Glioblastoma-Astrocyte Connexin 43 Gap Junctions Promote Tumor Invasion

Abstract

Glioblastoma Multiforme (GBM), classified as WHO grade IV astrocytoma, is the deadliest adult cancer of the central nervous system. An important contributing factor to poor survival rates in GBM is extensive invasion, which decreases the efficacy of resection and subsequent adjuvant therapies. These treatments could be markedly improved with increased resolution of the genetic and molecular initiators and effectors of invasion. Connexin 43 (Cx43) is the principal astrocytic gap junction (GJ) protein. Despite the heterogeneity of GBM, a subpopulation of cells in almost all GBM tumors express Cx43. Functional GJs between GBM cells and astrocytes at the tumor edge are of critical interest for understanding invasion. In this study we find that both in vitro and in ex vivo slice cultures, GBM is substantially less invasive when placed in a Cx43-deficient astrocyte environment. Further, when Cx43 is deleted in GBM, the invasive phenotype is recovered. These data strongly suggest that there are opposing roles for Cx43 in GBM migration. We find that Cx43 is localized to the tumor edge in our ex vivo model, suggesting that GBM-astrocyte GJ communication at the tumor border is a driving force for invasion. Finally, we find that by a Cx43-dependent mechanism, but likely not direct channel-mediated diffusion, miRNAs associated with cell-matrix adhesion are transferred from GBM to astrocytes and miR-19b promotes invasion, revealing a role for post-transcriptional manipulation of astrocytes in fostering an invasion-permissive peritumoral niche.

Introduction

Glioblastoma Multiforme (GBM) is the most common and most aggressive adult brain cancer. GBM prognoses are dismal, with a 5-year survival rate of below 5 percent¹. Poor GBM outcomes are driven by almost certain recurrence², resulting from cellular heterogeneity³, resistance chemo/radiotherapeutics^{4, 5}, and extensive invasion at the tumor periphery⁶. There has yet to be substantial progress made limiting GBM invasion, and therefore it is necessary to identify and refine our understanding of molecular targets that play a role in the “decision” of GBM cells to infiltrate surrounding tissue.

Glioblastoma invasion occurs preferentially along brain tracts, namely white matter and vasculature, but also through the parenchyma⁷. Initiation and duration of invasion have many determinants including chemotaxis within the peritumoral space⁸ and adhesion between GBM cells and extracellular matrix (ECM) moieties^{9, 10}. Evidence suggests communication, both diffusive and direct between cancerous cells and their non-malignant neighbors promotes GBM invasion. Gap junction (GJ)-mediated communication has been shown to play a part, but its role is incompletely understood ^{11, 12}.

Though it exhibits a high degree of cellular heterogeneity, by its most broad definition, GBM is a WHO grade IV astrocytoma, and as such, has a large proteomic overlap with native astrocytes. GBM expresses Connexin 43 (Cx43) and Connexin 30 (Cx30)^{13, 14}, astrocyte GJs, and functional GBM-GBM and GBM-astrocyte GJs have been observed in vitro and in vivo¹⁵. In human tumor biopsies, Cx43 expression decreases as astrocytoma grade increases¹¹ and its distribution within the tumor may change^{12, 16}. A decrement in Cx43 expression with increasing tumor grade indicates a correlation between Cx43 loss and invasion^{17, 18}. Conversely, Cx43-dependent GBM-astrocyte communication is a contributing factor to GBM invasion^{19, 20}. Data presented here reveal that Cx43 serves opposing roles in GBM invasion, where loss of expression in the tumor core and increased expression or redistribution of Cx43 at the tumor-astrocyte interface synergistically promotes infiltration.

One way by which GBM cells manipulate their adjacent and adjoining non-cancerous neighbors is through miRNA transfer²¹. miRNAs are post-transcriptional repressors and can greatly alter cellular phenotype. Many miRNAs, most notably hsa-miR-21, are upregulated in GBM tumors, and are detectable in the blood of GBM patients²². miRNAs are found in tumor exosomes and there is some evidence that cancer-associated miRNAs can be transferred via exosomes or other means to non-cancerous cells²³, suggesting GBM may be responsible for manipulating translational profiles of nearby cells to promote an invasion-permissive environment²⁴.

In this study we utilize both an in vitro and novel ex vivo organotypic slice culture model to demonstrate the coupling of GBM cell lines with astrocytes and its contribution to GBM invasion. Deleting Cx43 in astrocytes severely diminishes invasion of U87 cells in our ex vivo model. Additionally, we report Cx43-dependent transfer of specific miRNAs, and a role for hsa-miR-19b in progression of an invasive GBM phenotype.

Methods

Cell lines

U87 (ATCC, #HTB-14. From Dr. Jeffrey Segall, Albert Einstein College of Medicine. Obtained and authenticated 2018, P5) and murine GBM line GL261 (Gift, Jeff Segall, obtained and authenticated 2018, P2) were maintained in DMEM (Gibco, #11965–092) plus 10% fetal bovine serum (FBS) (Gibco #10437028) and 1% antibacterial-antimycotic (AA) (Gibco, #15240062). U87-MG spontaneously form tumor spheroids at 7–10 days in culture. Immortalized wild-type cortical astrocytes (IWCA) and immortalized Cx43 knockout cortical astrocytes (IKOCA)²⁵ were maintained in astrocyte medium, low-glucose DMEM (Gibco, #11885084) supplemented with 10% FBS and 1% penicillin-streptomycin (Gibco, #15140122). IWCA and IKOCA were used from passage 6 to 12. Cell lines were not tested for mycoplasma.

Primary astrocytes

Wild-type (WT) primary murine cortical astrocytes were obtained from C57BL/6J (Jackson Laboratory, C57BL/6J, #000664) pups at post-natal day 1 using standard dissociation protocol approved by the Einstein IACUC (#20180302). Briefly, pups were decapitated, and whole brains were removed and placed in ice-cold 1X phosphate buffered saline (PBS) (Gibco, #10010023). Cerebral cortices were dissected out after removal of meninges and chopped into fine pieces. Tissue was digested in 0.05% trypsin (Gibco, #25300054), moderately agitating the solution, and allowing to rest for 5 min. An equal volume of astrocyte culture media was added to the dissociation tube for resuspension. Samples were centrifuged at 2000 rpm for 10 minutes. Supernatant was removed, cells were resuspended in culture medium, and plated in T25 tissue culture flasks (Falcon, #353014). 14 days post-plating samples were shaken at 37°C, 180 rpm (INFORS HT, Ecotron, #CH-4103) for 12 hours to detach residual microglia and oligodendrocyte precursor cells. Supernatant was removed and cells were washed 3 times with 1X PBS before fresh medium was added. Cells rested for 3 days before splitting for experiments. After isolation, primary astrocytes were maintained in astrocyte medium and used for experiments up to passage 3.

Novel U87 cell lines

Three U87-MG cell lines were developed for these studies; cytosolic mCherry (U87^wt), mCherry CRISPR Cx43 deletion (U87^{Cx43 KO)}, and mCherry CRISPR scrambled guide RNA (sgRNA) (U87^scram). Cell lines were established by lentiviral transduction (LVP). For cytosolic mCherry, LVPs (Genecopoeia, #LP441-025) were added to 70–80% confluent U87-MG cultures at MOI 2, 5, and 10 in media supplemented with 10 μg/mL polybrene (Sigma Aldrich, #TR-1003-G). At 2 days, media was replaced with selection media, DMEM with 10% FBS 5 μg/mL puromycin (Sigma, #P9620). Cells were maintained in selection media for >14 days and individual colonies were selected for monoclonal lines. A GL261 mCherry cell line was also established via the U87^wt protocol. For each CRISPR line, 70–80% confluent U87-MG cultures were transduced with Cas9 LVPs (generated with Genecopoeia, #CP-LvC9NU-01) at MOI of 5 and 10. Selection was performed at 2 days post-transduction with 300 μg/mL G418 (Gibco, #10131035). Colonies were selected at >14 days. Stable Cas9 lines were then transduced with Cx43 targeted sgRNA (Genecopoeia, #HCP288502-LvSG03-3-B) or scrambled sgRNA (Genecopoeia, #CCPCTR01-LvSG03-B) at MOI of 5 and 10. Clones were established by 14 days with 5 μg/mL puromycin. CRISPR LVPs were generated by transfection of HEK293T cells for 4–7 days with collection of cell culture media. Lentiviral titer was obtained by Lenti-X qRT-PCR Titration Kit (Takara, #631235). Cx43 expression levels were evaluated by Western Blot as described.

Western Blotting

Cx43 CRISPR deletion was confirmed by Western blot. U87^scram and U87^{Cx43 KO} cells were cultured to confluence in triplicate 60mm petri dishes (Falcon, #353004). Cell lysates were obtained by scraping in lysis buffer (dH2O, NaHCO3 (1 mM), PMSF (2 mM), Na3VO4 (1 mM), EDTA (5 mM), Protease Inhibitor 1X) and frozen at −80°C at least 12 hours. Cell lysates were thawed on ice and quantified for total protein concentration by BCA kit (Pierce, #23228). Protein was mixed at equal volume with 2X Laemmli buffer (Bio-Rad, #1610737) to a concentration of 20 μg per well and boiled for 5 min prior to loading. 25 μL of sample was loaded into running gel (Bio-Rad 4%−20% graded, #456–1096) and run for 90 minutes at 35mA (Mini Transblot Cell, Bio-Rad, 153BR). Protein was transferred via electrophoresis Bio-Rad, #1620177) at 300mA for 2 hours. Blots were blocked for 30 min in 5% skim milk (Chem Cruz, #sc-2325) and 0.05% Tween 20 (Sigma Aldrich, #P1379) and incubated with Cx43 primary antibody (Sigma Aldrich, #C6219) and GAPDH primary antibody (Fitzgerald, #10R-G109a) in 5% skim milk overnight at 4C. HRP-conjugated secondary antibodies for respective primary antibodies (Santa Cruz, #sc-2004 and 2005), incubated at room temperature for 1 hour, were used to visualize protein. Blots were incubated with substrate (Millipore, # WBKLS0500) for 5 min and imaged via chemiluminescence.

Imaging homotypic Cx43 junctions

Standard culture media was replaced 2 hours prior to transfection. For each transfection, cells were plated in 30mm petri dishes (Falcon #35300). Lipofection was performed by Transit LT1 (Mirrus, #MIR2300) per manufacturer protocol. 2μg DNA, 100 μL OptiMem (Gibco, #31985062), and 5 μL transfection reagent were incubated for 20–30 min to form lipid complexes and added dropwise to each dish. U87-MG were transfected with C-terminal BFP-tagged Connexin 43 (Cx43-EBFP WT). IWCA and IKOCA were transfected with C-terminal GFP-tagged Cx43 (Cx43-msfGFP WT)^26–28. 3 days post-transfection media was changed, and cells were mixed at equal concentration into confocal well plates (Ibidi, #80826, ibitreat). Cells were imaged 24 hours post-plating by confocal microscopy (Zeiss, LSM 880 with Airyscan) with 63x oil objective. Live cell Imaging solution contained phenol free DMEM (Gibco, #31053028), 10% FBS, 1% pen-strep, and 25mM HEPES (Gibco, #15630106).

Parachute dye transfer

Acceptor cell populations were plated in 30mm petri dishes and grown to 80–100% confluence, approximately 1 to 2 days. Donor cells were harvested and centrifuged to remove media. Donor cells were resuspended to approximately 1×10^6 cells/mL in media with 10 μM calcein-AM (Molecular Probes, #C3100MP) and 5 μM DiI (Sigma, #42364) and incubated for 30 min. Post-dye incubation, donor cells were centrifuged and rinsed to remove dye-containing media twice, and resuspended to 1 × 10^6 cells/mL. 20 μL of donor cells were “parachuted” to acceptor cell cultures. 1–2 hours post parachute cells were imaged for spread of Calcein dye to acceptor cells by fluorescence microscopy (Nikon TE300 with RT Monochrome Spot camera) 10x (0.3 NA, 15.2 mm WD) and 20x objective (0.4 NA, 3.0 mm WD). For mCherry CRISPR lines, no DiI was used to label the parachuted cells, instead co-labeled cells represent acceptor cells. To inhibit GJ function, carbenoxolone (CBX) (Sigma, #C4790) at 100 μM was applied for the duration of culture.

Transwell invasion assay

In vitro invasion of GBM cells in monoculture and in coculture with astrocytes was assessed by transwell. 200 μL of standard culture media was added to Matrigel-coated invasion chambers with 8 μm pores (Fisher Scientific, Corning #354480) for 2 hours at 37°C to rehydrate. Hydrated invasion chambers were placed into 24-well tissue culture plates. 1 mL standard culture media (10% FBS) was added to the lower chamber of the transwell setup. 1×10⁴ cells for each condition were added to the upper chamber in serum-free media and incubated for 24 hours. Media in the bottom and top of the transwell was removed and the top of the transwell membrane was swabbed to remove non-invaded cells. Membranes were fixed in 4% paraformaldehyde (PFA) (Sigma, #1004960700) and rinsed 3 times in PBS. GBM cells expressed cytosolic mCherry to for visualization by fluorescence microscopy. 5 representative regions were imaged for each transwell. Control experiments were repeated with uncoated (sans Matrigel) transwell chambers (Fisher Scientific, Corning #3422). Invasion percentage was calculated by normalizing the number of invaded cells (coated) for each condition to number of migrated cells (uncoated). Invasion index was calculated by normalizing to the invasion percentage of invasive breast cancer line, MDA-MB-231 (ATCC, #HTB-26). For pharmacologic blockade experiments, CBX at 100 μM was applied for the duration of culture.

3D Matrigel invasion assay

3D tumor invasion controls were performed in a 3:1, Matrigel:culture media, matrix. In brief, Matrigel (Fisher Scientific, Corning #CB-40234A) was thawed at 4°C for 1 hour and mixed on ice with chilled cell culture medium (DMEM, 10% FBS, 1% AA). 250 μL of cell medium was added to an 8-well confocal imaging dish (Ibidi) and set at 37C for 1 hour. At 1 hour, cytosolic mCherry expressing U87 scrambled and Cx43 KO spheroids were injected by fine tip surgical tweezers into the set matrix. Spheroids were imaged at 24, 48, and 72 hours via fluorescence microscope (Nikon TE300). Samples were fixed in 4% PFA and imaged via confocal microscopy (Zeiss LSM880) in z-stack. Spheroids in matrigel were quantified for outgrowth by the ratio of their core (P_C) and total perimeter (P_T) and invasion radius (r_I = P_T/2π − P_C/2π).

Organotypic slice culture and ex vivo invasion assay

For slice culture experiments, 6–8 week old C57BL/6J mice (Jackson Labs, see above) and crossbreeds of Gfap-Cre (Jackson Labs, B6.Cg-Tg(Gfap-cre)77.6Mvs/2J, #024098) and Cx43 floxed (f/f) (Jackson Labs, B6.129S7-Gja1tm1Dlg/J, #008039) mouse lines were euthanized by isoflurane and decapitated. Whole brains were dissected and stored in ice cold oxygenated high-sucrose artificial cerebral spinal fluid (ACSF, 2.5 mM KCl, 1 mM CaCl₂*2H₂O, 4 mM MgSO₄*7H₂O, 4mM MgCl₂*6H₂O, 1.6 mM NaH₂PO₄*H₂O, 26 mM NaHCO₃, 20 mM glucose, 215 mM sucrose) used for slicing. Brains were fixed to a vibratome cutting platform (D.S.K Micoslicer DTK-2000) with adhesive (Locktite Ultragel #45208). Once affixed, brains were submerged in ice cold high-sucrose ACSF and 350–400 μm whole brain coronal slices were obtained at speed setting of 4 out of 10. Slices were placed in 33°C preheated recovery ACSF (124 mM NaCl, 2.5 mM KCl, 26 mM NaHCO₃, 10 mM glucose, 1 mM NaH₂PO₄*H₂O, 1.3 mM MgSO₄*7H₂O, 2.5 mM CaCl₂*2H₂O) and allowed to recover at room temperature for 1 hour. After recovery, slices were placed on 0.4 μm pore 6-well culture inserts (Sigma, #PICM03050), DMEM supplemented with 25% FBS, 1% pen-strep, and 25mM HEPES, was added beneath the slices, and 2 droplets of media were added to hydrate each slice. After 24 hours the media was replaced.

At 24 hours post-slicing, 250–300 μm tumor spheroids were implanted into the cortex of brain slices using fine tipped surgical tweezers. For pharmacologic blockade experiments, CBX at 100 μM was applied for the duration of culture with daily media replacement. Tumor-injected slices were rinsed 3 times in PBS and fixed at times 0, 24, 48, and 72 h in 4% PFA overnight. Fixed slices were rinsed 3 times in PBS and cleared by SeeDB²⁹. In brief, tissues were treated with 5% alpha-thioglycerol (Sigma, #M6145) plus increasing concentrations (%wt/vol) of Fructose (Sigma, #1053230250); 20%, 40% for 4–8 hours, 60% and 80% for 8 hours, 100% for 12 hours, and 115% for 12–24 hours. Cleared samples were rinsed and stored in PBS for imaging. Invaded spheroids were imaged in z-stacks at 3 μm intervals at 10x by confocal microscopy (Zeiss LSM880). ImageJ (NIH) was used for image processing analysis. Maximum projection of all z-planes was used for tracing to account for all branches. Perimeter of the tumor core was traced as were any invasion branches. Tracing was performed by ImageJ (Segmentation > Simple Neurite Tracer). Traces were analyzed for branches and used for Sholl analysis. Invasion radius (see above, r_I), was also calculated via ImageJ. Sholl analysis was performed on isolated invasion paths as a function of distance from tumor core edge. Path trace images were thresholded (ImageJ > Image > Adjust > Threshold) and Sholl analysis (ImageJ > Analyze > Sholl > Sholl Analysis (from image)) was performed with a step size of 5 μm.

All animal experiments were performed in accordance with relevant guidelines and protocols approved by the institutional animal care and use committee (IACUC) at Albert Einstein College of Medicine.

Tissue Immunofluorescence

Organotypic slices, post-invasion experiments, were fixed in 4% PFA at 2–8°C overnight and rinsed 1 hour in PBS. Slices were cryoprotected by treatment in graded solutions of 10%, 20%, and 30% glycerol (Sigma, #G5516) in PBS. Slices were embedded in OCT compound (Fischer Scientific, #23–730-571) then stored at −80°C overnight. 10 μm slices were obtained by cryotome (Thermo Fisher Scientific, Cryostar NX70) and affixed to glass slides (Fisher Scientific, #12–550-15). Slices were permeabilized/blocked in 0.4% Triton-X (Sigma, #X100) and 2% BSA (Sigma, #A9647), and then incubated in primary antibodies 1:500, GFAP (Sigma, #G9269) or Cx43 (Sigma, #C6219) in blocking solution at 4°C overnight. Cells were washed 3x for 15 min in PBS and incubated with secondary antibody 1:100 (Alexaflour488-conjugated donkey anti-rabbit, A32790A) and 300 nM DAPI (Sigma, #D9542) in blocking solution for 1 hour at room temperature. Cells were washed for 15 min at room temperature with 1X PBS and mounted (TFS, Prolong Gold #P10144) on coverslips (Fisher Scientific, #12–548-5M). Imaging was performed by confocal microscopy (LSM, 880) with 10x objective. Distribution of Cx43 along the tumor edge was quantified by line scan in ImageJ. Lines were drawn starting 100 μm from the tumor perimeter to 100 μm inside the tumor core, orthogonal to the incident edge. Mean fluorescence intensity was quantified by the plot profile function.

miRNA sequencing

Bulk miRNAseq was used to assess the Cx43-mediated transfer of miRNAs as potential signaling molecules implicated in invasion of GBM along tracts within astrocyte-rich brain regions. U87 cells labeled with cytosolic GFP were cocultured with IWCA and IKOCA cells in triplicate for 24 hours. After coculture cells were trypsinized and collected for sorting. GFP positive and negative cells were sorted by FACS (BD Biosciences FACSAria II) and small RNA was isolated from each cell population via isolation kit according to manufacturer protocols (TFS, #AM1560). Samples for miRNAseq were sent for sequencing and analysis (LC Sciences, Houston, TX).

miRNA transfer and transfection

Fluorescein-tagged miR-19b mimic (Qiagen #339173-YM00470545-ADB) was transfected into U87^scram or U87^{Cx43 KO} at 150 nM via lipofection (TransIT LT1, Mirrus). After 48 hours, transfected cells were rinsed 3 times and cocultured 1:1 with IWCA or IKOCA for 24 hours. miRNA localization was imaged via confocal microscopy (Zeiss LSM 880). Transfection of miR-19b mimic and inhibitor (Qiagen #339121-YI04101488-ADC) was performed in a similar fashion for ex vivo slice culture experiments, with U87 tumor spheroids alone (injected at 48 hours post-transfection), or full slices post-spheroid injection transfected at 150 nM miRNA.

Biased internalization

GJ plaques are degraded though internalization of regions of the plaque, consisting of entire terminally-mated channels³⁰. One pathway by which miRNA may be transferred to astrocytes from GBM is biased internalization of the connexosome. To evaluate biased internalization of Cx43 plaques coculture of cells expressing C-terminal fluorescently tagged Cx43 was performed. IKOCA and IWCA cells were transfected with GFP-tagged Cx43 and in isolation and reseeded for coculture in confocal 8 well-plates (Ibidi). U87 cells were transfected with BFP-tagged Cx43 in isolation and then added simultaneously to confocal well plates with either IWCA or IKOCA at a 1:1 ratio. Uptake of Cx43 plaques with expression of both tagged constructs was imaged by confocal microscopy in live cell imaging at 24 hours post-seeding and quantified with ImageJ. These experiments were repeated after pharmacologic blockade with 100 μM CBX. Further pairings of C-terminal truncated Cx43 mutants at residue 258 (m258) were tested to examine the role of CT phosphorylation sites in biased Cx43 endocytosis²⁷.

Statistics and Data Analysis

For groups of three or more, comparisons were made by one-way ANOVA with post hoc Tukey’s test. Comparison between two groups was done by student’s t test. miRNA sequencing data were analyzed by LC Sciences. In brief, unique miRNA sequences were mapped by LC Sciences algorithm, ACGT101-miR, in miRBase 22.0 by BLAST search. Differentially expressed miRNAs were identified in pairwise comparison by Chi-squared test. Predicted miRNA targets were identified using TargetScan and Miranda 3.3a. Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) were also performed by LC Sciences by Hypergeometric equation. Two and three principal component analyses were used to identify patterns in the dataset and Pearson correlations were performed to compare miRNA levels between cell types and their respective cocultures.

Data Availability Statement

All data will be made available upon request to authors. Raw and analyzed miRNA sequencing data can be obtained through request and will also be made available by contacting LC Sciences (ref #7248). miRNA sequencing data has also been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession number #GSE185474.

Results

GBM and astrocytes readily form Cx43 plaques and functional junctions

Homotypic heterocellular Cx43 plaque formation was observed after transfection of fluorescently tagged Cx43 constructs (Supplemental Figure 1). U87 cells readily formed Cx43 plaques with each other and with astrocytes within 24-hour post-seeding. Additionally, CRISPR deletion was successful in U87^{Cx43 KO} cells (Supplemental Figure 2).

A parachute dye-transfer assay was employed to evaluate the function of Cx43 junctions formed between U87 cells and astrocytes in vitro (Figure 1A). At two hours post-parachute, dye was transferred between U87^wt cells themselves (14.8 ± 1.8 cells/donor cell) and inhibited by CBX (1.1 ± 0.6 cells/donor cell), indicating gap junctional transfer. Dye was also transferred to IWCA cells, but not IKOCA cells, 24.4 ± 2.7 and 0.0 ± 0.0 cells/donor cell, respectively. Similarly, Calcein loaded U87^wt transferred dye readily to primary astrocytes, and transfer was blocked by CBX, 3.3 ± 0.4 and 0.0 ± 0.0 cells/donor cell. Finally, U87^{Cx43 KO} failed to transfer dye, while their scrambled sgRNA control demonstrated junctional diffusion, 0.9 ± 0.1 and 7.4 ± 0.6 cells/donor cell (Figure 1B–E and Supplemental Figure 3A). Robust dye transfer was shown between GL261 and both immortalized (5.3±1.5 coupled cells/region) and primary astrocytes (4.6±1.0 coupled cells/region) (Supplemental Figure 4A,B). GJ formation was also verified by immunostaining (Figure 1F).

Figure 1: GBM cells form functional Cx43 gap junctions with astrocytes and Cx43 plays opposing roles in GBM invasion in vitro.

(a) Illustration of parachute assay for assessment of gap junction (GJ) formation, showing Calcein-loaded donor cells plated diffusely on a monolayer of acceptor cells. (b-c) Two-channel fluorescence micrographs of intercellular Calcein diffusion at 1.5–2 hours post-parachute. Scale bars 25 μm. (d) Quantification of GJ coupling for cell types in monoculture. (e) Quantification of GJ coupling between U87 cells and astrocytes. (e) Confocal micrograph of Cx43 immunostaining in IWCA-U87^wt coculture, demonstrating formation of Cx43 plaques. Cx43 labeled in green, with U87 expressing cytosolic mCherry (red). Scale bar 20 μm, inset 4 μm. n ≥ 6. Data shown as mean ± SEM, ****p<0.0001. (g) Schematic of in vitro invasion assay. Matrigel coated transwell inserts with 8 μm pores were used to evaluate U87 invasion under various conditions. (h) Fluorescence micrographs of mCherry-labelled U87^wt invasion at 24 hours in coculture with wild-type and Cx43 knockout astrocytes. Scale bars 20 μm. (i) Invasion index for U87^wt cells, showing increase in invasion when U87-U87 communication is blocked and in coculture with wild-type astrocytes. (j) Fluorescence micrographs of mCherry-labelled U87^{Cx43 KO} or scrambled sgRNA control cells (U87^scram) at 24 hours in monoculture or coculture with immortalized astrocytes. Scale bars 20 μm (k) Invasion indices for U87^{Cx43 KO} and scrambled control. n = 3 for all experiments. Data shown as mean ± SEM. *p<0.05, **p<0.01, ***p<0.001.

GBM invasion increases upon coculture with Cx43 expressing astrocytes

A transwell invasion assay was used to quantify GBM invasion as a function of Cx43 communication (Figure 1G). Invasion index, normalized to that of the highly invasive breast cancer line MDA-MB-231, was used to quantify relative invasion of U87 cells under mono and coculture conditions. Invasion index of U87 monocultures increased 2-fold (0.7 ± 0.1 vs 1.9 ± 0.2, p = 0.008) with pharmacologic blockade of connexins. Invasion significantly increased in U87^wt coculture with IWCA and primary WT astrocytes, but did not increase in coculture with IKOCA, 0.3 ± 0.1. p = 0.95. U87^{Cx43 KO} invasion index (2.0 + 0.2) was significantly higher relative to U87^wt (p = 0.002) and U87^scram controls (0.9 ± 0.1, p = 0.016). U87^{Cx44 KO} also showed an increase in invasion relative to scrambled control when cocultured with IWCAs and IKOCAs 1.5 ± 0.1 (p = 0.013) and 1.4 ± 0.02 (p = 0.003), respectively. (Figure 1H–K and Supplemental Figure 3B). Similarly, GL261 invasion index increased ~2-fold (1.2 ± 0.2 GL261 monoculture vs 2.2 ± 0.2 astrocyte coculture, p = 0.027) when cocultured with WT astrocytes which was blocked by CBX (0.7 ± 0.1, p = 0.001) (Supplemental Figure 4C,D), whereas CBX GL261 alone treated with CBX showed no effect.

GBM spheroids incorporate into sliced cultured cortical tissue

An ex vivo slice culture assay was used to better approximate in vivo matrix composition and tissue architecture, diagrammed in Figure 2A. GFAP labeling demonstrated that most of the U87 implanted spheroid cells expressed GFAP, and GFAP expressing astrocytes at the tumor periphery colocalized with invading tumor cells (Figure 2B). Additionally, U87^wt spheroids formed Cx43 plaques with neighboring astrocytes in WT slices (Figure 2C). Cx43-deleted U87 showed no Cx43 expression up to 72 hours after implantation (Figure 2D,E). U87 spheroids did not form plaques with neighboring astrocytes in slices from mice deficient in astrocytic expression of Cx43 Gfap-Cre Cx43^f/f; however, Cx43 expression localized to the edge of the tumor (Figure 2F,G). The corona of Cx43 expression at the tumor edge is unique to U87 tumors in astrocyte Cx43 null slices, whereas Cx43 expression is more uniform throughout the tumor in Cx43-competent slices. Cx43 distribution at the primary tumor edge is shown in Figure 2H.

Figure 2: Ex vivo slice culture recapitulates the invasive phenotype of GBM.

(a) Schematic of U87 spheroid implantation into cortical tissue of murine whole brain coronal slices (350–400 μm). Scale bars 200 μm. (b) Confocal micrograph of GFAP immunostaining in wild-type murine brain with U87^wt spheroid implantation, showing intratumor labelling and colocalization of astrocytes at the invasive edge (inset). U87^wt expressing cytosolic mCherry shown in red and GFAP staining shown in green. Scale bars, 200 μm, top, and 50 μm, bottom. (c) Confocal micrograph of Cx43 immunostaining in wild-type (Cx43^+/+) murine brain with U87^wt spheroid implantation. Colocalization of GBM-astrocyte Cx43 was found at sites of invasion (inset). U87 cells shown in red, Cx43 in green, and DAPI stained cell nuclei in blue. Scale bars, 200 μm, top, 20 μm bottom. (d) Confocal micrograph of Cx43 Immunostaining for U87^{Cx43 KO} spheroids implanted in wild-type slice. Scale bar 200 μm. (e) Confocal micrograph of Cx43 Immunostaining for U87^{Cx43 KO} spheroids implanted in Gfap-Cre Cx43^f/f slice. (f, g) Confocal micrographs of Cx43 Immunostaining for U87^wt spheroids implanted in Gfap-Cre Cx43^f/f slices showing localization of Cx43 at the tumor edge for cohesive tumors and interaction with blood vessels. Cx43 shown in green, U87 in red, and DAPI in blue. Scale bars 100 μm. (h) Mean distribution of Cx43 staining in wild-type and Cx43 KO conditions at the tumor or edge. Line scans were performed from 100 μm outside of the tumor edge to 100 μm within the tumor core. In double knockout conditions no Cx43 was present. In double wild-type conditions Cx43 was expressed in astrocytes adjacent to the tumor and highly expressed in a uniform distribution within the tumor. For wild-type U87 in Cx43^f/f slice there was a clear localization of Cx43 to the tumor edge.

GBM invasion is greatly decreased in Cx43 knockout brain

U87 cells invade cortical tissue in an anisotropic pattern, as compared to the uniform radial expansion seen in a 3D Matrigel matrix (Supplemental Figure 5). When CBX was bath applied every 12 hours to inhibit GJ-mediated intercellular coupling, invasion was modestly, but significantly reduced (Figure 3A). Invasion radius (r_i) (146.4 ± 9.8 vs 94.3 ± 14.8 um, p = 0.016), total branches (69.4 ± 11.9 vs 45.3 ± 4.6 μm, p = 0.33), and intersections of branches radiating out from the tumor core were all reduced at 72 hours by CBX (Figure 3B–D). This demonstrates that total invasion, as well as tortuosity of invasion, were reduced by pharmacologic blockade of all connexin communication in the slice. Sholl schematic is shown in Supplemental Figure 6

Figure 3: GBM invasion in wild-type tissue is modestly decreased by GJ blockade and modestly increased by GBM Cx43 knockout.

(a) Maximum projection confocal micrographs of U87 spheroid invasion at 24-hour intervals, for control and 100 μm CBX conditions. Scale bars 200 μm. (b) Invasion radii at 72 hours post-implantation, mean ± SEM. (c) Number of invasive branches over time, mean ± SEM. (d) Sholl analysis of invasion extent and complexity at 72 hours. Intersections of invasive branches as a function of distance from the edge of the primary tumor, mean ± SEM. Data indicate a small, but significant decrease in invasion with GJ blockade at 72 hours. (e) Maximum projection confocal micrographs of U87^{Cx43 KO} and U87 scrambled sgRNA control spheroids implanted in wild-type slices over time. Scale bars 200 μm. (f) Invasion radius at 72 hours post-implantation, mean ± SEM. (g) Number of invasive branches over time, mean ± SEM. (h) Sholl analysis of branch intersections as function of distance from the tumor edge at 72 hours. Shown as mean ± SEM. Cx43 knockout in tumor cells had no effect on complexity and branching of tumor or invasiveness but showed a small effect on extent of tumor invasion. n ≥ 7 for all experiments. *p<0.05, ****p<0.0001.

Deletion of Cx43 in U87 cells alone profoundly enhanced the extent of invasion relative to WT or scrambled control (Figure 3E). At 72 hours average invasion radius for U87^scram was 74.31 ± 9.71 compared to 171.8 ±14.1 μm for U87^{Cx43 KO} (p < 0.0001), indicating Cx43 deletion in GBM increased distance of invasion (Figure 3F). However, number of branches was unchanged (Figure 3G), 56.2 ± 5.1, vs 64.7 ± 5.2, p = 0.282. Intersections of branches radiating from the tumor core (Figure 3H) were also similar (Supplemental Table 1).

Striking differences were found when GBM invasion was examined in Gfap-Cre Cx43^f/f slices, i.e. astrocyte specific Cx43 KO (Figure 4A). Both WT and scrambled sgRNA control U87 spheroids showed markedly lower invasion at 72 hours in Gfap-Cre Cx43^f/f slices (r_i = 55.6 ± 7.2 μm, p <0.0001, and 57.3 ± 10.2 μm, p <0.0001, respectively, Figure 4B,F;Total branches = 16.7 ± 3.0, p = 0.025, and 19.3 ± 3.8, p = 0.005, respectively, Figure 4C,G). The highly invasive phenotype seen in U87 cells in Gfap-Cre Cx43^+/f slices (r_i = 205.3 ± 23.5 μm; upper panel Figure 4B) but not Cx43^f/f slices was recovered in Cx43^f/f slices by Cx43 deletion in U87 cells (r_i = 170.3 ± 13.6 μm) as shown in Figure 4E–H. These data indicate that detachment of GBM cells from the primary tumor driven by loss of GBM-GBM junctions is a necessary step for invasion, and formation of GBM-astrocyte junctions is a contributing factor to this process. A summary of Sholl analysis data is shown in Supplemental Table 1. Slice culture invasion data from GL261 also show Cx43 dependence, with a marked decrease of invaded cells (28.7 ± 6.3 Vs 5.0 ± 1.8, p = 0.008) and invasion radius (141.4 ± 14.3 μm vs 40.6 ± 10.9 μm, p = 0.0005) with astrocyte-targeted Cx43 knockout (Supplemental Figure 4E–G).

Figure 4: Knockout of astrocyte Cx43 greatly limits GBM invasion.

(a) Maximum projection confocal micrographs of wild-type U87 spheroid invasion over time in Gfap-Cre Cx43^+/f and ^f/f slices. (b) Invasion radii at 72 hours post-implantation, mean ± SEM. (c) Number of invasive branches over time, mean ± SEM. (d) Sholl analysis of invasion extent and complexity at 72 hours. Intersections of invasive branches as a function of distance from the edge of the primary tumor, mean ± SEM. Data show a clear decrement in extent and complexity of GBM outgrowth in knockout slices. (e) Confocal micrographs of U87^{Cx43 KO} and U87 scrambled sgRNA control spheroids implanted in Gfap-Cre Cx43^f/f slices over time. (f) Invasion radius at 72 hours post-implantation, mean ± SEM. (g) Number of invasive branches over time, mean ± SEM. (h) Sholl analysis of branch intersections as function of distance from the tumor edge at 72 hours. Shown as mean ± SEM. Data strongly indicate that GBM invasion is limited by astrocyte Cx43 knockout, but the invasive phenotype is recovered when GBM Cx43 is also removed. n ≥ 4 for all experiments. *p<0.05, ****p<0.0001. Scale bars 200 μm.

miRNA sequencing of GBM and Astrocytes in coculture

To determine which miRNAs are transferred from GBM to astrocytes via Cx43-dependent pathway, we performed miRNAseq on IWCA and IKOCA cell lines cocultured with U87^wt. We found 295 conserved miRNAs highly expressed in GBM, of which 68 were enriched in IWCAs cocultured with U87. From these, we found 11 miRNAs (Figure 5A) that were transferred to IWCA but not IKOCA. Analysis of differentially expressed miRNAs in IWCAs after coculture with U87^wt showed enrichment for genes encoding protein and metal binding, cytoskeleton, mitochondrion and cytoplasmic vesicles (Figure 5B). Pathways enriched in coculture include cancer pathways (specifically RAS, MAPK, PI3K-Akt), focal adhesions, cytokines and endocytosis, and nervous system development (Figure 5C).

Figure 5: Cx43-dependent miRNA transfer from GBM to neighboring astrocytes.

(a) Workflow for identification of Cx43-mediated miRNA transfer. miRNA sequencing data generated from immortalized astrocyte lines IWCA and IKOCA cocultured (CC) with U87 GBM cells. (b) Gene ontology enrichment statistics for IWCAs cocultured with U87 cells, identifying shifts implicated in gene families of interest including, transport, cytoplasmic vesicles, cytoskeleton, nervous system development, and protein binding. (c) Pathway enrichment statistics for IWCAs cocultured with U87 cells, identifying pathways of interest including, pathways in cancer, focal adhesion, endocytosis, and apoptosis. n = 3 for all experiments. (d) Fluorescence micrographs of fluorescein-tagged hsa-miR-19b localization post-coculture of transfected U87^scram or U87^{Cx43 KO} and immortalized astrocytes. miRNA shown in green and U87 shown in red. Scale bars 20 μm. (e) Proportion of miR-19b found in astrocytes post-coculture with fluorescent miRNA mimic transfected U87. n≥ 6, ****p<0.0001.

Similarity of miRNA content between monocultures and the samples obtained from IWCA and IKOCA cocultured with U87 cells (IWCA_Co and IKOCA_Co) and from U87 cells cocultured with astrocyte cell lines (U87_IWCA and U87_IKOCA) was evaluated by Pearson correlation and principal component analysis (PCA) as shown in Supplemental Figure 7. Pearson analysis showed high similarity between monocultures of U87 cells and both IKOCA and IWCA cell lines (0.76 and 0.61, respectively). There was also substantial similarity between U87_IWCA and U87_IKOCA (0.999) and high correlation between both IWCA_Co and IKOCA_Co (.998). Correlation was much lower between monocultures and cocultures of respective cell types (0.11 for IWCA, 0.08 for IKOCA, 0.35 and 0.37 for U87 vs U87_IWCA and U87_IKOCA, respectively).

PCA was used to describe variance in miRNA expression across samples. PCA results were consistent with Pearson correlations, finding very high overlap of U87_IWCA and U87_IKOCA, as well as IWCA_Co and IKOCA_Co suggesting a coculture specific shift in both GBM cells and astrocytes independent of Cx43 expression. Like Pearson analyses, a significant divergence was found between cocultures and monocultures of each cell type, and between cocultured U87 cells and cocultured astrocyte populations.

miR-19b is transfer from GBM to astrocytes and contributes to GBM invasion

Fluorescein-tagged miR-19b mimic transfected into U87 was efficiently transferred to IWCA (83.0 ± 2.7%) at 24 hours coculture. When Cx43 was deleted in either cell type transfer to astrocytes was significantly decreased; IKOCA (44.2 ± 4.3%), U87^{Cx43 KO} (45.1 ± 4.8%), and both U87^{Cx43 KO} and IKOCA (29.4 ± 8.5%) (Figure 5D). Transfection of entire slices with miR-19b mimic markedly increased invasion of U87^wt spheroids implanted in Gfap-Cre Cx43^f/f slices at 72 hours (r_i = 253.9 ± 12.7 μm, branches = 61.9 ± 8.3), while transfection of spheroids alone did not significantly increase invasion (r_i = 115.2 ± 13.4 μm, branches = 31.6 ± 3.0) (Figure 6A–D). Conversely, treatment with miR-19b inhibitor in both slices and spheroids blunted invasion under previously invasive double WT conditions. Invasion radius for slice and spheroid conditions was limited to 51.3 ± 7.1 μm and 30.3 ± 7.5 μm, respectively, and branching was limited to 18.25 ± 2.6 and 29.2 ± 5.8 (Figure 6E–H). Sholl analysis data can be found in Supplemental Table 2.

Figure 6: miR-19b contributes to GBM invasion.

(a) Maximum projection confocal micrographs U87^scram spheroid invasion in Gfap-Cre Cx43^f/f slices with miR-19b mimic transfected into full slices post-spheroid implantation or transfection of tumor spheroids prior to implantation at 72 hours. Scale bars 200 μm. (b) Invasion radii at 72 hours post-implantation, mean ± SEM. (c) Number of invasive branches at 72 hours, mean ± SEM. (d) Sholl analysis of invasion extent and complexity at 72 hours. Intersections of invasive branches as a function of distance from the edge of the primary tumor, mean ± SEM. Data demonstrate invasive phenotype recovery with miR-19b treatment of in situ Gfap-Cre Cx43^f/f astrocytes. (e) Maximum projection confocal micrographs of U87^scram spheroid invasion in wild-type slices at 72 hours. miR-19b inhibitor transfected into whole slice post-implantation or into tumor spheroids prior to implantation. Scale bars 200 μm. (f) Invasion radii at 72 hours post-implantation, mean ± SEM. (g) Number of invasive branches at 72 hours, mean ± SEM. (h) Sholl analysis of invasion extent and complexity at 72 hours. Intersections of invasive branches as a function of distance from the edge of the primary tumor, mean ± SEM. Inhibition of miR-19b significantly limits invasion. n ≥ 6 for all experiments. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

Biased internalization

The analysis of miRNA species induced in WT but not Cx43 KO astrocytes identified a population whose presence presumably depended on transfer between the GBM and host tissue. Among mechanisms by which GJs may mediate transfer of molecules between cells is endocytosis, where the normal turnover of Cx43 involves internalization of segments of GJ plaques together with small volumes of cytoplasm. To determine dynamics of GBM-astrocyte GJ internalization, we cocultured U87 transfected with BFP-labeled Cx43 with IKOCA transfected with GFP-labeled Cx43. U87 and the transfected astrocyte cell line formed both homocellular and heterocellular GJ plaques (Figure 7A), and at 12–24 hours vesicles containing tagged Cx43 appeared within the cells. Experiments in which cells were transfected with different fluorescent protein tags demonstrated that internalized vesicles contained GJ components contributed by both cells. To determine the extent to which uptake of the vesicles was biased toward one cell type or the other in U87-astrocyte cocultures, we quantified the fraction of GJ vesicles within each cell. In homocellular plaques, distribution was roughly symmetrical, with U87-U87 plaques at 47.6 ± 2.8% and IWCA-IWCA plaques at 47.7 ± 5.5% (Figure 7B). By contrast, in U87-astrocyte cocultures internalization of plaques was biased heavily toward astrocytes, with 81.3 +/− 4.3% of endocytosed GJ complexes found within astrocytes. Bias was not altered by CBX Cx43 blockade (86.2 + 3.2%, p = 0.428) (Figure 7C,E). Additionally, bias was abolished by Cx43 C-terminal truncation in astrocytes (77.9 ± 6.0% toward astrocytes in WT-WT condition and 93.3 ± 2.3% with m258 truncation mutant in U87 vs 42.3 ± 4.4% for truncation in astrocytes and 28.5 ± 6.3% for truncation in both) as show in Figure 7D and Supplemental Figure 8.

Figure 7: Endocytosis of GBM-astrocyte Cx43 plaques is biased toward astrocytes.

(a) Confocal micrographs of Cx43-msfGFP U87 cells cocultured with Cx43-EBFP2 IWCA cells. Scale bars 20 μm. (b) Quantification of Cx43 plaque internalization bias for astrocyte-GBM, GBM-GBM, and astrocyte-astrocyte cultures, demonstrating no-bias for monocultures and bias toward astrocytes in coculture. Internalized plaques indicated by white arrows. n ≥ 5 for all experiments. Data shown as mean ± SEM. **p<0.01. (c) Percent of GFP+/BFP+ Cx43 endocytic vesicles in astrocytes as compared to U87 cells within each region of interest. Scale bars 50 μm. n ≥ 11 for all experiments. Data shown as mean ± SEM. No statistical difference was observed. Biased internalization experiments were repeated with pairwise variants to demonstrate internalization was independent of fluorescent tag type (C-terminal tagged msfGFP or EBFP2). (d) Percentage of Cx43 plaques internalized into astrocytes for pairwise cocultures of wild-type N-terminal tagged or truncation mutants (m258) in U87 cells and IKOCA. n ≥ 7, ****p<0.0001. (e) Two channel confocal micrographs of U87-IKOCA cocultures with msfGFP-tagged Cx43 expressed in IKOCA cells and EBFP2-tagged Cx43 expressed in U87 cells. Experiments were performed with and without 100 μM of Cx43 blocker carbenoxolone (CBX). Internalized plaques indicated by white arrows.

Discussion

We found that U87 cells efficiently form functional Cx43 GJs with each other and with both immortalized and primary astrocytes dissociated and in slice culture. Despite its incredibly diverse cell population, GBM is principally an astrocytoma³¹. Our findings are consistent with previous data showing expression of Cx43, the predominant astrocytic connexin, in GBM cells, and U87 cells specifically³². Of note is the distribution of Cx43 in vitro as compared to our ex vivo model. While it has been observed that Cx43 expression persists in astrocytomas with increasing grade, it is often diminished in GBM¹¹. Our data, however, suggest that there may be a Cx43 localization in higher grade tumors toward the outer edge. These data are consistent with the overall finding that GBM-astrocyte Cx43 junctions at the tumor periphery are a driver of invasion in Cx43 expressing tumors. The canonical role of Cx43 in GBM is increasing GBM-GBM cell communication, resulting in tumor growth suppression. However, a subpopulation of Cx43-expressing cells at the boundary is responsible for the exact opposite. This redistribution of Cx43 to the periphery implies that core tumor cells are less well adhered to each other, also increasing their proclivity for outward migration. Redistribution of Cx43 in the primary tumor may also be a consequence of cell death in the necrotic core. GBM are densely packed cell clusters with a high metabolism and poor, disordered circulation, which increases toxic metabolite concentration, initiates cell death, and alters intercellular communication. Localization of Cx43 at the boundary is more pronounced for spheroids implanted in astrocyte Gfap-Cre Cx43^f/f slices, where inability to form GJs with surrounding astrocytes may create an anti-invasive outer lamella of coherent GBM-GBM junctions.

The opposing roles played by Cx43 in GBM invasion present a challenge when investigating Cx43 as a therapeutic target for GBM. An effective approach would need to selectively delete Cx43 in astrocytes, as pharmacologic blockade would have only moderate short-term efficacy and a minimal impact on overall survival. Moreover, preserving GJ functionality in the tumor while blocking it in the host would be expected to optimize Bystander cell killing by irradiation or nucleoside transfer in HSV-tk transduced cells³³. Alternatively, treatment would need to isolate the GBM-astrocyte Cx43-mediated communication pathway, and its associated molecules.

Although our studies reveal Cx43 expression in tumor cells and host play opposing roles in GBM progression, these data are consistent with previous studies which have examined the role of Cx43 in tumorigenesis, extravasation, and invasion. GBM-GBM Cx43 influences primary tumor cohesion, either through intratumoral solute transport or stochastic effects on cell-cell adhesion. Loss of Cx43 has long been a canonical indicator of increased tumor aggressiveness¹². Despite this overall downregulation of Cx43 in high grade gliomas, many studies have shown that GBM tumors, heterogeneous as they are, do maintain low-level expression of Cx43, and Cx43-expressing cells form GJs with astrocytes¹⁵. Our findings also contribute to the growing body of literature that implicates GBM-astrocyte GJs in the invasion process^{6, 17, 19, 34}, and miRNA as signaling molecule.

miRNAs act as post-transcriptional repressors and therefore can have profound phenotypic effects. Our data show some miRNAs are transferred from U87 to astrocytes in a Cx43-dependent manner. Many miRNAs identified in our screen are overexpressed in GBM. hsa-miR-19, hsa-miR-378a, and others have been identified in sequencing studies of GBM^35–39. Some, like hsa-miR-335, are associated with cancer migration and invasion^40–42. hsa-miR-19, 378a, 335, and 340 have high predicted target scores for cadherins, integrins, focal adhesion kinases, and/or adhesion associated molecules⁴³. Down regulation of astrocyte adhesion at the basement membrane may provide an open corridor and potentiate invasion. The profound consequences of high-concentration treatment with a miR-19b mimic and its corresponding inhibitor observed here ex vivo necessitate further study in vivo. Some highly expressed miRNAs in U87 show up in IWCA but not IKOCA cells post-coculture, indicating Cx43-dependent miRNA shuttling. Notably, miRNAs such as hsa-miR-19 show GJA1 as a predicted target⁴³. If astrocyte Cx43 expression is suppressed this may also decrease astrocyte network coupling, increasing the likelihood of astrocyte apoptosis and detachment, and limit the spread of chemotherapeutics and radiation, modulating bystander effects³³.

Our principal component analysis (PCA) demonstrated a large shift in astrocyte miRNA expression in monoculture versus coculture. IWCA and IKOCA cocultured with U87 cells had strong similarity in PCA and Pearson coefficient. It appears that astrocytes, independent of Cx43, shift their miRNA expression with exposure to GBM, perhaps as a defensive mechanism. While this large shift occurs in both IWCA and IKOCA, there remains a marked suppression of invasion when astrocytic Cx43 is absent, therefore the subset of miRNAs differentially found in Cx43-expressing astrocytes may contribute to GBM circumventing astrocytic defenses.

While miRNAs, as linear molecules, may be able to traverse the GJ pore for direct exchange³⁴, miRNAs are an order of magnitude larger than the 1kDa size exclusion limit of Cx43 pores⁴⁴. Another means by which miRNA may be transferred preferentially from one cell to another is biased endocytosis of GJ plaques. GJ plaques are terminally mated, and entire plaques are taken up into one of two joined cells. Cytoplasmic material is engulfed by membrane invagination and taken up into the endocytic vesicle⁴⁵. The endosome is a signaling hub⁴⁶ and exchange of cytoplasmic contents from GBM to astrocytes may manipulate astrocyte behavior, creating an invasion-permissive environment along vascular tracts within the brain. Our data show the GJ plaque is predominantly endocytosed in astrocytes, roughly 75% of the time. The exact reason for this bias is unclear, but potential explanations include high metabolic rate of GBM cells⁴⁷ and selective Cx43 C-terminal modifications⁴⁸, consistent with our data demonstrating elimination of biased endocytosis in C-terminal truncation mutants. There are other proposed means of miRNA transit between cancer and neighboring cells, including exosomes⁴⁹, high density lipoprotein (HDL) transport⁵⁰, and tunneling nanotubes⁵¹. Exosomes can possess Cx43 in their lipid membrane, which may facilitate miRNA transfer and targeting to cells possessing homotypic Cx43⁵².

Conclusion

GBM, the deadliest adult CNS tumor, invades along tracts within the brain, rendering current treatment regimens ineffective long term. We find that deletion of Cx43 in astrocytes in situ greatly reduces invasion, demonstrating that GBM-astrocyte gap junctions contribute to initiation of invasion. Our data suggest that transfer of specific miRNAs, via biased endocytic uptake of Cx43 plaques, may be the culprit mechanism, leading to altered astrocyte translation of adhesion proteins. Our ex vivo slice culture invasion model provides a platform for further investigation of GBM-astrocyte interactions at the peritumoral interface, methods to block such interactions, and study of miRNA targets and their effects.

Implications:

Cx43-mediated communication, specifically miRNA transfer, profoundly impacts glioblastoma invasion and may enable further therapeutic insight.