Identification of ferroptosis-related lncRNAs
Through ''GENCODE'' database, we pinpointed 15071 and 11861 lncRNAs in the TCGA and ICGC datasets separately. Then, we obtained the expression matrix of 38 ferroptosis-related genes from each dataset. Pearson correlation analysis was first carried out to explore ferroptosis-related lncRNAs, thenceforth the univariate Cox regression was constructed to seek ferroptosis-related prognostic lncRNA (p < 0.05) in each dataset. Finally, we identified 23 ferroptosis-related lncRNAs which dramatically correlated with the survival among both datasets. The association between 23 ferroptosis-related lncRNAs and 38 ferroptosis-related genes was displayed in Fig. 2, the univariate Cox analysis of 23 ferroptosis-related lncRNAs was shown in Table 1.
Table 1
The 23 ferroptosis-related prognostic lncRNAs
lncRNA | TCGA | ICGC |
HR | HR.95L | HR.95H | p-value | HR | HR.95L | HR.95H | p-value |
OSMR-AS1 | 1.4915 | 1.0714 | 2.0763 | 0.0179 | 1.2615 | 1.0116 | 1.5732 | 0.0392 |
SLC16A1-AS1 | 1.4379 | 1.0252 | 2.0166 | 0.0354 | 1.2242 | 1.0028 | 1.4945 | 0.0469 |
SETBP1-DT | 0.8096 | 0.6639 | 0.9872 | 0.0369 | 0.7414 | 0.6194 | 0.8874 | 0.0011 |
AC097639.1 | 0.8032 | 0.6709 | 0.9617 | 0.0170 | 0.7965 | 0.6595 | 0.9621 | 0.0182 |
TRAM2-AS1 | 0.7956 | 0.6417 | 0.9863 | 0.0370 | 0.7370 | 0.5433 | 0.9998 | 0.0498 |
FOXN3-AS1 | 0.7658 | 0.6139 | 0.9552 | 0.0179 | 0.7542 | 0.5733 | 0.9921 | 0.0437 |
ZNF793-AS1 | 0.7639 | 0.6051 | 0.9645 | 0.0236 | 0.7920 | 0.6888 | 0.9107 | 0.0011 |
LINC02777 | 0.7626 | 0.5860 | 0.9923 | 0.0437 | 0.8125 | 0.6668 | 0.9900 | 0.0395 |
LINC00526 | 0.7624 | 0.5873 | 0.9898 | 0.0417 | 0.7570 | 0.6133 | 0.9345 | 0.0096 |
SLC25A5-AS1 | 0.7556 | 0.5987 | 0.9535 | 0.0182 | 0.6088 | 0.4681 | 0.7917 | 0.0002 |
ZNF582-AS1 | 0.7524 | 0.5734 | 0.9871 | 0.0400 | 0.8326 | 0.7038 | 0.9850 | 0.0326 |
AC073896.2 | 0.7519 | 0.5795 | 0.9755 | 0.0318 | 0.5455 | 0.3846 | 0.7736 | 0.0007 |
AC009506.1 | 0.7479 | 0.5960 | 0.9386 | 0.0122 | 0.6740 | 0.4885 | 0.9298 | 0.0163 |
SGMS1-AS1 | 0.7370 | 0.5463 | 0.9942 | 0.0457 | 0.7119 | 0.5489 | 0.9233 | 0.0104 |
AC138356.1 | 0.7351 | 0.5439 | 0.9936 | 0.0453 | 0.7396 | 0.5989 | 0.9135 | 0.0051 |
AC008966.1 | 0.7348 | 0.5746 | 0.9397 | 0.0140 | 0.7491 | 0.5969 | 0.9402 | 0.0127 |
LINC00242 | 0.7280 | 0.5712 | 0.9279 | 0.0103 | 0.6744 | 0.5436 | 0.8366 | 0.0003 |
ST7-AS1 | 0.7146 | 0.5571 | 0.9167 | 0.0082 | 0.7633 | 0.6110 | 0.9534 | 0.0173 |
WDFY3-AS2 | 0.7067 | 0.5573 | 0.8961 | 0.0042 | 0.7835 | 0.6359 | 0.9655 | 0.0221 |
ZNF710-AS1 | 0.6972 | 0.5354 | 0.9079 | 0.0074 | 0.7130 | 0.5704 | 0.8912 | 0.0030 |
PXN-AS1 | 0.6910 | 0.5026 | 0.9501 | 0.0229 | 0.6138 | 0.4550 | 0.8279 | 0.0014 |
AC036176.1 | 0.6817 | 0.5213 | 0.8915 | 0.0051 | 0.6414 | 0.5054 | 0.8140 | 0.0003 |
AC016876.1 | 0.6378 | 0.4634 | 0.8778 | 0.0058 | 0.6581 | 0.4901 | 0.8838 | 0.0054 |
lncRNAs marked with bold font were risky lncRNAs and others were protective lncRNAs. |
Construction Of The Fe-lpm In Pc
The LASSO-Cox analysis was performed using the expression matrix of the 23 ferroptosis-related prognostic lncRNA mentioned above. The Fe-LPM encompassed 8 ferroptosis-related lncRNAs in the TCGA dataset was identified (Fig. 3A, B), the coefficient of the Fe-LPM was shown in Fig. 3C. A risk score was computed for all patients in TCGA dataset, and patients were then split for survival analysis into high-risk and low-risk subgroups on the threshold of median risk scores. The principal component analysis (PCA) indicated that we do obtain a high degree of discrimination between high-risk and low-risk subgroups in the TCGA cohort (Additional file 1: Figure S1A, B). KM survival curves considered that patients from the high-risk subgroup had considerably worse clinical outcomes than those from the low-risk subgroup (Fig. 3D). The distribution of the risk score was shown in Fig. 3E. The forecasting capability of the Fe-LPM for overall survival was appraised by ROC curves, and the area under the curve (AUC) reached 0.70 at 1 year, 0.71 at 2 years, 0.75 at 3 years (Fig. 3F).
Validation Of The Fe-lpm In Pc
To verify the stability of the prognostic markers constructed in the TCGA dataset, patients in the ICGC validation set were also assigned into high-risk and low-risk subgroups relative to the median risk scores calculated with the same formula as that from the training set. PCA in ICGC datasets was shown in Additional file 1: Figure S1C, D. Similar to the conclusion in the TCGA cohort, patients with high-risk scores had shorter OS than the low‐risk ones in the ICGC dataset (Fig. 3G). The distribution of the risk score was shown in Fig. 3H. The AUC of the Fe-LPM in the ICGC cohort was 0.79 at 1 year, 0.71 at 2 years, 0.76 at 3 years (Fig. 3I).
Prognostic Analysis Of The Eight Fe-lpm
We implemented the univariate Cox analysis to estimate the prognostic role of the eight Fe-LPM. The result indicates that SLC16A1-AS1 is a risk factor with HR > 1, whereas SETBP1-DT, ZNF93-AS1, SLC25A5-AS1, AC073896.2, LINC00242, PXN-AS1 and AC036176.1 are protective factors (Fig. 4A). The heatmap exhibits that SLC16A1-AS1 expression increased with increasing risk score, while SETBP1-DT, ZNF93-AS1, SLC25A5-AS1, AC073896.2, LINC00242, PXN-AS1 and AC036176.1 expression decreased with increasing risk score (Fig. 4B). Also, the Kaplan-Meier curve certified that lower expression of SLC16A1-AS1 and higher expression of SETBP1-DT, ZNF93-AS1, SLC25A5-AS1, AC073896.2, LINC00242, PXN-AS1 and AC036176.1 corresponded with higher survival time in the TCGA cohort (Fig. 4C-J).
Clinical Relevance Of The Fe-lpm
In order to clarify the interdependency of risk scores and clinicopathological characteristics, we further scrutinize the clinicopathological information from the TCGA cohort. The result manifested that the risk score apparently corresponded with pathological T stage, TNM stage, grade and primary therapy outcome of PC patients (Fig. 5A-D), while the tumor site, pathological N stage, age and tumor size proved to be nonsensical (Fig. 5E-H).
Pathway enrichment analysis and GSEA analysis in the TCGA dataset
For exploring the underlying pathway among the high-risk and low-risk subgroups, 1213 differentially expressed mRNAs (DEmRNAs) between high-risk and low-risk groups were selected according to the criteria of | log2(Fold change) | > 1 and p < 0.05. These DEmRNAs were significantly enriched in chemical synaptic transmission, regulation of hormone levels, presynapse and regulation of ion transport (Additional file 1: Figure S2). The results of GSEA revealed that the genes in the low-risk subgroup were significantly enriched in fatty acid metabolism, peroxisome, oxidative phosphorylation, PI3K-AKT-mTOR signaling and lysosome (Fig. 6A-F).
The independence of the Fe-LPM in predicting OS in PC
To evaluate whether the Fe-LPM was an independent prognostic indicator for OS, univariate and multivariate Cox regression analyses were implemented. In the TCGA dataset, both univariate Cox analysis and multivariate Cox analysis showed that Fe-LPM was markedly correlated with survival status (Fig. 7A, B). To establish a predictive tool for quantitative analysis of OS in PC patients, we initiated a prognostic nomogram based on the pathological T stage, risk score, primary therapy outcome and age in the TCGA cohort (Fig. 7C). Calibration plots showed that the predictive concordance of the nomogram (Fig. 7D).
Construction Of The Cerna Regulatory Network
The ceRNA network comprising lncRNAs, miRNAs and mRNAs was constructed for the sake of systematically probing into the potential regulatory mechanism. According to the aforementioned 23 ferroptosis-related lncRNAs, 4 of 23 lncRNAs were filtered from the miRcode database, and 27 target miRNAs were distinguished. Then we use the MiRDB, miRTarBase, StarBase and TargetScan databases to extracted 57 target mRNAs on account of 27 miRNAs. At length, 4 lncRNAs-27miRNAs-57mRNAs ceRNA network was completed (Fig. 8A, Additional file 1: Figure S3). The pathway enrichment analysis of 57 target mRNAs elucidated that these mRNAs were enriched in lipid transport, anion transport and dephosphorylation (Fig. 8B-D).