Glioblastoma multiforme (GBM) is a late-stage astrocytoma brain tumour that is considered the most aggressive form of brain cancer. GBM tumours account for 12-15% of brain tumours diagnosed in North America (1). The incidence of GBM is 4 per 100,000 people and more commonly occurs in adult males than females (2). Only 4-7% of patients diagnosed with GBM will live more than 5-years (1, 3, 4). Despite decades of research, few advancements have been made to improve survival outcomes for GBM patients, which is partially attributed to a lack of appropriate model systems that accurately recapitulate the complex brain environment.

Brain tissue is a unique environment that contains high levels of astrocytes, proteoglycans/glycoproteins and hyaluronic acid (HA), while consisting of low levels of fibrous proteins (laminin, fibronectin, collagens). Brain tissue is softer than other organs, with the majority of fibrous proteins restricted to the basement membrane surrounding the vasculature (5, 6). Typical in vitro models of cancer cell invasion use a cocktail of extracellular matrix (ECM) proteins such as laminin and collagen (i.e. Matrigel, gelatin, etc.) that lack other brain specific components and thus do not represent a physiologically relevant substrate to model brain tumour cell invasion. Another unique feature of GBM invasion is that GBM cells seldom intravasate into the vasculature. As early as 1938, researcher Hans Joachum Scherer found glioma cells migrate along brain structures later termed “Scherer’s structures” (7). It is now well understood that brain tumour cells preferentially grow and invade along pre-existing structures including myelinated axons, blood vessels, and white matter tracts in the parenchyma, but can invade through the fluid-filled perivascular space surrounding the vasculature as well (8–11).

The organotypic brain slice invasion assay was first developed by Ohnishi and colleagues in 1998 and has been seminal for studying mechanisms of glioblastoma invasion using a biologically relevant scaffold (12). Several groups have since published a version of the ex vivo brain slice model using various methods for implanting tumour cells and imaging cellular motility. For instance, some groups use the technique of seeding tumour spheroids on top of brain sections (13–15), implanting tumour spheroids within the slice using a blunt-edge needle (16), or simply seeding tumour cells on top of the brain slice in a small droplet of medium (17). Other groups use alternative approaches such as cutting the brain slices in half and seeding tumour cells onto the membrane to create a “cell field” between the two hemispheres (18) or mixing tumour cells with ECM (i.e. Matrigel) and seeding the mixture into a cell spacer directly adjacent to the edge of the brain slice (19). Imaging strategies are also varied, ranging from taking a single image of the brain slice-spheroid interface to taking z-stack images through the brain slice on a confocal microscope. Importantly, many of these previous reports do not distinguish between migration along the top of the slice and invasion of cells into the slice.

Herein, we describe a universal technique for implanting tumour spheroids onto ex vivo brain slices without the use of Matrigel or other exogenous matrices that are not present in the brain environment. We also outline an improved imaging strategy for visualizing and quantifying cellular invasion into the brain slice. Our protocol involves embedding brain slices containing tumour spheroids into agar blocks that are re-sectioned onto slides to allow for the visualization of invasive projections below the tumour spheroids that would otherwise go undetected. The refinement of physiologically relevant models of GBM invasion will help to advance the development and testing of therapeutic strategies to reduce GBM cell motility and improve patient outcomes.

Modeling GBM invasion into ex vivo brain slices in the Z-direction can be technically challenging, particularly when attempting to distinguish between cellular migration along the top of the slice and tumour cell invasion into the slice. Thus, we sought to improve deep tissue imaging strategies such that visualizing cellular invasion into the brain slice is possible. We first mounted a whole 300 µm brain slice containing a GFP-expressing GBM spheroid onto a microscope slide, secured it in place with a coverslip over PBS, and then used the confocal microscope Z-stack function to capture 20 µm images in 10 focal planes ( Figure 3A ). The Z-stack images were used to create a single composite image of the brain slice in the X/Y direction. The ImageJ re-slice function was then used to generate a side view of the image stack across the center of the spheroid ( Figure 3A , yellow dotted line), which represents invasion into the brain slice in the Z-direction. Despite maximizing the imaging depth of the confocal microscope, this method of imaging resulted in low resolution images that are insufficient for visualizing any projections that may occur below the spheroid or for accurately determining the depth of the projections. We note that PBS was used to mount the coverslip and images were captured using an air lens rather than an immersion lens. However, we wanted to make a direct comparison between optical z-stacks captured using these parameters and the method outlined below.

To improve deep tissue visualization, we embedded brain slices containing GFP-expressing tumour spheroids into agar ( Figure 1C ), which allowed us to generate 200 µm sections perpendicular to the brain slice using the vibratome ( Figure 3B ). Each brain slice generates approximately 30 sections. The 200 µm sections were collected in order from top to bottom of the brain slice, flipped 90 degrees and rotated clockwise, and then placed flat onto a microscope slide. The confocal microscope was then used to image invasion into the brain slice in the Z-direction without the requirement for mounting medium or multiple Z-stack images ( Figure 3B ). We suggest implanting tumour spheroids on at least 3 different brain slices collected from the same brain to produce technical replicates. Averaging the area of invasive edge from these technical replicates will produce a single value for that murine brain. Biological replicates should be produced using 3 or more murine brains collected and processed on different experimental days. Overall, this imaging technique greatly improved the visualization of tumour cells invading deep into the brain slice, particularly cells below the spheroid that would otherwise go undetected using traditional imaging techniques.

We performed a series of experiments to test the viability of ex vivo brain slices under several conditions. Brain slices were harvested from 6-week-old, 8-week-old, and 22-week-old male C57BL/6J mice and then slice viability was analyzed using an alamarBlue™ metabolic assay to determine the role of mouse age on slice viability. We found slices harvested from 6-week-old mice were significantly more viable at 0 days in vitro (DIV) compared to the older mice ( Figure 4A ). Notably, 0 DIV refers to brain slices that were incubated at 37°C and 5% CO₂ for 2-4 hours until all slices were harvested, and then incubated in 10% alamarBlue™ containing medium for another 24 hours. While we limited the mouse age to a minimum of 6-weeks old to ensure the brains are large enough to collect at least 6 slices per brain, using younger mice may further increase viability as well. To compare various brain slice culture media, brain slices harvested from 6-week-old male C57BL/6J mice were cultured in serum-free DMEM-F12 medium (SFM), SFM supplemented with 1X B27 (a serum-free neuronal cell supplement), DMEM-F12 medium containing 5% or 25% FBS, and finally DMEM-F12 medium containing 5% or 25% heat-inactivated horse serum (HI-HS) for 4 days in vitro (DIV). This analysis revealed that slices maintained in 25% FBS or 25% HI-HS were significantly more viable than slices in serum-free DMEM-F12 medium or serum-free medium containing 1X B27 ( Figure 4B ). We also observed a lack of brain slice integrity and swelling that occurred when cultured in medium lacking FBS or HI-HS (data not shown). Next, we harvested brain slices from 6-week-old male C57BL/6J mice and cultured the slices in 25% FBS DMEM-F12 medium for up to 14 DIV. Brain slice viability was analyzed over time, which revealed that brain slices maintain greater than 50% viability by 4 DIV, as compared to 0 DIV brain slices ( Figure 4C ), and then viability drops substantially by 7 and 14 DIV. Brain slices fixed in 4% PFA were used as a positive control to determine maximum cell death measurable using alamarBlue™. Finally, each brain slice was collected in sequential order from 8-week-old mice to compare the viability of brain slices harvested from different regions of the brain. This revealed no difference in slice viability when comparing the first slice (Slice 1) to the last slice collected (Slice 18), which represent the slices closest to the occipital lobe and cerebellum, respectively ( Figure 4D ).

To demonstrate the use of our ex vivo brain slice invasion assay, we sequentially implanted GFP-expressing LN229 and LN18 spheroids onto murine brain slices between 0 and 8 days prior to imaging. We chose to sequentially implant all spheroids onto the same slice over time to reduce slice-to-slice variability ( Figure 5A ). The time course analysis revealed that approximately 25% of the volume of each LN229-GFP and LN18-GFP spheroid are within the brain slice on day 0 (i.e. 2 hours post-implantation), while nearly 100% of the volume of each spheroid was within the brain slice by days 3 and 5, respectively ( Figures 5B, C ). The ImageJ macro ( Figure 2 ) was used to quantify the invasive area, which revealed that LN229-GFP spheroids exhibited an increase in the total invasive area, while the invasiveness of LN18-GFP spheroids plateaued by 3 days in vitro (DIV) ( Figure 5D ). Furthermore, the average invasive distance, which was determined by measuring the distance of each particle (i.e area defined by ImageJ macro as greater than 10 pixels² in size) to the nearest edge of the ellipse, was found to increase up to 1 DIV and then plateaued from 1 DIV to 8 DIV ( Figure 5E ). These data show that our ex vivo brain slice invasion assay is sensitive enough to quantify differences in GBM invasion over time between cell lines.

To directly compare in vitro and ex vivo models of GBM invasion, we determined the invasiveness of GBM cells when invading into Matrigel versus ex vivo brain slices ( Figure 6 ). First, LN229-GFP and LN18-GFP spheroids were either embedded in Matrigel or seeded on top of a thin layer of Matrigel. This revealed that both LN229 and LN18 cells invade into the Matrigel or migrate along the top of the Matrigel collectively as sheets of cells ( Figures 6A, B ). Next, LN229-GFP and LN18-GFP spheroids were seeded on top of 300 µm thick organotypic brain slices and cultured ex vivo. Imaging the spheroids in the X/Y direction (described in Figure 3A ) revealed that GBM cells tend to migrate away from the edge of the spheroids collectively as both single cells and spindle-like structures along the top of the brain slice ( Figures 6A, B ). Similarly, imaging in the Z-direction (as described in Figure 3B ) shows a combination of single cells and spindle-like strands invading into the brain slice ( Figures 6A, B ). These data indicate GBM invasion into Matrigel in vitro does not accurately model the invasive behaviour of GBM cells into the brain, suggesting ex vivo brain slices are more suitable for studying mechanisms of GBM cell invasion.

Our version of the ex vivo brain slice invasion assay can also capture GBM tumour cells interacting with the surrounding brain environment. We found that GFP-tagged GBM cells invading away from the edge of the tumour spheroid can interact with αSMA-positive vasculature ( Figure 7A ) and GFAP-positive astrocytes ( Figure 7B ). These interactions would not be visible by imaging brain slices in the Z-direction using top-down confocal methods, and our system therefore provides the opportunity to more readily study glioma invasion along brain-specific structures.

For decades, GBM cell invasion has typically been modeled in vitro using chemotactic assays and synthetic basement membranes. These traditional assays lack environmental influences and may not accurately model the precise mechanisms of GBM cell invasion. Thus, there is a great need to transition from overly simplistic in vitro models to utilizing more physiologically relevant model systems to study GBM invasion, which will undoubtedly improve the development of more effective therapeutics for GBM patients. Here, we outline how to perform an ex vivo brain slice assay to more quantitatively model brain cancer invasion. Our version of the ex vivo brain slice invasion assay greatly improves upon the imaging techniques used to visualize invasion into the brain slice. This technique involves allowing GBM cells to invade into brain slices from spheroids, staining the brain slices with GFAP or other markers, embedding the stained slices in agar, and then re-sectioning the brain slices onto slides to allow visualization of the spheroid-brain slice interface in the Z-direction.

Our method provides an improved workflow for imaging tumour cell invasion depth using a confocal microscope, although we did not compare our methodology to imaging brain slices with a two-photon microscope. Two-photon microscopy is an alternative to confocal microscopy that offers increased resolution when imaging thick tissue sections, and can image tissue at depths that are unachievable using a confocal microscope (20). However, two-photon microscopes are highly specialized and expensive machines that are inaccessible to many labs. Alternatively, optical imaging depth can be improved by chemically clearing fixed tissue using a variety of treatment methods. Tissue clearing methods result in transparent tissue and increased imaging depth since light from the confocal microscope can penetrate into much thicker samples (21–23). Major disadvantages of tissue clearing include the use of harsh organic solvents, which may result in loss of lipids over time and tissue shrinkage, as well as fluorescent quenching (24, 25). Furthermore, tissue clearing processing can take weeks to complete. Our method offers a faster and more accessible way to image brain slices while maintaining staining quality and tissue architecture, as well as allowing quantification of tumour cell invasion.

Organotypic models of invasion offer many advantages over in vitro and in vivo assays. Seeding spheroids onto brain slices ex vivo requires far less technical expertise than performing intracranial implants to establish brain tumours in mice. Furthermore, less mice are required to perform ex vivo versus in vivo experiments highlighting ex vivo models as more cost-effective and ethical options (26). Lastly, organotypic brain slice cultures offer a 3D scaffold that contains tissue-specific ECM and normal cell populations that are often not present within in vitro models. Notably, others have shown that organotypic slices collected from fresh pancreatic ductal adenocarcinoma tumours could be cultured ex vivo for one week while maintaining immune cell populations of macrophages and T cells (27). While there are many benefits to using organotypic models of invasion, limitations still exist. For instance, while brain slices maintain vascular architecture ex vivo, the vessels do not maintain blood flow thus limiting the possibility of immune cell infiltration and tumour cell extravasation. Other limitations of ex vivo slice cultures include viability-dependent time constraints as well as the need to purchase specialized equipment (i.e., vibratome). Despite these limitations, ex vivo brain slice invasion assays offer many advantages over traditional two-dimensional chemotactic invasion assays and represent an important step forward for GBM research.

Glioblastoma tumours are highly invasive and typically invade within 1-2 cm from the primary tumour margins while rarely metastasizing to distant organs (8, 28). Notably, GBM cells can invade into brain tissue as single cells and/or in collective strands along pre-existing brain structures such as myelinated axons, blood vessels, or white matter tracts (7, 9, 11, 29–31). To determine the best method for modelling these distinct mechanisms of GBM cell invasion, we compared the invasive patterns of GBM cells when using an in vitro spheroid invasion assay and an ex vivo brain slice invasion assay. Interestingly, we found that LN229 and LN18 GFP-expressing GBM spheroids typically undergo collective invasion with cells invading in a diffuse, sheet-like pattern when either seeded on top of a thin layer of Matrigel or embedded within the Matrigel in vitro. Conversely, when the same GBM spheroids were implanted on top of murine brain slices ex vivo, we observed GBM cells migrating along the top of the slice and invading into the brain slice as both single cells and in strands of cells. These data therefore suggest in vitro models of invasion are lacking critical components necessary to facilitate collective invasion and that modeling GBM invasion using ex vivo brain slices better mimics the modes of invasion observed in human GBM tumours.

A major advantage of the organotypic brain slice invasion assay is the ability to implant cell line-derived or patient-derived tumour spheroids anywhere on the brain slice with a high level of precision (i.e. spheroids can be implanted only along the corpus collosum, only within white matter, etc.). The invasion assay presented herein models tumour cell invasion within the cerebral cortex, although this model could theoretically be used to evaluate invasion within the occipital lobe or cerebellum as well. However, we do not recommend using this method to model meningeal invasion since there is limited meningeal surface area present in the coronal brain slice sections. Also, we have found spheroids placed along the edge of the brain can detach and/or produce tumour cell migration along the outside of the brain slice and onto the membrane beneath the brain slice. Nevertheless, the ex vivo brain slice invasion assay also offers a high-throughput and time-effective option for measuring the effects of genetic manipulation (i.e., siRNA, shRNA, CRISPR/cas9, etc.) or therapeutic intervention (i.e., chemotherapy, immunotherapy, or radiation studies) on GBM cell invasion. For instance, tumour spheroids can either be pre-treated or exposed to various therapeutics once implanted onto the brain slice to analyze the effects of treatment on cancer cell survival, proliferation, and motility. Lastly, this model also provides the ability to study tumour cell-brain cell interactions, such as tumour cells interacting with astrocytes and/or nearby vasculature.

Taken together, our data highlight the importance of transitioning from in vitro models to organ-specific ex vivo models of invasion to better recapitulate features of the innate tumour microenvironment and surrounding tissue environment. This will result in an improved understanding of the precise mechanisms that drive GBM cell invasion and lead to the advancement of therapeutic options available to brain cancer patients.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The animal study was reviewed and approved by University of British Columbia (UBC) Animal Care Committee in accordance with Canadian Council on Animal Care Guidelines.

LD, CP, and KB conceptualized the project. LD designed and performed the experiments, analyzed and interpreted the data presented, and wrote the final manuscript. RS conceptualized and wrote the ImageJ macro code for analyzing brain slice invasion. KT assisted with fixing brain slices and MH helped generate tumour spheroids for implantation. CP and KB secured funding and edited the final manuscript. All authors contributed to the article and approved the submitted version.