Adolescent multi-omics and Mendelian randomization reveal transdiagnostic molecular mechanisms in psychiatric disorders.

Glioblastoma (GBM), isocitrate dehydrogenase (IDH) wildtype, and astrocytoma, IDH-mutant grade 4, are highly aggressive brain tumors according to the World Health Organization (WHO) classification (). The molecular profile of the tumor significantly impacts the efficacy of treatment with temozolomide, particularly the methylation status of the O6-methylguanine-DNA methyltransferase (MGMT) promoter (). The MGMT protein counteracts the DNA alkylation damage caused by temozolomide, leading to resistance to this treatment (). Methylation of the MGMT gene promoter region leads to decreased expression of MGMT, which correlates with improved survival in GBM and astrocytoma patients, indicating added benefit from temozolomide (, ). Early identification of tumor biomarkers therefore plays a crucial role in the management of malignant high-grade gliomas (HGG). The gold standard to test for MGMT promoter methylation (MGMTpm) involves obtaining tumor tissue and quantitative assessment with methods such as pyrosequencing (). However, there is additional value in predicting the MGMT status preoperatively. If a tumor is suggested to be non-MGMTpm, this could affect the neurosurgical approach, potentially involving a more aggressive gross total resection to improve survival (). Magnetic resonance imaging (MRI) has previously been used to predict MGMTpm with varying performance, including conventional, diffusion, and perfusion imaging techniques (). Molecular imaging using chemical exchange saturation transfer (CEST) MRI is also emerging as a potential non-invasive approach. CEST imaging is based on the principles of chemical exchange of selectively saturated protons, resulting in a measurable decrease in the bulk water magnetization (, ). Indirect detection of mobile proteins and peptides via CEST can be achieved using amide proton transfer-weighted (APTw) imaging (, ). According to the current CEST consensus guidelines (), APTw contrast is generated using prolonged saturation with relatively high B amplitude (~2 µT), ensuring sufficient labeling of amide protons at +3.5 ppm relative to the water resonance frequency. The APTw signal has previously been utilized in glioma grading (–), to distinguish tumor recurrence from treatment-related changes (–), and to predict overall survival (, ). The CEST signal at +2.0 ppm (CEST@2ppm), and the APTw/CEST@2ppm ratio (herein referred to as Ratio) are other metrics that may aid in tumor differentiation (–). Moreover, CEST post-processing techniques such as fluid suppression (FS) have been introduced to reduce the signal contribution from fluid-rich compartments such as necrotic or cystic tissues, while retaining the signal in semi-solid tissue compartments (, ). FS has previously been used in the distinction of GBM and brain metastases (), in the molecular profiling of diffuse gliomas (), and in the assessment of radionecrosis (). As summarized in a review article (), CEST imaging has shown promise in predicting tumor biomarkers such as IDH mutation (, , ), 1p19q co-deletion (, ), p53 overexpression () and Ki-67 index (). However, findings regarding MGMTpm have been inconsistent. Some studies have found higher APTw values in non-MGMTpm tumors using histogram analysis, max APTw% signal or visual assessment (–). Conversely, other studies have reported no significant associations between the CEST metrics and MGMTpm (, , , ). To the best of our knowledge, the CEST@2ppm signal, the Ratio, and the utilization of FS have not been used previously in MGMTpm prediction. Notably, some earlier studies may have been hampered by small or skewed sample sizes (, ), or the time-consuming task of whole tumor segmentation (, , ), possibly limiting translation to clinical practice. Automated segmentation of the tumor regions using machine learning (ML) based models may streamline the workflow in larger patient samples and reduce the need for manual delineation (). Some models that are publicly available for this purpose, and employ a graphical user interface, are DeepBraTumIA () and Raidionics (). Both models have been evaluated for their clinical utility, especially on preoperative well-demarcated lesions and peritumoral edema (, ). The prognostic value of the automatic volumes has also been suggested in survival analyses and longitudinal assessments (, ). Some have used the segmentations from DeepBraTumIA and Raidionics to compare volumetric and radiomics-based methods for MGMTpm prediction using conventional MRI (, ). These models may therefore have additional benefits in CEST imaging as well, which has not previously been investigated.