Construction of a novel immune-related lncRNA signature and its potential to predict the immune status of patients with hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is the fourth most common malignant tumor globally, and its incidence is increasing annually with poor survival [1, 2]. The primary relevant risk factors associated with the development of HCC include viral hepatitis, alcoholic liver disease, nonalcoholic fatty liver disease, aflatoxin exposure [3]. In recent years, chemotherapeutics have achieved encouraging results in treating HCC, especially immune checkpoint inhibitors (ICIs) [4]. With the recent success of clinical trials of immunotherapy, such as Checkmate 040, Keynote-224, and IMbrave150, ICIs such as nivolumab, pembrolizumab, and atezolizumab plus bevacizumab have been approved for the treatment of HCC [5‐7].

Long noncoding RNAs (lncRNAs) are nonprotein-coding RNAs with a transcription length of more than 200 nucleotides [8]. Because lncRNAs are abundant, they often participate in human physiological processes and are closely related to the development of diseases [9‐11]. In addition, lncRNAs have the ability to interact with molecules such as DNA, RNA, or protein to play enhancing or inhibitory roles [12]. Studies have reported that lncRNAs may participate in tumorigenesis through various molecular mechanisms [13, 14]. Recent work has demonstrated that lncRNAs can promote the malignant phenotypes of cancer by changing the genome or transcriptome level and varying the immune microenvironment [15]. LncRNAs can activate immune cells by expressing related genes, which leads to immune cells infiltrating tumors [16].

Immune cell infiltration markers in tumors show prospective predictive and prognostic value for tumor diagnosis, treatment, and survival evaluation [17‐19]. Moreover, because lncRNAs have a close relationship with tumor immunity, the study of lncRNAs in combination with tumor immunity will help to establish these markers. The researches of Hong [20], Wei [21], and Qu [20] constructed models to predict the prognosis of HCC, pancreatic cancer, and clear cell renal cell carcinoma based on the immune-related lncRNAs (irlncRNAs) and risk scores, which have certain accuracy in predicting the prognosis of tumor patients. Our study built a novel signature constructed by irlncRNAs to predict HCC patients’ prognosis. IrlncRNAs, such as LINC01138, THUMPD3-AS1, AL365203, TBX2-AS1, have been confirmed to be related to the prognosis of HCC patients [21‐24]. LINC01138 can promote cell proliferation, tumor invasion, metastasis and enhance protein stability by interacting with arginine methyltransferase 5 (PRMT5) [21]. TBX2 hypermethylation was associated with increased HCC risk [24]. While AC092535, FAM99A, AL355802, etc., have not been reported in HCC.

Generally, the accuracy of a tumor prediction model based on the combination of two biomarkers is better than that composed of a single gene [25]. To date, few models have been used to study the predictive role of lncRNAs and tumor immune-related cells in HCC [26, 27]. This study used a novel modeling algorithm, which does not require specific expression-level data, through pairing and iteration to establish an irlncRNA signature. Subsequently, we evaluated the diagnostic effect, predictive value, immune cell infiltration into tumors, and chemotherapy efficacy of this signature in HCC patients.

It is necessary to improve the accuracy of prognostic markers for HCC patients. LncRNAs are closely related to normal physiological activities and the development of diseases [10, 30]. Furthermore, studies have demonstrated that lncRNAs play vital roles in tumor development and antitumor processes [31‐33]. Recent studies have focused on investigating the potential relationship between coding genes and noncoding RNAs to predict patient prognosis with cancers [20, 34]. Unfortunately, the majority of these signatures were generated with the specific expression levels of transcripts. In our research, we ignored the specific expression levels of lncRNAs and utilized ir-gene pairing to generate a practical model with a combination of lncRNAs.

First, we downloaded the original information of lncRNAs from the TCGA database, and then a differential coexpression analysis was performed to catalog the DEirlncRNAs. The lncRNA pairs were verified by an improved cyclic single pair method along with 0 or 1 matrix. Second, univariate analysis and modified LASSO penalty regression were performed to determine DEirlncRNA pairs, procedures including cross-validation, multiple repetitions, and random stimulation. Then, we gained the optimum model by examining each AUC value of the ROC curve, and the optimum cutoff point was determined according to the AIC value of each point on the AUC to distinguish the high-risk and low-risk groups in the HCC dataset. Finally, the model was evaluated according to various parameters, such as survival rate, clinicopathological features, tumor-infiltrating immune cells, checkpoint-associated molecules, and chemotherapeutics.

The origin of lncRNAs may have the following four sources, mutation of a protein-coding gene, chromosomal rearrangement, duplications, and transposon insertion [35, 36]. Current research reveals that the phenotypic characteristics of lncRNAs regulation of cancer mainly include: cell proliferation, growth inhibition, cell migration, cell immortalization, angiogenesis, and cell viability [37]. The relationship between lncRNAs and tumors has received increasing attention [37‐39]. Deng et al. established a model to predict HCC patient survival [40]. The method utilized in this study does require data on the specific expression level of each lncRNA; only pairs with high or low expression levels need to be detected. Therefore, the model is practical and straightforward in distinguishing high-risk or low-risk clinical cases. The lncRNAs included in this model are related to ir-genes; Therefore, these lncRNAs may regulate the immune microenvironment and the activation of immune cells.

Our research reveals that some of the DEirlncRNAs included in the modeling play vital roles in the malignant phenotype of many cancers, such as MYLK−AS1 [41, 42], THUMPD3 − AS1 [22], and DSCR8 [43], especially in the development of HCC. MYLK−AS1 promotes angiogenesis and HCC progression by targeting the miR-424-5p/E2F7 axis and activating the VEGFR-2 signaling pathway [42]. THUMPD3 − AS1 was associated with the cell cycle and can be used as a prognostic marker in hepatitis B virus-related HCC patients [22]. Wang et al. revealed that DSCR8 promotes the progression of HCC by activating the Wnt/b-catenin signaling pathway [44]. The established model can identify new biomarkers for further tumor-related studies.

To achieve better accuracy and effectiveness of risk prediction, this study used the improved method of the LASSO penalty model [45]. In addition, we determined the maximum value for an optimal model by calculating each AUC value and then compared it with other clinicopathological characteristics, further improving the modeling process. The AIC value was used to obtain the ideal cutoff point for model fitting; the median value was not used to discriminate risk. After using this new method to differentiate high-risk and low-risk groups, survival outcomes and univariate and multivariate analyses of clinicopathological features were reevaluated. Moreover, the sensitivity of chemotherapy drugs commonly used to treat HCC treatment was analyzed. The relationship between high-risk and low-risk groups and immune cell infiltration into tumors and the relationship between high-risk and low-risk groups and immune checkpoint-related genes were also studied, and the results indicated that this modeling algorithm has a good clinical application prospects.

The immune checkpoint blockade reaction is closely related to tumor-infiltrating immune cells [46]. Our research used seven commonly recognized methods to identify infiltrating immune cell to investigate the relationship between risk scores and tumor-infiltrating immune cells, including XCELL [47, 48], TIMER [49, 50], QUANTISEQ [51, 52], MCPCOUNTER [53], EPIC [54], CIBERSORT-ABS, and CIBERSORT [55, 56]. Due to the defects and complexity of these algorithms, they are rarely compared with each other. Through integration analysis, our findings show that DEirlncRNA pairs have a positive correlation with tumor-infiltrating immune cells such as B cells, neutrophils, and macrophages but are negatively correlated with CD8+ T cells, CD4+ T cells, and monocytes. Wang et al. demonstrated that the immune score can predict the efficacy of immunotherapy and chemotherapy [57]. IrlncRNA SATB2-AS1 can affect the tumor immune cell microenvironment and inhibit colorectal cancer metastasis [41]. LncRNA-EGFR can stimulate T regulatory cell differentiation and promote immune evasion in HCC [42]. Our model suggests that high risk is related to sensitivity to chemotherapy drugs such as methotrexate, rapamycin, bleomycin, doxorubicin, gemcitabine, mitomycin, and paclitaxel but not sensitivity to sorafenib. We believe that immunotherapy is more effective than traditional chemotherapy, mainly because immunotherapy can activate immune cell functions and promote tumor resistance by triggering active immunity. Tumor mutations can cause a large number of neoantigens to be released, which can be recognized by T cells and cause many immune cells to infiltrate into tumors [58‐60].

We acknowledge that our study has some limitations. First, the research data were based on public databases. Some data were incomplete, such as some clinicopathological features and the sensitivity of drugs commonly used in the treatment of HCC; for instance, lenvatinib and oxaliplatin have not been analyzed. Second, the constructed model needs external verification because the expression level of each sample differs, which may lead to an unreliable final model. Third, this study did not analyze the expression level of these lncRNAs in immune cells. However, this research uses various methods to verify the new modeling algorithm and optimizes and analyzes it. Despite the lack of external data validation, from the analysis results, our model was acceptable. However, this study will be more convincing when external validation is performed. Therefore, our team will recollect clinicopathological data for subsequent studies and enlarge the sample size for further verification.

Our research shows that an innovative signature established by irlncRNAs that does not require data on the expression levels of lncRNAs to predict HCC patient prognosis and may contribute to identification of patients who can benefit from antitumor immunotherapy.