Pancreatic cancer is one of the deadliest tumors in the world with a poor prognosis. Therefore, an accurate prediction model is needed to identify patients at high risk of pancreatic cancer to tailor treatment and improve the prognosis of these patients.
We obtained The Cancer Genome Atlas (TCGA) pancreatic adenocarcinoma (PAAD) RNAseq data from the UCSC Xena database, identified immune-related lncRNAs (irlncRNAs) through correlation analysis, and identified differences between TCGA and normal pancreatic adenocarcinoma tissues. DEirlncRNA) from TCGA and genotype tissue expression (GTEx) of pancreatic tissue. Further univariate and lasso regression analyzes were performed to construct prognostic signature models. We then calculated the area under the curve and determined the optimal cutoff value for identifying patients with high- and low-risk pancreatic adenocarcinoma. To compare clinical characteristics, immune cell infiltration, immunosuppressive microenvironment, and chemotherapy resistance in patients with high- and low-risk pancreatic cancer.
We identified 20 DEirlncRNA pairs and grouped patients according to the optimal cutoff value. We demonstrated that our prognostic signature model has significant performance in predicting the prognosis of patients with PAAD. The AUC of the ROC curve is 0.905 for the 1-year forecast, 0.942 for the 2-year forecast, and 0.966 for the 3-year forecast. High-risk patients had lower survival rates and worse clinical characteristics. We also demonstrated that high-risk patients are immunosuppressed and may develop resistance to immunotherapy. Evaluation of anticancer drugs such as paclitaxel, sorafenib, and erlotinib based on computational prediction tools may be appropriate for high-risk patients with PAAD.
Overall, our study established a new prognostic risk model based on paired irlncRNA, which showed promising prognostic value in patients with pancreatic cancer. Our prognostic risk model may help differentiate patients with PAAD who are suitable for medical treatment.
Pancreatic cancer is a malignant tumor with a low five-year survival rate and high grade. At the time of diagnosis, most patients are already in advanced stages. In the context of the COVID-19 epidemic, doctors and nurses are under enormous pressure when treating patients with pancreatic cancer, and patients’ families also face multiple pressures when making treatment decisions [1, 2]. Although great advances have been made in the treatment of DOADs, such as neoadjuvant therapy, surgical resection, radiation therapy, chemotherapy, targeted molecular therapy, and immune checkpoint inhibitors (ICIs), only about 9% of patients survive five years after diagnosis [3]. ], 4]. Because the early symptoms of pancreatic adenocarcinoma are atypical, patients are usually diagnosed with metastases at an advanced stage [5]. Therefore, for a given patient, individualized comprehensive treatment must weigh the advantages and disadvantages of all treatment options, not only to prolong survival, but also to improve quality of life [6]. Therefore, an effective prediction model is necessary to accurately assess a patient’s prognosis [7]. Thus, appropriate treatment can be selected to balance the survival and quality of life of patients with PAAD.
The poor prognosis of PAAD is mainly due to resistance to chemotherapy drugs. In recent years, immune checkpoint inhibitors have been widely used in the treatment of solid tumors [8]. However, the use of ICIs in pancreatic cancer is rarely successful [9]. Therefore, it is important to identify patients who may benefit from ICI therapy.
Long non-coding RNA (lncRNA) is a type of non-coding RNA with transcripts >200 nucleotides. LncRNAs are widespread and constitute about 80% of the human transcriptome [10]. A large body of work has shown that lncRNA-based prognostic models can effectively predict patient prognosis [11, 12]. For example, 18 autophagy-related lncRNAs were identified to generate prognostic signatures in breast cancer [13]. Six other immune-related lncRNAs have been used to establish the prognostic features of glioma [14].
In pancreatic cancer, some studies have established lncRNA-based signatures to predict patient prognosis. A 3-lncRNA signature was established in pancreatic adenocarcinoma with an area under the ROC curve (AUC) of only 0.742 and an overall survival (OS) of 3 years [15]. In addition, lncRNA expression values vary among different genomes, different data formats, and different patients, and the performance of the predictive model is unstable. Therefore, we use a novel modeling algorithm, pairing and iteration, to generate immunity-related lncRNA (irlncRNA) signatures to create a more accurate and stable predictive model [8].
Normalized RNAseq data (FPKM) and clinical pancreatic cancer TCGA and genotype tissue expression (GTEx) data were obtained from the UCSC XENA database ( https://xenabrowser.net/datapages/ ). GTF files were obtained from the Ensembl database ( http://asia.ensembl.org ) and used to extract lncRNA expression profiles from RNAseq. We downloaded immunity-related genes from the ImmPort database (http://www.immport.org) and identified immunity-related lncRNAs (irlncRNAs) using correlation analysis (p < 0.001, r > 0.4). Identification of differentially expressed irlncRNAs (DEirlncRNAs) by crossing irlncRNAs and differentially expressed lncRNAs obtained from the GEPIA2 database (http://gepia2.cancer-pku.cn/#index) in the TCGA-PAAD cohort (|logFC| > 1 and FDR) <0.05).
This method has been reported previously [8]. Specifically, we construct X to replace the paired lncRNA A and lncRNA B. When the expression value of lncRNA A is higher than the expression value of lncRNA B, X is defined as 1, otherwise X is defined as 0. Therefore, we can obtain a matrix of 0 or – 1. The vertical axis of the matrix represents each sample, and the horizontal axis represents each DEirlncRNA pair with a value of 0 or 1.
Univariate regression analysis followed by Lasso regression was used to screen prognostic DEirlncRNA pairs. The lasso regression analysis used 10-fold cross-validation repeated 1000 times ( p < 0.05), with 1000 random stimuli per run. When the frequency of each DEirlncRNA pair exceeded 100 times in 1000 cycles, DEirlncRNA pairs were selected to construct a prognostic risk model. We then used the AUC curve to find the optimal cutoff value for classifying PAAD patients into high- and low-risk groups. The AUC value of each model was also calculated and plotted as a curve. If the curve reaches the highest point indicating the maximum AUC value, the calculation process stops and the model is considered the best candidate. 1-, 3- and 5-year ROC curve models were constructed. Univariate and multivariate regression analyzes were used to examine the independent predictive performance of the prognostic risk model.
Use seven tools to study immune cell infiltration rates, including XCELL, TIMER, QUANTISEQ, MCPCOUNTER, EPIC, CIBERSORT-ABS, and CIBERSORT. Immune cell infiltration data were downloaded from the TIMER2 database (http://timer.comp-genomics.org/#tab-5817-3). The difference in the content of immune-infiltrating cells between the high- and low-risk groups of the constructed model was analyzed using the Wilcoxon signed-rank test, the results are shown in the square graph. Spearman correlation analysis was performed to analyze the relationship between risk score values and immune-infiltrating cells. The resulting correlation coefficient is shown as a lollipop. The significance threshold was set at p < 0.05. The procedure was performed using the R package ggplot2. To examine the relationship between the model and gene expression levels associated with the immune cell infiltration rate, we performed the ggstatsplot package and violin plot visualization.
To evaluate clinical treatment patterns for pancreatic cancer, we calculated the IC50 of commonly used chemotherapy drugs in the TCGA-PAAD cohort. Differences in half inhibitory concentrations (IC50) between high- and low-risk groups were compared using the Wilcoxon signed-rank test, and the results are shown as boxplots generated using pRRophetic and ggplot2 in R. All methods comply with relevant guidelines and norms.
The workflow of our study is shown in Figure 1. Using correlation analysis between lncRNAs and immunity-related genes, we selected 724 irlncRNAs with p < 0.01 and r > 0.4. We next analyzed the differentially expressed lncRNAs of GEPIA2 (Figure 2A). A total of 223 irlncRNAs were differentially expressed between pancreatic adenocarcinoma and normal pancreatic tissue (|logFC| > 1, FDR < 0.05), named DEirlncRNAs.
Construction of predictive risk models. (A) Volcano plot of differentially expressed lncRNAs. (B) Distribution of lasso coefficients for 20 DEirlncRNA pairs. (C) Partial likelihood variance of the LASSO coefficient distribution. (D) Forest plot showing univariate regression analysis of 20 DEirlncRNA pairs.
We next constructed a 0 or 1 matrix by pairing 223 DEirlncRNAs. A total of 13,687 DEirlncRNA pairs were identified. After univariate and lasso regression analysis, 20 DEirlncRNA pairs were finally tested to construct a prognostic risk model (Figure 2B-D). Based on the results of Lasso and multiple regression analysis, we calculated a risk score for each patient in the TCGA-PAAD cohort (Table 1). Based on the results of lasso regression analysis, we calculated a risk score for each patient in the TCGA-PAAD cohort. The AUC of the ROC curve was 0.905 for the 1-year risk model prediction, 0.942 for the 2-year prediction, and 0.966 for the 3-year prediction (Figure 3A-B). We set an optimal cutoff value of 3.105, stratified the TCGA-PAAD cohort patients into high- and low-risk groups, and plotted the survival outcomes and risk score distributions for each patient (Figure 3C-E). Kaplan-Meier analysis showed that survival of PAAD patients in the high-risk group was significantly lower than that of patients in the low-risk group (p < 0.001) (Figure 3F).
Validity of prognostic risk models. (A) ROC of the prognostic risk model. (B) 1-, 2-, and 3-year ROC prognostic risk models. (C) ROC of prognostic risk model. Shows the optimal cut-off point. (DE) Distribution of survival status (D) and risk scores (E). (F) Kaplan-Meier analysis of PAAD patients in high- and low-risk groups.
We further assessed differences in risk scores by clinical characteristics. The strip plot (Figure 4A) shows the overall relationship between clinical characteristics and risk scores. In particular, older patients had higher risk scores (Figure 4B). In addition, patients with stage II had higher risk scores than patients with stage I (Figure 4C). Regarding tumor grade in PAAD patients, grade 3 patients had higher risk scores than grade 1 and 2 patients (Figure 4D). We further performed univariate and multivariate regression analyzes and demonstrated that risk score (p < 0.001) and age (p = 0.045) were independent prognostic factors in patients with PAAD (Figure 5A-B). The ROC curve demonstrated that the risk score was superior to other clinical characteristics in predicting 1-, 2-, and 3-year survival of patients with PAAD (Figure 5C-E).
Clinical characteristics of prognostic risk models. Histogram (A) shows (B) age, (C) tumor stage, (D) tumor grade, risk score, and gender of patients in the TCGA-PAAD cohort. **p < 0.01
Independent predictive analysis of prognostic risk models. (AB) Univariate (A) and multivariate (B) regression analyzes of prognostic risk models and clinical characteristics. (CE) 1-, 2-, and 3-year ROC for prognostic risk models and clinical characteristics
Therefore, we examined the relationship between time and risk scores. We found that risk score in PAAD patients was inversely correlated with CD8+ T cells and NK cells (Figure 6A), indicating suppressed immune function in the high-risk group. We also assessed the difference in immune cell infiltration between the high- and low-risk groups and found the same results (Figure 7). There was less infiltration of CD8+ T cells and NK cells in the high-risk group. In recent years, immune checkpoint inhibitors (ICIs) have been widely used in the treatment of solid tumors. However, the use of ICIs in pancreatic cancer has rarely been successful. Therefore, we assessed the expression of immune checkpoint genes in high- and low-risk groups. We found that CTLA-4 and CD161 (KLRB1) were overexpressed in the low-risk group (Figure 6B-G), indicating that PAAD patients in the low-risk group may be sensitive to ICI.
Correlation analysis of prognostic risk model and immune cell infiltration. (A) Correlation between prognostic risk model and immune cell infiltration. (BG) Indicates gene expression in high and low risk groups. (HK) IC50 values for specific anticancer drugs in high and low risk groups. *p < 0.05, **p < 0.01, ns = not significant
We further assessed the association between risk scores and common chemotherapy agents in the TCGA-PAAD cohort. We searched for commonly used anticancer drugs in pancreatic cancer and analyzed differences in their IC50 values between high- and low-risk groups. The results showed that the IC50 value of AZD.2281 (olaparib) was higher in the high-risk group, indicating that PAAD patients in the high-risk group may be resistant to AZD.2281 treatment (Figure 6H). In addition, the IC50 values of paclitaxel, sorafenib, and erlotinib were lower in the high-risk group (Figure 6I-K). We further identified 34 anticancer drugs with higher IC50 values in the high-risk group and 34 anticancer drugs with lower IC50 values in the high-risk group (Table 2).
It cannot be denied that lncRNAs, mRNAs, and miRNAs widely exist and play a crucial role in cancer development. There is ample evidence supporting the important role of mRNA or miRNA in predicting overall survival in several types of cancer. Undoubtedly, many prognostic risk models are also based on lncRNAs. For example, Luo et al. Studies have shown that LINC01094 plays a key role in PC proliferation and metastasis, and high expression of LINC01094 indicates poor survival of pancreatic cancer patients [16]. The study presented by Lin et al. Studies have shown that downregulation of lncRNA FLVCR1-AS1 is associated with poor prognosis in pancreatic cancer patients [17]. However, immunity-related lncRNAs are relatively less discussed in terms of predicting overall survival of cancer patients. Recently, a large amount of work has been focused on building prognostic risk models to predict the survival of cancer patients and thereby adjust treatment methods [18, 19, 20]. There is growing recognition of the important role of immune infiltrates in cancer initiation, progression, and response to treatments such as chemotherapy. Numerous studies have confirmed that tumor-infiltrating immune cells play a critical role in the response to cytotoxic chemotherapy [21, 22, 23]. The tumor immune microenvironment is an important factor in the survival of tumor patients [24, 25]. Immunotherapy, especially ICI therapy, is widely used in the treatment of solid tumors [26]. Immune-related genes are widely used to construct prognostic risk models. For example, Su et al. The immune-related prognostic risk model is based on protein-coding genes to predict the prognosis of ovarian cancer patients [27]. Non-coding genes such as lncRNAs are also suitable for constructing prognostic risk models [28, 29, 30]. Luo et al tested four immune-related lncRNAs and built a predictive model for cervical cancer risk [31]. Khan et al. A total of 32 differentially expressed transcripts were identified, and based on this, a prediction model with 5 significant transcripts was established, which was proposed as a highly recommended tool for predicting biopsy-proven acute rejection after kidney transplantation [32].
Most of these models are based on gene expression levels, either protein-coding genes or non-coding genes. However, the same gene can have different expression values in different genomes, data formats and in different patients, leading to unstable estimates in predictive models. In this study, we built a reasonable model with two pairs of lncRNAs, independent of the exact expression values.
In this study, we identified irlncRNA for the first time through correlation analysis with immunity-related genes. We screened 223 DEirlncRNAs by hybridization with differentially expressed lncRNAs. Second, we constructed a 0-or-1 matrix based on the published DEirlncRNA pairing method [31]. We then performed univariate and lasso regression analyzes to identify prognostic DEirlncRNA pairs and construct a predictive risk model. We further analyzed the association between risk scores and clinical characteristics in patients with PAAD. We found that our prognostic risk model, as an independent prognostic factor in PAAD patients, can effectively differentiate high-grade patients from low-grade patients and high-grade patients from low-grade patients. In addition, the AUC values of the ROC curve of the prognostic risk model were 0.905 for the 1-year forecast, 0.942 for the 2-year forecast, and 0.966 for the 3-year forecast.
Researchers reported that patients with higher CD8+ T cell infiltration were more sensitive to ICI treatment [33]. An increase in the content of cytotoxic cells, CD56 NK cells, NK cells and CD8+ T cells in the tumor immune microenvironment may be one of the reasons for the tumor suppressive effect [34]. Previous studies showed that higher levels of tumor-infiltrating CD4(+) T and CD8(+) T were significantly associated with longer survival [35]. Poor CD8 T cell infiltration, low neoantigen load, and a highly immunosuppressive tumor microenvironment lead to lack of response to ICI therapy [36]. We found that risk score was negatively correlated with CD8+ T cells and NK cells, indicating that patients with high risk scores may not be suitable for ICI treatment and have a worse prognosis.
CD161 is a marker of natural killer (NK) cells. CD8+CD161+ CAR-transduced T cells mediate enhanced in vivo antitumor efficacy in HER2+ pancreatic ductal adenocarcinoma xenograft models [37]. Immune checkpoint inhibitors target cytotoxic T lymphocyte associated protein 4 (CTLA-4) and programmed cell death protein 1 (PD-1)/programmed cell death ligand 1 (PD-L1) pathways and have great potential in many areas. Expression of CTLA-4 and CD161 (KLRB1) is lower in high-risk groups, further indicating that patients with high-risk scores may not be eligible for ICI treatment. [38]
To find treatment options suitable for high-risk patients, we analyzed various anticancer drugs and found that paclitaxel, sorafenib, and erlotinib, which are widely used in patients with PAAD, may be suitable for high-risk patients with PAAD. [33]. Zhang et al found that mutations in any DNA damage response (DDR) pathway can lead to poor prognosis in prostate cancer patients [39]. The Pancreatic Cancer Olaparib Ongoing (POLO) trial showed that maintenance treatment with olaparib prolonged progression-free survival compared with placebo after first-line platinum-based chemotherapy in patients with pancreatic ductal adenocarcinoma and germline BRCA1/2 mutations [40]. This provides significant optimism that treatment outcomes will improve significantly in this subgroup of patients. In this study, the IC50 value of AZD.2281 (olaparib) was higher in the high-risk group, indicating that PAAD patients in the high-risk group may be resistant to treatment with AZD.2281.
The forecasting models in this study produce good forecasting results, but they are based on analytical forecasts. How to confirm these results with clinical data is an important question. Endoscopic fine needle aspiration ultrasonography (EUS-FNA) has become an indispensable method for diagnosing solid and extrapancreatic pancreatic lesions with a sensitivity of 85% and specificity of 98% [41]. The advent of EUS fine-needle biopsy (EUS-FNB) needles is mainly based on perceived advantages over FNA, such as higher diagnostic accuracy, obtaining samples that preserve histological structure, and thus generating immune tissue that is critical for certain diagnoses . special staining [42]. A systematic review of the literature confirmed that FNB needles (especially 22G) demonstrate the highest efficiency in harvesting tissue from pancreatic masses [43]. Clinically, only a small number of patients are eligible for radical surgery, and most patients have inoperable tumors at the time of initial diagnosis. In clinical practice, only a small proportion of patients are suitable for radical surgery because most patients have inoperable tumors at the time of initial diagnosis. After pathological confirmation by EUS-FNB and other methods, standardized non-surgical treatment such as chemotherapy is usually chosen. Our subsequent research program is to test the prognostic model of this study in surgical and nonsurgical cohorts through a retrospective analysis.
Overall, our study established a new prognostic risk model based on paired irlncRNA, which showed promising prognostic value in patients with pancreatic cancer. Our prognostic risk model may help differentiate patients with PAAD who are suitable for medical treatment.
The datasets used and analyzed in the current study are available from the corresponding author on reasonable request.
Sui Wen, Gong X, Zhuang Y. The mediating role of self-efficacy in emotional regulation of negative emotions during the COVID-19 pandemic: a cross-sectional study. Int J Ment Health Nurs [journal article]. 2021 06/01/2021;30(3):759–71.
Sui Wen, Gong X, Qiao X, Zhang L, Cheng J, Dong J, et al. Family members’ views on alternative decision making in intensive care units: a systematic review. INT J NURS STUD [magazine article; review]. 2023 01/01/2023;137:104391.
Vincent A, Herman J, Schulich R, Hruban RH, Goggins M. Pancreatic cancer. Lancet. [Journal article; research support, NIH, extramural; research support, government outside the US; review]. 2011 08/13/2011;378(9791):607–20.
Ilic M, Ilic I. Epidemiology of pancreatic cancer. World Journal of Gastroenterology. [Journal article, review]. 2016 11/28/2016;22(44):9694–705.
Liu X, Chen B, Chen J, Sun S. A new tp53-related nomogram for predicting overall survival in patients with pancreatic cancer. BMC Cancer [journal article]. 2021 31-03-2021;21(1):335.
Xian X, Zhu X, Chen Y, Huang B, Xiang W. Effect of solution-focused therapy on cancer-related fatigue in colorectal cancer patients receiving chemotherapy: a randomized controlled trial. Cancer nurse. [Journal article; randomized controlled trial; the study is supported by a government outside the United States]. 2022 05/01/2022;45(3):E663–73.
Zhang Cheng, Zheng Wen, Lu Y, Shan L, Xu Dong, Pan Y, et al. Postoperative carcinoembryonic antigen (CEA) levels predict outcome after colorectal cancer resection in patients with normal preoperative CEA levels. Center for Translational Cancer Research. [Journal article]. 2020 01.01.2020;9(1):111–8.
Hong Wen, Liang Li, Gu Yu, Qi Zi, Qiu Hua, Yang X, et al. Immune-related lncRNAs generate novel signatures and predict the immune landscape of human hepatocellular carcinoma. Mol Ther Nucleic acids [Journal article]. 2020 2020-12-04;22:937 – 47.
Toffey R.J., Zhu Y., Schulich R.D. Immunotherapy for pancreatic cancer: barriers and breakthroughs. Ann Gastrointestinal Surgeon [Journal Article; review]. 2018 07/01/2018;2(4):274–81.
Hull R, Mbita Z, Dlamini Z. Long non-coding RNAs (LncRNAs), viral tumor genomics and aberrant splicing events: therapeutic implications. AM J CANCER RES [journal article; review]. 2021 01/20/2021;11(3):866–83.
Wang J, Chen P, Zhang Y, Ding J, Yang Y, Li H. 11-Identification of lncRNA signatures associated with endometrial cancer prognosis. Achievements of science [magazine article]. 2021 2021-01-01;104(1):311977089.
Jiang S, Ren H, Liu S, Lu Z, Xu A, Qin S, et al. Comprehensive analysis of RNA-binding protein prognostic genes and drug candidates in papillary cell renal cell carcinoma. pregen. [Journal article]. 2021 01/20/2021;12:627508.
Li X, Chen J, Yu Q, Huang X, Liu Z, Wang X, et al. Characteristics of autophagy-related long non-coding RNA predict breast cancer prognosis. pregen. [Journal article]. 2021 01/20/2021;12:569318.
Zhou M, Zhang Z, Zhao X, Bao S, Cheng L, Sun J. Immune-related six lncRNA signature improves prognosis in glioblastoma multiforme. MOL Neurobiology. [Journal article]. 2018 01.05.2018;55(5):3684–97.
Wu B, Wang Q, Fei J, Bao Y, Wang X, Song Z, et al. A novel tri-lncRNA signature predicts survival of patients with pancreatic cancer. REPRESENTATIVES OF ONKOL. [Journal article]. 2018 12/01/2018;40(6):3427–37.
Luo C, Lin K, Hu C, Zhu X, Zhu J, Zhu Z. LINC01094 promotes pancreatic cancer progression by regulating LIN28B expression and the PI3K/AKT pathway through sponged miR-577. Mol Therapeutics – Nucleic acids. 2021;26:523–35.
Lin J, Zhai X, Zou S, Xu Z, Zhang J, Jiang L, et al. Positive feedback between lncRNA FLVCR1-AS1 and KLF10 may inhibit pancreatic cancer progression through the PTEN/AKT pathway. J EXP Clin Cancer Res. 2021;40(1).
Zhou X, Liu X, Zeng X, Wu D, Liu L. Identification of thirteen genes predicting overall survival in hepatocellular carcinoma. Biosci Rep [journal article]. 2021 04/09/2021.
Post time: Sep-22-2023