In vivo fluorescence resonance energy transfer imaging reveals differential activation of Rho-family GTPases in glioblastoma cell invasion

The most common and aggressive human primary brain tumor is the glioblastoma, which invades extremely rapidly and culminates in the death of patients usually within a year after diagnosis (Louis et al., 2007). Glioblastoma was previously called glioblastoma ‘multiforme’, reflecting its histopathological divergence in size, shape, karyotype, etc. Glioblastomas invade either around the vascular space or along neuronal fibers into the brain parenchyma (Bellail et al., 2004; Furnari et al., 2007). Among many experimental models of human glioblastoma, the allograft model of rat C6 glioblastoma cells is used most extensively (Grobben et al., 2002). The C6 glioblastoma cells implanted into syngeneic Wistar rats share many histological hallmarks with human glioblastoma and preferentially migrate along neuronal fibers and through the perivascular region, which resembles the spread of human glioblastoma.

Various types of cancer cell exhibit diverse invasion morphologies with great plasticity not only in the tissues but also in the three-dimensional (3D) substrate (Sahai and Marshall, 2003; Friedl and Wolf, 2010). Rho-family GTPases are considered to play the central role in the regulation of invasion (Sahai et al., 2007; Croft and Olson, 2008; Sanz-Moreno et al., 2008), probably through cytoskeletal reorganization (Sahai and Marshall, 2003; Kurokawa and Matsuda, 2005; Pertz et al., 2006; Machacek et al., 2009). It has also been documented that coordinated activation and/or antagonistic action of Rho-family GTPases determine the invasion modes of various cancer cell types in the 3D environment (Sahai et al., 2007; Croft and Olson, 2008; Sanz-Moreno et al., 2008). However, the activity of Rho-family GTPases in vivo or in 3D substrate has never been successfully shown in mammals. Consequently, the spatial regulation of Rho-family GTPase in cancer tissues is unknown.

We and others have been developing biosensors based on the principle of Förster resonance energy transfer (FRET) for Ras-superfamily GTPases (Aoki and Matsuda, 2009). These biosensors, which are collectively called Ras and interacting protein chimeric unit (Raichu), have been successfully used to visualize the spatiotemporal regulation of Rho-family GTPases within MDCK cells (Kurokawa and Matsuda, 2005), zebrafish (Miyagi et al., 2004) and Drosophila melanogaster (Kamiyama and Chiba, 2009). However, there was a fatal flaw in the application of the current FRET biosensors to 3D imaging: stable transfectants of FRET biosensors are not readily obtained when fluorescent proteins derived from Aequorea victoria are used as the FRET donor and acceptor proteins (Aoki and Matsuda, 2009). To circumvent this problem, we have developed FRET biosensors with a teal fluorescent protein (TFP) as a FRET donor and succeeded in establishing C6 glioblastoma cell lines expressing FRET biosensors. With these newly developed FRET biosensors, we demonstrate that two different invasion modes of glioblastoma cells can be achieved by the activity balance of Rac1 and Cdc42 versus RhoA. This spatially biased activity is at least partly dependent on Zizimin1 (also known as DOCK9), a Cdc42-specific guanine nucleotide exchange factor.

C6 glioblastoma cells expressing GFP as a marker were grafted into rat brains and after 7 days were observed ex vivo by the brain slice culture technique. The transplanted C6 glioblastoma cells looked poorly delineated macroscopically (Fig. 1A). However, under a two-photon microscope, glioblastoma cells could be seen to have advanced preferentially along blood vessels at the invasion front (Fig. 1B,C), and had often penetrated the brain parenchyma from the perivascular region (arrowheads in Fig. 1C). Time-lapse images revealed that glioblastoma cells migrating along blood vessels (Fig. 1D, arrowhead; supplementary material Movie 1) and those invading the brain parenchyma (Fig. 1E; supplementary material Movie 2) exhibited striking differences, not only in their morphology but also in their mode of invasion. Glioblastoma cells migrating along blood vessels were spindle shaped with a single pseudopodium extending toward the direction of movement (Fig. 1D, arrowhead; supplementary material Movie 1), but glioblastoma cells invading the brain parenchyma extended multiple pseudopodia, which continuously probed in various directions (Fig. 1E; supplementary material Movie 2). One of these protrusions then became ‘favored’ and determined the direction of cell invasion; however the mechanisms underlying the selection of protrusion are not known. To quantify the difference in the two invasion modes, we measured the velocity and net displacement of the glioblastoma cells in each invasion mode at the periphery of the tumor mass (Fig. 1F). Both the velocity and net displacement of glioblastoma cells migrating along the perivascular region were larger than those of glioblastoma cells invading the parenchyma (Fig. 1G,H). These observations demonstrated two modes of glioblastoma invasion: (1) glioblastoma cells advance along blood vessels straightforwardly and rapidly (Fig. 1I); (2) some glioblastoma cells penetrate the brain parenchyma, extending multiple pseudopodia (Fig. 1J) eventually filling the inter-blood vessel space (Fig. 1K).

To elucidate the role of Rho-family GTPases in the regulation of the two different invasion modes of glioblastoma cells, we developed new FRET biosensors that could be expressed stably in glioblastoma cells. For this purpose, we modified FRET biosensors for Rac1, Cdc42 and RhoA (Itoh et al., 2002; Yoshizaki et al., 2003) by adopting teal fluorescent protein (TFP) and a yellow fluorescent protein variant, Venus, as donor and acceptor fluorescent proteins, respectively (supplementary material Fig. S1). We found that the sensitivity and the dynamic rages of the new biosensors were almost comparable with those of the previously reported biosensors. Then, we established C6 glioblastoma cells stably expressing these biosensors, and grafted them into rat brains. Glioblastoma cells extending multiple pseudopodia at the periphery of the tumor mass grossly exhibited higher Rac1 activity than those inside the tumor mass (Fig. 2A–C), and this activity gradient was independent of the expression level of the biosensors (Fig. 2C). When viewed at higher magnification the glioblastoma cells invading the parenchyma with multiple pseudopodia (arrowheads in Fig. 2D,E) exhibited higher Rac1 activity than those advancing along blood vessels (arrows in Fig. 2D,E). A similar observation was obtained for Cdc42 (Fig. 2F–I). In clear contrast, RhoA activity was lower in the glioblastoma cells invading the parenchyma with multiple pseudopodia than in those advancing along blood vessels (Fig. 2J–M). To visualize the difference in the activities of Rho GTPases more clearly, we extracted Rac1-high (red) and Rac1-low (blue) signals from the IMD image (supplementary material Fig. S2B,F,J). We further quantified the FRET ratio by two methods. First, cells at the perivascular region and intraparenchymal region were marked manually and the FRET ratio (FRET/TFP) was calculated and compared (supplementary material Fig. S2C,D,G,H,K,L). Second, regions were automatically set depending on the distance from the blood vessels. The FRET ratio in each region was calculated and quantified (supplementary material Fig. S3). Both methods confirmed our observation that Rac1 and Cdc42 activities were lower and RhoA activity was higher at the perivascular region than at the intraparenchymal region, and vice versa. Although various artifacts should be taken into consideration in the intravital FRET imaging, this inverse correlation between the activities of Rac1 and Cdc42 and those of RhoA, partially ignores the possible artifacts caused by the difference in cell density, depth of tissue, etc. In short, these observations suggest two interesting possibilities. First, glioblastoma cells penetrating the parenchyma might be subjected to stimulants that activate Rac1 and Cdc42 but suppress RhoA. Second, each cell might stochastically exhibit a different level of Rho-family GTPase activities. In this scenario, cells with high Rac1 and Cdc42 but low RhoA activities penetrate the brain parenchyma, whereas cells with low Rac1 and Cdc42 but high RhoA activities persist to advance along the blood vessels.

To test these two possibilities, we performed a 3D spheroid invasion assay (Gaggioli et al., 2007). C6 glioblastoma cells were cultured in suspension to form spheroids and embedded in 3D Matrigel. These C6 glioblastoma cells within the spheroid invaded into Matrigel in several directions (Fig. 3A). Interestingly, the glioblastoma cells with multiple pseudopodia led the other spindle-shaped cells during invasion (arrowheads and arrows in Fig. 3B, respectively). Magnified FRET imaging revealed that these glioblastoma cells leading the invasion had higher Rac1 activities (Fig. 3C–G; supplementary material Fig. S4 and Movie 3) and Cdc42 activities (Fig. 3H–K; supplementary material Fig. S4) than the trailing glioblastoma cells. By contrast, such intercellular divergence was not clear for RhoA activity (Fig. 3L–O; supplementary material Fig. S4). This pattern of activity gradient was similar to that found in the rat brain (Fig. 2), except that low RhoA activity in the multipod cells was not recapitulated in this spheroid invasion assay. Importantly, only a limited number of glioblastoma cells initiated the invasion of the gel, which should have been uniform in its concentration of growth factors. Therefore, these observations argue for the second hypothesis: namely, selected glioblastoma cells with higher Rac1 and Cdc42 activities invaded the gel, guiding those with lower Rac1 and Cdc42 activities. Under higher magnification, the activities of Rac1, Cdc42 and RhoA were all higher in the pseudopodia than in the cell bodies (Fig. 3P–R, respectively) as reported in conventional 2D culture dishes (Kurokawa and Matsuda, 2005). Thus, the failure to detect the intercellular divergence of RhoA activity was not due to the insensitivity of Raichu–RhoA under the 3D condition.

To verify the involvement of Rac1 and RhoA in formation of pseudopodia and invasion of gels, we performed a spheroid invasion assay in the presence of inhibitors of the Rac1 or RhoA pathway. NSC23766, which inhibits Rac1 activation, collapsed the pseudopodia and inhibited invasion of the gels (Fig. 3S, middle panel; supplementary material Movie 4). By contrast, Y27632, an inhibitor of the RhoA-dependent kinase ROCK, promoted the elongation of pseudopodia, which were thinner than those of the control cells. Notably, these pseudopodia could not accelerate the invasion (Fig. 3S, lower panel; supplementary material Movie 4). We examined the effect of NSC23766 and Y27632 on the global activities of RhoA, Rac1 and Cdc42 by pull-down analysis (supplementary material Fig. S5A). None of the Rho-family GTPases was inhibited by Y27632. NSC23766 inhibited only Rac1 activity, as expected. We also analyzed Rac1, Cdc42 and RhoA activities in the spheroid invasion assay in the presence of these inhibitors (supplementary material Fig. S5B). NSC23766 abolished glioblastoma invasion, concomitant with the global decrease in Rac1 activity. In the presence of NSC23766, the leading cells with high Cdc42 activity were lost; however, the Cdc42 activity was not altered in the central cell clusters. Y27632 did not affect the activities of RhoA, Rac1 or Cdc42. All these findings suggested that glioblastoma cells with higher Rac1 activity formed multiple pseudopodia and invaded the gel, guiding the other cells with lower Rac1 activity. The role of the RhoA–ROCK pathway was not clear in this assay.

We next looked for a guanine nucleotide exchange factor (GEF) that is responsible for the high Rac1 and Cdc42 activity in the invading glioblastoma cells with multiple pseudopodia. In preliminary shotgun transcriptome sequencing and microarray analyses, we identified GEFs for Rac1 and/or Cdc42 expressed in both C6 glioblastoma cells and human glioblastoma tissues. Among them, we focused on Zizimin1. Zizimin1 was originally identified as a Cdc42-specific GEF (Meller et al., 2002) and has been shown to regulate neurite outgrowth (Kuramoto et al., 2009). Because Cdc42 controls the formation of astrocytic processes and cell polarity in migrating astrocytes (Etienne-Manneville and Hall, 2001), and could activate Rac1 in a hierarchical manner (Nobes and Hall, 1995), Zizimin1 could cause the heterogeneity in Rac1 and/or Cdc42 activity and, thereby, the two glioblastoma invasion modes.

To investigate the possibility that Zizimin1 contributes to glioblastoma invasion, we developed C6 glioblastoma cells stably expressing three different shRNAs against rat Zizimin mRNA (Zizimin1 knockdown cells: Zizimin-KD-A, -B and -C) or control shRNA (Fig. 4A,B). A spheroid invasion assay revealed that Zizimin1 knockdown greatly inhibited the formation of pseudopodia and invasion of C6 glioblastoma cells (Fig. 4C; supplementary material Fig. S6 and Movie 5). Thus, we speculated that glioblastoma cells with high expression and/or activity of Zizimin1 exhibited high Rac1 and Cdc42 activities and guided the other glioblastoma cells. To test this possibility, C6 glioblastoma cells expressing shRNA and GFP were mixed with wild-type C6 glioblastoma cells expressing a red fluorescent protein, dKeima, to form spheroids, and embedded in 3D Matrigel. The C6 glioblastoma cells expressing the control shRNA and the wild-type C6 glioblastoma cells (green and red cells, respectively, in Fig. 4D and supplementary material Fig. S7 and Movie 6) were equally capable of forming multiple pseudopodia and invading the gel. By contrast, Zizimin1 knockdown cells (green in Fig. 4E and supplementary material Fig. S7 and Movie 6) could not invade the gels at the front, but only trailed the preceding wild-type C6 glioblastoma cells (red in the same figures and movies).

We quantified the proportion of the control and Zizimin1 knockdown cells in the leading cells in the spheroid assay (Fig. 4F). The C6 glioblastoma cells expressing the control shRNA accounted for 50% of leading cells, whereas Zizimin1 knockdown cells accounted for approximately 20%. The velocities of wild-type and Zizimin1 knockdown cells in the spheroid invasion assay were also measured for up to 16 hours (Fig. 4G,H). The wild-type C6 glioblastoma cells moved into the gels faster than the Zizimin1 knockdown cells in the first 8 hours (Fig. 4G; supplementary material Movie 6); however, Zizimin1 knockdown cells were able to move into the gels by trailing the leading cells in the second 8 hours (Fig. 4H; supplementary material Movie 6). We examined whether the effect of shRNA could be cancelled by the expression of Zizimin1. The C6 glioblastoma cells stably expressing EGFP and shRNA against rat Zizimin1 were transfected with expression vectors of human Zizimin1 and histone-H1-conjugated mCherry, and subjected to the spheroid invasion assay (supplementary material Fig. S8). We found that 60% of the leading cells expressed the marker protein, mCherry, indicating that the effect of shRNA could be cancelled by the expression of human Zizimin1.

To further verify the role of Zizimin1 in the formation of pseudopodia in 3D gels, we developed C6 glioblastoma cells stably expressing the Lyn N-terminal sequence-tagged fragment of the mammalian target of rapamycin (LDR) and FK506 binding protein (FKBP)-fused Zizimin1 DHR2 domain (Inoue et al., 2005; Aoki et al., 2007). Rapamycin-induced membrane translocation of the DHR2 domain of Zizimin1 rapidly induced pseudopodia in C6 glioblastoma cells (Fig. 4I; supplementary material Movie 7), showing that Zizimin1 could play a pivotal role in the formation of pseudopodia through its GEF activity.

To examine whether Zizimin1 knockdown suppresses Rac1 and Cdc42 activities of C6 glioblastoma cells in the 3D spheroid invasion assay, we introduced control or Zizimin-KD-A shRNA into C6 glioblastoma cells stably expressing Raichu biosensors, and the Zizimin1 knockdown cells were marked with histone-H1-conjugated mCherry (supplementary material Fig. S9). The Zizimin1 knockdown cells were left behind and exhibited lower Rac1 and Cdc42 activities than the control cells leading the invasion. These findings suggest that Zizimin1 promotes formation of pseudopodia and concomitant invasion of C6 glioblastoma cells through Cdc42 activation in the 3D Matrigel condition.

Next, to investigate the role of Zizimin1 in glioblastoma progression in vivo, we inoculated rat brains with control or Zizimin1 knockdown glioblastoma cells. Although the Zizimin1 knockdown cells replicated as efficiently as the control cells in the culture dish (Fig. 4J), the Zizimin1 knockdown cells ostensibly grew more slowly than the control cells in brain tissues (Fig. 4K). Histologically, the control glioblastoma cells diffusely invaded the brain tissue from the perivascular region (Fig. 4L,M), whereas Zizimin1 knockdown glioblastoma cells were well demarcated from the surrounding brain tissues (Fig. 4N) and packed densely around the perivascular region (Fig. 4O). These results suggested that formation of pseudopodia by Zizimin1 was required for C6 glioblastoma cell progression in vivo, and that Zizimin1 drove glioblastoma cells around the perivascular region to invade the brain parenchyma.

To investigate whether human glioblastoma cells also exhibited a gradient of Rho-family GTPase activity during invasion of the gel, we introduced Raichu biosensors into three human glioblastoma cell lines, U251MG, LN229 and U87MG, and performed the 3D spheroid invasion assay (Fig. 5A–C). FRET microscopy revealed that cells leading the invasion exhibited higher Rac1 and Cdc42 activity than the trailing cells in all human glioblastoma cell lines. However, the intercellular gradient for RhoA activity varied depending on the cell type. The leading U251MG cells had slightly higher RhoA activity than the trailing cells (Fig. 5A); the gradient was not clear in LN229 cells (Fig. 5B); and in U87MG cells the RhoA activity of the leading cells was lower than that of the trailing cells (Fig. 5C). These findings suggest that the human glioblastoma cells with higher Rac1 and Cdc42 activity guide the other human glioblastoma cells with lower Rac1 and Cdc42 activity, as in rat C6 glioblastoma cells, and that RhoA activity might play different roles depending on the glioblastoma cells.

Next, to investigate whether Zizimin1 played an important role in human glioblastoma cell invasion as in rat C6 glioblastoma cells, we established human glioblastoma cells stably-expressing shRNA against human Zizimin1 mRNA (Fig. 5D,E). Knockdown of Zizimin1 in U251MG cells did not have any effect in the 3D spheroid invasion assay (Fig. 5F, upper panels). However, knockdown of Zizimin1 in LN229 and U87MG cells impeded formation of pseudopodia and invasion of the gel (Fig. 5G,H, respectively, upper panels). To reconstitute in vivo human glioblastoma tissue consisting of cells with various expression and/or activity levels of Zizimin1, glioblastoma cells with or without Zizimin1 shRNA were mixed to form spheroids, and a 3D invasion assay was performed. We found that Zizimin1 knockdown LN229 or U87MG cells could only follow the wild-type glioblastoma cells into the gels (Fig. 5G,H, respectively, lower panels). Again, we could not find any effect of Zizimin1 shRNA in U251MG cells (Fig. 5F, lower panels). These findings indicate that Zizimin1 also plays a pivotal role in formation of pseudopodia and invasion in some human glioblastoma cells, as in rat C6 glioblastoma cells.

Previous histological studies have shown that C6 glioblastoma cells invade through the perivascular space and along neuronal fibers of the brain parenchyma, as do the genuine human glioblastoma cells (Grobben et al., 2002). Ex vivo time-lapse images taken with conventional fluorescence microscopy have further highlighted the importance of the perivascular growth of C6 glioblastoma cells (Farin et al., 2006). Recently, Winkler et al. used a two-photon microscope to observe the invasion of mouse glioblastoma cells and reported that the perivascular glioblastoma cells moved faster than did the glioblastoma cells in the parenchyma (Winkler et al., 2009). We also confirmed these findings for C6 glioblastoma cells by using a two-photon microscope, and further demonstrated that glioblastoma cells exhibited two distinct modes of invasion depending on whether they were located in the perivascular region or parenchyma (Fig. 1). Glioblastoma cells migrating in the perivascular region had an elongated spindle shape that is consistent with mesenchymal migration (Friedl and Wolf, 2010). By contrast, most of the glioblastoma cells in the brain parenchyma resembled normal astrocytes. Interestingly, time-lapse microscopy revealed that some glioblastoma cells dynamically changed their invasion paths from the perivascular region to the intraparenchymal region, and vice versa. These findings motivated us to examine the mechanistic aspect of the switch between the two invasive modes of glioblastoma, because flexible alteration of the two invasion modes in individual glioblastoma cells could lead to effective progression of glioblastomas.

FRET imaging demonstrated that glioblastoma cells extending multiple pseudopodia at the invasion front exhibited higher Rac1 and Cdc42 activities and lower RhoA activity compared with glioblastoma cells at the perivascular region (Fig. 2). To the best of our knowledge, this is the first report to show the intercellular activity gradient of small GTPases in living cancer tissues, although previous studies using inhibitors and mutants have suggested that the differential use of Rho-family GTPases determines the mode of cancer cell invasion (Sahai and Marshall, 2003; Sanz-Moreno et al., 2008; Yamazaki et al., 2009; Friedl and Wolf, 2010). The gradient of Rho-family GTPase activity within the glioblastoma tissue strongly supports the idea that the two glioblastoma invasion modes are correlated with the activities of Rho-family GTPases. Interestingly, recent studies by Panopoulos et al. showed that mesenchymal migration of glioblastoma cells does not require Rac1 activity on 2D substrates (Panopoulos et al., 2011). This observation suggests that glioblastoma cells do not require Rac1 activity to move without marked resistance on dishes or around blood vessels. At the same time, our observations suggested two scenarios for the increased Rac1 and Cdc42 activities: high activities of Rac1 and Cdc42 might be induced within cells invading the brain parenchyma, or cells with high Rac1 and Cdc42 activities might be driven from the perivascular space into the parenchyma.

Three lines of evidence from the 3D invasion assay support the second scenario, i.e. C6 glioblastoma cells with high Rac1 and Cdc42 activities preferentially invade the parenchyma. First, we found that C6 glioblastoma cells with high Rac1 and Cdc42 activities guided the other cells to invade the gel (Fig. 3; supplementary material Fig. S4). Second, inhibition of Rac1 activity by NSC23766 reduced the size and number of pseudopodia and suppressed the invasion (Fig. 3S; supplementary material Movie 4). We confirmed this observation with C6 glioblastoma cells expressing a dominant-negative mutant of Rac1 (data not shown). Third, Zizimin1 knockdown decreased the Rac1 and Cdc42 activities, reduced the number of cellular protrusions in C6 glioblastoma cells and suppressed the invasion in 3D Matrigel cultures (Fig. 4; supplementary material Figs S7 and S9). Intriguingly, cells with reduced Zizimin1 expression could follow the wild-type C6 glioblastoma cells in 3D culture, indicating that high Cdc42 activity is required for the invasion of the gel but not for the migration itself (Fig. 4G,H; supplementary material Movie 6).

We previously reported that Rac1 and Cdc42 activities are high in lamellipodia (Itoh et al., 2002). Thus, cells with multiple pseudopodia could contribute to high Rac1 and Cdc42 activities in cells at the invasion front. However, as shown in the magnified FRET images (Fig. 3), the leading cells exhibited high Rac1 and Cdc42 activities not only in the pseudopodia but also in the cell bodies, in comparison with the trailing cells. Thus, we concluded that the high Rac1 and Cdc42 activities in the leading cells were not caused by the presence of multiple pseudopodia.

The relevance of our finding to the development of human glioblastoma is partially validated by the use of human glioblastoma cell lines. We found that invasion-leading cells exhibited higher Rac1 and Cdc42 activity than the trailing cells in human glioblastoma cells (Fig. 5A–C), as in rat C6 glioblastoma cells, and that knockdown of Zizimin1 in LN229 and U87MG cells impeded cell invasion of the gel (Fig. 5G,H). Because knockdown of Zizimin1 in U251MG cells did not alter the invasive phenotype, we consider that the contribution of Zizimin1 to glioblastoma cell invasion might be cell-type dependent. In fact, C6, LN229 and U87MG cells expressing shRNA against Zizimin1 were found to still be capable of invading the gel when examined for longer times (data not shown), suggesting that there are other molecules cooperating with or substituting for Zizimin1. Notably, using fluorescence-based cell sorting analysis and a pull-down assay (data not shown), knockdown of Zizimin1 was not found to affect the activities of Rac1 and Cdc42 in the C6 glioblastoma cells grown on culture dishes. This observation strongly suggests that the role played by Zizimin1 is specific to the glioblastoma cells growing in 3D conditions. In addition, although Zizimin1 is highly expressed in brain (Meller et al., 2002), we found that it is not upregulated in human glioblastoma cells compared with normal brain tissue (data not shown). To the best of our knowledge, there is no previous report showing a positive or negative correlation between the Zizimin1 mRNA expression level and cell proliferation, cell invasion or the prognosis of glioblastoma patients (Phillips et al., 2006; Shirahata et al., 2007). Therefore, we could speculate that Zizimin1 might be regulated primarily by the protein activity rather than by the gene expression in glioblastoma cells and that a glioblastoma cell population should contain a small fraction of cells with high Zizimin1 activity that is highly invasive in brain parenchyma. Actually, in our observations of rat brain, only a small number of glioblastoma cells produced multiple pseudopodia and penetrated brain parenchyma (Fig. 1), as was reported in a previous study in which only a small fraction (~5%) of cancer cells showed high invasiveness in vivo (Giampieri et al., 2009).

In summary, we demonstrated that invasion modes of glioblastoma cells can be controlled by the balance of the activity of Rho-family GTPases and a Cdc42-specific GEF, and Zizimin1 appears to play an important role in the formation of multiple pseudopodia and in invasion of the brain parenchyma. Our results provide further information about the nature of heterogeneity in cancer cell populations that are composed of various cells with different fingerprints of invasiveness.

C6 rat glioblastoma cells, U251MG, LN229 and U87MG human glioblastoma cells, and Cos7 African green monkey SV40-transformed kidney fibroblast cells were obtained from the American Type Culture Collection and cultured in DMEM containing 10% FBS. The prototype FRET biosensors for Rac1, RhoA and Cdc42 were described previously (Itoh et al., 2002; Yoshizaki et al., 2003). In the Raichu–Rac1 biosensor (2227x), Raichu–Cdc42 biosensor (2219x) and Raichu–RhoA biosensor (1523x) used in this study, CFP was replaced with the teal fluorescent protein as described previously (Yoshiki et al., 2010). A pCX4 retroviral vector was used to express the FRET biosensors and fluorescent proteins in each glioblastoma cell line as described previously (Takaya et al., 2007). The infected cells were single-cell cloned before further experiments unless described otherwise.

All animal care measures and experiments complied with Japanese community standards on the care and use of laboratory animals, which were approved by Kyoto University. C6 glioblastoma cells (5×10⁵ cells/5 μl PBS) were stereotactically transplanted into the right subcortex of 3-week-old male Wistar rats. After 7 days, tumor-bearing rats were killed with carbon dioxide and sodium pentobarbital, and decapitated. In preliminary experiments, Texas-Red-conjugated dextran (70 kDa; Invitrogen, Carlsbad, CA) was injected into the left ventricle before decapitation to visualize blood vessels; however, we omitted this process in later experiments because we could easily distinguish blood vessels by the morphologies and distribution of cells at invasion borders. The whole brains were quickly removed, cut vertically and mounted on the stage of a Vibroslice Tissue Cutter (Campden Instruments, Loughborough, UK) filled with ice-cold artificial cerebrospinal fluid containing 135 mM N-methyl-D-glucamine, 1 mM KCl, 1.2 mM KH₂PO₄, 20 mM choline bicarbonate, 10 mM glucose, 1.5 mM MgCl₂ and 0.5 mM CaCl₂. Coronal brain slices (300 μm thick) were cut and transferred onto a plastic coverslip and incubated in slice culture medium at 37°C in a humidified atmosphere containing 5% CO₂ as described previously (Tanaka et al., 2008). The slice culture medium consisted of 77% MEM, 20% HBSS, 3% HEPES, 6.5 mg/ml glucose, 6.5 mg/ml L-glutamine and N2 supplement (all from Invitrogen). For tumor size calculation, brain samples were prepared by perfusion fixation. A coronal section at the injection site was prepared for each sample and imaged under a fluorescence stereomicroscope (Carl Zeiss Meditec Inc., Dublin, CA).

Cultured brain slices were maintained in an incubation chamber (Tokai Hit, Shizuoka, JAPAN) and imaged using a 20× water-immersion objective (XLUMPLFL20; Olympus Optical Co., Tokyo, Japan) on a BX61WI/FV1000 upright microscope (Olympus) equipped with a 440 nm laser diode (Olympus) and a Mai-Tai Ti:sapphire laser (Spectra Physics, Mountain View, CA). The excitation wavelength for two-photon excitation was 850 nm. We used a RDM650 IR-cut filter, a DM570 dichroic mirror and two emission filters: BA510-550 for GFP and BA570-625 for Keima or Texas Red, respectively. All filters were purchased from Olympus. For FRET imaging with a confocal laser scanning unit, we used a 440 nm laser diode and the following filter sets purchased from Olympus: an excitation dichroic mirror, DM405-440/515, a second dichroic mirror, SDM510, and two emission filters, BA465-495 for TFP and BA520-550 for FRET. To obtain the intensity profile of the Raichu biosensors, we used the FV1000 lambda-scan program with a 440 nm laser diode and a beam splitter BS/20/80 (Olympus), and an SPD (Olympus) emission detecting system. Fluorescence intensities at every 5 nm wavelength with 10 nm bandwidth were obtained from 470 nm to 550 nm.

Acquired images were analyzed with MetaMorph software (Universal Imaging, West Chester, PA). For 3D reconstruction of the images and calculation of the cell migration speed in a 3D environment, the original confocal images were analyzed with Imaris Software (Bitplane AG, Zürich, Switzerland). The tumor area was calculated according to the fluorescence signal with MetaMorph software.

The organotypic culture system was set up as previously described (Gaggioli et al., 2007). Briefly, cells were embedded in Matrigel (BD Biosciences, Bedford, MA) at a concentration of approximately 6 mg/ml. For spheroid invasion assays, 1×10⁶ cells in 1 ml serum-free CO₂-independent medium (Invitrogen) were agitated overnight in a 12-well plate coated with poly(2-hydroxyethyl methacrylate) (Sigma, St. Louis, MO) to form small aggregates. The aggregates were embedded in Matrigel, maintained in complete medium and observed under a two-photon or a confocal microscope for up to 18 hours in an incubation chamber. Y27632 and NSC23766 were obtained from Calbiochem (La Jolla, CA) and used at final concentrations of 20 μM and 100 μM, respectively.

pSuper.retro.puro vector (OligoEngine, Seattle, WA) was used for the expression of short hairpin RNA (shRNA). shRNAs for rat Zizimin1 were kind gifts from Hironori Katoh, Graduate School of Biostudies, Kyoto University (Kuramoto et al., 2009). An shRNA for human Zizimin1 (5′-TGGTTCCGGTTTAGATAGCTA-3′) was obtained from the Genome Network Project at the University of Tokyo (Hirata et al., 2009). GAPDH was used as an internal control. The following primer sets were used: for rat Zizimin1, forward 5′-CATGCAGGATGTCCATTTCA-3′, and reverse 5′-CGTGTCATACAGGTGGGCTA-3′; for rat GAPDH, forward 5′-GAGTCTACTGGCGTCTTCAC-3′, and reverse 5′-GTTCACACCCATCACAAACA-3′; for human Zizimin1, forward 5′-TCCGTGTAGTGTGCAACCAT-3′, and reverse 5′-CAGTCCCACCAAGAAGTGGT-3′; for human GAPDH, forward 5′-GAGTCCACTGGCGTCTTCAC-3′, and reverse 5′-GTTCACACCCATGACGAACA-3′. The conditions for PCR were 45 cycles of 95°C for 10 seconds for denaturation, 56°C for 10 seconds for annealing and 72°C for 10 seconds for extension.

The activities of Rac1, Cdc42 and RhoA in C6 glioblastoma cells were measured by the Bos' pull-down method (Aoki et al., 2005). Briefly, cells were harvested in ice-cold lysis buffer (50 mM Tris, pH 7.5, 100 mM NaCl, 2 mM MgCl₂, 1% Nonidet P-40, 10% glycerol, 1 mM dithiothreitol) containing GST–PAK–CRIB for Rac1 and Cdc42 or Rhotekin–RBD for RhoA. The cleared lysates were incubated with glutathione–Sepharose beads for 30 minutes at 4°C. The washed beads were boiled in sample buffer, and both the bound proteins and total cell lysates were analyzed by SDS-PAGE followed by immunoblotting. Precast SDS–polyacrylamide gels were purchased from Wako Pure Chemical Industries (Tokyo, Japan). Bound antibodies were detected with secondary antibodies conjugated with IRDye680 or IRDye800 and analyzed with an Odyssey Imager system (LICOR, Lincoln, NE). Anti-Rac1 and anti-Cdc42 mouse monoclonal antibodies were purchased from BD Transduction Laboratory (San Diego, CA). Anti-RhoA mouse monoclonal antibody was purchased from Santa Cruz Biotechnology (Santa Cruz, CA). Anti-human DOCK9 mouse monoclonal antibody was a kind gift from Martin A. Schwartz (Yale University) and anti α-tubulin mouse monoclonal antibody was purchased from Calbiochem.

When two groups were compared, a two-tailed unpaired Student's t-test was applied.

We thank H. Katoh, A. Miyawaki, T. Akagi, J. Miyazaki and M. A. Schwartz for the plasmids and antibodies. Y. Inaoka, K. Hirano, R. Sakai and N. Nonaka are also to be thanked for their technical assistance. We are grateful to the members of the Matsuda Laboratory for their helpful discussions.