. C3⁺ fibroblasts-derived Gas6 may enhance the invasive ability of cervical cancer cells through activating the AXL pathway. Clinical analysis further showed that C3⁺ fibroblasts are associated with poor prognosis in cervical cancer patients. This syngeneic orthotopic mouse model closely mimics the tumor microenvironment, providing a reliable preclinical platform for investigating metastatic mechanisms and evaluating novel therapeutic strategies for cervical cancer.

Methods

Cell culture

The U14-Luc, and Hela cell lines were cultured in an atmosphere of 5% CO2 at 37℃ in complete DMEM (Procell, China) supplemented with 10% FBS (Yeasen Biotechnology, China) and 1% penicillin-streptomycin (Beyotime Biotechnology, China). All cell lines were authenticated by short-tandem repeat profiling within six months prior to study initiation and were cultured for no longer than one month in culture.

Preparation and infection of luciferase virus

Lentiviral particles were produced by co-transfecting HEK293T with the pSPAX2, pMD2.G and luciferase expression plasmid by using PEI [19]. The cell cultured medium was replaced after 8 h post-transfection, and the supernatants containing lentiviruses were collected 48 h later. The viral particles were filtered by the 0.45 μm filter (Guangzhou Jet Bio-Filtration, China), then separately packaged in 1 ml and stored at -80℃ until used. Luciferase tagged U14 cell line (U14-Luc) was generated by infecting with the lentiviral particles above. Infected cells were selected by 8 µg/ml G418 for 7 days and luciferase expression was confirmed by IVIS Spectrum In vivo Imaging System (PerkinElmer, USA).

Cervix orthotopic injection

Anesthesia and perioperative monitoring

5-week-old C57BL/6 mice were anesthetized by intraperitoneal injection of tribromoethanol at a dose of 20 µL/g body weight (Prosperich, China). Anesthetic depth was confirmed prior to surgery by the absence of the pedal withdrawal reflex in response to a toe pinch test, in accordance with established guidelines [20, 21]. Throughout the procedure, Perioperative monitoring was performed at 10-minute intervals and included assessment of respiratory rate, body temperature (maintained at 36–38 °C using a thermostatically controlled heating pad), and repeated evaluation of the pedal withdrawal reflex [22]. Ophthalmic ointment was applied to both eyes immediately after anesthesia induction to prevent corneal desiccation. After surgery, animals were recovered in the Trendelenburg position (head-down tilt) on a pre-warmed surface until fully ambulatory.

Surgical exposure and standardized cervical scraping

Following confirmation of a stable surgical anesthetic plane, each mouse was placed in the dorsal lithotomy position. Under strict aseptic conditions, the lower abdominal wall, vulva, and vaginal canal were thoroughly disinfected with povidone-iodine, followed by 75%(v/v) ethanol. The vaginal canal was gently dilated using fine forceps under direct visualization, and the uterus was externalized by gentle traction to expose the uterine cervix. The external cervical os was then scraped using the tip of a 31G needle held at a 45° angle to the cervical surface. Scraping was performed in a single unidirectional motion for 8–10 passes over a stroke length of approximately 1–2 mm, consistent with the dimensions of the murine external cervical os. The adequacy of scraping was assessed visually: the intact cervical epithelium appears uniformly pale pink with a smooth surface, whereas a properly prepared cervix exhibits marked mucosal edema, surface erosion, and a matte appearance. This visual endpoint was used as the standardized criterion for proceeding to injection. Inadvertent deep scraping resulting in frank hemorrhage (estimated > 50 µL by visual assessment) was avoided. If substantial bleeding occurred, hemostasis was achieved by applying gentle pressure using a sterile cotton-tipped applicator placed through the vaginal canal; The uterus was then repositioned, and the animal observed for 5–10 min prior to proceeding.

Intracervical injection and bleb formation

The submucosal space between the external cervical os and the posterior vaginal fornix was targeted for injection. A 20 µL volume of the Matrigel–cell suspension (containing 5 × 10⁵ U14-Luc cells; BD Biosciences) was injected using a 31G needle mounted on a Hamilton syringe. Because a small volume of injectate may partially reflux, prior cervical scraping was essential to create a textured adherent surface that anchors the Matrigel-based suspension and promotes engraftment. Following injection, the uterus was carefully repositioned into the peritoneal cavity, and the vagina was gently retracted and dilated for 5–10 min to permit Matrigel polymerization at body temperature. Successful bleb formation was defined as the presence of a discrete, dome-shaped Matrigel deposit encapsulating the external cervical os, which was visible upon re-examination after vaginal closure. Bleb retention was confirmed by the absence of extruded injectate at the vaginal introitus following release of the vaginal dilating forceps, indicating that the Matrigel had sufficiently polymerized and adhension to the cervical surface. Finally, the uterus was carefully returned to the abdominal cavity, and mice were maintained in a Trendelenburg position (hindlimbs elevated relative to the head) during recovery from anesthesia.

Management of failed bleb formation

Bleb failure was defined as the extrusion of visible injectate from the vaginal opening upon uterine repositioning or vaginal closure. In such cases, re-injection was avoided to prevent additional tissue trauma. The animal was assigned a designated failure code, which was recorded separately from the successful injection cohort. Animals with confirmed bleb failure rarely developed orthotopic tumors on subsequent bioluminescence imaging.

Study endpoints

The primary endpoints were the four pre-specified harvest time points: days 7, 14, 21, and 28. Humane endpoints were pre-specified prior to study initiation in accordance with institutional IACUC guidelines and international recommendations [20–22]. The following criteria were used to determine when an animal required immediate humane euthanasia: (i) body weight loss ≥ 20% relative to initial body weight recorded at study entry; (ii) inability to reach food or water due to locomotor compromise; (iii) immobility, persistently hunched posture, or severe lethargy; (iv) labored or irregular respiratory distress; (v) primary tumor size exceeding 2.0 cm in any single dimension or estimated tumor volume ≥ 2,000 mm³; (vi) persistent hypothermia (rectal temperature < 35 °C for > 2 h); (vii) tumor ulceration with active or uncontrolled bleeding; and (viii) moribund state with minimal or absent response to external stimuli. For metastatic disease phases in which clinical signs may be subtle, supplementary assessments were incorporated. These included body condition scoring (BCS ≤ 2.0 as an additional trigger threshold), standardized behavioral observation (spontaneous activity, grooming behavior, and exploratory movement), and bioluminescence imaging review for signals emerging outside the primary tumor site. Animals meeting any of the above criteria were humanely euthanized by carbon dioxide inhalation followed by cervical dislocation.

In vivo / Ex vivo organ bioluminescence imaging

In vivo bioluminescence imaging was performed using the AniView Phoenix Full-Spectrum In Vivo Animal Imaging System (Guangzhou Biolight Biotechnology, Guangzhou, China). Ex vivo organ bioluminescence imaging was performed using the IVIS Spectrum In Vivo Imaging System (PerkinElmer, Waltham, Massachusetts, USA). Prior to in vivo imaging, the abdominal fur was completely removed by shaving to minimize signal attenuation. Mice were anesthetized by isoflurane inhalation (2.0–2.5% for induction; 1.5–2.0% for maintenance in 100% O₂ at 1 L/min). D-luciferin potassium salt (150 mg/kg body weight; Absin Bioscience, Shanghai, China) was administered intraperitoneally 10 min prior to image acquisition to allow substrate distribution and peak bioluminescence emission. All images were acquired with a fixed exposure time of 1 min, and bioluminescence signals were quantified as total photon flux (photons/second; p/s) within standardized regions of interest (ROIs) drawn over the abdominal and thoracic regions.

For ex vivo organ bioluminescence imaging, D-luciferin potassium salt (150 mg/kg, i.p.) was administered 10 min prior to euthanasia. Immediately following euthanasia, the uterus and cervix (primary tumor site), bilateral inguinal and para-aortic lymph nodes, lungs, liver, spleen, kidneys, and rectum were systematically harvested and arranged on a dark non-reflective surface for imaging. Ex vivo imaging was initiated within 10 min of organ harvest to preserve substrate activity and signal integrity.

Definition of bioluminescence positivity threshold

At each harvest time-point (Days 7, 14, 21, and 28), organs from sham-operated control mice injected with Matrigel/PBS without tumor cells (n = 3 per time-point) were processed and imaged in parallel under identical acquisition conditions. Organ-specific background average radiance values derived from sham animals were used as the reference threshold for each organ type and time-point. An organ was scored as bioluminescence metastasis-positive if both of the following criteria were met: (i) the ROI average radiance clearly exceeded the corresponding sham organ background under identical acquisition conditions; and (ii) the signal was confirmed to originate from the organ parenchyma rather than from cut surfaces, adjacent tissue, or harvest-related contamination.

Histological confirmation

All bioluminescence signal-positive organs were sectioned and subjected to hematoxylin and eosin (H&E) staining for histological confirmation of metastatic tumor involvement. To evaluate the false-negative rate of bioluminescence imaging, organs from one or two bioluminescence-negative animals were also examined histologically for each organ type at each time point. Histological sectioning was performed using a five-section interval protocol (20 μm inter-section interval).

Tumor volume measurement

On each pre-designated dissection day (Day 7, 14, 21, and 28), in vivo bioluminescence imaging was performed immediately prior to sacrifice to document tumor status and provide a preliminary estimate of tumor burden. Animals were subsequently euthanized, and the uteri were surgically excised and photographed for gross morphological evaluation. Tumor dimensions — including the maximum longitudinal diameter (L) and the maximum transverse diameter (W) — were measured from gross specimen images using ImageJ software (National Institutes of Health, Bethesda, MD, USA). Tumor volume was calculated according to the following formula: V = ½ × L × W².

Histology and hematoxylin-eosin staining

Following euthanasia, organs including the heart, liver, spleen, lungs, kidneys, intestines, uterus and ovaries were excised and subjected to ex vivo bioluminescence imaging. Organs were then fixed in 4% paraformaldehyde (Servicebio Technology, China) at room temperature for more than 24 h. Fixed tissues were dehydrated, paraffin-embedded, sectioned and stained with Hematoxylin-eosin. All sections were anonymized with coded identifiers prior to distribution; The codes were held by a third party and revealed only after both pathologists had completed their independent evaluations. Whole-slide digital images were then acquired using a Digital Pathology Slide Scanner (PRO) (KONFOONG BIOINFORMATION TECH, China) and independently interpreted by two board-certified pathologists in a blinded manner to ensure the objectivity and reproducibility. Discrepancies between the two evaluators were resolved by consensus review.

Multiplexed immunohistochemistry

Sections of formalin-fixed, paraffin-embedded (FFPE) tissues from mice (normal cervix, primary tumors, and lymph node metastasis tumors) and human cervical tissue sections from the SYSUSH cohort (including uterine prolapse hysterectomy specimens and cervical cancer resections) were dewaxed in xylene, rehydrated through a graded ethanol series, and subjected to antigen retrieval using EDTA buffer (pH 9.0). Following the aforementioned pretreatment steps, sections were sequentially incubated with the following primary antibodies at 4 °C overnight: SPP1 (Proteintech, 22952-1-AP), CD8α (CST, #98941), AXL (Proteintech, 13196-1-AP), GAS6 (Proteintech, 13795-1-AP), PanCK (Proteintech, 26411-1-AP), and COL1A1 (Huabio, HA722517). Subsequently, sections were incubated with polymer-HRP-conjugated secondary antibodies at room temperature for 1 h, followed by reaction with fluorescent tyramide substrate for 5–10 min using Multiplex Fluorescence Immunohistochemistry Kit (AiFang biological, AFIHC025, China). Nuclei were counterstained with DAPI. Finally, fluorescence images were acquired using a Digital Pathology Slide Scanner (PRO) (KONFOONG BIOINFORMATION TECH, China) with corresponding laser scanning settings.

Sample collection and single-cell suspension isolation

The normal cervix and tumors from both primary and metastatic sites employed in this study were collected from mice which underwent total hysterectomy. Three animals per group were used to minimize sample variation. The mapping of individual mice to sample condition and library identifier is provided in Table S1. After surgery, fresh tissue samples were collected and preserved in the GEXSCOPETM Tissue Preservation Solution (Singleron, Germany) for subsequent processing. Tissue samples were initially washed with phosphate-buffered saline (Servicebio Technology, China), then cut into small pieces of 1–2 mm, and digested using Enzyme cocktail (5 mL DMEM supplemented with 2 mg/mL collagenase I, 1 mg/mL collagenase II, 2 mg/mL collagenase IV, 0.25 mg/mL hyaluronidase, 1 mg/mL neutral protease, and 0.05 mg/mL DNase I, containing Ca²⁺, Mg²⁺, and Zn²⁺) at 37 °C with oscillation for 40 min. Following digestion, the samples were passed through a 70-µm sterile strainer (Servicebio Technology, China), and the supernatant was removed by centrifugation at 300 g for 5 min at 4 °C. The cell pellet was then treated with 1 mL red blood cell lysis buffer (Beyotime Biotechnology, China), and incubated for 10 min at 4 °C to lyse the red blood cells. Lastly, the single-cell suspension was collected, resuspended in PBS, adjusted the final cell concentration to 1,000–2,000 cells/µL. and subjected to further analysis.

Single-cell RNA sequencing and raw data preprocessing

Cellular suspensions were loaded on a DNBelab C series Single-cell instrument that generates single-cell droplet. Libraries were generated and sequenced from the cDNAs with DNBelab C series High-throughput Single-cell RNA Library Preparation Set V3.0 (TaiM 4). After the droplet generation, the Cell Beads, which have captured the cell mRNA and droplet index information, are filtered and collected for subsequent sequencing library construction experiments. The Cell Beads were then subjected to a series of steps, including droplet-RT cleaning up & cDNA amplification, library construction and sequencing fragment selection. Using the DNBSEQ-T7RS paired-end sequencing mode, the constructed libraries undergo high-throughput sequencing. DNBC4tools (version 2.1.3) was used to quality control, alignment and counts quantification for the raw FASTQ files. Reads with low-quality barcodes and UMIs were filtered out and then mapped to the reference genome (GRCm39). Reads uniquely mapped to the transcriptome and intersecting an exon at least 50% were considered for UMI counting. Before quantification, the UMI sequences would be corrected for sequencing errors, and valid barcodes were identified based on the EmptyDrops method [23]. The cell by gene matrices were produced via UMI counting and cell barcodes calling.

Clustering and cell-type annotation

After cell filtering, the resulting feature and count matrix were then imported into Seurat software (v4.4.0) for quality control, normalization, scaling, dimensionality reduction and clustering. The matrix was used to build a Seurat object with min.cells = 3 and min.features = 100. Doublets were identified and removed in each sample separately using Scrublet (v0.2.3) [24]. Subsequent quality control filtering was performed in merged data to retain only high-quality cells with nFeature_RNA > 250, nFeature_RNA < 6000, nCount_RNA > 500, percent.mt < 10, and percent.hb < 1. Data normalization was performed using the LogNormalize method with a scale factor of 10,000. The top 2,000 highly variable genes were identified using the variance stabilizing transformation (VST) method. To remove technical and biological confounding effects, scaling was performed with regression of nuisance variables including nCount_RNA and percent.mt, using all genes as features. To integrate data across multiple samples, the Seurat integration workflow was applied. Briefly, the object was split by sample, and integration anchors were identified using FindIntegrationAnchors, followed by IntegrateData. After scaling and PCA, Harmony was applied to further remove batch effects attributable to sample origin. For visualization of clusters, UMAP and t-SNE were then generated using the Harmony-corrected reduction.

The log-normalized matrices were then loaded on SingleR R packages for cell type annotation, which based on correlating gene expression of reference cell types with single-cell expression. The FindAllMarkers function of the Seurat package was used to identify differentially expressed genes (DEGs) between clusters (Table S2). Parameters for FindAllMarkers included min.pct = 0.1, logFC.threshold = 0.25, and only pos = T. To complement and validate the SingleR annotations, the identified marker genes for each cluster were additionally queried against the CellMarker 2.0 database (http://bio-bigdata.hrbmu.edu.cn/CellMarker/) [25].

Non-parametric Wilcoxon rank-sum test was performed to calculate p-values, with p-values < 0.05 after Bonferroni correction considered significant. Established markers were used, including epithelial cells (Krt18, Krt14, Stmn1 and Mki67 positive), fibroblast cells (Dcn, Col3a1, and Col1a1 positive), endothelial cells (Pecam1, Cdh5, and Flt1 positive), neutrophils (S100a8, S100a9 and Cd44 positive), macrophages (Lyz2, Apoe and Cd74 positive), DCs (Cd83, Ccr7 and Cst3 positive), B cells (Cd79a, Ms4a1 and Ptprc positive), and T cells (Cd3d, Cd3e positive).

Differential gene expression and functional enrichment analysis

The wilcoxauc function of presto R package (v1.0.0) was used to identify DEGs of U14 between primary and metastatic niches, as well as DEGs of C3⁺ Fibroblast between normal cervix and primary tumor. To assess differences in pathway activity across each cell type, the Gene Set Enrichment Analysis (GSEA) was employed to calculate pathway activity differences for each cell type. The R package clusterProfiler (version 4.14.6) was used to perform gene set enrichment analysis and retrieved the gene sets from the KEGG database.

CNV estimation

Cells defined as epithelial cells, U14 cells, and T cell were used as references to identify somatic CNVs with the R package inferCNV (version 1.22.0). The analysis was performed with a denoising cutoff of 0.1, clustering cells by their group labels enabled, and Ward’s D2 method used for hierarchical clustering. The analysis mode was set to subclusters to enable subcluster-level CNV detection, with spatial denoising applied to reduce noise and a hidden Markov model employed to predict CNV states. Each cell was scored according to the extent of the CNV signal, defined as the mean of the squares of the CNV values across the genome. Putative malignant cells were defined as those with a CNV signal > 0.05 and CNV correlation > 0.5.

CytoTRACE

CytoTRACE (v0.3.3), a computational method, was used to predict the differentiation state of cells from scRNA-seq data by counting the number of detectably expressed genes per cell. The resulting CytoTRACE scores were visualized using the plotCytoGenes function to display the top 10 genes correlated with differentiation status, and the plotCytoTRACE function to project the scores onto the UMAP embedding, with cell phenotypes (group and epitype combination) used for coloring. All other parameters were kept at their default values.

Cell–cell communication analysis

Cellular interaction analysis was conducted with CellChat (version 1.6.1), a public repository of ligands, receptors and their interactions, which enables systematic analysis of cell–cell communication at the molecular level. A CellChat object was created using the create CellChat function, with cell types as the grouping variable. The mouse CellChat database (CellChatDB.mouse) was loaded and assigned to the object. After subsetting the expression data, overexpressed genes and overexpressed interactions were identified using identify OverExpressedGenes and identify OverExpressedInteractions, respectively. The membrane, secreted and peripheral proteins were annotated. Based on the interaction and the normalized cell matrix achieved by Seurat normalization, the rank and mean significance were calculated (p-value < 0.05). Default parameter settings were applied for CellChat analysis, including permutation number = 100 and p-value threshold = 0.05 for significant interaction inference.

Boyden chamber assays

Cell migration and invasion were assessed using 24-well Boyden chambers (NEST, China) equipped with 8 μm pore-size inserts. For murine experiments, U14 cells (5 × 10⁴) were used; for human experiments, HeLa (human, HPV18-positive) and C33A (human, HPV-negative) cells (1 × 10⁵) were used. Cells were suspended in serum-free DMEM and seeded into the upper chamber. The lower chamber contained DMEM supplemented with 10% FBS (migration assay) or 20% FBS (invasion assay). Inserts were left uncoated for migration assays or pre-coated with growth factor-reduced Matrigel (BD Biosciences, USA) for invasion assays. Cells suspended in 200 µL of serum-free DMEM were seeded into the upper chamber. For AXL pathway activation experiments, recombinant human GAS6 (TargetMol, China) was added to lower chambers at a final concentration of 500 ng/mL. For AXL inhibition experiments, cells were pre-treated with the selective AXL inhibitor Bemcentinib (TargetMol, China) at a final concentration of 2 µM, and the inhibitor was maintained in lower chambers throughout the assay. All chambers were incubated at 37 °C in a 5% CO₂ atmosphere for 12 h. Following incubation, non-migrating cells on the upper surface of the membrane were removed by gentle swabbing. Cells that had traversed the membrane were fixed in 4% paraformaldehyde for 20 min and stained with 0.005% crystal violet solution (Sigma-Aldrich, USA) for 15 min. Representative images were captured using a Nikon phase-contrast microscope, and migrated or invaded cells were quantified using ImageJ software (version 1.54 g; NIH, USA).

Survival analysis

Survival analyses of C3⁺ fibroblast were conducted with the cervical and endocervical cancer (CESC) data from GEPIA3 (https://gepia3.bioinfoliu.com/). The top 20 differentially expressed genes (DEGs) of C3⁺ fibroblasts were used to construct a prognostic signature for patient survival, with the optimal high/low cutoff values determined by the quartile method. Specifically, the top 20 DEGs were: Dcn, Col1a2, Gsn, Col1a1, Serping1, Col3a1, Htra3, Sparc, Mmp2, C3, Dpt, Mgp, Cygb, Timp2, Clec3b, Col6a1, Plac9, Bgn, Igfbp4, Fbln1.

Human cohort description

The SYSUSH cohort comprised formalin-fixed, paraffin-embedded (FFPE) tissue samples from 3 normal cervices tissues obtained from patients undergoing hysterectomy for benign gynecological conditions and 6 locally advanced cervical cancer specimens (FIGO stages III–IV) obtained from patients who underwent surgical resection at The Seventh Affiliated Hospital (Shenzhen), Sun Yat-sen University. None of the cancer patients received neoadjuvant chemotherapy or radiotherapy prior to surgery. Detailed clinical and pathological characteristics — including age, FIGO stage, histological subtype, and treatment history — for each patient are provided in Table S3. The study was approved by the Institutional Review Board of The Seventh Affiliated Hospital (Shenzhen), Sun Yat-sen University (Approval No. KY-2024-036-01), and written informed consent was obtained from all participants.

Statistical analysis

All quantitative data are presented as mean ± SEM or mean ± SD, as specified in the corresponding figure legends. The exact number of biological replicates (n) is stated for each figure, with biological replicates defined as independent animals (for in vivo experiments) or independent experimental repetitions performed on separate days (for in vitro experiments), clearly distinguished from technical replicates.

For comparisons between two independent groups, a two-tailed unpaired Student’s t-test was used when the data met normality and equal-variance assumptions (assessed by Shapiro-Wilk test and F-test, respectively). For comparisons involving more than two groups, one-way ANOVA was performed, with Tukey’s post-hoc test applied for all pairwise comparisons for comparisons against a single control group; the correction method used is specified in each figure legend. For figures involving only two-group comparisons, no additional multiple comparison correction was applied, and this is stated in the corresponding legends. Survival curves were generated using the Kaplan-Meier method and compared by the log-rank test; hazard ratios and their 95% confidence intervals are reported. Effect sizes, including mean differences or Cohen’s d for continuous outcomes and hazard ratios for survival outcomes, are reported with 95% confidence intervals where feasible. For single-cell RNA-seq differential expression analysis, significance was assessed by non-parametric Wilcoxon rank-sum test with Bonferroni correction applied for multiple comparisons; adjusted p < 0.05 was considered significant. No additional multiple comparison correction was applied for independent in vitro or in vivo experimental comparisons, and this is explicitly noted in the respective figure legends. All statistical analyses were performed using GraphPad Prism (version 9.0) or R software (version 4.1.0). A p < 0.05 was considered statistically significant throughout, with significance levels indicated as: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Results

Construction of the U14-Luc orthotopic mouse model

To establish a tractable murine model permitting real-time monitoring of tumor progression, the U14 cervical cancer cell line [18], known for its high tumorigenicity, was transduced to stably express luciferase. The luciferase tagged U14 cell line (U14-Luc cell line) was then used to establish the orthotopic cervical tumor model, allowing for real-time monitoring the tumor growth and metastasis by bioluminescence imaging (Fig. 1A).

Fig. 1.

Orthotopic Model construction process and body weight monitoring in mice . A Schematic depicting the generation of the U14-Luc cervical carcinoma cell line and the orthotopic mouse model. U14 cells were transduced with a luciferase-expressing lentiviral vector to generate the U14-Luc cell line (left). A total of 5 × 10⁵ U14-Luc cells suspended in 20ul of Matrigel were injected into the cervix of mice to establish an orthotopic cervical tumor model (centra). Subsequent bioluminescence imaging enabled real-time monitoring of primary tumor growth, local invasion, lymphatic spread, and distant organ metastasis (right). Created in https://www.BioRender.com. B Orthotopic model construction process. Under sterile conditions, the lower abdomen, vulva and vaginal canal were disinfected with povidone–iodine then 75% ethanol (a). Sequential vaginal retraction elicited complete uterine prolapse and exposed the cervical OS (b–c). After epithelial disruption with a needle, 5 × 10⁵ U14-Luc cells in chilled Matrigel were injected intra-cervically (d). C Body weight (g) growth curve of sham group (n=12) and model group (n=18) were recorded from day 3 to day 28 post-injection

On the day of injection, the lower abdominal wall, vulva, and vaginal canal of C57BL/6 mice were disinfected thoroughly (Fig. 1Ba). Secondly, the vaginal walls were sequentially grasped and gently retracted until complete artificial prolapse of the uterus was achieved, allowing full visualization of the cervical os (Fig. 1Bb). Thirdly, the cervix was gently scraped with a needle to disrupt the epithelium barrier (Fig. 1Bc), U14-Luc cells were resuspended in Matrigel and then injected intra-cervix. 5–10 min was needed to wait till Matrigel bleb formation, which is important for successful model construction (Fig. 1Bd).

Among 24 mice injected for the longitudinal cohort, primary tumors were confirmed by bioluminescence imaging in 18 animals, yielding a take rate of 75% (18/24). First detectable signal was observed at Day 7 in 17 of 18 tumor-bearing mice, and at Day 14 in the remaining mouse; no additional mice converted after Day 14. The median time to first detectable signal was 7 days (range: 7–14 days), and the mean total photon flux of tumor burden was 6.56 × 10⁴ p/s (SD: 5.32 × 10⁴ p/s). Additionally, the primary tumor take rate across independent batches ranged from 53.3% to 75.0% (mean: 62.8%, Table S4).

Over the 28-day body weight monitoring period, the model group showed no significant loss of weight compared to the sham group (Fig. 1C, Table S5), indicating that the orthotopic injection of U14-Luc cells to establish orthotopic cervical cancer hardly induces cachexia of tumor-bearing mice, but allowing the stable tumor progression with metastatic characterizes, which is similar to human cervical cancer. Except for one mouse, showing gradual weight loss and dying on day 20 after injection. Further examination confirmed the predominant reason of mortality was multi-organ metastasis (Fig. 3B, Table S6). Accordingly, body weight surveillance is regarded as a critical component of longitudinal monitoring, as any unexplained weight loss or manifestation of cachexia frequently signals the onset of distant organ metastasis.

Fig. 3.

Multi-directional metastasis patterns of the orthotopic U14-Luc cervical cancer model. A Representative ex vivo bioluminescence images and tabulated data quantify rectal invasion and lymph node or pulmonary metastasis at day 7 (n = 5), 14 (n = 5), 21 (n = 5) and 28 (n = 3). B Representative gross pathology and ex vivo images of a mouse showing dramatically loss of weight, illustrating multi-organ metastasis, including urinary system, rectum, lymph node, lung, liver, spleen

Primary tumor growth and invasion patterns in the U14-Luc orthotopic cervical cancer model

In the U14-Luc orthotopic mouse model, tumor formation was first detectable by bioluminescence imaging on day 7. Serial in vivo imaging of a representative mouse maintained until day 28 showed a progressive increase in signal intensity over time (days 7, 14, 21, and 28) (Fig. 2A). Consistently, the excised tumor volumes measured at the same four time points also increased over time (Fig. 2B, Table S7). Histological examination was further performed to confirm tumor progression. Uterus isolated from U14-Luc-injected mice or sham mice were divided into three contiguous segments for evaluation: the upper segment (from the ipsilateral uterine horn and ovary to the uterine–cornual junction), the middle segment (from the uterine body to the external cervical os), and the lower segment (from the external cervical os to the vaginal orifice) (Fig. 2C). HE staining of tissues derived from U14-Luc-injected mice revealed marked cellular enlargement, hyperchromatic nuclei and pronounced nuclear pleomorphism, which are hallmarks of malignancy, not observed in control tissues (Fig. 2D and E).

Fig. 2.

Tumor growth and invasion patterns in the U14-Luc orthotopic cervical cancer model. A Representative images of serial in vivo bioluminescence images of sham group (Left) and model group (Right) from day 7 to day 28. B Tumor volume progression in the orthotopic cervical cancer model over time. Tumor volumes were measured at Days 7, 14, 21, and 28 post-injection (n = 5 mice for Days 7, 14, 21; n=3 mice for Day 28). C. Schematic representation of uterine segment sectioning and embedding strategy for histological analysis, depicting coronal sections of the upper segment and transverse sections of the middle and lower segments. Created in https://www.BioRender.com. D H&E staining depicts the tissue architecture of serially sectioned uterine segments in sham-operated mice. (Left) Schematic of tissue embedding orientation: coronal for upper, transverse for middle/lower. (Right) Corresponding histology reveals distinct anatomical features across uterine regions. Created in https://www.BioRender.com. E. Representative images of HE stained segmental uterine sections from tumor-bearing mice at day 7, 14, 21 and 28 illustrate the stepwise invasion trajectory of orthotopic cervical carcinoma

Although prominent bioluminescence was observed on in vivo imaging at day 7, no grossly discernible masses or nodules were evident. Nevertheless, manual palpation revealed indurated foci localized to the external cervical os and adjacent vaginal fornix. Histological examination demonstrated that tumor cells were retained at the injected position, in the vicinity of the external cervical os and vaginal fornix (Fig. 2Ea). By day 14, continued tumor expansion resulted in modest protrusion and enlargement of the external cervical os and vaginal fornix. Histological assessment revealed a marked increase in tumor cells that remained restricted to the external cervical os and upper vaginal segment, without extension into the mid-vagina, although focal disruption of the basal layer was evident (Fig. 2Eb). By day 21, tumor had enlarged, occupying the upper vaginal canal, and focal central necrosis was also observed. Notably, tumor cells were detected in both the middle and lower segment of uterus, indicating invasion into the mid-to distal vaginal segment (Fig. 2Ec). By day 28, gross examination revealed near-total transmural tumor infiltration with complete effacement of the vaginal lumen (Fig. 2Ed). Histological evaluation demonstrated mucosa-to-serosa neoplastic extension and confluent central necrosis.

U14-Luc orthotopic model recapitulates cervical cancer metastasis

To assess the metastatic pattern of U14-Luc orthotopic model, ex vivo bioluminescence imaging was performed on 18 tumor-bearing mice over time. Rectal invasion was first detected on day 7. At this timepoint, histology confirmed that tumor cells were retained at the injection site near the external cervical os/vaginal fornix (Fig. 2Ea). Consistent with progressive spread, bioluminescence signals in regional lymph nodes were observed by day 14, and lung metastasis by day 21 (Fig. 3A). Compared to earlier time points (days 7 and day 14), a significant increase in rectal invasion and lymph node metastasis tumor was observed at later stages (days 21 and day 28), with five out of eight mice showing these metastatic events. This pattern suggests a transition from sporadic early-stage spread to more consistent late-stage progression driven by tumor expansion and transmural infiltration.

Three out of eight tumor-bearing mice were first detected pulmonary metastases after 21 days, indicating that during the late phase of model establishment, tumor cells had invaded into the vasculature and subsequently colonized in the pulmonary parenchyma.

Collectively, among the 18 tumor-bearing mice subjected to ex vivo bioluminescence imaging, rectal invasion was identified in seven, LNMT in six, and lung metastasis in three, with two exhibiting all three sites of metastasis (Table S6). Notably, one mouse that experienced progressive weight loss and showed a moribund state on Day 20 (one day prior to the scheduled harvest) was humanely euthanized. Subsequent ex vivo bioluminescence imaging revealed widespread multi‑organ metastasis involving the urinary system, rectum, lymph nodes, lungs, liver, and spleen (Fig. 3B).

Single-cell transcriptomic landscape of PT and LNMT in the cervical cancer orthotopic mouse model

To comprehensively characterize the tumor microenvironment in our orthotopic mouse tumor model, scRNA-seq was performed on PTs and paired LNMTs (n = 3 per group), using normal cervical tissues of mouse as the control group (Fig. 4A). A total of 76,360 cells passed quality control, from which eight distinct cell types were identified using classical cell markers (Fig. 4B and C, Table S8). These cell types included epithelial cells (Krt18, Krt14, Stmn1 and Mki67), fibroblasts (Dcn, Col3a1, and Col1a1), endothelial cells (Peacam1, Cdh5, and Flt1), neutrophils (S100a8, S100a9 and Cd44), macrophages (Lyz2, Apoe and Cd74), dendritic cells (Cd83, Ccr7 and Cst3), B cells (Cd79a, Ms4a1 and Ptprc), and T cells (Cd3d, CD3e). To determine the enrichment of cell types in PT and LNMT, the proportional abundance of these cell types across the different tissue samples was analyzed (Fig. S1A). Epithelial cells, fibroblasts, endothelial cells and neutrophils were enriched in PT. In contrast, Dc cells and T cells, the well-known cellular components of normal lymph node, were enriched in the LNMT microenvironment compare to PT (Fig. 4D). These data suggested that the cell types in the microenvironment of LNMT differed from those in PT.

Fig. 4.

Single-cell transcriptomic landscape of PT and LNMT in the cervical cancer orthotopic mouse model. A Workflow diagram illustrating the experimental and computational pipeline for single-cell profiling of the orthotopic cervical cancer model. Single-cell suspensions were generated from six mice at day 30 post intra-cervix injection. Normal cervix (n = 3), and paired primary tumor (PT, n = 3) and lymph node metastasis tumor (LNMT, n = 3), a total of nine samples were collected and subjected to barcoding and sequencing. Subsequent analyses included cell type annotation, intercellular communication, differential gene expression, and functional enrichment. Created in https://www.BioRender.com. B UMAP plot of all single cells from the 9 murine samples (normal cervix, PT, and LNMT; n = 3 per group). Eight distinct cell types were identified. C Dot plot of marker genes defining the eight identified cell populations. D Bar plot showing proportional differences of various cell types across normal cervixs, PT and LNMT (n = 3, per group). E UMAP plot of distribution of epithelial cells in the 3 tissue types (normal cervix, PT, and LNMT; n = 3 per group), distinguishing normal cervical epithelial cells from malignant epithelial cell (U14) populations. F Dot plot of marker genes defining the normal cervical epithelial cells and malignant epithelial cell (U14). G CytoTRACE-predicted differentiation potential of malignant (U14) versus normal epithelial cells in normal tissues, PT and LNMT

To further validate the successful establishment of the model at the single-cell level, epithelial cells were classified and re-clustered using a signature gene set of U14 cells Table S9) [26]. A distinct separation was observed between normal cervical epithelial cells and U14 cells within the orthotopic tumor samples (Fig. 4E). Furthermore, the expression levels of markers including Krt18, Krt8, Vim, Krt14, Krt5, and Krt6a enabled the discrimination between normal cervical epithelium and U14 cells (Fig. 4F). CytoTRACE and copy number variation (CNV) analyses of normal cervical epithelium and U14 cells further demonstrated that U14 cells exhibit a higher stemness/CNV compared to normal cervical epithelium (Fig. 4G, Fig.S1B-E).

SPP1 suppresses CD8 + T cells infiltration in LNMT

Since lymph nodes are central hubs for immune cell circulation and activation, we next investigate how metastatic tumor cells survive within this environment. Cell-cell communication analysis revealed that T cells were the predominant signal receivers in LNMTs (Fig. 5A). Compared to PTs, T cells in LNMT received substantially higher levels of incoming signals (Fig. 5B), promoting us to focus on U14 cells in metastatic lesions as the key signal-emitting cells and to identify the signaling pathways mediating their interactions with T cells. Five major signaling pathways—VEGF, ANNEXIN, SPP1, LAMININ, and MIF—were mainly emitted by U14 cells, with SPP1-related signaling being the predominantly pathway received by T cells (Fig. 5C). In LNMT, U14 cells exhibited stronger SPP1 signaling output compared to macrophages, another cell type previously reported as a major source of SPP1 (Fig. 5D) [27–29]. The specific ligand-receptor pairs mediating this interaction were identified as SPP1–(ITGA4 + ITGB1), SPP1–(ITGA5 + ITGB1), SPP1–(ITGAV+ ITGB1), SPP1–(ITGAV+ ITGB3), and SPP1–CD44, with the SPP1–CD44 pair representing the highest-confidence interaction between U14 cancer cells and T cells (Fig. 5E).

Fig. 5.

SPP1 suppresses CD8⁺ T cells infiltration in LNMT. A. CellChat analysis of LNMT interactions, highlighting T cells as the predominant signal-receiving populations. B Circle plot from CellChat analysis depicting cell-cell interactions that are enhanced (red) or weakened (blue) in LNMT compared to the PT. Arrows indicate the direction of signaling. C CellChat heatmap depicting the landscape of signaling pathways in LNMT. Left: outgoing signaling patterns; Right: incoming signaling patterns. D Heatmap of SPP1 signaling in the LNMT, identifying U14 cells as the primary sender and T cells as the key receiver. E Bubble plot depicting SPP1 ligand-receptor interaction pairs between U14 cancer cells and indicated recipient cell populations in LNMT, inferred by CellChat analysis. Bubble size reflects communication probability; color indicates interaction strength (red: maximum; blue: minimum). Only interactions with p < 0.01 are shown. F Representative H&E and multiplex immunofluorescence (mIHC) images of LNMT zones with mild (upper) or severe (lower) tumor infiltration, stained for DAPI (blue), pan-CK (yellow), SPP1 (red), and CD8 (green; scale bar = 100 μm). G Quantification shows significantly higher SPP1⁺pan-CK⁺ cell abundance [Mean diff (95% CI): 7.720 (5.654, 9.785)] and lower CD8+ T cell density [Mean diff (95% CI): -0.8674 (-1.046, -0.6889)] in zones of severe tumor infiltration [n = 3 mice; unpaired t-test; ***p < 0.001]. H Representative H&E and mIHC images of PT and LNMT sections stained for pan-CK (yellow) and SPP1 (red; scale bar = 100 μm). I Quantification shows SPP1⁺pan-CK⁺ cell abundance [Mean diff (95% CI): 1.652 (0.6503, 2.654)] and SPP1 signal intensity [Mean diff (95% CI): 0.5315 (0.09059, 0.9725)] are both significantly elevated in LNMT compared to PT (n = 3 mice; unpaired t-test; *p < 0.05)

Multiplexed immunohistochemistry (mIHC) of LNMT from this model revealed higher SPP1 expression within metastatic lesions, accompanied by decreased CD8⁺ T cells infiltration, compared to adjacent lymph node region, leading to an immunosuppressive microenvironment (Fig. 5F and G, Fig.S2A). Moreover, upregulation of SPP1 was observed in LNMTs, compared to PTs, indicating tumor cells metastasized to lymph nodes may actively secrete SPP1 to inhibit T cell function and infiltration (Fig. 5H and I).

C3⁺fibroblasts promote cervical cancer invasion and metastasis via the GAS6-AXL pathway

We next sought to investigate whether C3⁺ fibroblasts promote tumor invasion and metastasis via the GAS6-AXL pathway. Cell-cell communication analysis showed that stromal cells, including fibroblasts and endothelial cells, exhibited the strongest outgoing interactions in PTs, indicating their central role in tumor microenvironment (Fig. 6A). Endothelial cells primarily facilitate the supply of essential nutrients to tumor cells and form metastatic tracks during invasion and metastasis, while the contribution of fibroblasts in regulating lymph node metastasis of tumor cells in this model remains unclear. Therefore, fibroblasts were further subjected to re-cluster. We obtained 4 subclusters of fibroblasts through unsupervised re-clustering (Fig. 6B): C3⁺ Fibroblasts (Col1a1, Dcn, Mgp and C3.etc), Myofibroblasts (Actg2, Cnn1, and Tagln.etc), Crip1⁺ fibroblasts (Acta2 and Crip1.etc) and Stmn1 fibroblasts (Stmn1, Col17a1 and Spink5.etc) (Fig. 6C). It is noteworthy that the proportion of C3⁺ fibroblasts was significantly enriched in PTs compared to normal cervical tissue (Fig. 6B), suggesting that C3⁺ fibroblasts may play a critical role in modulating the microenvironment within the primary tumor microenvironment. We next examined how C3⁺ fibroblasts communicate with tumor cells. C3⁺ fibroblasts interact with epithelial cells through several signaling pathways, including GAS, TENASCIN, THBS, and COLLAGEN (Fig. 6D and E, Fig. S3), but communicate with U14 cells (the malignant epithelial cells in the primary site) exclusively via the GAS pathway (Fig. 6F). Differential expression analysis of key GAS pathway genes (such as Gas6 and Axl) across different cell types confirmed that C3⁺ fibroblasts were the primary source of the ligand GAS6, while U14 cells highly express its receptor AXL (Fig. 6G). The mIHC staining for C3, COL1A1, GAS6, AXL, and pan‑CK further demonstrated enhanced GAS6-AXL distribution in PTs from this model compared to the normal tissues (Fig. 6H-I, Fig. S2B), supporting active pathway engagement in PTs.

Fig. 6.

C3⁺ fibroblasts promote cervical cancer invasion and metastasis via the GAS6-AXL pathway. A. CellChat analysis of the primary tumor. Bubble plot displays interaction counts (Left); Circle plot illustrates intercellular communication networks (Right), highlighting fibroblasts and endothelial cells as the predominant signal-sending populations; B.UMAP plot of distribution of four clusters of fibroblasts (colored by clusters) in normal cervix and primary tumor; C. Dot plot showing expression of marker genes characterizing the four identified fibroblast subpopulations; D. CellChat heatmap depicting the landscape of outgoing signaling patterns in primary tumor. Ten signals are predominantly emitted by C3⁺ fibroblasts as the major signaling source. E. CellChat heatmap depicting the landscape of incoming signaling from C3⁺ fibroblast in normal cervix and primary tumor. Compared to the normal cervix, GAS signaling from C3⁺ fibroblasts to epithelial cells exhibits a more pronounced enhancement in LNMT. F. Heatmap of GAS signaling in LNMT, identifying U14 cells as the primary sender and T cells as the key receiver; G.Violin plot showing gene expression levels of key molecules in the GAS signaling pathway within PT; H.Representative H&E and multiplex immunofluorescence staining of normal cervix and PT. Compared to normal cervix, PT demonstrate upregulated AXL in epithelial cells and increased GAS6-AXL interaction compared to normal tissue. Red arrows indicated the interaction of GAS6-AXL. (n=3, Scale bar = 100μm). I. Quantification of AXL⁺pan-CK⁺ tumor cell abundance [Mean diff (95% CI): 8.035 (5.445, 10.62)] (left) and GAS6⁺C3⁺COL1A1⁺ fibroblast abundance [Mean diff (95% CI): 2.133 (0.7910, 3.474)] (right) in normal cervix and primary tumor (PT) tissues by mIHC. Both populations are significantly elevated in PT compared to normal cervix (n = 3 mice; unpaired t-test; *p < 0.05, ***p < 0.001)

To directly validate the causal role of GAS6-AXL signaling in cervical cancer cell invasiveness within the model system, we first performed Boyden chamber migration and invasion assays in U14 murine cervical cancer cells. GAS6 treatment significantly promoted both migration and invasion of U14 cells compared to untreated controls, and this enhancement was completely abrogated by co-treatment with the selective AXL inhibitor Bemcentinib, confirming AXL-dependent signaling (Fig. 7A and B). Consistently, conditioned medium derived from GAS6‑overexpressing cells (but not control medium) also promoted U14 cell migration and invasion, recapitulating the effect of recombinant GAS6 (Fig. S4A, 4B).

Fig. 7.

GAS6-AXL axis-mediated promotion of HeLa cell migration and invasion. A. Representative crystal violet-stained Transwell migration (upper) and invasion (lower) assay images of U14 cells treated with vehicle control, GAS6 (500 ng/mL), Bemcentinib (2 µM), or GAS6 + Bemcentinib. Scale bar = 200 μm. B. Quantification of U14 cell migration and invasion from panel A. GAS6 significantly enhanced migratory and invasive capacity, effects fully abolished by AXL inhibition with Bemcentinib (n = 15; one-way ANOVA with Tukey's multiple comparisons test; **p < 0.01, ****p < 0.0001). C. Representative crystal violet-stained Transwell migration (upper) and invasion (lower) assay images of HeLa cells treated with vehicle control, GAS6 (500 ng/mL), Bemcentinib (2 μM), or GAS6 + Bemcentinib. Scale bar = 200 μm. D. Quantification of HeLa cell migration and invasion from panel C. GAS6 significantly promoted migration and invasion, which were significantly reversed by Bemcentinib (n = 15; one-way ANOVA with Tukey's multiple comparisons test; **p < 0.01, ***p < 0.001, ****p < 0.0001). E. Representative H&E and multiplex immunofluorescence images of normal cervix and cervical cancer tissues from the SYSUSH cohort (normal cervix n = 3; cervical cancer n = 6), stained for DAPI (blue), pan-CK (yellow), AXL (green), COL1A1 (grey), C3(purple), and GAS6 (red). Cervical cancer tissues demonstrate elevated AXL expression in pan-CK⁺ epithelial cells and increased GAS6⁺C3⁺COL1A1⁺ fibroblast abundance compared to normal cervix. Scale bar = 200 μm. F. Quantification of relative AXL⁺pan-CK⁺ tumor cell abundance [Mean diff (95% CI): 8.626 (3.153,14.10)] (upper) and GAS6⁺C3⁺COL1A1⁺ fibroblast abundance [Mean diff (95% CI): 5.135 (2.830,7.441)] (lower) in normal cervix versus cervical cancer tissues from the SYSUSH cohort. Both populations are significantly elevated in cervical cancer (unpaired t-test; **p < 0.01). G. Kaplan-Meier overall survival analysis stratified by C3⁺ fibroblast signature expression (high vs. low) in the TCGA-CESC cohort (n = 303; upper) and the independent external GSE52903 cohort (n = 55; lower). Cutoff determined by the quartile method. High C3⁺ fibroblast signature expression is significantly associated with shorter overall survival in both cohorts [log-rank test; TCGA-CESC: HR = 2.01, 95% CI (1.02,3.98), p = 0.0403; GSE52903: HR = 1.36, p = 0.0001]

To assess the translational relevance of these findings in human cervical cancer, we extended this analysis to the human HPV-positive cervical cancer cell line HeLa. Consistent with observations in U14 cells, GAS6 treatment significantly enhanced HeLa cell migration and invasion, and these effects were similarly abolished by Bemcentinib co-treatment (Fig. 7C and D). Complementary validation in the human HPV-negative cell line C33A yielded concordant results, further suggesting that GAS6-AXL-driven invasion may operate independently of HPV status (Fig. S4C-4 F). Together, these multi-line in vitro data provide direct causal evidence that GAS6 promotes cervical cancer cell invasiveness through AXL activation.

To corroborate these findings in human clinical tissues, we performed multiplex immunofluorescence (mIHC) on the SYSUSH cohort, comprising normal cervical tissues (n = 3) and locally advanced cervical cancer specimens (FIGO stages III–IV; n = 6). Co-staining for AXL, pan-CK (tumor cells), GAS6, C3, and COL1A1 (C3⁺ fibroblast markers) was performed and quantified using the Halo image analysis platform. The relative abundance of AXL⁺pan-CK⁺ tumor cells was significantly elevated in cervical cancer tissues compared to normal cervix (Fig. 7E, upper panel in F). Similarly, the density of GAS6⁺C3⁺COL1A1⁺ fibroblasts was markedly higher in cervical cancer tissues than in normal controls (Fig. 7E, lower panel in F), providing protein-level confirmation that both components of the GAS6-AXL signaling axis — the ligand-producing C3⁺ fibroblast population and the AXL-expressing tumor cell population — are concurrently enriched in human cervical cancer.

To evaluate the prognostic significance of the C3⁺ fibroblast gene signature, we performed survival analysis using the C3⁺ fibroblast marker gene set in two independent datasets. In the TCGA-CESC cohort, high C3⁺ fibroblast signature expression was significantly associated with shorter overall survival compared to low expression (log-rank p = 0.0403; HR = 2.01; Fig. 7G, upper). To confirm these findings in an independent external dataset, we further validated the signature in GSE52903, where high C3⁺ fibroblast signature expression was again significantly associated with inferior overall survival (log-rank p < 0.0001; HR = 1.36; Fig. 7G, lower). The consistent prognostic association across two independent datasets — one internal (TCGA-CESC) and one external (GSE52903) — supports the clinical relevance of C3⁺ fibroblast enrichment in human cervical cancer.

Collectively, these findings — spanning murine in vitro models, human cell line validation across HPV-positive and HPV-negative backgrounds, protein-level mIHC confirmation in human clinical tissues, and dual-cohort survival analyses — reinforce that C3⁺ fibroblasts act as key stromal drivers of cervical cancer invasion and metastasis by activating the GAS6-AXL signaling axis, and that their enrichment is associated with poor clinical outcomes in human disease.

Discussion

An ideal animal model of cervical cancer should faithfully recapitulate the phenotypic characteristics, immunocompetent microenvironment, temporal progression and metastatic behavior of the human disease, while simultaneously ensuring biosafety, reproducibility, and operational convenience throughout its establishment. In this study, we established a syngeneic orthotopic mouse model using U14 cells stably expressing luciferase, which allows for a real-time, longitudinal monitoring of tumor growth and metastatic spread. Tumor engraftment within the cervical mucosa was confirmed by in vivo bioluminescence and histological examination as early as day 7. The entire procedure was performed under direct visualization, ensuring precise injection placement, operational simplicity, high tumorigenic efficiency and reproducibility. Importantly, this model recapitulates the three major metastatic routes observed in clinical cervical cancer: contiguous invasion of pelvic viscera (especially the rectum), lymphogenous spread to regional nodes, and hematogenous dissemination to distant sites, such as lung. Moreover, unlike subcutaneous xenograft models [26], the orthotopic tumor model enables the incorporation of resident normal cervical epithelial cells, stromal cells, and immune cells into the research scope, thereby more faithfully recapitulating the stepwise progression of tumor development as a result of dynamic interactions among multiple cell types, including both normal and malignant populations. This model fills a methodological gap: it provides a reliable and tractable platform for studying the mechanisms of cervical cancer metastasis and evaluating potential therapeutic strategies.

High-resolution sequencing technologies, such as single-cell RNA sequencing, have significantly advanced the understanding of the tumor microenvironment. In this study, single-cell RNA sequencing was employed to characterize differences between PTs and paired LNMTs in this mouse model. We found that stromal cells in the primary site, particularly C3⁺ fibroblasts, which exhibiting marked enrichment in the PTs, enhance the invasive and metastatic capabilities of U14 cells via the GAS6-AXL signaling pathway, thereby acting as key orchestrators of the microenvironment (Figs. 6 and 7) [30, 31]. AXL, a member of the receptor tyrosine kinase family, has been documented to interact with its ligand GAS6, primarily participating in tumor cell proliferation, epithelial-mesenchymal transition, cancer stemness maintenance, and therapy resistance [32–34]. These findings are consistent with our results, which demonstrate that abnormal activation of AXL in tumor epithelial cells—compared to normal cervical epithelial cells—enhances their proliferative, invasive, and metastatic capacities. Interestingly, a study on prostate cancer revealed that GAS6–AXL activation can induce a dormant state in bone-metastasized prostate cancer cells, which may confer stronger therapy resistance [35]. However, our present study did not explore the activation status of the GAS6–AXL axis at metastatic sites in depth. Whether it constitutes a potential therapeutic target to reduce postoperative recurrence after lymph node dissection in cervical cancer warrants further exploration and represents a viable path for drug development. Moreover, although mIHC on human tissue and public survival analyses support clinical relevance of GAS6–AXL signaling, cross-species single-cell comparisons with human cervical tumors will be needed to evaluate conservation of CAF programs and tumor–immune interactions at cellular resolution.

Previous studies have reported that SPP1 is primarily secreted by tumor-associated macrophages and fibroblasts. SPP1⁺ macrophages have been shown to directly drive T cells into a state of stress response through the SPP1/CD44/PI3K/AKT signaling pathway [29, 36], thereby promoting liver metastasis in colorectal cancer. Conversely, androgen deprivation therapy-induced SPP1⁺ myofibroblastic cancer-associated fibroblasts can confer resistance to androgen deprivation therapy (ADT) in prostate cancer (PCa) via an SPP1-ERK paracrine mechanism [37]. These collective findings indicate that stromal-derived SPP1 is associated with T cell dysfunction, impaired anti-tumor immunity, and tumor cell drug resistance. However, much less is known about tumor cell-derived SPP1 and its role in shaping the immune microenvironment. Only a few studies have revealed that tumor cell-secreted SPP1 influences pancreatic cancer cell fate or polarizes macrophages toward tumor-promoting phenotypes [38, 39]. Here, our study demonstrates that, in LNMT, U14 cells actively release SPP1 signals to suppress lymph node–resident CD8⁺ T cells for tumor immune evasion (Fig. 5). It’s reported that SPP1 potentially induce exhaustion via CD44-PI3K-AKT pathway or A2AR signaling, thereby facilitating immune evasion and promoting their own colonization within lymph nodes, independently of tumor-associated macrophages [27–29, 40]. Importantly, these immune-related observations were obtained in the context of the HPV-negative U14 model and should be regarded as hypothesis-generating; whether analogous SPP1-mediated immune evasion mechanisms operate in HPV-positive human cervical cancer warrants further investigation in appropriate model systems and well-characterized clinical cohorts.

The U14 cell line used in this study is derived from methylcholanthrene-induced murine cervical squamous carcinoma and lacks high-risk HPV infection. Accordingly, our orthotopic model is representative of HPV-negative, chemical carcinogen-induced cervical cancer, and is well-suited for investigating general metastatic mechanisms — including local invasion, lymph node dissemination, and lung colonization — as well as tumor-stroma interactions within a fully immunocompetent microenvironment. However, the model does not recapitulate HPV-driven oncogenesis — specifically, E6/E7-mediated degradation of p53 and Rb — and therefore should not be used to draw conclusions about HPV-specific oncogenic pathways or therapeutic responses to HPV-targeted agents. Persistent high-risk HPV infection, mediated by the E6/E7 oncoproteins, is recognized as the primary driver of cervical carcinogenesis [41]; the U14 cell line was originally derived from a methylcholanthrene-induced murine cervical carcinoma and harbors neither HPV nor murine MmuPV1 genomes [42]. Consequently, the stepwise progression from HPV-infected cervical epithelium through cervical intraepithelial neoplasia to invasive carcinoma cannot be recapitulated at the etiological level. The model is best regarded as a metastasis research platform rather than a surrogate for HPV-positive human cervical cancer. Notably, previous studies have successfully introduced HPV16 or HPV18 E6/E7 transgenes into U14 cells to generate HPV-expressing variants (e.g., E6⁺U14, E7⁺U14, and U14/LV-HPV18E6E7) [15, 43]. The orthotopic injection methodology and single-cell analytical pipeline established in the present study can therefore be directly extended to these HPV-transfected derivatives, offering a tractable future path for modeling HPV-driven cervical cancer metastasis within an immunocompetent orthotopic setting. Future efforts will accordingly focus on the stable genomic integration of HPV16 E6/E7 coding sequences into the U14-Luc genome, thereby generating an isogenic E6/E7-positive subline suitable for in vivo interrogation of hrHPV-driven invasive and metastatic properties.

Our conclusions primarily address metastasis-associated tumor-stroma and tumor-immune interactions in an immunocompetent syngeneic setting, and should be interpreted with caution when extrapolating to hrHPV-positive human disease. The multi-level validation presented — spanning HPV-negative and HPV-positive human cell lines, protein-level confirmation in the SYSUSH cohort, and dual-cohort survival analyses in TCGA-CESC and GSE52903 — provides converging preliminary evidence that the C3⁺ fibroblast–GAS6–AXL axis identified in our murine model may be conserved in human cervical cancer biology. Nonetheless, several important limitations temper the translational interpretation of these findings. The in vivo perturbation of C3⁺ fibroblast–GAS6–AXL axis was not performed, and the SYSUSH cohort is modest in size. Furthermore, the survival analyses are derived from publicly available transcriptomic datasets without individual-level clinicopathological correlation. Accordingly, these human data should be regarded as hypothesis-generating rather than definitive. Larger, prospectively designed, HPV-stratified cohorts incorporating longitudinal survival follow-up and spatial multi-omic profiling will be necessary to establish the clinical utility of this stromal signature and to determine whether the GAS6-AXL axis represents a viable therapeutic target across the full biological spectrum of human cervical cancer. Additionally, the inability to clearly identify or physically access deep-seated lung metastases at Day 30 — despite detectable fluorescence signals in vivo — prevented tissue collection and precluded single-cell analysis of hematogenously disseminated tumor cells. Future studies involving extended observation periods may facilitate the acquisition of macroscopic lung metastatic lesions amenable to single-cell profiling.

The cross-species experimental design, which combined a murine syngeneic in vivo model with in vitro validation in U14 (murine, HPV-negative), C33A (human, HPV-negative), and HeLa (human, HPV18-positive), strengthens the conclusion that the C3⁺ fibroblast–GAS6–AXL axis is a conserved driver of cervical cancer invasiveness, independent of species origin or HPV status (Fig. 7, Fig. S4). Nevertheless, this three-line panel does not constitute an exhaustive survey, and future studies employing additional patient-derived and HPV-stratified models are warranted to define the full mechanistic scope. The human tissue and survival analyses should be regarded as hypothesis-generating given the modest cohort size; larger, HPV-stratified cohorts are needed to confirm translational significance. Finally, the inability to access deep-seated lung metastatic lesions at Day 30 precluded their inclusion in single-cell profiling; extended observation periods in future studies may enable hematogenous metastasis to be characterized at single-cell resolution.

In conclusion, beyond its utility as a tractable model that recapitulates local invasion and metastasis for preclinical therapeutic evaluation, this study provides an in-depth single-cell atlas of the primary and metastatic tumor microenvironment in cervical cancer. This atlas establishes a foundational resource for elucidating underlying metastatic mechanisms and identifying potential therapeutic targets. The GAS6-AXL and SPP1-mediated immune evasion axes identified here represent candidate targets warranting further investigation in both HPV-negative and HPV-positive disease contexts, as well as in larger, prospectively designed human cohorts.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Zhang C directed and conceived the project. Zhan J led this project, and performed most of the experiments and analyzed data. Ye J performed animal experiments. Liu D analyzed single cell sequencing data. Peng X, Lu W, Liu X, Li M and Zhong L provided experimental support. Zhang C, Zhong L and Zhan J wrote the manuscript and Li T helped to revise and proofread the manuscript.

Funding

This work was supported by Shenzhen Medical Research Fund (Grant No. A2303011); Shenzhen Science and Technology Program (Grant No. JCYJ20250604143519025, JCYJ20240813150204006); Guangdong Basic and Applied Basic Research Foundation (Grant No. 2024A1515030292); Open Funds of State Key Laboratory of Oncology in South China (Grant No. HN2025-01) and Funding of Shenzhen Clinical Research Center for Gastroenterology (Gastrointestinal Surgery) (LCYSSQ20220823091203008).

Data availability

The raw single-cell RNA sequencing data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2026), China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA041826) and are publicly accessible at https://ngdc.cncb.ac.cn/gsa. All other data generated or analyzed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the Seventh Affiliated Hospital of Sun Yat-sen University (No# KY-2024-036). All animal procedures followed the Guide for the Care and Use of Laboratory Animals and the Principles for the Utilization and Care of Vertebrate Animals and were approved by the Institutional Animal Care and Use Committee, Sun Yat-Sen University (Approval NO. SYSU-IACUC-2024-001650).

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.