The 2021 WHO (CNS5) classification defines adult-type diffuse gliomas primarily by IDH status. IDH-wildtype GBM (~90% of cases) is diagnosed when a diffuse astrocytic glioma harbors TERT promoter mutation, EGFR amplification, or combined chromosome 7 gain and chromosome 10 loss, even in the absence of microvascular proliferation or necrosis (1,10). IDH-mutant astrocytoma is graded as CNS WHO grade 4 in the presence of CDKN2A/B homozygous deletion, necrosis, or microvascular proliferation (1,10). H3 K27M-mutant diffuse midline glioma is a distinct, highly aggressive entity occurring mainly in children and adolescents (10). These criteria illustrate that molecular classification is no longer determined solely by IDH mutation and MGMT methylation, but by a combination of diagnostic, prognostic and predictive markers (10).

MGMT promoter methylation is a predictive biomarker rather than a diagnostic one: It identifies patients most likely to benefit from TMZ, and methylated patients show a mOS ~4-6 months longer than unmethylated patients (11,12). IDH mutation is both diagnostic and prognostic: IDH-mutant tumors carry a mOS 6-8 months longer than IDH-wildtype GBM and may respond to IDH1/2 inhibitors (13,14). A 2023 multicenter cohort study reported that patients with IDH-wildtype GBM with MGMT promoter methylation treated with TMZ plus immunotherapy achieved a mOS of 18.5 months vs. 14.3 months in unmethylated patients (11). In addition, circ-SETDB1 was recently identified as an independent prognostic factor in IDH-wildtype GBM (12).

Single-cell RNA sequencing has revealed that GBM is composed of transcriptionally plastic cellular states rather than fixed subtypes. The landmark integrative analysis identified four recurrent states, neural progenitor cell-like, oligodendrocyte progenitor cell-like, astrocyte (AC)-like and mesenchymal (MES)-like, that coexist within individual tumors and interconvert dynamically (7,15). The MES-like state is enriched in hypoxic and necrotic regions, upregulates TGF-β pathway genes, suppresses CD8⁺ T-cell activation, enhances invasion and reduces chemo-radiosensitivity (15). Recent spatial transcriptomics and single-cell multi-omics studies further show that lineage plasticity and cellular-state transitions are key drivers of treatment resistance and represent one of the most active areas of GBM research (16,17).

Temporal heterogeneity is equally important. A 2025 longitudinal study showed that GBM cells undergo epigenetic reprogramming under TMZ pressure, leading to MGMT promoter demethylation and upregulation of MGMT expression, converting TMZ-sensitive cells into TMZ-resistant cells and driving early progression after an initially effective response (16,18).

5-aminolevulinic acid (5-ALA) fluorescence-guided surgery is the most mature precision technique. A 2023 meta-analysis of 12 randomized controlled trials reported that 5-ALA increased the gross total resection rate by 32% and extended mOS by 2.1 months compared with conventional surgery (19-21). Nevertheless, the benefit is limited to patients in whom complete resection of the enhancing tumor is achievable, and fluorescence-guided surgery does not address the diffuse infiltrative component. Intraoperative rapid molecular testing has also advanced: A 2024 study described an intraoperative MGMT promoter methylation assay completed within 30 min that can guide postoperative chemoradiotherapy planning (22).

Whether to extend adjuvant TMZ beyond six cycles remains controversial. A 2022 phase III trial reported that extending adjuvant TMZ from 6 to 12 cycles in MGMT-methylated GBM prolonged median progression-free survival (PFS) from 6.4 to 10.2 months (23); however, this finding has not been consistently reproduced, and prolonged TMZ increases toxicity and cost without a universal overall survival benefit. Extended adjuvant TMZ should therefore be regarded as an individualized option rather than a new standard. Hyperbaric oxygen therapy combined with chemoradiotherapy has been reported to improve the objective response rate (ORR) by 15% by increasing tumor oxygenation and TMZ cytotoxicity (24), but the supporting evidence is limited and preliminary, and this approach should be interpreted cautiously.

TTFields deliver low-intensity, intermediate-frequency alternating electric fields that disrupt mitosis. In the pivotal EF-14 randomized trial, adding TTFields to maintenance TMZ in newly diagnosed GBM extended mOS from 16.0 to 20.9 months (25), and TTFields are now incorporated into major management guidelines for eligible patients. Real-world data from 2020-2025 generally support the feasibility of TTFields in routine practice, although adherence, scalp toxicity and cost remain practical limitations (25).

Immune checkpoint inhibitors have transformed several solid tumors but have largely failed in GBM. The randomized CheckMate 143 trial found no survival benefit of nivolumab over bevacizumab in recurrent GBM (26), and this negative result is now understood to reflect profound immune suppression, T-cell exhaustion, antigen heterogeneity, corticosteroid-induced immunosuppression and poor T-cell infiltration in the GBM microenvironment (27,28). Combination strategies aim to overcome these barriers: A 2024 phase II trial reported that nivolumab plus bevacizumab achieved an ORR of 22% and a mOS of 9.8 months in recurrent GBM [vs. 6.2 months for monotherapy (29)], but confirmatory phase III data are still lacking.

CAR-T cell therapy: Intracranial delivery is used to bypass the BBB. A 2025 phase I trial of intracranial EGFRvIII-directed CAR-T cells in 12 patients with recurrent GBM reported partial responses in 4 patients and a median PFS of 5.3 months (30). Dual-target (EGFRvIII + IL13Rα2) CAR-T cells may reduce antigen-escape resistance (31,32). However, CAR-T therapy in GBM faces major challenges, including antigen escape, limited T-cell persistence, neurotoxicity and delivery limitations, and efficacy remains unproven in randomized trials.

Tumor vaccines: A 2023 phase III trial reported that a dendritic cell vaccine added to standard therapy extended mOS to 19.3 months in IDH-wildtype GBM (33), and a 2025 phase I trial of an mRNA vaccine targeting GBM-specific mutations reported induction of tumor-specific immune responses with acceptable safety (34,35). These results are encouraging but require independent confirmation.

Targeted therapy has shifted from conventional signal-pathway inhibition toward metabolic inhibition and antibody-drug conjugates (ADCs). Depatuxizumab mafodotin, an EGFRvIII-directed ADC, achieved an ORR of 18% in EGFRvIII-positive GBM in a phase II trial (36,37). However, EGFR-targeted therapies have generally been disappointing in GBM despite frequent EGFR alterations, largely because of inadequate BBB penetration, pharmacokinetic limitations and redundant signaling pathways (38-40). A 2025 phase I trial of the hexokinase 2 (HK2) inhibitor OMO-103 plus TMZ in MGMT-unmethylated GBM extended median PFS by 2.9 months (23); again, this is an early-phase signal requiring confirmation.

Oncolytic viruses both lyse tumor cells and stimulate antitumor immunity and are increasingly combined with radiotherapy or immunotherapy. A 2024 phase II trial of oncolytic HSV-1 (T-VEC) plus radiotherapy in newly diagnosed GBM reported a mOS of 17.6 months (41), and a 2025 phase I trial of an oncolytic adenovirus expressing a PD-L1 inhibitor achieved an ORR of 25% in recurrent GBM (42,43). These agents remain investigational.

TMZ cytotoxicity depends on DNA alkylation, and enhanced DNA repair is a major resistance mechanism. MGMT-mediated direct repair: IDH-wildtype GBM upregulates MGMT transcription through EZH2-mediated H3K27me3 modification after TMZ exposure (44); the EZH2 inhibitor tazemetostat plus TMZ increased apoptosis of MGMT-high GBM cells by 42% in preclinical models (44,45). PARP-mediated base excision repair: a 2025 phase I trial reported that the PARP inhibitor olaparib plus TMZ extended median PFS to 7.1 months (vs. 4.2 months in controls) in MGMT-unmethylated GBM (46). ~30% of recurrent GBM harbor BRCA2 mutations, which activate homologous recombination and may predict response to PARP inhibitors (46).

Histone modification, non-coding RNA regulation and DNA methylation dynamics all contribute to acquired resistance. GBM upregulates HDAC6 after radiotherapy, stabilizing DNA repair proteins; the HDAC6 inhibitor ricolinostat plus radiotherapy extended survival in tumor-bearing mice (47). The circular RNA circRNA-0001234 upregulates MDR1 by sponging miR-214 in drug-resistant GBM (18). In IDH-mutant GBM, TET2 silencing induces global hypomethylation and activates invasion-related genes (18). These are predominantly preclinical observations.

GBM cells adapt metabolically to therapeutic pressure. TMZ upregulates HK2 by 2.3-fold, enhancing glycolysis and reducing reactive oxygen species to evade apoptosis; the HK2 inhibitor 2-DG plus TMZ reduced GBM cell survival by 60% in vitro (48,50). Recurrent GBM overexpresses glutaminase (GLS), which maintains GSC stemness; the GLS inhibitor CB-839 plus CAR-T cells increased GSC clearance in preclinical models (51). Radiotherapy activates fatty acid synthase (FASN); the FASN inhibitor orlistat plus radiotherapy reduced tumor volume in animal models (49).

Tumor-associated macrophages dominate the GBM TME and secrete IL-10, inducing CD8⁺ T-cell exhaustion; anti-IL-10 plus anti-PD-1 increased the ORR to 28% in a phase I study of recurrent GBM (52,53). GBM-induced ACs secrete CXCL12, promoting invasion and TMZ resistance; the CXCR4 inhibitor plerixafor plus TMZ extended survival in mice (54). GBM-secreted MMP9 remodels the extracellular matrix; the MMP9 inhibitor marimastat may enhance oncolytic virus penetration (55,56).

GSCs are self-renewing, multidrug-resistant cells central to recurrence. Notch1 signaling maintains stemness through Hes1; the Notch inhibitor DAPT plus radiotherapy reduced GSC proportion by 80% in preclinical models (57). EZH2 maintains stemness by repressing SOX2 negative regulators; tazemetostat plus CD133-ADC eliminated 90% of GSCs in vitro and in vivo (44). Hypoxia-inducible factor 1α (HIF-1α) upregulates ABCG2 in hypoxic GSCs, enhancing drug efflux; the HIF-1α inhibitor belzutifan plus TMZ increased GSC drug sensitivity in preclinical studies (58,59).

A key emerging concept is that these mechanisms are mutually reinforcing. Hypoxia stabilizes HIF-1α, which simultaneously drives metabolic reprogramming (glycolysis), GSC stemness (ABCG2 upregulation) and TME immunosuppression (TGF-β, IL-10), while DNA damage from TMZ and radiotherapy feeds back to activate epigenetic and metabolic adaptations (23,60). A schematic summary is provided in Fig. 1. Targeting single nodes of this network is therefore unlikely to be curative, and rational combination strategies that simultaneously address DNA repair, metabolism, immunosuppression and stemness are more likely to succeed.

PubMed, Embase and Web of Science were searched on 15 December 2025 for records published between 1 January 2020 and 15 December 2025. The following search string was adapted for each database: (‘glioblastoma’ OR ‘GBM’) AND (‘molecular classification’ OR ‘molecular subtyping’ OR ‘IDH’ OR ‘MGMT’) AND (‘targeted therapy’ OR ‘immunotherapy’ OR ‘tumor treating fields’ OR ‘drug resistance’ OR ‘tumor microenvironment’). No language restriction was applied at retrieval; non-English records were excluded after screening because of data-integrity concerns.

Study selection. After duplicate removal, two reviewers (L.S. and S.S.) independently screened titles and abstracts, and then full texts, against predefined inclusion and exclusion criteria. Disagreements were resolved by discussion or by a third reviewer (Y.G.). Inclusion criteria were: i) Original research, clinical trials or high-quality reviews published in authoritative oncology/neuro-oncology journals; ii) studies with clear experimental or clinical data on GBM molecular pathology, therapy or resistance; and iii) trials reporting phase, population and efficacy endpoints. Exclusion criteria were: i) case reports, conference abstracts and unpublished gray literature; ii) studies with incomplete data, unclear conclusions or low methodological quality; and iii) studies focused on non-GBM glioma subtypes.

Screening results and quality assessment. The search identified 1,247 records, of which 261 duplicates were removed, leaving 986 records for screening. Of these, 68 non-English records were excluded after title/abstract screening because of data-integrity concerns. The 772 records excluded at the title/abstract stage comprised 360 records on non-GBM glioma subtypes or out-of-scope topics, 162 case reports, 98 conference abstracts or gray-literature items, and 84 records with incomplete or unclear data. A further 128 records were excluded at the full-text stage, comprising 47 with incomplete data, 36 with low methodological quality, 20 with unclear conclusions, 14 case reports/conference abstracts confirmed on full text, and 11 non-GBM glioma subtypes. The remaining 86 studies were synthesized (Fig. 2). Inter-reviewer agreement was substantial (Cohen's κ=0.82). Methodological quality was assessed by study design, sample size and endpoint definition; findings from randomized trials were weighted above single-arm early-phase trials, which were in turn weighted above preclinical studies. The main clinical and pre-clinical studies included in this review are summarized in Table II.