Construction of an oxidative stress-related lncRNAs signature to predict prognosis and the immune response in gastric cancer

Oxidative stress, as a characteristic of cellular aerobic metabolism, plays a crucial regulatory role in the development and metastasis of gastric cancer (GC). Long noncoding RNAs (lncRNAs) are important regulators in GC development. However, research on the prognostic patterns of oxidative stress-related lncRNAs (OSRLs) and their functions in the immune microenvironment is currently insufficient. We identified the OSRLs signature (DIP2A-IT1, DUXAP8, TP53TG1, SNHG5, AC091057.1, AL355001.1, ARRDC1-AS1, and COLCA1) from 185 oxidative stress-related genes in The Cancer Genome Atlas (TCGA) cohort via random survival forest and Cox analyses, and the results were subsequently validated in the Gene Expression Omnibus (GEO) dataset. The patients were divided into high- and low-risk groups by the risk score of the OSRLs signature. Longer overall survival was detected in the low-risk group than in the high-risk group in both the TCGA cohort (P < 0. 001, HR = 0.43, 95% CI 0.31–0.62) and the GEO cohort (P = 0.014, HR = 0.67, 95% CI 0.48–0.93). Next, multivariate Cox analysis identified that the risk model was an independent prognostic characteristic (HR > 1, P = 0.005), and time-dependent receiver operating characteristic (ROC) curve analysis and nomogram analysis were utilized to evaluate the predictive ability of the risk model. Next, gene set enrichment analysis revealed that the immune-related pathway, Wnt/\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upbeta$$\end{document}β-catenin signature, mammalian target of rapamycin complex 1 signature, and cytokine‒cytokine receptor interaction was enriched. High-risk patients were more responsive to CD200, TNFSF4, TNFSF9, and BTNL2 immune checkpoint blockade. The results of qRT‒PCR further proved the accuracy of our bioinformatic analysis. Overall, our study identified a novel OSRLs signature that can serve as a promising biomarker and prognostic indicator, which provides a personalized predictive approach for patient prognosis evaluation and treatment.

Construction and validation of the OSRLs predictive signature. The interaction among oxidative stress-related DEGs and lncRNAs was extracted by the "limma" package. The screening criteria were correlation coefficient |R 2 |> 0.2 and P < 0.001. Afterward, a random survival forest (RSF) algorithm was applied via the "random Forest SRC" package to acquire the OSRLs in GC patients. Next, multivariate Cox regression was exploited to construct an OSRLs predictive signature via the "survival" package. The prognosis-related risk score of each GC patient was calculated as follows: Coef i and x i represent the corresponding coefficients and the expression levels of lncRNAs, respectively. For survival analysis, the median value of the risk score was used as the optimal cutoff point to classify GC patients into a high-risk group (HRiG) and a low-risk group (LRiG). The KM method was applied to assess the survival rate of the HRiG and LRiG of GC patients through the R "survival" and "survminer" packages. Moreover, the GSE66229 dataset was estimated with the same risk formula and median risk scores as the training set to assess the predictive ability of the OSRLs signature. Independent predictive characteristics among the OSRLs signature and clinicopathological variables (age, sex, clinical stage, tumor-node-metastasis (TNM) stage, and risk score) were identified by Cox regression analysis in the "survival" package. The R "rms" package was exploited to construct a prognostic nomogram to evaluate the survival rate of GC patients. Subsequently, the calibration curve was used to predict performance in the prognostic nomogram.

Enrichment analysis and tumor immune microenvironment analysis of the OSRLs prediction signature.
To determine the underlying molecular mechanisms and functional pathway of the OSRLs signature, the "GSEA" package was utilized to analyze the degree of enrichment in different pathways between the HRiG and LRiG, and the "CBNplot" package was used for visualization 17 . The "ssGSEA" package used a single sample gene set to evaluate 16 immune infiltration cell scores and 13 immune function scores. The half-maximal     Table S8). The relationships between OSRLs and mRNA are shown in Supplementary  Table S9. The risk score of each patient was calculated by the following formula: ( Table 1). Later, we utilized the OSRLs signature to construct a prognostic nomogram to visualize the survival rate of each patient and evaluated the 3-and 5-year overall survival probabilities for GC patients (Fig. 4D). A calibration curve was utilized to evaluate the predictive performance of the nomogram. Figure 4E and F show that the predicted and true values of the calibration curve were linearly correlated. The above results indicated that the predictive signature and nomogram had good performance, accurate prediction and discriminative ability in GC prediction, which can play an important role in clinical management.
Gene set enrichment analysis of the OSRLs signature. Gene enrichment analysis of predictive signature showed that the LRiG and HRiG patients were mainly related enriched in oxidative stress and immune pathways. For instance, the hallmark hypoxia ( Fig. 5A,D) represented genes upregulated in response to low oxygen levels, and these genes were enriched in the HRiG. The hallmark reactive oxygen species pathway ( Fig. 5B) was enriched in the LRiG; the related genes include the key gene thioredoxin (TXN) (Fig. 5E), which together with other enzymes constitutes scavenging enzyme systems involved in regulating mitochondrial ROS and protecting cells from oxidative stress 18 . As an upstream gene of ubiquinone oxidoreductase subunit A6 (NDUFA6), TXN may affect mitochondrial fitness by regulating the expression of NDUFA6 19 . The hallmark Wnt β-catenin signaling ( Tumor immune microenvironment analysis. To confirm the interaction between OSRLs and the tumor immune microenvironment, we utilized the "ssGSEA" package to analyze differences in 16 immunerelated cells and 13 immune-function scores in subgroups. As shown in Fig. 6A, mast cells, as cells involved in the immune response, were significantly differentially enriched between subgroups (P < 0.05) (Fig. 6B). Major histocompatibility complex class I (MHC class I) had higher expression levels in the LRiG than in the HRiG (P < 0.05) (Fig. 6C). This may be because MHC class I assist CD8 + T cells in eliminating malignant cells and providing long-term protective immunity. In contrast to the results for MHC class I, parainflammation, which is an inflammatory response by tissue cells under various stresses or abnormal functions, was more highly enriched in high-risk patients (P < 0.05) (Fig. 6D). Later, a heatmap was used to estimate the interaction between immune checkpoints and the risk model (Fig. 6E), and the expression of the CD200, TNFSF4, TNFSF9, and BTLN2 checkpoints was significantly increased in HRiG patients ( Fig. 6F-I). Finally, the measurement standard for exploring the relationship between the OSRLs predictive signature and chemotherapy drugs (the half-maximal inhibitory concentration, IC 50 ) was assessed. HRiG patients were more sensitive to methotrexate and TrKA inhibitors (Fig. 7A,B), while LRiG patients were more sensitive to rucaparib, bryostatin, embelin, and palbociclib

Discussion
GC is the third most common cause of cancer-related death in the world. Its 5-year survival rate is only 20% to 30%, indicating that it poses serious harm to physical health. Previous studies have shown that tumor cells produce more ROS than normal cells due to mitochondrial and metabolic disorders. To prevent the occurrence and metastasis of cancer, antioxidant therapy targeting GC may be a new approach to the treatment of GC 20 . However, only a few studies have focused on the prognostic characteristics of oxidative stress. In contrast, more studies have focused on the mechanism and treatment of oxidative stress in GC. Some therapeutic approaches mostly target only tumor cells with little attention to the tumor microenvironment (TME). With the rapid development of bioinformatics technology, there is increasing evidence that lncRNAs contribute to carcinogenesis and tumor development and can be used as a predictive feature associated with oxidative stress 21 . However, no report has been published about the prognostic significance of OSRLs in GC. Therefore, it is imperative to select oxidative stress-related biomarkers involved in GC development, construct a more efficient risk prediction model for GC, improve the prognosis of GC and seek new targets for microenvironment-targeted therapy.
In this study, we identified 185 oxidative stress-related DEGs between the TCGA-STAD dataset and oxidative stress-related gene set by bioinformatics methods. Next, to identify enriched pathways of oxidative stress-related DEGs, we used GO and KEGG pathway analysis and integrated these results, yielding two gene sets. The GO set of DEGs was significantly enriched in "response to oxidative stress" (P = 8.82 × 10 −30 ) in the BP category, which is consistent with the research direction in this study. The top three terms of the remaining set were enriched in "lipid and atherosclerosis" (P = 1.50 × 10 −9 ), "drug metabolism-cytochrome P450" (P = 4.79 × 10 −9 ), and "IL-17 signaling pathway" (P = 2.89 × 10 −10 ). In the KEGG analysis, atherosclerosis, as a chronic inflammatory disease, was found to be affected by the state of oxidative stress, and ROS was found to play a crucial role in inflammatory responses, cell growth, apoptosis, and vascular homeostasis, all of which may be relevant to cancer 22 . Cytochrome P450 (CYP450) is a hemoglobin superfamily that plays a significant role in drug detoxification, cellular metabolism, and homeostasis 23 . In this family, CPY2E1 was found to be involved in the occurrence and development of liver cancer because it can generate high levels of ROS 24 . The final term "IL-17 signaling pathway" has been verified to promote oxidative stress-induced hepatocyte apoptosis through the Nrf2/keap1 signaling pathway 25 . This indicated that oxidative stress-related genes were activated by the IL-7 signaling pathway, which influenced GC progression 26 . The mechanism of oxidative stress-related genes in GC still needs to be further explored.
A previous study indicated that lncRNAs play a significant role in the oxidative stress of cancer patients. Wang et al. indicated that lncRNA H19 and HULC activated oxidative stress by H 2 O 2 and glucose oxidase to regulate CCA cell migration and invasion 27 . Similarly, researchers have indicated that lncRNA NEAT1 is upregulated by (-)-epigallocatechin-3-gallate (EGCG)-induced oxidative stress, increasing cisplatin intake in lung cancer treatment 28 . However, the explanation of lncRNAs interaction with oxidative stress in cancer, especially in GC was awfully insufficient are unclear. In this regard, considering the abundance of oxidative stress-related genes, Pearson correlation analysis was used to explore OSRLs. Then, the RSF algorithm was exploited to discriminate OSRLs and obtained a total of 34 lncRNAs related to prognosis in GC. Finally, Cox regression was used to identify DUXAP8, TP53TG1, SNHG5, AC091057.1, AL355001.1, ARRDC1-AS1, DIP2A-IT1, and COLCA1 to construct the prognostic signature of OSRLs. Among them, DUXAP8 can promote GC development by epigenetically repressing PLEKHO1 expression by binding EZH2 and SUZ12 29 . Similarly, lncRNA TP53TG1 inhibits the activation of the PI3K/Akt signaling pathway by binding to the human tumor protein CIP2A, which leads to the inhibition of GC cell proliferation and survival 30 . Unlike the first two lncRNAs, the role of SNHG5 in tumors varies depending on gene copy variation (deletion or amplification), transcription factors, histone modification, or DNA methylation differences in gastric patients, it can either promote or suppress tumor growth 31 . In the study of glioma, SNHG5 promotes tumor growth by targeting E2F3, and E2F3 and E2F1 are both transcription factors of the E2Fs encoding gene family and are associated with poor prognosis of gastric cancer. Current study shows that E2F1 interacts with ARRDC1-AS1, so we infer that ARRDC1-AS1 has a similar function to SNHG5, and promotes the growth of gastric cancer cells by targeting E2F1. The remaining 4 lncRNAs (COLCA1, AC091057.1, www.nature.com/scientificreports/ www.nature.com/scientificreports/ AL355001.1, and DIP2A-IT1) have no relevant research in cancer. Therefore, it is necessary to verify the function of 8 lncRNAs through experiments, to determine the potential of OSRLs as prognostic biomarkers of GC that understand the mechanism of GC to develop therapeutic drugs for GC. DUXAP8 is a biomarker and therapeutic target of various cancers that is upregulated in GC, promotes cell proliferation and migration, and then accelerates the development of GC 32 , and this study verified the upregulation of DUXAP8 in GC. Previous studies have shown that lncRNA TP53TG1 can not only inhibit the development of GC by regulating the stability of CIP2A34 30 , but also play a promoting role in cancer development. For example, TP53TG1 increased the sensitivity of non-small cell lung cancer cells to DNA damaging agents by regulating the miR-18a/PTEN axis 33 . Interestingly, the qRT-PCR results of this study also showed that lncRNA www.nature.com/scientificreports/ TP53TG1 was highly expressed in GC and could be a potential risk factor rather than a protective factor. Similarly, lncRNA SNHG5 has been found to promote tumorigenesis and metastasis in a variety of cancers, while other studies have indicated that it suppresses tumorigenesis 34 . In the present study, SNHG5 was upregulated in GC cells, indicating its role as a risk factor. However, some studies have also shown that the expression of lncRNA SNHG5 was lower in GC cells than in healthy cells and benign gastric disease cells 35 . DIP2A-IT1 was upregulated in the analysis of DEGs in osteosarcoma tissue transcripts. But, in the present study, the expression of lncRNA DIP2A-IT1 was not significantly different between the GC group and normal gastric cells. Then, we found 4 novel  www.nature.com/scientificreports/ lncRNA biomarkers in GC and verified them experimentally. One of them was the prognostic protective factor lncRNA AC091057.1, which was associated with 68.87% of mRNAs in our study. The expression of AC091057.1 was experimentally verified and was consistent with the results of a pancreatic cancer study indicating it as a protective factor in cancer 36 . In a diffuse large B cell lymphoma (DLBCL) study, the knockdown of ARRDC1-AS1 exacerbated proliferation, inhibited apoptosis, and promoted invasion and migration 37 . Therefore, we presumed that the high expression of ARRDC1-AS1 in GC indicates its role as a tumor suppressor. In another study of coronary artery endothelial cells, the high expression of COLCA1 was stimulated by oxidized low-density lipoprotein, which regulates the level of oxidative stress in cells, thus leading to a sustained inflammatory response in cells 38 . This was consistent with the high expression of COLCA1 in our experiment. The remaining lncRNA, AL355001.1, was detected in recent years, and no relevant experiment has proven its expression in cancer cells.
Our study was the first to experimentally verify the expression of AL355001.1, which was higher in GC cell lines than in normal cell lines. Overall, the results indicate that these 8 lncRNAs can be used as prognostic factors for GC patients, consistent with our bioinformatics analysis results. Increasing evidence has shown that oxidative stress plays an important role in the tumor immune microenvironment. As highlighted by Wu et al. ROS restricts the cytosolic translocation of SUMO-specific protease 7 and affected the metabolism and functional activity of CD8 + T cells, which weakens antitumor activity in vivo 39 . To confirm the enriched pathways, we used GSEA of genes enriched in the HRiG and LRiG and integrated those results, yielding two aspects. One side indicated that tumor hallmarks were associated with ROS in both the HRiG and the LRiG. For example, the Hallmark hypoxia and Hallmark reactive oxygen species pathways were closely associated with the oxidative stress-related signature in this study. On the other hand, GSEA showed that cytokine-cytokine receptor interaction, cell cycle, Wnt/-β catenin signaling, and mTORC1 signaling were strongly connected to antitumor immunity and decreased oxidative stress in GC. Based on this, we speculated that antitumor immunity and oxidative stress were closely related in GC. Previous studies have shown that cytokines enhance the expansion and persistence of CAR-T cells and enhance their function in the immunosuppressive TME 40 . In addition, the Wnt pathway is an upstream pathway that participates in the regulation of the cell cycle, tumor, and other pathways 41 . The canonical Wnt signaling pathway is activated by the mitotic CDK14/ cyclin Y complex via phosphorylation of the LRP6 coreceptor, which leads to anti-inflammatory signaling that inhibits tumor growth 42 . Moreover, mTOR is located at the core of tumor-related signaling pathways 43 and plays a key regulatory role in the cell cycle 44 . The importance of mTORC1 in regulating innate and adaptive immunity has been widely recognized; for example, it regulates immune tolerance related to regulatory T cells 45 . Overall, we found a strong association between the predictive signature and antitumor immunity, providing new biomarkers for cancer immunotherapy.
Immunomodulation has been indicated to play an important role in cancer treatment. Immune cell infiltration can seriously influence cancer progression and the response to immunotherapy and is related to the prognosis of cancer patients. Patients with obvious tumor clinical manifestations are usually in the tumor escape phase; that is, tumor cells with reduced immunogenicity grow to a certain extent and surpass the ability of the body's immune response to avoid an antitumor immune response 46 . This may affect the levels of immunosuppressive and immune response cells. Therefore, we predicted a significant difference in immunotherapy in HRiG/LRiG patients. The outcomes showed that mast cells were prominent in the HRiG, so the mast cells might directly suppress immunity by promoting angiogenesis and the infiltration of mast cell subsets to facilitate different degrees of tumor development 47 . Immune checkpoint blockade, as one of the methods of immunotherapy, has been used to improve the prognosis of patients with malignant tumors 48 . The expression levels of CD200, TNFSF4, TNFST9, and BTNL2 were higher in HRiG patients, and these patients may benefit from immune checkpoint blockade to enhance the immune response or inhibit oxidative stress, which could improve prognosis in the HRiG. In recent years, some studies have reported that TPE-DPA-TCyp 49 , CPT 50 , MitoCAT-g 51 , TSEOP 52 are useful as polymer prodrugs or nano-delivery systems of antitumor immune drugs to modulate tumor oxidative stress. These findings regarding oxidative stress provide a new idea for tumor immunotherapy and a safe and economical method for the treatment of tumor patients.
In general, OSRLs, which were validated by qRT-PCR experiments, have the potential to be biomarkers for predicting the overall survival rate of STAD patients. Notably, the signature might regulate immune infiltration levels as well as immune function. Therefore, the mechanisms and relationships among oxidative stress, lncRNAs, immunity, and GC deserve further exploration and validation. We believe that the 8 OSRLs signature can guide research on the biological behavior of GC and its clinical prognosis. However, there are some limitations to this study: (1) The study used one external validation set, and more external validation sets are required to ensure the validity of the model. (2) It was necessary to further verify the mechanism of OSRLs in GC by performing experiments.