Transcriptional Factors for Epithelial–Mesenchymal Transition Are Associated with Mesenchymal Differentiation in Gliosarcoma

Abstract

Gliosarcoma is a rare variant of glioblastoma characterized by a biphasic pattern of glial and mesenchymal differentiation. It is unclear whether mesenchymal differentiation in gliosarcomas is because of extensive genomic instability and/or to a mechanism similar to epithelial–mesenchymal transition (EMT). In the present study, we assessed 40 gliosarcomas for immunoreactivity of Slug, Twist, matrix metalloproteinase‐2 (MMP‐2) and matrix metalloproteinase‐9 (MMP‐9), which are involved in EMT in epithelial tumors. Nuclear Slug expression was observed in >50% of neoplastic cells in mesenchymal tumor areas of 33 (83%) gliosarcomas, but not in glial areas (P < 0.0001). Nuclear Twist expression was observed in >50% of neoplastic cells in mesenchymal tumor areas of 35 (88%) gliosarcomas, but glial tumor areas were largely negative except in four cases (P < 0.0001). Expression of MMP‐2 and MMP‐9 was also significantly more extensive in mesenchymal than in glial tumor areas. None of 20 ordinary glioblastomas showed Slug or Twist expression in >10% neoplastic cells. Thus, expression of Slug, Twist, MMP‐2 and MMP‐9 was characteristic of mesenchymal tumor areas of gliosarcomas, suggesting that mechanisms involved in the EMT in epithelial neoplasms may play roles in mesenchymal differentiation in gliosarcomas.

INTRODUCTION

The epithelial–mesenchymal transition (EMT) is the conversion of cells with an epithelial phenotype into cells with a highly motile fibroblastoid or mesenchymal phenotype 23, 33. EMT is a critical mechanism in embryonic development, chronic inflammation and fibrosis 10, 33, 36, and is considered to play a key role in tumor progression 6, 25, 41 and metastasis 38, 42. Several transcription factors, including Slug, Snai1, Twist, matrix metalloproteinases (MMPs), have been reported to be involved in EMT 11, 26, 27, 33.

Gliosarcoma is a variant of glioblastoma that is characterized by biphasic neoplastic tissue composed of glial and mesenchymal components (14). The mesenchymal component shows long bundles of spindle cells with malignant features, and abundant deposition of collagen. Despite the presence of these two distinct histological components, gliosarcomas are considered to be monoclonal tumors, as genetic analyses have shown that glial and mesenchymal tumor areas are usually genetically identical in terms of TP53 mutations, PTEN mutations, p16^INK4a deletion, CDK4 amplification and MDM2 amplification 2, 28.

It is unclear whether mesenchymal differentiation in gliosarcoma simply reflects the extensive genomic instability of glioblastomas and/or whether mesenchymal differentiation is caused by a mechanism similar to that involved in EMT in epithelial neoplasms 16, 33. Yang et al (37) showed that Slug expression facilitates cellular invasion and angiogenesis as well as increasing spindle morphology in a mouse intracranial human glioma. Mikheeva et al (19) revealed that Twist1 promotes invasion through mesenchymal changes in glioblastoma.

In the present study, we carried out immunohistochemistry for Slug, Twist, matrix metalloproteinase‐2 (MMP‐2) and matrix metalloproteinase‐9 (MMP‐9) in 40 gliosarcomas, to assess whether there is evidence of expression of genes associated with EMT, suggesting that mechanisms similar to EMT may be operative in gliosarcomas.

MATERIALS AND METHODS

Tumor samples

Formalin‐fixed tissue samples of 40 gliosarcomas and 20 primary (de novo) glioblastomas were obtained from the Department of Neuropathology, University Hospital Zürich, Switzerland; the Institute of Neuropathology, University Hospital Munster, Germany; the Institute of Neuroscience, Bordeaux, France; the Department of Neuropathology, University Hospital Rome, Italy; and the Department of Pathology, Gunma University, Japan. Gliosarcomas from patients with recurrent disease were not included in this study. The median age at histological diagnosis was 59 ± 11 years (range, 32–82 years) and the male/female ratio was 1.22. Survival data were available for 16 patients. Mean follow‐up time was 10 months, and 2 out of 16 (13%) patients were still alive at the time of the study. Mean overall survival among the 14 patients who died was 8.5 ± 7.5 months (range, 3–29.7 months).

Sections were routinely stained with hematoxylin and eosin (HE), glial fibrillary acidic protein (GFAP) and reticulin. The glial areas with astrocytic features showed GFAP expression without reticulin staining. The mesenchymal areas showed typical reticulin expression in bundles of neoplastic spindle cells.

Immunohistochemistry

Immunohistochemistry was carried out using the Ultravision Quanto detection system (Thermo Scientific, Fremont, CA, USA) or Vectorstain ABC kit (Vector Laboratories, Burlingame, CA, USA). Briefly, paraffin sections were deparaffinized in xylene and rehydrated through a graded series of ethanol dilutions. Endogenous peroxidase activity was blocked by incubation in 0.3% hydrogen peroxidase for 30 minutes. After pretreatments performed according to manufacturer's instructions, sections were incubated at room temperature for 1 h with primary antibody, and visualized with diaminobenzidine. Antibodies were as follows: rabbit monoclonal to Slug (9585, Cell Signalling, Boston, MA, USA; dilution, 1:100), mouse monoclonal to Twist (ab50887, Abcam, Cambridge, UK; dilution, 1:250), mouse monoclonal to MMP‐2 (MS804, ThermoFisher Scientific, Cheshire, UK; dilution, 1:100) and rabbit polyclonal to MMP‐9 (RB9234, ThermoFisher Scientific; dilution, 1:500).

Immunoreactivity to each antibody was evaluated in glial and mesenchymal tumor areas separately. Staining was defined as nuclear (Slug, Twist) or cytoplasmic (MMP‐2, MMP‐9) immunoreactivity in neoplastic cells, and was scored as: −, <10% of positive cells; +, 10%–50%; ++, 51%–90%; and +++, >90%. The number of neoplastic cells with immunoreactivity for Slug, Twist, MMP‐2 or MMP‐9 was counted in 15–20 fields per tumor area by light microscopy at ×400 magnification. For Twist, vascular epithelial cells on the same histological slide were used as internal positive controls. Neoplastic cells with expression of Twist that was similar or higher than that in vascular epithelial cells were considered to be positive for Twist.

To assess co‐localization of Slug and GFAP, or Twist and GFAP, in tumor cells, double immunohistochemical staining was performed with antibodies to GFAP and Slug, or GFAP and Twist, in four gliosarcomas. Sections were first immunostained for Slug or Twist. The epitopes were then visualized with diaminobenzidine (brown staining). After completion of the first staining, sections were treated in Linblock (AbCys, Paris, France) for a total of 4 minutes at room temperature to remove the Slug or Twist antibody complex. Sections were then incubated with antibodies to GFAP overnight at 4°C and visualized with Vector VIP (purple staining; Vector Laboratories). After washing sections with phosphate‐buffered saline, a slight green counterstaining of the nuclei was performed for 5 minutes in a 0.5% solution of methyl green (Gurr, Bucks, UK).

Statistical analyses

Chi‐squared test or Fisher's exact test were used to assess the group differences analysis of qualitative features, and a Kaplan–Meier curve with log‐rank test was used for survival analysis. For comparative analysis of immunohistochemical expression in all protein groups (Slug, Twist, MMP2 and MMP9), data were analyzed using post hoc Bonferroni–Dunn tests to compare individual protein groups. We declared statistical significance at the traditional threshold of P = 0.05. However, when multiple comparisons were made, we also included Bonferroni‐corrected P‐values.

RESULTS

Immunohistochemistry to Slug, TwisT, MMP‐2 and MMP‐9

Nuclear Slug expression was observed in >50% neoplastic cells in mesenchymal tumor areas of 33 (83%) gliosarcomas (Table 1; Figure 1). In contrast, neoplastic cells in glial tumor areas were largely negative (P < 0.0001).

Table 1.

Fraction of immunoreactive cells to moleculaes associated with EMT in glial and mesenchymal tumor areas of gliosarcomas

	Slug		Twist		MMP‐2		MMP‐9
	Glial	Mesenchymal	Glial	Mesenchymal	Glial	Mesenchymal	Glial	Mesenchymal
Gliosarcomas
GS1	−	+	−	+	−	−	−	−
GS2	−	++	−	++	−	+	−	+
GS3	−	+++	−	+	−	++	−	−
GS4	−	+	−	++	−	+	−	−
GS5	−	+	−	++	+	+	−	+
GS6	−	+++	+	+++	+	++	−	+++
GS7	−	++	−	+++	−	+	+	+++
GS8	+	+++	−	+++	+	+	+	+
GS9	−	++	−	+++	−	+	−	+
GS10	−	−	−	+++	−	++	+	++
GS11	−	+++	++	+++	−	+++	+++	+++
GS12	−	+	−	+++	−	++	++	++
GS13	−	+++	++	+++	−	+	+	+++
GS14	−	+++	−	+++	+	+++	−	+++
GS15	−	++	−	+++	+	+	−	++
GS16	−	+++	−	+++	−	++	−	+++
GS17	−	+++	−	+++	−	++	−	+++
GS18	−	+++	−	+	+	+++	+	++
GS19	−	++	−	+++	+	+	+	++
GS20	−	+++	−	+++	−	+++	−	+++
GS21	−	+++	−	+++	+	++	+	++
GS22	−	+++	+	+++	−	++	−	++
GS23	−	++	−	+	−	++	+	++
GS24	−	++	−	+++	+	++	++	++
GS25	−	++	−	+++	−	++	−	++
GS26	−	+++	−	+++	+	+	+	+++
GS27	−	+++	−	+++	+	+	+	+
GS28	−	+++	−	+++	−	++	−	+++
GS29	−	+	−	+++	+	+	++	+++
GS30	−	++	−	+++	−	++	+	++
GS31	−	+++	−	+++	+	+	+	++
GS32	−	+++	−	+++	+	+++	−	+
GS33	−	+++	−	+++	−	++	++	++
GS34	−	+++	−	++	−	++	++	+++
GS35	−	+++	−	+++	+	++	+	++
GS36	−	++	−	+++	−	++	+	++
GS37	−	++	−	+	+	++	+	+
GS38	−	+	−	+++	−	++	−	+++
GS39	−	+++	−	+++	−	++	−	++
GS40	−	++	−	+++	−	++	−	+
Glioblastomas
G1	−		−		−		−
G2	−		−		−		−
G3	−		−		+		+
G4	−		−		+		+
G5	−		−		−		−
G6	−		−		+		+
G7	−		−		−		−
G8	−		−		−		−
G9	−		−		−		−
G10	−		−		+		−
G11	−		−		+		−
G12	−		−		−		−
G13	−		−		+		+
G14	−		−		+		−
G15	−		−		+		−
G16	−		−		++		+
G17	−		−		−		−
G18	−		−		−		−
G19	−		−		−		−
G20	−		−		+		−

The number of Slug‐, Twist‐, MMP‐2‐ and MMP‐9‐positive cells in immunohistochemistry was estimated using a four‐tiered scale; −, <10%; +, 10%–50%; ++, 51%–90%; +++, >90%.

Figure 1.

A border between mesenchymal and glial tumor areas of gliosarcomas. Note that Slug (A), Twist (B), MMP‐2 (C) and MMP‐9 (D) are selectively expressed in mesenchymal tumor areas of gliosarcomas.

Nuclear Twist expression was observed in >50% of neoplastic cells in mesenchymal tumor areas of 35 (88%) gliosarcomas (Table 1). In contrast, neoplastic cells in glial tumor areas were largely negative, except for four cases (P < 0.0001).

Cytoplasmic MMP‐2 expression was detected in >50% neoplastic cells in mesenchymal tumor areas of 26 (65%) gliosarcomas, but not in glial areas of any gliosarcoma (Table 1) (P < 0.0001). Cytoplasmic MMP‐9 expression was seen in >50% of neoplastic cells in mesenchymal tumor areas of 29 (73%) gliosarcomas, but in glial tumor areas of only six cases (15%, P < 0.0001) (Table 1).

In gliosarcomas, Bonferroni's correction revealed significant correlations between expression of Twist and MMP‐9 (P = 0.0002), Slug and MMP‐2 (P = 0.0005), and MMP‐2 and Twist expression (P < 0.0001), but there was no significant correlation between MMP2 and MMP9 (P = 0.16), MMP9 and Slug (P = 0.035), or Slug and Twist (P = 0.09).

We correlated survival of patients with gliosarcoma and the fraction of mesenchymal cells expressing Slug, Twist, MMP‐2 or MMP‐9 (>90%, 10%–90% or <10%). There was no significant difference in survival among gliosarcoma patients whose tumors showed different fractions of Slug, Twist, MMP‐2 and MMP‐9 immunoreactivity (data not shown).

None of the 20 ordinary glioblastomas showed Slug or Twist expression in >10% neoplastic cells (Table 1). The fraction of cells expressing MMP‐2/MMP‐9 in ordinary glioblastomas was similar to those in glial tumor areas of gliosarcomas (Table 1).

Double staining of GFAP and Slug, or GFAP and TWIST

In the majority (36 cases; 90%) of gliosarcomas, the mesenchymal components typically lacked GFAP expression. In four cases, focal mesenchymal tumor areas with co‐expression of GFAP and reticulin were detected (Figure 2); double staining demonstrated co‐expression of GFAP and Slug in 2 of these cases and co‐expression of GFAP and Twist in three cases (Figure 2).

Figure 2.

Consecutive sections of the mesenchymal tumor area of a gliosarcoma co‐expressing reticulin (A), Slug (B) and Twist (C). Note GFAP expression (D) in a focal area. Double stainings show co‐expression of GFAP (purple, cytoplasms) and Slug (brown nuclei) (E), or GFAP (purple, cytoplasms) and Twist (brown nuclei) (F) at the single‐cell level.

Analyses of TCGA data

We assessed mRNA expression of Slug, Twist, MMP‐2 and MMP‐9 in glioblastomas in the TCGA database (http://cancergenome.nih.gov). Z scores of > +2 were considered as overexpression. A small fraction of glioblastomas (4%) showed Slug (Snai2) overexpression, and this was significantly associated with poorer outcome of glioblastoma patients (mean survival, 11.6 vs. 14.3 months; P = 0.032). Twist1 overexpression was found in 9% of glioblastomas; this was not prognostic for patient survival. Overexpression of MMP‐2 and MMP‐9 was detected in 3.6% and 1.4% of glioblastomas, respectively.

We correlated mRNA expression of Slug, Twist, MMP‐2 and MMP‐9 in glioblastomas with their transcriptional expression profiles, that is, proneural, neural, classical, mesenchymal subgroups, reported by Verhaak et al (35). The mesenchymal subgroup was characterized by gene expression typical of mesenchymal tumors, but molecules involved in EMT, such as Slug, Twist, MMP‐2 or MMP‐9, were not used to define the mesenchymal subgroup (35). Slug (Snai2) and Twist1 were significantly overexpressed in mesenchymal subgroups compared with the other subgroups (proneural, neural, classical) (5/56, 9%, P = 0.007 for Slug; and 13/56, 23%, P = 0.0001 for Twist). Overexpression of MMP‐9 was found in 2/138 cases, and these two positive cases showed a mesenchymal expression profile. Overexpression of MMP‐2 was found in 3 of 56 cases in the mesenchymal subgroup, and 2 of 54 in the classical subgroup, but in no tumors in the neural (29 cases) or proneural (56 cases) subgroups.

DISCUSSION

We show that expression of Slug, Twist, MMP‐2, MMP‐9 is characteristic of mesenchymal tumor areas of gliosarcomas, suggesting that similar mechanisms involved in the EMT in epithelial neoplasms play roles in mesenchymal differentiation in gliosarcoma.

In gliosarcoma, GFAP expression in glial cells and reticulin staining in mesenchymal‐shaped cells are typically mutually exclusive (14). However, we may encounter foci of GFAP‐expressing fibroblastic‐like neoplastic cells in mesenchymal tumor areas in gliosarcomas. In the present study, four gliosarcomas with such foci of GFAP‐expressing cells in mesenchymal tumor areas showed co‐expression of GFAP/Slug and GFAP/Twist (Figure 2). These foci may represent the initial step in glial–mesenchymal transition in gliosarcoma.

Slug was first identified in the neural crest and in mesodermal cells emigrating from the primitive streak in chick embryos and is expressed by cells undergoing EMT during embryonic development (20). Overexpression of Slug has been reported in a variety of neoplasms, including gastric cancer (29), ovarian cancer 4, 39, breast cancer 4, 17, bladder cancer (40) and lung cancer (8). Yang et al (37) subcutaneously injected U251 glioblastoma cells overexpressing Slug using Slug lentivirus, into nude mice, and found that glioblastomas overexpressing Slug displayed spindle‐like morphology, suggesting mesenchymal differentiation (37).

Twist is a helix‐loop‐helix transcription factor and plays a central role in determining cell fate in mesoderm 12, 43. Overexpression of Twist has been observed in a variety of tumors, including gastric cancer (29), lung cancer (8), breast cancer (17), bladder cancer (40), ovarian cancer (39) and choroid plexus papillomas (7). In SNB19 glioblastoma cells, overexpression of Twist1 produced mesenchymal cellular changes (19). Furthermore, Twist1 significantly increased invasion in orthotopic xenotransplants and increased expression of genes that are involved in mesenchymal phenotypes in SNB19 and T98G glioblastoma cells (19).

MMPs are proteolytic enzymes that play key roles in tumor invasion, metastasis and angiogenesis (3). MMP‐2 5, 26, 34 and MMP‐9 26, 32, 34 are often up‐regulated during EMT. MMP‐2 and MMP‐9 up‐regulation has been associated with poor prognosis in various tumors 13, 18.

In the present study, nuclear Slug expression was observed in >50% of neoplastic cells in mesenchymal tumor areas of 33 (83%) gliosarcomas, but not in glial areas. Nuclear Twist expression was observed in >50% of neoplastic cells in mesenchymal tumor areas of 35 (88%) gliosarcomas, but glial tumor areas were largely negative except in four cases. None of the 20 ordinary glioblastomas showed Slug or Twist expression in >10% neoplastic cells. Similarly, the fractions of neoplastic cells expressing MMP‐2 and MMP‐9 in ordinary glioblastomas were similar to those in the glial tumor areas of gliosarcomas, that is, much less extensive than in mesenchymal tumor areas of gliosarcomas. Consistent with this finding, TCGA data show that overexpression of Slug, Twist, MMP‐2 and MMP‐9 in glioblastomas is rare (1.4%–9% of all cases).

The other widely used markers for EMT in epithelial tumors appear to be loss of E‐cadherin and up‐regulation of N‐cadherin (23). We carried out E‐cadherin immunohistochemistry to assess whether there are differences in E‐cadherin expression between glial and mesenchymal tumor areas of gliosarcomas (data not shown). However, strong E‐cadherin expression was not detectable in any of the glioblastomas or gliosarcomas analyzed, and therefore it was not possible to reliably assess whether there was down‐regulation of E‐cadherin in mesenchymal tumor areas of gliosarcomas. It has been also reported that, distinct from EMT in epithelial neoplasms, TWIST1 did not generate an E‐ to N‐cadherin “switch” in glioblastoma cells (19).

Several studies have shown that glioblastomas may be classified according to their expression profiles 24, 35. Verhaak et al (35) classified 204 glioblastomas from the TCGA project into four subgroups, that is, mesenchymal, proneural, neural and classical; the mesenchymal subgroup, for example, was characterized by expression of genes typical to mesenchymal tumors. However, Slug, Twist, MMP‐2 or MMP‐9 were not included among the genes defining mesenchymal subgroup of glioblastomas (35). Our analyses using the dataset from Verhaak et al (35) showed that Slug and Twist were significantly overexpressed in mesenchymal subgroups in glioblastoma compared with the other subgroups (proneural, neural, classical), and that MMP‐2 and MMP‐9 were also selectively expressed in glioblastomas with a mesenchymal phenotype.

There is evidence that EMT is associated with tumor metastasis in human cancer 9, 38, 42. Extracranial metastasis of glioblastomas is extremely rare (31), although it may occur in cases of secondary glioblastoma with longer survival (15). Beaumont et al reported extracranial metastasis in 18 of 162 (11%) gliosarcomas (1), which was more frequent than in ordinary glioblastomas (0.4%–0.5%) 22, 31. Sites of metastasis of gliosarcoma were the lung, liver and lymph nodes (1). Interestingly, metastatic foci from gliosarcomas were typically composed of tumor cells showing mesenchymal differentiation, but not of glial cells 1, 21, 30.

In summary, we show that mesenchymal differentiation in gliosarcoma is associated with up‐regulation of several transcriptional factors known to be involved in EMT. Our results suggest that mesenchymal differentiation in gliosarcomas is at least partly attributable to mechanisms similar to EMT, and gliosarcoma may be a useful model to study glial–mesenchymal transition in glioblastoma.