Importance of GFAP isoform-specific analyses in astrocytoma.

Gliomas are a heterogenous group of malignant primary brain tumors that arise from glia cells or their progenitors and rely on accurate diagnosis for prognosis and treatment strategies. Although recent developments in the molecular biology of glioma have improved diagnosis, classical histological methods and biomarkers are still being used. The glial fibrillary acidic protein (GFAP) is a classical marker of astrocytoma, both in clinical and experimental settings. GFAP is used to determine glial differentiation, which is associated with a less malignant tumor. However, since GFAP is not only expressed by mature astrocytes but also by radial glia during development and neural stem cells in the adult brain, we hypothesized that GFAP expression in astrocytoma might not be a direct indication of glial differentiation and a less malignant phenotype. Therefore, we here review all existing literature from 1972 up to 2018 on GFAP expression in astrocytoma patient material to revisit GFAP as a marker of lower grade, more differentiated astrocytoma. We conclude that GFAP is heterogeneously expressed in astrocytoma, which most likely masks a consistent correlation of GFAP expression to astrocytoma malignancy grade. The GFAP positive cell population contains cells with differences in morphology, function, and differentiation state showing that GFAP is not merely a marker of less malignant and more differentiated astrocytoma. We suggest that discriminating between the GFAP isoforms GFAPδ and GFAPα will improve the accuracy of assessing the differentiation state of astrocytoma in clinical and experimental settings and will benefit glioma classification.

of GFAP as a biomarker for astrocytoma that is still used to date (Dunbar & Yachnis, 2010). In the healthy human brain, GFAP is mainly expressed in mature astrocytes (Middeldorp & Hol, 2011). Therefore, in clinical as well as fundamental experimental settings, high GFAP expression is believed to mark more differentiated, less malignant tumors. However, more recently GFAP expression was observed in the radial glia of the developing human brain and in adult neural stem cells of the adult brain (Middeldorp & Hol, 2011;Roelofs et al., 2005; van den Berge et al., 2010), showing that GFAP is also expressed in immature, nondifferentiated CNS cells. Since then, GFAP is often used to mark cells with stem cell characteristics in glioma and to target neural stem cells to induce gliomagenesis in animal models (Kwon et al., 2008;J. Chen et al., 2012;Bradshaw et al., 2016;Guichet et al., 2016;Kanabur et al., 2016;Jiang et al., 2017;Welker, Jaros, An, & Beattie, 2017). In addition, GFAP is up-regulated in non-neoplastic astrocytes that become reactive in response to the growth of the tumor and do not reflect the differentiation state of neoplastic cells (Gullotta, Schindler, Schmutzler, & Weeks-Seifert, 1985;Yoshii et al., 1992;H. Y. Yang, Lieska, Glick, Shao, & Pappas, 1993). Therefore, high GFAP levels in tumor specimens may not be a direct indication of a less malignant, more differentiated astrocytoma subtype. Indeed, our recent studies in which we determined the expression of different GFAP isoforms show that higher levels of the alternative splice variant GFAPδ relative to the canonical variant GFAPα are associated with a higher malignant and less differentiated astrocytoma subtype (Stassen et al., 2017). In vitro studies that show a higher malignant gene expression profile and changes in astrocytoma malignant behavior in cells with higher levels of GFAPδ relative to GFAPα, as observed in neurogenic stem cells of the healthy brain (Middeldorp & Hol, 2011;Roelofs et al., 2005;van den Berge et al., 2010), further support the hypothesis of GFAP as a marker of more than lower malignant astrocytoma (Moeton et al., 2014;Stassen et al., 2017).
In order to investigate this hypothesis, we systematically reviewed all existing literature on GFAP expression in patient material of astrocytoma (grade I-IV classified according to the WHO 2007 system or earlier), IDH1 wild-type (IDHwt) glioma, and IDH1 mutated glioma without a 1q19p codeletion (IDHmut noncodel; classified according to the WHO 2016 system). We included studies that determined the presence of GFAP in control brains and astrocytoma tissue, in astrocytoma of different malignancy grades, in different areas of the tumor, in blood of astrocytoma patients, in proliferating or invasive cells, and studies that describe the morphology of GFAP expressing cells. We conclude that a strong correlation of GFAP to astrocytoma malignancy is absent, that the GFAP positive population is highly heterogeneous, and that distinguishing between the GFAP isoforms GFAPδ and GFAPα might improve the assessment of the differentiation state and malignancy of the tumor and identify subpopulations of GFAP expressing astrocytoma cells.

| RESULTS
We categorized 88 studies that quantified or described GFAP mRNA or protein expression in human astrocytoma patient material based on the methods that were used and comparisons that were made (Tables 1-8).
The results of the studies in each category are discussed below.

| GFAP expression is increased in astrocytoma versus healthy human CNS tissue
Two early studies already address one of the most important questions considering GFAP in astrocytoma; is the GFAP protein in astrocytoma different from GFAP in the healthy brain, in respect of its immunochemical characteristics and expression level? Delpech et al. (1978) used radial immunodiffusion analysis to demonstrate that GFAP in healthy human brain extracts is immunochemically similar to GFAP in astrocytoma extracts (Delpech et al., 1978). This agrees with the study of Dittmann et al. (1977) who show, by rocket immunoelectrophoresis, that there are no immunochemical differences between GFAP isolated from astrocytoma and the adult human brain. However, GFAP isolated from fetal human brain has different characteristics in this assay when compared to GFAP from astrocytoma or adult human brain (Dittmann et al., 1977). Both studies, and many studies that followed hereafter, show that GFAP expression in astrocytoma is increased compared to healthy brain tissue (Table 1) both at the protein (Delpech et al., 1978;Dittmann et al., 1977;Mauro et al., 1991;Narayan et al., 1986;Palfreyman et al., 1979) and mRNA (Bien-Moller et al., 2018;Laczko et al., 2007) level. In contrast, a more recent study that used two-dimensional (2-DE) gel electrophoresis followed by mass spectrometry, shows a decrease in the 50 kDa canonical GFAP protein in grade IV astrocytoma (although this protein is increased in grade III astrocytoma; Chumbalkar et al., 2005). Interestingly, a second GFAP protein of different mass and isoelectric point was detected in this study, which is upregulated in both grade III and grade IV astrocytoma. This suggests the expression of different GFAP proteins in these tumors, which potentially are isoforms, differently phosphorylated proteins, or degradation products. A variety of GFAP proteins with different molecular weights are described in a second 2-DE gel electrophoresis study that compares astrocytoma to healthy brain tissue, and the results show a higher expression of a 49 kDa GFAP protein and proteins ranging from 36 to 49 kDa in astrocytoma (Narayan et al., 1986). In a third study, 36 kDa and 50 kDa GFAP proteins are detected in astrocytoma, whereas only the 36 kDa GFAP protein is detected in control white matter tissue (Luider et al., 1999). Together, these studies show that the canonical GFAP protein is increased in astrocytoma compared to healthy brain tissue, but also that the detection of different GFAP forms (either splice isoforms, differentially phosphorylated forms, or degradation products) can be potentially used to discriminate between astrocytoma and healthy tissue, and astrocytoma of different malignancy  & Schiffer, 1991 Grade I (n = 1), Grade II (n = 5), Grade III (n = 6), Grade IV (n = 5) All tissues in triplicate Normal human brain tissue (n = 5) IHC on tissue microarrays (TMA) Significant higher level of GFAP in astrocytoma compared to normal brain tissue. Significant effect? Yes Laczko et al., 2007 Low grade (n = 9), Grade IV (n = 20)  (Donev et al., 2010;Gullotta et al., 1985;Oh & Prayson, 1999). Only two out of the 20 studies specify the malignancy grade in which GFAP negative areas or tumors are found. Takenaka et al. (1985) describe GFAP negative tumors in astrocytoma of all grades (Takenaka et al., 1985). van der Meulen et al. (1978) describe GFAP immunoreactivity in all grade I and II astrocytomas, the presence of GFAP negative areas in grade III, and show that four out of 17 grade IV astrocytomas are GFAP negative.
Quantification of differences in GFAP expression between astrocytoma grades was not performed in the studies listed in Table 2. Nevertheless, the observation of a decreasing number of GFAP positive cells with increasing astrocytoma grade is frequently mentioned (Cras et al., 1988;Cruz-Sanchez et al., 1992;Gullotta et al., 1985;Oh & Prayson, 1999;Royds et al., 1986;van der Meulen et al., 1978;Xing et al., 2016). One of the studies, however, describes stronger staining of GFAP in grade III compared to lower grade astrocytoma (Cruz-Sanchez et al., 1992).  van der Meulen et al., 1978 Low and high grade (n = 15) GFAP positive cells and processes in all astrocytoma. Chronwall, McKeever, & Kornblith, 1983 Grade I, II, and III (n = 58) 54/58: Very high and high expression levels 4/58: Positive but low levels Gullotta et al., 1985 Grade I (n = 9), grade II (n = 15), grade III (n = 14), grade IV (n = 6) Grade I: 6/9, grade II: 12/15, grade III: 8/14, grade IV: 2/6 Of all positive samples, nine cases showed diffuse, and 18 cases showed partial immunostaining. Takenaka et al., 1985 Low (n = 6) and high (n = 6) grade GFAP positive cells in all astrocytoma. Yung, Luna, & Borit, 1985 Astrocytoma (n = 71) GFAP reactivity in all tumors. Herpers, Ramaekers, Aldeweireldt, Moesker, & Slooff, 1986 Low and high grade (n = 13) GFAP reactivity in all tumors. Royds, Ironside, Taylor, Graham, & Timperley, 1986 Grade I (n = 12), grade III (n = 9), grade IV (n = 24) GFAP reactivity in all tumors. Cras, Martin, & Gheuens, 1988 Low and high grade (n = 66) GFAP reactivity in all tumors. Cruz-Sanchez et al., 1992 Grade I (n = 5), grade 2 (n = 5), grade III ( Studies in which a scoring system is used to quantify the level of GFAP immunostaining are listed in Tables 3 and 4. In Table 3, data of studies are shown in which GFAP is quantified without applying statistics to test for the significance of differences observed between astrocytomas. In six out of eight of these studies, there are no clear differences between GFAP scores in low-and high-grade astrocytoma. One study reports lower GFAP immunostaining scores in grade III and IV compared to I and II (Peraud et al., 2003) and in contrast, another study describes higher GFAP scores in grade IV compared to grade I astrocytoma (Colin et al., 2007). In the studies listed in Table 4, statistical testing is used to determine the significance of the differences in GFAP immunostainings between astrocytoma of different grades. Two out of 11 studies that compare grade I and II to grade III and grade IV, show significant higher GFAP scores in grade I and II astrocytoma (Hlobilkova et al., 2007;L. Yang et al., 2014). In one study, a significant decrease in GFAP levels with increasing astrocytoma grade is reported (Laczko et al., 2007) and a fourth study finds a significant correlation of GFAP levels to malignancy grade and a significant difference between grade II and IV astrocytoma (Schwab et al., 2018). Seven out of the 11 studies failed to show significant differ-   (Rasmussen et al., 1980) both report a high variability in expression. In two other studies that quantified protein levels, GFAP is higher in grade II compared to grade IV astrocytoma (Jacque et al., 1978;Odreman et al., 2005), which reached statistical significance when tested (Odreman et al., 2005). In addition, a third study reports higher GFAP levels in grade I and II compared to grade III and IV astrocytoma but only show western blot evidence for one sample of each grade (Peraud et al., 2003). In short, studies that analyzed astrocytoma homogenates neither generate consistent results on the correlation of GFAP expression to astrocytoma malignancy grades and additional analyses in larger cohorts are needed. In our own previously published study in which we used RNA sequencing data of 310 patients available at TCGA we do show a strong Grade II (n = 5), Grade III (n = 10), Grade IV (n = 26) Negative, single cells, cluster of cells (20-50%), 50%-90% of cells are positive, Almost 100% are positive Grade I and II: >50% of the cells Grade III and IV: Single cells or negative Peraud et al., 2003 Grade I and II (n = 5), Grade IV (n = 4) Scores from 0 to 3 9/9 score 2 or 3 Tan, Magdalene Koh, & Tan, 2006 Grade II (n = 10), Grade III (n = 11), Grade IV (n = 5) Percentage of positive cells All astrocytoma examined ranged from 5% to 100% of positive cells. Rousseau et al., 2006 Grade I (n = 8), Grade IV (n = 8)

| GFAP is heterogeneously expressed in astrocytoma
The above-described studies show a large variation in outcomes. To a certain extent, variation between studies is explained by the different methods of analysis, grouping of patient samples, sample sizes, scoring systems, antibodies used, quality of tissues, and variation in staining intensity within and between tumors (Cuny et al., 2002). Moreover, the complexity of GFAP positive cell morphology and staining patterns hamper quantification of immunohistochemistry data of the number of GFAP positive cells to the total number of cells (Tanaka et al., 2008).
Besides these methodological issues, the inconsistent correlation of GFAP expression to astrocytoma malignancy might results from Grade I and II (n = 10), Grade III and IV (n = 29) <25%, 25-50%, 50-80% Positive staining found in all tumors. No difference between grades.
No difference between malignancy grades, or recurrent versus primary tumors.
No Stan et al., 1999 Grade I (n = 7), Grade II (n = 13), Grade III (n = 7), Grade IV (n = 23) Yes, p = 0.027 Hlobilkova et al., 2007 Grade I (n = 1), Grade II (n = 5), Grade III (n = 6), Grade IV (n = 5) Score heterogeneity in localization and function of GFAP positive astrocytoma cells within and between tumors. Local variability of GFAP immunostaining was highlighted in many studies (Cras et al., 1988;Cruz-Sanchez et al., 1992;Gullotta et al., 1985;Herpers et al., 1986;Royds et al., 1986;Sharpe & Baskin, 2016;Tascos et al., 1982;van der Meulen et al., 1978;Velasco et al., 1980). Thus, as was already noted in 1985 (Gullotta et al., 1985) and many times thereafter (Hashemi et al., 2014;Sembritzki et al., 2002) analysis of different areas of the same tumor is necessary to determine GFAP expression levels of a tumor. This is emphasized by two studies that specifically analyzed regional differences within astrocytoma grade IV. Nagashima, Suzuki, Asai, and Fujimoto (2002)  staining in these areas has been described as well. While these characteristics are observed in all astrocytoma grades, some have been observed more frequently in low-or in high-grade astrocytoma. GFAP negative microcystic (Gullotta et al., 1985;Velasco et al., 1980) and necrotic areas (Tascos et al., 1982), and focal GFAP staining are more often seen in higher grade astrocytoma (Cras et al., 1988;Cruz-Sanchez et al., 1992), whereas diffuse GFAP staining (Cruz-Sanchez et al., 1992;Royds et al., 1986) and a dense fibrillary network with clear staining of processes (Peraud et al., 2003;Zamecnik et al., 2004) are more often seen in lower grade astrocytoma with one study specifically reporting on shorter processes in grade III and grade IV astrocytoma (Zamecnik et al., 2004). However, as GFAP positive neoplastic cells are intermingled with GFAP positive reactive astrocytes, determining the neoplastic origin of these thin and complex processes is

Described differences between astrocytoma grades Reference
Grade III (n = 5), Grade IV (n = 8) 1. Immunofluorescence found in the cell body and processes of malignant cells. 7. Peri-vascular staining. 8. GFAP expression aligning the inner surface of the plasma membrane.

More often seen in grade III
and specifically grade IV astrocytoma Cras et al., 1988 Low and high grade (n = 66) 4. Focal GFAP expression only. 6. Diffuse/homogeneous GFAP staining.
4. More often seen in high astrocytoma. 6. More often in low-grade astrocytoma.
1. Staining of processes more often observed in low-grade astrocytoma Peraud et al., 2003 (Continues) complicated (Gullotta et al., 1985). Although most studies report to be able to discriminate between these two cell types, and distinctions in the morphology of reactive and neoplastic astrocytes have been clearly described (Yoshii et al., 1992), quantification of process-rich GFAP positive areas might lead to an overestimation of GFAP levels in neoplastic cells. Additional immunohistochemistry of for example the 300 kDa intermediate filament associated protein (IFAP) specifically expressed in neoplastic astrocytes (H. Y. Yang et al., 1993), and the more recently identified ATRX, that is lost in neoplastic but present in reactive astrocytes (Mellai et al., 2017), might be helpful in distinguishing these cell types.
In accordance with the variability in cell morphology and immu-  (Schiffer, Giordana, Germano, & Mauro, 1986) and a lack of GFAP coexpression with the proliferation marker Ki67 in 15 low grade and 10 grade IV astrocytomas (Kros, Schouten, Janssen, & van der Kwast, 1996). In contrast, low levels of proliferating GFAP-expressing cells are detected in another study that reports coexpression of GFAP and Ki67 in 8.8% (±13.6%) of the total number of Ki67 expressing cells in low-grade astrocytoma (Tanaka et al., 2008). In grade IV astrocytoma, Ki67 is expressed in GFAP positive cells as well, although there are significantly more Ki67 cells that are negative for GFAP (Takeuchi, Sato, Ido, & Kubota, 2006). Similarly, lower numbers of argyrophilic nucleolar organizer regions (Ag-NORs), an indicator of proliferation rate, are present in GFAP positive cells compared to GFAP negative cells in astrocytoma of all grades (Kajiwara et al., 1992). This is confirmed in a second study that also shows a significantly lower number of Ag-NORs in GFAP positive cells compared to GFAP negative cells, although the number of Ag-NORs in GFAP positive cells was highly variable (Hara et al., 1991;Kajiwara et al., 1992). GFAP positive cells with high numbers of Ag-NORs were described as well (Kajiwara et al., 1992). In contrast, a more recent study shows Ki67-GFAP coexpression in astrocytoma and describes that 97% of Ki67 positive cells in low grade and 74% in high-grade astrocytoma is GFAP positive.
The number of Ki67-GFAP double positive cells is significantly higher in low-grade astrocytoma compared with high-grade astrocytoma (L. Yang et al., 2014). According to these studies, GFAP is expressed in both proliferating and nonproliferating cells. Local differences in the distribution of these cells might again account for the variation in Three studies determined GFAP expression in invading parts of astrocytoma. In areas of grade IV astrocytoma that invade cortical and white matter tissue, both GFAP positive and GFAP negative cells are present (Schiffer et al., 1986). In addition, invading astrocytoma cells into connective tissue (i.e., meningeal invasion) are marked by increased GFAP expression in the invading part compared to the noninvading part (Herpers, Budka, & McCormick, 1984;Nakopoulou et al., 1990). These studies further emphasize the large variation in GFAP expressing cells in astrocytoma that most likely causes the lack of a strong correlation of GFAP to astrocytoma malignancy based on current published literature. The identification of GFAP expressing subtypes of cells will be necessary to determine the role of GFAP in astrocytoma malignancy.

| Differential GFAP isoform expression to distinguish astrocytoma subtypes
In the healthy human brain, the GFAP positive subtype of neurogenic stem-cell like cells can be distinguished by the expression of a GFAP isoform that results from the process of alternative splicing, GFAPδ
2. Dense fibrillary network mainly in grade I and II astrocytoma 2. Shorter processes in grade III and IV astrocytoma Zamecnik et al., 2004 Grade I (n = 37), Grade II (n = 11) 2. Dense meshwork of GFAP positive cellular processes, stained over their full length, including thin processes. 3. GFAP negative microcystic (small cell) areas. Tanaka, Sasaki, Ishiuchi, & Nakazato, 2008 n = number of cases. (Middeldorp & Hol, 2011;Roelofs et al., 2005;van den Berge et al., 2010). A few studies have determined the expression level of the GFAPδ isoform in astrocytoma of different grades as well. The studies in Table 7 report that GFAPδ expression is rarely observed in the healthy brain (Choi et al., 2009;Heo et al., 2012), but is detected in reactive gliosis (Andreiuolo et al., 2009), and is increased in astrocytoma of all grades when analyzed by immunohistochemistry (Choi et al., 2009;Heo et al., 2012). In contrast, RNA quantification of GFAPδ shows increased levels in only three out of eight and decreased levels in five out of eight grade IV astrocytoma compared to control tissue (Blechingberg et al., 2007). Interestingly, studies that find differences in GFAP expression between astrocytoma grades report on a decrease of general GFAP levels with increasing astrocytoma grade, but higher levels of GFAPδ are found in grade IV astrocytoma compared to grade I (Andreiuolo et al., 2009), grade II (Brehar et al., 2014), and grade I, II, and III (Choi et al., 2009). In one of these studies, general GFAP levels are quantified and show a decrease in grade IV compared to grade III, II, I, and control tissue (Choi et al., 2009). In grade I, II, and III spinal cord astrocytoma, GFAPδ expression also increases with increasing grade (Heo et al., 2012). Interestingly, GFAPδ levels are significantly associated with a rounder cell morphology and fewer cellular processes (Choi et al., 2009;Heo et al., 2012), and with highly invasive grade IV astrocytoma (Brehar et al., 2014). In addition, two grade II astrocytoma that are categorized as highly Grade III (n = 8), Grade IV (n = 1), Frontal cortex control (n = 1) Quantitative PCR GFAPα, GFAPδ, and GFAPκ expression is decreased in 5/8 and increased in 3/8 astrocytoma compared to control tissue. The GFAPκ/GFAPδ ratio is increased in all astrocytoma compared to control tissue. Blechingberg et al., 2007 Grade I (n = 4), Grade IV (n = 2), Control healthy and epileptic tissue

Immunohistochemistry; No quantification
GFAPδ is coexpressed with GFAP in reactive astrocytes. GFAPδ expressed in 1/4 grade I astrocytoma, focal expression in 3/4 grade I tumors, but mostly negative. Grade IV: Strong focal GFAPδ expression Andreiuolo et al., 2009 Grade I (n = 3), Grade II (n = 5), Grade III (n = 4), Grade IV (n = 4), Control autopsy material (n = 4) Immunohistochemistry; Quantification of mean grey value after outlining the cell Increased pan GFAP and GFAPδ expression in astrocytoma compared to control tissue. Pan GFAP levels significantly increase from grade I to grade III and are decreased in grade IV astrocytoma. GFAPδ expression is mostly undetectable in control cells and increases with increasing astrocytoma grade. GFAP positive control cells are stellate-shaped with well-developed processes. Grade I and II: Stellate polygonal or round cells Grade III and IV: Round and spindle shaped cells GFAPδ is mainly observed in cell bodies not in processes. Inverse correlation of GFAPδ expression intensity and the amount of processes (round>polygonal>stellate).
invasive show strong GFAPδ expression (Brehar et al., 2014). These studies indeed indicate that GFAPδ can be used to identify astrocytoma subpopulations of cells as well and suggest that GFAP expressing cells with different functions (e.g., proliferating, quiescent, invasive, and static) consist of a different combination of GFAP protein isoforms. The detection of a second GFAP alternative splice variant, GFAPκ, in RNA isolated from grade IV astrocytoma further supports this hypothesis (Blechingberg et al., 2007). We recently showed that quantification of the relative level of GFAPδ to GFAPα, the GFAPδ/GFAPα ratio, using RNA sequencing data obtained from the cancer genome atlas (TCGA) indeed indicates that low-and highgrade astrocytoma express different combinations of GFAP variants.
In astrocytoma grade IV, the GFAPδ/GFAPα ratio was significantly higher compared to grade II and III (WHO 2007;Stassen et al., 2017).

| GFAP as a blood biomarker
The most consistent results on the relationship of GFAP to astrocytoma malignancy have been generated by the analysis of blood or cerebrospinal fluid (CSF) of astrocytoma patients. An early study already shows that GFAP levels in CSF can be used to distinguish astrocytoma from other types of tumors and healthy controls (Szymas, 1985), but there was no follow up study. As shown in Table 8, subsequent studies link GFAP detection in blood to grade IV astrocytoma specifically. Eight out of 12 studies report on a significant association of GFAP levels detected in serum (Baumgarten et al., 2018;Gállego Pérez-Larraya et al., 2014;Ilhan-Mutlu et al., 2013;Jung et al., 2007;Kiviniemi et al., 2015;Tichy et al., 2016), in microparticles (Sartori et al., 2013), and in mono-nucleated cells (Muller et al., 2014) isolated from blood of grade IV astrocytoma patients in comparison to lower grade astrocytoma, nonglial tumors, other neurological diseases, and healthy controls. One of these studies reports detectable GFAP levels in blood of grade III astrocytoma patients as well, but levels in grade IV astrocytoma patients were significantly higher (Kiviniemi et al., 2015). Similarly, GFAP positive exosome numbers are increased in grade III and IV astrocytoma compared to healthy controls (Galbo et al., 2017). Two of the 12 studies report on higher GFAP levels in plasma of grade IV patients, but no statistics were performed (Vietheer et al., 2017) or no statistical significance was reached (Lange et al., 2014). High GFAP levels in blood of grade IV astrocytoma patients is in contrast with the most often described lower GFAP levels in higher grade astrocytoma tissue (Tables 3-5). However, although one study also finds higher GFAP expression levels in tumor cells of patients with high GFAP serum levels (Tichy et al., 2016), . These studies suggest that GFAP serum levels are related to brain damage and cell death induced by, in, or near the tumor. This is supported by studies that have linked high GFAP serum levels in patients with traumatic brain injury (Bazarian et al., 2018;Thelin et al., 2017). Increased levels of GFAP positive microparticles can be observed from 7 days up to 7 months after surgery (Sartori et al., 2013), although another study that measured GFAP serum levels 6 weeks after surgery, does not show an increase in GFAP and for some patients the GFAP serum levels were even lower compared to levels before surgery (Vietheer et al., 2017). GFAP levels in blood might be induced by regrowth of the tumor, as levels of GFAP microparticles at 7 months compared to 7 days are increased in blood of patients with subtotal compared to The GFAPδ/α ratio distinguishes astrocytoma subpopulations. Overview of low-(left panel) and high-grade (right panel) astrocytoma and differences in the heterogeneous GFAP positive cell population. High-grade astrocytoma (right panel) is characterized by increased mitosis and cell density, necrosis (black area) and vascularization (red vessels). Invasive astrocytoma cells use white matter tracts, blood vessels and meninges as a surface to migrate on (Claes, Idema, & Wesseling, 2007). GFAP levels in blood are specifically associated with grade IV astrocytoma. In both high-and low-grade astrocytoma, the GFAP positive cell population is highly heterogeneous and contains cells with various functions (e.g., proliferating, quiescent, invasive, and static). GFAP negative areas are more often found in the center of high-grade tumors (orange arrows). The GFAPδ isoform distinguishes astrocytoma subpopulations of cells (a, b), and as the GFAPδ/α ratio is increased in grade IV astrocytoma, this subpopulation is most likely larger in these tumors (b). GFAP protein and GFAP positive cells in blood of patients are associated with high-grade astrocytoma and might contain different levels of GFAP isoforms (c). Similarly, invading cells that, for example, invade the meninges (connective tissue) might consist of a specialized GFAP network that equips them for this behavior (d) gross-total resections of the tumor (Sartori et al., 2013). Moreover, increased levels after 7 months of surgery compared to 4 months are seen in patients with radiological disease progression (Sartori et al., 2013). Furthermore, the significant negative correlation of preoperative GFAP serum levels with the time until tumor recurrence (progressionfree survival [PFS]) for grade III and IV astrocytoma patients (Kiviniemi et al., 2015) indicates that factors in the biology of the tumor contribute to the GFAP serum levels, rather than surgical damage only. Although, most studies did not find a significant correlation to either progressionfree survival or survival of patients (Gállego Pérez-Larraya et al., 2014;Ilhan-Mutlu et al., 2013;Jung et al., 2007;Muller et al., 2014;Sartori et al., 2013;Vietheer et al., 2017). In addition, high GFAP levels in blood of patients prior to treatment are associated with epidermal growth factor receptor amplified (EGFRvIII) tumors within grade IV astrocytoma (Muller et al., 2014), with IDHwt tumors within grade III and IV astrocytoma (Kiviniemi et al., 2015) and with higher levels of Ki67 (cell division marker) expression in tumor cells (Kiviniemi et al., 2015). Interestingly, one study has isolated GFAP expressing cells from blood of grade IV astrocytoma patients and showed that they contain astrocytoma spe- subpopulations. As these variants, as shown for GFAPδ, differentially correlate to the malignancy of the tumor, the current use of commercial GFAP antibodies that recognize all isoforms most likely masks a consistent correlation of GFAP to astrocytoma malignancy grade. Discrimination between GFAP variants, as we show here for GFAPδ and GFAPα, helps to identify different types of GFAP positive cells that could improve the assessment of astrocytoma differentiation and malignancy. We hypothesize, as summarized in Figure 1, that GFAP is expressed in heterogenous astrocytoma cells with a low malignant, more differentiated and noninvasive phenotype, as well as a high malignant, stem-cell like more invasive phenotype. Higher levels of GFAPδ are expressed in neurogenic stem-cell like cells of the healthy brain (Middeldorp & Hol, 2011;Roelofs et al., 2005;van den Berge et al., 2010) and in higher malignant astrocytoma (Andreiuolo et al., 2009;Brehar et al., 2014;Choi et al., 2009;Heo et al., 2012), and the GFAPδ/α ratio is increased in grade IV astrocytoma (Stassen et al., 2017). Therefore, cells with a high GFAPδ/α ratio might be the high malignant, stem-cell like more invasive cells of the GFAP cell population. These cells are present at lower numbers in low-grade astrocytoma and could potentially induce progression into higher malignancy grades. Differences in malignant behavior of cells with a high and low GFAPδ/α ratio support this hypothesis (Moeton et al., 2014;Stassen et al., 2017) and future studies should focus on unravelling the isoform-specific function in astrocytoma malignancy. In conclusion, information is lost when the expression of different GFAP isoforms is neglected and can be deceiving when GFAP is used to determine the differentiation state of a cell in experimental and clinical settings.
Therefore, future studies need to focus on further identifying the GFAP positive cell population and make use of the possibility to discriminate between GFAP variants that could be fruitful to diagnosis and to the understanding of the molecular basis of glioma.