Large-scale transcriptome analysis identified novel ferroptosis-related gene signatures for predicting the prognostic of patients with lung adenocarcinoma

Background: Lung cancer is the most common of all malignant tumors. Traditional tumor staging has general sensitivity and specificity in predicting patient prognosis. Ferroptosis is a novel form of non-apoptotic form of cell death promoted by lipid peroxidation. Ferroptosis may be involved in the malignant progression of tumors through the regulation of glutathione depletion or the P53 signaling pathway. However, there are fewer studies on the impact of ferroptosis on tumor prognosis. Methods: We summarized 22 molecules that regulate ferroptosis. The transcriptome and corresponding clinical data were obtained from the Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO) databases. The Wilcoxon test was utilized to analyze the expression of ferroptosis-related genes. We established risk score signatures, and all patients divided into two groups. Survival curves and multifactorial Cox regression analyses were performed to explore the prognostic value of ferroptosis on lung adenocarcinoma (LUAD). Results: We found that most ferroptosis-related genes are specifically expressed in LUAD tissue. We established a six‑mRNA signature and a seven‑IncRNA signature. All the models showed high predictive performance (AUC: 0.66–0.8), and patients in the low-risk group had higher overall survival than those in the high-risk group (P<0.05). The risk score is an independent risk factor for predicting the prognosis of LUAD. Conclusions: Our study demonstrates the critical role of ferroptosis in LUAD, and it is expected to supply a reference for the prognostic stratification of LUAD. prognosis of lung adenocarcinoma. Our study explains the critical role of ferroptosis in lung adenocarcinoma, and it is expected to


Background
Lung cancer is one of the most common malignancies in the world. It is estimated that lung cancer-related deaths account for nearly 23% of all cancer-related deaths in 2020 [1]. Non-small cell lung cancer (NSCLC) accounts for 80%-85% of all lung cancers, mainly including lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) [2]. In clinical practice, traditional tumor staging is often used to predict the prognosis of lung cancer patients. However, patients with the same tumor stage sometimes have different prognoses due to individual differences [3]. Over the past decade, it has been recognized that the development of lung cancer is a complex process involving multiple factors [4]. Different molecular typing affects patients' drug response and clinical prognosis [5,6]. Therefore, looking for the new molecular typing of lung cancer is expected to be a better choice of adjuvant treatment strategies and predict the prognosis of patients.
With the development of information technology, whole gene sequencing has become an effective method for analyzing gene expression profiles. Tumor databases established around the world provide the basis for the discovery of new tumor biomarkers and prognostic signatures. For example, based on the Cancer Genome Atlas (TCGA) database, many new molecules were found to be involved in the malignant progression of tumors, and different molecular signatures can be used to predict the prognosis of tumors [7]. Immunotherapy is an effective treatment for lung cancer, and tumor signatures based on immune-related genes can effectively predict the survival and recurrence of patients [8,9]. In addition, there are other signatures for lung cancer prognosis such as autophagy, RNA methylation [10,11]. However, there are few studies about the prognosis signatures of cell death.
In recent years, ferroptosis has been identified as a new form of programmed cell necrosis [12]. It is morphologically, biochemically, and genetically distinct from apoptosis, necrosis, and autophagy. Ferroptosis is a non-apoptotic form of cell death promoted by lipid peroxidation [13,14]. During mammalian development, increased accumulation of glutamate, iron, and polyunsaturated fatty acids, and increased consumption of nicotinamide adenine dinucleotide phosphate (NADPH) suggests that ferroptosis may be a necessary physiological function for body growth [12]. More studies have shown that ferroptosis is associated with a variety of pathological cell deaths. For example, some degenerative pathological changes are cell deaths due to a reduced ability to repair lipid peroxides, and ferroptosis may inhibit tumor progression by removing cells with abnormal nutrient absorption [14][15][16]. However, these studies are still in their infancy.
Ferroptosis was discovered in the study of erastin kills RAS-mutated tumor cells in 2012 [12]. A growing number of subsequent studies have confirmed that ferroptosis is involved in multiple processes of tumorigenesis and development. Nuclear factor erythroid2-related factor 2 (NRF2) inhibitors can inhibit the expression of Metallothionein 1G (MT-1G), increase glutathione consumption and lipid peroxidation, and promote sorafenib-induced ferroptosis [17]. This confirmed that NRF2 is a gene that suppresses ferroptosis. Tumor Protein P53 (TP53) is the most studied molecule of promoting ferroptosis, and TP53 is also recognized as a tumor suppressor protein that inhibits tumor progression through mediated cell cycle arrest, apoptosis, and senescence [18,19].
There are few studies on the impact of ferroptosis affecting the prognosis of cancer patients. Through the study of literature, we summarized 22 molecules that regulate ferroptosis (Table S1). In this study, we systematically analyzed the expression of these molecules in LUAD. We further searched for ferroptosis-related IncRNAs. Based on the expression of these genes, we constructed prognostic signatures to confirm the clinical value of ferroptosis in LUAD.

Datasets Acquisition
The mRNA expression profile and corresponding clinical information of LUAD were downloaded from the TCGA database(https://cancergenome.nih.gov/). As the training set, mRNA expression data included 59 cases of para-cancer tissues and 535 cases of tumor tissues. Four hundred and eighty-seven clinical cases with a follow-up period of more than 30 days and less than 10 years were obtained. Clinical information included age, gender, stage, T, N, M, overall survival (OS), and survival status (Table S2).
The test set data was downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) under the accession number (GSE42127). It is used to test the reliability of the prognostic signature constructed by the training set.
One hundred and twenty-six clinical cases were obtained. Clinical information included age, gender, stage, OS, and survival status (Table S3).

Bioinformatic analysis
After normalizing transcriptome data from the training and test sets, we extracted expression data of 22 genes related to ferroptosis. Expression levels of 22 molecules were analyzed in the TCGA database, classified according to lung adenocarcinoma tissues and adjacent tissues. Heatmap and box plot were drawn to observe the expression distribution of these ferroptosis-related molecules.
After removing the paracancerous cases in the training set, all the genes were fitted in a stepwise multivariate Cox proportional regression model to screen out risk genes and construct a prognostic signature. The risk score for each case was calculated using the following formula, where Coefi is the coefficient, and xi is the expression value of each selected risk gene. All cases were divided into high-risk and low-risk groups according to the median value of risk score. The Kaplan-Meier curve was used to analyze the difference in survival between the two groups, and the ROC curve was used to test the effectiveness of the prognostic signature. Univariate and multivariate regression was used to analyze the independent prognostic value of risk scores.

= ∑ * =1
Pearson correlation coefficient was calculated for correlations between the lncRNAs and ferroptosis-related genes. A lncRNA with a correlation coefficient (|R 2 | > 0.3 and P < 0.05) was considered to be an ferroptosis-related lncRNA. Univariate Cox regression analysis was used to screened out lncRNAs related to overall survival (OS) with P values < 0.01. The prognostic signature was constructed based on the expression of these lncRNAs, and patients also were divided into the high-risk group and low-risk analysis was performed with 1000 iterations. P-value (P < 0.05) and normalized enrichment score (NES) were used to find the most relevant KEGG pathways.
Wilcoxon test was used to compare differences in the expression of ferroptosis-related genes between tumor tissues and paracancerous tissues. The Kaplan-Meier method was used to compare the OS in different groups of patients. Pearson correlation coefficient is used to analyze the correlation between ferroptosis-related genes and lncRNAs. P < 0.05 was considered statistically significant.

Expression of ferroptosis-related genes in lung adenocarcinoma
In the first step, we analyzed the expression levels of 22 ferroptosis-related genes in tumor tissues and paracancerous tissues. The expression of seven ferroptosis-related genes was lower in tumor tissues than that in the adjacent tissues (p<0.05). The expression of fourteen ferroptosis-related genes was higher in tumor tissues than that in the adjacent tissues (p<0.05). In general, the expression of most ferroptosis-related genes is significantly different in cancer and paracancerous tissues, as shown in Figure   1A,B. The result suggests that ferroptosis may be involved in the malignant progression of LUAD.

Construction of the six-mRNA signature for predicting patient prognosis
To explore the prognostic role of ferroptosis in LUAD, we constructed a six-mRNA signature (ACSL4, GPX4, NCOA4, VDAC1, ISCU, CISD1), and the coefficients of each gene are shown in Table 1. All patients were divided into high-risk and low-risk groups based on the median risk score. The results showed that high expression of ACSL4, VDAC1, and CISD1 have worse survival in patients with LUAD. In contrast, high expression of GPX4, NCOA4, and ISCU have better survival in patients with LUAD. The survival curve showed a significant difference in the survival rates of the two groups (p<0.05) (Fig. 2A). The ROC curve verifies the high sensitivity and specificity of the signature (AUC:0.67~0.71) (Fig. 2B). Further, univariate and multivariate Cox regression results suggest that the signature can be an independent indicator for predicting the prognosis of patients with LUAD (p <0.05) (Fig. 2C-D).

Validation of the six-mRNA signature in predicting patient prognosis
Using the risk formula, we calculated the risk score for each patient in the test set. All cases in the test set were divided into high-risk and low-risk groups based on the median risk in the training set. The results of the test set also showed a significant difference in survival between the two groups (p <0.05), and high predictive effect of the signature (AUC:0.66~0.8) (Fig. 3A-B). Univariate and multivariate Cox regression also supported an independent risk role for the signatures in the test set (p<0.05) ( Figure 3C-D). Overall, these results confirm that ferroptosis affects the prognosis of LUAD, and the six-mRNA signature we constructed has high predictive value. IncRNAs, the expression of one IncRNA was lower in tumor tissues than that in paracancerous tissues, and the expression of four IncRNAs was higher in LUAD tissues than that in paracancerous tissues(p<0.05) (Fig. 4A-B). These results suggest that ferroptosis-mediated malignant progression of LUAD may be regulated by IncRNAs.
In the seven-IncRNA signature, all patients were also divided into high-risk and lowrisk groups, and the coefficients for each IncRNA are shown in Table 2. The seven-IncRNA signature also showed a significant difference in survival between the two groups (p <0.05), and high sensitivity and specificity (AUC:0.7~0.72) (Fig. 4C-D).
The risk score can also be an independent risk factor to predict patient prognosis through univariate and multivariate Cox regression (p <0.05) (Fig. 4E-F). This result provides an IncRNA-related biomarker for the prognosis of LUAD and broadens the way for subsequent research.

Gene set enrichment analysis
We performed KEGG analysis of gene expression in patients at six-mRNA and seven-IncRNA signature groups. The results of different groups consistent. In general, cell cycle and DNA replication were significantly enriched in the high-risk group.
Asthma and intestinal immune network for IGA production were significantly enriched in the low-risk group (Fig.5). These results provide a reference for further functions and mechanisms of ferroptosis in LUAD.

DISCUSSION
Over the past few decades, people recognized that the development of lung cancer is a complex process involving multiple factors, and the factors that affect the prognosis of lung cancer patients are various [4,20]. patients [21]. Another study established a six-gene survival risk model related to lung adenocarcinoma microenvironment and confirmed the prognostic value of the model for patients with LUAD [22]. Cell death is essential for the development and homeostatic maintenance of the organism, and tumor cell death has its own characteristics. In this study, we focus on ferroptosis, a new form of programmed cell death, and explore its impact on lung adenocarcinoma and clinical prognostic value.
Although studies have shown an important role for ferroptosis in human disease, the exact mechanism is not yet clear. Through the study of literature, we summarized 22 molecules that regulate ferroptosis and constructed a six-mRNA prognostic signature.
Doll's study found that ACSL4 mediates ferroptosis sensitivity by forming a lipid component of the cell [23]. Nagakannan's study found that VDAC1 protein maintains calcium homeostasis and reactive oxygen species (ROS) levels in mitochondria and is closely associated with the occurrence of ferroptosis [24]. Gao's study found that autophagy inhibitors or knock-down of NCOA4 blocked the degradation of ferritin and inhibited the occurrence of ferroptosis [14]. GPX4, CISD1, ISCU negatively modulate ferroptosis. Yang's study found that GSH catalyzes the reduction of lipid peroxides as a cofactor of GPX4, protecting cells and cell membranes from peroxides and ultimately inhibiting the occurrence of cellular ferroptosis [25]. Yuan's study found that inhibiting CISD1 expression causes iron accumulation in mitochondria and enhances lipid peroxidation, thereby promoting ferroptosis [26]. Du's study found that ISCU overexpression increased GSH levels and attenuated ferroptosis [27].
IncRNA dynamically regulates the transcription and translation of genes, and the prognostic model based on IncRNAs is now a popular research topic [28]. Zhao's study found that lncRNA-CCAT1 knockdown suppressed tumor proliferation and induced apoptosis. High expression of lncRNA-CCAT1 was related to tumor growth and reduced survival rate, which can be a novel marker for lung cancer [29]. Ju's study systematically found prognosis-related lncRNAs, miRNAs, and mRNAs and constructed a prognosis-related competing endogenous RNA network [30]. Zhou's study screened for autophagy-associated IncRNAs in lung adenocarcinoma and constructed a thirteen-IncRNA signature, which showed a high clinical predictive effect [31].
Ferroptosis is also regulated by IncRNA. Wang's study found that LINC00336 is upregulated in lung cancer and inhibits ferroptosis by functioning as a competing endogenous RNA [32]. In this study, we screened for ferroptosis-related IncRNAs affecting the prognosis of lung adenocarcinoma and constructed a prognostic signature with high clinical value. Among these risk molecules, LINC00324 was the most widely studied. For example, LINC00324 prevents breast cancer progression by modulating miR-10b-5p [33]. LINC00324 exerts tumor-promoting functions in lung adenocarcinoma via targeting miR-615-5p/AKT1 axis [34]. LINC00324 also serves as a biomarker for gastric cancer and rectal adenocarcinoma [35,36]. Jiang's study found that EBLN3P promotes the recovery of the function of impaired spiral ganglion neurons by competitively binding to miR-204-5p and regulating TMPRSS3 expression [37].
Ferroptosis was first found in tumor cells with RAS mutations, and a growing number of subsequent studies have demonstrated that ferroptosis is involved in multiple biological processes in tumors [12]. Kuang's paper on the relationship between iron and lung cancer mentions that the Intracellular iron depletion by the iron chelator or ROS can inhibit ferroptosis. In converse, ferric ammonium citrate, or iron chloride hexahydrate increase intracellular iron levels and the sensitivity of ferroptosis [38].
Several studies have tried to explore the therapeutic value of ferroptosis in tumors. The common ferroptosis inducers erastin and RAS selective lethality factor (RSL3) are not suitable for tumor therapy due to their pharmacological properties. However, FDAapproved drugs such as sorafenib, sulfasalazine, and artesunate have been shown to induce ferroptosis [39,40]. As far as current research is concerned, the mechanism of ferroptosis cannot be clearly elucidated, but its great potential for tumor research can already be identified.

Conclusion
Through the study of ferroptosis, we summarized twenty-two genes that regulate ferroptosis, and further searched for ferroptosis-related IncRNAs. The expression of most genes in tumor tissues differed from that in adjacent tissues, which suggested that ferroptosis may play a vital role in the malignant progression of lung adenocarcinoma.
We respectively constructed a six-mRNA signature and a seven-IncRNA signature for prognostic prediction of lung adenocarcinoma, and the results of both signatures were consistent. Survival differences were significant in different risk groups and the ROC curve showed high specificity and sensitivity. The risk score of prognostic signatures can be an independent risk factor for the prognosis of lung adenocarcinoma. Our study explains the critical role of ferroptosis in lung adenocarcinoma, and it is expected to supply a reference for the prognostic stratification and treatment strategy development of lung adenocarcinoma.

Acknowledgements
We thank our anonymous reviewers for their valuable comments on this manuscript, which have led to much many improvements to the article.

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Ethics approval and consent to participate
Not applicable.

Consent for publication
Not applicable.