Altered cytoskeletal status in the transition from proneural to mesenchymal glioblastoma subtypes

Glioblastoma is a highly aggressive brain tumor with poor patient prognosis. Treatment outcomes remain limited, partly due to intratumoral heterogeneity and the invasive nature of the tumors. Glioblastoma cells invade and spread into the surrounding brain tissue, and even between hemispheres, thus hampering complete surgical resection. This invasive motility can arise through altered properties of the cytoskeleton. We hypothesize that cytoskeletal organization and dynamics can provide important clues to the different malignant states of glioblastoma. In this study, we investigated cytoskeletal organization in glioblastoma cells with different subtype expression profiles, and cytoskeletal dynamics upon subtype transitions. Analysis of the morphological, migratory, and invasive properties of glioblastoma cells identified cytoskeletal components as phenotypic markers that can serve as diagnostic or prognostic tools. We also show that the cytoskeletal function and malignant properties of glioblastoma cells shift during subtype transitions induced by altered expression of the neurodevelopmental transcription factor SOX2. The potential of SOX2 re-expression to reverse the mesenchymal subtype into a more proneural subtype might open up strategies for novel glioblastoma treatments.


Results
Cytoskeletal organization of the glioblastoma cell lines. To study the organization of filamentous actin (F-actin), the cells were stained with fluorescently labeled phalloidin and analyzed by fluorescence microscopy. Stress fibers were visible in all of the cell lines, but they differed with regard to F-actin organization patterns and cell morphology (Fig. 1a,b). On this basis, the cells could morphologically be grouped into different categories: fibroblastic, intermediate, epithelioid. In addition, two cell lines (U-343MGa and U-343MGa CL2:6 grew in clusters, which made characterizations difficult because it was not always possible to identify individual cells.
Cells of cell lines U-2975MG and U-2987MG had a few lamellipodia, as 41% and 36% of the cells, respectively, and some had a few actin arcs, as 29% and 14% of the cells, respectively. In addition, in these cells the stress fiber bundles appeared shorter and thicker than for the other cell lines (Fig. 1a,b, Supplementary Tab. S1). These characteristics are indicative of an epithelioid phenotype. In contrast, cell lines U-2982MG and U-2990MG showed fibroblastic phenotypes according to their high aspect ratios (2.86, 4.65, respectively) and low circularity (0.46, 0.40, respectively). In addition, these cells had an abundance of stress fibers, which spanned the entire cell bodies (Fig. 1a,c,e, Supplementary Tab. S2). Cell lines U-343MGa 31L, U-2995MG, and U-2997MG showed phenotypes of an intermediate type, with relatively rounded cell shapes (circularity, 0.65, 0.69, 0.59, respectively) and intermediate areas (1.145, 1.157, 1.582 µm 2 , respectively), and with more actin arcs and lamellipodia (Fig. 1a,c,d, Supplementary Tab. S1). Cell line U-343MG had the largest cells (1.744 µm 2 ), which were cuboid in shape (circularity, 0.46) and had fewer actin arcs, which were only seen for 12% of these cells (Fig. 1a-
Vimentin is known to be expressed in mesenchymal cells and has been used as a diagnostic and prognostic marker in different cancers 27 . Vimentin was expressed in all glioblastoma cell lines analyzed in this study. However, the differences in the vimentin subcellular distributions classified the cells into two main groups. The first group comprised cell lines U-2987MG, U-2995MG, and U-343MGa 31L, where vimentin IFs occupied < 45% of the total cell area (Fig. 2b, Supplementary Tab. S3), and was localized mainly to the perinuclear area and on top of the nucleus (Fig. 2a). The second group comprised cell lines U-2982MG, U-2990MG, U-2997MG, and U-343MGa Cl2:6, where vimentin IFs spread over > 45% of the cell body (Fig. 2b, Supplementary Tab. S3), and the filaments were arranged in denser networks (Fig. 2a).
Two other IF proteins were also analyzed in these glioblastoma cell lines: glial fibrillary acidic protein (GFAP) and nestin. Cell lines U-2987MG and U-343MGa fully expressed both GFAP and nestin, whereas cell lines      www.nature.com/scientificreports/ U-2995MG, U-2997MG, and U-343MG expressed neither GFAP nor nestin (Fig. 2c, Supplementary Fig. S1a, b). Few cells in cell lines U-2975MG, U-2982MG, and U-2990MG (14%, 7% and 32% of the cells respectively) expressed nestin but not GFAP, whereas for cell lines U-343MGa Cl2:6, 21% of the cells expressed GFAP, but did not express nestin (Fig. 2c, Supplementary Fig. S1a, b). Cell line U-343MGa 31L did not express nestin, however GFAP was express to some extent but not in a fibrillary form.
To visualize MT organization, the cells were stained with an antibody against α-tubulin. For cell lines U-2975MG, U-2990MG, U-2982MG, U-2995MG, U-343MG, and U-343MGa, the MTs spanned the entire body of the cells ( Supplementary Fig. S2a), as a pattern of distribution that is often seen in cells with mesenchymal morphology. In contrast, for cell lines U-2987MG, U-2997MG, U-343MGa 31L, and U-343MGa Cl2:6, MTs appeared to be wrapped around the nucleus, and to have a more tangled organization in the rest of the cytoplasm; this is a MT organization that is often seen in cells with epithelial morphology (Supplementary Fig. S2a). Tubulin acetylation is indicative of MT stability, and acetylation of α-tubulin at lysine 40 is under the control of α-tubulin acetyltransferase 1 and histone deacetylase 6 28 . Cell lines U-343MG, U-2987MG, and U-2995MG showed the lowest ratios of acetylated MTs (0.20, 0.21, 0.32, respectively) and acetylated MTs were concentrated to the perinuclear area and absent from the cell edges ( Supplementary Fig. S2a, b). Conversely, the ratios of acetylated MTs were the highest for cell lines U-2975MG, U-2990MG, U-2997MG, U-343MGa, and U-343MGa Cl2:6 (0. Organization of cell-cell junctions in the glioblastoma cell lines. We next studied the organization of cell-cell junctions in the glioblastoma cell lines, using an antibody against β-catenin. The localization of β-catenin differed between the cell lines. For cell lines U-343MGa 31L and U-2987MG, β-catenin was located at the cell borders and created a well-defined framework at the cell-cell junctions ( Supplementary Fig. S3a) and thin junction width (0.5 µm and 0.38 µm respectively) ( Supplementary Fig. S3b), which is a phenotype frequently observed in cells with epithelial morphology. A similar organization was seen in cell lines U-343MGa and U-343MGa Cl2:6, with thin junction width (0.31 µm in each) but this was less distinguishable due to their clustered growth pattern (Supplementary Fig. S3a and b). In contrast, in cell lines U-343MG, U-2975MG, U-2982MG, U-2997MG, and U-2995MG, β-catenin was also localized to the cell borders, but the β-catenin organization appeared less defined and displayed broad junctions and a more serrated pattern (0.64 µm, 0. Migratory and invasive capacities of the glioblastoma cell lines. We next studied the migration and Matrigel invasion of the cell lines. To investigate the collective migration in two dimensions, wound closure was monitored for wounds inflicted in monolayers of cells, using the Incucyte imaging technique. Cell line U-343MG had the highest migratory capacity, with wound closure in < 24 h; for cell lines U-2975MG and U-2982MG, over the same period, these showed relative wound closures of 89% and 77%, respectively (Fig. 3a). The properties of this collective migration were reflected in the differences for single cell migration. Cell lines U-343MG, U-2975MG, and U-2982MG migrated with speeds of 0.6 μm/min, 0.6 μm/min and 0.9 μm/min, respectively ( Fig. 3c), which agrees with the fibroblastic morphology. Cell line U-343MGa 31L was also relatively efficient for the collective cell migration, with relative wound closure of 87% after 24 h, although the single-cell velocity was low, at 0.15 μm/min (Fig. 3a,c). Cell line U-2990MG had a relative wound closure of 76% at 24 h, and migrated at 0.4 μm/min, whereas cell line U-2995MG showed relatively high wound closure, at 90% after 24 h, but slow single cell migration velocity, at 0.2 μm/min (Fig. 3a,c). Cell lines U-2987MG and U-2997MG were less motile, with relative wound closures of 57% and 63%, respectively, and single cell migration speeds of 0.2 μm/min and 0.3 μm/min, respectively (Fig. 3a,c). In contrast, cell lives U-343MGa and U-343MGa Cl2:6 had the lowest migratory capacity (Fig. 3a). In addition, even though cell lines U-343MG and U-2982MG migrated at high speeds, the directness of their migration was lower compared to cell lines U-2995MG and U-2997MG (Fig. 3b).
Another decisive factor for three-dimensional migration is the degradation of the extracellular matrix by cancer cells. To study this property, we analyzed the degradation of fluorescently labeled gelatin by these glioblastoma cell lines. Surprisingly, for some of the cell lines the gelatin degradation was 'opposite' to their Matrigel invasiveness. The most invasive were cell lines U-343MG and U-2982MG, and these showed low gelatin degradation, while cell lines U-343MGa 31L and U-2990MG were most efficient in the gelatin degradation, but they were not very invasive (Fig. 3d,e, Supplementary Tab. S7). Cell lines U-343MGa and U-343MGa CL2:6, which showed the least migration for all of these cell lines, were also the least invasive and the least efficient at gelatin degradation (Fig. 3a,d,    www.nature.com/scientificreports/ processes, and the collective migration efficiency does not always correlate with single speed migration, cell invasion, or gelatin degradation.

Two dimensional hierarchical clustering of cell lines vs parameters
SOX2 and SFRP2 have roles in regulation of subtype transition of the glioblastoma cell lines. As indicated above, based on their morphometric, migratory, and invasive properties, the glioblastoma cell lines were distributed along an axis, and they showed a predominant epithelioid morphology (cell lines U-343MGa 31L, U-343MGa, U-2975MG, U-2987MG, U-2995MG, U-2997MG) or a predominantly fibroblastic morphology (cell lines U-343MG, U-343MGa Cl2:6, U-2982MG, U-2990MG) (Fig. 3f). These data also indicated that cell lines U-2982MG and U-2987MG differed significantly from each other in terms of their migratory and invasive properties. The cell lines also differed significantly in their cytoskeleton organization. Previous studies have shown that based on its gene expression signature, cell line U-2982MG can be stratified as a mesenchymal subtype glioblastoma 19 . In contrast, cell line U-2987MG shows a proneural signature, because of high expression of the transcription factors SOX2 and PROX1 19,29 . Transcriptome analysis by RNAseq or gene arrays reported previously indicated that two separate groups of glioblastoma cell lines can be distinguished 19,25 . For genes related to mesenchymal properties, including migration, invasion, and chemoresistance, cell lines U-2975MG and U-2982MG show greater expression than cell lines U-2987MG, U-2990MG, U-2995MG, and U-2997MG. Examples of such genes are GREM1, FN1, LUM, SPARC , and COL1A1 [30][31][32][33][34] . The transcription factor SOX2 has been shown to regulate glioblastoma gene expression, to shift the cells from proneural subtype to mesenchymal subtype, and SOX2 can be suppressed by SFRP2 29 . To understand more about the critical factors that are decisive for the two major glioblastoma subtypes, we used two cell lines that were produced in an earlier study: cell line U-2982MG with stable overexpression of SOX2 (U-2982MG/SOX2), and cell line U-2987MG with stable overexpression of SFRP2 (U-2987MG/SFRP2). To study the dynamics of cytoskeletal structure during glioblastoma subtype transition, the cell line model used included cell lines U-2982MG, U-2982MG/YFP, U-2982MG/SOX2, U-2987MG, U-2987MG/ YFP, and U-2987MG/SFRP2, which thus represented two cell types with mesenchymal subtype gene expression signature (i.e. U-2982MG and U-2987MG/SFRP2), two cell types with proneural subtype gene expression signature (i.e., U-2982MG/SOX2, and U-2987MG), and two transfection control cell lines (i.e., U-2982MG/YFP and U-2987MG/YFP) (Fig. 3f) 29 . Cell line U-2982MG/SOX2 had more cells with actin arcs and fewer lamellipodia (57%, 37% of the cells, respectively), as compared to cell line U-2982MG (35%, 61%, respectively) ( Fig. 4a-e, Supplementary Tab. S8). In contrast, cell line U-2987MG/SFRP2 had gained a fibroblast-like, mesenchymal appearance, with long stress fibers and with lamellipodia in 52% of the cells, as compared to cell line U-2987MG with 36% ( Fig. 4a-e, Supplementary Tab. S8). Cell line U-2982MG/SOX2 showed a more tangled MT network compared to control cell line U-2982MG, whereas for cell line U-2987MG/SFRP2, the MT network appeared straight and was less wrapped around the nucleus compared to control cell line U-2987MG/YFP (Fig. 5a).
We then analyzed the expression and localization of the IF proteins nestin, GFAP, and vimentin in these cell lines. Cell line U-2987MG/SFRP2 did not express GFAP anymore, contrary to control cell line U-2987MG/YFP, while cell line U-2982MG/SOX2 fully expressed GFAP contrary to cell line U-2982MG and U-2982MG/YFP ( Fig. 5d; Supplementary Fig. S4, Supplementary Tab. S10). Compared to the control cell lines, there was lower nestin expression for cell line U-2987MG/SFRP2 (37% of cells), and higher nestin expression for U-2982MG/ SOX2 cells (100% of cells) (Fig. 5d; Supplementary Fig. S4b, Supplementary Tab. S10). The SOX2 and SFRP2 expression in cell lines U-2982MG and U-2987MG, respectively, did not significantly impact vimentin expression in terms of its area or its distribution, in agreement with the transcriptomic data (Fig. 5b,c, Supplementary Tab. S10) 29 .

SOX2 and SFRP2 have roles in regulation of cell-cell junction integrity in the glioblastoma cell lines.
Cell-cell junctions were visualized with an antibody against β-catenin. For control U-2987MG/YFP cells, β-catenin localized to well-defined cell-cell junctions and at the cell borders with an average thickness size of 0.5 µm (Supplementary Fig. S5a and b). For control cell line U-2982MG/YFP, the cell-cell junctions appeared serrated with an average thickness of 0.85 µm (Supplementary Fig. S5a and b). The localization of β-catenin in cell line U-2987MG/SFRP2 was more similar to control cell line U-2982MG/YFP and the junctions were thicker (0.58 µm), while β-catenin localization in cell line U-2982MG/SOX2 was more similar to control cell line U-2987MG/YFP and the junctions were thinner (0.67 µm) (Supplementary Fig. S5a).
In summary, SFRP2 expression in cell line U-2987MG switched key characteristics of the epithelioid cell line into a fibroblastic phenotype. In contrast, SOX2 expression in cell line U-2982MG resulted in conversion of the fibroblastic phenotype into an epithelioid phenotype (Supplementary Fig. S6). www.nature.com/scientificreports/

Discussion
The overall aim of the current study was to determine whether information on cytoskeletal organization and dynamics could provide clues about differences between glioblastoma subtypes. In a previous study on malignant mesothelioma, we reported consistent differences in the organization of F-actin and intermediate filaments that reflected the aggressiveness of the cells from the malignant mesotheliomas of the patients 24 . In the glioblastoma cell lines here, the F-actin organization differed visibly between the cell lines that were of a more epithelioid subtype compared to those of a more mesenchymal subtype. Thus, cell lines U-343MGa 31L, U-2987MG, and U-2995MG were more circular and had shorter stress fibers, and cell lines U-343MG, U-2975MG, U-2982MG, and U-2990MG were more elongated and had long stress fibers that spanned the entire cell body. This information allowed us to group the cell lines according to an epithelial morphology or a mesenchymal morphology. Cell line U-2987MG is positioned toward a predominantly epithelioid morphology, cell line U-2982MG toward a predominantly fibroblastic morphology, and the other cell lines fit into intermediate positions according to the two-dimensional hierarchical clustering (Fig. 3f). Epithelial-to-mesenchymal transition is characterized by different morphological alterations, which include loss of cell-cell junctions. These intercellular attachment points are made up of different proteins, including cadherins and β-catenin, and they are linked to actin filaments. We observed clear differences in β-catenin organization between the different cell lines here. The clearest differences were seen between cell lines U-2982MG and U-2987MG, and between cell lines U-343MG and U-343MGa 31L. Cell lines U-343MGa 31L and U-2987MG had strong accumulation of β-catenin at the borders between adjacent cells, which is indicative of an epithelioid morphology. In contrast, cell lines U-343MG and U-2982MG had broad and serrated β-catenin-positive areas, which indicates mesenchymal cell characteristics. We did not detect any expression of β-catenin in the nuclei. In summary, β-catenin localization is a clear indicator of epithelioid versus fibroblastic glioblastoma morphology.
A previous analysis of malignant mesothelioma cells demonstrated correlation between migratory potential and tubulin acetylation 24 . Cells with a high migratory index generally have a lower ratio of acetylated tubulin. This is in line with the prevailing concept that acetylated MTs are more stable than non-acetylated MTs, and migrating cells have been shown to have less stable MTs 28,35 . Surprisingly, we did not find such a correlation in glioblastoma cells. Despite clear differences in cell migration, we did not find any major differences in the ratios of acetylated MTs, with the exception of cell line U-343MG, which was highly motile, and cell line U-2987MG, which showed low collective and single-cell migration. Thus based on these results, tubulin acetylation is not an ideal prognostic marker for glioblastoma migratory and invasive potential.
One important observation from this study is that there were significant differences in the migratory properties of the different glioblastoma cell lines. Mesenchymal cell lines U-343MG and U-2982MG had high migratory abilities in the single-cell migration, wound closure, and Matrigel invasion assays. In contrast, cell lines U-343MGa 31L and, in particular, U-2987MG had much lower migratory potentials in all of these assays. Interestingly, the cell migration did not correlate with the matrix degrading activities; indeed, this appeared to be largely the opposite, whereby the least migratory cells showed the most efficient matrix degradation. We also reported this concept in our studies of malignant mesothelioma 24 . The gelatin degradation assay is mainly dependent on invadopodia formation and activity. In contrast, the Matrigel used in the invasion assays is composed of collagen I, collagen IV, and laminin, which together with fibronectin, proteoglycans and hyaluronic acid, constitute the main components of the glioblastoma extracellular matrix environment 36 . This suggests that these two categories of glioblastoma use different modes to spread, and in a cancer setting, they need different treatments to counteract their migratory potential and invasive potential.
We have shown that glioblastoma cells of epithelioid and mesenchymal phenotypes have clear differences in their cytoskeletal fingerprints, which makes it possible to stratify them into subtypes. Cell line U-2987MG had obvious epithelioid properties, and was a representative of this subtype, whereas cell line U-2982MG had typical mesenchymal characteristics, and was a representative of this subtype. Cell line U-2987MG had a low migratory index, which was seen as a low migratory speed for single-cell migration, and for wound closure and Matrigel invasion. In contrast, cell line U-2982MG showed efficient wound closure and invasion, and these cells migrated with the highest speed of all cell lines tested. Importantly, we identified that the two established regulatory elements SOX2 and SFRP2 have key roles in the subtype cytoskeleton specification. During differentiation from a proneural to a mesenchymal subtype, SOX2 expression is lost. Interestingly, re-expression of SOX2 in the mesenchymal type cell line U-2982MG resulted in regression of the phenotype into an epithelioid cell type, based on the cytoskeletal characteristics and migratory behavior. Expression of SFRP2 in cell line U-2987MG resulted in transition to a mesenchymal phenotype with a more specific cytoskeletal organization increasing collective migration and invasion, as SFRP2 has been shown to act as a SOX2 antagonist in glioblastoma cells 29 . In conclusion, elucidating different mechanisms in migration and invasion as well as the potential of SOX2 re-expression to reverse the mesenchymal subtype into a more proneural subtype, should thus open a strategy to alternative treatments in glioblastoma. Moreover, as it has been promising in other type of cancer, further investigations in cytoskeletal reorganization in glioblastoma during tumor progression could be interesting to use as diagnostic tool.

Methods
Cell culture. All cell line were taken from lab Monika Nistér, Karolinska Institutet, Stockholm, Sweden (and from her former supervisor Bengt Westermark, Uppsala University, Uppsala, Sweden). Briefly, the U-343MG and U-343MGa cell lines were established from the same human glioblastoma multiform biopsy. U-343MGa Cl2:6 and U-343MGa 31L cells were derived from the U-343MGa cells 25,37,38 . Cell lines U-2975MG, U-2987MG, U-2995MG, and U-2997MG were diagnosed as grade 3-4 according to the World Health Organization, and cell lines U-2982MG and U-2990MG were diagnosed as grade 4 26  www.nature.com/scientificreports/ U-2987MG/YFP, and U-2987MG/SFRP2 were establish by transduction of SOX2 or SFRP2 overexpression lentivirus vectors into cell lines U-2982MG and U-2987MG, respectively 29 . YFP was introduced into both cell lines U-2982MG and U-2987MG in the same way, as controls. All of the cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% decomplemented fetal bovine serum, and penicillin/streptomycin (1000 U/mL and 1000 μg/mL, respectively). The cell lines were maintained at 37 °C in a humidified incubator with an atmosphere of 5% CO 2 .
Antibodies. The following commercial antibodies and reagents were used: mouse monoclonal anti-αtubulin antibodies (T9026); mouse monoclonal anti-vimentin antibodies (V6630); mouse monoclonal antiacetylated α-tubulin antibodies (T6793); tetramethyl rhodamine isothiocyanate (TRITC)-conjugated phalloidin ( www.nature.com/scientificreports/ Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.