Pancreatic cancer(PC) is a highly invasive digestive system tumor, with a 5-year survival rate of less than 10%. However, the mechanism of its occurrence remains unclear. Metabolic alterations in malignant tumor cells have been found to occur throughout the processes of uptake, synthesis, and degradation, making metabolic reprogramming one of the characteristics of malignant tumors. Disturbances in lipid metabolism have also been observed in pancreatic cancer cells. To explore the relationship between lipid metabolism and the survival of pancreatic cancer patients, we constructed a prognostic analysis model related to lipid metabolism using public databases such as TCGA and GEO. This model includes two risk genes (PLAAT2 and PTGES) and five protective genes (PEMT, CYP46A1, LTC45, TMEM86B, and LIPE). We validated the model using a validation dataset and found that it had good predictive performance. We also discovered distinct differences in immune levels and drug susceptibility between the high and low-risk groups. In summary, this model is helpful for predicting the survival of pancreatic cancer patients and provides insights into the identification of new therapeutic targets for pancreatic cancer.
Research Article
Metabolic Reprogramming：Application of lipid metabolism related genes in the risk model of clinical prognosis of pancreatic cancer
https://doi.org/10.21203/rs.3.rs-3744911/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
metabolic reprogramming
lipid metabolism
pancreatic cancer
risk score
prognosis
Pancreatic cancer is a malignant tumor of digestive system with hidden onset, rapid progression, extremely high degree of malignancy and poor prognosis. Although considerable achievements have been made in the treatment of pancreatic cancer in recent years, its prognosis has not been significantly improved, and its 5-year survival rate is still less than 10%[1]. According to The analysis of SEER (The Surveillance, Epidemiology and End Results) database of the National Cancer Institute of the United States, only about 20% of pancreatic cancer cases can be treated with comprehensive treatment focusing on surgery[2], and the median survival time can reach 30 to 40 months[3]. More than 50% of pancreatic cancer patients have distant metastasis at the first diagnosis and lose the opportunity for surgery, and the median survival is only 10 to 13 months[4], but non-surgical treatment is also faced with difficulties. With the application of bioinformation technology in clinical medicine more and more widely, the biomarkers related to the prognosis of pancreatic cancer are still insufficient[5].
Metabolic reprogramming is a recognized hallmark of malignant tumors, where cancer cells adapt their metabolic state to different stages of their survival environment in order to achieve biological objectives such as proliferation, invasion, migration, and colonization[6–8]. This reprogramming can also disrupt anti-tumor immunity by influencing organismal metabolism[9]. Lipids are key components of cell membranes and are involved in the processing of translated proteins. Previous studies have discovered significant disruptions in lipid metabolism in pancreatic cancer cells[10], Additionally, emerging treatment approaches based on targeting pancreatic cancer metabolic pathways have shown promising results[11, 12]. Therefore, metabolic therapy holds potential as a novel avenue for pancreatic cancer treatment.
This study utilized bioinformatics analysis tools to examine pancreatic cancer-related data from public databases. The aim was to analyze the relationship between gene expression characteristics and clinical features, identify lipid metabolism-related genes associated with prognosis, establish a survival prediction model for pancreatic cancer patients, and identify potential targets for metabolic therapy.
2.1 Data acquisition and preprocessing
The research process diagram is shown in Fig. 1. The TCGA-PAAD dataset (n = 182) was downloaded from The Cancer Genome Atlas database (TCGA, https://www.cancer.gov/ccg/research/genome-sequencing/tcga). The dataset of normal pancreatic tissue (n = 167) was downloaded from the Genotype-Tissue Expression database (GTEx, https://xenabrowser.net/). After classification, there were 178 pancreatic cancer data and 171 normal tissue data. The GSE57495 dataset (n = 63) was downloaded from the Gene Expression Omnibus database (GEO, https://www.ncbi.nlm.nih.gov/geo/). The PACA-AU and PACA-CA datasets (n = 273) were downloaded from the International Cancer Genome Consortium (ICGC, https://dcc.icgc.org/) database. All the mentioned datasets were downloaded simultaneously with RNA-seq data and clinical data for further research.
A set of lipid metabolism-related genes (LMRGs, n = 742) was obtained from the Molecular Signature Database (MsigDB, https://ngdc.cncb.ac.cn/databasecommons/database/id/1077) (REACTOME_METABOLISM_OF_LIPIDS).
The TCGA-PAAD and GTEx datasets were merged using the R software package inSilicoMerging. Batch effects were removed using the method described by Johnson et al.[13] to obtain a batch-effect-free matrix, which was used as the training set. The PACA-AU and PACA-CA datasets from the ICGC database were processed in the same manner.
2.2 Screening for differentially expressed genes
The R software package limma (version 3.40.6) was utilized to perform differential analysis on the training dataset, resulting in the identification of differential expression genes (DEGs). Specifically, 1133 upregulated genes and 1098 downregulated genes were found (FC = 1.5, FDR < 0.01) as shown in Fig. 2A. By intersecting the overall DEGs with LMRGs, a total of 72 lipid metabolism-related differential expression genes (LMRDEGs) were obtained, as illustrated in Fig. 2B. The expression heatmap for LMRDEGs can be seen in Fig. 2D. To visualize the gene-gene interaction network among LMRDEGs, the String website (https://cn.string-db.org/) was employed, and the resulting network is displayed in Fig. 2C.
2.3 Functional enrichment analysis of differential genes
We further analyzed these 72 differentially expressed LMRGs using the GO and KEGG pathways to explore their functions. Our analysis revealed that the differentially expressed LMRGs mainly exhibited catalytic activity in pathways related to lipid metabolism, with a predominant role in organelles such as the endoplasmic reticulum (as shown in Fig. 3A, 3B, 3C, 3D). This finding is consistent with the widely recognized dysregulation of lipid metabolism in pancreatic cancer cells[9].
2.4 Construction of prognostic risk model
In this study, we utilized the R software package glmnet to integrate survival time, survival status, and gene expression data. We employed the least absolute shrinkage and selection operator (LASSO) Cox regression model to evaluate differential expression of LMRGs in the training set. Additionally, a 10-fold cross-validation was performed to identify the key genes most relevant to pancreatic cancer prognosis (Table 1). lambda value was set to 0.115585430868722, resulting in the final model consisting of 7 genes (Fig. 4A, 4B). The formula for this model is as follows:
$$\text{R}\text{i}\text{s}\text{k}\text{S}\text{c}\text{o}\text{r}\text{e}={\sum }_{i=1}^{n}(Coef \dot{i}\times Gene i)$$
Coef i represents the coefficient of gene i, while Gene i denotes the transcript per million (TPM) value of gene i.
2.5 Evaluation and validation of prognostic risk models
The optimal cutoff value for RiskScore in the training group was calculated using the maxstat package in R. Based on this, patients were divided into high and low groups. Further analysis of the prognostic differences between the two groups was conducted using the survival package in R, specifically the survfit function. The logrank test method was employed to assess the significance of the prognostic differences between different groups. Ultimately, a significant prognostic difference was observed when p༜0.05 (p = 6.7e-7), based on which a Kaplan-Meier curve was generated (Fig. 5A). The same method was applied to the validation datasets GSE57495 (Fig. 5D) and ICGC (Fig. 5G), revealing a significant reduction in overall survival for patients with high RiskScore.
ROC analysis was conducted using the pROC package in R (version 1.17.0.1) to determine the AUC. ROC curves were generated for both the training and validation groups. The AUCs for the training group at 365, 1095, and 1825 days were 0.69, 0.82, and 0.89, respectively (Fig. 5B). In the validation group GSE57495, the AUCs at 365, 1095, and 1825 days were 0.68, 0.73, and 0.81, respectively (Fig. 5E). In the validation group ICGC, the AUCs at 365, 1095, and 1825 days were 0.80, 0.75, and 0.71, respectively (Fig. 5H). These results demonstrate satisfactory predictive performance of the model in determining prognosis.
To further explore the effectiveness of risk scoring, we also analyzed the relationship between different risk scores and patient follow-up time, events, and expression changes of various genes (Fig. 5C, F, I). It was observed that as the risk score increased, the patient's survival rate significantly decreased. Furthermore, higher risk scores were associated with a denser distribution of death outcomes, indicating the accuracy and reliability of risk scoring in predicting prognosis and survival of pancreatic cancer patients. Additionally, we identified PEMT, CYP46A1, LTC45, TMEM86B, and LIPE genes as protective factors, as their expression showed a decreasing trend with an increase in risk score. Conversely, PLAAT2 and PTGES genes were identified as risk factors, as their expression exhibited an increasing trend with an increase in risk score.
To validate the model's applicability, we conducted a stratified analysis on pancreatic cancer patients in the training group based on clinical characteristics, including age, gender, and tumor grading. The results revealed that patients in the high-scoring group had significantly shorter overall survival (OS) than those in the low-scoring group (p < 0.01), as shown in Fig. 6. Consequently, we believe this model is both feasible and reliable for predicting the survival of patients in different strata.
2.6 Construction and verification of nomogram
In this study, we integrated age, survival time, survival status, and risk score data using the R package "rms". We utilized the Cox method to establish a nomogram and assess the prognostic significance of these features in the sample (Fig. 7A). The results indicate that the risk score has better predictive performance compared to age, gender, tumor grade, and stage. We predicted the individual's survival probability at 1, 3, and 5 years, and plotted the calibration curve (Fig. 7B).
2.7 Tumor infiltrating immune cell analysis
In order to investigate the relationship between risk score model and tumor immunity, we estimated the stromal cells and immune cells in pancreatic cancer tissues using the ESTIMATE algorithm in the R software package IOBR based on the RNA-seq expression matrix of TCGA-PAAD. We measured the levels of stroma (stromal score), immune cell infiltration (immune score) and tumor purity in each sample, and plotted the boxplots as shown in Fig. 8A. The results showed that the immune score and tumor purity in the high-risk group were significantly lower than those in the low-risk group, while there was no significant difference between the two groups in terms of stromal score. Furthermore, to better understand the interaction between pancreatic cancer cells and the immune microenvironment, we analyzed the differences in immune cell infiltration status of 22 immune cell subtypes using the CIBERSORT algorithm, and plotted the boxplots as shown in Fig. 8B. The results showed that memory B cells, Treg cells, and macrophages were more abundant in the high-risk group, while monocytes and mast cells were more abundant in the low-risk group, and there were no significant differences in immune cell infiltration between the two groups. In summary, we found that different risk groups of pancreatic cancer accurately reflect the level of immunity, and patients in the high-risk group have an overall lower immune response level than those in the low-risk group.
2.8 Prediction of potential drug targets
In order to identify new potential drug targets, we used the R software package pRRophetic to assess the drug sensitivity of two patient groups and generated a box plot as shown in Fig. 8C. The results indicate that patients in the low scoring group exhibit significantly higher sensitivity to AICAR, Nilotinib, PF-4708671, rTRAIL, YK-4-279, and ZM-447439 compared to the high scoring group. These findings provide valuable guidance for our future work.
Pancreatic cancer, known as one of the most aggressive and deadliest malignancies in the digestive system, has made significant progress in terms of etiology, diagnosis, and treatment since the 20th century. Despite improvements in survival rates, it continues to pose a serious threat to people's health, placing a heavy burden on society and families primarily due to a lack of early detection methods and effective treatments. Therefore, there is an urgent need to seek effective diagnostic, prognostic, and therapeutic approaches. In recent years, many researchers have focused on identifying biomarkers, prognosis indicators, and prognostic models for pancreatic cancer[14, 15], with the potential to serve as targets for clinical treatment and benefit pancreatic cancer patients. Similarly, our constructed prognostic model has shown promising performance in predicting the prognosis of patients with Pancreatic cancer.
Metabolic reprogramming has emerged as a hallmark of malignancies, and growing evidence suggests that dysregulated lipid metabolism is associated with the development and progression of various cancers, including gastric cancer, prostate cancer, lung cancer, and pancreatic cancer[9, 16–18]. In pancreatic cancer cells, enzymes involved in fatty acid and cholesterol synthesis are significantly upregulated[19], cholesterol intake is enhanced[9], and excessive uptake of free cholesterol is ultimately converted to cholesterol esters by overexpressed acyl-CoA cholesterol acyltransferase-1 (ACAT-1), leading to cholesterol storage within pancreatic cancer cells[20]. The role of fatty acids in pancreatic cancer is complex, with different types of fatty acids exerting different effects on its development[21]. While saturated and monounsaturated fatty acids promote pancreatic cancer growth, polyunsaturated fatty acids have a dual impact on pancreatic cancer. Therefore, some studies[22, 23] aim to exploit the metabolic characteristics of pancreatic cancer to improve its prognosis. Furthermore, metabolic reprogramming also influences the pancreatic cancer's immunosuppressive microenvironment and immune therapy resistance[9].
In this study, we applied bioinformatics analysis methods to systematically investigate the expression levels of genes related to lipid metabolism in samples from the TCGA-PAAD and GTEx datasets. This allowed us to identify differentially expressed genes (DEGs) that exhibited significant differences in expression between pancreatic cancer and normal tissues, indicating the crucial role of lipid metabolism dysregulation in the development of pancreatic cancer. Through PPI network and functional enrichment analysis of the DEGs, we found that they are primarily involved in processes such as cholesterol metabolism, fatty acid metabolism, steroid metabolism, and the PPAR signaling pathway. These DEGs primarily function as enzymes in the endoplasmic reticulum. Subsequently, we constructed a novel prognosis analysis model based on seven lipid metabolism-related DEGs using LASSO regression analysis. This model includes two risk genes (PLAAT2 and PTGES) and five protective genes (PEMT, CYP46A1, LTC45, TMEM86B, and LIPE). We also assessed the predictive performance of the model using two external datasets (ICGC and GSE57495), and the results demonstrated favorable predictive performance in both the training and validation sets. Furthermore, our predictive model revealed that pancreatic cancer patients in the low-risk group had significantly longer overall survival (OS). This finding was consistent not only in the TCGA dataset but also in the two external validation datasets from the ICGC and GEO databases. We further developed a Nomogram and calibration curve, both of which demonstrated that the predictive performance of our model was significantly superior to conventional methods such as TNM staging, age, sex, and grade.
We used this model to divide the TCGA-PAAD dataset into high and low-risk groups and analyzed their immune features. In the high-risk group, there was a higher infiltration of memory B cells, Treg cells, and macrophages. In contrast, the low-risk group showed a higher infiltration of monocytes and mast cells. Combining the ESTIMATE results, we hypothesized that the overall immune response level in the high-risk group of patients was lower than that in the low-risk group, which is consistent with the immunosuppressive characteristics of pancreatic cancer. Additionally, we utilized the R package pRRophetic to analyze the drug sensitivity of these two groups of patients and identified six potential small molecule compounds with higher sensitivity in the low-risk group. These compounds are ICAR, Nilotinib, PF-4708671, rTRAIL, YK-4-279, and ZM-447439. ICAR, an adenosine analog and AMPK activator, regulates glucose and lipid metabolism and inhibits the production of pro-inflammatory cytokines and iNOS[24]. Nilotinib, a second-generation tyrosine kinase inhibitor, selectively targets the Bcr-Abl protein in cancer cells with abnormal chromosomes, inhibiting cancer cell proliferation and also inhibits the activity of KIT and PDGFR kinases[25]. PF-4708671, a highly efficient and specific S6K1 inhibitor, suppresses S6K1 activity and induces phosphorylation. In vitro experiments have shown that it significantly inhibits the proliferation and invasion capacity of A549, SK-MES-1, and NCI-H460 cells, causing cell cycle arrest in the G0-G1 phase and inhibiting tumor cell proliferation, exhibiting anti-tumor activity[26]. rTRAIL, a recombinant human tumor necrosis factor-related apoptosis-inducing ligand variant, induces cell apoptosis[27]. YK-4-279, an inhibitor of RNA helicase A and oncogenic transcription factor EWS-FLI1 binding, inhibits the growth of ESFT cells and induces apoptosis[28]. ZM-447439 effectively induces cell apoptosis by promoting DNA division and activating caspase3 and 7[29]. Nilotinib has been widely used in the treatment of chronic myeloid leukemia and Parkinson's disease. In conclusion, these drugs may provide new therapeutic strategies for pancreatic cancer patients and guide our future research.
However, our study has certain limitations. We have not yet elucidated the molecular mechanisms and pathways of the relevant genes in the model, nor have we established the specific relationship between small molecules and pancreatic cancer cells. These issues will be the focus of our future work.
This study establishes a prognostic model for predicting the prognosis of pancreatic cancer patients based on two risk genes (PLAAT2 and PTGES) and five protective genes (PEMT, CYP46A1, LTC45, TMEM86B, and LIPE). The results suggest that these genes could serve as potential biological markers for predicting the survival status of PC patients.
Availability of data and materials
The data that supports the findings of this study is available from the corresponding authors upon reasonable requests.
Declaration of competing interest
There were no financial conflicts of interest.
Acknowledgment
We appreciate the strong support of the China Natural Science Foundation (Approval No. 81902421 to L.Z.) for this study.
Ethics approval and consent to participate
This study conducted relevant analysis using public databases and did not involve ethical review or written informed consent forms.
- Neoptolemos JP, Kleeff J, Michl P, Costello E, Greenhalf W and Palmer DH. Therapeutic developments in pancreatic cancer: current and future perspectives. Nat Rev Gastroenterol Hepatol 2018; 15: 333-348.
- Riall TS and Lillemoe KD. Underutilization of surgical resection in patients with localized pancreatic cancer. Ann Surg 2007; 246: 181-182.
- Yu Q, Jobin C and Thomas RM. Implications of the microbiome in the development and treatment of pancreatic cancer: Thinking outside of the box by looking inside the gut. Neoplasia 2021; 23: 246-256.
- Conroy T, Desseigne F, Ychou M, Bouché O, Guimbaud R, Bécouarn Y, Adenis A, Raoul J-L, Gourgou-Bourgade S, de la Fouchardière C, Bennouna J, Bachet J-B, Khemissa-Akouz F, Péré-Vergé D, Delbaldo C, Assenat E, Chauffert B, Michel P, Montoto-Grillot C and Ducreux M. FOLFIRINOX versus gemcitabine for metastatic pancreatic cancer. N Engl J Med 2011; 364: 1817-1825.
- Lomberk G, Dusetti N, Iovanna J and Urrutia R. Emerging epigenomic landscapes of pancreatic cancer in the era of precision medicine. Nat Commun 2019; 10: 3875.
- Schild T, Low V, Blenis J and Gomes AP. Unique Metabolic Adaptations Dictate Distal Organ-Specific Metastatic Colonization. Cancer Cell 2018; 33: 347-354.
- Ohshima K and Morii E. Metabolic Reprogramming of Cancer Cells during Tumor Progression and Metastasis. Metabolites 2021; 11:
- Ward PS and Thompson CB. Metabolic reprogramming: a cancer hallmark even warburg did not anticipate. Cancer Cell 2012; 21: 297-308.
- Qin C, Yang G, Yang J, Ren B, Wang H, Chen G, Zhao F, You L, Wang W and Zhao Y. Metabolism of pancreatic cancer: paving the way to better anticancer strategies. Mol Cancer 2020; 19: 50.
- Sunami Y, Rebelo A and Kleeff J. Lipid Metabolism and Lipid Droplets in Pancreatic Cancer and Stellate Cells. Cancers (Basel) 2017; 10:
- Liu Q, Luo Q, Halim A and Song G. Targeting lipid metabolism of cancer cells: A promising therapeutic strategy for cancer. Cancer Lett 2017; 401: 39-45.
- Alistar A, Morris BB, Desnoyer R, Klepin HD, Hosseinzadeh K, Clark C, Cameron A, Leyendecker J, D'Agostino R, Topaloglu U, Boteju LW, Boteju AR, Shorr R, Zachar Z, Bingham PM, Ahmed T, Crane S, Shah R, Migliano JJ, Pardee TS, Miller L, Hawkins G, Jin G, Zhang W and Pasche B. Safety and tolerability of the first-in-class agent CPI-613 in combination with modified FOLFIRINOX in patients with metastatic pancreatic cancer: a single-centre, open-label, dose-escalation, phase 1 trial. Lancet Oncol 2017; 18: 770-778.
- Johnson WE, Li C and Rabinovic A. Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 2007; 8: 118-127.
- Wang H, Wang X, Xu L, Lin Y, Zhang J and Cao H. Identification of genomic alterations and associated transcriptomic profiling reveal the prognostic significance of MMP14 and PKM2 in patients with pancreatic cancer. Aging (Albany NY) 2020; 12: 18676-18692.
- Ma Q, Wu X, Wu J, Liang Z and Liu T. SERP1 is a novel marker of poor prognosis in pancreatic ductal adenocarcinoma patients via anti-apoptosis and regulating SRPRB/NF-κB axis. Int J Oncol 2017; 51: 1104-1114.
- Huang S, Guo Y, Li Z, Zhang Y, Zhou T, You W, Pan K and Li W. A systematic review of metabolomic profiling of gastric cancer and esophageal cancer. Cancer Biol Med 2020; 17: 181-198.
- Liu R-Z and Godbout R. An Amplified Fatty Acid-Binding Protein Gene Cluster in Prostate Cancer: Emerging Roles in Lipid Metabolism and Metastasis. Cancers (Basel) 2020; 12:
- Sun Z, Jiang Q, Li J and Guo J. The potent roles of salt-inducible kinases (SIKs) in metabolic homeostasis and tumorigenesis. Signal Transduct Target Ther 2020; 5: 150.
- Swierczynski J, Hebanowska A and Sledzinski T. Role of abnormal lipid metabolism in development, progression, diagnosis and therapy of pancreatic cancer. World J Gastroenterol 2014; 20: 2279-2303.
- Li J, Gu D, Lee SSY, Song B, Bandyopadhyay S, Chen S, Konieczny SF, Ratliff TL, Liu X, Xie J and Cheng JX. Abrogating cholesterol esterification suppresses growth and metastasis of pancreatic cancer. Oncogene 2016; 35: 6378-6388.
- Yu M, Liu H, Duan Y, Zhang D, Li S and Wang F. Four types of fatty acids exert differential impact on pancreatic cancer growth. Cancer Lett 2015; 360: 187-194.
- Huang BZ, Chang JI, Li E, Xiang AH and Wu BU. Influence of Statins and Cholesterol on Mortality Among Patients With Pancreatic Cancer. J Natl Cancer Inst 2017; 109:
- Hong JY, Nam EM, Lee J, Park JO, Lee S-C, Song S-Y, Choi SH, Heo JS, Park SH, Lim HY, Kang WK and Park YS. Randomized double-blinded, placebo-controlled phase II trial of simvastatin and gemcitabine in advanced pancreatic cancer patients. Cancer Chemother Pharmacol 2014; 73: 125-130.
- Višnjić D, Lalić H, Dembitz V, Tomić B and Smoljo T. AICAr, a Widely Used AMPK Activator with Important AMPK-Independent Effects: A Systematic Review. Cells 2021; 10:
- Blay J-Y and von Mehren M. Nilotinib: a novel, selective tyrosine kinase inhibitor. Semin Oncol 2011; 38 Suppl 1: S3-S9.
- Shum M, Houde VP, Bellemare V, Junges Moreira R, Bellmann K, St-Pierre P, Viollet B, Foretz M and Marette A. Inhibition of mitochondrial complex 1 by the S6K1 inhibitor PF-4708671 partly contributes to its glucose metabolic effects in muscle and liver cells. J Biol Chem 2019; 294: 12250-12260.
- Yu R, Albarenque SM, Cool RH, Quax WJ, Mohr A and Zwacka RM. DR4 specific TRAIL variants are more efficacious than wild-type TRAIL in pancreatic cancer. Cancer Biol Ther 2014; 15: 1658-1666.
- Erkizan HV, Kong Y, Merchant M, Schlottmann S, Barber-Rotenberg JS, Yuan L, Abaan OD, Chou T-H, Dakshanamurthy S, Brown ML, Uren A and Toretsky JA. A small molecule blocking oncogenic protein EWS-FLI1 interaction with RNA helicase A inhibits growth of Ewing's sarcoma. Nat Med 2009; 15: 750-756.
- Long Z-J, Xu J, Yan M, Zhang J-G, Guan Z, Xu D-Z, Wang X-R, Yao J, Zheng F-M, Chu G-L, Cao J-X, Zeng Y-X and Liu Q. ZM 447439 inhibition of aurora kinase induces Hep2 cancer cell apoptosis in three-dimensional culture. Cell Cycle 2008; 7: 1473-1479.
No competing interests reported.