Molecular and Cellular Complexity of Glioma. Focus on Tumour Microenvironment and the Use of Molecular and Imaging Biomarkers to Overcome Treatment Resistance

This review highlights the importance and the complexity of tumour biology and microenvironment in the progression and therapy resistance of glioma. Specific gene mutations, the possible functions of several non-coding microRNAs and the intra-tumour and inter-tumour heterogeneity of cell types contribute to limit the efficacy of the actual therapeutic options. In this scenario, identification of molecular biomarkers of response and the use of multimodal in vivo imaging and in particular the Positron Emission Tomography (PET) based molecular approach, can help identifying glioma features and the modifications occurring during therapy at a regional level. Indeed, a better understanding of tumor heterogeneity and the development of diagnostic procedures can favor the identification of a cluster of patients for personalized medicine in order to improve the survival and their quality of life.


Background
Glioma is the most diffused Central Nervous System (CNS) tumour in adults that accounts for the 75% of all the primary brain cancers [1]. Glioblastoma (GBM, WHO classification IV grade glioma), covers up to 54% of all gliomas and 16% of the total primary brain tumours [2]. The standard therapy relies on the Stupp protocol, officially proposed in 2005 [3,4]. The Stupp protocol is based on tumour surgical resection, followed by postoperative radiotherapy (RT), 60 Gy/30 fractions as standard, and concomitant plus adjuvant Temozolomide (TMZ) treatment [5]. Despite an increase in overall survival, the prognosis remains poor because of the development of GBM resistance to TMZ [6]. Alternative treatments are under evaluation, but the high intra-and inter-tumour heterogeneity of GBM and its intracranial localisation make the development of new therapies for this tumour a big challenge [7]. An important role in lesion heterogeneity is played by the tumour microenvironment (TME), composed by different cell populations, as immune cells, fibroblasts, precursor cells, endothelial cells, signalling molecules, and extracellular matrix (ECM) components [8,9]. In recent years, the use of new epigenetic molecules, such as non-coding RNA and microRNAs (miRNAs), has emerged

MicroRNA and GBM Microenvironment
MiRNAs are an endogenous class of about 20 nucleotides in length non-coding RNAs which are very abundant in the cells, with several regulatory functions. Indeed, miRNAs have a main role in the regulation of mRNA transcription by silencing or the degradation of target sequences. The complementarity of the mature miRNA sequence and the seed sequence (i.e., the sequence recognised by the miRNA on its putative target) determines the physiological role of the miRNA within the cell. Therefore, on the basis of the complementarity of the miRNA sequence with that of the seed sequence, the formation of a double strand allows the degradation of the target mRNA or its silencing due to the translation block [68,69]. The expression level of a selected miRNA could be easily determined by real time-PCR (RT-PCR) amplification, either on a tissue biopsy sample or on human biofluids [70,71]. Furthermore, in situ hybridization could be helpful in the localization of a specific miRNA inside the cells or tissues and in its quantification [72]. This feature allows miRNAs to be used either as diagnostic or prognostic biomarkers able to indicate the course of the pathology [73]. Moreover, the possibility of exogenous modulation on miRNAs makes them interesting molecules as a potential therapeutic target. Indeed, in vitro experiments on cell lines using synthetic oligonucleotides with a reverse and complement sequence of a miRNA could be used as an antisense sequence, to silence an endogenous molecule [74]. Vice versa, if a GBM has a decreased miRNA, the use of a synthetic sequence "mimicking" the silenced miRNA could re-establish its expression level, preventing the further development, growth, or metastasis of the GBM cells [17].
The main challenge in miRNA use is to identify the correct single or group of miRNAs with a primary role in GBM onset, development, or response to therapy. Bioinformatics in silico analyses help the biologists in the selection of miRNAs with a differential expression in GBM compared to normal tissues (diagnostic miRNAs) or those with a differential expression in different grade of the pathology, or those able to predict the prognosis of the patients (prognostic miRNAs) [75]. For instance, in 2013 Moller and colleagues proposed a miRNA signature of 256 upregulated miRNAs (notably miR-10b, miR-17-92 cluster, miR-21, and miR-93) and 95 downregulated miRNAs (including miR-7, miR-34a, miR-128, and miR-137) for the diagnosis of GBM [76,77]. Furthermore, individual miRNAs have been correlated with different glioma stages, such as in a work of Malzkorn and colleagues, where 12 upregulated miRNAs (miR-9, miR-15a, miR-16, miR-17, miR-19a, miR-20a, miR-21, miR-25, miR-28, miR-130b, miR-140, and miR-210) and two downregulated miRNAs (miR-184 and miR-328) were found to be associated to the spontaneous progression of GBM from the WHO grade II to the WHO grade IV [78].  [112] miR-148a and miR-31 Up Promotes angiogenesis factor inhibiting hypoxia-inducible factor 1 (FIH1), and HIF1α and Notch. [113] 3. Chemoresistance and Treatments

Resistance to Temozolomide
Lack of response to TMZ may be present from the beginning of therapy or acquired during the treatment. The reasons are complex and only partially known, but a non-complete response favours tumour relapse and recurrence [114]. The MGMT promoter methylation status is a key parameter for counteracting TMZ cytotoxicity. Low MGMT promoter methylation entails a higher enzyme activity, a phenomenon which helps to acquire resistance to TMZ [115].
Cells lacking MGMT or which have defective Homologous Replication (HR) are responsive to TMZ treatment [116]. For this reason, MGMT is indicated as a biomarker for TMZ treatment response in a clinical setting [116][117][118][119].
Another mechanism by which TMZ affects DNA repair systems is mediated by N3-methyladenine and N7-methylguanine adducts, which are Base Excision Repair process (BER) targets. In fact, these residues induce Poly (ADP-ribose) polymerase (PARP) nuclear enzyme activation, which, by the PARylation (poly-ADP-ribose chain synthesis), recruit three components on the DSBs areas (XRCC1, pol-beta, and DNA ligase), by triggering the BER activation [120]. Since the PARP plays a key role in the BER mechanism, PARP inhibitors (PARPi) have been proposed and studied in combination to TMZ treatment for overcoming the currently defined resistance to TMZ [121].
Briefly, in a recent study of Higuchi et al., TMZ sensitivity has been restored by treating mismatch repair mechanism (MMR)-deficient resistant gliomas with PARPi by a BER-independent mechanism [120].
Beyond DNA, post-translational modifications on histones are also able to play a pivotal role in mediating gene expression, and this phenomenon might help to elucidate why MGMT is not always an efficient response biomarker to TMZ treatment. In fact, in a work of Chen et al. in 2018, an enhancer was described to activate MGMT gene expression, despite its highly methylated promoter. In detail, they have identified an enhancer sequence positioned between the promoters of MGMT and of Ki67 (a widely recognised marker of cell proliferation, also in GBM) which was able to activate not only MGMT expression but also Ki67 expression, by favouring GBM proliferation, both in vitro and in vivo. Moreover, the authors have also shown that the acetylation levels of the nucleosome H3K27, which borders the K-M enhancer, are capable of promoting MGMT expression in spite of its methylated promoter. They have observed the abrogation or the delay of resistance to TMZ treatment through the inhibition of the K-M enhancer [122].
Another important factor involved in resistance to TMZ is represented by the BER system activity, which counteracts cytotoxic effects of TMZ: when BER is abrogated, TMZ is able to exert its cytotoxic power [123]. This year, Belter et al. have shown that TMZ is able to totally augment the rates of the main epigenetic marker contained in tumour cells' DNA, the 5-methylcytosine (m5C), in different GBM cell lines, by inducing a cellular hypermethylation state. During the reaction, S-adenosyl methionine (SAM) results in the unique Alkyl residues donor [124].
TMZ, beyond its alkylating effect, is able to induce other cellular mechanisms, such as the Endoplasmic Reticulum (ER) stress, Reactive Oxygen Species (ROS) production, and autophagy activation [125]. The tight relationship between the ER-stress and chemoresistance to TMZ has been already described, since TMZ toxicity also relies on promoting the chaperone protein folding role and on enhancing the degradation pathways, as the ER-mediated and the autophagic ones [126].
In a recent work, GBM MGMT-deficient cells with acquired resistance to TMZ treatment showed increased expression of dynein, cytoplasmic 2, heavy chain 1 (DHC2 or DYNC2H1), which is strictly related to TMZ-mediated cytoskeletal rearrangements [127].
In another recent study, the role of the bradykinin-bradykinin receptor B1 (BDKRB1) axis has been shown to modulate some GBM genes and pathways, as the Aquaporin-4 (AQP4/aqp4) expression and the calcium release within the cytosol. This process leads the mitogen-activated protein kinase (MAPK) pathway activation. The authors have demonstrated that the bradykinin/BDKRB1 increased the aqp4 expression through BDKRB1 mediated Ca2+ influx, which induces MEK4 phosphorylation and, consequently the downstream ERK1/2-NF-κB induction both in vitro and in vivo in human and murine cells. Overall, these data have suggested the key-role of bradykinin-BDKRB1-aqp4 action on cytoskeleton and morphological changes, as GBM oedema formation and its invasive and migratory power [128].

Pharmacological Strategies to Overcome GBM Resistance
Several therapeutic strategies have been developed in order to prolong patients' overall survival (OS) and to overcome GBM resistance. Some of these therapies have completed clinical trials or are under clinical evaluation alone or in combination with TMZ; others are still confined to a pre-clinical scenario.
The angiogenic inhibitor Bevacizumab (a humanized monoclonal antibody targeting VEGF) has been approved by the Food and Drug Administration (FDA), but not by the European Medicines Agency (EMA), for patients with recurrent GBM. Despite the initial expectation based on the mechanism of action of the drug, no effect on OS, and only an increase of response rate in a subset of patients has been reported [129]. However, in a phase III clinical trial (EORTC 26101), the addition of Bevacizumab to Lomustine in 432 GBM patients previously treated, showed an increase in the time to disease progression or death compared to Lomustine alone (median progression-free survival = 4.2 vs. 1.5 months, HR = 0.52, 95% confidence interval = 0.41-0.64) [130].
Lomustine is a cross-linking agent able to produce amino acid carbonylation. In a recent study, Herrlinger et al. showed in a phase III randomized trial that Lomustine (CCNU) associated with RT/TMZ prolonged OS in comparison with an RT/TMZ standard regimen (median OS of 48.1 months versus 31.4 months, respectively) [131,132].
The block of hyperactivated Epidermal Growth Factor Receptors (EGFRs) represents another potential strategy. Specific interaction between Nimotuzumab, a humanized anti-EGFR antibody, and the constitutively activated EGFRvIII receptor, was able to inhibit EGFR signalling intracellular cascade, thus abrogating tumour proliferation, angiogenesis, and chemoresistance. The drug has been evaluated in different clinical trials; however, its effect is still controversial [134]. Recently, Du and colleagues evaluated in a phase II trial (NCT03388372) the efficacy of Nimotuzumab associated with TMZ/RT and concomitant TMZ in patients with newly diagnosed GBM carrying EGFRvIII [135]. The median OS and PFS were 24.5 and 11.9 months, respectively. These data may be of interest for a phase III clinical randomized controlled trial [136].
The PARP family of enzymes is an attractive target for the sensitization of GBM to chemotherapy because of their activity on DNA repair and metabolism. PARP inhibitors (PARPi) have been evaluated in different clinical trials, in association with TMZ, for treating newly diagnosed and recurrent glioblastoma patients [120]. In preclinical studies the association of TMZ and Veliparib showed promising results in GBM models [137,138]. In addition, Veliparib increased the TMZ cytotoxic effect in the PDX of newly diagnosed GBM with high MGMT promoter methylation [139]. However, in a phase II clinical study, Veliparib added to an RT/TMZ standard treatment in unmethylated MGMT patients with newly diagnosed GBM resulted in no ameliorated outcomes compared to the standard regimen [140]. A phase III clinical trial (NCT02152982) in which this drug is added to TMZ in treating MGMT-methylated, newly diagnosed GBM patients, is still ongoing [141]. Another PARPi, Olaparib, has been included in a phase I dose escalation study (OPARATIC trial) in combination to the standard therapy in recurrent GBM patients. The results of the study have been encouraging, with 45% of patients showing a PFS at six months. At the moment, researches are focused on the identification of potential markers of treatment efficacy [142]. Anti-TIM-3 and Anti-PD-1 TIM-3 and PD-1 [169] Anti-TIGIT and anti-PD-1 TIGIT and PD-1 [170] * Combination with Temozolomide.

Role of miRNAs in GBM Chemoresistance
The current grading system, based on the histological classification, does not give any indication of the correct therapeutic approach for the patients [171]. The identification of drug sensitive patients with an affordable, reliable, and easy-to-use method could be helpful in addressing the patients to the right therapeutic option. In this view, miRNAs could be presented as indicative molecules for patient selection for a specific treatment. The choice of the right miRNA to be used as a drug-resistance associated molecule is still a challenge. The recent literature proposes several examples of miRNAs that could be used as potential biomarkers for the patients' classification.
Looking for "microRNAs and glioblastoma and chemoresistance" publications of the last five years in PubMed, we found 30 papers. Two have been shelved because they speak of long non-coding RNA, and one was withdrawn by the authors before the publication. Twenty-seven manuscripts were analyzed to understand the role of miRNAs in the chemoresistance development (summarized in Table 3 and Figure 1). Most of the publications describe the role of single miRNA in this process, without proposing a miRNA signature associated with chemoresistance.

In vivo Imaging for the Study of GBM Heterogeneity and Microenvironment
Imaging techniques cover a pivotal role in the management of GBM patients [211]. In particular, in the clinical scenario the combinations of diffusion-weighted MRI (DWI-MRI) and amino acidbased PET techniques are challenging diagnostic procedures for the clinical management of patients. In addition, the multimodal MR-PET imaging offers new insights for a better understanding of GBM heterogeneity [212,213]. In a clinical setting, the conventional MRI sequences include: T1-weighted alone or with Gadolinium contrast agents (T1w-Gd) to evaluate the blood-brain barrier (BBB) permeability and tumour vascularization and T2-weighted and fluid attenuation inversion recovery (FLAIR) to define tumour oedema, border, and presence of necrosis [212]. Emerging techniques give complementary information. Diffusion-weighted MRI (DWI) provides information about the reorganization of cellular tissue and tumour invasion, by calculating the apparent diffusion coefficient (ADC) and its variant diffusion-tensor MRI (DTI) permits to distinguish the dominant white matter tracts through the tractographic information and to obtain and foresee the GBM invasive power. Perfusion-weighted imaging (PWI), with the perfusion acquisition modalities Dynamic susceptibility contrast (DSC) and Dynamic contrast-enhanced (DCE), furnishes data on cerebral blood flow and cerebral blood volume [214]. In addition, arterial spin labelling (ASL), using magnetically labelled arterial blood water protons as an endogenous tracer, is able to evaluate perfusion. Finally, MR spectroscopy (MRS) allows for the identification of the regional distribution of tissue metabolites [215]. However, despite advances in methodologies, the identification of necrosis, tumour margin, recurrence, and grading still represents a critical issue [216]. The use of PET and selected radiopharmaceuticals has allowed to image in vivo different biological features of tissue, such as glucose metabolism, cell proliferation, tissue hypoxia, inflammation, and matrix Several papers proposed a single miRNA with a role in the development of glioblastoma resistance to Temozolomide; one of the most frequent reported in the literature is miR-1238. This miRNA was found upregulated in TMZ-resistant glioblastoma cell lines and their exosomes, and in clinical tissues and sera of patients; its overexpression was able to modulate the CAV1/EGFR pathway. The presence of this miRNA in exosomes could influence the behaviour of the surrounding tissue, leading to the spreading of the resistant phenotype to surrounding tumour cells [172]. It has been already discussed that exosomal miRNAs allow the communication of the tumour cells to the neighbouring ones, as occurs with exosomal miR-151. Indeed, exosomal miR-151a enhances chemosensitivity to TMZ in drug-resistant GBM, possibly controlling XRCC4-mediated DNA repair. XRCC4 encodes for a vital protein of the double strand break repairing process: XRCC4 protein forms with DNA ligase IV a heterodimer that covalently joins the broken DNA ends to help cell survival. XRCC4 expression levels are found significantly downregulated in glioma tissue, indicating crucial roles of XRCC4 in brain carcinogenesis [173].
The exosomal communication among cells has been also been proven to increase the tumorigenicity of the glioma stem cells, by demonstrating the release of miR-1587 from glioma-associated mesenchymal stem cells [174]. This miRNA is able to inhibit the hormone receptor co-repressor-1 (NCOR1), a tumour suppressor of glioblastoma cancer stem cells, after extracellular vesicular transfers [82].
Indeed, chemoresistance to TMZ in glioblastoma could also be mediated by the presence of several CSCs, which are reported to be CD133+ cells. This population is associated with poor prognosis and resistance to chemotherapy [175]. In CD133+ cells deriving from T98G and U87MG tumours, miR-29a is significantly downregulated. The miR-29a overexpression improved sensitivity to cisplatin treatment in these cells and suppressed tumour growth in animal model of GBM [176]. A miRNA of the same family, miR-29c, has been demonstrated to be a sensitizer for TMZ treatment of resistant glioblastoma. Indeed, miR-29c targets specifically Sp1, thus reducing MGMT expression. In this way, the reduction of this enzyme allows the increased efficacy of TMZ treatment. The overexpression of miR-29c correlates with a good prognosis, suggesting the use of circulating miR-29c low levels as a prognostic biomarker of resistance [177].
The same AKT/NF-κB pathway is also a final target of TMZ-upregulated miR-146b-5p [181]. By modulating the expression of tumour necrosis factor receptor-associated factor 6 (TRAF6) protein, miR-146b-5p enhanced the resistance against TMZ. TRAF6 is a signal transducer of TNFR family and Toll/interleukin-1 (IL-1) and its overexpression promotes the tumour cell proliferation in several solid tumours, i.e., breast, colon, and lung cancer [182]. TRAF6 is a known regulator of chemoresistance; it has been found upregulated in TMZ-resistant GBM tissues, and its overexpression increases the resistance of colon cancer cells to 5-Fluorouracil and to Bortezomib, respectively [183]. Overexpression of miR-146b-5p or TRAF6 knockdown significantly modulates the AKT/NF-κB pathway, decreasing the level of p-AKT and p-p65, activators of NF-κB [181]. The AKT/NF-κB pathway is involved in the regulation of autophagy, a process that emerging evidences suggests contributes to the development of TMZ resistance [184].
The pathway of PI3K/Akt is also target of miR-223: the exogenous expression of this miRNA promotes the survival of the cells exposed to TMZ by suppressing the PAX6 target gene, a known tumour suppressor in glioma cells [185,186].
In computational analyses it was shown that miR-93 and -193 could target Cyclin D1, a major regulator of cell cycle progression. Indeed, this protein regulates the progression of the cells from G1 to S phase. These two miRs also decreased cell cycling quiescence and induced resistance to TMZ [187].
The miR-7-5p has been described as a tumour suppressor in multiple cancers and it has been found significantly downregulated in TMZ resistant LN229 cells. Further experiments demonstrated that miR-7-5p has a role in the regulation of Yin Yang 1 (YY1), a transcription factor acting as a tumour promoter [188,189]. In glioblastoma, YY1 is able to interact with the NF-κB family member RelB, thus controlling both the proliferation of the cells as well as the pro-inflammatory cytokines leading to infiltration of glioma-associated macrophages [189]. The silencing of YY1 in cisplatin-resistant glioblastoma LN-229 cells was also obtained by overexpressing miR-186 which sensitizes the cells to drug treatment [190].
The miRNAs are also involved in the modulation of the hypoxia-induced autophagy process in different cancers. In glioblastoma, HIF1α increase and miR-224-3p decrease were simultaneously observed. It has been described that HIF1α knockout inhibited chemosensitivity by negatively regulating miR-224-3p expression under hypoxic condition, while hypoxia induced the expression of ATG5, a transcription factor possibly involved in the autophagic process. The researchers described a HIF-1α/miR-224-3p/ATG5 pathway that, in hypoxic condition, could influence the chemosensitivity of LN-229 cells to TMZ both in vitro and in the animal model [191].
Another miRNA involved in the control of autophagy is miR-519a. In U87 cells, made resistant to TMZ, it was found downregulated in comparison to the parental cell line. The exogenous overexpression of miR-519a dramatically enhanced TMZ-induced autophagy and apoptotic cell death through the inhibition of the STAT3/Bcl-2/Beclin-1 pathway [192].
The miR-138 is also upregulated in TMZ-resistant GBM, both in cell lines and in human tissue samples. BIM mRNA is the direct target of this miRNA, and miR-138/BIM axis promotes cell survival after TMZ treatment by regulating the autophagy process [193].
Another miRNA involved in the control of both chemoresistance and metastasis is miR-139, being able to control glial fibrillary acidic protein (GFAP) gene expression. GFAP encodes for a protein of the cytoskeleton, mainly expressed in the astrocytes. This protein is highly expressed in high-grade brain tumours and involved in aggressive phenotype [195]. The polymorphisms of GFAP, especially those affecting the miR-139 tumour suppressor seed, have been found and correlated with the malignant phenotype. Indeed, the lack of miR-139 binding on GFAP mRNA increases the expression of GFAP protein, which in turn favours the assembly and stabilisation of the cytoskeleton and increases the chemoresistance to cytotoxic drug [196].
Possible changes in miRNA processing, such as the dysregulation of miR-221/222 processing by apurinic/apyrimidinic endonuclease 1 (APE1) mRNA, could also impact on chemoresistance process. APE1 is a member of the base excision repair process; it acts as a master regulator of the cellular response to genotoxic damage. In response to oxidative damage, APE1 stimulates the DNA repair by promoting the initiation of transcription of SIRT1, of the multidrug resistance gene (MDR1) and the phosphatase and tensin homolog (PTEN) tumour suppressor [197,198]. The miR-221 and miR-222, belonging to a polycistronic cluster, are involved in the regulation of PTEN expression, and have been already associated with chemoresistance in breast and in lung cancer [199][200][201]. APE1, being able to bind double-stranded RNA, can bind a precursor form of miR-221/222, and the silencing of APE1 results in decreased maturation of the miR-221/222 cluster. Upregulation of APE1, leading to an increase in miR-221/222 cluster expression, has been associated with the development of radio-resistance, cancer growth stimulation and invasion by inhibiting PTEN and with the development of multidrug resistance and altered response to chemotherapy [200][201][202][203]. Thus, it seems that interfering with APE1/miR-221/222 cluster regulation could impact on the chemosensitization of the GBM cells.
The oestrogen responsive miR-191 has been correlated with the aggressive mesenchymal phenotype of GBM [204]. Its expression was found to be altered by the aromatase inhibitor Letrozol in glioma cells [205]. Aromatase enzymes are necessary for the synthesis of the oestrogen hormone which, after binding to oestrogen receptor (ER) alpha and beta, interacts with oestrogen responsive elements (ERE) in the promoter region of miR-191, mediating the chemoresistance effect observed in several cancers and in GBM. Letrozole treatment reduces oestrogen-mediated binding of the ER to the ERE, reducing miR-191 expression. The evaluation of miR-191 function as a potential anti-tumoral drug and as a sensitizer of GBM cells should be still evaluated [206]. miR-101 is significantly downregulated in TMZ-resistant GBM cells (TMZ resistant forms of U251 and A172) and human samples and over-expression of miR-101 sensitizes resistant GBM cells to TMZ through the downregulation of glycogen synthase kinase 3β (GSK3β). This gene encodes for an enzyme involved in the regulation of protein synthesis, glycogen metabolism, cell proliferation and survival, and chemoresistance [207]. Several studies show that GSK3β plays a critical role in the development and progression of various malignancies and that targeting GSK3β may represent a novel strategy for the treatment of chemoresistant cancers [208,209]. An attempt of description of miRNA profiles able to select TMZ-sensitive from resistant GBM patients was proposed in two papers. In 2015, by in silico analysis, the authors proposed seven miRNAs, four of which (miR-1280, miR-1238, miR-938, and miR-423-5p) overexpressed in TMZ-resistant samples and three of which (let-7i, miR-151-3p, and miR-93) downregulated in TMZ-sensitive GBM human samples [179]. These miRNAs were used to classify an independent dataset of samples, demonstrating that two over seven are able to correctly classify TMZ-resistant patients. No hypothesis was made on possible miRNA targets, that could justify their different expression in TMZ-sensitive or -resistant cells. In 2016, another miRNA profile was proposed on U251 and A172, made resistant to TMZ. In this paper, the authors analyzed microarray results of these two cell lines and found common miRNAs, associated with TMZ-resistant phenotype. In particular, they found miR-9, -182 and -221 to be upregulated, while miR-29c, -93, -101, and -130a were to be downregulated. These results were confirmed in 12 TMZ-resistant human tumour samples compared to primary tumours. Focusing on miR-101, the in silico analysis of possible targets revealed that this miRNA could control the expression of GSK3β, VEGF, and COX-2. In particular, the authors demonstrated that miR-101 is able to downregulate the serine/threonine protein kinase GSK3β. The loss of miR-101, observed in TMZ-resistant cells, caused an increase in GSK3β expression, leading to drug resistance development and the inhibition of apoptosis. This regulation could be possibly due to the control of GSK3β on the methylation of the MGMT promoter. Loss of miR-101 was correlated with poor GBM patient prognosis [210].

In Vivo Imaging for the Study of GBM Heterogeneity and Microenvironment
Imaging techniques cover a pivotal role in the management of GBM patients [211]. In particular, in the clinical scenario the combinations of diffusion-weighted MRI (DWI-MRI) and amino acid-based PET techniques are challenging diagnostic procedures for the clinical management of patients. In addition, the multimodal MR-PET imaging offers new insights for a better understanding of GBM heterogeneity [212,213]. In a clinical setting, the conventional MRI sequences include: T1-weighted alone or with Gadolinium contrast agents (T1w-Gd) to evaluate the blood-brain barrier (BBB) permeability and tumour vascularization and T2-weighted and fluid attenuation inversion recovery (FLAIR) to define tumour oedema, border, and presence of necrosis [212]. Emerging techniques give complementary information. Diffusion-weighted MRI (DWI) provides information about the reorganization of cellular tissue and tumour invasion, by calculating the apparent diffusion coefficient (ADC) and its variant diffusion-tensor MRI (DTI) permits to distinguish the dominant white matter tracts through the tractographic information and to obtain and foresee the GBM invasive power. Perfusion-weighted imaging (PWI), with the perfusion acquisition modalities Dynamic susceptibility contrast (DSC) and Dynamic contrast-enhanced (DCE), furnishes data on cerebral blood flow and cerebral blood volume [214]. In addition, arterial spin labelling (ASL), using magnetically labelled arterial blood water protons as an endogenous tracer, is able to evaluate perfusion. Finally, MR spectroscopy (MRS) allows for the identification of the regional distribution of tissue metabolites [215]. However, despite advances in methodologies, the identification of necrosis, tumour margin, recurrence, and grading still represents a critical issue [216]. The use of PET and selected radiopharmaceuticals has allowed to image in vivo different biological features of tissue, such as glucose metabolism, cell proliferation, tissue hypoxia, inflammation, and matrix metalloproteinases that are associated with the tumour or tumour microenvironment. Here, are discussed the main cancer hallmarks relevant for GBM and TME and the potential use of radiopharmaceuticals for their in vivo measurement both at clinical and preclinical levels, after a research of the publications of the last five years in PubMed. Radiotracers and targets are summarized in Table 4. In the following paragraphs, different biological features of cancer and tumour microenvironment are described together with specific radiopharmaceuticals already available or under development for the in vivo imaging.

Cancer and Metabolism
Altered glucose metabolism is a characteristic of invasive cancers. Since 1927, Otto Warburg noticed that cancer cells preferentially converted glucose into lactic acid instead of pyruvate even in presence of sufficient oxygen, for this reason the "Warburg Effect" is also called aerobic glycolysis [272,273]. On this basis, 2-deoxy-2-[fluorine-18]fluoro-D-glucose ([ 18 F]FDG) is the most used radiotracer in cancer; indeed, most cancers display increased glucose uptake and phosphorylation and there is a tight correlation between [ 18 F]FDG uptake and cell viability [217]. Unfortunately, the brain tumour imaging represents a unique challenge because of the high background uptake due to normal grey matter which makes it very difficult to efficiently visualize with [ 18 F]FDG PET [274].
Nevertheless, glioma shows several altered metabolic pathways which can be exploited with other markers to study cancer progression and to develop new metabolism-based therapeutic strategies [275]. Glutamine (Gln) is the second most abundant nutrient (after glucose) in blood circulation and supplies most of the carbon, nitrogen, and free energy and reduced equivalents necessary to support cell growth and division [276]. Gln is used by proliferating tumour cells to generate α-ketoglutarate which can be metabolised through the tricarboxylic acid (TCA) cycle to generate oxaloacetate (OAA) [277]. Tardito et al., analysing the metabolic role of Gln in sustaining the growth both in vitro (in human established GBM cell lines, in primary GBM stem-like cells, and in normal astrocytes) and in vivo (xenograft murine models and in GBM patients), found a tight collaboration between cancer cells and astrocytes. Indeed, Glutamine Synthetase (GS) is found in the majority of human GBM and its expression is associated with poor prognosis [278,279]. GS expression is greatly variable in tumour cells ranging from negative to high as in normal astrocytes. GS-positive astrocytes and/or GBM cells (potentially GSC) excrete Gln that supports the growth of GS-negative GBM cells [278]. Using imaging, Gln metabolism can be visualized with PET and 18 F-(2S,4R)4-fluoroglutamine ([ 18 F]F-Gln) or with Nuclear Magnetic Resonance (NMR) Spectroscopy [280]. In 2015, Venneti et al. characterised [ 18 F]F-Gln uptake both in healthy mice and in several orthotopic murine models of glioma. Compared to [ 18 F]FDG, [ 18 F]F-Gln uptake was not observed in the brain of healthy mice and in animals with a permeable blood-brain barrier or neuroinflammation. On the contrary, [ 18 F]F-Gln was able to identify very well both glioma margin in different murine models and tumor reduction after five days of treatment with TMZ. In addition, [ 18 F]F-Gln avidity was observed in patients with the progressive disease [218]. The ability to visualize orthotopic glioma was also confirmed by other groups [219]. Clinical safety and pharmacokinetics in several tumours among glioma were also assessed. In particular, high-grade glioma with IDH mutation showed high uptake of [ 18 F]F-Gln. Moreover, authors observed that [ 18 F]F-Gln-avid tumours were uniformly [ 18 F]FDG-avid but not vice versa [220].
At the moment, using NMRS is possible to quantify the abundance of the glutamate and glutamine pool defined as Glx and it has been observed an increase in Glx in high grade glioma or in invasive regions [281,282]. In other studies, authors have investigated the relationship between the presence of the IDH mutation and the presence of Glx and they found an inverse correlation between the presence of the oncometabolite 2-hydroxglutarate (2HG) and a significant decrease of Glx [283,284]. Heiland et al. associated the presence of specific metabolites to the subtypes of GBM on the basis of the gene mutations, in detail clustering of normalised Glx (nGlx) revealed two groups: one with high nGlx being attributed to the neural subtype and one with low nGlx associated with the classical subtype [285]. Considering the great importance to specifically identify Gln metabolism, Hangel et al. set up a reconstruction method, patch-based super-resolution (PBSR), in association with 7T MRSI to separate the peak of glutamate to that of Gln [286].
Additional metabolic pathways that have received considerable attention in GBM are fatty acid synthesis (FAS) and fatty acid ß-oxidation (FAO) [287,288]. In vivo studies performed in orthotopic mouse models of malignant glioma using 13 C show that acetate contributes over half of oxidative activity within these tumours [289,290]. Kant et al. observed that FAO "high" phenotype was enriched in mesenchymal tumours, whereas proneural tumours were mainly characterized by FAO "low" phenotype indicating that a specific treatment could be used based on the GBM phenotype [287]. In addition, inhibition of either fatty acid synthesis or beta-oxidation reduces proliferation of glioma cells [291,292]. [ 11 C]Acetate is a radiotracer that is related to enhanced lipid synthesis and it is rapidly transported into cells through monocarboxylic acid transporters (MCTs) and metabolised into acetyl-CoA [293]. Kim et al. evaluated that [ 11 C]Acetate PET/CT is able to differentiate between lowand high-grade glioma and, specifically, grade III from IV. Authors also found a correlation between the expression of acetylcoenzyme A synthetase (ACSS2), glioma grade and prognosis indicating ACSS2 as a potential drug target [221]. Finally, in a preclinical study Koyasu S et al. reported that U251 tumours carrying a mutation in IDH1 gene (R132H) display a higher uptake of [ 14 C]Acetate compared to U251 wt tumours [222].
In recent years, the tryptophan (TRP) metabolic pathway emerged as an important node both for the GBM progression and for resistance to therapy because it actively contributes towards immune suppression in particular in mesenchymal and classical molecular subtypes [294]. Tryptophan metabolism appears to play the most dominant role in immune tolerance in GBM, with over a two-fold increase in Tregs and a strong trend in decreased CD8+ T cells. TRP is an essential amino acid necessary for protein synthesis. The bulk of TRP is therefore available for metabolism via four pathways to produce physiologically important metabolites including serotonin, melatonin, tryptamine, and kynurenine (KYN). Over 95% of TRP is metabolized to KYN via the kynurenine pathway (KP). Through local TRP depletion and the production of the toxic metabolites, the KP fosters an immunosuppressive tumour microenvironment [295,296]. Three distinct enzymes are able to complete the initial and rate-limiting step of the metabolism of TRP to N-formylkynurenine (NFK): indoleamine 2,3-dioxygenase 1 (IDO1), indoleamine 2,3-dioxygenase 2 (IDO2), and tryptophan 2,3-dioxygenase 2 (TDO2) [224]. IDO1 is expressed in a wide variety of cell types, including dendritic cells, endothelial cells, macrophages, fibroblasts, and mesenchymal stromal cells, as well as in neurons and in cancer cells themselves [297]. Moreover, as observed by Ozawa et al., IDO1 is also highly expressed in glioma stem cells [298]. Given the importance of the kynurenine pathway in numerous cancers, it is increasingly recognised as a prominent target for cancer imaging and therapy. Currently, radiotracer alpha-[ 11 C]methyl-L-tryptophan ([ 11 C]AMT) is used for PET imaging of the IDO-mediated kynurenine pathway in clinical and preclinical research [223,224]. In two studies, Bosnyak et al. evaluated the ability of [ 11 C]AMT to differentiate between meningioma and grade II and grade III gliomas and the prognostic role of the radiotracer, finding that GBM patients with higher pre-treatment tumoral tryptophan uptake, expressed as tumour/cortex SUV-ratios, showed longer survival [225,299].
In two different studies, John et l. have used combined MRI and [ 11 C]AMT PET for discriminating non-enhancing GBM regions which are challenging to detect with MRI only. They have suggested the importance of the amino acid uptake as a key player in distinguishing metabolically active subregions in the tumour from the areas with necrosis and oedema, which correspond, respectively, to enhancing and non-enhancing tumour zones. In fact, GBM showed very heterogeneous regions within the tumour with high-uptake regions often extending into the non-enhancing brain with high cellularity and the non-resection of these areas predicted the site of post-treatment progression. Furthermore, they have suggested the high uptake of tryptophan in contrast-enhancing subregions as an independent imaging biomarker for detecting the prolonged survival in newly diagnosed GBM patients [212,226]. Guastella et al. observed that PDXs generated by injection in mice of GBM cells or tumour fragments derived from patients with different uptake of [ 11 C]AMT and expression of TRP pathway markers recapitulate the same TRP metabolic characteristics of the primary patient tumour underling as these models can be used to test new therapeutic approaches [224].
Nevertheless these interesting results, the use of [ 11 C]AMT is limited to a few PET centres due to its short half-life (20 min) and the complex radiosynthesis [300]. So, a number of fluorine-18 labelled tryptophan analogues have been developed for cancer imaging, among them L-1-[ 18 F]fluoroethyl-tryptophan (L-1-[ 18 F]FETrp). L-1-[ 18 F]FETrp uptake was tested in several cancer cell lines, both in vitro and in vivo, including 73C-glioma cells. Intracranial brain 73C-glioma transplanted mice confirmed the capability of L-1-[ 18 F]FETrp to pass the BBB and to efficiently visualize the tumour suggesting that L-1-[ 18 F]FETrp is a promising radiotracer for PET imaging of cancer [227].
The most widely used radiolabelled amino acid is the essential amino acid methionine, labelled with 11 C, which is taken up by the L-type amino acid transporters (LAT) 1 (SLC7A5) and 2 (SLC7A8) and incorporated into the protein. [ 18 F]fluoroethyltyrosine (FET) is a substrate for LAT1 and LAT2 and is widely used in place of methionine considering that it has a longer half-life (109 min) compared to that of C-11 (20 min) [301]. Both radiotracers are able to visualise low and high glioma but with difficulty it is possible to differentiate them. The radioligand [ 18 F]3 -deoxy-3 -fluorothymidine (FLT), substrate of thymidine kinase that is overexpressed during S-phase of cell cycle, reflects cell proliferation [302]. Since [ 18 F]FLT does not cross intact the BBB, it does not accumulate in low-grade tumours or stable lesions but it is detectable in high-grade (grade III or IV) tumours with a disrupted BBB [303]. It is used for imaging early treatment response and for predicting clinical outcome in brain tumours [231]. It is widely used in preclinical research to evaluate early response to standard or new therapeutic approaches [232][233][234][235][236].
Because of enhanced cell proliferation and resulting elevated levels of cell membrane synthesis during tumorigenesis, choline metabolism is also involved in brain cancer. Choline is a precursor for the synthesis of the phospholipid components of the cell membrane. Cellular choline is phosphorylated by choline kinase (CK) yielding phosphocholine (PCho), which further reacts with CTP to yield CDP-choline. The de novo synthesis (Kennedy pathway) of phosphatidylcholine (PC) then results from the reaction of CDP-choline with diacylglycerol. In the clinical setting, two imaging modalities allow for the examination of choline met abolism, namely proton magnetic resonance spectroscopy ( 1 H-MRS) and PET using [ 11 C]Choline. MRS provides non-invasive measurements of tissue concentrations of metabolites, such as total choline-containing compounds (tCho), total creatinine (tCr), and N-acetylaspartate (NAA). Increased tCho levels result from elevated cell membrane turnover and cellular density [237].
Wehrl et al. performed a comparative in vivo study assessing tCho levels by MRS using chemical shift imaging (CSI) and [ 11 C]Choline metabolism by PET with morphologic parameters, in a murine orthotopic glioma mouse model obtained with murine SMA-560 (spontaneous murine astrocytoma) glioma cells. Authors found an increase in [ 11 C]Choline uptake in the tumour paralleled to a decrease in tCho levels. Only a small overlap between tumour volumes, as identified by CSI and PET, was found; in detail, CSI highlights the areas of high choline concentration, which seems to be localized to the tumour rim, whereas [ 11 C]Choline PET identifies the regions of high choline turnover. Therefore, MRS and PET seems to give complementary information about choline metabolism [237].
MRS and DTI-MRI are used to identify metabolic changes in the invasive margin of glioblastomas and to monitor therapy response [281,304]. In a recent study, Takei et  F]FDG uptake was higher in the anaplastic astrocytoma (AA) IDH-wt group than in the IDH-mut group, whereas in GBM only [ 11 C]Methionine and [ 11 C]Choline uptake were significantly higher in IDH-wt group than in IDH-mut group suggesting that PET is able to differentiate IDH-wt and IDH-mut tumours [238].

Cancer and Hypoxia
Hypoxia is a key driver of tumour adaptation promoting tumour progression and resistance to therapy. In particular, hypoxia and hypoxia-related factors, such as HIF-1α, are associated with pseudo-palisading necrotic regions that protect stem cells niche, contributing to the aggressive profile of these tumours. HIF-1α protein expression in glioblastoma promotes angiogenesis by inducing VEGF [305]. For these reasons, it is pivotal to have therapies that counteract this situation and imaging techniques that allow one to identify hypoxic regions within the tumour. In a systematic review, Raccagni [239]. Among these, [ 64 Cu]ATSM results are very interesting for GBM. This radiotracer can be used both as an imaging and as a therapeutic agent, and it is considered as an independent prognostic biomarker of survival in high-grade glioma [240,241]. 64 Cu is a promising theranostic radionuclide owing to its suitable half-life (12.7 h) with multiple decay modes, including β + (18%), which is used for PET imaging,

Cancer and Inflammation
A large proportion of the tumour microenvironment consists of inflammatory infiltrate predominated by microglia and macrophages, which are thought to be subverted by glioblastoma cells for tumour growth [306]. Thus, GAMMs are logical therapeutic targets [307]. In addition to GAMMs, other inflammatory cells are present, such as tumour-infiltrating lymphocytes, neutrophils, and cancer-associated fibroblasts [308]. Inflammatory cells promote tumour growth, invasion, and tumour-to-therapy resistance thanks to the release of specific molecules and factors that favour anti-inflammatory activity (TGF-ß, ARG1 and IL10), tissue remodelling and angiogenesis (VEGF, MMP2, MMP9). In addition, inflammatory cells influence glioma-stem cells through the release of the matrix metalloproteinase MMP9 [309].
The translocator protein (TSPO; 18 kDa) is a peripheral benzodiazepine receptor, composed of a transmembrane multimeric protein complex of 18-kDa, situated in the outer mitochondrial membrane. TSPO is widely distributed in most peripheral organs. In addition, TSPO is also minimally expressed in resting microglial cells in the healthy brain but is substantially upregulated in reactive astrocytes and predominantly during the microglia activation process, due to neurodegenerative and neuroinflammatory diseases [310]. Moreover, most glioma cells express the TSPO. Different studies demonstrated a positive correlation between TSPO expression and grade of malignancy and a negative correlation with survival [311]. The most used TSPO radiotracer is [ 11 C](R)PK11195. Su et al. found that [ 11 C](R)PK11195 uptake in high-grade gliomas was significantly higher than in low-grade astrocytomas and low-grade oligodendrogliomas. TSPO in gliomas was expressed predominantly by neoplastic cells, and its expression correlated positively with uptake in the tumours indicating that [ 11 C](R)PK11195 can be used to stratify patients [245]. Considering the short half-life of 11  F]DPA-714 uptake, but also infiltrating glioma associated myeloid cells (GAMs) showed an important TSPO and MMP expression [246]. In another study performed in patients with glioma, authors found a strong relationship (r = 0.84, p = 0.009) between the [ 18 F]DPA-714 uptake and the number and activation level of GAMs. TSPO expression was mainly restricted to human leukocyte antigen D related-positive (HLA-DR+) activated GAMs, particularly to tumour-infiltrating HLA-DR+ myeloid-derived suppressor cells and tumour-associated macrophages indicating that TSPO radiotracer could be useful to study tumour microenvironment [247]. At the same time, in a preclinical model obtained with human GBM cells from patient, [ 18 F]DPA-714 was able to detect glioma infiltration into the contralateral brain earlier than [ 18 F]FET. In parallel, DWI MRI provides microstructure information. Authors also found an abundance of CD11b+ GAMs inside and around the tumour [248]. We have recently developed and characterised a new TSPO radiotracer, [ 18 F]VC701 in neuroinflammation models [249]. Similarly, in an orthotopic glioma model obtained with murine GBM cells, the GL261 ones, we observed an earlier TSPO uptake signal compared to [ 18 F]FLT uptake ( Figure 2 Furthermore, the radiotracer [ 18 F]GE-180 seems to provide interesting results in patients and the uptake correlates with the grade of tumour [250,251]. Other radiotracers are now under development to increase the specificity to TSPO [252,253]. .

Other Targets of Tumour Microenvironment
Prostate-specific membrane antigen (PSMA) is a type II transmembrane glycoprotein receptor with glutamate carboxypeptidase/folate hydrolase activity [312]. PSMA is highly expressed in prostate cancer and has recently emerged as a target for radionuclide imaging and treatment of this tumour [313]. PSMA is expressed not only in prostate cancer, but also in other solid tumours including brain tumours and correlates with the WHO grade. PSMA expression was found in endothelial cells associated with neovascularization and, for these reasons, can be a very interesting Furthermore, the radiotracer [ 18 F]GE-180 seems to provide interesting results in patients and the uptake correlates with the grade of tumour [250,251]. Other radiotracers are now under development to increase the specificity to TSPO [252,253].

Other Targets of Tumour Microenvironment
Prostate-specific membrane antigen (PSMA) is a type II transmembrane glycoprotein receptor with glutamate carboxypeptidase/folate hydrolase activity [312]. PSMA is highly expressed in prostate cancer and has recently emerged as a target for radionuclide imaging and treatment of this tumour [313]. PSMA is expressed not only in prostate cancer, but also in other solid tumours including brain tumours and correlates with the WHO grade. PSMA expression was found in endothelial cells associated with neovascularization and, for these reasons, can be a very interesting marker for monitoring anti-angiogenic drugs [314]. Several PET radiotracers for PSMA have been developed, such as [ 68 Ga]PSMA-HBED-CC, [ 18 F]DCFPyL, and [ 89 Zr]Df-IAB2M anti-PSMA minibody. At the moment, only few studies evaluating PSMA targeted imaging of gliomas/glioblastomas in clinical practice are currently available. Moreover, most of these studies are case reports or case series [255][256][257][258][259][260][261][262][263][264]315]. Although limited, these studies confirmed that high-grade glioma/glioblastoma are PSMA-avid compared to low-grade and PSMA radioligand mainly binds vascular endothelial cells. A more detailed review is present in the paper of Van de Wiele et al. [265]. This year, a preclinical study in which the authors compared the uptake of two PSMA radioligands ([ 68 Ga]-PSMA and [ 18 F]DCFPyL) in three rat glioma models (F98, 9L, or U87) has been published. Radiotracers' uptake was evaluated ex vivo with autoradiography and in vivo with PET. Uptake data were then correlated to immunofluorescence staining for blood vessels, microglia, astrocytes, and the presence of PSMA. Both radiotracers displayed a higher uptake at the rim of the tumours where CD11b staining demonstrated low presence of activated microglia but a strong GFAP staining indicating the presence of reactive astrocytes. Very few astrocytes were noted in the centre of the tumours. These data suggest that PSMA radioligand could also visualize astrocytes activation [266].
Matrix metallo-proteinases (MMPs) are other attractive biomarkers for tumour therapy and imaging. MMPs are linked to increased cell proliferation, tumour invasion, migration, and poor prognosis in glioma patients [316]. Moreover, MMPs facilitate microglia-mediated glioma invasion by degrading the basement membrane and proteins of the ECM. Besides that, MMPs affect the neuroinflammatory milieu by modulating the expression and activity of chemokines, inflammatory cytokines, growth factors, and receptor turnover [317]. Several molecules have been recently developed to study MMPs. For example, de Lucas et al. evaluated an immunoPET tracer for the membrane-type 1 matrix metallo-proteinase (MT1-MMP or MMP-14), [ 89 Zr]DFO-LEM2/15, in xenograft and orthotopic brain GBM models. This tracer displayed the higher uptake after 24 h post injection in tumours with high expression of MT1-MMP but a severe disruption of the BBB is needed to visualise intra brain tumours [267]. Zinnhardt  , which binds to MMP-2 in an orthotopic rat C6 glioma model. The uptake of the tracer in the normal brain is very low and a high accumulation was found in the glioma tissue. The tumour to normal brain ratio of [ 18 F]-FP-chlorotoxin was higher than that of [ 18 F]FDG with the maximum uptake at 90 min [269]. Kasten and colleagues designed an MMP-14-activatable dual PET/NIRF peptide probe for imaging and guiding resection of glioma. The peptide probe combined (1) a NIRF reporter and quencher pair separated by a peptide sequence (MMP-14 "substrate peptide") that is cleaved specifically by activated MMP-14 to release the quencher and allows visualization of the NIRF dye; and (2) [270].
A further PET radiotracer can be the Fibroblast Activation Protein (FAP) ligand.
Recently, Röhrich and colleagues have shown the potential of the targeting FAP-PET-based imaging ( 68 Ga-FAPI-01 and -04) in retrospectively studying 13 patients with wild type-IDH glioblastomas. FAP-positive signalling of extracranial tumours is attributed to activated Cancer Associated Fibroblasts (CAFs) that are located in the stromal compound of these tumours and overexpress FAP. In parallel, patients underwent multimodal MRI. FAP specific PET and rCBV MRI scans are modestly correlated and no correlation between the FAP-based PET and the ADC sequence has been observed, suggesting that FAP-specific PET could be useful in biopsy organization and discrimination between pseudo-progression and tumour progression after radiation therapy for GBM imaging [318].
In a paper of Henderson and colleagues, two recurrent GBM patients have been chosen for validating [ 18 F]Fluciclovine-based PET/CT imaging in GBM, a synthetic amino acid used as PET tracer. In this study, it has been reported the ability of this radiotracer, approved by FDA for prostate cancer care in 2016 and used since 2015 for gliomas treatment but not still approved, to be feasibly used for discriminating areas with the highest cancer recurrence from the ones where the most prevailing treatment efficacy is observable [228].
Considering the importance of the L-amino acid transporter 1 (LAT1), other radiotracers have been exploited and investigated.
In a paper by Verhoeven  F]FELP a very high and specific affinity for this amino acid transporter, this emerging tracer might be exploited for driving biopsy and primary brain tumour diagnosis, to plan and guide radiation treatment and for tumour recurrence and radio-necrosis discrimination after the starting therapy [229].

miRNA and Imaging
Several years of research correlating imaging features with histopathological data created a new branch of medicine: radiomics. Radiomics is an analysis approach to convert the image data into quantitative, mathematical output (features). Radiomics features are closely related to intra-tumour complexity and heterogeneity, representing the histopathological characteristics of the GBM. Imaging techniques could be helpful for tumour classification, analysis of the response to the treatment, and prognosis [319]. Radiogenomics is an emerging field that tries to link radiomics with the corresponding underlying genomic and epigenomic data. Radiogenomics analysis highlights the association between the radiomic quantitative data/features and molecular profile, including genomic, epigenomic, and metabolomic signatures, that are correlated with a clinical outcome [320]. The radiogenomics approach has been proposed for some types of cancer, from hepatocellular carcinoma to breast cancer [321,322]. Genomic data acquisition requires tissue biopsy, which is not always obtained in patients with GBM. Histopathological features, specific mutation pattern and molecular markers analysis revealed that GBM could be classified in four subtypes with different outcomes: proneural (the most favourable for the prognosis) neural, mesenchymal (the one associated with poor prognosis), and classical. Different subtypes react differently depending on treatment [323]. Radiogenomics can be applied to predict the GBM subtypes. Volumes of both contrast enhancement and necrosis are higher in GBM of the mesenchymal subtype compared to the proneural subtype. GBMs with less than 5% tumour enhancement are mostly of the proneural subtype.
Radiogenomics could also help in the prediction of the mutation status. Indeed, the IDH-mutant GBM is mainly localised in the frontal lobe and it is characterized by a higher percentage of non-contrast-enhancing part of the tumour and the presence of cysts on MRI. Left temporal lobe GBM are usually MGMT-methylated tumours, whereas MGMT-unmethylated GBM are more frequently found in the right hemisphere. 1p19q co-deletion is linked to classical oligodendroglial MRI characteristics (heterogeneous T2 signal intensity and calcifications). Tumours with EGFR amplification, mainly localised in the left temporal lobe, present a significant higher percentage of contrast-enhancement and T2/FLAIR hyperintensity compared with those lacking EGFR amplification. The relation between MRI features and the expression of other molecular markers (phosphatase and tensin homolog, PTEN; platelet-derived growth factor receptor, PDGFRA; cyclin-dependent kinase inhibitor 2A, CDKN2A; retinoblastoma 1, RB1; tumour protein 53, TP53) is under investigation [213].
Some attempts have been made in the description of GBM heterogeneity on the basis of epigenetic profile (DNA methylation), associating it with patients' survival, MRI, and quantitative pathology (tumour morphology, proliferative activity, and microenvironment). From this work it emerges that mesenchymal cell subtype, associated with the worst prognosis, is enriched in hypomethylated promoter regions in EZH2, KDM4A, RBBP5, and SUZ12, as well as in regulators of pluripotency (NANOG, SOX2, POU5F1) and enriched in hypermethylated chromatin promoter of CTCF, EZH2, and KDM4A. MRI data reveals that the highest number of immune cells was in tumours of the mesenchymal subtype, showing also increased tumour size, fewer vital tumour areas, and increased oedema, while histopathology revealed lower cell density and large necrotic areas [324].
In 2012, Elkhaled et al. demonstrated that there is concordance between the presence of D-2-hydroxyglutarate (2HG), as detected by proton high-resolution magic angle spinning ( 1 H HR-MAS) spectroscopy, and IDH1 mutation status, as determined by antibody staining and genetic sequencing, that was confirmed by Choi and colleagues [325,326]. These two studies revealed that there is a direct link between a genetic mutation and a behavioural or a metabolic phenotype observed by imaging analysis.
In 2018, a paper described a tentative correlation of data from quantitative MRI with transcriptomic profiles of 65 patients with primary glioblastoma [327]. Unsupervised clustering methods, confirming the data of Verhaak et al. revealed the existence of three different clusters, each associated with a different outcome [327,328]. Each cluster represented a distinct molecular classification of glioblastoma already described, classical type, proneural and neural types, and mesenchymal type. The first cluster was associated to the increased expression of anion channel activity, peroxisome, and the classical phenotype observed in Verhaak and colleagues' paper; the second cluster presented higher expression of genes of the mitosis, cell cycle and a proneural/neural subtype; the third cluster was characterized by altered expression of extracellular matrix molecules, defence response, immune signalling molecules, and the mesenchymal subtype molecules. These data suggest that radiomic data could provide information about molecular subtypes of GBM based on the transcriptomic profile [327,328].
Nowadays, a missing step is to demonstrate the existence of a correlation between the data extracted from MR imaging and the miRNA expression profiles.
Molecular imaging by the use of PET is increasingly being implemented into clinical practice for treatment planning and response monitoring in GBM. The potential clinical applications of PET include the monitoring of treatment response, as it could distinguish tumour recurrence from radiation necrosis or pseudoprogression. Correlation between different types of amino acid tracers in PET and molecular markers is currently under investigation [329].
A correlation between miRNA expression in tissue GBM and PET imagings has not been proven yet. Indeed, miRNAs free molecules or miRNAs associated to extracellular vesicles or exosomes released by GBM tumour have been considered as possible biomarkers in the liquid biopsy for GBM diagnosis [213].
Several miRNA profiles have been proposed for GBM diagnosis by liquid biopsy isolation (i.e., as proposed by Kopkova and colleagues and by Ebrahimkhani et al.), but no one has correlated their expression with PET imaging features [330,331].
Exosomal miR-21 is one of the plasma biomarkers proposed for GBM diagnosis and seems to be able to distinguish tumour progression from pseudoprogression or radionecrosis [332].
Recently, a noninvasive glioblastoma testing (NIGT) platform has been proposed, including a multimodal approach that combines the complexity and heterogeneity of GBM described by imaging techniques, with computational approaches, and the use of circulating biomarkers [213]. This could be especially helpful in the selection of patients in order to better address them to the best therapeutic approach.

Final Considerations and Future Direction
In this review, starting from the analysis of GBM cell heterogeneity, we focused on those aspects that sustain the development of TMZ resistance, in order to find out possible therapeutic solutions for clinical application. In particular, we highlighted the supporting role of microenvironment surrounding the tumour and the possible functions of several non-coding miRNAs, as possible targets for the development of new therapeutic strategies. In order to have a complete view of the gold standard treatment of GBM, we described the canonical treatment of TMZ and the main mechanisms involved in the development of resistance to TMZ. In addition, several possible alternative treatments to TMZ alone-based therapy have been discussed, including combinatorial pharmacological treatments, comprising radiotherapeutic approaches, antibodies, and miRNAs. To the view of the interesting role of miRNAs, new therapeutic strategies could include the combination of miRNAs and drugs, this field becoming an important opportunity to investigate [109,[333][334][335][336].
Moreover, we discussed a significant number of PET tracers that can help to visualise and monitor cancer cells and tumour microenvironments including GAMMs, endothelial cells, fibroblasts, astrocytes, and stem cells (Figure 3). These PET radiopharmaceuticals, particularly when associated with information deriving from MRI, may be relevant also for the identification of lesion heterogeneity and for the clusterisation of subsets of patients potentially responding to treatment under development.
In vivo diagnostic imaging data may be a useful platform to associate and monitor genetic and biological features of the tumour at sub-regional levels during time in a non-clinical model or associated with post-surgery analysis and outcome in patients. In turn, in vivo molecular imaging offers a potential tool to characterize the inter-and intra-tumour heterogeneity of GBM and clusterize the lesions. Moreover, new advanced technologies and analysis methods as combied PET/MRI scanners joined by textural features increase diagnostic accuracy of gliomas in the identification of specific genotypes guiding surgery.

24
In addition, PET tracers can be used for guiding targeted radionuclide therapy substituting the positron emitting radionuclide with a therapeutic beta-particle emitting radionuclide, either lutetium-177 or yttrium-90 as performed in other types of tumor. Although this application is not yet available or proof for glioma, the use of theranostic [177Lu]-PSMA has been already tested in brain metastasis indicating the potential feasibility for its application also in glioma [337].

Conclusions
In this review we have summarised the complex heterogeneity of GBM and the potential use of imaging to visualise differences at regional levels in biological features. We have described the main genetic and epigenetic mechanisms of drug resistance and the state-of-the-art research studies attempting to overcome it. It should be stressed that surgery remains the best treatment available. However, the delayed diagnosis and difficult identification of lesion margins and infiltrating cells limit its benefit. For these reasons, pharmacological as well as radiation therapies are necessary. Despite the efforts, information on the biology of GBM is still insufficient and the molecular classification limited to a selected portion of tumour. However, molecular and single-cell data In addition, PET tracers can be used for guiding targeted radionuclide therapy substituting the positron emitting radionuclide with a therapeutic beta-particle emitting radionuclide, either lutetium-177 or yttrium-90 as performed in other types of tumor. Although this application is not yet available or proof for glioma, the use of theranostic [177Lu]-PSMA has been already tested in brain metastasis indicating the potential feasibility for its application also in glioma [337].

Conclusions
In this review we have summarised the complex heterogeneity of GBM and the potential use of imaging to visualise differences at regional levels in biological features. We have described the main genetic and epigenetic mechanisms of drug resistance and the state-of-the-art research studies attempting to overcome it. It should be stressed that surgery remains the best treatment available. However, the delayed diagnosis and difficult identification of lesion margins and infiltrating cells limit its benefit. For these reasons, pharmacological as well as radiation therapies are necessary. Despite the efforts, information on the biology of GBM is still insufficient and the molecular classification limited to a selected portion of tumour. However, molecular and single-cell data suggest that treatment should act on common features, such as the metabolism or immune system, particularly using combined strategies. In this scenario, molecular imaging may help to better define GBM complexity and the modification occurring during therapy at regional level. This, together with biopsy or post-surgery in vitro single-cell imaging and bioinformatics, can help in the identification of cluster of patients with common biological traits and this is crucial for the proper development of novel pharmacological or radiation-based therapy.