Abstract Ovarian cancer (OC) usually progresses rapidly and is associated with high mortality, while a reliable clinical factor for OC patients to predict prognosis is currently lacking. Recently, the pathogenic role of neutrophils releasing neutrophil extracellular traps (NETs) in various cancers including OC has gradually been recognized. The study objective was to determine whether NETs-related biomarkers can be used to accurately predict the prognosis and guide clinical decision-making in OC. In this study, we utilized univariate and multivariate Cox regression to identify key prognostic features and developed a model with six NETs-related lncRNAs, selected via LASSO regression. The model’s predictive capability was assessed through Kaplan–Meier, ROC, and Cox analyses. To understand the model’s mechanisms, we conducted GO term analysis, KEGG pathway enrichment, and GSEA. We also analyzed gene mutation status, tumor mutation load, survival rates, and model correlation. Additionally, we compared immune functions, immune checkpoint expression, and chemotherapy sensitivity between risk groups. Besides, we validated the model’s predictive value using test data and tissues acquired from our institution. Finally, we performed in vitro and in vivo experiments to confirm the expression of model lncRNAs and the cellular level function of GAS5. We developed a model using six NETs-associated lncRNAs: GAS5, GBP1P1, LINC00702, LINC01933, LINC02362, and ZNF687-AS1. The model’s predictive performance, evaluated via ROC curve, was compared with traditional clinicopathological features. GO process analysis highlighted molecular functions related to antigen binding and immune system biological processes. Variations were observed in transcription regulators affecting immune response, inflammation, cytotoxicity, and regulation. We also predicted IC50 values for chemotherapeutic drugs (bexarotene, bicalutamide, embelin, GDC0941, and thapsigargin) in high- and low-risk groups, finding higher IC50 values in low-risk patients. The risk model’s robustness was validated using OC cells, tissues, and clinical datas. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-025-97548-5. Keywords: Neutrophil extracellular traps (NETs), LncRNA, Ovarian cancer, Prognosis, Targeted therapy Subject terms: Gynaecological cancer, Biotechnology, Cancer, Biomarkers, Health care Introduction Epithelial ovarian cancer, which constitutes approximately 90% of all ovarian cancer cases, ranks as the third most prevalent gynecological malignancy and is the leading cause of gynecologic cancer-related mortality^[36]1,[37]2. The primary reason for the high mortality rate is the extensive metastatic spread, primarily in the abdomen. The standard approach for treating patients with advanced epithelial ovarian cancer remains cytoreductive surgery and platinum-based chemotherapy^[38]3. Although the majority of patients with advanced stage disease (i.e., stage III-IV) initially respond to front-line therapy consisting of cytoreductive surgery and platinum-based chemotherapy, 60–80% of patients with advanced disease will experience recurrence^[39]4. In most cases, cancer recurrence occurs at progressively shorter intervals, and patients eventually develop drug resistance. Currently, the five-year survival rates for patients with advanced ovarian cancer remain below 50%^[40]5. This indicates that existing prognostic and predictive tools based on standard clinicopathological factors are still insufficient to accurately guide clinical decision-making. Therefore, there is a need for new biomarkers to predict the prognosis and drug sensitivity of ovarian cancer patients. Neutrophils, as the most prevalent leukocytes in the bloodstream of healthy humans, have increasingly been recognized for their dual roles in both tumor suppression and promotion^[41]6–[42]8. However, specific mechanism of the neutrophils acting on cancer cells remains to be explored. Recently, research showed that neutrophils releasing NETs can be detected in the omentum of women with early-stage ovarian cancer, and blockade of NETs can prevent omental metastasis^[43]9. Consequently, NETs, as critical factors in ovarian cancer metastasis, warrant comprehensive investigation. Released by neutrophils, NETs consist of their own nuclear and mitochondrial DNA decorated with cytosolic and granule proteins such as neutrophil elastase (NE) and myeloperoxidase (MPO). The discovery of NETs was first reported by Brinkmann in 2004^[44]10. NETs are formed in response to various factors, including pathogens^[45]11, activated platelets, and phosphonic myristate acetate (PMA)^[46]12. Recent experiments have validated that targeting PAD4^[47]13–[48]15 and DNase I^[49]16,[50]17 can effectively reduce the formation of NETs or neutralize their impact on tumor cells, thereby inhibiting tumor growth and metastasis. Additionally, promising results have been observed in studies targeting receptors involved in the interaction between tumors and NETs, such as TLR4-NE, tumor-specific integrins (α5β1 and ανβ3), and CCDC25 surface proteins^[51]18. NETs play significant roles in the biology of solid tumors, influencing oncogenesis, tumor growth, invasion, metastasis, and drug resistance. Targeting NETs holds significant potential in combating tumors and improving the prognosis of cancer patients. NETs-related genes are also considered as novel biomarkers associated with prognosis^[52]19–[53]23. Hence, we aim to determine whether factors associated with NETs can serve as valuable predictive and prognostic indicators to guide diagnosis and treatment strategies in OC. Research showed that by suppressing lncRNA MIR503HG, NETs promote non-small cell lung cancer metastasis^[54]24, which reminds us that lncRNA may have complex yet unrealized potential relation with NETs and NETs inserted proteins or other bioactive molecules. LncRNAs regulate transcription, mRNA processing, and nuclear organization^[55]25. Recent studies have provided increasing evidence of the significant roles lncRNAs play in human cancer^[56]26, by promoting tumor cell proliferation, invasion, and metastasis^[57]27–[58]29. Furthermore, lncRNAs may serve as novel therapeutic targets and potential biomarkers for diagnosing and forecasting human cancer^[59]30. For instance, LINK-A overexpression in ovarian cancer boosts cell migration and invasion via TGF-β1 upregulation^[60]31. Additionally, lncRNAs have been linked to chemotherapy resistance in the treatment of ovarian cancer^[61]32. Additionally, a lncRNA model related to NETs has been developed to accurately predict non-small cell lung cancer prognosis^[62]33. The objective of this study was to investigate the role of NETS-associated lncRNAs in predicting prognosis and treatment response in ovarian cancer, with the aim of enhancing long-term patient survival and quality of life. Materials and methods Origin and handling of the dataset Transcriptome RNA (RNA sequencing, RNA-seq) data was acquired from The Cancer Genome Atlas (TCGA, [63]https://portal.gdc.cancer.gov/) and Genotype-Tissue Expression Project ([64]https://gtexportal.org/home/) for a combined count of 379 tumor samples obtained from patients with ovarian cancer, along with 88 samples from healthy individuals. Additionally, we gathered the relevant clinical data. Following the exclusion of patients with a follow-up period of less than 10 days to prevent any bias caused by immortal time, our study ultimately included 371 individuals diagnosed with OC. We utilized the Perl programming language (version:5.30.1, [65]http://www.perl.org/) along with assistance from the Ensemble database ([66]http://asia.ensembl.org/index.html) to transform Ensemble IDs into a collection of gene symbols. Based on annotation information from the GENCODE database ([67]https://www.gencodegenes.org/), we categorized RNA molecules into messenger RNAs (mRNAs) or long non-coding RNAs (lncRNAs). The analysis was conducted utilizing the R statistical language (version 4.2.2, [68]https://www.r-project.org/). Construction of NETs-related LncRNAs risk model In order to build the predictive model, we selected 23 genes that have been linked to NETs according to previous research^[69]34–[70]36. By utilizing the ‘limma’ R package (version:3.60.6), we conducted Pearson’s correlation analysis and identified 128 lncRNAs associated with NETs genes. Selection criteria included |R| > 0.4 and p < 0.001, as stated in reference^[71]37. The associations between genes related to NETs and lncRNAs were displayed using the R software package ‘ggalluvial’ (version:0.12.5). The research group was made up of training (n = 186) and test (n = 185) sets, which were divided randomly and equally. Univariate Cox regression analysis was utilized to identify lncRNAs associated with prognosis in OC patients (p < 0.05). The ‘limma’ R package was employed to compute the lncRNAs associated with prognosis in ovarian cancer. We conducted logistic regression analysis using the ‘glmnet’ package (version:4.1-8), applying the Least absolute shrinkage and selection operator (LASSO) method. The optimal penalty parameter (λ) values were determined using ten-fold cross-validation. A prognostic signature with optimal performance was subsequently constructed through the utilization of multivariate Cox regression analysis. The RS was determined by the model using the following calculation: graphic file with name d33e382.gif Assessment and verification of prognostic predictive model In order to facilitate the application in clinical, the patients in the training set were categorized into two groups: high-risk and low-risk, based on the median risk score as a benchmark. To validate the precision of the created signature in forecasting prognosis, we employed the ‘survminer’ (version:0.5.0) and ‘survival’ (version:3.8.3) R bundles to produce Kaplan-Meier survival plots. Furthermore, we evaluated the accuracy and precision of forecasting the longevity of OC individuals at 1, 3, and 5 years utilizing the ‘timeROC’ R software (version:0.4) and time-varying receiver operating characteristics (tROC) curves as previously described^[72]38. The predictive model was used to calculate the risk score for patients in the validation set. We classified the patients into high-risk or low-risk groups by using the median risk score as a threshold. In addition, we conducted K-M survival analysis and time-dependent ROC analysis to precisely assess the predictive capacity and usefulness of the prognostic signature. By conducting survival analyses and utilizing the log-rank test, we examined how clinicopathological factors like age, stage, and grade, influenced overall survival outcomes, using the risk score as a basis. Cell culture, transfection and clinical samples collection The IOSE human ovarian tissue cell line and the A2780 and SKOV3 human ovarian cancer cell lines were purchased from Shanghai Zhong Qiao Xin Zhou Biotechnology co.Ltd. ID8 + luc cell line was purchased from Shanghai iCell Bioscience Inc. Cells were cultured in RPMI 1640 medium and DMEM medium supplemented with 10% bovine serum albumin, 1% streptomycin and penicillin. The specimens were kept in a moist incubator at 37 °C with a 5% CO2 level. The transfection process involved the use of riboFECT^TMCP reagent and Lipofectamine™ 2000 (Invitrogen, USA) to perform the introduction of siRNA and overexpression plasmid, following the guidelines provided by the manufacturer. The oligonucleotide sequences of the siRNAs are listed below: si-h-GAS5_1: GCAAGCCTAACTCAAGCCA; si-h-GAS_2: GGACCAGCTTAATGGTTCT. The GAS5 overexpression plasmid was constructed as follows: GCAAAGGACTCAGAATTCA. Between 2021 and 2023, a total of 20 sets of primary epithelial ovarian cancer tumor samples and corresponding adjacent tissue samples were gathered from patients who had cytoreductive surgery at the Second Affiliated Hospital of Harbin Medical University. Every patient had recently been diagnosed with epithelial ovarian cancer (EOC) and had not undergone any anti-tumor therapy before the surgical procedure. Prior to sample collection, informed consent was obtained from each patient after the study received approval from the Medical Ethics Committee (YJSKY2023-303) of the hospital. All researches were performed in accordance with relevant regulations, and statements confirming that informed consent were obtained from all participants or their legal guardians. For further analysis, the samples were kept at a temperature of − 80 °C. Cell colony formation, invasion and wound healing assays For colony formation experiments, normal ovarian cancer cell lines (1 × 10^3) and genetically engineered ovarian cancer cell lines overexpressed / knocked-down GAS5 (1 × 10^3) were incubated in 6-well plates at 37 °C with 5% CO2. The cells were plated in the center of the plates. The medium was replaced every three days for about 14 days. Following that, the colonies were treated, dyed using 1% crystal violet, and tallied. For cell invasion studies, Boyden chambers with 8 μm wells (Corning, USA) were coated with Matrigel (Corning, USA). After being cultured for 12 h, 4 × 10^4 cells were inoculated in the upper chamber without fetal bovine serum. The lower chamber was filled with a medium containing 10% fetal bovine serum. The chambers were incubated at 37 °C with 5% CO[2] and 95% air for 24 h. Non-migrated cells were removed from the upper chamber using a cotton swab. The migrated cells on the lower side of the membrane were fixed, stained, and counted under a microscope. For the wound healing experiment, cells were cultured for a duration of 12 h, then digested and placed in 6-well plates with a 10% FBS solution. After cells reached about 90% confluency, we made an incision on the surface of the cell using a 200 µl pipette tip, followed by three washes with PBS. Following this, the cells were placed in a medium that did not contain FBS. A microscope was used to capture images of the scratch at various time intervals. The calculation of cell migration involved subtracting the width at different time points from the width at 0 h, and then dividing the result by the width at 0 h. Animals and in vivo experiments The female C57BL/6 mice aged 6–8 weeks used in all animal experiments were obtained from Liaoning Changsheng Biotechnology co.Ltd. Mice were raised in a pathogen-free environment at the animal facility of Harbin Medical University. The Medical Ethics Committee of the Second Hospital of Harbin Medical University approved all research protocols with approval number YJSDW2023-006. All animal experiments were performed in accordance with the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. Systemic inflammation of C57BL/6 mice was induced by intraperitoneal injection in LPS group (n = 5) (Lipopolysaccharide, 10ug/mouse, Sigma), and corresponding amount of solvent was given in control group (n = 5). The LPS-induced inflammation model was established and 6 h later, intraperitoneal metastasis ovarian cancer models were created by intraperitoneally injecting 5 × 10^5 ID8 + luc ovarian cancer cells labeled with luciferase into two groups of mice^[73]39. The tumor metastasis burdens were visualization by bioluminescence imaging after eight weeks. Briefly, mice were anesthetized and injected intraperitoneally with firefly luciferase substrate D-luciferin (150 mg/kg), and optically imaged 5 minutes later using the IVIS 200 system (Xenogen). Before mice were sacrificed, tissues from the vivo tumors and peripheral blood samples were collected for further studies. Examining the levels of gene expression in cells and tissues The TRIzol reagent (Invitrogen, Waltham, MA, United States) was used to extract total RNA from cells, mouse tissues and clinical tissues. Using SYBR Green qPCR, the qRT-PCR method was employed to detect the levels of the six candidate lncRNAs. Afterward, the RNA was converted into cDNA using a reverse transcription kit (TaKaRa, Tokyo). Supplementary Table 1 contains the primers utilized in this investigation. An internal control was utilized to measure the level of GAPDH. A Quant Studio 3 (Thermo Fisher Scientific, Waltham, MA, USA) was used for all RT-qPCR reactions. The levels of expression were normalized using the relative quantitative method of 2 − ΔΔCT. ELISA analysis of NETs To detect the levels of NETs in murine plasma, the Mouse CitH3 ELISA Kit (Jingkang, Shanghai) was used. The experiments were conducted according to the manufacturer’s instructions. External validation To validate the prognosis, a group of patients from our hospital was included in the study as an independent external validation cohort. The risk score (RS) was calculated based on the expression levels of RT-qPCR and the signature formula. Patients were then categorized into high- and low-risk groups based on their risk score. To validate the prognostic signature, Kaplan-Meier survival analysis, univariate Cox regression analysis, and multivariate Cox regression analysis were performed for survival analysis, ensuring the validity and reliability of the findings. Construction and analysis of the competing endogenous RNA network To begin, the microRNA (miRNA) was downloaded from the miRcode database ([74]http://www.mircode.org/). Perl scripts were then utilized to predict lncRNA-miRNA interactions. Subsequently, mRNA target prediction databases such as TargetScan, miRDB, and miRTarBase were consulted to identify the mRNA targets of microRNA as previous literature reported^[75]40. The co-expression network of NETs-related lncRNAs, miRNAs, and mRNAs was constructed with Cytoscape 3.9.0 using a lncRNA-miRNA-mRNA ceRNA network. Principal component analysis and functional enrichment analysis PCA analysis was conducted using the ‘scatterploted’ R package (version:0.3–44) to identify patient subgroups at different risks. The ‘limma’ R package was used to analyze the differential expression of genes (DEGs) between these groups, applying criteria of |log2 fold change (FC)| > 0.4 and p < 0.05. To investigate the biological functions of the NETs-related DEGs, the present study performed a functional pathway enrichment analysis through GO analysis and KEGG pathway enrichment using the R ‘clusterProfiler’ software package (version:4.12.6). Additionally, gene set enrichment analysis (GSEA) was conducted to comprehensively analyze the differences in function enrichment between the high- and low-risk groups. Analysis of tumor mutational burden Somatic mutations were transformed into the mutation annotation format known as MAF. The R ‘maftools’ package (version: 2.20.2), which includes functions for summarizing, analyzing, and visualizing mutation data, was used for the analysis of somatic mutation data. To calculate tumor mutational burden (TMB), a perl script was used to divide the total number of somatic mutations by the length of exons (38 million). SsGSEA and immune correlation analysis The R ‘gsva’ package (version: 1.46.0) was utilized to analyze the correlation between immune cell types and their associated pathways. Additionally, survival analyses were conducted to evaluate the variations in survival rates among patients with high and low risk scores for different immune checkpoints. Correlation analysis for targeted therapy Chemotherapy is frequently employed for the treatment of OC. In our study, we utilized the R package ‘pRRophetic’ (version: 0.5) to estimate the chemotherapeutic response based on the half maximal inhibitory concentration (IC50) of various risk groups. To evaluate the relative disparities in IC50 among these groups, the Wilcoxon signed-rank tests were performed. Statistical analysis The Perl data language (version:5.30.1, [76]http://www.perl.org/), R software (version 4.2.2, [77]https://www.r-project.org/), and GraphPad version 8 were utilized for all data manipulation and statistical computations. The Wilcoxon signed-rank test or Kruskal-Wallis rank sum test can be used to compare two or multiple groups, respectively. A P value of 0.05 is considered statistically significant. Code availability All code for data cleaning and analysis associated with the current submission is available at [78]https://github.com/JingMwang/OV-NET-lncRNA. Results The study flowchart is illustrated in Supplementary Fig. 1. Cases were randomly assigned to either the training or validation cohort in a 1:1 ratio (Table [79]1). The training dataset was internally evaluated using both the validation set and the complete dataset. Afterwards, we proceeded to externally validate the findings by utilizing samples from 20 patients within our own hospital. Table 1. Clinical data of entire set patients. Characteristics Entire set Internal validation set Training set P value N = 371 Number (%) N = 185 Number (%) N = 186 Number (%) Age ≤ 65 255 (68.73%) 125 (67.57%) 130 (69.89%) 0.7106 >65 116 (31.27%) 60 (32.43%) 56 (30.11%) Grade G1 1 (0.27%) 1 (0.54%) 0 (0%) 0.4639 G2 42 (11.32%) 20 (10.81%) 22 (11.83%) G3 317 (85.44%) 160 (86.49%) 157 (84.41%) G4 1 (0.27%) 1 (0.54%) 0 (0%) Unknow 10 (2.70%) 3 (1.62%) 7 (3.77%) Stage I 1 (0.27%) 0 (0%) 1 (0.54%) 0.7177 II 22 (5.93%) 10 (5.41%) 12 (6.45%) III 290 (78.17%) 146 (78.92%) 144 (77.42%) IV 55 (14.82%) 29 (15.68%) 26 (13.98%) Unknow 3 (0.81%) 0 (0%) 3 (1.61%) [80]Open in a new tab Construction of NETs-related LncRNA prognostic model During the analysis of RNA-seq data, we utilized the GENCODE website to identify 128 lncRNAs associated with NETs out of the 16,876 lncRNAs annotated in TCGA. Additionally, we identified NETs-related genes using Pearson correlation analysis, considering a Pearson R value exceeding 0.4 and a p-value lower than 0.05. Based on the correlation between the NETs-related genes and the lncRNAs, we constructed a gene-lncRNA co-expression network using Pearson’s correlation coefficient. The Sankey diagram in Fig. [81]1A illustrates the relationship between them. To further investigate the potential prognostic value of the lncRNAs, we conducted a univariate Cox regression analysis to examine the correlation between the expression of lncRNAs associated with NETs and survival information. Consequently, we discovered 14 lncRNAs linked to the prognosis of ovarian cancer, which encompassed 10 lncRNAs with low risk (hazard ratio (HR) < 1) (Fig. [82]1B). In order to confirm if these 14 lncRNAs associated with NETs were implicated in the advancement of tumors, we compared their levels of expression in tumor tissue with that of normal tissues. Figure [83]1C’s boxplot indicates a significant distinction (p < 0.05) in the expression levels of lncRNAs associated with NETs between the two groups. To simplify the model, the LASSO-Cox regression analysis was conducted (Figs. [84]1D, E). Eventually, GAS5, GBP1P1, LINC00702, LINC01933, LINC02362, and ZNF687-AS1 were identified as prognostic genes through multivariable Cox regression analysis. Among these prognostic factors, GAS5, LINC00702, and LINC 01933 were considered as risk factors (HR > 1), while GBP1P1, LINC023262, and ZNF687-AS1 were considered as protective factors (HR < 1). Figure [85]1F displays the associations between these six potential lncRNAs and genes related to NETs, as shown by the heatmap. Furthermore, the risk score for each ovarian cancer patient was calculated based on the expression level of each prognostic lncRNA and its corresponding coefficient, using formula: graphic file with name d33e706.gif Fig. 1. [86]Fig. 1 [87]Open in a new tab Construction of the prognostic model. (A) Sankey diagram used to visualize the co-expression network of NETs-related lncRNAs. (B) The forest plot represents the results of the univariate Cox regression test. (C) 14 NETs-related lncRNAs were differentially expressed between tumor tissue and normal tissue. Normal tissue is shown in blue, while tumor tissue is shown in red. (D, E) LASSO Cox regression and cross-validation of 14 NETs-related lncRNAs. (F) An analysis of correlations between lncRNAs and genes associated with NETs. The NETs-related LncRNA model performed well in predicting prognosis of ovarian cancer patients Using the median risk score as the threshold, the patients in the testing set were categorized into high- and low-risk groups according to their anticipated risk scores. Risk curves and scatterplots were created to illustrate the relationship between the risk score and survival status of each patient with OC. As depicted in Fig. [88]2A and B, there was a rise in the mortality rate corresponding to the increase in the risk score. In the high-risk group, the heatmap of the six NETs-associated lncRNAs showed significant overexpression of GAS5, LINC00702, and LINC01933, whereas in the low-risk group, GBP1P1, LINC02362, and ZNF687-AS1 were overexpressed as protective factors (p < 0.05, Fig. [89]2C). Fig. 2. [90]Fig. 2 [91]Open in a new tab Evaluation of NETs-related lncRNA prognostic model. Distribution of risk scores, survival status, and heatmap of 6 risk lncRNAs of OC patients in (A–C) the training set. (D, E) The Kaplan-Meier survival analysis shows that high- and low-risk patients have different prognoses with respect to OS (D) and PFS (E). (F) Time-dependent ROC curves for 1-, 3-, and 5-year OS. (G–L) Assessment of prognostic ability of the model in the internal validation set. Consistently, the overall survival and progression-free survival of patients in the high-risk group were significantly worse than those in the low-risk group (p < 0.05, Figs. [92]2D, E). The sensitivity and specificity of the prediction model were evaluated using the ROC curve. Figure [93]2F displayed AUC values of 0.700, 0.646, and 0.716 at the first, third, and fifth years respectively, as indicated by the ROC curve. Consistent results were obtained from both the TCGA internal validation set and the entire set by performing identical analyses with the same formula and cutoff point, as depicted in Fig. [94]2G-K. In the internal validation set, Fig. [95]2L illustrates AUC values of 0.653, 0.590, and 0.592 for 1-year, 3-year, and 5-year survival, respectively. The results suggest that the model is highly accurate in forecasting the prognosis of individuals with ovarian cancer. Validation of LncRNA prognostic models To further validate the reliability of the novel prognostic risk scores, correlation analysis was performed between clinical characteristics and prognosis, along with stratified analysis between risk scores and clinical characteristics. KM survival analysis based on the 6-lncRNAs risk score models demonstrated that compared to the low-risk group, the high-risk group had worse overall survival (OS) in each subgroup (Fig. [96]3A-D). Despite potential limitations, our findings suggest that the prognostic signature based on NETs-related lncRNAs could serve as a dependable predictive tool for assessing ovarian cancer survival. Fig. 3. [97]Fig. 3 [98]Open in a new tab Construction and validation of the prognostic model. (A–D) Subgroup survival analysis adjusted by age, stage, and grade. (E, F) Forest plots summarizing the results of univariate and multivariate Cox analyses of risk scores and clinicopathological features. (G, H) AUC values of nomogram points at 1-year, 3-years, and 5-years were compared with other clinical parameters, including age, grade and stage in training set (G) and internal validiation set (H). (I) The final multivariate model resulted in a concordance index (c-index). (J) The nomogram was plotted based on the risk score and clinical characteristics. (K) Calibration curve of the nomogram. Forest plots indicated that the prognostic signature was found to be independently predictive of overall survival in ovarian cancer patients. Univariate Cox analysis in the training set revealed a strong correlation between age (p < 0.01) and risk score (p = 0.003) with survival in OC. The multivariate Cox regression analysis validated a noteworthy correlation between the hazard ratio (HR 1.060, 95% CI 1.007–1.116) and the survival of ovarian cancer (p = 0.027) (Fig. [99]3E). Moreover, in Fig. [100]3F, the risk score indicated a hazard ratio (HR) of 1.079 (95% confidence interval [CI] 1.027–1.134). The findings confirm the efficacy of the prognostic signature as a standalone predictor. According to the analysis of the area under the curve (AUC) and C-index curve, Figs. [101]3G-I demonstrate that the risk score surpassed other clinical parameters in terms of prediction performance. Furthermore, we developed a nomogram that integrates the risk score and clinicopathological factors (age, sex, and stage) to forecast the overall survival at 1, 3, and 5 years in patients with ovarian cancer (Fig. [102]3J). The calibration curve demonstrated accurate predictive ability for the probability of OS at 1-, 3-, and 5-years (Fig. [103]3K). Construction of a NETs-related CeRNA regulatory network in ovarian cancer To investigate the regulation of genes involved in NETosis by NETs-related lncRNAs acting as miRNA sponges in OC, we developed a comprehensive regulatory network known as the lncRNA-miRNA-mRNA ceRNA network. This network was created by integrating the lncRNA/mRNA network and miRNA-target network. We first obtained 48 target miRNAs for the six model lncRNAs from the miRcode database. Then, using three online tools and considering the differentially expressed genes (DEGs) between the two risk groups, we identified 33 targeting mRNAs. Finally, we visualized the co-expression network with the overlapped lncRNA-miRNA-mRNA relationship (Supplementary Fig. 2). Principal component analysis and functional enrichment analysis Principal component analysis (PCA) was utilized to analyze and visualize the distribution and discrimination of different patterns within the samples. In comparison to the all-gene set, the prognostic NETs-related lncRNA signature demonstrated a clearer distinction between low- and high-risk patients (Figs. [104]4A-D). To further examine the function of differentially expressed genes (DEGs), enrichment analyses based on Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were conducted. The GO enrichment analysis of DEGs focused on three aspects: biological processes (BPs), molecular function (MF), and cellular components (CCs). The top biological processes of DEGs were immunoglobulin production and B cell receptor signaling pathway. Molecular function analysis indicated that DEGs were primarily enriched in antigen binding and immunoglobulin receptor binding. In terms of cellular components, the top two terms were immunoglobulin complex and external side of plasma membrane (Fig. [105]4E). Pathways obtained through KEGG enrichment were visualized using a bubble map, revealing that the DEGs were associated with protein digestion and absorption as well as human papillomavirus infection (Fig. [106]4F). Additionally, the activated (top) and inhibited (bottom) signaling pathways in the high-risk and low-risk groups were analyzed using Gene Set Enrichment Analysis (GSEA) (Fig. [107]4G). Fig. 4. [108]Fig. 4 [109]Open in a new tab PCA and functional enrichment analysis. (A–D) PCA plots based on all genes, NETs-related coding genes, NETs-related lncRNAs and model lncRNAs. (E) GO biological process, cellular component, and molecular function enrichment. (F) KEGG pathway analysis results. (G) Gene set enrichment analysis (GSEA) was utilized to assess the enrichment of specific gene sets. Analysis of tumor mutational burden The waterfall plot displayed the top 15 most commonly mutated genes in patients with ovarian cancer. The mutation frequency was represented in the bar chart on the right (Supplementary Fig. 3A-B). By considering the TMB score, we categorized the patients into high-TMB and low-TMB classes. The box plots and scatter plots demonstrated a significant negative correlation between TMB and risk scores (RS) (Supplementary Fig. 3C-D). Furthermore, the analysis of survival outcomes indicated that the high-TMB group was associated with better survival rates. Subsequently, we divided the tumors into four groups based on TMB and risk scores. The survival curves revealed that patients with low-TMB and high-risk scores had a worse prognosis (Supplementary Fig. 3E-F). SsGSEA and immune correlation analysis To further analyze the association between different groups and immune status, ssGSEA was used to quantify the relation between the enrichment scores for different immune cell subsets and related functions or pathways. The box plot revealed that immune functions, including APC co-inhibition, APC co-stimulation, CCR, Check-point, Cytolytic-activity, HLA, Inflammation-promoting, MHC-class-I, T cell co-inhibition, T cell co-stimulation, and Type-II-IFN-Response, were more active in the low-risk group (p < 0.05, Fig. [110]5A). Furthermore, the aDCs, B cells, CD8 + T cells, DCs, NK cells, pDCs, Tfh cells, Th1 cells, Th2 cells, TILs, Treg, Neutrophils, and Macrophages were generally highly expressed in the low-risk group (p < 0.05, Fig. [111]5B). Figure [112]5C shows the differential expression of different immune checkpoint-related genes between the high and low-risk groups. Next, we employed seven distinct online analysis platforms to forecast the enrichment of immune cells in various risk categories and assessed the correlation between diverse immune cells and the risk score, Fig. [113]5D presents the result. Figure [114]5E revealed the immune cells associated with survival through the Kaplan-Meier survival analysis. Fig. 5. [115]Fig. 5 [116]Open in a new tab Assessment of immune cell infiltration, immune function signature, and immune checkpoint genes in different risk Groups. (A) Differential expression levels of immune function signature between the high- and low-risk groups. (B) Differences in the infiltration of immune cells between the high- and low-risk groups. (C) Differential expression of immune checkpoint genes between the high- and low-risk groups. (D) Seven distinct online analysis platforms to forecast the enrichment of immune cells in various risk categories. (E) Kaplan–Meier survival analysis identifies immune cells associated with survival. *ns: not significant, *p < 0.05, **p < 0.01, ***p < 0.001. Drug sensitivity prediction The use of medication has been acknowledged as an essential element in the management of individuals diagnosed with OC. To evaluate the practicality of the signature in guiding treatment decisions for OC, we examined the relationship between the risk score and the IC50 values of frequently prescribed drugs using the ‘PRRophetic’ R package as previous described^[117]41. The results indicate that bexatotene, bicalutamide, embelin, and thapsigargin have the potential to be considered as potential medications for the high-risk group (Supplementary Fig. 4A, B). While additional clinical validation is necessary, these forecasts could also assist in the assessment of prognosis and the development of personalized chemotherapy plans for ovarian cancer. An external validation was performed to validate the model The expression levels of 6 lncRNAs in IOSE, A2780, and SKOV3 cell lines were analyzed using a RT-qPCR assay. The results are presented in Fig. [118]6A as a bar graph. To validate these findings, we also examined the expression of these 6 lncRNAs in 20 pairs of OC samples and adjacent non-cancerous tissues from our hospital, along with prognostic information. Consistently, the clinical samples displayed a comparable pattern of expression levels for these lncRNAs, as depicted in Fig. [119]6B. Furthermore, to strengthen the reliability of our prognostic model, we included an external dataset of 20 patients with complete prognostic information, as depicted in Fig. [120]6C-E. Fig. 6. [121]Fig. 6 [122]Open in a new tab Validation of the NETs-related prognostic model. (A) An analysis of six candidate lncRNAs expressed in ovarian cancer and normal cells. (B) Twenty pairs of ovarian cancer and matched paracancer tissues qPCR results. (C, D) In univariate and multivariate Cox regressions, RS was found to be an independent factor in OC. (E) Overall survival was significantly associated with the RS according to K-M curves. (F, G) LPS induced NETs model mice developed extensive abdominal metastasis. (n = 5 in each group) (H) NETs model mice had higher levels of citH3 in vivo by ELISA. (I) Model lncRNAs expression altered significantly in NETs model. *p < 0.05, **p < 0.01, and ***p < 0.001. To explore the role of NETs-related lncRNAs in vivo, we constructed LPS-induced NETs model mice and found that compared with the control group, NETs model mice developed extensive abdominal metastasis in Fig. [123]6F and quantitative analyed in Fig. [124]6G. In addition, by cit-H3 ELISA experiment, we confirmed that the LPS group mice had higher levels of NETs in vivo in Fig. [125]6H. RNA expression in solid tumor tissues of mouse model was detected by RT-qPCR, in which the model lncRNAs altered significantly in Fig. [126]6I, similar to those of cell lines in vitro. Role of GAS5 in cellular function High-tumor GAS5 expression was found to be correlated with a poor prognosis in our model. However, previous studies have demonstrated that GAS5 actually hinders the progression of ovarian cancer cell proliferation, migration, and invasion. In order to investigate the precise role of GAS5 in ovarian cancer progression, we generated cell lines with A2780 and SKOV3 overexpression and knockdown in Fig. [127]7A. Our results indicate that GAS5 inhibits ovarian cancer progression in terms of cellular functionality, despite its identification as a poor prognostic factor in our risk model (Fig. [128]7B-D). Fig. 7. [129]Fig. 7 [130]Open in a new tab Function of model lncRNA-GAS5 in OC cell. (A) OC cells were transfected with GAS5 overexpression plasmids and siRNAs, efficiency was verified by RT-qPCR. (B) Colony formation assay of transfected OC cells. (C) Transwell invasion assay of transfected OC cells. (D) Wound healing assays of transfected OC cells. Discussion We constructed a prognostic model using the LASSO Cox regression method, which included 6 NETs-related lncRNAs, for better predicting prognosis and clinical response in OC patients. This model consisted of 3 risk factors (GAS5, LINC01933, LINC00702) and 3 protective factors (GBP1P1, LINC02362, ZNF687-AS1). GAS5 is generally considered a tumor suppressor in solid tumors. Previous studies have shown that the lncRNA GAS5 can decrease OC progression by upregulating WT1 and attenuating miR-23a^[131]42. In colorectal cancer, GAS5 inhibits progression by interacting with and triggering YAP^[132]43. Interestingly, GAS5 has been implicated as a promoting factor in our model, and PCR results demonstrated that GAS5 was decreasedly expressed in OC cells and tissues. Recent studies have suggested that the localization of a gene in the cytoplasm can play a role in anti-infection and activation of immune response, while its escape into the nucleus can inhibit DNA repair and facilitate tumorigenesis^[133]44. This may explain the phenomenon observed in our study. A literature search revealed that LINC00702 has been implicated in the advancement of ovarian cancer^[134]45 and malignant meningioma^[135]46. Previous studies have indicated that LINC02362 and GBP1P1 act as tumor suppressors in the pathogenesis of liver cancer^[136]47,[137]48. However, the roles and molecular mechanisms of LINC01933 and ZNF687-AS1 have not been extensively investigated. The outcomes of our model are nearly consistent with these prior studies. As demonstrated in previous studies, the combination of biomarkers has been found to exhibit higher accuracy compared to using a single biomarker alone^[138]49. Furthermore, the predicted survival probabilities based on the nomogram align well with the actual observations. Both univariate and multivariate Cox proportional hazard regression analyses confirmed the score as an independent prognostic factor. Additionally, PCA analysis revealed distinct differentiation between patients in the high- and low-risk groups. We conducted GO and KEGG analyses to study gene functions. GO analysis showed that differentially expressed genes were mainly linked to immune-related terms, while KEGG highlighted ECM receptor interaction as highly enriched. It is worth noting that recent studies have demonstrated the interdependence between the extracellular matrix and the immune system^[139]50. The extracellular matrix not only supports tissues and cells but also regulates immunity and aids in repair, with its dysregulation often causing chronic diseases. Therefore, therapies for immune-mediated diseases should consider both immune cells and their environment. GSEA indicated that ECM and cancer pathways were active in high-risk patients, suggesting a poor prognosis. A ceRNA network was constructed to evaluate the potential functions and interactions of model lncRNAs, highlighting GAS5’s role in tumor development through its binding to specific miRNAs and mRNAs. Due to limited data on other lncRNAs, they were not deeply explored. SsGSEA analysis revealed that the high-risk group had lower protective immune cell infiltration and significant differences in immune checkpoint gene expression compared to the low-risk group, indicating a poorer prognosis. Our study also identified bexatotene, bicalutamide, embelin, GDC0941, and thapsigargin as potential drugs for ovarian cancer, offering new insights into chemotherapy options. However, this study has several limitations. Firstly, we had to use the GTEx database as a supplement since the TCGA database lacks normal ovarian tissue sequencing results. This introduced a batch effect. Secondly, due to limited research on certain lncRNAs involved in the model construction, we were unable to find suitable external datasets for validation, which would have helped determine the model’s applicability and accuracy. Moreover, the precise mechanism that governs the correlation between lncRNAs associated with NETs and the prognosis of OC is still not understood. Specifically, the role of GAS5, which is considered a risk factor in the model but contradicts previous findings, requires further investigation. Therefore, future prospective multicenter studies with larger sample sizes are necessary to validate our study results and optimize prediction models for clinical practice. Furthermore, additional research is needed to explore the function of lncRNA GAS5 in ovarian cancer. Conclusion We established a NETs-related lncRNA risk model which has shown potential as reliable predictors of OC patients’ prognosis. Additionally, the risk signature derived from these lncRNAs can also predict the response to targeted therapy and immunotherapy. This model may help promote precision medicine and guide targeted treatments and improved patient outcomes. This valuable information can assist clinical doctors in making personalized decisions regarding diagnosis and treatment. Electronic supplementary material Below is the link to the electronic supplementary material. [140]Supplementary Material 1^ (1.7MB, doc) Acknowledgements