Endothelial transdifferentiation of glioma stem cells: a literature review

Abstract

Endothelial transdifferentiation represents a multifaceted process wherein glioma stem cells (GSCs) gradually adopt endothelial characteristics, marked by the expression of endothelial markers (CD31, CD34) and functional traits, while concurrently relinquishing their stem-like properties. This phenomenon is heterogenous in glioblastoma (GBM) samples, but holds importance in terms of prognosis. Typically occurring within hypoxic environments, particularly in perinecrotic regions, endothelial transdifferentiation is influenced by the secretome of neighboring cells, which orchestrates the activation of various signaling pathways including Notch during endothelial lineage commitment, PI3K/AKT, Wnt/β-catenin and epithelial-mesenchymal transition (EMT) during both commitment and maturation. Initially, GSCs organize into vascular-like channels resembling vasculogenic mimicry and express CD144; however, this signature diminishes as endothelial maturation progresses. GSC-derived endothelial cells (ECs) eventually integrate with normal ECs from the tumor periphery, yielding a mosaic pattern. Endothelial transdifferentiation plays a role in response to standard treatments such as temozolomide chemotherapy and radiotherapy.

Introduction

Glioblastoma (GBM) is the most frequent primary malignant tumor of the central nervous system. Despite the latest advancements in translational research, GBM still carries a poor prognosis. Surgical resection combined with radiation therapy (RT) offers a median overall survival (mOS) of 12 months, and the addition of temozolomide (TMZ), the standard chemotherapeutic drug, extends mOS to just 14.6 months [35]. Therapeutic alternatives such as anti-vascular endothelial growth factor (VEGF) antibodies, alkylating agents (lomustine, carmustine) or tumor treatment fields (strategy that utilizes alternating electric fields to interfere with mitosis) may be used in recurrent GBM, but with modest benefits [42]. In this context, the mechanisms that contribute to recurrence and resistance are fundamental to our understanding of GBM biology. Over the last decade, the tumor heterogeneity received extended attention. GBM is an intricate ecosystem in which tumor clones recruit non-tumor cells (astrocytes, oligodendrocytes, microglia, macrophages, endothelial cells etc.) and convert them into a pro-tumor phenotype [18]. Within this heterogeneous environment, glioma stem cells (GSCs) represent a particular subpopulation that is widely recognized for its pivotal role in resistance and recurrence.

Two models have been proposed to explain the origin of glioma stem cells (GSCs). Firstly, GBM cells can dedifferentiate and reprogram to the pluripotent state of GSCs under cellular insults (hypoxia, chemotherapy, radiotherapy) even in the presence of differentiation conditions [74]. Secondly, neural stem cells (NSCs) in the subventricular zone (SVZ) acquire driver mutations and transform into GSCs that will eventually migrate and develop into GBM tumors [75]. In addition to these, stemness and differentiation are just the extremities of a continuous gradient of cellular states [103], where GSCs and non-GSCs are interconvertible states among GBM cells [6].

Similar to NSCs residing in specialized niches in the SVZ, GSCs nest in particular areas and often display a metabolic plasticity in accordance with the conditions imposed by the surrounding microenvironment. For instance, in the perivascular niche, GSCs resemble a proneural phenotype, highly proliferative and dependent on oxidative phosphorylation metabolism, whereas in hypoxic conditions such as the perinecrotic areas, GSCs recapitulate a mesenchymal phenotype with a glycolytic metabolism and a quiescent, but migratory behavior [16, 64, 120]. While some authors define the perivascular niche as the direct contact between GSCs and endothelial cells (ECs) [114], this should exclude larger transport vessels that cannot function as exchange vessels and consequently create the same hypoxic conditions as in the perinecrotic niche [1]. All in all, even these states are interchangeable without genetic mutations, but rather epigenetic, transcriptional and post-transcriptional reprograming under external influences [46, 64].

Hence, defining GSCs proves a challenging assignment. The mandatory features of GSCs include self-renewal, persistent proliferation and tumor initiation upon transplantation. However, GSCs can also exhibit differences in multipotency, frequency within the tumor and stemness-associated markers (Table 1) [73]. The aforementioned represent a molecular repertoire that altogether orchestrates the differentiation-stemness switch and the interactions within the microenvironment. CD133 (Prominin-1) is a well-known stemness marker crucial for neurodevelopment. It activates the Wnt/β-catenin signaling pathway, promoting proliferation and differentiation towards the neural lineage [10]. In GBM, CD133 is linked not only to the Wnt/β-catenin pathway but also to the Notch, Sonic Hedgehog (SHH), c-Jun N-terminal kinase (JNK), and phosphatidylinositol 3-kinase (PI3K)/AKT pathways [51]. Nonetheless, CD133- GSCs were also described as a distinct subpopulation [12]. Other markers include intermediate filament proteins such as nestin, vimentin, and glial fibrillary acidic protein (GFAP) [11]. Nestin is a cytoskeletal protein expressed not only in neuroepithelial stem cells, but also in various non-neural stem and progenitor cells, including endothelial progenitor cells (EPCs); however, although it is an unreliable marker for distinguishing between stem cells, nestin is not expressed in mature ECs [111]. Vimentin is the primary intermediate filament in oligodendrocyte precursor cells (OPCs) and immature astrocytes. In mature astrocytes, vimentin is replaced by GFAP as the main cytoskeletal protein, though it can be re-expressed in reactive astrocytes. Another frequently used stemness marker is SRY (sex determining region Y)-box 2 (SOX2), a key transcription factor that contributes to the neuroepithelial fate of stem cells and to the epithelial-mesenchymal transition (EMT) in cancers [113]. Both SOX2 and octamer transcription factor 4 (OCT4) are essential for tumorigenicity and TMZ resistance [92]. Additionally, ATP-binding cassette sub-family G member 2 (ABCG2) is highly expressed in GSCs, while normal astrocytes lack its expression [21]. This protein is involved in self-renewal and the acquisition of stemness markers [141].

Table 1.

Stemness markers

Marker	Definition	Cell populations	Downstream targets and functions	References
CD133	Pentaspan transmembrane glycoprotein predominantly localized to lipid rafts within the plasma membrane	HSPCs EPCs NSCs GSCs	CD133 phosphorylation → PI3K/AKT activation → GSK-3/β-catenin → self-renewal, proliferation, survival; Src phosphorylation → EMT → migration; Stem-cell-adjacent cell interactions; Regulation of microvilli structure and dynamics;	[93] [86] [143]
Nestin	Class VI intermediate filament protein involved in cytoskeletal organization	HSPCs EPCs NSCs GSCs	Stabilization of βII-tubulin → G2/M progression → proliferation, survival; Regulation of actin polymerization, looser interaction with the ECM → migration; PI3K/AKT activation → self-renewal, proliferation, survival;	[138] [61] [80]
Vimentin	Class III intermediate filament protein involved in cytoskeletal organization	HSPCs EPCs NSCs GSCs	Cell-ECM adhesions: interaction with integrins → Ccd42 activation → cytoskeletal reorganization → motility; Vimentin phosphorylation → Jagged1 interaction → Notch activation → vascular wall remodeling; Regulation of TGF-β1/Slug/EMT signaling → ECM remodeling; MT1-MMP translocation to plasma membrane → ECM remodeling; ERK phosphorylation → proliferation;	[102]
GFAP	Class III intermediate filament protein	NSCs GSCs	Focal adhesion, cell: ECM interaction; Proliferation;	[91]
SOX2	High-mobility group transcription factor involved in embryonic development and cell fate determination	HSPCs EPCs NSCs GSCs	Transcriptional activator and repressor → pluripotency; Wnt/β-catenin activation → self-renewal, proliferation, survival; Notch regulation → self-renewal; PI3K/AKT/mTOR activation → self-renewal, proliferation, survival;	[70] [39] [83]
OCT4	Homeodomain transcription factor of the Pit-Oct-Unc family	HSPCs EPCs NSCs GSCs	Maintenance of stemness; Control of pluripotency; Self-renewal, proliferation;	[79, 92, 94]
ABCG2	ATP-binding cassette (ABC) transporter	HSPCs EPCs NSCs GSCs	Maintenance of undifferentiated state; Efflux pump involved in cellular protection for toxic substances, multidrug resistance;	[62]

ABCG2 ATP-binding cassette subfamily G member 2, Cdc42 cell division control protein 42 homolog, ECM extracellular matrix, EMT epithelial-to-mesenchymal transition, EPCs endothelial progenitor cells, ERK extracellular signal-regulated kinase, GFAP glial fibrillary acidic protein, GSCs glioblastoma stem cells, GSK-3 glycogen synthase kinase 3, HSPCs hematopoietic stem and progenitor cells, MT1-MMP membrane type 1 matrix metalloproteinase, mTOR mammalian target of rapamycin, NSCs neural stem cells, OCT4 octamer-binding transcription factor 4, PI3K phosphoinositide 3-kinase, SOX2 SRY-box transcription factor 2, TGF-β transforming growth factor beta, VEGFR2 vascular endothelial growth factor receptor 2

Many of these markers are also critical for NSCs, reinforcing the hypothesis that GSCs originate in the SVZ. Differentiating between GSCs, NSCs, OPCs, and EPCs proves to be laborious due to the major overlap between the stemness markers in these populations. For example, GSCs and NSCs/OPCs cannot be distinguished based solely on stemness marker expression, as they share similar markers. Their differences are functional, driven by genetic alterations. Compared to NSCs, GSCs sustain proliferative signals (EGFR and PDGFRα overexpression), resist growth suppression (TP53 mutations), evade apoptosis (PTEN mutations), and exhibit limitless replicative potential (TERT promoter mutations) [9].

Stemness markers and their localization within specific niches (perinecrotic, perivascular) are commonly employed for recognition of GSCs in vivo, although their identification should be further validated by the presence of genetic abnormalities characteristic of GBM cells. Isolation of GSCs by flow cytometry based on surface markers does not ensure the purity of the GSC population and may overlook GSCs with tumorigenic potential that do not express the putative markers [89]. These surface markers mediate interactions within the TME, and once dissociated from the tumor microenvironment, GSCs start to lose their informational content [100]. After serum treatment, GSCs differentiate towards a neural lineage by up-regulating β-III tubulin (TUJ1), microtubule-associated protein 2 (MAP2) and GFAP, while downregulating CD133 [27]. For this reason, GSCs are usually cultured in media enriched in epithelial growth factor (EGF) and basic fibroblast growth factor (bFGF) to inhibit their spontaneous differentiation. However, this can favor the selection of subpopulations with a biased receptor profile, which likely restricts the original tumor’s heterogeneity [53, 131]. Additionally, long-term passaged GSCs are susceptible to genetic drift and selection of proliferative populations, leading to variations in phenotypic marker expression and behavior [131]. It is very difficult to reproduce the complexity of the niche conditions in which bone morphogenetic proteins (BMPs), Wnt, Notch and SHH pathways maintain the behavior of the stem cell pool. Standard in vitro studies of GSCs neglect the interaction with GBM cells, ECs, immune cells, ECM components, and fail to reproduce nutrient and hypoxic gradients [147]. Thus, in vitro data should not be over-interpreted in terms of phenotypic marker expression and cell behavior.

GSCs-derived ECs and host ECs can collaborate to influence the histological features of GBM vasculature. Histologically, GBM is characterized by hypercellularity, nuclear atypia and palisading necrosis [32]. The vascular network is composed of disorganized, glomeruloid and tortuous blood vessels, which exhibit high permeability, disruptions in the basement membrane, and abnormal pericyte coverage [29]. Poor perfusion caused by this abnormal vasculature leads to both hypoxia and reduced drug delivery. While hypoxia drives the development of treatment-resistant phenotypes, suboptimal drug concentrations further limit treatment efficacy [151]. Several mechanisms contribute to GBM vascularization. Firstly, tumor cells can migrate along the blood vessels, surround the blood–brain barrier (BBB) without detaching the pericytes, or perturb the BBB and attach to ECs, thus incorporating the pre-existing vasculature into the tumor [72]. In addition to this vessel co-option mechanism, GBM utilizes sprouting angiogenesis, where branches of the pre-existing vasculature grow towards the tumor mass [110]. During this process, the loss of pericyte-EC contact destabilizes the basement membrane and prompts ECs to adopt a mesenchymal phenotype [134]. These migratory and proteolytic ECs (‘tip’ cells) navigate the vascular path, while proliferative ECs (‘stalk’ cells) assemble the neovessel. Vasculogenesis is similar to sprouting angiogenesis, but involves blood vessel formation by bone marrow-derived circulating endothelial progenitors (EPCs) [50]. Another mechanism is vessel intussusception where pre-existing vessels enlarge and bifurcate [28]. Lastly, vasculogenic mimicry (VM) involves GBM cells rearranging to form functional, vascular-like channels that eventually connect to the existing network.

Despite the general view that ECs originate from the mesoderm [34], NSCs were demonstrated to differentiate towards an endothelial phenotype both in vitro and in vivo [144]. This important observation led to the investigation of a similar behavior in GSCs, known as endothelial transdifferentiation (endo-transdifferentiation). Since transdifferentiation is usually defined as an irreversible conversion between two differentiated cells [118], GSCs constitute an intermediary phase between GBM cells and GBM-derived ECs. One of the main stimuli for GBM cell dedifferentiation is hypoxia, which induces CD133, CD15, Nestin, SOX2, Oct4 and Nanog in GBM cells [137]. Additionally, CD133- GBMs cultured in differentiation media acquire CD133 expression following TMZ treatment [74]. Similar to hypoxia, TMZ increases the expression of both HIF-1α and HIF-2α. HIF knockdown inhibits the interconversion between non-stem GBM cells and GSCs [74]. Radiotherapy also contributes to dedifferentiation. Patient-derived differentiated GBM cells exposed to subtoxic RT doses (3 Gy) upregulate stemness markers (CD133, Nanog, Olig2, SHH, EZH2) via survivin, an anti-apoptotic protein. Notably, this effect was observed only in cells cultured in stem cell medium, but not in differentiation medium, arguing that RT alone is insufficient for dedifferentiation [31]. GSCs are localized within specialized niches, such as the perinecrotic and perivascular niches. In the perinecrotic area, hypoxia promotes generation of proangiogenic factors, promoting GSC involvement in neovessel formation. In the perivascular niche, both hypoxia and direct interactions with ECs may enhance GSC recruitment into the vasculature. Although endothelial transdifferentiation and VM are distinct processes, they share some similar pathogenetic mechanisms, leading to occasional misuse of the terms. As recently illustrated by Maddison et al., endothelial transdifferentiation should be demonstrated by the presence of endothelial markers (CD31, CD34) [85].

In general, ECs can be identified based on their cobblestone morphology in monolayer cultures and the presence of several endothelial markers (Table 2) including von Willebrand factor (vWF), endothelial nitric oxide synthase (eNOS) or CD31 (PECAM-1) [45]. CD34 is also found on the surface of mature ECs as well as on EPCs and hematopoietic cells [45]. Another commonly used marker for ECs is CD144 (CDH5/VE-cadherin), which is located at intercellular junctions and regulates vascular permeability, proliferation, and modulation of the VEGFR2 functions [127]. While VEGFR2 (KDR/Flk-1) and Tie receptors (Tie1, Tie2) are critical for the endothelial homeostasis, they are not EC-specific, being also expressed by GBM cells [37] and pericytes [128]. Endoglin (CD105), a receptor of the transforming growth factor beta (TGF-β) family, is expressed by proliferating, active ECs that form disorganized vessels, as seen in GBM [101]. Complementary to these surface markers, ECs display notable functions such as tube formation or uptake of acetylated-low density lipoprotein (acLDL).

Table 2.

Molecular markers of the endothelial lineage

Marker	Cells	Location	Function and known downstream pathways	References
CD34	HSPCs EPCs Mature ECs (microvascular)	Resting ECs: apical Activated ECs: EC-EC junction, filopodia	Proliferation and blocking of differentiation in HSPCs Adhesion of HSPCs to BM endothelium via E/P-selectin Maintaining quiescence in mature ECs Adhesion of naïve lymphocytes to mature ECs via L-selectin Resting ECs: apical localization → prevents opposing ECs adhesion Activated ECs: EC-EC junctions → permeability Cytoskeletal remodeling (filopodia) → migration	[96, 108] [122] [3] [59]
CD31	HSPCs EPCs Mature ECs Leukocytes, Platelets	EC-EC junction	EC-EC junction → barrier function Homophillic interaction with leukocytes → transmigration EC-ECM interactions CD31/SHP-2 complex dephosphorylates FAK, paxillin → EC migration CD31/SHP-2 complex dephosphorylates β-catenin → β-catenin signaling; formation of adherens junctional complexes	[108] [26] [13] [160]
CD144	EPCs Mature ECs	EC-EC junction	EC-EC junction → barrier function Binds and maintains VEGFR2/FGFR1 dephosphorylated → quiescence, survival Binds TGF-βR2 → activates ALK5/Smad2/3 → inhibits migration, proliferation Connects to the actin cytoskeleton via β-catenin	[15] [43] [129]
CD105	HSPCs EPCs (minor subset) Mature ECs	Cell surface	Inhibits TGF-β1/Smad3 → proliferation	[108] [48]
VEGFR2	ECs EPCs	Cell surface	Very strong mitotic signals cPKC/RAF/MEK/ERK1/2 → migration/proliferation SRC/FAK/Paxillin → Permeability CD144 internalization → Permeability PI3K/AKT/mTOR → survival, vasodilation (via eNOS)	[108] [123]
VEGFR1	HSPCs EPCs Mature ECs	Cell surface	Higher affinity for VEGF, but very weak mitotic signal	[108]
Tie2	HSPCs EPCs Mature ECs M2 monocytes	Resting ECs: EC-EC junction Migrating EC: EC-ECM	Ang1 interaction → AKT activation → survival, quiescence, barrier integrity Inflamed endothelium: Ang2 is antagonist for Tie2 → actin stress fibers → destabilize endothelial monolayers Non-inflamed endothelium: Ang2 is weak agonist for Tie2	[112]
Tie1	HSPCs EPCs Mature ECs	Resting ECs: EC-EC junction Migrating EC: EC-ECM	Orphan receptor (lacks a known ligand) Co-receptor for Tie2: required for full activation of Tie2 by Ang proteins	[112] [67]
vWF	EPCs Mature ECs Megakaryocytes	Weibel-Palade bodies	Binds fVIII, GP Ib, GP IIb/IIIa, heparin, collagen → hemostasis	[108]

ALK5 activin receptor-like kinase 5, Ang angiopoietin, BM bone marrow, cPKC conventional protein kinase C, EC endothelial cell, ECM extracellular matrix, EPC endothelial progenitor cell, ERK1/2 extracellular signal-regulated kinase 1/2 FAK, focal adhesion kinase, fVIII, coagulation factor VIII, FGFR1 fibroblast growth factor receptor 1, GP glycoprotein, HSPC hematopoietic stem and progenitor cell, MEK mitogen-activated protein kinase kinase, mTOR, mammalian target of rapamycin, PI3K phosphoinositide 3-kinase, RAF, rapidly accelerated fibrosarcoma kinase, SHP-2 Src homology region 2-containing protein tyrosine phosphatase-2, TGF-β1 transforming growth factor beta 1, TGF-βR2 transforming growth factor beta receptor 2, Tie1 tyrosine kinase with immunoglobulin-like and EGF-like domains 1, Tie2 TEK receptor tyrosine kinase, VEGFR vascular endothelial growth factor receptor, vWF von Willebrand factor

The concomitant presence of stemness and endothelial antigens (Table 3) is suggestive of transdifferentiation, but can be misleading due to the co-expression of CD133, CD34 and VEGFR2 in EPCs [157]. Typically, early EPCs in the bone marrow or shortly after entering the circulation are CD133 + /CD34 + /VEGFR2 + , while circulating EPCs are CD34 + /CD31 + /CD144 + /VEGFR2 + , start expressing vWF and may lose CD133 [55]. In the mouse lung, a population of resident EPCs identified as CD45-/CD31 + /VEGFR2- was able to differentiate to CD45-/CD31 + /VEGFR2 + cells and then to ECs, suggesting that VEGFR‐2 marks a later stage of endothelial progenitor development [148]. During neovessel formation in adults, Patel et al. identified a subpopulation of EPCs as quiescent CD31-low/VEGFR2-/CD105- cells and definitive differentiated ECs as CD31 + /VEGFR2 + /CD105 + cells [104]. Regarding the expression of other stemness markers, vimentin plays a crucial role in EPCs, as its knockout in embryonic stem cells (ESCs) impairs spontaneous endothelial differentiation by disrupting the sequential acquisition of key markers. These include the early mesodermal marker Brachy-T, the lateral plate mesodermal marker VEGFR2, and endothelial-specific markers such as Tie2, CD31, and CD144 [14]. Endothelial nestin expression is observed in undifferentiated EPCs before they exit the cell cycle and is downregulated as EPCs mature into ECs in vitro [111]. Isolated circulating mesangioblasts express pluripotency markers Klf4 and c-Myc, low levels of Oct3/4, but lack Sox2; however, lentiviral transduction of Sox2 enhances their ability to differentiate into ECs in vitro [71]. During in vitro endothelial differentiation of ESCs, the endothelial markers CD31, CD144, and VEGFR2, CD144 appeared to coincide with an increase in SOX2 expression, suggesting that SOX2 may be present in EPCs. Finally, ABCG2 is expressed not only in GSCs and NSCs but also in EPCs and mature ECs participating in the formation of the BBB [78, 115]. However, the presence of glioma-specific genetic aberrations in intratumoral ECs (monosomy 10/heterozygous PTEN deletion, polysomy 7/EGFR amplification) can distinguish GSC-derived ECs from their bone marrow-derived counterparts [157].

Table 3.

Immunophenotypes corresponding to GSC-derived ECs

Immunophenotypes	Cell characteristics	References
Nestin + /CD31 +	In vitro: Can be generated by exposing Nestin + /CD133 + cells to hypoxia; confers chemoresistance to neighboring GSCs; can revert to GSC by culturing in stem cell medium Human GBM samples: represent 85% ECs in hypoxic regions; frequently co-expressed with ETV2 + ; high CD31 expression mostly associated with low Nestin expression; low CD31 expression mostly associated with high Nestin expression	[158] [154]
Nestin + /ABCG2 +	Human GBM samples: present in tumor vessel wall	[40]
Nestin + /CD105 +	Human GBM samples: CD105 and Nestin localized in different regions of the same ECs in the tumor area; overlapped immunoreactivity in the peritumor area	[121]
CD133 + /VEGFR2 +	In vitro: co-express EphA2, CD144, laminin5γ2 In vivo: co-implanting with GSCs results in larger and more vascularized tumors; can initiate tumors whose vasculature is mainly derived from tumor cells	[38] [152]
CD133 + /CD31 +	Human GBM samples: localize near SOX2-/CD31 + , SOX2-/CD105 + , CD133-/CD31 + ECs, in glomeruloid tufts, microvascular proliferations, invasive front; high CD31 expression mostly associated with low C133 expression; low CD31 expression mostly associated with high CD31 expression; co-express nestin, lipocalin2, selectin, endothelin 1 and 3, VEGF-C, eNOS	[56] [38] [154]
CD133 + /CD34 +	In vitro: cell population increased by TMZ	[8]
CD133 + /CD144 +	In vitro: Can be generated by exposing CD133 + /CD144- cells to the GBM secretome; cultivation in endothelial medium results in downregulation of CD144 and upregulation of CD105, CD31, CD34, VEGFR2; can revert to GSC by culturing in stem cell medium; cell population increased by TMZ	[139] [56] [8]
GFAP + /CD31 +	In vitro: display tumorigenic activity; cultivation in endothelial medium results in the loss of tumorigenic activity Human GBM samples: represent 18% of the total vascularization; co-express vWF, CD144	[109] [90]
GFAP + /CD34 +	In vitro: can be generated as small population after cultivation of CD133 + GSCs in transdifferentiation medium Human GBM samples: represent 14% of the total vascularization	[40] [90]
SOX2 + /CD31 +	Human GBM samples: tend to localize near SOX2-/CD31 + , SOX2-/CD105 + , CD133-/CD31 + ECs	[56]
SOX2 + /CD34 +	In vitro: cell population increased by TMZ	[8]
SOX2 + /CD105 +	In vitro: cell population increased by TMZ Human GBM samples: tend to localize near SOX2-/CD31 + , SOX2-/CD105 + , CD133-/CD31 + ECs	[56] [8]
CD105 +	In vitro: co-express CD31, VEGFR2, vWF Human GBM samples: similar genetic aberrations (EGFR amplification, Cep7) as the parent tumors; form glomeruloid-like structures	[139]
ABCG2 + /CD34 +	Human GBM samples: appeared in tumor vessel wall in clinical GBM specimens	[40]

ABCG2 ATP-binding cassette sub-family G member 2, CD cluster of differentiation, Cep7 chromosome 7 centromere amplification, EC endothelial cell, EGFR epidermal growth factor receptor, eNOS endothelial nitric oxide synthase, EphA2 ephrin type-A receptor 2, ETV2 ETS variant transcription factor 2, GBM glioblastoma, GFAP glial fibrillary acidic protein, GSC glioblastoma stem cell, Laminin5γ2 laminin subunit gamma 2, SOX2 SRY-box transcription factor 2, TMZ temozolomide, VEGF-C vascular endothelial growth factor C, VEGFR2 vascular endothelial growth factor receptor 2, vWF von Willebrand factor

In this review, we collected data from the literature on the effect of endothelial transdifferentiation in the formation of vascular structures in glioblastoma. We identify the signaling pathways involved in this mechanism and delineate with regards to the phases of endo-transdifferentiation. Additionally, we addressed contradictory findings, highlight gaps in the literature, and suggest possible directions for future research.

Endothelial cells in GBM: tumoral and non-tumoral

The isolation of patient-derived GSCs using a clonogenic assay involves culturing enzymatically dissociated GBM cells in stem cell medium to promote the formation of tumor spheres. These cells are heterogeneous, containing both CD133 + and CD133- populations. While both populations have similar tumorigenic potential, CD133 + GSCs tend to form more vascularized tumors [27]. Differentiated GBM cells cannot generate ECs in culture media optimized for EC, whereas GSCs (CD133 + /CD31- cells) transdifferentiate into functional GBM-derived ECs capable of tube formation, LDL uptake and eNOS expression [109]. Under hypoxic and endothelial conditions, GSCs gain higher expression of endothelial markers [17]. Once acquiring an endothelial signature, CD34^high GSCs generate more proliferative, necrotic, undifferentiated tumors compared to CD34^low/− GSCs. Ultrastructural analysis of these GSC-derived ECs using transmission electron microscopy reveals a flat phenotype with few rough endoplasmic reticula, developed Golgi bodies, numerous lysosomes, mitochondria, glycogenosomes, and pinocytosis vesicles [23, 156]. Endothelial transdifferentiation appears to be irreversible beyond a certain point, as GBM-derived ECs (CD34 + /CD31 + /CD133-) retain their endothelial markers and functions even when cultivated back in stem cell medium [22]. However, Nestin + /CD31 + cells can revert to the GSC state in serum-free culture, suggesting the existence of intermediate states with phenotypic plasticity [158]. These seemingly contradictory findings underscore the need for comprehensive studies to fully decipher the molecular signature of these cells.

In GBM, necrotic areas arise due to vessel occlusions or the imbalance between highly proliferative GBM cells and the ECs bearing a lower proliferation rate [114]. Tumor-derived ECs preferentially form in these hypoxic cores, activating molecular pathways that promote an endothelial fate. In a co-culture system, Zheng et al. demonstrated that GSCs transdifferentiate into Nestin + /CD31 + cells under hypoxia [158]. Moreover, GSC-derived ECs influence neighboring GSCs, resulting in larger spheres and more generations compared to GSCs in mono-culture or when co-cultured with normal ECs. In a mouse model, Ding et al. managed to isolate VEGFR2 + /CD133 + cells and VEGFR2 + /CD133- cells and co-implanting VEGFR2 + /CD133 + ECs with GSCs in a tumor-naïve mouse resulted in larger tumors with more developed vascularization [38]. Therefore, transdifferentiated ECs provide conditions that support the generation of a stem cell pool, which can be further recruited into the vascular network.

Normal ECs seem to play context-dependent roles in the endo-transdifferentiation process. Co-cultivation of GBM cells with human brain microvascular ECs (HBMECs) induces downregulation of VEGF receptor 1 (VEGFR1), Notch4, FGF1, FGF2, TGFβR1, and matrix metalloproteinases (MMPs), indicating that normal ECs might inhibit transdifferentiation [124]. Conversely, co-cultivation of GBM cells with human umbilical vein ECs (HUVECs) resulted in EC-derived vascular branches that later englobed GBM cells into the vascular network [117]. In this setting, while ECs expressed both Tie2 and VEGFR2, the recruited U251 cells did not. Notably, inhibition of Tie2 and VEGFR2 prevents the formation of GBM-derived blood channels, suggesting the role of the VEGF pathway in the recruitment of the neighboring tumor cells by ECs. The observed differences in GBM cell and EC behavior across these studies may be attributed to variations in experimental conditions. Smith et al. cultured KNS42 cells in DMEM/F12 and HBMECs in RPMI-1640, whereas Shaifer et al. co-cultured U251 cells with HUVECs in Matrigel with EGM. Smith et al. proposed that a high EC-to-GBM cell ratio inhibited GBM cell recruitment by reducing selection pressure for neovessel formation [124]. However, the findings of Shaifer et al., who employed a 4:1 HUVEC-to-U251 ratio, contradicted this hypothesis [117]. A more plausible explanation is the difference in culture media, as RPMI-1640 was originally formulated for non-adherent cells, such as hematopoietic cells, whereas Matrigel with EGM more accurately recapitulates the physiological environment required for EC culture.

Carlson et al. found that ECs associated with the tumor are extremely heterogenous, comprising five distinct clusters with particular endothelial markers [20]. Another research group also identified five distinct clusters with varying tumor tissue localization and functional characteristics. Two clusters were found at the tumor periphery, exhibiting either a quiescent or immune-activated profile, while three clusters were located within the tumor core, displaying proangiogenic, intermediate, or immune-activated phenotypes [145]. Understanding how these individual clusters interact with GBM cells, particularly GSCs, is crucial for comprehending the interplay between ECs and GSCs.

Endothelial state: a continuous maturation gradient

Due to the inter- and intratumor heterogeneity, endothelial transdifferentiation varies in extent among patients. For instance, in EGFR-amplified GBM specimens, El Hallani et al. found that ECs carrying EGFR amplifications were present in only 3 out of 10 GBM samples [41]. However, in GBM specimens with mutant TP53, Ricci-Vitiani et al. showed that 20–90% of cells lining the vascular lumen shared the same genetic abnormalities as the tumor cells [109]. As a potential explanation for these differences, TP53 mutations and EGFR amplifications are generally considered mutually exclusive, with studies reporting a limited co-occurrence of these alterations ranging from 20.7% to 42.8% [66, 76]. The circumscribing cells described by Ricci-Vitiani et al. were positive for CD31 and variably co-expressed GFAP (10–76%) [109]. Mei and colleagues reported that CD34 + /GFAP + and CD31 + /GFAP + cells were absent in more than half of GBM samples, but in 70% of positive co-staining specimens, these cells formed vascular structures and comprised 14–18% of the total vascularization [90]. In a CRISPR/Cas9-mediated native GBM model (PTEN, TP53, and NF1 deletion), vessel morphometry showed that the host vessels density and branching points are unchanged throughout time [20]. Instead, CD31 + GBM-derived cells are dynamically involved in the vascularization. Isolating GBM-derived from host-derived ECs showed that the transdifferentiated ECs contribute to less than 1% of the total population. In vitro, patient-derived CD133 + GSCs cultured in transdifferentiation medium predominantly exhibits a CD34 + /GFAP- signature, while the number of CD34 + /GFAP + and CD34-/GFAP + is reduced [40]. This indicates that these endothelial/glial hybrids could be GBM-derived ECs retaining glial markers, but the hypothesis of a common precursor for both endothelial and glial lineages cannot be totally excluded. Data from larger GBM cohorts may offer us a broader image of these endothelial/glial phenotypes and more complex in vitro models could elucidate the dynamics of these potential intermediate states over time.

Strikingly, when taking stemness into consideration, CD31 co-localizes with CD133 and Nestin, but stemness expression is inversely proportional to the endothelial signature [154]. CD144 is expressed in both tumor and vascular cells in most GBM samples, showing scattered and clustered patterns in tumor cells [87]. The endothelial signature progressively increases from differentiated GBM cells (CD133 − /CD144 −) to GSCs (CD133 + /CD144 −) to GSC-derived ECs (CD133 + /CD144 +) and finally to CD133 − /CD144 + cells [56]. Co-cultivation of GBM cells with CD133 + /CD144- cells results in the switch of the latter mentioned towards CD133 + /CD144 + cells, implicating the GBM secretome in the initiation of lineage commitment [139].

In human GBM specimens, Yao et al. observed Periodic acid-Schiff (PAS) + lumina containing red blood cells and lined by two distinct cell populations: CD34 + cells and Nestin + /CD34- cells [152]. A similar pattern of CD34 + and CD34- anastomosis was also noticed by Chen et al. [24]. Another study investigated the presence of ABCG2 in GBM samples and found its co-localization with CD34 or Nestin in the intimal layer of vessels [40]. While ABCG2 + /Nestin + cells lacking CD34 are more likely to be GSCs, ABCG2 + /CD34 + cells cannot be definitively identified as GSC-derived ECs, as this marker signature is also observed in EPCs. All in all, these findings suggests that GSCs group into vascular channels and assemble the basement membrane even before the expression of endothelial markers. Therefore, vasculogenic mimicry could be in fact an intermediary step during transdifferentiation.

Additionally, SOX2 + /CD31 + , CD133 + /CD31 + , and SOX2 + /CD105 + ECs tend to localize near SOX2-/CD31 + , SOX2-/CD105 + , and CD133-/CD31 + ECs [56]. This phenomenon can be explained by GBM-derived ECs recruiting normal ECs, with evidence supporting both the recruitment of regular ECs into the tumor vasculature and the co-option of GSCs by regular ECs [117, 125]. However, another explanation could be the complete loss of stemness in GSC-derived ECs. While SOX2-/CD31 + and CD133-/CD31 + cells could indeed be normal ECs, SOX2-/CD105 + cells are more likely tumor-derived proliferating ECs. This is supported by the fact that, beside the endothelial morphology and signature (CD31, VEGFR2, vWF), the majority of CD105 + cells display similar genetic aberrations (EGFR amplification, Cep7) as the parent tumors [139]. In other tumors, CD105 was demonstrated to be a marker of activated, proliferating ECs, which is why normal brains lack CD105 [33]. Cultivating CD133 + /CD144 + GSCs in endothelial medium results in the downregulation of CD144 and upregulation of CD105, CD31, CD34, and VEGFR2, suggesting that GSCs transdifferentiate towards an endothelial precursor that later on will become an active, proliferating, mature EC [139].

Within the perinecrotic area, both GBM cells and CD31 + ECs express hypoxia-inducible factor (HIF)-1α, and 85% of ECs in the hypoxic regions are Nestin + /CD31 + [158]. In vitro, Zheng et al. revealed that Nestin + GSCs co-localize with Nestin + /CD31 + cells in a hypoxic co-culture system mimicking the perivascular niche [158] and Smith et al. found that CD105 is expressed at the interface between the hypoxic/necrotic core and the proliferative area [124]. Different surface markers are associated with various vascularization behaviors; for example, CD133-/CD144 + cells form capillary-like vascular structures in 3D gel, while primary CD105 + cells form glomeruloid-like structures, reflecting the vascularization patterns observed in GBM samples [139]. In a mouse model, Ding et al. observed that CD133 + /CD31 + cells were located in the glomeruloid tufts, microvascular proliferation areas and the invasive front [38].

In rodent models with human GSC-derived tumors, 70% of the CD31 + cells in the tumor core were GBM-derived, while almost all CD31 + cells in the periphery were host-derived [109]. Additionally, Yao et al. found that 34% of CD31 + cells in GSC-derived tumors were of human origin, compared to almost all CD31 + cells being host-derived in tumors formed by U87 cells [152]. However, in a lentivirus murine model, Soda et al. found that 10–25% of ECs are tumor-derived, participating in up to 37% of the vascularization in the central area and in 2–12% vessels from the periphery. A mosaic pattern was noticed, in which GBM-derived ECs and host ECs cooperate to form vessels. This is supported by Wang et al. who showed that some CD31 + cells and CD144 + /CD133- cells are genotypically normal, therefore representing normal ECs [139]. However, nestin is expressed in both GBM-derived ECs and host-derived ECs [125], probably as a marker of EPCs recruited to participate in the vessel formation. Chen et al. also proved this phenomenon in rats harboring C6 tumors by injecting magnetically-labeled EPCs that formed vessels from the periphery to the center of the tumor [23]. Thus, the vascular network generated by GSCs in hypoxic niches are partially tumoral in origin and connects with the periphery vessels which originated from sprouting angiogenesis or vasculogenesis. Notably, tumoral and non-tumoral ECs do not fuse in vivo [109, 125].

Taken together, these data suggest that under the process of transdifferentiation, GSCs progressively lose their stemness signature (CD133, Nestin, SOX2) while enriching endothelial markers (CD31, CD34, CD105) via an immature EC progenitor state (CD144 +). The co-expression of these markers recapitulates a maturation gradient (Fig. 1). Indeed, gene-expression analysis revealed that the GBM-derived ECs situate between GBM cells and normal ECs [109]. In a GEMM (PTEN, TP53, NF1 deletion), tumor-associated ECs (co-opted vessels) and tumor-derived ECs presented over 400 differentially expressed genes [20]. In GSC-derived ECs, there was an enrichment in migratory genes associated with a more infiltrative and angiogenic phenotype. Additionally, compared to CD133-/CD31 + , isolated CD133 + /CD31 + cells are enriched in nestin, lipocalin2, selectin, endothelin 1 and 3, VEGF-C, and eNOS, possible meditators of GBM tumorigenicity [38].

Fig. 1.

The maturation gradient of GBM-derived ECs. Under the influence of GBM cells, GSC progressively transdifferentiate to an immature EC-like phenotype (CD133 + /CD144 +) that will further maturate into an active EC (CD105 +). The active ECs proliferate and migrate in order to connect the transdifferentiated network with the pre-existing vessels in a mosaic pattern. CD cluster of differentiation, EC endothelial cell, GBM glioblastoma, GSC glioma stem cell, SOX2 SRY (sex determining region Y)-box 2, VEGFR2 vascular endothelial growth factor receptor 2, Wnt5A Wnt family member 5A

In a single cell RNA sequencing of neurosurgical samples (4 GBMs, 3 low grade gliomas, 7 non-malignant), Xie et al. magnetically sorted cells for CD45 + leukocytes depletion and enrichment of CD31 + ECs and PDGFRβ + MCs [146]. Cells were separated in 10 major clusters, including ECs, tumor cells and a small cluster labeled as “proliferating cells”. While the authors concluded that ECs could not be generated by GBM cells, based on the absence of genomic alterations in the EC cluster, there are several arguments supporting that GSC-derived ECs could in fact be identified in the proliferating cluster. Firstly, dot-plot heatmaps indicate that the proliferative cluster possess an intermediate endothelial signature (CLDN5, ABCG2, FLT1, SLC2A1, PECAM1) situated between tumor cells and ECs, while no other cluster presented this property. Secondly, uniform manifold approximation and projection (UMAP) reveals a subpopulation in the proliferative cluster with similar expression of CD144 and CD31 as the EC cluster. Thirdly, 19.5% of the cells in the proliferative cluster presented genomic alterations as the tumor cluster, while in other clusters this phenomenon was nearly absent. Lastly, the signature of the proliferating cluster as revealed by the dot-plot heatmap was marked by upregulation of MKI67, TOP2A, CENPF. Another group isolated 9520 ECs from 19 tissues in healthy domestic pigs and separated 21 EC clusters, with a small EC cluster harboring proliferation genes (CENPF, CENPE, TOP2A, TPX2) [136]. Further research is needed to study this proliferative cluster, as it may contain the GSC-derived ECs.

GBM-derived ECs are capable of tumor initiation given purified CD31 + /CD144 + cells can be grafted and give rise to vascularized anaplastic tumors, but if cultured in endothelial medium, they fully differentiate and lose tumorigenicity [109]. Therefore, GBM-derived ECs are distinct from normal mature ECs, but a complete differentiation could lead to a bona fide EC phenotype, at least unable to generate tumors. What are the precise moment and signature of the GSC-derived ECs when they are irreversibly transformed remains an unanswered question.

Molecular pathways involved in endothelial transdifferentiation

Endothelial transdifferentiation occurs in two phases: endothelial lineage commitment and endothelial maturation. Lineage commitment represents the initial decision of the GSC to adopt an endothelial identity, marked by early endothelial-specific gene expression, while maturation is the functional refinement of the ECs, allowing them to participate in angiogenesis.

In GBM, Wang et al. demonstrated that the conversion of CD133 + /CD144- to CD133 + /CD144 + cells (lineage commitment) is reduced by γ-secretase inhibition or NOTCH1 knockdown, whereas the progression from CD133 + /CD144 + to CD105 + ECs (endothelial maturation) is blocked by anti-VEGF antibodies or VEGFR2 knockdown [139]. Additionally, VEGF stimulation of a SOX2 + /CD133 + patient-derived subpopulation of GSCs results in the formation of a confluent monolayer of spindle-shaped cells that lose SOX2 and CD133 expression, while gaining CD34 immuno-positivity, suggesting a role for VEGF in the final phase of transdifferentiation [90]. Although these cells exhibit stemness markers, the CD144 signature was not studied, leaving open the possibility that the studied SOX2 + /CD133 + GSCs might harbor an intermediary endothelial identity that responds to VEGF and consequently assumes the mature EC signature. Indeed, isolated GSCs from U87 cells can already express several EC markers such as CD144, VEGFR2, laminin5γ2, and Ephrin Receptor A2 (EphA2) [152]. In these GSCs, VEGF stimulation induces Y1175 phosphorylation in addition to the constitutive auto-phosphorylation of VEGFR2 at the Y1059 position, resulting in the activation of mitogen-activated protein kinase (MAPK)/extracellular signal-regulated kinases (ERK1,2) and PI3K/AKT pathways [152]. This behavior is similar to normal ECs, in which Y1059 phosphorylation enhances VEGFR2 kinase activity, while Y1175 phosphorylation is involved in proliferation, migration, and cell permeability [140]. VEGF stimulation also upregulates VEGFR2 and CD144, while VEGFR2 knockdown reduces CD144 expression and tube formation despite VEGF stimulation, suggesting a role for VEGF in the early phase of lineage commitment [152]. Lastly, MAPK/ERK1,2 and PI3K/AKT can be activated by other growth factors, such as insulin-like growth factor 2 (IGF2) or FGF [4, 95]. Smith et al. showed that VEGF inhibition results in a mild reduction of tube formation, while FGFR inhibition leads to a severe impairment of the vascular structures [124]. In this model, U87 cells presented upregulated FGFR1, VEGFR1, TGFβR1, and overexpressed ligands including angiopoietin 1 (ANG1), hepatocyte growth factor (HGF), VEGF-C and IGF1 [124]. Since Notch and VEGF signaling pathways are not mutually exclusive, it is challenging to delineate these phases based on their stimuli alone. Therefore, we distinguished these phases according to CD144 expression (Fig. 1).

Molecular pathways involved in the endothelial commitment

The primary trigger of endothelial commitment on the way to transdifferentiation is hypoxia. Interestingly, hypoxia induces both CD133 and a disintegrin and metalloproteinase with thrombospondin motifs 1 (ADAMTS1) in U87, U251, U373, but not in T98G cells [116]. HIF-1 binds to the ADAMTS1 promoter, increasing its expression [52]. While CD133 suggests the acquisition of stemness features, ADAMTS1 is correlated with the endothelial signature since ADAMTS1 knockdown leads to downregulation of CD144, EphA2, VEGFR1 and VEGFR2. Moreover, co-culturing U87, U251 or U373 with HUVECS results in a predominantly GBM-derived vascular network, while T98G-HUVECs co-culture develops a HUVEC-based vascularization. ADAMTS1 mediates transdifferentiation through its proteolytic activity against insulin-like growth factor binding protein 2 (IGFBP2) [88]. In GBM cell cultures, both full-length IGFBP2 and its C- and N-terminal fragments are present. However, in tumor samples, only the N-terminal fragment is constantly expressed. IGFBP2 binds to integrin α5β1, activating the focal adhesion kinase (FAK)/ERK pathway, which will further activate the specificity protein 1 (SP1) transcription factor. Then, SP1 interacts with the CD144 promoter and stimulates its transcription. Consequently, GBM cells overexpressing IGFBP2 form more vascular tubes, show higher migratory and invasive abilities, and express higher levels of CD144 and MMP2 [82]. ADAMTS1 cleaves IGFBP2, releasing IGF2, which binds to IGF1R and enhances GBM cell migration [88]. In physiological conditions, IGF signaling increases the number of endothelial progenitors and enhances ECs migratory and tube-forming abilities [7]. In GBM cells, IGF1R phosphorylation modulates the activity of PI3K/AKT pathway [130]. However, ADAMTS1 also inhibits proliferation, migration and angiogenesis in HUVECs [135]. Besides HIF-1, VEGF can induce ADAMTS1 overexpression in ECs [149], and ADAMTS1 binds to VEGF, blocking VEGFR2 phosphorylation [60], participating in a negative loop in the modulation of the angiogenic activity. Based on this data, ADAMTS1 likely contributes to the endothelial lineage commitment while maintaining an inactive, quiescent state of the ECs. Further studies are needed to understand the role of IGF1R in the endothelial transdifferentiation.

Hypoxia may also trigger the ETS variant transcription factor 2 (ETV2) overexpression, as ETV2-positive cells are more abundant in the hypoxic tumor core [154]. In embryonic stem cells, HIF-1α binds to the ETV2 promoter, leading to its upregulation, followed by Notch1 signaling [132], which is consistent with Notch1’s role as shown by Wang et al. [139]. ETV2 overexpression in GBM cells results in the cells acquiring endothelial-like morphology and higher expression of VEGFR2, CD144 and CD31 [154]. This process is independent of VEGF-A, bFGF and EGF, although these factors can double the number of GBM-derived ECs in vitro. In human samples, ETV2 is often co-expressed with stemness and endothelial markers, as well as EGFR amplification. Additionally, ETV2 knockdown in patient-derived GBM tumor cells is associated with smaller tumors and the absence of proliferative GSC-derived ECs (hCD31 + /Ki67 +) in murine models. Although ETV2 is a downstream target of Notch, Wnt, BMP, and VEGF pathways, the mechanism of its upregulation in GBM remains unexplored [25, 99]. ETV2 downregulates Wnt family member 3 (WNT3) and Glycogen synthase kinase 1 (GSK1), both responsible for the neural lineage commitment, and upregulates of FLI1A, VEGFR2, Tie2, CD34 and CD144, suggestive of transdifferentiation towards the intermediate EC phenotype. Other studies have shown that LIM Domain Only 2 (LMO2), Delta Like Canonical Notch Ligand 4 (DLL4) and Notch1 are also possible downstream target genes of ETV2 [99].

Compared to CD133- cells, CD133 + cells exhibit higher expression of LMO2 [68]. Induction of patient-derived GSCs differentiation with serum-containing media significantly downregulates LMO2, while overexpression of LMO2 in premalignant astrocytes leads to increased levels of CyclinD1, Nestin and CD133, suggesting the adoption of a stemness signature. LMO2 overexpression activates both JAG1 (encoding Jagged canonical Notch ligand 1) and CD144 promoters, facilitating their transcription. Consequently, LMO2 overexpression results in the upregulation of VEGFR2, CD144, CD34 and vWF. The exact mechanism of how LMO2 form a transcriptional complex in this context remains unknown. Hypoxia also directly stimulates CD144 expression via HIF-1α and HIF-2α, which bind to the promoter of CD144 [87]. Additionally, C6 cells cultured in endothelial medium under hypoxic conditions show upregulation of HIF-1α, Notch1, VEGFR2 and its phosphorylated form [23]. Thus, hypoxia promotes endothelial commitment, in part via LMO2, which acts both directly by regulating the CD144 promoter, and indirectly by activating the Jag1/Notch signaling pathway.

As previously mentioned, GSCs group into vascular-like channels prior to expressing an EC signature. MMP production and the presence of the basement membrane suggest that extracellular matrix (ECM) remodeling is an early event in the process of transdifferentiation [124, 155]. Collagen, a major component of the basement membrane, plays a significant role in transdifferentiation. For instance, collagen-VI-alpha-1 (COL6A1), a marker of high-grade glioma, is mainly located in the perivascular areas [133]. In vitro studies show that COL6A1 knockdown decreases CD31 expression in GSC-derived ECs under hypoxic conditions [49]. Collagen-IV is also expressed in the perivascular regions of GBM [98]. Prolyl 4-hydroxylase (P4H), a critical enzyme for collagen synthesis, is upregulated by hypoxia (via HIF-1α) in GBM cells, leading to the overexpression of EMT markers (Snail, Vimentin, MMP2, MMP9) [159, 161]. In human samples, P4HA1, a key component of P4H, positively correlates with glioma grade, proliferation, microvessel density, and endothelial markers (CD31, CD34, VEGF-A, VEGF-B, VEGFR2), while negatively impacting survival [49, 159]. P4HA1 knockdown in GSCs reduces collagen IV and VI levels, as well as tube formation and length. How collagen influences endothelial transdifferentiation is not known, but studies on another P4H isoenzyme, namely P4HA2, demonstrate that its knockdown inactivates the PI3K/AKT pathway and downregulates EMT markers (Snail, Slug, Twist1) [77].

A small proportion of induced GSCs (via transduction of dominant-negative TP53 and constitutively activated AKT in c-Myc immortalized NSCs) expressed CD144 despite NSC maintenance conditions [56]. AKT is critical for GSCs acquiring CD144, as p53DN-AKT-Myc-NSCs adopt a CD133 + /CD144 + signature, while p53DN-Myc-NSCs only express CD133. These CD133 + /CD144 + cells also express CD31, CD34, CD105, Tie2, VEGFR2, and vWF, indicating a role for PI3K/AKT in both lineage commitment and maturation. Additionally, inhibition of mammalian target of rapamycin (mTOR) decreases the CD133 + /CD144 + population and their endothelial functions. Moreover, silencing of mTOR, downstream of AKT, downregulates HIF-1α, MMP2, and MMP14, and inhibits VM under both normoxic and hypoxic conditions [57]. In comparison, mTOR inhibition in endothelial precursors represses CD144 and vWF [19], although the exact mechanism by which mTOR contributes to the overexpression of CD144 remains to be determined.

AKT activation also triggers the epigenetic activation of distal-less homeobox 5 (DLX5) and suppression of paired box 6 (PAX6), both collaborating to induce transcriptional activation of the WNT5A locus [56]. ChIP-seq analysis revealed a transition from H3K27 trimethylation to acetylation for WNT5A and DLX5, while PAX6 exhibited increased repressive H3K27me3 marks. WNT5A expression strongly correlates with the mTOR/S6K pathway but is inversely associated with neural and stemness markers. Furthermore, WNT5A downregulates CD133 and ALDH1A1 through the GSK3β/β-catenin/EMT signaling pathway, although the precise mechanism remains unclear. WNT5A secretion is higher in GSC-derived ECs compared to undifferentiated GSCs, and its knockdown results in reduced tube formation, migration, and downregulation of CD31, CD34, MMP9, and MMP14 [22]. In vivo studies demonstrate that WNT5A blockade decreases vessel density and Ki67 + GSC-derived ECs. Additionally, co-culturing GSCs with shWNT5A-GSC-derived ECs results in reduced tube formation, lower levels of EC markers and higher levels of stemness markers in the GSCs, suggesting that GSC-derived ECs are recruiting GSCs. WNT5A binding to Frizzled 4 (FZD4) activates GSK3β/β-catenin pathway, enhancing EC signature and migration markers in GSC-derived ECs. Blocking FZD4 reduces mesenchymal markers (Vimentin, Snail, Twist) and increases E-cadherin expression while maintaining a stem-like morphology.

The roles of PI3K/AKT and MAPK/ERK1/2 were also revealed by treating GBM cells with BMP9, which deactivates these pathways and consequently downregulates the stemness signature (CD133, Nestin, SOX2) [107]. During endothelial transdifferentiation, BMP9 diminishes the number, migratory potential and tube-formation ability of GSC-derived ECs. PI3K/AKT and Ras/MAPK pathways also target ETV4, a transcription factor that cooperates with SP1 to promote the expression of β1,4-Galactosyltransferase 5 (β4Gal-T5) [63], which positively correlates with the intranuclear Notch intracellular domain (NICD) [30]. Notch1 is β1,4-galactosylated by β4Gal-T5, facilitating the interaction between Notch1 and Galectin-3. This galactose-dependent interaction activates NICD and promotes its nuclear translocation [65]. Furthermore, B4GALT5 knockdown in patient-derived GSCs inhibits tube formation, decreases tube length and CD31 expression in vitro, and reduces tumor dimensions and increases survival in an orthotopic mouse model.

Transducing GSCs with the activated form of Notch1 results in the upregulation of TAL1-PP22, a truncated isoform of TAL1, whereas the full-length isoform is normally expressed in HUVECs [47]. Overexpression of TAL1-PP22 is associated with a decline in the GSC growth, independent of SOX2 and OLIG2, suggestive of a role in differentiation. TAL1 + GSCs resemble an endothelial phenotype, expressing CD31 and CD144. Although co-immunoprecipitation reveals an interaction between TAL1-PP22 and LMO2, the importance of this interaction is yet to be elucidated. TAL1-PP22 lacks the transcriptional domain, and there are indications from other studies that this isoform might inhibit the transcriptional activity of the formed complex [47].

In sprouting angiogenesis, ‘tip’ cells express DLL4, which activates Notch1 signaling in ‘stalk’ cells. This DLL4-Notch1 pathway inhibits VEGFR2 expression, preventing the acquisition of the ‘tip’ phenotype [2]. Conversely, ‘stalk’ cells express Jag1, which counteracts DLL4-Notch1 signaling and enhances VEGFR2 expression. Notch1 is critical for the vascular hierarchy, as its inactivation disrupts tube formation and stable junctions. Thus, Notch1 maintains a balance between proliferative ‘stalk’ and proteolytic, migratory ‘tip’ phenotypes. It is possible that the Jag1-Notch1 interaction upregulates VEGFR2 in neighboring cells, promoting a migratory phenotype, while further VEGF stimulation leads to the active, proliferating CD105 + signature. ‘Stalk’/’tip’ EC polarization has not been explored in endothelial transdifferentiation of GSCs. However, Zheng et al. found that Jag1 and DDL4 are widely expressed in hypoxic/necrotic areas of GBM samples [158]. In hypoxic conditions, co-culturing GSCs with both tumor and non-tumor ECs results in higher expression of both Jag1 and DLL4 in ECs, more so in GSC-derived ECs than in HBMECs. Blocking Jag1 and DDL4 also reduces proliferation and sphere formation ability of the neighboring GSCs. Although Jag1 and DDL4 are also present in soluble form, suggesting an autocrine activation of Notch, the impact upon GSC-derived ECs was not studied.

Therefore, Notch1 significantly contributes to endothelial commitment, but growth factors may also play a crucial role, especially with the rise in VEGFR2 expression (Fig. 2). Once more, isolating GSCs solely based on CD133-positivity could select GSCs already undergoing transdifferentiation. Thus, employing a broader array of markers alongside quantitative techniques is imperative to isolate "pure" GSCs (CD133 + /CD144-) and differentiate them from endothelial-committed CD133 + /CD144 + cells.

Fig. 2.

Major pathways involved in the endothelial lineage commitment of GSCs. ADAMTS1 a disintegrin and metalloproteinase with thrombospondin motif 1, AKT AKT serine/threonine kinase, β4Gal-T5 beta-1,4-galactosyltransferase 5, CD144 cluster of differentiation 144, DLL4 delta like canonical Notch ligand 4, ERK extracellular regulated MAP kinase, ETV ETS variant transcription factor, FAK focal adhesion kinase, HIF-1α hypoxia-inducible factor 1-alpha, IGF2 insulin-like growth factor 2, IGFBP2 insulin like growth factor binding protein 2, IGFR1 insulin-like growth factor receptor 1, Jag1 jagged canonical Notch ligand 1, LMO2 LIM domain only 2, MAPK mitogen-activated protein kinase, mTOR mammalian target of rapamycin, P4HA1 prolyl 4-hydroxylase A1, PI3K phosphatidylinositol 3-kinase, VEGF vascular endothelial growth factor, VEGFR2 vascular endothelial growth factor receptor 2

Molecular pathways involved in the endothelial maturation

Endothelial maturation is the transition from a CD133 + /CD144 + progenitor state to an active CD105 + mature phenotype, characterized by downregulation of CD133, CD144 and upregulation of CD31, CD34, VEGFR2, CD105 [139]. CD133 + /CD144- and CD133 + /CD144 + cells show no difference in the expression level of CD105, suggesting that CD105 upregulation is a hallmark of maturation [56]. Since CD31, CD34 and VEGFR2 are also upregulated during endothelial commitment, it is likely that some similar pathways continue to be active during maturation. Additionally, pathways involved in endothelial maturation could also be seen during lineage commitment but activated to a lesser degree due to the different receptor signature. Notably, in vitro conversion of CD133 + /CD144 + to CD105 + ECs was observed after cultivation in endothelial medium (supplemented with VEGF) and it was inhibited by the VEGF/VEGFR2 blockade [56, 139]. Therefore, the enrichment in VEGFR2 is responsible for entering maturation. Once GSC-derived ECs lose their stemness markers, this process became irreversible, and cells retain their endothelial signature even if cultured back in stem cell medium [22].

Some mechanisms observed in normal angiogenesis can explain the crosstalk between VEGFR2, CD144 and CD105. Firstly, CD144 normally keeps VEGFR2 in an unphosphorylated state, maintaining the cell quiescence [140]. Additionally, CD144 loss activates FGFR, which upregulates the expression of VEGFR2. Thus, loss of CD144 leads to an increase of both total VEGFR2 and phosphorylated VEGFR2 levels. Activated VEGFR2 further enhances the internalization of CD144, contributing to a feedback loop that sustains its own activation. VEGFR2 promotes cell proliferation and MMPs secretion via the MAPK/ERK1/2 pathway. Loss of CD144 and gain of CD105 also promote a switch in TGF-β signaling from an antiproliferative to a proliferative effect [108].

Another explanation for CD144 loss is the activation of the EMT program. As previously mentioned, PI3K/AKT/mTOR drives WNT5A expression that will further activate GSK3β/β-catenin/EMT signaling pathway via FZD4 [22]. In GBM cells, TGF-β and ERK pathways also contributes to EMT [119]. Consequently, these signaling cascades activate transcription factors which promote the synthesis of Snail1/2, Slug and Twist. The EC signature is likely driven by Snail1 (targeting CD31 and CD34) and Twist (targeting MMP9 and MMP14). Both Snail and Twist bind to the CD144 promoter, downregulating its transcription [84]. Additionally, MMP14 cleaves CD144 in the EC membrane.

These observations suggest that PI3K/AKT and WNT/β-catenin/EMT are key effectors of endothelial maturation, too (Fig. 3). Since CD105 is a primary marker for active ECs, further studies should investigate the role of WNT5A in CD105 expression. Notably, rapamycin reduces CD105 expression, indicating a possible involvement of mTOR. While embryonic stem cells in 3D cultures express CD105, NSCs in the same conditions do not, although OCT4 overexpression in NSCs induces CD105 expression [124]. In liver cancer, OCT4 activates the AKT/Nuclear factor kappa B (NF-κB) p65 pathway, and silencing of OCT4B1 reduces CD105 expression [79]. Notch inhibition in GBM explants decreases CD105 + cells, though this does not rule out Notch's early role in transdifferentiation [54]. In a U87-based 3D culture model, CD105 expression was unaffected by VEGF or FGFR inhibition, suggesting that this mechanism may become VEGF-independent and other paracrine factors in the GBM secretome can contribute to endothelial maturation [124]. In summary, VEGF induces CD105 expression and downregulates CD144, with potential signaling pathways involved including PI3K/AKT and EMT; however, these pathways can be activated by various other upstream signals [97].

Fig. 3.

The interplay between epigenetic mechanisms and epithelial-to-mesenchymal (EMT) pathway in endothelial maturation. AKT AKT serine/threonine kinase, Ang1 angiopoietin 1, CD cluster of differentiation, DLX5 Distal-Less Homeobox 5, FZD4 Frizzled class receptor 4, GSK-3β glycogen synthase kinase 3, H3K27-me3/-ac trimethylation/ acetylation of lysine 27 on histone, HIF-1α hypoxia-inducible factor 1-alpha, LRP low density lipoprotein receptor-related protein, MMP matrix metalloproteinase, mTOR mammalian target of rapamycin, PAX Paired Box, PI3K phosphatidylinositol 3-kinase, VEGF vascular endothelial growth factor, VEGFR2 vascular endothelial growth factor receptor 2, Wnt5A Wnt family member 5A

VEGFR2 controversy: not all GSCs act in the same manner

The main issue with the definition of endothelial maturation in the context of GSC transdifferentiation is that this process was firstly observed in vitro where is driven by VEGF/VEGFR2 signaling pathway. In several GBM patient samples and in vivo GBM models, many cells with a typical endothelial signature and localization do not express VEGFR2, which casts doubt on the role of VEGFR2 in maturation. The main explanation is related to the TME, which includes various other growth factors or cytokines with a role in angiogenesis.

In a lentiviral murine model (constitutively activated HRAS and AKT in TP53^+/− transgenic mice), Soda et al. found that VEGFR1, VEGFR2 and VEGFR3 were absent in most GBM-derived ECs, while VEGFR2 was present in the majority of host ECs. Additionally, FGFR1 was expressed in all subsets of both GBM-derived and host ECs [125]. GSCs from this model secreted a basal level of VEGF when cultured in NSC medium, but under differentiation conditions, they secreted three times more VEGF in normoxia and up to ten times more in hypoxia. In GSCs, hypoxia also elevated the expression of endothelial markers (vWF, CD31, and VEGFR2 in a limited subpopulation), promoted the adoption of an EC-like morphology, and enhanced tube formation. VEGF is a potent chemotactic signal, proportional to the VEGFR2 expression [152]. Thus, the VEGF secretion of GSC-derived ECs could stimulate the VEGFR2-expressing host ECs and the few VEGFR2 + GSC-derived EC to migrate and contribute to the mosaic pattern of the neovessels. Similarly, Deshors et al. cultured patient-derived GSC lines in endothelial medium (EGM-2), where cells exhibited increased CD31 expression and loss of stemness, with significantly higher levels of VEGFR2 observed only in GSC-derived ECs from one of the three lines tested [37]. Despite variations in VEGFR2 expression, all GSC-derived ECs displayed tube formation similar to HUVECs, LDL uptake, and enhanced migration towards a VEGF gradient, suggesting the potential presence of an alternative receptor for integrating VEGF chemotactic signals. Compared to Wang et al. that originally discovered the role of VEGFR2 in maturation using sorted CD133 + /CD144 + cells, Deshors et al. used sphere-formation assay to isolate GSCs, which can generate a more heterogeneous cell population. Thus, the divergence in results can be attributed to the use of distinct cell lines from various patients, highlighting both intra- and intertumoral heterogeneity as factors contributing to the discrepant behavior of different cell populations.

To comprehend the emergence of the divergent VEGFR2 + and VEGFR2- populations, several mechanisms observed in GBM vascular networks require elucidation. Huang et al. observed that CD31 + ECs isolated from GBM samples exhibit decreased VEGFR2 levels but increased expression of N-cadherin, alpha smooth muscle actin (α-SMA), and fibroblast-specific protein 1 (FSP-1) [58]. Therefore, the absence of VEGFR2 may indicate an EC transition towards a mesenchymal phenotype characterized by disruptions of intercellular junctions and increased vascular permeability, as well as cytoskeletal remodeling to promote a migratory and invasive behavior [153]. Exposure of ECs to the GBM secretome triggers c-Met phosphorylation, leading to the activation of downstream targets NF-κB and ETS-1, the latter being expressed irrespective of hypoxia [69]. Similarly, ECs in GBM secrete platelet-derived growth factor (PDGF) that acts in an autocrine manner upon PDGFR-α and PDGFR-β, leading to NF-κB activation [81]. Both NF-κB and ETS-1 promote FSP-1 overexpression, although the underlying mechanisms remain unclear [58]. Additionally, NF-κB binds to the SNAIL promoter and upregulates Snail, which binds to the VEGFR2 promoter, inhibiting its expression [81]. Indeed, exposure of HBMECs to glioma conditioned medium induces a phenotypic transition towards a disorganized architecture of spindle-shaped cells, endothelial marker reduction, mesenchymal signature upregulation, intercellular protein loss and increased HGF levels [58, 81]. Functionally, these cells exhibit increased proliferation, migration, and monolayer permeability, characteristics retained upon re-culture in normal medium, indicating an irreversible transformation possibly mediated by the autocrine feed-back loops. Despite adopting a mesenchymal phenotype, these ECs maintain endothelial functions (tube formation and LDL uptake). In rodent models, deletion of c-Met or PDGFR-β does not affect tumor growth but promotes vascular network normalization (reduced tortuosity, minimal hemorrhagic necrosis, pericyte coverage) [58, 81]. Although endothelial-to-mesenchymal transition has not been studied in transdifferentiated ECs, the involvement of similar EMT transcription factors in both processes provides a potential explanation for how VEGFR2 may be lost during this transition [153].

On the other hand, hypoxia and nutrient deprivation could trigger the production of reactive oxygen species (ROS) in tumor cells and consequently induce autophagy. In vivo, bevacizumab decreases vessel density, but promotes the formation of VM channels, while chloroquine (used as autophagy inhibitor) reduces VM channels without altering vessel density [142]. Therefore, VEGF is critical for angiogenesis, but not for VM. Instead, ROS promote the overexpression of autophagy protein 5 (ATG5), which is positively correlated with both VM and pVEGFR2 in GBM samples. Indeed, knocking down ATG5 results not only in reduced levels of VEGFR2, but also in diminished phosphorylation at Y1175 position. This is in line with previous research showing that VEGFR2 activation could contribute to lineage commitment, although other pathways such as Notch are more important [152].

Transdifferentiation in relation to GBM treatment

The importance of endothelial transdifferentiation was illustrated in a 3D mathematical model proposed by Yan et al., which predicted that targeting GSC-derived ECs is critical for the development of novel combinatorial regimens aimed at tumor eradication [150]. Current treatment strategies not only fail to address transdifferentiation but may also exacerbate its occurrence.

Following radiation exposure, GSC-derived ECs exhibit Tie2 overexpression without alterations in CD31 or stemness/neural markers, thereby enhancing migratory capabilities towards VEGF and tube formation [37]. Since mRNA and protein levels of ANG1 and ANG2 in GSC-derived ECs are not influenced by radiotherapy, Tie2 is not activated in an autocrine manner. Several GBM cell lines (U87, ADF) express angiopoietins and induce in vitro angiogenesis when co-cultivated with HUVECs, suggesting that GBM itself could provide ligands for Tie2 activation in irradiated GSC-derived ECs [5]. Tie2 primarily acts through the PI3K/AKT pathway rather than the MAPK/ERK signaling pathway. As explained above, PI3K/AKT is crucial for endothelial transdifferentiation. Inhibition of Tie2 signaling by regorafenib decreases CD31 expression and tube formation in irradiated patient-derived GSCs [36]. Additionally, Notch inhibition has been demonstrated to enhance the efficacy of RT, with the combination of RT and γ-secretase inhibition resulting in reduced proliferation, neurosphere formation, and CD133 + cells compared to either treatment alone, although no effect upon transdifferentiation was investigated [54].

TMZ induces the generation of CD133 + /CD144 + cells not only in CD133 + cells, but also in CD133- cells, outlining its role in both dedifferentiation and endothelial fate [8]. Moreover, TMZ increases CD133 + /CD34 + and SOX2 + /CD34 + cell populations, and under therapeutical stress, there is a notable co-expression of SOX2 + and CD105 + , suggesting the development of activated, proliferating ECs. In vivo, TMZ exposure induces the endothelial signature in a time- and dose-dependent manner. In hypoxic conditions, GSCs display a higher survival rate to TMZ when co-cultivated with Nestin + /CD31 + cells or HBMECs, indicating endothelial transdifferentiation confers chemoresistance to the neighboring GSCs [158].

Since hypoxia plays an important role in lineage commitment, therapeutic strategies that combat this mechanism could inhibit endothelial transdifferentiation. However, the main problem is that hypoxic areas are either avascular or characterized by leaky vasculature, which prevents active drug concentrations from reaching these areas. In a phase II clinical trial involving 24 patients with recurrent GBM, small-molecule PT2385 (an HIF-2α inhibitor) did not achieve objective radiographic responses, had a modest impact on outcomes (mOS, 7.7 months), and demonstrated high variability in pharmacokinetics between patients [126]. Another significant pathway involved in transdifferentiation is Notch. Targeting Notch signaling with the γ-secretase inhibitor RO4929097 also proved ineffective in a phase II trial involving 47 patients with recurrent GBM, yielding a 6-month progression-free survival (PFS) rate of only 4% [105]. Disappointing results were also observed when attempting to target the PI3K/AKT/mTOR pathway. In a phase II study of the oral PI3K inhibitor PX-866 in recurrent GBM, 73% of patients experienced disease progression [106]. To date, the only FDA-approved targeted therapy with a possible role in transdifferentiation is the VEGF inhibitor bevacizumab. In a randomized clinical trial, newly diagnosed GBM patients were treated with concomitant chemoradiotherapy followed by either bevacizumab or placebo. While PFS was longer in the bevacizumab arm, no significant difference in mOS was noted (15.7 vs. 16.1 months) [44]. Although these molecules have failed to bring major benefits, strategies to normalize vascularization, multidrug combinations, and strategies for better drug penetration into tumor tissue are necessary to combat transdifferentiation in GBM.

Conclusions

Endothelial transdifferentiation, wherein GSCs assume an endothelial phenotype to facilitate tumor neovascularization, is influenced by the tumor microenvironment, notably hypoxic conditions in perinecrotic regions and the secretome of neighboring cells. This process is not constant in GBM, but many associated markers indicate a poor prognosis. Given that current treatments can promote endothelial transdifferentiation, it is more likely to occur in recurrent GBM. Initially, GSCs form vascular-like channels that integrate into the vascular network, accompanied by a gradual reduction in the stemness signature and an increase in endothelial markers and functions. During the initial phase of endothelial commitment, GSCs upregulate CD144 expression, which is subsequently downregulated in a maturation phase. These processes are tightly regulated by Notch, PI3K/AKT/mTOR, MAPK/ERK, and EMT pathways. Notch appears pivotal in lineage commitment, while PI3K/AKT/mTOR and Wnt/β-catenin/EMT are involved in both phases, though specific activation mechanisms remain incompletely understood. The shift in balance between phases could be attributed to alterations in growth factor receptors, and the quantitative impact of these pathways. Conflicting findings due to tumor heterogeneity necessitate clarification through more sophisticated disease models and larger patient cohorts.

Abbreviations

ABCG2

ATP-binding cassette sub-family G member 2

acLDL

Acetylated-low density lipoprotein

ADAMTS1

A disintegrin and metalloproteinase with thrombospondin motifs 1

ANG

Angiopoietin

ATG5

Autophagy protein 5

BBB

Blood–brain barrier

bFGF

Basic fibroblast growth factor

BMP

Bone morphogenetic protein

COL6A1

Collagen-VI-alpha-1

DLL4

Delta like canonical notch ligand 4

DLX5

Distal-less homeobox 5

EC

Endothelial cell

EGF

Epithelial growth factor

EMT

Epithelial-mesenchymal transition

eNOS

Endothelial nitric oxide synthase

EPC

Endothelial progenitor

EphA2

Ephrin receptor A2

ERK

Extracellular signal-regulated kinases

ETV

ETS Variant transcription factor

FAK

Focal adhesion kinase

FSP-1

Fibroblast-specific protein 1

FZD4

Frizzled 4

GBM

Glioblastoma

GFAP

Glial fibrillary acidic protein

GSC

Glioma stem cell

GSK

Glycogen synthase kinase

HBMEC

Human brain microvascular endothelial cell

HGF

Hepatocyte growth factor

HIF

Hypoxia-inducible factor

HUVEC

Human umbilical vein endothelial cell

IGF

Insulin-like growth factor

IGFBP2

Insulin-like growth factor binding protein 2

JNK

c-Jun N-terminal kinase

LMO2

LIM domain only 2

MAP2

Microtubule-associated protein 2

MAPK

Mitogen-activated protein kinase

MMP

Matrix metalloproteinase

mTOR

Mammalian target of rapamycin

NF-κB

Nuclear factor kappa B

NICD

Notch intracellular domain

NSC

Neural stem cell

OCT4

Octamer transcription factor 4

P4H

Prolyl 4-hydroxylase

PAX6

Paired box 6

PDGF

Platelet-derived growth factor

PI3K

Phosphatidylinositol 3-kinase

PTEN

Phosphatase and tensin homolog

ROS

Reactive oxygen species

RT

Radiation therapy

SHH

Sonic Hedgehog

SOX2

SRY (sex determining region Y)-box 2

SP1

Specificity protein 1

SVZ

Subventricular zone

TGF-β

Transforming growth factor beta

TMZ

Temozolomide

TUJ1

β-III tubulin

VEGF

Vascular endothelial growth factor

VM

Vasculogenic mimicry

vWF

Von Willebrand factor

α-SMA

Alpha smooth muscle actin

β4Gal-T5

β1,4-Galactosyltransferase 5

Author contributions

A.B. and S.S. designed the study. A.B., S.I.F., A.I.F., O.S. and S.S. participated in writing and editing the manuscript. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Availability of data and materials

Not applicable.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.