OC clinical data (n = 587), OC RNA sequencing profiles (n = 379), and normal ovarian epithelial tissue RNA sequencing profiles (n = 88) were obtained from the Cancer Genome Atlas (TCGA) (15) and GTEx databases (16). We excluded patients with OC without RNA sequencing and survival time data, and finally, only 374 patients were retained for subsequent analysis. At a ratio of 3:7, the OC patients were divided into two sets (the training cohort and testing cohort) using the “caret” package in R software (17). Meanwhile, lncRNAs and protein-coding genes were identified based on annotation documents from the GENCODE database (18). Additionally, 33 PRGs were extracted based on previous studies (9).

Mutation and copy number variation (CNV) data, including 436 OC patient samples, were downloaded from the TCGA database. In addition, the mutation frequencies of the 33 PRGs in patients with OC were generated using the “maftools” package (19). The locations of CNV alterations in the 33 PRGs on the 23 chromosomes were plotted using the RCircos package.

Prognostic lncRNAs (p < 0.05) were screened using a univariate Cox regression analysis. These prognostic lncRNAs were further incorporated into multivariate Cox and LASSO regression analyses to identify the lncRNAs involved in signature construction (20). We used the appropriate λ to build the model and control for the complexity of the LASSO regression. The risk score was calculated as follows: risk score for OS =

For clinical utility, the comparative value (2^−Δct) was calculated from RT-qPCR results and used for score calculation in our hospital cohort, and the score was further standardized and simplified to generate a risk score (21). The risk score was calculated as follows: Risk score (RT-qPCR) = (Score − Min)/Max.

The risk score for each OC patient was summed using the above formula. Risk signatures for predicting survival were assessed using AUC and decision curve analysis (DCA) curves (22). We calculated the risk score of each patient by determining their cutoff value using the receiver operating characteristic (ROC) curve, which is used to select the “high-risk” and “low-risk” groups. The Kaplan–Meier survival analysis and log-rank test suggested that the high-risk groups had shorter survival times than the low-risk groups.

A network of PRLs with co-expressed genes was constructed using the Cytoscape software (23). Gene enrichment analysis was performed on differentially expressed ARGs using related packages in R software. Meanwhile, we used the “limma” package to explore the differential expression of lncRNAs participating in PRL signatures between normal samples (GTEx) and OC samples (TCGA) (24).

To explore the differences in immune cell infiltration, we simultaneously used six algorithms [TIMER (25), CIBERSORT (26), quanTIseq (27), MCP-counter (28), xCell (29), and EPIC (30)] to estimate the abundance of immune cells in the different risk groups distinguished by the PRL signature. Moreover, we used the ssGSEA algorithm to quantify immune functions and pathways between the low- and high-risk groups (31). More importantly, we explored the expression levels of immune checkpoint-related genes in the different risk groups.

In this study, we used cell culture, transfection, CCK-8, and qRT-PCR for in-vitro assays. The Shanghai Cell Institute Country Cell Bank provided the cell lines SKOV-3, A2780, HO-8910PM, and IOSE80. GenePharma generated and annealed small interfering RNA (si-RNA-1/2/3) oligos for DICER-AS1 and a general negative control. Following the manufacturer’s instructions, each siRNA duplex was transfected into cells using Lipofectamine^® 2000 (Invitrogen, Carlsbad, CA, USA). A2780 and SKOV-3 cells were seeded in 96-well plates, with or without DICER-AS1 knockdown. After 0, 1, 2, 3, and 4 days of cell culture, cell viability was determined using the CCK-8 assay (Dojindo, Tokyo, Japan). In addition, the DICER-AS1 primers were as follows: DICER-AS1-F, 5′-CGAAGAAATGGAATAACTTCCAAC-3′; DICER-AS1-R: 5′-TTGGTCCAAACACAGAAGATC-3′. The details of these methods are provided in (32, 33). The experimental results related to this study were provided by Qingquan Zhang, Nantong University.

OC tissues were obtained from patients who had undergone surgery at the Affiliated Hospital of Nantong University. In the cohort, 10 pairs of tissues were obtained from patients with OC, between 2017 and 2020. The study was authorized by the Ethical Committee of Affiliated Hospital of Nantong University (2021-K150-01).

All statistical analyses were performed using R software (v.4.0.1). Detailed statistical methods for transcriptome data processing were described in the above section.

First, we explored the expression of 33 PRGs in OC and normal ovarian epithelial tissues based on mRNA expression in the TCGA and GTEx datasets. Interestingly, the heatmap ( Figure 1A ) and boxplot ( Figure 1B ) revealed that all PRGs, except IL16 and CASP1, were aberrantly expressed in OC patients (p < 0.05). More specifically, these results may indicate the activation of pyroptosis-related pathways in OC. In addition, we explored the correlations between the 33 PRGs. The results revealed that NLRP3 showed the strongest positive correlation with NLRC4 (r = 0.75), and PYCARD showed the strongest negative correlation with PLCG1 in patients with OC (r = −0.49), as shown in Figure 1C . It is also worth noting that we explored the incidence of CNVs and mutations in the 33 PRGs in OC patients from the TCGA database. Figure 2A shows the locations of the CNV alterations on the chromosomes for the 33 PRGs. Mutations were present in 44/436 OC samples, and missense mutations were the most common variant classification, as shown in Figure 2B . The results also showed that NLRP3 had the highest mutation frequency among the 33 PRGs. We also investigated the frequency of CNV alterations and found that 33 PRGs showed widespread CNV alterations. Copy number amplification was present in all 33 PRGs except GPX4 and ELANE, with up to 60% amplification in GSDMC. Not surprisingly, copy number deletions were also present in most PRGs, with GPX4 approaching 60% ( Figure 2D ).

Given the significant role of lncRNAs in the regulation of OC mechanisms, we performed a Pearson correlation analysis (|cor| > 0.3, p < 0.001) on 13,832 lncRNAs and 33 PRGs from the TCGA gene expression raw data. Ultimately, we screened 250 PRLs for subsequent bioinformatic analysis. The networks are shown in Figure 2C .

A total of 374 OC patients with matched transcriptome data from the TCGA database were downloaded to explore the association between the expression of the above 250 PRLs and OS. To construct and validate the signature, we randomly divided 374 patients with OC into a validation cohort (110 patients) and a training cohort (264 patients). Subsequently, 19 PRLs (p < 0.05) were significantly correlated with survival in the univariate Cox regression analysis in the training cohort, as shown in Figure 3A . We aimed to avoid the occurrence of collinearity in the high-dimensional transcriptome data; therefore, a LASSO regression analysis ( Figures 3B, C ) was used to further screen eight PRLs, which constituted the prognostic risk signatures of the PRLs. We determined the coefficient of each PRL using the multivariate Cox regression analysis, as shown in Figure 3D . Finally, by combining the expression of eight PRLs and the regression coefficients, the risk scores of patients with OC were calculated as follows: risk signature = (−0.106 * DICER1-AS1) + (−0.325 * MIR600HG) + (−0.428 * AC083880.1) + (−1.100 * AC109322.1) + (−0.401 * AC007991.4) + (0.395 * IL6R-AS1) + (0.448 * AL365361.1) + (−0.636 * AC022098.2), as shown in Figure 3E .

To explore the prognostic value of the PRL signatures, we divided the training and validation cohorts into high- and low-risk groups based on the median values of the risk scores of the training cohort. In each group, the survival times of high-risk patients were significantly shorter than those of low-risk patients (p < 0.05), as shown in Figures 4A, E . ROC curves demonstrated the predictive effectiveness of the risk signature. The results showed that in the training cohort, the AUCs for predicting 1- ( Figure 4B ), 3- ( Figure 4C ), and 5-year survival ( Figure 4D ) were 0.702, 0.647, and 0.533, respectively. The AUCs for predicting 1- ( Figure 4F ), 3- ( Figure 4G ), and 5-year survival ( Figure 4H ) were 0.691, 0.611, and 0.505 in the validation cohort, respectively. In addition, heatmaps showed the expression of eight PRLs in the different groups ( Figures 5A, B ). In the survival analysis of each lncRNA participating in the risk signature, we found five protective factors, namely, AC083880.1 ( Figure 5D ), AC109322.1 ( Figure 5F ), AC007991.4 ( Figure 5G ), IL6R-AS1 ( Figure 5I ), and AC022098.2 ( Figure 5J ), and only one risky factor, MIR600HG ( Figure 5E ), in the TCGA cohort. Survival analysis of the protective factors showed that PRL overexpression may lead to a longer survival time. In contrast, individuals with upregulated expression of risk factors may have shorter survival times. Although there was no significant difference in survival time among the subgroups of AC022098.2 ( Figure 5C ) and the DICER1-AS1 ( Figure 5H ) group, this is also of concern.

First, we showed the overall distribution of different clinicopathological factors in the high- and low-risk groups via the expression of eight PRLs in the form of a heat map, as shown in Figure 6A . Next, a correlation analysis was performed between the different risk groups and the corresponding clinical characteristics. As presented in Figure 6 , increased risk scores were significantly related to multiple factors, including age ( Figure 6B ) and FIGO stage ( Figure 6C ) (p < 0.05). It should also be noted that the different risk groups were not statistically correlated with the following clinical characteristics: grade ( Figure 6D ) or residual size ( Figure 6E ). To investigate the independent prognostic value of the risk signature in OC patients, the univariate Cox analysis revealed that the risk score was a high-risk factor (p < 0.001; Figure 6F ). Furthermore, a further multivariate Cox analysis showed that only the risk score was independently associated with OS (p < 0.001), implying that the risk signature may be an independent prognostic predictor for OC patients ( Figure 6G ). To further verify the prognostic value of our signature, we performed risk scores for each OC patient in the GSE26193 dataset using the same formula. The results showed that our signatures also had an excellent survival prediction ability in the external dataset, and the OS of the high-risk group was shorter than that of the low-risk group ( Figures S1A, B ). In addition, univariate and multivariate Cox analyses also revealed that risk score was an independent prognostic factor ( Figures S1C, D ).

To further explore the predictive efficiency of the PRL signature on different clinical characteristics from the TCGA cohort, the following clinical variables were used for analysis: age (≤65 and >65 years), FIGO stage (I–II and III–IV), pathological grade (G1–2 and G3–4), and residual tumor size (R0 and non-R0). In the survival analysis of the clinical subgroups, the survival time of the high-risk group was significantly shorter than that of the low-risk group ( Figures 7A, B ), advanced stage group ( Figure 7D ), G3–G4 group ( Figure 7F ), and residual tumor size subgroup ( Figures 7G, H ; p < 0.05). In addition, there was no statistical difference in survival in the I–II ( Figure 7C ) and G1–G2 ( Figure 7E ) groups, likely due to the small sample size of early stage patients.

Considering that the formula for PRL signatures is complicated, nomograms can intuitively be applied to clinical work; therefore, we visualized the risk signature based on the above risk formula. We combined indicators commonly used in clinical work to construct two visual prognostic models using different packages in R software based on the same risk formula, as shown in Figures 8A, B . The calibration curve of the nomogram showed that the prediction curves were close to the standard curve in the two cohorts, which indicates that the predicted survival rate is closely related to the actual rates at 1, 3, and 5 years, as shown in Figures 8C – H .

Interestingly, the DCA curve and the ROC curve showed that the PRL signature was significantly better at predicting survival time than traditional clinical characteristics in the training cohort ( Figures 9A – D ) and validation cohort ( Figures 9E – H ; p < 0.05). In addition, to further compare the accuracy of the PRL risk signature, we compared the prognosis signatures of patients with OC in other studies. The results were also exciting: in the C-index for predicting the TCGA cohort, the risk signature of our study showed better predictive value than the glycolysis-related gene signature established by Zhang et al. (34), the glycolysis-related lncRNA signature established by Zheng et al. (35), and the DNA methylation-driven gene signature established by Zhou et al. (36), as shown in Figure S2 .

To explore the potential biological functions and pathways of the eight PRLs in the signature, we screened out genes that were co-expressed with the eight PRLs. Finally, we selected 80 mRNAs based on the cutoff value stated in the Materials and Methods section, as shown in Figure 10C . GO enrichment analysis ( Figure 10A ) showed that more than 80 mRNAs were mainly related to natural killer cell-mediated cytotoxicity in the BP section, the external side of the plasma membrane in the CC section, and protein phosphatase in the MF section. KEGG enrichment analysis ( Figure 10B ) showed that related mRNAs were enriched in the NOD-like receptor signaling pathway. In addition, we explored the above mRNAs in the PPI network ( Figure 10C ). Finally, as shown in Figure 10D , we performed topological analysis of the genes in the PPI network using Cytoscape and identified the top 5 genes: CD27, CD8A, CXCR3, IRF1, and IGLL5.

Considering the important correlation between the BRCA1/2 gene in OC patients and maintenance therapy after cytoreductive surgery, we further analyzed the different risk groups for any associations between their risk signatures and BRCA1/2 gene mutations. According to the mutation data in the TCGA database, we found that there was a statistically significant difference in BRCA1 gene mutation numbers between the high- and low-risk groups in the TCGA cohort (p = 0.023) ( Figure 11A ). However, mutations in this gene were not statistically significant in the TCGA training and testing sets ( Figures 11B, C ). However, in the analysis of BRCA2 mutations, we did not find significant differences among the groups ( Figures 11E – G ). Meanwhile, we combined risk groups with mutations in the BRCA1/2 genes for survival analysis. The results showed no statistically significant differences between the four groups, as shown in Figures 11D, H .

To comprehensively explore the relationship between different risk groups and immune cell infiltration, we plotted a heatmap of immune cell infiltration based on six algorithms (TIMER, CIBERSORT, quanTIseq, MCP-counter, xCell, and EPIC), as shown in Figure 12A . Interestingly, analysis of immunologic function confirmed significant differences between the low- and high-risk groups for immunological function other than MHC class I (p > 0.05), as shown in Figure 12B . Finally, it is worth noting that there were significant differences in the expression of some commonly used clinical immune checkpoints between the different risk groups, as shown in Figure 12C .

lncRNA DICER-AS1 was filtered as a candidate molecule in which to perform cell function assays. To further illustrate the prognostic value of DICER-AS1, we performed a meta-analysis of DICER-AS1 by combining several datasets (TCGA, GSE18520, GSE19829, GSE26931, and GSE63885) using a random effects model. Among the results of the meta-analysis of OS, DICER-AS1 was shown to be a high-risk factor for survival in OC patients [HR = 1.27 (1.07–2.07)], as shown in Figure 13A . In three OC cell lines (A2780, SKOV-3, and HO-8910PM), real-time qPCR analysis revealed that DICER-AS1 mRNA expression was substantially upregulated compared with normal ovarian epithelial cells (IOSE80), as shown in Figure 13B . We also assessed the efficiency of siRNAs targeting DICER-AS1 and discovered that siRNA-1 was most efficiently transfected into SKOV-3 cells ( Figure 13C ); in addition, siRNA-2 was most efficiently transfected into A2780 cells ( Figure 13E ). We initially performed CCK-8 experiments, which revealed that the downregulation of DICER-AS1 expression significantly reduced the proliferation of OC cells ( Figures 13D, F ).

As shown in Figures 14A – H , we found that four lncRNAs (AC007991.4, AC022098.2, AC1093322.1, and AC083880.1) were highly expressed in tumor tissues as compared with normal tissues by differential analysis based on the TCGA and GTEx databases. Meanwhile, AL365536.1, DICER1-AS1, IL6R-AS1, and MIR600HG were overexpressed in normal tissues. Notably, DICER1-AS1 was overexpressed in OC cell lines. Subsequently, RT-qPCR was used to detect the expression of the eight lncRNAs in 10 pairs of tissues. The results were consistent with the prediction results in public databases ( Figure 14J ). Moreover, four patients were considered high risk and six patients were low risk, and their OS rates were also significantly different ( Figure 14I ).