The sequencing data of 10,534 TCGA pan-cancer samples and 15,776 GTEx normal samples were downloaded from the UCSC Xena database. In particular, for breast cancer, raw transcriptomic count data and the normalized mRNA profile fragments per kilobase of exon per million reads (FPKM) of 115 TNBC patients and 113 normal tissues were retrieved.

The logistics algorithm was used to model the data so that the predictive ability of GLS expression for breast cancer could be further improved. Subsequently, we established a nomogram for predicting the diagnostic probability of breast cancer using the expression of GLS.

The tumor microenvironment condition was assessed quantitatively by calculating the levels of stromal and immune cell infiltration using the expression profiles obtained from the TCGA dataset. This was done by the R package “ESTIMATE” in which the stromal score, immune score, and ESTIMATE score were calculated (16).

For the analysis of immune cell infiltration, we used the CIBERSORT algorithm which is a strong and robust immunoinformatic method to evaluate the immune cell population based on bulk RNA transcriptome. This was achieved by the R package “immunedeconv” (17).

The single-cell transcriptomic profile of the GSE114727 dataset (18) was recorded in the Gene Expression Omnibus (GEO) database. The visualization was done by the online toolkit TISCH (http://tisch.comp-genomics.org/) (19).

The breast cancer cell line, MCF7, was cultured in Huazhong University (April 2023). The medium was specially designed (i.e., DMEM + 10% FBS + 20ng/ml EGF + 0.5ug/mL Hydrocortisone + 10ug/mL Insulin + 1% NEAA + 1% P/S) and regularly changed when the medium colour turned yellowish. The culturing flasks were saved consistently at 37°C in a humidified incubator containing 5% CO2.

The vectors, si-NC and si-GLS were transfected into MCF7 cell line by Liposome 2000 transfection agent (Invitgen, USA) according to the manufacturer’s protocol. Sequences of si-NC was: UUCUCCGAACGUGUCACGUTT ACGUGACACGUUCGGAGAATT, and sequences of si-GLS was: GAUGGUGUCAUGCUAGACAAATT UUUGUCUAGCAUGACACCAUCTT.

The siRNA-transfected MCF7 cells were fully digested, inoculated into 96-well culture plates at 2 x 10³ cells per well and incubated at 37°C. The absorptivity was measured every day using the CCK-8 kit, for a total of seven days of incubation. The above procedures were carried out at 450 nm on the advice of the reagent supplier and four secondary wells were retained for the experimental sessions.

We treated the MCF7 cell line with cell lysis buffer at 100°C for a total of 10 min with cells that had been previously rinsed with cold PBS to extract total protein. Prepared proteins were added to a 10% SDS/PAGE grid system and equal amounts of proteins were electrolysed at constant pressure (i.e., 200V) and then transferred to a PVDF membrane. After sealing the membrane with 5% skimmed milk powder for 1h, we incubated it overnight at 4°C with primary antibodies, followed by 1.5h incubation with secondary antibodies. Then, the PVDF membrane strips were washed and visualized, the relative protein abundance was assessed using Image Lab analysis software and ImageJ.

R Studio software (v 4.1.0) was used for bioinformatics analytics and GraphPad Prism 5.0 (GraphPad Software, Inc., San Diego, CA) was used for semi-quantification of the real-time PCR and western blot results. The software was also used to generate the relevant charts. The Chi-square test or the Wilcoxon signed-rank test was conducted to calculate the P-values. The correlation analyses were done under the Spearman method. The area under the curve (AUC) and the receiver operating characteristic (ROC) curve were analyzed to determine the predictive accuracy. P-value < 0.05 was deemed statistically significant.

Pan-cancer analysis is a widely used and efficient approach to identify the diagnostic value of a specific gene in regard of human tumors (20, 21). Therefore, we first conducted a pan-cancer analysis based on the transcriptomic data from the TCGA database, the result of which revealed that the expression of the GLS gene was significantly down-regulated in diverse cancer types, including but not limited to GBM, KICH, KIRC, and KIRP, etc. ( Figure 2A ), particularly in the breast cancer (abbreviation: BRCA or BC, Figure 2B ). As the TCGA database lacks normal controls, we further combined the transcriptomic data from the GTEx repository for comparison. As a result, the previous observations were confirmed ( Figures 2C, D ).

We used the TCGA dataset alone ( Figure 3A ) and combined it with the GTEx database data ( Figure 3B ) to plot the receiver operating curves (ROC curves), through which we found that the ROC-AUC values were high (in the case of solely used TCGA dataset, ROC-AUC value = 0.68, and in the case of TCGA-GTEx combination, ROC-AUC value = 0.75), suggesting that GLS may serve as an important diagnostic biomarker breast cancer. Then, we used the logistics algorithm to optimize a model aiming for better diagnostic prediction and constructed a GLS-based nomogram ( Figure 3C ), with a C-index = 0.680, hindering it possessed a very good clinical diagnostic ability, which once again supported our idea.

By knocking down the GLS expression via the siRNA techinique, we found that the down expression of GLS promoted the vitality of tumor cells ( Figure 3D ).

As we know, the emt pathway has a very important role in cancer deterioration. From the very general aspect, the up regulation of emt pathway is regarded as one of the potential bases of tumor metastasis. On the other hand, E-adherin and N-cadherin, the 2 key proteins in the emt pathway are thought as the most active players that anchor the tumor tissues to their place of origin. Namely, when the 2 proteins are down regulated, the emt pathway will be activated, and so more likely to have tumor metastasis occurred. Our observation showed that after intervening with si-GLS, the expression levels of both E-cadherin and N-cadherin were significantly decreased, indicating a promoting effect on the pathway, and so the metastasis of the tumor ( Figures 3E, F ).

A genetic signature coined as PAM50 composed of 50 genes is commonly used clinically to divide breast cancer into several different molecular subtypes, and even now is still being a critical indicator in the treatment of breast cancer (22). Here we used GLS instead of this signature ( Figure 4A ) and found that the classification can be performed more efficiently, although the classification is somehow rougher (Luminal type A, Luminal type B, and Her2 type were indistinguishable, thus hereby we termed it as mixed type). The ROC analyses of the comparison between Basal type and mixed type ( Figure 4B ), between Basal type and Normal type ( Figure 4C ), and between Normal type and mixed type ( Figure 4D ) were also conducted. As a result, we found that such GLS-based distinguishments were of high accuracy with high ROC-AUC values, especially in the case of Basal type and mixed type, the ROC-AUC value was up to 0.854.

PR-positive cancer cells may require progesterone to grow. Consequently, these cells may stop growing or die when treated with substances that block the binding and action of progesterone. Therefore, patients with positive PR in the short term have better treatment prospects. In this study, we found that GLS expression was significantly reduced in PR-positive patients than in PR-negative patients ( Figure 4E ). This conjecture was then confirmed by the subsequent ROC analysis results (AUC-value = 0.681, Figure 4F ).

Similar to PR status, ER-positive cancer cells may require estrogen to grow. These cells may stop growing or die when treated with substances that block the binding and action of estrogen. Therefore, patients who are ER-positive in the short term have better treatment prospects. We found that GLS expression was also significantly less in ER-positive patients than in PR-negative patients ( Figure 4G ). This conjecture was also confirmed by the subsequent ROC analysis results (AUC-value = 0.757, Figure 4H ).

Taking them altogether, based on the PR/ER status, it seems that the low expression of GLS may exert adverse effects on the treatment in the short term. In addition, its expression level in breast cancer is lower than that in normal tissues, as such, finding ways to increase its expression level may be a new adjuvant treatment strategy.

We first assessed the association of GLS with the tumor microenvironment by using the ESTIMATE algorithm and found that it was significantly associated with the stromal cells and immune cells ( Figure 5A ). This conclusion was further strengthened by the group comparison between high- and low-GLS expression ( Figure 5B ). Then, we performed ssGSEA analysis on a variety of immune cells and the expression of GLS. Subsequently, it was found that GLS extensively affected the infiltration of almost all immune cell types ( Figure 5C ). This conclusion was also further strengthened by the group comparison of the expression level of GLS ( Figure 5D ), as well as its visualization at single-cell transcriptome ( Figure 5E ), which further highlighted the important impact of GLS on the immune microenvironment of breast cancer.

Through the ssGSEA algorithm, we investigated the functionalities of GLS in breast cancer. It was found that except for DNA repair ( Figure 6H ), its expression was positively associated with the rest pathways ( Figures 6A-G, I ).

In addition, we supplemented an analysis of drug sensitivity targeting the GLS gene based on the GDSC repository ( Supplementary Figure S1 ), from that we found KIN001−102, Phenformin, WZ3105, Vinorelbine, and SB52334 were positively related to GLS expression. They might have certain value to be tested in the future.