Ferroptosis-Related Long Non-coding RNA Predicts Prognosis Model of Hepatocellular Carcinoma

Background: Hepatocellular carcinoma (HCC) is the most common malignancy globally, and ferroptosis is an iron-dependent cell death process. Furthermore, aberrant expression of long non-coding RNAs (lncRNAs) driving HCC development and progression has increased attention. Materials and Methods: We collected lncRNA expression proles associated with ferroptosis from The Cancer Genome Atlas (TCGA) and FerrDb databases and clinicopathological and overall survival (OS) information to determine the association between ferroptosis-related lncRNAs(FRlncRNAs) and survival of HCC patients by co-expression analysis. A prognostic lncRNA model of 22 differentially expressed lncRNAs was constructed using Cox regression analysis and the LASSO algorithm. Kaplan-Meier analysis revealed that a high-risk lncRNAs prole was associated with poor prognosis in HCC. Our risk assessment model outperformed conventional clinical data in predicting the prognosis of HCC. Result: GSEA revealed immune and tumor-related pathways in individuals in the high- and low-risk groups. In addition, TCGA showed that T cell functions, including B cells, Cytolytic, macrophages, MHC-class-I, mast cells, neutrophils, NK cells, helper T cells, Type-I-IFN, and Type-II-IFN, were signicantly different between high and low-risk groups. Immune checkpoints such as TNFSF18, IDO2, CD276, NRP1, and TNFSF4 were also differentially expressed between the two risk groups. Conclusions: Our ndings provide a robust prognostic and immune response prediction model for HCC patients based on lncRNAs associated with ferroptosis. for the treatment of tumors. This study rst identied novel models of prognostic lncRNAs associated with ferroptosis based on the TCGA dataset. We then explored the role of immune inltrating cells in the tumor microenvironment and immune checkpoint inhibitors in the prognosis of HCC. The results of this study identied potential biomarkers and therapeutic targets in the ferroptosis pathway.


Introduction
HCC is the most common pathological type of primary liver cancer and is the fth most common malignancy worldwide, ranking third in cancer-related mortality (1). HCC is a complex process caused by multiple factors, including chronic hepatitis virus, heavy alcohol consumption, and chronic hepatitis due to nonalcoholic fatty cirrhosis (2). Patients with early-stage HCC can have surgical resection or liver transplantation to effectively control cancer progression and prolong survival. However, more than twothirds of patients will experience recurrence (3), and many patients are often diagnosed at an intermediate to the advanced stage. Most patients have minor surgical resection due to age or physical status limitations. Interventional therapy, radiotherapy, targeted therapy, and local ablation have limited effect, and poor prognosis remains (4). Therefore, it is crucial to explore the potential molecular mechanisms and cellular signaling pathways in the pathogenesis of HCC, to seek early diagnosis and early treatment, as well as to study the expressed genes with prognostic value and to construct a model of predictive characteristics.
In recent years, Ferroptosis has grown by leaps and bounds in research. It is an iron oxidation-dependent and lipid peroxidation-mediated programmed cell death by oxidatively reactive oxygen species (ROS) (5).
The importance of the role of ferroptosis has been demonstrated in the regulation of metabolism and redox biology, affecting the pathogenesis and treatment of cancers, including prostate cancer, gastric cancer, and HCC. Recently, ferroptosis induction has been a promising therapeutic strategy for malignancies resistant to conventional therapy (6, 7). Zhang et al found that tumor suppressor (BAP1) inhibited cystine uptake by suppressing SLC7A11 expression, leading to elevated lipid peroxidation and ferroptosis (8). Sun et al found that p62 expression blocked nuclear factor degradation of erythroid 2related factor (NRF2) and enhanced NRF2 nuclear accumulation through mis re of kelch-like ECHassociated protein 1 (9). Liu et al reported ferroptosis and immune correlation and found that this prognostic feature could be used to screen HCC patients for immunotherapy and targeted therapies (10).
Long non-coding RNAs (LncRNAs) are non-coding transcripts more signi cant than 200 nucleotides in length that can modulate the expression of various cancer-related genes. Recently, Sun et al found that in HCC cells, high levels of lncRNA GA-binding protein subunit (GABPB1) antisense RNA 1 enhanced erastininduced ferroptosis by blocking translation of the GABPB1 and inhibiting peroxidase-5, leading to inhibition of cellular antioxidant capacity and cell survival (11). Zhang et al analyzed the relationship between ferroptosis and tumor mutations in HCC to construct a ferroptosis-related gene model that may bridge the gap between ferroptosis nuclear tumor mutations, which offers the possibility of individualized treatment for HCC patients (12). This study constructed a model of prognosis-related lncRNAs based on the TCGA database. The role of ferroptosis-related mRNAs, N6-methyladenosine (m6A) mRNA, and immune response in HCC prognosis was explored (13).

Enrichment analysis of ferroptosis-related genes
We obtained a total of 84 DEGs associated with ferroptosis (13 down-regulated and 71 upregulated; Additional Table 3).GO enrichment analysis revealed that BP participated in cellular response to chemical stress, oxidative stress, cellular response to oxidative stress. MF mainly regulated the production of the apical part of cells, focal adhesion, melanosome, pigment granule. CC was mainly upregulated in oxidoreductase activity, acting on NAD(P)H, organic anion transmembrane transporter activity, antioxidant activity. KEGG pathway based analysis revealed the over-expressed genes were mainly involved in Pathways of neurodegeneration−multiple diseases, Chemical carcinogenesis−reactive oxygen species, MicroRNAs in cancer, Lipid and atherosclerosis, Central carbon metabolism in cancer, Serotonergic synapse, Fluid shear stress, and atherosclerosis, Ferroptosis (Fig. 1, Fig. 2 and Additional Table 4).

The ferroptosis-based lncRNAs prognostic signature
We identi ed 1072 lncRNAs associated with ferroptosis (Additional Table 5 Table 6). Therefore, we calculated risk scores and constructed prognostic signatures for lncRNAs.

Prognostic characteristics and prognostic value of the model
Kaplan-Meier analysis veri ed that patients with TCGA-LIHC in the high-risk group had signi cantly lower survival times at 1, 3, and 5 years than those in the low-risk group (P < 0.001, Fig. 3A). Moreover, the timedependent ROC curve for survival prediction of the risk score model had an AUC of 0.811 at 1 year, which was more speci c and sensitive in predicting the prognosis of HCC than standard clinicopathological features (Fig. 3B, 3E). In addition, time-dependent ROC analysis showed that the AUC of the risk score model was 0.811 at 1 year, 0.752 at 3 years, and 0.692 at 5 years (Fig. 3D). Prognostic curves and scatter plots were used to represent the risk score and survival status of each HCC patient, and using patient risk survival status plots, we found that patient risk score was negatively correlated with survival of HCC patients and that most deaths were present in the high-risk group (Fig. 3C).
Next, univariate Cox analysis showed that lncRNAs-based characteristics (HR: 1.106, 95% CI: 1.081-1.132) and tumor stage (HR: 1.680, 95% CI: 1.369-2.062) were independent prognostic factors in HCC patients (Fig. 4A). Multivariate Cox analysis showed that lncRNAs characteristics (HR: 1.098, 95% CI: 1.072-1.125) and tumor stage (HR: 1.585, 95% CI: 1.282-1.958) were also independent prognostic risk factors for HCC patients (Fig. 4B). the relationship between lncRNAs and mRNAs is shown in Figure 5. We also analyzed the association heatmap between prognostic features and clinicopathological manifestations of FRlncRNAs (Fig. 6). In addition, the predictors in the nomogram included the novel risk score model and clinicopathological features were stable and accurate (Fig. 7). In this combined nomogram, the risk score model proved to play the most excellent role in these clinically relevant variables and thus can be applied with the clinical prognostic assessment of HCC patients. These studies demonstrate that this novel model of lncRNAs associated with ferroptosis can be reliably used as an independent prognostic factor for HCC patients.
GSEA enrichment analysis of risk scores GSEA showed that most of the prognostic features of lncRNAs associated with ferroptosis regulate immune and tumor-related pathways such as ubiquitin-mediated proteolysis, homologous recombination, RIG-1-like receptor signaling pathway, colorectal cancer, FC gamma R-mediated phagocytosis, TGF-β signaling pathway, Notch signaling pathway, T cell receptor signaling pathway, regulation of autophagy, Toll receptor signaling pathway, Aminoacyl tRNA biosynthesis, and RNA degradation ( Fig. 8and Additional Table ).

Immunoassay and Gene Expression
To determine whether the model is related to tumor immunity, we evaluated the risk score of HCC based on TIMER, CIBERSORT, CIBERSORT-ASS, QUANTISEQ, MCPCOUNTER, XCELL, and EPIC algorithms with an immunoreactive heatmap of tumor-in ltrating immune cells. The heatmap showed (Fig. 9), B cells, T cells, neutrophils granulocytes, macrophages, and medullary dendritic cells, including regulatory T cells, memory T cells, M0 macrophages, M2 macrophages, CD8+ T cells, lymphocytes, and CD4+ Th2 cells were signi cantly different between the high-and low-risk groups. Correlation analysis of immune cell subpopulations and related functions in ssGSEA based on the TCGA-LIHC database showed that immune cells including B cells, Cytolytic, macrophages, MHC-class-I, mast cells, neutrophils, NK cells, helper T cells, Type-I-IFN, and Type-II-IFN were signi cantly different between high and low-risk groups were evident between them (Fig. 10). Given the importance of immune checkpoint blockade based therapeutic strategies in HCC, we further explored the differences in immune checkpoint expression between the high and low-risk groups, and we found signi cant differences in the expression of TNFSF18, IDO2, CD276, NRP1, and TNFSF4 between the two groups of patients (Fig. 11). The comparison of m6A-related mRNA expression between those two groups showed RBM15, HNRNPC, YTHDC1, YTHDF1, WTAP, METTL3, ALKBH5, YTHDF2, and FTO were signi cant (Fig. 12).

Discussion
In previous studies, models of lncRNAs for predicting prognosis have been validated in many cancers, such as gastric cancer(23) and head and neck squamous cell carcinoma. Ferroptosis can overcome the resistance of malignant cells to chemotherapy and promote the clearance of defective cells. Therefore, it can be a new approach for the treatment of tumors. This study rst identi ed novel models of prognostic lncRNAs associated with ferroptosis based on the TCGA dataset. We then explored the role of immune in ltrating cells in the tumor microenvironment and immune checkpoint inhibitors in the prognosis of HCC. The results of this study identi ed potential biomarkers and therapeutic targets in the ferroptosis pathway.
Overall, our analysis identi ed 84 DEGs associated with death by iron. KEGG analysis further revealed that these genes are mainly involved in Pathways of neurodegeneration-multiple diseases, Chemical carcinogenesis-reactive oxygen species, MicroRNAs in cancer, Lipid and atherosclerosis, Central carbon metabolism in cancer, Serotonergic synapse, and Fluid shear stress and atherosclerosis, Ferroptosis. A recent study found that the overactivation of PI3K-AKT-mTOR signaling mutations protected cancer cells from oxidative stress and ferroptosis through SREBP1/SCD1-mediated lipid formation (24). Sun et al reported that Fin56 is a type 3 ferroptosis inducer and Torin2 is a potent mTOR inhibitor used to activate autophagy, which has synergistic effects on the cytotoxicity of bladder cancer cells (25). It was recently found that ischemia-reperfusion (I/R) induced upregulation of miR182-5p and miR378a-3p led to ferroptosis activation in kidney injury through downregulation of GPX4 and SLC7A11(26). Metformin was found to induce ferroptosis through upregulation of miR324-3p in a mouse breast cancer's xenograft model (27). As well as isoglochidonine induced ferroptosis through miR122-5p/TP53/SLC7A11 pathway protects neuronal cells from brain hemorrhage induced ferroptosis (28). In brief, in this study, 22 differentially expressed lncRNAs were identi ed as independent prognostic factors in HCC. Recent studies have revealed that KDM4A-AS1 promotes deubiquitination of AR and AR splice variants (AR/AR-Vs) to protect them from MDM2-mediated degradation of ubiquitin protectors. In addition, KDM4A-AS1 enhanced enzalutamide resistance in desmoplastic prostate cancer by inhibiting AR/ AR-Vs degradation and antisense oligonucleotide drugs targeting KDM4A-AS1 signi cantly reduced enzalutamide-resistant tumor growth (29). LINC00205, a novel transcriptional gene Yin Yang-1 (YY1), regulates lncRNA that can accelerate the proliferation of HCC cells by sponging miR26a-5p to promote the expression of CDK6 (30). In gastric cancer, LIN01224 sponges miR193a-5p and targets upregulation of CDK8 to accelerate the malignant transformation of gastric cancer (31). Knockdown of LUCAT1 promoted miR375 expression in tongue squamous cell carcinoma (TSCC) cells, and low expression of mir-374 was associated with poor prognosis in TSCC; thus, LUCAT1 promoted TSCC cell proliferation, cell cycle, and migration by targeting miR375 (32). MIR210HG recruits DNA methyltransferase 1, which promotes methylation of the CACNA2D2 promoter region, and overexpression of CACNA2D2 signi cantly inhibited non-small-cell lung cancer cell proliferation (33). LncRNA MIR44435-2HG binds to and inhibits the decapentaplegic protein (DSP), leading to activation of WNT/β-catenin signaling and a cascade of epithelial-mesenchymal transition in gastric cancer cells (34). Overexpression of LncRNA muskelin1 antisense RNA (MKLN1-AS) enhances the stability of Yes-associated transcriptional regulator 1 (YAP1) and enhances proliferation, metastasis, and invasion of hepatocellular carcinoma cells (35). lncRNA NRAV promotes respiratory syncytial virus production by sponging miR-509-3p pro le (36). Silencing LncRNA PRRT3-AS1 activates peroxisome proliferator-activated receptor γ (PPARγ), thereby blocking the mTOR signaling pathway to inhibit prostate cancer cell proliferation and promote apoptosis and autophagy (37). LncRNA SNHG4 suppresses METTL-3-mediated transcriptional activator 2 (SATA2) mRNA at m6A levels by promoting LPS-induced in ammation in human lung broblasts and mouse lung tissue in vitro and in vivo (38). LncRNA SNHG12 promotes colon carcinogenesis and progression by regulating the miR-15a/PDK4 axis (39). STAT1-inducible lncRNA ZFPM2 antisense RNA1 (ZFPM2-AS1) promotes the development of colon cancer by regulating the target of miR-653 gene GOLM1 to reverse the inhibitory effect of miR-653 on the proliferation and metastasis of HCC cells (40). However, there are a few studies on the role of FRlncRNAs in HCC prognosis. The results of our study may provide valuable perspectives for future cancer control.
This study explored ferroptosis biomarkers that help predict the prognosis of HCC, which could provide a reliable reference for the treatment of the disease. However, our model is mainly based on bioinformatics studies, still lacks validated experimental validation of these indicators. Therefore, further validation using different cohorts is necessary. Given that our ndings have not been validated using clinical samples, the reliability of our results cannot be fully guaranteed. Overall, the prognostic indicators established by the model need further validation.

Conclusion
In conclusion, we constructed a model of FRlncRNAs predicting prognosis and immune response in HCC patients, which was strongly associated with a risk score, survival time, and clinical data of cancer. Thus, the ndings suggest that our model of FRlncRNAs provides a personalized, predictive tool for prognosis and immune response in HCC patients

Data sources and clinical information
RNA-sequence data from 424 samples and associated clinical data were extracted from the TCGA database, including 374 HCC tissues and 50 normal liver tissues. An overview of relevant clinical aspects of HCC patients can be found in Supplementary Additional Table1, with a total of 377 clinical data available for further analysis (Table1). Clinical data collected for HCC patients included gender, age, grade, TNM stage, survival status, and survival time.

Identi cation of FRlncRNAs and enrichment analysis
The corresponding ferroptosis-related genes were downloaded from FerrDb (14), a web-based consortium providing a comprehensive and up-to-date database of ferroptosis markers, regulatory molecules, and associated diseases. Ultimately, we identi ed 382 ferroptosis-related genes (Additional Table 2), of which Diver: 150, Suppressor: 109, and Marker: 123. These associated genes were subjected to Pearson's test to check the relationship between FRlncRNAs and HCC. The association was considered signi cant if the correlation coe cient |R2| > 0.5 and P < 0.001. First, we explored the functions of up-and down-regulated ferroptosis-related differentially expressed genes (DEGs). We used Kyoto Encyclopedia of Genes and Genes (KEGG) data to assess the biological pathways associated with DEGs. The functions of biological processes (BP), molecular functions (MF), and cellular components (CC) regulated by differentially expressed FRlncRNAs were further analyzed according to gene ontology (GO) using the cluster pro le, GOplot package in R language. Table 1 The clinical characteristics of patients in the TCGA database

Construction and validation of a model of FRlncRNAs
FRlncRNAs with prognostic value were screened using Lasso-penalized Cox regression analysis. 376 lncRNA-seq samples remained in the nal cohort for analysis by excluding those with unknown survival times (n=1) based on the clinical data obtained above. A genetic model containing biomarkers useful for predicting prognosis was identi ed using the "glmnet" package in R, and a risk score was calculated for each sample in all data sets based on this model. Based on the risk score = alncRNA1 × lncRNA1 expression + blncRNA2 × lncRNA2 expression +⋯ + nlncRNAn × lncRNAn expression. To assess the model's predictive power for prognostic risk, we operated characteristic curve (ROC) for 10-year survival were analyzed using the "timeROC" in R language. The RNA was divided into low-and high-risk groups based on median scores. The prognostic signi cance of this model for HCC was explored using Kaplan-Meier analysis. Finally, univariate and multivariate Cox regression analyses were performed to assess whether the model showed good predictive power independent of other clinicopathological characteristics.
The predictive nomogram The "regplot" package in R was used to collate clinical data and construct a nomogram integrating prognostic features to predict the 1-, 3-, and 5-year OS of HCC patients.

Function Enrichment Analysis
Potential functional pathways involved in the model of ferroptosis lncRNAs were de ned based on gene set enrichment analysis (GSEA). GSEA was performed on the KEGG dataset c2.cp.kegg.v7.4.symbols.gmt in java GSEA4.1.0 based on the optimal cut-off value, dividing the TCGA data into high-risk and low-risk groups to identify enrichment pathways between the high-risk and low-risk groups. Statistical signi cance was set at P < 0.05, and false discovery rate (FDR) q < 0.25 was considered statistically signi cant.
Immunity in ltrates analysis and gene expression Also, the TIMER(15), CIBERSORT (16), CIBERSORT-ABS (17), QUANTISEQ(18), MCPCOUNTER (19), XCELL (20), and EPIC (21) algorithms were compared to assess FRlncRNAs models between high-and low-risk groups based on cellular components and cellular immune responses. Heatmaps represented the differences in immune responses under different algorithms. Also, single-sample gene set enrichment analysis (ssGSEA) was used to quantify tumor-in ltrating immune cell subpopulations between the two groups and to assess their immune function.

Statistical Analysis
All statistical analyses were performed using the R Language and its corresponding software packages.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. AdditionalTable.xlsx