a–f,In vivo microscopy (3D) of 6 different GBMSC lines (all non-codeleted for 1p/19q, andIDH wild-type) reveals abundant formation of ultra-long membrane protrusions in the mouse brain: T1 (a), T269 (b), T325 (c), S24 (d), WJ (e), and P3 (f) (z-dimensions from 200–500 μm depth). Insets show the boxed areas in the corresponding images in higher magnification, covering a proportion of thez-dimension. Per cell line, two time points are shown, adapted to their growth speedin vivo (T269, P3 fast; T1, S24 intermediate, T325 and WJ slow).g, 3D image of a S24 astrocytoma (injection of a 1:1 mixture of either GFP- or RFP-positive cells), revealing multiple ultra-long and very thin membrane protrusions (arrows) in the live mouse brain. Note that membrane tubes partly run in parallel.h, CGH-profile of the S24 GBMSC line demonstrating chromosomal alterations typical for GBM (chromosome 7 gain, 10 loss).i, Chromosome 7 FISH analysis of one S24 GBMSC in the main tumour area demonstrates polyploidy: 90% ofn = 100 analysed cells in the main tumour area were clearly polyploid for chromosome 7, indicating that implanted S24 GBMSCs give rise to tumours genetically identifiable as glioblastomas.j, Whole mouse brain coronar sections at day 171 after S24 injection showing two main features of glioblastoma growth: diffuse brain invasion in a typical dissemination pattern (left image), and a solid, angiogenic core identified by haemorrhagic changes of the main tumour area (right bright field image).k, Increasing angiogenesis in this tumour is further demonstrated by dynamicin vivo MPLSM.l, Actin-rich S24 GBMSC tip, invading into the brain (single plane images; schematic drawing below).In vivo MPLSM:a–g,k,l.

a, Number and length of protrusions during tumour progression (S24 tumours;n = 77–120 cells inn = 3 mice).b, MPLSM images of S24 GBMSCs genetically expressing green fluorescent protein (GFP, green) linked to different cellular/molecular components.c, Confocal immunohistochemistry (maximum intensity projections) of human nestin (green, allows specific detection of S24 GBMSC-related structures in the mouse brain), and different other cellular and molecular factors (red, co-stainings). The degree of expression of the factor in tumour cell-derived membrane tubes is indicated in the right lane. −, no signal in membrane tubes, (+), positive signal in some membrane tubes, +, positive signal in all membrane tubes.In vivo MPLSM,a,b.

a, 3D reconstruction of membrane microtubes in a T325 astrocytoma over 3 days (in vivo MPLSM). Arrowheads, stable main tube; arrows, dynamic side tubes.b, Example of a very stable T325 GBMSC membrane microtube (arrowheads), followed over 126 daysin vivo; MIP,z-dimension 48 μm.c, Scanning electron microscopy (SEM) image of two photoconverted membrane microtubes (arrows) and a nucleus of a non-photoconverted brain cell (N).d, 3D reconstruction of serial SEM images (22.29 μm (xy) × 4.62 μm (z) = 102.9 μm³) illustrating the membrane contours.e, Maximum speed of mitochondria in S24 membrane tubes versus tumour cell soma (n = 10 per group,t-test, red lines show means).f, 3D reconstruction of serial SEM sections of the membrane microtube (red) and the two axons (green), which are shown inFig. 1f.g, 3D image of the genetic Tlx mouse glioma model, with abundant membrane microtubes connecting single stem-like astrocytoma cells (z-dimension 83 μm).In vivo MPLSM:a,b,e,g. *P < 0.05.

a, Graphs illustrating two theoretically possible ways of intercellular connections by membrane tubes in a model of two tumour cell populations marked with 2 different fluorescent proteins. In hypothesis 1, tumour cells remain connected after cell division with their ancestors. In this case, only connections between cells of the same colour are expected (GFP–GFP (green) or RFP–RFP (red)). In hypothesis 2, tumour cells only connect to unrelated glioma cells. Here, 50% of connections would be between cells of different colour (GFP–RFP or RFP–GFP (grey)), and 25% of the same colour (GFP–GFP (green) and RFP–RFP (red)), respectively.b, Quantification of the real data set, where a 1:1 mixture of either GFP or RFP expressing S24 GBMSCs (S24GFP/S24RFP) was co-injected into the mouse brain, revealing that both potential mechanisms are in place (n = 164 connections inn = 3 mice).c, 3D image (70 days after injection) of a co-implantation of GFP- and RFP-expressing S24 GBMSCs. Quantification revealed that both large fluorophores (which cannot pass gap junctions) never colocalized in cell somata or in TMs (n ≥ 2,500 astrocytoma cells analysed).d,e, Examples of 3D images of membrane tube connections between individual, non-related astrocytoma cells that differently express GFP or RFP (arrows ind ande).f, Example of a 3D image of same-colour connections between two RFP-positive cells (arrows).g, Scanning electron microscopy image of a S24 spheroid. Left, yellow colour marks cell bodies, arrowheads point to membrane microtubes; right, high magnification of tubes with direct membrane contact (arrow).h, 3D images of a perivascular T325 astrocytoma cell (arrows), which first utilizes a TM to explore the perivascular niche (D45–D73) until it moves to the explored region, and remains in a strict perivascular position until day 255. A second cell (arrowhead) is quiescent until D129 and is embedded into a vascular loop formation, which persists after disappearance of the main cell soma.i, MIP of a TM-containing S24 GFP astrocytoma cell which enters a perivascular position over time (arrow), and another one which remains in its non-vascular (parenchymal) position over 105 days (arrowhead).In vivo MPLSM,c–f,h,i; 50–650 μm deep in the brain.

a, 3D image (in vivo MPLSM) of a BT088 oligodendroglioma xenograft tumour growing in the mouse brain, inset shows the boxed area in a higher magnification. Cells are rounded, TMs are scarce.b, Quantification of TM lengths of BT088 oligodendroglioma cells (left), and S24 astrocytoma cells (right), at day 60 after tumour implantation.n = 3 animals per entity.c,IDH1^R132H immunohistochemistry of the contralateral brain hemisphere (macroscopically tumour-free) of a patient deceased from a WHO III astrocytoma.d, Staining of resected primary glioblastomas (n = 3, non-codeleted,IDH wild-type) with a mutation-specific antibody against theirBRAF^V600E mutation reveals the existence of long tumour-cell-derived membrane microtubes in these tumours. Representative image.e, ExemplaryIDH1^R132H immunohistochemistry of gliomas morphologically classified as oligoastrocytoma, with (left) or without (right) 1p/19q codeletion.f, Maximum microtube length of oligoastrocytomas with 1p/19q codeletion (OA CODEL;n = 31 patients) and without (OA NON-CODEL;n = 9 patients).g, Maximum microtube length of tumours morphologically classified as astrocytomas but with 1p/19q codeletion (“A” CODEL;n = 6 patients), or classified as oligodendrogliomas but without 1p/19q codeletion (“O” NON-CODEL;n = 9 patients).In vivo MPLSM:a,b.

a, Example of a calcium wave involving TMs of GBMSCs in a tumour region; measurement by the genetically encoded sensor Twitch-3 that allows ratiometric calcium measurements via FRET. Shown is an overlay of cpVenus^CD and CFP channels. Yellow colour reflects low, red colour high calcium concentrations. Right: ratios of single sections of one TM illustrating the propagation of a calcium wave along the TM.b, MIP (10 slices) of the region shown inFig. 2b (red cells, astrocytes; green cells, tumour cells without Rhod-2AM signal; yellow cells, tumour cells with Rhod-2AM signal).c, Exemplary heat map of intercellular calcium wave (ICW) communications between T325 astrocytoma cells transfected with the genetically-encoded calcium sensor GCaMP3.d, Heat map of the region shown inSupplementary Video 4 (small molecule calcium indicator Fluo-4AM).e, Frequency of calcium peaks recorded during brain superfusion with extracellular saline (ES-control) versus 100 μM carbenoxolone (CBX) in GBMSCs (blue box) and normal brain astrocytes (red box);n = 3 mice per group;t-tests.f, Analysis of baseline-normalized synchronicity (see Methods for details) of calcium signals between S24 GBMSC glioma cells versus those between normal brain astrocytes. Different pharmacological blockers of main propagation mechanisms of ICWs were tested: inositol triphosphate was blocked by 2-APB, cellular ATP receptors by the nonselective purinergic 2 receptor antagonist suramin, and gap junctions were blocked by CBX (glioma cells,t-tests; astrocytes, Mann–Whitney tests). ES, extracellular saline used as control.g, 3D images (z-dimension 180 μm) of SR101 microinjected tumours, without (control, upper image) and with co-injected CBX (lower image; area of injection: circles) 120 min. after injection. Red cells, normal brain astrocytes. Graph, corresponding quantification of SR101-fluorescence (n = 4,962–5,676 cells inn = 3 mice per group; Mann–Whitney test).h, 3D images of a non-TM-connected S24 tumour cell (S24tdTomato), loaded with the gap-junction permeable dye Lucifer yellow via electroporation.i, 3D images of TM-connected S24 tumour cells (S24tdTomato) after dye transfer into one of the TM-connected cells.j, Quantification of Lucifer yellow fluorescence intensity in the neighbouring cells next to the electroporated cell (n = 4 sections fromn = 2 mice;n = 64 TM-connected versusn = 42 non-TM connected cells quantified;t-test).k, Western blot analysis of Cx43 protein expression in 4 GBMSC and 2 oligodendroglioma stem-like (OSC) cell lines.l, Immunohistochemistry demonstrating the localization of different connexins in S24 GBMSCs; no clear TM-related expression, and/or localization at TM crossings could be observed.m, Proportion of TM-devoid (0 TMs) versus TM-rich (>4 TMs) cells in shControl versus shCx43 tumours 20 and 40 days after tumour implantation (n = 3 mice per group, ANOVA, Tukey’s post hoc test).n, Kaplan–Meier survival plot of animals implanted with shCx43 vs. shControl S24 GBMSCs (log rank test).a–g,m, Acquired byin vivo MPLSM.h,i,l, Confocal microscopy images. For gel source data, seeSupplementary Fig. 1. Scale bars show s.d. *P < 0.05, ***P < 0.001.

a, 5 days after initiation of radiotherapy (3×7 Gy), nuclear fragmentation characteristic for apoptosis (arrow) can be detected in a proportion of cells. Green, nuclear staining by H2B–GFP transduction; red, S24 cell cytoplasm.b, Representative 42 day time course of a distinct tumour microregion, followed after start of radiotherapy (day 0). TM-connected cells (two examples are marked with black asterisks) show long-term survival; note that surviving cells show an increase in the number of their TMs.n = 3 mice per group.c, Exemplary heat maps of calcium transients (Rhod-2AM) of a sham treated (left) and radiated GBMSC tumour region (right).d, Relative changes of all cells (left) and subgroups of TM-connected versus non-connected GBMSCs of shControl versus shCx43 tumours after sham/radiotherapy (n = 3 mice per group,t-tests).e–h, Ratiometric measurements of basal calcium levelsin vivo.e, Mean ratios of fluorescence intensities of the FRET partners cpVenus^CD and CFP, before, and after two days of radiation (2 × 7 Gy) in TM-connected cells (n = 3 mice per group; Mann–Whitney test).f, Fluorescence intensities (normalized by the mean intensities of the corresponding data sets) in TM-connected cells for the two FRET partners illustrated by a scatter blot (black dots represent analysed cells at the day before radiotherapy, red dots 2 days after initiation of radiotherapy); linear regression revealed similar correlation strengths at the two time points (n = 3 mice), reflecting very homogenous calcium levels in the astrocytoma cells before and after radiotherapy.g, Mean ratios of fluorescence intensities of the FRET partners before and after two days of radiation (2 × 7 Gy) in non-connected cells,n = 3 mice; Mann–Whitney test.h, Normalized fluorescence intensities in non-TM-connected cells for the two FRET partners. Here, linear regression revealed highly homogeneous basal calcium levels only before radiotherapy, while after radiotherapy the linear correlation was lost, illustrating heterogeneous calcium levels in the analysed cells. (n = 3 mice per group). All data acquired within vivo MPLSM. GBMSCs, S24 cell line.

Biological function analysis of 1p/19q non-codeleted (n = 124) versus 1p/19q codeleted (n = 70) human gliomas of the TCGA database was performed using Ingenuity Pathway Analysis. All tumours analysed wereIDH mutated (GCIMP+).a, Bar plot of the top differentially regulated downstream biological functions.b, Heat map of downstream biological functions. The map is colour coded: more intense orange means more activation in 1p/19q non-codeleted tumours (compared to codeleted tumours), blue the other way round. Note the activation of “cellular movement” and “cell-to-cell signaling” in non-codeleted tumours.c, Results of the analysis of canonical pathways in 1p/19q non-codeleted versus codeleted gliomas. Higher positive z-score: upregulated in 1p/19q non-codeleted versus codeleted gliomas; higher negative z-score: upregulated in 1p/19q codeleted gliomas versus non-codeleted gliomas.

a, TrkA, TrkB, NGF and NT-4 protein expression detected by immunohistochemistry in 1p/19 codeleted versus non-codeleted human gliomas (n = 8 each,t-tests, allIDH mutated).b, Western blot analysis of GAP-43 protein expression of different glioma cell lines. OSC, oligodendroglioma stem-like cell lines.c, GAP-43 western blot of 4 GBMSC lines cultured under non-adherent, stemlike (SC +) versus differentiating, serum-containing, adherent (SC –) conditions.d,In vivo 3D images of S24 shControl versus shGAP-43 GBMSCs (left) and quantification of TM side branches 20 days after implantation (n = 60 cells inn = 5/6 mice,t-test).e, Spheroid invasion assay from S24 shControl versus shGAP-43 cells in a gel matrix, and the corresponding quantification (t-test).f,In vivo tumour cell invasion distance within 24 h of S24 shControl versus shGAP-43 GBMSC tumours (n = 3 mice, Mann–Whitney test).g,In vivo proliferation dynamics in the main tumour area (volume of 0.037 mm³;n = 4 mice, Mann–Whitney tests).h, Fraction of TM-connected cells at day 20 in these tumours (n = 164 cells inn = 6 mice,t-test).i, Western blot analysis of Cx26 (expressed in normal astrocytes), Cx31 and Cx37 (both located on chromosome 1p), and Cx43 protein expression in shGAP-43 GBMSCs versus shControls. Of note, the GAP-43 knockdown leads to a Cx43 protein reduction of 89%, while expression of the other connexins was not reduced.j, T2 MRI images of S24 shControl versus shGAP-43 tumours, 72 days after tumour implantation. Quantifications ofn = 6 animals per group (t-test).k, Kaplan–Meier survival plot of S24 shControl versus shGAP-43 tumour-bearing mice (log rank test).l, Exemplary brain sections with nestin immunohistochemistry of S24 shControl versus shGAP-43 tumours 60 days after radiotherapy. Note that in shGAP-43 tumours, only small remnants of tumour cells can be detected by the tumour cell-specific staining. Regions with highest tumour cell densities (boxes) were quantified for proliferation index (Ki-67-positive cells/all cells;n = 3 animals;t-test).m, Overexpression of GAP-43 in BT088 oligodendroglioma cells results in protein levels similar to that in GBMSCs.n–p, GAP-43 overexpression in BT088 oligodendroglioma cells leads to an increase in TM numbers (n,n = 80 cells inn = 3 mice per group), more TM branches (o,n = 40 cells inn = 3 mice per group), and a higher invasion capacity (p,n = 75 cells inn = 3 mice per group;t-tests) 14 days after tumour injection. Scale bars show s.d. Red lines show means.In vivo MPLSM,d,f–h,n–p. For gel source data, seeSupplementary Fig. 1. *P < 0.05, **P < 0.01, ***P < 0.001.

Anatomical and molecular mechanisms of TM-driven tumour dissemination and network function in astrocytomas. MV, microvesicles; mito, mitochondrion; ER, endoplasmic reticulum; MT, microtubules.

This file contains Supplementary Figures 1 (gel source data) and Supplementary Figure 2 (MRI source data of animals not shown in the main or Extended Data Figures).  This file was updated on 11 November 2015 to correct old references. (PDF 942 kb)

This file contains Supplementary Table 1 which contains results of the differential gene expression analysis of(A) 1p/19q non-codeleted /IDH mutated (n=124) vs. codeleted /IDH mutated (n=70), and(B) 1p/19q non-codeleted /IDH wild-type (n=56) vs. codeleted /IDH mutated (n=70) grade II and III gliomas of the TCGA database. logFC, log fold-change; logCPM, log counts per million; FDR, false-discovery rate adjusted p value. A positive logFC value means relative overexpression in 1p/19q non-codeleted gliomas; a negative logFC value means relative overexpression in codeleted gliomas.GJA1: gene encoding connexin 43 protein.GAP43: gene encoding GAP-43 protein. (XLS 2022 kb)

a) Brain invasion of S24-GFP GBMSCs that were implanted at day 0 into the mouse brain, and followed from day 13 to 62 byin vivo MPLSM in the same brain microregion. Note extension of ultra-long cellular protrusions at the invasive front. Green, S24-GFP cells; red, brain microvessels (TRITC dextran angiography).b) High-magnification time-lapsein vivo MPLSM of one astrocytoma cell reveals that protrusions arborize, and demonstrate a scanning behavior. The box shows a region where the two upper protrusions are extended, the lower is retracted. (MP4 2853 kb)

z-stacks of three different astrocytoma mouse models, to illustrate the 3D morphology of microtubes, and microtube-interconnected cellular networks:a) S24-GFP GBMSC xenografts after 60 days of growth in a mouse brain;b) T269-GFP GBMSC xenografts after 102 days of growth in a mouse brain,c) Genetic mouse model of astrocytoma, where a tumor cell subpopulation with stem-like properties is identified by GFP expression driven by the promotor of the nuclear receptor tailless (day 105 after tumor induction). Depth is given for focal planes. All images: in vivo MPLSM. (MP4 5797 kb)

Confocal microscopy (z-stack) of aIDH1-R132H immunohistochemical staining of a patients’ WHO III° astrocytoma. (MP4 1431 kb)

Images were acquired by time-lapse in vivo MPLSM. Detection of tumor cell calcium transients in S24 GBMSCs by brain superfusion with the small molecule calcium indicator Fluo4-AM (green). At the end, the RFP-expressing GBMSCs of this region are shown to demonstrate the cellular density and morphology of tumor cells. (MP4 2857 kb)