Identication of a nine ferroptosis-related lncRNA prognostic signature for lung adenocarcinoma

Recently, mounting of studies has shown that lncRNA affects tumor progression through the regulation of ferroptosis. The current study aims to construct a robust ferroptosis-related lncRNAs signature to increase the predicted value of lung adenocarcinoma (LUAD) by bioinformatics analysis. The transcriptome data were abstracted from The Cancer Genome Atlas (TCGA). Differentially expressed lncRNAs were screened by comparing 535 LAUD tissues with 59 adjacent non-LAUD tissues. Univariate Cox regression, lasso regression, multivariate Cox regression were conducted to design a ferroptosis-related lncRNA signature. This signature’s prognosis was veried by the log-rank test of Kaplan-Meier curve and the area under curve (AUC) of receiver operating characteristic (ROC) in train set, test set, and entire set. Furthermore, univariate and multivariate Cox regression were used to analyze its independent prognostic ability. The relationship of the ferroptosis-linked lncRNAs' expression and clinical variables was demonstrated by Wilcoxon rank-sum test and Kruskal-Wallis test. Gene set enrichment analysis (GSEA) was performed to signaling pathways it may involve.

Worldwide, lung cancer has remained the leading cause of cancer incidence as well as mortality among cancers, with 2.1 million new cases along with 1.8 million deaths estimated in 2018, which represented about 1 in 5 (18.4%) cancer deaths [1]. Among the types of lung cancer, non-small-cell lung cancer (NSCLC) comprises the most frequent, which is responsible for an estimated 85% of all the lung cancer cases [2]. Notably, lung adenocarcinoma (LUAD) constitutes the most frequent histological subtype of NSCLC, which accounts for about 40-70% cases [3,4]. Although the current treatment of LUAD has made signi cant progress, the prognosis is still very poor, with an average 5-year survival rate of 15% [5]. In clinical practice, individualized treatment has attracted mounting attention. Therefore, investigating promising prognostic signatures along with potential targets is considered as an essential phase to achieving this goal.
Ferroptosis is a type of cell death that is characterized by high production of lipid ROS (L-ROS) as a result of inactivation of cellular glutathione (GSH)-dependent antioxidant defenses. This form of cell death is iron-dependent and differs from apoptosis, classic necrosis, ferroptosis, and other forms of cell death [6,7]. Extensive studies found that ferroptosis was associated with the initiation of multiple diseases, including kidney injury, blood circulation diseases, conditions of the nervous system, and ischemiareperfusion injury [8]. Scholars have suggested that ferroptosis may be adaptive strategy used for eliminating cancerous cells and hence prevent cancer development in situations of infections, cellular stress, and nutrient de ciency [9]. Increasing studies have shown that many factors were involved in regulating ferroptosis in lung cancer. For example, some inducers include erianin [10], lncRNA-P53RRA [11], concurrent mutations of STK11 and KEAP1 [12], erastin/sorafenib [13], acetaminophen [14], Zinc [15], dihydroartemisinin [16], MT1DP (lncRNA), ginkgetin [17], inhibitors include LINC00336 [18], FSP1 [19,20], NFS1 [21], EGLN1 [22]. In addition, a study showed ferroptosis inducers may enhance the sensitivity of radiotherapy [23]. Hence, it is essential to discover ferroptosis-linked biomarkers that can be applied as valuable early diagnostic as well as prognostic indicators for LAUD.
Long non-coding RNAs (lncRNAs) is a class of non-coding RNAs with more than 200 nucleotides long that have apparently little or no protein-coding ability [24]. LncRNAs regulate critical biological functions related to growth of cells and survival, allosteric regulation of enzyme activities, chromatin modi cations, and genomic imprinting [25]. Besides, a mounting number of studies have chronicled that lncRNAs affect cancer progression and predict dismal prognosis in diverse cancer types by modulating ferroptosis. For example, p53 related lncRNA (P53RRA) promotes apoptosis and ferroptosis of cancerous cells by activating the p53 pathway [11]. LncRNA GABPB1-AS1 regulates the status of oxidative stress in context of erastin-triggered ferroptosis in HepG2 hepatocellular carcinoma cells [26]. LncRNA-linc00336 suppresses ferroptosis in lung cancer tissues by acting as a competing endogenous RNA [27]. Linc00618 accelerates ferroptosis via inhabiting vincristine (VCR) and lymphoid-speci c helicase (LSH) /SLC7A11 in leukemia [28]. In non-small cell lung cancer cells, LncRNA-MT1DP enriched on folate-modi ed liposomes promotes erastin-triggered ferroptosis by modulating the miR-365a-3p/NRF2 axis [29]. Hence, it is critical to explore the pivotal lncRNAs closely linked to ferroptosis along with prognosis in LAUD.
This study is the rst to propose a predictive model of lncRNA related to ferroptosis genes in LAUD.
Herein, we explored the expression of lncRNAs in LAUD from The Cancer Genome Atlas (TCGA) and identi ed ferroptosis-associated lncRNAs with prognostic potential. We constructed and veri ed a nine ferroptosis-correlated lncRNA biosignature with the ability to estimate the survival prognosis of LAUD patients.

Data download and processing
The transcriptome data (Cases  Table 1). Patients with no follow-up time and follow-up time shorter than 30 days were excluded from this study.

Development, veri cation, and assessment of prognostic biosignature
We utilized the R language 4.0.1version "caret" package to randomly classify the entire data set (Additional le 1) with FRlncRNAs expression pro les into two sets (train set (Additional le 2) and test set (Additional le 3)), and conducted univariate Cox regression for FRlncRNAs in the train group (P < 0.05). Lasso regression analysis was utilized to minimize over tting using the "glmnet" package [31] (P < 0.05). Afterward, multivariate Cox regression was employed to develop the optimal prognostic risk model and leveraged "coxph" and "direction = both" functions of the R language "survival" package [32] (P < 0.05). Then, the prognostic lncRNA signature's risk score constituting multiple lncRNAs was developed by summing up the product of each lncRNA with its corresponding coe cient. Additionally, the Proportional Hazards Assumption was tested in the Cox model. Similarly, on the basis of the previous training set's risk score formula, we applied it to the testing set as well as the entire set as validation.
This model was employed to explore each patient's survival prognosis by the Kaplan-Meier curve along with the log-rank test on the basis of the median of risk score, namely low-risk group and high-risk group in the train set, test set, entire set. The lncRNA signature's predictive power was explored by computing the AUC of 3 years using the ROC curve by the "survival ROC" package [33].
To further enhance the prognostic signature's credibility, we conducted a strati ed survival prognostic analysis on gender, age, clinical stage, postoperative tumor status, KRAS status, EGFR status, ALK status, ECOG score.
Independent and prognostic ability of the lncRNA signature Multivariate Cox regression and univariate Cox regression analyses were conducted to analyze the independent and prognostic ability of the lncRNA signature (Additional le 4). The clinical parameters include age, gender, clinical stage, T stage, lymph nodes as well as distant metastasis. Besides, compared with clinical variables, The ROC curve was employed to explore whether the lncRNA biosignature has better predictive power. The "rms" package was employed to construct the nomogram according to the multivariate Cox regression result (P < 0.05). To further investigate whether the ferroptosis -associated lncRNAs are involved in LAUD development, we explored the relationship of the ferroptosis-linked lncRNAs' expression with clinical variables using the Wilcoxon rank-sum test and Kruskal-Wallis test.
GSEA analysis of the lncRNA signature.
Gene set enrichment analysis (GSEA4.1.0) downloaded from https://www.gseamsigdb.org/gsea/index.jsp website was employed to identify the biological function of the prediction model [34]. Based on the median expression of lncRNA signature riskScore in 568 tumor samples, we divided them into low and high-risk groups for Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis of GSEA. The abundant signaling cascades in each phenotype were based on the normalized enrichment score (NES), the nominal (NOM) P-value as well as the false discovery rate (FDR). FDR < 25% and NOM P-value < 5% serve as a standard for inclusion.

| Statistical Analysis
R software 4.0.3 version and attached packages were employed to conduct data analyses. All the statistical analyses were two-sided. P < 0.05 signi ed of statistical signi cance.

Results
Screening of ferroptosis-related lncRNAs in LAUD.
Comparing LAUD tissues with adjacent non-LAUD tissues, 1224 differentially expressed lncRNAs were found, of which 1044 are up-regulated and 180 are down-regulated (Additional le 5). The correlation results between 259 ferroptosis-related genes and differentially expressed lncRNAs shown that there are 195 ferroptosis-related lncRNAs (FRlncRNAs) (Additional le 6).
Construction, validation, and evaluation of an nine ferroptosis-related lncRNAs prognostic signature The entire set (N = 477) with 195 FRlncRNAs expression data was randomized into the test set (N = 237) and train set (N = 240). In the univariate Cox regression assessment, 22 FRlncRNAs modulated the overall survival of the patients in the train set (Fig. 1a). Lasso regression was used for further analysis to eliminate over tting lncRNAs, and the 14 lncRNAs we obtained were used for the subsequent multivariate Cox regression analysis (Fig. 1b-c)  According to the median value of the risk score, results of the Kaplan-Meier curves demonstrate that the high-risk group has a remarkably dismal overall survival (OS) in contrast with the low-risk group in the train set (P = 8.66E-06), test set (P = 2.766E-04), and entire set (P = 7.533E-09) (Fig. 2a-c). The train set shows three years' OS for patients with high and low-risk group were 38.3% and 73.3%, respectively. The test set is 41.3% and 79.3%, respectively. The entire set is 40.9% and 78.4%, respectively. The AUC of three years dependent ROC for the seven-lncRNA biosignature achieves 0.754, 0.716, and 0.738 respectively in the train set, test set, and entire set (Fig. 2d-f), which demonstrate the good performance of the model in estimating the LAUD patients' OS. The mortality rate was higher in patients with high-risk scores relative to those with low-risk scores in the three sets ( Fig. 2g-i). The seven lnRNAs' (AC099850.3, NAALADL2-AS2, AL844908.1, AL365181.2, FAM83A-AS1, LINC01116, AL049836.1) expression of signature were lower in low-risk group compared to the high-risk group in cluster heat map, SMIM25 and C20orf197 oppositely (Fig. 2j-l).
It is worth noting that AC099850.3, FAM83A-AS1 and LINC01116's high expression of this lncRNA signature also has a worse OS than low, C20orf197 oppositely (Fig. 3). The association of the seven lncRNAs with ferroptosis genes is shown by network diagram in Fig. 4. In addition, we strati ed according to various clinical factors (gender, age, clinical stage, postoperative tumor status, KRAS status, EGFR status, ALK status, ECOG score) and applied the prognostic model to OS detection, which is shown in Fig. 5, the results shown that the signature has good predictive signi cance for LAUD patients in most strati cation factors, and part of results are not satisfactory (P > 0.05), which might be due to there are not enough samples in these strati cations.
Independent prognostic analysis of the nine ferroptosis-associated lncRNAs signature and its correlation with clinical variables.
The Univariate Cox regression assessment demonstrated that the lncRNA biosignature risk score was evidently correlated with the patients' OS (hazard ratio HR = 1.003, con dence interval 95%CI = 1.001-1.006, P = 0.009) ( Table 2). Moreover, the multivariate Cox regression analysis demonstrated that the lncRNA biosignature risk score remained independent with OS considering other conventional clinical factors including Lymph-node status, the clinical stage, distant metastasis, and T stage (HR = 1.004, 95% CI = 1.002-1.007, P = 0.001). Meanwhile, clinical stage was demonstrated as an independent prognostic index. Compared to clinical variables, this signature risk score's ROC curves of three years demonstrate the largest AUC value (0.737) (Fig. 6).
Functional enrichment analysis of the nine ferroptosis-related lncRNAs signature.
GSEA analysis is used to discover potential biological functions of the nine ferroptosis-associated lncRNAs signature of LAUD ( Fig. 8 and Table 3

Discussion
Lung cancer is one of the leading causes of cancer-related death globally, while LAUD ranks rst in the proportion of lung cancer subtypes [35]. Although the current treatment methods have made great advancements, the prognosis is still very poor. Ferroptosis is differs from other types of cell death in terms of biochemically and morphologically and has been shown to regulate cancer development [6]. More and more reports have documented that lncRNA plays a very important role in regulating gene expression and regulation in tumor [25,36]. In addition, many lncRNAs in uence the progression of LAUD by regulating ferroptosis. However, there are no reports on that prognostic model of lncRNA related to ferroptosis was constructed. Although two previous genetic prognostic models of ferroptosis have been reported in hepatocellular carcinoma [37] and glioma [38], our study is the rst to report the study of ferroptosis-related lncRNA prognostic models in LAUD In the present study, we downloaded ferroptosis genes from FerrDb, and used the R language and its attached packages to nd differentially expressed lncRNAs related to ferroptosis (FRlncRNAs). We randomly grouped all the patients into train set as well as the test set, then a nine ferroptosis-related lncRNAs signature model (AC099850.3, NAALADL2-AS2, AL844908.1, AL365181.2, FAM83A-AS1, LINC01116, AL049836.1, SMIM25 and C20orf197) was established through univariate Cox regression, Lasso regression, as well as multivariate Cox regression in the train set. At the same time, the biosignature was veri ed in the test set as well as the entire set. On the basis of the median risk score, the Kaplan-Meier curves revealed that the high-risk group had an evidently dismal overall survival relative to the low-risk group in the three data sets and various clinical strati cation factors. Assessment of the biosignature for OS in the three sets by ROC curve exhibited well predictive value. The Univariate Cox  [45]. The examples we have cited are only the tip of the iceberg, and the relationship between lncRNA, ferroptosis, and LAUD has also been well demonstrated in this study from a new perspective.
Among these lncRNAs of the signature, some studies have shown that AC099850.3 is also used as an autophagy-related lncRNA signature model in hepatocellular carcinoma as well as oral and oropharyngeal squamous cell carcinoma [46,47]. Benoist GE et al revealed that patients with NAALADL2-AS2 high-expression showed a longer time to progression [48].
Xiao G et al. discovered FAM83A-AS1 promoted LAUD cell migration as well as invasion via targeting miR-150-5p as well as modifying MMP14 [49], Shi R et al. found FAM83A-AS1 facilitated LUAD proliferation and invasion by increasing FAM83A expression [50]. He J et al. revealed long noncoding RNA FAM83A-AS1 promotes the progression of hepatocellular carcinoma by binding with NOP58 to promote the mRNA stability of FAM83A [51]. Huang GM et al. re ected lncRNA FAM83A-AS1 aggravates the malignant development of esophageal cancer by binding to miR-495-3p [52]. LINC01116 has been studied as an oncogene in many tumors, such as LAUD, its overexpression promotes LAUD proliferation and metastasis [53], contributes to ge tinib resistance in NSCLS through regulating IFI44 [54], results in resistance of LAUD to cisplatin via the EMT process [55]. Leng X et al. indicated that SMIM25 (Aliases LINC01272) promoted gastric cancer metastasis through regulating EMT process [56]. The remaining lncRNAs have not seen relevant reports in previous studies, which are worthy of further research.
Our current study also has some limitations. First, we use the data in the TCGA database as the starting point for research; although the model has been internally veri ed, it is still needed for further veri cation in external data; second, TCGA's race is mainly white (75%), and whether the model ts other race needs further veri cation. Third, the analysis of the lncRNA expression of the model and the KEGG function enrichment analysis by the GSEA model requires further cell function experimental analysis.

Declarations
Xiwen Tong downloaded the lncRNA and mRNA expression information, Xiwen Tong and Guodong Yang constructed lncRNA signature model and performed the statistical analysis using R language software, and wrote the rst draft of the manuscript. Guanghui Yi and Yujiao Zhang revised the manuscript. Guanghui Yi contributed conception and design of the study and checked the manuscript.

Funding
The present study was supported by the Huanggang Municipal General Project of China (grant no. XQYF2020000016).

Ethics approval and consent to participate
LncRNA and mRNA sequencing pro les were obtained from the TCGA data portal, which is a publicly available dataset. Therefore, no ethics approval is needed.

Consent for publication
All listed authors have actively participated in the study and approved the submitted manuscript.