Disulfidptosis-related lncRNAs signature predicting prognosis and immunotherapy effect in lung adenocarcinoma

Introduction

Lung cancer is the second most common form of malignant tumour, with the highest mortality rate among all malignant tumours [1, 2]. According to 2021 statistics, lung cancer accounts for approximately 22% of cancer-related deaths, with a five-year survival rate of only 21% [3]. In recent years, medical technologies such as surgery, chemotherapy, radiotherapy, and targeted therapy have advanced in the treatment of lung cancer [4]. However, early-stage lung cancer, especially peripheral lung adenocarcinoma (LUAD) with imaging features of ground-glass nodules, is difficult to detect early due to its lack of obvious symptoms. Therefore, most LUAD patients have already reached the intermediate or late stage at the time of diagnosis, losing the best opportunity for surgery [5]. For this group of people, the primary treatment methods are chemotherapy, radiotherapy, molecular targeted therapy, and immunotherapy [6]. However, despite the diverse treatments available, there has been no significant improvement in the overall survival rate of LUAD to date. These treatment methods have many shortcomings, including high side effects and poor tolerance in relapsed patients [7]. Therefore, the development of new early diagnostic methods, more sensitive biomarkers, and treatment targets is urgently needed for LUAD.

In 2023, Professor Gan Boyi’s team reported a new mode of cell death called ‘disulfidptosis’. This form of cell death is induced by a shortage of NADPH under glucose starvation [8]. The shortage leads to the overexpression of the cystine transporter solute carrier family 7 member 11 (SLC7A11) in cancer cells, which induces abnormal accumulation of disulfide (such as Cystine, etc.) in cells, ultimately leading to cell death. In this research mechanism, the process of cell reduction of cystine to cysteine is blocked due to an insufficient supply of NADPH caused by glucose starvation. This blockage leads to the production of a large number of disulfide bonds between actin molecules, inducing disulfide stress. The stress activates the Rac/WAVE regulatory complex-7 subunit of actin-related protein 2/3 (Arp2/3) signal pathway, causing disarray of the cell skeleton and ultimately inducing disulfidptosis. Currently, research on disulfidptosis is focused on cancer cell lines that overexpress SLC7A11 and are under glucose starvation [9, 10]. However, the role and mechanism of disulfidptosis in LUAD are still unclear.

Long non-coding RNA (lncRNA) is defined as RNA longer than 200 nucleotides. Although lncRNA is not involved in protein translation, it plays a very important role in gene regulation. LncRNA can affect the translation and stability of cytoplasmic mRNA and influence signal transduction pathways by regulating the function of chromatin and the function of membrane-free nucleosomes, thereby affecting gene expression in various biological physiological, and pathological environments [11, 12]. In particular, abnormal lncRNA expression may be associated with multiple biological events, such as ferroptosis [13]. LINC00336 reduces ferroptosis in lung cancer through competitive endogenous RNA activity [14]. LncRNA P53RRA induces tumor suppression by isolation of nuclear p53 and promotes ferroptosis and apoptosis [15]. As a newly discovered mode of cell death, disulfidptosis has different causes than programmed cell death such as apoptosis [16], ferroptosis [17], and pyroptosis [18]. Therefore, identifying disulfidptosis-related lncRNAs associated with LUAD prognosis is critical to developing accurate prognostic assessment and treatment.

In this study, we established and validated a highly predictive disulfidptosis-related prognostic model for LUAD. By constructing risk scores, we analyzed the clinical value in predicting clinical prognosis, immune induction, tumor mutation compliance, and immunotherapy response. Our study may provide a promising tool for predicting the prognosis of LUAD patients and offer new theoretical references for elucidating the molecular mechanisms of LUAD.

Methods

Results

Discussion

With the continuous deepening of tumor research, LUAD has made positive progress in detection, diagnosis, and treatment. However, this disease remains one of the most deadly malignancies due to the complexity of its molecular and genetic processes [22]. As an emerging biomarker, lncRNAs play an important role in the occurrence and development of different tumors, including LUAD [23]. In addition, many lncRNAs participate in the occurrence, development, and drug resistance of malignant tumors and have been identified as novel biomarkers and therapeutic targets for tumor diagnosis and treatment [24–26]. Nevertheless, it is still uncertain whether DRG-lncRNAs can predict LUAD patients’ prognosis. Therefore, we developed a prediction model for LUAD and improved patient survival rates by screening for DRG-lncRNAs.

To prevent overfitting, we used LASSO regression to reduce the dimensionality of the data. We identified 10 DRG-lncRNAs (AL365181.3, AL365181.2, AL606489.1, SNHG12, LINP1, AC106045.1, GSEC, AL590666.2, AC090559.1, and AC012615.1) that are closely associated with the OS of LUAD patients. Some of these DRG-lncRNAs have been previously reported and are closely linked to the occurrence and progression of tumors. Chen et al.’s research team found that LINP1 can increase pancreatic cancer metastasis through adsorbing microRNA-491-3p [27]. In addition, studies have shown that SNHG12 is upregulated in NSCLC tissues and cells, and facilitated immune escape through the HuR/PD-L1/USP8 axis [28]. In other LUAD-related prognostic models, AL365181.3 and AL365181.2 have been identified as indicators that can accurately predict prognosis, and their overexpression is considered a protective factor for the prognosis of LUAD patients [29, 30]. However, their specific mechanisms of action in LUAD are not yet fully clear and require further study. Research on the remaining DRG-lncRNAs is limited, and although little is known about them, their importance should not be underestimated.

In addition, we constructed a prognostic model that predicts the survival of LUAD patients based on 16 DRGs, which can be independently used as a prognostic indicator for LUAD. This model accurately divides patients into two groups: low-risk group and high-risk group. In the overall cohort, the low-risk group showed a better prognosis. In addition, we evaluated the effectiveness of this prognostic model in predicting prognosis by constructing ROC curves, and the results showed that the AUC values of the risk score for 1 year and 3 years were 0.700 and 0.630, respectively. To date, several evaluation methods have been used to assess the prognosis of patients with LUAD. In a study by Li et al., the investigators constructed a prognostic model of LUAD based on ferroptosis-related genes. In their study, the AUC value predicting 1-year survival was 0.698, which is lower than our prognostic model [31]. Therefore, the 10 DRG-lncRNAs model we studied could be a good prognostic model for LUAD.

One of the major challenges in immune therapy for LUAD patients is the lack of understanding of tumor heterogeneity, complexity, and immune evasion mechanisms. Additionally, specific biomarkers to evaluate the benefits of tumor immune therapy are insufficient. Therefore, the discovery of new immune targets and prognostic indicators is of great importance [32]. Several clinical-pathological studies have shown a correlation between gene mutations and immune therapy response [33, 34]. In this study, the high-risk group showed a significantly increased frequency of mutation events and TMB levels. This finding suggests that LUAD patients may benefit from tumor immune therapy, such as programmed cell death protein 1 (PD-1) inhibitors. Because patients with higher TMB levels have significantly improved response rates to immune checkpoint inhibitors compared with patients with lower TMB levels [35, 36]. In addition, studies have shown that the effectiveness of immune therapy depends on the coordinated responses of innate and acquired immune cells [37]. Furthermore, tumor-infiltrating immune cells may be a valuable prognostic tool in cancer treatment. In this study, we used seven algorithms such as CIBERSORT, CIBERSORT-ABS, XCELL, MCPcounter, QUANTISEQ, EPIC, and TIMER to study the potential of the 10 DRG-lncRNAs model to reflect the immune microenvironment status of LUAD and focused on using the CIBERSORT algorithm to detect the relative proportions of 22 different types of immune cells in tumor tissues in the TCGA collection. The results showed a significant increase in the number of activated CD4 memory T cells, resting NK cells, and resting macrophages in the high-risk group, confirming the role of the 10 DRG-lncRNAs model in regulating tumor immune cell infiltration. These findings suggest that the proposed model in this study can accurately predict the immune therapy efficacy of LUAD patients.

Metabolic reprogramming is one of the important features of cancer, which typically leads to increased uptake of nutrients crucial in biosynthesis and bioenergetics processes by cancer cells, such as glucose and amino acids including glutamine [38]. Cancer cells primarily achieve this through upregulating transporters for glucose and amino acid uptake. Accordingly, some cancer cells undergo cell death upon glucose or amino acid restriction while normal cells can survive under the same conditions. This nutrient dependency provides potential metabolic vulnerabilities for targeted therapy against cancer [39]. The research by Gan Boyi’s team indicates that elevated expression of SLC7A11 promotes metabolic vulnerability resulting in disulfidptosis, which may be an effective strategy for treating tumors.

With the introduction of the concepts of precision medicine and personalized therapy and the deepening of the application of machine learning in the medical field, researchers have also begun to try to apply these concepts and methods to the diagnosis, personalized medication and prognosis evaluation of cancer patients [40]. Patient clinical information, pathological information, imaging results, laboratory test results, epidemiological characteristics, genomic and proteomic data, and other data can be used as predictors for tumor prediction models [41]. The synthesis of many different data types for model construction is a trend in current research [42]. Machine learning methods have greatly accelerated the interpretation of medical big data, especially genetic and genomic data. Guyon et al. used a support vector machine (SVM) approach to construct a predictive model by selecting a subset of characteristic genes to distinguish between tumors and normal tissues [43]. Statnikov et al. systematically and comprehensively evaluated several major multi-class diagnostic models using gene expression data obtained from gene chips, and the results showed that multi-class support vector machines were the most effective classifiers [44].

In this study, we screened DRGs-lncRNAs prognostic prediction models associated with prognosis using the GEO database and COX proportional hazards regression analysis. We used the training set data to optimize the prediction model, and the test set data was used to verify the prediction effect of the model internally. We further performed statistical analysis on the function of the prognostic genes in the DRGs-lncRNAs prediction model and used GO and KEGG of the clusterProfiler package to enrich these significantly related genes, respectively. Subsequently, we validated the predictive model with tumor cell lines. In the future, if predictive models are to be applied to clinical practice, it is necessary to further validate the model in many clinical samples.

This study has some limitations. Firstly, the data source of this study was single, and the amount of data included was not large, so the analysis results may have some bias. Secondly, in order to ensure the correctness of the prognosis model, we needed to further confirm the prognosis model in other independent cohorts to ensure its accuracy. Thirdly, functional experiments should be performed to further elucidate the underlying molecular mechanisms for predicting the effects of DRG-lncRNAs.

Conclusions

In conclusion, this study established a novel model of DRG-lncRNAs that can independently predict the OS of LUAD patients. Through analysis of immune cell infiltration and drug sensitivity, it was demonstrated that DRG-lncRNAs markers are closely related to immune cell infiltration and chemotherapy drug sensitivity. The DRG-lncRNAs risk prediction model constructed in this study can serve as an important method for predicting whether LUAD patients can benefit from immunotherapy and chemotherapy.

Supplementary Materials

Supplementary Tables