https://orcid.org/0000-0002-6025-7479
Abstract
Glioblastoma cells assemble to a syncytial communicating network based on tumor microtubes (TMs) as ultra-long membrane protrusions. The relationship between network architecture and transcriptional profile remains poorly investigated. Drugs that interfere with this syncytial connectivity such as meclofenamate (MFA) may be highly attractive for glioblastoma therapy.
In a human neocortical slice model using glioblastoma cell populations of different transcriptional signatures, three-dimensional tumor networks were reconstructed, and TM-based intercellular connectivity was mapped on the basis of two-photon imaging data. MFA was used to modulate morphological and functional connectivity; downstream effects of MFA treatment were investigated by RNA sequencing and fluorescence-activated cell sorting (FACS) analysis.
TM-based network morphology strongly differed between the transcriptional cellular subtypes of glioblastoma and was dependent on axon guidance molecule expression. MFA revealed both a functional and morphological demolishment of glioblastoma network architectures which was reflected by a reduction of TM-mediated intercellular cytosolic traffic as well as a breakdown of TM length. RNA sequencing confirmed a downregulation of NCAM and axon guidance molecule signaling upon MFA treatment. Loss of glioblastoma communicating networks was accompanied by a failure in the upregulation of genes that are required for DNA repair in response to temozolomide (TMZ) treatment and culminated in profound treatment response to TMZ-mediated toxicity.
The capacity of TM formation reflects transcriptional cellular heterogeneity. MFA effectively demolishes functional and morphological TM-based syncytial network architectures. These findings might pave the way to a clinical implementation of MFA as a TM-targeted therapeutic approach.
The capacity of TM formation varies between the transcriptional subtypes of glioblastoma.
TM formation depends on axon guidance molecules.
MFA leads to a functional and morphological breakdown of TM-based connectivity.
Glioblastoma cells assemble to a syncytial communicating network based on tumor microtubes (TMs). Here, we report that the tumor cells’ capacity of TM formation reflects transcriptional cellular plasticity and is dependent on axon guidance molecules. Further, we show that meclofenamate (MFA) leads to a functional and morphological breakdown within TM-based spatial network architectures. This breakdown is driven by MFA-mediated inhibition of intercellular cytosolic traffic via gap junctions as well as a downregulation of adhesion and axon guidance molecule signaling which spawns a reduction in TM length and reduces the ability to form cell-cell contacts. Loss of glioblastoma communicating networks was accompanied by a failure in the upregulation of genes that are required for DNA repair in response to temozolomide (TMZ) treatment and culminated in profound treatment response to TMZ-mediated toxicity. Taken MFA as a FDA-approved drug, these findings harbor the potential of bridging the idea of a TM-targeted therapeutic approach from bench to bedside.
Despite intensive research for decades, glioblastoma has remained to rank among the most malignant human neoplasms with a median overall survival time of merely about 17 months.1–3 Since the introduction of the alkylating agent temozolomide (TMZ) to adjuvant radiotherapy in 2005,4 only few improvements in glioblastoma therapy have been reported,5,6 and TMZ still constitutes the cornerstone of chemotherapy for glioblastoma.4 However, in recent years, groundbreaking discoveries both on the morphological and the transcriptional level have been made which are fundamentally revising our understanding of glioblastoma pathogenesis and tumor biology and raise hope for novel conceptual therapeutic implementations.
It has been shown that glioblastoma cells exist in a defined subset of transcriptionally diverse cellular states that recapitulate neural progenitor-like (NPC-like), oligodendrocyte progenitor-like (OPC-like), astrocyte-like (AC-like), and mesenchymal-like (MES-like) molecular signatures.7 Further, glioblastoma cell populations have recently been reported to span from injury response to developmental programs on the transcriptional single-cell level.8 With regard to the capacity of a single glioblastoma cell to regenerate within all transcriptional states, therefore, enabling potential transcriptional adaption and plasticity of subclonal cell populations, intratumoral heterogeneity is considered a major hallmark in fostering therapy resistance and predisposing patients to inferior clinical outcomes.7,9 Morphologically, glioblastoma cells have been shown to be linked in a tight three-dimensional intercellular network arrangement based on ultra-long and thin membrane protrusions—so-called tumor microtubes (TMs)—that extent into the surrounding brain and tumor tissue in order to facilitate long-distance communication.10 On the microscopical level, these intercellular contacts are known to be supported by connexin-43-based gap junctions and the resulting syncytial composition has been suggested to contribute to overcome therapy-induced damage as well as glioblastoma invasion and proliferation capacity.10,11 In addition to their role in facilitating glioblastoma homotypical (between individual tumor cells) cellular networking, TMs have been discovered to interact with neurons as part of the tumor microenvironment via AMPAergic neurogliomal synapses which integrate the tumor cells into neural circuits and promote malignant growth by additional direct synaptic activity.12,13
In the present study, we provide evidence that the tumor cells’ capacity of TM formation varies between the transcriptional subtypes of glioblastoma and depends on axon guidance molecules. Further, utilizing a highly clinically relevant human neocortical glioblastoma slice model, we show that meclofenamate (MFA)—a U.S. Food and Drug Administration (FDA)-approved nonsteroidal anti-inflammatory drug (NSAID)—leads to both a functional and morphological breakdown within TM-based network architectures. This breakdown is driven by MFA-mediated inhibition of intercellular cytosolic traffic via gap junctions, functional connectivity as well as a downregulation of axon guidance molecule signaling. As a result, MFA-induced destruction of glioblastoma communicating networks culminated in a profound treatment response to standard chemotherapeutic agent TMZ. With regard to MFA as a clinically approved drug, these data might pave the way to an instant subsequent clinical implementation of a TM-targeted therapeutic approach in the context of current translational glioblastoma research.
Methods
Ethical Approval
Human sample preparation and analysis were approved by the local ethics committees of the Universities of Freiburg and Ulm (Freiburg: protocol 100020/09 and 472/15_160880; Ulm: 162/10) with written informed consent obtained from all subjects. The studies were approved by the respective institutional review board.
2D Cell Cultures and Viral Transduction
Primary human glioblastoma cell populations (BTSC#1, BTSC#G35, BTSC#168, BTSC#233) were purified from surgical specimens as previously described.14 For details regarding cell culture and transduction with the lentiviral particle rLV.EF1.ZsGreen1-9 and LV-CAG-GCaMP6f, see Supplementary Material.
Human Organotypic Slice Culture and Tumor Injection
We made use of a human neocortical slice model which was recently described in detail.15 In case of tumors deeper localized without infiltrating the cortex, we used cortical regions that were mandatorily removed to access the deeper localized tumor. These cortex cubes were resliced into 300 µm thick slices and cultured as recently described.15 Tumor injection and imaging were carried out as described in the Supplementary Material. Simple Neurite Tracer (SNT) plugin16 implemented in ImageJ was used to visualize cell-cell connections. We transformed the traces into a 3D matrix and visualized the network pattern in plotly (R software).
Cell Morphology and Cell Connectivity Analysis
Computational analysis of cell morphology and cell-cell contact was analyzed using Cell Profiler, Ilask, and R software. For details on the experimental setup, see Supplementary Material.
Calcein Dye Transfer
To examine intercellular cytosolic traffic, the redistribution of calcein fluorescent dye that passes through connexin-43 connected cells was measured. For details on the experimental setup, see Supplementary Material.
Calcium Imaging
For assessment of cellular network connectivity, transfected BTSC#233-GCaMP6f cells were recorded and analyzed by MATLAB and R software. For details on the experimental setup, see Supplementary Material.
RNA Sequencing
After cells were treated for 48 hours, RNA was purified using the RNeasy Mini Kit (Qiagen, Hilden, Germany). For the preparation of RNA sequencing on the MinION Sequencing Device (Oxford Nanopore Technologies, Oxford, UK) with MinKNOW software, the PCR Barcoding Kit and cDNA-PCR Sequencing Kit were utilized according to the manufacturer’s instructions. For a detailed description of data postprocessing, see Supplementary Material.
Animal Experiments
Experimental setup of murine experiments as well as details on magnetic resonance and PET-CT imaging are outlined in the Supplementary Material.
Supporting methods, information on statistics, and a reagent list can be found in the Supplementary Material.
Results
Cell-Cell Connections and Tumor Cell Membrane Protrusions Vary Between the Transcriptional Subtypes of Glioblastoma
To decipher links between cellular morphology and gene expression subtypes, we extracted single-cell morphological features (Supplementary Figure S1) and performed hierarchical clustering. Along with the first and second principal components, we observed two major subgroups spanning from closely connected cells characterized by multiple TMs to less connected cells and the absence of TMs (Figure 1A). Unexpectedly, a heterogeneity of these morphological subgroups with a varying distribution of high and low connected cells was found within the examined cell lines (Figure 1B). Next, we performed RNA sequencing and aligned all cell lines into the 4-subtype classification (Neftel) (Figure 1C), or the most recent cell line-based classification of glioblastoma which encompasses a gradient from developmental to injury-related transcriptional programs8 (Figure 1D). By rearranging the fraction of high-connected cells accordingly to the transcriptional phenotype, we observed an association between high-connected cells and the expression of developmental programs (Figure 1D). This was confirmed through the Neftel classification in which lineage differentiation states such as NPC-, OPC-, or AC-like were overrepresented in the group of high-connected cells (Figure 1B, C).
Cell-cell connections and tumor cell protrusions vary between the transcriptional subtypes of glioblastoma.
(A) Analysis of cell morphology of 6 different cell populations. A scatterplot illustrates clustering and PCA dimensional reduction. Colors indicate the cluster and different cell populations are illustrated with distinct shapes. (B) Barplot of distribution of cell populations within high- and low-connectivity clusters. (C) Scatter plot indicating the transcriptional state of each cell population in accordance with the classification of Neftel et al. (D) Alignment of cell populations (indicated by color as in B and C) along the gradient between injury response and developmental programs.8 Barplots at the bottom indicate the distribution of transcriptional subtypes in the high-connectivity cluster. (E) Scatter plot of axon guidance genes obtained from scRNA-seq dataset of Neftel et al. with 4-state classification (left) and gradient between injury response and development (right). (F) Gene expression plots of the published dataset15 from two representative patients showing the spatial distribution of injury response and developmental signature. (G) Correlation analysis of gene expression signatures and expression of axon guidance molecules. (H) Workflow illustration. Resected cortex specimens were sectioned into slices and ZsGreen-tagged primary glioblastoma cell populations (n = 3) were injected. After 7 days, imaging and 3D reconstruction of microtube formation were performed. (I) 3D modeling of network architecture of three primary glioblastoma cell populations. Points indicate cell somata and populations represent detected TMs.
AC-like, astrocyte-like; BTSC#168, BTSC#1, BTSC#233, primary glioblastoma cell populations; MES-like, mesenchymal-like; NPC-like, neural progenitor-like; OPC-like, oligodendrocyte progenitor-like; PCA, principal component analysis; TM, tumor microtube.
In the development of the central nervous system, directed migration of neurons as well as spatial configuration of astrocytes are mediated by a specific class of proteins summarized as axon guidance molecules. Osswald and colleagues10 assumed that axon guidance molecules might be involved in the regulation of forming cell-cell contacts, hence, we hypothesized that the different network architectures might be caused by a heterogeneous presence of axon guidance molecules. In order to verify our hypothesis, we analyzed gene expression profiles of all major classes (Ephrin: EPHA4/2; Semaphorins: SEMA3E, SEMA6A, SLIT: SLIT1, and Netrin: NTNG1) of known axon guidance molecules using the publicly available scRNA-seq dataset from Neftel and colleagues7 (Figure 1E). We mapped the cells into the 4-state classification and along the development-injury response gradient. Both classifications suggest that axon guidance molecules were stronger expressed in lineage differentiated states (NPC-, OPC-, or AC-like) with enriched expression of developmental programs. To investigate to what extent the presentation of injury response and developmental programs are heterogeneous distributed in space, we analyzed the published dataset from Ravi et al.,17 which showed a large heterogeneity of both transcriptional programs (Figure 1F), with a stronger correlation (R2 = 0.48, P < .001), of axon guidance expression and the developmental signature (Figure 1G).
To investigate the cellular morphology directly in the human brain with a preserved neural environment, we injected three of our primary patient-derived glioblastoma cell populations (BTSC#168 and BTSC#233 with high percentage of connected cells, BTSC#1 with low percentage of connected cells) into human neocortical slices, as recently described,15,18 workflow is illustrated in Figure 1H. After a period of 7 days, we performed multiphoton laser-scanning microscopy and traced the membrane protrusions as described by Osswald and colleagues10 in order to model the network architecture of different transcriptional glioblastoma subtypes (Figure 1I). BTSC#168 and BTSC#233 displayed a low response of injury transcriptional programs and showed a pronounced sparse density and a high number of cell-cell connections based on membrane protrusions; areas of higher cellular density formed palisading shapes similar to reactive astrocytes (Figure 1I). The cell population with the highest activation of injury responses and mesenchymal gene expression showed a cluster-like architecture, resulting in multiple cell hubs with high cellular density (Figure 1I). These results largely overlap with our findings from 2D cell-layer cultures. Our data suggest that morphological intercellular connectivity—reflected by the numbers of membrane protrusions as well as cellular density in regions occupied by tumor cells—differs depending on the transcriptional glioblastoma subtype. We hypothesize that cellular morphology and the capacity of forming membrane protrusions also known as TMs are subject to transcriptional heterogeneity and depend on axon guidance molecules.
MFA Causes Functional Decoupling of Glioblastoma Cells
TM formation has been proposed as a major hallmark for the evolution of a gap junction-coupled syncytial communicating network.10 In a recent study, MFA was used to alter glioblastoma intercellular connectivity based on the inhibition of calcium signaling along with preformed tumor networks.12 The actual mechanism by which MFA impairs intercellular functional coupling has been ascribed to a potential ability to block connexin-43-based gap junctions, but this has not been conclusively clarified, yet. We therefore aimed to investigate to what extent MFA might affect gap junction-mediated intercellular cytosolic traffic as well as electrical coupling and functional network evolution in glioblastoma. We quantified gap junction-mediated cytosolic traffic within tumor networks by real-time imaging of fluorescence-guided cell-to-cell transfer of calcein—a fluorescent molecule that is only able to spread from cell-to-cell through intercellular gap junctions, workflow is depicted in Figure 2A. For MFA treatment, we observed a significant reduction of calcein receiver cells after up to 150 minutes (Figure 2B, C; Supplementary Figure S2). In line with these findings of a MFA-mediated functional inhibition of intercellular cytosolic exchange via gap junctions, calcium imaging of glioblastoma cells containing a virally transduced genetically encoded calcium indicator (GCaMP) was performed. We used a glutamate stimulation of 100 µM for 10 minutes to analyze the network communication in tumor cells. After 5 minutes of recording, we added MFA which revealed a reduction of calcium signaling after MFA treatment (Figure 2E). The glioblastoma networks showed properties reminiscent from neuronal networks, with bursts of synchronized activity, with all the cells participating in the network oscillation (Figure 2E). In order to simplify the functional structure observed within the cellular network, it is easiest to represent them as graphs, where the interacting cells are designated as nodes and their functional interactions as edges. We defined functional interactions as subsequent calcium events, detected in adjacent cells within a narrowly defined radius (50 µm) and timeframe (2.5 seconds). When the cellular nodes were filtered based on connectivity, measured by the frequency of edges passing each node, we were able to re-identify the “activity hubs” of the glioblastoma network (Figure 2F). The correlation analysis of frequency of interaction vs nodal connectivity results in the coefficient of determination, R2 of 0.87, which reached the criteria of a scale-free biological network (Figure 2G). After MFA treatment, we observed a significant reduction in the coefficient of determination (R2 = 0.734), which suggests that the electrical interactions post-signaling inhibition is random, lacking structural organization and activity (Figure 2G, H).
Meclofenamate causes functional decoupling of glioblastoma cells.
(A) Workflow illustration. Donor cells contain calcein, a fluorescent dye that passes only through connexin-43-connected cells. (B) Fluorescence images of BTSC#G35 were acquired by the IncuCyte® S3 Live-Cell Analysis System. Fluorescence intensity enables to distinuish between donor cells (green in the center) and receiver cells (whole red). (C) Quantification of the increase of receiver cells after MFA treatment compared to the control. Experiments were performed in triplicate with BTSC#G35. Data are given as mean ± SEM. P values are determined by the Wilcoxon matched-pairs signed rank test. (D) Workflow illustration. Calcium imaging of GCaMP transduced glioblastoma cells. (E) Recordings of BTSC#233 glioblastoma cells stimulated with glutamate (100 µM, 10 minutes) solely (left panel) and the same cells after glutamate stimulation and MFA (right panel). Change of fluorescence is given in ΔF/F(t)). Synchronous bursts are highlighted in red boxes. (F) Measurement of cellular connectivity in the same conditions (Ctr + glutamate, MFA + glutamate, BTSC#233) for 20 minutes. Lines indicate the connections between cells defined by consecutive and spatially dependent events. The estimated connectivity is illustrated as thickness of the lines. Cells in which the first local speeding event was generated are colored in green and defined as hubs. Hub-associated cells mainly acting as receivers are colored in red. (G) Estimation of scale-free topology as a measurement of functional networks. Scale-free topology is estimated by the R2 of the log10(connectivity) and log10(frequency of events). (H) Histogram of the number of cells with a different number of events.
MFA Causes Reduction of NCAM and NETRIN Signaling
Based on the fact that MFA inhibits functional coupling of glioblastoma cells, we next analyzed gene expression changes caused by MFA treatment. We identified 301 differentially expressed genes (DEGs) in the BTSC#168 cell populations (Padj < .01), 452 DEGs in BTSC#233, and 361 DEGs in the BTSC#G35 cell populations of which 163 genes are commonly differentially expressed (Figure 3A, B). A gene set enrichment analysis identified a significant reduction of two pathways that are involved in neural development and cell-cell connections: the neural cell adhesion molecule (NCAM) pathway (MSigDM REACTOME NCAM1_INTERACTIONS) and the NETRIN-signaling pathway that is involved in axon guidance (MSigDB REACTOME_NETRIN_1_SIGNALING) (Figure 3C, E). NCAM1, a major important adhesion molecule in the brain was significantly downregulated as well as SEMA3A, a protein that is expressed by astrocytes of the spinal cord and is required for proper motor neuron and sensory neuron circuit organization19 (Figure 3E). Additionally, we observed a significant upregulation of MAPK activity and WNT signaling under MFA treatment (Figure 3D), which was also suggested by enhanced phosphorylation on the protein level (Figure 3F). Based on the findings that axon guidance molecules were mainly observed in cells of pronounced expression of injury response programs and the knowledge of tumor cells dynamically changing their cellular state, we measured the adaptation of signature genes and found a shift toward an enriched injury response gene expression profile in response to MFA treatment (Figure 3G).
Meclofenamate causes reduction of NCAM and NETRIN signaling.
(A) Volcano plots of differentially expressed genes (DEGs) in BTSC#168, BTSC#233, and BTSC#G35. Red dots indicate significant DEGs defined as Padj < .05. (B) Venn diagram shows the number of DEGs of each cell population separately and the number of shared DEGs. (C) Gene set enrichment analysis (GSEA) of downregulated pathways after MFA treatment. (D) GSEA of upregulated pathways after MFA treatment. (E) Barplot of NCAM1 and SEMA3A in BTSC#233. (F) Protein array of MFA-treated cells. (G) Scatter plot showing the enrichment of signature genes in accordance with the classification of Richards et al. indicating the dynamic change of cellular states following MFA treatment.
BTSC#233, BTSC#G35, primary glioblastoma cell populations; MFA, meclofenamate.
MFA Demolishes TM-Based Intercellular Network Morphology
With regard to our findings of TM formation to be dependent on axon guidance molecule expression as well as the observation that MFA treatment significantly downregulates signaling pathways involved in axon guidance, we asked to what extent MFA might impair TM formation and therefore alter spatial glioblastoma network architecture morphology. In order to address these questions, we injected the cell populations BTSC#168 and BTSC#233 into human cortical brain slices and treated the slices with MFA. MFA treatment resulted in a breakdown of intercellular network architectures (Figure 4A, B), which was reflected by a reduction in the number of TM-based intercellular connections per cell (Figure 4C), as well as a reduction in TM length, whereas the number of TMs per cell was not significantly affected (Figure 4D). Further, MFA-mediated impairment of glioblastoma intercellular TM networks yielded a decrease of tumor bulk expansion compared to untreated control populations (Figure 4E).
Meclofenamate demolishes tumor microtube-based intercellular network morphology.
(A) 3D reconstruction of TMs formation in human slice cultures for BTSC#168 and BTSC#233 both for MFA and control. TMs length is color-coded as depicted in the heatmap. (B) Stacked fluorescence multiphoton images of the slice cultures. (C) Quantitative statistics showing the number of cellular connections for MFA treatment and control. P values are determined by one-way ANOVA. Mean ± SD is depicted. (D) Violin plots displaying the TMs length for BTSC#168 and BTSC#233 (n = 350), and barplots depicting the average number of TMs per cell after MFA treatment compared to the control. Results were averaged from triplicates and are presented as the mean ± SEM. P values are determined by unpaired t test. (E) Quantification of relative tumor expansion at days 1, 3, and 7 after injection of primary glioblastoma cell populations. Mean ± SD is depicted. P values are determined by one-way ANOVA. *, **, and **** denote P < .05, P < .01, and P < .0001.
BTSC#168, BTSC#233, primary glioblastoma cell populations; MFA, meclofenamate; TM, tumor microtube.
Reduced Connectivity Due to MFA Treatment Results in Profound TMZ Sensitivity
We propose that MFA inhibits cell-cell contacts by two mechanisms: on the one hand, a direct connexin-43 inhibition which was also reported by several authors,12,20 on the other hand, a downregulation of adhesion molecules and proteins required for axon guidance which reduces the cells’ capacity to form morphological cell-cell connections. As intercellular points of contact are not only points of adhesion, but also origins of complex signaling pathways that, among others, mediate survival,21 we asked to what extent the cellular network is required for resistance against chemotherapy and if these networks contribute to repair damaged cells. As reported by others,22 TMZ treatment was accompanied by a significant upregulation of genes mediating DNA repair which is required to overcome TMZ-induced cell death (Figure 5A, B). However, for combined treatment, we observed a failure in the upregulation of these genes (Figure 5B), which culminated in significantly increased cell death rates (Figure 5C–F), and revealed a tendency for a survival benefit for combination treatment in a murine in-vivo setting for the chosen observation time period of 8 weeks after treatment onset (Figure 5G, H; Supplementary Figure S3). Analysis of morphological tumor networks in the neocortical glioblastoma slice model confirmed a breakdown of TM length as well as a significant reduction of tumor volumes for combination vs TMZ single treatment (Figure 5I–L).
Reduced connectivity due to meclofenamate treatment results in profound temozolomide sensitivity.
(A) Heatmap of gene expression for BTSC#233 and BTSC#G35 after TMZ mono-treatment and additional MFA administration. Heatmap is based on significant DEGs wherein low expressed genes are indicated in blue and high expressed genes in red. (B) GSEA of DNA damage pathway (MSigDB). TMZ mono-treatment is marked in yellow and TMZ + MFA combination treatment in green. For statistical analysis, see Methods section. (C) Line plot displaying the time-dependent increase in the number of cells under different treatment conditions for BTSC#G35. Data points are given as mean ± SEM. P values are determined by the Wilcoxon matched-pairs signed rank test. (D) Workflow illustration. Cells were treated for 144 hours and stained with propidium iodide (PI) for FACS analysis. DNA fragmentation served as readout for cell death. (E) Representative scatter plots from flow cytometric analysis of BTSC#G35. The SubG1 peak is accentuated within the corresponding histograms and the mean percentage of DNA fragmentation is depicted below. (F) Barplots showing the percentage of specific DNA fragmentation of PI-stained nuclei as a surrogate for cell death in two primary glioblastoma cell populations under different treatment conditions. Data are given as mean ± SEM. P values are determined by the Mann-Whitney U test. (G) Representative 11C-MET PET scans of mice brains illustrate the uptake of this tracer following treatment with TMZ and TMZ + MFA compared to the controls. Scale bar 2.5 mm. Scale for SUV shown on the left side and background/tumor SUV is shown next to the PET images. (H) Kaplan-Meier curves showing percentage survival in days. BTSC#G35 cells were orthotopically implanted into NOD-SCID mice. One-way ANOVA with Bonferroni correction for multiple testing, ***P < .001 and ****P < .0001. (I) Workflow illustration. Resected cortex specimens were sectioned into slices and ZsGreen-tagged primary glioblastoma cell populations were injected. After 7 days, imaging and 3D reconstruction of microtube formation were performed. (J) Stacked fluorescence multiphoton images of the slice cultures injected with BTSC#233 and treated with TMZ alone and the combination of TMZ + MFA. (K) Violin plots with integrated boxplots displaying the microtube length for BTSC#233 after TMZ treatment and additional MFA administration (n = 144). P values are determined by unpaired t test. (L) Quantification of relative tumor volume for BTSC#233. Mean ± SEM is depicted. P values are determined by unpaired t test. *, **, and **** denote P < .05, P < .01, and P < .0001. BTSC#G35, BTSC#168, BTSC#233, primary glioblastoma cell populations; DEGs, differentially expressed genes; GSEA, gene set enrichment analysis; FACS, fluorescence-activated cell sorting; MFA, meclofenamate; TMZ, temozolomide; PI, propidium iodide; SSC-A, side scatter area.
Discussion
Current insights into cancer biology comprise the perception of a three-dimensional tumor architecture that drives malignancy by structurally and functionally encoded information.23,24 The importance of mutual interactions of tumor cells within an established network is attributed to many critical functions for maintaining and promoting malignancy. Of note, tissue architecture has been suggested to imply both the consequence and the cause of physiological and pathological tissue morphogenesis, development, and differentiation—an understanding of cancer that offers the therapeutic concept of fighting a tumor’s spatial network morphology.25,26 In glioblastoma, recent evidence suggests this syncytial spatiality to be driven by TMs as ultra-long characteristic membrane protrusions that enable glioblastoma cells to interconnect into a complex multicellular network reminiscent of a functional organ.10,27 Here, we report that the spatial growth pattern and morphological properties of glioblastoma cells in relation to their transcriptional origin are crucial to understand the complex network structures. To preserve the natural tissue architecture of a human brain and avoid an interspecies bias, we made use of a neocortical slice model which has been evaluated most recently.15,18 TM tracing of glioblastoma cells in neocortical slices yielded a complex three-dimensional tumor architecture that ranged from a spatial sparse-dense pattern of tumor cells with a large number of TM-based intercellular connections in AC-like cells to a cluster-based architecture in MES-like cell populations. Similar findings have not been reported in the literature so far, highlighting the gap of knowledge of transcriptional-morphological relationships in glioblastoma. We found that axon guidance molecules which are known to mediate neural axon guidance, branching morphogenesis, and spatial configuration of astrocytes28–30 were differentially expressed across the dynamical transcriptional states of glioblastoma. With regard to an observed enrichment of developmental programs in highly connected cells as also suggested in a previous work by Xie et al.,31 our data suggest that the tumor cells’ capacity of TM formation reflects transcriptional cellular heterogeneity and is dependent on the expression of axon guidance molecules. Recently, evidence was shown that damaged cells can be repaired or rebuilt by use of cellular connections which was found to be an important mechanism for therapy resistance of glioma.10
Using MFA, we observed a demolition of spatial network morphology on a global scale which was reflected by a profound reduction of TM length as well as a quantitative reduction of TM-based cell-cell connections. So far, MFA was reported as a connexin-43 blocker, however, detailed information regarding its transcriptional changes and impact on morphology are unknown. By exploring the transcriptional changes due to MFA treatment we identified a significant reduction of the NETRIN- and NCAM-signaling pathways suggesting that MFA treatment causes dysregulation of axon guidance and cell adhesion resulting in morphological and functional disbalance. These axon guidance molecules play an important role in tissue development, branching morphogenesis, and the evolution of three-dimensional structures within organogenesis.32,33 In glioblastoma, NETRIN-1 has been shown to induce the growth of thin and elongated capillary sprouts therefore driving blood vessel outgrowth and recruitment in the course of malignant angiogenesis.34 To overcome spatial formation and cellular connections, a loss of these mechanisms is essential to leave these tumor cells more vulnerable and open to attack. SEMA3A, another known axon guidance molecule, that was significantly downregulated under MFA treatment in our study, promotes glioblastoma cell dispersal by modulating substrate adhesion as well as to elevate vascular permeability resulting in enhanced tumor-induced vascular leakage.34,35 Our data suggest a novel function of axon guidance molecules in glioblastoma biology—that is for intercellular TM formation and, in consequence, for the establishment of a three-dimensional network architecture.
We propose MFA as the first TM-targeted FDA-approved drug that prevents TM morphogenesis and outgrowth resulting in a morphological breakdown of glioblastoma spatial interconnecting cellular arrangements. In line with these findings, we observed a MFA-induced shift of transcriptional gene expression toward an enriched level of MES-like or injury response cellular signatures which in turn confirmed our initial observation of a reduced TM-based connectivity in MES-like spatial network architectures.7 Cells within a network were assumed to be in a protected environment, thus their ability to move outside the network is reduced.11 Environmental stress such as metabolic imbalances or tumor treatments causes increased stress signals (MAPK pathway) which leads to a transcriptional shift toward a more mesenchymal state36 which enables the cells to escape the network.
Beyond the observed profound effects on glioblastoma morphology, MFA treatment induced an impairment of intercellular electrical coupling as previously reported12 as well as a significant reduction of gap junction-mediated intercellular cytosolic traffic. These findings are in line with several reports that indicate MFA to functionally block gap junctions in various physiological cell entities.37,38 Previous reports have shown that connexin-43 knockdown in primary glioblastoma cells resulted in reduced intercellular TM-based morphological connectivity10 and in-vivo experiments have shown MFA to reduce tumor cell proliferation and tumor growth.12 We therefore propose MFA-mediated destruction of glioblastoma networking to be the result of both a direct connexin-43 inhibiting effect and the abovementioned transcriptional changes. We assumed that the destruction of network structures is associated with a higher vulnerability of unconnected cells to chemotherapy. When exposed to alkylating chemotherapy, reduced functional and morphological connectivity due to MFA treatment was accompanied by a failure in the upregulation of DNA repair signaling which culminated in increased sensitivity to TMZ-mediated toxicity. These findings add to the growing perception of a TM-based network architecture to protect tumor cells against antitumoral effects of cytotoxic therapy.10,11,39,40 Though the exact mechanisms are far from being understood, it has been speculated that syncytial network arrangements might facilitate single tumor cells to prevent critically high increases of intracellular calcium- or drug concentrations.10,11 Glioblastoma cells might also hijack nonmalignant astrocytes of the tumor microenvironment via gap junction-mediated intercellular signaling to contribute to the protective syncytial structures as it has been demonstrated in mouse models of brain metastasizing breast and lung cancer.20 Further, TMs have been shown to interact with neurons through neurogliomal AMPAergic synapses which integrate the tumor cells into neural circuits and foster the malignant growth of these tumors by synaptic activity.12,13 Our analysis showed that the combined therapy resulted in an improved response to TMZ in all cell lines, but with significant differences within the different cell lines with improved response of cells showing increased activation of developmental programs. We assume that MFA provokes a transcriptional shift toward injury response programs which leads to a better response to TMZ, which explains the overall response to the combined therapy regardless of the transcriptional phenotype.
In summary, we suggest that the capacity of TM formation as a morphological construction unit for the evolution of glioblastoma spatial network architectures is dependent on proteins required for axon guidance and branching morphogenesis therefore unveiling a novel role for axon guidance molecules in glioblastoma biology. Further, we show that MFA leads to both a morphological and a functional breakdown within these network arrangements and that this breakdown is driven by MFA-mediated inhibition of TM outgrowth and intercellular communication via two mechanisms: on the one hand, a direct connexin-43 inhibition resulting in significantly reduced intercellular gap junction-mediated cytosolic traffic, on the other hand, a downregulation of adhesion as well as axon guidance molecules which spawns a reduction in TM length and reduces the ability of forming cell-cell contacts. Loss of TM-based network arrangements was accompanied by a failure in the upregulation of genes that are required for DNA repair in response to TMZ treatment and culminated in a profound sensitivity to TMZ-mediated cell death. With regard to MFA as a FDA-approved drug, these findings harbor the potential of bridging the idea of fighting glioblastoma’s spatial TM-based network architectures from bench to bedside. Registered as a NSAID in the United States, MFA is applied for the treatment (pain relief) of bone/joint diseases (such as rheumatoid arthritis, osteoarthritis, and acute pain shoulder), dysmenorrhea, and fever and adverse events are the ones common to NSAIDs. Furthermore, systemic administration of MFA has been reported to exhibit CNS-mediated anticonvulsive effects in murine epilepsy models41 suggesting MFA to be capable of sufficiently crossing the blood-brain barrier. However, up to date, there is a lack of studies evaluating concentration levels of MFA within the human brain which seems to be of utmost importance in view of a translational scope of MFA in the course of an expanded oncological field of application. Against this backdrop, there is an ongoing US trial using MFA monotherapy in patients with recurrent/progressive brain metastases from solid primary tumors (NCT02429570). Based on the experimental findings of the present work, a nation-wide phase I/II trial of MFA/TMZ combination therapy in relapsed MGMT-methylated glioblastoma (“MecMeth” EudraCT2021-000708-39) is being initiated in Germany which will measure concentrations of MFA within the malignant human tumor and evaluate safety as well as feasibility of a combined MFA/TMZ approach and may gain first insights into the efficacy of MFA as the potential first clinically feasible TM-targeted drug.
Funding
M.S. was funded by Bonfor and the Familie Mehdorn Stiftung. M.S. was supported by a junior research program within the Mildred-Scheel School of Oncology (MSSO) Cologne-Bonn (project-ID: 70113307) funded by the German Cancer Aid. D.H.H. was funded by the Else Kröner-Fresenius-Foundation. The work is part of the MEPHISTO project funded by the BMBF (Bundesministerium für Bildung und Forschung) (project number: 031L0260B). M.C. was funded by The Norwegian Cancer Society (grant #190170) for the in-vivo experiments.
Conflict of interest statement. No potential conflicts of interest were disclosed by the authors.
Authorship statement. Conceptualization and design: M.S., U.H., D.H.H. Development of methodology: M.S., A.L.P., V.M.R., B.O.E., A.D., D.H.H. Data acquisition: M.S., A.L.P., L.V., B.O.E., J.K., P.W., N.N., P.d’E., D.Z., P.S. Slice model experiments: V.M.R. Bioinformatical analysis: M.S., A.L.P., U.G.H., D.H.H. Murine experiments: M.A.R, S.S., P. Ø. E., M.C. Visualization: M.S., A.L.P., D.H.H. Writing—original draft: M.S., A.L.P., L.V., D.H.H. Writing—review and editing: M.S., A.L.P., B.O.E., M.M.L., U.G.H., E.G., T.P., P.S., O.S., M.A.W., J.B., H.V., A.W., U.H., D.H.H. Supervision: M.S., M.A.W., H.V., A.W., U.H., D.H.H. Funding acquisition: M.S., D.H.H.
Data Availability Statement
The authors declare that the data within this manuscript are available in the main and/or supplementary figures or from the corresponding authors upon reasonable request.
References
Author notes
These authors shared the first authorship.
These authors shared the last authorship.
