Multiparametric Magnetic Resonance Imaging Correlates of Isocitrate Dehydrogenase Mutation in WHO high-Grade Astrocytomas

Purpose and background: Isocitrate dehydrogenase (IDH) mutation and O-6 methyl guanine methyl transferase (MGMT) methylation are surrogate biomarkers of improved survival in gliomas. This study aims at studying the ability of semantic magnetic resonance imaging (MRI) features to predict the IDH mutation status confirmed by the gold standard molecular tests. Methods: The MRI of 148 patients were reviewed for various imaging parameters based on the Visually AcceSAble Rembrandt Images (VASARI) study. Their IDH status was determined using immunohistochemistry (IHC). Fisher’s exact or chi-square tests for univariate and logistic regression for multivariate analysis were used. Results: Parameters such as mild and patchy enhancement, minimal edema, necrosis < 25%, presence of cysts, and less rCBV (relative cerebral blood volume) correlated with IDH mutation. The median age of IDH-mutant and IDH-wild patients were 34 years (IQR: 29–43) and 52 years (IQR: 45–59), respectively. Mild to moderate enhancement was observed in 15/19 IDH-mutant patients (79%), while 99/129 IDH-wildtype (77%) had severe enhancement (p-value <0.001). The volume of edema with respect to tumor volume distinguished IDH-mutants from wild phenotypes (peritumoral edema volume < tumor volume was associated with higher IDH-mutant phenotypes; p-value < 0.025). IDH-mutant patients had a median rCBV value of 1.8 (IQR: 1.4–2.0), while for IDH-wild phenotypes, it was 2.6 (IQR: 1.9–3.5) {p-value = 0.001}. On multivariate analysis, a cut-off of 25% necrosis was able to differentiate IDH-mutant from IDH-wildtype (p-value < 0.001), and a cut-off rCBV of 2.0 could differentiate IDH-mutant from IDH-wild phenotypes (p-value < 0.007). Conclusion: Semantic imaging features could reliably predict the IDH mutation status in high-grade gliomas. Presurgical prediction of IDH mutation status could help the treating oncologist to tailor the adjuvant therapy or use novel IDH inhibitors.


Introduction
The management of high-grade gliomas has undergone a paradigm shift with the addition of molecular parameters to morphological features in the World Health Organization (WHO) 2016 classification [1]. The modern-day treatment practices of gliomas are now based on molecular biomarkers to have a biologically homogenous treatment group to study newer interventions in clinical trials. Although previously categorized under grade 4 gliomas as an isocitrate dehydrogenase (IDH) mutant and wildtype, glioblastomas are now considered biologically and molecularly separate entities: glioblastoma IDH-wildtype and IDH-mutant grade 4 astrocytoma [2]. Molecular biomarkers such as IDH and MGMT (O-6 methylguanine methyltransferase) have allowed oncologists to personalize treatment and prognosticate the disease better than it used to be a decade earlier [3].
Despite multimodal therapy that includes gross total resection (GTR), radiotherapy, and chemotherapeutics such as temozolomide (TMZ), the prognosis for grade 4 gliomas is dismal with a median survival of only 14 months, and <5% of patients survive beyond five years [1]. This calls for a more concerted approach to understanding the disease biology and the factors associated with its outcome. Literature studies have shown that age, performance status, and treatment-related factors can prognosticate these tumors. However, these factors lack the accuracy to predict response to therapy [4]. With a better understanding of biology and the advent of newer molecular techniques, researchers have been able to show unique biomarkers that could predict treatment response and prognosticate these tumors with a high degree of accuracy, paving the way for a more personalized treatment approach. The two molecular biomarkers of significant interest that have translated into clinical practice are IDH and MGMT, which are responsible for the epigenetic alterations in grade 4 gliomas. Evaluating these biomarkers have become a norm in tailoring therapy and disease prognostication [5].
IDH plays an essential role in the Kreb's cycle by converting isocitrate to alphaketoglutarate. Mutated IDH converts alpha-ketoglutarate into 2-hydroxyglutarate, an oncometabolite responsible for the epigenetic changes in gliomas and associated with improved prognosis. Therefore, IDH-mutants have a better prognosis compared to wild phenotypes. Thus, IDH is a diagnostic and prognostic biomarker for gliomas [6]. Mutated IDH also drives increased methylation in gliomas [7]. MGMT is a DNA (deoxyribonucleic acid) repair enzyme, detoxifying temozolomide (TMZ)-induced DNA damage. Clinically, high MGMT protein expression has been associated with therapeutic resistance to DNAalkylating agents apart from having prognostic significance [8].
Typically, these markers can only be assessed from the tumor tissues of the surgical specimen. However, it is well-known that grade 4 gliomas are genetically heterogenous, and targeting may lead to erroneous interpretations [9]. Therefore, using high-quality noninvasive magnetic resonance imaging (MRI) parameters to distinguish grade 4 gliomas with different genetic compositions (radiogenomics) has gained importance for prognostication and therapy selection [10]. Imaging studies have used VASARI (Visually AcceSAble Rembrandt Images) semantic features for the molecular subgrouping of grade 4 gliomas [11].
Given the inherent tumor heterogeneity on histopathology and the universal availability of MRI, we envisaged multiparametric semantic MRI features in a large cohort of grade 4 glioma patients to classify them based on the IDH status as confirmed by gold standard immunohistochemistry (IHC) with or without gene sequencing.

Patient Selection
This was a retrospective chart review of 148 grade 4 glioma patients in whom we had the baseline MRI and IDH mutational status. The study was carried out after approval from the Institutional Ethics Committee (IEC). One hundred and forty-eight patients were analyzed after assessing the eligibility criteria: availability of baseline pre-treatment MRI and IDH mutational status in our hospital's electronic medical records (EMR).

MR Imaging Parameters
We adopted the standard VASARI imaging features with an expanded set of parameters to study their association with IDH status [12]. The VASARI features and extended imaging parameters are provided in Supplementary Table S1.

Mutational Analysis
Patients' surgical pathology reports from electronic medical records (EMR) were used for IDH mutational status. IDH status was available for all 148 grade 4 glioma patients. The standard institutional practice is to perform immunohistochemistry for IDH status, which was considered the gold standard and confirmed by Sangers sequencing for IDH1R132 and IDH2R172 loci if required.

Statistical Analysis
Statistical analysis was performed using the Statistical Package for the Social Sciences software (SPSS, ver. 21). Descriptive statistics such as frequencies and percentages were used for the categorical variables, and mean and standard deviation (SD) for the continuous variables. Univariate analyses were done using the Fisher's exact test or the chi-square test. Multivariate stepwise logistic regression analyses were conducted to evaluate the risk factors for IDH mutation status. Variables with a p-value of < 0.05 on the univariate analysis were used as inputs for the multivariable analysis. A p-value < 0.05 was considered statistically significant.

Patient Demographic and Tumor Characteristics
From the neuro-oncology database, 148 patients were taken up for the study. Most grade 4 glioma patients were IDH-wildtype (n = 129), while IDH-mutant phenotype was observed in only 19 patients, and their representative images are provided in Figures 1 and 2. The male-to-female ratio was 2:1 (males: 99 and females: 49). The median age of patients who harbored IDH mutations was significantly lower than the patients who were IDH-wildtype. Specifically, for the IDH-mutant group, median age was 34 years (interquartile range, IQR: 29.5-43), while for the IDH-wildtype group, median age was 52 years (IQR: 45-59). We observed that 13/19 (68.4%) IDH-mutant patients had left hemispheric predominance than 55/129 IDH-wildtype patients (42.6%). The frontal lobe was predominantly involved in IDH-mutants, while multilobar involvement was more common in IDH-wildtypes. The patient and tumor demographic profiles are summarized in Table 1.     representing an IDH-wildtype glioblastoma. Images a-e reveal a heterogeneous peripheral rimenhancing right frontoparietal mass with significant internal necrosis and hemorrhage. Image f is a DTI image that shows subcortical and adjacent parenchymal infiltration. Images g and h reveal significant hyperperfusion.

MRI Parameters of IDH-Wildtype vs. Mutant Phenotype Tumors
Univariate analysis of the semantic MRI features of IDH-mutant vis-à-vis IDH-wildtype grade 4 gliomas have been summarized in Table 2.
On multivariate analysis, using a stepwise logistic regression, the MRI parameters of necrosis of > 25% on CE-MRI (p-value < 0.001) and rCBV cut-off of > 2.0 (p-value 0.007), independently correlated for IDH-wildtype phenotype while the other semantic features that were significant on the univariate analysis lost their significance on the multivariate analysis (Table 3). Figure 12 shows the graphical abstract for this study.         Other parameters such as the proportion of the enhancing tumor, leptomeningeal spread, satellite lesions, calvarial remodeling, and ependymal invasion did not correlate with the IDH status. These are summarized in Supplementary Table S2.
On multivariate analysis, using a stepwise logistic regression, the MRI parameters of necrosis of > 25% on CE-MRI (p-value < 0.001) and rCBV cut-off of > 2.0 (p-value 0.007), independently correlated for IDH-wildtype phenotype while the other semantic features that were significant on the univariate analysis lost their significance on the multivariate analysis (Table 3). Figure 12 shows the graphical abstract for this study.

Discussion
This study attempts to identify the semantic imaging features on multiparametric MRI that best define IDH mutation status against the gold standard immunohistochemistry to predict these mutations. To the best of our knowledge, this study includes one of the largest cohorts of grade 4 gliomas with a description of its imaging morphology at diagnosis and radiopathological correlation.
In our cohort, we found that rCBV and the proportion of necrosis accurately predicts

Discussion
This study attempts to identify the semantic imaging features on multiparametric MRI that best define IDH mutation status against the gold standard immunohistochemistry to predict these mutations. To the best of our knowledge, this study includes one of the largest cohorts of grade 4 gliomas with a description of its imaging morphology at diagnosis and radiopathological correlation.
In our cohort, we found that rCBV and the proportion of necrosis accurately predicts the IDH mutation status on multivariate analysis. The median rCBV for IDH-mutants was 1.8 (IQR: 1.4-2.0), which was significantly lower than the IDH-wildtypes with a median of 2.6 (IQR: 1.9-3.5) on univariate analysis. On multivariate analysis, the rCBV cut-off of <2.0 was able to differentiate IDH-mutants from IDH-wildtype tumors. Our results corroborate with various other studies [13][14][15][16]. Additionally, studies have shown a higher nCBV (normalized CBV) in IDH-wildtypes [17]. The hemodynamic tissue signature segmentation model explains the reduced rCBV in mutant phenotypes [18]. Literature studies have attributed the lower rCBV in IDH-mutants to reduced angiogenesis in the tumor tissue while a positive association of IDH-wildtype tumors have been attributed with increased necrosis [19]. In our study, the percentage of necrosis in IDH-mutant cohorts was <25% of the entire tumor volume, while in IDH-wild phenotypes necrosis was >25%. These findings corroborate the results of several studies [13,14,20]. Park et al. observed that a cut-off necrosis of <33% of the tumor was commonly associated with IDH-mutants [20]. The increased necrosis and non-enhancing tumors in wildtype phenotypes have been attributed to augmented intratumoral hypoxia that results from the activation of the coagulation pathway and intravascular thrombosis leading to excessive tumor necrosis [13,14].
We observed that a mild to moderate and patchy enhancement was commonly associated with the IDH-mutant phenotype tumors. Similarly, IDH-wildtype tumors had severe or intense rim enhancement, a notable feature documented in other literature studies [14][15][16]. Molecular-and genetic-level research provides few valid reasons for these significant associations. It has been observed that IDH-mutant tumors have reduced expression of vascular endothelial growth factor and thereby, reduced vascular permeability leading to reduced enhancement in mutant subtype tumors compared to wildtype tumors [15]. Minimal to mild contrast enhancement in IDH-mutants was attributed to a lack of microvascular proliferation resulting in reduced neoangiogenesis, vascular permeability, and contrast enhancement [13]. Secondly, IDH-wildtypes are known to have increased expression of hypoxia-inducible factor-1α (HIF-1α), which is associated with an increased expression of proangiogenic factors such as vascular endothelial growth factor-A (VEGF-A) and platelet-derived growth factor-A (PDGF-A), responsible for neoangiogenesis leading to increased contrast enhancement [14]. Carrillo et al. showed an increased association of rim enhancement in MGMT unmethylated patients [15].
The absence of peritumoral edema or edema volume less than tumor volume was commonly associated with IDH-mutants, while peritumoral edema volume equal to or more than tumor volume was associated with IDH-wildtypes. Similar results were documented in studies by Lasocki et al. [11] and Patel et al. [16]. Lasocki's group obtained a cut-off value of 33% to differentiate IDH-mutant from IDH-wildtype tumors. IDH-mutants were associated with the presence of cysts, which is in concordance with other studies [21].
In the present study, left lobe predominance was associated with IDH-mutant tumors, which concurs with the observations of a prior survey by Ellingson et al. [22]. Left hemispheric predominance is a common finding in MGMT-methylated tumors. In our cohort, all the IDH-mutant patients had MGMT methylation. Although literature studies have no proper explanation for this phenomenon, studies have demonstrated that MGMT-methylated tumors have a left hemispheric predominance [22] and most of the MGMT-methylated tumors are IDH-mutant subtypes. Further research needs to be conducted to understand the predisposition of IDH-mutated and MGMT-methylated tumors toward the left hemisphere. Subcortical involvement in our study was primarily seen in IDH-mutants (94.7%). A plausible explanation for this association could be the preponderance of deep white matter involvement by wildtype tumors. Dural enhancement was absent in all IDH-mutants and was present in 24% of IDH-wildtype tumors, probably due to increased vascularity and dural invasion leading to increased dural enhancement. These observations in our study are novel and yet to be documented in the literature.
The T2-FLAIR mismatch sign was predominant in IDH-mutants, characterized by homogeneous or heterogeneous signal intensity on T2WI and relative central hypointensity on FLAIR with a peripheral hyperintense rim [16]. However, in the present study, 15.8% of IDH-mutants and 3.9% of IDH-wildtypes had a peripheral hyperintense rim.
Being a retrospective study, we acknowledge that our results are limited by the inherent bias associated with all retrospective studies. Study designs with still larger sample sizes or prospective studies are needed in this domain to address the inequalities between IDH-mutant and wildtype cohorts. Although a single institutional study, the results are derived from a large cohort of patients who had uniform preoperative MRIs and underwent molecular studies with the internationally recommended gold standard test. We obtained a significant correlation between IDH mutational status and various imaging parameters that corroborate the literature studies. We documented a few new and unique features such as enhancement patterns-rim/nodular/patchy/solid, dural enhancement, and subcortical involvement, which have been highlighted in our study.

Conclusions
Semantic imaging features reliably predicted the IDH status in patients diagnosed with grade 4 gliomas. The correlations were more robust when an advanced technique such as perfusion was employed. Tumor necrosis of <25% and rCBV values of <2.0 stood out as independent imaging surrogates for IDH mutation. With the evolution in glioma therapeutics, the advent of newer strategies, and the ongoing trials for targeted therapies, we envisage a need for molecular predictions based on fast, non-invasive, and easy-to-adopt semantic radiological features shortly to select outpatients for targeted therapeutic interventions.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10.3 390/jpm13010072/s1, Table S1: VASARI imaging parameters with its modifications, and Table S2: MR imaging parameters which did not significantly correlate with IDH status. 2. Tumor laterality: The side at which the tumor is epicentered, namely -Right, Left and Central or bilateral 3. Enhancement pattern: It is calculated based on subjective qualitative assessment of the degree of contrast enhancement of the entire or part of the tumor on post contrast T1w sequences comparing to pre contrast T1w sequence.
It is further subdivided into 3. Homogenous/heterogenous: Whether the tumor is enhancing homogenously or heterogeneously in its entire portion.
4. Mild/Moderate and severe: Qualitative visual assessment of the degree of contrast enhancement of the entire tumor or part of the tumor.
5. Rim/Nodular/Solid/Patchy: Rim enhancement is provisionally defined as peripheral rim enhancement with central non-enhancing areas. Nodular enhancement is provisionally defined as enhancing nodule with remaining non-enhancing areas. Solid enhancement is provisionally defined as near homogenous or heterogenous enhancement of the entire tumor. Patchy enhancement is provisionally defined as non-uniform distribution of enhancing and non-enhancing areas within the tumor.
6. Proportion enhancing: It is a subjective qualitative assessment of the proportion of the enhancing tumor and is provisionally grouped into four categories namely -No enhancing part, <25% of tumor, 25-50% of tumor and >50% of tumor. 7. Eloquent brain involvement: Involvement of the eloquent cortex or involvement of the underlying subcortical white matter of the eloquent cortex.
Following eloquent cortex were included in our study, -Wernicke's, Broca's, Motor and Vision. 8. Necrosis: It is the portion of the tumor which is not enhancing on post contrast sequence and is T1 hypointense and T2 hyperintense with irregular borders. It is provisionally grouped into four subjective categories, namely -No necrosis, <25% of tumor, 25-50% of tumor and >50% of tumor. 9. Edema: It is defined as the T2 and FLAIR hyperintense areas/ region surrounding the tumor. It is provisionally grouped into four subjective categories, namely -No edema, less than tumor volume, equal to tumor volume and greater than tumor volume. 10. Hemorrhage: It is defined as the area of GRE/SWI blooming within the tumor or any area of T1 hyperintensity or T2 hypointensity. 11. Cysts: It is defined as relatively well-defined round or oval areas of T2 hyperintensity and T1 hypointensity, paralleling the CSF intensity in all sequences with a smooth, thin and regular enhancing or non-enhancing rim. Internal septations can be present or absent.
12. Multifocal/multicentric: -Focal: It is defined as single tumor localized to one region of the brain parenchyma.
-Multifocal: It is defined as at least one not contiguous lesion with the dominant tumor and is outside the region of edema surrounding the dominant tumor. It can be either enhancing or non-enhancing.
This can be defined as those resulting from dissemination or growth or spread via commissural or other pathways, or via CSF channels or local metastases.
-Multicentric: Two or more widely separated lesions in different lobes or hemispheres that are not attributed to one of the above-mentioned pathways of spread.
-Gliomatosis: It refers to generalized neoplastic transformation of the white matter of most of a hemisphere.
13. Tumor size: It is defined as the maximum longest diameter of the tumor on axial plane (anteroposterior and transverse dimensions) and the craniocaudal dimension measured on coronal or sagittal plane. However, the single largest dimension was only taken in to account. It was further subdivided into -Maximum dimension less than 2 cm, between 2-5 cm and more than 5 cm. 14. Satellite lesions: It is defined as an area of enhancement within the region of edema surrounding the index lesion and is not contiguous in any part with the major tumor mass.
15. Leptomeningeal spread: It is defined as the enhancement of the overlying pia/arachnoid layer either in continuity or not in continuity with enhancing or nonenhancing tumor. 16. Midline shift: It is defined as the mass effect with resulting midline shift of the brain to contralateral side. It was further subdivided into -No midline shift, mild midline shift of less than 5 mm, moderate midline shift between 5-10 mm and severe midline shift of more than 10 mm. 17. Tumor crossing midline: It is defined as either the enhancing or non-enhancing tumor crosses into contralateral hemisphere through white matter commissures (not including the herniated ipsilateral parenchyma).
18. Calvarial remodelling: It is defined as erosion of inner table of skull either directly by the tumor or indirectly by the mass effect.
19. Diffusion restriction: It is defined as either presence of restricted diffusion or absence of restriction or facilitated diffusion.
-Restricted diffusion: Hyperintense on diffusion sequence and hypointense on ADC maps. 23. Tumor margins: The margin characteristics of the tumor are provisionally grouped into three subjective categories, namely -Well defined: All the margins were smooth and well defined.
-Ill defined: Most of the margins were ill-defined and irregular.
-Well defined with areas of focal infiltration: Most of the margins were smooth and well defined with few areas of ill-defined and irregular margins.
25. rCBV value: The value of relative cerebral blood volume is calculated based on the dynamic susceptibility contrast MRI.
26. Closeness to subventricular zone: The distance of the closest margin of the tumor from the ventricular margin. It is subdivided into -Distance of the tumor less than 5 mm and more than 5 mm. 27. Dural enhancement: It is defined as the enhancement along the dural lining in case of a cortical based tumor.
28. Presence of vessel on T2w images: Presence or absence of vessels within the tumor on T2 weighted images.