A novel 3D nanofibre scaffold conserves the plasticity of glioblastoma stem cell invasion by regulating galectin-3 and integrin-β1 expression

Glioblastoma Multiforme (GBM) invasiveness renders complete surgical resection impossible and highly invasive Glioblastoma Initiating Cells (GICs) are responsible for tumour recurrence. Their dissemination occurs along pre-existing fibrillary brain structures comprising the aligned myelinated fibres of the corpus callosum (CC) and the laminin (LN)-rich basal lamina of blood vessels. The extracellular matrix (ECM) of these environments regulates GIC migration, but the underlying mechanisms remain largely unknown. In order to recapitulate the composition and the topographic properties of the cerebral ECM in the migration of GICs, we have set up a new aligned polyacrylonitrile (PAN)-derived nanofiber (NF) scaffold. This system is suitable for drug screening as well as discrimination of the migration potential of different glioblastoma stem cells. Functionalisation with LN increases the spatial anisotropy of migration and modulates its mode from collective to single cell migration. Mechanistically, equally similar to what has been observed for mesenchymal migration of GBM in vivo, is the upregulation of galectin-3 and integrin-β1 in Gli4 cells migrating on our NF scaffold. Downregulation of Calpain-2 in GICs migrating in vivo along the CC and in vitro on LN-coated NF underlines a difference in the turnover of focal adhesion (FA) molecules between single-cell and collective types of migration.


Results
NF network production and physical characterization. The CC is the favourite route to the contralateral hemisphere of glioblastoma cells 17 . Figure 1a,b highlight the three-dimensional anatomic organization of the heterotypic fibres in the trunk of the CC. To better understand, characterize and target migrating glioblastoma cells on the CC, we designed a NF network which could be made of aligned or non-aligned fibres (Fig. 1c,d).
The purpose of this model is to be able to study the impact of the spatial and mechanical properties of a fibrous microenvironment. PAN NF have been selected for their biocompatibility and resistance to biodegradation that would interfere with a mechanistic study. Moreover, the spatial design and mechanical properties of PAN NF are easily tuneable. Fourier transformed infrared (FTIR) spectroscopy ( Fig. 1e) was used to discriminate the functional groups of the stabilized PAN 18 . Commercial PAN contains traces of (free) water (3622 and 1626 cm −1 ) and bands at 2940 cm −1 (CH 2 , CH stretching), 2242 cm −1 (nitrile groups), 1453 cm −1 (δCH 2 ), 1356 cm −1 (CH bending), 1249 cm −1 (χ CH 2 ) and 1072 cm −1 (C-C stretching). After stabilization and oxidation, the spectrum shows a strong reduction of the nitrile band at 2242 cm −1 , broad weak bands from 3100 to 3600 cm −1 attributed to OH and NH stretching, a very strong band at 1589 cm −1 (C = C stretching and NH bending), a strong band at 1372 cm −1 (NH and CH bending) and a weak band at 807 cm −1 (C = C-H rocking in the aromatic plane). When subjected to UV light, NFs appear fluorescent. They emit in red, green and at a much lower extent in blue and infrared (data not shown). The NFs were decorated with LN. Spots of LN deposit can be seen discontinuously distributed along the NF + LN (Fig. 1f, white arrows). The diameter of the NF of 0.65 ± 0.082 µM (Fig. 1g) is similar to the diameter of the axons in the CC (0.64 ± 0.42 µm) 19 . The porosity of the aNF scaffold predominantly ranges between 0.5 µm² and 7 µm² (Fig. 1h). This creates a confinement situation which is required to trigger linear glioma migration.
The NFs constitute a suitable tridimensional environment for GIC adhesion and migration. Gli4 cells express several typical markers of neural precursors, are multipotent and generate tumours in vivo 20 . When seeded on 2D in differentiation medium, neurospheres (NS) adhere to the support and cells migrate away from the NS (Fig. 2a). Analysis of the repartition of the Gli4 cells, within the fibre scaffold shows that cells migrate deeply into it (Fig. 2b). Gli4 cells feature an extensive and flat network of stress fibres when plated in 2D conditions (Fig. 2c), whereas Gli4 seeded on NF appear as elongated cells presenting filopodia while no stress fibres are visible (Fig. 2d,e). In 2D, vinculin spots localize at the leading edges of the lamellipodia (Fig. 2f). On NF, Gli4 cells send cellular extensions in various directions to attach to several fibres in a tridimensional manner (Fig. 2g,h, yellow arrows). Rather than focal, adhesion sites are spread along filopodia, which also encircle the fibres on NF (Fig. 2g,h). It has previously been reported that astrocytic brain tumour cells extend ultra-long membrane protrusions that they use as routes for brain invasion 16 . These data show that NFs constitute a pertinent permissive tridimensional microenvironment for glioblastoma cell migration and adhesion.
Gli4 migrate individually or collectively in the presence or absence of LN on PAN NF. During glioblastoma progression, the secretion of ECM molecules such as LN, fibronectin or hyaluronic acid increase significantly 13 and the microenvironment is modified by deposing these molecules. For this reason, we address the question whether or not LN-coated fibres modify GIC migration.
In the absence of LN, Gli4 cells migrate collectively by forming aggregates composed of tens of tightly associated cells (Fig. 3a,c,e). On the contrary, Gli4 do not aggregate on NF + LN and migrate as single cells (Fig. 3b,d,f). Cells are rounded at the centre of the cell mass (Fig. 3a, white star), while at the border they are bipolar and have a filopodial protrusion (Fig. 3a, arrowheads). Because cadherin expression and actin cytoskeleton remodelling were reported to control collective migration 21 , we analysed the organization of the actin cytoskeleton and the N-cadherin-mediated adherent junctions. On NFs, actin cytoskeleton-specific shows that most of the cells are rounded and appear to form a continuum within aggregates (Fig. 3c). Collectively migrating Gli4 express N-Cadherin at the plasma membrane (Fig. 3d). Thus N-Cadherin-mediated adherent junctions are maintained during collective Gli4 migration. On the contrary, Gli4 behave as single cells on NF + LN, they are bipolar (Fig. 3d,f, arrowheads).
Using human ex-vivo GBM slice cultures, it has been found that GBM tumours that have their EGFR gene amplified are more invasive than those that have not, suggesting that EGF signalling stimulates GBM cell migration. In line with this notion is that migration of this type of tumour was slowed down by the tyrosine kinase inhibitor, Gefitinib 22 . Subsequently, clinical expectations raised by this report were unfortunately not met by     Parallel Perpendicular  www.nature.com/scientificreports www.nature.com/scientificreports/ targeting the EGFR signalling cascade with Gefitinib 23 . Several hypotheses can be forwarded to explain the discrepancy between the in vitro and clinical results, one of which pertains to the experimental context. To verify the clinical relevance of our NF culture design, we tested the effect of gefitinib on migrating GICs in the presence of an EGFR stimulation background. It was found that Gefitinib is indeed without effect if GICs migrate individually on LN-coated fibres, but that is very efficient in reducing the speed of collectively migrating cells on uncoated fibres (Fig. 3g,h).

Gli4 haptokinesis and haptotaxis result from the orientation and coating of the NF scaffold.
To investigate the impact of the ECM topography on GIC migration, we compared the direction of migration of Gli4 cells on aNF and naNF functionalized or not with LN. First of all, our results show that on aNF, Gli4 cells migrate predominantly according to the orientation of the NF (Fig. 4a). We observed a significant higher number of migrating cells in the direction parallel to the NF than in the perpendicular direction, particularly when LN is present (Fig. 4b,c).
A proliferation test indicated the lower proliferation rate of Gli4 cultured on aNF with respect to 2D (Fig. 4d). Transcriptome analysis revealed that 98 genes involved in cell cycle progression as well as members of the kinesin family were expressed by at least a factor of 1.5X less in cells grown on NF than on 2D, suggesting a shift to the invading, less proliferative mesenchymal phenotype of GIC cells in a NF context (Fig. 4e). Accordingly, the expression of mesenchymal and pre-metastatic stem cell markers such as ATXN1, ALCAM, CD9, ITGA7 and CD44 are maintained in NF vs 2D culture conditions (Supplementary Figure 1). We may conclude from this analysis that Gli4 cultured in the NF microenvironment retain their mesenchymal phenotype. Moreover, the 3.3X overexpression of CHI3L1 (Supplementary Figure 1) in NF could promote as reported for liver cancer cell migration and invasion 24 .

Cellular adhesion to the ECM and FA dynamics of Gli4 cells differ between conventional 2D
planar surfaces and aNF. Important proteins implicated in migration are the multifunctional modulators of cell FA such as galectins and integrins. FAs play a central role in migration by controlling the dynamic cycles of attachment and detachment to the ECM 25 . FAs are considered as prototypical integrin-mediated cell-ECM contact sites that link cellular actin cytoskeleton to the ECM scaffold 26 . Therefore, our next aim was to compare the expression of integrin β1, α6 and galectin-3 in the different culture conditions. On aNF, the protein levels of galectin-3 and integrin β1 were respectively 6 and 2.6 fold higher in cells grown on LN-coated fibres than on non-coated fibres ( Fig. 5a and Supplementary Figure 2). On 2D, only galactin-3 expression is somewhat reduced by LN coating (Fig. 5a and Supplementary Figure 2). In contrast, the protein level of integrin α6 was 2.5 times lower on LN-coated fibres whereas its level in 2D increases with LN ( Fig. 5a and Supplementary Figure 2). Figure 5b shows that integrin β1 is localized at the plasma membrane of Gli4 migrating on aNF + LN (white arrowheads), while it is distributed in the cytoplasm and around the nucleus in Gli4 growing on 2D + LN. Note that on NF + LN, integrin β1 membrane staining is localized in the attachment points with the NF (Fig. 5b, white arrowheads).
According to these previous data we decided to study the expression of FA components such as calpain-2, vinculin, FAK and phosphorylation of FAK and Talin and cleaved Talin expressions in Gli4 cells on the aNF +/− LN and 2D +/− LN supports ( Fig. 5a and Supplementary Figure 2). Calpain-2 expression is 2 fold higher in Gli4 cells plated on 2D + LN than in the other conditions. Immunofluorescence confirmed that calpain-2 is upregulated in 2D + LN compared to aNF + LN (Fig. 5b). The expression levels of vinculin are equal in aNF +/− LN and 2D +/− LN (Fig. 5a). Compared to 2D + LN, FAK expression increases on aNF + LN, whereas its phosphorylation ratio decreases. Talin expression and cleavage are increased in 2D + LN compared to all other conditions. We therefore conclude that FA dynamics are affected by the dimensionality and functionalization of the microenvironment.
GliT migrate collectively irrespective of LN coating and exhibit a different pattern of expression of cell adhesion and signalling proteins compared to Gli4 cells. We examined a second primary glioblastoma cell line with a different signature, GliT. Transcriptome analysis comparing Gli4 and GliT in NS culture yielded 784 genes differentially expressed by at least a factor 2. P53, EGF, PDGF, and NGF signalling pathways were upregulated in Gli4 with respect to GliT, while the axon guidance and TGF-β pathways as well as FA and cell adhesion genes were downregulated (Supplementary Table 1). In addition, quantification of the number of collective migration events per NS on NF +/− LN for Gli4 and GliT indicates that the shift from collective to single cell migration by the addition of LN is not as strong in GliT as in Gli4 (Fig. 5c,d). Apparently this collective behaviour is important for GliT cells as we found that their survival was compromised if they were seeded on NF as dissociated cells at low density (data not shown).
GliT also differs from Gli4 in the expression of proteins involved in the interaction with the ECM (Fig. 5d and Supplementary Tables 1, 2). For instance, galectin-3 protein levels decreased by 1.7 fold in GliT when plated on aNF + LN in comparison to 2D + LN ( Fig. 5f and Supplementary Figure 2), whereas we found a 6-fold increase in the same conditions with Gli4. Integrin α6 expression in GliT increases on aNF +/− LN relative to 2D +/− LN (Fig. 5e and Supplementary Figure 2), whereas we observed a downregulation for Gli4 on NF + LN. Integrin β1 NF. Unpaired Student's t-test. (e) Heat map of the 98 genes associated with cell cycle and differentially expressed between Gli4 cells cultivated in either 2D or NF +/− LN. The list alongside the heatmap comprises 'cell cycle' genes that are expressed differently between the 2D and NF samples. Significance was estimated with Student's t-test (n ≥ 4).  www.nature.com/scientificreports www.nature.com/scientificreports/ protein levels were not different between aNF +/− LN and 2D +/− LN for GliT cells (Fig. 5e and Supplementary  Figure 2), whereas for Gli4 a higher integrin β1 protein expression was found on LN-coated NF than on uncoated NF. Immunofluorescence analysis of calpain-2 expression (Fig. 5f) shows equal expression of this enzyme in both 2D and on NF in GliT, contrary to Gli4. Also, GliT underexpress CAMK2D and overexpress CAMK2N1 with respect to Gli4 (Supplementary Table 2). CAMK2N1 is an inhibitor of the Ca 2+ /calmoduline-dependent protein kinase (CaMKII), which controls a range of cellular processes including cell adhesion and migration 27 . Altogether, our results show that Gli4 and GliT cells exhibit a different response of integrins -β1 -α6 and galectin-3 to ECM modifications.

Integrin β1 and galectin-3 are upregulated in Gli4 cells in the CC whereas calpain-2 is downregulated in invasive Gli4 cells in vivo.
When transplanted in the brains of nude mice, Gli4 are highly invasive 20,28 (Fig. 6a). They invade the cortex, proceed through the fibrous white matter tracts of the CC and reach the contra-lateral hemisphere (Fig. 6a). GliT cells in contrast, are almost not invasive and form a bulky tumour 28 (Fig. 6a, area delineated by the dashed line). The expression of integrin β1 (Fig. 6b) and galectin-3 (Fig. 6c) in invasive Gli4 cells is higher when they migrate along the aligned myelinated fibres of the CC (Fig. 6c, yellow arrow) than in the cerebral cortex (white arrow) or the striatum. Calpain-2 is not expressed by invasive Gli4 cells in the CC nor in the cortex (Fig. 6d). Interestingly, GliT inside the tumour bulk express calpain-2 (Fig. 6d), while GliT migrating away from the tumour mass do not express calpain-2 (Fig. 6d). These data indicate that integrin β1 and galectin-3 expression depends on the cerebral microenvironment and determines invasiveness. Calpain-2 expression seems to be inversely correlated to the invasive potential of GIC in vivo.

Discussion
After the stabilization step, FTIR analysis allowed us to determine that the proportion of pyridone (pyridine ring C 5 N bearing a carbonyl function) is very low in our compound given the absence of the carbonyl absorption band. The predominant functional groups (Fig. 1e) show that dehydrogenation and aromatization reactions occurred, in agreement with the colour and the fluorescence of the NF. The functional groups of the stabilized PAN NF promote hydrogen bond formation onto the cell surface, thus creating a biocompatible environment favouring cell attachment.
We have overcome a limiting problem encountered with electrospun NF consisting of the poor cellular infiltration or in-growth 3 . This is due to the small dimensions of the microenvironment which hamper the cells to insinuate their nucleus and cell body in the spaces separating adjacent NF. In the case of our PAN NF, alignment is not a requirement to create interstitial spaces permissive to cellular infiltration. This is likely due to the chemical nature, filament spacing and diameter of the material. These features allow the GICs to fully embed in a tridimensional microenvironment thereby abolishing the apico-basal polarity imposed by 2D. Also, our results show that the orientation of the NF governs the direction of migration of Gli4 cells much in the way that glioblastoma cells in vivo are guided by axon alignment in the CC. The addition of LN on aNF improves migration of Gli4 cells (Fig. 4b) as was reported previously for migration in the perivascular space in vivo 29 . The continuous actin cytoskeleton and N-Cadherin-mediated formation of cell-cell adherent junctions are responsible for the coordination between cells during collective Gli4 migration on uncoated NF (Fig. 3d). These cell-cell junctions at the leading edge as well as in lateral regions and inside the moving cell group are characteristic of collective migration 21 . Collective migration is a hallmark of epithelial cancer 30 , in contrast to the individual, mesenchymal, type of migration 31,32 . Similar to what has been described for ErbB2-positive breast cancer and lung tumour cells 33,34 , we propose that invading glioblastoma stem cells use collective migration to overcome anoikis.
Glioblastoma cells adapt outstandingly to various environments during their migration inside the brain. In our NF system, integrin β1 and galectin-3 overexpression coincides with a shift from collective to single cell migration (Figs 3 and 5). Unexpectedly, no difference in galectin 3 and integrin β1 mRNA content was discerned between 2D and NF conditions, even though a clear difference in protein levels was observed. This suggests that the regulation of expression of these proteins occurs after transcription 35,36 . Galectin 3 and integrin β1 overexpression in glioblastoma cells migrating along the white matter is typical of mesenchymal single cell migration 8,32 . Their overexpression by cells cultured on NF coated with LN and by cells migrating in vivo along the CC (Figs. 3 and 5) might indicate resistance to anoikis and a less proliferative (Fig. 4d), more metastatic phenotype 37 . Galectin-3 also acts as a multifunctional modulator of cell-cell and cell-ECM adhesion 38 , in particular promoting adhesion to LN 39 . The interaction of LN with Mgat5-modified N-glycans at the cell surface of mammary carcinoma cells stimulates α5β1-integrin activation and translocation to fibrillar adhesions 40 . Upregulation of the integrin β1 adhesion network at the plasma membrane as we observed with Gli4 cells grown on aNF + LN, not rarely goes hand in hand with a similar concerted raise in N-cadherin adherence (Fig. 3f) 41 . Both networks are mayor actors in to cell migration 42 , radio-resistance 43 and immune-resistance 44 . The decrease of integrin-α6 in Gli4 cultured on NF + LN (Fig. 5a), is probably due to a loss of stemness of Gli4 cells in favour of a more invasive phenotype 45 . Our transcriptome analysis (Supplementary Figure 1) suggests that its role in heterodimer formation is taken by integrin α7.
While integrin β1, FAK and galectin 3 are overexpressed in cells grown on NF + LN in comparison to 2D + LN, neither an increase in calpain2 expression nor an increase in pFAK/FAK and talin/total talin ratios are observed when changing from NF -LN to + LN. These data indicate a stronger adhesion of Gli4 cells to the NF + LN independent of an increase of FA turnover. In 2D, LN promotes FA turnover without an increase of integrin β1, FAK and galectin 3 expression. In conclusion, the different responses (adhesion, turn-over and signalling) of Gli4 FA to LN are dependent on the physical properties of the support.
GliT maintains collective migration even on LN-coated NF (Fig. 5). This is in line with the observation that GliT migrate collectively in the CC, while Gli4 adopt a single cell mode (Fig. 6). We also found that calpain-2 is expressed by GliT cells staying inside the bulky tumour, while Gli4 cells migrating in the CC do not express www.nature.com/scientificreports www.nature.com/scientificreports/ calpain-2 (Fig. 6). With our NF model the expression of calpain-2 decreased for Gli4 grown on NF + LN in comparison to 2D + LN but not for GliT grown in the same conditions. Consistent with their maintenance of collective migration, GliT do not vary galectin 3 and integrin β1 expression in 2D and NF conditions. Moreover, Talin cleavage is not different in GliT grown in 2D NF +/− LN (Supplementary Figure 2). Calcium influxes are a unique characteristic increasing the invasiveness of glioblastoma cells. Calpain 2 is a key downstream effector of calcium ions, facilitating glioblastoma cell invasion 46 . As preliminary data, we recorded the occurrence of local calcium transients in filopodia of GliT cells collectively migrating on aNF + LN (Video 1). These transients are known to stabilise filopodia, but do not lead to a highly invasive phenotype. Also, according to our transcriptomic data, Gli4 and GliT in NS culture express various amounts of the ECM-related genes coding for LN, ECM receptor integrins and cell adhesion molecules (Supplementary Figure 2 and supplementary Table 2).
We propose that ECM attachment and sensing as well as the signal transduction pathways on which these depend are integrated in a different fashion in Gli4 than in GliT. It results in different migration modes, possibly due to different affinities to ECM proteins. Remarkably, all these different signal interactions sum up to complete different migrating behaviours of glioma cells in vivo i.e. (Fig. 6).
In conclusion, we have developed a new electrospun NF scaffold suitable for glioblastoma adhesion and migration in a 3D microenvironment akin mesenchymal migration. It allows for discrimination between the migration potential of different glioblastoma stem cells. The PAN NF matrix represents a valuable tool to study the role of the ECM in GIC migration. It can also serve as a relevant in vitro platform for compounds targeting PDL-1 whose expression is dependent on the interaction of FA with the ECM.

Methods
Cell culture. Gli4 and GliT were obtained as described by Guichet et al. 20 . Gli4 and GliT were cultured according to the NS protocol elaborated by Guichet et al. 20 . In DMEM/F12 medium supplemented with glucose, glutamine, insulin, N2, Epidermal Growth Factor and Fibroblast Growth Factor, (proliferation medium) GBM cells form NS, in DMEM/F12 medium supplemented with the same components as for the proliferation medium except from growth factors and heparin replaced by foetal bovine serum (0.5%), (differentiation medium) GBM cells migrate 20 . Prior to cell seeding, 2D or NF were either functionalized or not with poly-D-lysine added overnight 20 and then the addition of LN (sigma L2020) (0.05 mg/mL) one hour at 37 °C. MTT test was used to evaluate cell viability as described 47 . To obtain NS of the same size, we used Corning ® 96 well round bottom ultra-low attachment microplates coated with a covalently bond hydrogel (Corning 7007) (Supplementary Material 1). Dissociated GBM NS cultured in proliferation condition, were seeded in each well and remained in culture during 2 days until formation of single neurosphere due to sedimentation. Then GICs NS were deposited on the top of NF and left it to migrate during 5 days (Supplementary Material 1).
NF production and characterization. aNF and naNF were produced by electrospinning using a solution of 10 wt% PAN (Aldrich, average Mw 150,000) dissolved in N,N-Dimethylformamide (DMF, Aldrich >99%, molecular biology grade). For the fabrication of the NF scaffold, a needle was used to project the polymer, which was collected on an electrode located 15 cm from the needle. A voltage of 20 kV was applied. To produce aNF, the collecting electrode was a rotating drum (2000 rpm), whereas a non-mobile metal disk was used to produce naNF. After electrospinning, the NF scaffold was thermally stabilized in a chamber furnace (250 °C, 2 h dwell, 2 °C. min −1 heating rate). ATR-FTIR spectra were recorded on a Perkin-Elmer Spectrum 100.
NF cryosectioning: For cryosectioning, the NF were included in OCT before freezing. Section thickness was 14 µm (Supplementary Material 2). GBM orthotopical xenotransplantation. Gli4 and GliT othotopic xenotransplantation were done as previously described 28 . After 3 months, the animals were sacrificed and brains were removed and post-fixed in 4% PFA and then immersed successively in 7%, 15% and 30% sucrose. Afterwards, the brains were included in an optimal cutting temperature (OCT) compound, snap-frozen and cryosectioned in 14 µm thick slices. 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.
Migration quantifications were done using ZEN 2012 software in counting number of cells, using dapi staining, from 200 µm away from the border of the NS to the last observed migrating cell. Alternatively, migration capacity was quantified by measuring an area of migration. To measure migration areas, we subtracted the area of the NS containing non-migrating cells from the total area where cells were detected. Calcium imaging. After 6 days of migration, GliT were incubated before acquisition with 2 mM Fluo-4 AM (Invitrogen) and Pluronic (F-127) (Life technologies) in differentiation conditions. Time lapse calcium imaging