Non-invasive neural stem cells become invasive in vitro by combined FGF2 and BMP4 signaling

Neural stem cells (NSCs) from the developing forebrain can recapitulate in vitro the proliferation and differentiation of all major cell types in the central nervous system (CNS) (Davis and Temple, 1994; Johe et al., 1996; Raff et al., 1983; Ray et al., 1997; Reid et al., 1995). To correctly form the cortex layers NSCs are involved in complex migration patterns (Borello and Pierani, 2010; Guillemot, 2005; Liu, 2011; Rakic et al., 2009; Sur and Rubenstein, 2005). In contrast, only sparse data exist on invasion within the forebrain, although invasion may be critical for correct migration and generation of distinct forebrain structures, such as the interhemispheric mesenchyme (IHM) (Sailer et al., 2005). Indeed, invasion is a complex multistep process that requires proteolytic activities for degradation of extracellular matrix, followed by migration that provides substrate adhesion and traction for the cells to move (Sánchez-Tilló et al., 2012). Invasion has been well described for tumor cells, including gliomas (Nakada et al., 2007), but NSCs are not considered to be invasive. In the present study, we investigated whether NSCs may become migratory and invasive in defined culture conditions. Multiple signaling pathways are active during midterm rat forebrain development, including fibroblast growth factor (FGF), bone morphogenetic protein (BMP), Wingless-type proteins (Wnt) and Notch signaling (Ford-Perriss et al., 2001; Gaiano, 2008; Grove and Fukuchi-Shimogori, 2003; Hébert and Fishell, 2008). We focused on FGF2 and BMP4 signaling because FGF2 is one of the standard mitogens for NSCs, and BMPs are involved in dorsalization and NC induction in the caudal neural tube (Lee and Jessell, 1999; Panchision and McKay, 2002; Sauka-Spengler and Bronner-Fraser, 2008). Our findings reveal a novel mechanism to induce both migration and invasion in normal NSCs, with a molecular expression pattern resembling the highly invasive nature of malignant gliomas. To our knowledge, this study is the first describing similarities between induction of stem cell invasion and glioma invasion pathways.

Multipotent NSCs isolated from the E14.5 rat cortical ventricular zone (CVZ) were expanded in FGF2 during 4 days and passaged. NSCs were then tested whether they were invasive by studying their ability to cross an extracellular (EC) matrix containing basal membrane (BM) extract during a 3 day period with different growth factors (Fig. 1A). FGF2-expanded NSCs were further cultured without any factor or with BMP4 alone (F/- or F/B, respectively). They did not show any invasive character (Fig. 1B,D,F). FGF2-expanded NSCs followed by continued culture in FGF2 (F/F) showed a low frequency of crossing the matrix (Fig. 1C,F). We found that the sequence of FGF2 followed by FGF2 and BMP4 (F/FB) effectively caused NSCs to cross a BM/EC matrix (Fig. 1A–F), with both factors being necessary to induce invasion. Interestingly, the freshly isolated cells did not respond to the combination of FGF2/BMP4, whereas FGF2-expanded cells did (supplementary material Fig. S1). This indicates that naïve cells have to be exposed to FGF2 factors for a period of time to allow the reprogramming of gene expression. Therefore, F/FB describes a specific sequence/code to induce invasion in otherwise non-invasive NSCs.

Next, we tested whether the invasion of NSCs was associated with migration. NSCs migration was investigated in an outgrowth assay: E14.5 CVZ explants containing NSCs were cultured under various conditions as indicated (Fig. 1G–Q). Explants were identically cultured as above for the invasion experiments. Control explants without factor did not show any significant cell outgrowth (Fig. 1G), indicating that there is very little spontaneous migration. Also the combinations BMP4/BMP4 (B/B), F/-, F/B, BMP4/FGF2 (B/F) and the combinations with epidermal growth factor (EGF) EGF/EGF (E/E) and EGF/EGF+BMP4 (E/EB) did not show any strong migration (Fig. 1I–N,Q). Relative to control (168±19 µm), F/F caused a migration of 864±47 µm, F/FB of 1687±123 µm and FB/FB of 2541±97 µm (Fig. 1H,O,P, respectively, and 1Q). Some of the outgrowth of the F/F group was probably due to strong proliferation, as the cells were round, dividing and tiny (supplementary material Fig. S2B). The F/FB and FB/FB cells at the explant edge, in contrast, demonstrated a flat elongated morphology with a leading edge as expected of migratory cells (supplementary material Fig. S2I–L). Single cells in the F/FB and FB/FB groups moved without any obvious particular direction. The overall migration was centrifugal from the explant center. Taken together, these results demonstrate that the F/FB sequence/code induced both the strongest invasion and a very strong (visible without a microscope) migration.

Next, we characterized the invasive cells on the protein level. The F/FB cells crossing the BM/EC matrix (on the bottom of the chamber) were stained for various genes expressed in the dorsoventral axis. We observed that 95.9% of the cells expressed Podoplanin (PDPN) and 92.2% expressed p75NGFR (Fig. 2A–D), but not Sox10 and Olig2 (data not shown). We could not test Snail1/2 because we did not find an antibody that worked reliably in rat tissue.

PDPN has been shown to be expressed in various tumors during migration and metastasis, modulating the actin cytoskeleton (Astarita et al., 2012). As PDPN was expressed in invasive cells (Fig. 2A,B), we screened for it as marker for invasive cells under various conditions in CVZ explants (Fig. 2E–N). We found the highest PDPN expression in the F/FB and FB/FB groups (Fig. 2O,P, respectively). Only explants with the combination of BMP4 and FGF2 contained PDPN-positive cells (Fig. 2O). Notably, EGF, an important mitogen in NSCs, could not substitute for FGF2. This corroborates the data above that FGF2 is both mitogenic, as expected, but also confers the competence for invasion in response to BMP4.

In order to characterize the invasive cells on the molecular level, we performed a time course analysis of gene expression using quantitative reverse transcriptase-polymerase chain reaction (qRT-PCR). To reproduce the invasive condition used above (F/FB), cortical NSCs were first expanded in FGF2 and then exposed to FGF2/BMP4. RNA was harvested for qRT-PCR after 9, 18, 24, 48 or 72 h in order to compare the gene expression with and without BMP4. Since BMP4 is known to be dorsalizing, we screened for genes in the dorsoventral axis including developmental invasion genes (Astarita et al., 2012; Nieto, 2002; Raica et al., 2008). Developmental invasion genes were upregulated during the F/FB time course, starting at 9 h (Fig. 3A–C). Among these, Snail1 showed the strongest upregulation of over 70-fold after 48 h (Fig. 3A). Snail3 was not expressed (data not shown). BMP4 also induced the expression of Tnc, Sparc and Hey1 (Fig. 3D–F), genes known to be highly expressed in human gliomas (Hirata et al., 2009; Hulleman et al., 2009; Rempel et al., 1998). Interestingly, Tnc, which is known to strongly promote glioma invasion (Hirata et al., 2009), was upregulated over 100-fold after 18 to 72 h of BMP4 exposure in NSCs. Further, Pax6, Gli3 and Msx1, transcription factors expressed in the dorsal forebrain (Furuta et al., 1997; Quinn et al., 2009), were significantly upregulated after 24 to 72 h of BMP4 exposure (Fig. 3G–I). P75NGFR (Ngfr), which is found in the NC and in the mesenchyme of the forebrain (Dupin and Coelho-Aguiar, 2013; Morrison et al., 1999; Sailer et al., 2005; Wong et al., 2006), was upregulated over 15-fold at 72 h (Fig. 3J, compare with Fig. 4M). Sox2 was expressed at high absolute levels, but did not significantly change during the time course (supplementary material Fig. S3). The induction of these dorsal genes was paralleled by the downregulation of genes from the ventral pallidal region Gsh1/Gsx1, Gsh2/Gsx2, Sox10, Olig2 and Pdgfra (PDGFRα), and of Egfr (Fig. 3K–P).

Based on upregulation of over 10-fold, we defined Snail1, Pdpn, Tnc, Sparc and Ngfr as the key invasion genes in normal stem cells. To compare these genes from normal stem cells with malignant gliomas and to validate NSC invasion as a tumor model, we tested the five key invasion genes on primary GBM samples. Compared to normal brain controls, the mean expressions of these genes were all upregulated in the most malignant grade IV gliomas (Fig. 3Q). Interestingly, the same gene combination was almost conformly expressed in all tested human glioma samples (Fig. 3R).

Since migration and proliferation are antagonistic cellular behaviors (Schiffer et al., 1997), we tested whether invasion affected proliferation. We compared proliferation and cell cycle arrest in NSCs exposed to the combinations F/F (no migration) and F/FB (inducing migration). We found that the cell cycle inhibitor p21 was more highly expressed in F/FB as compared to F/F (Fig. 4A,B,G). Interestingly, the cell proliferation marker Ki67 was similarly expressed in both groups, indicating that the cells in F/FB group had a slower cell cycle time (Fig. 4C,D,I). As a consequence, the density of cells was more than 2-fold higher in F/F compared to F/FB (Fig. 4H). This observation is consistent with the notion that Snail-dependent tumor invasion is correlated with reduced proliferation (Nieto, 2002; Nieto, 2009). This dichotomy is illustrated in supplementary material Fig. S4, showing the time course of gene expression inducing migration/invasion and reduction of cell cycle speed.

The reduced cell density in F/FB was not due to increased apoptosis, since similar low levels of about 2% of NSCs expressed activated caspase 3 in the F/F and F/FB groups (Fig. 4E,F,J).

It has been shown that BMP4 signaling promotes differentiation of brain tumor stem-like cells (Piccirillo et al., 2006). To investigate if invasive/migratory NSCs undergo differentiation or retain their stem cell fate, we performed immunostaining of NSCs treated with the combinations F/F, F/B or F/FB for nestin, MAP2 and GFAP (Fig. 4K–V). FGF2 is known to preserve the stem cell fate. Indeed, 96±1.8% NSCs were nestin-positive in F/F (Fig. 4K). A similar proportion was found for cells treated with F/FB (90±4.2%, Fig. 4M). In contrast, only 49±13.0% were nestin-positive with F/B (Fig. 4L). A small proportion of NSCs expressed the neuronal marker MAP2 in F/F and F/FB compared to F/B (0.1±0.1% and 3.1±2.2%, compared to 9.6±1.7%, respectively, Fig. 4O–R). Whereas no GFAP-positive cells were observed in F/F (Fig. 4S), few cells (9.0±4.6%, Fig. 4U) expressed this astrocytic marker in F/FB and 30.0±2.0% in F/B (Fig. 4T). Interestingly, F/FB therefore promoted invasion, but not differentiation, whereas F/B promoted differentation, but not invasion. This observation in NSCs is consistent with the findings in brain tumor stem-like cells (Piccirillo et al., 2006).

The results demonstrate that otherwise non-invasive stem cells become invasive although the forebrain does not have any obvious need for invasion. In the posterior neural tube dorsal precursors have to become invasive to form the NC (Dupin and Sommer, 2012). NC derivatives, such as melanocytes, show controlled invasion to form additional cell types within an already existing tissue. For the anterior neural tube, however, it is unclear why it contains potentially invasive cells. The last NC cells emigrate from the mouse neural tube at E9.5 (Serbedzija et al., 1992) and only from regions caudal to the mid-diencephalon (Couly and Le Douarin, 1987; Le Douarin and Kalcheim, 1999). Hence, our experiments demonstrate that NSCs from the forebrain retain the potential to become invasive resembling the EMT of NC cells, even after NC formation of the posterior neural tube. Our results support the hypothesis that the choroid plexus mesenchyme is derived from forebrain NSCs. Consistent with this view, Snail transcription factors have been found in the choroid plexus mesenchyme (Marin and Nieto, 2004). It remains to be shown in lineage-tracing experiments if, when and how this happens during forebrain development in vivo.

We reported previously that pericytes of the choroid plexus express p75NGFR (Sailer et al., 2005), a receptor also found expressed by invasive cells. Thus, the ability to become invasive may be necessary to contribute to mesenchymal perivascular cells within the brain. Likewise, NSCs may also differentiate into peripheral Schwann cells within the adult CNS (Nazarenko et al., 2012).

Gliomas represent the majority (∼80%) of all primary malignant brain tumors and all of them, in particular glioblastoma multiforme (GBM), are strongly invasive (Schwartzbaum et al., 2006). Increasing evidence supports the hypothesis that gliomas derive from a cancer stem cell (Das et al., 2008). Genes encoding the following proteins are expressed in gliomas: Tenascin C (Hirata et al., 2009; Mariani et al., 2001), Hey1 (Hulleman et al., 2009), SPARC (Rempel et al., 1998), Snail1 and Snail2 (Han et al., 2011; Yang et al., 2010), FGFR1 (Morrison et al., 1994), BMPR1a (Yamada et al., 1996), EGFR (Hegi et al., 2012), PDGFRα (Andrae et al., 2008; Nazarenko et al., 2012), Sox2 (Annovazzi et al., 2011), Podoplanin (Mishima et al., 2006), Gli3 (Shahi et al., 2008) and p75NGFR (Johnston et al., 2007). All these glioma-related genes were also expressed in normal NSCs during their transformation into invasive stem cells in the present study, including the constitutively expressed Fgfr1 and Bmpr1a genes (Panchision et al., 2001; Vescovi et al., 1993). Further, invasion was negatively correlated to proliferation, an observation that has also been observed for GBM stem cells (Piccirillo et al., 2006; Vescovi et al., 2006). This study, therefore, unveils an unexpected similarity between invasive NSCs and invasive glioma cells at the transcriptional level. On the basis of these observations we propose that NSC invasion may be used as a model to understand glioma invasion.

If this NSC model is applied to glioma invasion, it suggests that the dorsoventral developmental identity is deregulated in adult glioma cells. This developmental transition between non-invasive (ventral) neuroepithelium to invasive (dorsal) forebrain mesenchyme may be falsely reactivated in adult gliomas. Our results suggest that blocking FGF and BMP signaling may interfere with invasion and migration. These observations are both relevant to understand normal forebrain as well as cancer development.

All animal procedures followed the ‘Principles of laboratory animal care’ (NIH publication No. 86-23, revised 1985) and were approved by the Animal Welfare Committee of Basel City (Swiss Guidelines for the Care and Use of Animals). NSCs were isolated from E14.5 timed-pregnant Sprague-Dawley rats (Harlan, Horst, The Netherlands) and cultured as described (Johe et al., 1996; Sailer et al., 2005). Noon of day of plug detection equaled E0.5. Adult pregnant rats (8 to 12 weeks old) were deeply anesthetized in isoflurane aerosol before decapitation, cesarean section and dissection of the CVZ. E14.5 rat forebrain was dissected free of meninges and a cortex section of 800 µm by 1600 µm along the length of the medial ganglionic eminence (MGE) was isolated and either used for stem cell or explant preparations. The CVZ started at the anterior MGE pole and followed the MGE anteroposteriorly at a distance of 800 µm from the lateral edge of the MGE. Explants were cut by microsurgical needle to a diameter size of 500 µm. Further trituration yielded single stem cells as previously described (Johe et al., 1996; Kim et al., 2003; Sailer et al., 2005).

NSCs and CVZ explants were cultured in DMEM/F12 (with N2, omitting progesterone) and on poly-ornithine/fibronectin-coated dishes, as described (Sailer et al., 2005). Medium containing growth factors was replaced every other day. The explants were washed twice with basal medium between the first and second phases of treatment. The following concentrations of growth factors were used: 10 ng/ml recombinant human (rh) FGF2 (146 aa), 10 ng/ml rhBMP4, 10 ng/ml rhEGF (all from R&D Systems; with BSA as carrier protein) and bovine pancreatic insulin at 25 µg/ml (Sigma).

Cell invasion was tested using the CytoSelect cell invasion assay (Cell Biolabs) with 8 µm pore size. The inserts are coated with EC/BM matrix. The matrix is formed by 200 µl 0.4 mg/ml-matrigel that is lyophilized on a 60 mm² bottom area (manufacturer specifications). The EC/BM matrix consists of laminin (56%), collagen type IV (31%), entactin (8%) and heparan sulfate proteoglycan (<5%) (Kleinman et al., 1982). Cells were added to the upper chamber and incubated for 72 h in medium with or without growth factors as indicated in the legends of the figures. Growth factors were identical in top and bottom chambers. The non-invaded cells from the upper side of the filter were scraped using moist cotton swabs. The invaded cells in the reverse side of the filter were stained with CellMask Orange plasma membrane stain (CMO, 1∶1000, Invitrogen), antibodies or with Crystal violet, lysed and quantified at OD 560 nm.

Migration distance was measured from the explant edge to the outermost 200×200 µm² area containing at least 50 cells. We used Fiji/ImageJ v1.45i software (Schindelin et al., 2012) to define the outermost cell using a tangent to the explant surface with a perpendicular radial line, using four sample distances per explant.

Explants or NSCs were fixed and stained with fresh cold 4% paraformaldehyde (PFA)/PBS as described (Sailer et al., 2005). Briefly, primary antibodies were diluted in 5% NGS in PBS, applied after 1 h blocking in 0.2% Triton X-100/10% NGS in PBS and incubated overnight at 4°C. After washing in PBS, appropriate secondary antibodies coupled with fluorescent dyes (Cy3, 1∶300, from Jackson Immunoresearch, or Alexa Fluor 488, 1∶500 from Molecular Probes) were applied for 1 h at RT. Primary antibodies were as follows: anti-Caspase3a (rabbit, 1∶100, Chemicon), anti-MAP-2 (mouse, 1∶500, Millipore), anti-nestin (rabbit, 1∶500, Novus), anti-GFAP (rabbit, 1∶1000, DAKO), anti-p21/WAF1 (rabbit, 1∶800, Abcam), anti-p75NGFR (rabbit, 1∶200, Millipore), anti-PDPN (mouse, 1∶100, Acris). Nuclei were labeled with 0.25 µg/ml DAPI (Sigma) for 15 min and slides were mounted in Vectashield (Vector Labs). Images were photographed using a DMRE (Leica Microsystems) microscope combined with an F-view camera (Soft Imaging) or with an Axiovert 25 microscope (Carl Zeiss) combined with a C-5060 camera (Olympus).

The human brain material was obtained from the Neurosurgery clinic, University Hospital of Basel, Switzerland, in accordance with Swiss law and the Ethical Committee Beider Basel. Patient consent was obtained for every sample. Seven human glioma tissue samples (GBM, grade IV) and two epileptogenic tissue samples were collected after resection surgery. The samples were immediately frozen in liquid nitrogen before further RNA extraction.

For gene expression analysis, rat NSCs or human glioma tissue samples (GBM grade IV), and control human brain tissue (epileptogenic brain tissue) were harvested in Trizol reagent (Invitrogen) for isolation of RNA using RNeasy Mini Kit (Qiagen) and treated with DNase I (Turbo DNA-free Kit, Ambion). First strand cDNA was created using the QuantiTect Reverse Transcription Kit (Qiagen). All rat and human specific primers are commercially available (QuantiTect Primer Assays, Qiagen). Quantitative RT-PCR was performed on 300 ng to 1 µg of total cDNA per well using the QuantiFast SYBR Green PCR kit (Qiagen). Each sample was run in triplicate on the Bio-Rad detection system (C1000 and CFX96, Bio-Rad). The comparative cycle threshold [C(t)] method was used to calculate relative quantification of gene expression. Each measurement was made in duplicate and expressed relative to the internal control GAPDH [ΔC(t)]. The following formula was used to calculate the relative amount of the transcripts in the FGF2-treated conditions samples and the FGF2+BMP4 conditions: ΔΔC(t) = ΔC(t)FGF2+BMP4−ΔC(t)FGF2. The formula 2E-ΔΔC(t) expressed the fold-change for the FGF2+BMP4-treated relative to FGF2-treated samples. For human glioma samples, the relative amounts of transcripts were calculated relative to the mean amounts of the two epileptogenic tissue samples.

Samples were analyzed using the Student's t-test for unpaired data doublets and one way ANOVA for unpaired groups. Values describe means ± s.e.m. Statistical analysis was performed using Prism 6.0 (Graphpad).

We would like to thank Elisabeth Taylor for technical assistance. We thank Jean-Louis Boulay, Martin E. Schwab and Daniel Bodmer for discussions on the manuscript. The work is dedicated to Prof. Hanna Keidel.