Skip to main content
Download PDF

Radiogenomic profiling of global DNA methylation associated with molecular phenotypes and immune features in glioma

BMC Medicine volume 22, Article number: 352 (2024) Cite this article

Abstract

Background

The radiogenomic analysis has provided valuable imaging biomarkers with biological insights for gliomas. The radiogenomic markers for molecular profile such as DNA methylation remain to be uncovered to assist the molecular diagnosis and tumor treatment.

Methods

We apply the machine learning approaches to identify the magnetic resonance imaging (MRI) features that are associated with molecular profiles in 146 patients with gliomas, and the fitting models for each molecular feature (MoRad) are developed and validated. To provide radiological annotations for the molecular profiles, we devise two novel approaches called radiomic oncology (RO) and radiomic set enrichment analysis (RSEA).

Results

The generated MoRad models perform well for profiling each molecular feature with radiomic features, including mutational, methylation, transcriptional, and protein profiles. Among them, the MoRad models have a remarkable performance in quantitatively mapping global DNA methylation. With RO and RSEA approaches, we find that global DNA methylation could be reflected by the heterogeneity in volumetric and textural features of enhanced regions in T2-weighted MRI. Finally, we demonstrate the associations of global DNA methylation with clinicopathological, molecular, and immunological features, including histological grade, mutations of IDH and ATRX, MGMT methylation, multiple methylation-high subtypes, tumor-infiltrating lymphocytes, and long-term survival outcomes.

Conclusions

Global DNA methylation is highly associated with radiological profiles in glioma. Radiogenomic global methylation is an imaging-based quantitative molecular biomarker that is associated with specific consensus molecular subtypes and immune features.

Peer Review reports

Background

The imaging markers identified from the radiomic profiles with comprehensive traits of cancers have been shown to predict clinical outcomes with high accuracy [1,2,3,4,5]. A number of studies have made efforts to develop genome-wide hallmarks of cancer, which could systematically assess molecular alterations and improve the early management and therapeutic intervention for cancers [6, 7]. Recent radiogenomic studies have explored the linkage between tumor imaging phenotypes and molecular alterations that are valuable in clinical decision-making [8,9,10,11,12]. These efforts have promoted personalized cancer care by recognizing the radiomic signatures that were associated with specific molecular phenotypes, including mutations of isocitrate dehydrogenase (IDH) [13], epidermal growth factor receptor (EGFR) [14], and telomerase reverse transcriptase (TERT) [15], disturbing pathways [16, 17], and microsatellite status [18, 19].

Among these molecular hallmarks, DNA methylation-based markers have reshaped the clinical decisions in glioma [20]. IDH mutation-related DNA methylation patterns, such as glioma-CpG island methylator phenotype (G-CIMP), are highly associated with the prognosis of glioma and are commonly used to stratify patients in the WHO classification [21,22,23]. Besides the well-elucidated function of the hypermethylated promoter in gene silencing, such as O-6-methylguanine-DNA methyltransferase (MGMT) [24, 25], several studies have reported the role of global DNA methylation loss in genetic alterations and oncogenesis in glioma [26,27,28]. Recent studies also reported the association of global DNA methylation with immune response [28] and intratumoral heterogeneity [29] in glioma. Furthermore, global DNA methylation has been utilized as a marker for phenotyping in gliomas [30, 31]. However, the association of global DNA methylation with current molecular phenotypes and clinical outcomes remains to be explored.

Despite the significance of global DNA methylation and related molecular phenotypes, there is still a lack of a fast and non-invasive approach to evaluate global DNA methylation with cost-effective and less labor-intensive procedures. Considering that the radiomic feature is an emerging non-invasive marker for tumor phenotyping, we performed this radiogenomic analysis to characterize the connection between radiological images and multi-omic profiles in gliomas. In this study, we generated a comprehensive radiogenomic map that could align radiological images to quantitative molecular profiles. To profile the global DNA methylation level, we analyzed the methylation array probes targeting genomic repetitive elements. In addition, we devised two novel enrichment approaches to enable the radiological interpretation of molecular profiles. The novel radiological insights into the specific molecular alterations in tumors may provide the understanding of molecular alteration in shaping the tumor features that decide the radiological images and facilitate the imaging marker-driven clinical trials.

Methods

Study patients

To analyze the association of imaging features with molecular profiles in tumors, we analyzed glioma patients with molecular and magnetic resonance imaging (MRI) profiles according to the research protocol approved by the Institutional Review Board of The Sixth Affiliated Hospital of Sun Yat-sen University. The molecular and imaging profiles along with clinicopathological information and treatment outcomes were acquired from the data release of The Cancer Imaging Archive [32] and The Cancer Genome Atlas [33, 34]. The patients with glioblastoma (GBM) or low-grade glioma (LGG) and available pre-operative multimodal MRI (mMRI) were included. The patients with inconsistent MRI scanners or without complete sequences of MRI data were excluded. Finally, a total of 146 patients were included for subsequent radiogenomic analysis (Additional file 1: Fig. S1). The baseline characteristics of study patients were shown in Additional file 2: Table S1.

Molecular and immunological characterization

We calculated the tumor mutation burden (TMB) of each case by dividing the total number of mutations by the size of the coding region [35]. The intra-tumor heterogeneity (ITH) score evaluated at the genomic level based on somatic mutations and copy number variation (CNV) profiles of bulk tumors was acquired from previous studies [36]. The well-characterized molecular subtypes for gliomas, including pan-glioma DNA methylation subtype [37], pan-glioma RNA expression subtype [37], reverse-phase protein array (RPPA) subtype [38], transcriptional subtype [39, 40], and glioma subtype [37], were obtained from previous studies. In addition, we analyzed the three subtypes classified by Brat et al. with the IDH mutation and 1p/19q codeletion status: IDH-mutant glioma with 1p/19q codeletion, IDH-mutant glioma without 1p/19q codeletion, and IDH-wild glioma [33]. Furthermore, we characterized the MGMT promoter methylated status [41], ATRX mutation [37] and TERT expression [42], which were shown to have biological and clinical significance. The fractions of 22 tumor-infiltrating immune cells in the tumor microenvironment of each sample were estimated by CIBERSORT [43]. The detailed information for each molecular phenotype and marker was shown in Additional file 2: Table S2.

Global DNA methylation profiling

The DNA methylation data profiled by Illumina Infinium HumanMethylation450 (450 K) BeadChip array was obtained and analyzed as we previously described [44, 45]. To determine the global DNA methylation level of each sample, we analyzed the methylation probes for which at least 90% of sequences (≥ 45 bp) mapped to the young subfamilies of long interspersed nuclear element-1 (LINE-1) and Alu [28, 46]. The documented probes targeted to Alu and LINE-1 repeated sequences were obtained as previously described [46], and we averaged the beta values of these probes to estimate the global DNA methylation of each tumor.

Radiomic feature extraction

The MRI images were used to extract radiomic features for GBM and LGG as previously described [32, 47]. Four MRI sequences were available, including T1-weighted (T1WI), gadolinium-enhanced T1-weighted (T1WI + Gd), T2-weighted (T2WI), and T2-fluidattenuated inversion recovery (T2-FLAIR) images (Additional file 3: Supplementary Method 1 [47,48,49,50,51,52]). The extracted radiomic features of regions of interest (ROI), including enhancing region of the tumor core (ET), non-enhancing region of the tumor core (NET), and peritumoral edema (ED), were obtained from the previous studies using the software “Cancer Imaging Phenomics Toolkit” (CaPTK) [53, 54]. As a result, a total of 726 radiomic features comprising of 5 subgroups were obtained from each individual case: (1) intensity features; (2) histogram-based features; (3) volumetric features, including volume, spatial, morphologic, and solidity; (4) textural parameters, including gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), gray-level size zone matrix (GLSZM), and neighborhood gray-tone difference matrix (NGTDM); (5) tumor growth model (TGM) parameter [55] (Additional file 2: Table S3).

Framework for mapping radiogenomic signature to molecular profile

The patients were randomly divided into the training and validation sets with a size ratio of 7:3. We applied the multi-step pipeline that integrates correlation or ANOVA test followed by machine-learning-based feature selection to construct models for each molecular feature. For continuous variables, including global DNA methylation loss, TMB, and ITH, the Pearson correlation test was performed to select the radiomic features linearly related to molecular features for 9 molecular feature-specific radiomic sets (MRS9) matrix; then, a liner regression optimized by the elastic-net method was applied to further select the features and construct MoRad models with liner combination of these features by their coefficients. For categorical variables, the ANOVA or t-test was performed for each molecular feature to select the molecular phenotype-specific radiomic features for the MRS9 matrix; then, a multinominal regression optimized by the least absolute shrinkage and selection operator (LASSO) method was applied to further select radiomic features and generate MoRad models that could classify each patient into the specific molecular phenotype. We have shown that this pipeline could generate well-performed predicting models in the previous studies [2, 56] by eliminating the potential redundance, overfitting, and bias induced by confounding features in model construction. Finally, radiomic models for profiling each molecular feature were constructed based on selected radiomic features and their coefficients with elastic net regression or LASSO approaches (Additional file 3: Supplementary Method 2).

Radiomic ontology (RO) and radiomic set enrichment analysis (RSEA)

To provide radiological insights into the molecular profiles of tumors, we devised two tools called “RSEA” and “RO” analysis. Both RO and RSEA evaluate radiomic and molecular data at the level of radiomic sets. We developed the RSEA tool on the basis of Subramanian et al.’s Gene Set Enrichment Analysis methodology [57]. We first hierarchically classified extracted radiomic features into four predefined radiomic sets based on prior clinical and radiological knowledge and CaPTK instruction [53, 54]. Briefly, they were categorized into three subtypes: (1) image sequences, including T1, T1Gd, T2, and FLAIR; (2) lesion location, including ET, ED, and NET; (3) algorithm features, including intensity, histogram, volumetric, textural, and TGM parameter. Detailed information and description of the radiomic sets were provided in supplementary files (Additional file 2: Table S4). To perform RSEA, the 726 radiomic features were initially ranked based on the coefficients of the correlation with each molecular feature. The goal of RSEA was to determine whether the members of a specific radiomic set were randomly distributed throughout the feature list or primarily found at the top or bottom, in which case the radiomic set was believed to correlate with the molecular variable. The normalized enrichment scores (NES) and corresponding statistical significance were calculated by increasing a running-sum statistic when we encountered a radiomic feature in the specific set and decreasing it when we encountered radiomic features not in the set as we traversed the entire list. It reflects the degree to which a specific radiomic set was overrepresented at the extremes (top or bottom) of the entire list. In the RO analysis, the selected radiomic features were tested for their associations with specific radiomic sets using the hypergeometric test. We calculated the statistical significance of which radiomic sets appeared more frequently than would be expected by chance when examining the sets annotated to the input features. As the detailed feature matrix is not required, RO is applicable when the feature values are not accessible and convenient for quick enrichment as a primary screening. A full description and discussion of the methodology and collected radiomic sets for running RSEA and RO were provided in the supplementary materials.

Statistical analysis

The group comparisons were performed using the t-test or Mann–Whitney U test according to the variable distribution. Pearson’s test was performed for correlation analysis. The Kaplan-Meier analysis was used to estimate and compare the long-term survival. The principal component analysis (PCA) was conducted using the “ggord” package. All the statistical analyses were conducted with the R software version 4.0.0 (http://www.R-project.org). The statistical significance levels were all set to be 0.05 with two sides.

Results

Analysis of radiomic signatures associated with molecular profiles

The concept and design of this study were shown in Fig. 1a. A total of 136,154, 131, and 16 radiomic features were identified to be linearly related to Alu methylation, LINE-1 methylation, TMB, and ITH, respectively. For the molecular subtypes, we identified 264, 178, 156, 296, and 137 most varied radiomic features among subtypes for the pan-glioma DNA methylation subtype, pan-glioma RNA expression subtype, transcriptional subtype, glioma subtype, and RPPA subtype, respectively. As a result, MRS9 was generated (Additional file 2: Table S5). Overall, the radiomic features in MRS9 were well correlated with molecular profiles (Fig. 1b).

Fig. 1
figure 1

Landscape of radiomic and molecular features in glioma. We developed the MoRad tool with machine learning algorithm to bridge the gap between visible images and invisible molecules, including the genomic landscape of methylation, mutation, and transcriptomic and proteomic profiles (a). By using MoRad, we demonstrated the radio-molecular trajectories mapping the molecular features to radiological images. The union of primary screening features of ten MoRad models gained 436 radiomic features and the heatmap was created for an overview of the radiomic atlas (b). Overall, the unsupervised hierarchical clustering based on the union features showed the connection between the molecular profiles and radiomic features. LINE-1, long interspersed nuclear element-1; TMB, tumor mutation burden; ITH, intra-tumor heterogeneity; CL, classical; ME, mesenchymal; NE, neural; PN, proneural; RPPA, reverse-phase protein array

Full size image

Strong association of radiological features with DNA methylation

To further test the association of radiomic signatures with molecular profiles, we sought to map the radiomic features to molecular profiles. We developed the fitting model for each molecular feature using the MRS9. Overall, the generated MoRad models performed well for profiling each molecular feature, in which the molecular features profiled by MoRad were significantly correlated with their actual status profiled by array or sequencing, including mutational, methylation, transcriptional, and protein profiles (Fig. 2 and Additional file 1: S2).

Fig. 2
figure 2

Development and validation of MoRad tool for molecular profiling by using radiomic features. The correlation analysis showed that the MoRad-profiled methylation levels were highly correlated with array-profiled methylation levels for Alu and LINE-1 in the training, validation and whole sets. This correlation was further confirmed in the IDH-mutation and IDH-wild subgroups for the MoRad-Alu and MoRad-L1 models

Full size image

Among the molecular profiles, MoRad models have a remarkable performance in detecting DNA methylation profiles. The array-profiled global DNA methylation was linearly correlated with those profiled by radiogenomic signatures in the training (Alu, r = 0.86, P < 0.001; LINE-1, r = 0.79, P < 0.001; Fig. 2) and validation (Alu, r = 0.67, P < 0.001; LINE-1, r = 0.7, P < 0.001) sets (Fig. 2). In addition, this linear correlation was confirmed in both IDH-mutant and IDH-wild gliomas (Fig. 2). Of interest, the linear association of radiogenomic signatures with TMB was inferior to that with global DNA methylation (Additional file 1: Fig. S2). Expectedly, the MoRad models could well discriminate the methylation subtypes of LGm2 and G-CIMP-low with area under curves (AUCs) of 0.970 and 0.973, respectively (Additional file 1: Fig. S2). Taken together, we uncovered a strong association between radiological features and DNA methylation and constructed the radiogenomic signature that could be used to quantitatively map global DNA methylation.

Radiological annotation of the molecular features with devised ontology analysis framework

To provide radiological insights into the molecular profiles, we devised two algorithms and applied them in the integrating analysis of radiomic and molecular data (Fig. 3a). The RO enrichment analysis for DNA methylation-specific radiomic features showed that global DNA methylation could be mapped to the volumetric and textural features in the enhanced region of gliomas (Fig. 3b and Additional file 2: Table S6). In addition, TMB could be mapped to the volumetric and textural features of peritumoral edema in the T1WI + Gd images (Fig. 3b and Additional file 2: Table S6). However, no radiomic sets were significantly enriched in the RO analysis for other molecular features.

Fig. 3
figure 3

Radiomic set enrichment analysis (RSEA) and radiomic ontology (RO) analysis. The workflow of RSEA and RO analysis (a). The RO enrichment for the top correlated radiomic features with molecular-profiled features showed that global DNA methylation variation was associated with textural and volumetric abnormal of enhanced gliomas, which could be observed more clearly in T1- weighted images (b). TMB could be mapped to the volumetric and textural features of ED in the T1WI + Gd images (b). The RSEA indicated that global DNA methylation variation was mostly associated with textural features in enhanced gliomas of T2 series images with the highest normalized enrichment scores (NES) (c). The TMB and ITH were found to be reflected by the textural features. TMB was featured in the enhanced region and ED from both T1WI and T2WI images and ITH was featured in both the enhanced and non-enhanced regions (c). The RO analysis of the most common feature sets included in at least 7 MoRad models showed that all molecular characterization could be reflected by Busyness NGTDM and volumetric abnormalities that were most apparent in the enhanced region of gliomas (d). The RO analysis for the unique feature sets of specific MoRad models showed the exclusive feature set enriched for transcriptional subtype, glioma subtype, pan-glioma DNA methylation subtype, pan-glioma RNA expression subtype, and the intratumor heterogeneity (d). The dotted line represented the cut-off P value of 0.05. ITH, intra-tumor heterogeneity; T1WI, T1-weighted; T1WI + Gd, gadolinium-enhanced T1-weighted; T2WI, T2-weighted; ET, enhancing region of the tumor core; ED, peritumoral edema; NET, non-enhancing region of the tumor core; TGM parameter, tumor growth model parameter; NGTDM, neighborhood gray-tone difference matrix; GLSZM, gray-level size zone matrix; GLN, gray-level nonuniformity

Full size image

Next, we investigated the MRS9 with the RSEA tool. We found that the global DNA methylation level could be represented in the enhanced region of T2WI with textural variation (Fig. 3c and Additional file 2: Table S7). Moreover, TMB and ITH could be represented by the textural features. However, TMB was featured in the enhanced region and peritumoral edema from both T1WI and T2WI images, while ITH was featured in both the enhanced and non-enhanced regions. (Fig. 3c and Additional file 2: Table S7).

We further explored the unique and common radiological features among MRS9. The global DNA methylation-specific radiogenomic signatures shared several common features with other molecular profile-specific radiogenomic signatures (Additional file 1: Fig. S3). Next, we performed RO analysis for the most common features across all the molecular profiles. Interestingly, we found that most molecular profiles could be represented by the enhanced region with volumetric and textural alterations (Fig. 3d and Additional file 2: Table S8). We also performed RO analysis for the unique radiogenomic features that were exclusive for each molecular profile. We found that the DNA methylation subtype had unique features in the non-enhancing region with the volumetric and histogram-based alterations, which is distinct from the unique features for mutational, transcriptional, and protein subtypes (Fig. 3d and Additional file 2: Table S8). Taken together, we provided the radiological annotation for global DNA methylation and showed its exclusive radiological features in the non-enhancing region compared with those of other molecular profiles.

Radiogenomic signatures for global DNA methylation are associated with clinicopathological and immunological features

We next investigated the association of MoRad-profiled molecular features with clinicopathological features and clinical outcomes. The radiogenomic global methylation was defined as the MoRad-profiled methylations of Alu and LINE-1, which was found to be successively decreased as the histological grade increased (all P < 0.05; Fig. 4a). The radiogenomic global methylation loss was more prevalent in GBM compared with astrocytoma, oligodendroglioma, and oligoastrocytoma (all P < 0.05; Fig. 4b). Moreover, the radiogenomic global methylation was significantly lower in unmethylated-MGMT, ATRX-mutation, TERT-expressed, and IDH-wild tumors (all P < 0.05; Fig. 4c, e). In the comparison among DNA methylation-based molecular phenotypes, including glioma subtype and pan-glioma DNA methylation subtype, the radiogenomic global methylation was significantly lower in the well-documented phenotypes with lower genome-wide methylation such as classic-like, mesenchymal-like, G-CIMP-low, and LGm4-6 subtypes (all P < 0.05; Fig. 4c, e). Interestingly, in the gene expression-based phenotypes, we found radiogenomic methylation loss was more prevalent in the RPPA-K1, classic-like, and LGr4 subtypes (all P < 0.05; Fig. 4c, e).

Fig. 4
figure 4

Association of MoRad models with current cancer omics and clinical information. The MoRad-profiled methylation levels with MoRad-Alu and MoRad-L1 models were both significantly different among the groups defined by histological grades (a) and classification (b). The MoRad-profiled methylation levels were significantly different among consensus molecular subtypes (c, e). The Kaplan–Meier analysis showed that both low MoRad-Alu and MoRad-L1 groups had longer survival than high groups (d, f). Both the MoRad-Alu and MoRad-L1 were negatively correlated with the ratio of M2 macrophage in tumor microenvironment. The MoRad-L1 was also negatively correlated with the ratio of follicular T helper cell, lymphocyte, resting memory CD4 + T cell, and CD8 + T cell (g). The range between the lower whisker and lower edge of the box covers the minimum to the lowest 25% of the data values, and the range between the upper whisker and upper edge of the box covers the highest 25% to the maximum of the data values. *, P < 0.05

Full size image

Furthermore, we analyzed the prognostic value of global DNA methylation in gliomas. We found that radiogenomic global methylation loss was significantly associated with unfavorable survival outcomes (all P < 0.001; Fig. 4). Finally, we investigated the association of immune response with global DNA methylation. As a result, we found that the radiogenomic global methylation loss was correlated with a specific glioma microenvironment featured by abundant infiltration of lymphocyte, M2 macrophage, follicular T helper cell, resting memory CD4 + T cell, and CD8 + T cell (Fig. 4g). Taken together, we demonstrated the comprehensive associations of global DNA methylation with clinicopathological and molecular features.

Discussion

In the current study, we mapped the features in radiological images to molecular profiles in glioma. We uncovered the innate associations between visible images and invisible molecules using matched MRI images and molecular profiles of diffuse gliomas. We found that radiological images could represent molecular features, including DNA methylation and gene expression. Among them, the MoRad signature could accurately reflect the global DNA methylation. Two approaches called RO and RSEA with tractable pipelines were devised and demonstrated to yield readable radiological annotation of global DNA methylation. We found that global DNA methylation could be reflected by the volumetric and textural features in the enhanced region of gliomas. We also demonstrated the associations of global DNA methylation with clinicopathological, molecular, and immunological features, including histological grade, mutations of IDH and ATRX, MGMT methylation, multiple methylation-high subtypes, tumor-infiltrating lymphocytes, and long-term survival outcomes.

Previous studies have reported that IDH-1 mutation could induce the distinct increase of genome-wide methylation of gliomas probably by inhibiting DNA demethylases, which is associated with the G-CIMP and favorable survival outcome [37, 42]. LGGs are prevalent to carry IDH mutations (65–90%), while the vast majority of primary GBMs retain wild-type IDH [13, 39]. In our study, patients with IDH mutations had significantly higher radiogenomic global methylation, which is consistent with previous studies. Moreover, we performed the analysis in the IDH-mutation (IDH-MUT) and IDH-wild type (IDH-WT) groups separately to clarify that the association of DNA methylation with radiological features was stable independent of IDH mutation.

The MoRad achieved an inferior performance in profiling TMB compared with that in profiling global DNA methylation. We speculated that DNA methylation may play a superior role in deciding the heterogeneity in radiological images. The RSEA analysis indicated that TMB was associated with the imaging changes in peritumoral edema which was in concordance with previous studies that the peripheral tumor area was important in determining TMB and immunotherapy response [58, 59]. The NET and ET features were both enriched for the ITH feature sets in the RSEA analysis, suggesting that the involvement of tumor heterogeneity may not be limited in the enhanced region, which can be supported by the findings in the previous study [60].

Radiogenomic profiling of global DNA methylation is widely associated with multiple molecular subtypes. In the pan-glioma DNA methylation subtype, the patients with higher MoRad-profiled global DNA methylation levels were mostly enriched in the LGm1-3 groups that mostly exhibit IDH mutation and increased genome-wide methylation [37]. In the methylation-based glioma subtype, higher MoRad-profiled global DNA methylation was recognized in the Codel and G-CIMP-high groups, which were mostly hypermethylated with prolonged survival [37]. In addition, the G-CIMP-low patients had the lowest MoRad-profiled global DNA methylation level among the three groups. In the pan-glioma RNA expression cluster, the MoRad-profiled global methylation level was significantly lower in the LGr4 group that has been recognized to exclusively harbor IDH-WT tumors. Taken together, these association analyses further validate the MoRad models and demonstrate its biomedical significance.

DNA methylation loss has been well recognized as a surrogate marker for advanced tumor phenotype and unfavorable outcomes in various types of tumors [61, 62]. The strong association of global DNA methylation with radiomic signatures in gliomas may reveal the critical role of genomic-wide methylation alteration in shaping radiological image by regulating tumor growth and differentiation. Previous studies have discussed the epigenetic alteration in glioma and regarded it as an important event in the development of glioma [63, 64]. By using MoRad in profiling the epigenetic landscape, we found that patients with global DNA methylation loss tended to have higher malignancy grades and unfavorable outcomes, which was concordant with current knowledge [65]. Furthermore, several radiological and pathological studies have uncovered the heterogeneity in the enhanced region with microvascular proliferation and/or pseudo-palisading necrosis in GBMs [66]. Our study validated these reports by finding that patients with GBMs have the lowest MoRad-profiled global DNA methylation level among four histologic classifications.

Interestingly, we found that the loss of DNA methylation determined by MoRad was associated with abundant infiltration of follicular T helper cell, lymphocyte, resting memory CD4 + T cell, and CD8 + T cell. The global DNA methylation loss might serve as a source of genetic alteration and thus be correlated with an elevated burden of neoantigen [26], which is a potential explain for the association between T cell infiltration and global DNA methylation loss. In addition, we found that the infiltration of macrophages was negatively correlated with DNA methylation. Kong et al.’s study found that global DNA methylation loss in various kinds of cancers, including LGG and GBM, could reactivate the transposable elements such as LINE and stimulate immune cell infiltration via MHC or antiviral responses. The infiltrated immune cells, including macrophages, may exert a complicated impact on glioma. As described in the previous studies, intense infiltration of M2 macrophages could facilitate glioblastoma growth by secreting abundant pleiotrophin to stimulate glioma stem cells [67, 68], which may account for unfavorable survival outcomes of patients with radiogenomic global methylation loss.

To bridge the gap between invisible molecules and visible images, it is crucial to interpret the high-dimension radiomic information. Therefore, we developed the “RO” and “RSEA” tools to perform radiomic feature enrichment analysis, which could translate the algorithm-based radiomic features into human-readable imaging terms that closely relate to biomedical settings to help understand the radiogenomic trajectories and better guide clinical practice. The RO and RSEA analysis for MRS9-L1 and MRS9-Alu feature sets suggested that the variation of global DNA methylation was mainly related to volumetric alteration along with some textural changes, which were prone to be observed in enhanced tumor region of T1WI and T2WI images. Biologically, the enhanced tumor region implied leakage of contrast agents caused by severely disrupted blood–brain barrier and disordered vascularization, which were commonly seen in advanced gliomas [69].

Recent landmark studies have completed comprehensive molecular characterization for glioma to discriminate patients with different clinical outcomes after specific targeted therapy by stratifying them based on molecular profiles [33]. In our study, we constructed MoRad models to profile five molecular phenotypes based on epigenomics, genomics and proteomics. As a result, MoRad was able to discriminate the subtypes with good performance. This may provide a non-invasive imaging marker panel for stratifying patients to facilitate cancer treatment and biomarker-based clinical trials. However, there are some limitations in our study. The sample size of our study was limited, and the generalization performance of the models would be improved by training our pre-trained MoRad models in a larger dataset with diverse and balanced histological subtypes of gliomas. Although we have validated the models in the validation cohort, their performance should be further tested in the external cohorts.

Conclusions

In this study, we performed a comprehensive analysis of molecular and radiomic profiles and uncovered the innate associations between radiological images and molecular features in gliomas. MoRad could particularly profile the genomic methylation level with high accuracy, indicating the vital role of genomic methylation in shaping tumor features that decide the radiological images. We demonstrated how radiomic analysis could be applied as a valuable tool to profile the molecular features and put them into the current perspectives of cancer omics and clinical information.

Availability of data and materials

The “MoRad” package with the codes for the algorithms of each radiomic model for molecular profiling and RO and RSEA analysis is available at https://github.com/hyu020/MoRad. The sample codes for molecular characterization using MoRad models could be found in the supplementary materials.

Abbreviations

AUC:

Area under curves

CaPTK:

Cancer Imaging Phenomics Toolkit

CNV:

Copy number variation

ED:

Peritumoral edema

EGFR:

Epidermal growth factor receptor

ET:

Enhancing region of the tumor core

GBM:

Glioblastoma

G-CIMP:

Glioma-CpG island methylator phenotype

GLCM:

Gray-level co-occurrence matrix

GLRLM:

Gray-level run-length matrix

GLSZM:

Gray-level size zone matrix

IDH:

Isocitrate dehydrogenase

IDH-MUT:

IDH-mutation

IDH-WT:

IDH-wild type

ITH:

Intra-tumor heterogeneity

LASSO:

Least absolute shrinkage and selection operator

LGG:

Low-grade glioma

LINE-1:

Long interspersed nuclear element-1

MGMT:

O-6-methylguanine-DNA methyltransferase

mMRI:

Multimodal MRI

MRI:

Magnetic resonance imaging

MRS9:

9 Molecular feature-specific radiomic sets

NET:

Non-enhancing region of the tumor core

NES:

Normalized enrichment scores

NGTDM:

Neighborhood gray-tone difference matrix

PCA:

Principal component analysis

RO:

Radiomic ontology

ROI:

Regions of interest

RPPA:

Reverse-phase protein array

RSEA:

Radiomic set enrichment analysis

T1WI:

T1-weighted

T1WI + Gd:

Gadolinium-enhanced T1-weighted

T2-FLAIR:

T2-fluidattenuated inversion recovery

T2WI:

T2-weighted

TERT:

Telomerase reverse transcriptase

TGM:

Tumor growth model

TMB:

Tumor mutation burden

References

  1. Dercle L, Zhao B, Gönen M, Moskowitz CS, Firas A, Beylergil V, et al. Early readout on overall survival of patients with melanoma treated with immunotherapy using a novel imaging analysis. JAMA Oncol. 2022;8(3):385–92.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Zhuang Z, Liu Z, Li J, Wang X, Xie P, Xiong F, et al. Radiomic signature of the FOWARC trial predicts pathological response to neoadjuvant treatment in rectal cancer. J Transl Med. 2021;19(1):256.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  3. Patel RV, Yao S, Huang RY, Bi WL. Application of radiomics to meningiomas: a systematic review. Neuro Oncol. 2023;25(6):1166–76.

    Article  PubMed  PubMed Central  Google Scholar 

  4. Schniering J, Maciukiewicz M, Gabrys HS, Brunner M, Blüthgen C, Meier C, et al. Computed tomography-based radiomics decodes prognostic and molecular differences in interstitial lung disease related to systemic sclerosis. Eur Respir J. 2022;59(5):2004503.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  5. Fan M, Cui Y, You C, Liu L, Gu Y, Peng W, et al. Radiogenomic signatures of oncotype DX recurrence score enable prediction of survival in estrogen receptor-positive breast cancer: a multicohort study. Radiology. 2022;302(3):516–24.

    Article  PubMed  Google Scholar 

  6. Liu Q, Zhang J, Guo C, Wang M, Wang C, Yan Y, et al. Proteogenomic characterization of small cell lung cancer identifies biological insights and subtype-specific therapeutic strategies. Cell. 2024;187(1):184-203.e28.

    Article  PubMed  CAS  Google Scholar 

  7. Chan TA, Yarchoan M, Jaffee E, Swanton C, Quezada SA, Stenzinger A, et al. Development of tumor mutation burden as an immunotherapy biomarker: utility for the oncology clinic. Ann Oncol. 2019;30(1):44–56.

    Article  PubMed  CAS  Google Scholar 

  8. Perez-Johnston R, Araujo-Filho JA, Connolly JG, Caso R, Whiting K, Tan KS, et al. CT-based radiogenomic analysis of clinical stage I lung adenocarcinoma with histopathologic features and oncologic outcomes. Radiology. 2022;303(3):664–72.

    Article  PubMed  Google Scholar 

  9. Beig N, Singh S, Bera K, Prasanna P, Singh G, Chen J, et al. Sexually dimorphic radiogenomic models identify distinct imaging and biological pathways that are prognostic of overall survival in glioblastoma. Neuro Oncol. 2021;23(2):251–63.

    Article  PubMed  Google Scholar 

  10. Su GH, Xiao Y, You C, Zheng RC, Zhao S, Sun SY, et al. Radiogenomic-based multiomic analysis reveals imaging intratumor heterogeneity phenotypes and therapeutic targets. Sci Adv. 2023;9(40):eadf0837.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  11. Jiang L, You C, Xiao Y, Wang H, Su GH, Xia BQ, et al. Radiogenomic analysis reveals tumor heterogeneity of triple-negative breast cancer. Cell Rep Med. 2022;3(7):100694.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  12. Skingen VE, Hompland T, Fjeldbo CS, Salberg UB, Helgeland H, Ragnum HB, et al. Prostate cancer radiogenomics reveals proliferative gene expression programs associated with distinct MRI-based hypoxia levels. Radiother Oncol. 2023;188:109875.

    Article  PubMed  CAS  Google Scholar 

  13. Kim M, Jung SY, Park JE, Jo Y, Park SY, Nam SJ, et al. Diffusion- and perfusion-weighted MRI radiomics model may predict isocitrate dehydrogenase (IDH) mutation and tumor aggressiveness in diffuse lower grade glioma. Eur Radiol. 2020;30(4):2142–51.

    Article  PubMed  Google Scholar 

  14. Rossi G, Barabino E, Fedeli A, Ficarra G, Coco S, Russo A, et al. Radiomic detection of EGFR mutations in NSCLC. Cancer Res. 2021;81(3):724–31.

    Article  PubMed  CAS  Google Scholar 

  15. Ivanidze J, Lum M, Pisapia D, Magge R, Ramakrishna R, Kovanlikaya I, et al. MRI features associated with TERT promoter mutation status in glioblastoma. J Neuroimaging. 2019;29(3):357–63.

    Article  PubMed  Google Scholar 

  16. Park JE, Kim HS, Park SY, Nam SJ, Chun SM, Jo Y, et al. Prediction of core signaling pathway by using diffusion- and perfusion-based MRI radiomics and next-generation sequencing in isocitrate dehydrogenase wild-type glioblastoma. Radiology. 2020;294(2):388–97.

    Article  PubMed  Google Scholar 

  17. Sun Q, Chen Y, Liang C, Zhao Y, Lv X, Zou Y, et al. Biologic pathways underlying prognostic radiomics phenotypes from paired MRI and RNA sequencing in glioblastoma. Radiology. 2021;301(3):654–63.

    Article  PubMed  Google Scholar 

  18. Song XL, Luo HJ, Ren JL, Yin P, Liu Y, Niu J, et al. Multisequence magnetic resonance imaging-based radiomics models for the prediction of microsatellite instability in endometrial cancer. Radiol Med. 2023;128(2):242–51.

    Article  PubMed  Google Scholar 

  19. Wang Q, Xu J, Wang A, Chen Y, Wang T, Chen D, et al. Systematic review of machine learning-based radiomics approach for predicting microsatellite instability status in colorectal cancer. Radiol Med. 2023;128(2):136–48.

    Article  PubMed  PubMed Central  Google Scholar 

  20. Louis DN, Perry A, Wesseling P, Brat DJ, Cree IA, Figarella-Branger D, et al. The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro Oncol. 2021;23(8):1231–51.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  21. Malta TM, Sabedot TS, Morosini NS, Datta I, Garofano L, Vallentgoed W, et al. The Epigenetic Evolution of Glioma Is Determined by the IDH1 Mutation Status and Treatment Regimen. Cancer Res. 2024;84(5):741–56.

    Article  PubMed  CAS  Google Scholar 

  22. Capper D, Jones DTW, Sill M, Hovestadt V, Schrimpf D, Sturm D, et al. DNA methylation-based classification of central nervous system tumours. Nature. 2018;555(7697):469–74.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  23. Malta TM, de Souza CF, Sabedot TS, Silva TC, Mosella MS, Kalkanis SN, et al. Glioma CpG island methylator phenotype (G-CIMP): biological and clinical implications. Neuro Oncol. 2018;20(5):608–20.

    Article  PubMed  CAS  Google Scholar 

  24. Kinslow CJ, Mercurio A, Kumar P, Rae AI, Siegelin MD, Grinband J, et al. Association of MGMT promoter methylation with survival in low-grade and anaplastic gliomas after alkylating chemotherapy. JAMA Oncol. 2023;9(7):919–27.

    Article  PubMed  PubMed Central  Google Scholar 

  25. Shields JA, Meier SR, Bandi M, Mulkearns-Hubert EE, Hajdari N, Ferdinez MD, et al. vrk1 is a synthetic-lethal target in VRK2-deficient glioblastoma. Cancer Res. 2022;82(21):4044–57.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  26. Cadieux B, Ching TT, VandenBerg SR, Costello JF. Genome-wide hypomethylation in human glioblastomas associated with specific copy number alteration, methylenetetrahydrofolate reductase allele status, and increased proliferation. Can Res. 2006;66(17):8469–76.

    Article  CAS  Google Scholar 

  27. Briand J, Nadaradjane A, Bougras-Cartron G, Olivier C, Vallette FM, Cartron PF. Diuron exposure and Akt overexpression promote glioma formation through DNA hypomethylation. Clin Epigenetics. 2019;11(1):159.

    Article  PubMed  PubMed Central  Google Scholar 

  28. Jung H, Kim HS, Kim JY, Sun JM, Ahn JS, Ahn MJ, et al. DNA methylation loss promotes immune evasion of tumours with high mutation and copy number load. Nat Commun. 2019;10(1):4278.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Klughammer J, Kiesel B, Roetzer T, Fortelny N, Nemc A, Nenning KH, et al. The DNA methylation landscape of glioblastoma disease progression shows extensive heterogeneity in time and space. Nat Med. 2018;24(10):1611–24.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  30. Babaeenezhad E, Moradi Sarabi M, Rajabibazl M, Oraee-Yazdani S, Karima S. Global and regional DNA methylation silencing of PPARγ associated with glioblastoma multiforme pathogenesis. Mol Biol Rep. 2023;50(1):589–97.

    Article  PubMed  CAS  Google Scholar 

  31. Weller M, Felsberg J, Hentschel B, Gramatzki D, Kubon N, Wolter M, et al. Improved prognostic stratification of patients with isocitrate dehydrogenase-mutant astrocytoma. Acta Neuropathol. 2024;147(1):11.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  32. Clark K, Vendt B, Smith K, Freymann J, Kirby J, Koppel P, et al. The Cancer Imaging Archive (TCIA): maintaining and operating a public information repository. J Digit Imaging. 2013;26(6):1045–57.

    Article  PubMed  PubMed Central  Google Scholar 

  33. Brat DJ, Verhaak RG, Aldape KD, Yung WK, Salama SR, Cooper LA, et al. Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas. N Engl J Med. 2015;372(26):2481–98.

    Article  PubMed  CAS  Google Scholar 

  34. McLendon R, Friedman A, Bigner D, Van Meir EG, Brat DJ, M. Mastrogianakis G, et al. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature. 2008;455(7216):1061–8.

    Article  Google Scholar 

  35. Forschner A, Battke F, Hadaschik D, Schulze M, Weißgraeber S, Han CT, et al. Tumor mutation burden and circulating tumor DNA in combined CTLA-4 and PD-1 antibody therapy in metastatic melanoma - results of a prospective biomarker study. J Immunother Cancer. 2019;7(1):180.

    Article  PubMed  PubMed Central  Google Scholar 

  36. Carter SL, Cibulskis K, Helman E, McKenna A, Shen H, Zack T, et al. Absolute quantification of somatic DNA alterations in human cancer. Nat Biotechnol. 2012;30(5):413–21.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  37. Ceccarelli M, Barthel FP, Malta TM, Sabedot TS, Salama SR, Murray BA, et al. Molecular profiling reveals biologically discrete subsets and pathways of progression in diffuse glioma. Cell. 2016;164(3):550–63.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  38. Chen F, Chandrashekar DS, Varambally S, Creighton CJ. Pan-cancer molecular subtypes revealed by mass-spectrometry-based proteomic characterization of more than 500 human cancers. Nat Commun. 2019;10(1):5679.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  39. Verhaak RG, Hoadley KA, Purdom E, Wang V, Qi Y, Wilkerson MD, et al. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell. 2010;17(1):98–110.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  40. Wang Q, Hu B, Hu X, Kim H, Squatrito M, Scarpace L, et al. Tumor evolution of glioma-intrinsic gene expression subtypes associates with immunological changes in the microenvironment. Cancer Cell. 2018;33(1):152.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  41. Wick W, Weller M, van den Bent M, Sanson M, Weiler M, von Deimling A, et al. MGMT testing–the challenges for biomarker-based glioma treatment. Nat Rev Neurol. 2014;10(7):372–85

    Article  PubMed  CAS  Google Scholar 

  42. Eckel-Passow JE, Lachance DH, Molinaro AM, Walsh KM, Decker PA, Sicotte H, et al. Glioma groups based on 1p/19q, IDH, and TERT promoter mutations in tumors. N Engl J Med. 2015;372(26):2499–508.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  43. Newman AM, Steen CB, Liu CL, Gentles AJ, Chaudhuri AA, Scherer F, et al. Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat Biotechnol. 2019;37(7):773–82.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  44. Cao X, Li W, Wang T, Ran D, Davalos V, Planas-Serra L, et al. Accelerated biological aging in COVID-19 patients. Nat Commun. 2022;13(1):2135.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  45. Yu H, Wang X, Bai L, Tang G, Carter KT, Cui J, et al. DNA methylation profile in CpG-depleted regions uncovers a high-risk subtype of early-stage colorectal cancer. J Natl Cancer Inst. 2023;115(1):52–61.

    Article  PubMed  CAS  Google Scholar 

  46. Zheng Y, Joyce BT, Liu L, Zhang Z, Kibbe WA, Zhang W, et al. Prediction of genome-wide DNA methylation in repetitive elements. Nucleic Acids Res. 2017;45(15):8697–711.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  47. Bakas S, Akbari H, Sotiras A, Bilello M, Rozycki M, Kirby JS, et al. Advancing the cancer genome atlas glioma MRI collections with expert segmentation labels and radiomic features. Sci Data. 2017;4:170117.

    Article  PubMed  PubMed Central  Google Scholar 

  48. Rohlfing T, Zahr NM, Sullivan EV, Pfefferbaum A. The SRI24 multichannel atlas of normal adult human brain structure. Hum Brain Mapp. 2010;31(5):798–819.

    Article  PubMed  Google Scholar 

  49. Smith SM. Fast robust automated brain extraction. Hum Brain Mapp. 2002;17(3):143–55.

    Article  PubMed  PubMed Central  Google Scholar 

  50. Doshi J, Erus G, Ou Y, Resnick SM, Gur RC, Gur RE, et al. MUSE: MUlti-atlas region Segmentation utilizing Ensembles of registration algorithms and parameters, and locally optimal atlas selection. Neuroimage. 2016;127:186–95.

    Article  PubMed  Google Scholar 

  51. Smith SM, Brady JM. SUSAN—a new approach to low level image processing. Int J Comput Vision. 1997;23(1):45–78.

    Article  Google Scholar 

  52. Nyúl LG, Udupa JK, Zhang X. New variants of a method of MRI scale standardization. IEEE Trans Med Imaging. 2000;19(2):143–50.

    Article  PubMed  Google Scholar 

  53. Davatzikos C, Rathore S, Bakas S, Pati S, Bergman M, Kalarot R, et al. Cancer imaging phenomics toolkit: quantitative imaging analytics for precision diagnostics and predictive modeling of clinical outcome. J Med Imaging (Bellingham). 2018;5(1):011018.

    PubMed  Google Scholar 

  54. Pati S, Singh A, Rathore S, Gastounioti A, Bergman M, Ngo P, et al. The Cancer Imaging Phenomics Toolkit (CaPTk): technical overview. Brainlesion. 2020;11993:380–94.

    PubMed  PubMed Central  Google Scholar 

  55. Zwanenburg A, Vallières M, Abdalah MA, Aerts H, Andrearczyk V, Apte A, et al. The Image Biomarker Standardization Initiative: standardized quantitative radiomics for high-throughput image-based phenotyping. Radiology. 2020;295(2):328–38.

    Article  PubMed  Google Scholar 

  56. Zou Q, Wang X, Ren D, Hu B, Tang G, Zhang Y, et al. DNA methylation-based signature of CD8+ tumor-infiltrating lymphocytes enables evaluation of immune response and prognosis in colorectal cancer. J Immunother Cancer. 2021;9(9):e002671.

    Article  PubMed  PubMed Central  Google Scholar 

  57. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA. 2005;102(43):15545–50.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  58. Sun R, Limkin EJ, Vakalopoulou M, Dercle L, Champiat S, Han SR, et al. A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. Lancet Oncol. 2018;19(9):1180–91.

    Article  PubMed  CAS  Google Scholar 

  59. He B, Dong D, She Y, Zhou C, Fang M, Zhu Y, et al. Predicting response to immunotherapy in advanced non-small-cell lung cancer using tumor mutational burden radiomic biomarker. J Immunother Cancer. 2020;8(2):e000550.

    Article  PubMed  PubMed Central  Google Scholar 

  60. Lue KH, Wu YF, Liu SH, Hsieh TC, Chuang KS, Lin HH, et al. Intratumor heterogeneity assessed by 18F-FDG PET/CT predicts treatment response and survival outcomes in patients with hodgkin lymphoma. Acad Radiol. 2020;27(8):e183–92.

    Article  PubMed  Google Scholar 

  61. Heyn H, Vidal E, Ferreira HJ, Vizoso M, Sayols S, Gomez A, et al. Epigenomic analysis detects aberrant super-enhancer DNA methylation in human cancer. Genome Biol. 2016;17:11.

    Article  PubMed  PubMed Central  Google Scholar 

  62. Van Tongelen A, Loriot A, De Smet C. Oncogenic roles of DNA hypomethylation through the activation of cancer-germline genes. Cancer Lett. 2017;396:130–7.

    Article  PubMed  Google Scholar 

  63. Chaligne R, Gaiti F, Silverbush D, Schiffman JS, Weisman HR, Kluegel L, et al. Epigenetic encoding, heritability and plasticity of glioma transcriptional cell states. Nat Genet. 2021;53(10):1469–79.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  64. Rahme GJ, Javed NM, Puorro KL, Xin S, Hovestadt V, Johnstone SE, et al. Modeling epigenetic lesions that cause gliomas. Cell. 2023;186(17):3674-85.e14.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  65. Dodgshun AJ, Fukuoka K, Edwards M, Bianchi VJ, Das A, Sexton-Oates A, et al. Germline-driven replication repair-deficient high-grade gliomas exhibit unique hypomethylation patterns. Acta Neuropathol. 2020;140(5):765–76.

    Article  PubMed  CAS  Google Scholar 

  66. Johnson KC, Anderson KJ, Courtois ET, Gujar AD, Barthel FP, Varn FS, et al. Single-cell multimodal glioma analyses identify epigenetic regulators of cellular plasticity and environmental stress response. Nat Genet. 2021;53(10):1456–68.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  67. Shi Y, Ping YF, Zhou W, He ZC, Chen C, Bian BS, et al. Tumour-associated macrophages secrete pleiotrophin to promote PTPRZ1 signalling in glioblastoma stem cells for tumour growth. Nat Commun. 2017;8:15080.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  68. Andersen BM, Faust Akl C, Wheeler MA, Chiocca EA, Reardon DA, Quintana FJ. Glial and myeloid heterogeneity in the brain tumour microenvironment. Nat Rev Cancer. 2021;21(12):786–802.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  69. Hu LS, Hawkins-Daarud A, Wang L, Li J, Swanson KR. Imaging of intratumoral heterogeneity in high-grade glioma. Cancer Lett. 2020;477:97–106.

    Article  PubMed  PubMed Central  CAS  Google Scholar 

Download references

Acknowledgements

Not applicable.

Funding

This work was supported by the Project 5010 of Clinical Medical Research of Sun Yat-sen University-5010 Cultivation Foundation (No. 2018026, YL), the Sixth Affiliated Hospital of Sun Yat-sen University Clinical Research- “1010” Program (MH; YL), the National Natural Science Foundation of China (No. 81972245, YL; No. 82173067, YL; No. 81902877, HY; No. 82272965, HY), the Natural Science Foundation of Guangdong Province (No. 2022A1515012656, HY; No.2021A1515010134, MH), the Science and Technology Program of Guangzhou (202201011004, HY), the Scientific Research Project of the Sixth Affiliated Hospital Of Sun Yat-Sen University (2022JBGS07), the Talent Project of the Sixth Affiliated Hospital of Sun Yat-sen University (No. P20150227202010251, YL), the Excellent Talent Training Project of the Sixth Affiliated Hospital of Sun Yat-sen University (No. R2021217202512965, YL), the Fundamental Research Funds for the Central Universities, Sun Yat-sen University (No. 23ykbj007, HY), the Program of Introducing Talents of Discipline to Universities (YL), the Program of Guangdong Provincial Clinical Research Center for Digestive Diseases (2020B1111170004), and National Key Clinical Discipline (2012).

Author information

Author notes

  1. Zhuokai Zhuang, Jinxin Lin, and Zixiao Wan contribute equally to this work.

Authors and Affiliations

  1. Department of General Surgery (Colorectal Surgery), The Sixth Affiliated Hospital, Sun Yat-Sen University, Guangzhou, Guangdong, China

    Zhuokai Zhuang, Jinxin Lin, Jingrong Weng, Yumo Xie, Meijin Huang & Yanxin Luo

  2. Guangdong Institute of Gastroenterology, Guangzhou, Guangdong, China

    Zhuokai Zhuang, Jinxin Lin, Zixiao Wan, Jingrong Weng, Ze Yuan, Yumo Xie, Xiaolin Wang, Meijin Huang, Yanxin Luo & Huichuan Yu

  3. Guangdong Provincial Key Laboratory of Colorectal and Pelvic Floor Diseases, The Sixth Affiliated Hospital, Sun Yat-Sen University, Guangzhou, Guangdong, China

    Zhuokai Zhuang, Jinxin Lin, Zixiao Wan, Jingrong Weng, Ze Yuan, Yumo Xie, Xiaolin Wang, Meijin Huang, Yanxin Luo & Huichuan Yu

  4. Biomedical Innovation Center, The Sixth Affiliated Hospital, Sun Yat-Sen University, Guangzhou, Guangdong, 510655, China

    Zhuokai Zhuang, Jinxin Lin, Zixiao Wan, Jingrong Weng, Ze Yuan, Yumo Xie, Xiaolin Wang, Meijin Huang, Yanxin Luo & Huichuan Yu

  5. Ministry of Education, Key Laboratory of Human Microbiome and Chronic Diseases (Sun Yat-Sen University), Guangzhou, Guangdong, China

    Zhuokai Zhuang, Jinxin Lin, Zixiao Wan, Ze Yuan, Xiaolin Wang, Meijin Huang, Yanxin Luo & Huichuan Yu

  6. Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Cancer Epidemiology, Peking University Cancer Institute, Beijing, 100142, China

    Zongchao Liu

  7. Department of Radiology, The Sixth Affiliated Hospital, Sun Yat-Sen University, Guangzhou, Guangdong, 510655, China

    Peiyi Xie

  8. Image and Minimally Invasive Intervention Center, Sun Yat-Sen University Cancer Center, State Key Laboratory of Oncology in South China, and Collaborative Innovation Center for Cancer Medicine, Guangzhou, China

    Siyue Mao

  9. Department of Neurosurgery and Pituitary Tumor Center, The First Affiliated Hospital, Sun Yat-Sen University, Guangzhou, China

    Zongming Wang

Authors
  1. Zhuokai Zhuang
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Jinxin Lin
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Zixiao Wan
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Jingrong Weng
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Ze Yuan
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Yumo Xie
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Zongchao Liu
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Peiyi Xie
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Siyue Mao
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Zongming Wang
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Xiaolin Wang
    View author publications

    You can also search for this author in PubMed Google Scholar

  12. Meijin Huang
    View author publications

    You can also search for this author in PubMed Google Scholar

  13. Yanxin Luo
    View author publications

    You can also search for this author in PubMed Google Scholar

  14. Huichuan Yu
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

HY, ZZ, ZW and JL conceived of and established the machine learning framework for MoRad; HY, ZZ, SM and PX conceived of and established the pipeline for Ontology analysis; HY and YL provided the framework for array-based global methylation analysis; ZZ and ZL collected the data of study cohorts and performed the data analyses under the supervision of HY; XW, YX, ZY, JW, JL, ZW, PX and SM assisted with the data sorting and analyses; ZZ, ZW, PX, SM, ZW, MH, YL and HY jointly interpreted the results; MH and YL supervised the whole project; ZZ, JL, XW, SM, ZW and HY wrote the original manuscript; all co-authors critically revised the manuscript and approved the final manuscript.

Authors’ Twitter handles

Twitter handles: @huichuan_yu (Huichuan Yu).

Corresponding author

Correspondence to Huichuan Yu.

Ethics declarations

Ethics approval and consent to participate

This is an observational study. The Institutional Review Board of the Sixth Affiliated Hospital of Sun Yat-sen University (Guangzhou, China) has confirmed that no ethical approval is required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

12916_2024_3573_MOESM1_ESM.pdf

Additional file 1: Fig. S1. The workflow of this study. Fig. S2. Comprehensive analysis of MoRad performance. Fig. S3. Upset plot for the overview of radiomic features in MRS9. Fig. S4. Association of MoRad-profiled TMB and ITH with molecular profiles and clinical information.

12916_2024_3573_MOESM2_ESM.zip

Additional file 2: Table S1. Baseline and molecular characteristics of study patients. Table S2. Description of molecular data used for analysis. Table S3. Summary of extracted radiomics features. Table S4. Collected radiomic sets: terms for RO and RSEA. Table S5. Molecular feature-specific radiomic sets. Table S6. RO for the MRS9. Table S7. RSEA for the MRS9. Table S8. RO for the union and unique features of MRS9.

12916_2024_3573_MOESM3_ESM.pdf

Additional file 3: Supplementary Method 1. Data Collection and Image Preprocessing. Supplementary Method 2. Feature Selection and Model Construction. Supplementary Method 3 RO and RSEA.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Zhuang, Z., Lin, J., Wan, Z. et al. Radiogenomic profiling of global DNA methylation associated with molecular phenotypes and immune features in glioma. BMC Med 22, 352 (2024). https://doi.org/10.1186/s12916-024-03573-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12916-024-03573-y

Keywords

BMC Medicine

ISSN: 1741-7015

Contact us