The role of molecular subtypes and immune infiltration characteristics based on disulfidptosis-related genes in ovarian cancer

Abstract

Ovarian cancer (OC) is the most fatal, gynecological malignancy. Compared with advanced ovarian cancer, the 5 year survival rate of early ovarian cancer is significantly improved, and predicting early detection and diagnosis is very important to improve the prognosis of OC. Recent research has found a new way of cell death: disulfidptosis. Under glucose starvation, abnormal accumulation of disulfide molecules such as Cystine in SLC7A11 overexpression cells induced disulfide stress to trigger cell death. Studies of disulfidptosis are still in their infancy and its role in ovarian cancer progression is unclear. In this study, we used a public database to detect the expression and mutations of disulfidptosis-related genes in OC. Cluster analysis was performed based on disulfidptosis-related genes, and disulfidptosis differential expression genes were analyzed. A prognostic risk model was constructed using three disulfidptosis-related genes, and the reasons for differences in prognosis were explored through immune infiltration analysis and drug sensitivity analysis. The prognostic characteristics of transcriptome based on disulfidptosis-related genes are closely related to the prognosis of OC patients. Finally, quantitative polymerase chain reaction (RT-qPCR) was used to detect the expression of three prognostic genes in clinical OC samples.Our study establishes a link between disulfidptosis and OC, providing new ideas for personalized and precise treatment of OC.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-024-01489-w.

Introduction

Ovarian cancer is the eighth most common cancer in women worldwide. It is also the fifth most fatal type of cancer in women。With an incidence of 3.4% and a mortality rate of 4.7%, it affects more than 300,000 women each year and kills about 152,000 women from ovarian cancer [1]. In China, data show that there were about 53,342 new cases and about 37,519 deaths in 2020, accounting for 17.63% and 18.10% of global cases, respectively, highlighting the serious threat posed by this disease to women's health and survival [2]. The 5 year overall survival rate for early-stage ovarian cancer is about 92%, compared to 29% for advanced ovarian [3]. Because the disease has no typical signs and symptoms in its early stages, more than 70% of ovarian cancer patients have progressed to advanced ovarian cancer by the time they are diagnosed [4]. The 5 year survival rate of advanced ovarian cancer is less than 30%, so early diagnosis is important to improve the prognosis of OC [5].

Recent studies have found that glucose starvation-induced high-expression cancer cell death of SLC7A11 does not belong to any of the known types of cell death. This new type of cell death can neither be inhibited by drugs commonly used to inhibit cell death (such as apoptosis, cell necrosis, and autophagy inhibitors), nor can it be prevented by knocking out disulfide/apoptosis key genes. Reducing agents for disulfide stress, such as DL-Dithiothreitol (DTT), 2-Mercaptoethanol (2ME), and Tris (2-carboxyethyl) phosphine (TCEP), can completely inhibit glucose starvation-induced cell death in SLC7A11high cells. In addition, thiol oxidants (diamine and diethyl maleate) promote cell death in SLC7A11high cells under glucose starvation and lead to a sharp accumulation of disulfide molecules within cells. This type of cell death modality is named disulfidptosis [6].

Although the number of confirmed disulfidptosis-related genes (DRGs) is small, a considerable number of studies have shown a potential relationship between DRGs and the development of OC. SNAI2 knockdown promotes ferroptosis in ovarian cancer cells. The mechanism is that SNAI2 binds directly to the promoter of SLC7A11, promotes SLC7A11 expression, and inhibits ferroptosis. Studies have shown that SLC7A11 and GPX4 expression levels are positively correlated with platinum resistance in patients with epithelial ovarian cancer (EOC). High co-expression of SLC7A11 and GPX4 may be an important independent prognostic factor for platinum-based resistance in EOC patients. In another study, HRD1 interacted with SLC7A11 in OC cells, and HRD1 regulated stability and ubiquitination in OC. HRD1 inhibits tumor formation in OC and promotes ferroptosis by enhancing degradation of SLC7A11 [7–9]. SLC7A11 plays an important role in regulating the progression of OC. Other DRGs are also closely related to the development of ovarian cancer. Significantly enhanced expression of FLNA protein predicts a poor prognosis in ovarian cancer patients [10], ovarian cancer cell lines with high FLNA expression are prone to cisplatin resistance, and reducing the expression of FLNA can restore the sensitivity of ovarian cancer cells to cisplatin [11]. MYH9 is a useful independent prognostic marker for epithelial ovarian cancer and is expected to be a target for precision treatment of ovarian cancer in the future [12]. The actinin-4 gene in cancer is also a candidate oncogene associated with poor prognosis and chemoresistance in tumors [13]. A deeper understanding of disulfidptosis-related genes may provide more accurate prognostic biomarkers for ovarian cancer and may reveal new targets for more treatment in ovarian cancer patients. However, there have been no studies to explore in depth whether disulfidptosis-related genes are associated with the prognosis of ovarian cancer.

In this study, we used 2 public databases, 4 sets of ovarian cancer datasets to detect the expression and mutation of DRGs in OC. Cluster typing was carried out based on disulfidptosis-related genes. The differential genes of two cluster were analyzed. Finally, we used three disulfidptosis-related genes to construct an OC prognostic risk model and discussed the reasons for the difference in prognosis from the aspects of immune infiltration analysis and drug susceptibility analysis.

Materials and methods

Data sources and processing

Download RNA-seq data and patient data from both the TCGA database (https://cancergenome.nih.gov/) and the GEO database (http://www.ncbi.nlm.nih.gov/geo/). A combined cohort of 909 patients was analyzed in this study, including data from four datasets: TCGA-OC, GSE26712, GSE19829, and GSE32062. Additionally, downloading simple nucleotide variation (SNV) data for TCGA-OC patients from the UCSC Xena database (http://xena.ucsc.edu/). The FPKM values were transformed into transcripts per kilobase million (TPM), which were treated as comparable to transcripts obtained from the GEO microarray. The measurement of gene expression profiles was achieved by utilizing TPM estimation, followed by log2-based transformation. We utilized Strawberry Perl (version 5.26) to perform ID conversion on the four datasets. Furthermore, we combined and performed batch correction on mRNA expression data from the four datasets using the “limma” and “sva” packages [14], respectively. The twenty genes SLC7A11, GYS1, NDUFS1, NCKAP1, LRPPRC, SLC3A2, RPN1, ACTN4, ACTB, CD2AP, CAPZB, DSTN, FLNA, FLNB, IQGAP1, MYH1O, MYL6, MYH9, PDLIM1, TLN1 are disulfidptosis-related genes (DRGs) retrieved from the currently available publications [6, 15] for drug sensitivity analysis, we obtained the ‘GDSC2_Res.rds’ and ‘GDSC2_Expr.rds’ files from the CCLE database (https://sites.broadinstitute.org/ccle/) [16].

Consensus clustering analysis of DRGs

The “ConsensusClusterPlus” package was utilized for categorizing patients into separate subtypes associated with DRGs based on their gene expression profiles [17]. Performing clustering analysis using the “PAM” algorithm with Euclidean distance as the distance measure. Eighty percent of the samples were randomly sampled and repeated 1000 times. The ideal value of K, representing the number of molecular subtypes, was determined using the Proportion of Ambiguous Clustering (PAC) and Cumulative Distribution Function (CDF) to identify the most appropriate K value.

The Relationship between Molecular Subtypes and Clinical Characteristics and Prognosis of Ovarian Cancer

We compared the clinical features (including age, stage, and tumor grade) among different molecular subtypes using the ‘pheatmap’ package for visualization. Survival time and status were integrated from four datasets (909 patients), and the overall survival (OS) differences among different subtypes were evaluated based on the Kaplan–Meier method. We utilized the “limma” package to identify genes that exhibit differential expression (DEGs) between different molecular subtypes associated with double sulfur death.

Correlation enrichment analysis

We used the ‘c5.go.symbols.gmt’ and ‘c2.cp.kegg.Hs.symbols’ files from the database named MsigDB (https://www.gsea-msigdb.org/gsea/msigdb) [18] to conduct gene set variation analysis (GSVA) and gene set enrichment analysis (GSEA) [19], analyzing the biological functional differences between high-risk and low-risk group and the biological functional differences associated with DRGs. To investigate the potential biological functions of DEGs, Gene Ontology (GO) [20] and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses [21] were performed using the ‘clusterProfiler’ package [22].

Construction of disulfidptosis-related prognostic signature in ovarian cancer

The modeling process in this study involved performing univariate Cox regression analysis ultizing the “limma” package. The analysis focused on differentially expressed genes (DEGs) to identify prognostic-related genes as candidate genes. Subsequently, based on the expression profiles of the candidate genes, patients were classified into different subtypes using cluster analysis for further analysis. Further selection of candidate genes was then performed through multivariable Cox regression analysis, leading to the identification of target genes. Remove samples with missing expression of target genes, and finally include 772 patients to construct the model. The cohort of patients was randomly divided into a training set and a validation set in a ratio of 7:3. The risk score was calculated using the following formula: score = “Σ (Expi * coefi)”, where Σ denotes the sum from i=1 to N, and N represents the count of genes that were selected for analysis.

In this study, patients were divided into high-risk and low-risk groups based on the median of their risk scores. Using the risk stratified data, we generated risk curve plots and risk heatmap using the “pheatmap” package [17]. Survival curves and ROC curves of the model were created using the “survival” package to evaluate the clinical predictive value of the model. We also performed a drill-down analysis on the impact of each gene on the prognosis model using SHAP value dependence analysis based on single features. SHAP values can quantify the impact of the expression levels of each gene on the patient’s prognosis, thereby making this model more effective in guiding medical practice and medical diagnosis [22]. Finally, the model was evaluated for its predictive effect on different clinical characteristics, including age, sex, pathological stage, and grade.

The analysis of tumor microenvironment and immune-related factors

We downloaded the source code “CIBERSORT.R” and reference data files from CIBERSORT official website (https://cibersort.stanford.edu/). By combining the gene expression data from the four datasets and running the source code, we obtained an input sample file called “CIBERSORT-Results.txt,” which provides information on the infiltration of 22 immune cell types in each tumor sample [23].Additionally, based on the gene expression matrix, we utilized the “estimate” package to evaluate the differences in stromal, immune cell abundance, and tumor purity between the two groups of patients [24].

Mutation data processing and Tumor mutation burden

Using SNV data from 407 ovarian cancer patients in the TCGA database, the tumor mutation burden (TMB) was calculated for each patient, where TMB = the number of non-synonymous mutations/exome chip size (approximately 38Mb) [25, 26]. A Perl script was used to obtain the number of non-synonymous mutations in the sequencing data for each patient, which was then adjusted using the above formula. The “GenVisR” package was used to draw gene mutation waterfall plots and the “survival” package was used for survival analysis.

Drug sensitivity analysis

Using the ‘oncoPredict’ package, sensitivity scores for 198 drugs were calculated for 909 patients, allowing for a comparison of the therapeutic effects of targeted biologics between the high and low-risk groups.

Analysis of quantitative reverse TranscriptionPolymerase chain reaction (qRT-PCR)

Both ovarian cancer and adjacent non-cancerous tissues used in this study were derived from patients with gynecological ovarian cancer after surgery in the Guangdong Maternal and Child Health Hospital during 2020–2023. All samples were reviewed by the Ethics Committee and informed consent from OC patients was obtained. We extracted RNA from the specimen using a TRIzol reagent (Ambion, USA) and then reverse-transcribed it into cDNA using a quantitative reverse transcription kit (Promega, USA). Quantitative PCR (qPCR) is a technique for measuring DNA content in sampl in real time. Real-time fluorescence quantitative qPCR assay was performed with the help of SYBR-Green (Vazyme, China) and expression levels were standardized to -actin levels. The primers are shown in Table S6.

Statistical analyses

All data analyses were performed using the following software: R language (version 4.3.0) and Strawberry Perl (version 5.30.0). P-value < 0.05 was considered statistically significant.

Results

Basic information of ovarian cancer patients

A total of 909 patients were included in 4 datasets GSE26712, GSE19829, GSE32062, and TCGA-OC, all of whom had clinical data on survival time and survival status.The details are shown in Tables 1 and 2.

Table 1.

Information of four datasets

Datasets	Platform	Country	Numbers of patiens
GSE26712	GPL96	US	185
GSE19829	GPL570	US	42
GSE32062	GPL6480	US	260
TCGA	Illumina HiSeq	US	422

Table 2.

General information of patients in four datasets

	GSE26712	GSE19829	GSE32062	TCGA-OC
Age (mean ± SD)	Na	56.7 ± 11.8	43.6 ± 28.3	57.3 ± 14.3
Stage (n,%)
I	Na	0	0	Na
II	Na	1 (0.24%)	0	Na
III	Na	35 (83.33%)	204 (78.46%)	Na
IV	Na	6 (14.29%)	56 (21.54%)	Na
Unknow	Na	0	0	Na
Grade (n,%)
I	Na	1 (0.24%)	0	Na
II	Na	9 (21.43%)	131 (50.38%)	Na
III	Na	32 (76.19%)	129 (49.62%)	Na
Unknow	Na	0	0	Na
Fustat (n,%)
Alive	24 (12.97%)	19 (45.24%)	67 (23.85%)	160 (37.91%)
Death	129 (69.73%)	23 (54.76%)	193 (47.23%)	262 (62.09%)
Unknow	32 (17.30%)	0	0	0

Expression and mutation of 20 disulfide-related genes

Figure 1A shows that these 20 DRGs have low mutation rates in ovarian cancer, and FLNB, MYH9, NCKAP1, TLN1, LRPPRC, FLNA, IQGAP1, MYH10 mutation rates are in the top 8. In the copy number variation analysis of DRGs (Fig. 1C), we found that ACTN4, FLNA, IQGAP1 and RPN1 had obvious copy number increase characteristics, while CAPZB, FLNB, ACTB, PDLIM1, GYS1Z COPY NUMBER WAS SIGNIFICANTLY MISSING. Figure 1B shows the localization of DRGs on chromosomes. According to the median expression of 20 DRGs, they were divided into high and low expression groups, and 8 genes were statistically significant in survival between two groups (P < 0.05) (Fig. 1E, Table S1) by KM analysis. Figure 1D illustrates the interaction relationships among 20 DRGs.

Fig. 1.

Expression and mutation profile of DRG in OC. A Mutation frequency of 20 DRGs in the TCGA-OC cohort of 422 patientst. B The locations of CNV alterations in DRGs across 23 chromosomes,The red dots represent gain, while the blue dots represent loss. C The DRGs in TCGA-OV chohrt show instances of gene copy number gain and gene copy number loss. D Interactions among DRGs in OC. Each node represents a gene, the size of the node corresponds to the significance level (p-value), indicating the strength of the association between the gene and prognosis.The red and blue connecting lines represent positive and negative interactions between genes, respectively. E Kaplan–Meier curve demonstrate a significant relationship between the expression of 8 DEGs and prognosis

Identification of DRG subtypes in OC

In order to further investigate the correlation between the expression of 20 DRGs and ovarian cancer, we used cluster analysis to obtain the optimal K value (Fig. 2A, Figure S1A-G), as shown by the lower slope of the CDF curve (Fig. 2B), and divided the patients into two groups A and B when k = 2 (Fig. 2A, Table S2).At the same time, subtype A showed a significant survival advantage (Fig. 2C). After excluding samples with missing clinical data, we compared the clinical and pathological characteristics of 302 patients with A subtype and B subtype (Fig. 2F). Patients with subtype B had significantly higher stage and lower general expression levels of SLC7A11 (p < 0.05). GSVA showed that subtype A was mainly enriched in signaling pathways between neurons, cell death, and cell cycle regulation related pathways. Subtype B is mainly related to organ development and morphological establishment, biological metabolism, and immunomodulatory pathways. Notably, the B subtype is also enriched in the peroxisome proliferator-activated receptor (PPAR) signaling pathway, which is one of the important regulatory mechanisms for the formation and breaking of disulfide bonds (Fig. 2E). Finally, we utilized the CIBERSORT tool to find that subtype A has higher levels of activated dendritic cells, and subtype B has higher levels of macrophages, NK cells, neutrophils, and plasma cells (Fig. 2D), indicating that disulfidptosis may have an impact on the tumor microenvironment.

Fig. 2.

Clinical features, biological functions, and immune infiltration differences between two DRG subtypes. A Cluster analysis results based on DRGs with K = 2. B Cumulative Distribution Function (CDF) curves for k = 2 to 9. C Kaplan–Meier curve showing the overall survival (OS) differences between the two subtypes of DRGs. D GSVA heatmap showcases the biological enrichment pathways between two subtypes of DRGs,in which Red color indicates positive activation of the pathway, while blue color represents negative inhibition of the pathway. E Heatmap illustrates the relationship between expression data of 20 DRGs and clinical pathological features of 302 ovarian cancer patients. F differences in the infiltration levels of 23 immune cells between the two subtypes of DRGs (*p < 0.05, **p < 0.01, ***p < 0.001)

Identification of gene subtypes based on gene expression patterns associated with DEGs

In order to further investigate the underlying biological behaviors associated with two DRG subtypes, we firstly identified 24 DEGs between subtypes A and B, including SLC7A11, SLC5A1, LYPD1, CLU, CHST1, ID4, TMOD1, BBOX1, GJB1, HOXA5, DEFB1, SFRP1, CYP4B1, KLK7, PAEP, CCNA1, CLDN10, SERPINA5, PGAM2, MMP7, SST, IL3RA, ICAM4, IL9R (Table S3). In the GO analysis, these DEGs were relatively enriched in stable biological processes such as reproductive development, alveolar development, and cell maturation (Fig. 3A, B), and the JAK-STAT signaling pathway was involved in the KEGG enrichment analysis, suggesting that DEGs were involved in regulating cell growth, differentiation, and apoptosis(Fig. 3C,D). In our study, we further narrowed down the selection to 10 genes from the DEGs using COX model. These 10 genes were found to be closely associated with prognosis (Table 3). Additionally, through clustering analysis, we classified patients into different disulfidptosis-related genes subtypes (Fig. 3E, Table S4). From the KM curve, it can be observed that patients in the subtype A had longer survival time (P < 0.05) (Fig. 3F). Furthermore, the boxplot revealed significant statistical differences in the expression patterns of 10 DRGs between these two gene subtypes (Fig. 3G).

Fig. 3.

Gene subtype analysis based on DEGs. A, B GO enrichment analyses between the two gene subtypes. C, D KEGG enrichment analyses between the two gene subtypes. E The result of the clustering analysis was the division of patients from the four datasets into two gene subtypes (K = 2). F KM analysis compares the survival time of patients between the two gene subtypes. G Expression levels of 20 DRGs were compared between the two gene subtypes. The number of m5C peaks in HCT15 cell and sh-NSUN2 HCT15 cell on each mRNA. Most mRNAs have only one methylation peak

Table 3.

Univariate analysis showing 10 DEGs related to the OS time

Id	HR	HR.95L	HR.95H	P value
SLC7A11	0.900891	0.841984	0.96392	0.002486
LYPD1	0.828253	0.740291	0.926667	0.001004
ID4	0.799233	0.664473	0.961323	0.017377
BBOX1	0.904026	0.828972	0.985874	0.022508
GJB1	0.842333	0.761418	0.931847	0.000869
HOXA5	1.067494	1.002331	1.136892	0.042112
SFRP1	1.112865	1.008302	1.228273	0.033655
CYP4B1	0.879322	0.815373	0.948287	0.000843
CLDN10	0.890529	0.814335	0.973853	0.011068
SST	0.947319	0.900311	0.996782	0.037152

Construction of disulfidptosis-related prognostic signature in ovarian cancer

Ten candidate genes were screened by univariate COX model (Fig. 4A), and three genes for risk model construction (Table S5) were identified after further multivariate COX regression analysis. Risk score = (− 0.097*expression of SLC7A11) + (− 0.152*expression of LYPD1) + (− 0.082*expression of CYP4B1).After the single-feature analysis of SHAP, it can be obtained that the lower the gene expression of the gene SLC7A11, the higher the SHAP value, that is, the higher the risk of death of the patient, which is consistent with the previous analysis results. Based on the purpose of gene data smoothing, after log2 processing the gene expression amount, we can further obtain that when the gene expression of SLC7A11 is specifically < 1, it will have a more negative impact on the prognosis of ovarian cancer patients. When the expression of gene CYP4B1 is less than 1 or the expression amount is greater than 8, it is the most unfavorable to the prognosis of ovarian cancer patients. When the gene LYPD1 is highly expressed, the SHAP value will increase, that is, the risk of death will increase, indicating that the high expression of gene CYP4B1 will adversely affect the prognosis of ovarian cancer patients (Fig. 4F). A series of subsequent prognosis and risk analyses were carried out on the risk models constructed from the total set (n = 771), training set (n = 540) and validation set (n = 231) (Fig. 4G–K, Figure S2A-E, Figure S3A-E). The analysis results of the three-cohort data were consistent, showing better prognosis in low-risk patients (Fig. 4G, Figure S2A, Figure S3A), and the value of the model in predicting patient prognosis at years 1, 3, and 5, respectively, is shown in Fig. 4H, Figure S2B and Figure S3B. As the risk increases, so does the number of ovarian cancer deaths (Fig. 4 J, K, Figure S2D, E, Figure S3D,E). Introducing clinical data of patient age, tumor stage and tumor grade, the model still had independent predictive value in univariate Cox regression analysis (Fig. 4L), while there was no significant difference in multivariate Cox analysis (P > 0.05, Fig. 4M). Additionally, we found that DRG subtypes, gene subtypes, and high/low-risk groups were closely related (Fig. 4B–D), with better prognosis for DRG subtype A, gene subtype A, and low-risk group, while worse prognosis for DRG subtype B, gene subtype B, and high-risk group. Figure 4E showed the expression differences of 20 DRGs between high and lowrisk groups.

Fig. 4.

Construction of disulfidptosis-related prognostic signature. A Prognosis-related DEGs were identified by Cox univariate regression analysis. B Sankey diagram illustrating the patient scores and survival status in the risk model for all subtypes. C, D Comparison of risk scores between DRG subtypes and gene subtypes. E Expression of 20 DRGs in the high and low-risk groups. F Predicting patient prognosis based on the relationship between SHAP values and the expression of three target genes. (The x-axis represents gene expression levels (log2 transformed), and the y-axis represents SHAP values. Higher SHAP values indicate a greater probability of patient death). G Kaplan–Meier curve shows different overall survival (OS) between two risk score groups. H ROC curve demonstrates the predictive accuracy of the risk model for patient prognosis at 1, 3, and 5 years. I Expression of 3 DEGs in the high and low-risk groups. J, K Ranked dot and scatter plots showing the Risk score distribution and patient survival status, respectively. L, M COX analysis was performed to discover the independent predictive ability of the risk model for ovarian cancer patient prognosis

TMB analysis and survival analysis of TMB

The analysis results of immune cell profiling in the high-risk and low-risk groups indicate a negative correlation between activated dendritic cells, memory activated CD4 + T cells, follicular helper T cells, and Risk_score. On the other hand, there is a positive correlation between memory resting CD4 + T cells, γδ T cells, and Risk_score (Fig. 5A). Figure 5B shows how closely the 3 DRGs are associated with the level of infiltration of these immune cells. The immuneScore in the low-risk group is higher compared to the high-risk group (p < 0.05) (Fig. 5C). The waterfall plot shows all gene mutations between two risk groups (Fig. 5D, E). After calculating and comparing the TMB in all ovarian cancer patients, we observed a significant disparity in TMB between the two risk groups (Fig. 5F). Figure 5G shows that as the risk score increases, patients have a higher tumor mutation burden, and this difference in TMB levels significantly impacts patient survival (Fig. 5H). Next, we further explored the association between risk score and TMB and found a negative correlation between the two. Survival analysis indicated that patients with high TMB and low risk had the best survival stutas (Fig. 5I). Finally, we found through GSEA analysis that the main enrichment functions of the high-risk group were adipocytokine signaling pathway, extracellular matrix receptor interaction, neuroactive ligand receptor interaction, and immune system function and regulation, indicating that high-risk patients may be involved in the development of ovarian cancer and the formation of tumor microenvironment through the above pathways and functions. The enrichment pathway in the low-risk group involved graft immunity, antigen processing and presentation, and biological processes associated with flagella and cilia (Fig. 5J, K).

Fig. 5.

TMB analysis and immune microenvironment analysis. A The linear relationship between risk scores and the levels of immune cells in tumor patients. B The correlation between immune cell abundance and three genes in the risk model. C Correlations between Risk score and Stromal Scores, Immune Score and ESTIMATE Score. D, E Waterfall plots summarizing the mutation status of high and low-risk patients. F The difference in tumor mutation burden between high and low-risk score groups. G Correlation between risk score and TMB. H Kaplan–Meier curves of high and low TMB groups. I Kaplan–Meier curves of four groups classified by risk score and TMB. J The enriched KEGG pathways in two risk groups of patients, respectively. K The enriched GO pathways in two risk groups of patients, respectively

Drug susceptibility analysis

With the “oncoPredict” package, we analyzed 198 drugs and found that 92 drugs showed significant sensitivity differences (p < 0.05) between the two groups (Figure S4). Notably, targeted drugs Olaparib (Fig. 6A), Talazoparib (Fig. 6B), Cediranib (Fig. 6C), Foretinib (Fig. 6D), and first-line ovarian cancer chemotherapy drugs Oxaliplatin (Fig. 6E), Docetaxel (Fig. 6F) demonstrated better treatment effects in the high-risk patient group, whereas Afatinib (Fig. 6G), Lapatinib (Fig. 6H), and KRAS (G12C) Inhibitor-12 (Fig. 6I) showed better treatment effects in the low-risk group.

Fig. 6.

Relationships between DRG Score and Drug susceptibility

Single-cell level research

Through the analysis of the dataset GSE118828, we found that these three key genes are not only commonly highly expressed in malignant cells (Fig. 7A–F), but also SLC7A11 and CYP4B1 are highly expressed in endothelial cells (Fig. 7C, D, F), and LYPD1 and CYP4B1 are also highly expressed in myofibroblast cells (Fig. 7D–F). These findings suggest that these genes may also play a role in stromal cells outside of malignant cells.

Fig. 7.

Expression of prognostic related genes. The results showed that the expression of (A) LYPD1, (B) SLC7A11, (C) CYP4B1 in clinical OC tissues and adjacent normal tissues

Real-time quantitative reverse transcription PCR (qRT-PCR)

To determine whether 3 disulfidptosis prognostic genes are differentially expressed in ovarian cancer tissues, we used qRT-PCR to analyze the expression of each gene in 3 pairs of clinical OC tissues and adjacent normal tissues.The results showed that the expression level in OC tissues was different from that in normal tissues (Fig. 8A–C).

Fig. 8.

BVHggh

Discussion

OC is the tumor with the highest mortality rate among gynecologic tumors, of which platinum resistance is the main reason for treatment failure in ovarian cancer patients. Platinum-resistant ovarian cancer patients are those whose disease relapses within 6 months of achieving a complete response (CR) or partial response (PR) with first-line platinum-based chemotherapy [27], for whom chemotherapy drugs have very limited efficacy [28], and who respond less than 30% to subsequent platinum-based chemotherapy regimens [29].Therefore, there is an urgent need for new and more effective strategies to improve clinical outcomes in ovarian cancer patients. Disulfidptosis opens new possibilities for cancer treatment. The metabolism of disulfide is closely related to the regulation of redox status, which is also closely related to the proliferation and metabolism of tumor cell [30]. Cancer cells increase their survival and tolerance by altering intracellular redox [31]. The study found that some chemotherapy drugs, such as cisplatin and paclitaxel, or targeted drugs and immune checkpoint inhibitors, can exert their anticancer effects through intracellular disulfide reactions [32]. Therefore, inducing disulfidptosis in cancer cells is expected to become a promising new direction for tumor treatment.

The treatment plan for advanced OC patients often requires individualized customization. The same treatment plan can produce different treatment outcomes and prognoses. Currently, two research has explored the potential relationship between DRGs and ovarian cancer. In comparison to the study by Jin M and his team [33], our study includes a more comprehensive set of genes, particularly incorporating four key inhibitory genes: SLC7A11, SLC3A2, RPN1, and NCKAP1, as well as two core actin-related proteins: GYS1 and NDUFS1. Additionally, we have employed a more scientific and comprehensive approach in the selection of gene characteristics. We first used LASSO analysis combined with univariate and multivariate COX regression analysis, a method suitable for analyzing large datasets and capable of avoiding overfitting. We also further explained the impact of three key genes on the prognosis of OC patients through SHAP analysis, which is a machine learning method that can overcome the black box problem of machine learning by quantifying the degree to which each clinical indicator affects patient prognosis through interpretable machine learning methods. Thus, we can intuitively see the specific relationship between the expression levels of these three genes and patient prognosis. When the SHAP value is negative, i.e., when the gene expression level of SLC7A11 < 1 (after log2 transformation), the prognosis of ovarian cancer patients shows a significant deterioration trend. At the same time, when the expression level of CYP4B1 is less than 1 or greater than 8, and the gene LYPD1 is highly expressed, the prognosis of ovarian cancer patients is the worst. This analysis result is consistent with the risk coefficient of the model, further verifying the accuracy of the model construction. Compared with simple linear analysis, SHAP analysis is more in line with clinical reality and has greater guiding significance for clinical practice and is worth further exploration. Compared to the methods used by Cong Y and his team [34], we have used the expression levels of disulfidptosis-related genes to additionally perform subgroup classification of ovarian cancer patients. Different subtypes are involved in varying degrees of immune cell infiltration, molecular functions, and prognoses. Patients in cluster A had better prognosis than those in cluster B, and patients in cluster B had significantly higher stage classification and lower SLC7A11 expression levels, as well as richer immune cell infiltration. Next, we identified 24 differentially expressed genes between the two subtypes, which could be divided into two gene clusters, with significant prognostic differences between the gene clusters and corresponding to DRG cluster relationships. In the analysis of immune-related outcomes, low-risk patients exhibited a higher tumor mutation burden, indicating a potential increased sensitivity to immunotherapy. The risk model we finally constructed demonstrated greater clinical value in the assessment of drug sensitivity tests, we focused on the observation that Oxaliplatin, Docetaxel, Olaparib, Talazoparib, and Cediranib showed better therapeutic effects in high-risk patients. Oxaliplatin, Docetaxel, and Olaparib are currently the first-line chemotherapeutic agents for ovarian cancer, while Olaparib and Talazoparib are the preferred treatment options for patients with BRCA-mutated ovarian cancer. Cediranib, as a VEGF receptor inhibitor, is believed to reduce the oxygen and nutrient supply in the tumor microenvironment, thereby increasing the sensitivity of tumor cells to PARP inhibitors. Multiple clinical studies have shown that the combination of Cediranib with Olaparib in ovarian cancer demonstrates a synergistic antitumor effect and good tolerability [35–38]. Other targeted drugs such as Afatinib, Lapatinib, and KRAS (G12C) Inhibitor-12, among others, suggest better therapeutic effects in the low-risk group. These findings are more conducive to the screening of individualized patients and guiding tumor treatment. In summary, our data reveal the potential functions of disulfide death-related genes and their predictive ability for the prognosis of ovarian cancer (OC).

SLC7A11 in OC and its relationship with prognosis have been confirmed [7–9]. CYP4B1 is thought to be a biomarker associated with long-term survival and chemotherapy resistance mechanisms in ovarian cancer [39].Elevated urine LYPD1 mRNA levels are a positive prognostic factor in patients with ovarian cancer [40]. LYPD1 is widely expressed in primary and metastatic ovarian cancer. Anti-LYPD1/cd3 T cell-dependent bispecific antibodies (TDBs) can be used to redirect T cell responses to LYPD1-expressing ovarian cancer in PBMC recombinant immunodeficient mouse and human CD3 transgenic mouse models, thereby inducing potent LYPD1 polyclonal T cell activation and target-dependent killing, ultimately producing an effective in vivo anti-tumor response [41].

Our research results indicate that there are different immune microenvironments among DRG subtypes in terms of immune cell infiltration, which can affect the occurrence, development, and treatment efficacy of tumors. Patients with subtype A have higher levels of activated dendritic cells (ADCs), and the biggest feature of ADCs is that they can stimulate the proliferation of naïve T cells and kill tumor cells [42], and some studies have found that the number, activity, and high degree of infiltration of ADCs may be associated with a better prognosis of OC [43] and an open, randomized phase II trial demonstrated that dendritic cell-based anticancer vaccines extended OS by 13.4 months in ovarian cancer patients [44]. However, patients with subtype B have higher levels of macrophages, NK cells, neutrophils, and plasma cells, and the presence of macrophages can promote the growth, invasion, and metastasis of ovarian cancer and inhibit the immune response, suggesting that it is associated with a poor prognosis in cancer patients [45, 46], researchers have found that modulating tumor-associated macrophages (TAMs) polarization by targeting monoamine oxidase A (MAOA) can enhance the effectiveness of cancer immunotherapy [44].

Finally, we performed qRT-PCR detection of 3 disulfidptosis related genes associated with prognosis in patients with OC. These results demonstrate the accuracy of our first differential analysis, improve the credibility of subsequent studies, further validating the predictive power of our model.

Conclusion

In summary, we clustered genes related to disulfidptosis, divided OC patients into two subtypes, and then analyzed the differences between the two subtypes. Subsequently, three genes were selected through regression analysis to quantify the relationship characteristics between the genes and prognosis, and an OC prognosis prediction model was constructed. Immunological and drug sensitivity analyses were performed to explore personalized stratification screening for OC patients. The results show that the predictive model we built can predict the prognosis of OC patients. Although our disulfidptosis-related prediction model has outstanding ability to identify the immune status of OC patients and predict patient prognosis, this study still has certain limitations. Data analysis based on public database data can lead to deviations between predictions and actual conditions. In the future, more data on OC patients will need to be used to further validate the usefulness and accuracy of the model.

Author contributions

Conceptualization: YL, ZP H Methodology: RR Y, XS H YT W Formal analysis: ZF W Writing-original draft: YL, RR Y Writing-review & editing: All authors Supervision: XS H Project administration: YL.

Funding

This study was supported by Guangzhou Science and Technology Foundation of China (No.202002030174).

Data availability 

The datasets generated during and/or during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by Committee on Medical Ethics of Guangdong Women and Children Medical Hospital and the study was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. No ethical approval was required in accordance with the TCGA and GEO Research Data User Agreement.

Consent for publication

No consent for publication was required since our manuscript adhered to the TCGA and GEO Research Data User Agreement.

Competing interests

The authors declare no competing interests.