Expression of Glial Cell Line–derived Neurotrophic Factor and Its Receptors in Glioblastoma

Abstract

Glioblastoma is the most frequent and aggressive primary brain cancer in adults, and the prognosis is poor. The neurotrophic factor glial cell–derived neurotrophic factor (GDNF) and its receptors, which are involved in neuronal development, have in experimental studies been suggested to drive tumorigenic processes in glioblastoma, but the role and expression in glioblastoma in patients is under-investigated. The aim of this study was to investigate the expression of GDNF, GDNF family receptor 1–4 (GFRA1–4), and the downstream REarranged during Transfection (RET) receptor in human glioblastoma tissue by RNA in situ hybridization, immunohistochemistry, and immunofluorescence. Expression was quantified by software-based classifiers. The results showed that GDNF was expressed in approximately 10% of tumor cells. The GFRA1 receptor was widely expressed in tumor cells, often colocalizing with the astrocytic tumor cell marker glial fibrillary acidic protein (GFAP), and in a smaller fraction of tumor cells expressing the stem cell markers oligodendrocyte transcription factor 2 (OLIG2) and SRY-Box Transcription Factor 2 (SOX2). The GFRA2 receptor expression was very limited, whereas expression of GFRA3, GFRA4, and RET, respectively, was almost absent. In conclusion, GDNF and its primary receptor GFRA1 were expressed in patient glioblastoma tissue. Potential clinical value needs further investigation.

Introduction

Glioblastoma is the most aggressive primary brain tumor and one of the most lethal cancer types overall. Despite multimodal treatment consisting of maximal surgical resection, radiotherapy, and chemotherapy, median overall survival remains around 12–15 months.^1–5 Thus, novel therapeutic targets are strongly needed for glioblastoma. In a number of preclinical glioblastoma studies, glial cell–derived neurotrophic factor (GDNF) has been suggested to stimulate glioblastoma proliferation, migration, and angiogenesis, and it has been suggested to be involved in resistance to chemotherapy.^6–16 However, these findings have not yet been translated into novel therapeutic strategies in patients. Since the expression and cellular localization of GDNF and in its receptors in glioblastoma is relatively unknown, the aim of this study therefore was to investigate the expression of GDNF, the GDNF family receptors alpha 1–4 (GFRA1–4), and the REarranged during Transfection (RET) receptor in human glioblastoma. In the central nervous system, GDNF ligands and receptors have been found primarily to be expressed in specific populations of neurons in substantia nigra, striatum, hippocampus, cerebellum, and cortex. ¹⁷ Functionally they have been demonstrated to play an important role in neuronal development and maintenance.^17,18 The GDNF family of ligands consists of GDNF, neurturin, artemin, and persephin. The ligands bind to receptors belonging to the GFRA family, and each ligand has specific affinity for each of the receptors. Thus, GDNF binds to GFRA1, neurturin binds to GFRA2, artemin binds to GFRA3, and persephin binds to GFRA4. Some extent of crosstalk among the ligands and receptors has been demonstrated in laboratory experiments,^19–24 among which GDNF has been demonstrated to bind to GFRA2 and GFRA3, and neurturin and artemin to GFRA1.^25–27 The vast majority of studies done on the GDNF ligands and receptors are either in vitro or in vivo studies in mice and rats focusing on cells in the nervous system. ¹⁷ Different types of GDNF ligand-receptor downstream signaling have been demonstrated in experimental studies. The primary downstream receptor is the RET receptor, which has been shown to increase neurite survival, neurite growth, and neuronal differentiation.^26,28 GDNF, GFRA1, and RET, respectively, have each been shown to be essential for murine renal and enteric nervous development, based on studies with knockout models.^29,30 Other experimental studies have pointed to GDNF and GFRA1 binding to neural cell adhesion molecule (NCAM), which was shown to increase migration in schwann cells and axonal growth in hippocampal and cortical neurons.^31,32

The knowledge about GDNF and GFRA1 expression and signaling in human glioblastoma tissue is very limited. One study has previously investigated the expression by immunohistochemistry on frozen glioblastoma tissue in up to 14 samples, where both proteins were to be expressed in cells that morphologically were interpreted as tumor cells. ³³ Two other studies showed presence of GDNF protein in human glioma bulk tissue by enzyme-linked immunosorbent assay (ELISA)^8,34 and immunofluorescence. ⁸

In this study, we examined the expression of GDNF, GDNF family receptors GFRA1–4 and RET in human glioblastoma tissue, including specific expression in the different cell types: tumor cells, tumor stem–like cells, microglia-macrophages, and endothelial cells. These cell types were chosen due to their importance in glioblastoma progression. Glioblastoma stem–like tumor cells have been shown to be associated with resistance to chemotherapy and radiotherapy, and aggressive tumor growth.^35–38 Microglia-macrophages are abundant in glioblastoma tumors making up a large part of the cell population in the tumor microenvironment (30% or even more). The interactions of these cells with tumor cells and other cells in the tumor microenvironment play an important role in tumor progression as previously reviewed by our group ³⁹ and other groups.^40–43 Glioblastomas are highly vascularized and endothelial cells interact with tumor cells and cells in the tumor microenvironment, and assist in promoting tumor growth.^40,41

For our detailed investigation, we used in situ hybridization, immunohistochemistry, and double immunofluorescence protocols, and we quantified the expression levels by threshold-based software-based classifiers.^44–49

Methods

Patient Inclusion

Frozen glioblastoma samples were obtained from 27 consecutive patients diagnosed at Odense University Hospital between January 1, 2016 and December 31, 2017, and formalin-fixed paraffin-embedded (FFPE) glioblastoma samples were obtained from 10 consecutive patients diagnosed at Odense University Hospital between January 1, 2018 and December 31, 2018. Surgery was performed at the Department of Neurosurgery, Odense University Hospital followed by diagnostic work at the Department of Pathology, Odense University Hospital.

Immunohistochemistry on Frozen Glioblastoma Tissue

Fresh tumor tissue from glioblastoma patients was stored at −80C upon arrival to the Department of Pathology, Odense University Hospital, less than 30 min after removal from the patient. The frozen tissue blocks were cut into 5-µm sections and sections from the patients were stored at −80C. Before the staining procedure, the sections were thawed and fixated in 4% neutral-buffered formalin for 10 min. An antibody directed against GDNF (Table 1) was used, and the DAB-Optiview detection system (Roche Diagnostics, Rotkreuz, Switzerland) was used for antibody detection. Hematoxylin was used for nuclei counterstaining. The Ventana Discovery Ultra Platform (Roche Diagnostics) was used to perform the staining. Frozen human cerebellum tissue samples were used as antibody control tissue with Purkinje cells serving as a positive control and neurons in the cerebellar granular layer serving as a negative control (Appendix Fig. A1).

Table 1.

List of Antibodies Used, Protocols and Tissues Used for Positive and Negative Controls, Respectively.

Antibody Target	Catalog Number	Company	HIER Pretreatment	Dilution	Positive Control	Negative Control
GDNF	ab18956	Abcam	None	1:200	Purkinje cells, neurons in cerebellar medulla	Cells in cerebellar granular layer
GFRA1	sc-271546	Santa Cruz Biotechnology	CC1-buffer 48 min, 100C for single stain, CC1-buffer 32 min, 100C for double stain (48 min vs 32 min yielded identical results in control stainings)	1:800	Renal tubuli, adrenal gland tissue, ganglion plexes in colon muscular layer	Liver tissue, glandular cells in colon
GFRA2	HPA024704	Atlas Antibodies	CC1-buffer 32 min, 100C	1:50	Ganglion plexes in colon muscular layer, some cells in Hodgkin lymphoma, adrenal gland tissue, germinal centers in tonsil tissue	Muscular layer in colon, esophagus squamous epithelial (ESE) cells
GFRA3	ab8028	Abcam	CC2-buffer 32 min, 91C	1:200	Ganglion plexes in colon muscular layer, some cells in ventricular mucosa and duodenal mucosa, islet cells in pancreas, thyroid carcinoma	Colon smooth muscle tissue, ESE cells
GFRA4	ab183077	Abcam	CC1-buffer 32 min, 100C	1:200	Islet cells in pancreas, thyroid carcinoma	Liver tissue, ESE cells
RET	ab 214791	Abcam	CC1-buffer 48 min, 100C	1:4000 +amplification	Ganglion plexes in colon muscular layer, thyroid carcinoma	Colon smooth muscle tissue, squamous epithelial cells in portio uteri
GFAP	Z033429-2	Dako	CC1-buffer 32 min, 100C	1:4000	Astrocytic tumor cells, astrocytes, ganglion plexes in colon muscular layer	Epithelial cells in colonic mucosa, endothelial cells in glioblastoma
IBA1	019-19741	Wako Pure Chemical Corporation	CC1-buffer 32 min, 100C	1:2000	Microglia and macrophages in glioblastoma	Glioblastoma tumor cells
OLIG2	18953	Immuno-Biological Laboratories	CC1-buffer 32 min, 100C	1:200	Some astrocytic tumor cells in glioblastoma	ESE cells
SOX2	MAB2018	R&D Systems	CC1-buffer 32 min, 100C	1:200	Some astrocytic tumor cells in glioblastoma	ESE cells
CD34	AC-0082	Cell Marque	CC1-buffer 32 min, 100C	1:50	Endothelial cells in vessels in glioblastoma	ESE cells, astrocytic tumor cells in glioblastoma
NCAM	156R-96	Cell Marque	CC1-buffer 32 min, 100C	1:250	Astrocytic tumor cells in glioblastoma	Endothelial cells in glioblastoma

Immunohistochemistry on FFPE Glioblastoma Tissue

Tumor tissue from glioblastoma patients was formalin-fixed in 4% neutral-buffered formalin for approximately 24 hr. The tissue was then paraffin-embedded and cut in 3-µm-thick sections that were subjected to heat-induced epitope retrieval (HIER) pretreatment with CC1 (Roche Diagnostics) or CC2 buffer (Roche Diagnostics) with times and temperatures differing for each antibody used, based on protocol optimization (Table 1). For the RET antibody, detection amplification using the OptiView Amplification Kit (Roche Diagnostics; cat. no. 760-099) was performed. The DAB-Optiview detection system and the Ventana Discovery Ultra Platform were used for antibody detection and staining procedures, respectively, and hematoxylin was used for nuclear counterstaining. Different types of control tissue were used for the different antibodies (Table 1, Appendix Figs A2–4. Negative control tissue is included for the GFRA1 antibody). Different GDNF antibodies and protocols were tested on FFPE tissue, without obtaining valid staining results (Appendix Table A1).

Immunofluorescence Double Staining on FFPE Glioblastoma Tissue

Sections of tumor tissue were prepared as described above. First, one primary antibody was added to the tissue, and then the Discovery Cy5 kit and Discovery OmniMap anti-rabbit/anti-mouse horseradish peroxidase (HRP) (Roche Diagnostics) were added for antibody detection. Hereafter, a second round of HIER treatment was done, and the second primary antibody was added. Then the Discovery FAM kit and Discovery OmniMap anti-rabbit/anti-mouse HRP were added for detection of the second primary antibody. Vectashield mounting medium including DAPI (Vector Laboratories, Newark, CA) was applied afterward, counterstaining the nuclei. Please see Table 1 for the entire list of antibodies used.

RNAscope Assays on FFPE Glioblastoma Tissue

Sections of FFPE tumor tissue were treated according to the manufacturer’s assay protocol (RNAscope 2.5 HD detection reagents; ACDBio, Newark, CA; cat. no. 322310) and an additional amplification step (TSA-Plus DIG; PerkinElmer, Waltham, MA; cat. no. NEL748001KT) was added to the protocol. Briefly, sections were heated to 60C, treated with pre-treatment reagent, heated to 40C and then cooled to room temperature. Hereafter, slides were deparaffinized in xylene, hydrated in ethanol and dried. A PAP-pen was used to mark the tissue section. Sections were then again treated with pre-treatment reagent, washed, treated with protease reagent, washed, and then the GDNF probe (RNAscope Probe-Hs-GDNF; ACDBio; cat. no. 437871) was added to the sections for hybridization overnight. Hereafter, six rounds of amplification and washing steps where performed, and the additional TSA amplification was added before liquid permanent red was added for detection of signals. All steps were performed manually. Positive controls were neurons in the periphery of glioblastoma tissue and islet cells in endocrine pancreatic tissue, and liver tissue served as a negative control (Appendix Fig. A1).

Double Staining with RNAscope Probe and Antibodies

The RNAscope assay procedure was performed as described above. Afterward, the tissue sections were incubated with primary antibody followed by secondary binding with the EnVision kit (DAKO; cat. no. K4003; Glostrup, Denmark) and detection with Teal (Roche Diagnostics; cat. no. 760-247) or DAB (Agilent, Santa Clara, CA; cat. no. K3468). Thus, messenger RNA (mRNA) became visible as purplish red dots and protein as cyan or brown staining.

Software-based Cell Classification and Quantification

Slides stained by a chromogenic reaction were scanned and digitalized using the NanoZoomer 2.0 HT digital image scanner (Hamamatsu, Hamamatsu City, Japan). The digitalized slides were imported into the Visiopharm software V2018.9.4 (Visiopharm, Hoersholm, Denmark) and classifiers, different for each marker, for measuring area fractions or count fractions in tumor tissues were created via a threshold-based approach (example in Appendix Fig. A5A).

Slides stained by immunofluorescence were put on the tray of a Leica DM6000B microscope equipped with an Olympus DP72 camera and a Ludl motorized stage connected with the Visiopharm software. Brightfield images were acquired at ×1.25 magnification and regions of interests (ROIs) encompassing relevant tissue were drawn manually for each slide. For each ROI, 15 systematic randomly selected images were obtained at ×20 magnification. Training of a classifier and colocalization measurements in Visiopharm was done using a threshold-based approach. Colocalization was determined as area [for glial fibrillary acidic protein (GFAP)] or count (for IBA1, OLIG2, and SOX2) of each specific marker colocalizing with GFRA1 stain divided by total area or count of marker of interest (example in Appendix Fig. A5B).

Statistical Analysis

Differences in expression among the receptors GFRA1–4 and RET were investigated with one-way analysis of variance (ANOVA) with Bonferroni multiple comparisons test. Differences in expression in the GFRA1 double stainings were investigated with the Kruskal-Wallis test with Dunn’s multiple comparisons test, as these data points collectively had a non-Gaussian distribution. Statistical analyses were performed with GraphPad Prism version 9.3.1 (GraphPad Software, Boston, MA).

Results

GDNF Staining

GDNF protein expression was investigated within frozen tissue sections since it was not possible to detect GDNF in FFPE tissue by immunohistochemistry despite testing six different antibodies. In total, GDNF was detected in 22 of the 27 stained tumors (Fig. 1) and it was found to be expressed in many different areas of the tumor tissue, including areas with high cellularity, pseudopalisading cells surrounding necroses, and tumor cells around microvascular proliferations (Fig. 1A–C). Occasionally, weak to moderate staining was also seen in endothelial cells in vascular structures. However, tissue quality and background staining varied between tumors, preventing measuring of the amount of staining with a software-based classifier. Likewise, reliable double immunofluorescence staining for GDNF and other markers could not be established. Collectively, we assessed manually that the mean area fraction of GDNF expression was approximately 10%, with large variations in the stained sections.

Figure 1.

Expression of GDNF in glioblastoma tissue. Expression was seen in tumor cells (A), including tumor cells surrounding necrotic areas (B) and perivascular tumor cells, and occasionally in endothelial cells (C). GDNF mRNA signal expression was likewise seen in tumor cells (D), tumor cells surrounding necrotic areas (E) in perivascular tumor cells, and in some cases in endothelial cells (F). The presence of GDNF mRNA in tumor cells was confirmed with double staining for GNDF mRNA and the astrocytic tumor marker GFAP (G), and co-expression was also seen in perinecrotic tumor cells (H) and perivascular tumor cells (I). Double staining for GDNF mRNA and the microglia/macrophage marker IBA1 did mostly not show co-expression, but occasionally GDNF mRNA was seen in microglia/macrophages in viable tumor tissue (J), close to necrotic areas (K) and close to vascular structures (L). Scale bars represent 50 µm in main images and 25 µm in inserts. Asterisks mark necrotic areas, and double asterisks mark vascular structures. Arrowheads mark cells with co-expression.

To ensure more precise detection, we also assessed GDNF mRNA expression in FFPE tumor tissue sections from 10 glioblastoma patients by in situ hybridization using the RNAscope assay. GDNF mRNA signal expression was widely present in the assessed glioblastoma tissue sections, most frequently in cells which morphologically appeared to be tumor cells. In some slides, signals were present in pseudopalisading tumor cells and in microvascular proliferations, both occasionally in endothelial cells and in perivascular tumor cells (Fig. 1D–F). Digital quantification showed a mean of 1670 mRNA signals per square millimeter tissue, ranging from 375 to 4598 (Fig. 2A).

Figure 2.

Quantitative analyses by threshold-based software classifiers were performed for GDNF mRNA signals, for receptor stainings and for double stainings for GFRA1 and each of the five different markers GFAP, IBA1, OLIG2, SOX2, and CD34. Count fractions of GDNF mRNA signals (A). Area fractions of receptor stainings were calculated as positive stained tissue area versus total stained tissue area, and GFRA1 area fractions were significantly higher than area fractions for the other receptors (p<0.001) (B). Cell colocalization fractions of GFRA1 and different markers. The co-expression fraction for GFRA1 and GFAP was significantly higher than for the other double stainings (p<0.001), and GFRA1-OLIG2 co-expression and GFRA1-SOX2 co-expression, respectively, were each significantly higher than GFRA1-IBA1 co-expression and GFRA1-CD34 co-expression, respectively (p<0.001) (C). In all of the graphs, mean value and standard deviations are shown.

For more precise characterization of the cells expressing GDNF mRNA, double staining was performed for GDNF mRNA and the astrocytic tumor cell marker GFAP, and the microglia-macrophage marker, ionized calcium-binding adapter molecule 1 (IBA1), respectively. Due to the varying sizes of the GDNF mRNA signals and thus varying distances to membranous and cytoplasmic antibody staining, it was not possible to make accurate software-based classifiers for measuring colocalization in cells for GDNF mRNA-GFAP protein, and GDNF mRNA-IBA1 protein, respectively. Nonetheless, upon double staining with GDNF mRNA-GFAP protein, it was visually confirmed that many of the GDNF signals were located in tumor cells with positive GFAP staining (Fig. 1G–I). Double staining of GDNF mRNA-IBA1 protein revealed a very limited number of cells with co-expression (Fig. 1J–L).

Receptor Staining

We performed staining for receptors in the same 10 glioblastoma tissue sections used for GDNF mRNA staining. We observed pronounced GFRA1 membranous and cytoplasmic immunoreaction (Fig. 3A–F) in major tumor areas. Strong antibody reaction was seen in pleomorphic tumor cells (Fig. 3A–C), including in gemistocytic tumor cells (Fig. 3D), pseudopalisading cells surrounding necrotic areas (Fig. 3E), and in cells surrounding vessels and microvascular proliferations (Fig. 3F). Likewise, staining was observed in cells with neuronal morphology (not shown). The mean area fraction of the staining was 30.7%, ranging from 7.2% to 56.2%. GFRA2 staining was overall negative (Fig. 3G–I), but a few populations of gemistocytic tumor cells had a positive membrane reaction (Fig. 3G), and apart from that, rare positive cytoplasmic reaction in cells was seen in areas with high cellularity (Fig. 3H). Positive reaction in gemistocytic cells surrounding microvascular proliferations was seen in one tumor (Fig. 3I). Neurons in the tumor periphery were positive (not shown). The mean area fraction of the GFRA2 staining was 3.4%, ranging from 0.3% to 7.5%. Staining for GFRA3 (Fig. 4A–C) demonstrated weak to moderate cytoplasmic reaction in very few cells (Fig. 4A), while the vast majority of the tumor tissues was negative (Fig. 4B and C). The same was observed for GFRA4 staining, with very few cells showing weak cytoplasmic reaction (Fig. 4D), but most tissue samples being negative (Fig. 4E and F). RET staining also showed very limited staining with positive cytoplasmic reaction in a very small number of cells (Fig. 4G and I), and most of the tissue being negative (Fig. 4H), although slightly more positive reaction was observed compared with staining for GFRA3 and GFRA4, respectively. Mean area fractions were 0.5% for GFRA3 (ranging from 0.01% to 2.4%), 0.1% for GFRA4 (ranging from 0.02% to 0.4%), and 1.0% for RET (ranging from 0.1% to 3.6%). Receptor staining quantifications are shown in Fig. 2B. One-way ANOVA analysis with Bonferroni multiple comparisons showed that the GFRA1 area fractions were significantly higher than GFRA2 (p<0.001), GFRA3 (p<0.001), GFRA4 (p<0.001), and RET (p<0.001), respectively, whereas the differences among the other area fractions were not significant.

Figure 3.

GFRA1 (A–F) and GFRA2 (G–I) staining in glioblastoma tissue. GFRA1 was expressed in pleomorphic tumor cells (A), giant tumor cells (B), proliferating tumor cells (C), gemistocytic tumor cells (D), pseudopalisading cells surrounding necroses (E), and tumor cells in a perivascular area (F). GFRA2 was expressed in gemistocytic tumor cells (G). No positive reaction was seen in pseudopalisading cells surrounding necroses (H) and positive reaction in tumor cells in perivascular areas was very limited (I). Scale bars represent 50 µm in main images and 25 µm in inserts. Asterisks mark necrotic areas and double asterisks mark vascular structures.

Figure 4.

GFRA3 (A–C), GFRA4 (D–F), and RET (G–I) staining of glioblastoma tissue. GFRA3 and GFRA4 expression was mostly absent, but very few cells expressed GFRA3 (A) or GFRA4 (D), respectively. RET staining were mostly negative, but very small numbers of positive cells were observed in some areas (G, I). The image examples include cellular tumor (A, D, G), pseudopalisading cells surrounding necroses (B, E, H), and perivascular areas (C, F, I). Scale bars represent 50 µm in main images and 25 µm in inserts. Asterisks mark necrotic areas and double asterisks mark vascular structures.

Due to NCAM being an alternative receptor for the GDNF-GFRA1 complex, we performed a chromogenic NCAM staining on a glioblastoma tissue microarray with eight tumor tissue cores and saw as expected strong reaction in all cores in almost all of the tumor tissue (Appendix Fig. A6).

Based on high expression of the GFRA1 receptor and low expression of the other receptors, we proceeded with double immunofluorescence staining for GFRA1 and different cellular markers (Fig. 5A–F). GFAP-GFRA1 reaction often overlapped partially or completely (Fig. 5A–B). Co-expression was observed in areas with both high and low cellularity, in tumor cells with different morphology including gemistocytic tumor cells, in pseudopalisading perinecrotic tumor cells and in perivascular tumor cells. Mean colocalization area fraction was 67.6% (ranging from 40.1% to 97.3%) (Fig. 2C). Double staining with stem cell markers showed small subsets of cells with co-expression of OLIG2-GFRA1 as well as SOX2-GFRA1 (Fig. 5C and D). Often, when co-expression was present, cells were clustered together, but occasional double-positive single tumor cells were also seen. Mean count fraction for double staining of OLIG2 and GFRA1 was 17.5% (ranging from 0.0% to 89.4%), whereas mean count fraction for double staining of SOX2 and GFRA1 was 11.0% (ranging from 0.0% to 68.6%) (Fig. 2C). Colocalization of IBA1-GFRA1 was rarely seen, but a few IBA1 positive microglia/macrophages did also express membranous GFRA1 (Fig. 5E). Mean IBA1-GFRA1 double staining count fraction was 0.5% (ranging from 0.0% to 4.0%) (Fig. 2C). Co-expression of the endothelial marker CD34 and GFRA1 was not observed. However, in some areas, strong GFRA1 expression was present in tumor cells surrounding blood vessels (Fig. 5F). Mean double staining area fraction was 0.0% (Fig. 2C). Using the Kruskal-Wallis test with Dunn’s multiple comparisons, the GFRA1 and GFAP double stainings showed significantly higher co-expression fractions than each of the other double stainings (p<0.001 for each of the comparisons). Also, GFRA1 and OLIG2 and GFRA1 and SOX2 co-expression fractions, respectively, were significantly higher than the co-expression fractions of GFRA1 and IBA1, and GFRA1 and CD34, respectively (p<0.001).

Figure 5.

Immunofluorescent double staining with GFRA1 and different markers in glioblastoma tissue. Staining for GFRA1 (green) and astrocytic tumor cell marker GFAP (red) (A, B), tumor stem cell marker OLIG2 (purple) (C), tumor stem cell marker SOX2 (purple) (D), microglia/macrophage marker IBA1 (red) (E), and endothelial cell marker CD34 (red) (F). Images with single channels are shown in the left and middle, and merged channels are shown to the right. Scale bars 50 µm in main images and 25 µm in inserts. Arrowheads mark cells with co-expression.

Discussion

In this study, we demonstrated expression of the neurotrophic factor GDNF and its main receptor GFRA1 in glioblastoma. The receptors GFRA2–4 and the downstream receptor RET were expressed at a very low level.

Positive GDNF immunoreaction was seen in most of the frozen tissue samples. Frozen tissue was chosen for investigating the GDNF expression, after extensive testing of multiple different GDNF antibodies had failed to show any positive reaction in different FFPE tissue samples including control tissue. A previous study also detected GDNF using immunohistochemistry in frozen human glioblastoma tissue. ³³ A total of 11 grade IV glioma tissue samples with unknown isocitrate dehydrogenase (IDH) mutation status were assessed for GDNF, and strong staining was found in six of these, and weak staining in the other five. The expression was proposed to be present in tumor cells, based on morphology. In another study, an immunofluorescent GDNF staining of one human glioblastoma tissue sample also shows GDNF protein expression. ⁸ Since GDNF is a neurotrophic factor that is secreted, we wanted to confirm our findings using in situ hybridization for GDNF mRNA. This also showed positive GDNF expression in tumor cells. Furthermore, for the first time, we confirmed the frequent presence of GDNF mRNA in cells positive for the tumor cell marker GFAP by double staining, thereby providing compelling evidence that GDNF is produced in tumor cells and secreted from these cells. In line with this, GDNF mRNA expression was demonstrated in patient glioma by tissue bulk analysis,^8,50 and experimentally GDNF was also detected on the mRNA level in rat C6 ⁵¹ and human LN-229 and A-172 glioblastoma cells ¹¹ and the protein level in human U251 glioblastoma cells ⁶ and murine GL261 glioblastoma cells. ⁸

We only rarely observed GDNF in cells with microglia-macrophage morphology. We used a double staining protocol and found that GDNF mRNA was only rarely expressed in IBA1-positive microglia-macrophages detected by immunohistochemistry. Both our group and other groups have used IBA1 as a reliable marker for microglia-macrophages.^52–54 This finding was surprising, as GDNF mRNA and protein have been detected in microglia and macrophages in previous experimental studies in rats and mice, and shown to influence the growth of tumor cells.^8,55–59 Based on our results, the translational value of these studies should be reconsidered and confirmed with cells of human origin.

Morphologically, we further observed some labeling of endothelial cells in vascular formations in staining for GDNF mRNA and protein. Previous experimental studies show conflicting results regarding GDNF expression in endothelial cells. GDNF was not detected by ELISA in one study assessing cultured human umbilical vein endothelial cells (HUVECs), ⁶⁰ whereas GDNF was detected by immunofluorescence in vessels in murine ischemic skeletal muscular tissue, ⁶¹ and via a cytokine array in cultured HUVEC, ⁶² respectively, by others. Also, several experimental studies detected GFRA1 protein in cultured HUVEC by immunocytochemistry, ⁶⁰ in vessels in rat cerebral cortical tissue by immunofluorescence, ⁶³ and in cultured human brain microvascular endothelial cells by Western blotting. ⁶⁴ For the first time, we investigated the expression of GFRA1 in cells positive for the endothelial marker CD34 in vessels in human glioblastoma tissue, and we did not see GFRA1 expression in the endothelial cells. However, in some cases, we did see GFRA1 expression in tumor cells adjacent to endothelial cells. The findings suggest that endothelial-derived GDNF mainly signal in a paracrine manner, by affecting GFRA1 on the tumor cell surface. This may also apply to the perivascular tumor stem cell niche.

We investigated the expression of GDNF family receptors and found pronounced expression of GFRA1, very limited expression of GFRA2, and barely any expression of GFRA3, GFRA4, and RET. To our knowledge, this is the first study to investigate the expression of GFRA2–4 and RET in human glioblastoma tissue. GFRA1 protein expression has been shown in human glioblastoma tissue in one previous study, by immunohistochemistry where quantification of expression was not performed. ³³ Furthermore, GFRA1 mRNA has been found by reverse transcriptase quantitative polymerase chain reaction (RT-qPCR) in human bulk glioblastoma tissue. ¹¹ Several studies have shown GFRA1 expression in cultured rat C6 glioblastoma cells^15,51,65 and murine GL261 glioblastoma cells as well as cultured human glioblastoma cells. ⁸ To our knowledge, this is the first time that GFRA1 expression has been quantified in glioblastoma tissue. As a novel aspect we demonstrated that a minor fraction of cells expressing GFRA1 co-expressed stem cell markers OLIG2 and SOX2, respectively. These findings suggest that GFRA1 may play a role in tumor stemness in glioblastoma, which is an important feature in tumor aggressiveness. Future studies focusing on the role of GFRA1 in glioblastoma tumor stemness are thus warranted to elucidate whether the GDNF-GFRA1 pathway could be a therapeutic target, as targeting the pathway may both eliminate tumor cells being part of the stem cell niche and the more differentiated tumor cells.

Expression of GFRA2 was low in the assessed glioblastoma tissue samples. One study found GFRA2 mRNA to be expressed at low levels in bulk glioblastoma tumor tissue, whereas higher GFRA2 mRNA expression was seen in control adult and fetal brain tissues, respectively, using RT-qPCR assays on glioblastoma bulk tumor tissue from a total of 13 patients. ¹¹ Similar mRNA analysis in rat glioblastoma C6 cells presented no GFRA2 expression. ⁵¹ Another study using Western blotting found very weak GFRA2 protein expression in murine GL261 glioblastoma cells and in cultured human patient-derived glioblastoma tumor cells. ⁸ Thus, in line with previous studies, we found very limited expression.

We detected RET expression in a very limited number of cells in the tumor tissue and in neurons, but RET expression was otherwise absent. This finding is interesting, as RET has been described as the primary receptor for the GDNF-GFRA1 complex in the healthy and inflammatory nervous system. The absence of RET in this study points toward RET-independent GDNF-GFRA1 signaling taking place in human glioblastoma tissue, a signaling cascade known to be of importance in the brain.^17,27 This may explain why the few clinical trials conducted in glioblastoma patients testing the effect of multi-target kinase inhibitors, which include inhibition of RET, have failed to show a survival benefit.^66,67 Results from one experimental study point toward NCAM being the primary co-receptor for the GDNF-GFRA1 complex in cultured rat C6 glioblastoma cells. ⁵¹ This study found that knockdown of GFRA1 or NCAM significantly reduced cell migration, whereas knockdown of RET did not, and further that protein expression of the protein RhoA was reduced upon GFRA1 or NCAM knockdown, but not upon RET knockdown. We demonstrated that NCAM was widely expressed in glioblastoma tissue. Although this has only been investigated functionally in the single study mentioned above, it is a possibility that the GDNF-GFRA1 complex mainly induces downstream signaling by the NCAM receptor.

In this study, we showed a pronounced expression of GDNF mRNA and GFRA1 protein, respectively, in tumor cells in human glioblastomas. A strength to point out in our study was the thorough use of control tissues to make sure that our staining protocols were valid. Moreover, we used an RNAscope GDNF assay to validate and further investigate GDNF expression identified by immunohistochemistry. Limitations include the limited number of tissue sections investigated and a lack of reporting of clinical data. Comparison of staining levels using a sizeable glioblastoma tissue cohort with clinical data is warranted to identify potential prognostic value of GDNF and GFRA1 expression. Another limitation is that we did not perform functional experiments in this study. Since functional studies exist but few expression data are available in human tissue, we focused on descriptive and quantitative studies of expression of GDNF and its receptors using software-based quantifications, rather than doing manual expression scoring. Our results nonetheless point toward the GDNF-GFRA1 interaction in glioblastoma mainly being based on an autocrine mechanism, where GDNF is secreted from tumor cells and binds to GFRA1 receptors on the same or other tumor cells. There may also be secretion from endothelial cells/vessels, which may support the perivascular niche with tumor stem-like cells earlier proposed to be critical in glioblastoma.^68–70 So far, to our knowledge, no clinical trials have been conducted for treatment targeting the GDNF-GFRA1 signaling pathway specifically in brain tumors or in any other types of cancer disease. Further preclinical and functional studies are warranted to investigate whether this justified.

Conclusion

We conclude that GDNF and GFRA1 are predominantly expressed in tumor cells within glioblastoma tissue. GFRA1 is consistently expressed in tumor cells regardless of their stem cell phenotype, whereas GFRA2 shows only sparse expression. The other GDNF receptors, including RET, exhibit very low or no expression. Future preclinical studies are needed to reveal the potential therapeutic value of targeting GDNF and GFRA1.

Appendix A

GDNF antibodies tested unsuccessfully in formalin-fixed paraffin-embedded (FFPE) tissue. Varying intensities of nonspecific reactions were seen, and valid, specific reactions could not be obtained using different staining protocols.

Table A1.

Catalog Number	Company	Concentrations Tested
AF-212-NA	R&D Systems	1:25, 1:50, 1:100, 1:400
ANT-014	Alomone Labs	1:100, 1:200, 1:400, 1:800
Ab223347	Abcam	1:50, 1:100, 1:200
Ab18956 (worked successfully in frozen tissue)	Abcam	1:100, 1:200, 1:400, 1:800

For all of the antibodies above, the different antigen retrieval protocols listed below were tested. The staining and detection procedures were otherwise the same as described in the methods section for immunohistochemistry on FFPE tissue.

• CC1 buffer (cat. no. 06414575001; Roche Diagnostics) for 32 min at 100C.

• CC1 buffer for 48 min at 100C.

• CC2 buffer (cat. no. 05279798001; Roche Diagnostics) for 32 min at 91C.

• Protease 1 (cat. no. 05266688001; Roche Diagnostics) treatment for 8 min.

• CC1 buffer for 32 min at 100C + protease 1 treatment for 8 min.

• Protease 3 (cat. no. 05266718001; Roche Diagnostics) treatment for 4 min + CC1 buffer for 32 min at 100C.

Control tissue used:

• Tissue microarray blocks containing human tissues from a wide range of different organs and some cancer specimens, including lymph node, tonsil, colon, duodenum, gallbladder, skin, esophagus, liver, pancreas, lung, thyroid gland, kidney, placenta, adrenal gland, cerebellum, cervix uteri, corpus uteri, skeletal muscle, pulmonal adenocarcinoma, and glioblastoma.

Figure A1.

Positive and negative controls for the GDNF antibody [a neuron in glioblastoma tissue (A) and a cerebellar Purkinje cell (B) as positive controls. Neurons in the cerebellar granular layer (E) as a negative control] and for the GDNF RNAscope probe [neurons in glioblastoma tissue (C) and islet cells in endocrine pancreatic tissue (D) as positive controls. Hepatocytes in liver tissue (F) as a negative control]. Scale bars 25 µm.

Figure A2.

Positive and negative controls for the GFRA1 antibody (positive controls: A, neuron, B, adrenal gland tissue, C, ganglion cells in colon, D, renal tubuli. Negative controls: E, hepatocytes in liver tissue, F, glandular cells in colon tissue). Scale bars 25 µm.

Figure A3.

Positive controls for GFRA2 (A, neuron, B, ganglion cells in colon), GFRA3 (C, neurons, D, islet cells in endocrine pancreatic tissue), GFRA4 (E, neuron, F, islet cells in endocrine pancreatic tissue), and RET (G, neurons, H, ganglion cells in colon). Scale bars 25 µm.

Figure A4.

Positive controls for OLIG2 (A, IDH mutated astrocytoma tumor cells), SOX2 (B, glioblastoma tumor cells), CD34 (C, endothelial cells in liver tissue), and IBA1 (D, macrophages in splenic tissue). Scale bars 50 µm.

Figure A5.

Examples of classified images using the Visiopharm software. Chromogenic GFRA1 stained slide to the left and classified image to the right, where nuclei are marked blue, GFRA1-negative tissue is marked red, and GFRA1-positive tissue is marked yellow (A, scale bar 100 µm). Fluorescent double stained slide with GFRA1 (green) and GFAP (red) to the left, and classified image to the right, where nuclei are marked blue, GFRA1 staining is marked green, GFAP staining is marked red, and co-staining for GFRA1 and GFAP is marked yellow (B, scale bar 50 µm, insert, scale bar 20 µm).

Figure A6.

NCAM staining of glioblastoma tissue (A) and magnification (B). Positive staining is present in tumor cells, whereas endothelial cells in vascular structures are negative. Scale bars 100 µm (A) and 50 µm (B).