Prognostic Value of Choline and Other Metabolites Measured Using 1H-Magnetic Resonance Spectroscopy in Gliomas: A Meta-Analysis and Systemic Review

Glioma is the most prevalent primary central nervous system malignant tumor, with high heterogeneity observed among different grades; therefore, non-invasive prediction of prognosis could improve the clinical management of patients with glioma. 1H-magnetic resonance spectroscopy (MRS) can estimate metabolite levels non-invasively. Multiple studies have investigated its prognostic value in gliomas; however, no consensus has been reached. PubMed and Embase databases were searched up to 20 October 2022 to identify studies investigating the prognostic value of metabolites using 1H-MRS in patients with glioma. Heterogeneity across studies was evaluated using the Q and I2 tests, and a fixed- or random-effects model was used to estimate the combined overall hazard ratio (HR). Funnel plots and Begg tests were used to assess publication bias. Higher choline levels were associated with shorter overall survival (HR = 2.69, 95% CI, 1.92–2.99; p < 0.001) and progression-free survival (HR = 2.20, 95% CI, 1.16–4.17; p = 0.02) in all patients; however, in pediatric gliomas, it showed no significant correlation with overall survival (HR = 1.60, 95% CI, 0.97–2.64; p = 0.06). The estimated choline level by 1H-MRS could be used to non-invasively predict the prognosis of patients with adult gliomas, and more studies are needed to evaluate the prognostic value of other metabolites.


Introduction
Gliomas are the most prevalent primary malignant tumors of the central nervous system (CNS) [1]. Despite the utilization of standard Stupp therapy combined with multiple chemotherapies, targeted therapy, immunotherapy, and tumor-treating field treatment, the overall survival (OS) of patients with glioma remains poor, varying among different World Health Organization (WHO) grades of glioma, with a 5-year survival rate of 7% for glioblastoma, the most aggressive subtype of gliomas [2]. Since there are highly heterogeneous malignancy and survival characteristics among gliomas, the development of a non-invasive tool to predict the prognosis of patients may help with the management of gliomas and improve the survival time and quality of life of patients.
Metabolic alterations within the tumor microenvironment are essential characteristics of cancer because of the high proliferation rate and demand of cancer cells [3]. Multiple studies have shown that metabolic changes in gliomas are associated with tumor grades [4], indicating that metabolites might be potential predictive biomarkers of prognosis. 1 Hmagnetic resonance spectroscopy (MRS) can detect various metabolites non-invasively and can provide estimated levels of choline (Cho), creatine (Cr), N-acetyl aspartate (NAA), lactate (Lac), and lipids [5]. Recent studies have shown that 2-hydroxyglutarate (2-HG), which is a product of isocitrate dehydrogenase (IDH)-mutant glioma, plays a key role in cancer metabolism reprogramming [6], with particular attention to metabolic alterations in the tumor microenvironment of glioma.

Literature Search and Selection of Studies
The current meta-analysis was conducted in accordance with the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [19]. PubMed and EMBASE were used to search for studies exploring the prognostic value of metabolites measured using MRS in gliomas up to 20 October 2022. The following search strategies were used: ((MR spectroscopy) OR (MRS)) AND (glioma) AND (survival) AND ((predict) OR (prognostic risk)). We also searched for potential studies by screening citations of the included studies and reviews. The search was limited to studies published in English. The protocol and systematic search strategy of the review are documented online (CRD42022368691) in the International Prospective Register of Systematic Reviews Registry (PROSPERO).

Eligibility Criteria
The search results were first screened for titles and abstracts, and further evaluated based on a full-text review. Three authors (Y.S., D.L., and Z.K.) independently assessed the search results for study inclusion, discussed potentially controversial studies, and reached an agreement.
Studies were considered eligible for inclusion if all the following criteria were met: (1) patients had preoperative or postoperative in vivo 1 H-MRS, (2) patients had histopathologically confirmed WHO grade 2-4 glioma, (3) patient outcomes were defined as OS or progression-free survival (PFS), and (4) the prognostic value of metabolites was measured using 1 H-MRS and evaluated with hazard ratio (HR) and 95% confidence interval (95% CI), or if there was sufficient data to calculate HR and 95% CI.
Studies were excluded if they entailed any of the following criteria: (1) a review article or conference abstract, (2) letters, editorials, and comments, (3) animal or in vitro studies, (4) studies with partially overlapping cohort data, and (5) studies assessing treatment response. For studies with overlapping data, the study with the completed study results was selected.

Data Extraction and Quality Assessment
Three authors (Y.S., D.L., and Z.K.) independently extracted information using a standardized extraction form, including study and patient characteristics, and MRI characteristics of the selected studies. Firstly, the study and patient characteristics were obtained: author, year of publication, country, study design, number of patients, glioma subtypes, WHO grade of glioma, primary or recurrent gliomas, the median age of patients, and male/female ratio of patients. Secondly, MRI characteristics were obtained: magnetic strength, MRS techniques, echo time (TE/ms), software for postprocessing of MRS imaging, the timing of MRS, metabolites utilized to predict prognosis, the cutoff value, and the assessment of outcomes. (The latter included HR and the corresponding 95% CI of PFS and/or OS. If they were not given, particularly in an article, the essential data that were used to estimate them were collected. The estimation methods used were reported by Tierney et al. [20]). If the results of the univariate and multivariate analyses were both stated in a previous study, the multivariate analysis was included in the analysis.
The quality of the enrolled studies was assessed using Newcastle-Ottawa Scale (NOS) for cohort studies [21]. Cohort selection (score 0-4), comparability (score 0-2), and outcome (0-3) were independently evaluated by two authors (Q.L. and H.X.), with a total score of 0-9 for each study. If a disagreement occurred, a third author (Y.W.) assessed the study and reached a consensus.

Statistical Analyses
The heterogeneity of HRs across studies was evaluated using Q and I 2 statistics. If there was no significant heterogeneity across the studies (I 2 < 50%, p > 0.1), fixed-effects models were used for the combined risk estimates. If there was significant heterogeneity (I 2 ≥ 50%, p ≤ 0.1), random effects models were used. Sensitivity analysis was used to assess the potential reasons for heterogeneity and verify the reliability of the results. Potential publication bias was qualitatively evaluated using a funnel plot and quantitatively evaluated using the Begg test. Review Manager 5.4.1 (The Cochrane Collaboration, London, UK, 2020) and STATA version 17.0 (StataCorp LLC, College Station, TX, USA) were used for statistical analyses. A p-value < 0.05 was considered statistically significant.

Literature Search
A total of 170 studies were identified through a database search, and after removing duplicates, 135 studies were screened based on titles and abstracts and evaluated based on full-text articles according to our eligibility criteria. A total of 121 studies were excluded for the following reasons: 10 were conference abstracts, 49 were not in the fields of interest, 31 were reviews, 5 were animal studies, 2 were letters/study protocols, 2 included central nervous system (CNS) tumors other than gliomas, 14 had no available or calculable HR and corresponding 95% CI, 6 assessed treatment response, 1 analyzed partially overlapping patient cohorts, and 1 utilized HRMAS-NMR ( Figure 1). In total, 14 studies with 568 patients investigating the prognostic value of 1 H-MRS were included in the final meta-analysis and systemic review.

Study Characteristics
The NOS scores for the 14 high-quality studies ranged from 7-9 (Supplementary Table S1). These studies were published between 2000 and 2022, of which six were conducted in the USA, two in China, one in Canada, one in Germany, one in Japan, one in Mexico,

Study Characteristics
The NOS scores for the 14 high-quality studies ranged from 7-9 (Supplementary Table S1). These studies were published between 2000 and 2022, of which six were conducted in the USA, two in China, one in Canada, one in Germany, one in Japan, one in Mexico, one in Poland, and one in Sweden. A total of 568 patients were included in the current analysis and review, with a median age ranging from 13.5 to 57 years. Some studies focused on pediatric gliomas, particularly diffuse intrinsic pontine glioma (DIPG), and others enrolled patients with adult glioma, including astrocytoma, oligodendroglioma, and glioblastoma, varying from WHO grades 2-4 ( Table 1). The parameter and post-processing software used in the selected studies varied, and the magnetic strength and echo time (TE) chosen were mostly 3T and 144ms, respectively. All studies utilized 1 H-MRS and point-resolved spectroscopic selection (PRESS) to detect metabolites, including Cho, lactate, Cr, NAA, glycine, and 2-HG, of which nine studies considered the prognostic value of Cho/Cr and/or Cho/NAA. Thirteen studies evaluated the predictive value of metabolites for OS and four studies investigated PFS (Table 2).

Choline and Progression-Free Survival
In total, three studies evaluated the associations between Cho and PFS [5,13,17], two studies used Cho/NAA as the predictive parameter, one study assessed Cho/Cr, and one study assessed the volume of regions with Cho/NAA > 2. The Q test and I 2 statistic showed p < 0.1, I 2 = 82%; therefore, a random-effects model was used. The results showed no significant association between Cho and PFS (HR = 1.64, 95% CI, 0.82-3.28; p < 0.001). Since heterogeneity was observed, sensitivity analysis was conducted, and when studies using Cho/NAA and Cho/Cr were included, no heterogeneity was observed across studies (p = 0.21, I 2 = 36%) and better results were obtained, suggesting that Cho was associated with shorter PFS (HR = 2.20, 95% CI, 1. 16-4.17; p = 0.02) (Figure 4).

Choline and Progression-Free Survival
In total, three studies evaluated the associations between Cho and PFS [5,13,17], two studies used Cho/NAA as the predictive parameter, one study assessed Cho/Cr, and one study assessed the volume of regions with Cho/NAA > 2. The Q test and I 2 statistic showed p < 0.1, I 2 = 82%; therefore, a random-effects model was used. The results showed no significant association between Cho and PFS (HR = 1.64, 95% CI, 0.82-3.28; p < 0.001). Since heterogeneity was observed, sensitivity analysis was conducted, and when studies using Cho/NAA and Cho/Cr were included, no heterogeneity was observed across studies (p = 0.21, I 2 = 36%) and better results were obtained, suggesting that Cho was associated with shorter PFS (HR = 2.20, 95% CI, 1.16-4.17; p = 0.02) (Figure 4). p < 0.1, I 2 = 82%; therefore, a random-effects model was used. The results showed no significant association between Cho and PFS (HR = 1.64, 95% CI, 0.82-3.28; p < 0.001). Since heterogeneity was observed, sensitivity analysis was conducted, and when studies using Cho/NAA and Cho/Cr were included, no heterogeneity was observed across studies (p = 0.21, I 2 = 36%) and better results were obtained, suggesting that Cho was associated with shorter PFS (HR = 2.20, 95% CI, 1. 16-4.17; p = 0.02) (Figure 4). .

Publication Bias
Potential publication bias was assessed using a funnel plot and Begg's test ( Figure 5). The funnel plot showed moderate asymmetry; however, Begg's test showed no evidence of publication bias among the included studies (p = 0.0856).

Publication Bias
Potential publication bias was assessed using a funnel plot and Begg's test ( Figure 5). The funnel plot showed moderate asymmetry; however, Begg's test showed no evidence of publication bias among the included studies (p = 0.0856).

Discussion
This study suggests that Cho levels estimated by 1 H-MRS have great prognostic value in gliomas, both for OS and PFS. Subgroup analysis showed that Cho level is less predictive of survival in pediatric gliomas. Lower Cho/NAA, the volume of regions with Cho/NAA > 2, and Cho/Cr are all associated with shorter OS. Previous studies have

Discussion
This study suggests that Cho levels estimated by 1 H-MRS have great prognostic value in gliomas, both for OS and PFS. Subgroup analysis showed that Cho level is less predictive of survival in pediatric gliomas. Lower Cho/NAA, the volume of regions with Cho/NAA > 2, and Cho/Cr are all associated with shorter OS. Previous studies have confirmed that Cho is associated with membrane synthesis and degradation. NAA represents a neuronal function, and Cr participates in energy metabolism [22,23]. Higher levels of both Cho/NAA and Cho/Cr indicate CNS tumor malignancy and distinguish the WHO grades of gliomas [4]. A recent study conducted by Gao et al. [16] suggested the Cho/Cr ratio as a biomarker for cellular proliferation and that it could be used to predict the prognosis of glioma. Additionally, Pucci et al. [24] showed that Cho could promote the proliferation of glioblastoma cells by activating the AKT and ERK pathways, suggesting that it contributes to the aggressiveness of glioblastoma. Despite the use of MRS as a non-invasive estimate of Cho, quantitative features from CHO PET were also investigated and found to be effective in distinguishing the WHO grades of glioma [25] and predicting molecular alterations in glioma [26].
In pediatric gliomas, particularly DIPG, Hipp et al. [12] and Yamasaki et al. [13] found that Cho/NAA was prognostic. Yet, Cho/Cr showed no significant value in predicting OS and PFS, which might partially explain why our analysis showed no prognostic value for Cho in pediatric gliomas. However, further studies regarding the function of Cho in pediatric gliomas are lacking, and more studies are warranted to determine the use of Cho as a predictive biomarker and explain the underlying molecular mechanism.
In the WHO 2021 guidelines for CNS tumors, molecular alterations have been regarded as an essential part of glioma classification [27], and 2-HG holds considerable importance because it is the oncometabolite of IDH mutation [28]. 2-HG competitively inhibits α-ketoglutarate-dependent dioxygenases, leading to epigenetic disorders and cell differentiation blocks, and can also improve HIF-1 α level, induce angiogenesis, and mediate the escape and metastasis of tumor cells [29]. Consistent with clinical experience, Tiwari et al. [3] found that lower survival risk was associated with higher levels of 2-HG and the presence of IDH mutations. Interestingly, Autry et al. [18] revealed that higher levels of 2-HG/Cr indicated a significant reduction in PFS, which contradicts our clinical impression. However, this study enrolled patients with IDH mutations, suggesting that in patients with IDH mutation, 2-HG levels might play an adverse role in survival, and a more complicated metabolic network might be involved in glioma growth.
Recently, more investigations on metabolic heterogeneity of glioblastoma utilizing MRS have been conducted. Grande et al. applied MRS in glioblastoma stem-like cells and detected multiple metabolites in vitro, indicating an important role of mitochondrial fatty oxidation in energy supplements in glioblastoma stem-like cells [30]. They also discovered specific metabolic signatures of glioblastoma stem-like cells after stressful treatments via MRS detection [31]. All the studies suggest more advanced utilization of MRS in metabolic investigations and clinical practice, and further investigations could be conducted.
Our study has some limitations. Firstly, we included several retrospective studies, and the baseline characteristics of enrolled patients were not steady, hence making it vulnerable to selection bias. The variety of device parameters and selected samplings would result in potential bias and heterogeneity, and well-designed prospective studies may be conducted to further evaluate the prognostic value of MRS. Additionally, since the studies selected in our meta-analysis were between 2000 and 2022, there are improvements in the management of glioma-induced heterogeneity. Although we conducted a subgroup analysis for essential factors, other confounding factors remained. Finally, owing to the limited number of studies, several subgroups could not be included in the meta-analysis, such as the predictive value of Cho for PFS, the prognostic value of other metabolites, and the predictive value of Cho in pediatric low-grade glioma and high-grade glioma. Thus, more studies to investigate the remaining problem are warranted. Therefore, further studies are needed to determine other metabolites measured by 1 H-MRS as prognostic biomarkers.

Conclusions
The estimated level of Cho by 1 H-MRS could be used to non-invasively predict the prognosis of patients with adult gliomas, and its use in predicting the survival time of pediatric patients with glioma should be carefully considered. More studies are needed to evaluate the prognostic value of other metabolites such as 2-HG, glycine, and lactate.