Ferroptosis‐related gene NOX4, CHAC1 and HIF1A are valid biomarkers for stomach adenocarcinoma

Abstract Ferroptosis is a regulated cell death nexus linking metabolism, redox biology and diseases including cancer. The aim of the present study was to identify a ferroptosis‐related gene prognostic signature for stomach adenocarcinoma (STAD) by systematic analysis of transcriptional profiles from The Cancer Genome Atlas (TCGA), GEO and a clinical cohort from our centre. We developed a predictive model based on three ferroptosis‐related genes (CHAC1, NOX4 and HIF1A), gene expression data and corresponding clinical outcomes were obtained from the TCGA database, and the reliability of this model was verified with GSE15459 and 51 queues in our centre. ROC curve showed better predictive ability using the risk score. Immune cell enrichment analysis demonstrated that the types of immune cells and their expression levels in the high‐risk group were significantly different from those in the low‐risk group. The experimental results confirmed that NOX4 was upregulated and CHAC1 was downregulated in the STAD tissues compared with the normal stomach mucosal tissues (p < 0.05). In sum, the ferroptosis‐related gene signature can accurately predict the outcomes of patients with STAD, providing valuable insights for personalized treatment. As the signature also has relevance to the immune characteristics, it may help improve the efficacy of personalized immunotherapy.

advanced stomach cancer combines neoadjuvant chemoradiotherapy, molecular-targeted therapy and immunotherapy. Moreover, the dramatic development of immune checkpoint inhibitors, such as CTLA-4 and PD-1, suggests amazing therapeutic effects in clinical efficacy. However, therapy conditions are not eligible for most STAD patients, which suggests that more studies on the molecular mechanisms' elucidation and identifying useful biomarkers for immune checkpoint inhibitors are still urgently needed for cancer immunotherapy.
The Stockwell BR laboratory first proposed the concept of ferroptosis in 2012. 2 Ferroptosis is a novel form of programmed cell death that is distinct from apoptosis, necroptosis and autophagy in terms of its genetics, cell morphology and biochemical function. 3 This process is significantly distinctive because of the catastrophic accumulation of reactive oxygen species (ROS) and abnormal iron metabolism. An initial characterization of the mechanism triggering ferroptosis is cysteine depletion, which leads to the exhaustion of glutathione (GSH) intracellularly. 4 Hence, it is conceivable that a complex interplay that regulates the different cancer cell susceptibilities to ferroptosis would be a fruitful area in cancer research. Many studies have confirmed that many genes are involved in the initiation and execution of ferroptosis in cancers. [5][6][7] By bioinformatic analysis of clinical information derived from The Cancer Genome Atlas (TCGA) database, we found that the ferroptosis-related genes CHAC1, NOX4 and HIF1A might be the valid indicators for predicting the outcomes of STAD patients. To confirm our speculation, we used R scripts and website tools to conduct several bioinformatics analyses to investigate the clinicobiological function of CHAC1, NOX4 and HIF1A and the therapeutic potential of CHAC1, NOX4 and HIF1A in stomach cancer. Our speculation was verified in clinical specimens through immunohistochemistry (IHC).
Based on the above findings, we conclude that a ferroptosis-related prognostic model constructed by CHAC1, NOX4 and HIF1A might be a reliable prognostic signature for STAD patients.

| Data acquisition and processing
The RNA-seq profile and clinical information of 300 STAD samples, and 30 normal samples were downloaded from TCGA website (https://portal.gdc.cancer.gov/) on 23 April 2021. Then, the GTF annotation file was used to convert the Ensembl gene ID to the gene symbol. This TCGA-STAD cohort was set as the training group for this study. The GEO dataset GSE15459, comprised of genome-wide mRNA expression profiles of 192 primary stomach adenocarcinoma (STAD) tissues, was the validation set in our study. A total of 144 ferroptosis-related genes were retrieved from the manually curated database as possible regulators of ferroptosis FerrDb (http://www. zhoun an.org/ferrdb).

| Identification of significantly different genes
The 'LIMMA' R package was used to identify significantly different genes (SDGs) by the Wilcoxon test. The cut-off values were determined according to the parameters p < 0.01 and false discovery rate <0.01.

| Functional enrichment analysis
The 'clusterProfiler' R package was used to analyse the functional enrichment of significantly differentially expressed genes, including bio-

| Construction and verification of the ferroptosis-related prognostic model
Ferroptosis-related genes that were highly correlated with the prognosis of STAD were screened by univariate Cox regression analysis and a least absolute shrinkage and selection operator (LASSO) Cox regression model. Then, each STAD sample risk score was calculated by the following formula: Coef represents the gene coefficient and X reflects the gene expression level. We divided the samples into high-and low-risk groups bounded by the median of the risk score.
Kaplan-Meier survival analysis and time-dependent ROC curve analysis were applied to evaluate the prognostic capability of the signature. We evaluated whether the clinical characteristics and risk scores were risk factors for the prognosis of STAD by univariate Cox regression analyses and calculated the hazard ratio with the 'survival' R package. Then, the same procedures were applied to the GSE15459 data to verify the reliability of the prognostic model. Furthermore, we designed a nomogram to estimate the 1-, 3-and 5-year survival probability, and the risk score was used as one of the prognostic factors. To compare the calculated rates with the probabilities predicted by the nomogram, a calibration curve describing the 3-year overall survival (OS) was plotted. The nomogram and calibration curves were constructed by the 'rms' R package.

| Verification of the clinical samples
Fifty-one pairs of archived fresh frozen tumour specimens from STAD patients who underwent surgery from August 2016 to September 2017 in our centre were included in this study as another validation set. Basic clinical information was collected from all enrolled patients, including age, sex and tumour staging (according to the eighth edition of the American Joint Committee on Cancer (AJCC) TNM staging system for gastric cancer). These clinical characteristics are shown in Table 1. All aspects of this study were approved by the ethics committee of the Qingdao University School of Medicine. Informed consent was obtained from each participant.
Fifty-one freshly frozen tumour specimens from archived STAD patients were immunohistochemically stained (TMA). The expression patterns and levels of CHAC1 NOX4 and HIF1A were determined using IHC assays according to the manufacturer's instructions (Cell Signaling Technology). In brief, the paraffin-embedded slides were baked for 1 h at 68°C before xylene deparaffinization and subsequent rehydration through graded ethanol (100% and 95%). Fragments per kilobase of transcript per million fragments mapped (FPKM) were used to calculate the RNA expression levels in tissues derived from the TCGA database.

| Evaluation of the tumour microenvironment (TME) and infiltrated immune cells
To estimate the amount of stromal and immune cells in the tumour tissues, expression data (ESTIMATE) analysis was conducted to calculate the stromal score, immune score, ESTIMATE score and tumour purity of each tumour sample. 8,9 The Cibersort 10

| Statistical analysis
In this study, we used Strawberry Perl for Windows (Version 5.18.2) to organize the data and used Student's t test to screen for differentially expressed genes between tumour and normal tissues.
Pearson's correlation analysis was applied to compare the correlations between two sets of data and to calculate the correlation coefficient. R (4.1.0) was applied for all statistical analyses and graphing, and p < 0.05 was considered statistically significant.

| Differential expression of ferroptosis-related genes and functional enrichment analysis
RNA-seq data of ferroptosis-related genes were extracted from 300 genes were significantly differentially expressed (p < 0.05; Figure 1A).
Thirteen of these genes were upregulated, while the other 21 genes were downregulated in the STAD samples ( Figure 1B,C).
We further investigated the 34 genes' corresponding biological functions and pathways in ferroptosis by GO and KEGG functional enrichment analysis (Figure 2A-D). The GO results show that these genes are strongly enriched in oxidative stress response and inflammatory response, including response to oxidative stress (BP), regulation of leukocyte and lymphocyte activation (BP) and oxidoreductase activity acting on NADPH (MF). The KEGG results showed that the ferroptosis, HIF-1 signalling and platinum drug resistance pathways were significantly enriched.

| Establishment and verification of a ferroptosis-related prognostic model in STAD
We eliminated 14 patients who were lost to follow-up. According  Figure 3A. The results show that in the high-risk group, death states are denser, NOX4 and HIF1A are highly expressed, and CHAC1 expression is reduced.
Kaplan-Meier survival curves showed that the survival probability of STAD patients with a high-risk score was significantly lower than that of patients with a low-risk score. To evaluate the efficiency of the prognostic model, we created a receiver operating characteristic curve (ROC). The area under the curve (AUCs) for 1, 3 and 5 years was 0.67, 0.7 and 0.75 respectively ( Figure 3A). These results demonstrated that this model was a powerful predictor of STAD patient outcomes.
F I G U R E 3 Survival and ROC analysis. Risk score distribution, survival overview, heatmap of key genes, Kaplan-Meier overall survival curves and time-dependent ROC curves in TCGA (A), GSE15459 (B), and our cohort (C). As the risk score increased, more patients died We then applied univariate Cox regression analysis to verify whether the ferroptosis-related prognostic signature was an independent prognostic factor for patients with STAD. The tumour staging and ferroptosis-related prognostic signature were significantly correlated with OS (p < 0.001; Figure 4A,B).

| Construction of the nomogram
Nomograms are commonly used to intuitively evaluate patient prognosis in oncology. 11 Here, we constructed a nomogram to graphically depict a statistical prognostic model that generates a probability of cancer death for a given individual with STAD. To estimate the survival probability of the STAD patients, we integrated some clinicopathological factors, including age, sex, stage, and T, N and M stages, as well as the prognostic characteristics of the ferroptosis-related genes to construct a nomogram ( Figure 5A).
It can be applied to predict the 1-, 2-and 3-year survival probability. Additionally, the calibration curve showed that the 3-year actual survival was highly consistent with the predicted values ( Figure 5B), indicating that the nomograph is reliable and accurate.
This may be helpful for clinicians to make decisions and personalize treatment for STAD patients.

| Assessment of the immune microenvironment in STAD
We further investigated the correlation between the ferroptosisrelated prognostic signature and immune infiltration in STAD. By conducting ESTIMATE analysis, we found that the risk score of the ferroptosis-related prognostic signature was significantly positively correlated with the stromal score, immune score and ESTIMATE score (p < 0.05) ( Figure 6A-C). In addition, there was no obvious correlation between tumour purity and the risk score ( Figure 6D).
Then, we used a deconvolution algorithm based on support vector regression and Cibersort immune analysis to determine the type  Figure 6F). We also compared the component differences in immune cells between the two groups ( Figure 6E). The results showed that the plasma cells, follicular helper (TFH) T cells, regulatory T cells (Tregs), monocytes, M2 macrophages and DCs activated in the high-risk group were significantly different from those in the low-risk group. In addition to monocytes and M2 macrophages, the infiltration of other macrophages was significantly lower in the highscore group than in the low-score group. These results suggested that the ferroptosis-related prognostic signature was associated with immune infiltration.

| CHACI and NOX4 were differentially expressed in our human STAD cases
To further support our findings, we evaluated CHAC1, NOX4 and  (Table S1, Figure 7A-C). Then, we applied Image-Pro Plus 6.0 to estimate the expression levels of the three genes in each sample and plotted a boxplot to intuitively show the expression differences ( Figure 7D-F). We found that the proportion of samples with high expression of NOX4 in STAD tissues was significantly higher than that in paracancerous tissues, while CHAC1 showed the opposite trend. However, there was no significant difference in HIF1A expression between the two groups, this may be due to a bias in sample distribution. Next, the estimated expression level was substituted into the prognostic model to verify the reliability of this signature.

| DISCUSS ION
Ferroptosis, driven by unrestricted lipid peroxidation, is a newly defined form of regulated cell death. 12 It is involved in various pathophysiological conditions of several diseases, such as metabolic  Subsequently, by applying the ESTIMATE and Cibersort algorithms, we studied the TME's immune cell infiltration. We found that M2 macrophages, which promote tumour growth and invasion, 27 were significantly increased in the high-risk group. A recent study demonstrated that follicular helper T cells, which were significantly increased in the low-dose group, exerted antitumour effects in a CD8 + -dependent manner. Moreover, the presence of TFH is essential for the efficacy of PD-1/PD-L1 therapy. 28 However, Tregs that can prevent any overt immune response 29 were highly expressed at low levels. The estimation analysis indicated that the degree of immune cell infiltration was significantly positively correlated with the risk score. This result may be related to the large proportion of tumour-associated macrophages (TAMs) in the TME. By regulating tumour cell metabolism, TAMs can promote tumour growth. 30 Thus, the risk score calculated by the ferroptosis-related prognostic model can represent the TME of STAD patients to a certain extent.

| CON CLUS ION
In conclusion, our study describes a ferroptosis-related gene-based prognostic model for STAD that is significantly correlated with OS, clinical characteristics, TMB and the tumour microenvironment. Moreover, the reliability of the prognostic model and the expression differences of related genes in STAD were verified through our clinical samples.
However, this study was performed at a machine learning level, and further in vivo and in vitro experiments are needed to validate the findings.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.