Abstract
The malignant brain cancer glioblastoma (GBM) contains groups of highly invasive cells that drive tumor progression as well as recurrence after surgery and chemotherapy. The molecular mechanisms that enable these GBM cells to exit the primary mass and disperse throughout the brain remain largely unknown. Here we report using human tumor specimens and primary spheroids from male and female patients that glial cell adhesion molecule (GlialCAM), which has normal roles in brain astrocytes and is mutated in the developmental brain disorder megalencephalic leukoencephalopathy with subcortical cysts (MLC), is differentially expressed in subpopulations of GBM cells. High levels of GlialCAM promote cell–cell adhesion and a proliferative GBM cell state in the tumor core. In contrast, GBM cells with low levels of GlialCAM display diminished proliferation and enhanced invasion into the surrounding brain parenchyma. RNAi-mediated inhibition of GlialCAM expression leads to activation of proinvasive extracellular matrix adhesion and signaling pathways. Profiling GlialCAM-regulated genes combined with cross-referencing to single-cell transcriptomic datasets validates functional links among GlialCAM, Mlc1, and aquaporin-4 in the invasive cell state. Collectively, these results reveal an important adhesion and signaling axis comprised of GlialCAM and associated proteins including Mlc1 and aquaporin-4 that is critical for control of GBM cell proliferation and invasion status in the brain cancer microenvironment.
SIGNIFICANCE STATEMENT Glioblastoma (GBM) contains heterogeneous populations of cells that coordinately drive proliferation and invasion. We have discovered that glial cell adhesion molecule (GlialCAM)/hepatocyte cell adhesion molecule (HepaCAM) is highly expressed in proliferative GBM cells within the tumor core. In contrast, GBM cells with low levels of GlialCAM robustly invade into surrounding brain tissue along blood vessels and white matter. Quantitative RNA sequencing identifies various GlialCAM-regulated genes with functions in cell–cell adhesion and signaling. These data reveal that GlialCAM and associated signaling partners, including Mlc1 and aquaporin-4, are key factors that determine proliferative and invasive cell states in GBM.
- AQP4
- astrocyte
- brain cancer
- extracellular matrix
- hepaCAM
- megalencephalic leukoencephalopathy with subcortical cysts
Introduction
Initiation, growth, and progression of the malignant cancer glioblastoma (GBM) are tightly coupled to blood vessels in the brain microenvironment (Wirsching et al., 2021). Stem-like GBM cells (GSCs) localize to vascular niches where they exploit local cues for self-renewal and differentiation (Testa et al., 2022). GBM develops hallmark blood vessel pathologies including florid angiogenesis as well as hemorrhage and edema because of breakdown of the intratumoral blood–brain barrier (BBB; Rodríguez-Camacho et al., 2022). GSCs often disperse throughout the brain parenchyma using extracellular matrix (ECM)-rich vascular basement membranes (Sayegh et al., 2014). Hence, identifying and selectively targeting pathways that couple GBM cells to blood vessels may be an effective strategy to block invasive growth and enhance patient responsiveness to chemotherapy.
To characterize targetable pathways used by GBM cells to thrive in blood vessel niches and invade through the brain, we mined open-source databases to identify genes that (1) show enriched expression in perivascular astroglial cells, (2) display high expression in glioma cells during progression to GBM, and (3) have putative links to cerebrovascular pathophysiology. These efforts have led to identification of glial cell adhesion molecule (GlialCAM), encoded by the hepatocyte cell adhesion molecule (HepaCAM) gene. GlialCAM is a 50 kDa single-pass type I transmembrane glycoprotein that is predominantly expressed in the brain and liver (Favre-Kontula et al., 2008). GlialCAM suppresses hepatocyte proliferation, and its expression is downregulated in hepatocellular carcinoma cells, suggesting tumor suppressor-like functions (Moh et al., 2005a). In the brain, GlialCAM is normally expressed in astrocytes where it has roles in ion homeostasis (Brignone et al., 2015), BBB physiology (Gilbert et al., 2021), and synaptic excitation (Baldwin et al., 2021). A significant role for GlialCAM was revealed during genomic analysis of patients with the neurodevelopmental disorder megalencephalic leukoencephalopathy with subcortical cysts (MLC; Leegwater et al., 2001). Most MLC patients contain recessive mutations in the MLC1 gene, which encodes a 38 kDa astrocyte-expressed protein with eight transmembrane domains (Brignone et al., 2019). Using genetically engineered mice, we have recently shown that Mlc1+ astrocytes juxtapose blood vessels, regulate endothelial barrier integrity, and are critical for maintaining mammalian BBB physiology (Morales et al., 2022).
The extracellular IgG-like domains of GlialCAM mediate interactions with Mlc1 as well as components of the dystrophin–glycoprotein complex (DGC) including aquaporin-4 in astrocyte end feet (Capdevila-Nortes et al., 2015). We have reported previously that Mlc1 protein is overexpressed in human GBM (Lattier et al., 2020) and is a molecular marker for the classical GBM subtype, which is defined in part by epidermal growth factor receptor (EGFR) overexpression and wild-type TP53 status. Silencing MLC1 expression in human GSCs leads to defective self-renewal in vitro and impaired invasion in preclinical mouse models (Lattier et al., 2020). Reduced Mlc1 protein expression in GSCs is associated with the hyperactivation of receptor tyrosine kinase (RTK) signaling pathways, particularly those involving Axl (Lattier et al., 2020). Potential functions for GlialCAM in brain tumors and its possible links to Mlc1, however, remained unexplored.
Here, we have analyzed functions for GlialCAM in human GBM using resected cancer tissue samples, patient-derived tumor stem cell spheroids, and xenograft mouse models. Our data reveal that GlialCAM is highly expressed in GBM cells within the tumor core. In contrast, GBM cells with low levels of GlialCAM are more dispersive along blood vessels and white matter tracts in the surrounding brain tissue. Quantitative RNA sequencing (RNAseq) identifies various GlialCAM-regulated genes with functions in cell adhesion and signaling, including AQP4, which encodes the water channel aquaporin-4. Indeed, cell-based assays and small molecular inhibitors of aquaporin-mediated water transport reveal that GlialCAM and aquaporin-4 coordinately balance GBM cell proliferation and invasion. In summary, these data reveal that GlialCAM, aquaporin-4, and associated signaling partners are key factors that control GBM cell polarity and invasion in the brain parenchyma.
Materials and Methods
Human GBM cells.
Approval for the use of human specimens was obtained from the Institutional Review Board (IRB) at the University of Texas MD Anderson Cancer Center. The IRB waived the requirement for informed consent for previously collected residual tissues from surgical procedures stripped of unique patient identifiers according to the Declaration of Helsinki guidelines. GSCs were cultured from freshly resected human tumors (20), and were grown and maintained in complete GSC media; DMEM Ham's F12 50/50 medium containing l-glutamine (catalog #10-090-CV, Corning), supplemented with 1× B27 supplement (catalog #17504-044, Thermo Fisher Scientific), 20 ng/ml EGF (catalog #PHG0313, Biosource), 20 ng/ml bFGF (catalog #PHG0021, Biosource), and 1× penicillin-streptomycin-antimycotic solution (catalog #MT30004CI, Corning). When GSCs developed neurosphere-like spheroids in culture, they were passaged by dissociation using 50 μl Accutase (catalog #AT104, Innovative Cell Technologies) per 1 × 106 cells and maintained in complete GSC media. Genomic validation of GSCs was performed by DNA short tandem repeat profiling in a Cancer Center Support Grant-funded Characterized Cell Line Core Facility. GSCs were routinely tested for mycoplasma using commercially available kits (Thermo Fisher Scientific), and only those cells deemed mycoplasma free were used for experiments. GSCs were infected overnight with concentrated pGIPZ lentivirus at a multiplicity of infection of 1.0. The following clones were used for GlialCAM shRNA: V3LHS_413349, V3LHS_413351, and V3LHS_413352 versus control pGIPZ containing RFP (Dharmacon, GE). HEK293T cells were transfected with GlialCAM ORF using the Precision LentiORF (pLOC) vector versus control pLOC (Dharmacon, GE). All HEK293T cells were transfected using Lipofectamine 3000 reagent (Thermo Fisher Scientific).
3D ECM invasion assays.
CytoSelect 24-well cell invasion assay, basement membrane (catalog #CBA-110, Cell Biolabs) was used for in vitro cell invasion assays (colorimetric) with nontargeting (NT) shRNA (n = 3) and GlialCAM shRNA (n = 3) GSCs. ECM-coated cell culture inserts (pore size, 8 µm) were seeded with 1 × 105 viable GSCs in complete GSC medium, and the lower chambers were filled with chemoattractant, 10% FBS (catalog #F0926, Sigma-Aldrich) containing DMEM Ham's F12 50/50 medium, supplemented with 1× antibiotic-antimycotic solution, followed by incubation at 37°C and 5% CO2 for 24 h. Invasive cells degraded the ECM proteins and passed through the pores of the membrane of the cell culture insert, while noninvading cells remained inside the insert and were removed with a cotton swab. Invasive cells adhering to the outer side of the inserts were stained and quantified by measuring OD (absorbance) at 560 nm (Synergy HTX multimode reader, BioTek).
Live cell time-lapse microscopy experiments.
GSCs were obtained by dissociating spheroid cultures with Accutase (catalog #A6964, Sigma-Aldrich), and live cell counts were determined by a Vi-CELL Auto Cell Viability Analyzer (Beckman Coulter). Live cells (between 2 × 104 and 7.5 × 104) were plated on each quadrant of 35 10 mm four-chambered glass bottom (175 ± 15 µm) tissue culture dishes (CELL View; catalog #627870, GREINER BIO-ONE) coated with laminin (Engelbreth-Holm-Swarm murine sarcoma basement membrane; catalog #L2020, Sigma-Aldrich), in 1× B27 supplement (catalog #17504-044, Thermo Fisher Scientific) and 1× penicillin/streptomycin/actinomycin solution (catalog #MT-30 004-CI, Corning) containing phenol red-free Ham's DMEM/F-12 50/50 medium (catalog #21041025, Thermo Fisher Scientific). Adherent cells were incubated for 8–10 h at 37°C and 5% CO2 and in a humidified environment before the start of imaging. A Leica TCS SP8 confocal microscope with resonant scanner, and a humidified chamber with 5% CO2 maintained at 37°C (Advanced Microscopy Core, Department of Epigenetics and Molecular Carcinogenesis, The University of Texas MD Anderson Cancer Center) was used to capture z-stack images of GFP+ live adherent GSCs from three independent fields of view per quadrant and per cell type at a time lapse of 15 min for up to 24–36 h. Single-cell migration was tracked and quantified by TrackMate, a single-particle tracking plugin in Fiji (ImageJ, NIH). Total distance traveled by cells, their mean velocity and directionality of migration were calculated for each cell type by Chemotaxis and Migration Tool 2.0 (Ibidi). Trajectories of 2D cellular migration were plotted. Cell aggregation indices were determined by using the measurement module in Fiji application by quantifying hourly cell cluster area, mean GFP signal, and cluster density between 12 and 24 h. All parameters of live-cell time-lapse imaging, image analyses, and quantitation were kept constant across all fields of view and cell types.
RNA sequencing and bioinformatics.
Spheroids from NT shRNA (n = 3), MLC1 shRNA (n = 3), and GlialCAM shRNA (n = 3) GSC6-27 cultures were washed in cold 1× PBS, and total RNA was extracted using the RNeasy Plus Mini Kit (catalog #74134, Qiagen). Following RNA quality validation [input resistance, ≥7], samples were sequenced using NextSeq500 high-output 75 nt phycoerythrin flow cell instrument (Illumina Next Generation Sequencing Core facility, Advanced Technology Genomics Core, MD Anderson Cancer Center) by generating 44 million paired end reads for each sample. Sequenced reads fastq files were aligned with Star/2.6.0b, the number of reads per gene were counted with HTSeq/0.11.0, and the read counts were normalized with “DESeq2.” Plots, hierarchical clustering, and principal component analysis (PCA) were used for quality assessment, and normalized log2 read counts were fitted with negative binomial GLM and tested by Wald statistics. Regularized log-transformed data were used for visualization, clustering, and PCA. Beta-uniform mixture model was used for determining p-values from Wald statistical tests. We analyzed an average of ∼45 million paired end reads generated for each of the GSC samples (n = 3 NT shRNA, MLC1 shRNA, or GlialCAM shRNA). Sequenced reads fastq files were mapped to human reference genomes using Tophat (Kim et al., 2013). Raw reads were calculated using HTseq (Anders et al., 2015). Differentially expressed genes, defined as those having logFC > 2 and false discovery rate (FDR) < 0.01, were obtained with EdgeR package analyses (Robinson et al., 2010). The signal-to noise metric was used to calculate the gene expression differences between cell samples. Kyoto Encyclopedia of Gene and Genome (KEGG) pathways were compiled from MsigDB (Subramanian et al., 2005). Unsupervised gene set enrichment analysis (GSEA) for all the KEGG pathways was performed using GSEA. Normalized enrichment scores and FDR values were calculated under a 1000-fold permutation. We queried the National Center for Biotechnology Information (NCBI)-Gene Expression Omnibus (GEO) dataset GSE131928 that has transcriptomic profiles of 24,131 single cells derived from 28 IDH1 wild-type adult and pediatric GBM patients (Neftel et al., 2019). Of the 7930 single cells sequenced by the Smartseq2 platform, we analyzed GlialCAM, MLC1, and AQP4 gene expression profiles in ∼5000 GBM single cells isolated from 20 adult GBM patient tumors. Cells that did not show detectable expression of any of our genes of interest were excluded from analysis.
Experimental mice.
Mouse studies were reviewed and approved before experimentation under the guidelines of the Institutional Animal Care and Use Committee and The University of Texas MD Anderson Subcommittee on Animal Studies, both AAALAC accredited institutions. Male nude (NCR ν/ν) mice were purchased from The Jackson Laboratory and were used for all GSC implantation experiments. Randomization was not used since all mice used were the same age and sex. Seven-week-old mice were anesthetized, and a Hamilton syringe was used to dispense 1.5 × 105 GSC6-27 cells infected with pGIPZ with NT shRNAs or GlialCAM shRNAs (n = 10 mice/cell type). The sample size was selected to ensure power analyses using α = 0.05 and an effect size = 0.4 for comparing the two groups using one-way ANOVA. Mice were killed at 16 weeks postinjection, perfused with 4% paraformaldehyde (PFA)/PBS, and brains were sectioned for experimental analyses. Animals were excluded from the analysis if death occurred before 16 weeks after injection.
Reverse-phase protein arrays.
GSCs (with or without GlialCAM shRNAs) were washed twice in ice-cold PBS, then lysed in radioimmunoprecipitation assay (RIPA) or reverse-phase protein array (RPPA) lysis buffer containing 1% Triton X-100, 50 mm HEPES pH 7.4, 150 mm NaCl, 1.5 mm MgCl2, 1 mm EGTA, 100 mm NaF, 10 mm sodium pyruvate, 1 mm Na3VO4, 10% glycerol, and a cocktail of protease and phosphatase inhibitors (Roche Diagnostics) for 20–30 min with frequent mixing on ice; then they were centrifuged at 14,000 rotations per minute (rpm) at 4°C for 15 min to isolate the detergent-soluble protein supernatant. The protein concentration was determined using the Pierce BCA Protein Assay Kit (catalog #2327, Thermo Fisher Scientific). The optimal protein concentration of lysates for RPPA is ∼1.2 μg/μl. Lysates were denatured in 4% SDS/2-mercaptoethanol (2-ME) sample buffer (35% glycerol, 8% SDS, 0.25 m Tris-HCl, pH 6.8; no β-mercaptoethanol) for 5 min at 95°C. Lysates were stored at −80°C and subsequently analyzed in the RPPA core facility at MD Anderson Cancer Center. Samples were serially diluted and probed with 466 antibodies and arrayed on nitrocellulose-coated slides.
ECM adhesion assays.
Cell adhesion assays were performed using ECM cell adhesion array kit (CytoSelect 48-Well Cell Adhesion Assay; catalog #CBA-070, Cell Biolabs). Spheres were dissociated using Accutase (catalog #AT104, Innovative Cell Technologies) followed by centrifugation [1000 rpm, 5 min, room temperature (RT)]. A viable cell count was obtained (Vi-CELL Analyzer, Beckman Coulter), and cells were gently resuspended in 4 ml of assay buffer. Subsequently, 150 µl of the cell suspension containing 1 × 106 viable GSCs in serum-free media (GSC media lacking growth supplements) were added to each well and incubated for 90 min at 37°C and 5% CO2. The nonadherent cells were gently removed by rinsing each well four times with 1× PBS. The resulting adherent cells were fixed and stained at RT for 10 min. Cells were washed four times with deionized water. Stain was extracted from each well, and absorbance readings were taken at 560 nm (Synergy HTX Multimode Reader, BioTek).
Immunofluorescence.
Fixed brain samples were embedded in 4% agarose, sectioned (100 µm thickness) on a vibratome and stored in 1× PBS at 4°C. Alternatively, brains were processed for paraffin embedding and sectioning. Sections were permeabilized and blocked with 10% donkey serum in PBS-T (1× PBS supplemented with 0.1–0.25% Triton X-100) for 1 h at RT, followed by an overnight 4°C incubation with primary unconjugated antibodies diluted in the blocking solution. Immunofluorescence analyses were performed with the following primary antibodies: anti-CD31 (catalog #AF3628, R&D Systems); anti-GFP (catalog #ab290 or #ab13970, Abcam; catalog #GFP-1020, Aves Labs); anti-GFAP (catalog #Z0334 DAKO; or catalog #NBP1-05198, Novus Biologicals); anti-Iba1 (catalog #0919741, WAKO); and anti-AQP4 (catalog #HPA014784, Sigma-Aldrich). The sections were then washed with PBS-T and incubated with secondary antibodies (dilution, 1:500) in the blocking solution for 1 h. Sections were again washed three times with PBS-T, then briefly washed with PBS. The sections were mounted on pretreated microscope slides, sealed using VECTASHIELD Antifade Mounting Medium with DAPI (Vector Laboratories) and kept at 4°C until imaging. Confocal images were acquired using FLUOVIEW FV3000 microscope (Olympus) with 10×, 20×, and 30× objectives. All comparative images were taken with the same laser power and gain settings to make qualitative and quantitative comparisons between staining levels in different samples. Multiple fields were imaged from biological replicates.
GBM cell fluorescence staining.
Chamber slides (Nunc Lab-Tek, Thermo Fisher Scientific) were coated with 1× PBS containing fibronectin (1:60; catalog #F0895, Sigma-Aldrich) overnight at 4°C. Chambers were rinsed thrice with 1× PBS; 400 µl of 1× PBS containing 2% BSA (catalog #SH30574, HyClone) was added to each well and incubated at RT for 1 h. GSC spheroids in culture were dissociated by Accutase (catalog #AT104, Innovative Cell Technologies), and single cells were resuspended in DMEM Ham's F12 50/50 medium containing 2% BSA. Viable cell count was obtained (Vi-CELL Analyzer, Beckman Coulter) and 5 × 103 viable GSCs were added to each well containing BSA-DMEM Ham's F12 50/50 medium followed by incubation at 37°C and 5% CO2 for 2 h. Adherent GSCs were gently rinsed (5 min, at RT) with 1× PBS, fixed with 4% PFA in 1× PBS at RT for 10 min and rinsed thrice (5 min each, at RT) with ice-cold 1× PBS. Cells were permeabilized with 0.1% Triton X-100 for 10 min at RT and rinsed twice with 1× PBS. Cells were blocked with 1% BSA in 1× PBS for 30 min at RT and incubated (4°C) overnight with the following primary unconjugated antibodies: anti-paxillin (1:100; catalog #AHO0492, Thermo Fisher Scientific); anti-GlialCAM (1:100; catalog #MAB4108, R&D Systems); anti-GFP (1:1000; catalog #ab13970, Abcam); anti-MLC1 (1:200; catalog #NBP1-80073, Novus); and anti-aquaporin-4 (1:250; catalog #HPA014784, Sigma-Aldrich) diluted in blocking buffer. Cells were rinsed three times (5 min each, at RT) with wash buffer (0.1% BSA in 1× PBS) and incubated (at RT) for 45 min with fluorochrome-conjugated secondary antibodies; anti-chicken-Alexa Fluor 488 (1:500; catalog #703–545-155, Jackson ImmunoResearch), anti-mouse-Alexa Fluor 594 (1:500; catalog #715–585-150, Jackson ImmunoResearch), and anti-rabbit-Alexa Fluor 594 (1:500; catalog #711–585-152, Jackson ImmunoResearch) diluted in blocking buffer. Additionally, Phalloidin-iFluor 647 (1×; catalog #ab176759, Abcam) was used (in combination with secondary antibody solutions) to stain actin filaments. Cells were rinsed thrice (5 min each, at RT) with wash buffer, rinsed twice with 1× PBS, and finally the chamber slides were sealed using VECTASHIELD Antifade Mounting Medium with DAPI (Vector Laboratories). Confocal Images were acquired using an FLUOVIEW FV3000 microscope (Olympus) with 30× and 60× oil-immersion objectives. All images were captured with the same laser power and gain settings to make comparisons between staining levels in different samples. Multiple fields of view were imaged from biological replicates.
Immunohistochemistry.
Immunohistochemical staining was performed on formalin-fixed, paraffin-embedded mouse brain tissue. The tissue was deparaffinized at 65°C and rehydrated in decreasing concentrations of alcohol. Heat-induced antigen retrieval was performed using a Na-citrate buffer at pH 6. The tissue was blocked in 10% serum of the same species as the secondary antibody for 1 h at room temperature. The tissue was then incubated overnight at 4°C in primary unconjugated antibodies diluted in blocking buffer. Immunohistochemical analyses were performed with the following primary antibodies: rabbit anti-GlialCAM (1:200; catalog #18177–1-ap, Proteintech); anti-Mlc1 antibody (catalog #NBP1-81555, Novus Biologicals); anti-aquaporin-4 (catalog #16473–1-ap, Proteintech); and human-specific goat anti-Vimentin (catalog #AF2105, R&D Systems). Slides were then washed in 1× PBS and incubated with biotinylated secondary antibodies (1:500–1:1000) in the blocking buffer for 1 h at RT. Following another series of washes in PBS, the slides were incubated in ABC Reagent (catalog #PK-4000, Vector Laboratories) for 30 min at RT. Following more washes with PBS, ImmPACT DAB Substrate (catalog #SK-4105, Vector Laboratories) was added for 5–10 min before rinsing the slides in double-distilled H2O and counterstaining with hematoxylin for 10–15 s. After rinsing in cold tap water, the slides were then dehydrated in increasing concentrations of alcohol and mounted using mounting media and stored until images were collected. Light microscope images were acquired using a microscope (model BX43, Olympus) with a 20× objective.
Immunoblotting.
Whole-cell lysates were collected under normal culture conditions by lysing spheres in RIPA buffer (50 mm Tris-HCl pH 8.0, 150 mm NaCl, 1% sodium deoxycholate, 1% Triton X-100, 0.1% SDS, 1% NP-40, and 1 mm EDTA) containing protease and phosphatase inhibitors (catalog #A32955, #A32957, Thermo Fisher Scientific) to obtain soluble protein fractions. Total protein was measured by Pierce BCA Protein Assay Kit (catalog #23227, Thermo Fisher Scientific), and then denatured at 95°C for 10 min in 4× Laemmeli buffer (catalog #1610747, BIO-RAD) containing 2.5% 2-ME (catalog #M6250, Sigma-Aldrich). A total of 30–75 μg of protein was resolved on 4–15%, 10%, or 12% Tris-glycine gels. Immunoblotting was performed with nitrocellulose membranes (catalog #1620112, #1620115, BIO-RAD), blocked using Odyssey TBS-based blocking buffer (LI-COR) or 5% milk in 0.1% Tween-20 containing 1×-TBS (TBST), and then incubated with specific primary antibodies diluted in blocking buffer overnight at 4°C. The following primary antibodies were used for immunoblotting: anti-GlialCAM (1:500; catalog #18177-1-ap, Proteintech); anti-GlialCAM (1:1000; catalog #MAB4108, R&D Systems); anti-GlialCAM (1:1000; catalog #ab300571, Abcam); anti-MLC1 (1:500; catalog #ab186436, Abcam); anti-Src (1:1000; catalog #2109, CST); anti-phospho-Src-pY416 (1:1000; catalog #2101, CST); anti-phospho-Src-pY527 (1:1000; catalog #2105, CST); anti-p130Cas (1:1000; catalog #13846, CST); anti-phospho-p130Cas-pY410 (1:1000; catalog #4011, CST); anti-phospho-FAK-pY925 (1:1000; catalog #3284, CST); anti-phospho-paxillin-pY118 (1:1000; catalog #2541, CST); anti-phospho-tyrosine pY99 (1:1000; catalog #sc-7020, Santa Cruz Biotechnology); anti-integrin β1 (1:2000; catalog #ab183666, Abcam); anti-aquaporin-4 (1:1000; catalog #16473–1-ap, Proteintech); and anti-aquaporin-4 (catalog #HPA014784, Sigma-Aldrich). Target proteins were normalized to total cellular/housekeeping proteins: anti-α-actinin (1:3000; ab18061, Abcam); anti-β-actin (1:3000; A5441, Sigma-Aldrich; and 1:3000; catalog #BS-0061R-TR, Bioss), and anti-GAPDH (1:2000; catalog #G8796, Sigma-Aldrich). Blots were incubated with fluorochrome-conjugated secondary antibodies (1:15,000 dilutions; IRDye 800CW goat anti-rabbit and IRDye 680RD goat anti-mouse, LI-COR) in the dark at RT for 30 min or with horseradish peroxidase-conjugated secondary antibodies (1:5000; goat anti-rabbit; catalog #1705046, BIO-RAD; and 1:5000; and goat anti-mouse; catalog #1706516, BIO-RAD) diluted in 5% milk-TBST, incubated for 1 h at RT. Dual-channel infrared scan and quantitation of immunoblots were conducted using the Odyssey CLx Infrared Imaging System (LI-COR) with Image Studio (version 5.2; LI-COR) for detecting fluorochrome-conjugated secondary antibodies. For HRP-conjugated secondary antibodies, signals were detected with an enhanced chemiluminescence reagent (catalog #NEL-104001EA, PerkinElmer) and developed in a ChemiDoc Imaging System (BIO-RAD).
Image acquisition, analysis, and quantitation.
Immunofluorescence images were acquired using an Fluoview FV3000 Confocal Laser Scanning Microscope (Olympus). Multidimensional acquisition was conducted using z-stacks with 2.5 µm slicing intervals at a scan rate of 4 ms/pixel with a resolution of at least 1024 × 1024 pixels per slice and digitally compiled in FV31S-SW (version 2.4.1.198). Image acquisition parameters, including exposure time, laser power, gain, and voltages, were fixed for each imaging channel. Immunohistochemistry-labeled images were captured using an Olympus BX43 light microscope. Using ImageJ, all images were scaled (in µm) as per the objective lens used for acquisition to measure signals and cellular parameters. Acquired image z-stacks were projected for maximum intensity to include all signals, but the “auto-threshold” module was used to include only cellular signals for quantitation and exclude nonspecific background or noise. Fluorescence signal intensity of channels was measured using the standard “color histogram” module. Cell counts, mean fluorescence signals from single cells and cellular peripheries, and major and minor axes for defining cell shape were analyzed using the “analyze particles” algorithm (De et al., 2022).
Statistical analyses.
Quantitation of confocal images was performed using ImageJ software (NIH). GraphPad Prism 9.0 was used to plot mean values and all data points (n = ≥3 ± SEM per group, unless otherwise indicated) to compare between experimental and control samples and to determine statistical differences by unpaired Student's t test and one-way or two-way ANOVA (Tukey's post hoc test, Bonferroni's test, or Dunnett's test analyses, where applicable) at 95% confidence intervals (α value = 0.05). Statistics were considered significant at *p < 0.05, ** p < 0.01, *** p < 0.001, and **** p < 0.0001.
Results
GlialCAM promotes GSC self-renewal but suppresses migration and invasion in vitro
We first investigated GlialCAM and Mlc1 protein expression by immunoblotting detergent-soluble lysates from human GSCs. Cultured GSCs grow as free-floating spheroids and drive tumor initiation, invasion, and recurrence after therapy in vivo (Gimple et al., 2019). As shown in Figure 1A, there were detectable levels of GlialCAM in five of six human GSC samples analyzed. In contrast, Mlc1 protein was expressed at more variable levels, with only three of five GSCs showing detectable expression by immunoblotting (Fig. 1A). We selected GSC6-27 spheroids for more detailed mechanistic analyses, since these cells express high levels of GlialCAM and Mlc1 protein (Fig. 1A) and generate proliferative and infiltrative brain tumors after implantation in xenograft mice (Chen et al., 2018). To manipulate endogenous GlialCAM and Mlc1 protein levels, GSCs were stably infected with pGIPZ lentiviruses expressing GFP as well as targeted shRNAs (n = 2 shRNAs per gene). A pGIPZ lentivirus expressing GFP and NT shRNA was used as a negative control. The shRNAs targeting GlialCAM resulted in a variable reduction (20% and 80%) in protein expression (Fig. 1B–D). Both shRNAs targeting MLC1 resulted in an 80% reduction in protein expression as determined by immunoblotting detergent-soluble cell lysates and by immunofluorescence with anti-Mlc1 antibodies (Fig. 1B,C,E). Silencing the expression of the MLC1 gene did not result in a significant reduction in the expression of the GlialCAM protein and vice versa (Fig. 1B–E).
GlialCAM and Mlc1 promote GSC proliferation and spheroid formation. A, Detergent-soluble lysates from six different primary human GSC spheroid cultures were immunoblotted with anti-GlialCAM and anti-Mlc1 antibodies. Note the detectable levels of GlialCAM protein in five of six GSC samples, whereas Mlc1 protein is detected in three of five samples. B, pGIPZ lentivirus-expressed shRNAs were tested for GlialCAM (n = 2) and Mlc1 (n = 2) gene silencing in infected GSC6-27 cells. The specific shRNAs cause diminished GlialCAM and Mlc1 protein expression to varying levels, as determined by immunoblotting. Note that silencing GlialCAM does not significantly impact Mlc1 expression, and vice versa. As controls for these experiments, we used GSC6-27 cells infected with pGIPZ expressing NT shRNAs. C, Adherent GSCs expressing control shRNAs (left) or shRNAs targeting GlialCAM (middle) or Mlc1 (right) were analyzed for GlialCAM (green) or Mlc1 (red) expression. Scale bars, 50 µm. D, E, Quantitation of GlialCAM (D) and Mlc1 (E) protein levels in GSCs following shRNA-mediated gene silencing (one-way ANOVA, Dunnett's multiple-comparisons test: *p < 0.05, **p < 0.01, ****p < 0.0001). F, GSC6-27 cells infected with pGIPZ lentivirus-expressing control NT shRNAs formed spheroids that were obviously larger by 7 d in culture than spheroids formed from cells expressing GlialCAM shRNAs (middle) or MLC1 shRNAs (bottom). Scale bars, 100 µm. G, The diameter of newly formed spheroids was recorded each day, revealing that spheroids with reduced GlialCAM were significantly smaller compared with control GSC6-27 cells that express GlialCAM (unpaired Student's t test for comparison, n = 4, ****p < 0.0001). H, The percentages of newly formed spheres (>20,000 µm2) were recorded daily for 7 consecutive days. GSC6-27 cells expressing GlialCAM shRNAs showed obviously reduced spheroid formation with significantly smaller cross-sectional areas (ANOVA and Tukey's post hoc analysis for comparison, n = 4, ****p < 0.0001). I, The diameter of newly formed spheroids from GSC6-27 cells infected with pGIPZ lentivirus-expressing control NT shRNAs was recorded each day, revealing that spheroids with reduced Mlc1 were significantly smaller compared with control GSC6-27 cells that express Mlc1 (unpaired Student's t test for comparison, n = 4, ****p < 0.0001). J, The percentages of newly formed spheres (>20,000 µm2) were recorded daily for 7 consecutive days. GSC6-27 cells expressing MLC1 shRNAs showed obviously reduced spheroid formation with significantly smaller cross-sectional areas (ANOVA and Tukey's post hoc analysis for comparison, n = 4, ****p < 0.0001).
GSC6-27 cells were next tested in sphere formation assays to determine GlialCAM-dependent and Mlc1-dependent functions in GSC proliferation and self-renewal. Analysis of spheroid formation revealed a major reduction in growth capacities of GBM cells expressing GlialCAM or Mlc1 targeting shRNAs (Fig. 1F). In addition, spheroid sizes were significantly reduced in GSCs expressing reduced GlialCAM or Mlc1 compared with control cells (Fig. 1G,I). After 7 d in culture, cross-sectional areas of nearly 40% of control GSC spheroids were ≥2 × 104 µm2 compared with only 10% or less of GSC spheroids with low levels of GlialCAM or Mlc1 expression (Fig. 1H,J).
Next, we investigated functions for GlialCAM and Mlc1 in live cell motility using time-lapse video microscopy. Compared with control GSCs that express Mlc1 and GlialCAM (Fig. 2A), cells expressing reduced levels of GlialCAM show enhanced motility (Fig. 2B). Quantitative analysis of motility patterns revealed that the loss of GlialCAM resulted in increased total distance migrated by these cells as well as elevated cellular velocity (Fig. 2D,E). In contrast, GSCs with low Mlc1 levels displayed reduced motility (Fig. 2C) based on the quantification of total distance traveled (Fig. 2D,E). We also investigated GlialCAM-dependent and Mlc1-dependent cell invasion in three-dimensional ECM assays. GSC6-27 cells with low levels of GlialCAM showed increased invasion through ECM-coated transwells (Fig. 2F). In contrast, GSC6-27 cells with low levels of Mlc1 showed diminished invasion through ECM-coated transwells (Fig. 2F), which is consistent with our previously published studies (Lattier et al., 2020). Last, analysis of time-lapse fluorescent imaging data over 24 h revealed that GSCs expressing GlialCAM showed intercellular adhesion behaviors leading to the formation of multicellular clusters. In contrast, GSCs with low GlialCAM levels did not form distinct clusters and instead migrated as individual cells that remained more motile and displayed limited intercellular adhesions (Fig. 2G).
GlialCAM and Mlc1 differentially control GSC motility and invasion. A–C, Migration patterns for GSCs expressing NT shRNAs (A), GlialCAM shRNAs (B), and MLC1 shRNAs (C). D, E, Quantitation of GSC migration distance (D) and velocity (E; one-way ANOVA, Tukey's post hoc multiple-comparisons test: *p < 0.05, **p < 0.01, ****p < 0.0001). F, GSC6-27 cells expressing GlialCAM shRNAs showed enhanced invasion through basement membrane-coated transwells compared with NT shRNA control cells (unpaired Student's t test for comparison, n = 3, **p < 0.01). In contrast, GSC6-27 cells expressing MLC1 shRNAs showed diminished invasion through basement membrane-coated transwells compared with NT shRNA control cells (unpaired Student's t test for comparison, n = 3, **p < 0.0001). G, Quantitation of GSC aggregation based on live-cell imaging analyses over 12 and 24 h. Compared with NT shRNA GSCs, aggregation indices are lower for cells with shRNAs targeting GlialCAM and Mlc1. Statistical comparisons are shown for NT shRNA versus GCAM and MLC1 shRNAs using two-way ANOVA and Bonferroni's multiple-comparisons test: **p < 0.01, ****p < 0.0001.
Analysis of GlialCAM-dependent and Mlc1-dependent gene expression profiles in human GSCs
Quantitative RNAseq was performed (n = 3 samples per shRNA) to identify GlialCAM-dependent and Mlc1-dependent gene expression in human GSCs (Fig. 3A,C). Among the top 50 GlialCAM-regulated genes, we found the water channel AQP4, which shows a nearly twofold increase in expression (n = 3, p < 0.00,001) in GSCs with low levels of GlialCAM (Fig. 3B). Although statistical filtering (FDR, 0.05; and fold change >2 or <2) did not identify AQP4 in the top 50 Mlc1-regulated genes (Fig. 3C,D), AQP4 showed a fivefold decrease in expression (n = 3, p < 0.00001) in GSCs with low levels of Mlc1 (Fig. 3E). Further analysis of AQP4 mRNA expression showed inverse levels in GSCs expressing GlialCAM shRNAs versus Mlc1 shRNAs (Fig. 3E). To validate the RNA sequencing results, we analyzed aquaporin-4 protein levels in cultured GSCs. Cells with low levels of GlialCAM expressed higher amounts of aquaporin-4 and displayed more polarized morphologies (Fig. 3F,G). In contrast, GSCs expressing reduced levels of Mlc1 showed reduced levels of aquaporin-4 and displayed more rounded and nonpolarized morphologies (Fig. 3F,G).
Identification of GlialCAM-dependent and Mlc1-dependent gene expression in GSCs. A, Heat map showing the top 50 differentially expressed genes in GSCs expressing NT shRNAs versus GlialCAM shRNAs. Differentially expressed genes were identified using the EdgeR package with adjusted p-value cutoffs <0.05 and log2 fold changes >2. B, Relative levels of the top 50 differentially expressed protein-coding genes in GSCs expressing GlialCAM shRNAs compared with NT control shRNAs. Genes showing log2 fold change >0 are overexpressed in GSCs expressing GlialCAM shRNAs, whereas genes with log2 fold change <0 show lower expression in GSCs expressing GlialCAM shRNAs. FC, Fold change. C, Heat map showing the top 50 differentially expressed genes in GSCs expressing NT shRNAs versus MLC1 shRNAs. Differentially expressed genes were identified using the EdgeR package with adjusted p-value cutoffs <0.05 and log2 fold changes >2. D, Relative levels of the top 50 differentially expressed protein-coding genes in GSCs expressing MLC1 shRNAs compared with NT control shRNAs. Genes above the 0 line are overexpressed in GSCs expressing MLC1 shRNAs, whereas genes below the 0 line show lower expression in GSCs expressing MLC1 shRNAs. FC, Fold change. E, Comparison of AQP4 mRNA levels in GSCs expressing GlialCAM shRNAs versus MLC1 shRNAs. Note the inverse level of AQP4 expression (unpaired Student's t test, n = 3; ****p < 0.0001). F, Adherent GSCs expressing control shRNAs (left) or shRNAs targeting GlialCAM (middle) or Mlc1 (right) were analyzed for GlialCAM (green) or aquaporin-4 (red) protein expression. Note the elevated levels of aquaporin-4 in cells expressing lower levels of GlialCAM, whereas cells expressing reduced levels of Mlc1 have diminished levels of aquaporin-4. Scale bars: 50 µm. G, Quantitation of aquaporin-4 protein expression in GSCs expressing control shRNAs or shRNAs targeting GlialCAM or Mlc1 (unpaired Student's t test for comparison: ***p < 0.001, ****p < 0.0001).
GSEA was next performed and lists of KEGG pathways (Subramanian et al., 2005) were compiled, comparing control GSCs to GSCs with low levels of GlialCAM and Mlc1. KEGG pathways that are upregulated in GSCs in a GlialCAM-dependent manner include a vasopressin water reabsorption pathway related, in part, to the higher expression of AQP4 (Fig. 4A,C). In contrast, we detect a Mlc1-dependent decrease in the KEGG pathway signature for vasopressin water reabsorption related to the downregulation of AQP4 expression (Fig. 4B,D). There are 44 gene signatures for the KEGG vasopressin water reabsorption pathway, with 13 (29.6%) and 21 (47.7%) of these genes showing core enrichment in GSCs with low levels of GlialCAM and Mlc1, respectively. Interestingly, 9 of the 13 genes that showed positive core enrichment for the KEGG vasopressin water reabsorption pathway in GSCs with low levels of GlialCAM, showed negative core enrichment in GSCs with low levels of Mlc1 (Fig. 4E). In GSCs with low levels of GlialCAM, AQP4 has the highest rank metric score of 18.05 with a running enrichment score of 0.293. In contrast, in GSCs with low levels of Mlc1, AQP4 has the lowest rank metric score of −26.94 with a running enrichment score of 0.004. In addition to the vasopressin water reabsorption pathway, other KEGG pathways showing inverse enrichment correlations include ECM receptor adhesion, gap junctions, and hematopoietic cell lineage (Fig. 4A,C).
Analysis of KEGG pathways linked to GlialCAM and Mlc1 in human GSCs. A, Pathways were identified by GSEA based on differential levels of GlialCAM expression in GSCs. The normalized enrichment score (NES) and the log-transformed (-Log) q-values are shown for the top 20 KEGG pathways. B, Pathways were identified by GSEA based on differential levels of Mlc1 expression in GSCs. The NES and the log-transformed (-Log) q-values are shown for the top 20 KEGG pathways. C, Analysis of the enrichment plot for the upregulated vasopressin water reabsorption pathway in GSCs with low levels of GlialCAM reveals that among the 13 core-enriched genes, AQP4 has the highest rank metric score of 18.05 with a running enrichment score of 0.293. D, Analysis of the enrichment plot for the downregulated vasopressin water reabsorption pathway in GSCs with low levels of Mlc1 reveals that among the 21 core-enriched genes, AQP4 has the least rank metric score of −26.94 with a running enrichment score of 0.004. E, Comparison of the 9 genes in the KEGG vasopressin water reabsorption pathway showing core enrichment in GSCs expressing GlialCAM shRNAs or Mlc1 shRNAs. Note the inverse enrichment levels for all 9 genes, with positive enrichment in GSCs with low levels of GlialCAM and negative enrichment in GSCs with low levels of Mlc1.
Functional connections between GlialCAM and aquaporin-4 in GSCs
Our data thus far reveal that GlialCAM plays important roles in promoting cell proliferation and suppressing cell motility/invasion (Fig. 2) and that AQP4 is upregulated in cells with low levels of GlialCAM. Interestingly, aquaporin-mediated water influx is critical for control of cytoskeletal dynamics and cell shape changes necessary for polarization and motility (Loitto et al., 2002). Therefore, we next analyzed GlialCAM and aquaporin-4 connections in GSC polarity and migration using in vitro scratch-wound assays (Liang et al., 2007). Compared with control cells that express GlialCAM (Fig. 5A), enhanced migration from the wound edge (Fig. 5B) and invasive capacities of GSCs expressing low levels of GlialCAM were largely ameliorated following treatment with [2-(nicotinamide)−1,3,4-thiadiazole], TGN-020, a small-molecule inhibitor of aquaporin-4 water transport (Fig. 5A,B,D; Igarashi et al., 2011). TGN-020 is reported to be highly selective for aquaporin-4 versus other aquaporins (Huber et al., 2009). Compared with control GSCs expressing GlialCAM, cells expressing low levels of GlialCAM displayed AQP4-dependent changes in cell shape indicating alterations in cell polarization critical for water influx-mediated cell motility and invasion. (Fig. 5C). In GSCs expressing reduced levels of GlialCAM, we also detected more pronounced F-actin-rich filopodia as determined by staining cells with phalloidin (Fig. 5E). Treatment of GSCs with TGN-020 ameliorated the GlialCAM-dependent differences in filopodial number and length (Fig. 5F,G), but did not impact aquaporin-4 protein levels (Fig. 5H).
GlialCAM regulates GSC polarity and invasion via aquaporin-4 water transport. A, B, Monolayers of GFP+ GSC6-27 cells were analyzed for GlialCAM-dependent polarity and migration using scratch-wound assays. At 24 h after wounding, cells were labeled with anti-GlialCAM (red) and anti-aquaporin-4 (cyan) antibodies. Note that compared with control GSCs (A), cells expressing low levels of GlialCAM have increased migratory capacities showing enhanced polarization and migration away from the wound edge (B). Scale bars, 30 µm. C, Adherent GSC6-27 cells expressing control shRNAs or shRNAs targeting GlialCAM were treated with TGN-020 (10 µm, 24 h) and then were analyzed for alterations in morphology. There are significant AQP4-dependent shape changes following TGN-020 treatment selectively in cells expressing GlialCAM shRNAs (unpaired Student's t test for comparison: ***p < 0.001, *p < 0.05). D, GSC6-27 cells expressing control shRNAs or shRNAs targeting GlialCAM were treated with TGN-020 (10 µm, 30 min) followed by analysis for three-dimensional invasion through ECM-coated transwell filters. Compared with control cells, GSC6-27 cells expressing diminished levels of GlialCAM show increased invasive capacities. Treatment with the aquaporin-4 inhibitor leads to diminished invasion of GSCs in a GlialCAM-dependent manner (unpaired Student's t test for comparison, n = 3; **p < 0.01, *p < 0.05). E, Adherent GFP+ GSC6-27 cells expressing control NT shRNAs (left) or GlialCAM shRNAs (right) were treated with TGN-020 (10 µm, 24 h) and then labeled with phalloidin (cyan) and DAPI (red). Scale bars, 10 µm. F, G, GSC6-27 cells treated with TGN-020 were analyzed for numbers of filopodia (F) and filopodial length (G). Note cells expressing reduced GlialCAM levels have longer and increased numbers of filopodia, and this is ameliorated following TGN-020 treatment (unpaired Student's t test for comparison; *p < 0.05, ****p < 0.0001). H, GSC6-27 cells expressing control shRNAs or shRNAs targeting GlialCAM were treated with TGN-020 (10 µm, 24 h) or control buffer. Cells were then analyzed for aquaporin-4 protein expression based on immunofluorescence signal, revealing elevated levels of aquaporin-4 following silencing of GlialCAM. Note that levels of aquaporin-4 protein do not significantly change following treatment with TGN-020 (unpaired Student's t test for comparison; ****p < 0.0001).
We next performed RPPA analysis to identify potential GlialCAM-regulated adhesion and signaling pathways involved in GSC motility and invasion. RPPA is an antibody-based high-throughput quantitative system to study protein-signaling cascades (Zajec et al., 2021). The top 25 proteins that were differentially expressed (upregulated and downregulated) and/or phosphorylated in a GlialCAM-dependent manner are shown in Figure 6A. To confirm the RPPA results, detergent-soluble lysates from cells (n = 3 per shRNA) were next analyzed by immunoblotting for confirming select pathway components, particularly related to focal adhesion signaling. These include Src at the activating tyrosine 416 (Y416) along with enhanced tyrosine phosphorylation of focal adhesion kinase (FAK) at Y925 (Fig. 6B,C). Phosphorylation of FAK at Y925 is reported to promote cell polarity and migration (Deramaudt et al., 2011). In addition, antibodies recognizing the Src substrate paxillin revealed elevated phosphorylation at pY118 in cells expressing reduced levels of GlialCAM (Fig. 6B,C). Integrin β1 protein was also upregulated in GSC6-27 cells in a GlialCAM-dependent manner (Fig. 6B,C). GSC6-27 cells expressing control shRNAs or shRNAs targeting GlialCAM were next plated on various ECM proteins including fibronectin, laminin, collagens, and fibrinogen. Reduced GlialCAM expression correlated with increased adhesion to fibronectin but not to the other ECM proteins (Fig. 6D), suggesting activation of the fibronectin receptor α5β1 integrin.
GlialCAM controls ECM adhesion and signaling in GSCs. A, Bar graph summarizing select proteins that show statistically significant differences (fold change over control NT cells) in expression and/or phosphorylation in GSCs expressing GlialCAM shRNAs. Shown are the top 25 proteins displaying reduced (red) or elevated (green) expression and/or phosphorylation in GlialCAM shRNA cells (n = 3) compared with NT shRNA cells (n = 3). B, Immunoblots of detergent-soluble lysates from GSC6-27 cells expressing GlialCAM shRNAs versus control NT shRNAs show differential expression and/or phosphorylation of select ECM adhesion and signaling proteins. In lysates from cells expressing low levels of GlialCAM, note the increased expression of β1 integrin as well as higher levels of tyrosine phosphorylation of signaling proteins such as Src, Fak, and paxillin. C, Heat map from immunoblot data summarizing select proteins that show significant differences in expression and/or phosphorylation in GSCs expressing GlialCAM shRNAs versus control NT shRNA cells. Proteins displaying reduced (white) or elevated (red) expression and/or phosphorylation are shown. D, GSC6-27 cells expressing control shRNAs or GlialCAM shRNAs were added to tissue culture wells coated with the indicated ECM proteins, and cell adhesion was quantified. Note that reduced GlialCAM expression leads to increased cell adhesion selectively to fibronectin. Differences in cell adhesion to other ECM proteins are not statistically significant (ns). ANOVA and Tukey's post hoc analysis was used for comparisons, n = 3; ****p < 0.0001.
Genetic inhibition of GlialCAM leads to enhanced GBM cell invasion in vivo
We have previously reported that Mlc1 promotes GBM cell invasive growth in vivo (Lattier et al., 2020); however, roles for GlialCAM in GBM initiation and progression were not reported. Therefore, we next focused on understanding the functions for GlialCAM in tumor growth and invasion in vivo. GSC6-27 cells infected with pGIPZ virus expressing GFP and NT shRNAs (n = 5) or GlialCAM shRNAs (n = 9) were intracranially injected into the striatum of NCR ν/ν mice (Fig. 7A). Animals were monitored for tumor-related neurologic deficits over a 16 week period, and when the first animal showed deficits, all mice were subsequently killed. Anti-GFP antibodies were used to label fixed brain sections to monitor patterns of tumor cell growth and invasion. Tumor cells with low levels of GlialCAM showed robust patterns of invasion across the corpus callosum and into contralateral (noninjected) hemisphere, compared with control cells (Fig. 7B–I). Based on multiple images captured from different regions of the fixed and immunolabeled xenograft brain samples, on average, 254 GFP+ cells were found in the corpus callosum and 182 GFP+ cells were found in the contralateral hemispheres of mice injected with GSCs expressing low levels of GlialCAM. This is compared with 101 and 48 GFP+ cells in the same brain regions, respectively, of mice injected with control GSCs. Compared with control GSCs (Fig. 7E), cells with low levels of GlialCAM were more elongated and polarized, showing robust invasion in the corpus callosum and more infiltrative growth in the noninjected hemisphere (Fig. 7I). Quantitation of GFP signal intensity in the corpus callosum confirmed GlialCAM-dependent increases in cell invasion (Fig. 7J) as well as differences in GBM cell morphology (Fig. 7K).
GlialCAM suppresses cancer cell invasive growth in xenograft models of GBM. A, Experimental schema for analyzing GlialCAM-dependent tumor growth and invasion in xenograft mouse models of GBM. B–I, Coronal sections through the striatum of mice harboring xenograft tumors formed from human GSC6-27 cells expressing control shRNAs (B–E) or GlialCAM shRNAs (F–I) were H&E-stained (top panels) or immunofluorescently labeled with anti-GFP and anti-CD31 antibodies (middle and bottom panels) to visualize cancer cells (green) and blood vessels (red), respectively. Shown are images of the injected hemisphere (B, F), the corpus callosum (C, E, G, I), and the opposite (contralateral) hemisphere (D, H). Note that cells expressing GlialCAM shRNAs showed more robust invasion through the corpus callosum and distal growth in the opposite hemisphere. The boxed regions (B–D, F–H) in the middle panels are shown at higher magnification in the bottom panels. Scale bars: top panels, 50 µm; middle panels, 100 µm; bottom panels, 20 µm. E and I are magnified images of NT shRNA (C) and GlialCAM shRNA (G) xenograft tumors, highlighting perivascular GBM cell invasion in the corpus callosum. Scale bars: E, I, 50 µm. J, K, Quantitation of GSC6-27 cell invasive growth in vivo, revealing GlialCAM-dependent increases in cell invasion through the corpus callosum and tumor development in the contralateral hemisphere compared with cells expressing control NT shRNAs. Note that enhanced GFP expression in the corpus callosum confirms the presence of a higher number of invasive GSCs with reduced GlialCAM levels compared with GSCs expressing NT shRNAs that express normal GlialCAM protein levels (J; unpaired Student's t test for comparison; n > 100 cells; ****p < 0.0001). These invasive GSCs displayed marked changes in cellular shape (K; unpaired Student's t test for comparison, n > 100 cells; ****p < 0.0001).
Analysis of GlialCAM and aquaporin-4 protein expression in human GBM tissue
We next performed immunohistochemical staining on noncancerous human brain tissue as well as human GBM samples to interrogate spatial patterns of GlialCAM and aquaporin-4 protein expression in situ. In the noncancerous human cerebral cortex, GlialCAM protein was detected in astrocyte end feet that juxtaposed cerebral blood vessels (Fig. 8A). Anti-GlialCAM immunostaining of primary GBM samples taken from non-necrotic tumor core regions revealed expression in cancer cells, although enrichment around blood vessels was less apparent compared with noncancerous brain samples (Fig. 8B). Analysis of the Ivy GAP database (Puchalski et al., 2018), which contains spatial expression patterns from multiple human GBM samples, revealed high GlialCAM mRNA levels mainly within the tumor core and comparatively lower levels at the invasive leading edge (Fig. 8C). Aquaporin-4 protein was also detected in perivascular astrocyte end feet in the noncancerous human cerebral cortex (Fig. 8D). Elevated levels of aquaporin-4 expression were detected in the GBM core with some enrichment in perivascular tumor cells (Fig. 8E). AQP4 mRNA was more abundant in infiltrative and invasive GBM cells as well as in tumor cells of the GBM core (Fig. 8F).
Analysis of GlialCAM, Mlc1, and aquaporin-4 expression in human GBM samples. A, B, Anti-GlialCAM immunohistochemical staining of fixed noncancerous human brain (A) and resected human GBM tissue samples (B). Note that GlialCAM protein is enriched around blood vessels in the normal brain, whereas there is more diffuse expression of GlialCAM in tumor cells within GBM samples. Scale bars, 50 µm. C, Mining the Ivy GAP database for spatial differences in gene expression patterns reveals high levels of GlialCAM mRNA in the tumor core, with reduced levels at the invasive leading edge. D, E, Anti-aquaporin-4 immunohistochemical staining of fixed noncancerous human brain (D) and resected human GBM tissue sample (E). Note the enrichment of aquaporin-4 protein surrounding blood vessels in the normal brain, whereas there is elevated expression of aquaporin-4 in tumor cells within the GBM sample. Scale bars, 50 µm. F, Mining the Ivy GAP database for spatial differences in gene expression patterns reveals high levels of AQP4 mRNA in the infiltrating and invasive tumor regions as well as in cells of the tumor core. G, H, Correlation AnalyzeR software was used to identify genes that are coexpressed with GlialCAM (G) and AQP4 (H) in human brain cancer samples. Genomic data from the ARCHS4 platform were used for analysis. I, Querying ∼5000 cells from a GBM scRNAseq dataset from 28 different tumors (GSE131928; Neftel et al., 2019) reveals variable coexpression levels of GlialCAM and aquaporin-4 in different subpopulations of tumor cells. J, K, Analysis of GlialCAM/aquaporin-4 gene expression levels in more than ∼1500 double-positive tumor cells, reveals higher levels of aquaporin-4 compared with GlialCAM (unpaired Student's t test for comparison; ****p < 0.0001). L, High correlative expression between GlialCAM and aquaporin-4 in GBM cells (r = 0.98, ***p < 0.0001). M, Querying ∼5000 cells from the same GBM scRNAseq dataset from 28 different tumors (GSE131928; Neftel et al., 2019) also reveals variable coexpression levels of GlialCAM and Mlc1 in different subpopulations of tumor cells. N, O, Analysis of GlialCAM/Mlc1 gene expression levels in more ∼1500 double-positive tumor cells, reveals significantly higher levels of Mlc1 expression compared with GlialCAM (unpaired Student's t test for comparison; ****p < 0.0001). P, Lack of correlative expression between GlialCAM and Mlc1 in GBM cells (r = 0.21, ****p < 0.0001).
The Correlation AnalyzeR database (Miller and Bishop, 2021) was next queried to determine expression correlations between GlialCAM and AQP4 as well as to identify genes that are coexpressed with GlialCAM or AQP4 in human brain cancer samples. As shown in Figure 8, G and H, there was a significant correlation between AQP4 and GlialCAM/HepaCAM in human brain cancer samples. Next, we queried an open-source dataset containing gene expression profiles for single GBM cells isolated from 28 different human tumors (Neftel et al., 2019). As shown in Figure 8I, of the ∼5000 GBM cells analyzed, GlialCAM and aquaporin-4 were coexpressed (GlialCAM+/AQP4+) in ∼29% of cells, compared with ∼25% of cells that expressed only GlialCAM (GlialCAM+/AQP4-). Analysis of the 1471 double-positive cells (GlialCAM+/AQP4+) revealed that AQP4 mRNA expression levels were significantly higher compared with GlialCAM mRNA levels (Fig. 8J,K), although GlialCAM and AQP4 coincident expression was highly correlative in GBM cells (Fig. 8L). GlialCAM and Mlc1 were coexpressed (GlialCAM+/MLC1+) in ∼30% of cells, compared with ∼10% of cells that showed GlialCAM only (GlialCAM+/MLC1–). Analysis of double-positive (GlialCAM+/MLC1+) cells revealed that MLC1 mRNA expression was also significantly higher compared with GlialCAM mRNA levels (Fig. 8N,O); however, unlike AQP4 and GlialCAM we did not detect a significant coincident expression between Mlc1 and GlialCAM (Fig. 8P).
Collectively, these various in vitro and in vivo data reveal that GlialCAM is critical for balancing cell growth and invasive cell states via regulation of adhesion and signaling pathways as well as gene expression events, particularly involving the water channel aquaporin-4 (Fig. 9). High levels of GlialCAM promote GBM cell proliferation and intratumoral cell–cell adhesion, and dampen invasion, whereas cells with low levels of GlialCAM are more motile and invasive due, in part, to increased aquaporin-4 expression as well as enhanced focal adhesion and integrin signaling dynamics. Hence, targeting components of the GlialCAM adhesion and signaling network may be an effective strategy to selectively target proliferative versus invasive cell states in patients with GBM.
A model for GlialCAM regulation of GBM cell–cell adhesion and proliferation versus invasion in the brain microenvironment. A, GlialCAM is expressed in proliferative GBM cells, where it regulates cell–cell adhesion between cancer cells in the main tumor mass. B, Reduced GlialCAM expression promotes GBM cell invasion throughout the brain parenchyma. Invasive GBM cells display reduced cell–cell adhesion and enhanced cell–ECM adhesion associated with elevated signaling pathways involving integrins, aquaporin-4, and focal adhesion proteins. We propose that GlialCAM-dependent water transport by aquaporin-4 contributes to shape changes that facilitate GBM cell polarity and invasion in the brain parenchyma. This figure was created and adapted from BioRender.com.
Discussion
A central finding from this study is that GlialCAM promotes GBM cell–cell adhesion, thus limiting dispersal from the primary tumor mass. GBM cells with low levels of GlialCAM display reduced cell–cell adhesion, increased cell–ECM interactions, and enhanced focal adhesion signaling. These events collectively promote dispersal from the primary mass and invasion throughout the brain parenchyma. Our data also reveal that the water channel aquaporin-4 (Nagelhus and Ottersen, 2013) is a critical regulator of GBM cell shape, polarity, and invasion. Inhibition of aquaporin-4 water transport results in cell morphology changes and modifications in F-actin assembly and disassembly dynamics at the leading edge. We propose that GlialCAM negatively modulates water transporter activities, possibly by regulating localization of aquaporin-4 to the plasma membrane. Downregulation of GlialCAM leads to elevated aquaporin-4 functions, increased water transport, and enhanced cell migration and invasion. These results from genetic manipulation of GSCs are supported by data using TGN-020, a pharmacologic agent. At low doses, TGN-020 has been reported to be a selective inhibitor for aquaporin-4 water transport functions (Huber et al., 2009). At the concentrations of TGN-020 we have used, however, we cannot rule out roles for other aquaporin water transporters in GSC invasive growth.
In GSCs with low levels of GlialCAM, we detect increased tyrosine phosphorylation of Src and FAK, higher levels of β1 integrin, and enhanced adhesion to fibronectin. The cytoplasmic domains of integrin β subunits are involved in the recruitment of effector proteins, including talins and kindlins, that promote inside-out integrin activation (Ginsberg, 2014). GlialCAM contains an ∼125 aa cytoplasmic tail with largely unknown functions (Moh et al., 2005a; Favre-Kontula et al., 2008). It is possible that intracellular signaling effectors that bind to GlialCAM suppress inside-out integrin activation and dampen ECM adhesion in GBM cells. Alternatively, the extracellular IgG domains of GlialCAM may promote interactions with integrins to limit adhesion to the ECM. GSCs with low levels of GlialCAM show enhanced adhesion to fibronectin, which is a molecular marker for the mesenchymal GBM subtype (Kabir and Apu, 2022). The major cell surface receptor for fibronectin is α5β1 integrin (Hou et al., 2020). A prior report has shown that targeting β1 integrin increases the efficacy of antiangiogenic therapies in GBM (Carbonell et al., 2013) with more specific targeting of α5β1 integrin blocking invasive growth and activating apoptosis (Martinkova et al., 2010). We are currently determining whether the GlialCAM extracellular IgG domains or the intracellular signaling tail differentially regulate α5β1 integrin–ECM affinity and/or focal adhesion signaling functions. Caveolin-1 (Cav1) couples β1 integrin to the tyrosine kinase Fyn in lipid rafts (Wary et al., 1998) to activate the Ras–Erk pathway (Hynes, 2002). Cav1 can also activate α5β1 integrin in GBM cells, which stimulates invasive growth (Martin et al., 2009). It will be important to determine whether GlialCAM levels in GBM cells control the balance between α5β1 integrin signaling in lipid rafts to promote proliferation versus activation of α5β1 integrin in focal adhesions to promote invasion.
A prior report has coupled Mlc1 to EGFR activation in cultured astrocytes (Lanciotti et al., 2016) and in low-grade human gliomas harboring IDH1 R132H gene mutations (Leventoux et al., 2020). Hence, it is enticing to speculate the loss of GlialCAM expression in GBM cells causes an imbalance in RTK expression and/or signaling, possibly involving Mlc1. However, the RPPA analyses identify connections among GlialCAM, integrins, and focal adhesion components, which do not overlap with RPPA pathways regulated by Mlc1 (Lattier et al., 2020), suggesting that GlialCAM may function independently of Mlc1 in GSCs. In some GBM spheroid cultures, there are low endogenous levels of GlialCAM and Mlc1. It will be important to investigate aquaporin-4 expression levels and determine functional links to GlialCAM-Mlc1 in GSC self-renewal, migration, and ECM adhesion in vitro as well as in GBM initiation, growth, and invasion in vivo. Finally, we do not detect correlative expression between GlialCAM and Mlc1 RNAs, unlike the strong correlative expression between GlialCAM and aquaporin-4 RNAs in subsets of GBM cells. It will be important to further study these correlative gene expression results and establish mechanistic connections using the in vitro spheroid cultures and the xenograft mouse models of GBM.
A recent report has shown that genetic inhibition of GlialCAM/hepaCAM leads to aberrant astrocyte patterning in the mouse brain (Baldwin et al., 2021). Our data showing that reduced GlialCAM protein levels in GBM cells correlate with more invasive behaviors suggest that spatial positioning of GBM cells may also be determined by GlialCAM, as summarized in Figure 9. The relative functions for astrocytes in the GBM microenvironment remain mixed, with some data supporting tumor-promoting roles whereas other studies show tumor-suppressive roles (Krawczyk et al., 2022). It will be important to determine whether levels of GlialCAM in GBM cells control interactions with astrocytes and/or other glial cell types that impact tumor malignancy. Oligodendrocytes also express GlialCAM, and one route for GBM cell invasion in the brain is along oligodendroglia-rich white matter tracts. Hence, GlialCAM may facilitate adhesion between GBM cells and myelinating oligodendrocytes to promote invasion along white matter.
In most healthy adult tissues and organs, GlialCAM is expressed at very low levels, with loss of expression often correlating with cell transformation. Indeed, studies of epithelial cancers of the prostate, lung, and colon have shown that GlialCAM has tumor suppressor-like functions, with diminished GlialCAM levels leading to increased cell proliferation and metastasis (Moh et al., 2005b; Shao et al., 2016; Geng et al., 2017; Deng et al., 2019). These data are somewhat different from our studies of GBM, indicating that GlialCAM functions are influenced by cellular context, oncogenic gene mutations, and possibly other factors in the tumor microenvironment. It will be important to determine the mechanisms that control GlialCAM expression in different GBM cell populations and understand how these events are regulated during tumor progression and recurrence. A prior report in U373 human glioma cells has shown that forced expression of GlialCAM suppresses growth and induces differentiation (Lee et al., 2009). Unlike GSCs, we have found that adherent GBM cell lines such as U373 and U87 express very low or undetectable levels of endogenous GlialCAM as determined by analysis of mRNA levels in the NCBI-GEO database and immunoblotting GBM cell lysates (A. De, J. M. Lattier, and J. H. McCarty, unpublished observations). These results call into question the pathophysiological relevance for forced overexpression of GlialCAM and its impact on highly passaged GBM cell lines.
Genetic deletion of Aqp4 in mouse astrocytes leads to reduced migration (Saadoun et al., 2005) because of major cytoskeletal alterations (Nicchia et al., 2005). In astrocytes, aquaporin-4 is a component of the DGC, where it regulates water transport at the neurovascular unit (Waite et al., 2012). GlialCAM interacts with components of the DGC in astrocytes and patients with MLC disease develop brain edema and astrocyte end feet swelling indicative of hyperactive water transport (López-Hernández et al., 2011). Hence, it is enticing to speculate that GlialCAM loss-of-function leads to increased aquaporin-4 expression and enhanced water transport in astrocytes. Based on our data, we also propose that reactive astrocytes respond to brain pathologies by downregulating GlialCAM and activating aquaporin-4. AQP4 is a marker for GSCs (Xie et al., 2022) and is linked to self-renewal and invasion (Simone et al., 2019). Hence, the increased AQP4 expression in GSCs expressing low levels of GlialCAM may suggest a less differentiated cellular state. Aquaporin regulation of water influx at the leading edge of the cell, where focal adhesions are assembled and disassembled, is critical for cytoskeletal dynamics and shape changes during polarization and migration (Loitto et al., 2002). Indeed, intracellular trafficking of aquaporin-2 in kidney cells is critical for water homeostasis and is dependent on interactions with components of the actin cytoskeleton (Holst et al., 2021). Our data for pharmacologic inhibition of aquaporin-4 water transport supports functional connections between GlialCAM, aquaporin-4, and focal adhesion signaling proteins in invasive GBM cells.
Last, therapeutic agents that target angiogenesis, such as bevacizumab, have failed to yield significant improvements in overall patient survival (Lauko et al., 2022). In many patients with GBM, bevacizumab treatment unexpectedly leads to a pathologic burst in invasion after tumor recurrence (Keunen et al., 2011). We propose that downregulation of GlialCAM may be one mechanism used by GBM cells to inhibit cell–cell adhesion, exit the primary tumor mass, and disperse throughout the brain in response to antiangiogenic therapies. Bevacizumab is used in the clinic mainly to treat edema associated with hyperpermeable blood vessels in patients with GBM. We speculate that the GlialCAM and aquaporin-4 complex may be linked to VEGF signaling in the GBM microenvironment. Hence, targeting components of the GlialCAM and aquaporin-4 complex may prove to be an effective strategy for anti-invasive therapy.
Footnotes
The time-lapse imaging experiments were funded by Cancer Prevention & Research Institute of Texas (CPRIT) Core Facility Support Grant RP170628 and were conducted in the Advanced Microscopy Core laboratory. The following National Cancer Institute-funded Cancer Center Support Grant Core Facilities were also instrumental in data acquisition: the shRNA and ORFeome Core, the Research Histopathology Facility, the Flow Cytometry and Cellular Imaging Facility, and the Sequencing and Microarray Facility. We thank Dr. Sabbir Khan and Dr. Sanjay Singh (MD Anderson Cancer Center) for technical assistance and advice with experiments.
The authors declare no competing interests.
- Correspondence should be addressed to Joseph H. McCarty at jhmccarty{at}mdanderson.org
