Integrated clinical characteristics and omics analysis identifies a ferroptosis and iron-metabolism-related lncRNA signature for predicting prognosis and therapeutic responses in ovarian cancer

The OC-clinical data, OC-RNA sequencing profiles, and normal ovarian epithelial tissue RNA sequencing profiles were obtained from The Cancer Genome Atlas (TCGA) [17] and GTEx database [18] using UCSC Xena. We excluded OC patients without RNA sequencing, survival time, or repeat sequencing, and finally, only 374 patients were retained for subsequent analysis. At a ratio of 3:7, the total OC patients were divided into two sets (training set and testing set) using the caret package in R software. Meanwhile, lncRNAs and protein-coding genes were identified based on annotation documents of the GENCODE database [19]. In addition, 296 FIRGs (Table. S1) were extracted based on previous studies [20], including ferroptosis regulators, ferroptosis markers, ferroptosis pathway, Iron uptake and transport, and Iron ion homeostasis, etc. It is worth mentioning that somatic mutation data were also obtained from the TCGA database, and homologous recombination repair (HRR) related genes were obtained from the previous reference [21].

In exploring differences in immune cell infiltration, we simultaneously used 6 algorithms (TIMER, CIBERSORT, QUANTISEQ, MCP-counter, XCELL, and EPIC) to estimate the abundances of immune cells in different risk groups distinguished by FIRLs signature. Moreover, we used the ssGSEA algorithm to quantify immune functions and pathways. More importantly, we also explored immune checkpoint-related gene expression levels in different risk groups. Finally, GO and KEGG functional enrichment analysis of FIRLs signature was conducted.

The IC50 was calculated using pRRophetic package in R software, and the chemotherapeutic medications were obtained from the Genomics of Drug Sensitivity in Cancer (GDSC) database [22].

To highlight the substantial prognostic value of FIRLs signatures, we compared the efficacy of other signatures from different references. Zhang et al. identified a glycolysis-related gene signature for OC patients, including ISG20, CITED2, PYGB,

IRS2, ANGPTL4, TGFBI, LHX9, PC, and DDIT4 [23]. Zhou et al. identified a DNA methylation-driven genes signature, including PON3, MFAP4, AKAP12, and BHMT2 [24]. Moreover, Zheng et al. developed a risk stratification system based on glycolysis-related lncRNAs, including AC133644.2, CTD-2396E7.11, CTD-3065, J16.9, LINC00240, and TMEM254-AS1 [25]. In 374 patients with OC, we performed Lasso-Cox analysis on the above genes to calculate the corresponding risk score. Finally, C-index was used to compare the predictive ability of the different models.

We randomly divided 374 OC patients into a testing set (110 patients) and a training set (274 patients) to construct and validate the signature. Subsequently, 14 lncRNAs (p < 0.01) were significantly correlated with the survival by univariate Cox regression analysis in the training set, as shown in Fig. 2a. We aimed to avoid the occurrence of collinearity of transcriptome data, and LASSO regression analysis was used to screen out further 8-lncRNAs, which constituted a prognostic risk signature of FIRLs (Table.1, Fig. 2b). Finally, combining the expression of 8-FIRLs and regression coefficients in multivariate Cox regression analysis, the risk score of OC patients is calculated as follows: Risk score = (-0.354 *AC138904.1) + (-0.325 * AP005205.2) + (-0.428 * AC007114.1) + (-1.100 * LINC00665) + (-0.401 * UBXN10-AS1) + (0.395 * AC083880.1) + (0.448 * LINC01558) + (-0.636 * AL023583.1). Taken together, our data showed that an 8-FIRLs signature was derived in the training set.

We calculated the risk score of OC patients in the testing set and training set according to the above formula. According to the median value in the training set, the population in each group was divided into a high-risk group and a low-risk group. Firstly, the PCA analysis confirmed a risk signature's classification ability in the testing set and training set, as shown in Fig. 3a, e. Subsequently, Kaplan–Meier survival analysis showed that OS of the high-risk group was significantly shorter than that of the low-risk group (training set: p < 0.05, as shown in Fig. 3b; testing set: p < 0.05, as shown in Fig. 3f), which indicates that FIRLs signature has an excellent predictive value. We evaluated the predictive sensitivity and specificity of FIRLs signature by ROC curve. The AUC of the training set and the testing set at 1, 3, and 5 years reached 0.732, 0.684, 0.711 and 0.634, 0.577, 0.525, respectively, as shown in Fig. 3c, g. In addition, the heatmap of 8 FIRLs expressions in the high- and low-risk groups is shown in Fig. 3d, h. Finally, we performed univariate and multivariate Cox regression analysis of FIRLs signature and clinical characteristics in total patients. The results showed that FIRLs signature is an independent prognostic factor for OC patients (p < 0.001), as shown in Fig. 4a, b. What is exciting is that the DCA curve and ROC curve showed that FIRLs signature to predict the median survival time is significantly better than traditional clinical characteristics in testing set and training set, as shown in Fig. 4c-f. Taken together, our data showed that FIRLs signature has a superior clinical benefit for OC patients.

Considering that the formula of the FIRLs signature is complicated in routine clinical work, the nomogram [26] can intuitively apply to clinical work, so we visualized the risk signature based on the above risk formula. As shown in Fig. 5a-b, we plotted two nomograms based on the same risk formula. Moreover, the calibration curve of the nomogram showed that the prediction curves are close to the standard curve in the testing set and training set, which indicates that the predicted survival rate is closely related to the actual rates at 1, 3, and 5 years, as shown in Fig. S1. Taken together, our data showed that our nomograms could intuitively apply to clinical work.

We used the chi-square test to study whether the high- and low-risk groups based on FIRLs signature is involved in the development of OC. Our heatmap showed that there were significant differences between the high-risk group and the low-risk group in FIGO staging (P < 0.01) and residual tumour size (P < 0.05), as shown in Fig. 6a. To further explore the predictive efficiency of FIRLs signature in different clinical characteristics (Fig. 6b-i), the following clinical variables were used for analysis: age (≤ 60 and > 60), FIGO stage (I—II, and III—IV), pathological grade (G1- 2 and G 3–4), residual tumour size (R0/R1 and > R1). In the remaining subgroups except for the FIGO I-II (p = 0.058) and G1-G2 subgroups (p = 0.193), the results revealed that FIRLs signature has prognostic significance between different risk groups. Particularly worth mentioning is that the OS of patients in the high-risk group was significantly lower than that of the low-risk patients in most subgroups (P < 0.05). Taken together, our findings revealed that the FIRLs signature plays a pivotal role in predicting the prognosis in patients with OC.

We screened out the 40 FIRGs co-expressed with 8-LncRNAs, as shown in Fig. 7a. KEGG enrichment analysis showed that related mRNAs were enriched in ferroptosis, mitophagy, and autophagy pathways, etc., as shown in Fig. 7b. GO enrichment analysis showed that 40 mRNAs were mainly related to response to nutrient levels and epithelial cell apoptotic process in BP section, mitochondrial outer membrane and protein kinase complex in CC section, and protein serine/threonine kinase activity and long-chain fatty acid-CoA ligase activity in MF section (Fig. 7c). In addition, we explored the expression of 8 lncRNAs in clinical samples. As expected, most of the lncRNAs (7/8) in OC samples were up-regulated except for AP005205.2, as shown in Fig. S2. Taken together, our data showed that GO and KEGG analysis verified the relationship between FIRLs signature and iron metabolism from another perspective.

To comprehensively explore the relationship between different risk groups and immune cell infiltration, we plotted the heatmap of immune infiltration based on 6 algorithms. Specific immune cells differed significantly among risk subgroups, such as Macrophages, T cells, NK cell resting, etc. (Fig. 8a). Interestingly, analysis of immunologic function confirmed significant differences between low- and high-risk groups for other immunological functions except for cytolytic activity, HLA, inflammation-promoting, MHC class I, and Type_I_IFN response (P > 0.05), as shown in Fig. 8b. Meanwhile, the boxplot showed immune checkpoints mRNA were up-regulated in the high-risk group compared to the low-risk group, as shown in Fig. 8c. Taken together, our data showed that FIRLs signature was correlated with immune cell infiltration and immunotherapy to a certain extent.

We investigated the drug sensitivity of chemotherapeutic agents often used in clinics in different risk subgroups. The IC50 values of 7 chemotherapeutic medicines were quantified in OC patients, and 5 were statistically different between risk groups. In detail, the IC50 levels of Docetaxel (Fig. 9a), Doxorubicin (Fig. 9b), Etoposide (Fig. 9c), Paclitaxel (Fig. 9d) and Gemcitabine (Fig. 9e), Bleomycin (Fig. 9f), and Cisplatin (Fig. 9g) were significantly higher in high-risk group (P < 0.05). It indicated that the OC patients in the low-risk group distinguished by FIRLs signature were more sensitive to the above chemotherapeutics. In addition, considering the critical correlation of HRR-related genes in OC patients for the maintenance therapy, we further analyzed the association of different risk groups with HRR-related genes. We found that BRCA1, BRCA2 and CDK12 were the top three genes (Fig. 10a). Mutations in BRCA1 and CDK12 were not statistically significant in TCGA-cohort; however, there was a statistically significant difference in BRCA1 gene mutation (Fig. 10b). Specifically, the low-risk group had a higher frequency of BRCA1 mutations. Meanwhile, we combined risk groups with mutations in BRCA1, BRCA2 and CDK12 for survival analysis. The results showed a statistically significant difference between the four groups, as shown in Fig. 10c-e. Taken together, our data showed that 7 chemotherapeutic agents and PARP inhibitors might have a better effect on patients in the low-risk group.

To highlight the prognostic value of FIRL signatures, we compared the efficacy of other signatures from different references. Zhang et al. identified a glycolysis-related gene signature for OC patients, and Zhou et al. identified a DNA methylation-driven genes signature. Moreover, Zheng et al. developed a risk stratification system based on glycolysis-related lncRNAs. In 374 patients with OC, we performed Lasso-Cox analysis on the above signatures to calculate the corresponding risk score and showed ROC analysis of 1, 3, and 5 years (Fig. 11a-d). The C-index value showed that the 8-FIRLs signature had the most robust predictive performance (Fig. 11e). However, it should not be ignored that other risk signatures can also stratify the risk of OC patients.