Introduction

The year 2012 was a milestone in the expanding knowledge of midline gliomas with the identification of histone gene mutations [1,2,3]. Diffuse midline glioma (DMG) as an entity of tumors was described for the first time in the WHO classification of CNS tumors 2016 (4). According to cIMPACT-NOW (the Consortium to Inform Molecular and Practical Approaches to CNS Tumor Taxonomy—Not Official WHO) update 2, the entity DMG, H3K27M-mutant should be used only for the tumors that are diffuse, i.e., infiltrating, midline (involving midline structures, e.g., thalamus, brainstem, spinal cord, etc.), gliomas (astrocytic lineage), and are H3K27M mutant [5]. The newly defined entity includes tumors previously denoted as diffuse intrinsic pontine glioma or brainstem glioma along with other gliomas arising from the thalamus and spinal cord with occasional cases of tumors involving the hypothalamus, ganglio-capsular region, third ventricle, pineal region, cerebellum, and cerebellar peduncles [4, 6,7,8,9]. These midline tumors carry a similar recognized mutation (lysine to methionine substitution) at the 27 residues of the N-terminal tail of histone H3 variants, H3.3 (gene H3F3A), and H3.1 (gene HIST1H3B/C) [2, 10]. With the growing awareness of the mutations in the pathway of DMG tumorigenesis (H3 K27, TP53, ACVR1, PDGFRA, EGFR, EZHIP), the newer terminology has been given to the DMGs, i.e., “diffuse midline glioma, H3 K27-altered” in the recent 2021 WHO classification of CNS Tumors [11].

H3K27M-mutant DMGs (M-DMGs) clinically exhibit an aggressive course and display a worse overall prognosis with a short median survival interval of almost 9 to 11 months from the time of diagnosis regardless of the site of the tumor as compared to their wild-type counterparts (WT-DMGs) [2, 12]. Histologically, similar to other gliomas, M-DMGs can also show morphological features of grade II to grade IV gliomas. However, in view of their overall dismal prognosis, they have been classified as WHO grade IV neoplasms, whether or not they fall in the histologically defined category of high-grade tumor [4, 6, 8, 13]. The critical anatomical location and the infiltrative nature of the tumor in the eloquent areas of the brain limit any meaningful surgical resection. Radiotherapy, the current treatment of choice, is largely palliative. No survival benefit has been demonstrated by the adjuvant chemotherapy in prior trials [14]. A biopsy is cautiously refrained as the tumor involves eloquent areas of the brain, underscoring the value of imaging features in the diagnosis of DMG [7, 15]. Most of the studies have reported that DMGs exhibits variable radiologic appearances without distinctive imaging characteristics that could differentiate M-DMGs from WT-DMGs [6, 15,16,17]. As the occurrence of H3K27M mutation may foresee an aggressive clinical course of the tumor while concurrently giving a prospect for novel targeted treatment methods such as panobinostat and JMJD3, search for the imaging surrogates of these molecular alterations would be useful (6).

Diffusion-weighted imaging (DWI) has been extensively investigated to determine the grade as well as the molecular status (e.g., IDH mutation status) of the gliomas. The tumor cellularity and the aggressiveness of the lesion have been related to the extent of diffusion within and surrounding the tumor (18,19,20,21). Dynamic susceptibility contrast perfusion-weighted imaging (DSC-PWI) offers the functional/metabolic information of gliomas non-invasively measuring cerebral hemodynamics at the scale of microcirculation. Apart from its ability to discriminate among grades of the gliomas, it could also predict their progression over time as well as their IDH mutation status (19, 22,23,24,25).

To the best of our knowledge, there is only a single study by Chen et al. (2019) that has reported the use of DWI features to differentiate the mutant from WT-DMGs, especially with a limited sample size of 38 cases (19 each, M- and WT-DMGs) (15)(15). Further to this, none of the previous reports has utilized DSC-PWI for the probability assignment of mutational status in the DMGs. We hypothesize that advanced MRI may hold diagnostic value in characterizing DMG and potentially predict mutation status. In this work, we studied the advanced MRI modalities, viz., diffusion and perfusion MRI features of DMG patients retrospectively. Our study aimed at assessing these advanced MRI surrogates to predict the H3K27M mutation status in DMGs non-invasively.

Materials and methods

This retrospective observational study was performed in compliance with the institutional research protocols after institutional research board approval. We retrospectively reviewed our hospital’s database of patients with DMG from July 2016 to July 2020.

Patient cohort

The patients included in the study had MRI suggestive of DMG, involving various midline structures of the brain, including the septal region, thalamus, brainstem, cerebellum, middle cerebellar peduncles, and pineal region. These were patients in whom MRI was performed before any biopsy, surgical resection, or the initiation of other kinds of therapy. The intervention was executed within one month of the MRI examination. Patients with hemispheric and spinal cord gliomas, biopsy-proven WHO grade I pilocytic astrocytomas, non-glioma tumor subtypes, and inconclusive biopsy were excluded from the study. Biopsy-proven DMG cases with MRI study of poor diagnostic quality were excluded. A total of 94 cases satisfied the inclusion criteria, of which 48 patients had M-DMGs, and 46 patients had WT-DMGs (Fig. 1).

Fig. 1
figure 1

Inclusion and exclusion criteria. Flow diagram depicting the selection process and cases included in the study (n, number of patients)

Full size image

Histopathology and immunohistochemistry

Surgical resection had been performed in 71 of 94 patients, and 23 patients underwent stereotactic biopsy. The Neuropathologist had reviewed all the tumors and categorized them as phenotypic low-grade diffuse astrocytomas (grade II), anaplastic astrocytomas (grade III), glioblastomas (GBM) (grade IV), and diffuse midline gliomas-H3K27M-mutant (grade IV). The formalin-fixed paraffin-embedded Sects. (4 μm) from the blocks had been collected on silane-coated slides, and the VENTANA BenchMark automated staining system (VENTANA BenchMark-XT) was used for immunohistochemistry (IHC). For a short time, the sections had been subjected to antigen retrieval and subsequently incubated with primary and secondary antibodies. Hematoxylin was used for counterstaining. The antibody used was H3K27me3 (Millipore, 07–449; 1:100) (H3.3K27Mme3, Malaysia, RM192, 1:100) with appropriate positive or negative controls incorporated in each batch of staining.

Image acquisition

MRI scans had been performed on the 1.5 (Aera 1.5 T, Siemens Medical Systems, Erlangen, Germany) or 3.0 (Achieva 3 T, Philips Medical Systems, Best, Netherlands) Tesla MR scanners using 32-channel head coil as per standard operating procedure with/without sedation. MRI protocol for the diffusion and perfusion data acquisition included the following sequences: diffusion tensor imaging in the axial plane (oriented along anterior commissure-posterior commissure line) with a spin echo-planar sequence (6 directions, TR/TE: 2424/72 ms, section thickness = 4 mm, intersection thickness = 1 mm, NEX 1.0, matrix size 230 × 230, b-values: 0 and 1000 s/mm2 with diffusion gradients encoded in six orthogonal directions to generate 3 sets of diffusion-weighted maps, i.e., the isotropic trace DW, ADC, and FA maps). DSC-PWI was a single-shot gradient-recalled T2*-weighted echo-planar imaging sequence [EPI factor: 41, flip angle: 75°, TR/TE: 1606/40 ms, section thickness: 4 mm, intersection thickness: 1 mm, NEX:1.0, FOV: 220 × 220 mm]. For DSC perfusion, during the first 7 phases, images had been acquired before injecting the contrast material to establish a pre-contrast baseline. In the 8th phase, gadopentetate dimeglumine in a dose of 0.1 mmol/kg body weight and 5 ml/s injection rate was injected intravenously with an MRI-compatible power injector, followed by 20-ml saline chase at the same injection rate. Then, a series of 40 dynamics, comprising of 1200 images (30 sections) were obtained in 1 min 9 s.

Image analysis

The MR diffusion and raw perfusion data were digitally transferred from the picture archiving and communication system to and processed in Philips IntelliSpace Portal (ISP) software version 9.0. Diffusion data of some of the cases were analyzed on the SIEMENS workstation additionally.

For tumor, apparent diffusion coefficient (ADC) values and multiple ROIs of 10–20 mm2 (as per an earlier study by Darbar et al. (26)) were drawn on the ADC maps, and the ROI with the lowest mean value was taken (min ADC) and normalized with the contralateral normal-appearing posterior limb of the internal capsule (PLIC) to get the normalized ADC (nADC). The lowest value was chosen for ADC as it has been observed to show the highest accuracy for distinguishing tumor grade and has been used in various previous studies owing to a better interobserver agreement (26, 27). The ROIs were placed in the solid portions of the tumor, avoiding obvious areas of necrosis, cysts, hemorrhage, or blood vessels which might influence the ADC values. For peritumoral ADC (PT ADC) values, ROIs were placed in the peritumoral T2/FLAIR hyperintense signal, or within 1 cm of the tumor margin, for cases without peritumoral hyperintensity. The PT ADC was also normalized with the contralateral normal PLIC (nPT ADC) (Fig. 2). ADC was expressed in mm2/s.

Fig. 2
figure 2

a and b depict the placement of tumoral, peritumoral, and internal capsule diffusion ROIs in Siemens workstation (a) and Philips ISP (b)

Full size image

For interpretation of the perfusion data, whole-brain rCBV maps were generated by applying a single compartmental model with manual arterial input function (AIF) as well as leakage correction (LC) Boxerman-Weiskoff model (28). Either of the middle cerebral arteries was selected for manual AIF analysis. Multiple ROIs of dimensions of approximately 30–40 mm2 were drawn on the enhancing/diffusion restricted/homogeneous aspect of the tumor. The ROI size was chosen as per our departmental protocol and in accordance with the previous studies (29, 30). The maximum relative CBV (rCBV; as determined by the manual AIF method and expressed as milliliters of blood per 100 g of brain tissue) and leakage-corrected and leakage-uncorrected rCBV (as determined by the LC method) of the tumor ROI were extracted. The maximum rCBV values were selected for quantitative analysis as this approach has been shown to provide the best interobserver and intraobserver reproducibility in previous studies (31, 32). Normalization of the rCBV was performed by calculating ratio of maximum tumor rCBV to the maximum CBV of the contralateral unaffected white matter (nrCBV/ uncorrected nrCBV) to minimize the interindividual/inter-scanner variation in rCBV values (Fig. 3). Similarly, normalized values of tumoral rCBF (cerebral blood flow), corrected rCBV, and K2 values were also calculated. K2 is a permeability-weighted leakage coefficient, derived from the LC model which can be considered an imaging biomarker of microvascular permeability (leakiness) (28).

Fig. 3
figure 3

a shows the placement of ROIs for the tumor perfusion analysis in the normal-appearing white matter and the tumor. b depicts the perfusion curve and the derived values of various perfusion parameters

Full size image

Intergroup analysis of the imaging variables parameters was done between M- and WT-DMG groups, irrespective of tumor grade and location. Furthermore, subgroup analysis for location (thalamic/brainstem) and grade IV M-DMGs and WT-DMGs was also performed.

Statistical analysis

Data was collated offline in a Microsoft Excel 2007 spreadsheet in a de-identified manner and was analyzed using R software version 3.5.2. Interval scale data were presented as means and standard deviations, and nominal scale data as frequencies and percentages. Between-group analysis of interval scale data was conducted using non-parametric Mann–Whitney U test and of nominal scale data, using chi-square test with or without Yate’s correction. For finding cut-off of interval scale variable levels to predict histone mutation status, receiver operating characteristic (ROC) curve analysis was conducted, and the best cut-off was found using Youden’s method. In case of conflict of multiple levels being found by Youden’s method, higher specificity of prediction was used as the benchmark for appropriate level selection. P < 0.05 was considered statistically significant.

Results

Forty-eight (51%) out of the 94 cases were M-DMG (21 males/27 females), while 46 patients (49%) had WT-DMGs (27 males/19 females). The mean age of patients in the M-DMG group was 23.04 ± 13.74 years, while in the WT group was 33.83 ± 18.55 years (P = 0.0000007). The gender distribution was not significantly different between the two groups. In our study, the majority of the tumors were located in the thalamus (60%), followed by the brainstem (25%) and other locations (15%; cerebellum, middle cerebellar peduncles, pineal region, and septum pellucidum). The majority of the brainstem gliomas (18/24) were mutation-positive (P = 0.001) (Table 1). Perfusion data were available in 64 cases, out of which 34 cases were M-DMG.

Table 1 Patients’ demographic data and location of the mutant and WT DMGs
Full size table

Significantly lower PT ADC (mutant—1.1 ± 0.33, WT—1.23 ± 0.34; P = 0.033) and nPT ADC (mutant—1.64 ± 0.48, WT—1.83 ± 0.54; P = 0.040) were noted in the M-DMGs. Among the various perfusion parameters evaluated, the rCBV (mutant—25.17 ± 27.76, WT—13.73 ± 14.83; P = 0.018), nrCBV (mutant—3.44 ± 2.16, WT—2.39 ± 1.25; P = 0.049), rCBF (mutant—266.15 ± 189.26, WT—181.91 ± 167.97; P = 0.017), and uncorrected nrCBV (mutant—3.5 ± 2.05, WT—2.54 ± 1.56; P = 0.019) were statistically different between the two groups, being significantly higher for the M-DMG group (Figs. 4 and 5).

Fig. 4
figure 4

a and c depict the box plots of the distribution of the various diffusion parameters of the tumor, internal capsule, and peritumoral region between the H3 K27M-mutant and WT groups and thalamic subgroup respectively. b and d represent the ROC curves for the significantly different diffusion parameters between the overall M- and WT-DMGs and between the thalamic subgroup respectively

Full size image
Fig. 5
figure 5

a and c depict the box plots of the distribution of the various perfusion parameters of the tumor and normal-appearing white matter between the H3 K27M-mutant and WT groups and thalamic subgroup respectively. b and d represent the ROC curves for the significantly different perfusion parameters between the overall M- and WT-DMGs and between the thalamic subgroup respectively

Full size image

Within the subgroup of thalamic tumors, both the min ADC (mutant—0.76 ± 0.18, WT—0.89 ± 0.24; P = 0.042) and PT ADC (mutant—1.19 ± 0.37, WT—1.3 ± 0.38; P = 0.050) as well as the nADC (mutant—1.11 ± 0.27, WT—1.31 ± 0.36; P = 0.036) and nPT ADC (mutant—1.74 ± 0.53, WT—1.93 ± 0.57; P = 0.049) showed a significant difference, being lower for the M-DMGs than the WT-DMGs. Among the perfusion parameters of the thalamic tumors, the nrCBV (P = 0.043), rCBF (P = 0.021), nrCBF (P = 0.013), uncorrected rCBV (P = 0.009), uncorrected nrCBV (P = 0.008), and corrected rCBV (P = 0.06) showed significant difference, the variables being significantly higher for the M-DMGs (Figs. 4, 5, and 6) (online resource).

Fig. 6
figure 6

Examples of H3K27M-mutant and wild-type DMGs. a–e A 32-year male with bithalamic H3K27M-mutant DMG shows a well-defined, heterogeneously T2 hyperintense tumor. DWI with ADC map shows striking diffusion restriction within the tumor. rCBV map and the curve of the DSC-PWI show that the tumor is markedly hyper-perfused. f–j A 43-year male with left thalamic H3 K27M WT-DMG depicts a poorly marginated T2 hyperintense tumor with associated perilesional edema. DWI with ADC map shows no diffusion restriction within the tumor. rCBV map and the curve of the DSC-PWI show that the tumor is hypoperfused

Full size image

No intergroup differences were noted in the analyzed variables within the subgroup of brainstem DMGs. Furthermore, we did not find any significant difference in the values of FA on DWI and K2 on PWI between the groups (online resource).

ROC curve analysis

We validated our results with the ROC curve analysis. It showed that the PT ADC (cut-off—1.245; AUC 0.627) and nPT ADC (cut-off—1.853; AUC—0.623) were significant independent predictors of H3K27M mutational status in DMGs. In thalamic subgroup nADC (cut-off—1.129; AUC—0.664), PT ADC (cut-off—1.185; AUC—0.651), and nPT ADC (cut-off—1.853; AUC—0.653) demonstrated a significant independent predictive value for H3K27M mutational status (Fig. 4, Table 2).

Table 2 Measurement of cut-off, specificity, sensitivity, and AUC of DWI parameters for assessing the H3 K27M status (overall and thalamic subgroup)
Full size table

Similarly, nrCBV (cut-off—1.83; AUC—0.645) and uncorrected nrCBV (cut-off—2.28; AUC—0.673) could independently discriminate the M- and WT-DMGs with optimal sensitivity and specificity. Among the thalamic tumors, nrCBV (cut-off—1.83; AUC—0.703), nrCBF (cut-off—2.73; AUC—0.738), and uncorrected nrCBV (cut-off—2.27; AUC—0.759) demonstrated a significant independent predictive value for H3K27M mutational status (Fig. 5, Table 3).

Table 3 Measurement of cut-off, specificity, sensitivity, and AUC of DSC-PWI parameters for assessing the H3 K27M status (overall and thalamic subgroup)
Full size table

Discussion

We observed that the peritumoral region of M-DMGs exhibited lower ADC values than those of the WT-DMGs. This, at least in part, indicates higher cell density in the peritumoral area of the M-DMGs. However, the tumor ADC values were not significantly different between the M- and WT-DMGs. This is in contrast to observations of Chen et. al., who have reported that both the minimal tumor ADC and PT ADC of the M-DMGs were significantly lower than the WT-DMGs (15). On subgroup analysis of thalamic tumors, we observed that both the minimal tumor and PT ADC, as well as the normalized tumor and PT ADC, were significantly lower for the M-DMGs than the WT-DMGs, while no such difference was evident when analyzed for brainstem DMGs alone. This location ought to be contextualized for the discordance of our findings with those of Chen et al. as well, wherein the majority of the M-DMGs (n = 12/19, 63.2%) were thalamic while only 15.8% (n = 3/19) were brainstem tumors (15). The difference in the number of brainstem gliomas in our study could possibly have skewed the diffusion results. This again represents the significance of the effect of location on the intrinsic characteristics of the DMGs. Between grade IV M- and WT-DMGs, PT ADC and nPT ADC of WT-DMGs were found to be significantly higher than that of M-DMGs. The greater cellular proliferation and surrounding parenchymal infiltration by the tumor cells in M-DMGs than their WT counterparts could be a possible explanation for this finding; both the tumors being matched in their high-grade histomorphology.

The present study also revealed that the rCBV ratio (or nrCBV) and uncorrected rCBV ratio (or uncorrected nrCBV) were significant independent variables correlating to the histone mutation status in DMGs. Both these parameters were significantly high (P = 0.049 and 0.019 respectively) in the mutant as compared to the WT-DMGs. Thalamic M- and WT-DMGs exhibited significant differences in terms of nrCBF and corrected rCBV also, in addition to the nrCBV and uncorrected nrCBV. These parameters with optimal sensitivity and specificity can serve as biomarkers of the mutational status of DMGs (Table 3). The H3K27M mutation drives the overexpression of growth factors like PDGF and VEGF due to amplification and recurrent mutations of signaling genes like ACVR1, PI3K, and RTKs. These later mutations promote neoangiogenesis (33, 34), which could be the possible reason for the M-DMGs being perfused higher than the WT-DMGs as witnessed in this study. Piccardo et al. also observed in their study that on comparing the mutant and WT-DMGs (irrespective of the tumor histologic grade), the arterial spin labeling (ASL)–derived rCBF was higher in the M-DMGs (35). The lack of intergroup difference in the perfusion results in the brainstem DMGs could indicate that the influence of this mutation in different anatomical locations may be variable, and these tumor subsets can be studied in a more comprehensive manner.

Our study has some general and study-specific limitations. Firstly, there was a lack of clinical follow-up data regarding the survival and prognosis of the patients. Secondly, we did not perform subgrouping of the histone of the H3.3/H3.1 gene group that could have been interesting. Thirdly, the data were acquired on two scanners with different field strengths. Previous reports (36,37,38) had suggested that the ADC values were comparable in gliomas as well as in head and neck cancers when measured using 1.5-T and 3-T scanners. Their results led us to compare our results among different field strengths. We have tried to address this limitation by normalizing the DWI and PWI parameters with the uninvolved white matter structures; however, we acknowledge that this factor might have an impact on the generalizability of the results. Lastly, it was a single-institution retrospective study. Future studies with more extensive clinical data are needed to establish the generalizability of the results and recognize various other aspects of M-DMGs, like outcome and management strategies, which are not discussed in this paper.

Conclusion

This work, a first of its kind, has explored the diagnostic utility of advanced MRI modalities, viz., perfusion MRI and diffusion MRI in patients with M- and WT-DMGs. The diffusion MRI features (tumor and PT ADC and normalized tumor and nPT ADC) and perfusion MRI features (normalized rCBV, rCBF, and uncorrected rCBV) can potentially discriminate between the M-DMGs and WT-DMGs. The diffusion within the tumor as well as the PT region of the M-DMGs is lower as compared to the WT-DMGs, and the M-DMGs are markedly hyper-perfused as compared to the WT-DMGs. Larger studies are, however, needed to further investigate these results.