ECM stiffness regulates glial migration in Drosophila and mammalian glioma models

The formation of the brain is one of the most fascinating processes in development. Once neurons are born, they settle and never divide again. By contrast, glial cells retain mitogenic and motile properties. Glial motility requires intensive interaction with the extracellular environment. On the one hand directional cues have to be perceived; on the other hand, a dynamic regulation of cell-cell or cell-matrix contact is needed to allow glial migration (Friedl and Gilmour, 2009; Klämbt, 2009).

In Drosophila, glial migration has been extensively studied in several parts of the nervous system (Klämbt, 2009). During embryogenesis, peripheral glia migration is in part controlled by Netrin signaling and the modulation of adhesive forces is provided by the expression of the NCAM homolog Fasciclin2 (von Hilchen et al., 2010; Sepp et al., 2000; Silies and Klämbt, 2010; Silies and Klämbt, 2011). During larval development, the forming visual system provides an easily accessible model system with which to study glial migration. Photoreceptor neurons emerge in the eye imaginal disc (eye disc), which is connected to the brain by the optic stalk. All retinal glial cells are derived from the optic stalk glia and subsequently move onto the eye disc (Choi and Benzer, 1994; Rangarajan et al., 1999; Silies et al., 2007b). The proliferation and subsequent differentiation is controlled by a sequential activation of the fibroblast growth factor (FGF) receptor and only the sustained overactivation of the FGF receptor leads to the formation of massive solid tumors (Franzdóttir et al., 2009). Whereas FGF-receptor activity promotes glial proliferation, activity of Neurofibromin2 counteracts glial proliferation in flies as well as in human neurofibromatosis (Asthagiri et al., 2009; Reddy and Irvine, 2011).

Owing to the inherent mitogenic potential of glia, glial tumors are common in humans and gliomas are among the most deadly types of cancer (Furnari et al., 2007). The most aggressive form is glioblastoma [World Health Organization (WHO) grade IV], an astrocytic tumor with a median survival of 16 months (Furnari et al., 2007). Owing to their highly invasive nature, glioblastomas cannot be completely resected, and even though multimodal therapeutic approaches comprising surgical resection followed by radiation and chemotherapy are applied, prognosis is dismal. In recent years, a number of molecular pathways contributing to glioma formation have been identified (Riemenschneider et al., 2010; Verhaak et al., 2010). Amplification of the epidermal growth factor (EGF) receptor occurs in 40% of all glioblastomas and contributes to glial proliferation (Libermann et al., 1985; Louis et al., 2007). Platelet-derived growth factor (PDGF) receptor pathway activation via genomic amplification or overexpression is often linked to glioma progression and worse prognosis (Brennan et al., 2009; Ozawa et al., 2010). Accordingly, inhibitors of the PDGF receptor pathway, such as imatinib or dasatinib, are clinically explored in individuals with glioblastoma (Morris and Abrey, 2010; Paulsson et al., 2011).

Here, we have employed a Drosophila glioma model (Read et al., 2009; Witte et al., 2009) where glial cell motility is evoked by activation of the PDGF receptor homolog PVR. Cell migration is in part regulated by extracellular matrix (ECM) stiffness, which is mediated through Lysyl oxidase (Lox) activity. Drosophila encodes two genes with such enzymatic activity (Molnar et al., 2005). Both genes appear to be regulated through PVR and PVR is not able to evoke ectopic glial migration when their function is removed. We use atomic force microscopy (AFM) to show that Drosophila Lox regulates the stiffness of the ECM. Moreover, we demonstrate a positive-feedback loop via Integrin signaling that ensures a rigid ECM in the vicinity of migrating cells. We further use a mouse xenograft model: following transplantation, glioma cells rapidly colonize the host brain; this colonization is suppressed when Lox activity is blocked by a specific inhibitor. Our results highlight the relevance of ECM rigidity in the modulation of glioma cell migration and suggest novel therapeutic targets.

During Drosophila eye development, photoreceptor neurons develop in the eye disc, whereas retinal glial cells are born in the CNS and migrate onto the eye disc through the so-called optic stalk (Fig. 1A,C). At the end of larval development, about 350 glial cells populate every eye disc (Silies et al., 2007b).

Glial expression of activated PVR (λPVR), the Drosophila PDGF- and VEGF-receptor homolog, resulted in increased migration with streams of glial cells moving across the eye disc in young third instar larvae (Figs 1 and 3). In addition, we noted a reduction in glial cell number (15%, n=30 discs) and an increase in the size of glial nuclei (57%, n>200 cells). Depending on the migration pattern, we defined three different phenotypic classes (Fig. 1). In about 50% of the eye discs (n=43) we observed streams of glial cells on both the basal and the apical side of the eye disc (Fig. 1D). In 40% we noted glial migration along the Bolwig's nerve on the peripodial membrane of the eye imaginal disc, whereas in the remaining discs abnormal glial migration was noted on the basal side (Fig. 1E,F).

Pan-glial expression of activated FGF receptor results in dramatically increased glial number, but does not induce glial migration (Franzdóttir et al., 2009). Thus, signaling downstream of receptor tyrosine kinases diverts either to promote cell division or to stimulate cell migration. PVR-induced cell migration is manifested already very early in development and can be seen in second instar larvae (supplementary material Fig. S1A,B). In addition, concomitant misprojections of the photoreceptor axons are seen in ∼40% of the eye discs dissected from third instar larvae (n=23, supplementary material Fig. S1C,D). Within the brain, glial cells expressing activated PVR are organized in an abnormal multilayered fashion (supplementary material Fig. S1E,F). Importantly, PVR signaling appears to primarily instruct cell migration. However, phenotypes were only apparent when animals were grown at 18°C, whereas a culture temperature of 25°C resulted in embryonic/early larval lethality.

To further understand how PVR activity is transduced to increase cell motility, we followed a genetic approach and screened for genes that would alleviate the consequences of PVR over-activation by using a cell-type-specific knockdown of gene functions via Gal4-mediated expression of dsRNA (Dietzl et al., 2007). We generated a tester fly strain in which the Gal4-mediated expression of activated PVR is blocked by concomitant ubiquitous Gal80 expression {UAS-λPVR/UAS-λPVR; repo-Gal4 UAS-CD8GFP/TM6 tub-Gal80}. When these tester flies are crossed to wild-type flies, Gal80 function is removed genetically (supplementary material Fig. S2) and all non-TM6 flies, which can be recognized by the dominant marker Tubby, die due to the activation of PVR mediated by repo-Gal4 activity in glial cells.

We next crossed the tester flies to flies carrying different UAS-dsRNA constructs. The non-TM6 progeny of these crosses were then scored for a shift in the lethal phase towards pupal stages. Following screening of 2000 lines, we identified several genes that, upon silencing, shifted the lethal phase to late third instar or even pupal stages (Table 1). Silencing of PVR function in this paradigm also shifts the lethal phase to the pupal stage and serves as a control in the screen. We noted a shift in lethal phase upon silencing several components of the hedgehog signaling pathway,which has previously been associated with glial migration (Rangarajan et al., 2001) (Table 1). Hedgehog binding to Patched blocks its negative action on Smoothened (Briscoe and Thérond, 2013; Ingham et al., 2011) and thus it was surprising that knockdown of hedgehog or patched both shift the early lethality induced by glial PVR expression. In agreement with these findings, we were able to rescue the early lethal phenotype of PVR expression upon panglial co-expression of Patched. We also identified several members of the RTK and Wingless signaling pathways. The largest group of genes affecting PVR-induced lethality regulates transcription via chromatin regulation (Table 1). The strongest suppression effects were noted upon silencing of Bap60 and lox (Table 1). As Bap60 might simply affect UAS-mediated PVR expression, we focused on an analysis of the role of lox.

lox encodes a Lysyl oxidase, which in mammals is known to exert ECM crosslinking functions to increase ECM stiffness (Lucero and Kagan, 2006). Moreover, lox expression promotes breast cell metastasis, as increasing ECM stiffness promotes cell migration (Levental et al., 2009; Moore et al., 2010; Ng and Brugge, 2009).

In Drosophila, perineurial glial cells migrate below the neural lamella, a dense ECM surrounding the entire nervous system (Stork et al., 2008; Xie and Auld, 2011). We therefore set out to test whether modulation of ECM rigidity suppresses glial cell migration evoked by expression of activated PVR. A well-established specific inhibitor of Lox activity is BAPN (β-aminoproprionitrile). This small compound irreversibly blocks the catalytic properties of Lox family members (Smith-Mungo and Kagan, 1998; Tang et al., 1983). When we reared the flies on food containing 1.25 mM BAPN, we also noted a shift of the PVR-induced early larval lethality to late larval stages, with few animals reaching pupal stages (1-4 pupae/small vial with four females and two males; for crossing scheme, see supplementary material Fig. S2). This is never observed when PVR-overexpressing flies are fed untreated food, when lethality in early larval stages is found. This supports the notion that lox acts downstream of activated PVR and points towards the notion that lysyl oxidase function promotes migratory phenotypes evoked by PVR expression.

The Drosophila genome encodes two lox-related genes (Molnar et al., 2005). Whereas lox is weakly expressed throughout all developmental stages, lox2 is expressed mostly in adults. Lox is a predicted 38.5 kDa large secreted protein, most closely related to the mammalian Loxl4 protein (Molnar et al., 2005,, 2003) (Fig. 2A). In Drosophila S2 cells, N- or C-terminally-tagged Lox protein of the predicted size is quantitatively found in the supernatant, suggesting that no proteolytic processing occurs in S2 cells (Fig. 2B,C). When we treated the supernatants with Endo H or PNGase F, which removes N-linked glycosylation, we noted only a slight change, indicating a moderate glycosylation of the protein (Fig. 2C). When we followed expression of a Myc-tagged Lox protein during Drosophila development in a ubiquitous manner using the da-Gal4 driver we found unprocessed secreted Lox protein in the larval hemolymph (Fig. 2D). Interestingly, we could also detect secreted Lox protein in the hemolymph when we expressed Lox specifically only in glial cells using the repo-Gal4 driver (Fig. 2D).

To verify that lox plays a decisive role in regulating PVR-induced migration, we generated a lox mutant by imprecise excision of a P-element. In the allele lox^ΔA38, the entire lox-coding sequence is removed without affecting the neighboring loci (supplementary material Fig. S3A). Flies homozygous for this mutation are viable and show no obvious external phenotypes. Likewise, the larval visual system forms with no discernible abnormalities (Fig. 3B). Moreover, pan-glial overexpression of lox does not interfere with normal glial migration (Fig. 2E,F).

Surprisingly, the lox^ΔA38 mutation does not shift the lethal phase associated with glial expression of activated PVR. Given that the Lox inhibitor BAPN suppresses the PVR induced lethality, this may suggest that redundantly acting genes prevent a phenotypic rescue. The deduced Lox2 protein is highly related to the Lox protein in sequence (Fig. 2A). It carries two SRCR domains similar to the human Loxl2 protein (Molnar et al., 2005). A transposon induced Drosophila lox2 mutation is available (supplementary material Fig. S3B). The transposon insertion disrupts the open reading frame and mostly likely removes the catalytic function of Lox2. lox2 mutants are fertile and no abnormal phenotypes were detected in the eye disc. By contrast, lox lox2 double mutant animals are male sterile but no defects in glial migration onto the eye imaginal disc were noted (n=13, Fig. 3C). However, we noted that the optic stalk is shortened and thicker in 50% of the cases (7/13). To demonstrate this phenotype, we analyzed intact third instar brains with the eye imaginal disc attached. In a single confocal section, this tubby-like optic stalk phenotype can be identified (Fig. 3C′, arrows). These phenotypes suggest at least partially redundant functions of lox and lox2.

When activated PVR is expressed in a lox lox2 double mutant background, the lethal phase is shifted to late larval stages with rare animals reaching pupal stages, as observed following BAPN treatment (see above). Importantly, the glial overmigration caused by expression of activated PVR onto the eye disc is suppressed (Fig. 3D,E, n=7). In addition, the optic stalk appears thicker than normal, resembling that of lox lox2 double mutants (Fig. 3C′,E). In agreement with these observations, we found that lox and lox2 mRNA expression is regulated by PVR activity. Interestingly, activated PVR is able to induce lox expression during embryogenesis, whereas lox2 expression is activated only in larval tissues (Table 2).

Lox oxidizes peptidyl lysine residues to α-aminoadipic-δ-semialdehyde (e.g. of collagen), which permits covalent cross-linking and thus lox genes are crucial for ECM maturation (Lucero and Kagan, 2006). In humans, cutis laxa, a rare congenital disorder, is associated with lox deficiency and is characterized by a lack of collagen monomer cross-linking, which affects the elasticity and thus the rigidity of the ECM. In breast cancer, lox dysfunction has been implicated in metastasis (Levental et al., 2009).

During Drosophila eye development, glial cells migrate along a thick ECM called neural lamella (Silies et al., 2007b). We therefore assayed the elastic properties of the ECM in an area of the eye disc that is contacted by migrating glia using atomic force microscopy (AFM; Fig. 4A). The force required to indent the AFM tip about 200 nm into the neural lamella was measured and Young's modulus, which describes the elasticity of the matrix, was calculated.

In wild type, Young's modulus is around 400 Pa (median 384; mean 426, both values are indicated; n=488 independent measurements in ten imaginal discs; Fig. 4B,C, all measurements are summarized in supplementary material Table S1). A pronounced reduction in rigidity was noted in flies homozygous for the excision allele lox^ΔA38 (222; 221; n=165, four imaginal discs; Fig. 4C). By contrast, pan-glial overexpression of lox resulted in increased rigidity in the overlying ECM (431; 522; n=170, four imaginal discs; Fig. 4C) when compared with wild-type or repo-Gal4 control imaginal discs.

Given the decrease in ECM stiffness noted in lox mutants, we analyzed the ultrastructure of the neural lamella in wild type and lox mutants by transmission electron microscopy. The ECM generated in lox mutants appeared structurally similar to the wild-type ECM. When we analyzed the ECM of lox and lox lox2 mutant larvae, we noted no significant increase in its thickness. Areas with reduced and increased ECM thickness are found in about equal numbers (Fig. 4D-G, arrows). For quantification, we determined the actual size of the ECM at 24 positions per optic stalk, defined by randomly placing a star on the TEM image and measuring the thickness of the ECM at the respective intersections (Fig. 4H,I). The ECM thickness varies from 150 nm to 700 nm (n=5 optic stalks per genotype). In summary, this shows that ECM rigidity, but not its apparent structure, is crucially regulated by lox.

Most cell-matrix interactions are mediated by Integrin receptors and we therefore asked whether the rigidity of the ECM is also influenced by integrins. Glial-specific knockdown of myospheroid encoding a β-Integrin subunit also resulted in a softer more-flexible matrix, demonstrating that Integrin signaling not only mediates the adhesion to the ECM but also signals towards the regulation of matrix rigidity (184; 211; n=42; two imaginal discs; Fig. 4C). In addition, pan-glial suppression of myospheroid expression by RNAi suppressed migration of glial cells onto the eye disc (supplementary material Fig. S4). A similar reduction in ECM stiffness was noted in mutants lacking the focal adhesion kinase (fak), which is involved in integrin signaling (178; 327; n=205; six imaginal discs; Fig. 4C). However, glial migration onto the eye disc does not appear to be affected. Moreover, when we co-expressed the Drosophila integrins PS1 and PS2 [multiple edematous wings (mew) and inflated (if)] in glial cells, the overlying ECM turned stiffer (411; 479; n=225; seven imaginal discs; Fig. 4C).

The modulation of ECM stiffness in response to Integrin signaling prompted us to determine a possible link between integrin and PVR function. PVR activation significantly induces expression of myospheroid (Table 2), but silencing of myospheroid did not suppress the lethal phenotype induced by PVR expression. However, overexpression of Integrin causes an upregulation of lox and lox2 expression predominantly in the larvae, whereas overexpression of lox causes an upregulation of myospheroid expression (Table 2). These data suggest that a feedback loop is established that ensures the stiff extracellular matrix of actively migrating cells (see Discussion).

The above data demonstrate the crucial role of ECM modulation during glial migration in response to PVR activation in Drosophila. Because, in humans, PDGF receptor activation is often linked to glioma progression and to a worse prognosis (Brennan et al., 2009; Ozawa et al., 2010), we analyzed the expression of human LOX genes. The human genome encodes LOX and four members of the LOX-like family (Lucero and Kagan, 2006; Molnar et al., 2003). LOX, LOXL1 and LOXL4 are expressed in the normal brain, where they are found mostly in neurons, vessel walls and some glia of the white matter (Fig. 5A-C). In glioblastoma biopsy tissue, however, an increase of LOX, LOXL1 and LOXL4 expression is detected (n=17 patients, Fig. 5D-F). In addition, we noted expression of LOX genes in different glioblastoma cell lines (human glioma cell lines U343 and A172 or rat C6-SPGFP cells; Fig. 5G-J).

As glioma cells are able to secrete extracellular matrix components (Gladson, 1999), Lox expression by glioma cells in cell culture may be linked to their migration. To prove this hypothesis, we performed a scratch assay (see Materials and Methods). Following knockdown of expression of LOXL4, which is most similar to Drosophila lox, cell migration is slightly reduced (expression level about 12% in U343 and 5% in A172 cells after 24 h; and 15% after to 96 h of culture; Fig. 5K). However, when we stimulated U343 cell migration with 40 ng/ml PDGF-BB, relative migration was significantly decreased in the LOXL4-depleted cells (Fig. 5H). In conclusion, lox-like genes appear to have a function during cell migration in the mammalian system.

To directly test whether inhibition of the activity of the Lox-like enzymes also suppresses glioblastoma invasion, we then turned to a mouse xenograft model. The glioma cell line C6 is derived from a rat glioma and is widely used (Benda et al., 1968). To follow the C6 cells after stereotactic injection, we used the GFP expressing variant C6-SPGFP (Tatenhorst et al., 2005). C6-SPGFP cells (4×10⁴) were transplanted into identical positions of the striatum of female nude mice (n=10). Ten and 15 days later, tumor size was measured by MRI to document the increase in tumor volume (Fig. 6A,C). Subsequent histological analysis revealed extensive diffuse infiltration of GFP-positive C6 glioma cells into the host brain (Fig. 6D,F). In the transplanted C6-SPGFP cells, expression of Loxl1 and Loxl4 is upregulated compared with normal rat brain cells and can be easily detected (Fig. 6G,H,J). A similar upregulation of the expression of lox-family genes has been described for other astrocytoma cell lines (Laczko et al., 2007).

To test whether lox inhibition restricts migration of the rat glioma cells, as observed in the Drosophila model, we administered daily intraperitoneal doses of the Lox inhibitor BAPN (100 µg/g body weight in PBS), which also crosses the blood-brain barrier (Martin et al., 1991). A control cohort of mice was injected with PBS only. Mice injected with C6-SPGFP cells and daily BAPN doses developed gliomas that were indistinguishable from those in controls by MRI, with comparable proliferation indices (Fig. 6K,L). However, the extent of infiltration of glioma cells into the host brain was dramatically suppressed upon BAPN treatment (Fig. 6D,E). A six- to sevenfold lower number of tumor cells or small tumor cell clusters was detected in the infiltration zone of BAPN-treated compared with control mice (P=0.008, Fig. 6F,I). In conclusion, BAPN inhibits glioma cell migration but does not affect tumor cell proliferation.

Cell migration is of prime relevance during the development of the nervous system and is often triggered by the activation of the PDGF receptor. The Drosophila PDGF-receptor homolog PVR is required in a number of cell types for their normal migratory responses. In the fly ovary, pvr instructs the migration of border cells in part by influencing actin dynamics (Duchek et al., 2001; Fulga and Rørth, 2002; Rosin et al., 2004; Wang et al., 2006). In addition, migration of hemocytes and some embryonic glial cells depends on PVR activity (Cho et al., 2002; Janssens et al., 2010; Sears et al., 2003; Wood et al., 2006). In line with these findings, expression of activated PVR efficiently induces glial migration and generates phenotypes resembling a metastatic tumor cell phenotype (Vidal and Cagan, 2006; Witte et al., 2009). Likewise, PDGF-receptor activity instructs cell migration during normal mammalian development, but also during tumor metastasis (Hoch and Soriano, 2003; Jones and Cross, 2004; Shih and Holland, 2006).

Interestingly, the effects of activated PVR on glial motility appear to be a specific feature of this receptor tyrosine kinase, as activation of the related FGF-receptor results in extensive proliferation but does not trigger cell migration. Here, we used a genetic suppressor screen to better understand PVR-induced motility. We identified components of hedgehog and wingless signaling pathways. As both loss and gain of patched function is able to suppress the early lethality induced by PVR activation in glia, we suspect that the precise level of hedgehog signaling is crucially important during glial development. This might also account for the apparently different glial phenotypes associated with hedgehog (Hummel et al., 2002; Rangarajan et al., 2001).

Another component identified by our RNAi screen was lox. Surprisingly, we found that in contrast to the expression of lox^dsRNA, lox mutants do not suppress the PVR induced phenotype. As the Lox inhibitor BAPN is able to suppress PVR mediate lethality, we also analyzed the second Drosophila gene encoding a Lysyl oxidase (lox2). Recently, it was shown that Lox-mediated matrix stiffening also promotes breast and colorectal cancer metastasis (Baker et al., 2011; Levental et al., 2009). Increased ECM rigidity also favors migration of cultured glioma cells (Ulrich et al., 2009). In addition, we found that integrins affect the rigidity of the ECM, possibly by regulating lox mRNA expression. Thus, matrix rigidity often correlates with malignancy, suggesting more general implications of matrix rigidity and cell migration (Krndija et al., 2010; Paszek et al., 2005).

Cells exert forces during migration and thus favor a rigid surrounding (Moore et al., 2010). A number of receptors connect the cells to the ECM. Most prominent are members of the Integrin family, which not only provide mechanical coupling to the matrix, but also convey signals into the cell (Bökel and Brown, 2002; Ginsberg et al., 2005; Moore et al., 2010). The reduction in ECM stiffness following myospheroid knockdown or in fak mutants points to a role of Integrin signaling in regulating lox expression. Indeed, we observed an increase of lox expression in an Integrin gain-of-function situation (Table 2). Similarly, we noted a decrease in the level of lox mRNA expression in fak mutants. Thus, Lox and Integrin might act in a positive-feedback loop enforcing a local extracellular environment favoring migration (Fig. 7).

In the mammalian system, integrins may suppress tumor metastasis (Ramirez et al., 2011). Generally, however, integrins are thought to be important mediators of cell adhesion and constitute attractive therapeutic targets to treat tumor progression (Weaver et al., 1997). In line with these findings, Integrin signaling is also key in regulating glioma cell migration (D'Abaco and Kaye, 2007; Fukushima et al., 1998). As BAPN can suppress glioma cell migration in vivo, this may indicate that the Lox/Integrin feedback loop identified in Drosophila is evolutionarily conserved to couple ECM stiffness to glioma cell migration.

The regulation of ECM stiffness via Lox is not required for viability. In humans, loss of lox leads to recessive connective tissue disorders (Smith-Mungo and Kagan, 1998), whereas no abnormal phenotypes were so far detected in Drosophila mutants (Molnar et al., 2005). In humans and Drosophila, lox activity is upregulated in wound repair (Smith-Mungo and Kagan, 1998; Stramer et al., 2008), reflecting the need for increased migration of cells into the wound area. Thus, Lox might be primarily involved in the regulation of plastic responses encountered during injury and disease.

Flies have a relatively simply structured nervous system; however, relatively few glial cells share pronounced motility with their mammalian counterparts (Klämbt, 2009). The Drosophila glioma model used here has disclosed a crucial role for Lox-related genes in maintaining the migratory state of mammalian glioma cells. Indeed, the cancer clinical genomics database ‘Repository of Molecular Brain Neoplasia Data (Rembrandt)’, which comprises a large number of glioma patient data (http://rembrandt.nci.nih.gov; Madhavan et al., 2009), demonstrates a significant correlation between the upregulation of LOX and LOXL1 and overall survival. As BAPN treatment can block the diffuse infiltration of glioblastoma cells in a mouse model, new therapeutic strategies for a currently incurable disease might be in reach.

All crosses were performed on standard food at 25°C unless indicated otherwise. The lox^ΔA38 excision mutant was generated from P[XP]d09952. The following strains were used: repo-Gal4, repo4.3-Gal4, UAS-λhtl, UAS-λtop, UAS-λPvr, UAS-inflated UAS-mew [provided by V. Auld (University of British Columbia, Vancouver, Canada); B. Jones (University of Mississippi, Oxford, USA); M. Leptin (University of Cologne, Cologne, Germany); B. Shilo (Weizmann Institute, Rehovot, Israel); P. Rørth (University of Copenhagen, Copenhagen, Denmark); N. Brown (University of Cambridge, Cambridge, UK)]; lox^d09952 (Exelixis collection, Harvard University, Cambridge, USA); and lox2^MI06311, tub>Gal80, UAS-CD8GFP, w¹¹¹⁸ (Bloomington Stock Center, Bloomington, USA). UAS-dsRNA strains were from Bloomington, VDRC or NIGFly. For details see Table 1. The different tagged UAS-based lox transgenic fly strains were generated according to standard procedures.

DNA cloning was conducted using Gateway-based vectors (Invitrogen). All constructs were verified by DNA sequencing. Proteins were extracted from cells and analyzed by western blot as described previously (Bogdan et al., 2005). For the glycosidase assay, proteins were extracted by TCA precipitation and deglycosylated by Endo H (Roche) or PNGase F (New England Biolabs) according to the manufacturer's instructions. We used RNase B as a positive control.

Total RNA was isolated from cultured cells or tissue with GenElute (Sigma-Aldrich) or TRizoL (Invitrogen). cDNA templates were synthesized using SUPERscript II polymerase with Oligo(dT)_12-18 primers and RNaseOUT (Invitrogen) following the manufacturer's instructions. qRT-PCR was performed using SYBR Green for monoplex reaction (Bioline), TaqMan Gene Expression Assays for duplex reaction (Applied Biosystems) and Realplex Mastercycler (Eppendorf) following the manufacturer's instructions. The analysis of the data was conducted using the REST 2009 software (Qiagen; Pfaffl et al., 2002). For primers see methods in the supplementary material.

Fixation and treatment of tissues for immunohistochemistry on third instar larval brains and eye imaginal discs were performed as described previously (Silies et al., 2007a; Stork et al., 2008). For a list of antibodies used, see methods in the supplementary material. LOX, LOXL1 and LOXL4 staining was evaluated semi-quantitatively by scoring the percentage of stained cells [0 (absent), 1 (<10%), 2 (10-50%), 3 (51-80%), 4 (81-100%)], as well as staining intensity [0 (absent), 1 (weak), 2 (strong)]. Both scores were multiplied resulting in a maximum staining score of eight.

U343 or A172 glioblastoma cells were transfected 1 day after plating using two siRNAi sequences (LOXL4_1 und LOXL4_4 siRNA, Qiagen) or control siRNA not able to bind known RNA sequences (AllStar negative siRNA, Qiagen). After 24 h the confluent cell layer was scratched with a sterile 100 µl pipette tip. The same area was documented after 0 h, 24 h, 48 h and 72 h, and analyzed using ‘analySIS FIVE’ (Olympus) and TScratch software (ETH Zürich, Switzerland).

Fixation and sectioning was performed as described previously (Stork et al., 2008). All measurements were made using Photoshop. Five different animals were analyzed for each genotype.

Glioblastomas (grade IV, WHO), obtained by biopsy (frozen, n=10; formalin-fixed and paraffin-embedded, n=17), and non-pathologic frontal lobe brain tissue, obtained by autopsy (frozen, n=2, formalin-fixed and paraffin-embedded, n=2), were examined. Ethical approval was obtained from the local Ethical Committee of the Ärztekammer Westfalen-Lippe, Germany (Az.2007-420-f-S).

C6-SP-GFP cells (4×10⁴ in 2 µl PBS) were transplanted into the right striatum of 20 anesthetized female NMRI nude mice (Elevage Janvier) aged 6 weeks using a stereotactic frame (Narashige). Cell suspensions were injected at 4 mm anterior to the interaural line, 2.5 mm lateral to the midline, at a depth of 3.5 mm into the brain. One day before transplantation, intraperitoneal injection of BAPN (100 µg/g/day in 50 µl of PBS) was started in 10 animals, while 10 control mice were injected with 50 µl of PBS. Intraperitoneal injection of BAPN or PBS was continued daily. On day 10 and day 15, MRI was carried out. At day 15 after injection, brains were prepared. Five brains of each group were immediately frozen in liquid nitrogen, while five brains of each group were fixed in formalin and embedded in paraffin. Ethical approval was obtained from the local Ethical Committee of the Ärztekammer Westfalen-Lippe, Germany (Az.2007-420-f-S).

MRI measurements were performed on a 3T Philips Achieva System. A 3D gradient echo dataset was acquired in 8.50 min scan time using SPIR fat suppression and the following parameters: isotropic resolution, 200 µm; reconstructed to 90×90×200 µm³; field of view (FOV), 20×14×8 mm³; echo time/repetition time (TE/TR), 5.5/28 ms; excitation pulse angle, 35°; averages, 6. As contrast agent, Gd-DTPA (Magnevist) was applied at 0.5 mmol/kg body weight through a tail vein catheter 5 min prior to measurement start. Hyperintense-appearing tumor tissue was segmented semi-manually using Amira (Visage Imaging) and the volume was calculated.

Serial coronal sections (3 µm) were made from paraffin-embedded brain tissue of xenotransplanted mice (n=5 for BAPN and control mice). Whole sections were photographed using an automated Corvus high resolution stage control. Areas of tumor core and infiltration zone were quantified using Analysis software and a fixed threshold of light intensity. Fluorescent cell clusters detected outside the tumor core were investigated with respect to size (µm²) and number. We used ImageJ to determine the nuclear sizes. In wild type, almost all glial nuclei appeared round. By contrast, glial nuclei appeared more oval upon expression of activated PVR. The longest diameter of a glial nucleus was measured in ImageJ and was taken as size value.

Comparison of tumor volume (MRI investigation) was analyzed using two-way ANOVA [treatment as between-subject factor and timepoints (10 days and 15 days after implantation) as within-subject factor]. Number of GFP-positive tumor cell clusters in the infiltration zone was compared by Mann–Whitney U-test. Results of the migration assay were analyzed by Student's t-test. P<0.05 was considered significant. Experiments were performed independently three times. All statistical analyses were performed using IBM SPSS Statistics 18 version 18.0.

We are grateful to V. Auld, S. Baumgartner, N. Brown, K. Csiszar, B. Jones, M. Leptin, M. Mink, D. Montell, R. Palmer, P. Rørth, B. Shilo, the Bloomington and Harvard Drosophila stock collections, and to the VDRC and NIG-FLY collection of UAS-dsRNA strains for providing antibodies and fly strains needed in this study. We thank F. Kiefer, R. Adams, P. Kain and W. E. Berdel for discussion and comments on the manuscript. We acknowledge support from members of the Klämbt lab throughout the project.