Gliomas are the most common primary brain tumors, characterized by high mortality and morbidity rates (1). According to the 2021 World Health Organization (WHO) Central Nervous System (CNS) classification, adult-type diffuse gliomas are categorized into astrocytomas (isocitrate dehydrogenase mutant [IDH-mt], 1p/19q non- codeletion), oligodendrogliomas (IDH-mt, 1p/19q codeletion), and glioblastomas (IDH wild-type, [IDH-wt]) (2). IDH-wt gliomas are classified as grade IV, oligodendrogliomas as grade II to III, and astrocytomas range from grade II to IV. Glioma grading influences treatment approaches, with high-grade gliomas typically managed by maximal surgical resection followed by adjuvant radiotherapy and chemotherapy, while low-grade gliomas are treated based on the extent of resection and patient factors such as age to determine postoperative adjuvant therapy (3). The new classification guidelines highlight the importance of genotypes and molecular characteristics. Research indicates that patients with 1p/19q codeletion respond better to radiotherapy and chemotherapy, resulting in improved prognosis (4). Additionally, O-6-methylguanine-DNA methyltransferase (MGMT) promoter methylation predicts a better response to temozolomide and enhances survival (5). However, molecular typing often relies on pathological diagnosis, which is invasive, prone to sampling errors, and costly.

MRI is the most commonly used preoperative diagnostic tool for gliomas. Grade IV gliomas frequently exhibit ring enhancement on T1-weighted images, whereas grade II and III gliomas typically show no enhancement, making it challenging to distinguish these grades on imaging. Most studies focus on the comparison between low-grade gliomas (grade II) and high-grade gliomas (grades III and IV), or between lower-grade gliomas (grades II and III) and higher-grade gliomas (grade IV), while the identification of grade III gliomas remains relatively vague. Diffusion-weighted imaging (DWI) has been used to predict MGMT promoter methylation and 1p/19q codeletion (5, 6). However, apparent diffusion coefficient (ADC) measurements are often based on subjective regional delineation, and the heterogeneity of gliomas may introduce selection bias in region of interest settings. Some studies have found that the methylated MGMT promoter type exhibited larger ADC values, while others reported no differences between methylated and unmethylated types (7, 8). Dynamic susceptibility contrast (DSC) and dynamic contrast-enhanced (DCE) imaging have also demonstrated value in predicting MGMT promoter methylation and 1p/19q codeletion, but both methods require contrast agent injection (9–12).

The degree of tumor metabolism is often positively correlated with malignancy. High-grade gliomas exhibit vigorous cell proliferation, angiogenesis or vascular disruption, and an accumulation of more acidic metabolic byproducts in the extracellular space. Persistent hypoxia, increased glycolysis, and heightened acidity in tumors can affect tumor invasiveness and alter gene expression (13, 14). Therefore, characterizing tumor metabolism and the acidity of the tumor microenvironment is a feasible method for grading and predicting molecular subtypes.

Chemical exchange saturation transfer (CEST) imaging is an MRI technique that enhances the detection of low- concentration biomolecules by exploiting the chemical exchange properties between molecules and water protons (15). The exchange rates of certain protons are pH-dependent, making this technique useful for assessing tissue pH, which is crucial for evaluating the tumor microenvironment (16). Previous studies have demonstrated that CEST imaging of amine protons in glutamine molecules can be used as a noninvasive pH-weighted MRI technique for human and preclinical investigations of malignant gliomas (17). Amide proton transfer (APT) imaging is a relatively mature CEST technology, with amide protons in tissues serving as the primary source of the APT signal (18). Studies have shown that APT imaging holds potential for the differential diagnosis, grading, molecular typing, and prognostic evaluation of gliomas (19–24). However, APT imaging based on magnetization transfer asymmetry analysis can overlook confounding factors, including intrinsic semi-solid magnetization transfer (MT) asymmetry and low-field relayed nuclear Overhauser effect (NOE) signals. Methods such as multi-pool Lorentzian analysis and inverse Z-spectrum analysis have been proposed to enhance CEST quantitative analysis (25, 26). Multi-pool Lorentzian analysis decomposes the Z-spectrum into five components: amide, NOE, amine, DS, and MT. Amide and amine represent mobile proteins/peptides and creatine, respectively. The DS signal is related to water proton concentration and tissue relaxation time, while the MT signal originates from immobile macromolecules. The NOE signal comes from the aliphatic and olefinic components of various metabolites, including mobile proteins, peptides, and lipids. Multi-pool Lorentzian analysis has demonstrated potential value in glioma grading and the diagnosis of IDH and 1p/19q genotypes (25, 27).

In this study, we quantitatively describe glioma metabolism and the pH characteristics of the tumor region based on CEST imaging using multi-pool Lorentzian and pH analyses. We explore their value in assessing glioma grade, IDH mutation, 1p/19q codeletion, and MGMT promoter methylation. Additionally, histogram analysis was employed to better characterize tumor heterogeneity.

This study investigated glioma grading and molecular genotyping using CEST-based pH assessment and micro- metabolic profiling within the context of the 2021 WHO CNS classification. Our results indicate that multi-pool Lorentzian analysis and pH-weighted analysis demonstrate diagnostic performance in grading gliomas and ingenotyping for IDH mutation status, 1p/19q co-deletion, and MGMT promoter methylation status.pH_weighted imaging can characterize the acidic microenvironment of tumors. We found that high-grade gliomas, IDH-wt gliomas, and 1p/19q non-codeleted gliomas exhibited higher pH_weighted values compared to low-grade gliomas, IDH-mt gliomas, and 1p/19q codeleted gliomas. Tumor cells preferentially convert glucose to lactic acid even in the presence of oxygen, resulting in excessive lactic acid production. Additionally, poor vascularization in these tumors leads to hypoxic conditions that further drive glycolysis and acid production (15). The higher metabolic activity in more invasive gliomas results in hypoxia and the accumulation of acidic metabolic products, leading to larger pH_weighted values (31). As shown in Figure 5 , gliomas with and without 1p/19q codeletion are difficult to differentiate on T2-FLAIR and T1-enhancement images. However, pH_weighted imaging can visually highlight differences between them and better reflect tumor heterogeneity. In the central tumor region, acidity is significantly increased, while in the peritumoral edema zone, tumor acidity is relatively lower. In more invasive IDH-wt gliomas, both the central tumor region and the peritumoral edema zone exhibit higher acidity, partially explaining their greater invasiveness. Therefore, we believe that pH_weighted imaging is a promising biomarker for glioma grading and subtyping analysis.

MT and DS respectively represent the content of semi-solid molecular tissue and water molecules. Grade III gliomas exhibited higher DS and lower MT compared to grade II gliomas. DS is related to tissue water proton density. Research indicates that high-grade gliomas tend to have higher vascular endothelial growth factor (VEGF) expression (32). VEGF is known as a potent growth factor for vascular endothelial cells, playing a crucial role in tumor growth and invasion by promoting the proliferation and migration of tumor vascular endothelial cells, increasing tumor vascular permeability, and inducing tumor lymphangiogenesis (32, 33). The higher VEGF expression corresponds with more severe edema, resulting in higher DS (25). However, our findings showed that DS was lower in grade IV gliomas compared to grade III gliomas, possibly due to differences in ROI selection. In grade III gliomas, where most cases did not show enhancement on T1-weighted images, the entire T2 hyperintense region was selected as the ROI. In contrast, grade IV gliomas were characterized by selecting the enhanced T1 area and excluding the peritumoral edema zone. MT primarily originates from immobile macromolecules, such as proteins and polysaccharides, and may serve as an indicator of white matter integrity (34). MT was higher in grade II gliomas, likely due to their retention of more normal brain tissue structure and composition. In grade IV gliomas, elevated cell density may lead to increased levels of proteins, polysaccharides, and other components within the tumor region, resulting in a higher MT effect. Furthermore, MT was associated with 1p/19q codeletion in grade II gliomas, with 1p/19q codeletion gliomas showing lower MT ( Figure 4 ). Further subclassification of 1p/19q codeletion and non-codeletion within low-grade gliomas is meaningful, as the identification of 1p/19q codeletion in IDH-mt gliomas maybe influenced by histological grading. Distinguishing 1p/19q subtypes in grade II/III gliomas can facilitate more precise treatment planning and efficacy assessment in preoperative or postoperative follow-up.

Amide and NOE reflect the content of amide protons and macromolecules such as lipids within the tissue. Significant differences were observed in amide and NOE signals between grade III and IV gliomas, as well as between IDH-wt and IDH-mt gliomas. IDH-wt gliomas are typically more aggressive and have higher cellular density compared to IDH-mt gliomas. This aggressive phenotype is associated with an increased proliferative rate and elevated protein synthesis (23, 35). In contrast, IDH-mt leads to the production of the oncometabolite 2- hydroxyglutarate (2-HG), which results in abnormal methylation of DNA and histones, affecting gene expression and cell differentiation (36). The higher concentration of proteins and peptides in IDH-wt gliomas likely contributes to a stronger amide and NOE signal. We found that the diagnostic performance of amide is superior to MTR_3.5. This maybe because MTR_3.5 is influenced by signals such as NOE and DS, and therefore cannot reflect a purer source of the amide signal.

Our study also found that MGMT promoter unmethylated gliomas typically exhibit higher amide signals. MGMT promoter methylation in gliomas is associated with reduced protein expression, which may impact the expression of downstream proteins. Therefore, CEST may serve as a useful imaging biomarker for predicting MGMT methylation status, consistent with previous findings (37). In contrast to previous studies where APT could not predict MGMT promoter methylation, possibly due to smaller sample sizes (26), our results suggest that MTR_3.5, despite being affected by multiple factors, can predict MGMT promoter methylation more effectively than the relatively pure amide signals.

In our study, the amine signal showed no significant differences in the grading and molecular classification of gliomas. Notably, Zhu et al.’s research also found no differences in the amine signal between IDH wild-type and mutant gliomas (27). We believe that there are two possible reasons for this result. Firstly, the amine signal has been assumed to mainly represent the contribution from creatine amine protons. However, amine signal obtained through Lorentzian fitting frequently overlaps with other rapidly exchanging pools, like glutamate, making it difficult to isolate them under 3T conditions. Creatine provides phosphate through phospho-creatine for adenosine triphosphate synthesis in the cell energy requirement. Tumor has reduced creatine and tumor creatine further reduces with tumor progression presumably due to elevated energy deficiency (38). There is building evidence that alterations to glutamate homeostasis in gliomas play an important role in diffuse glioma cell survival and increased extra-cellular glutamate causes excitotoxicity to peri-tumoral structures and promotes tumor invasion in pre-clinical studies (39). The complex variations in the contents of various components within the tumor lead to fluctuations in the amine signal. Secondly, our study was conducted using a 3T MRI scanner. Due to the relatively fast exchange rate of the amine signal and its fitting being close to the water peak at 2 ppm, the z-spectrum characteristics may not be distinct enough, making the Lorentzian fitting more challenging. In summary, the amine signal in glioma research may be influenced by various factors. Further research may need to explore more sensitive techniques or methods to better understand the role of the amine signal in gliomas. Based on the above discussion, the combination of multi-pool Lorentz analysis and pH analysis based on CEST demonstrates good performance in improving the grading and IDH gene typing of glioma. MT and pH_weighted can effectively identify 1p/19q codeletion in grade II and grade III gliomas, respectively. MTR_3.5 demonstrates potential effect in identifying MGMT promoter methylation. This technology can be implemented on standard MRI equipment, with a scanning time of approximately 5 minutes being clinically feasible. It does not require additional injection of contrast agents, making it relatively safe. In our study, we chose to perform the scans before the injection of the contrast agent to avoid the influence of the contrast agent on the CEST effect (40).

However, our study has several limitations. Firstly, although existing literature has demonstrated that, within lower irradiation power ranges, multi-pool Lorentzian fitting offers superior quantification accuracy compared to the three-frequency offset method and the Lorentzian-Dipolar (LD) method (38). However, multi-pool Lorentzian fitting has several limitations. In situations where the resonance frequencies of different signals are closely spaced or mixed, Lorentzian fitting may struggle to effectively differentiate between these signals. For example, the wide ‘MT’ peak could have multiple contributions especially the NOE(-1.6), which have attracted many interests in recent years (41–44). However, we could not resolve these components precisely in our analysis. Such spectral overlap can lead to inaccuracies in the fitting results, adversely impacting the quantification of specific signals, such as those from amines or other metabolites (27). Besides, Lorentzian fitting exhibits high sensitivity to background noise, particularly when signal intensities are low. The presence of background noise can interfere with the fitting process, resulting in erroneous parameter estimates. Additionally, successful Lorentzian fitting requires careful selection of initial parameters and fitting ranges. Inappropriate parameter choices can lead to convergence on local minima, thus compromising the accuracy and reliability of the fitting results. These limitations underscore the importance of judiciously selecting appropriate fitting methods and parameters in practical applications to ensure the reliability and validity of the results. Secondly, for pH assessment, although research indicates that amine proton-based CEST imaging (with a resonance frequency of approximately 3.0 ppm) can provide pH-weighted image contrast and may serve as an important imaging biomarker for human brain gliomas (17). The measured CEST contrast depends on various technical factors, including the shape, duration, length, amplitude and repetition time of the saturation pulse, and the strength of the scanning field, and the concentration of amine protons. Additionally, the image SNR can affect pH measurements (45). Furthermore, exchangeable protons from other proteins or macromolecules within the tissue may also influence the amine signal (27).Further research is needed to standardize CEST scanning protocols and post-processing techniques to optimize signal acquisition and data fitting. Additionally, larger-scale clinical studies are required to investigate pH variations among different tumor grades and molecular subtypes across the entire tumor. Thirdly, the sample size is relatively small, particularly for the 1p/19q expression status subgroup. Further validation in a larger cohort is necessary. As a single-center study, there are inherent limitations such as reduced generalizability and potential biases. Multi-center studies are needed to validate and expand upon these findings. Lastly, due to time constraints, only 2D single-slice imaging was performed, which might have missed important pathological regions due to intra-tumoral heterogeneity. Implementing 3D acquisition to cover the entire tumor could address this issue.

In summary, our findings indicate that quantitative assessment of tumor metabolism and microenvironment acidity through multi-pool Lorentzian analysis and pH-weighted analysis can serve as indicators for glioma grading, and for predicting IDH mutations, 1p/19q codeletion, and MGMT promoter methylation status. These metrics not only provide valuable insights into tumor subgroups but also reflect the heterogeneity within tumors.