Glioblastoma: A Pathogenic Crosstalk between Tumor Cells and Pericytes

Cancers likely originate in progenitor zones containing stem cells and perivascular stromal cells. Much evidence suggests stromal cells play a central role in tumor initiation and progression. Brain perivascular cells (pericytes) are contractile and function normally to regulate vessel tone and morphology, have stem cell properties, are interconvertible with macrophages and are involved in new vessel formation during angiogenesis. Nevertheless, how pericytes contribute to brain tumor infiltration is not known. In this study we have investigated the underlying mechanism by which the most lethal brain cancer, Glioblastoma Multiforme (GBM) interacts with pre-existing blood vessels (co-option) to promote tumor initiation and progression. Here, using mouse xenografts and laminin-coated silicone substrates, we show that GBM malignancy proceeds via specific and previously unknown interactions of tumor cells with brain pericytes. Two-photon and confocal live imaging revealed that GBM cells employ novel, Cdc42-dependent and actin-based cytoplasmic extensions, that we call flectopodia, to modify the normal contractile activity of pericytes. This results in the co-option of modified pre-existing blood vessels that support the expansion of the tumor margin. Furthermore, our data provide evidence for GBM cell/pericyte fusion-hybrids, some of which are located on abnormally constricted vessels ahead of the tumor and linked to tumor-promoting hypoxia. Remarkably, inhibiting Cdc42 function impairs vessel co-option and converts pericytes to a phagocytic/macrophage-like phenotype, thus favoring an innate immune response against the tumor. Our work, therefore, identifies for the first time a key GBM contact-dependent interaction that switches pericyte function from tumor-suppressor to tumor-promoter, indicating that GBM may harbor the seeds of its own destruction. These data support the development of therapeutic strategies directed against co-option (preventing incorporation and modification of pre-existing blood vessels), possibly in combination with anti-angiogenesis (blocking new vessel formation), which could lead to improved vascular targeting not only in Glioblastoma but also for other cancers.


Introduction
Glioblastoma Multiforme (GBM) is a highly invasive brain cancer, with prominent vascular involvement, characterized by twisted blood vessel [1] and infiltration along external vessel walls [2], which makes it resistant to treatment. Evidence from a rat GBM model has shown that early tumor vasculature forms by cooption of pre-existing brain blood vessels and precedes new vessel formation (angiogenesis) [3]. Vessel co-option also occurs during metastasis of other tumors, as recently demonstrated for the spread of breast cancer into the brain [4]. Furthermore, co-option is also responsible for tumor recurrence and metastasis following antiangiogenic therapies, both in GBM and in other types of cancer [5]- [8]. Therefore, vessel co-option is likely to be a principle cause of malignancy, which occurs during tumor initiation/progression, metastasis and re-initiation after treatment. However, in contrast to angiogenesis that is well understood, the cellular and molecular bases of vessel co-option in tumors are currently unknown. The normal brain microvasculature is made up of narrow tubes (capillaries), consisting of endothelial cells surrounded by contractile pericytes, which function normally to regulate vessel tone and morphology [9], [10]. Because pericytes are located on the abluminal wall of blood vessels, they are good candidates for a role in mediating vessel co-option by tumor cells. Brain pericytes are pluripotential cells with stem cell properties [11]- [13], similar if not identical to the mesenchymal stem cells that occupy an equivalent perivascular location in bone marrow. There is a growing realization that, in addition to their critical role in maintaining blood vessel integrity and controlling blood flow, pericytes are also key players in other aspects of brain homeostasis and disease. For example, evidence suggests that they are regulators of innate immunity and, depending on the context, can mediate not only pro-inflammatory functions associated with host defense [14], but also the anti-inflammatory response to malignant tumors such as human GBM, which includes the inhibition of T cell function and local immunosuppression [15]. Consistent with a role in normal cerebral immunity, purified brain pericytes have been shown to be interconvertible with macro-phages [16] and to behave as macrophage-like cells in culture, by phagocytosing plastic beads [17] and by secreting inflammatory cytokines such as IL-1b, TNF-a and IL-6. Moreover, pericytes play an additional role in maintaining a proper function of the brain-immune interface, by controlling the migration of leukocytes in response to inflammatory mediators [18]. Given that immune cells contribute to tumor progression [19], pericytes could therefore provide a critical node for local control of both vessel co-option and immune system modulation. Within established tumors, blood vessels are often dysmorphic, with abnormal pericyte coverage and either atypical or absent endothelium [20]. Recent research, aimed to understand the possible function of pericytes in tumor progression, has emphasized their role in new vessel formation during angiogenesis [21]. In co-culture experiments, pericytes have been shown to modulate the angiogenic response of endothelial cells to glioma cells [22]. Furthermore, the recent discovery that GBM stem cells can trans-differentiate into tumor pericytes during the process of angiogenesis [23] further emphasizes the contribution of perivascular cells to tumor growth. While these findings underline the role of pericytes in the establishment of new vessels, very little is known about pericyte function in tumor infiltration. It is now recognized, for example, that perivascular tumor invasion occurs in some types of cancers [24]. Recent discoveries in breast and colon carcinomas have proposed that paracrine crosstalk between tumor and stromal cells is able to promote tumor growth and motility [25], [26]. Nevertheless, up to now no information exists about the cell biology of tumor cell/pericyte interaction, either in GBM or in cancer in general.
Here we use mouse xenografts in combination with live imaging techniques of brain explants and GBM-cell/pericyte co-cultures on laminin-coated deformable silicone substrates to investigate the fundamental cellular mechanisms by which GBM cells exploit the surrounding vascular niche to promote tumor survival and invasion.

Results
First, we challenged human GBM cells (U87) with mouse brain slices where blood vessels were pre-labeled with black ink ( Figure 1A, B). This assay revealed a remarkable ability of GBM cells to pull pre-existing blood vessels into the graft. Vessel cooption happens quickly (within 15 hours) with conversion of normal capillaries into highly twisted structures, demonstrating a capacity for rapid vascular network acquisition, even in the absence of new vessel formation. Next, we used GFP-actin labeling and 2-photon live imaging to identify the intrinsic GBM cellular mechanisms involved in vessel-co-option ( Figure S1; Figure 1C-F). We found that, 6 hours after seeding, tumor cells in contact with blood vessels can convert straight segments (up to ,60-70 mm long) into hairpin bends ( Figure 1C and Movie S1). Vessel re-organization is linked to the appearance of tumor cell protrusions, that we called 'flectopodia' (flectere, to bend, and podós, foot), characterized by an unusual, discontinuous (moniliform) organization of actin, with highly dynamic beads (0.6-3.0 mm in diameter), moving bidirectionally. Flectopodia are 7-30 mm long, present in 9% of GFP-actin + -cells on vessels (n = 266), with the ability to elongate for up to 20 mm or more, at a rate of 14-99 mm/hour, and to form a bent vessel segment in 30 minutes. A gap of approximately 10 mm separates the GFP-actin labeling from the vessel lumen, suggesting that flectopodia are in close contact with perivascular cells. Flectopodia-associated bending often involves a pair of GFP-actin labeled cells, cooperating via a cytoplasmic bridge. During the process, while the leading cell performs the re-arrangement, the lagging cell translocates the bent segment ahead of the first cell ( Figure 1D-F; Movie S1). This suggests that flectopodia mediate the ongoing ratchet-like recruitment of host vessels at the co-option front of the tumor.
Next, to validate and extend our ex-vivo observations, we established a xenograft model that recapitulates human GBM in mice ( Figure 1G) and used it to confirm the presence of flectopodia and vessel co-option in situ. High resolution-microscopy of sections from GFP-actin-labeled xenografts identified flectopodia-like extensions as early as 2 days ( Figure 1H). In situ hybridization at 7 days for the activated mouse (m)-pericyte marker Rgs5, a gene controlling tumor vasculature remodeling [27], [28], showed the presence of m-Rgs5 + -cells surrounding abnormally dilated vessels throughout the graft ( Figure 1I-K). m-Rgs5 + -pericytes are also detected in the infiltrating margin of 1month xenografts ( Figure 1L-N). Taken together these data strongly suggest that host brain perivascular cells are a target cell type for GBM vessel co-option and modification throughout tumor progression.
To confirm that GBM cells interact with pericytes, GFP-actin labeled U373 and U87 cells were co-cultured with brain slices prelabeled in situ for pericytes, using either a DsRed-transgene reporter for the pericyte marker NG2 [9] or a fluorescent Dextran tracer (Methods). Our results show that tumor cells on vessels interact with Dextran (Dextran Labeled Pericytes, DLPs) or NG2-DsRed labeled perivascular cells (Figure 2A, B; Figure S2 A-G). Unexpectedly, we also found tumor-derived cytoplasm within the cortex of the target pericyte ( Figure 2B and Figure S2 G), implying a role for cytoplasmic transfer in the co-option process (see also Movie S2).
Our identification of pericytes as a specific GBM cell target raised the possibility that the altered blood vessel morphology in tumors could be caused by deregulated pericyte contraction. To test this, we established an in vitro system using isolated mouse brain pericytes (Methods) cultured on deformable silicone substrates [29], with the novel coating of human-laminin to reproduce the blood vessel basal lamina that houses perivascular cells in vivo ( Figure 2C). Pericytes in vitro express NG2 and Figure 1. GBM cells co-opt and modify blood vessels in-vivo. A, Scheme showing seeding of human-GBM cells (grey spots) onto mouse brain slices. B, White arrowheads point to abnormal blood vessels (black-Ink), co-opted by fluorescence-labeled cells (MiRu + , red). Asterisks indicate agglomerated co-opted vessels in the body of the graft. C, Maximum projection (top) and 4D-reconstruction (bottom)-video frames (respectively) showing GFP-actin human GBM cells (green) re-arranging blood vessels. Frames have been selected to visualize flectopodia with GFP-actin-beads (green, white arrows) bending (yellow arrowheads) a previously straight vessel (DiI-red; blue arrowheads). D-F, A highly schematic cartoon of vessel structure before (D), during (E) and after (F) flectopodia-mediated co-option (based on Movie S1). 1 and 2, co-operating tumor cells (green), linked by cytoplasmic bridge; dashed-lines, recruited/modified vessel segment; black and blue arcs, which show the advance of GBM cells on the vessel, are analogous to the expanding tumor margin. G, Scheme of GBM-hanging drop xenografts. H, GFP-actin-U373 cell (green) in striatum of 2 dayxenograft, contacting host vessel (DiI-red) through a flectopodium (arrow) with moniliform-actin (white arrowhead). I, arrowheads point to Ink-filled, dilated vessels in 7 day-U87 graft. Activated perivascular cells (ap, Rgs5 + , blue) are visible on co-opted, modified vessels (cv, black-Ink) at the expanding edge (red dashed-outlines) of both 7day-(J-K) and 1 month-(M-N) U87-xenografts. Note the difference in diameter between normal (black arrowhead) and co-opted (red arrowhead) vessels in K. Scale bars: 30 mm (B), 10 mm (C, H), 200 mm (J, L), 40 mm (K), 25 mm (N). doi:10.1371/journal.pone.0101402.g001 Figure 2. GBM cells target pericytes and modify their contractility. GFP-actin-GBM cells (green) contacting an NG2-DsRed + -pericyte (A, red iso-surface in magnified box) and a DLP (B, blue; confocal section) through flectopodia (arrowheads indicate moniliform actin). Note the presence of GFP-actin within the DLP (merged channels, inset in B). v, vessels (DiD-blue in A; Ink-filled-grey in B). C, Scheme showing pericyte (colored cells) invivo (top; BV, blood vessel), in-vitro (middle) and on silicone-substrate (bottom). Wrinkling is associated with high aSMA-expression (red-color). D and E (boxed area in G), DIC-optic video-frames of the same field before (D) and after (E) GBM cell addition to pre-plated pericytes. Pericytes alone display attributes consistent with stem (CD44; Vimentin; Nestin), contractile (a-smooth muscle actin, aSMA) and macrophage (CD68 and phagocytosis) potential ( Figure S2 H). Two days after plating, cells generate compression forces (indicative of vasoconstriction activity [29]) visible, using Differential Interference Contrast (DIC) imaging, as wrinkles in the silicone sheet ( Figure 2D). These are organized around local nodes of higher contractility (Movie S3), correlated with the expression of aSMA protein, a key determinant of pericyte contraction ( Figure S2 I-K'). Remarkably, aSMA is also enriched in DLPs in vivo, suggesting that they may represent strategic nodes for the regulation of brain vessel tone ( Figure  We then investigated if pericyte wrinkling-activity was affected by GBM cell addition. Although neither U373 nor U87 cells alone deform the substrate, in co-cultures they induce nearby pericytes to generate both new wrinkles and destabilize pre-existing ones (Figure 2E-G; Movie S4; Figure S2 P-P'). Quantification of the contractile activity with and without GBM cells, by tracking the behavior of identified wrinkles ( Figure S2 O'', P''), revealed a difference in the way the wrinkles move in space and time (E2-E1, P = 0.02, where E1 and E2 are defined as track-straightness of each wrinkle end, Methods). Moreover and interestingly, including the track-straightness of the wrinkle center (C) and plotting all the wrinkle data in a 3D-scatter graph ( Figure 2I), showed that GBM cells abolish the formation of wrinkles that drift laterally or drift laterally and pivot, typical of pericytes in nodes and inter-nodes (point clusters 1 and 2, respectively), leaving only the less organized activity characteristic of anti-nodes (point cluster 3; Movie S3). In conclusion, therefore, our data provide strong evidence that tumor cells can corrupt the intrinsic contractility of brain pericytes.
Next, we investigated the cellular mechanisms employed by GBM cells to induce pericyte dysfunction. Live imaging of U373 and U87 GBM cell/pericyte co-cultures on silicone-laminin substrates identified long extensions (maximum length 81632 mm) ( Figure 3A; Figure S2 P, P'), characterized by a discontinuous distribution of both cytoplasmic varicosities and GFP-actin ( Figure 3B, C). Moreover, co-transfection assays in vitro showed that the small GTPase Cdc42, a principle regulator of cell polarity and actin cytoskeletal organization [30], is co-localized with the actin beads within cytoplasmic varicosities ( Figure S3 A) and the native protein is visible within the GBM cell extensions on silicone substrate ( Figure 3D). Importantly, analysis of human CD44, a cancer-associated cell surface adhesion molecule [31], revealed that the edges and tips of GBM cell elongations contact the target pericyte ( Figure 3E, F). Remarkably, tumor cell projections predict the locations where the wrinkling pattern is changing ( Figure S4 A-C''; Movie S5 and Figure 3C). Taken together, these converging lines of evidence strongly suggest that the cellular extensions seen on silicone substrates are similar, if not identical, to the flectopodia described above (Figure 1, 2).
Detailed analysis showed that flectopodia are characterized by alternating phases of extension and retraction ( Figure 3G-L). Surprisingly, we found that, during retraction, cytoplasmic fragments (3-8 mm in diameter), corresponding to varicosities located originally at the advancing tip ( Figure 3I'-K') can be transferred into the target pericyte ( Figure 3H, L and L'; Movie S6; Figure S4 D). These data are compatible with the GBM cellcytoplasmic fragments positive for h-CD44 identified inside the contacted pericyte ( Figure 3E, F) and corroborate cytoplasmic transfer observed in brain slices ( Figure 2B; Figure S2 E, G). Moreover, this is supported by the mixing of GBM cell-derived Cherry-tagged Cdc42 and host DLP cytoplasms in xenografts ( Figure S3 D-G). Taken together, our data indicate that co-option involves pericyte up-take of cytoplasmic micro-domains released by flectopodia.
Unexpectedly, our data suggest that GBM cells can use pericytes also as fusion-partners. First, MiniRuby-labeled (MiRu + ) U373 grafts into unlabeled hosts generate clusters of strongly MiRu + -cells, which express mouse Rgs5 but lack human centromeric (h-cen) DNA ( Figure S5 A, B). Furthermore, grafting Dextran MiRu + -U373 or U87 cells, into mouse brains or slices harboring DLPs, resulted in a range of differentially doublelabeled derivatives ( Figure S5 C-R). Among these, a novel cell type, that we called GDH (GBM cell/DLP Hybrid), is particularly striking due to its intense double labeling and location on highly constricted/dilated vessels, far beyond the tumor margin ( Figure 3M; Figure Figure S6 K-M), indicating that vessel hyper-contractility is linked to oxidative/nitrative stress [32]. GDH cells retain their strategic position even in advanced GBMxenografts ( Figure S6 N-U) and, interestingly, shed Cdc42 +particles in the lumen of dilated vessels in 7-day grafts, which are also found in sinusoidal vessels in 1-month-tumours ( Figure S6 X, Y).
Remarkably, time-lapse confocal imaging on silicone substrates showed that the double-labeled cytoplasm, characteristic of GDH cells, could result from a fusion-like process that leads to the merger of a cell pair. This process, involving the interaction of a round GBM cell with a raised pericyte, occurs in three steps over 2-3 hours ( Figure 3N and Movie S7). A preliminary cell/cell contact-phase (1-hour) is followed by tumor cell-cytoplasmic transfer and coalescence of the cell bodies (1-hour), with final translocation of the presumptive hybrid-derivative. When both GBM cells and pericytes were loaded with different colored Dextrans and combined on glass, we found cohorts of differentially double-labeled progenies ( Figure 3O, P; Figure S7), some with aberrantly sized nuclei, indicative of abnormal ploidy. Timecourse analysis showed that loss of h-CD44 and h-Nestin is already complete 48 h after cytoplasmic mixing (data not shown), produce stable drifting wrinkles (red arrows) that are de-stabilized by GBM cells. White and yellow arrowheads indicate the appearance and disappearance of wrinkles, respectively. Dashed line marks the upper-limit of GBM cell population, transposed from F and G, which show the low magnification of FR Dextran-labeled GBM cells (white false-color in F and magenta in G), plated on cultured pericytes. Time in minutes. H, Traces of two wrinkles, produced before (i) and after (ii) U87-GBM cell-addition, revealing the spatial evolution and colored to indicate lengthening (violet to green) or shortening (green to red) for each time-frame (numbers). I, 3D-plot summarizing the wrinkling behavior of pericytes, either alone (red points, n = 40) or with U87-GBM cells (green points, n = 23). Note the lack of green points in clusters 1 and 2. E1, E2 and C: track-straightness of the ends (E) and center (C) of each wrinkle. Scale bars: 10 mm (A, B), 30 mm (D), 100 mm (G). doi:10.1371/journal.pone.0101402.g002 supporting our cell-hybrid data in mouse xenografts ( Figure S5). Thus, these results identify pericytes, for the first time, as a specific GBM cell-target for the production of fusion-like hybrids, with the potential to generate novel malignant cell variants, such as the hyper-contractile GDH cells, strategically located to maintain a hypoxic penumbra at the invasive edge of the tumor.
Next, considering that flectopodia are actin-based extensions from highly polarized cells, we reasoned that inhibition of the actin GTPase Cdc42 might block vessel co-option. Immunohistochemistry on GBM cells seeded onto brain slices showed that Cdc42 is enriched in flectopodia ( Figure 4A). Reducing Cdc42 in tumor cells, using either siRNA (iCdc42) or the specific Cdc42-inhibitor Secramine-A [33], results in shortened extensions to vessels and in reduced angle of vessel bending in brain slices ( Figure 4B-E). Secramine-A also decreases the likelihood of a bend occurring where a tumor cell is attached to a blood vessel ( Figure S8 A, B). To test Cdc42-function at the tumor/host margin, GBM cellpellets, with and without inhibition for Cdc42, were grafted either separately or adjacent to each other into brain slices ( Figure 4F). Importantly, while wild type-cells pull some vessels into the graft and use others for radial migration, iCdc42-treated cells show little affinity for blood vessel and no co-option ( Figure 4F-H; Figure S8 C, C'). Co-opted vessels present abnormal constrictions, dilations and localized hairpin bends, while vessels adjacent to iCdc42grafts maintain a straight morphology ( Figure 4F1 and F2, respectively).
We then showed that CD44, a GBM marker with fusogenic properties [34], is enriched at vessel contact sites and cooperates with Cdc42 in vessel co-option/modification in brain slices. Our data demonstrated that knocking down CD44 (by shDNA) in combination with Cdc42 increases the inhibitory effect of iCdc42 alone on flectopodia-length and the angle of vessel bending, with a reduction in the number of glomeruloid-like structures by 70% ( Figure S8 D-H). Additionally, iCdc42 shifts tumor cell-phenotype from ensheathing/re-arranging vessels, to a loosely associated state, a tendency amplified when CD44 is also reduced ( Figure S8 I, J). Taken together, these data suggest that Cdc42 and CD44 act synergistically during flectopodia-induced vessel co-option/modification.
Subsequently, we tested the effect of iCdc42-GBM cells on pericyte behavior on silicone/laminin substrates. In addition to the initial pericyte activation induced by wild-type GBM cells ( Figure  S9 A-B'', E-F'', M), confocal live imaging showed a further pericyte-transformation into hyper-activated macrophage/dendritic-like cell phenotype, capable of killing and engulfing iCdc42-treated tumor cells, with concomitant overall reduction in wrinkling activity ( Figure 4I; Movies S8, S9; Figure S9 C-D'', G-H'', I-L). In summary, this study uncovers a switching role for Cdc42 not only in flectopodia-mediated vessel co-option, but also in suppressing the activation of pericytes into cytotoxic, macrophage-like phenotypes.
We next investigated whether iCdc42-treatment could block vessel co-option and promote an immune response in vivo. Sevenday xenografts of wild type-GBM cells show high levels of h-CD44 and h-Nestin, with m-Rgs5 + -cells around dilated, co-opted vessels ( Figure 4J). In contrast, iCdc42 tumors appear to be compromised. Three days after grafting, implants present only a thin h-CD44 +shell and an intense m-Rgs5 + -core, with no evidence for h-Nestin +cells or vessel co-option ( Figure 4K; Figure S9 P). By 7-days, the h-CD44 + -shell and the m-Rgs5 + -core are reduced or even absent ( Figure 4L), with an accumulation of vimentin + -microglia at the implantation site, demonstrating an increased host phagocytic response (Figure S9 Q). Taken together, our data indicate that Cdc42 activity in GBM cells favors tumor-establishment over clearance.
Targeting the Cdc42/CD44/actin/pericyte/hypoxia axis (demonstrated above), to block vessel co-option in patients, depends critically on the underlying mechanism being conserved from mouse to human. Strikingly, we found that 88% (7/8) of an unbiased sample of 8 primary human GBM tumors show abnormally elevated levels of CDC42 and CD44, restricted to perivascular locations on abnormal blood vessels at the boundary between tumor tissue and normal looking brain ( Figure S10 A-C), a feature reproduced in an independent GBM biopsy (Figure S10 E). Interestingly, these two genes are part of a coordinately expressed group of genes (synexpression group), that may function together in tumorigenesis, including Hypoxia induced factor-1 alpha (HIF1A), the actin binding protein Transgelin (TAGLN, SM22) and Platelet-derived growth factor receptor beta (PDGFRB), markers of abnormal tumor vessels [35] and pericytes [36], respectively (Figure S10 A-C). Notably, all the tumors (2/8) that recur 6 months after radio-and chemotherapies correlate to those cases where expression of these genes are greatest. Taken together, this validates the Cdc42/CD44/actin/pericyte/hypoxia axis as a desirable target for GBM therapy.
Overall, our results suggest a 2-signal model for GBM progression that involves two distinct tumor cell-derived signals, which act on contractile brain pericytes ( Figure 5 and Figure S11 A-D). The first signal (signal-1) causes pericyte activation and conversion to phagocytic, macrophage-like cells. In contrast, signal-2, which is flectopodia-dependent and requires active Cdc42 function, promotes vessel co-option by engaging contractile pericytes. The combination of these events generates fusion-like hybrids. In the presence of both signals, the tumor expands by continuous co-option and diversification, while, in the absence of signal-2, it is cleared by the unrestricted generation of cytotoxic cells, derived from the activation of contractile pericytes.

Discussion
Recent advances in glioblastoma research have shown that GBM is propagated by perivascular stem-like cells [37] and that . GBM cell/pericyte interaction involves flectopodia and cytoplasmic mixing. A, A flectopodia-like extension (arrow) from a FR labeled-GBM cell (magenta) contacts wrinkling pericytes (arrowhead) on silicone-laminin substrate (grey, DIC-optics). B, GFP-actin-beads in a presumptive flectopodia correlate with varicosities (arrowheads in insets). C, A beaded GFP-actin-extension (U87 cell, white arrowhead) induces altered wrinkling of pericytes (red arrowhead). D, Cdc42 protein (green, white arrowhead in the magnification) partially co-localizes with FR-dextran (FR, magenta, yellow arrowhead) as dots (0.5 mm in diameter) in the extension of a GBM cell. Fixed co-cultures show human CD44 protein in GBM cell flectopodia-like extensions (E-F, cyan, arrows) and in cytoplasmic particles (asterisks and arrowhead in magnification) in target pericytes (phalloidin, red). G-L, Time-lapse analysis of a U87 cell extending and retracting flectopodia (red and white dashed-arrows, respectively) and shedding terminal varicosities (asterisks, and magnifications in H, L and L'). The cell of interest was outlined and filled with a transparent yellow color using Photoshop. M, Double-labeled GDH cells (arrowheads) on constricted (co) and dilated (di) vessel segments (7 day-xenograft). N, Stepwise fusion-like process of a GBM cell (magenta, arrowheads) with a pericyte (arrows), resulting in a migratory cell-derivative (asterisk). Co-cultured GBM cells (MiRu + , red) and pericytes (FlEm + , green) show partial (O, arrow) or complete (P, white arrowheads) co-labeled cytoplasm, 12 or 48 hours after replating, respectively.  . Cdc42-inhibition in GBM cells blocks flectopodia-mediated co-option and activates innate immunity. A, Cdc42 is present in U373 cell-flectopodia in brain slices (arrows). B, GFP-actin-labeled tumor cell (green; red color indicates transfection of the oligonucleotide siRNA control) co-opting a bent vessel (red arrowheads). C, A non-polarized, iCdc42-treated, GFP-actin + -U373 cell (yellow, arrowhead, indicates double labeling of green [GFP] and red [negative control for transfection]), on a straight vessel (Ink-filled, black-filled-white arrowheads). D-E, Graphs of the stemness can be transferred to 'host' cells during tumor progression [38]. Our discovery of a role for pericytes in vessel co-option provides a plausible explanation for these salient features of GBM. Recruitment of activated-pericytes may determine the perivascular location of tumor propagating cells, with transfer of stemness resulting from a fusion-like process and/or cytoplasmic inheritance. The requirement for ongoing co-option at the expanding tumor margin also suggests a reason for the failure of anti-angiogenic therapies, which target only new vessel formation [39].
In contrast to current cancer models that emphasize mutation and clonal expansion from a single-cell of origin [40], our work implies that glioblastoma is a dual cell of origin-disease, where tumor diversification is driven by GBM/perivascular cell coupling (Figure S11 E, F; Discussion S1).
A novel finding of this work is the discovery of flectopodia, as an important GBM cell specialization with the ability to alter vessel morphology. Flectopodia show a unique organization of actin in beads, which seems to reflect the ongoing trafficking of tumor cytoplasm into the cortex of the recipient pericyte. We found that vessel co-option and flectopodia formation and function are dependent on the activity of Cdc42 in tumor cells. Although the precise mechanism of Cdc42-action in this process remains to be determined, it likely relates to its well-established role as an evolutionary conserved organizer of the cortical actin cytoskeleton during cell polarization [30]. This suggests a number of possibilities. First, Cdc42 could act to polarize GBM cells towards the target and to initiate pericyte contact. Second, Cdc42 also controls the focal development of matrix-degrading structures called podosomes, found in normal cell types, such as macrophages and smooth muscle cells [41], and of functionally similar structures called invadopodia in several metastatic tumors [42] including glioma [43]. In spite of the dissimilarities in the actin organization in flectopodia and invadopodia (beads, described here, versus short filaments in bundles [44], respectively), it is intriguing to speculate that flectopodia represent the tissue-version of the in-vitro invadopodia. Alternatively, they might be GBM extensions highly specialized for the interaction with vessel pericytes. Finally, and perhaps more interesting, GBM cell-Cdc42 may act directly within the target pericyte to modify its behavior, as suggested by our demonstration that it is one of the proteins transferred (Figures S3 D-G and S7 B). In smooth muscle cells, physiological contraction in response to vasoconstrictor stimuli requires signaling through Rho-GTPase and Rho-kinase (ROCK), which act specifically by remodeling aSMA and myosin containing stress fibers. Smooth muscle cells also contain a second pathway that leads to actin polymerization and contraction, which is localized in the cortical cell cytoplasm and is mediated, through integrin activation by mechanical stress and contraction stimuli, by Cdc42 [45]. Since contractile pericytes behave similarly to smooth muscle cells and since overexpression of activated-Cdc42 is linked to hyper-contractility of the acto-myosin cortical cytoplasm [46], it effects of iCdc42-treatment on the length of GBM cell extensions and the angle of vessel bending (asterisk, see Methods for length/angle-grouping). D, n = 15 (iRNA and iCdc42); controls, n = 22. E, n = 13 (controls and iCdc42). F, Two juxtaposed U373-grafts, with wild-type (MiRu + , red) or iCdc42 (FlEm + , green) cells. Magnifications display 3D rendering of Ink-filled co-opted convoluted (1, arrows) and non-co-opted straight (2, arrowheads) vessels, respectively. G-H, Quantitative analysis of graft/host margin interaction in short-term slice implants, incorporating both individual and juxtaposed grafts; n = 15 (all controls); n = 11 (iCdc42, G), n = 7 (iCdc42, H). I, Video-frames illustrating two macrophage-like cells (white arrowheads and arrow, respectively) pursuing and destroying a GBM cell (yellow arrows) (see also Movie S8). Time in minutes. U373-wild type (J, 7-day) or iCdc42 (K, 3-day and L, 7-day) xenografts analyzed for the indicated markers. Tc and dtc, core and degenerating tumor-core; white arrows and arrowheads show infiltrating and smooth margin, respectively; dashed outlines mark the original graft. Scale bars: 10 mm (A-C, F1, I), 100 mm (F, J). doi:10.1371/journal.pone.0101402.g004 is interesting to speculate, therefore, that local transfer of GBM-Cdc42 could cause inappropriate contraction of the cortical cytoskeleton, with consequent reorganization of blood vessel morphology. Our identification of Cdc42 as a molecular switch that drives flectopodia-mediated GBM cell/pericyte interaction reveals a new role for Cdc42 in addition to its previously proposed function in GBM cell migration [47], [48]. Interestingly, Cdc42 is implicated in regulating altered cell morphology downstream of TP53 [49] and PTEN [50], two tumor-suppressor genes mutated or inactivated in glioblastoma [51]. Hence, this suggests the intriguing possibility that Cdc42-mediated flectopodia formation results from inactivation of these two key regulatory genes.
Our findings also support pre-existing evidence indicating a direct molecular interaction of Cdc42 and CD44 in association with tumor cell-actin cytoskeleton [31], by demonstrating that they synergize in altering vessel architecture in brain slices. In particular, our data suggest that the adhesion protein CD44 could be involved in anchoring the cell to the vessel, facilitating the transfer/bending process.
The presence of Nitrotyrosine co-localized either within, or close to, fusion-like derivatives (GDH cells) suggests that these cells could represent a focal source of oxygen and nitrogen free radicals that might cause long lasting alterations in pericyte contraction. This is reminiscent of the Nitrotyrosine-containing pericytes that control the capillary 'no-reflow' phenomenon in ischemia reperfusion models of stroke [52]. The location of GDH cells on constricted and dilated blood vessels beyond the tumor margin indicates that they may contribute in creating the hypoxic penumbra, which surrounds the tumor core and which may promote tumor progression. It has also been suggested that reactive Nitrogen species and their derivatives, such as Nitrotyrosine, provide a ''chemical barrier'' that maintains both a tumorspecific immune response and supports the escape phase of cancer [53]. The presence of Nitrotyrosine associated with GDH cells, in our work, therefore implicates these strategically placed tumor cell/pericyte derivatives as part of such an immunological barrier.
Our work provides additional support for the idea that brain pericyte are part of the innate immune system and can behave as macrophage-like cells which possess phagocytic activity [17]. We have shown, here, that isolated brain pericytes in vitro express the macrophage [54] and general phagocytic [55] marker CD68 and are able to engulf fluorescent beads. Furthermore, the cell morphologies of our pericytes co-cultured with GBM cells on laminin-coated silicone substrates present a similarity to M1 and M2 polarized macrophage phenotypes described by others [56]. Specifically, our round and/or spindle-shaped pericytes, induced by wild-type tumor cells, resemble the anti-inflammatory M2, Tumor Associated Macrophages (TAMs) involved in tumor promotion. In contrast, 'dendritic-like' morphologies found in our pericyte co-cultures with iCdc42-GBM cells, which are able to pursue and phagocytose them, are more similar to pro-inflammatory, tumoricidal M1 macrophages. Thus, pericytes are highly plastic cells that can respond to GBM-mediated signals by changing their contractility, detaching from the substrate and acquiring a pro-or anti-tumor activity. Although it remains to be determined how our proposed two signal-model ( Figure 5) fits with the well characterized signaling systems involved in tumor immunology, we speculate that it could involve inflammatory cytokine networks and purinergic signaling ( Figure S11 G, H).
In conclusion, our findings reveal that brain pericytes not only provide the materia prima for GBM progression, through blood vessel co-option and immune suppression, but remarkably may also represent its 'Achilles' heel'. Taken together, Cdc42-inhibitors (possibly in combination with anti-angiogenic drugs) may prove therapeutically beneficial not only against primary and recurrent glioblastoma, but also other tumors that co-opt blood vessels during brain metastasis [57].

Human tumor cells and in vivo labeling
Human glioblastoma cell lines U87-MG and U373-MG were purchased from American Type Culture Collection (ATCC) and European Cell Culture Collection (ECACC), respectively, and grown in a-MEM medium with 10% fetal bovine-serum (Invitrogen). Cells for imaging were transfected with: GFP-actin plasmid (pEGFP-actin, Clontech), encoding a fusion of EGFP and human cytoplasmic beta actin; EGFP-plasmid alone (pCAGGS-GFP); Cherry-Cdc42 plasmid, encoding an N-terminal fusion of mCherry and human Cdc42 (GeneCopoeia); GFP-RR (for GFP targeting to the outer mitochondrial membrane, kind gift from Dr. Nica Borgese, CNR, University of Milano, Italy) and GFP-beta3 integrin (kind gift from Dr. Victor Small, Institute of Molecular Biotechnology, Vienna, Austria). For chemical transfection, we used Superfect reagent (Qiagen) according to manufacturer's instructions. Dextran-labeled U87 and U373 cell pellets (biotinylated mini-Ruby, MiRu, Fluoro-Emerald, FlEm, or Alexa-Fluor-647: D3312, D7178 and D22914, respectively, Invitrogen/ Molecular Probes) were prepared in 20 ml hanging drops [58], containing 45,000 cells, 17.5 ml growth medium and 2.5 ml of Dextran stock (0.5 mg in 40 ml H 2 O). Cultures were incubated for 48 hours and cell pellets were washed prior to use, either as grafts into mouse brain slices or as implants into mouse brains. In some experiments, single cell suspensions of Dextran-labeled cells were generated by trypsinization and trituration of labeled cell pellets. GBM cells for 2-day mouse xenografts were transfected with h-Cherry-Cdc42 and used to prepare hanging drops. In some cases GFP-actin transfected cells were co-labeled with CMTMR (5-6-4-Chloromethyl-Benzoyl-Amino-Tetramethylrhodamine).
Pericyte isolation and in vitro co-cultures. Mouse brain pericytes for co-culture experiments were isolated according to the method of Oishi et al. [59]. Pericytes and GBM cells were labeled separately with different color Dextrans, as hanging drops. Prior to co-culture, cell aggregates were dispersed to a single cell suspension with trypsin, mixed at a ratio of 1:1 with GBM cells and plated on glass coverslips for 15 hours. To identify hybrid cells, co-cultures were trypsinized, replated and checked for double-labeling 12 and 48 hours later.

In vivo vessel visualization
To label cerebral blood vessels with DiI or DiD (Invitrogen/ Molecular Probes), mice (Harlan Laboratories) were anaesthetized with Ketamine (100 mg/kg) and Xylacine (10 mg/kg) and 0.8 ml of a 0.05 mg/ml solution of DiI or DiD (1:10 dilution of 0.5 mg/ ml-DiI/ethanol in 30% sucrose) was injected into the left atrium of the heart, followed 2 minutes later by transcardial perfusion with PBS. To label with ink, anaesthetized mice were transcardially perfused with 15 ml of a 20% solution of black drawing Ink (Pelikan), in Krebs buffer. For intravital experiments, mice were anesthetized with liquid isoflurane (Esteve Veterinaria) and DiI was injected in the retro-ocular venous plexus.

In vivo pericyte labeling
Mouse brain pericytes were labeled in vivo using a modified version of a method described by Hirase et al. [61]. Briefly, 0.5 ml of a fluorescent Dextran (Fluoro-Emerald, FlEm, D-1820; Cascade Blue, D-1976; or Alexa-Fluor 647-conjugated [Far Red, FR], D22914; 10 KD, Molecular Probes) (250 mg/ml in 1.5% BSA) was applied to the tip of a hand-pulled heat-sealed glass Pasteur pipette stub (external diameter of approximately 0.5 mm, length of 1.5 cm) and allowed to dry for 20 min at 37uC. For each mouse, two holes were drilled in the skull (1.3 mm in front of the bregma and 2 mm from the midline; 3 mm behind the bregma and 3.5 mm from the midline) and Dextran-labeled glass stubs were inserted into the brain to a depth of 2.5 mm, via pilot holes made using a 28-gauge needle, and left in place for 10 min. This protocol resulted in the strong labeling of the meninges and of a subset of brain pericytes. Mouse brains were implanted with tumor cell aggregates or used for brain slice cultures 5 days after the labeling.

Preparation of live brain slices and explants
Following in vivo labeling of vessels with Ink, DiI or DiD, brains of ICR or NG2DsRedBAC transgenic mice [62], designated as Tg (Cspg4-DsRed.T1)1Akik/J by the Jackson Lab (kind gift from Dr. Dirk Dietrich, Clinic for Neurosurgery, University Clinic, Bonn, Germany), were embedded in 4% low-gelling temperature (LGT)-agarose and cut into 250 mm-vibratome sections or into approximately 1 mm-thick slices, using a razor blade, and collected into ice-cold Krebs solution. Brain sections were incubated with U87 or U373 cell suspensions in a rotating tube for 12 hours, prior to imaging or fixation. Thick slices for grafting were transferred to 3.5 cm petri dishes, held in place with 4% LGT-agarose and equilibrated in culture medium at 37uC for 30 min before cell grafting. Explants for imaging were placed in glass bottom dishes (MatTek Corporation) and secured using 4% LGT-agarose and a glass coverslip. All explants were maintained in a-MEM medium containing 10% FBS.

Preparation of in vitro silicone substrates and co-cultures
Flexible silicone substrates were prepared as previously described [29], [63]. Briefly, 50 ml of silicone oil (viscosity 60,000 cSt, Code 1838, Sigma) was applied to a 12 mm-glass coverslip using a 1 ml-syringe barrel. The silicone-coated coverslip was then centrifuged at 1,000 rpm for 2 minutes in a 12 well microplate (using absorbent paper discs as cushions and to soak up the excess liquid) and then polymerized by heating for 2 seconds over a small yellow Bunsen flame. After 2 hours sterilization with Ultraviolet light (UV), they were finally coated for 1 hour, at room temperature, with 200 ml of human-laminin protein (0.08 mg/ml, Millipore; also used for coating glass coverslips at 10 mg/ml), and finally rinsed with PBS. For co-cultures with GBM cells, pericytes alone were plated on silicone substrates for about 48 hours and imaged by confocal microscopy for approximately 6 hours, to determine the fields for the addition of GBM cells. The same selected fields were imaged for 8 hours after starting the cocultures.

Xenografts and thick slice implants
Cell pellets, prepared as hanging drops, were grafted into Athymic Nude mice (Foxn1 nu , Harlan Laboratories) or into brain slices. Xenografts (1 pellet/mouse) were introduced into the right hemisphere through a small craniotomy (2 to 3 mm from the midline, approximately 1 mm behind the bregma) at 2.5 mm depth, using a stereotactic apparatus and a Pasteur pipette handpulled to an internal diameter of 0.38 mm. This produced grafts that integrated into the striatum, the cortex or the hippocampus. Mice, 2-days to 3-months post grafting, were perfused using a mixture of 20% Ink and 4% PFA. Brains were either embedded in agarose and cut at 120 mm using a vibratome, or embedded in paraffin-wax (standard procedures) and cut at 7 mm. For thick slice-grafts, cell pellets were manipulated using tungsten needles and pushed inside the region of the striatum. For grafting into NG2DsRedBAC brain slices, hanging drop cell pellets were injected through the edge of a 1 mm-coronal slice using a hand pulled Pasteur pipette to leave a tumor cell implant extending through the striatum. Slices were fixed between 12 and 72 hours after grafting and cut at 30 mm in wax.

Gene expression analysis on Human Glioblastoma
Tumors. Resected Human GBM tumors were analyzed for tumor associated gene expression markers, by mining the Allen Brain Atlas database of in situ hybridization data on Glioblastoma tumor sections. An additional Human GBM biopsy was used for immunohistochemistry (provided by Dr. J. Sola, Hospital 'Virgen de la Arrixaca' Murcia, University of Murcia, Spain).
Ethics Statements. All experiments with mice were in accordance with the Spanish law 32/2007 of November 7 th , for the care, use, transport, experimentation and sacrifice of animals. The use of human GBM biopsies was approved by the Ethical Committee of Clinical Investigation.

Imaging fixed and live specimens
A TCS-SP2-AOBS laser scanning spectral inverted Confocal Microscope (fitted with temperature and CO 2 control; Leica Microsystems, Barcelona, Spain) was used for analysis of fixed and live brain explants and for live imaging of cells cultured on glass. For live imaging of silicone substrates, transmission Differential Interference Contrast (DIC) images were collected, simultaneously with confocal fluorescence images. A fluorescence automated DM6000B microscope and a MZ16FA Fluorescence Stereomicroscope (for wide-field microscopy), running Leica Application Suite (LAS) AF6000 Software (version 2.0.2), equipped with a DFC350-FX (monochrome) or DC500 (color) digital cameras (all purchased from Leica Microsystems, Barcelona, Spain) were used to analyze fixed cells, brain explants and histological sections from xenografts. For live imaging of brain explants, we used a TCS SP2 Acousto-Optical Beam Splitter (AOBS) scanning multiphoton system, with an inverted electronically controlled DM-IRE2 microscope (Leica), equipped with temperature and CO 2 control. For intravital imaging, we used a TCS SP2 RS-scanning multiphoton system with an upright DM LFSA-microscope (Leica), equipped with a special bridge (Bridge 500) housing a mouse head-holder (Luigs & Neumann Feinmechanik und Elektrotechnik GmbH, Ratingen, Germany). For this purpose, a small window was made in the skull over the neocortex of an anesthetized mouse. GBM cells were injected onto the brain surface and the window was then closed with a glass coverslip and held in place with silicone grease. Inverted and upright multiphoton microscopes were connected to either a Millenia-Tsumani or a Mai Tai HP i:Sapphire picosecond laser (SpectraPhysics, Mountain View, CA, USA). All images were collected using internal spectral detectors and LCS Lite Software

Statistical analyses
For statistical analysis, we used Microsoft Excel (2003) and MATLAB Software. Tests were aimed to compare either features and behavior between control and iCdc42 and/or shCD44treated GBM cells in brain slices or pericyte wrinkling behavior, with and without GBM cells, on deformable substrates. For CD44interference, we co-transfected shCD44-plasmid with GFP-actin and included only green cells in the analysis. Normal or nonnormal distributions were evaluated using Kolmogorov-Smirnov (K-S) test for non-parametric data (one-sample). K-S test for non normal data sets was used for comparing wrinkling behavior of the same region, before and after adding GBM cells, and length of protrusions of GBM cells interacting with blood vessels in brain slices. Length of protrusions (mm) has been grouped as follows: 0(0), 1(0-7), 2(7- (0u) = 0%; iCdc42: n = 13, median = 0, range: 0-160u, fraction (0u) = 54%; iCdc42 shCD44: n = 13, median = 0, range: 0-108u; fraction bending-angle (0u) = 62%. For evaluation of bending events: controls: n = 29, total fraction of cell-associated bent vessels = 83%; Secramine A-treated: n = 29, total fraction = 36%. Wilcoxon Mann Whitney test for non-parametric data was used for: vessel co-option in thick slices (n = number of grafts in a total of 14 thick slices). Chi Square (x2) test for descriptive statistic was used to compare differences in the frequency of vessel bending (bending events) and differences in glomeruloid-like structure formation in brain slices (controls: n = 36, total fraction of cellassociated glomeruloids = 36%; treated: n = 31, total fraction = 6%). Student's t-test for parametric data was used for the number of infiltrating cells on vessels in thick slices (n = number of grafts in a total of 14 thick slices). K-S and t-tests were two-tailed. Significance value (P,0.05) was adjusted to avoid inflated type-1 error: a = 0.05/4 (for brain slices); a = 0.05/2 (for thick slices). Protrusion length was counted only for GFP-actin labeled cells contacting vessels, either directly or through extensions, using Leica Application Suite (LAS) AF6000 software. For statistical analyses on protrusion length, a length = 0 mm was assigned to each cell lacking protrusions. To measure the angle of vessel bending, each bent segment was manually traced from the microscope screen onto an acetate sheet and the angle of deviation, from vessel axis, was measured using a protractor. A bending angle degree = 0 was assigned to straight vessels. For tracking selected wrinkles on silicone substrates, xyzt-data from confocal videos were imported into Imaris Software (5.7 version, Bitplane) and used to generate 4D movies. Identified wrinkles were tracked using the Filament Tracking package on the z-series frame offering the sharpest focus. Specifically, we tracked the displacement of the ends of each wrinkle (designated as e1 and e2), for 5 wrinkles across each area of 0.03 millimeters squared. Among the various statistical parameters offered by Imaris Software for the analysis of each track (e1 and e2), we selected the value defined as 'track straightness' (E1 and E2, indicating track displacement/ track length), which represents a numerical measurement of how the ends of the wrinkle can move in space and time. With pericytes alone, the track straightness is generally higher than in co-culture with tumor cells. This reflects the tendency for wrinkles to either: 1) drift laterally or drift laterally and pivot, which represents the majority of the pericytes located in the regions surrounding the nodes and in the regions connecting them (internodes); 2) increased local movement and/or less lateral displacement, shown by a relatively small percentage of pericytes located in the antinode regions. In the presence of tumor cells, the reduction in the wrinkle track straightness indicates the second behavior as the main modality of pericyte contraction. Imaris and Leica Application Suite programs were used for cell measurements and quantification in brain explants and on silicone substrates. The 3D-scatter plot used to illustrate wrinkle data distribution was obtained using MATLAB Software.  Table showing the labeling characteristics and phenotypes for cells located in the tumor core (Type-1fusion-like, Type-1f) and at the infiltrating margin (Type-1, Type-2 and GDH), for both U373 and U87 GBM xenografts. P-R, Confocal pictures of brain slices, pre-labeled with FlEm-Dextran for pericytes (green) and seeded with MiRu + -U373 cells (red). Examples of novel recombinant cell types, that resemble those found in mouse xenografts: P shows a type-1 like-cell located on a blood vessel (yellow arrow) and associated with a FlEm +particle (white arrowhead in magnification); type-1f-like cells are also present, either closely apposed (arrowheads in Q) or partially co-labeled (white arrow in the confocal section, and asterisk in inset). R, Double labeled cells (FlEm + and MiRu + ), similar to GDH cells, are also visible. Arrows point to the almost complete co-localization of the two Dextrans. Vessels   Signal-1 activates pericytes to trigger the innate immune system and phagocytic response. Signal-2, by co-opting naïve, contractile pericytes, leads to vessel cooption and remodeling, and acts as an immunosuppressor by limiting the number of naïve pericytes available to be converted into macrophage-like cells. Signal-1 plus signal-2 also leads to the production of fusion-like hybrids. In the absence of Cdc42 (C), the constraint on pericyte activation is relieved, leading to rapid tumor clearance. E, Hypothetical scheme showing putative recombinant cell types generated by flectopodia-mediated transfer into naïve pericytes (TP1, TP2, left hand side), or by resolution of the tumor cell/pericyte-'fusion hybrid' (Hy1-Hy4, right hand side). The phenotype of the resulting cells is hypothesized to be specified by the independent segregation of cytoplasmic and nuclear determinants, where the behavioral phenotype depends primarily on the cytoplasmic factor. (D and F) Key legends. G, Signal 1 could involve pro-inflammatory molecules (red) released by GBM cells (green), which activate contractile pericytes to a macrophage-like, oncolytic phenotype (M1-type, white). During this acute inflammatory process, extracellular ATP (e-ATP), for instance, could promote the M1-type polarization and assembly of the P2X7R/ NLRP3 inflammasome [64]. Signal 2, which opposes signal 1 and requires GBM cell/pericyte contact through flectopodia, results in the altered contractility of the pericyte and the localized clustering of F-actin in the pericyte cytoplasm ( Figure 3F and Figure S11 H). This clustering could be due to the flectopodia-dependent transfer of F-actin beads into the target pericyte. Alternatively, or in addition, it may involve the production of PPi, from extracellular ATP [64], [65]. ATP hydrolysis involves the activity of the ectonucleotidases CD39 and CD73, present on pericytes [66] and GBM cells [67], respectively, in combination with the E-NPP1, which is enriched in glioma stem cells [68]. F-actin clusters could in turn inactivate the inflammasome and convert macrophage-like pericytes from M1 (inflammasome active) to M2-like type (M2?) (uncoupled P2X7R/NLRP3, inflammasome inactive). ATP hydrolysis also produces Adenosine, which promotes immunosuppression [69] and cell fusion [70]. Signal 2, therefore, may act through F-actin clustering and ATP hydrolysis to prevent the formation of M1-type macrophage-like pericytes and instead maintain a M2-type macrophage-like phenotype, thus supplying a chronic, low level inflammation-state which can never properly resolve.

Supporting Information
(EPS) Movie S1 Flectopodia mediate large-scale vessel rearrangement. 4D-reconstruction of a two-photon video, showing GFP-actin U373 cell-flectopodia (green) with beaded actin (white arrows) inducing a vessel re-arrangement (red, DiI; white arrowheads, bent vascular structures). GFP-actin transfected cells are also labeled with CMTMR (red). Another GFP-actin-cell (yellow arrow) co-operates in translocating modified vessel segments. Yellow arrowheads, actin beads in the initial cytoplasmic bridge (initial blue arrows); red arrows, fusion of the first cellflectopodia; magenta arrow, coalescence between the pre-existing flectopodia and the leading edge of the second cell; final blue arrows, cytoplasmic bridge with reverse flow of actin beads after merging of the two cell extensions. Movie S3 Isolated brain pericytes show contractile activity on laminin-coated silicone substrates. Flat pericytes generate parallel moving-wrinkles (red arrowheads, in the regions delimited by the red dashed lines), organized around highly contractile nodes (red arrows). Yellow dashed lines outline the internode regions where wrinkles move and pivot (yellow arrowheads), while the green arrowhead points to a raised, nonwrinkling, activated pericyte in an anti-node region (approximated by the green dashed boundary). Dtime Movie S7 GBM cells merge with brain pericytes on laminin-coated silicone substrates. Magnification of a detail from Video 4, showing the step-by-step sequence of a merging between a tumor cell and a pericyte. The FR dextran + -cytoplasm from a U87 cell (white arrowheads) is progressively transferred (yellow arrowheads) into an adjacent unlabeled pericyte (white arrows). Asterisks indicate the merging cell product, which then migrates away. Dtime: 10 minutes; time: 1 hour 40 min; rate: 1 fps. (AVI) Movie S8 GBM cells are destroyed by activated phagocytic pericytes. An iCdc42-U87 GBM cell (yellow arrows) is pursued and caught by a non-contractile, macrophage-like, activated pericyte (white arrowheads), which appears to contain cytoplasm from a previous engulfed FR + -GBM cell (yellow arrowheads). Blue arrowheads indicate the sequence of enveloping and destruction of the tumor cell, with the resulting uptake of the residual GBM cell into the phagocytic cell-cytoplasm (blue arrowheads). A second activated pericyte (white arrows) participates in the immune reaction. Dtime: 10 minutes; time: 8 hours; rate: 3 fps.
Movie S9 Inhibition of Cdc42 activity (iCdc42) in GBM cells induces pericyte immune response on laminincoated silicone substrates. An iCdc42-U373 GBM cell (yellow arrows) is chased by an activated host pericyte (white arrowheads), which finally engulfs it (blue arrowheads). The result of another phagocytic event is shown in the upper part of the video (second blue arrowheads). Note that the entire field is characterized by dendritic-like cells connected in a network by long thin extensions (red arrowheads). Dtime: 10 minutes; time: 8 hour; rate: 4 fps. Time between frames: 10 min; total time: 8 h.