4.1 Genetic and transcriptomic criteria for the diagnosis and subclassification of glioblastoma

The number of published articles on glioblastoma and its genetics has increased exponentially during the last decade [24]. The 2021 WHO Classification of CNS Tumors defines three genetic parameters for diagnosing glioblastoma, IDH-wildtype: TERT promoter mutation, epidermal growth factor receptor (EGFR) amplification, and the combined gain of entire chromosome 7 and loss of entire chromosome 10 [17]. However, this neoplasm can be further classified into molecular subtypes, which can impact disease progression and clinical practice.

In 2016, Verhaak et al. [25]classified glioblastomas into four subtypes based on their molecular features: neuron, astrocyte, oligodendrocyte, and cultured astrocytic gliomas. Detecting these subtypes relies on the different therapeutic approaches required for each patient and their impacts on tumor progression [26]. Recently, Neftel et al. [27] identified four heterogeneous cellular states using single-cell RNA-sequencing and validated the intratumoral heterogeneity present in GBM and the relevance of this subtyping. They classified the development of neural signatures into neural-progenitor-like, oligodendrocyte-progenitor-like, astrocyte-like, and mesenchymal-like states [26].

The neural subtype is derived from astrocytes and oligodendrocytes and expresses neuron-related genes, whereas the proneural subtype exhibits the characteristics of oligodendroglial cells and develops in young patients [25]. The classical subtype possesses astrocytic features and expresses neuron precursor and stem cell markers, whereas the mesenchymal subtype shows characteristics of cultured astrocytic gliomas [28].

Verhaak's latest update for reclassifying gliomas removed the neural subtype because it is problematic for identifying primary and recurrent gliomas owing to its ongoing genomic signature changes [29]. Further, a study dedicated to the evolution of the tumor determined that the vast majority of The Cancer Genome Atlas (TCGA) subtypes were from a proneural-like precursor and switched to a mesenchymal-like state in a differentiation process regulated by TNF-α/NF-κB signaling or ASCL1. To sum up, Neftel et al.’s studies [27]and the recent update of the WHO CNS tumor classification reflect the fluidity of GBM's transcriptional states and the influence of the tumor microenvironment (TME) on the development and transition from one subtype to another [22].

Jankowska et al. [30] classified glioblastoma subtypes based on immunochemical expression and concluded that the classical subtype is represented by TP53 mutation, which makes this subtype highly sensitive to classical radiotherapy plus chemotherapy with adjuvant TMZ. The mesenchymal subtype shows NF1, PTEN, AKT, MET, and TRADD mutations. Further, PDGFRA, IDH1, TP53, HIF, and OLIG2 mutations are characteristic of the proneural subtype. Another study revealed the neuronal markers for identifying and profiling neural glioblastomas, such as NEFL, GABRA1, SLC12A5, and SYT1 (Table 2) [31].

In parallel to this classification, Herrera-Oropeza et al. [32] performed a multi-omics analysis of driver genes. They concluded that mesenchymal subtype development was related to the upregulation of the MGMT promoter and the downregulation of ATRX, H3F3A, TP53, and EGFR. Complementary information was provided for the proneural subtype, characterized by the overexpression of MKI67 and OLIG2, and the classical subtype by the overexpression of EGFR, NES, VIM, and TP53. The characterization of differential molecular characteristics of histologically similar tumors is relevant to improve the diagnosis of GBM in patients. Besides, the determination of expression profiles is useful for creating progression models and enhancing the prognosis of each tumor subtype.

There is a rising tendency to classify gliomas based on their mRNA sequencing and the clustering of samples with computational programs. Using the “Consensus Cluster Plus” package for R v4.0.3, Cai et al. [33] investigated the reclassification of glioma based on the expression levels of Gβ/γ genes from TCGA and the Chinese Glioma Genome Atlas (CGGA) datasets. The result was a differential distribution map correlated with the samples analyzed from these two chosen databases. The Gβ/γ heterodimer can activate the Erk1/2 pathway by inducing the overexpression of guanine nucleotide-binding protein beta 4 (GNB4), which results in the transformation of epithelial and mesenchymal cells into glioma cells. After clustering, they obtained three subgroups: GNB2, GNB3, and GNB5. GNB2 appeared to be the best indicator of malignant tumors, especially in patients with IDH-mutated, non-codeleted 1p/19q low-grade gliomas (LGGs). This subgroup is characterized by high M0/M2 cell infiltration levels and is highly associated with the immunosuppressive phenotype, thus demonstrating enhanced PI3K-Akt/JAK-STAT pathways and high levels of tumor-associated macrophages (TAMs) and M2 macrophages. Therefore, the GNB2 subgroup would represent the immunosuppressive phenotype in gliomas. Each subgroup has a unique tumor-related pathway that can answer the selection of a chemotherapeutic drug and enhance the glioma prognosis by choosing the right target [33].

An analysis of the differentially expressed genes (DEGs) revealed 110 upregulated genes and 75 downregulated genes in the GBM samples. This observation identified a four-protein prognostic signature (SLC12A5, CCL2, IGFBP2, and PDPN) for the segregation of patients into high- and low-risk groups and for the estimation of survival time. These were observed via a weighted gene correlation network analysis (WGCNA) algorithm, a strategy that has also been used to determine disease-related genes in other oncologic diseases [34]. Another DEG screening found 662 DEGs in patients with GBM and concluded that DECR1, POLR2F, HDAC1, and PDIA3 could be the critical genes related to the overall survival (OS) time of patients with GBM [35].

A WGCNA study comparing the transcriptome and proteome of glioblastoma, IDH-wildtype tumors found six proteomic modules correlated with survival, but none of the identified RNA modules did [36]. After performing a Kaplan–Meier analysis, 11 proteins were revealed to have a significant association with survival, despite not being significant at the RNA level. Owing to the apparent lack of correlation between RNA and survival, previously established single-cell-based signatures were used to define the dominant cell subpopulation of each tumor analyzed. This study revealed that mesenchymal and neural progenitor cell-like subpopulation signature genes correlated with shorter survival, whereas oligodendrocytic precursor cell-like and astrocytic subpopulation signature genes correlated with more prolonged survival. Gene Ontology (GO) enrichment analysis from the proteomic and single-cell-based signature data revealed that lysosomal activity and amino and nucleotide sugar metabolism were enriched in a cluster of genes and proteins correlated with short survival [36].

Transcriptomic and proteogenomic profiling techniques have been further developed for the classification of gliomas; these could lead to a robust and objective method for the stratification of patients and improve survival prediction, as commented in the previous survival [36].

After a regular medical diagnosis of glioblastoma is achieved via MRI, computed tomography, or biopsy followed by blood analysis, patients await a more personalized and guided treatment [38]. The repertoire of molecular biomarkers characterized by transcriptomics is quite robust. However, only a few are critical for a detailed diagnosis. Further research is needed to classify glioblastomas into subtypes and grades and to estimate survival rates. The most relevant biomarkers described in this section are the mutation of TP53 as the molecular feature of classical glioblastoma, the PDGFRA mutation for the proneural subtype, the presence of GNB2 as an indicator of aggressive tumors, and the cell subpopulation signature as a measure for survival.

4.2 Age- and sex-related patterns in molecular classification

Molecular classification can also reveal an age-related pattern of biomarkers for glioma. This evidence is reflected in a previous computational clustering study that revealed H3F3A, AHNAK2, SOX1, SUSD2, and KMT2C were the most mutated genes in young-age patients, PIK3CA and TERT were the most mutated genes in middle-age patients, and RYR2 was the most mutated gene in old-age patients. Furthermore, two mutations were relevant for young- and middle-age groups: BCORL1 (as an indicator for HGGs) and KMT2D, whereas three mutational events on TERT, PTEN, and NF1 were more frequent in old-age patients [39].

The characterization of these new biomarkers could provide a more refined molecular classification of HGG/LGGs between age groups when added to the IDH1 mutational status and TERT methylating status. However, these data are based on the IV WHO classification of CNS tumors, and thus they should be updated [18].

RNA-sequencing and real-time polymerase chain reaction (RT-qPCR) quantification followed by Kyoto Encyclopedia of Genes and Genomes (KEGG) and gene ontology (GO) analyses revealed a sex-related pattern that emerged from the differential expression of hub genes in the cell cycle, DNA replication, and the Fanconi anemia pathway. These DEGs were mainly enriched in women because the involved pathways mediated progesterone release, which leads to oocyte maturation. Four strongly correlated genes (CCNB1, CDC6, KIF23, and KIF20A) were upregulated in the glioma samples and mediated cell cycle, ATP-binding, and DNA replication. The CCNB1 protein accelerates mitosis and promotes tumoral invasion, thus suggesting a recurrent role in GBM [40]. CDC6 encodes an enzyme that mediates mitosis via E2F regulation [41]. KIF23 is highly expressed in malignant tumors, and KIF20A promotes reverse transport from the Golgi complex to the endoplasmic reticulum and the presentation of the major histocompatibility complex class I (MHC-I), thereby disguising the tumor from immune response and maintaining its proliferation [42, 43]. These last two hub genes might be potential biomarkers for GBM diagnosis, especially in women [44].

These molecular approaches aim to specify a pattern of DEGs between age or sex groups. On the one hand, distinctive DEG clusters between age groups have been observed that could be useful for classifying the tumors of the CNS following the most recent criteria, especially to distinguish the pediatric-type diffuse LGGs and HGGs. On the other hand, an RNA-sequencing study led to the characterization of four hub genes highly related to glioma samples, two of them being possible new biomarkers for GBM diagnosis in women: KIF23 and KIF20A.

4.3 Epigenetic-based characterization of glioblastoma and prognostic value

Survival prediction can be discussed from an epigenetic perspective. The most frequently observed molecular feature is the status of the MGMT promoter, whose methylation level correlates with the tumor's prognosis and is considered a universal marker to evaluate TMZ sensitivity in glioma chemotherapy. In fact, the MGMT promoter methylation level is more significant than grade or 5-hydroxymethylcytosine (5hmC) for age-related prognosis [14, 45, 46].

Conventional chemotherapy with TMZ as adjuvant treatment is an inductor of DNA damage and leads to genetic alterations in the glioma cells, which adapt to the drug dose and develop resistance when the MGMT promoter is hypermethylated [45].

Methylation profiling is another interesting strategy to stratify GBM tumors. DNA methylation-based GBM subtypes seem related to local T-cell infiltration [47]. In fact, these immunological characteristics lead to the classification into four methylation subgroups: IDH, RTK I, RTK II, and mesenchymal tumors.

Interestingly, IDH methylation groups have the lowest CD3+ T-cell infiltration and a low PD-1 expression. Mesenchymal subtype tumors have the highest CD3+/ CD8+ T-cell infiltration. An increased PD-1 expression along with higher levels of CD8+ infiltration results from radiochemotherapy, suggesting that CD8+ T-cells might evolve to an anergic phenotype and activate the immunosuppressive response. Consequently, the mesenchymal subtype might become more aggressive against immune response after conventional therapy. Thus, this information could help identify patients suitable for specific immunotherapy trials [47].

DNA methylation-based diagnosis could support the histological diagnosis of GBM by combining the transcriptomic and methylation patterns of tumor samples and measuring the methylation degree of the CpG islands [45]. Using the MethylMix algorithm, Wang et al. [48] revealed six highly methylated genes (ANKRD10, BMP2, LOXL1, RPL39L, TMEM52, and VILL) that could be used for the molecular subclassification of GBM. The methylation signature is an independent factor that might predict high- and low-risk glioblastomas and overall survival.

Another epigenetic modification contributing to cancer proliferation is the aberrant methylation of histones, a process regulated by histone methyltransferases. An active research field in GBM therapy relies on applying histone deacetylase inhibitors (HDACIs) to improve the patient's OS. An excellent example of the application of HDACIs to treat GBM is a phase II/III trial designed with HDACIs + gene-mediated cytotoxic immunotherapy (G-MCI) or gene-editing treatment mediated by zinc-finger, CAS enzyme, and new-generation sequencing findings [45]. A phase II study tested valproic acid (VPA), an HDAC inhibitor, in newly diagnosed patients showing improved overall survival outcomes and lower toxicity. VPA sensitized glioblastoma cells to radiation in 81% of the patients, thus increasing the effectiveness of standard radiotherapy [49].

It has been observed that receptor tyrosine kinases I (RTK-I) subtype GBM show global hypomethylation. The SOX10 gene, linked to chromatin remodeling and therapy resistance in melanoma, is hypomethylated and overexpressed in RTK-I subtype GBM. Repression of this gene in an in vivo syngeneic graft GBM mouse model resulted in epigenetic alterations, a phenotypic switch to a mesenchymal subtype, and increased tumor cell invasion. In this case, the RTK-I subtype is related to better overall survival than the mesenchymal subtype [50].

Some super-enhancers involved in the regulation of cell identity genes show subtype-specific enrichment. Consequently, the status of the enhancer landscape plays an essential role in determining tumor subtype identity in GBM, and their enrichment could serve as a biomarker for the molecular diagnosis of specific subtypes [50].

Alternative-splicing profiling represents a novel technique for glioblastoma classification. ANXA7, MARK4, MAX, USP5, WWOX, BIN, RON, and CCND1 have been suggested as altered biomarkers serving as functional targets for personalized treatment depending on the heterogeneity of the phenotype and genotype of each patient [51-53]. Moreover, SNRPB, a vital element of the spliceosome complex SmB/B′ implicated in DNA repair and chromatin remodeling, might be a potential target for novel therapies [54]. Additionally, CELF2, a regulator of splicing events, could be a valuable predictor of the prognosis, along with the IDH status and the zinc-finger motif deletions (3′ ZNF domain alterations) [55].

In summary, the MGMT promoter methylation level is an indicator of age-related gliomas, prognosis, and immunoresistance. However, this is not the only biomarker identified by epigenetic changes. Hypermethylation, hypomethylation, and alternative splicing play an essential role in tumor heterogeneity. The latest clinical trials focused on the methylation of histones and the evaluation of CELF2 as a potential predictor of prognosis. Epigenetic changes are a hallmark of cancer, and alternative splicing is one of their most frequent manifestations. These subtle changes lead to a wide heterogeneity of phenotypes, making epigenomic profiling and characterization two essentials for personalized prognosis.

4.4 Noncoding RNA's role in the progression of glioblastoma

Noncoding RNAs (ncRNAs), such as microRNAs (miRNAs) and long ncRNAs (lncRNAs), have interesting regulatory effects on GBM. These nucleic acids can potentially modify the expression levels of proteins involved in the proliferation and migration of tumor cells, like metalloproteinases, cytokines, and growth factors [56]. GBM cells exchange miRNA molecules with oligodendrocytes and endothelial cells within the TME. These molecules can promote angiogenesis and cell differentiation, but some work as tumor suppressors [57].

Identifying miRNA with highly altered expression in glioma provides another method of analyzing patient samples via microarray. The diagnosis can be made by detecting only three miRNAs: miR-4763-3p, miR-1915-3p, and miR-3679-5p. The first and second miRNAs appear oncogenic and are higher in patients with diffuse glioma, while the last might be a tumor suppressor because of its lower levels in patients. Although this could be a promising technique, the current results seem inefficient in discriminating diffuse gliomas from healthy tissue. Nevertheless, these three serum miRNAs represent a powerful tool for GBM diagnosis in combination with histological and molecular characterization [58].

The lncRNA MIR4435-2 Host Gene (MIR4435-2HG) is upregulated in GBM tissues. Besides, higher expression of this lncRNA correlated with shorter OS. MIR4435-2HG targets miR-1224-5p, which inhibits TGFBR2. The inhibition of this receptor results in a diminished cell invasive potential compared to MIR4435-2HG overexpressing U87 cells. This result agrees that MIR4435-2HG knockdown resulted in the inhibition of cell proliferation and increased cell apoptotic rates in U87 and U251 cell lines. Furthermore, this lncRNA can be found in other tumors (e.g., gastric and colorectal cancer), in which its upregulation is also linked to poor [59].

The small nucleolar RNA host gene 12 (SNHG12) is overexpressed in TMZ-resistant GBM samples after TMZ treatment. Hypomethylation of the promoter region of this lncRNA induces transcriptional activation of SNHG12 by the SP1 transcription factor. SNHG12 acts as a molecular sponge for miR-129-5p, increasing the expression of MAPK1 and E2F7, which regulate TMZ-induced cell apoptosis and cell proliferation. Even though tumor heterogeneity implies that each patient presents distinct differentially expressed lncRNAs, ncRNAs are promising biomarkers that could have relevant clinical significance [60].

In brief, noncoding RNAs seem to play a role in tumoral proliferation. lncRNAs exhibit a poor prognosis linked to higher apoptosis rates, whereas miRNAs can be targeted for tumor suppression. MIR4435-2HG and SNHG12 are highly expressed in glioblastomas, increasing the tumoral genotypes and heterogeneity. Targeting these molecules could prevent aggressive or high-grade glioblastomas.

4.5 Computational methods for glioblastoma diagnosis

Deep convolutional radiomics features of diffusion tensor imaging (DTI) [61] and machine-learning assisted dynamic susceptibility contrast-magnetic resonance imaging (DSC-MRI) are novel constructional methods that are worth mentioning. These methodologies provide a better molecular classification of gliomas [62, 63] and subtypes of GBM based on analyzing pathological images and their computational modeling via a deep learning method integrating different biomarkers [37, 64]. Deep learning machine analysis is based on computational artificial intelligence that learns from data samples and builds up neural network models that elucidate the diagnosis, decision-making, and clinical predictions related to GBM therapy [65].

An example of computational modeling is described in a study by Randles et al. [66], where the authors investigated the dynamics of glioblastoma stem cells within the perivascular niche as they designed a computational model on the Vulcan supercomputer, which let them examine different treatments and their outcomes. Each simulation analyzed the spatial distribution and interactions between cells, giving a fitness value to each cell. Following the motion and the spatial landspace of these cells, the supercomputer could determine tumor growth through time. Thus, they concluded that giving chemotherapy with TMZ right before radiotherapy improved survival because of the timing glioblastoma stem cells spread in space. In this manner, the immunoresistant response to TMZ was more effectively blocked.

Computational models can interpret both proteogenomic and metabolomic characterizations of GBM. Wang et al. [67], using computational analysis, revealed how PTPN11 and PLCG1 are signaling hub genes in RTK-altered tumors, how immune cells characterize GBM subtypes, and how histone H2B acetylation is a biomarker for classical glioblastoma. The processing of data collection and interpretation was facilitated by a non-negative matrix factorization for multi-omics subtyping, the use of iPROfun for covariates, and a deep learning histopathology image analysis.

An enormous variety of studies can be approached by computational modeling, a potential tool for long-period analysis and multiple condition evaluations. The computational model is designed in terms of the population census and the experimental conditions. Nevertheless, deep-learning methods make a difference. Advances in processing and refining data compilations might help researchers head in a direction when giving a diagnosis and prognosis to GBM patients.

The list of possible molecular biomarkers for diagnosing and classifying gliomas is endless. The following table (Table 3) collects those molecular targets for diagnosis, prognosis, and the personalized treatment mentioned in the previous sections:

5.1 GBM-related molecular pathways

Tumor heterogeneity deviates from the “cell niche” regulation and develops from several signaling and immunosuppressive pathways interceptions. Control of the deactivation of cell proliferation, self-renewal, and differentiation of glioma-initiating cells is mediated by the Wnt, Notch, and TGF-β signaling pathways. The mesenchymal subtype is characterized by an overexpression of TGF-β and vascular endothelial growth factor (VEGF) pathways and attenuation to both Wnt and Notch signaling as well as the expression of YKL40, a specific biomarker for this subtype. On the contrary, Notch and Wnt signaling pathways were prominently activated in the proneural subtype [68]. There is differential activation of GICs specific to each GBM subtype. The concurrency of TGF-β signaling and lower activation of both Notch and Wnt signaling pathways suggests that targeting GIC subtypes might improve clinical outcomes.

Apart from these pathways, p53 signaling remains essential for immortality by amplifying murine double minute 2 (MDM2), which binds the TP53 gene and inhibits its regulatory role in mutations. The retinoblastoma protein (Rb) pathway is also crucial for regulating the cell cycle and proliferation. Rb protein inhibits the E2F transcription factor, which stimulates the transcription of genes involved in the progress from the G1 to S phase during mitosis [69]. These two latest pathways control the cell cycle and their targeted interception might mitigate the invasiveness and migration of glioblastoma cells.

There is an opposing interplay between the IDH1 mutation and the Wnt/β-catenin pathway [70]. The Wnt signaling pathway is crucial in cell proliferation, migration, and apoptosis. However, this pathway inhibits glycogen synthase kinase-3β (GSK-3β), an inflammation and cell membrane signaling regulator. IDH1 mutation is related to a better response to cytotoxic therapy and longer survival in GBM patients.

The PI3K/Akt pathway is involved in the phosphorylation of GSK-3β, which leads to the nuclear transport of β-catenin. This transport promotes the activation of STAT3, an oncogenic transcriptional factor involved in GBM growth, stimulation of cyclin D1 and c-Myc (related to angiogenesis and proliferation), and overexpression of MMP-2/9, which induces cell invasion [71, 72]. A clinical trial based on the combined treatment using sulindac + LY294002 [73] aims to inhibit PI3K for the blockade of GBM invasion. Another biological drug inhibiting invasion is celecoxib, a PI3K inhibitor that can also diminish Akt signaling [71]. The concomitant reduction in tumor proliferation is accompanied by increased cell death [72].

The most remarkable epigenetic silencing of the Wnt pathway occurs because of the hypermethylation of soluble frizzled-related protein (FRP) genes. FRPs create a receptor complex that binds to Wnt ligands and consequently activates the AXIN/APC/GSK-3β complex via phosphorylation. This last step promotes the accumulation of β-catenin in the cytosol and leads to the activation of RTKs, therefore, the stimulation of the HIF-1α via the PI3K/Akt pathway. HIF-1α is a hypoxia factor that enhances the Warburg effect by overproducing glycolytic enzymes, such as LDH-A. The final result of FRP silencing is the inhibition of glucose metabolism in the glioma cells [74, 75].

In recent studies, IDH1-R132H mutation was found to be correlated with better prognosis owing to the decreased expression of the Wnt/β-catenin pathway. This result could be explained by the lower intracellular glutathione (GSH) levels due to the reduced availability of NADPH, an essential cofactor in the oxidative carboxylation of α -ketoglutarate, and higher levels of reactive oxygen species, which induce apoptosis and reduce cell proliferation. IDH1 is an independent predictor of improvement in the clinical outcomes of TMZ therapy. As mentioned above, IDH1-mutated tumors correlate with a better prognosis for low-/high-grade gliomas. Consequently, this type of mutation in patients with glioma reduces proliferation and induces apoptosis [76].

The Notch signaling pathway regulates cell migration, differentiation, apoptosis, self-renewal, and homeostasis. This pathway consists of four cytoplasmic receptors (Notch 1-4) and their ligands, Jagged-1, Jagged-2, and DII 1-4. The expression level of Notch 1, predominantly expressed in neurons, astrocytes, precursor/ependymal, and endothelial cells could be related to the GBM survival period. Notch signaling activity might be useful to predict the overall survival and tumor resistance. Results with a novel therapeutic antibody, functionally validated with a computational-guided approach, suggest that Notch signaling via Hes1/Hey1 targeted genes could be a druggable and clinically relevant target in GBM. Brontictuzumab (BRON) is the first humanized anti-Notch 1 blocking antibody directed against cell surfaces to diminish tumor cell invasion [77].

The Hedgehog (Hh) signaling pathway plays a crucial role in embryogenesis and tumorigenesis; furthermore, this pathway plays a pivotal role in tissue repair and regeneration. The terminal effectors of the Hh pathway in glioma are glioma-associated oncogene homolog 1 (GLI1) zinc-finger transcription factors. An alternative-splicing, truncated variant, tGLI1, is expressed in most GBM samples, but it is undetectable in normal brain cells. This tGLI1 is a gain-of-function variant able to activate several genes not regulated by GLI1. The targeted genes upregulated by tGLI1 include VEGFR1, VEGF-A, VEGF-C, TEM7, HPSE, CD24, and CD44, thus promoting glioblastoma cell proliferation, migration, invasion, and angiogenesis [78]. The aberrant role of the Hh pathway leads to the need to understand the impact of GLI variants, potentiating the development of novel therapies that stop metastasis.

In summary, one of the main reasons that glioblastomas are so heterogeneous is that they can modulate the core regulatory signaling pathways involved in immune response, apoptosis, cell growth, proliferation, and migration. The evolution of glioma depends on the upregulation or downregulation of three main pathways: TGF-β, Wnt, and Notch.

5.2 The role of the RTK/PI3K/Akt/mTOR axis in glioblastoma

A relevant part of the research devoted to molecular-targeted therapy of GBM focuses on identifying intrinsic biomarkers in the RTK/PI3K/Akt/mTOR, JAK-STAT3, and RAS/RAF/MEK pathways as well as the p53 and cell cycle regulation pathways. The RTK/PI3K/Akt/mTOR pathway regulates cell growth, metabolism, and survival in gliomas (Figure 1) [46]. mTOR kinase functions in two complexes: as the nutrient sensor of the cell regulating cell growth (mTOR complex 1 + protein RAPTOR) and coordinates the cytoskeleton's organization and Akt activation via phosphorylation (mTOR complex 2). A remarkable distinction between normal and glioma cells is the loss of function of phosphatase and tensin homolog (PTEN). Consequently, the deactivation of PTEN results in increased Akt activity that triggers mTOR activity that enhances cell proliferation [79].

On the contrary, RTKs activate PI3K and lead to the activation of Akt depending on the phosphorylation of protein kinases 1 and 2 (PDK1/2). Thus, the hyperactivation of Akt is pertinent in understanding why glioma cells are permanently proliferating, changing their metabolism, and promoting a cancer phenotype. Resistance to TMZ treatment stems from the role of autophagy in glioma cells induced by the inhibition of this last pathway [79].

The relevance of the hyperactivation of this axis relies on the control of glioblastoma cell survival. This survival is characterized by changes in metabolic, cell cycle, and cell growth pathways and is translated into radiotherapy + TMZ chemotherapy resistance.

6.1 A closer look at metabolic changes in glioma cell

The conventional classification of LGGs, based on the IDH1/2 mutational status, leads to the metabolic characterization of differentiated astrocytes, one of the cell types that could possibly originate from GICs, as previously described. The traditionally defined IDH-mutated astrocytoma represents the best example of altered metabolism within the tumoral heterogeneity of gliomas [84].

The brain is highly dependent on glucose intake to function correctly. Glioma cells adapt their metabolism according to glucose availability, which gives them extra resistance to hypoxia or altered redox situations. Selective pressure on tumors makes them overexpress glucose transporters, such as GLUT1/3, on their plasmatic membranes, regulated by the hypoxia factor HIF-2 α. Even when glucose levels are low, HIF-1 α guarantees the upregulation of hexokinase 2 (HK2), increasing the glycolytic pathway. Furthermore, many gliomas are characterized by the loss of PTEN function, which causes the constitutive activation of Akt1 and the stabilization of PFKP [85, 86].

MYC is a proto-oncogene that promotes a bidirectional flow in the mitochondrial transport of lactate-pyruvate. Deletion of MPC1 and the accumulation of LDH-A leads to the transformation of pyruvate into lactate, which enhances the Warburg effect [87].

The next modifying step comes with the modulation of the Krebs cycle by extracting C atoms from it (cataplerosis) or introducing C atoms (anaplerosis) to it. These reversible reactions play a crucial role in the de novo biosynthesis of fatty acids, amino acids, and nucleotides. Intermediate metabolites, such as citrate and α-KG, escape oxidation and serve as precursors for fatty-acid biosynthesis and aspartate/glutamate synthesis. Aspartate initiates nucleotide biosynthesis, while glutamate provides the C-skeleton of non-essential amino acids. Hence, α-KG DNA repair and demethylating activities become highly inhibited by the overproduction of 2-HG (sensitizing IDH-mutant cells to PARP-1 inhibition and NAD⁺ deficiency), which also affects the transamination of this compound into keto acids and glutamate [88]. The marginal extensions of astrocyte-like glioma cells contain high levels of cytosolic citrate, especially in the pseudopodia, where limited access to glucose leads to the uptake of acetate and oxidation, mediated by the ACSS2 enzyme [89].

Lipid metabolism is also altered in GBM. The marked metabolic heterogeneity of GBMs allows the use of this altered lipid metabolism to mark GBM stem and non-stem cells in separate tumor niches [88]. GBMs can use ketone bodies and fatty acids to maintain growth, thus allowing their progression during ketogenic diet therapy [90]. Two essential enzymes mediate the biosynthesis of fatty acids, acetyl-CoA carboxylase (ACC) and fatty-acid synthase (FASN). FASN can be used as a biomarker since it is enriched in GBM-derived EVs [91]. ACC and FASN are regulated by SREBP-1, which responds to the EGFR-PI3K-Akt1 signaling pathway. By increasing the EGFR signaling, SREBP-1 favors the tumor evolution of GBSCs into a proliferative status by synthesizing long-chain ω3/6 polyunsaturated fatty acids. Meanwhile, the marginal and hypoxic regions store fatty acids via FABP3/7, which binds to polyunsaturated fatty acids (PUFAs) in a particular structure, defined as pseudopalisades, a hallmark of GBM [92, 93]. A super-enhancer in GBM and GSCs promotes PUFA synthesis. These PUFA maintain EGFR signaling and membrane organization in GSCs. This observation suggests that dual targeting of EGFR and PUFA metabolism could be a novel potential therapeutic approach for glioblastoma management [94].

Regarding nitrogen metabolism, glioblastoma cells show both altered expression and activity of the amino acid transporter SLC7A11 [95], key enzymes involved in glutamine metabolism (glutamine synthetase and glutaminase) [96, 97] and cysteine metabolism [98]. The resulting balance of nitrogen metabolism gives rise to cataplerosis (a decreased availability of carbon atoms to enter the Krebs cycle oxidative pathway) [84].

Glutamine dependence exhibited by some tumor cells has motivated the development of therapeutic approaches based on the metabolism of this amino acid, and GBM is no exception. A phase I clinical trial combines a glutaminase inhibitor, Telaglenastat (CB-839), with radiotherapy and TMZ chemotherapy in IDH1-mutant astrocytomas. Telaglenastat may stop tumor growth by blocking the enzymes needed for this process (NCT03528642).

Altered metabolism is a consequence of tumor heterogeneity and is favored by the complexity of the tumor niche, which is highly vascularized, with infiltrating M2 lymphocytes and TAMs, and a heterogeneous cell population. This characterization of gliomas is exemplified by the study of the mutation in IDH1/2. Knowledge relative to the altered metabolism shown by glioma cells is important to determine how these metabolic changes can affect the development of the tumor and to find new therapeutic targets. Several key intermediates and enzymes from the four main metabolic pathways previously described are discussed and suggested for glioblastoma targeting and treatment. Targeting PTEN could reduce the glucose intake in glioblastoma cells, while targeting PUFA and EGFR might affect the storage of lipids. Finally, the design of inhibitors in the synthesis mechanism of 2-HG might recover the DNA-repairing system.

8.1 Immunotherapy methods

Microglia are the principal antigen-presenting cells in the CNS. GBSCs escape the immune system by increasing STAT3 expression, which translates into the upregulation of Wnt and TGF- β pathways (NCT01904123), resulting in the secretion of immunosuppressive factors, such as TGF β-2 and interleukins IL-10. This cell-to-cell mediated release is regulated by activating the cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) in regulatory T cells (Tregs). Additional help comes from TAMs that show high levels of programmed cell death-ligand 1 (PD-L1). PD-L1 binds to receptors on the surface of T cells, restricting their activity [69, 113].

Based on this previously exposed mechanism, the most cited approach in immunotherapy involves blocking TAMs by creating suitable antibodies against CTLA-4 and PD-L1 [35]. The double blockade of CTLA-4 and PD-L1 is another suitable checkpoint for immunotherapy trials (NCT03233152, NCT04145115, NCT04003649). PD-L1 was targeted by a phase I clinical trial that evaluated the effects of nivolumab plus ipilimumab for recurrent GBM. Its purpose was to compare this new treatment's adverse effects and efficacy with bevacizumab monotherapy (NCT02017717) [69]. Another interesting approach is combined therapy comprising PD-1/PD-L1 inhibitors plus radiotherapy or antibodies targeting CTLA-3, TIM-3, LAG-3, IDO, or OX-40 (NCT02658981, NCT04003649) [108, 113]. Antibodies that block the PD-1/PD-L1 interaction or the T-cell response are only effective in some patients. A series of ongoing trials targeting PD-1 aim to restore immunogenicity, especially CD8 and CD4 T-cell responses (NCT02287428, NCT04201873, NCT03422094, NCT03899857).

The most recent interventions are the development of highly personalized cytomegalovirus (CMV) therapy and neoantigen vaccines. CMV protein pp65 causes a robust CD8+ T-cell response that benefits survival, especially for CMV transferred into dendritic cells (NCT03688178). This last intervention serves as an adjuvant for vaccination, the same way CpG oligonucleotides and granulocyte-macrophage colony-stimulating factor (GM-CSF) do. Neoantigen vaccines show potent antitumoral activity by inducing both CD8+ and CD4+ T cell responses. However, before these strategies reach clinical practice, further research and optimization are required [112].

To sum up, most immunotherapy therapies in development to treat GBM aim to avoid the inhibition of T cells, mediated by Tregs and TAMs, using antibodies against CTLA-4 and PD-L1. However, new strategies that could provide personalized treatment with a robust immune response, such as CMV therapies and neoantigen vaccines, constitute a promising research area.

8.2 Molecular-targeted therapy

9.1 Immunotherapy and molecular-targeted therapies

Immunotherapy and molecular-targeted therapies are two major trends in the treatment of glioblastoma. Immunotherapy has not provided positive results thus far due to the immunoresistance of CNS tumors. Even so, the development of new approaches, such as tumor-specific neoantigen-based and nucleic acid-based vaccines, could be a step toward personalized GBM treatment. In fact, an increase in the number of clinical trials based on nucleic acid-based vaccines may be expected thanks to the remarkable advances in this strategy achieved during the last years. Implementing multiple-antigen targeting could improve vaccine therapy by increasing OS and PFS, two benchmarks currently failing in recent trials. Most commonly found molecular biomarkers might be targeted simultaneously: EGFR, IL-10, CD27, SOX40 and WT1, combined with tumor-specific antigen vaccines, thus providing a more individualized treatment [112, 115]. The challenges of a multifaceted approach remain overwhelming, but the potential benefits that may result from them are substantial.

9.2 Circulating biomarkers for GBM diagnosis and prognosis

Circulating tumor cells, EVs, and circulating nucleic acids and proteins are candidate biomarkers for glioma diagnosis, although none of them have been approved for clinical practice thus far [31]. Circulating tumor nucleic acids are a source of comprehensive information relative to glioblastoma cells' genome, while circulating miRNAs can be used to find potential targets related to tumorigenesis and proliferation and to elucidate the grade and response to glioma chemotherapy [124, 125].

As previously mentioned, liquid biopsy can be a helpful tool for diagnosing and developing personalized treatments. Besides being the brain immersed in CSF, this body fluid could be used as a source of tumor metabolites and other biomarkers [126]. This routinely applied clinical intervention could amplify the chances of identifying specific targets with minimal sample invasion. However, further studies with larger cohorts and the standardization of detection approaches are needed to enhance the potential of liquid biopsy applied to GBM [127].

9.3 Mathematical models and IA applications on diagnosis

Mathematical models may provide useful information about GBM cell invasion dynamics of prognostic value, which could facilitate the prediction of tumor progression [128]. MRIs contain structural and functional information for diagnosis and can be analyzed by machine-learning algorithms.

Artificial intelligence applied to MRI data could be relevant in the future for diagnosing and prognosticating patients with GBM. However, as gliomas are relatively rare, it is challenging to generate a large amount of clinical data necessary to develop these tools [129]. Nonetheless, certain models and algorithms have already reached diagnostic accuracies higher than 80% [130]. These approaches could also become an excellent non-invasive method for identifying GBM subtypes [131].

9.4 Nanodelivery for GBM treatment

Many pharmacological therapies are inefficient due to the hydrophilic nature of the compound and its low capacity to go through the BBB. Nanodelivery is designed to optimize drug pharmacokinetics across the BBB. The design of nanoparticles (NPs) includes proper factors to assimilate the therapeutic agent and enhance its outcome. Hence, a wide range of nanodelivery systems is being investigated in interventional trials. CMV therapy might be tested for nanodelivery, although there are uncertain values for toxicity and immunogenicity that should be solved [109, 132].

A current trial of siRNA NPs describes an example of liposomal carriers [119]. Another interesting application of nanodelivery comes with autophagy modulators found in natural products, such as resveratrol, curcumin, capsaicin, and others. The limited availability of these phenolic compounds relies on their low solubility. This can be overcome by administering NPs that form complexes and interactions to enhance their solubility and potential efficacy [133].

The promising outcomes from nanodelivery approaches may enhance specific targeting, the reversion of drug resistance, a reduction in the adverse effects, and a more prolonged circulation time [134]. Nevertheless, nanodelivery is still a pilot-stage field that should be further researched and documented before approval. The lack of treatments based on nanodelivery relies on the extended testing they require, the lack of standardization of the nanotoxicological assays, and the manufacturing costs of the techniques [135].