1 Introduction

Esophageal adenocarcinoma is a common malignant tumor of the distal esophagus and gastroesophageal junction [1]. Esophageal adenocarcinoma is often complicated with severe malnutrition [2]. Esophageal adenocarcinoma and local metastatic lymph nodes often invade and compress critical anatomical structures in the mediastinum (e.g., trachea, vertebral body, recurrent laryngeal nerve, and pericardium), resulting in fatal complications such as aspiration pneumonia. In China, esophageal cancer accounts for about half of the world, of which less than 10% are esophageal adenocarcinoma [3]. Surgery, radiotherapy, chemotherapy, and immunotherapy are the primary treatments. The mortality rate remains elevated despite the increasing diversity of treatment. In the United States, the 5-year survival rate is only 17%. In China, the 5-year survival rate for early esophageal cancer is up to 30%, but for advanced esophageal cancer, it is only about 5% [4]. Outcomes have long been judged using TNM staging. There is an essential difference in survival between early, locally advanced, and advanced esophageal adenocarcinoma. Nevertheless, TNM staging relies only on anatomical information obtained from imaging and does not consider the characteristics of the tumor itself (e.g., tumor biological characteristics, sensitivity to radiotherapy and chemotherapy, and tumor immune status). For this reason, it is challenging to predict outcomes accurately.

In addition to surgery and concurrent chemoradiotherapy, immunotherapy is becoming a critical part of esophageal adenocarcinoma treatment. The Checkmate-649 study showed that nivolumab combined with chemotherapy resulted in higher objective response rates, more prolonged progression-free survival, and overall survival (OS) in patients with esophageal adenocarcinoma with a CPS (Combined Positive Score) of 5 or more [5]. The OS of patients with esophageal adenocarcinoma is strongly influenced by the treatment effect of immune checkpoint inhibitors (ICIs), which depend on tumor immune status [6]. Precise regulation of immune gene expression modulates this immune status of tumors. Long non-coding RNA (lncRNA) accounts for 80% of the human transcriptome, participates in 70% of the gene expression regulation, and positively or negatively regulates by binding with DNA or RNA. Several lines of evidence suggest that lncRNAs regulate the expression and function of immune genes. Lnc-DC regulates dendritic cells' differentiation and activation profile by binding to STAT3 in the cytoplasm and preventing its dephosphorylation and inactivation by tyrosine phosphatase SHP1 [7]. Lnc-MAF-4 facilitates Th1-cell differentiation and inhibits Th2-cell differentiation by directly inhibiting MAF, a Th2-cell transcription factor [8]. However, the effect of a single lncRNA on the regulation of an immune gene on the overall immune function is unknown. Therefore, investigators are exploring lncRNAs related to the expression of immune genes in tumor tissues using bioinformatics analysis. Outcomes in several solid tumors have been calculated using risk models based on the potential regulatory relationship between these lncRNAs and immune genes. Studies showed that the composition of immune infiltrating cells in the tumor microenvironment and the expression levels of genes related to immunosuppressive pathways in tumor tissues are associated with the therapeutic effect of ICIs [9,10,11]. Analysis of the correlation between immune-related lncRNAs and these two factors may be helpful to predict the efficacy of ICIs and enhance the value of survival prediction models.

By analyzing the transcript data of esophageal adenocarcinoma specimens and adjacent normal tissue specimens in The Cancer Genome Atlas (TCGA), we generated a risk model based on an immune-related lncRNA (irlncRNA) pair signature and calculated the ability of the model to predict outcomes. Then, based on this risk model, we performed predictive analyses of chemotherapy drug sensitivity in various risk groups using the pRRophetic package in R software. In addition, we computed the differences in the composition of immune infiltrating cells and the expression of genes related to the immunosuppressive pathway in various risk groups.

2 Material and Methods

2.1 Acquisition and Preparation of Transcriptome Data and Analysis of Differentially Expressed IrlncRNAs

Fragments per kilobase million and clinical data of 78 esophageal adenocarcinoma tumor tissue samples and 9 adjacent normal tissue samples were obtained from TCGA-ESCA project (https://portal.gdc.cancer.gov). Inclusion criteria were: (1) esophageal adenocarcinoma and adjacent normal tissue; (2) relevant inspection completed; (3) test results are complete. Exclusion criteria were: (1) recurrent or metastatic tissues; (2) those with a history of anti-tumor treatment, including immunotherapy, targeted therapy, radiotherapy and chemotherapy; (3) those with other types of pathogenic diseases, such as severe trauma. 78 cases of esophagus of primary site, 70 cases of GENIE-MDA and 8 cases of GENIE-JHU, 68 male and 10 female, without vital status were reported. Merged data in the genetic Ensembl format were annotated using gene transfer format files (http://asia.ensembl.rog). Messenger RNAs (mRNAs) and lncRNAs were distinguished according to the corresponding naming rules in the gene transfer format files. A list of genes updated in July 2020 was downloaded from the IMPORT database (https://www.immport.org/), including genes deemed to be involved in the regulation of immune function. We calculated correlations between the expression levels of lncRNAs and immune-related genes. LncRNAs with correlation coefficients greater than 0.4 and P-values less than 0.001 were chosen as irlncRNAs. Then, differentially expressed irlncRNAs in 78 esophageal adenocarcinoma tissue samples and 9 adjacent normal tissue samples were analyzed. IrlncRNAs with log fold-changes greater than 2 and false discovery rates less than 0.05 were designated as differentially expressed immune-related lncRNAs (DEirlncRNAs). Correlation analysis and differential expression analysis were performed using the limma package. The 50 lncRNAs with the most significant differences were presented in a heat map (Fig. 1A).

Fig. 1
figure 1

Identification of DEirlncRNA: (A) heatmap of 50 most significant lncRNAs; (B) volcano map of 339 DEirlncRNAs

Full size image

2.2 Pairing DEirlncRNA

All possible pairs of DEirlncRNAs (lncRNA-A and lncRNA-B) were identified and assessed based on the following definition: (I) if the expression level of lncRNA-A was higher than lncRNA-B, the pair was defined as lncRNA-A|lncRNA-B = 1; (II) if the expression level of lncRNA-A was less than lncRNA-B, the pair was defined as lncRNA-A|lncRNA-B = 0; (III) for a given pair if lncRNA-A|lncRNA-B = 1 in more than 80% of all samples, the pair was considered invalid; and (IV) if lncRNA-A|lncRNA-B = 0 in more than 80% of all samples, the pair was also considered invalid. Valid DEirlncRNA pairs were identified by applying these definitions (lncRNA-A|lncRNA-B = 1 accounted 20% at most, and lncRNA-A|lncRNA-B = 0 accounted 20% at most) and used to construct a matrix with each entry corresponding to the respective lncRNA-A|lncRNA-B value.

2.3 Establishing a Risk Model for Evaluating Risk Score

The survival status and survival times were fused with DEirlncRNAs pairs. The fused matrix was analyzed using univariate Cox regression to obtain DEirlncRNAs pairs associated with survival, of which the independent variables include genes with statistically significant differences, and the dependent variable is survival status. Least absolute shrinkage and selection operator (LASSO) regression was used to identify the optimal DEirlncRNAs pairs related to outcome. Ten-fold cross-validated LASSO regression was performed with a P-value of 0.05. LASSO regression was performed 1000 cycles, and each cycle was randomly stimulated 1000 times. We recorded the frequency of each pair in the LASSO regression model with 1000 repetitions and selected the pairs with a frequency of more than 100 to perform subsequent analysis. Partial likelihood deviance was plotted against the lambda logarithm in the tenfold cross-validation (Fig. 2A). LASSO coefficient profiles are shown in Fig. 2B. Finally, these DEirlncRNAs pairs were analyzed using Cox proportional hazard region analysis, and a value at risk model was constructed. The patient’s value at risk can be calculated using the following formula: risk score = ∑ coef (DEirlncRNA pairsi) × expr (DEirlncRNA pairsi), where coef (DEirlncRNA pairsi) and expr (DEirlncRNA pairsi) represent the survival correlation coefficient of the DEirlncRNA pair and the DEirlncRNA pair matrix, respectively.

Fig. 2
figure 2

Construction and ROC curves of the lncRNA pair signature: (A) partial likelihood deviance was plotted against the lambda logarithm in the tenfold cross-validation; (B) LASSO coefficient profiles; (C) Forest maps of the 11 DEirlncRNA pairs identified by univariate Cox regression analysis; (D) Forest maps of the 11 DEirlncRNA pairs identified by multivariate Cox regression analysis; (E) the cutoff point of 1-year ROC is calculated by AIC; (F) the 1-, 2- and 3-year ROC curves; (G) a comparison between the 1-year and clinical characteristics ROC curves

Full size image

Receiver operating characteristic (ROC) curves were drawn for 1, 2, and 3 years. The Akaike information criterion values of every point of the 1-year ROC curve were calculated to identify the maximum inflection point (defined as the optimal cutoff point, of which the Youden index was the most), which we used to classify the cohort into high- and low-risk groups for further validation. The R package used for these calculations was survival ROC.

2.4 Clinical Evaluation by Risk Assessment Model

Kaplan–Meier analysis was used to compute differences in survival between the high- and low-risk groups. Survival curves were drawn to visualize these differences, of which the abscissa was time (years), and the ordinate was survival probability. We performed the Chi-square test to analyze the determine between the model and different clinical characteristics. We performed the Wilcoxon signed-rank test to calculate the differences in risk scores among groups with these characteristics. Block diagrams were drawn to visualize differences. We conducted univariate and multivariate Cox regression analyses between risk score and clinical characteristics to determine whether the risk model could be used as an independent prognostic factor. A forest map was drawn to display the results. The R packages used for this section included limma, survival, survival ROC, glmnet, survminer, ComplexHeatmap, and guppbr.

2.5 Investigating the Relationship Between Drug Sensitivity and the Risk Model

We utilized the pRRophetic R package and the Wilcoxon signed-rank test to compare the 50% inhibitory concentrations in high- and low-risk groups.

2.6 Investigating the Relationship Between Tumor-Infiltrating Immune Cells and the Risk Model

We used the software packages TIMER [12], CIBERSORT [13], XELL [14], QUANTISEQ [15], MCPcounter [16], and EPIC [17] to analyze immune infiltrating cell content variation between high- and low-risk groups. The Wilcoxon signed-rank test was performed, and the results were displayed in a box chart. The relationship between risk score and distribution of immune infiltrating cells was analyzed using Spearman correlation analysis. The significance threshold was set to P < 0.05. The R packages used for this section were limma, scales, ggplot2, and ggtext.

2.7 Investigating the Relationship Between ICI-Related Genes and the Risk Model

We used limma and ggpubr R packages to compare expression levels of ICI-related genes between high- and low-risk groups. Violin diagrams were drawn to visualize the results.

3 Results

3.1 Identification of DEirlncRNAs

We analyzed transcript data from 78 esophageal adenocarcinoma tissue samples and 9 adjacent normal tissues. We measured the expression levels of lncRNAs that correlated with the expression of known immune function-related genes in all samples. We identified 339 irlncRNAs with altered expression in tumor tissues and adjacent normal tissues using differential analysis (Supplementary Table 1), of which 14 irlncRNAs were downregulated, and 325 were upregulated in tumor tissues (Fig. 2B).

3.2 Construction of DEirlncRNA Pairs and the Risk Model

First, using an iteration loop and a 0-or-1 matrix screening among 335 DEirlncRNAs, 40,531 valid DEirlncRNA pairs were identified. We correlated these DEirlncRNA pairs with the survival data and found 26 DEirlncRNA pairs correlated with OS. Then, 26 DEirlncRNA pairs associated with survival were identified using LASSO regression analysis, and 17 were retained (Supplementary Table 2). Finally, the univariate and multivariate Cox regression analyses were carried out with the 17 DEirlncRNA pairs, and 11 DEirlncRNA pairs were obtained that could be used to construct the risk model (Fig. 2C, D). The coefficient values of the 11 DEirlncRNA pairs identified by univariate Cox regression analysis were − 0.218, − 0.196, 0.247, 0.223, − 0.205, − 0.207, − 0.193, 0.341, − 0.182, − 0.217, − 0.236, respectively, and the coefficient values of the 11 DEirlncRNA pairs identified by multivariate Cox regression analysis were − 0.215, − 0.199, 0.251, 0.236, − 0.212, − 0.204, − 0.195, 0.342, − 0.189, − 0.215, and − 0.231, respectively. Based on the risk model constructed by these 11 DEirlncRNA pairs, a ROC curve of 1-year survival was plotted, with an area under the curve (AUC) of 0.904 and a cutoff of 0.009 (Fig. 2E). We plotted the ROC curves at years 2 and 3, with AUCs 0.987 and 0.977, respectively (Fig. 2F). These AUCs were greater than 0.9, indicating that the risk model established predicts survival well. By comparing the ROC curves of this model with other clinical features, we found that the model was far better at predicting survival than T staging, N staging, and TNM staging (AUCs were 0.516, 0.583, and 0.614, respectively) (Fig. 2G). We calculated the risk value of 78 patients with esophageal adenocarcinoma according to the risk model (Supplementary Table 3).

3.3 Clinical Evaluation Using the Risk Assessment Model

Based on the previously determined cutoff value, we classified 78 patients with esophageal adenocarcinoma into a high-risk group (n = 28) and a low-risk group (n = 50). RiskScores and survival for each case are shown in Fig. 3A, B. Outcomes for the low-risk group were better than those of the high-risk group. Based on the survival curve, we determined that survival in the low-risk group was significantly better than the high-risk group (Fig. 3C). The median OS was 0.65 years in the high-risk group and 3.85 years in the low-risk group.

Fig. 3
figure 3

Risk assessment model for prognosis prediction: (A and B) risk scores (A) and survival outcome (B) of each case are shown; (C) patients in the low-risk group experienced a longer survival time tested by the Kaplan–Meier test, median OS was 0.65 years (0.58–1.16 years) in the high-risk group and 3.85 years (2.70-NA years) in the low-risk group

Full size image

A Chi-square test was used to calculate correlations between T stage, N stage, M stage, TNM stage and risk score (Fig. 4A and B). We also determined whether T stage, N stage, TNM stage, and risk score could be used as indicators to predict survival. Using univariate Cox regression analysis, we found that TNM stage, M stage, and risk score correlated with survival (Fig. 4C). Multivariate Cox regression analysis showed that only risk score was an independent predictor of survival in patients with esophageal adenocarcinoma (Fig. 4D).

Fig. 4
figure 4

Clinical characteristics evaluation by the risk model: (A) a strip chart aggregating common clinical characteristics; (B) associations between: gender, T stage, N stage, M stage, clinical stage and risk scores, * present p < 0.05; (C) univariate and (D) multivariate Cox regression analyses of risk score and clinical characteristics. Only risk score [p < 0.001, HR = 2.087, 95% CI (1.612–2.700)] presented as an independent prognostic predictor by multivariate Cox regression

Full size image

3.4 Relationship Between Risk Models and Drug Susceptibility

Concurrent chemoradiotherapy is an essential neoadjuvant or radical treatment for locally advanced esophageal cancer, while chemotherapy is the primary treatment for patients with advanced esophageal adenocarcinoma. We hope to improve the accuracy of our prediction model by identifying chemotherapeutic agents that correlate with risk models by selecting sensitive agents. This analysis was performed using the pRRophetic package (Supplementary Table 4). The list of agents includes traditional chemotherapeutic drugs and several small molecule inhibitors (commercially available and otherwise). There was an association between the 50% inhibitory concentrations of camptothecin, cytarabine, methotrexate, and cisplatin and the risk value; however, there was no association with docetaxel, doxorubicin, or paclitaxel (Fig. 5). Small molecule inhibitors against CDK1 (RO. 3306), Bcl-2 (TW37), SRC (KIN001.135), and VEGFR1 (axitinib) were associated with risk score (Supplementary Fig. 1).

Fig. 5
figure 5

The relationship between risk model and chemotherapeutic agent sensitivity: the model acted as a potential predictor for chemosensitivity as high-risk scores were related to were related to a lower IC 50 for (A) Camptothecin, (B) Cisplatin, (C) Cytarabine and (D) Methotrexate; IC50 values of (E) Docetaxel, (F) Doxorubicin and (G) Paclitaxel did not differ between high- and low-risk groups

Full size image

3.5 Relationship Between the Risk Model and Immunoinfiltrated Cells and Genes Related to ICIs

The Checkmate-649 study is a multicenter, phase III, randomized controlled trial of first-line treatment for patients with unresectable locally advanced or advanced gastric, gastroesophageal junction, and esophageal adenocarcinomas [5]. A total of 211 patients with esophageal adenocarcinoma were included. The combination of nivolumab and chemotherapy prolonged the median OS in patients with esophageal adenocarcinoma compared with chemotherapy alone (12.3 m vs. 11.6 m, hazard ratio = 0.81). The success of this study changed the first-line treatment mode of advanced esophageal adenocarcinoma, especially those with CPS scores greater than or equal to 5, highlighting the value of immunotherapy for these patients. Immune infiltrating cells and the expression of genes related to immunosuppression have a potential role in predicting the efficacy of immunotherapy. Therefore, it is vital to explore the relationship between immune infiltrating cells, immunosuppressant-related genes, and the risk model of esophageal adenocarcinoma. We found that M2 macrophages, T cell CD4+ Th2, and T cell CD4+ naive correlated with risk scores (Fig. 6A). We also identified genes linked to ICI efficacy by consulting the literature (Supplementary Table 5) and analyzed the correlation between expression levels of these genes and the risk model. We found that the expression levels of TNF, IL23A, CANX, CXCL5, and DNMT1 were lower in the low-risk group (Fig. 6B).

Fig. 6
figure 6

Estimation of tumor-infiltrating cells and immune checkpoint inhibitor-related molecules by the risk model: (A) patients in the high-risk group were more positively associated with tumor-infiltrating immune cells such as common lymphoid progenitor, neutrophil, T cell CD4 + Th2, mast cell resting and eosinophil, whereas they were negatively associated with T cell CD4 + naïve, common myeloid progenitor, macrophage M2, B cell plasma and mast cell activated, as shown by Spearman correlation analysis. B The expression level of TNF, IL23A, CANX, CXCL5 and DNMT1 in the low-risk group was lower than in the high-risk group (* represented p < 0.05, *** represented p < 0.001)

Full size image

4 Discussion

The OS in esophageal adenocarcinoma is relatively poor because its signs and symptoms are vague and non-specific, and the tumor is invasive. Patients with early-stage esophageal adenocarcinoma may have the opportunity of cure with endoscopy, surgery, or concurrent radiotherapy and chemotherapy. However, most patients have distal metastasis or local adjacent organ involvement at the time of diagnosis. These patients often do not have the opportunity to undergo radical treatment. Currently, the most crucial outcome predictor is the TNM stage. The 5-year survival was 80.5% in patients with the least proportion of stage I. Five-year survival decreased with a progressive increase in the stage. For stages II, III, and IV, survival rates were 45.1%, 17.6%, and 2.1%, respectively [1]. Traditional TNM, which relies solely on anatomical information derived from imaging, does not incorporate any biological features of the tumor and, therefore, cannot distinguish between indolent and aggressive tumors. Imaging cannot predict the sensitivity to chemotherapy agents. In the era of precision cancer treatment, imaging’s role in judging the prognosis and guiding the choice of treatment options fails to meet clinicians’ needs. By contrast, biomarkers (especially those with prognostic value), as a complement to tumor pathology classification of tumor and TNM staging in clinical prognosis, may prove more beneficial for diagnosis, treatment, and outcome prediction. The Checkmate-649 study showed that nivolumab combined with chemotherapy improved OS and objective response rates in the overall population. Subgroup analysis showed that the OS hazard ratio was 0.69 in patients with a CPS of 5 or higher verse 0.94 for patients with a CPS less than 5 [5]. The study results reflect the therapeutic value of ICIs in patients with advanced esophageal adenocarcinoma. The biomarker PD-L1 might also predict the efficacy of nivolumab in esophageal adenocarcinoma. In the era of cancer immunotherapy, we hope to construct an individualized outcome prediction model that can accurately predict survival and guide the formation of individualized treatment strategies.

LncRNAs are critical for post-transcriptional regulation of gene expression. They regulate gene expression through direct binding to critical sites of mRNAs and interfere with the regulation of mRNAs expression by miRNAs. Our previous study showed that lncRNA-based ceRNA networks could predict survival in patients with esophageal adenocarcinoma [18]. Several lines of evidence suggest that lncRNAs regulate the expression of immune genes and function. Precise regulation of immune gene expression is essential for complete immune function. Outcome-related irlncRNA signatures have been described in tumors such as liver, bladder, and colon cancer [19,20,21]. These irlncRNAs can be used as biomarkers for outcomes in cancer patients. The potential usefulness of the irlncRNA risk score model as a valid response and outcome predictor needs to be validated in patients with esophageal adenocarcinoma [22]. Gao X et al. report screening of tumor grade-related mRNAs and lncRNAs for Esophagus Squamous Cell Carcinoma, and they found that a total of 1864 tumor grade-related mRNAs (846 positively related and 1018 negatively related) and 552 tumor grade-related lncRNAs (331 positively related and 221 negatively related) were obtained. However, there are no reports of valid DEirlncRNA pairs in esophageal adenocarcinoma. This study is innovative. We constructed a risk model consisting of 11 irlncRNA pairs in the present study. Some lncRNAs in our risk model were considered oncogenic factors. Yang et al. found that GK-IT1 expression was higher in ESCC (esophageal squamous cell cancer) than adjacent normal tissues and positively correlated with the clinical stage and shorter survival [23]. Shen et al. showed that GASAL1 displayed a high expression in HCC (hepatocellular carcinoma) cells, and knockdown of GASAL1 led to inhibited cell proliferation, migration, and progression [24]. Wu et al. demonstrated that expression of DLEU2 was elevated in non-small cell lung cancer tissues, and DLEU2 promoted the tumorigenesis and invasion of non-small-cell lung cancer by sponging miR-30a-5p [25]. These lncRNAs are all related to genes mediating immune function; therefore, these lncRNAs may also affect the function of tumor-infiltrating cells. The MMP25-AS1/hsa-miR-10a-5p/SERPINE1 axis might affect the changes in the tumor immune microenvironment and the development of kidney renal clear cell carcinoma by affecting the expression of chemokines (CCL4, CCL5, CXCL13, and XCL2) [26]. The 1-year ROC curve in the present study reflects the excellent sensitivity and specificity of the risk model, with an AUC of 0.942. To validate the model’s predictive value for survival prognosis, we performed a risk stratification survival curve based on the model and found that the median OS of the high-risk population was 0.65 years, while it was 3.85 years for the low-risk population (P < 0.05). Multivariate Cox regression analysis showed that the risk score calculated based on the risk model was an independent predictor of OS.

Immune infiltrating cells are an essential component of the tumor microenvironment, and their distribution in the tumor microenvironment directly reflects the immune state of the tumor. The type and number of infiltrating immune cells in the tumor microenvironment are believed to reflect the effect of immunosuppressive therapy. The Keynote-001 study showed that tumors with more infiltration of CD8+ T cells responded better to pembrolizumab treatment than tumors with less infiltration. Some infiltrating immune cells are resistant to ICIs [27]. Recently, tumor-associated macrophages were demonstrated to directly limit PD-1 blockade by removing anti-PD-1 antibodies from PD-1+ CD8+ T cells in an FcgR-dependent manner [28]. Considering that ICIs play an increasingly important role in esophageal adenocarcinoma, we explored whether there is a difference in the distribution of immune infiltrating cells between high- and low-risk populations. We used seven well-established methods to predict tumor-infiltrating cells and found that CD4+ Th2 cells positively correlated with risk scores. Tomoyuki et al. reported that the percentage of CD4+ Th2 cells in peripheral blood of patients with gastrointestinal tumors was significantly greater than that of healthy people. Difference analysis showed that the proportion of CD4+ Th2 cells in high-risk populations was higher than in low-risk populations (Supplementary Fig. 2A). Therefore, we believe that CD4+ Th2 cells negatively regulate the microenvironment of esophageal adenocarcinoma. In addition, CD4+ naive cell was negatively correlated with risk score; differentiation analysis showed that their proportion in the low-risk population was higher (Supplementary Fig. 2B). Su et al. found that the number of Treg cells in the microenvironment of breast cancer was differentiated from CD4+ naive cells [29]. By reducing the expression of PITPNM3 and blocking the aggregation of CD4+ naive cells in the tumor microenvironment, the number of Treg cells in the tumor microenvironment can be substantially reduced, and proliferation of tumors can be inhibited. The role of CD4+ naive cells in the microenvironment of esophageal adenocarcinoma has yet to be explored.

We analyzed the expression levels of immunomodulatory genes in tumor tissues with risk stratifications and found that expression levels of TNF, IL23A, CANX, CXCL5, and DNMT1 were lower in low-risk patients. Zheng et al. showed that outcome positively correlated with expression of CANX in colon cancer. Inhibition of mir-148a-3p increased the expression of CANX, which increased the expression of MHC-1 on the cell surface and considerably enhanced the immune attack mediated by CD8+ cells [30]. Deng et al. found that patients with non-small cell lung cancer with low expression levels of CXCL5 responded more to ICIs than those with high levels [31]. Huang et al. found that the immunomodulatory agent DNA methyltransferase inhibitor (DNMTi) led to demethylation of the PD-L1 promoter and increased radiotherapy-induced PD-L1 upregulation via interferon β, and DNMTi strongly enhanced the response to combined treatment with radiotherapy and anti-PD-L1 inhibitors [32]. These studies show that the genes we have determined related to risk models are associated with immunotherapy effectiveness. We believe that combining our risk model and these gene expression levels could predict the OS in esophageal adenocarcinoma accurately and have potential value in predicting ICI efficacy.

This study has several limitations. The seven-irlncRNA pair signature was only validated in TCGA. More patient datasets from several cancer centers are needed to verify the prognostic performance of the 11-irlncRNA pair signature. These valid DEirlncRNA pairs still need to be verified by clinical trials, and how to guide the clinical treatment of esophageal adenocarcinoma to prolong the survival period still needs a lot of in-depth research.

5 Conclusion

We identified a novel prognostic immune-related lncRNA pair signature for esophageal adenocarcinoma. We also found that the risk score-based groups had different immune statuses, drug sensitivity, and immune-mediated gene expression. These findings may offer insights into the prognostic evaluation of esophageal adenocarcinoma and create a basis for personalized treatment plans.