The Lipid Metabolites Prole in Patients with Pancreatic Cancer: A Controlled Study of Cancer Tissue Versus Matched Para-Cancer Tissue

Background and objectives: Most patients with pancreatic cancer were diagnosed at a late stage because identifying pancreatic cancer early was dicult. Therefore, it was of great signicance to screen specic biomarkers in early stage of pancreatic cancer. The aim of this study was to demonstrate the prole and characteristics of lipid metabolites in patients with pancreatic cancer and to correlate the expression level of these metabolites with the tumor. Methods: A total of 9 tissue samples from patients with pancreatic cancer were collected and divided into cancer group and para-cancer group according to different sites. All patients’ samples were performed a metabolomics analysis based on Liquid Tandem Chromatography Quadrupole Time of Flight Mass Spectrometry. Results: PCA based on lipidomics analysis could clearly distribute in different regions. OPLS-DA analysis could lter out the irrelevant variables in metabolites and obtain more reliable information about the intergroup differences of metabolites. The volcano plot was used to visualize all variables with VIP>1 and presented the important variables with P<0.05 and |FC|>2. Euclidean distance matrix was calculated for the quantitative values of the differential metabolites, and the differential metabolites were clustered by using the full linkage method and heat map was used to demonstrate. The different metabolites of each group were compared and analyzed by radar map, the corresponding ratio of the quantitative value of differential metabolites were calculated. Heatmap of correlation analysis showed the correlation coecient between different metabolites. Bar plot visualized the variation degree and classication information of metabolites. Bubble plot showed the variation degree, difference signicance degree and classication information of metabolites. Conclusion: The tissue samples of pancreatic cancer had the characteristics of lipidomics, and the difference of lipid metabolites could be used as potential tumor markers of cancer.


Introduction
Pancreatic cancer was one of the most common malignancies of the digestive system in China. Although the incidence of pancreatic cancer was not high, it was the seventh leading cause of cancer death [1].
Despite advances in surgical techniques and adjuvant therapy, the approach of mortality to morbidity has not changed over the past 40 years. Furthermore, less than 5 percent of patients with pancreatic cancer had a survival rate of more than 5 years, and even among those who could perform radical resection, the 5-year survival rate was only about 20 percent [2]. J. Ferlay et al. believed that pancreatic cancer would exceed breast cancer as the third leading cause of cancer death in the future [3]. The reason for such a low 5-year survival rate for pancreatic cancer was the lack of speci c biomarkers for early diagnosis in clinical practice. Therefore, it was of particular importance to look for diagnostic hallmark of signi cance for pancreatic cancer.
The pancreas was an important organ for regulating lipid metabolism in the body. Thus, lipids and their metabolites might be used as indicators of health or disease. Lipidomics, proposed by Han in 2003, studied all lipid molecules in the body and their role in the regulation of protein expression and gene expression [4]. It has been reported that lipidomics was widely used in the study of biomarkers for various tumors (such as ovarian cancer, prostate cancer, breast cancer, etc.) [5][6][7]. Recently, studies have reported that the plasma concentrations of arachidonic acid, lysophosphatidylcholine, phosphatidylcholine (34: 2) and phosphatidylethanolamine (26:0) were increased in patients with early pancreatic cancer using lipidomics [8]. Since lipid metabolism was involved in the proliferation of pancreatic cancer cells, the detection of such lipid content in plasma might be helpful for the early diagnosis of pancreatic cancer. However, in pancreatic diseases, especially pancreatic cancer, lipidomics studies of pancreatic tissue samples have not been comprehensive.
Although our previous studies have suggested that differences in serum lipid metabolites might serve as potential biomarkers for early diagnosis of pancreatic cancer [9], tissue samples represented changes in the local environment of the tumor and were less affected by systemic factors. We aimed to investigate the predictive value of the local tissue lipid metabolites pro le in patients undergoing radical resection of pancreatic cancer. In this study, the differences of lipid metabolites between tumor tissues and adjacent tissues of 9 patients pancreatic cancer were analyzed and compared by metabonomics, so as to nd therapeutic targets for pancreatic cancer.

Methods
All participants signed the informed consent form, which was reviewed and approved by the Clinical Research Ethics Committee of Shanghai Fourth People's Hospital A liated to Tongji University School of Medicine (No. 2019057-001).

Patients
In total, 9 patients with pancreatic adenocarcinoma that had been treated and accept surgery at the General Surgery, Shanghai Fourth People's Hospital A liated to Tongji University School of Medicine, between October 2018 and March 2019 were enrolled in the study. All the 9 patients with pancreatic cancer were diagnosed by CT or MRI before operation and underwent laparoscopic pancreaticoduodenectomy, and postoperative pathology con rmed ductal adenocarcinoma of the pancreas. The mean age of the patients was 63, with 6 males and 3 females. In the clinical stage, 2 cases were T2N0Mx, 3 cases were T2N1Mx, 3 cases were T2N2Mx, and 1 case was T3N1Mx. Fresh frozen tumor tissues were selected as the experimental group, and adjacent tissues were selected as the control group. The exclusion criteria were for the diagnosis of benign tumors, chronic pancreatitis, combined with other tumors, and the recent 6 months of radiotherapy and chemotherapy.
The TripleTOF 5600 mass spectrometer was used for its ability to acquire MS/MS spectra on an information-dependent basis (IDA) during an LC/MS experiment. In this mode, the acquisition software (Analyst TF 1.7, AB Sciex) continuously evaluates the full scan survey MS data as it collects and triggers the acquisition of MS/MS spectra depending on preselected criteria. In each cycle, the most intensive 12 precursor ions with intensity above 100 were chosen for MS/MS at collision energy (CE) of 45 eV (12 MS/MS events with accumulation time of 50 msec each). ESI source conditions were set as following: Gas 1 as 60 psi, Gas 2 as 60 psi, Curtain Gas as 30 psi, Source Temperature as 600 ℃, Declustering potential as 100 V, Ion Spray Voltage Floating (ISVF) as 5000 V or -3800 V in positive or negative modes, respectively.

Data preprocessing and annotation
An in-house program was developed using R for automatic data analysis. The raw data les (wiff format) were converted to les in mzXML format using the 'msconvert' program from ProteoWizard (version 3.0.19282).Then, the mzxML les were loaded into Lipid Analyzer for data processing. Peak detection was rst applied to the MS1 data. The Cent Wave algorithm in XCMS was used for peak detection, with the MS/MS spectrum, lipid identi cation was achieved through a spectral match using a MS/MS spectral library.

Statistical analysis
All of gures were performed by SIMCA sofware (version 15.0.2, Sartorius Stedim Data Analytics AB, Umea, Sweden). Statistical analysis of partial data was performed using SPSS software version 23.0 (IBM Corporation, 2015, USA). All continuous variables that followed a normal distribution were showed as the mean ± standard deviation and were compared using the Student's t-test. A bilateral probability (p) value < 0.05 was considered statistically signi cant. Pearson method was used to calculate the correlation coe cient of the quantitative value of differential metabolites. All patients performed a metabolomics analysis based on Liquid Tandem Chromatography Quadrupole Time of Flight Mass Spectrometry (LC-QTOFMS).

Principal component analysis
Principal component analysis (PCA) was a statistical method that converts a set of observed possible correlated variables into linearly independent variables (i.e., principal components) through orthogonal transformation. PCA could reveal the internal structure of data and transform a multivariate data set into low-latitude data presented with a small number of principal components (PC). SIMCA software (V15.0.2, Sartorius Stedim Data Analytics AB, Umea, Sweden) was used to conduct Data processing LOG conversion plus center (CTR) formatting process, and then automatic modeling analysis [10]. PCA based on lipidomics analysis could clearly distinguish the difference between pancreatic cancer and adjacent tissues ( Figure 1). The lipid metabolism of the two groups was obviously distributed in different regions, indicating that the lipid metabolism of the two groups had their own unique characteristics. The Xcoordinate PC [1] and Y-coordinate PC [2] represented the scores of the rst and second PC respectively, and the color and shape of scatter points represented samples experimental grouping. The samples were all within the 95% con dence interval (Hotelling's T-Squared Ellipse).

Orthogonal projections to latent structures -discriminant analysis
In high-dimensional data, variables contained not only differential variables related to categorical variables, but also a large number of undifferentiated variables that may be related to each other. The difference variables would be dispersed to more principal components because of the in uence of related variables, so better visualization and subsequent analysis could not be carried out. Orthogonal projections to latent structures -discriminant analysis (OPLS-DA) could lter out orthogonal variables unrelated to categorical variables in metabolites, and analyzed non-orthogonal variables and orthogonal variables respectively, so as to obtain more reliable information about the intergroup differences of metabolites and the degree of correlation between the experimental group. SIMCA software (V15.0.2, Sartorius Stedim Data Analytics AB, Umea, Sweden) was used to perform LOG conversion and UV formatting on the Data. First, OPLS-DA modeling analysis was performed on the rst principal component, and 7-fold cross validation was used to verify the quality of the model. Then the validity of the model was evaluated by R2Y (the model's interpretability to the categorical variable Y) and Q 2 (the model's predictability) obtained after cross-validation. Finally, the permutation test was used to randomly change the arrangement order of the classi cation variable Y for several times to obtain different random Q 2 values, and further tested the validity of the model. To investigate the statistical differences in lipid metabolites between the pancreatic cancer group and the para-cancer group, a multivariate analysis model OPLS-DA was used. Data from the pancreatic group and the para-cancer group were distributed in two opposite regions in the OPLS-DA model analysis (Figure 2A). In the gure, the X-coordinate T [1]P represented the predicted principal component score of the rst principal component, the Y-coordinate T [1]O represented the orthogonal principal component score, and the shape and color of scatter points represented different experimental groups. The results of the OPLS-DA score chart showed that the two groups of samples differed signi cantly. All samples were within the 95% con dence interval (Hotelling's T-Squared Ellipse).
The permutation test established the corresponding OPLS-DA model to obtain the R 2 and Q 2 values of the random model by randomly changing the ranking order of the categorical variable Y for several times, which played an important role in avoiding the over tting of the test model and evaluating the statistical signi cance of the model. Q 2 values of the random model were all smaller than Q 2 values of the original model. The intercept of Q 2 regression line and vertical axis was less than zero; at the same time, as the retention degree of displacement decreased gradually, the proportion of Y variable of displacement increased, and Q 2 of the random model decreased gradually. This indicated that the original model had good robustness. The original model could better explain the difference between the two groups of samples ( Figure 2B). The abscissa represented the permutation retention of the permutation test (the proportion consistent with the original model's Y variables; the point at which the permutation retention was equal to 1 was the value of R2Y and Q2 of the original model). The vertical coordinate represented the value of R2Y or Q2, the green dot represented the value of R2Y obtained by the permutation test, the blue square represented the value of Q2 obtained by the permutation test, and the two dotted lines represented the regression lines of R2Y and Q2 respectively. The original model R2Y was very close to 1, indicating that the established model conformed to the real situation of the sample data. The original model Q2 was close to 1, indicating that if new samples were added to the model, a more approximate distribution could be obtained. In general, the original model could better explain the difference between the two groups of samples. Q2 values of the random model were all smaller than Q2 values of the original model. The intercept of Q2 regression line and vertical axis was less than zero. At the same time, as the retention degree of displacement decreased gradually, the proportion of Y variable of displacement increased, and Q2 of the random model decreased gradually. This indicated that the original model had good robustness and no over tting phenomenon exists.

Univariate analysis
Student's t-test P value was less than 0.05, fold change was greater than 2 and variable importance in the projection (VIP) of the rst principal component of the OPLS-DA model was greater than 1 when the differential metabolites from the cancer group and the para-cancer group were selected.

Volcano plot
The results of screening differential metabolites in the cancer group versus para-cancer group were visualized in the form of a volcano plot (Figure 3). Each point in the volcano plot represented a metabolite, the horizontal coordinate represents the fold change (FC) of each substance in the comparison group (log2 FC), the vertical coordinate represented the P-value of the Student t test (-log10 P-value), and the size of scatter point represented the VIP value of the OPLS-DA model. The larger the size of scatter point was, the greater the VIP value was. The scatter color represented the nal screening result. The signi cantly up-regulated metabolites were shown in red, the signi cantly down-regulated metabolites were shown in blue, and the non-signi cantly different metabolites were shown in gray.
Heat map of hierarchical clustering analysis Euclidean distance matrix was calculated for the quantitative values of the differential metabolites, and the differential metabolites were clustered by using the full chain method, and the heat map was used to demonstrate ( Figure 4). The abscissa in the gure represented different experimental groups, the ordinate represented the different metabolites compared in this group, and the color blocks at different positions represented the relative expressions of metabolites at corresponding positions.

Radar chart
We calculated the corresponding ratio of the quantitative value of differential metabolites, and took the logarithmic transformation of base 2, which was shown in red in the gure, and the corresponding content trend change was displayed in the radar chart ( Figure 5).

Heatmap of correlation analysis
For each group of comparison between the cancer group and the para-cancer group, we calculated the correlation coe cient of the quantitative value of the different metabolites. Pearson's method was used for calculation, and it was presented in the form of headmap ( Figure 6). The horizontal and vertical coordinates in the gure represented the different metabolites compared in this group, the color blocks at different positions represented the correlation coe cient between metabolites at corresponding positions, red represented positive correlation and blue represented negative correlation. At the same time, the nonsigni cant correlation was marked with a cross.

Bar plot
The bar plot of lipid group visualized the results of cancer group and para-cancer group by using the change degree of metabolite content and classi cation information (Figure 7). Each column in the lipid column represented a class of metabolites. The ordinate of the gure represented the relative change percentage of the content of various substances in this group ratio. If the relative change percentage of content was 0, it meant that the content of this substance was the same in both groups. The percentage change in relative content was positive, it indicated that the content of this substance was higher in the cancer group. A negative percentage of the relative change in content indicated a higher content of the substance in the para-cancer group. The abscissa of the column chart of lipid group represented the lipid classi cation information.

Bubble plot
The bubble plot was visualized by the degree of metabolite content change, difference signi cance and classi cation information of the cancer group and the para-cancer group (Figure 8). Each point in the bubble represented a metabolite. The size of the point represented the P-value of the student's T-test (-log10 P-value). The bigger the dot, the smaller the p-value. Gray points represented non-signi cant differences with a P-value not less than 0.05, and colored points represented the p-value was less than 0.05 (different colors marked according to lipid classi cation). The abscissa of the bubble plot represented the relative change percentage of the content of each substance in the group (for substances with great change in content, the relative change percentage of the content of other substances was marked on the corresponding abscissa scale). The relative change percentage of the content was 0, indicating the same content of the substance in the two groups. The relative change percentage of content was positive, indicating that the content of this substance was higher in the cancer group. A negative percentage change in relative content indicated a higher content of the substance in the paracancer group. The ordinate of the bubble plot represented the lipid classi cation information. The black line at the bottom showed the distribution density of the metabolite (a line represented a metabolite).

Discussion
The insidious onset, inconspicuous early symptoms, rapid progression and high fatality rate of pancreatic cancer brought great di culties to the early detection, diagnosis and treatment of pancreatic cancer [11]. Although CA199 played an important role in the diagnosis of pancreatic diseases, it was less sensitive to early pancreatic cancer and precancerous lesions. CA199 did not serve as a biomarker for the screening of asymptomatic population, but only for the screening of symptomatic patients or clinical differential diagnosis of other diseases [12]. Therefore, nding speci c biomarkers for pancreatic cancer has been a hot topic in pancreatic cancer research.
Lipids were not only involved in the regulation of a variety of life processes, including energy conversion, material transport, information recognition and transmission, etc., but abnormal lipid metabolism was also closely related to some diseases, such as diabetes, Alzheimer's disease (AD) and the occurrence and development of tumors [13][14][15]. With the development of fast and high-throughput analytical techniques, especially the extensive application of liquid chromatography tandem mass spectrometry (LC-MS/MS) technology, the robust analysis of various trace lipids in samples was achieved through rapid, highthroughput and high-precision analysis, large-scale and comprehensive lipid analysis, as well as metabolic and functional research [16]. Conventional analytical methods included PCA, OPLS-DA, etc [17].
The pancreas was an important organ of lipid metabolism in human body and the study of lipid omics could help to understand the changes of pancreas function. It is reported that the concentration of arachidonic acid, lysophosphatidylcholine, phosphatidylcholine (PC) and phosphatidyl ethanolamine (PE) in the plasma of patients with early pancreatic cancer was found to be increased by the method of lipid group analysis [8,9]. However, lipid metabolites with differences in pancreatic cancer and adjacent tissues were screened out through tissue samples, and the results showed that PC, PE were signi cantly increased in cancer tissues, which was basically consistent with previous plasma studies.
PC and PE all belonged to glycerolipids, whichwere the main components on the surface of cell membranes and organelles. For tumor cells, endogenous lipid synthesis not only provided energy for the proliferation of tumor cells, but also provided lipid components of the cell membrane for the proliferation of tumor cells. Studies have shown that the expression of PC and PE in serum of liver cancer patients was different in different TNM stages by LC-MS analysis [18]. Our study showed that the content of PC and PE in pancreatic cancer tissues was higher than that in para-cancer tissues. This also con rmed that glycerolipids including PC and PE played an important role in the development of tumors. At the same time, our study also further demonstrated that PC and PE could be used as potential biomarkers for tumors.
Tumor cells exhibited metabolic plasticity, providing a selective advantage for tumor cells to survive and proliferate in harsh microenvironments such as hypoxia, acidosis, and malnutrition. In this microenvironment, tumor cell lipid synthesis increased [19]. It has been reported that changes in tumor metabolism and accumulation of metabolites could lead to local immunosuppression in tumor microenvironment. The correlation analysis between the proportion of immune cells and the expression of lipid metabolism genes in tumor microenvironment showed that there were differences in the expression of immune-related lipid metabolism genes, suggesting a potential interaction between lipid metabolism and immune response [20]. Our previous study suggested abnormal lipid metabolism in the serum of pancreatic cancer, and this study further veri ed this study through tissue samples, which laid a foundation for the later study on the molecular mechanism of lipid metabolism and pancreatic cancer [9].
There was a certain relationship between lipid metabolism and in ammatory reaction. Lipid metabolism disorder could enhance oxidative stress and affect chronic in ammatory reaction. Since in ammatory cytokines tended to be synthesized from activated macrophages and these cells did exist in fat cells in cancer patients with weight loss, immune cells or fat cells might be involved in regulating energy pathways and lipid mobilization. In adipocytes, TNF-γ inhibited lipoprotein lipase activity, resulting in triglyceride uptake and reduced lipid deposition, leading to cachexia. Abnormal lipid metabolism was closely related to digestive tract tumors, especially colorectal cancer [19,21,22]. Hyperlipidemia and obesity were high risk factors for colorectal cancer. The increased level of free fatty acids in serum might be related to the occurrence of oxidative stress and the enhancement of lipid toxicity [23]. Currently, the effects on dyslipidemia associated tumorigenesis, transfer mechanism was not clear, so the need for more in-depth research, to explore new targets for cancer treatment. Therefore, it was a new therapeutic idea to explore the therapeutic target of pancreatic cancer by analyzing the abnormal lipid metabolite spectrum.
The advantage of this study was that the most complete analysis of lipid metabolites was performed on tissue samples from patients with pancreatic cancer. The study also had some limitations. We need to further study the molecular mechanism of lipid metabolites and the occurrence and progress of pancreatic cancer. Comparative analysis of pancreatic tissue samples from pancreatic cancer tissues, para-cancer, and healthy controls was also lacking. Further prospective studies and analyses with larger sample sizes are needed to con rm these clinical predictors. Further prospective studies and larger sample size analysis are needed to identify these associated lipid metabolites.

Conclusions
The tissue samples of pancreatic cancer had the characteristics of lipidomics, and the difference of lipid metabolites could be used as potential tumor markers of pancreatic cancer.  Score scatter plot of PCA model for cancer group versus para-carcinoma tissue group. Figure 2