Identication of cross-talk between m6A/m5C regulators and ferroptosis associated with immune inltration and prognosis in pan-cancer

Although it has been recognized that m6A/5mC methylation and ferroptosis play critical roles in different types of cancers, little is known about the relationship between them. In addition, there is also a growing appreciation that m6A/5mC methylation and ferroptosis may affect immune cell inltration and activation. This study aimed to reveal the extensive cross-talk between epigenetic modication and ferroptosis. A total of 31 cancer type-specic datasets in TCGA were individually collected by the publicly available web servers for multiple bioinformatic analyses of m6A/5mC regulators and ferroptosis-related genes. Intriguingly, m6A/5mC regulators and ferroptosis-related genes were identied to have considerable global coverage and prognostic signicance across multiple cancer types. Moreover, m6A/5mC regulators showed interactive potential with ferroptosis-related genes, and genomic alteration of ferroptosis-related genes coupled with m6A/5mC regulators, at least in pancreatic cancer. Furthermore, m6A/5mC regulators and ferroptosis-related genes were found to be signicantly associated with TILs. Finally, m6A/5mC regulators and ferroptosis-related genes exhibited functionally related to each other or co-regulated by TF or non-coding RNA. Together, m6A/5mC methylation and ferroptosis show a wide-ranging connection, and a combination strategy of epigenetic and ferroptosis therapies with ICP inhibitors may benet more cancer patients in the future.


Introduction
The most reproducibly mapped and well-known nucleotide methylation, notably in the forms of 5methylcytosine (5mC) in DNA and N6-methyladenosine (m6A) in mRNA, is of great functional signi cance to the fundamental biological systems, especially in the regulation of gene expression [1].
Recent research advances emphasize the biological importance of m6A methylation as a highly dynamic and readily reversible post-transcriptional modi cation in nuclear pre-mRNA splicing, nuclear export, microRNA processing, translation initiation, and RNA stability [2]. 5mC DNA methylation, a conserved and abundant epigenetic modi cation, plays broad and critical roles in various biological processes, including gene expression regulation, cellular differentiation, and stress responses [3]. A recent study has established a cross-talk between 5mC DNA methylation and m6A mRNA methylation in pan-cancer [4].
Intriguingly, m6A and 5mC regulators exhibited comparable levels of mutation frequency, signi cant cooccurrences of genetic alterations, similar gene expression patterns, and functionally related to each other or co-regulated. Ferroptosis is a recently described type of programmed cell death driven by iron-dependent lipid peroxidation that differs from other forms of cell death, such as apoptosis and necrosis, in morphology and mechanisms [5]. Multiple productive lines of studies have suggested that ferroptosis plays a pivotal role in the development and progression of cancer [6][7][8] . For example, stearoyl-CoA desaturase-1 (SCD1) in cancer cells and fatty acid-binding protein 4 (FABP4) in tumor microenvironment cooperatively protect from oxidative stress-induced ferroptosis and promote tumor recurrence [9] . Ferroptosis can also play an essential role in suppressing tumor metastasis as melanoma cells from the lymph are more resistant to ferroptosis and, thus, can form more metastases than those in the blood [10] . Most recently, it was reported that a therapy-resistant, high-mesenchymal state depends on the glutathione peroxidase 4 (GPX4) pathway to evade ferroptosis, suggesting that induction of ferroptosis represents an emerging strategy for innovative anti-tumor drug discovery [11] . The role of ferroptosis in cancer may also involve multiple cell types within the tumor microenvironment. Immunotherapy activated CD8 + T cells could downregulate the expression of two subunits of glutamate-cystine antiporter system Xc−, solute carrier family 7 member 11 (SLC7A11) and SLC3A2 via releasing interferon-gamma (IFNgamma), resulting in the accumulation of lipid peroxidation and ferroptosis in cancer cells [12].
Emerging evidence suggests that m6A/5mC methylation plays a vital role in regulating ferroptosis. Exosomal miR-4443 facilitates cisplatin resistance in non-small cell lung carcinoma by modulating FSP1 m6A modi cation-mediated ferroptosis via METTL3 [13] . YTHDC2 inhibits SLC7A11 and SLC3A2 in an m6A-dependent manner and is believed to be an endogenous ferroptosis inducer in lung adenocarcinoma [14,15] . However, due to the limitations in methodology, these studies have been limited to only one or two epigenetic regulators. To the best of our knowledge, the "cross-talk" between epigenetic modi cations and ferroptosis in pan-cancer has not been systematically investigated. Therefore, a comprehensive understanding of the extensive cross-talk between epigenetic regulators and ferroptosis will contribute to our understanding of epigenetic regulation in ferroptosis and the development of potential therapeutic strategies for controlling cell death and survival by mediating the reversibility of ferroptosis.
This study explored the expression levels and mutation frequencies of epigenetic regulators and ferroptosis-related genes in pan-cancer samples from The Cancer Genome Atlas (TCGA) cohort. Next, we identi ed the potential relationship between epigenetic regulators and ferroptosis-related genes from several aspects, including co-expression, functional states at single-cell resolution, immune cell in ltration, transcription factors (TFs), and ceRNA regulation. This study provides essential insights into the interaction between epigenetic regulators and ferroptosis-related genes in cancer and paves new ways for related therapeutic targets.

The expression and prognostic analysis by GEPIA across different cancer types
The Gene Expression Pro ling Interaction Analysis (GEPIA, http://gepia2.cancer-pku.cn, version 2) is an open-access web-based tool for rapid and customizable exploration of RNA sequencing expression data of 9736 tumors and 8587 normal samples derived from the TCGA and the Genotype-Tissue Expression (GTEx) projects [16] . In this study, GEPIA was utilized to calculate the differential expression and prognostic indexes of m6A/5mC regualators and ferroptosis-related genes. One-way ANOVA was used to identify the differential expression of m6A/5mC regulators and ferroptosis-related genes with |log2FC| values > 1 and q values < 0.01. Overall survival (OS) or disease-free survival (DFS, also called relapse-free survival (RFS)) analyses of m6A/5mC regulators and ferroptosis-related genes were assessed using the log-rank test for Kaplan-Meier methods with a 50% (Median) cut-off for both low and high expression groups. A univariate Cox proportional hazards regression model was adopted to calculate the hazard ratio, and p-value < 0.05 was used as a threshold in ranking the results.

Construction of protein-protein interactions (PPIs) by STRING
STRING (Search Tool for the Retrieval of Interacting Genes/Proteins, https://string-db.org, version 11.0) was used to construct PPIs between m6A/5mC regulators and ferroptosis-related genes. This well-known online interaction repository includes direct physical interactions and indirect functional associations [17] . Combined scores were computed by combining the probabilities from numerous sources, including highthroughput experimental data, mining of databases and literature, and predictions based on genomic context analysis. The con dence score ranked from 0 to 1, with 1 indicating the highest possible con dence. Pearson correlation between m6A/5mC regulators and ferroptosis-related genes was carried out using the package corrplot in R. The circos plot was generated through the circlize package. . The data of gene-level is stored with available clinical information, including OS, progression-free survival (PFS), DFS, and disease-speci c survival (DSS). In this study, CVCDAP was used to visualize and compare genetic alterations of m6A/5mC regulators and ferroptosis-related genes in pan-cancer [19] . To determine whether alterations in m6A/5mC regulators and ferroptosis-related genes affected the OS and PFS of pancreatic cancer patients, we used the cBioPortal to evaluate the survival data of patients with or without genetic m6A/5mC regulators and ferroptosis-related genes alterations from four pancreatic cancer studies. Co-occurrence and mutual exclusivity of genetic alterations between inquired m6A/5mC regulators and ferroptosis-related genes were determined by log2 odds ratio, p-value, and q value, and results with q value < 0.05 were selected. In pancreatic cancer, OS and PFS were individually investigated to compare the prognostic differences between altered and unaltered groups. The Log-rank test was used for hypothesis testing.

Functional states analysis at single-cell resolution by CancerSEA
The functional states of m6A/5mC regulators and ferroptosis-related genes in various cancer types were analyzed by CancerSEA (http://biocc.hrbmu.edu.cn/CancerSEA/). CancerSEA is the rst integrative database aimed to decode different functional states of cancer cells at a single-cell resolution, which covers 14 functional states (including stemness, invasion, metastasis, proliferation, EMT, angiogenesis, apoptosis, cell cycle, differentiation, DNA damage, DNA repair, hypoxia, in ammation, and quiescence) of 41,900 cancer single cells from 25 cancer types [20] . Correlations between the gene of interest and functional state in different single-cell datasets were ltered by a correlation strength > 0.3 and a false discovery rate (FDR) (Benjamini & Hochberg) < 0.05.

TIMER database analysis
Tumor Immune Estimation Resource 2.0 (TIMER2.0) is an integrative resource for comprehensively evaluating tumor in ltration of immune cells across various cancer types (http://timer.cistrome.org/) [21] . TIMER2.0 utilizes a deconvolution statistical method to estimate the abundance of 36 subtypes of tumor-in ltrating immune cells, including B cell, B cell memory, B cell naive, B cell plasma, cancer associated broblast, class-switched memory B cell, common lymphoid progenitor, common myeloid progenitor, endothelial cell, eosinophil, granulocyte-monocyte progenitor, hematopoietic stem cell, macrophage, macrophage M1, macrophage M2, mast cell, monocyte, myeloid dendritic cell, myeloid dendritic cell activated, neutrophil, NK cell, plasmacytoid dendritic cell, T cell CD4 + (non-regulatory), T cell CD4 + central memory, T cell CD4 + effector memory, T cell CD4 + memory, T cell CD4 + naive, T cell CD4 + Th1, T cell CD4 + Th2, T cell CD8 + , T cell CD8 + central memory, T cell CD8 + effector memory, T cell CD8 + naive, T cell gamma delta, T cell NK and T cell regulatory (Tregs). We analyzed the correlation between m6A/5mC regulators and ferroptosis-related genes expression and the abundance of immune in ltrates using the gene module. The gene expression level was displayed with log2 RSEM.

TF regulatory network
The hTFtarget database (http://bioinfo.life.hust.edu.cn/hTFtarget) has integrated TF-target regulations and epigenetic modi cation information from large-scale of ChIP-Seq data of human TFs (7,190 experiment samples of 659 TFs) in 569 conditions (399 types of cell line, 129 classes of tissues or cells, and 141 kinds of treatments) to predict accurate TF-target regulations [22]. The Cistrome Data Browser (DB, http://cistrome.org/db) maps the genome-wide locations of transcription factor binding sites, histone post-translational modi cations, and regions of chromatin accessible to endonuclease activity, which contains approximately 47,000 human and mouse samples [23]. The hTFtarget and Cistrome database were used to predict TFs of m6A/5mC regulators and ferroptosis-related genes. Pearson correlation among TF, m6A/5mC regulators and ferroptosis-related genes was carried out using the package corrplot in R. The circos plot was generated through the circlize package.

Results
3.1. The expression and prognostic landscape of m6A/5mC regulators and ferroptosis-related genes across different cancer types We rst examined the expression pro les of 41 m6A/5mC regulators and 60 ferroptosis-related genes in multiple cancers by analyzing GEPIA (Supplement Table 1A-B). Pan-cancer expression analysis results indicated signi cant deregulation of m6A/5mC regulators and ferroptosis-related genes in the majority of malignancies ( Fig. 1A-B). The expression of m6A/5mC regulators and ferroptosis-related genes have a signi cant degree of heterogeneity across cancers. THYM, PAAD, DLBC and CHOL showed the most signi cant changes in the expression of m6A/5mC regulators and ferroptosis-related genes. Surprisingly and contrary to expectations, the expression of m6A and 5mC regulators showed no signi cant difference in KIRP.

Interaction between m6A/5mC regulators and ferroptosis-related genes
In view of the critical role of both m6A/5mC regulators and ferroptosis-related genes in the prognosis of cancer patients, the logical next step seems to investigate whether there is a potential interplay between m6A/5mC regulators and ferroptosis-related genes. Network analysis was carried out using PPI data from STRING database, a robust functional association between m6A/5mC regulators and ferroptosisrelated genes was observed (Fig. 4A). Since almost all m6A/5mC regulators and ferroptosis-related genes were unregulated in PAAD, pancreatic adenocarcinoma was chosen for further study. Importantly, our analysis based on TCGA data depicted a strong correlation between the expression of m6A/5mC regulators and ferroptosis-related genes in PAAD (Fig. 4B).
The global genomic alteration landscape of m6A/5mC regulators and ferroptosis-related genes in pancancer was analyzed using CVCDAP online analyzing tool (Fig. 5A-B). The detailed correlation between each m6A/5mC regulators and ferroptosis-related genes was individually analyzed in PAAD, and statistically signi cant relationship was presented in Supplementary Table 2. For example, SQLE was found associated with KIAA1429 where they shared 24 variants in 860 patient samples. The genomic alterations of m6A/5mC regulators showed general co-occurrence rather than mutual exclusivity with ferroptosis-related genes. In fact, a total of 1654 signi cant associations between two genes among m6A/5mC regulators and ferroptosis-related genes were observed in this analysis, all of which showed co-occurrence but not mutual exclusivity. Furthermore, integrated prognostic analyses of OS and PFS indicated that integrated genomic alterations of m6A/5mC regulators and ferroptosis-related genes were signi cantly unfavorable for multiple prognoses of patients with pancreatic cancer (Fig. 5C-F). 3.3. Relationship between m6A/5mC regulators and ferroptosis-related genes with cancer single-cell functional states Heterogeneity associated with different functional phenotypes of tumor cells has been one of the major challenges for cancer diagnosis and effective cancer treatment. Recent advances in single-cell sequencing (scRNA-seq) technology have provided a tool for dissecting cellular heterogeneity, unraveling cell status and identifying subpopulation structures across different cell types at the cellular level. Functional correlation analysis of cancerSEA showed that the functional phenotypes of m6A/5mC regulators and ferroptosis-related genes were positively correlated with DNA damage and DNA repair (r > 0.3, p < 0.05) (Fig. 6A-B), which suggested that they may participate in the same processes of tumorigenesis and advancement of most types of cancer (Fig. 6C-D). Moreover, we also found that m6A/5mC regulators and ferroptosis-related genes were positively correlated with invasion in CML, CRC, OV, and PC.

Associations between m6A/5mC regulators, ferroptosis-related genes and cancer immunity
The associations between m6A/5mC regulators, ferroptosis-related genes and tumor microenvironment were investigated in the 36 cell types via Tumor IMmune Estimation Resource (TIMMER) for each cancer type. Pan-cancer analysis results indicated signi cant associations between m6A/5mC regulators, ferroptosis-related genes and tumor microenvironment in the majority of malignancies ( Fig. 7 and Supplementary Table 3). We found that some m6A/5mC regulators and ferroptosis-related genes displayed much similarities in immune-cell in ltrating pro les. 3.5. Transcriptional regulatory network for m6A/5mC regulators and ferroptosis-related genes In order to shed light on the mechanism involved in the regulation of m6A/5mC regulators and ferroptosis-related genes, we rst tried to nd out transcription factors regulating m6A/5mC regulators and ferroptosis-related genes and established a TF-gene regulatory network. We obtained 290 TF from the Cistrome Database and hTFtarget overlaps, and a network was generated (Fig. 8A). Correlation test function was utilized to test the correlations with cutoff criteria set as the correlation coe cient >0.7 and p <0.001 in PAAD (Fig. 8B). The results revealed that a large number of m6A/5mC regulators and ferroptosis-related genes were regulated by POLR2A (Fig. 8C), and correlation test function showed POLR2A was also positively correlated with the expression of m6A/5mC regulators and ferroptosisrelated genes in PAAD (Fig. 8D).
3.6. LncRNA-miRNA-mRNA ceRNA regulatory network To investigate the role of ceRNA triplets (lncRNAs-miRNAs-mRNAs) and the competitive patterns of m6A/5mC regulators and ferroptosis-related genes, we constructed a ceRNA network. As a result, a total of 854 pair-wise interactions among 239 nodes were identi ed, consisting of 69 lncRNAs, 110 miRNAs, 33 ferroptosis-related genes, and 27 m6A/5mC regulators (Fig. 9A). Among the ceRNA networks, XIST was the lncRNA that regulates the most miRNA and mRNA, so we take it as the core to construct the ceRNA network. The network consisted of 73 nodes and 223 edges, and the degrees of the top 5 nodes (ACSL4, TET3, YTHDF3, IGF2BP1, and ZBTB33) were 16, 16, 16, 15, and 12, respectively (Fig. 9B).

Discussion
There is increasing experimental evidence shows DNA/RNA methylation regulators may form an important and complex cellular regulatory network in ferroptosis, and a considerable cross-talk between them [28]. While there is no systematic study to investigate whether a cross-talk between m6A/5mC regulators and ferroptosis-related genes exists. Here, we revealed global alterations of m6A/5mC regulators and ferroptosis-related genes at transcriptional and genetic levels, and their mutual correlation in pan-cancer. Our results revealed the prognostic signi cance, functional status, tumor-in ltrating immune cells, transcription factors, and ceRNA regulatory network of m6A/5mC regulators and ferroptosis-related genes in multiple cancers. Understanding the cross-talk between epigenetic modi cation and ferroptosis may provide important insights into the mechanisms underlying tumorigenesis.
FTO-mediated m6A demethylation in tumor cells elevates the transcription factors c-Jun, JunB, and C/EBPβ, which allows the rewiring of glycolytic metabolism to evade immune surveillance [29]. Another study revealed the role of m6A in dendritic cell (DC) activation, in which METTL3-mediated m6A of CD40, CD80, and TLR4 signaling adaptor TIRAP transcripts enhanced their translation in DC for stimulating T cell activation and strengthening TLR4/NF-κB signaling-induced cytokine production [30]. Ferroptotic cells might release distinct " nd me" signals, HMGB1, prostaglandin E 2 (PGE 2 ), 5-hydroxyeicosatetraenoic acid (5-HETE), oxidized phospholipids (oxPLs), which will attract antigen presenting cells (APCs) and other immune cells to the site of ferroptotically dying cells [31]. Although the role of m6A/5mC regulators and ferroptosis-related genes in the immune microenvironment have been studied individually, however, how m6A/5mC regulators and ferroptosis-related genes play a crucial role in coordinating immune system responses needs to be further investigated. Our study provides a comprehensive insight for revealing the signi cant role of m6A/5mC regulators and ferroptosis-related genes in the tumor immune microenvironment. Interestingly, we found that FANCD2 and UHRF1 expression were positively correlated with in ltrating levels of T cell CD4 + Th2 in a wide variety of malignant neoplasms. UHRF1 is a critical factor that binds to interstrand crosslinks (ICLs). In turn, this binding is necessary for the subsequent recruitment of FANCD2, which allows the DNA repair process to initiate [32]. The exact mechanism of how UHRF1 and FANCD2 contribute to the regulation of immune microenvironment, whether directly or indirectly, needs to be experimentally veri ed.
Meanwhile, some m6A/5mC regulators and ferroptosis-related genes were regulated by the same TF.
EP300-induced H3K27 acetylation activation increased ALKBH5 expression, promoted uveal melanoma (UM) cell proliferation, migration, invasion, and decreased apoptosis in vitro [33]. CREB suppressed lipid peroxidation by binding the promoter region of glutathione peroxidase 4 (GPX4), and this binding could be enhanced by E1A binding protein P300 (EP300) [34]. In addition to TF, the m6A regulators and ferroptosis-related gene expression changes were guided by non-coding RNAs, such as long non-coding RNAs (lncRNAs) and miRNAs. MALAT1 was found to upregulate IGF2BP2 via m6A modi cation recognition by competitively binding to miR-204, conferring a stimulatory effect on proliferation, migration and invasion of TC cells, which was accompanied by weakened tumor growth and cell apoptosis [35]. MALAT1 induced KEAP1 downregulation, leading to NRF2 stabilization and activation mediated human umbilical vein endothelial cells (HUVEC) protection against H 2 O 2 [36]. NFE2L2 and ZBTB4 both were regulated by miR-17-5p in tumor [37,38] . METTL3 interacted with the microprocessor protein DGCR8 and positively modulated miR-873-5p mature process in an m6A-dependent manner. Further experiments showed that miR-873-5p could regulate Keap1-Nrf2 pathway against colistin-induced oxidative stress and apoptosis [39]. How multiple ncRNAs or TFs coordinately control the expression of m6A regulators and ferroptosis-related genes is a question that needs to be answered in future experiments.
There are still major questions that need to be urgently addressed in the near future: For instance, whether or how m6A/5mC methylation and ferroptosis connect and coordinate in immune microenvironment and immunotherapy, and how ncRNAs contributes to m6A/5mC methylation and ferroptosis via regulation of target gene activity? How m6A/5mC modi cations affect the function of speci c ferroptosis-associated genes? Most of these questions should be analyzed in detail for the construction and re ning of the conception and system of targeting m6A/5mC regulators as well as ferroptosis-based therapy. Therefore, identifying distinct m6A/5mC modi cation patterns in the tumor immune microenvironment will provide insights into the interactions of m6A/5mC RNA methylation on anti-tumor immune response and facilitate more effective precision immunotherapy strategies. Ferroptosis-mediated TIME regulation is anticipated to open a new research eld at the frontier of anticancer immunity.

Conclusions
In conclusion, we have shown that an extensive cross-talk between epigenetic regulators and ferroptosis in several aspects. Epigenetic and ferroptosis regulators are promising cancer therapeutic targets, and an epigenetic and ferroptosis regulators-targeting strategy should be synergized with ICB to fully harness the power of TIME and obtain a maximum clinical bene t for future cancer immunotherapy. Coordinated efforts are required to determine optimal therapeutic combinations and to apply both immune-pro ling and genomic-pro ling technologies to develop a personalized treatment.

Author Statement
Wumin Dai and Yingli Zhang conceived and designed the study. Xia Li, Yongyi Chen, Wangang Gong and Ying Su participated in the acquisition, analysis, and interpretation of all data. Wumin Dai wrote the paper. Yingli Zhang edited the manuscript. All authors reviewed the manuscript and gave nal approval to submit the manuscript.  comparing the groups with different expression levels of m6A/5mC regulators in pan-cancer (TCGA tumors). Red blocks represent m6A/5mC regulators unfavorable to survival, blue blocks represent m6A/5mC regulators favorable to survival, and the ones with outer wireframe indicate signi cant in uence. Mantel-Cox test was used for the hypothesis tests, and the Cox proportional hazard ratio was included in the survival plots. A p value < 0.05 was considered to be statistically signi cant.  The correlation between m6A/5mC regulators and ferroptosis-related genes. (A) The STRING database was used to analyze the correlation between m6A/5mC regulators and ferroptosis-related genes and Cytoscape was used to display the PPI network. The red circles represent the 5mC regulators, the yellow circles represent the m6A regulators, and the green circles represent the ferroptosis-related genes. The yellow connecting lines represent the connection between 5mC regulators and ferroptosis-related genes, the gray connecting lines represent the connection between m6A regulators and ferroptosis-related genes. (B) The expression correlation between m6A/5mC regulators and ferroptosis-related genes in PAAD. The red connecting lines indicate that the m6A regulator is positively correlated with ferroptosis-related genes, and the green connecting lines indicate that the m6A regulator is negatively correlated with ferroptosisrelated genes. individually collected and subjected to PFS analysis. The time period covers progression-free status. Red curves represent the altered groups, and blue curves represent the unaltered groups. A log-rank test was used for the hypothesis test, and a p-value < 0.05 was considered to be statistically signi cant.  The associations between m6A/5mC regulators or ferroptosis-related genes and the tumor microenvironment in pan-cancer. (A) The associations between m6A/5mC regulators and the tumor microenvironment in pan-cancer. (B) The associations between ferroptosis-related genes and the tumor microenvironment in pan-cancer. The yellow boxes represent a positive correlation between m6A/5mC regulators or ferroptosis-related genes with the immune cell, and the blue boxes represent a negative correlation between m6A/5mC regulators or ferroptosis-related genes with the immune cell.

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