To investigate changes in metabolic rate-limiting enzymes in GC, we systematically screened the transcriptome profiles of 111 rate-limiting enzymes between tumor and adjacent normal tissues from a database reported previously (Wang et al., 2019). All these genes were listed in the Supplementary Table S1. A total of 29 differentially expressed transcripts were identified (p < 0.05): 21 upregulated and 8 downregulated transcripts (Figure 1A). We then utilized Cox regression analysis with the LASSO algorithm to determine the effects of these transcripts on clinical prognosis, and five genes (SOAT1, GNE, UCK2, DCK, and GAD1) were selected according to the minimum criteria and the regression coefficients (Figure 1B). Based on the expression levels of the five genes, the following formula was derived to calculate the risk score for predicting the prognosis of each patient: Risk score = (0.0360 × expression value of SOAT1) + (−0.0006 × expression value of GNE) + (−0.0052 × expression value of UCK2) + (−0.0087 × expression value of DCK) + (−0.0700 × expression value of GAD1) (Figure 1C). By the risk score formula, the patients were divided into the low-risk (n = 152) and high-risk (n = 151) subgroups based on the mean risk score. The expression level of SOAT1 was higher in the high-risk group, while the expression levels of the other four genes were higher in the low-risk group. In addition, the high-risk group patients often had a more advanced TNM stage (p = 0.021), higher tumor grade (p = 0.016), and higher incidence of lymph node metastasis (LNM) (p = 0.025) (Figure 1D). Moreover, the patients in the high-risk group had shorter overall survival (OS) time than those in the low-risk group (Figure 1E). Among the five genes, high expression level of SOAT1 was significantly correlated with poor prognosis in GC patients, and the expression of the other four genes was not associated with the survival of GC patients (Figure 1F; Supplementary Figures S1A–D). Furthermore, Kaplan-Meier plotter data also confirmed that high level of SOAT1 was associated with poor survival in patients with gastric adenocarcinoma patients (Supplementary Figure S1E). Moreover, the expression level of SOAT1 was significantly associated with clinicopathological features, such as TNM stage (p < 0.001) and LNM (p < 0.001) (Figures 1G,H). However, SOAT1 expression had no statistically significant association with tumor grade, invasion depth or distant metastasis status (Supplementary Figures S1F–H). Taken together, these results revealed that the expression of SOAT1 was upregulated in GC and it might be an independent prognostic risk factor in GC patients.

To further confirm the expression of SOAT1 in GC, we first examined the SOAT1 expression levels in 34 GC cancerous tissues and paired adjacent noncancerous tissues. We found that the mRNA levels of SOAT1 were significantly higher in cancerous tissues (Figure 2A) and the protein levels of SOAT1 were significantly higher in 9/10 (90%) cancerous tissues than the corresponding noncancerous tissues (Figure 2B). We then utilized an independent validation cohort to further investigate SOAT1 expression and its relationship with clinical outcome using immunohistochemistry (IHC) staining in a GC tissue microarray (TMA). Similarly, these results showed that the expression of SOAT1 was significantly increased in cancerous tissues compared with the matched normal tissues (n = 160, p < 0.01; Figures 2C,D, Supplementary Figure S2B). Furthermore, the protein expression of SOAT1 in the GC cohort was significantly correlated with LNM status (p = 0.001), TNM stage (p = 0.029) and differentiation degree (p = 0.006); however, SOAT1 expression had no statistically significant correlation with depth of invasion (p = 0.195), distant metastasis (p = 0.328), tumor diameter (p = 0.697) or histological type (p = 0.366) (Supplementary Table S2). Moreover, SOAT1 protein expression was greatly increased in advanced-stage and LNM-positive patients (Figures 2E,F). Kaplan-Meier survival analysis showed that GC patients with high level of SOAT1 had worse OS (p = 6.558e-04) (Figure 2G). Simultaneously, univariate Cox regression analysis revealed that TNM stage (HR = 5.817, 95% CI: 3.114–10.836), depth of invasion (HR = 2.499, 95% CI: 1.808–3.454), LNM (HR = 2.832, 95% CI: 2.189–3.665) and SOAT1 (HR = 2.437, 95% CI: 1.567–3.791) were significantly associated with the survival in patients with GC (Supplemental Figure S2A). Furthermore, by using multivariate Cox regression analysis, we found that SOAT1 was one of the independent predictors of the prognosis of GC patients HR = 1.802 (1.116–2.911) (Figure 2H). To further evaluate the predictive ability of the SOAT1 expression level, we conducted receiver operating characteristic curve (ROC) analysis, and the area under the curve (AUC) of SOAT1 and LNM were 0.778 and 0.797, respectively (Figure 2I). Collectively, these data revealed that SOAT1 expression level was significantly increased in GC tissues and that SOAT1 might be an independent prognostic risk factor for GC.

To elucidate the function of SOAT1 in GC cells, we first investigated SOAT1 expression in GC cell lines and normal gastric mucosal cells (GES-1) by western blot, and it was shown that its expression was higher in most of GC cells than GES-1 cells (Figure 3A). Subsequently, we established stable SOAT1-overexpressing GC cells in AGS cells with relative low expression of SOAT1 (AGS-SOAT1) (Figure 3B). As shown in Figure 3C, overexpression of SOAT1 significantly increased the colony formation efficiency of AGS cells. We also knocked down SOAT1 by two specific siRNAs in MGC-803 and BGC-823 cells and confirmed that both the mRNA and protein levels of SOAT1 were markedly reduced after transfection for 72 h (Figures 3D,E; Supplementary Figures S3A–B). As expected, knockdown of SOAT1 obviously decreased the colony formation efficiency (Figure 3F; Supplementary Figure S3C). All these results confirmed that SOAT1 promoted the proliferative ability of GC cells.

In addition, transwell analysis was performed to determine the role of SOAT1 in GC cell migration and invasion. The results showed that over expression of SOAT1 promoted the migration and invasion of AGS cells (Figure 3G). Conversely, knockdown of SOAT1 in MGC-803 and BGC-823 cells exerted the opposite effects (Figure 3H; Supplementary Figure S3D). Collectively, these data indicate the critical role of SOAT1 in promoting GC cell proliferation, migration and invasion.

Avasimibe, a potent small molecule inhibitor of SOAT1 (Figure 4A), has been proven to exert anticancer effects against many tumors, including glioblastoma (GBM), hepatocellular carcinoma (HCC) and prostate cancer (PCa) (Tardif et al., 2004; Bemlih et al., 2010; Jiang et al., 2019; Liu et al., 2021). Herein, we investigated whether avasimibe has an inhibitory effect on GC cells. As shown by the cell viability curve, avasimibe inhibited the proliferation of GC cell lines in a dose-dependent manner. The IC50 values of avasimibe in MGC-803 and BGC-823 cells at 24 h were 13.489 and 25.377 μM, respectively. In addition, the respective IC50 values at 48 h were 8.811 and 14.208 μM (Figures 4B,C; Supplementary Figures S4A,B). Consistent with the CCK-8 assay results, avasimibe also decreased the colony-forming capacity of GC cells in a dose-dependent manner (Figure 4D).

Next, we explored the effects of avasimibe treatment on the cell metastasis ability in GC cell lines. The wound healing assay showed that avasimibe significantly reduced the migration capability of BGC-823 cells (Supplementary Figure S4C). In addition, we performed a transwell assay, which showed significant dose-dependent decreases in the numbers of migrated and invaded MGC-803 and BGC-823 cells after treatment with avasimibe (Figure 4E; Supplementary Figure S4D). Collectively, these results indicated the critical inhibitory effect of avasimibe on GC cell proliferation and metastasis.

It has been reported that SOAT1 can convert excess free cholesterol into cholesteryl esters for storage in lipid droplets (Xu et al., 2020). We first detected the level of cholesterol in GC cells, and the results showed that esterified cholesterol was significantly reduced in SOAT1 knocked-down GC cells or avasimibe treatment (Figures 5A,B). In addition, we evaluated whether the inhibition of SOAT1 affected the formation of lipid droplets by Nile red and oil red O staining. The results showed that knocking down SOAT1 could significantly reduce the number of lipid droplets in GC cells (Figure 5C, Supplementary Figures S5A,B), and avasimibe treatment had a similar effect (Figures 5D–G).

To further confirm that SOAT1 regulates lipid biosynthesis and catabolism, we examined the expression levels of a panel of lipid metabolism-related genes in SOAT1-knockdown and avasimibe-treated GC cells. Intriguingly, both SOAT1 knockdown and avasimibe treatment reduced the expression of cholesterol metabolism genes (HMGCR, SREBP1, and SREBP2) and fatty acid biosynthesis genes (FASN, ACC, and SCD1) (Figures 5H,I; Supplementary Figures S5C,D). In addition, the expression of SREBP1, SREBP2, and FASN were strikingly reduced at the protein level (Figures 5J–K). Taken together, these results suggested that dysregulated SOAT1 accelerates the esterification of cholesterol and the synthesis of lipids in GC cells.

It has been demonstrated that lipids are required as energy sources and cellular signaling molecules, which are crucial for cancer lymphangiogenesis and lymph node metastasis (Ma Y. et al., 2018; Lee et al., 2019). In addition, the bioinformatic analysis using TCGA and our TMA data suggested that the expression level of SOAT1 was significantly higher in LNM-positive GC tissues (Figures 1H, 2F). Thus, we speculated whether SOAT1 could promote lymph node metastasis of GC cells. VEGFC, VEGFR3, and LYVE-1, which play key role in lymph node metastasis of multiple malignancies, therefore we used online bioinformatics tools (http://gepia.cancer-pku.cn/) to study the relationship between SOAT1 and VEGFC, VEGFR3, and LYVE-1 expression. The results showed that the expression level of SOAT1 was significantly positively correlated with VEGFC, VEGFR3, and LYVE-1 (Figures 6A–C). Next, we further examined whether SOAT1 has an effect on tumor-induced lymphangiogenesis. As shown in Figures 6D,E and Supplementary Figures S6A,B, HLECs treated with conditioned medium (CM) derived from SOAT1-knockdown or avasimibe treatment GC cells significantly reduced the lymphatic capillary formation and the migratory capability of HLECs compared with CM derived from the corresponding vector cells. In addition, compared with CM from the control group, CM from SOAT1-overexpressing GC cells significantly promoted these biological processes (Supplementary Figure S6C). As expected, qRT-PCR data showed that the mRNA levels of VEGFC and VEGFD were significantly decreased in both SOAT1-knockdown and avasimibe-treated GC cells (Supplementary Figures S6D–G). Furthermore, we evaluated the secretion level of VEGFC using ELISA, and found that the VEGFC protein levels were significantly decreased in conditioned medium collected from GC cells which knockdown of SOAT1 or treated with avasimibe compared to conditioned medium from the corresponding control cells (Figures 6F,G; Supplementary Figures S6H,I). In addition, the expression of VEGFC were also decreased at the protein levels (Figure 6H). Intriguingly, the IHC staining results showed that the expression of SOAT1 was scarcely detected in adjacent noncancerous tissues, slightly increased in LNM-negative GC tissues and strongly upregulated in LNM-positive GC tissues, and positively correlated with the density of microlymphatic vessels, as indicated by the number of LYVE-1-positive microvessels (Figures 6I,J). Taken together, these findings indicated that inhibition of SOAT1 suppresses lymphangiogenesis and that avasimibe has the potential therapeutic effect for GC patients with lymph node metastasis.

SREBP1 and SREBP2 are both important signaling molecules that relate to tumor lymph node metastasis (Heo et al., 2020; Li et al., 2020). To investigate whether SREBP1 and SREBP2 involved in SOAT1-mediated lymphangiogenesis and VEGFC production, we pharmacologically inhibited SREBP1 and SREBP2 by fatostatin, which displays antitumor activity in cancers by downregulating SREBP- mediated metabolic pathways (Li et al., 2014). As shown by the cell viability curve, fatostatin inhibited the proliferation of AGS cells in a dose-dependent manner and the IC50 value is 30.822 µM (Figure 7A). Intriguingly, the results showed that fatostatin could significantly reversed the promoting effects of SOAT1 overexpression on the migration and the lymphatic tube formation of GC cells (Figures 7B–D). In addition, ELISA and qRT-PCR results suggested that the expression of VEGFC were recovered after treated with fatostatin (Figures 7E,F). Collectively, these results indicating that blocking SREBP1 and SREBP2 pathway inhibits SOAT1-mediated lymphangiogenesis in gastric cancer.

Human GC cell lines AGS cells were purchased from the American Type Culture Collection (ATCC). HLECs, HEK-293T, BGC-823, HGC-27, MGC-803, SGC-7901, MKN-74, and MKN-45 were obtained from the Type Culture Collection of the Chinese Academy of Sciences (Shanghai, China). HEK-293T cells were cultured in DMEM (Biological Industries, Cromwell, CT, United States), AGS cells were cultured in F12K medium (Biological Industries, Cromwell, CT, United States), and the other cells were cultured in RPMI-1640 medium (Biological Industries, Cromwell, CT, United States). All cells were cultured with 10% fetal bovine serum (FBS; Biological Industries, Cromwell, CT, United States), 100 U/ml penicillin (Invitrogen), 100 μg/ml streptomycin (Invitrogen) and incubated in 5% CO₂ at 37°C.

A total of 34 pathologically confirmed GC tissues and the corresponding adjacent noncancerous fresh frozen tissues were collected from patients treated with radical gastrectomy at the Nanjing Drum Tower Hospital, the Affiliated Hospital of Nanjing University Medical School (Nanjing, Jiangsu, China). All patients provided written informed consent, and all these tissues were obtained for further qRT-PCR, western blot and immunohistochemistry assay.

A total of 160 pathologically confirmed GC patient tissues from Nanjing Drum Tower Hospital were obtained for TMA construction, and the expression of SOAT1 was detected by IHC staining. The chips were stained and scanned by Servicebio (Wuhan, Hubei, China) according to a standard protocol. Stanning of SOAT1 was scored by the two pathologists blind to the clinical data by applying a semi-quantitative immunoreactivity score (IRS), IRS of 0–6 and IRS of 8–12 were classified as low and high expression of SOAT1, respectively.

SOAT1 si-RNAs were designed and synthesized by RiboBio (Guangzhou, China), and the sequence was showed as follows: si-SOAT1#1: 5′-TAATGGTCGAATTGACATAA-3′; si-SOAT1#2: 5′-TTG​AAC​TCA​AGT​ACC​AGC​CTT​C-3′. The vector expressing SOAT1 (pcDNA3.1-SOAT1) as well as the blank pcDNA3.1-vector were purchased from Hanbio Bio-technology Company (Shanghai, China). SiRNA and plasmid were transfected with DharmaFECT4 (GE Healthcare) and Lipofectamine 3000 (Invitrogen, Carlsbad, CA, United States). Virus-containing supernatant was collected 48 h after transfection and was added to cell when confluence reached 70%, 48 h later, transduced cells were selected with 1 μg/ml puromycin.

Total RNAs were extracted from human GC tissues and associated non-cancerous tissues using TRIzol reagent (Invitrogen, Carlsbad, CA, United States) according to the manufacturer’s instructions. The reverse transcription reaction (RT) was performed with Reverse Transcription kit (Vazyme, Nanjing, China). The RT-PCR reactions were performed with a SYBR Green PCR Kit (Vazyme, Nanjing, China), measured in triplicate and performed on an Applied Biosystems 7900HT sequence detection system (Applied Biosystems). GAPDH was used as an internal control for mRNA. The relative expression levels of the target genes were calculated using the comparative 2−ΔΔCt method. All primers used in this study were listed in Supplementary Table S3. All results were obtained from three independent experiments performed in duplicate.

The Western blot protocol was performed as previously described(Wang et al., 2020). The antibodies used were listed as follows: anti-SOAT1 (Immunoway, YN1370); anti-SOAT1 (Abcam, ab39327); anti-VEGFC (Proteintech, 14517-1-AP); anti-FASN (Proteintech, 10624-2-AP); anti-SREBP1 (Proteintech, 66875-1-AP); anti-SREBP2 (Proteintech, 28212-1-AP); and anti-GAPDH (Proteintech, 60004-1-Ig). Phenylmethylsulfonyl fluoride was purchased from Selleck (Houston, TX, United States).

The antiproliferative effect of avasimibe and fatostatin (MedChemExpress, Shanghai, China) was evaluated with a Cell Counting Kit-8 kit (CCK-8, Vazyme, Nanjing, China). One day before avasimibe treatment, GC cells were seeded in 96-well plates (5,000 cells/well). The cells were treated with avasimibe for 24 h or 48 h. After the indicated time, 10 µL per well of CCK-8 solution was added and incubated at 37°C for 1 h. Absorbance was recorded at 450 nm, and five independent assays were carried out.

For the plate colony formation assay, GC cells were seeded in 6-well plates or 12-well plates (1,000 cells/well in 6-well plates and 500 cells/well in 12-well plates) and incubated for 10–14 days. The medium with or without avasimibe was changed every other day. Then, the cells were fixed with 4% paraformaldehyde for 15 min and stained with crystal violet for 1 h. Images were acquired with a digital camera, and three independent assays were carried out.

GC cells were seeded in 12-well plates, following by 0.2 µM oil acid (OA) (Sigma, United States) for 24 h. Then the cells were fixed with 4% paraformaldehyde solution. For Oil red O staining, cells were incubated with 60% isopropanol for 15 min before stanning by Oil red O working solution. The Oil red O working solution was prepared by diluting the Oil red O stock solution with distilled water at a ratio of 3:2, followed by filtration. For Nile red staining, cells were sequentially stained with 0.05 μg/ml Nile red (Sigma, United States), washed with PBS twice and then stained with DAPI (Beyotime, Shanghai, China). Images of the cells were acquired by fluorescence microscopy. All the operations were performed in a dark environment.

HLECs were seeded in 96-well plates (2 × 10⁴/well, precoated with 50 µL of Matrigel, Corning Life Sciences, Bedford, MA, United States), containing medium obtained from tumor cells and cultured for 12 h. Images of lymphatic tubes were acquired using a fluorescence microscope and quantified by measuring the number and area of the completed tubule structures.

Cell migration and invasion abilities were evaluated by transwell chambers (Corning Life Sciences, Bedford, MA, United States). Briefly, a total of 5 × 10⁴ GC cells suspended in media without FBS were seeded in the upper chambers coated with or without 50 μL of Matrigel (BD Biosciences). Then, 600 μL of culture medium containing 10% FBS was added to the lower chambers. After incubation at 37°C for 12 h, the cells in the lower chambers were fixed with 4% paraformaldehyde for 30 min, stained with crystal violet (Beyotime, Shanghai, China) for 30 min. Finally, three random fields were microscopically examined, and the number of cells was determined by photoshop software.

GC cells were seeded into six-well plates and grown to 90% confluence. The cell layer was scrapping with a 10-μL sterile pipette tip and washed with three times with PBS to remove the floating and detached cells, then cultured in 1% FBS medium and images were acquired under a microscope at multiple time points (0, 6, 12, and 24 h).

The cell culture supernatant was collected, secreted VEGFC was quantified using the Human VEGFC Quantikine ELISA Kit (Cat. No. E-EL-H1600c, elabacience) according to the manufacturer’s instructions. Briefly, GC cells seeded in 6-well culture plates were cultured in the growth medium until 90% confluence. The cells were washed three times with phosphate-buffered saline (PBS) and cultured in serum-free medium for 24 h. The cell culture supernatant or VEGFC standards were added to the 96-well plates coated with polyclonal antibody specific for human VEGFC in triplicate and incubated for 1.5 h at 37°C. Then, the Biotinylated detection Ab working fluid was added to each well and incubated for 1 h at 37°C. Afterwards, the well was washed five times, HRP conjugate working solution was added and incubated for 30 min at 37°C. Then, substrate solution was added to each well, and incubated for 30 min at 37°C in the dark. Then, the stop solution was added to each well. Absorbance was determined at 450 nm. All assays were performed in triplicate.

Cells were collected by centrifugation for 10 min (4°C, 1,000×g) and resuspended in 200 μL extracting solution, then ultrasonic crushing in an ice bath was performed for 30 s. Supernatants were combined and centrifuged for 20 min at 10,000 × g (4°C) and placed on ice for measurement. Total cholesterol and unesterified cholesterol were quantitated using the manufacturer’s protocol of a Total Cholesterol and Cholesterol ester Fluorescence Determination Kit (Cat. No. E-BC-F032, elabacience). Amount of cholesterol ester were determined by subtracting the amount of unesterified cholesterol from total cholesterol.

A total of 111 human rate-limiting metabolic enzymes were obtained from the rate-limiting enzyme database according to a previous study (Wang et al., 2019). The RNA expression data and clinical information of GC patients were obtained from TCGA (https://tcga-data.nci.nih.gov/tcga/). Differentially expressed genes were screened with the R package “limma.” The expression level of prognostic associated rate-limiting metabolic enzymes between cancerous and normal samples was displayed via package “heatmap” and “ggplot,” respectively. Univariate Cox regression analysis was performed to identify prognostic associated rate-limiting metabolic enzymes. Package “glmnet” was used to perform LASSO Cox regression model to select optimal weighting coefficients via penalized maximum likelihood and build a prognostic signature. The formula of the risk score for the prediction of GC patients’ prognosis was as follows: risk score = the sum of the multivariate Cox regression coefficient ratio of each mRNA multiplied by the expression level of each mRNA. For survival analysis, overall survival was defined as the time from first treatment to death for any cause, Kaplan-Meier method and log-rank test were used to detect potential prognostic factors. For clarify relationship of SOAT1, clinicopathological characteristics, and prognosis, univariate Cox regression analysis was performed to find out the independent factors correlated with OS. AUC was employed to demonstrate the sensitivity and specificity of different variables by risk estimation. The “pROC” package was used to perform ROC curve and analyze AUC. All statistical tests were two sided and p < 0.05 were significant and each experiment was carried out in at least triplicate.