To examine the status of aerobic glycolysis in metastatic breast cancer cells with ambient oxygen, we first determined the lactate production and glucose consumption in four different breast cancer cell lines. As shown in Supplementary Figure S1A, three triple-negative breast cancer (TNBC) cell lines, SUM159, MDA-MB-231 and BT549, all with high level of SIPA1 expression, consumed more glucose and produced more lactate than MCF7, a non-TNBC cell line with low SIPA1 expression (Supplementary Figures S1A–S1C), indicating that a high level of SIPA1 expression might be responsible for high aerobic glycolysis in breast cancer cells. To verify the correlation between SIPA1 and aerobic glycolysis, we established MDA-MB-231 and BT549 cell lines expressing a low level of SIPA1 and an MCF7 cell line overexpressing a high level of SIPA1 (MCF7/SIPA1) (Figure 1A and Supplementary Figure S2A), and we found knocking down SIPA1 decreased the levels of lactate production and glucose consumption, whereas forced expression of SIPA1 exhibited the opposite effects in these cells (Figures 1B,C and Supplementary Figures S2B–C). Furthermore, we examined the transcriptional levels of glycolysis-related genes of parental and SIPA1-knockdown MDA-MB-231 cells, as well as those of parental and SIPA1-overexpressing MCF7 cells by RNA sequencing. As shown in Figure 1D, knocking down SIPA1 down-regulated mRNA levels of nearly all glycolysis-related genes in MDA-MB-231 cells, whereas overexpressing SIPA1 up-regulated them in MCF7 cells. It was then confirmed by qRT-PCR assay that SIPA1 modulated the transcription levels of the glycolysis-related genes in MDA-MB-231, BT549 and MCF7 cells (Figure 1E and Supplementary Figure S3). The expression of key enzymes HK2 and LDHA was detected by Western blotting and we confirmed that knocking down SIPA1 significantly down-modulated the expression of HK2 and LDHA in MDA-MB-231 cells, whereas overexpressing SIPA1 up-regulated them in MCF7 cells (Figure 1F). Since LDHA is a critical enzyme for the conversion of pyruvate to lactate at the final step of aerobic glycolysis, we set out to determine whether LDH activity was modulated by SIPA1 in breast cancer cells. As shown in Figures 1G,H, knocking down SIPA1 in MDA-MB-231 cells significantly decreased both extracellular and intracellular LDH activities, whereas forced-expression of SIPA1 in 231si cells increased both LDH activities. In MCF7 cells, overexpressing SIPA1 also greatly increased both LDH activities (Figures 1G,H).

Additionally, it is worthy of note that HIF-1α and c-myc have been reported to serve as dominant regulators of glycolysis by targeting multiple glycolysis-related genes in a variety of cancers (Gatenby and Gillies, 2004). The expression levels of HIF-1α and c-myc were, however, not affected by SIPA1 both in MDA-MB-231 and MCF7 cells (Figure 1C). Taken together, SIPA1 might be a modulator of aerobic glycolysis under normoxia in breast cancer cells.

When breast cancer cells were incubated with glucose as an exclusive carbon source, as shown in Figure 2A (left panel), the cellular ATP level was increased by knocking down SIPA1 in MDA-MB-231 and restored by forced expression of SIPA1 in 231si cells. A decrease in the level of ATP was also observed in MCF7 cells overexpressing SIPA1 (Figure 2B, left).

Since overexpressing SIPA1 in breast cancer cell lines increased glucose consumption and lactate production (Figures 1D,E), but decreased ATP production (Figures 2A,B), it is likely that there might exist a SIPA1-induced energy production switch from respiration to fermentation in breast cancer cells. To address this hypothesis, we treated cells with the mitochondrial ATPase inhibitor oligomycin, and found that 100 nM oligomycin treatment decreased ATP levels in breast cancer cells (Figures 2A,B, left panel). Notably, the ATP level in 231si cells was significantly decreased to one-sixth of its original value after oligomycin treatment, suggesting that a significant amount of ATP (about 80%) was produced by respiratory pathway in 231si cells, whereas as much as 70% ATP were produced via aerobic glycolysis in MDA-MB-231 cells or 231si cells with forced-expression of SIPA1 (Figure 2A, right panel). Similarly, about 50% ATP were derived from respiratory processes in parental MCF7 cells, whereas about 20% ATP were attributed to mitochondrial respiration in MCF7/SIPA1 cells (Figure 2B, right panel). These results strongly suggested that SIPA1 switched energy sources from respiration to fermentation.

We further examined the effect of SIPA1 on the mitochondrial ATP production. When cells were cultured with sodium pyruvate as a sole carbon source, the level of ATP production was significantly higher in SIPA1 low-expression cells than in SIPA1 high-expression cells (Figures 2C,D, green bars). It is, therefore, likely that a high level of SIPA1 expression down-modulated the intracellular ATP generation by switching the ATP source to aerobic glycolysis (Figures 2C,D, right panel). These results suggest that SIPA1 suppressed oxidative phosphorylation in mitochondria of breast cancer cells.

To address whether the alteration of ATP source by SIPA1 affects the sensitivity of cells to glycolytic and mitochondrial inhibitors, we treated the SIPA1-overexpressing or -knockdown breast cancer cells with 10 mM oxamate (an LDH inhibitor) or 50 nM oligomycin. As shown in Figure 2E, suppression of SIPA1 significantly increased the proliferation of MDA-MB-231. When MDA-MB-231 cells were treated with oxamate, the proliferation was significantly decreased, whereas the proliferation of MDA-MB-231 was not affected by the addition of oligomycin. Conversely, treatment of 231si cells with oligomycin greatly suppressed the cell proliferation, while oxamate had no effect on the proliferation of 231si cells. Additionally, overexpressing SIPA1 in MCF7 cells significantly reduced the cell proliferation. Interestingly, oligomycin treatment greatly decreased the proliferation of parental MCF7 cells, while no effect was observed for oxamate treatment. On the contrary, oxamate significantly suppressed the proliferation of MCF7/SIPA1 cells, but oligomycin failed to exhibit the suppressive activity (Figure 2F).

Pyruvate dehydrogenase kinase 1 (PDK1) is a key switch of tricarboxylic acid (TCA) cycle in mitochondria and inhibits mitochondrial oxidation of glucose by blocking the catalytic activity of pyruvate dehydrogenase (PDH) that converts pyruvate to acetyl-CoA (Dupuy et al., 2015). Interestingly, knocking down SIPA1 down-regulated PDK1 at both mRNA and protein expression levels in MDA-MB-231 and BT549 cells, whereas overexpression of SIPA1 up-regulated PDK1 expression in 231si and MCF7 cells (Figures 2G,H and Supplementary Figure S2). It is thus likely that a high level of SIPA1 expression alters the glucose metabolism from respiration to aerobic glycolysis by promoting the expression of glycolysis-related gene products and inhibiting the mitochondrial TCA cycle via up-regulating PDK1.

In order to elucidate the pathway by which SIPA1 modulates the expression of glycolysis-related genes and PDK1, we analyzed RNA-seq data from two pairs of breast cancer cell lines: parental and SIPA1-knowndown MDA-MB-231 cells and parental and SIPA1-overexpressing MCF7 cells. Differentially-expressed genes (DEGs) analysis revealed that 4,972 genes were down-regulated and 5,726 genes were up-regulated by knocking-down SIPA1 in MDA-MB-231 cells. In case of overexpressing SIPA1 in MCF7 cells, 1,423 genes were up-regulated and 1,352 genes were down-regulated. The DEGs which were positively correlated with SIPA1 expression in both cell lines were selected, from which 432 bilateral candidate genes were extracted (Figure 3A). Then 432 genes were subjected to GO enrichment analysis, and two annotations named “extracellular matrix organization” and “response to hypoxia” with significant enrichment score and statistically significant p-values were noted among the top 9 annotations (Figure 3B). We plotted the protein-protein interaction (PPI) network of the DEGs nominated in the annotation of “response to hypoxia” involved in cellular hypoxic responses as well as aerobic glycolysis processes. EPAS1 that encodes HIF-2α was identified as a hub gene to link its downstream genes including TGFB1, VEGFA and CA9 in sight of the PPI network (Figure 3C). The mRNA levels of EPAS1 as well as TGFB1, VEGFA and CA9, were confirmed to be positively correlated with those of SIPA1 in these two pairs of breast cancer cells (Figure 3D) and BT549, another TNBC cell line (Supplementary Figure S3). Meanwhile, knocking-down SIPA1 decreased HIF-2α expression in MDA-MB-231 cells, and overexpressing SIPA1 in MCF7 cells increased HIF-2α expression at a protein level (Figure 3E).

It was previously demonstrated that SIPA1 up-regulated the promoter activity of certain target genes in breast cancer cells (Zhang et al., 2015; Wang et al., 2020). In this study, we also demonstrated that SIPA1 could be located in the nuclei of MCF7/SIPA1 and three TNBC cells (Supplementary Figure S4). Dual luciferase reporter assay further revealed that overexpression of SIPA1 and its C-terminal region (540-1042aa), but not the N-terminal region (1-539aa), increased the transcriptional activity of the EPAS1 promoter transcriptional activity by three-fold, compared to controls (Figure 3F). We then conducted a chromatin immunoprecipitation (ChIP) assay to examine the direct interaction between SIPA1 and the EPAS1 promoter. Three pairs of specific primers corresponding to three segments of the EPAS1 promoter, Seg 1, Seg 2, and Seg 3 were employed (Figure 3G, upper panel). As shown in the lower panel of Figure 3G, Seg 2 (−1,400 to −1,240 bp) and Seg 3 (−893 to −700 bp) were specifically amplified from the cell lysates immunopecipitatated with anti-SIPA1 antibody. Moreover, the direct interaction between SIPA1 protein and the EPAS1 promoter segments was observed by EMSA, in which GST-tagged C-terminal SIPA1 specifically pulled down the biotin-labeled Seg 2, (Figure 3H). It is thus likely that SIPA1 directly interacted with the EPAS1 promoter, enhanced its promoter activity and promoted HIF-2α expression in breast cancer cells.

We next established stable HIF-2α knockdown MDA-MB-231 cell lines (Figure 4A) to address the hypothesis that the increased dependence of SIPA1-overexpressing breast cancer cells on aerobic glycolysis was mediated by HIF-2α. It was demonstrated that knocking-down HIF-2α in MDA-MB-231 breast cancer cells significantly decreased the expression of glycolysis-related gene products, such as HK2, LDHA, SLC2A1 and PDK1 (Figure 4A), consisting with the suppression of mRNA levels of glycolysis- and hypoxia-related genes (Supplementary Figure S5). In addition, lactate production and glucose consumption were also decreased in MDA-MB-231 cells by knocking down of HIF-2α (Figures 4B,C). Essentially, the same results were observed in BT549 breast cancer cells (Figures 4D–F).

We then compared the invasiveness of the breast cancer cells with SIPA1 or HIF-2α knockdown with that of parental cells by using a transwell assay. As shown in Figure 4G, knocking down either SIPA1 or HIF-2α significantly inhibited the invasion of MDA-MB-231 cells in vitro and the tumor cell translocation was further inhibited by the treatment with 20 mM oxamate. BT549 breast cancer cells behaved similarly after gene manipulation and drug treatment (Figure 4H). These results suggest that SIPA1 could enhance the aerobic glycolysis by up-regulating the expression of HIF-2α to promote the invasion of breast cancer cells in vitro.

We then examined the role of SIPA1/HIF-2α-mediated aerobic glycolysis in breast tumor growth and metastasis in vivo. We set up four groups of mice xenografted with (1) MDA-MB-231 breast cancer cells followed by 0.9% NaCl treatment (n = 6), (2) MDA-MB-231 followed by oxamate treatment (n = 6), (3) 231si cells followed by 0.9% NaCl treatment (1#: n = 6, 2#: n = 6), and (4) 231si cells followed by oxamate treatment (1#: n = 6, 2#: n = 6) under the xenograft experimental workflow (Figure 5A). No significant difference was observed on the body weight of mice in all the four groups during the treatments (Supplementary Figure S6). For two groups of mice inoculated with MDA-MB-231 cells, oxamate treatment significantly suppressed the tumor volumes and weights, compared to the group without oxamate treatment. As for the mice inoculated with 231si cells, the tumor growth rate was significantly lower than that of mice xenografted with parental MDA-MB-231 cells (Figures 5B,C).

Furthermore, the level of lactate in MDA-MB-231 xenograft tumors was much higher than that in 231si xenograft tumors. On the other hand, oxamate treatment decreased the level of lactate both in MDA-MB-231 and 231si xenograft tumors (Figure 5D). Also, SIPA1 knockdown decreased the expressions of EPAS1, glycolysis-related products and PDK1 in tumor tissues at mRNA and protein levels (Figures 5E,F).

Regarding lung metastasis, the number of nodules on the surface of the lungs in 231si xenografted mice was significantly fewer than that in MDA-MB-231 xenografted mice. In addition, the oxamate treatment further reduced the number of lung nodules in MDA-MB-231 xenografted mice (Figure 5G). The histopathological images of HE staining revealed that the numbers of tumors and the area of each nodule were decreased in the lungs in mice received 231si cells, compared to those inoculated with MDA-MB-231 cells. In addition, the oxamate treatment reduced the number of tumor nodules (Figure 5H). These results indicate that knocking down SIPA1 decreased breast tumor metastasis, which might be associated with the suppression of glycolysis.

To determine the clinical relevance of the above findings in breast cancers, we analyzed the TCGA database and found that SIPA1 was up-regulated in some breast cancer subtypes including luminal, Her2+ and TNBC (Figure 6A). In addition, we found that SIPA1 was aberrantly expressed in TNBC cells when compared to the other subtypes in GSE41313 dataset (Figure 6B). We also found a positive correlation between mRNA levels of SIPA1 and EPAS1 genes in both datasets (Figures 6C,D). A survival analysis by the log-rank test indicated that patients with EPAS1-high tumors had significantly lower rates of survival probability than patients with EPAS1-low tumors in TCGA (Figure 6E). These data suggest that SIPA1 induced the expression of HIF-2α and the expression of HIF-2α in tumor cells might result in poor prognosis in breast cancer patients.

Taken together, SIPA1/HIF-2α axis could be a key regulator of aerobic glycolysis, contributing to the switch from oxidative phosphorylation to aerobic glycolysis in breast cancer cells under an ambient oxygen condition, leading to tumor invasion and metastasis (Figure 6F).

Plasmids of pcDNA3-SIPA1, pcDNA3-N-SIPA1 encoding N-terminal half part of SIPA1 (1-539aa) and pcDNA3-C-SIPA1 encoding C-terminal half part of SIPA1 (540-1042aa) were constructed as previously described (Ma et al., 2021). SIPA1 and EPAS1 shRNA nucleotides (Supplementary Table S1) were cloned into pLKO.1-GFP (RRID: Addgene_30323) to construct pLKO.1-GFP-SIPA1 and pLKO.1-GFP-EPAS1, respectively. Lentivirus packaging plasmids, psPAX2 (RRID: Addgene_12260) and pMD2. G (RRID: Addgene_12259), and a reporter plasmid pRL-TK (RRID: Addgene_11313) were purchased from Merck & Co. (Kenilworth, NJ). EPAS1 promoter region (-1618bp to transcription starting site) was cloned into pGL4 (RRID: Addgene_48744) to give pGL4-EPAS1. pGEX-C-SIPA1 was constructed to express GST-tagged C-terminal half part SIPA1 protein.

Human cell lines MDA-MB-231, MCF7, and HEK293T were purchased and authorized from China Center for Type Culture Collection (CCTCC, Wuhan, China). BT549 was purchased from Procell (Wuhan, China) with short tandem repeat authentication. SUM159 were gifted from Dr. Peijing Zhang (Yao et al., 2018). Cells were cultured according to the previous study (Ma et al., 2021). The stable knockdown cells were established following the previous study (Zhang et al., 2015).

To detect glucose consumption and lactate production, 5 × 10⁴ cells were placed into wells and incubated in modified RPMI1640 medium with 15 mM glucose in the absence of glutamine and pyruvate for 5 h. After being washed with PBS, the culture medium and cells were harvested separately. Concentrations of glucose and lactate in the culture medium were measured using a glucose assay kit and a lactate assay kit (Nanjing Jiancheng Bioengineering Institute, Nanjing, China) according to the manufacturer’s instructions, respectively. To measure lactate in tumor tissues, 30 mg tumor specimens were pulverized in liquid nitrogen and lysed in 300 µl NP40 lysis buffer. After centrifugation for 5 min at 13,000 rpm at 4oC, the supernatant was harvested and examined for lactate and protein concentrations.

For the determination of the intracellular lactate dehydrogenase (LDH) activity, 1 × 10⁵ cells were seeded into the wells and incubated in media for 24 h. After being washed with PBS, cells were harvested and examined for LDH activity with an LDH assay kit (Nanjing Jiancheng Bioengineering Institute) according to the manufacturer’s instructions. For extracellular LDH activity assay, 2 ×10⁵ cells were seeded into the wells and incubated with medium for 5 h, and then supernatants were collected and examined for LDH activity by using an LDH assay kit.

ATP levels in breast cancer cells were measured using an ATP assay kit (Beyotime, Nanjing, China) according to the manufacturer’s instructions. The amount of total ATP was calculated based on luminescence measured on a FlexStation 3 luminescence reader (Molecular Devices, San Jose, CA). To measure the levels of glycolytic ATP production, 5 × 10⁴ cells were seeded into wells and incubated in media containing 100 nM oligomycin (Thermo Fisher Scientific) for 5 h. After being washed with PBS, the cells were lysed and examined for ATP levels using an ATP assay kit according to the manufacturer’s protocol. To measure the ATP production in mitochondria, 5 × 10⁴ cells were seeded into wells and incubated in media containing 10 mM pyruvate in the absence of glucose and glutamine for 5 h. After being washed with PBS, the cells were lysed and examined for ATP levels and protein concentrations.

Quantitative real-time polymerase chain reaction (qRT-PCR) and chromatin immunoprecipitation PCR (ChIP-PCR) were conducted as previously described (Zhang et al., 2015). The specific primers are listed in Supplementary Tables S2 and S3.

Total RNA was isolated using TRIzol reagent (Thermo Fisher Scientific). Transcriptome sequencing was conducted by Novogene Co. Ltd. (Beijing, China). Gene expression levels for each transcript were estimated as the number of reads per kilobase of exon model per million mapped reads (RPKM) using only uniquely mapped reads in exonic regions. A gene was considered differentially expressed if its expression differed between any two samples with the fold change >2 and the p value <0.05. DAVID (RRID: SCR_001881) online tool (https://david.abcc.ncifcrf.gov) was used for Gene Ontology (GO) enrichment analysis and KEGG (RRID: SCR_012773) pathway enrichment analysis. STRING (RRID: SCR_005223) online tool (https://string-db.org) was used to assess the protein-protein interaction of cluster genes.

EPAS1 promoter region was amplified by PCR from MDA-MB-231 genomic DNA, and cloned into a pGL-4 basic luciferase expression vector (Promega, Madison, WI). Reporter assays were performed using HEK293T cells transfected with the indicated plasmids and analyzed using a Dual-Luciferase Reporter Assay kit (Promega). The luciferase activity was measured by FlexStation 3 (Molecular Devices). The expression levels were normalized with respect to those for cells co-transfected with a renilla plasmid.

Cell lyzates, cytoplasmic lyzates, and nuclear extracts were prepared as previously described (Zhang et al., 2015). Primary antibodies for SIPA1 (Abcam, ab85928, RRID: AB_1925436), LDHA (Abcam, ab47010, RRID: AB_1952042), HK2 (Abcam, ab104836, RRID: AB_10710018), SLC2A1 (Abcam, ab115730), HIF-2α (Abcam, ab207607, RRID: AB_2618694), HIF-1α (Abcam, ab179483), c-myc (Cell Signaling Technology, Danvers, MA, 9,402), PDK1 (Abcam, ab207450), and secondary antibodies (Cell Signaling Technology, 91196 and 7,074) were used for blotting, and β-tubulin (Absin Bioscience Inc., Shanghai, China, abs830032) was included as a control. The relative amount of protein was quantitated by an ImageJ software.

For immunohistochemistry, the xenograft tumor tissue slides underwent deparaffinization, rehydration, and antigen-retrieval and were then incubated with the primary antibody: HIF-2α (Abcam, ab207607), SLC2A1 (Abcam, ab115730) or PDK1 (Abcam, ab207450). Staining image was monitored using 80i fluorescence microscope (Nikon, Tokyo, Japan).

GST-tagged C-terminal half of SIPA1 was expressed in E. coli BL21 (DE3) and purified using glutathione-Sepharose beads. The purified protein (2 µg) was incubated with competitive biotin-labeled double-stranded DNA segments on ice for 5 min. Then, 1 µ mole of unlabeled double-stranded DNA segments, which had been amplified from EPAS1 promoter, were added to the mixture, and the protein/DNA mixture was incubated at room temperature for 30 min. The proteins bound to biotin-labeled Seg 2 were resolved by native 6.5% PAGE and imaged by immunoblot assay.

Cell growth and invasion in vitro was measured as described previously (Zhang et al., 2015).

All animal studies were conducted after approval by the Institutional Animal Care Committee of Huazhong University of Science and Technology. For xenograft experiments, female BALB/c nude mice were divided into two groups, and MDA-MB-231 and 231si cells (2 × 10⁶/mouse) were subcutaneously injected with matrigel (v/v = 3:1) into the right mammary pad for each group, respectively. Five days later, a subset of 6 mice in each group received an intraperitoneal injection with 15 mg oxamate or 0.9% NaCl. The administration was done every 24 h for the next 14 days. The tumor volume was determined every 24 h with vernier caliper according to the following formula: volume (mm³) = [width (mm)]² × [length (mm)]/2. Subsequently, all the tumors were surgically removed and weighed, and the wounds were stitched with surgical suture to maintain animals for 15 additional days. Finally, all the mice were sacrificed by cervical dislocation, and the lungs of each mouse were dissected and immersed in formalin, then analyzed by haematoxylin and eosin (HE) staining.

Dataset GSE41313 was downloaded from GEO (Riaz et al., 2013). In GSE41313, mRNA expression levels in 153 samples from 51 breast cancer cell lines were extracted from the general public license (GPL) 13158 platform (Affymetrix, Santa Clara, CA). Fragments per kilobase of exon model per million mapped fragments (FPKM) were used to compare gene expressions in different groups. Data acquisition and application from The Cancer Genome Atlas (TCGA) were performed in accordance with TCGA publication guidelines and data access policies (Chandrashekar et al., 2017).

The data were presented as mean ± s.d. Student’s t-test was used to evaluate p values. One-way or two-way ANOVA were used to compare multiple testing correction within multiple groups. A p-value of 0.05 or lower is considered to be statistically significant.

The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Chen et al., 2021) in National Genomics Data Center (CNCB-NGDC Members and Partners, 2021), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number HRA001265 that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human.