Exploring the role of disulfidptosis-related signatures in immune microenvironment, prognosis and therapeutic strategies of cervical cancer

Highlights

• •Experimental validation confirms the presence of disulfidptosis in cervical cancer cells.
• •Classification of cervical cancer patients into distinct molecular subclusters reveals variations in immune infiltration and prognosis.
• •The developed disulfidptosis-related prognostic model accurately predicts both prognosis and immunotherapy response in cervical cancer patients.
• •The association of YWHAG with advanced cervical cancer stages, tumor occurrence, development, and metastasis have implications for potential therapies.

Abstract

Background

Cervical cancer is characterized by a complex immunosuppressive tumor microenvironment (TME). Disulfidptosis is a recently identified form of programmed cell death that has emerged as a crucial factor in tumorigenesis. However, the research on the specific involvement of disulfidptosis within the TME is still in its early stages.

Methods

Under glucose starvation, SiHa and HeLa cells underwent experiments employing diverse cell death inhibitors and SLC7A11 knockdown to observe their impact on cell survival. TCGA-CESC cohort was subjected to consensus clustering for disulfidptosis-related clusters. Prognosis, function, immune infiltration, and differentially expressed genes (DEGs) evaluations among clusters were compared. A prognostic model based on DEGs and disulfidptosis regulator genes (DRGs) was constructed and internally and externally validated. The correlation between YWHAG and clinicopathological characteristics in cervical cancer patients was investigated at both the mRNA and protein levels. Proliferation and migration assays were performed to uncover the roles of YWHAG in cervical cancer.

Results

Experimental validation confirmed disulfidptosis in cervical cancer cell lines. Cervical cancer patients were classified into three clusters based on DRGs, showing notably improved prognosis and increased immune infiltration in cluster B. The developed disulfidptosis-related prognostic model effectively stratified patients into high- and low-risk groups. Low-risk patients exhibited more favorable responses to immunotherapy and improved overall prognosis. Additionally, YWHAG, recognized as a tumor-promoting gene, demonstrated active roles in enhancing the growth, migration, and invasion of cervical cancer cells.

Conclusion

Our research proposed a prognostic model for cervical cancer, probably contributing to tumor microenvironment traits and more potent immunotherapy strategy exploration.

List of abbreviations

DEGs

differential expressed genes

DRGs

disulfidptosis regulator genes

ICB

immune checkpoint blockade

SLC7A11

solute carrier family 7 member 11

TME

tumor microenvironment

GTEx

Genome Tissue Expression

CNVs

copy number variations

GEO

Gene Expression Omnibus

SCC

squamous cell carcinoma

AC

adenocarcinoma

FFPE

formalin-fixed and paraffin-embedded

GSVA

Gene Set Variation Analysis

LASSO

least absolute shrinkage and selection operator

OS

overall survival

PFS

progression-free survival

ROC

receiver operating characteristic

AUC

area under the ROC curve

TIDE

Tumour Immune Dysfunction and Exclusion

TCIA

The Cancer Immunome Atlas

ATCC

American Type Culture Collection

TCGA

The Cancer Genome Atlas

PI

propidium iodide

IHC

immunohistochemistry

SEM

standard error of the mean

Ferr-1

ferrostatin-1

Nec

necrostatin

CQ

chloroquine

DTT

dithiothreitol

F-actin

actin filament

PCA

principal components analysis

FDA

Food and Drug Administration

IPS score

immunophenoscore

TMB

tumor mutational burden

HISL

high-grade squamous intraepithelial lesion

LSIL

low-grade squamous intraepithelial lesion

NMPA

National Medical Products Administration

EMT

epithelial-mesenchymal transition

Introduction

Cervical cancer, ranked as the fourth most prevalent malignancy among women worldwide, presents a complex scenario of challenges in its management [1]. Although current treatment strategies including surgery, chemotherapy and neoadjuvant chemotherapy followed by radical surgery have significantly improved the prognosis of patients, the clinical outcome for patients with advanced cervical cancer remains suboptimal[2,3]. The burden intensifies with recurrence and metastasis, emphasizing the persistent challenge posed by cervical cancer [4]. Moreover, the inherent heterogeneity within cervical cancer adds a layer of complexity to treatment decisions, as cases sharing the same tumor stage may exhibit diverse clinical outcomes [5,6]. This heterogeneity highlights the limitations of the current staging system in accurately predicting prognosis. On the other hand, with the presence of clinical studies exploring immunotherapy for cervical cancer patients, the efficacy of immune checkpoint blockade (ICB) remains notably constrained, particularly in the early stage [7,8]. Consequently, there is an urgent need for precise biomarkers capable of accurately predicting the prognosis of cervical cancer and indicating patients' response to immunotherapy.

Recent advances in understanding programmed cell death have uncovered a novel mode termed disulfidptosis. This cellular process is triggered under glucose starvation conditions, particularly in cells expressing high solute carrier family 7 member 11 (SLC7A11). Disulfidptosis involves the aberrant accumulation of cystine or other disulfide compounds, depleting cellular NADPH supplies and inducing sulfur dioxide stress, ultimately leading to rapid cell death [9]. Due to its potential significance in cancer biology and tumor microenvironment (TME), especially in metastatic settings where cells may exhibit heightened vulnerability to disulfidptosis, there is a compelling need to investigate its role in cervical cancer [10,11].

Our primary objective was not merely to explore the correlation between disulfidptosis regulator genes (DRGs) and the prognosis, as well as TME traits in cervical cancer. Additionally, we aimed to develop an innovative prognostic model for cervical cancer. Furthermore, we evaluated the sensitivity of cervical cancer patients to immunotherapy based on the developed model. Substantiating the relevance of the selected gene, we performed proliferation and migration assays in vitro. The overarching goal of this study was to identify and provide novel prognostic biomarkers associated with disulfidptosis for potential targeted therapy in cervical cancer.

Materials and methods

Data resources

The mRNA expression (FPKM) and somatic mutation data for TCGA-CESC cohort (n = 304) were retrieved from the UCSC XENA platform and TCGA database [12]. Simultaneously, the normal cervical samples' mRNA expression (FPKM) was acquired from the Genome Tissue Expression (GTEx) portal. Comprehensive clinical data, including survival information and gene-level frequency of copy number variations (CNVs), were sourced from a prior study [13]. For external validation, the series matrix file of GSE52903 (n = 72) was downloaded from the Gene Expression Omnibus (GEO), and the processed expression matrix of GSE63514 (n = 128) was acquired from a previous study [14].

Tumor samples (tumor tissues and paired paracancerous tissues) from 18 squamous cell carcinoma (SCC) patients were collected in Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, and those paired tissues were for RNA extraction (HUST cohort). The tumor tissues from 34 SCC and 1 adenocarcinoma (AC) patients collected from the Women's Hospital, Zhejiang University School of Medicine, were preserved as formalin-fixed and paraffin-embedded (FFPE) (ZJU cohort). All pathological diagnoses were re-reviewed by an expert pathologist.

Consensus clustering for DRGs

Ten DRGs referred to the previous literature, signatures including GYS1, LRPPRC, NCKAP1, NDUFA11, NDUFS1, NUBPL, OXSM, RPN1, SLC3A2 and SLC7A1 [9]. Based on DRGs expression, consensus clustering was performed with the ConsensusClusterPlus package in R to identify disulfidptosis-related clusters [15].

Pathway enrichment and immune indices analysis

Gene Set Variation Analysis (GSVA) was applied using canonical KEGG pathway gene sets from MSigDB and immunologically relevant gene sets from ImmPort, followed by differential analysis using the R package limma to elucidate functional and immune differences among clusters [16], [17], [18]. Immune, stromal and ESTIMATE scores were calculated via the ESTIMATE algorithm [19]. CIBERSORTX-Absolute scores were obtained and processed as previously described [14].

Risk model construction and validation

Differentially expressed genes (DEGs) associated with disulfidptosis were identified using the R package limma [18]. The criteria for DEG analysis among clusters were set as an adjusted p-value < 0.05 and |log2Fold Change| > 0.3. Univariate Cox regression analysis was initially conducted to determine the prognostic significance of the combination of protein-coding DEGs and DRGs. Subsequently, a least absolute shrinkage and selection operator (LASSO) Cox regression model was further developed for both overall survival (OS)- and progression-free survival (PFS)-related genes using the R package glmnet. The optimal lambda (λ) was chosen from the LASSO model using 10-fold cross-validation with the minimum partial likelihood deviance. Finally, the risk score was calculated as follows:

$\text{Risk}\text{score}=\sum (\text{Coe}{\mathrm{f}}_{\mathrm{i}}*\text{Ex}{\mathrm{p}}_{\mathrm{i}})$

The Coef_i represents the regression coefficients and Exp_i represents the expression levels. The independent verification dataset also validates the risk model (GSE52903). Patients in different cohorts were dichotomized into high- and low-risk groups according to the median risk score. The area under the receiver operating characteristic (ROC) curve (AUC) was calculated to assess the model's predictive potential.

Immunotherapy analysis

Tumour Immune Dysfunction and Exclusion (TIDE) score and Immunophenscore from The Cancer Immunome Atlas (TCIA) were applied to predict the response to ICB between high- and low-risk patients [20,21]. A list of co-stimulatory and co-inhibitory immune checkpoint targets (https://www.sinobiological.com/research/immune-checkpoint/immune-checkpoint-targets) was referred to calculate the correlation of risk model gene expression/risk score and these immunomodulators.

Cell culture and siRNA transfection

The cell lines SiHa and HeLa were procured from the American Type Culture Collection (ATCC, USA). All cell lines were cultured at 37 °C in a 5 % CO2 incubator in a humidified environment and cultured with DMEM medium (Gibco, USA), routinely supplemented with 10 % fetal bovine serum (Everyday Green, China). All cell lines were authenticated by STR profiling. The following reagents were obtained from MedChemExpress: DL-dithiothreitol (HY-15,917), Z-VAD (HY-16658B), Necrostatin-1 (HY-15,760), Ferrostatin-1 (HY-100,579), Chloroquine (HY-17589A), Trolox (HY-101,445). All reagents were dissolved according to the manufacturer's instructions.

SLC7A11, YWHAG and control siRNA were synthesized by GenePharma (China). Transfection was performed using RNAiMX Lipofectamine (Thermo Fisher) following the manufacturer's instructions. The target sequences of siRNAs are listed in Table S1.

Cell lysis and western blotting

Samples were collected and lysed in cold RIPA lysis buffer supplemented with protease inhibitor cocktails. Protein samples were separated by 10 % SurePAGE gels (GenScript, USA) and transferred to 0.22 µm PVDF membranes (Bio-Rad, USA) using the eBlot L1 protein transfer system (GenScript, USA). Before SDS-PAGE analysis, all samples were heated to 95 °C for 10 min to ensure proper denaturation and sample preparation. Subsequently, the PVDF membranes were blocked with 5 % milk at room temperature to reduce non-specific binding. The membranes were then exposed and visualized using the ImageQuant LAS 4000mini system (Cytiva, Japan). The antibodies used are listed in Table S2.

Assay for proliferation and migration

After being transfected with siRNA for 24 h, the cells were seeded into 96-well plates. The CCK-8 kit (YEASEN, 40203ES76) was employed following the manufacturer's instructions. The absorbencies at 450 nm were measured by spectrophotometer reader (Thermo Fisher Scientific, USA)

Cells were seeded in 6-well plates (1000 cells per well) and cultured for 14 days. Colonies were fixed with glutaraldehyde (6.0% v/v), stained with crystal violet (0.5% w/v) and counted using ImageJ software.

The 24-well transwells (8 µm pore size, Falcon, USA) were used for cell migration and invasion assays. 5 × 10⁴/200 µL cells were cultured in a serum-free medium in the upper chambers, and a medium containing 10 % FBS was added to the lower chambers. Matrigel (Corning, USA) was added for cell invasion assays. After 16–24 h, cells adhered to the bottoms of the upper chambers were fixed, stained and counted. Migration and invasiveness were determined by counting cells in five random fields.

3 × 10⁴ cells were suspended in 70 µL of DMEM medium and seeded in an ibidi culture insert (ibidi, Germany). The incubation chambers were placed in a 12-well plate. When cells were at full confluency, the insert was removed and a serum-free medium was added. The wounds were photographed at 0 and 48 h, and the migration rate of the cells was expressed as relative gap closure using the ImageJ v10 software.

Cell death assay

Cells were first seeded in 12-well plates one day before the treatment. Following culture with a specific medium, with or without the relevant drugs, the cells were detached using trypsin and collected into 1.7 ml microtubes. After a wash with PBS, the cells were suspended in cold PBS containing 1 μg/ml of propidium iodide (PI). The measurement of dead cells, indicated by PI-positive staining, was carried out using a CytoFLEX S Flow Cytometer (Beckman Coulter). Subsequently, the data was analyzed using FlowJo software.

Fluorescent staining of actin filaments (F-actin)

Following treatment, cells in a chamber slide (Biosharp BS-18-RC) were washed once with PBS and fixed in 3.7 % paraformaldehyde for 5 min three times at room temperature with wash buffer (PBS with 0.1% Triton X-100). The cells were incubated with 1:100 diluted Actin-Tracker Red-Rhodamine (Beyotime, C2207S) in PBS (0.1% Triton X-100) at room temperature in the dark for 30 min. After washing three times with PBS (0.1% Triton X-100), cell nuclei were stained with DAPI for 5 min. All fluorescence images were captured using a confocal microscope (STELLARIS5, Leica).

RNA extraction and RT-qPCR analysis

Total RNA was extracted from cells or tissues using FastPure Cell/Tissue Total RNA Isolation Kit V2 (Vazyme, RC112–01). cDNA was synthesized with HiScript® III All-in-one RT SuperMix Perfect (Vazyme, Q333–01). qPCR was conducted using the ChamQ Universal SYBR qPCR Master Mix (Vazyme, Q711–02) by the LightCycler® 480 System (Roche, USA). The relative mRNA expressions were calculated by the 2^−ΔΔCt method normalized to Beta-Actin. Primer sequences are listed in Table S3.

Immunohistochemistry (IHC) staining and scoring

Paraffin-embedded tissue sections underwent deparaffinization in xylene, followed by hydration through graded alcohol-to-water series. Antigen retrieval was performed using Tris-EDTA buffer (pH 9.0) with a heat-induced protocol. Sections were treated with 10 % hydrogen peroxide for 15 min to inhibit endogenous peroxidase activity and blocked in 5 % Goat Serum (Fisher Scientific). They were subsequently incubated with the primary antibody Anti-14–3–3 gamma overnight at 4 °C (Table S2). After primary antibody incubation, sections underwent primary and secondary antibody incubation, with antibody binding detected using the DAB IHC staining system (Servicebo, G1211). Slides were scanned with an Ocus®20 microscope scanner with ×20 objective lens (Grundium, FI). The slide files were loaded into a project in QuPath software (version 0.4.4) for scoring, and quantification of cell positivity (mean DAB staining) was observed exclusively in tumor cells. QuPath output identified the H-score based on the extent and intensity of cell staining (0-3), multiplied by the percentage of cells positive, with a potential score ranging from 0 to 300 [22].

Statistical analyses

Statistical analyses and graphical visualization were performed using R (version 4.3.1) and GraphPad Prism 9 (version 9.5.0). The paired or unpaired student's t-test, one-way ANOVA followed by Dunnett's or Student's t-test, and Chi-square test were applied for statistical comparisons. Survival curves were plotted using the Kaplan-Meier method and compared using the log-rank test. Each experiment was repeated more than three times. Pearson's correlation coefficients were calculated to ascertain bivariate correlations. Data were shown as means ± standard error of the mean (SEM). All p-values were two-sided, with p < 0.05 indicating statistical significance unless otherwise stated.

Result

Disulfidptosis in glucose-starved cervical cancer cell lines

We initially examined whether cell death under glucose starvation in cervical cancer cells could be prevented by any known cell death inhibitors. Diverse cell death inhibitors were examined, including the ferroptosis inhibitor ferrostatin-1 (Ferr-1), the apoptosis inhibitor Z-VAD-fmk, the necroptosis inhibitor necrostatin (Nec), the autophagy inhibitor chloroquine (CQ), the antioxidant Trolox, and the reducing agent dithiothreitol (DTT). Among these, only treatment with the reducing agent DTT fully suppressed glucose starvation-induced cell death in SiHa and HeLa cells (p < 0.05, Fig. 1A). Notably, SLC7A11-knockdown completely abolished the cell death-promoting effect of glucose starvation in SiHa and HeLa cells (p < 0.05, Fig. 1B). Glucose starvation-induced striking changes in cell morphology that were characterized by cell shrinkage and F-actin contraction. Meanwhile, the knockdown of SLC7A11 inhibited the striking changes in cell morphology (Fig. 1C). Collectively, our data suggested that glucose starvation conditions promote disulfideptosis in cervical cancer cells.

Fig. 1.

Disulfidptosis in glucose-starved SiHa and HeLa cells. A. Cell death in SiHa and HeLa cells cultured in glucose-containing (+Glc) or glucose-free (-Glc) medium. With or without the indicated concentrations of DTT, Z-VAD, Nec, Ferr-1, CQ and Trolox for 10 h (SiHa cells) or 6 h (HeLa cells). B. Cell death in control and si-SLC7A11 SiHa and HeLa cells cultured in the indicated medium for 16 h (SiHa cells) or 6 h (HeLa cells). C. Fluorescent staining of F-actin with phalloidin in control and si-SLC7A11 SiHa and HeLa cells cultured in glucose-containing (+Glc) or glucose-free (-Glc) medium for 10 h (SiHa cells) or 6 h (HeLa cells). Nuclei were stained by 4,6-diamidino-2-pheylindole (DAPI). * p < 0.05, ** p < 0.01, *** p < 0.001.

Genetic characteristics of DRGs and clustering of molecular subclusters

Among 304 cervical cancer samples, 31 (10.2 %) carried mutations of DRGs, with the highest mutational frequency observed in LRPPRC (3.3 %) (Fig. 2A). Additionally, exploring CNV alternation frequency determined a higher incidence of CNV gain in the RPN1, whereas NDUFS1 and NDUFA11 exhibited a greater frequency of copy number loss (Fig. 2B). The expression levels of six genes (GYS1, LRPPRC, OXSM, RPN1, SLC3A2, SLC7A11) in DRGs were upregulated in cervical cancer samples compared to normal samples (Fig. 2C). By conducting consensus clustering based on the expression profile of DRGs, patients were divided into three subclusters, named DRG cluster A/B/C (Fig. S1A–D). Principal components analysis (PCA) effectively distinguished patients with three clusters (Fig. 2D). Compared to patients in clusters A and C, those in cluster B had a more favorable prognosis (all p < 0.05, Fig. 2E). The comparison of three clusters regarding clinicopathological parameters (age, FIGO stage, grade and histology) did not reveal any significant difference (Fig. 2F).

Fig. 2.

Landscape of DRGs and clustering of molecular subclusters. A. Somatic mutation frequency of DRGs in TCGA-CESC cohort. Each column represents individual patients. The number on the right indicates the mutation frequency in the DRGs. The right bar graph revealed the proportion of each variant type. B. CNV frequency of DRGs. The red and blue bars represent CNV amplification and deletion, respectively. C. mRNA levels of DRGs. D. Distribution of patients in three clusters. E. Kaplan-Meier curves for OS and PFS among three clusters. F. Expression profiles of DRGs and clinicopathological characteristics among three clusters. Amp, amplification, Del, deletion. * p < 0.05, ** p < 0.01, *** p < 0.001, ns, not significant.

Analysis of functional annotation and immune infiltration between DRG clusters

To gain a better comprehensive understanding of the biology of the clusters, we evaluated the functional enrichments. Cluster B, with a better prognosis, displayed a primary downregulation in cancer-related pathways, exemplified by beta-catenin/Wnt, TGF-beta signaling, Hedgehog signaling, Phosphatidylinositol signaling, and the ErbB signaling pathway. Simultaneously, it exhibited downregulation in signaling molecules and interactions, specifically those associated with adherens junctions, the regulation of the actin cytoskeleton, and ECM-receptor interaction. In contrast, cluster C displayed a predominant downregulation in pathways related to immune response and signaling activities, including Toll-like receptor signaling, NOD-like receptor signaling, JAK-STAT signaling, and Natural killer cell-mediated cytotoxicity pathways. Differences were also observed in pathways associated with metabolism among three clusters (Fig. 3A, Table S4).

Fig. 3.

Evaluation of molecular and TME characteristics in three DRG clusters. A. Heatmap showing the activation states of biological pathways in clusters. The purple represented activated pathways, and the green represented inhibited pathways. B. Boxplots showing the difference in Immune, Stromal, and ESTIMATE scores among clusters. C. Characteristics of immune responses in three clusters. D. Abundance of immune cells in clusters. * p < 0.05, ** p < 0.01, *** p < 0.001, ns, not significant.

Given the increasing emphasis in research on the vital role of TME in cancer development [23,24], we further explored the differences in immune signatures among the DRG clusters. Patients in cluster B had higher immune and ESTIMATE scores (all p < 0.05). In contrast, there was no significant difference in stromal scores among three clusters (Fig. 3B). Moreover, cluster B had higher interferon and TNF family receptor enrichment and lower TGF-beta family receptor expression. This association has been linked to promoting immune infiltration [25], [26], [27] (Fig. 3C). Specifically, cluster B showed higher proportions of infiltrating CD8+ T cells, activated memory CD4+ T cells, and M2 macrophages. In contrast, clusters A and C exhibited more infiltrating M0 macrophages and naïve B cells, respectively. In summary, there were discernible differences among three clusters regarding the immune cell infiltration characteristics of TME, with cluster B displaying superior adaptive immune cell infiltration and activation (Fig. 3D).

Establishment and validation of the disulfidptosis-related prognostic model

We identified 115 DEGs based on gene expression across three clusters. Univariate Cox regression analysis was performed to determine the prognostic significance of the combination of protein-coding DEGs and DRGs, resulting in 16 genes associated with OS and PFS (all p < 0.05). Subsequently, LASSO Cox analysis was applied to these genes, ultimately retaining eight disulfidptosis-related signatures based on the minimum partial likelihood deviation (Fig. S2A, B) and establishing a prognostic model related to disulfidptosis (Fig. 4A, Table S5). The Sankey diagram showed that most patients in cluster B were in the low-risk group, whereas cluster A had a majority in the high-risk group (Fig. 4B).

Fig. 4.

Construction and validation of the disulfidptosis-related prognostic model. A. The establishment of the disulfidptosis-related prognostic model. B. Sankey diagram for the DRG clusters and risk score. C. Expression profiles of risk genes and clinicopathological characteristics between high- and low-risk groups in the training and testing sets. D. Kaplan-Meier curves between risk groups in the training set and testing set. E. ROC curves predicting the sensitivity and specificity of 3-, 5- and 10-year survival based on the risk score. H. Forest plot of significant factors based on multivariate Cox regression analysis in the training and testing sets.

Subsequently, we used the TCGA-CESC cohort as the training set and the GSE52903 as the testing set. The risk score value was calculated according to the formula, and patients in both the training and testing sets were categorized into high- or low-risk groups based on the median risk score. Additionally, there were significant differences in the expression of risk genes but no difference in clinicopathological characteristics between high- and low-risk groups (Fig. 4C).

For survival analysis of TCGA-CESC cohort, patients in the high-risk group exhibited significantly shorter OS and PFS than those in the low-risk group. Moreover, the same conclusion was reached for OS in GSE52903 (Fig. 4D). In the training and testing sets, the 5-year AUC values for OS and PFS consistently reached 0.7 (Fig. 4E). To clarify the independent impact of the risk score on prognosis in cervical cancer patients, multivariable Cox regression models adjusted for typical clinicopathological parameters (age, FIGO stage, grade and histology) were performed. These models revealed that risk score remained independently associated with OS and PFS (all p < 0.05, Fig. 4F). Furthermore, there was no correlation between these clinicopathological parameters and risk score (Fig. S3A–D).

Prediction of immunotherapy efficacy by the risk score

Although immunotherapies for recurrent or metastatic cervical cancer have been approved by the US Food and Drug Administration (FDA), the response rate was limited [28]. This study explored the relationship between risk score and the response to immunotherapy. Our results showed that the percentage of responses in the low-risk group was higher than in the high-risk group (Fig. 5A). Furthermore, there were significant differences in immune dysfunction scores (Dysfunction) and immune rejection scores (Exclusion) between the two groups (Fig. 5B). Next, we evaluated the differences in the immunophenoscore (IPS score) between groups, which would reflect a hypothetical treatment with immune checkpoint inhibitors. Importantly, our results revealed significantly higher IPS scores in the low-risk group, who would be more likely to respond to combination blockade for CTLA-4 and PD-1 or for CTLA4 alone or PD-1 alone (all p < 0.05, Fig. 5C). The risk score negatively correlated with MHC I–related antigen-presenting molecules (all p < 0.05), indicating lower immunogenicity. Additionally, a positive correlation was observed between the risk score and higher expression of coinhibitors and immune checkpoint molecules. This suggested high-risk tumors expressed immune checkpoint molecules to evade immune killing post-stimulation (Fig. 5D) [29,30]. It has been recommended that patients with higher tumor mutational burden (TMB) were more responsive to immunotherapy [31]. However, no correlation between the high- and low-risk groups and TMB was found in this study (Fig. S3E).

Fig. 5.

The response to immunotherapy in high- and low-risk groups. A. Distribution of the immunotherapy responders and non-responders between high- and low-risk groups. B. Tumor immune dysfunction score and exclusion scores between high- and low-risk groups. C. Boxplots indicating the IPS score across the high- and low-risk groups. (i) anti-CTLA-4 (+), anti-PD-1 (-): patients who would receive immunotherapy with anti-CTLA-4 alone but not anti-PD-1; (ii) anti-CTLA-4 (-), anti-PD-1 (+): patients who would receive immunotherapy with anti-PD-1 alone, but not anti-CTLA-4; (iii) anti-CTLA-4 (+), anti-PD-1(+): patients who would receive combination immune checkpoint inhibition therapy. D. Heatmap showing correlation among risk genes/score, MHC molecules, costimulators and coinhibitors.

YWHAG was upregulated in cervical cancer and correlated with unfavorable clinicopathology

In the prognostic model, YWHAG mRNA expression showed the highest and positive association with risk. Additionally, the expression of the majority of MHC I-related antigen-presenting molecules and B7-H4 was negatively correlated with the expression of YWHAG (Fig. 5D), suggesting a potential relationship between YWHAG and the tumor immune-suppressive microenvironment and activation of antitumor immunity [32,33]. Nevertheless, it is yet to be fully understood how YWHAG affects cervical cancer and development biologically.

YWHAG mRNA expression was initially significantly elevated in SCC compared to normal and high-grade squamous intraepithelial lesion (HISL) samples (all p < 0.05). A marginal increase was also noted in SCC compared to low-grade squamous intraepithelial lesion (LSIL) (p < 0.1, Fig. 6A). Through RT-qPCR experiments in the HUST cohort, we further confirmed significant upregulation of YWHAG mRNA expression in cervical cancer samples compared to the corresponding adjacent normal tissues (p < 0.05, Fig. 6B). Next, we examined the correlation between YWHAG mRNA expression and clinicopathological characteristics in the TCGA-CESC cohort. The analysis revealed that increased YWHAG mRNA expression was approaching significance associated with the advanced FIGO stage (p < 0.1, Table 1). Corresponding to the change in mRNA, high protein expression of YWHAG, as validated by IHC, was associated with advanced FIGO stages and lymph node metastasis in the ZJU cohort (all p < 0.05, Table 1). The representative IHC staining images of YWHAG are shown in Fig. 6C. Our findings suggested the potential implication of YWHAG in the initiation and progression of cervical cancer, positioning it as a promising diagnostic and prognostic biomarker for the disease.

Fig. 6.

Validation of the role of the key gene YWHAG in SiHa and HeLa cells. A, Boxplots showing the correlations between YWHAG expression and severity of cervical lesion in GSE63514. B. The YWHAG mRNA level was detected by RT-qPCR 18 cervical cancer tissues and matched adjacent tissues (HUST cohort). C. IHC staining images showing low and high YWHAG expression in cervical cancer tissue from the ZJU cohort. D, E. RT-qPCR and Western blotting validated the knockdown efficacy of siRNAs targeting YWHAG in SiHa and HeLa cells. F. CCK8 assays were used to detect the viability of YWHAG knockdown cells. G. Colony formation assays were applied to evaluate cell proliferation ability. H, Transwell assays were performed to evaluate the migration and invasion abilities of SiHa and HeLa cells. I. Wound healing assays to show the cell migration abilities. † < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001.

Table 1.

Association of YWHAG expression and clinicopathological characteristics in TCGA-CESC and ZJU cohorts.

Characteristics	YWHAG expression		Total	p values
	High	Low
TCGA (mRNA level)	152	152	304
Age, n (%)
< 55	107 (70.4)	107 (70.4)	214 (70.4)	1.000
≥ 55	45 (29.6)	45 (29.6)	90 (29.6)
Histology, n (%)
AC/ASC	12 (8.1)	19 (14.1)	31 (11.0)	0.157
SCC	136 (91.9)	116 (85.9)	252 (89.0)
FIGO stage, n (%)
I-II	107 (72.8)	124 (82.7)	231 (77.8)	0.056
III-IV	40 (27.2)	26 (17.3)	66 (22.2)
Histologic grade, n (%)
G1-G2	73 (56.2)	80 (56.3)	153 (56.2)	1.000
G3-G4	57 (43.8)	62 (43.7)	119 (43.8)
LNM, n (%)
Negative	52 (65.0)	55 (69.6)	107 (67.3)	0.651
Positive	28 (35.0)	24 (30.4)	52 (32.7)
ZJU (Protein level)	18	17	35
Age, n (%)
< 55	10 (55.6)	11 (64.7)	21 (60.0)	0.733
≥ 55	8 (44.4)	6 (35.3)	14 (40.0)
FIGO stage, n (%)
I-II	6 (33.3)	12 (70.6)	18 (51.4)	0.043
III-IV	12 (66.7)	5 (29.4)	17 (48.6)
Histologic grade, n (%)
G2	10 (58.8)	14 (82.4)	24 (70.6)	0.259
G3	7 (41.2)	3 (17.6)	10 (29.4)
LNM, n (%)
Negative	8 (44.4)	14 (82.4)	22 (62.9)	0.035
Positive	10 (55.6)	3 (17.6)	13 (37.1)

YWHAG promoted cervical cancer cell proliferation and metastasis in vitro

A series of cell functional assays were performed to elucidate the biological roles of YWHAG in cervical cancer. We knocked down YWHAG using two siRNA in SiHa and HeLa cells. Subsequent RT-qPCR and western blotting analyses indicated a relatively efficient reduction in YWHAG levels in SiHa and HeLa cells (all p < 0.05, Fig. 6D, E). The proliferation capacity of the cells was assessed through CCK-8 and colony formation assays. Knockdown of YWHAG in SiHa and HeLa cells resulted in decelerating growth curves and forming fewer and smaller colonies (p < 0.05, Fig. 6F, G). Transwell and scratch wound healing assays were also conducted to measure cell migration and invasion capabilities. Knockdown of YWHAG significantly inhibited the invasion and migration abilities of SiHa and HeLa cells (p < 0.05, Fig. 6H, I). In conclusion, these results strongly indicated that YWHAG promoted the proliferation and migration of cervical cancer cells in vitro.

Discussion

The latest research has revealed a new mode of programmed cell death: disulfidptosis. This phenomenon occurs under glucose starvation conditions, leading to the aberrant accumulation of cysteine. This phenomenon, arising under glucose-starvation conditions, has captured significant attention due to its intricate connection with tumor metabolism and migration [9,10]. Building on this discovery, we experimentally validated disulfidptosis in cervical cancer cell lines, contributing a new layer to our comprehension of cell death mechanisms. The mechanisms of the interconnectivity between disulfidptosis and cervical cancer were further analyzed.

In our findings, cervical cancer patients were categorized into three molecular subclusters based on the expression of DRGs. These three subclusters exhibited significantly distinct characteristics of TME. Cluster B stood out from the others by harboring many CD8+ T cells and inhibiting the TGF-beta signaling pathway, both associated with immune activation and improved prognosis [34,35]. On the contrary, Cluster C displayed a predominant downregulation in pathways related to immune response and signaling activities. This suggested that clusters associated with disulfidptosis could differentiate the degree of immune infiltration in patients, and cancer-associated inflammation might play a crucial role in patient prognosis [36].

The current staging of cervical cancer, relying on clinical examination, imaging, and surgery, is limited by a rate of understaging or upstaging [37], [38], [39]. Given the reliance on FIGO stage for prognostication, exploring new biomarkers holds promise for more accurate prognostic information, addressing these limitations. Considering the relationship between disulfidptosis, TME and prognosis in DRG clusters, we constructed a disulfidptosis-related prognostic model based on DRGs and DEGs among three clusters. The patients in the low-risk group exhibited better prognoses in both internal and external datasets. Unlike many previous cervical cancer prognostic models trained and tested internally using the TCGA database, our validation extended to external datasets, providing a more comprehensive evaluation [40], [41], [42]. Furthermore, our risk model consistently achieved a 5-year AUC value of 0.7, distinguishing itself from many cervical cancer biomarkers that primarily predicted OS by excelling in predicting both OS and PFS [43], [44], [45]. At the same time, existing classifications struggle to accurately identify patients responding to immunotherapy like pembrolizumab and Cadonilimab, which gained FDA and the National Medical Products Administration (NMPA) of China approval for recurrent/metastatic cervical cancer [28,46,47]. The ongoing search for robust biomarkers in ICB treatment for cervical cancer is a subject of debate across various studies, encompassing factors like PD-L1 expression and TMB [48]. Our study indicated that patients in the low-risk group demonstrated more favorable treatment responses, whether they received anti-PD-1, anti-CTLA4, or a combination of both antibodies. A negative correlation between the majority of MHC-related antigen-presenting molecules and immune checkpoint genes such as CTLA4, TIGIT, and LAG3 with the risk score strengthened the notion that patients with a low-risk score were more likely to experience improved treatment responses to ICB. The findings collectively suggested that the risk score played a role in predicting immunotherapy response for cervical cancer patients.

YWHAG was identified as the most essential gene among the eight genes in the prognostic model. As a member of the 14–3–3 protein family, YWHAG played a critical role in mediating signal transduction by binding to proteins with phosphorylated serine/threonine residues. This molecular function extended to specific cellular signal transduction, cell cycle regulation, and the orchestration of programmed cell death [49]. Notably, our study and previous research indicated a significant positive correlation between YWHAG expression and the risk of cervical cancer [50]. Clinicopathological correlations further supported the essential nature of YWHAG in cervical cancer. Across the spectrum from LSIL to HSIL to SCC, there was a consistent elevation in YWHAG gene expression. Additionally, as the FIGO stage increased, there was a noteworthy escalation in YWHAG expression levels, indicating its potential role in disease progression. Moreover, YWHAG expression negatively correlated with most MHC-related antigen-presenting molecules and B7-H4, suggesting a potential link to immunotherapy efficacy [51,52]. Functionally, our experimental evidence revealed that the knockdown of YWHAG significantly impaired the proliferation, migration, and invasion of SiHa and HeLa cells, underscoring its crucial role in these critical cellular processes. Previous study has also illuminated YWHAG involvement in epithelial-mesenchymal transition (EMT) in tumors, promoting autophagy to shield tumor cells from oxidative damage. Silencing YWHAG in mouse models reduced primary tumor volume, prevented tumor metastasis, and increased median survival, emphasizing its functional significance in cancer progression [53]. In conclusion, the integrated findings from our investigation and previous research established YWHAG as a pivotal player in cervical cancer, influencing molecular signaling, immunomodulation, clinical outcomes, and critical cellular processes. This comprehensive understanding positioned YWHAG as a promising target for further research and potential therapeutic interventions in the context of cervical cancer.

While this study contributed valuable data for creating a novel prognostic model related to disulfidptosis in cervical cancer, several limitations should be acknowledged. Firstly, the TCGA-CESC cohort's data completeness was hindered by the lack of detailed supporting information for some patients, restricting a more comprehensive analysis of clinicopathological characteristics in individuals with CESC. Additionally, the scarcity of publicly available immunotherapy clinical trial data for cervical cancer hindered external validation of our clinical treatment predictions. We will gather clinical specimens to validate further and confirm our findings.

Conclusion

In summary, our study has comprehensively characterized the intricate relationship between disulfidptosis, biological functions, immune infiltration, and clinical outcomes in cervical cancer. The identified prognostic model and potential immunotherapy targets offered promising avenues for advancing precision medicine in managing cervical cancer. Furthermore, YWHAG was identified as a promoter of cervical cancer progression and migration, suggesting its potential as a therapeutic target for cervical cancer treatment. These findings have contributed to a deeper understanding of the disease and opened new avenues for more tailored and effective therapeutic strategies.

Availability of data and materials

Publicly available datasets were analyzed in this study. These can be found in The Cancer Genome Atlas and the NCBI Gene Expression Omnibus. The raw experimental data and analysis codes supporting the conclusions of this article will be made available by the corresponding author.

Ethics approval and consent to participate

This study was approved by the Ethics Committee of Women's Hospital, School of Medicine, Zhejiang University (granted number IRB-20,210,085-R) and the Ethical Committee of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology (approval number TJ-IRB2021207).

Consent for publication

Not applicable.

Funding

This work was supported by the National Key Research and Development Project of China (2021YFC2701200). The study was also endorsed by the Medical and Health Technology Program of Zhejiang Province (Grant No. WKJ-ZJ-2113). The project was funded by China Postdoctoral Science Foundation (2023TQ0289).

CRediT authorship contribution statement

Tianzhe Jin: Writing – original draft, Methodology, Conceptualization. Taotao Yin: Writing – original draft, Validation, Conceptualization. Ruiyi Xu: Resources, Methodology. Hong Liu: Resources, Methodology. Shuo Yuan: Validation, Supervision, Resources. Yite Xue: Validation, Resources. Jianwei Zhang: Writing – review & editing, Supervision. Hui Wang: Writing – review & editing, Supervision.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.