Comprehensive analysis for the immune related biomarkers of platinum-based chemotherapy in ovarian cancer

Abstract

Background

Ovarian cancer (OC) is one of the most lethal gynecological malignancies. This study aimed to identify biomarkers that were sensitive to platinum-based chemotherapeutic agents and can be used in immunotherapy and explore the importance of their mechanisms of action.

Methods

RNA-seq profiles and clinicopathological data for OC samples were obtained from The Cancer Genome Atlas (TCGA) and cBioPortal platform, respectively. Platinum-sensitive and platinum-resistant OC samples in the TCGA cohort were selected based on the clinical information. RNA-seq data for 70 OC samples withSingle-sample gene set enrichment analysis (ssGSEA) and unsupervised clustering were used to classify OC patients from the TCGA cohort into clusters with different proportions of infiltrating immune cells. ESTIMATE analysis was used to assess the immune landscape among clusters. Differential expression, univariate Cox regression, and LASSO regression analyses were performed to construct prognostic model. Spearman correlation analysis was conducted to investigate the correlations among immune checkpoint inhibitors (ICIs) and risk score, half-maximal drug inhibitory concentration (IC₅₀) and risk score.

Results

Using ssGSEA and unsupervised clustering, OC samples were divided into two clusters with different immune cell infiltration. Then, 1715 differentially expressed immune-related genes (DEIRGs) were identified between two clusters, 984 differentially expressed platinum-sensitive related genes (DEPSRGs) between 149 platinum-sensitive and 63 platinum-resistant OC samples were identified, and 5384 differentially expressed genes (DEGs) between 380 OC and 194 normal samples were detected from the TCGA cohort. Six biomarkers (GMPPB, SRPK1, STC1, PRSS16, HPDL, and SPTSSB) were detected to establish a prognostic model. The OC patients in the TCGA cohort were classified into high- and low-risk groups. The receive operating characteristic (ROC) curve was plotted and demonstrated that the prognostic model performed well with the area under ROC curve (AUC) greater than 0.6. The expressions of 5 ICIs, including CD200, TNFRSF18, CD160, CD200R1, and CD274 (PD-L1), were significantly different between two risk groups, and the risk score was significant negative associated with CTLA4, TNFRSF4, TNFRSF18, and CD274. Moreover, there were significant differences in IC₅₀ of 10 chemo drugs between two risk groups, patients in the high-risk group could be more resistant to po0tinib, dasatinib, and neratinib.

Conclusion

In summary, this study constructed a novel prognostic model based on six prognostic biomarkers, including GMPPB, SRPK1, STC1, PRSS16, HPDL, and SPTSSB, which can be utilized for predicting the prognosis of OC patients. These biomarkers were the potential therapeutic targets.

Introduction

Ovarian cancer (OC) is the most common malignant tumors of the female reproductive system,which is the third highest incidence of gynecologic tumors and the highest mortality rate [1]. About 90% of ovarian cancers are epithelial in origin Epithelial Ovarian Cancer (EOC) [2] and has the highest mortality rate among epithelial ovarian cancers [3]. The five-year OS of OC is only 40% [4]. Surgery, combined with platinum-based chemotherapy, is the main treatment for patients with OC. Chemotherapy is effective in 80% of patients with OC [5]. Different resistance mechanisms have been idetified, however, the mechanism of platinum resistance has not been completely elucidated. The development of resistance to chemotherapy in OC patients further complicates OC treatment outcomes [6]. Therefore, identifying biomarkers related to the occurrence, development and prognosis of OC is important for early diagnosis and treatment, as well as for finding new targets and treatment methods.

Tumor heterogeneity refers to the existence of subpopulations of cells, with distinct genotypes and phenotypes that may harbor divergent biological behaviors, within a primary tumor or between a primary tumor and its metastases [7]. The heterogeneity, including inter-tumor heterogeneity and intra-tumor heterogeneity, is significant in tumor progression and clinical choices [8].

It is generally accepted that platinum drugs bind to DNA and the DNA damage triggers apoptosis of tumor cells [9]. But the mechanisms that tumors development resistant to platinum drugs are multifactorial, including: reduced intracellular drug accumulation, intracellular inactivation of the agent, increased DNA repair, or impaired apoptotic signaling pathways [10]. Wang et al. reported miR-211 improved the prognosis of ovarian cancer patients by enhancing the chemosensitivity of cancer cells to platinum via inhibiting DNA damage response gene expression [11]. Knockout of both alleles of CTR1 reduces initial cDDP influx [12], enhances efflux [13], and renders cells highly resistant to cDDP. Meanwhile, platinum-based chemoresistance is a major obstacle to the treatment of high-grade serous ovarian cancer (HGS-OvCa). Fang Lei et al. showed that lncRNA plays an important role in platinum resistance in HGS-OvCa patients and delineated a lncRNA-mRNA co-expression network in HGS-OvCa patients exhibiting platinum resistance [14].Gao Ce et al. Based on the ovarian cancer cell line (SKOV3 and SKOV3\DDP) to study the potential role of lncRNA in EOC, and also constructed a lncRNA-mRNA co-expression network [15].

The past decade has witnessed the rapid development of immunotherapy. The immunotherapy enhances the body's response tumor natural immune defense, remodeling of immune microenvironment [16]. The current immunotherapy for EOC can be divided into three categories:Immune checkpoint inhibitors, Therapeutic vaccines and adoptive cellular immunotherapy.Some reseachers reported that PD-L1 expressions were strongly associated with TILs and stem cell markers in ovarian cancer [17]. Therefore, the establishment of biomarkers that are sensitive to platinum-based chemotherapy drugs and can be used for immunotherapy and the establishment of prognostic correlation models are of great significance for understanding the heterogeneity between OC at the molecular level and improving the treatment of these cases.

Materials and methods

Ovarian cancer data collection

Bulk RNA-seq data of 380 OC samples were obtained from The Cancer Genome Atlas data portal (TCGA,https://portal.gdc.cancer.gov/), and subsequently combined with clinical information downloaded from cBioPortal (https://www.cbioportal.org/) database, 365 OC samples with corresponding survival data were obtained. 194 normal ovarian samples were downloaded from Genotype-Tissue Expression (GTEx, https://commonfund.nih.gov/GTEx/) database. Then, 149 platinum-sensitive and 63 platinum-resistant OC samples in the TCGA cohort were detected according the platinum status. RNA-seq data for 70 OC samples associated with survival data from GSE63885 (platform: GPL570) dataset and RNA-seq data for 107 OC samples associated with survival data from GSE26193 (platform: GPL570) dataset were acquired from Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) database. 43 immune checkpoint inhibitors (ICIs) were obtained from previous study [18].

Grouping of OC samples by ssGSEA algorithm

Single-sample gene set enrichment analysis (ssGSEA) algorithm was utilized to calculate the relative infiltration levels of 28 immune cell types in each OC sample from the TCGA cohort using GSVA (Version 1.38.2) R package [19]. Unsupervised clustering was conducted on 380 OC samples from the TCGA cohort based on the relative abundance of 28 immune cell types using pheatmap (version 1.0.12) to classify OC samples into different clusters. Stromal score, immune score, and ESTIMATE score between two clusters were calculated by ESTIMATE algorithm using estimate (Version 1.0.13) R package [20].

Differential expression analysis

Differential expression analysis was conducted on immune high and immune low groups to screen differentially expressed immune-related genes (DEIRGs) using limma (Version 3.46.0) R package with P < 0.05 as the cut-off value [21]. Differentially expressed platinum-sensitive related genes (DEPSRGs) between 149 platinum-sensitive and 63 platinum-resistant OC samples were identified by limma R package with P < 0.05 as the threshold. Differentially expressed genes (DEGs) between 380 OC patients and 194 normal samples were detected by limma R package with P < 0.05 and logFC| > 0.5 as cut-off values. Candidate genes were obtained by intersecting DEIRGs, DEPSRGs, and DEGs using a Venn diagram.

Functional annotation of candidate genes

Metascape (https://metascape.org/gp/index.html#/main/step1) database was adopted to explore ontology terms of the candidate genes [22].

Establishment of prognostic model

To acquire genes in the TCGA cohort with prognostic value the candidate genes were enrolled in univariate Cox regression analysis with the cutoff value of P < 0.05. Subsequently, the results were enrolled in least absolute shrinkage and selection operator (LASSO) regression Cox analysis using glmnet (Version 4.1–1) R package to detect biomarkers [23]. Subsequently, the risk score of each sample was calculated by below formula:

$$Riskscore=\sum _{i=1}^{n}coef\left(gen{e}_i\right)\times \exp r\left(gen{e}_i\right)$$

coef: coefficient values of each gene; expr: expression level of each gene.

The median risk score was adopted to divide OC samples in the TCGA cohort into high- and low-risk groups. The Kaplan-Meier (KM) survival curve for overall survival (OS) in the high-and low-risk groups was plotted using the survminer (version 0.4.8) R package (http://finzi.psych.upenn.edu/library/survminer/html/ggsurvplot.html). Receiver operator characteristic curve (ROC) was drawn using survivalROC (Version 1.0.3) R package, and area under the ROC curve (AUC) was calculated to assess the diagnostic effectiveness. GSE26193 dataset was utilized to validate the prognostic model.

Construction and evaluation of a nomogram

Risk score and clinicopathological characteristics (age, grade, stage, and tumor residual) of OC patients in the TCGA cohort were enrolled in univeriate Cox regression analysis, P < 0.05 were considered statistically significance. Furthermore, multivariate Cox regression analysis was performed based on the results of univariate cox regression analysis to detect independent prognostic factors with P < 0.05. A nomogram was established based on the detected independent prognostic factors using rms (Version 6.2–0) R package to predict 1, 3, and 5 years survival probability of OC patients in the TCGA cohort (https://cran.r-project. org/web/packages/rms/). Calibration plot was performed to graphically evalue the prediction ability of the nomogram.

Immune heterogeneity

The expression levels of 43 ICIs were compared between high- and low-risk groups and visualized using ggpubr (Version 0.4.0). Spearman's correlation analysis was conducted to explore the correlations between risk score and ICIs. P < 0.05 was the cutoff value. Immunophenoscore (IPS) calculated scores were based on four categories: (1) immunosuppressive cells (SC), (2) MHC molecules (MHC), (3) effector cells (EC), and (4) immune Checkpoints (CP) [24], through which could be used to derive patient's immunoscore without bias by machine learning. Combined four categories to obtain the IPS z-score, a higher z-score indicating a more immunogenic sample. Besides, SC, MHC, EC, CP, and IPS z-scores between high- and low-risk groups were compared with P < 0.05 as the cutoff value.

The lncRNA-biomarker and TF-biomarker interaction network

To mine lncRNAs associated with biomarkers, spearman correlation analysis was conducted to explore the correlations between biomarkers and lncRNAs from the TCGA cohort (P < 0.01 and |Cor| > 0.3) to construct interaction networks. In order to further probe the upstream regulating factors of biomarkers, human transcription factors (TFs) of the related biomarkers were explored through NetworkAnalyst (https://www.networkanalyst.ca/) database, and selected TF-gene Interactions database ENCODE: Only peak intensity signal <500 and the predicted regulatory potential score <1 was used [25]. Moreover, we visualized the interaction network among lncRNA-biomarker and TF-biomarker using Cytoscape tool.

Drug sensitivity analysis

RNA-seq expression profiles, National Cancer Institute 60 (NCI-60) compound activity data, and half-maximal drug inhibitory concentration (IC₅₀) data of 163 drugs were downloaded from CellMiner (https://discover.nci.nih.gov/cellminer/loadDownload.do) database [26]. The relationships among risk score and IC50 of 163 drugs were explored by Spearman correlation analysis. P< 0.05 were consudered statistical significance.

Specimen collection

The samples from 20 ovarian cancer patients and 20 normal ovarian tissuses were collected from 2019 to 2022. The diagnosis of all patients was confirmed by surgery and pathology results in the Benxi Central Hospital. The written informed consent form was obtained prior to study initiation from each patient. Samples were promptly frozen in liquid nitrogen after surgical dissection and maintained at −80 °C deep cryogenic refrigerator until use.

Cell culture

Human ovarian cancer cell lines (SKOV3, A2780) and their resistant cell lines (SKOV3/DDP, A2780/DDP) were purchased from Cell Bank (Shanghai Institute for Biological Science). All cells were cultured in RPMI-1640 medium (Solarbio, Beijing, China) supplemented with 10% fetal bovine serum (FBS, Thermo Fisher, Wilmington, DE, USA) and 1% penicillin/streptomycin (Solarbio, Beijing, China) and maintained in a 37 °C humidified incubator with 5% CO2. For SKOV3/DDP and A2780/DDP cells, 50 nM DDP was added to the culture medium to maintain DDP resistant phenotype.

RNA extraction and real-time quantitative PCR (qRT-PCR)

Total RNA was isolated from tissues and cells using Trizol (Beyotime, Shanghai, China) according to the manufacturer's protocol. The complementary DNA (cDNA) was obtained using the specific reverse transcription kit (RR047A, Takara, China). Gene expression was determined using SYBR Premix Ex Taq (cat no. RR420A; Takara) and calculated using the 2^−ΔΔCt method. The expression of β-actin was used as internal control.

Cell transfection

Small interfering RNAs against six genes (si-GMPPB, si-SRPK1,si-STC1, si-PRSS16, si-HPDL, and si-SPTSSB), as well as their corresponding negative controls (si-NC) were synthesized by RiboBio (Guangzhou,China). Cell transfection was performed using Lipofectamine 3000 transfection reagent (Invitrogen) when cells reached to 60–80% confluence.

Transwell assay

To determine cell invasion ability, transwell chambers in 24-well plates were used. Briefly, transfected ovarian cells were suspended in serum free medium, and 2 × 10⁵ cells/per well were plated into the upper chamber (Corning, NY, USA), in which the upper surface of the filter was precoated with Matrigel (Solarbio, Beijing, China), whereas the bottom chamber contained RPMI-1640 complete medium with 10% FBS. After culture for 48 h, the invaded cells were stained and imaged. For transwell migration assay, similar procedure was performed except that the upper surface of the filter in the upper chamber was not coated with Matrigel.

Statistical analysis

All data were shown as the mean ± standard deviation from at least three independent experiments. Spearman analysis was used to analyze the relationship of associated factors. Statistical analysis was performed using SPSS 26.0. P < 0.05 was considered statistically significant.

Results

Construction of immune-associated OC subgroups and identification of 74 candidate genes

To quantify the proportions of 28 immune cell types in each sample from the TGGA cohort, ssGSEA was conducted and visualized with a heatmap (Fig. 1A). Through Unsupervised clustering analysis, the OC samples in the TCGA cohort were divied into cluster 1 (n = 234) and cluster 2 (n = 131) (k = 2), and the proportions of 28 immune cell types in cluster 1 were higher than that in cluster 2, namely immune high and immune low subgroups respectively. Stromal score, immune score, and ESTIMATE score of immune high subgroup were significant higher than that of low-immune subgroup (Fig. 1B).

Fig. 1.

Identification of 74 candidate genes in two immune-associated OC subgroups. A Infiltrating heatmap of 28 immune cell types in each sample from the TGGA. B Differences in stromal score, immune score, and ESTIMATE score in two immune-associated OC subgroups. C Volcano plot of differentially expressed immune-related genes (DEIRGs) between immune high and immune low subgroups. D Venn diagram of 74 candidate genes common to the three subtype classifications. E Gene Ontology (GO) enrichment analysis of the candidate genes. F Metascape enrichment network analysis of the 74 candidate genes.

In order to select the potential crutial genes of different immune subgroups, totals of 1715 DEIRGs between immune high and immune low subgroups were identified with 294 genes up-regulated and 1421 genes down-regulated in immune high subgroup (Fig. 1C, Supplementary Table 1). The expressions of top 50 DEIRGs were shown in Supplementary Fig. 1A. And meanwhile, a total of 5384 DEGs between 380 OC and 194 normal samples were detected with 1915 genes up-regulated and 3469 genes down-regulated in OC samples (Supplementary Fig. 1B, Supplementary Table 2). There were 984 DEPSRGs between 149 platinum-sensitive and 63 platinum-resistant OC samples with 254 genes up-regulated and 730 genes down-regulated in OC smaples (Supplementary Fig. 1C, Supplementary Table 3). Therefore, 74 candidate genes were obtained by intersecting 1715 DEIRGs, 984 DEPSRGs, and 5384 DEGs (Fig. 1D, Supplementary Table 4). The bar grouph demonstrated that candidate genes were significant correlated to 20 ontology terms including negative regulation of cell-cell adhesion, reactive oxygen species metabolic process, regulation of lipid localization, response to bacterium, and positive regulation of cytokine production et al.(Fig. 1E). Metascape enrichment network was created to visualize enriched terms, one node for each enriched term, connecting pairs of nodes with Kappa similarity above 0.3 (Fig. 1F).

Establishment of a prognostic model based on six biomarkers

Univariate Cox regression analysis was implemented to identify candidate genes associated with the prognosis of OC, and six candidate prognostic genes with P < 0.05 were detected, including GMPPB, SRPK1, STC1, PRSS16, HPDL, and SPTSSB (Fig. 2A). LASSO Cox regression analysis with 10-fold cross-validation was carried out to construct a prognostic model. When lambda_min = 0.01385, the optimal prognostic model was constructed, and six biomarkers (GMPPB, SRPK1, STC1, PRSS16, HPDL, and SPTSSB) with corresponding coefficient values were obtained (Fig. 2B). Risk score of each sample was calculated with below formula:

$$\begin{array}{c}Riskscore=\left(-0.13712\right)\times \exp ressionlevelofGMPPB+\left(-0.01692\right)\hfill \\ \times \exp ressionlevelofSRPK1+\left(-0.0479\right)\hfill \\ \times \exp ressionlevelofSTC1+\left(-0.04188\right)\hfill \\ \times \exp ressionlevelofPRSS16+\left(-0.07331\right)\hfill \\ \times \exp ressionlevelofHPDL+\left(-0.10781\right)\hfill \\ \times \exp ressionlevelofSPTSSB+\hfill \end{array}$$

Hence, 365 OC patients with clinicopathological data in the TCGA set were stratified into high- and low-risk groups by the median risk score (1.2427). KM survival curves demonstrated that OC patients in high-risk group showed significantly poorer OS compared to low-risk group, and the expression of six biomarkers in the high-risk group was lower than that in low-risk group (Fig. 2C, Supplementary Fig. 2). The AUC value of the risk model for survival prediction reached 0.692 at 1-year, 0.641 at 3-year, and 0.606 at 5-year, indicating the risk model had a favorable predictive value in both short- and long-term follow-up (Fig. 2C). And meanwhile, GSE63885 and GSE26193 datasets were utilized to validate the prognostic model, and consistent results were achieved (Supplementary Fig. 3A–H). It is noteworthy that the expression of all six biomarkers was up-regulated in the OC tissues rather than normal samples (Fig. 2D).

Fig. 2.

Search for the six biomarkers and construction of nomogram. A Univariate Cox regression analysis to identify candidate genes associated with the prognosis of OC. B Six biomarkers were screened by the least absolute shrinkage and selection operator (LASSO) model and cross-validation diagram for tuning parameter selection. C OC patients with clinicopathological data were subgroup into two groups, displaying the kaplan-meier (K-M) curves and receiver operating characteristic (ROC) curves to verify accuracy of the risk model in survival prediction. D The expression patterns of six biomarkers in OC and normal tissues. E Univariate and multivariate Cox regression analysis showed the independent prognostic factors for OS in OC patients. F The nomogram and the calibration curve for clinical utilize.

Next, Chi-square tests was performed to investigate the differences in clinicopathological characteristics between high- and low-risk groups. Baseline demographic and clinicopathological characteristics of OC patents in high- and low-risk groups in TCGA cohort were shown in Table 1. The results demonstrated strikingly differences in age at diagnosis, survival status, platinum status, tumor residual, therapy, and immune subgroups between high- and low-risk groups (P < 0.05). Moreover, results from the runk sum test suggested that age (≤ 60 vs. > 60), stage (Stage II vs. Stage III, stage II vs. Stage IV), immunity (high-immunity vs. Low-immunity), survival status (dead vs. alive), platinum status (sensitive vs. resistant), Therapy (CR vs. PR, CR vs. PD, and CR vs. SD), and tumor residual (no macroscopic VS. >20 mm) were significant differences in high- and low-risk groups (P < 0.05), demonstrating that six biomarkers were significantly associated with the prognosis of OC (Supplementary Fig. 4A and B).

Table 1.

Baseline demographic and clinicopathological characteristics of OC patents in high- and low-risk groups.

Characteristics	Total	High-risk group	Low-risk group	P-value
age(year)
Mean (SD)	59.7 (±11.3)	63.0 (±10.6)	56.6 (±11.1)	< 0.001
Vital
Alive	164 (44.9%)	63 (34.6%)	101 (55.2%)	< 0.001
Dead	201 (55.1%)	119 (65.4%)	82 (44.8%)
Stage
II	19 (5.2%)	6 (3.4%)	13 (7.1%)	0.18
III	288 (79.6%)	142 (79.3%)	146 (79.8%)
IV	55 (15.2%)	31 (17.3%)	24 (13.1%)
Grade
G2	42 (11.8%)	22 (12.4%)	20 (11.2%)	0.75
G3	315 (88.2%)	156 (87.6%)	159 (88.8%)
Platinum status
Resistant	63 (24.3%)	42 (36.2%)	21 (14.7%)	< 0.001
Sensitive	149 (57.5%)	57 (49.1%)	92 (64.3%)
Tooearly	47 (18.1%)	17 (14.7%)	30 (21.0%)
Therapy
Complete response	201 (68.6%)	78 (56.5%)	123 (79.4%)	< 0.001
Partial response	43 (14.7%)	26 (18.8%)	17 (11.0%)
Progressive disease	27 (9.2%)	21 (15.2%)	6 (3.9%)
Stable disease	22 (7.5%)	13 (9.4%)	9 (5.8%)
Tumor residual
> 20mm	69 (21.2%)	41 (24.6%)	28 (17.7%)	0.039
1–10 mm	169 (52.0%)	86 (51.5%)	83 (52.5%)
11–20 mm	26 (8.0%)	17 (10.2%)	9 (5.7%)
No macroscopic disease	61 (18.8%)	23 (13.8%)	38 (24.1%)
Cluster
Immunuty high	201 (55.1%)	119 (65.4%)	82 (44.8%)	< 0.001
Immunuty low	164 (44.9%)	63 (34.6%)	101 (55.2%)

Therefore, the clinicopathological characteristics was involved into the univariate and multivariate Cox regression analysis to investigate whether risk score was an independent prognostic indicator. Factors of tumor residual (HR, 1.163, 95% CI, 1.003–1.347, p = 0.045), and risk score (HR, 1.550, 95% CI, 1.194–2.012, P < 0.001) were selected as independent prognostic factors for OS in OC patients (Fig. 2E). A nomogram was constructed predicting the probability of survival for OC patients at 1, 3, and 5 years (Fig. 2F). The C-index value was 0.6135, suggesting favorable discrimination of the nomogram, and the calibration curve of the nomogram demonstrated a high consistency between the predicted and observed survival probability.

Relationships between prognostic model and immune microenvironment

To further investigate the correlations between risk score and ICIs, the current study analyzed the differences in the expression of 43 ICIs between high- and low-risk groups, and the results demonstrated that the expressions of CD200, TNFRSF18, CD160, CD200R1, and CD274(PD-L1) were significant differences between high- and low-risk groups (P < 0.05) (Fig. 3A). The risk score was significant negative associated with CTLA4, TNFRSF4, TNFRSF18, and CD274 (correlation coefficient (Cor) < −0.1, P < 0.05) (Fig. 3B). On the other hand, six biomarkers were significant negative associated with BTLA, KIR3DL1, CD200, CD86, CD40LG, HAVCR2, CD200, CD244, CD160, LAIR1, NRP1, and TNFSF14 (Cor < −0.1, P < 0.05) (Fig. 3C). Results for the differences in MHC, EC, SC, and CP scores between two risk groups showed that MHC score was significantly higher in high-risk group, while CP score was significantly lower in high-risk group compared to low-risk group (Fig. 3D), and IPS z-score was significantly lower in high-risk group than low-risk group (Fig. 3E). Spearman correlation analysis indicated that PRSS16 wre significantly prositive associated with IPS z-score and MHC, while negative associated with CP, HPDL were significantly prositive associated with SC, while negative associated with EC, GMPPB was negative associated with SC and CP (Fig. 3F).

Fig. 3.

Comparison of immune infltration in OC. A Comparison of expression of immune checkpoint inhibitors (ICIs) between two risk groups. B Correlation bubble chart for ICIs with risk score. C Correlation bubble chart for ICIs with six biomarkers. D Comparison the MHC molecules (MHC), effector cells (EC), immunosuppressive cells (SC) and immune Checkpoints (CP) scores between two risk groups. E Boxplot for Immunophenoscore (IPS) z-score between the two groups. F Spearman correlation analysis showed the relationship between six biomarkers and MHC, EC, SC, CP, average z-score (AZ).

The TF-biomarker interaction network and chemotherapy drugs targeting six biomarkers

In order to explore the underlying mechanism relevent to biomarkers in OC, 85 lncRNAs significantly related to biomarkers were selected and shown in Supplementary Fig. 5A. We noted that 36 lncRNAs (KTN1-AS1, LINC01976, KCTD21-AS1, LINC01816, etc.) were significantly linked to HPDL, 15 lncRNAs (GAS5, SNHG29, CIRBP-AS1, AC068580.3, etc.) were significantly correlated with GMPPB, 13 lncRNAs (LINC00239, LYRM4-AS1, LINC01269, etc.) were significantly correlated with SPTSSB, 9 lncRNAs (LINC01012, HCG18, KIFC1, etc.) were significantly associated with SPK1, and 9 lncRNAs (HCG27, LINC02315. LINC01339, etc.) were significantly related to PRSS16, and 3 lncRNAs (HIF1A-AS3, LUCAT1, AL731533.2) were significantly connected to STC1. And meanwhile, totally 138 TFs obtained from NetworkAnalyst database were took intersection with DEGs to obtain 11 differentially expressed TFs (DE-TFs) in OC, and the interaction networks for DE-TFs and lncRNAs with biomarkers were shown (Supplementary Fig. 5A and B). In this network, KDM5B regulates both HPDL, SRPK1, and PRSS16. BCL11A regulates both GMPPB, SRPK1, and PRSS16, and INSM2 and DDX20 regulate both SRPK1 and PRSS16.

Furthermore, CellMiner database was adopted to predict the response to chemotherapy for each sample in TCGA-OC cohorts. Pearson correlation analysis showed that 27 drugs were correlated to risk score, where the correlation coefficients between risk score and IC₅₀ of po0tinib, dasatinib, and JNJ-42756493 were greater than 0.4 (Supplementary Fig. 5C). Hence, the 11 chemo drugs with Cor > 0.3 were used to assesse the IC₅₀ for each sample in the TCGA cohort. There were significant differences in the estimated IC50 between high- and low-risk groups, in which patients in the high-risk group could be more resistant to po0tinib, dasatinib, and neratinib (Supplementary Fig. 5D).

Six biomarkers regulates the migration and invasion ability in resistant OC cells

Six biomarkers knockdown were confirmed by qRT-PCR.(Supplementary Fig. 6A–F). We uesd the transwell assays to investigate the migration and invasion ability of 6 biomaker. The results showed that si-GMPPB, si-HPDL, si-PRSS16, si-SPTSSB, si-SRPK1, si-STC1 inhibited the invasion and migration ability in SKOV3/DDP(Fig. 4A-L) and A2780/DDP(Fig. 4M-X).

Fig. 4.

Effect of six biomarkers knockdown on SKOV3/DDP(A-L) and A2780/DDP (M-X) . A Transwell examined the efects of GMPPB on cell invasion. B Transwell examined the efects of HPDL on cell invasion. C Transwell examined the efects of PRSS16 on cell invasion. D Transwell examined the efects of SPTSSB on cell invasion. E Transwell examined the efects of SRPK1 on cell invasion. F Transwell examined the efects of STC1 on cell invasion.G Transwell examined the efects of GMPPB on cell migration. H Transwell examined the efects of HPDL on cell migration. I Transwell examined the efects of PRSS16 on cell migration. J Transwell examined the efects of SPTSSB on cell migration. K Transwell examined the efects of SRPK1 on cell migration. L Transwell examined the efects of STC1 on cell migration.M Transwell examined the efects of GMPPB on cell invasion. N Transwell examined the efects of HPDL on cell invasion. O Transwell examined the efects of PRSS16 on cell invasion. P Transwell examined the efects of SPTSSB on cell invasion. Q Transwell examined the efects of SRPK1 on cell invasion. R Transwell examined the efects of STC1 on cell invasion.S Transwell examined the efects of GMPPB on cell migration. T Transwell examined the effects of HPDL on cell migration. U Transwell examined the effects of PRSS16 on cell migration. V Transwell examined the effects of SPTSSB on cell migration. W Transwell examined the effects of SRPK1 on cell migration. X Transwell examined the effects of STC1 on cell migration.

Discussion

Platinum-based chemotherapy is the essential treatment for OC [27]. Most patients with OC will experience repeated recurrence and gradually progress from platinum-sensitive to platinum-resistant [28]. However, the optimal therapy for patients with platinum-resistant OC is unknown. Therefore, understanding the molecular mechanisms of chemotherapy resistance in OC is critical for improving the poor prognosis.Meanwhile, some researches showed OC has been identified as an immunogenic tumor. Immunotherapy should be optimized to be included as part of OC therapeutics [29]. In this study, we explored the involvement of immunotherapy in platinum-resistant OC.

By using DEIRGs, DEPSRGs, and DEGs analysis, we found 74 candidate genes in the TCGA datasets. Then, after univariate Cox regression analysis, LASSO regression analysis, and multivariate Cox regression analysis, a six-gene risk score was established, including GMPPB, HPDL, PRSS16, SPTSSB, SRPK1 and STC1. GDP-mannose pyrophosphorylase B (GMPPB) gene encodes beta subunit of an enzyme which catalyses the formation of GDP-mannose required in the glycosylation of α-dystroglycan [30]. It had been reported to affect this glycosylation process [31]. GMPPB could be used as one of group impact factor for evaluating the prognosis of endometrial carcinoma [32]. HPDL, is a paralog of HPD (4-Hydroxyphenylpyruvate Dioxygenase). It is a protein coding gene for 4-hydroxyphenylpyruvate dioxygenase like protein [33]. HPDL was related to neurological diseases [34] and cancers [33]. PRSS16 encoded the hymus-specific serine protease (TSSP) [35], which is involved in CD4+ T cell maturation in the thymus, exerts a tumor suppressor activity [36]. SPTSSB is a small SPT subunit that stimulates SPT activity and confers acyl-CoA preference to the SPT catalytic heterodimer of SPTLC1 and either SPTLC2 or SPTLC3 [37]. In a study published by Li e al., castration-resistant prostate cancer xenografts significantly upregulated SPTSSB, both in response to castration as well as resistance to enzalutamide [38]. On the other hand, accumulation of dihydrosphingosine caused by an activating mutation in SPTSSB has been shown to lead to neurodegeneration [39]. SRPK1 is a protein kinase that specifically phosphorylates proteins containing serine–arginine-rich (SR) domains [40]. SRPK1 has been demonstrated to be involved in various physiological and pathological processes [41]. In OC, SRPK1 and lncRNA influenced the chemoresistance of OC cells to cisplatin [42,43]. STC1 is a glycoprotein associated with calcium and phosphorus metabolism [44]that has been detected in various human tissues [45]. In addition, dysregulated STC1 expression has been identified in various cancers [46]. In OC, STC1 serves an oncogenic role and promotes metastasis and chemoresistance [47]. In our study, we first found GMPPB, HPDL, PRSS16 and SPTSSB were high expression in OC tissues than normal tissues. Relevant studies in cisplatin resistance have been not reported.

Then, Next, we constructed the six-gene signature of the prognostic risk model by performing univariate/multivariate Cox regression and LASSO regression analyses. We found tumor residual and risk score were independent prognostic factors for OS in OC patients. The risk model we constructed had good predictive value for the 1-, 3- and 5-year survival rates of patients.After completing these works, we analyzed 43 ICIs expression between high- and low-risk groups. The risk score was significant negative associated with CTLA4, TNFRSF4, TNFRSF18, and CD274. Ovarian cancers are known to evade immunosurveillance and to orchestrate a suppressive immune microenvironment [48]. CTLA-4 downregulates immune responses and has been identified to be an immune checkpoint for cancer immunotherapy [49]. With the gradual deepening of the immune checkpoints of CTLA-4 and PD-1/PD-L1 and the approval of their blocking antibodies for the treatment of various tumors, other immune checkpoints as TIM3, TIGIT and LAG-3 also attract great attention [50]. In recent years, immunotherapies, particularly those targeting immune checkpoints, have been considered a new and effective strategy for ovarian cancer [51]. In addition, six biomarkers were associated with some ICPs, including BTLA, KIR3DL1, CD200, CD86, CD40LG, HAVCR2, CD200, CD244, CD160, LAIR1, NRP1, and TNFSF14, indicating that these genes could be an effective indicator for immune checkpoint blockage therapy. The higher the immunogenicity score, the greater is the immunogenicity of the epitope by IPS z-score. To identify genes that interact or affect six genes, we analyzed lncRNAs and transcription factors associated with genes. Some reports showed that immune-related lncRNA paired model for predicting the prognosis and immune-infiltrating cell condition in OC [52]. Through systematic analysis, we found platinum resistance-specific lncRNA-mRNA networks targeting key biomarkers. Wang et al. found SRPK1 participated in cisplatin resistance associated with lncRNA UCA1 in human OC cells [53]. Another reported showed that the UCA 1 / miR-99b-3p / SRPK 1 axis may be a novel target for the treatment of ovarian cancer [54].However, further functional analysis and the prognostic impact of related lncRNA on OC patients with platinum-based chemotherapy resistance need subsequent analysis, which can be supplemented as future research prospects.

Then, we uesd Pearson correlation analysis to check the relationgship between risk score and IC₅₀ of po0tinib, dasatinib, and JNJ-42,756,493_. Wen et al. found that combined treatment of sunitinib, dasatinib, and everolimus—results in simultaneous inhibition of multiple signaling pathways and a better anti-tumor activity than any single treatment [55]. Dasatinib may act as a sensitizer to restore the sensitivity of platinum drugs to OC cells [56].Neratinib could be used to treat the ovarian cancer patients with ErbB2 overexpression [55]. Our study found that patients in the high-risk group could be more resistant to po0tinib, dasatinib, and neratinib. It is worth that these chemotherapy drugs are not currently used for the patient population with OC, and a number of preclinical investigations based on a larger clinical cohort is warranted.

Finally, we analyzed the expression and biological function of six biomarkers in OC by experiments. qRT-PCR tests of the patient tissues confrmed that the expression of the six genes in cancer tissues were higher compared to normal tissues. Using OC cells, we demonstrated that knockdown of six genes inhibited the invasion and migration of cancer cells. A key issue that cannot be ignored is that, considering the considerable differences observed in immune cell populations and the small differences observed in knockdown studies, the decrease in the mRNA levels of the six biomarkers in OC tissues and the above-mentioned inhibitory ability are because the changes of these genes in tumor cells or in non-tumor cells within the tumor microenvironment cannot be clearly defined yet. Further cellular experimrnts and single-cell sequencing data are needed.

Our study demonstrated six genes as prognostic biomarkers for OC, highlighting their potential as predictive biomarker and an immunotherapy target.Our study is a retrospective study based on public data, and the clinical application of biomarkers requires data support from more samples. Future studies will be necessary to determine the mechanism responsible for promoting cisplatin resistance and predicting prognosis prediction in OC.

Funding

No funding.

Declarations

Ethics approval and consent to participate The studies involving human participants were reviewed and approved by the Ethics Committee of Benxi Central Hospital (Benxi, China). The patients/participants provided their written informed consent to participate in this study. Consent for publication Not applicable.

Author contributions

JXH propoesed the conceptualization. JL taked charge the data curation and formal analysis. YYL was in charge of project administration. CJY was in charge of methodology. JJL was in charge of writing original draft.

Declaration of Competing Interest

The authors declare that they have no competing interests. We don't have the third party to provide the reported work. We don't have any financial interest or relationship within the last 3 years.

Data availability

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