2.1 Construction and validation of CuRLsig in LUAD patients
A total of 504 CuRLs were screened out by Pearson correlation analysis (Table S1). Thirty-six candidate prognostic CuRLs, including 10 risk factors and 26 protective factors, were identified by UCRA (P < 0.05, Fig. 1A). Subsequently, LASSO-penalized Cox analysis was performed to reduce the overfitting and enhance the prognostic accuracy of these CuRLs (Fig. 1B, C). Twenty-four CuRLs were selected with minimum partial likelihood deviation (Log Lambda=-3.3). Finally, 13 CuRLs significantly correlated with the OS of LUAD patients were identified out by MCRA and designated as CuRLsig (Table 1). Principally, four CuRGs, PDHB, PDHA1, LITP2 and DLST, were negatively corelated with the CuRLsig, while another four CuRGs, NLRP3, MTF1, GLS and DBT, were positively associated with the CuRLsig (Figure S2). Based on the median value of CuRLsig-derived risk score, all LUAD patients were assigned into high- and low-risk groups in training, testing and TCGA sets (Figure S3A-C). Highly mortality was found in patients from high-risk scored group than those from low-risk scored group (Figure S3D-F). The heatmap of CuRLsig expression profiles in training, testing and TCGA sets were plotted (Figure S3H-J). Notably, PCA results of CuRLsig showed that the identified CuRLsig is powerful for discriminating the high-risk patients from low-risk patients (Figure S4).
To validate the predictive performance of CuRLsig for LUAD patients, Kaplan-Meier and ROC analyses were performed. In training set of patients, lower mortality was observed in low-risk scored group than that from high-risk scored group (median OS of 8.7 and 2.6 years, respectively; log-rank test, P < 0.001). The area under curves (AUC) of time-dependent ROC in training set were 0.789, 0.784 and 0.799 for 1-, 2- and 3-years OS predicted by CuRLsig, respectively (Fig. 1D). Similar results were obtained in testing and TCGA sets (Fig. 1E, F). These results support the potency of our CuRLsig on predicting the prognosis of LUAD patients.
Table 1
Cuproptosis-related lncRNA Signature
lncRNA | Coefficient | HR | HR-L (95%CI) | HR-H (95%CI) |
AC138965.1 | 0.526050707 | 1.692235959 | 1.049704964 | 2.728064209 |
LINC02605 | 0.226701734 | 1.254455651 | 1.058870861 | 1.486167048 |
AC092168.2 | 0.185038242 | 1.203264455 | 1.066713879 | 1.357294938 |
PLUT | 0.132577446 | 1.141767438 | 1.031084063 | 1.264332297 |
ARNTL2-AS1 | 0.113471892 | 1.120160403 | 0.987697738 | 1.270387975 |
AL161431.1 | 0.057111745 | 1.058774117 | 0.994621355 | 1.127064712 |
AC083806.2 | -0.130710829 | 0.877471477 | 0.734977543 | 1.047591455 |
AL133445.2 | -0.149653499 | 0.861006264 | 0.755743894 | 0.980929906 |
SMCR5 | -0.162072817 | 0.85037928 | 0.700802791 | 1.031880764 |
AC022272.1 | -0.178659912 | 0.836390298 | 0.730068157 | 0.958196469 |
AC005072.1 | -0.232238534 | 0.792757001 | 0.667191902 | 0.941953373 |
LINC01447 | -0.268292057 | 0.764684417 | 0.642139312 | 0.910615886 |
ADPGK-AS1 | -0.308677881 | 0.734417302 | 0.585019454 | 0.921967243 |
2.2 CuRLsig was a favorable prognostic factor with excellent prediction performance in OS of LUAD patients
UCRA and MCRA were further performed to determine whether the CuRLsig could be used as an independent prognostic indicator for LUAD patients. The results showed that CuRLsig-derived risk score and tumor stage were independent of other clinical characteristics on predicting the clinical outcomes (P < 0.05, Fig. 2A-C). So CuRLsig is an independent OS predictor for LUAD patients. To quantify the risk of each patient, we constructed a nomogram to predict OS in 1, 3, and 5 years by weighting gender, age, stage and risk value. From the nomogram, we can find that the survival probability of a low-risk patient (patient 30) in 1-year, 3-year and 5-year is 0.958, 0.836 and 0.685, respectively (Fig. 3A). The survival probability of another high-risk patient (patient 20) in 1-year, 3-year and 5-year were 0.756, 0.311 and 0.0847, respectively (Fig. 3B). Notably, the survival of low-risk patients was much longer than that of high-risk patients. C-index curves showed that the survival prediction by risk score was much superior to that by age, gender and stage (Fig. 3C), as well as all C-index were greater than 0.5 which reflects an good predictive performance of the nomogram. Meanwhile, the calibration plots of 1-, 3-, 5-year OS exhibited highly consistent between the actual survival rate and nomogram-predicted survival rate (Fig. 3D). Moreover, the AUCs of ROC for GIRlncSig-based risk score, age, gender and stage in TCGA set were 0.751, 0.537, 0.596 and 0.711, respectively (Fig. 3E). These results suggest that constructed nomogram possessed excellent practicable performance on OS prediction of LUAD.
To more intuitively estimate the advantage of the CuRLsig, we compared the prediction capacity of overall survival between our CuRLsig with literature-reported prognostic lncRNA models in LUAD patients. Li’s prediction model was composed of 7 immune-related LncRNAs (AC022784-1, NKILA, AC026355-1, AC068338-3, LINC01843, SYNPR-AS1 and AC123595-1) 27. In contrast, Jin’s prediction model was composed of another 7 immune-related lncRNA (AC092794.1, AL034397.3, AC069023.1, AP000695.1, AC091057.1, HLA-DQB1-AS1, HSPC324) 28. And, other 5-lncRNAs-constituted model (OGFRP1, ITGB1-DT, LMO7DN, NPSR1-AS1, PRKG1-AS1) was established by Zeng for predicting the OS of LUAD patients 29. As shown in Fig. 3F-H, the AUCs for 1-, 2-, 3-year survival prediction by our CuRLsig were 0.751, 0.712 and 0.718, respectively. Apparently, the OS-predication performance by our CuRLsig was more sensitive than that by other three models. These results suggest that our CuRLsig-derived risk model outperformed other models in predicting the overall survival of LUAD patients.
2.3 CuRLsig possessed credible prediction performance in PFS and DSS of LUAD patients
Furthermore, we found that CuRLsig could be used as an independent PFS and DSS predictor for LUAD patients (P < 0.05, Fig. 4A, B). Kaplan-Meier and ROC analyses demonstrated that lower incidence rate of PFS and DSS were observed in low-risk scored group than that from high-risk scored group (Fig. 4C, D top; log-rank test, P < 0.001). The AUC of time-dependent ROC in TCGA set were all greater than 0.62 for PFS and DSS predicted by CuRLsig (Fig. 4C, D bottom). Moreover, the ROC results of GIRlncSig-based risk score in PFS and DSS were obviously superior to those of age, gender and stage in TCGA set (Figure S5A, B). In addition, the prediction capacities of PFS and DSS of CuRLsig were also better than those of other three models (Fig. 4E, F). These results suggest that our CuRLsig has the credible prediction performance in PFS and DSS of LUAD patients.
2.4 GIRlncSig was a stable risk model in different clinical variables of LUAD patients
To validate the stability of our model, we performed stratification analysis between risk score and clinical variables. LUAD patients were first grouped by different clinical variables, and further stratified into high- and low-risk subgroups by risk score. We found that clinicopathological variables, like survival status, Stage, T- and N-stage, and tumor relapse, were significant differences in high- and low-risk group (Figure S6A). The percentage of clinical variables in T3-4, N1-3, Stage III-IV and relapse groups were higher in high-risk subgroup than those in low-risk subgroup. As well as patients in dead, T3-4, N1-3, Stage III-IV and relapse groups possessed higher risk scores than those in alive, T1-2, N0, stage I-II and non-relapse groups (Figure S6B-F). Kaplan-Meier survival analyses demonstrated that patients with low-risk score exhibited longer survival than those with high-risk score in all subgroups, excepting the pathologic M1-subgroup (log-rank test, P < 0.001; Fig. 5A-K). This may attribute to the fact that there are only 24 patients in M1-subgroup, which resulted in no survival-difference between high- and low-risk subgroups (log-rank test, P = 0.072; Fig. 5L). These results highlight the good stability of our GIRlncSig-based risk score model.
2.5 Biological Characteristics and Functions in Different Risk Groups
To determine the possible biological characteristics and functions in high- and low-risk groups of LUAD patients, enrichment analyses of GO, KEGG and cancer hallmark were performed using “GSVA” package. GO annotation is composed of three categories (Fig. 6A): biological process (BP), cellular component (CC) and molecular function (MF). CC results showed that the high-risk group presented enrichment of proteasome, cytochrome, respiratory chain and cyclin complex. While BP results displayed that the high-risk group enriched in redox homeostasis, mitochondrial ATP synthesis coupled electron transport and nucleic acid metabolism. Moreover, Heatmap of KEGG signaling pathways showed that proteasome, spliceosome, p53 signaling pathway and DNA damage and repair pathways were more enriched in high-risk group compared to that in low-risk group (Fig. 6B). Notably, further cancer hallmark analyses showed that the high-risk group were mainly enriched in the pathways like DNA repair, the targets of cancer genes, TNFα and MTORC1 signaling, hypoxia and glycometabolism (Fig. 6C). As the enriched biological and signal pathways are closely related to carcinogenesis process, these results suggest that the CuRLsig is essential to LUAD development.
2.6 Chemotherapeutic evaluation of LUAD patients with CuRLsig
To assess the predictive potential of CuRLsig on drug sensitivity, spearman correlation analysis was performed to estimate the IC50 of chemotherapeutics in high- and low-risk patients. A total of 65 chemotherapeutic drugs, including 5 resistance drugs and 60 sensitivity drugs, showed significant correlation with CuRLsig risk scores (Fig. 7A; Table S2). Notably, Drug resistances were mainly lied in SRC/PI3K/mTOR signals inhibitors for high-risk LUAD patients such as mTOR inhibitor Rapamycin and Phenformin, SRC inhibitor KIN001-135, and PI3K inhibitor YM201636 (Fig. 7B-E). While those high-risk LUAD patients were more sensitive to receptor tyrosine kinases signals inhibitors (Fig. 7F-I), like IGF1R and IR inhibitors (BMS-754807 and GSK1904529A) and VEGFR, PDGFR, KIT inhibitors (Tivozanib and Masitinib). In addition, some metabolism (OSU-03012), DNA replication (Cisplatin, Doxorubicin, Etoposide, Methotrexate, Mitomycin C, Gemcitabine, Camptothecin), genome integrity (Talazoparib and AG-014699), and mitosis (Epothilone B, Vinorelbine, Docetaxel, Vinblastine, VX-680, Ispinesib Mesylate, Paclitaxel)signals inhibitors were more sensitive in high-risk LUAD patients than those in low-risk LUAD patients (Figure. 7J-M). Overall, these results suggested that CuRLsig-derived score could be used for chemotherapeutic evaluation of LUAD patients.