Glioblastoma, a malignant brain cancer, is characterized by abnormal activation of receptor tyrosine kinase signalling pathways and a poor prognosis. Extracellular proteoglycans, including heparan sulfate and chondroitin sulfate, play critical roles in the regulation of cell signalling and migration via interactions with extracellular ligands, growth factor receptors and extracellular matrix components, as well as intracellular enzymes and structural proteins. In cancer, proteoglycans help drive multiple oncogenic pathways in tumour cells and promote critical tumour–microenvironment interactions. In the present review, we summarize the evidence for proteoglycan function in gliomagenesis and examine the expression of proteoglycans and their modifying enzymes in human glioblastoma using data obtained from The Cancer Genome Atlas (http://cancergenome.nih.gov/). Furthermore, we demonstrate an association between specific proteoglycan alterations and changes in receptor tyrosine kinases. Based on these data, we propose a model in which proteoglycans and their modifying enzymes promote receptor tyrosine kinase signalling and progression in glioblastoma, and we suggest that cancer-associated proteoglycans are promising biomarkers for disease and therapeutic targets.
Glioblastoma (GBM), the most common primary malignant brain tumour of adults, is characterized by abnormal activation of receptor tyrosine kinase (RTK) signalling pathways and diffuse invasion of tumour cells into the adjacent brain. Although advances in current therapeutic strategies have improved outcome for GBM patients, the current median survival for GBM remains < 2 years [1, 2]. A major impediment to treatment has been the diversity of oncogenic alterations present across tumours, as well as within an individual tumour [3-5].
Proteoglycans, including heparan sulfate and chondroitin sulfate proteoglycans (HSPG and CSPG, respectively), regulate the activity of many signalling pathways, as well as cell–microenvironment interactions. As a result of these diverse functions, proteoglycans and their modifying enzymes have been implicated in tumourigenesis in a number of cancers [6-9]. In brain cancer, data from our laboratory and other studies suggest that proteoglycans regulate multiple oncogenic pathways in tumour cells and promote critical tumour–microenvironment interactions [10-17]. Thus, proteoglycans and their modifying enzymes are potentially important therapeutic targets and biomarkers of GBM.
A number of fundamental studies important in our understanding of proteoglycan function have focused on processes outside of the central nervous system (CNS); thus, we begin the present review with an overview of these studies. We then summarize some of the known functions of proteoglycans in the brain and in cancer, and we examine the expression of proteoglycans and proteoglycan-related enzymes in human GBM. Because GBM can be stratified into biologically and clinically relevant subgroups [18-20], we investigate proteoglycan and proteoglycan-related gene expression across tumour subtypes. We further show a correlation between RTK amplification and specific alterations in proteoglycans. We propose a model in which proteoglycans and their modifying enzymes promote RTK signalling and progression in GBM. An improved understanding of proteoglycan function in brain tumours may reveal druggable therapeutic targets for this deadly disease.
Proteoglycans in development and disease
Proteoglycans consist of a core protein and covalently attached glycosaminoglycan (GAG) chains, including heparan sulfate (HS), chondroitin sulfate (CS) and keratan/dermatan sulfate. Proteoglycans can be inserted in the plasma membrane, glycophosphatidylinositol (GPI) anchored to the membrane, or secreted (Fig. 1). In the brain, the most abundant components of the extracellular environment include HSPGs, CSPGs and hyaluronan/hyaluronic acid (HA) [21, 22]. Although HA is not a proteoglycan, it is a GAG that is directly released extracellularly by hyaluronan synthase (HAS) enzymes located in the plasma membrane. Moreover, HA on neural stem cells and their differentiated progeny binds to a number of proteoglycans, including neurocan, aggrecan (ACAN) and versican (VCAN) .
As a result of their ability to interact with diverse partners, including soluble factors, membrane proteins and the extracellular matrix (ECM), proteoglycans can regulate processes ranging from ligand-mediated signalling involved in cell proliferation to cell adhesion and cell migration [22, 24, 25] (Fig. 1). In addition, the intracellular domain of some transmembrane proteoglycans can interact with the cytoplasmic domain of proteins and help regulate intracellular signalling. For example, the cytoplasmic tail of syndecan-4 directly interacts with α-actinin to help regulate cytoskeletal organization [26-28], and the cytoplasmic domain of syndecan-1 (SDC1) interacts with talin to modulate integrin signalling via a SDC1–integrin–insulin-like growth factor 1 receptor trimolecular complex .
As extracellular reservoirs for growth factors and as enablers of growth factor-receptor interactions, proteoglycans play a dual role in regulating ligand-mediated signalling. Proteoglycans can bind and sequester soluble ligands, which help to establish and maintain morphogen gradients [30-32]. For example, decreased HSPG synthesis disrupts normal localization of wingless (Wg), Decapentaplegic (Dpp) and hedgehog (Hh) and results in developmental defects in Drosophila [33, 34]. On the other hand, proteoglycans can act as co-receptors for ligand-mediated signalling. Indeed, HSPGs stabilize the fibroblast growth factor (FGF) ligand–receptor complex to promote FGF2 signalling [35-40].
The critical role for proteoglycans, including both the core protein and GAGs, in normal development and growth is illustrated by data from model organisms and from the study of human disease . In mice, a deficiency of perlecan (HSPG2), which is expressed in the basal lamina of the brain, can result in exencephaly or neuronal ectopias [42-44]. Mutations in human GPC3 (glypican-3) can result in Simpson–Golabi–Behmel syndrome, which is characterized by overgrowth of multiple tissues and tumour susceptibility . Mice genetically engineered to lack CSPG4/NG2 show multiple phenotypes ranging from delays in the production of mature oligodendrocytes to deficits in brown fat function and adult onset obesity [46, 47]. The data demonstrate proteoglycans play important roles in development and tissue homeostasis.
Proteoglycans and their GAG chains undergo extensive post-translational and post-synthetic enzymatic modifications to generate the necessary structural diversity important for their function. For HSPGs, a major determinant of the specificity and the affinity of ligand interactions is the sulfation pattern of the HS chains, particularly the 6O-sulfate (6OS) of glucosamine [24, 48-51]. Regulation of 6OS levels occurs during biosynthesis and post-synthetically by the extracellular sulfatases (SULFs) [48, 52]. Dhoot et al.  first identified the SULFs and demonstrated their ability to regulate Wnt signalling in myogenic progenitor cells in the quail embryo. Subsequently, the SULFs have been shown to regulate HSPG-dependent signalling by removing 6OS and liberating protein ligands from HSPG sequestration, including Wnts, FGF2, vascular endothelial growth factor (VEGF), glial cell line-derived growth factor and stromal cell-derived factor 1 [53-62].
The importance of HSPG sulfation in normal development is well illustrated by loss of function studies. Genetic ablation of heparan sulfate 6O-sulfotransferase (HS6ST)-1 is embryonic lethal and the effected placentas have a profound reduction in microvasculature . Furthermore, mouse embryonic fibroblasts derived from mice with ablation of HS6ST1 and HS6ST2 show reduced FGF-dependent signalling . This is consistent with the co-receptor function for proteoglycans in FGF2 signalling. There are two SULF genes in vertebrates (SULF1 and SULF2) and one or both have been implicated in a number of developmental processes, including brain development, oesophageal development, dentinogenesis and bone development [55, 65-67].
Taken together, alterations in proteoglycan core proteins, biosynthetic enzymes and the extracellular regulating enzymes are associated with a number of developmental anomalies and, in some cases, overgrowth or tumour predisposition syndromes.
Proteoglycan function in the brain
Proteoglycans are abundant in the brain and have known roles in normal development and neurologic disease. Both HSPGs and CSPGs are up-regulated in neurogenic brain regions where the regulation of growth factor signalling is critical [25, 68, 69]. SDC1 is highly expressed in the neural germinal zone of the developing cortex and can regulate the proliferation and maintenance of neural progenitor cells, at least partially through the modulation of Wnt signalling . CSPGs are expressed on neural progenitor cells and their enzymatic removal is associated with decreased proliferation in response to FGF2 and also alters cell differentiation in response to epidermal growth factor (EGF) .
The proteoglycan CSPG4/NG2, encoded by the cspg4 gene, is an example of a molecule with pleiotropic functions in the postnatal mammalian brain. CSPG4/NG2 is an integral membrane proteoglycan and has a large extracellular domain and a small cytoplasmic domain. CSPG4/NG2 is found on the surface of several immature progenitor cells, including oligodendrocyte progenitor cells (OPCs) and pericytes . On OPCs, the largest population of dividing cells in the adult brain, CSPG4/NG2 promotes cell proliferation and cell migration [46, 73, 74]. Early studies suggested that CSPG4/NG2 cooperates with platelet-derived growth factor receptor-α (PDGFRA), which mediates OPC proliferation in response to its ligand PDGF . Subsequent studies in mice lacking the cspg4 gene have demonstrated a role for CSPG4/NG2 in promoting the proliferation of PDGFRA-positive OPC at postnatal stages . Indeed, PDGFAA and FGF2 show high affinity binding to the CSPG4/NG2 core protein , suggesting that the proteoglycan may act as a reservoir and co-receptor for these growth factors, thereby modulating RTK signalling. In addition to influencing ligand bioavailability, direct interactions between CSPG4/NG2 and the RTK itself are considered to promote mitogenic signalling, as observed for FGFR1 and FGFR3 in pericytes and smooth muscle cells .
Recently, Sugiarto et al.  demonstrated that CSPG2/NG2 regulates EGF-dependent proliferation and self-renewal of OPCs. Moreover, CSPG4/NG2 itself exhibits polarized localization in OPCs prior to mitosis and is unequally inherited to the self-renewing progeny of OPCs but not the progeny destined to differentiate. The study by Sugiarto et al.  provides unprecedented evidence that OPCs divide asymmetrically to regulate self-renewal and differentiation. CSPG4/NG2 is also required to set-up OPC polarity in part by achieving asymmetric segregation of active epidermal growth factor receptor (EGFR). These data show not only that CSPG4/NG2 is a marker of polarity and self-renewing cell fate, but also that it may actively participate in regulating such fundamental processes as asymmetric progenitor divisions.
CSPG4/NG2 further interacts with cytoplasmic and ECM components such as collagen V and VI  and forms signalling complexes with α3β1 integrin and galectin-3, possibly acting as a co-receptor [74, 78]. Other partner molecules include matrix metalloproteinase-16, plasminogen and tissue-type plasminogen activator . Moreover, several binding partners of the C-terminal, PDZ-containing cytoplasmic domain include the multiple PDZ domain protein (MUPP1), the glutamate receptor interacting protein and syntenin-1 [80-82]. The interaction with glutamate receptor interacting protein suggests a function for CSPG4/NG2 in synapse formation, whereas its interaction with syntenin-1 provides connection to the cell cytoskeleton. In addition, differential phosphorylation of the CSPG4/NG2 cytoplasmic domain, and subsequent changes in cellular localization, are determinants of β1-integrin signalling in glioma [83, 84].
During brain development, the sulfation of HS on neural progenitor cells changes, and this may be important in regulating their differentiation . Indeed, undersulfation of HS on embryonic stem cells restricts their differentiation potential and blocks maturation along the neural lineage . In the ventral spinal cord, the SULFs establish a morphogen gradient for Shh, which is required for neural progenitors to switch from a neuronal to an oligodendroglial fate [58, 86, 87]. Knockdown of CSPG biosynthetic enzymes has shown that sulfation of CS chains is a critical determinant of cell migration from the ventricular zone into the cortical plate .
In addition to their role in development, proteoglycans also play important roles in response to nervous system injury and neurologic disease. In the CNS, CSPGs are up-regulated in response to injury and demyelination, and they can limit axonal regeneration and remyelination [89-91]. Interestingly, despite their inhibitory action on neural repair, exogenous application of CSPG fragments has been shown to reduce excitotoxicity and to repress neural cell death induced by glutamate analogues . Furthermore, in disease models, CSPG4/NG2 deficiency leads to reduced myelin repair as a result of the decreased proliferation of normally CSPG4/NG2+ cells, such as OPCs, pericytes and macrophages/microglia . The 6O-sulfation levels of HS are also important in regulating signalling in neural injury. In a recent study, increased HS sulfation was shown to promote the formation of the glial scar, suggesting that manipulation of HS sulfation may be therapeutically useful after CNS injury .
Proteoglycan alterations are clearly a normal response to CNS injury and the modulation of this response may be important for promoting repair or ameliorating disease.
Proteoglycans in cancer growth, angiogenesis and inflammation
Consistent with their diverse roles in normal growth and development, proteoglycans have been implicated in influencing several aspects of tumour biology, including cell proliferation, tumour cell adhesion and migration, inflammation, and angiogenesis [6, 8, 95, 96].
Changes in proteoglycan core proteins are observed in many cancers and are often associated with changes in cell signalling and invasion. For example, glypican-1 (GPC1) expression is elevated in human pancreatic cancer and the attenuation or ablation of GPC1 expression in pancreatic tumour cells can confer a decreased response to FGF2 and heparin-binding EGF-like growth factor, decreased downstream mitogen-activated protein kinase signalling, decreased tumour cell proliferation, decreased production of angiogenic factors, and decreased tumour growth and metastasis in vivo [97-99].
Alterations in proteoglycan biosynthetic and modifying enzymes are also associated with carcinogenesis. Heparanase (HPSE) specifically liberates biologically active GAG chains from HSPGs, and its activity is associated with increased tumourigenesis, angiogenesis and invasiveness in diverse cancers, including, multiple myeloma, breast and pancreatic cancer . In multiple myeloma, a number of oncogenic functions for HPSE have been identified, including direct effects on tumour cells and indirect effects on tumour-associated endothelial cells [101-103]. For example, HPSE can drive hepatocyte growth factor (HGF) signalling via its ability to increase the production of HGF and increase the production and shedding of SDC1, which forms a SDC1–HGF complex that is able to activate the c-Met receptor [101, 104].
As noted above, the levels of 6O-sulfation are critical for HSPG function, and altered SULF levels have been associated with a large number of human cancers, including brain, breast, nonsmall cell lung carcinoma, multiple myeloma, gastric, pancreatic cancer and hepatocellular carcinoma [8, 9, 56, 105-110]. In brain cancer, SULF2 has been directly implicated in driving tumourigenesis in murine and human malignant glioma [11, 111]. SULF2 protein is highly expressed in primary human GBM, and SULF2 levels are inversely related to HSPG 6O-sulfation in a murine model for GBM. Furthermore, the ablation of SULF2 decreases tumour growth, prolongs host survival and decreases the activity of PDGFRA, as well as related downstream signalling pathways .
Importantly, proteoglycans also regulate ligand-mediated signalling in non-neoplastic components of the tumour. In multiple myeloma, SDC1 shed from tumour cells that express high levels of HPSE promotes VEGF signalling in endothelial cells . Conditional knockout of N-acetylglucosamine N-deacetylase/N-sulfotransferase 1 (Ndst1) in endothelial cells alters their HSPG composition, decreases FGF2 and VEGF signalling in tumours, and reduces tumour angiogenesis in vivo . Interestingly, vessel formation during normal development was not affected in these mice, suggesting that proteoglycan involvement in angiogenesis is different in disease, providing an opportunity to specifically target tumour vascularity. In ovarian cancer, increased HSPG 6O-sulfation, as a result of increased HS6OST levels in tumour-associated endothelial cells, promotes VEGF and FGF signalling . CSPG4/NG2 knockout mice also have reduced vascularization of syngeneic tumours as a result of decreased integrin-dependent pericyte recruitment and altered ECM collagen deposition . In addition, proteoglycans have important roles regulating the immune response to cancer . Indeed, tumour-derived VCAN has been shown to activate macrophages and increase metastatic tumour growth in a model for lung carcinoma . Furthermore, the overexpression of HPSE results in chronic inflammation and is involved in immune cell recruitment in colitis-associated colon cancer . Thus, tumour proteoglycans can promote tumour progression via direct effects on tumour cells and via effects on endothelial cells and immune cells that contribute to tumour-associated angiogenesis and inflammation, respectively.
Diffuse glioma, which include astrocytoma, oligodendroglioma and oligoastrocytoma, are characterized by invasive growth of tumour cells into adjacent non-neoplastic brain, making complete surgical excision impossible. GBM, an astrocytoma, is the most malignant and most common type of diffuse glioma. Common to almost all GBM is dysregulation of RTK signalling pathways. Despite the importance of RTK signalling in GBM, therapies targeting RTKs have had little clinical success. One explanation may be that tumours are driven by the summation of multiple signalling inputs [3-5]. In addition, GBM is a molecularly heterogeneous disease and tumours can be stratified into molecularly defined subtypes with differences in signalling pathway activation [18-20, 118-121]. Abnormal RTK activity can be driven by alterations in receptor expression or by altered ligand availability. As described above, proteoglycans regulate the extracellular availability of oncogenic factors; receptor localization and activity; immune cell recruitment; and tumour cell migration, all of which contribute to tumourigenesis and are likely to be involved in GBM.
Collectively, the data relating to developmental biology and from other cancers suggest that proteoglycans have the potential to regulate multiple determinants of tumourigenesis in GBM. We therefore examined common proteoglycan alterations in GBM and whether they are associated with abnormalities of specific signalling pathways.
Altered proteoglycan expression in human GBM
A comprehensive analysis of the expression pattern of proteoglycans and their modifying enzymes in human GBM is lacking. Using data from The Cancer Genome Atlas (http://cancergenome.nih.gov/) , we investigated the expression of proteoglycan and proteoglycan-related genes across 424 human tumours. Multiple proteoglycan core proteins and related enzymes were differentially expressed in GBM tumours relative to normal brain (Fig. 2).
These included many genes previously implicated in promoting tumour cell invasion or tumour development, including GPC1, SDC1, HSPG2, CSPG4, PTPRZ1, CD44 and VCAN [12, 16, 99, 123-128]. As noted previously, GPC1, a GPI-linked HSPG, can regulate the signalling of a number of heparin-binding ligands, including FGF and VEGF [99, 129, 130]. In brain, breast and pancreatic cancer, increased expression of GPC1 on tumour cells or on tumour-associated endothelial cells is associated with alterations in RTK signalling and promotes tumourigenesis [16, 97, 99, 130, 131].
Of the membrane-associated proteoglycans CSPG4/NG2, PTPRZ1 and CD44 were the most up-regulated, and all three have been implicated in gliomagenesis [124, 132, 133]. CSPG4/NG2 has both cell autonomous and noncell autonomous functions in GBM and its expression has been associated with shorter survival [13, 114, 132]. PTPRZ1, expressed in the normal brain, can be modified by CS/DS chains and exists as three isoforms, full-length PTPRZ1-A, the truncated PTPRZ1-B and a secreted isoform comprising the ectodomain, known as phosphacan. In embryonic oligodendrocyte progenitors, PTPRZ1 acts to maintain self-renewal and suppress differentiation and this could have important implications in cancer . High CD44 expression is also common in GBM and it is one of a panel of genes used to define a subset of human GBM with particularly poor survival [20, 118].
Of the secreted proteoglycans VCAN was the most up-regulated. A member of the lectican family, VCAN contains an amino-terminal hyaluronan-binding domain, a C-terminal globular lectin domain that interacts with GAGs and other proteins, and a domain that can be modified with CS. VCAN is often up-regulated in response to brain injury and, in GBM, VCAN can promote transforming growth factor-β2 induced tumour cell migration [89, 127]. Consistent with the pleiotropic role for proteoglycans in cell signalling, alterations in VCAN expression are not exclusively associated with tumour promotion. In malignant melanoma, a VCAN splice variant interferes with the interaction of EGFR and CD44/ERBB2 and confers decreased tumour growth . Lumican, a small leucine-rich proteoglycan, was also up-regulated in GBM. Although alterations in lumican expression in cancer are common, their role may be tissue specific and include anti-tumour effects [95, 136, 137]. Although many proteoglycans are up-regulated in GBM, some are down-regulated, such as glypican-5. Interestingly, a single nucleotide polymorphism, associated with the down-regulation of glypican-5, was recently linked with an increased susceptibility to lung cancer in nonsmokers . Overall, both HSPG and CSPG core proteins are largely up-regulated in GBM relative to normal brain (Fig. 2A).
By contrast, the HS and CS biosynthetic enzymes appear to be differentially regulated (Fig. 2B). Although CS biosynthetic enzyme expression is predominantly up-regulated in GBM, HS biosynthetic enzyme expression, including HS6ST1–3, is predominantly down-regulated, with the striking exception of HS3ST3a1. Increased HS3ST3a1 expression has previously been reported in GBM cell lines . Because the sulfation status of HSPGs is a critical determinant for ligand interactions, these data suggest that increased 3O-sulfation levels and relatively low 6O-sulfation levels may be important in GBM. Moreover, the SULFs, which act to decrease HS 6O-sulfation levels in the extracellular environment, are up-regulated in GBM (Fig. 2B) [9, 11].
In addition, the extracellular HS modifying enzyme HPSE and the membrane integral biosynthetic enzyme HAS2 are up-regulated in GBM. As discussed above, HPSE is up-regulated in a number of cancers and its expression is associated with increased oncogenic signalling and the promotion of inflammation [100, 117]. In mammary epithelial cells, HAS2 regulates the epithelial–mesenchymal transition induced by transforming growth factor-β and this appears to be independent of extracellular hyaluronan . Increased HAS2 activity and HA production have been associated with malignant transformation and increased RTK activity [140-142].
GBM subtype-specific alterations in proteoglycan expression
GBM is a heterogeneous disease associated with multiple oncogenic alterations. The two most commonly altered RTKs in GBM are EGFR and PDGFRA [18, 143, 144]. A number of strategies have been proposed to subgroup GBM [19, 20, 118-121], and a common strategy proposed by Verhaak et al.  is based on gene expression analysis. Because the oncogenic pathways assumed to drive disease appear to differ across tumour subtypes, we hypothesized that proteoglycan alterations associated with these oncogenic pathways may be different between the tumour subtypes. Indeed, previous studies have demonstrated high expression of SULF2, which can promote PDGFRA signalling, in the proneural GBM subtype characterized by alterations in this signalling pathway .
The gene expression of proteoglycan and proteoglycan-modifying enzymes in 170 previously subtyped GBM  was examined. Consistent with the diverse role for proteoglycans in gliomagenesis, expression of multiple genes was altered in a subtype-specific manner. Representative examples are illustrated in Fig. 3, where the normalized expression score for each gene is compared across tumour subtypes.
For many proteoglycans and proteoglycan-modifying enzymes, we identified selective and significant up-regulation in specific GBM subtypes (Fig. 3A). SDC1 expression was specifically increased in the mesenchymal GBM subtype, which is characterized by high expression of genes in the tumour necrosis factor superfamily pathway, the nuclear factor-κB (NF-kB) pathway, and increased expression of inflammatory related genes [18, 145]. Interestingly, in a glioma cell line, SDC1 expression has been shown to be dependent on NF-κB activation . Taken together, these data suggest that cell surface SDC1 expression levels may reflect NF-κB activation, and it can be speculated that SDC1 function may be particularly relevant in GBM with NF-κB activation. In multiple myeloma, SDC1 can promote tumour growth and tumour angiogenesis, and high levels of shed SDC1 in the blood represent a poor prognostic indicator . SDC1 interactions with integrins can promote tumour invasion, as has been shown in breast cancer , and disruption of this interaction decreases tumour growth and angiogenesis . SDC1 can also modulate the inflammatory response via binding to chemokines, cytokines and integrins and, in models of allergic disease, shed SDC1 can attenuate inflammation [149-151]. Taken together, this suggests that SDC1 could modulate multiple tumourigenic pathways in GBM and this may be of particular relevance in mesenchymal GBM.
The classical and the proneural GBM subtype are characterized by high-level amplification of EGFR and alterations in the PDGFRA signalling pathway, respectively [18, 120]. CSPG4/NG2 is significantly elevated in both the classical and proneural subtypes relative to the remaining subtypes (Fig. 3A), and CSPG4/NG2 has been implicated in promoting both of these RTK signalling pathways [12, 75]. As reviewed above, CSPG4/NG2 in glial progenitor cells co-localizes with PDGFRA and EGFR and it is required for the normal signalling response to ligand [12, 15, 75, 152]. CSPG4/NG2 is also significantly up-regulated in low-grade oligodendroglioma , suggesting potential functions early in gliomagenesis. It is intriguing to speculate that sequential up-regulation of CSPG4/NG2 and EGFR or PDGFRα may further tumour prog-ression.
Our analysis also revealed interesting GBM subtype differences in the expression of the CSPG lecticans, including ACAN and BCAN (Fig. 3A), both of which are normally expressed in the brain and bind HA [153, 154]. ACAN was significantly increased in the proneural GBM subtype relative to the other subtypes. BCAN expression was also increased in the proneural subtype and decreased in the mesenchymal GBM subtype relative to the other subtypes. Striking intertumoural differences in BCAN expression have been previously reported in GBM, including up-regulation of BCAN in proneural GBM and selective down-regulation of BCAN in mesenchymal GBM [18, 20]. Indeed, down-regulation of BCAN is a signature alteration of the mesenchymal subtype used in our analysis . Similar to other proteoglycans, BCAN can be enzymatically cleaved, and its shed hyaluronan-binding domain can promote glioma cell invasion and EGFR signalling [155-159]. By contrast, VCAN was not differentially expressed between tumour subtypes, suggesting a function independent of tumour subtype (data not shown).
The extracellular sulfatases, SULF1 and SULF2, also demonstrate subtype-specific differences in expression (Fig. 3B). SULF1 expression is significantly decreased in classical GBM relative to the other subtypes. By contrast, SULF2 expression is significantly elevated in proneural GBM, which is characterized by alterations in PDGFRA signalling. In recent studies, we identified SULF2 as an important determinant of PDGFRA signalling in GBM . SULF1 and SULF2 are both extracellular sulfatases with similar reported substrate specificities; however, they exhibit partially non-overlapping tissue and cellular expression patterns [55, 160]. Because both SULFs can also undergo alternative splicing, a combination of differences in cellular localization and isoform heterogeneity may contribute to differences in SULF function in vivo .
The GBM subtypes in Fig. 3 are defined based on gene expression profiles not specific oncogenic alterations. Although these subtypes generally correspond to tumours with different patterns of RTK signalling , we wanted to determine whether specific oncogenic events are associated with alterations in proteoglycan or proteoglycan-modifying enzymes. Because EGFR and PDGFRA are the most common RTKs altered in GBM, we stratified tumours based on their amplification status and compared proteoglycan expression.
The expression of SULF1 and SULF2 was strikingly different in tumours with amplification of different RTKs (Fig. 4). SULF1 expression was significantly decreased in GBM with EGFR amplification relative to GBM without amplification, and this association held true at the mRNA expression level because SULF1 expression was significantly negatively correlated with EGFR expression (r = −0.3554) (Fig. 4A,B). By contrast, SULF2 expression was significantly elevated in PDGFRA-amplified GBM relative to tumours without amplification (Fig. 4C). SULF2 expression was also positively correlated with PDGFRα expression across all GBM tumours (r = 0.2115) (Fig. 4D). These data suggest that SULF2 may be particularly relevant in tumours with PDGFRA pathway activation, consistent with our previous studies in human glioma and in a murine model for glioma, in which the loss of SULF2 decreases PDGFRA signalling, decreases proliferation and prolongs mouse survival .
We performed a similar analysis with the proteoglycan core proteins and two representative examples are illustrated (Fig. 5). CSPG4/NG2 expression was significantly elevated in both PDGFRA amplified and EGFR amplified tumours relative to non-amplified tumours. These data are consistent with our subtype analysis (Fig. 3A), and suggest that joint alterations in CSPG4/NG2 and EGFR or PDGFRA may contribute to gliomagenesis. ACAN expression was significantly decreased in EGFR-amplified tumours but significantly increased in PDGFRA amplified tumours relative to non-amplified tumours, suggesting potential tumour-specific differences in ACAN function (Fig. 5).
By stratifying tumours based on RTK amplification status, it is clear that the SULFs may have different functions in GBM subgroups, and this consideration is important for future studies on SULF function in GBM.
Tools for dissecting proteoglycan function
Proteoglycans represent an excellent potential therapeutic target in GBM; however, determination of the mechanisms by which they promote disease is critical. This is especially important given the subtype-specific differences in proteoglycan expression and possibly function. Models for GBM that reflect the heterogeneity of the human disease are required for the study of proteoglycan function and for conducting preclinical studies of potential therapeutic agents.
Human GBM xenografts, isolated from primary tumours and sustained in vivo in mice, provide a robust model system for studying human disease. When passaged in mice, human xenografts are molecularly and phenotypically stable over time, they retain much of their genomic and transcriptional heterogeneity, and they can be transcriptionally stratified into GBM subtypes similar to human GBM [18, 162]. Indeed, using human GBM xenografts, we demonstrate diverse proteoglycan and SULF2 expression profiles (Fig. 6). In addition to human xenografts, murine neural stem cell-derived models for high-grade astrocytoma provide a genetically tractable, immunocompetent system for studying the role of proteoglycan alterations in disease [11, 163]. Although genetically engineered mouse models for GBM will also be useful, the ability to manipulate the genetics of the tumour and the non-neoplastic cells independently will be critical for dissecting the contribution of each to the oncogenic niche.
Potential clinical applications
As extracellular proteins, HSPGs and the extracellular enzymes that modify them, such as SULF2 and HPSE, are amenable to therapeutic targeting. Heparan sulfate mimetics, comprising highly sulfated oligosaccharides, inhibit both SULF and HPSE functions, and sequester HS-binding ligands, making them attractive candidates for GBM therapy across tumour subtypes [164-166]. In preclinical studies, heparan mimetics have effectively targeted multiple HSPG-dependent phenotypes, as indicated by their ability to decrease tumour growth, invasion, metastasis and angiogenesis [167, 168]. A phase II clinical trial for a HS mimetic in recurrent hepatocellular carcinoma has demonstrated safety and preliminary efficacy , and recent pre-clinical studies of a new rationally engineered HS mimetic, M402, suggest its potential as a therapeutic agent . Indeed, M402 is now in a Phase I/II clinical trial for metastatic pancreatic cancer [http://www.clinicaltrials.gov; NCT01621243]. In addition to potential direct anti-tumour effects, therapeutic targeting of proteoglycans could also modulate the tumour-associated immune response. This may be particularly useful in GBM because tumour-associated macrophages are abundant and are likely to contribute to tumour growth [145, 170]. Indeed, therapeutic targeting of HSPGs with heparan sulfate mimetics decreased myeloid-derived suppressor cell levels in a murine mammary carcinoma model . As a primary constituent of the brain extracellular environment, proteoglycans may impede diffusion of therapeutic agents in the brain and their disruption may improve therapeutic delivery to cancer cells . Consistent with this concept, enzymatic removal of CS improved the delivery of an oncolytic virus and increased its anti-tumour efficacy in an intracranial tumour model . Importantly, this treatment did not increase tumour cell invasion. Although heparan sulfate mimetics have not yet been tested in GBM, our data ( and Figures 1–6) suggest that HS mimetics or more selective inhibitors of proteoglycan function may have anti-tumour efficacy in GBM. Proteoglycan function is diverse and selective inhibitors may be essential, as suggested by a recent study in which single-chain variable fragment antibodies directed against endothelial HS elicited a pro-angiogenic response in primary human endothelial cells . This illustrates the critical importance of conducting pre-clinical studies in vivo aiming to test the therapeutic efficacy and safety of agents that target proteoglycans in human cancer.
Shed or secreted proteoglycans and their extracellular modifying enzymes can often be detected in the blood [174, 175]. Because these are often altered in cancer, changes in their blood levels may be useful as biomarkers of disease. In lung cancer patients, plasma levels of SDC1 are significantly higher than in controls . Moreover, in blood-based proteomic screens for cancer, biomarkers extracellular and cell-surface proteins, including proteoglycans and growth factors, are often identified . Proteoglycans and their modifying enzymes are present extracellularly, promote oncogenic signalling and are altered in GBM. If these alterations can be detected in the blood of patients, then they may be useful as tumour biomarkers.
Proteoglycans are a major constituent of the brain extracellular environment and regulate cell signalling and cell migration. In the present review, we summarize the data supporting a functional role for proteoglycans in brain cancer and demonstrate that proteoglycans and their modifying enzymes are altered in human GBM. Furthermore, we associate specific proteoglycan alterations with alterations in EGFR or PDGFRA signalling pathways. A mechanistic understanding of proteoglycan function in oncogenic signalling and tumour-microenvironment interactions in GBM is critical, and will likely lead to the identification of novel tumour biomarkers and druggable therapeutic targets.
We thank Noemi Andor for her helpful advice. We apologize to the many investigators whose articles we did not cite as a result of space constraints. This work was supported by the National Institutes of Health (U01CA168878 to J.J.P.; R01 NS081117 to J.J.P.; R01 CA164746 to C.P.; and R01 NS080619 to C.D.J. and C.P.). The authors declare that there are no conflicts of interest.