In present study, only the white racial was selected, so the prognostic model we constructed was more accurate to white racial populations. Considering guarantee the effectiveness of the model, the entire datasets were partitioned into training and test sets using Stata16/SE block-randomization script, the model was constructed using the training series, and validated in the test series and entire series. As shown in Table 1, the training series and the test series were randomly allocated. To avoid overfitting, the LASSO-Cox regression method was performed for further select lncRNAs. Considering clinical factors are also important in clinical practice, so the significant clinical variables, especially AJCC-TNM stage and age, were combined with risk score model to better predict the prognosis.
We screened 616 differentially expressed lncRNAs using TCGA-KIRC dataset in the training cohort, of which 334 were down-regulated and 282 were up-regulated. Among them, 12 lncRNAs (AC008556.1, AC012404.1, AC092296.1, AC099684.2, AC108752.1, AC131097.1, AL606519.1, FOXP4-AS1, LINC00261, LINC02446, LINC02475, and SPINT1-AS1) were associated with the prognosis of ccRCC (all P < 0.05). Then we constructed a prognostic 12-lncRNAs signature, the results proved that a higher risk score was correlated with inferior OS, suggesting that risk score was related to the prognosis of ccRCC. In addition, an outstanding consistency between the
real survival situation and prediction results through the prognostic nomogram were presented in the calibration plots for 1-, 3-, 5-, and 10-year OS. The 12-lncRNAs signature all remains an effective prognostic model, when stratified by age, gender, treatment type, and the AJCC-TNM stage, indicating that this lncRNAs signature was effective in patients with different clinical features. Then, 12-lncRNAs signature was combined with age and AJCC-TNM stage to better predict the prognosis. Finally, the prognostic effects of this combined model were validated in the test and entire cohorts, the results in two data sets were both consistent with the result in training data set, indicating broad applicability of this signature in ccRCC patients. As shown in the Kaplan-Meier plots for all cohorts, OS rate of subjects with low-risk scores was significantly higher than subjects with high-risk scores (P < 0.001). All the results suggesting that this prognostic model is suitable for estimating the OS of ccRCC patients.
The performance evaluation of combined model showed that concordance index (95%CI) in the training, test, and entire cohorts were 0.863(0.830–0.896), 0.863(0.814–0.912), and 0.841(0.812–0.870), respectively, the 5-year area under time-dependent receiver operator characteristic curve (95%CI) in the training, test, and entire cohorts were 0.923(0.879–0.967), 0.861(0.777–0.945), and 0.879(0.834–0.924), respectively. The results of three data sets were all greater than 0.800, indicating that this combined model has higher prediction ability.
For the prognostic model of RCC reported before, three important signatures were used for different types of patients. The University of California Los Angeles integrated staging system model integrated Fuhrman grade, Eastern Cooperative Oncology Group performance status, and TNM stage to predict survival for RCC patients, this model is superior to stage alone in differentiating the survival of patients and is simple to use, however, because of the heterogeneity of patients and treatments, it may be less accurate in the metastatic RCC patients [24, 25]. The Memorial Sloan-Kettering Cancer Center model used five pretreatment factors to predict the survival of metastatic patients [26, 27]. The International Metastatic Renal Cell Carcinoma Database Consortium model use six prognostic factors to predict the survival of metastatic patients [28]. All three prognostic signatures were stratified the subjects into three different risk groups including high-, intermediate-, and low-risk groups. These models had been validated before, the prognostic factors contained in these models are mostly biochemical indicators, whereas no biomarkers such as genes were included.
In the risk score model, 12 lncRNAs could be seen as potential prognostic factors. Among these, AC008556.1, AC012404.1, AC092296.1, AC099684.2, and SPINT1-AS1 are the protective biomarkers. However, the biological functions of AC012404.1, AC092296.1, and AC099684.2 have not been reported in previous research, only AC008556.1 and SPINT1-AS1 were studied before in other cancers. AC008556.1 (ENSG00000277013), as one of the novel S-phase-upregulated lncRNAs, was used in mechanistic studies to determine the role in cell-cycle progression [29]. Serine peptidase inhibitor, Kunitz Type 1 antisense RNA1 (SPINT1-AS1) is a member of serine protease inhibitors of the Kunitz family, has anti-cancer properties through inhibiting cell proliferation, invasion, migration, and decreased expression in many cancers [30]. Zhou et al. [31] confirmed that SPINT1-AS1 is up-regulated in the breast cancer cell lines and it can promote the migration and proliferation of breast cancer cells through regulating miR-let-7a/b/i-5p. However, SPINT1-AS1 is down-regulated in this research. Huang et al. [32]reported that the prognostic role of some lncRNAs are consistent in many tumors, while other lncRNAs may have different functions in different tumors. Thus, the biological function of SPINT1-AS1 in ccRCC need more experiments to explore.
High expression of AC108752.1, AC131097.1, AL606519.1, FOXP4-AS1, LINC00261, LINC02446, and LINC02475 were significantly correlated with an inferior prognosis of ccRCC. However, to date, the biological functions of AC108752.1, AC131097.1, AL606519.1, and LINC02475 were little known, no experimental studies have been performed in cancer before, only LINC02446, FOXP4-AS1, and LINC00261 were studied before in other cancers. LINC02446 (ENSG00000256039) was identified as Epithelial-Mesenchymal Transition-Related lncRNAs correlated with the progression and prognosis in bladder cancer patients [33]. Forkhead box P4 antisense RNA 1 (FOXP4-AS1) is a 24.727 kb lncRNA. Li et al. [34] have confirmed that FOXP4-AS1 expression is associated to the development of colorectal cancer (CRC), in vivo experiments demonstrated that silencing of FOXP4-AS1 in CRC cell lines can inhibit the ability of tumor cells to form tumors in nude mice. LINC00261 (ENSG00000236384) was found to behave as a tumor suppressor through activating DNA damage signaling pathway to arrest cellular division in lung adenocarcinoma [35]. Yan et al. [36] demonstrated that LINC00261 repressed colon cancer progression through regulating the Wnt and miR-324-3p pathway. However, the function and expression of LINC02446, FOXP4-AS1, and LINC00261 in ccRCC are still unknown.
The present study has several important strengths. Firstly, we found seven novel candidate prognostic biomarkers including AC012404.1, AC092296.1, AC099684.2, AC108752.1, AC131097.1, AL606519.1, and LINC02475 in ccRCC, the related research should be conducted to explore their biological functions. Furthermore, the significant clinical factors were combined with the 12-lncRNAs signature to make it more functional in clinical practice. Finally, the inclusion criteria were stricter in this study to decrease the confounding factors.
In the present research, though the prediction model elucidated good accuracy in white patients of ccRCC, some limitations should be acknowledged. Firstly, the research data were downloaded from public available TCGA datasets and no validation set was from a prospective cohort, so experimental studies should be prepared in our later research to further validate prognostic value of the model we constructed in ccRCC patients. Furthermore, the mechanism between the 12-lncRNAs and ccRCC prognosis is unclear, so more mechanism research should be conducted to explore the role of these 12 lncRNAs in ccRCC. We acknowledge that before this model applied in the clinic, it is necessary to conduct more experimental researches to further confirm this model.