Glutamate drives glioblastoma invasion in three-dimensional hyaluronic acid hydrogels

Abstract

Glioblastoma (GBM) tumors are characterized by an excess of extracellular glutamate, one important source of which is the tumor cells themselves. This abundance of glutamate promotes GBM proliferation, migration, and therapeutic resistance, and causes excitotoxicity in nearby neurons. However, despite glutamate’s clear role in promoting GBM aggression, the exact mechanisms through which excess glutamate drives these phenotypes, particularly 3D invasion, remain incompletely understood. To address this gap, we used a 3D brain-mimetic hyaluronic acid (HA) hydrogel to investigate the role of glutamate signaling in GBM 3D invasion. We demonstrate that inhibiting the glutamate N-methyl-D-aspartate receptor (NMDAR) reduces invasion from 3D tumorspheres, a result that is reproducible across multiple continuous culture models and a patient-derived xenograft cell line. We then conducted glutamate-driven invasion studies in 3D HA-based devices that can be micro-dissected to isolate and differentially analyze invasive and non-invasive cells. Transcriptomic analysis of invasive, non-invasive, and drug-treated populations of cells reveals that NMDAR inhibition suppresses several pathways associated with the mechanobiology of invasion, including matrix remodeling and collagen deposition. Correspondingly, supplementation with exogenous collagen VI partially rescued 3D invasion. Our work speaks to the potential value of biomaterial platforms for identifying the autocrine and paracrine mechanisms through which neurotransmitters fuel GBM invasion.

Introduction

Glioblastoma (GBM) is the most aggressive primary brain cancer, characterized by extensive infiltration of tumor cells into healthy brain tissue.¹ Current treatment involves surgical resection followed by radiation and chemotherapy; however, recurrence is nearly universal, and less than 5% of patients survive 5 years after diagnosis.¹ A better understanding of factors within the GBM microenvironment that drive invasion would improve the field’s understanding of GBM pathophysiology and potentially uncover novel biomarkers and therapeutic targets.

Neurotransmitters have emerged as important signaling molecules in the GBM tumor microenvironment.^2–6 In particular, in GBM there is an abundance of extracellular glutamate, the primary excitatory neurotransmitter in the central nervous system.⁷ Recent work has identified that neurons and glioma cells can form bona fide synapses which result in post synaptic currents in glioma cells that drive growth and invasion.^5,6 Outside of neuron-glioma synapses, other stromal cells such as astrocytes can contribute to glutamate production; however, a disproportionate share of the extracellular glutamate in GBM is produced by the tumor cells themselves.⁸ Indeed, glutamate concentrations in cerebrospinal fluid in glioma patients have been reported to range up to 400 μM, exceeding normal values by two orders of magnitude.⁸ Several factors result in this overproduction of glutamate by tumor cells, including overexpression of the glutamate/cystine antiporter SLC7A11/xCT, suppression and/or mutation of glutamate uptake machinery (e.g. EAAT transporters), and increased glutamine metabolism of which glutamate is a byproduct.^8,9

The high concentration of glutamate in the GBM microenvironment can promote disease progression in a variety of ways. For example, glutamate can promote tumor growth by causing excitotoxicity in nearby neurons.¹⁰ This results in neuronal cell death which creates additional space for the tumor to grow.^7,11 In addition, glutamate signaling to the cancer cells themselves can drive proliferation, migration, and therapeutic resistance.¹² Glutamate signals through two categories of receptors: ionotropic receptors, which are ligand-gated ion channels and metabotropic receptors, which are G protein-coupled receptors.⁸ Ionotropic glutamate receptors are particularly recognized for their role in driving disease progression and can be further classified into the N-methyl-D-aspartate receptor (NMDAR), alpha-amino-3hydroxy-5-methly-4-isoxazolepropionic acid receptor (AMPAR), and kainite receptors.⁸ Overexpression of calcium permeable AMPARs can facilitate tumor cell migration, invasion, and proliferation.^13–15 NMDAR activation can enhance resistance to chemotherapy and radiation, regulate redox homeostasis, and promote tumor cell proliferation.^16–19 NMDAR signaling has similarly been tied to migration of tumor cells.^19,20

Much of our understanding of how glutamate drives GBM invasion derives from either two-dimensional (2D) culture systems or highly simplified three-dimensional (3D) paradigms such as Boyden chamber assays.^19,20 While these models have offered valuable insights into GBM cell behavior, they do not capture the 3D nature of the GBM microenvironment. This is important because our laboratory and others have shown that mechanisms of cell migration differ radically in 3D environments due to the presence of steric constraints, the need for matrix remodeling, and the lack of dorsal-ventral polarization cues.^21–23 It therefore remains unclear how glutamate signaling through NMDAR affects invasion in 3D. Hyaluronic acid (HA) is a natural platform with which to study GBM invasion, because it is the most abundant extracellular matrix (ECM) component in brain tissue.²⁴ Mice lacking the HA receptor CD44 exhibit impaired glutamatergic signaling, raising the intriguing yet unexplored possibility that engagement of HA regulates glutamate-dependent neurotransmission.²⁵

In this study, we combine engineered 3D HA hydrogel invasion studies, transcriptomics, and bioinformatics analysis to investigate the role of glutamate in driving 3D invasion of GBM cells. We begin by showing that NMDAR inhibition reduces invasion in 3D tumorspheres. To gain mechanistic insight into this result, we deploy GBM cells in an HA-based invasion device that enables dissection and analysis of invasive and non-invasive fractions. Transcriptomic analysis of these fractions revealed that NMDAR inhibition results in downregulation of ECM factors, including several collagen subtypes. Correspondingly, treatment with recombinant collagen VI, which we previously showed to promote GBM invasion, can rescue the effect of NMDAR inhibition.²⁶ Overall, our study highlights the application of biomaterials platforms to investigate neurotransmitter signaling in GBM invasion.

Materials and Methods

Cell culture.

U251 MG and U87 MG human GBM cells were obtained from the University of California, Berkeley Tissue Culture Facility. These cells were cultured in Dulbecco’s modified Eagle medium (DMEM; Gibco) supplemented with 10% (v/v) fetal bovine serum (FBS; Thermo Fisher Scientific), 1% (v/v) penicillin/streptomycin (Gibco), 1% (v/v) MEM non-essential amino acids (Gibco), and 1% (v/v) sodium pyruvate (Gibco). GBM-43 cells²⁷ were received from Dr. Manish Aghi at UCSF and cultured in DMEM (Gibco) supplemented with 10% (v/v) fetal bovine serum (FBS; Thermo Fisher Scientific), 1% (v/v) penicillin/streptomycin (Gibco), and 1% (v/v) Glutamax (Thermo Fisher Scientific, 35-050-061). Cells were harvested using 0.25% trypsin-EDTA (Thermo Fisher Scientific) and passaged <30 times.

HA hydrogel synthesis and crosslinking.

Methacrylated HA (HA-Me) was synthesized as previously described.^28,29 The extent of methacrylation per disaccharide was quantified by ¹H nuclear magnetic resonance spectroscopy (NMR) and found to be ~90–100% for materials used in this study. To functionalize the HA-Me with integrin-binding motifs, cysteine-containing RGD peptide Ac-GCGYGRGDSPG-NH2 (Anaspec, AS-62349) was conjugated to HA-Me via Michael Addition at a concentration of 0.5 mmol/L. To form hydrogels, 6 wt% HA-RGD was crosslinked with protease-cleavable peptide (KKCG-GPQGIWGQ-GCKK, Genscript) in phenol red-free serum-free DMEM (Thermo Fisher Sci, 21-063-029) at 37 °C for 1 h. A final concentration of 4.43 × 10⁻³ M peptide crosslinker was used to crosslink the gels, resulting in a final 1.5 wt% HA gel to yield a shear modulus of ~300 Pa.

Hydrogel rheological characterization.

The shear modulus of the hydrogel was measured using an oscillatory rheometer as previously described (Anton Parr Physica MCR 310).³⁰

3D tumorsphere invasion assays.

Tumorspheres were generated as previously described by seeding ~200,000 cells per well in AggreWell plates to generate spheroids of ~100 μm in size (STEMCELL Technologies, 34415).³¹ Tumorspheres were harvested from AggreWell plates, resuspended in non-phenol red DMEM, and mixed with HA-RGD and peptide crosslinker to form hydrogel solution as described above. The gel solution was plated as 8.5 μL droplets (typically containing 1–10 spheres/droplet) in non-TC treated 48 well culture plates and incubated at 37 °C for 1 h. During crosslinking, the plate was flipped upside down every ~3–5 min for the first 30 min to ensure the tumorspheres did not settle to the bottom of the gel. Spheres were imaged with a 10x phase contrast objective on a Nikon Eclipse Ti2 microscope at day 0 and after invasion had occurred (2–7 days). Invasion area was measured using ImageJ and quantified as (A_f - A_i)/A_i where A_f = final spheroid area and A_i = initial spheroid area. Invasion was normalized to the control spheres. For statistical analysis spheroids were pooled across multiple independent replicates, where n is the number of spheroids. In general, trends and statistical significance held across drug treatments when statistics were performed within a given replicate.

Glutamate Assay.

The Glutamate-Glo^™ assay (Promega, J7021) was performed according to the manufacturer’s recommendations. Briefly, 10,000 cells were seeded in 10 μL HA hydrogels and treated with 400 μL of cell culture media. Every 24 h, 5 μL of conditioned media was removed, diluted in 95 μL of PBS, and stored at −20 °C. Control samples were taken directly from cell culture maintenance media stocks and diluted in PBS. To run the assay, 50 μL of diluted conditioned media in PBS was added to a white bottom 96 well plate. A standard curve of glutamate dilutions was prepared in PBS. 50 μL of reaction solution, prepared according to the manufacturer’s instructions, was added to each well. The plate was incubated for 1 h at room temperature before reading the luminescence on a plate reader (Tecan).

Cell viability assay.

Cells were plated at a density of 15,000 cells per well in 48-well plates and allowed to adhere overnight. Media was then replaced with dilutions of (+)-MK-801 hydrogen maleate (Millipore Sigma) or memantine hydrochloride (Millipore Sigma; M9292). For control wells media was supplemented with equivalent volume of the vehicle as used for the highest dosage (DMSO for MK-801 and water for memantine). After 48 h, the wells were washed with PBS to remove dead cells. The cells were then stained with calcein AM (ex/em 488 nm/15 nm) in PBS for 30 min. Cells were then imaged on a Nikon Eclipse Ti2 epifluorescence microscope, selecting two random areas per well. Using ImageJ, the number of cells was quantified, and relative cell viability was calculated by dividing the number of calcein-positive cells for each drug dilution by the average total number of calcein-positive cells in control wells. Inhibitors were subsequently deployed at the maximum dosage for which relative viability remained >95%.

Scratch wound assays.

15,000 cells were plated in 12 well plates and left to grow till confluency (2 to 3 days). At confluency, a scratch was generated by dragging a 1 mL pipette tip in a straight line across the well. Wells were washed with PBS to remove lifted cells before adding normal cell culture media or drug treatment. The scratch wounds were imaged on 10x Phase at 0 and 24 h. Scratch area was traced using ImageJ. The percentage of scratch closure was quantified as ((A_i – A_f)/A_i)*100 where A_f = final scratch area and A_i = initial scratch area.

Collagen gel invasion assays.

2 mg/mL collagen gels were generated as previously described.³¹ The gel solution was plated as 10 μL droplets in non-TC treated cell culture plates and incubated at 37 °C for 1 h before adding cell culture medium and drug treatment. Tumorspheres were imaged and invasion area calculated as described above for HA gels.

Invasion device fabrication and seeding.

Invasion devices were fabricated as previously described.²⁷ Devices were cultured for 35 days with media replenished every 4 days. MK-801 or control DMSO was spiked in the media every 2 days. Devices were imaged every 7 days.

Device dissection and RNA extraction.

To extract RNA, the tops of the devices were first carefully removed using a razor blade. For control devices, invasive cells were physically separated from the non-invasive “core” cells using a scalpel. Each fraction was transferred into 1.5 mL Eppendorf tubes. For MK-801 treated devices, the invasive fractions were not large enough to isolate separately from the core, so both fractions were pooled for these samples. Equivalent volume of 10k U/mL Hyaluronidase from bovine testes, Type IV-S (Sigma-Aldrich, H3884) was added to the volume of extracted cells + hydrogel. The gels were incubated on the shaker at 37 °C and 500 rpm for ~ 30 min until the hydrogel was fully degraded. The tubes were then spun at 130 × g for 3 min to pellet the cells. The supernatant was removed, and cells were washed with PBS before re-pelleting. Next, the RNA was extracted from the cell pellets using TRIzol Reagent (Invitrogen, 15596018) according to the manufacturer’s recommendations and as previously described.³²

RNA-sequencing and differential gene expression analysis.

RNA samples were sent to Novogene Corporation Inc. (Sacramento, CA) for library construction, quality control and sequencing as previously described.³² Libraries were sequenced using Illumina platforms. DEGs were identified using DESeq2 (https://github.com/thelovelab/DESeq2) and visualized in Rstudio (version 2024.04.2+764).

Pathway enrichment analysis.

Enrichr was used to perform enrichment analysis of gene lists using the Reactome pathway database.^33–36 Pathways were ranked by p adjusted value and plotted in GraphPad Prism.

Bioinformatic analysis of publicly available sequencing datasets.

We generated a list of “glutamate-related” genes by combining glutamate receptor and transporter genes with the KEGG glutamate metabolism pathway genes.³⁷ We then queried the IVYGAP database (https://glioblastoma.alleninstitute.org) for these genes.³⁸ The average z score for each tumor region was graphed in GraphPad Prism. Sequencing data from site-directed biopsies of human tumors was accessed via the NCBI’s Gene Expression Omnibus database (GEO GSE246697).²⁷ The GEO2R tool was used to download the top 250 DEGs comparing GBM patient tumor tissue taken from the core and edge. These DEGs were filtered for glutamate related genes and Log2FC was plotted in GraphPad Prism. Clustered single-cell RNA sequencing data from GBM patients was accessed via (http://gbmseq.org/).³⁹ Average normalized log2 counts per million were calculated for NMDAR receptor genes for core and peripheral neoplastic cells and plotted in GraphPad Prism.

Statistics.

Statistical analyses were performed in GraphPad Prism unless otherwise stated. For comparison between 2 groups Student’s t test or Wilcoxon’s test was performed. For comparison between 3 or more groups 2-tailed 1-way ANOVA and post hoc tests were performed. A p value of less than 0.05 was considered significant.

Data Availability.

The raw and processed bulk RNAseq data of MK801 and control invasion devices are deposited in the NCBI GEO Database (GSE308404).

Results

Glutamate related gene signatures are upregulated in the leading edge of human glioma

We first began by probing the IVY Glioblastoma Atlas Project database to determine if genes related to glutamate signaling are upregulated in invasive regions of the tumor.³⁸ We created a list of glutamate related genes by combining KEGG glutamate metabolism pathway genes with glutamate receptor and transporter genes.³⁷ We observed increased expression of glutamate receptor genes in the leading edge and infiltrating tumor regions, with particularly notable enhancement of NMDAR receptor genes (GRIN1, GRIN2A-C, GRIN3A) and several metabotropic receptors (GRM1–5, GRM7) (Fig. 1A). Additionally, several glutamate transporter genes (SLC1A6, SLC17A7, and SLC17A6) and glutaminase genes (GLS and GLS2) were enriched in the leading edge (Supplementary Fig. 1). To further investigate if glutamate related gene signatures are upregulated in the invasive front, we analyzed our previously published sequencing dataset from site-directed biopsies of human tumors.²⁷ GRIN1, which encodes for the essential subunit of NMDAR, emerged as one of the top differentially expressed genes between the edge and core samples (Fig. 1B). Because these prior two datasets were based on bulk sequencing data, we wondered if this increased expression of glutamate receptor genes was perhaps due to neurons and astrocytes present in the total sample. We therefore probed a clustered single-cell RNA sequencing dataset for expression of NMDAR genes in the tumor cells specifically.³⁹ Comparing neoplastic cells in the periphery versus tumor core, we observed that GRIN2A and GRIN2C are increased in the periphery (Fig. 1C). In summary, invasive GBM tumor cells upregulate glutamate receptors, especially NMDAR.

Figure 1.

Transcriptomic analysis of GBM tumors shows increased expression of glutamate genes at tumor edge. (A) IVYGAP analysis of glutamate receptor genes by tumor region.³⁸ (B) Expression of glutamate-related genes in core vs edge of patient site-directed tumor biopsies.²⁷ (C) Analysis of scRNAseq of neoplastic cells in the periphery vs core tumor for NMDAR gene expression.³⁹

Extracellular glutamate drives GBM invasion through NMDAR

Given that our bioinformatics analysis indicated increased expression of glutamate receptor genes in invasive regions of the tumor, we asked if we could drive invasion of tumorspheres in 3D HA hydrogels by supplementing cell culture media with additional glutamate. We performed tumor sphere invasion assays in HA gels of ~300 Pa (Supplementary Fig. 2A–B) with a continuous GBM cell line (U251) and a patient derived xenograft line (GBM43). Exogenous glutamate did not increase invasion in either cell type compared to the control (Fig. 2A–D). Given previous literature indicating that glioma cells secrete high levels of glutamate^7,8 and the bioinformatic analysis confirming increased expression of glutamate synthesizing enzymes and transporter genes at the leading edge (Supplementary Fig. 1, Fig. 1B), we hypothesized that the lack of response was due to already-saturating concentrations of glutamate in the culture media. We therefore measured glutamate production by single cells encapsulated in 3D gels in multiple GBM cell lines. GBM cells steadily secreted glutamate into the media over time, confirming their capacity for autocrine signaling. By 72 and 96 hours, glutamate concentrations in cell-conditioned media were significantly higher than in basal cell culture media, which already contains 200–300 uM glutamate (Fig. 2E–F; Supplementary Fig. 2C). Thus, the lack of exogenous glutamate-induced invasion likely results from pre-existing NMDAR saturation.

Figure 2.

Extracellular glutamate signaling through NMDAR drives invasion in HA hydrogels. (A) Quantification and (B) representative phase images of U251 invasion assay with glutamate supplementation (n = 23–51 spheres). (C) Quantification and (D) representative phase images of GBM43 invasion assay with glutamate supplementation (n = 24–34 spheres). (E) Measurement of glutamate concentration in cell conditioned media (CM) over time in U251 and (F) GBM43 cells (n = 6). (G,I) Quantification and (H,J) representative phase images of U251 invasion assay with MK801 (200 μM) and memantine (20 μM) treatment (n = 18–75 spheres). (K,M) Quantification and (L,N) representative phase images of GBM43 invasion assay with MK801 (50 μM) and memantine (20 μM) treatment (n = 75–121 spheres). Spheroid invasion results were pooled across at least 2–3 independent replicates. *P <0.05; ***P < 0.001; ****P < 0.0001.

Given the high preexisting levels of glutamate in the medium, we next asked if blocking NMDAR with small molecule inhibitors (MK-801 and memantine) could reduce tumorsphere invasion. Appropriate dosages of inhibitor were selected by performing cell viability assays to determine the highest dosage that would not reduce relative cell viability below 95% (Supplementary Fig. 2D–I). Treatment with MK801 and memantine both reduced U251 tumorsphere invasion by 64% and 42% respectively (Fig. 2G–J). We observed similar effects with both U87 and GBM43 tumorspheres with 43% and 40% reduction in invasion in U87 and 15% and 48% reduction in invasion in GBM43 tumorspheres for MK801 and memantine, respectively, confirming that the result is generalizable across culture models, including patient-proximal models (Fig. 2K–N; Supplementary Fig. 2J–K).

NMDAR inhibition slows invasion in 3D but not 2D

As mentioned earlier, 3D invasion differs from 2D migration because of the presence of steric constraints, the need for matrix remodeling, and the lack of dorsal-ventral polarization cues.^21–23 To determine if NMDAR-driven invasion depends on matrix dimensionality, we performed 2D scratch wound assays in the presence or absence of MK801 or memantine. Treatment with MK801 and memantine did not reduce scratch closure in U251, GBM43, or U87 cells (Fig. 3A–D, Supplementary Fig. 3A–B), indicating the 3D-specific nature of the effect. To investigate the specificity of this result to matrix type, we repeated tumorsphere invasion assays in collagen gels, where MK801 and memantine reduced tumorsphere invasion respectively by 58% and 40% in U251 cells and 39% and 36% in GBM43 cells (Fig. 3E–H). Interestingly, NMDAR inhibition of U87 cells in collagen gels did not decrease tumorsphere invasion; however, the interpretation of this result is complicated by the fact that U87 cells rapidly and extensively invade collagen within 4 hours, making it challenging to capture differences associated with the inhibitors (Supplementary Fig. 3C–D).

Figure 3.

NMDAR driven invasion is a 3D-specific effect. (A) Quantification and (B) representative phase images of U251 scratch assay with NMDAR inhibition (200 μM MK801, 20 μM memantine, n = 6). (C) Quantification and (D) representative phase images of GBM43 scratch assay with NMDAR inhibition (50 μM MK801, 20 μM memantine, n = 6). (E) Quantification and (F) representative phase images of U251 invasion assay in collagen gels with NMDAR inhibitors (200 μM MK801, 20 μM memantine, n = 34–47 spheres). (G) Quantification and (H) representative phase images of GBM43 invasion assay in collagen gels with NMDAR inhibitors (50 μM MK801, 20 μM memantine, n = 12–19 spheres). Spheroid invasion results were pooled across at least 2–3 independent replicates. ***P < 0.001; ****P < 0.0001.

NMDAR inhibition with MK801 reduces invasive growth in 3D HA invasion devices

To investigate the mechanistic basis through which NMDAR inhibition reduces 3D GBM invasion, we made use of our previously described microchannel invasion platform (Fig. 4A).²⁷ This device consists of a core channel of cells embedded within a 3D hydrogel. Over time cells invade out of the core channel, enabling physical separation of the invasive and non-invasive (core) cell fractions, which then may be subjected to differential analysis. We seeded GBM43 cells in these invasion devices in the presence or absence of MK801 and monitored invasion over time. By calculating changes in area of the invasive and core fractions, we determined that NMDAR inhibition with MK801 specifically reduces invasive growth (Fig. 4B–C) but does not alter the total or core growth (Supplementary Fig. 4A–B) compared to control devices, illustrating an effect on invasion versus tumor growth more generally. These results both recapitulate in HA invasion devices the phenomenology we had observed with tumorspheres and more clearly show that NMDAR inhibition reduces invasion, motivating us to move forward with analysis of invasive and non-invasive fractions.

Figure 4.

NMDAR inhibition decreases invasive growth in 3D HA invasion devices. (A) Schematic of 3D invasion device. (B) Invasive growth quantification and (C) representative phase images of GBM43 invasion devices with MK801 (50 μM, n = 6). *P <0.05; ***P < 0.001; ****P < 0.0001. Core regions are indicated as the space between light blue dashed lines while invasive regions are between light blue and purple dashed lines.

NMDAR inhibition causes transcriptional changes in GBM cells and downregulation of ECM remodeling pathways

We then isolated the core and invasive fractions of cells from the control devices and extracted RNA for bulk RNA sequencing (Fig. 5A). The MK801-treated devices did not have an invasive fraction large enough to extract separately, so the core and invasive fractions were pooled together, as we have done in a previous study.³² We identified over 2000 differentially expressed genes (DEGs) between the pooled MK801 group and the control core. We also identified over 1000 DEGs between the pooled MK801 group and the control invasive fraction (Supplementary Fig. 5A–C). Principal component analysis revealed distinctions between all three groups among the first principal component (PC1). Core versus invasive controls appear to separate along the first principal component (PC1) with substantial variation in PC1 across devices in the MK801 pooled samples (Supplementary Fig. 5D). Except for the core fraction from one control device, drug versus no treatment segregated samples across PC2.

Figure 5.

GBM43 transcriptional changes induced by MK801 treatment. (A) Schematic of sample extraction from invasion devices for bulk RNAseq. Transcriptomic analysis of 3D invasion devices showing volcano plots (left) and Reactome pathway analysis (right) for comparison of (B) invasive fraction versus core fraction of control devices, (C) entire pooled channel of MK801 devices versus core fraction of control devices, and (D) entire pooled channel of MK801 devices versus invasive fraction of control devices. * P_adj <0.05 ** P_adj < 0.01 *** P_adj < 0.001 ****P_adj < 0.0001.

To visualize changes in gene expression, we created volcano plots, applying a threshold of |Log2FC| > 1 and pvalue <0.05 (Fig. 5B–D). To identify pathways associated with the decrease in invasive phenotype in response to MK801 treatment, we performed Reactome pathway enrichment analysis with the DEGs (Fig. 5B–D). When comparing the control core samples to MK801 pooled samples, MK801 pooled samples showed downregulation of several cell-cycle pathways (Fig. 5C). Comparing the control invasive fractions to MK801 pooled samples, there was upregulation of apoptosis-related pathways and downregulation of several ECM-related mechanobiological pathways including Extracellular Matrix Organization, Integrin Cell Surface Interactions, Collagen Formation, Assembly of Collagen Fibrils, and Collagen Chain Trimerization (Fig. 5D).

We and others have previously shown the importance of ECM secretion and remodeling to GBM invasion in 3D.^26,40 Given that we observed an effect of NMDAR inhibition in invasion in 3D but not 2D, we specifically focused on differential gene expression within the Extracellular Matrix Organization pathway. When we compared the Log2FC changes in gene expression between the MK801-treated samples and control invasive fractions for genes in the Extracellular Matrix Organization pathway, we saw downregulation of several ECM proteins including fibronectin and a variety of collagens (Fig. 6A), including multiple collagen VI subunits. We previously showed that GBM cells secrete collagen VI to prime the 3D HA matrix for invasion.²⁶ We therefore asked if we could rescue the invasive phenotype of NMDAR-inhibited cells by treating with recombinant collagen VI. Collagen VI treatment partially rescued invasion in MK801-treated cells (Fig. 6B–C), indicating that NMDAR inhibition may suppress invasion by interfering with matrix secretion events key to 3D invasion.

Figure 6.

Supplementation with recombinant collagen VI can rescue NMDAR inhibition. (A) Heatmap of ECM organization pathway genes downregulated in the MK801 (50 μM) treated group compared to control invasion fraction. (B) Quantification and (C) representative phase images of GBM43 invasion assay with COL6 supplementation and MK801 treatment (n = 10–26 spheres). **P <0.01; ****P < 0.0001.

Discussion

By exploiting engineered biomaterials platforms, we have demonstrated that NMDAR inhibition suppresses 3D GBM invasion. This effect was specific to 3D matrices, including invasion-permissive collagen hydrogels, and most pronounced within HA hydrogels. Transcriptomic analysis within HA-based invasion devices revealed that NMDAR inhibition downregulates transcripts and pathways associated with the mechanobiology of invasion, including expression of collagen VI subunits, which we previously showed promotes 3D invasion.²⁶ Treatment with recombinant collagen VI partially rescued invasion in NMDAR inhibitor treated cells, suggesting that NMDAR inhibition suppresses invasion by impairing ECM secretion.

Our findings add to the growing body of literature surrounding glutamate signaling in cancer. NMDAR is emerging as a regulator of tumor growth, migration, and/or invasion in a variety of cancers including prostate, gastric, pancreatic, and breast cancer as well as melanoma and glioblastoma.⁴¹ In GBM, treatment with NMDAR inhibitors can sensitize GBM cells to chemotherapy and radiation and slow growth of tumors in situ.^16,17,42 NMDAR signaling has also been reported to drive migration in 2D and in transwell models.^17,20 Interestingly, we detected no effect of NMDAR inhibition in 2D scratch wound assays but did see robust decreases in invasion in 3D hydrogels. Further study should help determine whether these dimensionality-dependent differences derive from differences in tumor type, matrix platform, or other factors.

Our transcriptomics analysis indicated that NMDAR inhibition downregulates cell cycle pathways compared to control core samples. This finding is aligned with previous literature that implicates NMDAR signaling in regulating cell cycle both in healthy neurons and in diseased tissue.⁴³ We also observed downregulation of ECM and ECM remodeling proteins when comparing the control invasive and pooled MK801-treated samples. Previous studies have also demonstrated that NMDAR signaling modulates ECM deposition and remodeling. Matrix metalloproteinase expression is increased upon NMDA and glycine stimulation in U251 cells.¹⁹ Outside of GBM, knockout of GRIN1 in megakaryocytes resulted in decreased expression of collagens, fibronectin, matrix metalloproteinases, and lysyl oxidase.⁴⁴ NMDAR inhibition also reduced collagen deposition associated with glomerulosclerosis in rat kidneys.⁴⁵ Precisely how NMDAR inhibition interferes with matrix remodeling, however, remains an important open question. Investigation of these mechanisms would benefit from cell culture models that allow for direct stimulation of the NMDAR rather than relying on inhibition as we did in this study due to the presence of saturating levels of glutamate in the cell culture media.

NMDAR inhibition is beginning to be explored as a therapeutic strategy for limiting brain tumor progression. A phase 1 trial⁴⁶ determined that temozolomide chemotherapy in combination with mefloquine, memantine, or metformin was well tolerated. For combination therapy with memantine, the most common dose limiting toxicity was dizziness. This study was not, however, powered to determine treatment efficacy. Another clinical trial⁴⁷ involving patients with brain metastases determined that treatment with memantine after whole brain radiation had neuroprotective effects, reducing decline in memory, executive function, and processing time. Other clinical studies have tried to disrupt glutamate signaling in other ways by targeting the glutamate-cysteine antiporter with sulfasalazine or the AMPAR with talampanel.^48,49 Currently, a phase Ib/II trial is ongoing comparing chemoradiotherapy alone to chemoradiotherapy in combination with gabapentin, sulfasalazine, and memantine which respectively target glutamate synthesis, secretion, and signaling.⁵⁰ The impact of these inhibitors on patient survival remains unclear.

Future work should continue to dissect the underlying mechanisms of how NMDAR activation can drive tumor invasion in 3D. More broadly, biomaterials platforms can be applied to investigate glutamate signaling through other receptors in addition to NMDAR. While we chose to focus on NMDAR, other literature indicates important roles of AMPAR and metabotropic glutamate receptors in promoting GBM progression.^{13–15,51,52} Some research indicates that there may be crosstalk between glutamate receptors with NMDAR stimulation leading to increased expression of AMPAR.²⁰ It would be interesting to investigate if NMDAR inhibition could impact invasion by regulating activity of additional glutamate receptors. Further insight into the mechanisms of glutamate signaling in the cancer microenvironment can better inform how to target these processes to prevent growth and spread; and biomaterials are a promising platform to dissect these interactions.

Conclusion

While it is known glutamate can drive cancer progression, how extracellular glutamate signaling through the NMDAR affects invasion in 3D is understudied. Here we combine biomaterials platforms with bioinformatics analysis to investigate the role of NMDAR signaling in GBM invasion. We show that inhibition of the NMDAR robustly decreases invasion in tumorspheres and this effect is not dependent on matrix type. Our transcriptomics analysis reveals that NMDAR inhibition decreases ECM remodeling and collagen production pathways. This work speaks to the utility of engineered 3D culture platforms in dissecting the role of neurotransmitter signaling in glioblastoma, and cancer more broadly.

Impact Statement.

Glutamate is an abundant signaling molecule in the glioblastoma (GBM) microenvironment and important driver of disease progression; however, the mechanisms through which glutamate promotes invasion in 3D remain poorly understood. By combining engineered hyaluronic acid (HA) hydrogel platforms and transcriptomic analysis, we determine that blocking glutamate signaling through the NMDA receptor suppresses invasion in 3D by interfering with ECM remodeling processes. This work highlights the utility of biomaterials platforms in dissecting neurotransmitter signaling in GBM invasion.