Phostine PST3.1a Targets MGAT5 and Inhibits Glioblastoma-Initiating Cell Invasiveness and Proliferation Free

Control of cell surface glycosylation homeostasis occurs during cell proliferation, differentiation, and invasiveness and has been correlated with adaptation to the microenvironment and to disease development (1). Mannoside acetyl glucosaminyltransferase 5 (MGAT5) overexpression is associated with malignancies and correlates with cell migration, invasion, and epithelial–mesenchymal transition (EMT; refs. 2–4). MGAT5 located in the medial Golgi catalyses the addition of β1,6-N-acetylglucosmine (GlcNAc) to selectively generate tri (2, 2, 6)- and tetra- (2, 4, 2, 6)antenna-like oligosaccharides, intermediates of complex type N-glycans (5). Gliomas express highly variable levels of MGAT5 mRNA (6), and MGAT5 enzymatic activity changes through the course of glioma genesis (7, 8). This leads to a dynamic adhesion of the cells to the extracellular matrix (ECM), which appears crucial for glioblastoma multiforme (GBM) invasiveness (9, 10). GBM recurrent lesions are formed by heterogeneous but widely infiltrative small anaplastic cells responsible for the fatal outcome (11, 12).

Glycans and glycan mimetics have been successfully engineered as therapeutic agents against type II diabetes, viral and bacterial infectious diseases, and cardiovascular pathologies (13). No organ toxicity (14), brain malformation, cell number, and architecture (15) were observed in MGAT5^−/− mice in which N-glycan products were missing. Moreover, in these mice, mammary tumor growth and metastases induced by the polyomavirus middle T oncogene were drastically reduced (14). These results confirmed the initial observations showing that MDAY-D2 cells deficient in MGAT5 activity lose their metastatic potential (2).

We have established a strategy based on rational drug design to create glycomimetic compounds, phosphinosugars also called phostines, in which the hemiacetal group of a hexopyranose was replaced by a chemically and configurationally stable phosphinolactone group (16, 17) with the aim to interfere with glycosylation in cancer cells. In vitro screening of several derivatives bearing this scaffold described by Clarion and colleagues (16, 17) revealed potent activities. Among the active phostines, the compound PST3.1a (3-hydroxy-4,5-bis-benzyloxy-6-benzyloxymethyl-2-phenyl2-oxo-2λ5-[1,2]oxaphosphinane; Fig. 1A) was selected according to a versatile screening based on cellular phenotype assays linked to cell glycosylation alterations combined with enzymatic activity tests.

In this article, we decipher the mode of action of PST3.1a, selected for its ability to inhibit MGAT5, its stereoselectivity, and selectivity toward the CNS cluster within the NCI panel of cancer cell lines (17). Physiologic and signaling effects of PST3.1a were analyzed on two GIC lines (Gli4 and GliT) grown either in proliferation or differentiation conditions (10). The resultant effects were mostly similar to those triggered by siRNA-mediated inhibition of MGAT5 expression. The in vivo pharmacologic activity of PST3.1a was then evaluated in orthotopic graft models of GBM using Gli4 and GliT. Finally, using the NCI-60 cancer cell line panel, we showed that mitochondrial gene expression was negatively correlated with PST3.1a cytotoxicity, whereas the expression of genes involved in EMT was positively correlated with the response.

GBM stem cells were obtained using the classical nonadherent neurospheres (NS) isolation and culture method as described by Dromard and colleagues (18) and adapted by Guichet and colleagues for Gli4, Gli7 (10), and GliT. SNB75 cells were obtained in 2010 from Prof. Shoucheng Ning and Prof. Susan J. Knox (Stanford University, Stanford, CA). All cells were checked monthly for mycoplasma (LT07-418, Lonza) and used under passage 15.

All experiments involving animals were submitted to the local committee (Division Départementale de la Protection des Populations de l'Hérault) and approved under the Project Licence: C34-172-36. The lead experimenter holds a Level 1 Personal Licence under the reference: I-34UnivMontp-F1-12.

Phostine solutions were prepared in 100% DMSO and diluted to obtain in each condition a final percentage of 0.3% DMSO.

The MTT assay was carried out as described previously at 48 and 72 hours after drug addition (17). The number of cells per distance of migration was quantified using the Axio Imager software. Briefly, concentric arcs were defined from the center of the sphere with regular intervals (see inset, Fig. 5B), and the number of cells present between two successive arcs was counted, defining a number of cells by range of migration.

For migration experiments, inserts of Boyden chambers (BD Biosciences) were coated overnight with 20 μg/mL of either laminin, fibronectin, or vitronectin (BD Biosciences). For invasion, 100 μL of Matrigel (BD Matrigel TM Matrix, mouse EHS tumor, LDEV; BD Biosciences) was diluted 33 times in PBS and added onto the insert. Cells (50 × 10⁵) suspended in DMEM supplemented with 0.1% BSA were plated into the upper compartment and incubated for 24 hours with and without treatment. Cell culture medium (0.5 mL) containing 10% FCS was placed in the lower compartment to facilitate chemoattraction. Cells attached to the lower side were counted using a Zeiss axiophot microscope.

We developed a nanosuspension using the Nanoedge technology allowing an oral administration of doses up to 50 mg/kg. The average particle size was estimated at 300 nm with a polydispersity index around 0.15.

For the detection and quantification of PST3.1a in the brains, 3 mice were treated with 27 mg/kg of PST3.1a by oral administration. The animals were sacrificed 1 hour posttreatment by lethal administration of pentobarbital. After exsanguination and washing with 5 mL of NaCl, the brains were removed and stored at −80°C until shipment. The quantifications and analyses were performed by amatsiavogadro group.

Gli4 or GliT cells were dissociated and resuspended in PBS at 0.5 × 10⁵ cells per μL. Three microliters of cells (1.5 × 10⁵) were injected into the striatum (1 mm rostral, 2 mm lateral, and 2.5-mm depth from Bregma) of 6-week-old female NMRI-nude mice (Janvier Labs) under isoflurane anesthesia. At the end of the surgery, the remaining cells were seeded to check for cell viability. Transplanted animals were allocated randomly to each treatment group by the end of the surgery.

Ten weeks postsurgery, the animals were treated with 2 daily oral administrations of PST3.1a (15 mg/kg) for 10 consecutive days. Control animals received the same volume of excipients. The animals were sacrificed at the end of the treatment and the brains removed and processed for histochemistry.

For the survival study, 1.5 × 10⁵ Gli4 GIC cells were transplanted by stereotaxic injection into the striatum of nude mice. Ten weeks posttransplantation, 28 mice were randomized in four groups: control, PST3.1a, temozolomide, and PST3.1a + temozolomide. PST3.1a was daily administered by oral gavage: 2 × 25 mg/kg orally, for two cycles of 30 days separated 19 days. Temozolomide was administered intraperitoneally: 3 cycles of 4 administrations (40 mg/kg) every other day. Cycles were separated by 4 days.

Animals were sacrificed when signs indicating brain tumor development (prostration, hunched back, weight loss) were detected, and their brains were checked for tumor growth.

The enzymes, human recombinant MGAT3, MGAT5, and ectonucleoside triphosphate diphosphohydrolase3 (ENTPD3), were obtained from R&D Systems. The MGAT donor substrate, UDP-N-acetylglucosamine, was obtained from Sigma. The MGAT acceptor substrate, a biantennary N-linked core pentasaccharide, was obtained from V-Labs.

The activities of MGAT3 and MGAT5 were measured according to the procedure described by R&D Systems with slight modifications using the malachite green method (19) to determine the amount of Pi released by ENTPD3. PST3.1a was solubilized in DMSO, and activities were measured in the presence of 2% DMSO.

SiMGAT5 and scrambled siRNA were ordered from CliniSciences. Cells were transfected with 100 nmol/L of siRNA using an Amaxa Nucleofector with the Mouse Neural Stem Cell Kit (Lonza).

Gli4 cells transfected or not with siMGAT5 and treated with DMSO (controls) or PST3.1a were stained with the PHA-L lectin for binding to β1,6 N-acetylglucosamine moieties or with CD15 coupled to PE (CD15-PE). Biotinylated PHA-L (Vector Laboratories) was used at 4 μg/mL followed by streptavidin-FITC (CliniSciences, 1/500) in PBS + 0.1% sodium azide. Labeled cells were sorted using the BD Accuri C6 flow cytometer.

After a 48-hour treatment with or without PST3.1a, Gli4 or GliT cells were dissociated and plated as single cells in poly-HEMA–coated 96-well plates using a BD FACSAria flow cytometer, in the presence or in the absence of PST3.1a. The number of colonies in each condition was counted 3 weeks postplating.

Twenty micrograms of protein lysate was separated by SDS-PAGE. Immunoreactive protein bands were visualized using ECL Western blotting detection reagents (Bio-Rad). The ECL analysis system was used for detection in accordance with the manufacturer's instructions. Immunostainings for HuNu (Millipore) and DCX (Chemicon) were performed following standard protocols. The sections were mounted in Fluoromount medium.

Gli4 cells were first dissociated and seeded at 25,000 cells per mL in a 24-well plate in proliferation conditions. PST3.1a was added at 5 μmol/L, and neurosphere formation was monitored for 24 hours in a 37°C chamber with 5% CO₂. In the control video, acquisition was taken every 14 minutes. In the treated well, acquisition was taken every 3 minutes. The videos were reconstructed with the Cells and Maps software (bram.org/serf/CellsAndMaps.php) and then converted in MP4 format.

The presented experiments were carried out at least in biological triplicates (except for the in vivo experiments, which were done twice). EC₅₀ values were calculated using the Hill equation of the dose–log response curves using serf software. IC₅₀s were computed from the means of three different experiments [n = 3 (toxicity), 10 (invasion, migration) for each concentration] expressed as the percentage of control value in each experiment. Values are expressed as mean ± SEM or 95% confidence intervals.

The oxaphosphinanes core of PST3.1a is conformationally and stereochemically related to the structure of glucose (Fig. 1A) or more specifically to C-arylglycosides (e.g., canagliflozin, ertugliflozin). This structural analogy is supported by other series of phosphorus heterocycles embedding a phosphinolactone group as a surrogate of lactol group present in glycosides. The combination of the structural analogy and the phenotypic screening (16, 17) prompted us to evaluate its inhibitory effects on MGAT5 enzymatic activity. In our assays, using a synthetic pentasaccharide as glycan acceptor and UDP-GlcNac as the donor substrate, the apparent K_m value for MGAT5 was 14 mmol/L and the V_max 4 μmol/h/mg. PST3.1a showed an inhibitory effect against MGAT5 activity with an IC₅₀ of 2 μmol/L (Fig. 1B), whereas no inhibition of MGAT3 enzymatic activity was found (Fig. 1C).

Gli4 cells express several typical markers of neural precursors (CD133, CD15, Nestin, OLIG2, and SOX2) and are multipotent (10). They generate NS when cultivated in growth factor–containing medium (proliferative medium; ref. 10). The major N-glycan structures expressed by Gli4NS are presented in Supplementary Fig. S1A. This glycome was next compared with the glycomes of Gli4NS grown for 10 days in the absence or presence of 5 μmol/L PST3.1a (Supplementary Fig. S1B). Relatively to untreated cells, the abundance of ions at m/z 1,907, 2,081, 2,111, 2,285, and 2,489 were all significantly reduced in treated Gli4NS (Supplementary Fig. S1A and S1B). These data indicated that multibranched β1,6-GlcNAc mannose intermediate N-glycans were decreased by the treatment confirming the in cellulo MGAT5 inhibition by PST3.1a.

To measure whether the N-glycosylation pattern of Gli4NS was altered beyond the triantennary intermediate glycans, a glycoprofile was performed at 10 days using a set of lectins. The glycoprofile analysis confirmed a lower binding of GLI4NS-treated cells to PHA-L lectins (Fig. 1D; Supplementary Fig. S2). The status of glycan sialylation was estimated by measuring cell binding through lectin recognition of α2,3-and α2,6-linkages in N-glycans. Figure 1D indicates a lower binding of PST3.1a–treated cells to the MAA and SNA lectins. PST3.1a did not affect the binding to the fucosylated core recognized by the PSA lectin, indicating that PST3.1a treatment does not inhibit the fucosyl-transferase Fuc-TVIII. Finally, the Tn-antigen was mapped with the MPA, BPA, and AIA lectins. Although PST3.1a-treated cells bound to a lower amount to the MPA lectin (Supplementary Fig. S2), this differential binding to the Tn-antigen was not confirmed by the binding to BPA and AIA (Supplementary Fig. S2). These data indicate that O-glycosylation was not affected. Analysis of the O-linked glycan structures by the glycome analysis confirmed the absence of effect on O-glycosylation (Supplementary Fig. S3). A more complete analysis of the Gli4 glycoprofile presented in Supplementary Fig. S3 indicates that Gli4NS did not bind or bound very poorly to lectins recognizing polymeric GlcNAc (GSL-II, PWM, STA, UEA-II), terminal fucosylation (UEA-I) and glycan structures containing GalNAc (DBA, ECA) or αGal motifs (GSLI-β4, PAI-L).

After 10 days of treatment, the inhibition of MGAT5 activity by PST3.1a on Gli4NS could account for the observed reduction of β1,6 glycans on the glycome and glycolprofile analyses. However, the effects on multibranched β1,6-GlcNAc N-glycans analyzed by FACS indicate that during the first 48 hours of treatment, PST3.1a had no effect on the binding of PHA-L on Gli4NS (Fig. 1E).

We therefore monitored MGAT5 expression in Gli4NS treated or not with PST3.1a. Figure 2A indicates that MGAT5 expression first increases during the first 48 hours of PST3.1a treatment, and then decreases to reach control levels after 5 days (Fig. 2A; Supplementary Fig. S4). This could explain the absence of effect on PHA-L binding after 48 hours of treatment as observed by FACS analysis (Fig. 1E). To test this hypothesis, we inhibited the expression of MGAT5 by siRNA (Fig. 2B and E; Supplementary Fig. S4) to counteract its overexpression. As expected, inhibition of MGAT5 expression in Gli4NS-siMGAT5 led to lower binding of the PHA-L lectin after 48 hours (Fig. 1E). In addition, PST3.1a treatment of Gli4NS-siMGAT5 cells led to a further reduction in PHA-L binding, underlying the inhibitory effect of PST3.1a on remaining MGAT5 activity (Fig. 1E). Similar effects were observed with another GIC (GliT; Fig. 1E).

MGAT5 modulates the cell surface glycosylation level and markedly regulates many transmembrane receptor families, including the VEGF, ErbB (12), TGFβR (20) families, and integrins (21). The activation states of signaling pathways associated with these receptors were therefore evaluated in Gli4NS treated with PST3.1a or transfected with siMGAT5. SiMGAT5 but not PST3.1a treatment inhibits TGFβR signaling (Fig. 2A and B). Considering that in MGAT5^−/−, cells TGFβR signaling is correlated to MGAT5 expression (22), we can hypothesize that the lack of effect of PST3.1a on TGFβR signaling can be explained by the previously observed compensatory overexpression of MGAT5 protein in response to PST3.1a treatment. ERK (data not shown) and focal adhesion kinase (FAK; data not shown) phosphorylations were not affected by PST3.1a or by siMGAT5 treatments.

A defining characteristic of GIC is their ability to grow as neurospheres in serum-free medium, with cell surface characteristics permitting cell–cell interactions. We analyzed the effects of PST3.1a on the ability of GIC to form large NS (Fig. 2C). A 5-day treatment (5 μmol/L) had no effect on the growth of GLI4 and GliT GICs (data not shown). Quantification (Fig. 2D) confirmed that the total numbers of NS are increased in treated GICs, compared with controls. Large NS formation is not simply the result of cell proliferation but also involves cell/cell, NS/NS, and cell/NS interactions. A video microscopy analysis, performed over 24 hours following drug addition, shows that cell/cell, NS/NS, and cell/NS interactions are all affected by PST3.1a treatment (Supplementary Movies S1 and S2). This explains the dramatically reduced number of enlarged NS following a 5-day treatment. Similarly, siMGAT5 transfection confirmed that MGAT5 inhibition increases the number of small neurospheres (Fig. 2C and D). The effect of PST3.1a was significantly different on siCtrl- and siMGAT5-transfected cells (two-way ANOVA, P < 0.0001) concordant with MGAT5 as the target of PST3.1a.

We next tested the ability of GICs to generate clones in a clonogenicity test after PST3.1a treatment. Gli4 (Fig. 2E) and GliT (not shown) were pretreated or not with 2 μmol/L PST3.1a for 48 hours. The cells were then dissociated and plated as single cells with or without PST3.1a (2 μmol/L) in 96-well plates. Three weeks postplating, the number of colonies was reduced in PST3.1a-pretreated cells compared with non-pretreated cells. Interestingly, the number of colonies was not further diminished when cells were seeded as single cells in PST3.1a-containing medium. The expression of the CD15 glycan, a GBM stem cell enrichment marker (23), was reduced by siMGAT5 but not upon treatment with PST3.1a treatment (Fig. 2F). However, OLIG2, another marker associated with GBM stemness, was significantly increased by both PST3.1a treatment and siMGAT5 transfection (Fig. 2A and B).

GICs undergo a shift in their surface glycans when they enter differentiation (24). We therefore evaluated the impact of PST3.1a on GICs cultured as adherent cells on polyD-lysine/laminin (PDL/LN) in a differentiation medium (retrieval of growth factor, addition of 0.5% serum). We first measured the impact of PST3.1a on the formation of multibranched β1,6- linked N-acetylglucosamine on Gli4 and GliT differentiated cells (DC). Figure 3A and B indicates that reduced PHA-L binding is observed at nanomolar concentrations of PST3.1a, on both Gli4DC and GliTDC. We next analyzed the effects of PST3.1a and siMGAT5 on MGAT5 expression level in Gli4DC. Both treatments resulted in the reduction of MGAT5 expression (Fig. 3C). The effects of PST3.1a on signaling pathways were then studied on Gli4DC. Figure 3C indicates that SMAD2 phosphorylation was inhibited by both PST3.1a and MGAT5, whereas FAK phosphorylation was inhibited by PST3.1a and activated by siMGAT5. Erk phosphorylation was not affected by either approach (data not shown).

The consequences of these glycosylation and signaling effects were evaluated on cell migration in Boyden chambers (Table 1). PST3.1a inhibited the migration of Gli4, Gli7, GLIT, and SNB75 cells on matrix supports made from either fibronectin, vitronectin, or laminin (Table 1). IC₅₀ values for PST3.1a to inhibit migration of GICs were in a 1 to 30 nmol/L range.

PST3.1a cytotoxic activity has been reported previously (17). Here, we show that PST3.1a inhibits the proliferation of SNB75 and Gli4DC, Gli7DC, and GliTDC cells with IC₅₀ values between 1.7 and 2.6 μmol/L (Table 1). This antiproliferative effect triggered by the inhibition of MGAT5 enzymatic activity was confirmed by siMGAT5 in Gli4DC (Fig. 3D). The impact on migrating cells was also measured by wound healing (Fig. 4A) and by plating Gli4NS on PDL/LN for 4 days before the addition of 2 μmol/L of PST3.1a for 48 hours. Figure 4B shows that cells that migrate the farthest were the most sensitive to the treatment. Phalloidin staining of the treated migrating cells indicates a profound disorganization of the actin cytoskeleton (Fig. 4C). We next evaluated the inhibitory effect of PST3.1a on SNB75, Gli4-, Gli7-, and GliT invasion (Table 1; Supplementary Fig. S5). IC₅₀ values of PST3.1a (7 and 2.5 nmol/L for Gli4 and GliT, respectively) were similar to those observed for migration.

To evaluate the impact of PST3.1a on differentiation processes in GLI4, NSs were plated on PDL/LN coated slides and grown in differentiation medium in the presence or absence of 2 μmol/L PST3.1a. In the control wells, we observed that cells entering migration out of the sphere express DCX, a determinant of the infiltrative capacity of GBM (25). In contrast, DCX-positive cells in the PST3.1a-treated wells were blocked within the NS (Fig. 4D) and failed to spread and migrate outside the NS. Western blot quantification showed that DCX expression was increased in the presence of PST3.1a (Fig. 4E). As a control, when plated on PDL/LN, but cultivated in proliferation medium, DCX was not expressed and PST3.1a treatment did not induce its expression (Fig. 4D) in Gli4 cells. In addition, Galectin 3 expression was decreased by PST3.1a treatment. These data suggest that PST3.1a inhibits Gli4 migration, leading them to accumulate within the NS. This effect is connected to a strong disruption of the actin cytoskeleton.

PST3.1a at doses up to 50 μmol/L did not exhibit toxicity against nonproliferative astrocytes (17) or activated PBMCs (Supplementary Fig. S6). No mortality was observed in mice treated with oral administration of 50 mg/kg daily during 20 days. In addition, no weight loss, no obvious organ toxicity, and no behavioral alteration was revealed (data not shown). Thus, the pharmacologic activity of PST3.1a was next investigated using Gli4- and GliT-GICs in an orthotopic mouse graft model of the disease. A biodistribution study indicated that 1.5 hours after oral administration of 27 mg/kg of PST3.1a, concentration of the product into the brain reached 4.6 ± 2.2 ng/g. Figure 5A and B show that Gli4 cells are highly invasive as 85% of the ipsilateral and 65% of the contralateral hemispheres were invaded by grafted cells at the end of the experiment. Cell densities were counted across several areas of the cortex, corpus callosum, and striatum regions. Figure 5A–C indicates that in PST3.1a-treated animals, the overall Gli4-invaded surface was significantly reduced, and tumoral cell densities within invaded areas were also substantially lower than in untreated mice. Notably, in the striatum of the ipsilateral and contralateral hemispheres, the number of Gli4 cells in treated mice was reduced by 80% (Fig. 5A and B).

GliT cells are less invasive and form a tumor mass at the surface of the brain (Fig. 5D). The tumor compresses the brain, leading to overt clinical signs in control animals (prostration, weight loss) by the end of the experiment. In contrast, at the same time, no clinical symptoms were visible in PST3.1a-treated mice. We quantified the brain compression due to the tumor mass by measuring the dorso-ventral and left-right (Fig. 5E) axes of the brains. We found that the dorso-ventral axis was 28% shorter in the control group compared with PST3.1a-treated animals, owing to the reduction of the tumor mass in response to PST3.1a treatment (Fig. 5D and E). The left-right axis was similar in control and PST3.1a-treated animals (Fig. 5D and E).

On the basis of these first results demonstrating a strong anticancer potential of PST3.1ain vivo, we sought to determine whether PST3.1a could lead to better survival of these xenotransplanted animals. We thus established a survival curve of control versus treated mice (n = 7/group). Mice treated with PST3.1a demonstrated a significant survival advantage over those treated with vehicle (P = 0.0017), with a median survival of 108 days versus 79 days for the vehicle cohort (Fig. 5F). Mice treated with PST3.1a and temozolomide demonstrated a significant survival advantage over those only treated with temozolomide (P = 0.0287), with a median survival of 135 days versus 102 days for the vehicle cohort (Fig. 5F).

We have previously shown the selective antiproliferative activity of PST3.1a against the central nervous system (CNS) cluster within the NCI-60 collection of cancer cells (17) as published by Ross and colleagues (8) and a correlation between antiproliferative and antimigratory effects of phostines (16, 26). Although all NCI-60 cell lines express MGAT5 at the mRNA level, the antiproliferative effect was seen primarily on the CNS cluster (17). Therefore, we performed an in silico transcriptomic analysis, using CellMiner and DAVID of all 60 cell lines and identified a series of genes that were either positively or negatively correlated to cell growth in the presence or absence of PST3.1a (Fig. 6A). We then clustered the selected genes by pathways and function and found that a large number of genes (>150) associated with mitochondrial function are positively correlated with cell growth in the presence of PST3.1a. Consequently, these genes are negatively correlated to the cell sensitivity to the compound (Fig. 6B). On the contrary, genes associated with cell adhesion and cell interaction with the ECM were negatively correlated with cell growth (Fig. 6C). It is noteworthy that SNAI1 and N-cadherin (CDH2), two genes reflecting EMT progression, were negatively correlated to cell growth in the presence of PST3.1a. This finding suggests that PST3.1a preferentially targets mesenchymal-like cancer cells. This latter information supports our observation that of over 31 cell lines that express E-cadherin (E-CADH), according to the CellMiner transcriptomic analysis, only two are sensitive to PST3.1a cytotoxic effect at 10 μmol/L (17). To determine whether or not the absence of expression of E-CADH is a requirement to PST3.1a cytotoxicity, we overexpressed E- CADH in the sensitive E-CADH⁻ Gli4DC. E- CADH overexpression renders these cells resistant to PST3.1a (Fig. 6D). Conversely, the inhibition of E-CADH with a Crispr-Cas9 construct in the resistant pancreatic ductal cancer cell line PANC-1 renders these cells sensitive to PST3.1a (Fig. 6E).

The inhibition of MGAT5 enzymatic activity (Fig. 1B), coupled to the modular character of the N-glycosylation pathway (27), could account for the lectin-binding profile. Final glycan production is controlled by the regulated biosynthesis of intermediate glycans (27). Thus, lower cell binding to SNA and MAA lectins could be explained by the reduction of the tri- and tetra-antennary glycan motifs (Fig. 1D; Supplementary Fig. S1), which are substrates for α2,6-and α2,3-sialylation (4). These data are in accordance with the observation that MGAT5 enzymatic activity controls the number of α2,3-sialyl residues at the glycan termini of proteins by controlling the expression of the α2,3 sialyl transferase (ST3Gal-IV; ref. 28). The lower binding to PHA-E lectin indicated that bisecting tri-antennary glycans motifs could be reduced. Because MGAT3 was not inhibited by PST3.1a (Fig. 1C), we can conclude that either MGAT3 enzymatic activity was reduced or that the bisected tri-antennates glycans are less accessible.

SiMGAT5 transfection and PST3.1a treatment have different effects on MGAT5 expression in NS. This difference may account for the distinct effects on the TGFβ signaling pathways (Fig. 2A and B) and on CD15 expression (Fig. 2F). Nevertheless, the overexpression of OLIG2 in proliferative culture condition is observed in both treatments (Fig. 2A and B). These apparently contradictory data suggest that PST3.1a-treated cells maintain an apparent GIC expression profile identity, but the disruption of N-glycosylation induced by the treatment leads to phenotypic alterations strongly affecting their function. An overexpression of OLIG2 in proliferative culture conditions was previously reported to suppress the ability of GIC to form colonies in an anchorage-independent manner (29). Taken together, the data support the suggestion of Farahani and colleagues (30) that cell adhesion is crucial for GIC to keep their stemness character.

Signaling effector pathways impacted by PST3.1a also differ depending on cell growth conditions (Fig. 2A and B; Fig. 3C). Such a difference was already reported for SMAD2 activation in glioma cells treated with an integrin signaling inhibitor (31). In this latter study, when the expression of integrins was inhibited in nonsphere glioma cells, the TGFβ pathway was strongly impaired, whereas no such effect was observed in glioma cells cultured under sphere formation conditions. In our conditions, the TGFβ pathway is also only impaired in differentiation culture conditions after PST3.1a treatment (Figs. 2A and 3C), which could underlie the resulting effects on cell migration and proliferation owing to the key role of the TGFβ receptor in these two processes in glioma cells (32). In GBM, TGFβ signaling is associated with EMT, loss of E-cadherin expression (33) and poor prognosis (7). The impact of MGAT5 on TGFβ was already extensively reported (20, 22, 34). However, in the PyMT MGAT5^−/− cell line used in these studies, TGFβ signaling was associated with cell growth arrest. Inhibition of the activating phosphorylation of the FAK by PST3.1a only occurs in nonsphere glioma cells, suggesting that signaling centers involved in interactions between the ECM and integrins are altered by PST3.1a. Finally, PST3.1a treatment reduces galectin-3 expression in Gli4 cultivated in adherent conditions. This observation is coherent with the stimulatory effect of galectin3 on integrin-mediated activation of FAK, increasing cell motility (35). Interestingly, integrins, growth factors, and cytokine receptors are all glycoproteins whose activities are controlled by the high-avidity interactions of their N-glycans with a galectin lattice (36). Tight regulation of integrin turnover has been reported to be a key factor regulating cell migration and cytokinesis (37), and FAK inhibition was shown to inhibit actin assembly (38). We show that integrin signaling is sensitive to glycosylation dysregulation, as illustrated by the alteration of FAK phosphorylation (Fig. 3C) associated with the actin cytoskeleton disruption (Fig. 4C).

Transcriptomics analyses on CellMiner revealed that MGAT5 expression was necessary but not sufficient for a cytotoxic effect of PST3.1a on cell lines within the NCI-60 panel. Therefore, dependence on MGAT5 expression is not equivalent in all cell types. DAVID analysis revealed that physiologic states of cells are linked to PST3.1a responsiveness. First, it appeared that the EMT engagement of cancer cells is positively associated with responsiveness to PST3.1a. This analysis was confirmed by the loss of cytotoxic response to PST3.1a after reexpression of E-CADH in Gli4 (Fig. 6D). Second, the absence of response to PST3.1a can be linked to the energy metabolism status of the cell. Nonresponding cells express a set of genes, such as NDUFB3, UQCRC2 involved in oxidative phosphorylation. It was previously reported that triantennary N-glycan levels double in an MGAT5^−/− mammary tumor cell line, when supplemented with GlcNac (20). This feedback regulation system was shown to be associated with mitochondrial respiration (20, 39, 40).

Nanomolar range of PST3.1a concentrations was sufficient to trigger antimigration effects on fibronectin, whereas the cytotoxic effects of the molecule in proliferative conditions were in the millimolar range (Table 1). This implies that GIC migration is more dependent on β1,6-GlcNAc branching than is cell proliferation. An estimation of the Ki considering a strict competitive inhibition against the acceptor glycan predicts a value of 300 nmol/L, confirming the nanomolar range of activity of PST3.1a. In addition, the cellular and pharmacologic data are to be put in light to the biochemical characteristic of the linear pathway that initiates the GlcNac branching (4, 41). This branching is dependent on glycoprotein acceptor concentrations for the initial MGAT1 and MGAT2 steps, while MGAT5 activity is limited by UDP-GlcNac concentrations (4). We observed a diminution of the abundance of ions at m/z 1,662 and 1,836 (Supplementary Fig. S1). This observation suggests that the access of MGAT2 to its substrates could have been reduced either by an increase of MGAT3 activity, which blocks further branching (41), or by a decrease of α-mannosidase I and II activities. Moreover, we observed that MGAT5 expression was reduced under PST3.1a treatment in Gli4DC (Fig. 3C). In other words, the cell answer to MGAT5 inhibition could enhance the cellular and physiologic effects of PST3.1a and account for its nanomolar range of activity.

In summary, PST3.1a targets MGAT5. The antitumor effect relies on the MGAT5 dependency of a given cell type. This dependency is linked to cellular engagement in the EMT and its energetic metabolic status. The biological importance of these combined physiologic states, associated with glycosylation, opens up a new area of investigation for cancer therapy.

L. Clarion has ownership interest (including patents) in Phost'In. N. Bakalara is a consultant/advisory board member for Phost'In. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Z. Hassani, S. Turpault, S. Loiseau, L. Clarion, J.-N. Volle, D. Virieux, J.-L. Pirat, N. Bakalara

Development of methodology: Z. Hassani, A. Saleh, S. Turpault, J.-P. Hugnot, E. Uro-Coste, P. Legrand, M. Delaforge, S. Loiseau, J.-L. Pirat, N. Bakalara

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z. Hassani, W. Morelle, J. Vignon, J.-P. Hugnot, E. Uro-Coste, M. Delaforge, S. Loiseau, M. Lecouvey, J.-L. Pirat

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Z. Hassani, S. Khiati, W. Morelle, J. Vignon, M. Delaforge, J.-L. Pirat, H. Duffau, N. Bakalara

Writing, review, and/or revision of the manuscript: Z. Hassani, S. Khiati, W. Morelle, J. Vignon, E. Uro-Coste, J.-N. Volle, D. Virieux, J.-L. Pirat, H. Duffau, N. Bakalara

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Delaforge, N. Bakalara

Study supervision: M. Delaforge, J.-N. Volle, D. Virieux, N. Bakalara

Other (supervision of student upon analytic aspects of qualitative and quantitative measurements of "Phostin" and their metabolites in vivo and in vitro in body fluids and the development of analytic tools to improve their recovery): M. Delaforge

The authors thank Odile Sainte-Catherine for her technical support. We thank the Montpellier RIO Imaging (MRI) facility, the flow cytometry platform at the Institute for Regenerative Medicine and Biotherapy (IRMB) in Montpellier, and in particular Dr. Christophe Duperray for his expertise and the Réseau d'Histologie Expérimentale de Montpellier (RHEM). We also thank Prof. James W. Dennis (University of Toronto), Prof. Michael Peter Barrett (University of Glascow), and Dr. Guy Lenaers (University of Angers) for critical reading and language correction.

N. Bakalara, P. Legrand, M. Lecouvey, D. Virieux, J.N. Volle, J.L. Pirat, E. Uro-Coste, and A. Saleh were supported by the ANR-PCV (G2O), ANR Emergence-Bio (IDEPHOST), SATT (AxLR), INCA (formation des medecins), the Région Languedoc-Roussillon-Midi Pyrénées, and the ENSCM. L. Clarion, Z. Hassani, S. Loiseau, and S. Turpault were supported by Phost'In SAS. The Mass Spectrometry facility (MALDI-TOF and MALDI-TOF/TOF) is funded by the European Community (FEDER). The Région Nord-Pas de Calais (France) and the Université des Sciences et Technologies de Lille I support W. Morelle.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.