A novel ferroptosis-related lncRNA signature for prognosis prediction in gastric cancer

Background Gastric cancer (GC) is a common malignant cancer with a poor prognosis. Ferroptosis has been shown to play crucial roles in GC development. Long non-coding RNAs (lncRNAs) is also associated with tumor progression in GC. This study aimed to screen the prognostic ferroptosis-related lncRNAs and to construct a prognostic risk model for GC. Methods Ferroptosis-related lncRNAs from The Cancer Genome Atlas (TCGA) GC expression data was downloaded. First, single factor Cox proportional hazard regression analysis was used to select seven prognostic ferroptosis-related lncRNAs from TCGA database. And then, the selected lncRNAs were further included in the multivariate Cox proportional hazard regression analysis to establish the prognostic model. A nomogram was constructed to predict individual survival probability. Finally, we performed quantitative reverse transcription polymerase chain reaction (qRT-PCR) to verify the risk model. Results We constructed a prognostic ferroptosis-related lncRNA signature in this study. Kaplan-Meier curve analysis revealed a significantly better prognosis for the low-risk group than for the high-risk group (P = 2.036e-05). Multivariate Cox proportional risk regression analysis demonstrated that risk score was an independent prognostic factor [hazard ratio (HR) = 1.798, 95% confidence interval (CI) =1.410–2.291, P < 0.001]. A nomogram, receiver operating characteristic curve, and principal component analysis were used to predict individual prognosis. Finally, the expression levels of AP003392.1, AC245041.2, AP001271.1, and BOLA3-AS1 in GC cell lines and normal cell lines were tested by qRT-PCR. Conclusions This risk model was shown to be a novel method for predicting prognosis for GC patients. Supplementary Information The online version contains supplementary material available at 10.1186/s12885-021-08975-2.

shown that ferroptosis is associated with drug resistance [6,7]. Zhang, H et al. found that ferroptosis promoted drug resistance in GC [8], while Shin, D. et al. indicated that Nrf2 inhibition reversed resistance to GPX4 inhibitor-induced ferroptosis in head and neck cancers [9]. These results revealed that ferroptosis played an important role in cancer development.
Long non-coding RNAs (lncRNAs) are a class of transcripts longer than 200 nucleotides [10]. Increasing studies have shown that lncRNAs play a crucial role in tumor progression [11,12]. Recently, molecular risk signatures as prognostic predictors of cancer progression were investigated, especially lncRNA signatures [13,14]. However, the prognostic value of ferroptosis-related lncRNAs signature in GC has not been systematically explored.
In this study, we constructed a molecular signature model comprising four prognostic ferroptosis-related lncRNAs and assessed their prognostic ability for GC. This study showed that the prognostic ferroptosisrelated lncRNA signature was a reliable prognostic predictor for GC. Moreover, we explored the correlation between clinical factors and the four prognostic ferroptosis-related lncRNAs in this risk model. Finally, four lncRNAs (AP003392.1, AC245041.2, AP001271.1 and BOLA3-AS1) were validated in GC cell lines.

Identification and correlation analysis
A total of 40 ferroptosis-related genes were downloaded from MSigDB v7.2. First, we selected all lncRNA expression data from TCGA database as shown in Table 1.
Pearson correlation analysis was performed to determine the correlation between ferroptosis-related genes and all lncRNA expression data of the samples. Finally, ferroptosis-related lncRNAs were identified based on the Pearson correlation coefficient and p values (|Cor Pearson | > 0.4, p < 0.001). The correlation between ferroptosis-related lncRNAs and clinical characteristics was analyzed using the R package "ggpubr."

Prognostic risk signature construction
Single-variable Cox proportional risk regression analysis was performed to identify ferroptosis-related lncRNAs, which were significantly associated with OS in TCGA GC dataset. Multivariate Cox proportional risk regression analysis was then used to establish the prognostic model for GC. The results were plotted using a forest map with the R package "ggplot2." The risk score for each patient was calculated using the risk formula: explncRNA1*coef lncRNA1 + explncRNA2*coef lncRNA2 + … + explncRNAi*coef lncRNAi [15]. Kaplan -Meier plotter was used to analyze the different OS times between the high-risk and low-risk groups using the R package "survival." PCA PCA is a widely used tool for dimensionality reduction and feature extraction in the computer vision field [16]. The R package "scatterplot3d" was used to assess potential differences between the high-risk and low-risk groups.

Construction of nomogram
A nomogram was built according to all the independent prognostic factors using the R package "rms" (https://cran.rproject.org/web/packages/rms/index. html). A calibration plot curve analysis was performed to assess the consistency between the actual and predicted survival.

Cell culture and qRT-PCR
Human gastric epithelial cell line (GES-1) and GC cell lines (MKN-45 and AGS) were obtained from Tianjin Createch Biotechnology Co. LTD (Tianjin, China). All cells were maintained in RPMI 1640 medium (Invitrogen, Carlsbad, CA, USA) supplemented with 1% penicillin-streptomycin (Invitrogen) and 10% fetal bovine serum (Invitrogen) at 37°C and 5% CO 2 . qRT-PCR was performed according to the method described earlier [17]. Total RNA was extracted using TRIzol reagent (Invitrogen) and synthesized into cDNA using M-MLV reverse transcriptase (TaKaRa Bio, Japan) following the manufacturer's instructions. qRT-PCR was performed using SYBR Green assay (Roche, Switzerland). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) or U6 was utilized as an endogenous reference. The primers sequences are listed in Supplementary Table S1.

Statistical analysis
All data were analyzed using the R software (R version: 3.6.1) and the following packages: "limma," "survival," "Pheatmap," "ggpubr," and "survivalROC." The OS difference was determined using the Kaplan -Meier analysis method and logrank test. Statistical significance was set at p < 0.05.

Results
The workflow of the prognostic model analysis is illustrated in Fig. 1. In this study, we used the data of 407 GC patients from The Cancer Genome Atlas (TCGA) cohort (T = 375, N = 32).

Ferroptosis-related lncRNA identification in TCGA cohort
Forty ferroptosis-related mRNAs were selected based on one ferroptosis gene set (WP_FERROPTOSIS) from the

Construction of prognostic ferroptosis -related lncRNA signature
The clinical data of 334 GC patients were analyzed. Using univariate Cox regression analysis, seven prognostic ferroptosis-related lncRNAs were identified. The results were presented as a forest plot in Fig. 2A. Seven prognostic ferroptosis-related lncRNAs were further analyzed using multivariate Cox regression analysis. Finally, a 4-ferroptosis-related lncRNA signature model was established. The coefficients of each lncRNA were listed in Table 2.
The risk score was calculated as follows: Risk score = (− 0.509319782 × expression value of AP003392.1) + (0.235578881 × expression value of AC245041.2) + (− 0.313355736 × expression value of AP001271.1) + (0.525606422 × expression value of BOLA3-AS1). Heatmap library was used to evaluate the values of the risk scores. Based on the median risk score, we divided the patients into high-risk and low-risk groups and assessed the score's ability. The patients were ranked from low to high according to the risk score. We have shown the population follow-up time and gene heat-map by ranking as well (Fig. 2B -D).

Correlation between the four lncRNAs and clinical features
The correlation between the four prognostic ferroptosisrelated lncRNA signatures and clinicopathological characteristics was assessed. The expression of BOLA3-AS1 was significantly associated with T and age (Fig. 3A, E). Moreover, AP001271.1 expression was significantly associated with gender and grade (Fig. 3C, F). However, the expression of none of the four lncRNAs was significantly associated with N and M (Fig. 3B, D).

Differential analysis between high-risk and low-risk groups
Results of principal component analysis (PCA) are shown in Fig. 5. There were no significant differences between the high-risk and low-risk groups in the expression of all examined genes (Fig. 5A), and ferroptosisrelated genes (Fig. 5B), and lncRNAs (Fig. 5C); however, but there was a significant difference between the highrisk and low-risk groups in the expression of for the four lncRNAs (Fig. 5D) used in the prognostic model. This study revealed that the high-risk and low-risk groups were significantly different in terms of the ferroptosis signature.

Discussion
The term ferroptosis was coined in 2012 to describe iron-dependent cell death [19]. Ferroptosis is increasingly recognized to be associated with the prognosis of patients with GC and other cancers [20][21][22][23][24][25]. Several studies have focused on the effect of ferroptosis on tumor development and treatment. Carbone, M. et al.
found that regulating ferroptosis could be a new therapeutic approach in ovarian cancer [26], revealing the potential value of ferroptosis in guiding clinical decisions. Zhang et al. showed that ferroptosis was influenced by the differential regulation of transcription in liver cancer [27].
In this study, we first identified 173 ferroptosis-related lncRNAs, and then, 7 prognostic ferroptosis-related lncRNAs were further analyzed using multivariate Cox regression analysis. Finally, a four -ferroptosis-related lncRNA signature model was established from TCGA datasets. We also demonstrated that four ferroptosisrelated lncRNA signature was an independent risk factor for GC. This result indicated that the prognostic signature of the four ferroptosis-related lncRNAs could accurately predict the prognosis of GC patients.
Nomograms are widely used in prognostic prediction in oncology and medicine [45]. In the current study, a nomogram was constructed using AP003392.1, AC245041.2, AP001271.1, BOLA3-AS1, and the risk score. The nomogram showed a good performance in predicting 1-year, 2-year, and 3-year survival OS of GC patients, which may contribute to the promotion of individualized treatment of GC patients. To investigate the relationship between the four ferroptosis-related lncRNAs and clinical factors, we further analyzed the relationship between clinical features and the four lncRNAs. We found that the four prognostic ferroptosisrelated lncRNAs were associated with T-staging, sex, age, and grade.
Interestingly, Wang, S et al. built a novel prognostic nomogram based on five lnc-RNAs in clear cell renal cell carcinoma, comprising AC026992.2, AC245041.2, LINC00524, LINC01956, and LINC02080 [46]. Previous work on lncRNA profiling also revealed that BOLA3-AS1 was associated with higher-risk myelodysplastic syndrome, which played an important role in the development of blood lineages such as platelets, erythrocytes, and myeloid cells [47]. However, the roles of AP003392.1 and AP001271.1 are yet unknown. To explore the expression levels of AP003392.1, AC245041.2, AP001271.1, and BOLA3-AS1, qRT-PCR analysis was applied. We found that the expression of AP003392.1, AC245041.2, AP001271.1, and BOLA3-AS1 was upregulated in GC cells lines. Thus, despite the important prognostic value of ferroptosis-related lncRNA signature identified in this study, future experiments on lncRNAs components are required to elucidate their roles in GC.
Our study has some limitations. First, this was a retrospective study, and therefore it lacked novel clinical samples and data. Second, the mechanism and interrelationship among ferroptosis-related lncRNAs comprising this signature require further study. In conclusion, this study constructed a prognostic risk model consisting of four ferroptosis-related lncRNAs by analyzing the RNA-sequencing-based gene expression profiles of GC from TCGA database. This risk model has been established be independently associated with OS and facilitated the prediction of GC prognosis. However, the mechanisms underlying the interplay between ferroptosis and lncRNAs in GC require further investigation.

Conclusions
This risk model was shown to be a novel method for predicting prognosis for GC patients. Thus, the
Additional file 1. The primers sequences of four ferroptosis-related lncRNAs.