2.1 Patients and study design
This study was designed as a retrospective cohort study, which utilized data from the First Affiliated Hospital of Wenzhou Medical University and public databases (including TCGA (https: //cancergenome. nih.gov/) and Gene Expression Omnibus (GEO) datasets (https://www.ncbi.nlm.nih.gov/geo/)). The data collection and processing protocols were approved by the institutional ethics committee (Ethics Commission of the Faculty of Medicine of the Wenzhou Medical University). All procedures were carried out in accordance to BRISQ Guidelines for reporting research on human biospecimens. These selected GC patients were histopathologically diagnosed with primary GC and then received surgical treatment with or without regular chemotherapy. All patients were followed up until death or March 2021 (end of follow-up). Patients without a pathological diagnosis of primary GC, with gastric stromal tumor subtype, who had undergone prior therapy (chemotherapy, resection prior to enrolment), did not undergo the excision or did not have complete pathology, laboratory, and follow-up data were excluded. The following demographic, clinical and pathology data were used: T stage, N stage, M stage, pathological stage, tumor history, laboratory test results (age, sex, body-mass index (BMI), TG, HDL-C, Cho, TC, CEA, creatinine). Pathologists assessed the tumor stage according to the 7th edition of the AJCC TNM staging guidelines. Finally, a total of 458 patients were enrolled in the study. Hyperlipoidemia was defined as conform to more than one criteria : 1) Total cholesterol (TC) than or equal to 5.17mmol/l; 2) Triglyceride (TG) than or equal to 1.70mmol/l; 3) High-density lipoprotein (HDL) less or equal to 1.16mmol/l; 4) Low-density lipoprotein white (LDL) than or equal to 3.10mmol/l. Disease-free survival was defined as the time from diagnosis to tumor recurrence or occurrence of metastatic disease and overall survival as the time from diagnosis to disease-related death.
This study also utilized data from public database. We retrospectively selected GC gene expression and its clinicopathological data from the TCGA and Gene Expression Omnibus datasets. Raw microarray data Affymetrix were downloaded and normalized using the limma package. For validation, we also searched “gastric cancer” and “Homo sapiens” to March 2021 to select suitable chips in the GEO database. All chips with gene expression (containing at least 20 samples) from primary human gastric tumors and normal tissue were considered eligible, with no unique exclusion criteria being applied. The study contained 13 cohorts of samples from patients with GC: GSE12369, GSE13911, GSE2685, GSE26942, GSE26988, GSE29272, GSE37023, GSE54129, GSE65801, GSE66229, GSE79973, GSE84787 and TCGA STAD. Chips were summarised, together with accession numbers, in Table S1. In total, 1488 GC and 448 normal cases were acceptable for subsequent meta-analysis. The RNA-sequencing data were processed via R limma package, setting P≤ 0.01, fold change ≥ 1.5 as the cutoff line. The detailed working algorithm was demonstrated in Figure 1.
2.2 Construction Clinical Nomogram
The OS and PFS clinical nomograms were constructed based on the main prognostic factors to predict 1-, 3- and 5-year survival of each GC patients. A multivariable logistic regression analysis was applied to build nomogram. Each patient could sum up variable score and finally establish predictive measures of survival and relapse. The calibration curve for predicting 1-, 3- and 5-year OS and PFS indicated that the nomogram-predicted survival closely corresponded with actual survival outcomes. The survival analysis was conducted using rms, survival and survcomp package. Hazard ratios (HRs) and 95% confidence intervals (CIs) were recorded.
2.3 Establishment of the LASSO regression model and calculation of lipodystrophy risk score
We used the GSEA program to derive the enrichment scores of each lipid-metabolism Gene sets. The concrete gene lists of each lipid metabolism enrichment KEGG terms were listed in the table 2. The least absolute shrinkage and selection operator (LASSO) method which conducting with 100 iterations of 10-fold cross-validations to select the most optimal significant features and avoid overfitting, was selected to obtain the most significant genes for predicting lipid metabolism score. The features with non-zero coefficients were then selected. A lipodystrophy score was calculated for each patient via weighted by their LASSO Cox coefficients. Moreover, GSE15459, GSE26253, GSE62254 and GSE84437, which contain concrete GC patient survival information, were further employed to confirm the prognostic power of gene signature. We calculated the prognostic risk score for each patient, then K-M survival analysis were employed.
2.4 Co-expression Gene Network Based on RNA-seq Data and functional analysis
The Weighted correlation network analysis (WGCNA) was used to identify important co-expression modules and their enriched genes associated with GC lipid metabolism[9]. The proper soft threshold power (β) was chosen based on the scale-free topology criterion. The correlation between the modules and lipidscore was evaluated using Pearson’s correlation coefficient analysis. Two modules with the highest average gene significance scores among all genes in the modules were selected for further study. The connectivity degree of each node of the network was calculated by STRING database and reconstructed via Cytoscape software. Gene ontology (GO) enrichment analysis was performed with the DAVID platform.
2.5 Meta-analysis
To further confirm the accuracy of conclusions, we conducted meta-analysis via Review Manager. Meta-analysis estimated the error from the heterogeneity of platform. Continuous outcomes were estimated as standard mean difference with 95 % confidence interval. Continuous outcomes were estimated as standard mean difference with 95 % confidence interval (CI).
2.6 Colony Formation and Transwell Migration Assay
Human gastric tumor cell lines BGC823 and SGC7901 were cultured in 1640 medium (Gibco company) supplemented with 10% fetal bovine serum (FBS), 100U/mL Penicilin and 100µg/mL Streptomycin. Cells were cultured in an incubator with 5% CO2 at 37°C. The number of 1 × 103 BGC823 and SGC7901 were inoculated in six-well plates and incubated at 37℃ for 5 days. Cell colonies were finally fixed with 4% paraformaldehyde formaldehyde (Solarbio, Beijing, China) followed by staining with crystal violet (Sigma-Aldrich). The number of colonies was calculated. Transwell migration experiments were used to confirm the migration ability. 5 × 104 cells were added to the upper chamber placed in a 24-well plate, with serum-free medium. Meanwhile, medium containing 15% serum was added to the lower chamber. Taking cell images cell at 100× magnification.
2.7 Western blot, RT-PCR and Antibodies
An equal amount of proteins was subjected to SDS-PAGE. Proteins were transferred onto PVDF membranes, and the blots were incubated with the following different primary antibodies: Rabbit p-mTOR (S2448), p-AKT, β-actin from Cell Signaling Technology and IL6 from Proteintech. Anti-mouse and anti-rabbit antibodies were purchased from Santa-Cruz Biotechnology. All primary antibodies were confirmed to be reactive only to the targets by the manufacturer and used at 1:1000, and secondary antibodies were used at 1:5000. As for RT-PCR, the following primers were used:
ACLY forward, 5'-GACTTCGGCAGAGGTAGAGC-3', and reverse, 5'-TCAGGAGTGACCCGAGCATA-3';
GAPDH forward, 5'-TGTGGGCATCAATGGATTTGG-3' and reverse, 5'-ACACCATGTATTCCGGGTCAAT-3'.
ACACA forward, 5'-CAACAGTGGAGCAAGAATCGG-3' and reverse, 5'-TCACAATGGACAGAGTTGAGAGC-3'.
FASN forward, 5'-CCTGGCTGCCTACTACATCG-3' and reverse, 5'-CACATTTCAAAGGCCACGCA-3'.
SCD1 forward, 5'-GCGATATGCTGTGGTGCTTAATGC-3' and reverse, 5'- GGAGTGGTGGTAGTTGTGGAAGC -3'.
HMGCR forward, 5'-TGATTGACCTTTCCAGAGCAAG-3' and reverse, 5'-CTAAAATTGCCATTCCACGAGC-3'.
2.8 Quantification of free fatty acid and cholesterol
We prepared chloroform/methanol (2:1) for extracting lipids. The levels of free fatty acid and cholesterol were determined with EnzyChromTM free fatty acid and cholesterol kits (Bioassay Systems, Hayward, CA, USA).
2.9 Immunohistochemistry
20 GC specimens were collected, of which 10 hyperlipemia and 10 non- hyperlipemia GC tissues. Two researchers evaluated the staining results independently and scored the intensities of immunostaining as: 0 (negative), 1 (weakly positive), 2 (moderately positive) and 3 (strongly positive).
2.10 Statistical analysis
Statistical analysis was conducted using R software (v. 3.0.1; http:// www.Rproject.org), SPSS (v. 21.0.0.0; https://www.ibm.com), and GraphPad Prism 6. Kaplan-Meier curves and log-rank tests were used to predict OS and PFS in relation to lipid metabolism. Univariate and multivariate Cox regression analyses were performed to calculate corresponding hazard ratios (HRs) and 95% confidence interval (CI). All statistical tests were two-sided and P <0.05 was considered statistically significant.