Association of lipid biomarkers with clinicopathological features
According to our selection criteria, a total of 458 GC patients were included in this retrospective study. One hundred and three (22.5%) patients were female, and 355 (77.5%) were male; the median age was 64.3 (range, 29–87) years. About 38.9% of the patients only received surgery, while the others received both regular adjuvant postoperative therapy and surgery. The median follow-up for OS and PFS was 40.67 and 37.21 months respectively. Hyperlipemia is conferred as meeting one of the following items: the serum total TC, TG and LDL-C are increased and HDL-C is decreased. In our study, the proportions of patients with high- TC, TG, LDL-C and HDL-C were 26.64, 29.04, 25.33 and 27.51%, respectively. Totally, there were 288 patients diagnosed with hyperlipemia and 170 patients with non- hyperlipemia. Baseline clinicopathological characteristics were summarized in Table 1.
The results of univariate Cox hazards analysis was presented in Fig. 1 and Fig. S2. The univariate analysis demonstrated that HDL, TC and LDL were significant associated with PFS (Fig. 1A, P < 0.05). Although some univariate Cox hazard results of lipid factors did not reach statistical significance, we saw a clear trend and called the result promising and worth further studies. An OS and PFS nomogram were constructed to predict 1-, 3- and 5-year overall survival (Fig. 1B) and PFS (Fig. S2B) of GC patients. Total scores were summations of each variable based on the intersection of the vertical line. As shown in Fig. 1B and Fig. S2B, TC and LDL contributed the most risk points, whereas the other clinical information contributed much less. By sum of the total points from all variables combined with the location at the total point, we could obtain the probabilities of survival outcomes of 1-year, 3-year, and 5-year. In addition, decision curve analysis showed that both the predictive accuracy of prognostic nomograms for OS and PFS (Fig. 1C and Fig. S2C).
Identification of lipid metabolism signature in patients with GC
Gene Set Enrichment Analysis (GSEA) is a computational method that defined set of genes shows statistically differences between two biological states [9]. The significantly enriched lipid-related pathways included unsaturated fatty acids, fatty acid metabolism, steroid biosynthesis, ether lipid metabolism and glycosphingolipid biosynthesis lacto and neolacto serie. A total of 133 lipid-relative genes were significantly enriched in the pathways related to lipid metabolism and gastric cancer (Table 2). LASSO Cox regression was performed to identify the most important features in terms of predicting the survival of GC patients (Figs. 2A). By forcing the sum of the absolute value of the regression coefficients to be less than a fixed value, certain coefficients were reduced to exactly zero, and the most prognostic features (ACLY, ABCG4, ABCG1, FTO, ABCA2, IGFBP7) were identified with relative regression coefficients. The prognostic risk score model was established with the following formula: lipid score = expression level of ACLY × 0.47 + expression level of ABCG4× 0.133 + expression level of ABCG1 × 0.2238 + expression level of FTO× 0.274-expression level of ABCA2 × 0.385 + expression level of IGFBP7 × 0.210. All patients were divided into the high- and the low-lipid group using the median score as the cutoff line (Fig. 2B-D). K-M survival analysis showed that high risk group had significantly poorer OS than that of low (log-rank P < 0.001). In addition, following the univariate and multivariate analyses, the lipid signature also showed to be an independent prognostic factor in the GC cohorts (95%HR1.78–2.12, P < 0.001; 95%HR1.65–1.92 P < 0.001).
Evaluation of the prognostic lipid metabolism signature in external validation cohorts
To verify the accuracy of prognostic model identified by TCGA were also important in additional GC cases, we further selected eligible cohorts of GC cases from the GEO database (GSE15459, GSE26253, GSE62254 and GSE84437). As result, a total of 1357 GC cases were applied to evaluate the robustness of our model. Consistent with the results in the train cohort, the four survival analysis all showed that high lipid group had a worse prognosis than low one. The distribution of lipid-score, and survival information of patients were analyzed and showed in Fig. 3A-C. In brief, these external validation outcomes combination with prior studies demonstrated that our lipid signature were powerful enough to precisely discriminate high lipid score of GC patients.
Functional annotation and WGNCA of GC patients
To further identify the underlying biological characteristics in the lipid metabolism signature, WGCNA was performed, and the correlation of lipid score and module membership were analyzed. The soft threshold selection is shown in Fig. 4A. The yellow module had a significant p-value with lipid signature (Fig. 4B). The association between module membership and gene significance for each gene in the yellow module is shown in Fig. 4C. To better annotate the module function, we singled out the 20 central genes in the co-expression network whose MM > 0.8. As shown in the Fig. 4D, ACLY is the hub gene in the GC lipid regulation.
ACLY markedly elevated expression in the GC tissues
Genome data from TCGA suggested that ACLY gene is amplified in mostly GC cases (Fig. 5A). To further determine the ACLY mRNA expression in GC tissues, we examined the open GEO datasets which contain both GC specimens and normal specimens (Additional Table 1). In total, 1488 GC and 448 normal cases were enrolled in the subsequent meta analyses. Notably, the results showed that the ACLY mRNA expression exhibited a significantly increasing trend in group of GC specimens compared with normal specimens (Z = 4.34; p < 0.0001, Fig. 5B). These meta-analyze combination with prior outcomes mainfested that ACLY mRNA expression was significantly enhanced in GC tissues.
To evaluate the clinical significance of ACLY overexpression, we analyzed its protein expression by immunohistochemistry (IHC) in 30 GC patients. As shown in the Fig. 5C, GC patients with high ACLY expression were endowed with advanced pathological stage than those with low ACLY expression in cohorts. Essentially the same result was seen with HPA GC cohort (Fig. 5D). These results robustly demonstrate that ACLY was an independent prognostic predictor of poor survival in patients with GC.
ACLY increased the expression levels of fatty acid synthesis enzymes and AKT/mTOR signaling
To study the role of ACLY in the lipid metabolism of GC cells, we designed to measure the changes of lipid content in gastric cancer cells with relative higher and lower ACLY expression. As shown in the Fig. 6A-B, overexpression of ACLY significantly increased the levels of intracellular free fatty acid and triglyceride, while knockdown of ACLY markedly reduced the levels of those lipids. We further investigated the expression levels of key molecules involved in fatty acid metabolism (ACACA, FASN, SCD1, HMGCR) in GC tissue when ACLY was knocked-down or over-expressed. As shown in the Fig. 6C-D, the expression of ACLY resulted in significantly change expression levels of those lipogenic enzymes. Thus, we inferred that ACLY increased de novo fatty acid synthesis and cholesterol biosynthesis in GC tissues. To provide further support, the expression levels of ACLY and lipogenic enzymes were determined in 380 GC tissue samples from TCGA. Spearman rank correlation analysis indicated significantly positive correlations between the expression levels of ACLY and lipogenic enzymes of ACACA (r = 0.345, p < 0.0001), FASN (r = 0.413, < 0.0001), SCD1 (r = 0.300, < 0.0001) and HMGCR (r = 0.286, < 0.0001) (Fig. 6E).
It is consensus viewed that lipids are a broad church of hydrophobic biomolecules that participate in a wide array of metabolic pathways, and can influence cancer cell biology via a range of multiple oncogenic signaling pathways. Considering the mTOR pathway is a master regulator of cell growth and metabolism in response to nutrient signals, particularly lipid. A key example is the well-defined influence of PI3K-mTOR, which activates multiple oncogenic signaling pathways, such as AKT signaling, during tumor progression [10]. Considering that AKT/mTOR pathway has been well established to play a central role in the regulation of cell lipid metabolism [11, 12], we hypothesized that ACLY overexpression may activate AKT/mTOR pathway to promote cancer development. As expected, ectopic expression of ACLY was sufficient to robustly promote GC cell migration and transwell compared with control cells (Fig. 6F and G). ACLY knockdown significantly decreased the phosphorylation levels of AKT and mTOR, whereas in inverse, indicating that ACLY activates AKT/ mTOR signaling in GC cells (Fig. 6H-I). Thus, we came up with an assumption that ACLY increased GC progression by activating AKT/mTOR signaling.