Human breast cancer MCF7 and MDA-MB-231 cell lines were from American Type Culture Collection (ATCC; Manassas, VA, USA). Cells were cultured in DMEM supplemented with 10% fetal bovine serum (Biological Industries; Kibbutz Beit Haemek, Israel) and 1% penicillin-streptomycin (Gibco; Carlsbad, CA, USA) at 37°C in 5% CO₂ atmosphere as described in other study (Mertens-Talcott et al., 2007; Kim et al., 2010; Wardi et al., 2014). For hypoxic treatment, cells were maintained in an in vivo 200 hypoxia chamber (Ruskinn; Bridgend, UK) with 1% O₂/ 5% CO₂/ 94% N₂ atmosphere for 24 h as previously described(Nagpal et al., 2015; He et al., 2017).

Total proteins were extracted with T-PER Reagent (Thermo Scientific; San Jose, CA) as described previously (Tan et al., 2014). In brief, cells (~1 × 10⁷) were detached with trypsin, washed twice with ice-cold 1 × PBS (0.01 M phosphate buffer containing 0.15 M NaCl, pH 7.4), lysed with 1 mL T-PER Reagent containing protease inhibitor cocktail (Sigma-Aldrich; St. Louis, MO, USA) and phosphatase inhibitor cocktail (Sigma-Aldrich), incubated for 30 min on ice, homogenized, and centrifuged at 12,000 rpm for 15 min. The supernatant was collected and stored at −80°C. Protein concentration was determined by BCA assay (Beyotime; Haimen, Jiangsu, China).

Western blotting was performed as described previously (Tan et al., 2014). In brief, total proteins (30 μg) from normoxia- and hypoxia-treated samples were separated by 7.5% SDS-PAGE. Gels were transferred onto polyvinylidene difluoride (PVDF) membranes with Trans-Blot Turbo Transfer System (Bio-Rad; Hercules, CA). Membranes were soaked in 5% skim milk in TBST (20 mM Tris-HCl, 150 mM NaCl, 0.05% Tween 20, pH 8.0) for 2 h at 37°C, probed with primary antibodies against MGAT3 (1:500; ab135514; Abcam, Cambridge, MA, UK), fibronectin (1:1000; ab2413; Abcam), E-cadherin (1:10000; 610181; BD Biosciences, San Jose, CA, USA), β-catenin (1:5000; ab32572; Abcam), tubulin (1:5000; T7816; Sigma-Aldrich, St. Louis, MO, USA), HIF-1α (1:1000; 3716; Cell Signaling Technology, Beverly, MA, USA), GLUT1 (1:5000; ab40084; Abcam), AKT (pan) (1:1000; 4685; Cell Signaling Technology), and p-AKT (Ser473) (1:2000; 4060; Cell Signaling Technology) overnight at 4°C and incubated with appropriate HRP-conjugated secondary antibody. Specific bands were visualized with a Pro-light HRP Kit (Tiangen; Beijing, China).

Wound healing assay was performed as described previously (Castro et al., 2018). Migratory capacity of cells was determined by wound healing assay. In brief, cells (2 × 10⁵ per well, in a six-well plate) were cultured overnight and treated as described above. Three separate wounds were scratched with a pipette tip on the cell monolayer in each well, moving perpendicularly to a line drawn at the bottom of the plate. Cells were rinsed twice with 1 × PBS twice, added with fresh serum-free medium, and wounds at marked lines were photographed. After 24 h incubation at 37°C under normoxia or hypoxia, cells were washed with ice-cold 1 × PBS, and wound tracks were photographed and marked using ImagePro Plus software (Media Cybernetics; Silver Spring, MD, USA).

Cell proliferation was performed as described previously (Yu et al., 2008). Cells were plated in 96-well plates, and incubated 4 h with CellTiter 96 AQ_ueous. One Solution Cell Proliferation Assay (MTS) solution (Promega; Madison, WI, USA). MTS products in supernatant were transferred into 96-well microtiter plates, and absorbance at 490 nm was determined.

Colony formation was performed as described previously(Yu et al., 2008). Cells (2,500 per well) were plated in a 6-cm dish, and grown 1–2 weeks until small colonies were clearly seen. Medium was discarded, cells were rinsed twice with 1 × PBS, fixed with 2% fresh paraformaldehyde, stained with crystal violet solution, and photos were taken. 10% acetic acid solution (1 mL) was added to dissolve the crystal violet. Optical density at 595 nm (OD 595) was measured.

RNA isolation was performed as described previously (Tan et al., 2014). Cells (1 × 10⁵ per well in a six-well plate) were cultured and treated as described above. Total RNA was isolated using an RNApure Tissue Kit (CWBiotech; Beijing) as per the manufacturer's instructions.

Real-time PCR was performed as described previously (Tan et al., 2014). Total RNA was extracted as above. Primers were selected from qPrimerDepot (primerdepot.nci.nih.gov/) (Cui et al., 2007). First-strand cDNA was synthesized from total RNA using a ReverTra Ace-α First-strand cDNA Synthesis Kit (Toyobo; Osaka, Japan). Quantitative real-time PCR was performed by LightCycler-based SYBR Green I dye detection with UltraSYBR Mixture (CWBiotech). Gene expression was quantified by the 2^−ΔΔCT method (Livak and Schmittgen, 2001).

Lectin microarrays were constructed and analyzed as described previously (Qin et al., 2012; Yu et al., 2012). In brief, 37 commercial lectins from Vector Laboratories (Burlingame, CA), Sigma-Aldrich, and Calbiochem Merck (Darmstadt, Germany) were immobilized onto a solid support at high spatial density. Glycoprotein samples labeled with fluorescent dye Cy3 (GE Healthcare; Buckinghamshire, UK) were applied to the lectin microarrays, and the arrays were scanned with a GenePix 4000B confocal scanner (Axon Instruments; Union City, CA).

Lectin staining was performed as described previously (Tan et al., 2014). Cells were cultured in 24-well plates with sterilized coverslips to obtain monolayers with 70–80% confluence. Cells were washed with ice-cold 1 × PBS, immobilized with 2% fresh paraformaldehyde for 15 min at room temperature (RT), permeabilized with 0.2% Triton X-100 in 1 × PBS for 10 min at RT, and blocked with 5% BSA in 1 × PBS for 1 h at 37°C. Fixed cells were incubated with 15–20 μg/mL Cy3 fluorescein-labeled lectins (Con A, MAL-I, LCA, PHA-E) in 5% BSA for 3 h in the dark at RT, washed with 1 × PBS, stained with 20 μg/mL DAPI in 1 × PBS for 10 min at RT, washed again with 1 × PBS, and photographed with a fluorescence microscope (model Eclipse E600; Nikon; Tokyo, Japan).

N-glycans were separated as described previously (Tan et al., 2014). In brief, total proteins (2 mg) from each cell line were concentrated and desalted using a size-exclusion spin ultrafiltration unit (Amicon Ultra-0.5 10 KD, Millipore; Billerica, MA). Proteins were denatured with 8 M urea, 10 mM DTT, and 10 mM IAM (Sigma-Aldrich) and centrifuged. The sample was further digested with PNGase F (New England BioLabs; Ipswich, MA) overnight at 37°C. Released N-glycans were collected and lyophilized.

N-glycans were desalted as described previously (Tan et al., 2014). N-glycans were desalted using Sepharose 4B (Sigma-Aldrich) as described previously (Tan et al., 2014). In brief, Sepharose 4B in a microtube was pre-equilibrated with methanol/ H₂O (1:1 v/v) (MW) and 1-butanol/ methanol/ H₂O (5:1:1 v/v/v) (BMW). Glycans were dissolved in 500 μL BMW, added to the Sepharose 4B, and the mixture was shaken gently and washed with BMW. N-glycans were eluted with MW, and the eluent was collected and lyophilized.

Mass spectrometry analysis of N-glycans were performed as described previously (Tan et al., 2014). N-glycans were characterized by MALDI-TOF-MS (UltrafleXtreme, Bruker Daltonics; Bremen, Germany). Lyophilized N-glycans were resuspended in 10 μL MW, and 1 μL of the mixture was spotted onto an MTP AnchorChip sample target and air-dried. 1 μL of 20 mg/mL 2, 5-dihydroxybenzoic acid (DHB) in MW was spotted to recrystallize the glycans. Measurements were taken in positive-ion mode, and m/z data were analyzed and N-glycan structures were annotated using the GlycoWorkbench software program (code. google.com/p/glycoworkbench/). Relative intensity was analyzed and generated using FlexAnalysis software (Bruker Daltonics) based on MALDI-TOF-MS intensity. Relative proportion was calculated by accumulating the relative intensity of a given type N-glycan.

MGAT3 was amplified via PCR and linked to lentiviral overexpression vector pLVX-AcGFP1-N1 (Takara; Shiga, Japan). Lentiviral shRNA vector is constructed based on pLVX-shRNA2-Puro (Takara). Target sequences are listed below. Lentiviral vectors were packed in HEK293T via the packaging system, together with pMD2.G and psPAX2 (Addgene; Cambridge, MA, USA). Transfected MCF7 cells were selected and enriched by adding puromycin to culture medium.

Data are presented as mean ± SD. Statistical significance of differences between means was evaluated by Student's t-test at p < 0.05.

Previous studies using various cancer cell models have demonstrated induction of EMT by hypoxic conditions (Zhou et al., 2009; Copple, 2010). In the present study, we exposed breast cancer cell lines MCF7 and MDA-MB-231 to hypoxic environments. HIF-1α, an indicator of hypoxic conditions, was strongly expressed in hypoxia-treated cells (Figure 1A). The treated cells also showed loss of an epithelial character (E-cadherin) and acquisition of a mesenchymal character (fibronectin), typical hallmarks of EMT, and elevated expression of glucose transporter-1 protein (GLUT1), a putative intrinsic cellular marker of hypoxia regulated by HIF-1α (Figure 1A).

Hypoxia treatment (1% O₂) of the two cell lines for 72 h resulted in altered morphology in comparison with normoxia (21% O₂) treated cells. Hypoxia-treated cells showed elongation and flattening (Figure 1B), and strongly enhanced migratory capacity, particularly in MCF7 (Figure 1C).

Alterations in glycosylation have been observed in numerous physiological and pathological processes. To identify specific N-glycan alterations during hypoxia, we used MALDI-TOF-MS to compare glycomic profiles of normoxia- vs. hypoxia-treated cells. Compositional and structural features were indicated in representative MALDI-TOF-MS spectra with signal-to-noise ratios >5 (Figures 2, 3, Tables S1, S2; Supplementary Information).

Thirty-four individual N-glycan masses were detected in MCF7, and 19 in MDA-MB-231. Numbers of distinct N-glycans were: 33 in normoxia-treated MCF7, 25 in hypoxia-treated MCF7, 17 in normoxia-treated MDA-MB-231, and 16 in hypoxia-treated MDA-MB-231. N-glycan structures found exclusively under hypoxic conditions in each of the two cell lines were mostly complex or hybrid type rather than high-mannose type.

We examined quantitative differences in various N-glycan types under the two conditions, as shown by the relative proportions summarized in Table 1. Proportions of bisecting GlcNAc structures were reduced under hypoxia vs. normoxia (2.51 vs. 5.35% in MCF7; 3.62 vs. 5.65% in MDA-MB-231). Proportions under hypoxia vs. normoxia were also reduced for hybrid type N-glycans (2.94 vs. 4.34% in MCF7; 1.47 vs. 3.66% in MDA-MB-231) and for sialylation (0.43 vs. 0.76% in MCF7; 0.00 vs. 3.02% in MDA-MB-231). Proportions for fucosylation under hypoxia vs. normoxia were elevated in MCF7 (22.57 vs. 17.03%) but reduced in MDA-MB-231 (13.58 vs. 21.62%). Detailed information regarding substituents and branching patterns of N-glycans was obtained by MALDI-TOF/TOF-MS/MS (Figure S1).

Lectin microarray analysis was used to identify specific glycopatterns in hypoxia-treated vs. normoxia-treated cells. The microarray was used to analyze fine glycan structures of glycoproteins in cells under the two conditions contained 37 lectins (Table S3), including two negative controls (BSA) and one positive control (Cy3-BSA). Under hypoxia, significant differences (>1.5-fold or <0.67-fold) were observed for glycans recognized by 12 different lectins in MCF7 and by 22 different lectins in MDA-MB-231 (Figure 4A). In both cell lines, bisecting GlcNAc structures (recognized by PHA-E), core fucosylation (recognized by PSA), and T antigen (recognized by PNA) were significantly suppressed, whereas non-substituted α1,6-Man structures (recognized by HHL) was elevated (Tables 2, 3). A heatmap produced by complete hierarchical clustering and visualization using the HemI 1.0 software program (hemi.biocuckoo.org/down.php) (Deng et al., 2014) showed side-by-side clustering of normoxia- and hypoxia-treated cells in the dendrogram for both MCF7 and MDA-MB-231 (Figure 4A).

Glycan profiles under hypoxic conditions were further investigated and confirmed by histochemistry using PHA-E, MAL-I, LCA, and Con A (Figure 4B). Both MCF7 and MDA-MB-231 showed strong reduction of PHA-E and increase of Con A fluorescence signals. MCF7 showed reduction of MAL-I and MDA-MB-231 showed reduction of LCA signals, consistently with results of lectin microarray analysis. On the other hand, MDA-MB-231 microarray results using Con A were inconsistent with the elevated high-mannose N-glycan levels detected by mass spectrometry, lectin histochemistry, and microarray results using NPA and HHL. This apparent discrepancy may be due to random orientation of Con A following immobilization on glass slides (Qin et al., 2012). We examined expression of MGAT3 (which catalyzes synthesis of bisecting GlcNAc structures) in hypoxia-treated MCF7 and MDA-MB-231, and found that it was decreased at protein level (Figure 4C).

To elucidate the functional effects of reduced levels of bisecting GlcNAc structures and MGAT3 during hypoxia, we overexpressed MGAT3 gene in MCF7. Cells were stably transduced with a green fluorescent protein (GFP)-marked lentivirus carrying mock (MCF7/mock) or MGAT3 (MCF7/MGAT3-1/2). MGAT3 overexpression was confirmed by western blotting (Figure 5A) and PHA-E lectin blotting (Figure 5B), and resulted in increased expression of MGAT3 and bisecting GlcNAc structures. The elevation of MGAT3 level inhibited proliferation (Figure 5C), colony formation (Figures 5D,E), and migration (Figures 5F,H) of the cells.

Previous studies have shown that MGAT3 regulates activation of signaling pathways such as extracellular signal-regulated kinase (ERK) 1/2 or protein kinase B (AKT) (Miwa et al., 2013), and is involved in EMT and its reversed process, mesenchymal-epithelial transition (MET) (Pinho et al., 2012; Xu et al., 2012). We therefore examined possible effects of aberrant MGAT3 expression on relevant signaling pathway during hypoxia. In mock transfectants under hypoxia, reduced E-cadherin expression, increased fibronectin expression, and activation of AKT signaling were observed (Figure 5G). In MGAT3 transfectants under hypoxia, MGAT3 overexpression had significant blocking effects on downregulation of E-cadherin expression, upregulation of HIF-1α expression, and AKT signaling activation, and a slight blocking effect on upregulation of fibronectin expression (Figure 5G). These findings, taken together, indicate that MGAT3 overexpression inhibits hypoxia-induced EMT.

We silenced MGAT3 expression in MCF7 using MGAT3 shRNAs (MCF7/shMGAT3-1/2) or shNC (MCF7/shNC). MGAT3 knockdown was confirmed by western blotting (Figure 6A). Cell proliferation and migration were enhanced in MGAT3-shRNA transfectants (Figures 6B,C,E). Under hypoxia, MGAT3 knockdown in these cells promoted downregulation of E-cadherin expression and AKT signaling activation, but had no notable effect on β-catenin expression (Figure 6D).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The handling editor is currently co-organizing a Research Topic with one of the authors FG, and confirms the absence of any other collaboration.