3.2. Signature of Prognostic Ferroptosis-Related lncRNAs
24 ferroptosis-related lncRNAs were supposed to have a prognostic value for COAD by P < 0.05 based on univariate Cox. Subsequently, 17 ferroptosis-related lncRNAs were identified after LASSO regression (Fig. 1) and further incorporated into a multivariate Cox model, only finding that eight lncRNAs might have a potential impact on COAD prognosis (Fig. 2).
According to the hazard ratio and KM survival curves, 7 lncRNAs (ZEB1-AS1, AL354993.2, LINC02381, AP001505.1, NCK1-DT, CD27-AS1, and AC068580.3) predicted a poor prognosis, and SNHG16 predicted a good prognosis (Fig. 3).
All the 8 lncRNAs were applied to build up a ferroptosis-related lncRNA signature (Table 1). The calculation formula of risk score: risk score = (0.557*ZEB1-AS1)+(0.199*AL354993.2)+(0.183* LINC02381)+(0.079*AP001505.1)+(0.279*NCK1-DT)+(0.107*CD27-AS1)+(0.532*AC068580.3)-(0.138*SNHG16).
Table 1
Multivariate Cox results of lncRNAs
ID | Coefficient | HR | HR.95L | HR.95H | P value |
ZEB1-AS1 | 0.557 | 1.746 | 1.144 | 2.665 | 0.010 |
AL354993.2 | 0.199 | 1.220 | 0.978 | 1.523 | 0.078 |
LINC02381 | 0.183 | 1.200 | 0.977 | 1.475 | 0.082 |
AP001505.1 | 0.079 | 1.082 | 1.020 | 1.149 | 0.009 |
NCK1-DT | 0.279 | 1.322 | 1.021 | 1.711 | 0.034 |
SNHG16 | -0.138 | 0.871 | 0.759 | 1.000 | 0.050 |
CD27-AS1 | 0.107 | 1.113 | 1.032 | 1.201 | 0.006 |
AC068580.3 | 0.532 | 1.703 | 1.193 | 2.431 | 0.003 |
3.3. Prognostic Performance of Ferroptosis-Related lncRNAs Signature
417 patients were classified into high-risk and low-risk groups based on the median of risk score and survival status distribution was also visualized. The heatmap demonstrated expression profiles of 8 Prognostic Ferroptosis-Related lncRNAs which were consistent with coefficients in risk score formula (Fig. 4). ZEB1-AS1, AL354993.2, LINC02381, AP001505.1, NCK1-DT, CD27-AS1, and AC068580.3 were up-regulated in the high-risk group, and SNHG16 was down-regulated in the high-risk group. Kaplan-Meier survival analysis of risk score indicated that the low-risk group had longer OS than the high-risk group (P < 0.001), so the risk score might be related to survival of COAD patients (Fig. 5A). The AUC of ROC curve in 3-years survival was 0.737 and 0.785 in 5-years survival (Fig. 5B).
3.4. Prognostic Validation of of Signature Combined with Clinical Features
According to the forest map, the hazard ratios of risk score in univariate and multivariate Cox regression analyses were 1.354 (95% CI: 1.228–1.494, P < 0.001) and 1.343 (95%CI = 1.202 -1.500, P < 0.001), respectively (Table 2). Either in univariate or multivariate analysis, the risk score was of independent significance for prognosis (Fig. 6A and Fig. 6B).
Table 2
Clinical characteristics and risk scores of COAD using multivariate cox regression.
Variable | Coef | SE (coef) | Z | HR | HR.95L | HR.95H | P Value |
age | 0.036 | 0.012 | 3.056 | 1.037 | 1.013 | 1.061 | 0.002 |
stage | 0.678 | 0.447 | 1.517 | 1.971 | 0.820 | 4.736 | 0.129 |
gender | -0.021 | 0.254 | -0.081 | 0.980 | 0.595 | 1.612 | 0.935 |
T | 0.530 | 0.309 | 1.713 | 1.699 | 0.927 | 3.116 | 0.087 |
M | 0.279 | 0.591 | 0.472 | 1.321 | 0.415 | 4.205 | 0.637 |
N | 0.146 | 0.260 | 0.562 | 1.158 | 0.695 | 1.928 | 0.574 |
RiskScore | 0.295 | 0.056 | 5.214 | 1.343 | 1.202 | 1.500 | 1.85E-07 |
According to the 5-year multivariate factor ROC curves, risk score with an AUC of 0.751 was better than age, gender, stage, and TNM in risk prediction. As shown in illustration, risk score was the largest contributors to OS of COAD patients (Fig. 7). Besides risk score, age and TNM stage also had a certain ability to predict overall survival, while gender had a weak ability. Increased with stage, risk score signature have potential ability in predicting the clinical progression of COAD (Table 3 and Fig. 8). Next, risk score, age, stage, T, M and N were incorporated into the nomogram (Fig. 9). We divided the clinical sample data into three groups to obtain the calibration diagram, showing that the OS predicted by the model was in a good agreement with the observed clinically (Fig. 10). The C-index of the prognostic model was 0.795 (se = 0.028).
Table 3
clinicopathological factors and risk score of COAD by multivariate Cox regression.
Clinical | Group | n | Risk score | t | P |
Mean | SD |
age | ༞60 | 257 | 1.360 | 1.455 | 0.683 | 0.495 |
≤ 60 | 106 | 1.258 | 1.232 | | |
gender | female | 169 | 1.304 | 1.244 | -0.344 | 0.731 |
male | 194 | 1.353 | 1.513 | | |
Stage | 1–2 | 211 | 1.198 | 1.371 | -2.132 | 0.034 |
3–4 | 152 | 1.514 | 1.407 | | |
T | 1–2 | 72 | 1.017 | 0.826 | -2.990 | 0.003 |
3–4 | 291 | 1.408 | 1.491 | | |
M | 0 | 308 | 1.274 | 1.348 | -1.612 | 0.112 |
1 | 55 | 1.644 | 1.601 | | |
N | 0 | 219 | 1.182 | 1.351 | -2.498 | 0.013 |
1–2 | 144 | 1.556 | 1.43 | | |
3.5. Exploration of Ferroptosis-Related lncRNAs function
As shown in the correlation heat map (Fig. 11), there was some potential correlations between some prognostic ferroptosis-related lncRNAs and ferroptosis-related mRNAs. Significant correlations were displayed by scatterplots with correlation coefficient \(\left|R\right|\)>0.4 and P < 0.001 (Fig. 12). Coexpression network and Sankey diagram were used to clarify the interactions and risk types between 8 lncRNAs and ferroptosis-related mRNA (Fig. 13A and Fig. 13B), which assisted to explore the significant target biomarkers further. The ferroptosis-related lncRNAs were mainly enriched in tricarboxylic acid cycle, aerobic respiration and some other oxidative metabolites in GO analysis. (Fig. 14A ). KEGG analysis revealed that these lncRNAs may concentrate in some classical tumor pathways, such as P53 signal pathway and metabolism of carbohydrate, lipid and protein (Fig. 14B). Furthermore, we also found that the gene sets were mainly involved in the decomposition, rather than synthesis, of nutrients to provide good conditions for tumor invasion and metastasis.
4.Discussion
Ferroptosis has been involved in colorectal, breast, pulmonary, pancreatic, renal and hepatic cancers [5, 23–25]. Recent studies have found that some modulators such as p53 affect both apoptosis and ferroptosis, though the role of p53 in autophagy has been researched for decades while has been explored in recent years in ferroptosis [26]. Ferroptosis often functions independent of other known cell death pathways, even in the absence of key effectors of apoptosis, such as BAX, BAK, and caspases[5]. New bioinformatic technologies have dug out various prognostic biomarkers for numerous cancers [27–32]. For example, the lncRNA RAMS11 has been proved to be associated with colorectal cancer progression [33].
The specific function of SNHG16 in COAD has not been determined, but a research on prognostic autophagy-related lncRNAs of COAD mentioned that SNHG16 predicted a good prognosis and AC068580.3, CD27-AS1 and LINC02381 predicted a poor prognosis of COAD [34]. These are consistent with the findings in the present study, indicating that some lncRNAs may regulate the process of colon adenocarcinoma cells death through both autophagy and ferroptosis. These results also provide some evidence that autophagy and ferroptosis may be simultaneously regulated by some pathways, although both exhibiting evident morphological, biochemical and genetic distinctions [6]. Still, some articles suggest that SNHG16 can promote the proliferation, migration or invasion of colorectal tumor cells by sponging miRNAs, such as miR-132-3p, and modulating some pathways, such as miR-302a-3p/AKT [35, 36]. Therefore, how SNHG16 regulates COAD progression remains to be answered with more in-deep studies.
ZEB1-AS1 has been proved to promote COAD malignant progression through miR-455-3P/PAK2 axis and miR-101/ZEB1 axis [37–39]. The dysregulation of ZEB1-AS1 supports the development of tumors and may serve as therapeutic target for cancer [38]. LINC02381 may increase the viability and promote the migration of tumor cells via targeting miR-133b [40]. CD27-AS1 is a carcinogenic RNA that regulates CD27 in melanomagenesis [41]. While an expression microarray analysis has revealed that CD27-AS1 is down-regulated in cervical carcinoma [42]. Few researches have studied the role of the remaining four prognostic ferroptosis-related lncRNAs (AL354993.2, AP001505.1, NCK1-DT and AC068580.3) in tumor development. Therefore, more studies are needed to probe the specific mechanism between selected prognostic lncRNAs and ferroptosis in COAD.
In the present study, the signature based on the eight ferroptosis-related lncRNAs showed accuracy in predicting the prognosis of patients with COAD. Similar to some other articles, the high-risk group had a shorter overall survival (OS) than the low-risk group in our study. The ROC curves and forest map showed that the signature may be significant in predicting the prognosis. The results of C-index and Calibration curve revealed that our nomogram might serve as a potential clinical tool for COAD prognosis. The eight ferroptosis-related lncRNAs may be taken as significant biomarkers in the prognosis of COAD.
The selected lncRNAs were mainly enriched in oxidative metabolism processes in GO analysis, and classical cancer pathways (especially p53 pathway) and catabolism in KEGG analysis. Recent some researches have suggested that ferroptosis can act as a tumor suppressive mechanism, always invovling p53 pathway [5, 23–25]. Studies on P53 gene have revealed that both wild-type and mutant p53 can contribute to ferroptosis through downregulating the post-transcriptional levels of tumor associated antigens [43].
There are still some deficiencies in our research. First, the sample size of our study is only limited to the data downloaded from the database. Secondly, our signature should be verified with more clinical data. Third, prospective studies are needed to prove whether the selected lncRNAs have quantitative relationships with the survival time in COAD patients.
5.Conclusion
A signature based on eight ferroptosis-related lncRNAs is efficient in predicting the prognosis of patients with COAD. The eight lncRNAs may be novel prognostic biomarkers for COAD. The specific mechanism of how ferroptosis affects the invasion and metastasis of COAD remains to be researched.