Assessing the predictability of IDH mutation and MGMT methylation status in glioma patients using relaxation-compensated multipool CEST MRI at 7.0 T

Abstract

Background

Early identification of prognostic superior characteristics in glioma patients such as isocitrate dehydrogenase (IDH) mutation and O⁶-methylguanine-DNA-methyltransferase (MGMT) promoter methylation status is of great clinical importance. The study purpose was to investigate the non-invasive predictability of IDH mutation status, MGMT promoter methylation, and differentiation of low-grade versus high-grade glioma (LGG vs HGG) in newly diagnosed patients employing relaxation-compensated multipool chemical exchange saturation transfer (CEST) MRI at 7.0 Tesla.

Methods

Thirty-one patients with newly diagnosed glioma were included in this prospective study. CEST MRI was performed at a 7T whole-body scanner. Nuclear Overhauser effect (NOE) and isolated amide proton transfer (APT; downfield NOE-suppressed APT = dns-APT) CEST signals (mean value and 90^th signal percentile) were quantitatively investigated in the whole tumor area with regard to predictability of IDH mutation, MGMT promoter methylation status, and differentiation of LGG versus HGG. Statistics were performed using receiver operating characteristic (ROC) and area under the curve (AUC) analysis. Results were compared with advanced MRI methods (apparent diffusion coefficient and relative cerebral blood volume ROC/AUC analysis) obtained at 3T.

Results

dns-APT CEST yielded highest AUCs in IDH mutation status prediction (dns-APT_mean = 91.84%, P < 0.01; dns-APT₉₀ = 97.96%, P < 0.001). Furthermore, dns-APT metrics enabled significant differentiation of LGG versus HGG (AUC: dns-APT_mean = 0.78, P < 0.05; dns-APT₉₀ = 0.83, P < 0.05). There was no significant difference regarding MGMT promoter methylation status at any contrast (P > 0.05).

Conclusions

Relaxation-compensated multipool CEST MRI, particularly dns-APT imaging, enabled prediction of IDH mutation status and differentiation of LGG versus HGG and should therefore be considered as a non-invasive MR biomarker in the diagnostic workup.

Importance of the study

Non-invasive early identification of prognostic superior characteristics in glioma patients such as IDH mutation and MGMT promoter methylation status is of great clinical importance for determination of the appropriate therapy. Our study demonstrates that relaxation-compensated multipool CEST MRI at 7.0T enables non-invasive prediction of IDH mutation status and differentiation of lower- versus higher-grade tumors in patients with newly diagnosed untreated glioma. Consequently, the study forms a fundament for further clinical application of multipool CEST MRI as a non-invasive imaging biomarker in the preoperative diagnostic workup of glioma patients.

Gliomas account for most of primary brain tumors in adults.¹ Patients with high-grade glioma (HGG), especially World Health Organization (WHO) grade IV glioblastomas (GBMs), still have a dismal prognosis, even after resection and additive radiochemotherapy.² Therefore, early identification of patients with prognostic superior characteristics is of utmost importance. Survival in general, radiosensitivity, and chemosensitivity are highly dependent on both WHO grade and histopathological characteristics, particularly isocitrate dehydrogenase (IDH) mutation and O⁶-methylguanine-DNA-methyltransferase (MGMT) promoter methylation status.^3–6 Nevertheless, this information can currently only be provided by resection or biopsy, through histopathological analysis, although clinicians are in desperate need of this information from the very beginning of the disease. However, gadolinium contrast-enhanced (GdCE) multiparametric MRI, the key element of today’s diagnostic routine, is not capable of gaining reliable data about histopathological characteristics.^7–10 Promising results have been reported on in vivo detection of 2-hydroxyglutarate (2-HG) using magnetic resonance spectroscopy (MRS), possibly enabling determination of IDH status in glioma patients.^11,12 However, MRS-based detection of 2-HG is still challenging and therefore not widely used in clinical practice. Consequently, alternative MRI sequences that non-invasively provide reliable information on histopathological characteristics are urgently needed.

Chemical exchange saturation transfer (CEST) imaging might have the potential to cope with this problem. CEST contrasts are based on the spontaneous chemical exchange between solute-bound protons and protons of free water that to a large extent depends on the concentration of endogenous cellular proteins. Furthermore, CEST signals are sensitive to tissue microenvironment, such as pH, affecting proton exchange properties.^13,14 Consequently, through its sensitivity to local protein concentration,^13,15 intracellular pH,^16–18 and protein states,^19,20 CEST imaging provides complementary information to current standard MRI sequences. Recently, Jiang et al reported the predictability of IDH mutation status in grade II gliomas employing amide proton transfer-weighted (APTw) CEST MRI at 3T in a retrospectively analyzed study cohort.²¹ Furthermore, Jiang et al found a significant correlation between MGMT promoter methylation status and APTw CEST metrics.²² Similar to the aforementioned studies, most previous CEST approaches were based on magnetization transfer ratio asymmetry (MTR_asym) analyses.^13,23,24 However, the limitation of these approaches is that the MTR_asym contrasts have multiple contributions, including upfield and downfield relayed nuclear Overhauser effect (rNOE) signals, T₁ and T₂ relaxation times, and conventional semisolid magnetization transfer (MT) asymmetry.^25–28 Consequently, a proper separation of different CEST pools is essential to better understand the role of each effect in brain tumor imaging and to investigate its clinical diagnostic value. For CEST data with high spectral resolution obtained at 7T, it has been shown that this separation is feasible using Lorentzian fit analysis.^26,27 Furthermore, using the relaxation-compensated metric yields the apparent exchange-dependent relaxation rate (AREX), which is corrected for contributions of semisolid MT, T₂ relaxation (spillover), and T₁ relaxation times.²⁷ Ultimately, it has been shown that downfield rNOE signals can be eliminated, yielding the downfield rNOE suppressed amide proton CEST contrast (dns-APT).

In this prospective study, we investigated isolated APT- and NOE-mediated CEST effects and the predictability of histopathological features using the obtained multipool CEST metrics, in a large glioma patient study cohort at 7.0T. We hypothesized that the CEST contrasts may enable prediction of the molecular parameters IDH mutation, MGMT promoter methylation status, and differentiation of lower- versus higher-grade WHO tumors in patients with newly diagnosed gliomas.

Materials and Methods

Patients

This prospective study received approval by the local institutional review board committee. Thirty-eight patients with newly diagnosed untreated intracranial expanding lesions were included after the nature and possible consequences of the study were explained and written informed consent was obtained. Inclusion criteria were patients with MRI findings suspicious for glioma, previously untreated, age 18 years or older, and eligibility for 7T MRI.

In total, 31 patients had histopathological diagnosis of glioma (WHO grades II–IV) subsequent to the study examination; 5 patients had tumors of a nonglial entity (metastasis, lymphoma), and 2 patients with typical low-grade glioma (LGG) lesions did not undergo surgical resection or biopsy, since a watch-and-wait strategy was chosen. Consequently, those patients were excluded from statistical analyses (Table 1). Eleven patients of the study cohort had been previously included in methodical publications.^27,28

Table 1.

Glioma patient cohort included in this study*

Patient	Age	Sex	Diagnosis*	WHO Grade	MGMT	IDH
#1	46	f	GBM	IV	unmet	wt
#2	51	f	GBM	IV	n/a	n/a
#3	69	m	GBM	IV	met	wt
#4	60	m	GBM	IV	met	wt
#5	53	m	GBM	IV	met	wt
#6	53	f	GBM	IV	IND	wt
#7	52	m	GBM	IV	unmet	wt
#8	75	m	GBM	IV	IND	wt
#9	65	m	GBM	IV	unmet	wt
#10	65	m	GBM	IV	met	wt
#11	71	m	GBM	IV	IND	wt
#12	81	f	GBM	IV	met	wt
#13	54	m	GBM	IV	unmet	wt
#14	53	f	GBM	IV	met	wt
#15	63	f	GBM	IV	unmet	wt
#16	59	m	Gliosarcoma	IV	unmet	wt
#17	59	m	GBM	IV	unmet	wt
#18	61	m	GBM	IV	unmet	wt
#19	48	f	GBM	IV	met	wt
#20	50	m	GBM	IV	IND	wt
#21	50	m	GBM	IV	unmet	wt
#22	86	m	GBM	IV	met	wt
#23	41	f	GBM	IV	n/a	mut
#24	42	f	AA	III	met	mut
#25	27	f	AA	III	met	mut
#26	48	f	ODG	II	met	mut
#27	25	m	Astrocytoma	II	n/a	mut
#28	49	f	ODG	II	n/a	mut
#29	74	m	Astrocytoma	II	n/a	wt
#30	29	f	ODG	II	met	mut
#31	39	f	Astrocytoma	II	met	mut

*All diagnoses were histopathologically proven after surgical resection or tumor biopsy subsequent to CEST MRI at 7.0 T (AA = anaplastic astrocytoma, GBM = glioblastoma, ODG = oligodendroglioma). n/a = data not available, IND = result of histopathological analysis indeterminate. WHO grade II tumors were considered LGG and WHO grades III‒IV as HGG.

Histopathology

All diagnoses were histopathologically proven after surgical resection or tumor biopsy, according to the 2016 WHO classification of CNS tumors, by neuropathologists who were blinded to the MRI data. WHO tumor grade was obtained for all patients (6/31 = 19.4% LGG; 25/31 = 80.6% HGG), IDH1-R132H mutation status for 30/31 patients (8/30 = 26.7% IDH mutant [mut]; 22/30 = 73.3% IDH wildtype [wt]), and MGMT promoter methylation status in 26/31 cases (13/26 = 50.0% methylated [met]; 9/26 = 34.6% unmethylated [unmet]; 4/26 = 15.4% indeterminate). All histopathological findings of the study cohort are described in Table 1.

MRI Acquisition

Conventional MRI at 3T

The acquisition of conventional MRI exams was performed at 3T, employing the following protocol parameters: T1-weighted GdCE MRI (GdCE-T1) (echo time [TE] = 4.04 ms; repetition time [TR] = 1710 ms; field of view [FoV] in mm²: 256 × 256; matrix: 512 × 512; slice thickness: 1 mm), T2 fluid attenuated inversion recovery (TE = 135 ms; TR = 8500 ms; FoV: 230 × 172 mm²; matrix: 256 × 192; slice thickness: 5 mm), and T2-weighted MRI turbo spin echo (TSE) (TE = 86 ms; TR = 5550 ms; FoV: 229 × 172 mm²; matrix: 384 × 230; slice thickness: 5 mm). Apparent diffusion coefficient (ADC) maps were generated from diffusion-weighted imaging (DWI) employing a single-shot spin-echo echo-planar imaging (EPI) sequence, relative cerebral blood volume (rCBV) maps were generated from dynamic susceptibility contrast-enhanced (DSC) MRI performed after gadolinium prebolus injection using a T2*-weighted gradient-echo EPI sequence; both techniques have been described previously in detail.²⁹ DSC-MRI and DWI data were available in 26 of the 31 (~84%) glioma patients included. In 5 patients these sequences were not acquired due to time constraints in daily clinical practice. Coregistration of all images was performed using an automatic multimodal rigid registration algorithm of the Medical Imaging Interaction Toolkit.³⁰

CEST MRI at 7T and data post processing

CEST MRI was performed on a 7T whole body MRI scanner (Magnetom 7.0T; Siemens Healthineers) employing a custom-developed sequence based on a centric-reordered 2D single slice gradient echo (TE = 3.76 ms; TR = 7.6 ms; FoV: 200 × 175 mm²; matrix: 128 × 112; resolution: 1.56 × 1.56 × 5 mm³; flip angle: 10°) and a single channel transmit (Tx)/24-channel receive (Rx) 1H head coil as previously described.²⁷ In vivo CEST images were obtained following saturation by a train of 150 Gaussian-shaped radiofrequency pulses (pulse width = 15 ms, duration time = 10 ms, duty cycle = 60%, saturation time = 3.75 s) for 2 different B₁ amplitudes (1.0 μT and 0.6 μT) at 65 unevenly sampled saturation offsets dubbed Δω.²⁷ To achieve separation of APT- and NOE-mediated CEST effects and correction for conventional MT asymmetry, Z-spectra were fitted pixelwise by a multi-Lorentzian fit approach as recently reported by Windschuh et al.³¹ The fit model provides a specifically labeled (Z_lab) and a reference Z-spectrum (Z_ref) for each CEST effect, which are used to calculate the T₁ and T₂ relaxation-compensated and MT-corrected contrasts:

$$\text{AREX}=\left(\frac{1}{{\text{Z}}_{\text{lab}}}–\frac{1}{{\text{Z}}_{\text{ref}}}\right)\text{⋅}\frac{1}{{\text{ T}}_1}$$ (1)

An explicit derivation of the formula has been previously reported.²⁷ Ultimately, the downfield-rNOE-suppressed (dns)-APT imaging technique additionally removes downfield resonating rNOE contributions from the amide signal. The contrast was calculated at $\Delta \text{ω}$ = 3.5 ppm by using the ratio r_rNOE = 0.2 as proposed by Zaiss et al²⁸:

$$\text{dns-APT}\left(+\Delta \text{ω}\right)=\text{AREX}\left(+\Delta \text{ω}\right)-{\text{r}}_{\text{rNOE}}\cdot \text{AREX}\left(-\Delta \text{ω}\right)$$ (2)

B₀ and B₁ maps were acquired simultaneously by means of the water shift and B₁ (WASABI) method.³² B₁ correction was achieved employing the B₁-correction method proposed by Windschuh et al, yielding CEST contrasts corresponding to 0.6 μT saturation amplitude.³¹ The CEST scans at 2 B₁ amplitudes required approximately 4 min each, the WASABI scan about 2 min, and an additional T₁ map 1:20 min, yielding total measurement duration of approximately 11 min for B₀- and B₁-corrected, relaxation-compensated, multipool CEST MRI.

Data Analysis

The tumor region, defined as the whole area of abnormal signal intensity on T1-GdCE and T2-weighted images, was manually segmented (by A.R., with 9 years of experience in neuroradiology) for quantitative analysis in each patient. Necrotic tumor regions were thoroughly excluded. The approach of whole tumor segmentation and subsequent determination of signal mean and 90^th percentile of all included voxel intensities allowed reproducibility, in contrast to manual selection of multiple regions of interest. For NOE CEST contrasts, the 10^th percentile of included voxels was determined, since NOE-mediated effects are known to be decreased in tumor lesions.^26,33–35 Accordingly, mean value and 10^th percentile of all included voxels were determined for the ADC, while mean signal intensity and 90^th percentile were calculated for rCBV.

Statistics

Receiver operating characteristic (ROC) area under the curve (AUC) analyses were performed for the investigated CEST metrics (NOE mean, NOE 10^th percentile [NOE₁₀], APT mean, APT 90^th percentile [APT₉₀], dns-APT mean, and dns-APT 90^th percentile [dns-APT₉₀]) and the advanced imaging methods acquired at 3T (ADC mean, ADC 10^th percentile, rCBV mean, rCBV 90^th percentile) in order to assess the contrasts’ ability to predict (i) WHO tumor grade (LGG = grades I–II vs HGG = grades III–IV), (ii) IDH mutation (IDH1-R132H wt vs mut), and (iii) MGMT promoter methylation status (met vs unmet). Best sensitivity and specificity pairs were determined for each contrast and parameter with stronger weighting of specificity (condition: specificity >75%). Furthermore, mean signal intensity and standard deviation were calculated for each contrast and parameter category over all patients of the group (WHO grade: HGG/LGG, IDH: wt/mut, and MGMT promoter: unmet/met). Statistics were performed with SigmaPlot version 12.5 (SystatSoftware). For statistical analyses the level of significance was set to P < 0.05.

Results

Predictability of WHO Tumor Grade

Reliable prediction of WHO tumor grade (LGG vs HGG) was only possible using the dns-APT metric yielding areas under the ROC curve of 0.78 (P < 0.05, mean dns-APT) and 0.83 (P < 0.05, dns-APT₉₀) (Fig. 1). Best classification was achieved using the mean dns-APT metric with a test sensitivity and specificity of 71% and 100% (cutoff value = 1.88). All other CEST contrasts and rCBV analysis showed a tendency toward higher values in HGG tumors but did not allow for significant differentiation of lower- versus higher-grade glioma (P > 0.05). No trend was observed for the investigated ADC metrics.

Fig. 1.

Predictability of WHO tumor grade (HGG vs LGG) in newly diagnosed untreated glioma. (A, B) dns-APT allowed prediction of WHO tumor grade with highest AUC for the dns-APT₉₀ metric (0.83) and a test sensitivity/specificity of 0.63 (0.41–0.81)/1.00 (0.48–1.00) (P < 0.05). Two exemplary patients with HGG (GBM, c₁–g₁) and LGG (oligodendroglioma II, c₂–g₂) shown: c_i: GdCE T₁-w, d_i: T₂-w (TSE), relaxation-compensated multipool CEST MRI at 7T with separated APT (e_i), NOE (f_i), and dns-APT (g_i) effects (unit: %). The GBM patient shows only a small spot-like contrast enhancement in the tumor region (c₁), while a clear hyperintensity displays within the tumor at dns-APT imaging (g₁, white arrow).

Predictability of IDH Mutation Status

Higher signal intensity in tumors of patients with IDH-wt status allowed significant prediction of IDH-mut status at multipool CEST imaging with highest AUCs at dns-APT contrasts (mean APT = 0.88; P < 0.05, APT₉₀ = 0.90; P < 0.05, mean dns-APT = 0.92; P < 0.01, and dns-APT₉₀ = 0.98; P < 0.001). Mean NOE signal intensity yielded an AUC of 0.78 (P < 0.05) (Fig. 2). Furthermore, perfusion MRI showed significantly increased rCBV in IDH-wt tumors, yielding an AUC of 0.79 (P < 0.05). ADC values did not enable significant prediction of IDH-mut status. Best test performance was achieved using the dns-APT₉₀ metric with sensitivity and specificity of 95% and 100%, respectively, at a cutoff value of 2.86.

Fig. 2.

Predictability of IDH mutation status (IDH1-R132H wt versus mutation) in newly diagnosed untreated glioma. (A, B) APT and dns-APT CEST metrics allowed prediction of IDH mutation status with highest AUC for the dns-APT₉₀ metric (0.98) with a test sensitivity/specificity of 0.95 (0.77–1.00)/1.00 (0.59–1.00) (P < 0.0001). Two exemplary patients with newly diagnosed GBM IDH-wt (c₁ –g₁) and IDH-mut (c₂ –g₂) shown: c_i: GdCE T₁-w, d_i: T₂-w (TSE), relaxation-compensated multipool CEST MRI at 7T with separated APT (e_i), NOE (f_i), and dns-APT (g_i) effects (unit: %). A ring-like hyperintensity can be delineated in the periphery of the IDH-wt glioblastoma at dns-APT imaging (g₁, white arrow), while the IDH-mut GBM displays barely hyperintense at dns-APT (g₂, white arrow). The head of the caudate nucleus also displays hyperintense on dns-APT images (g₂, pink arrows).

Predictability of MGMT Promoter Methylation Status

None of the investigated CEST metrics at 7T and none of the advanced MRI methods at 3T allowed for significant prediction of MGMT promoter methylation in glioma patients (P > 0.05) (Fig. 3). Trends toward higher signal intensities in patients with unmethylated MGMT promoter were observed at all investigated CEST contrasts, while no trend could be found for ADC and rCBV values. The results of all ROC curve analyses are provided in Table 2.

Fig. 3.

Predictability of MGMT promoter methylation status (unmethylated vs methylated) in newly diagnosed untreated glioma patients. (A, B) None of the investigated CEST contrasts allowed for significant prediction of the MGMT promoter methylation in glioma patients. Two GBM patients with unmethylated (GBM MGMT−: c₁–g₁) and methylated (GBM MGMT+: c₂–g₂) promoter shown: c_i: GdCE T₁-w, d_i: T₂-w (TSE), relaxation-compensated multipool CEST MRI at 7T with separated APT (e_i), NOE (f_i), and dns-APT (g_i) effects (unit: %). A tendency toward higher signal intensities for patients with unmethylated MGMT promoter was observed for all investigated CEST contrasts.

Table 2.

Results of quantitative analyses for multipool CEST MRI at 7.0T and advanced MRI methods at 3.0T

WHO Tumor Grade (LGG vs HGG)
CEST MRI @ 7T	LGG (mean ± SD)	HGG (mean ± SD)	Cutoff	Sensitivity (95% CI)	Specificity (95% CI)	AUC	P-val.
NOE mean	7.69 ± 3.96	9.06 ± 3.21	8.95	0.61 (0.39–0.80)	0.83 (0.36–1.00)	0.66	0.24
NOE 10th pc	4.24 ± 2.59	4.22 ± 2.01	5.18	0.33 (0.16–0.55)	0.83 (0.36–1.00)	0.50	1.00
APT mean	3.07 ± 1.50	3.96 ± 1.32	3.66	0.79 (0.58–0.93)	0.80 (0.28–0.99)	0.76	0.07
APT 90th pc	4.54 ± 2.15	6.03 ± 2.26	5.42	0.67 (0.45–0.84)	0.80 (0.28–0.99)	0.73	0.11
dns-APT mean	1.47 ± 0.68	2.14 ± 0.85	1.88	0.71 (0.49–0.87)	1.00 (0.48–1.00)	0.78	0.0497
dns-APT 90th pc	2.37 ± 1.20	4.01 ± 1.85	3.62	0.63 (0.41–0.81)	1.00 (0.48–1.00)	0.83	0.0234
Advanced MRI methods @ 3T
rCBV	462 ± 405	914 ± 581	606.8	0.68 (0.45–0.86)	0.75 (0.19–0.99)	0.73	0.17
rCBV 90th pc	1069 ± 863	1945 ± 1024	1317	0.73 (0.50–0.89)	0.75 (0.19–0.99)	0.74	0.16
ADC mean	1168 ± 231	1117 ± 146	1189	0.36 (0.17–0.59)	0.80 (0.28–0.99)	0.53	0.85
ADC 10th pc	772 ± 220	749 ± 145	876.9	0.29 (0.11–0.52)	0.80 (0.28–0.99)	0.53	0.87
IDH status (IDH1-R132H wildtype vs mutation)
CEST MRI @ 7T	IDH mut (mean ± std)	IDH wt (mean ± std)	Cutoff	Sensitivity (95% CI)	Specificity (95% CI)	AUC	P-val.
NOE mean	6.03 ± 4.55	9.68 ± 2.15	8.95	0.62 (0.41–0.83)	0.88 (0.47–1.00)	0.78	0.02
NOE 10th pc	3.63 ± 2.94	4.29 ± 1.69	5.18	0.29 (0.11–0.50)	0.75 (0.35–0.97)	0.56	0.64
APT mean	2.30 ± 1.77	4.30 ± 0.80	3.66	0.86 (0.65–0.97)	0.86 (0.42–1.00)	0.88	0.0032
APT 90th pc	3.36 ± 2.43	6.67 ± 1.64	5.22	0.86 (0.65–0.97)	0.86 (0.42–1.00)	0.90	0.0019
dns-APT mean	1.10 ± 0.81	2.36 ± 0.61	1.88	0.81 (0.60–0.95)	1.00 (0.59–1.00)	0.92	0.0011
dns-APT 90th pc	1.69 ± 1.13	4.45 ± 1.53	2.86	0.95 (0.77–1.00)	1.00 (0.59–1.00)	0.98	0.0001
Advanced MRI methods @ 3T
rCBV	465 ± 390	1002 ± 567	851	0.67 (0.43–0.87)	0.86 (0.42–1.00)	0.79	0.0251
rCBV 90th pc	1133 ± 871	2104 ± 974	2072	0.61 (0.38–0.84)	0.86 (0.42–1.00)	0.79	0.0293
ADC mean	1190 ± 239	1107 ± 142	911	0.10 (0.01–0.30)	0.83 (0.36–1.00)	0.63	0.33
ADC 10th pc	833 ± 108	721 ± 167	737	0.53 (0.32–0.77)	0.83 (0.36–1.00)	0.72	0.11
MGMT promoter methylation status (methylated [+] vs unmethylated [−])
CEST MRI @ 7T	MGMT + (mean ± std)	MGMT - (mean ± std)	Cutoff	Sensitivity (95% CI)	Specificity (95% CI)	AUC	P-val.
NOE mean	7.34 ± 3.76	9.97 ± 3.09	10.12	0.56 (0.30–0.80)	0.84 (0.54–0.98)	0.68	0.15
NOE 10th pc	3.38 ± 1.88	4.86 ± 2.04	4.60	0.44 (0.20–0.70)	0.77 (0.46–0.95)	0.69	0.13
APT mean	3.35 ± 1.85	4.34 ± 0.95	4.73	0.44 (0.20–0.70)	0.75 (0.43–0.95)	0.68	0.17
APT 90th pc	5.01 ± 2.74	6.18 ± 1.10	6.59	0.44 (0.20–0.70)	0.75 (0.43–0.95)	0.62	0.34
dns-APT mean	1.86 ± 1.11	2.35 ± 0.69	2.71	0.22 (0.07–0.52)	0.75 (0.43–0.95)	0.61	0.39
dns-APT 90th pc	3.20 ± 1.93	4.03 ± 1.00	6.59	0.44 (0.20–0.70)	0.75 (0.43–0.95)	0.62	0.34
Advanced MRI methods @ 3T
rCBV	730 ± 582	986 ± 688	1167	0.50 (0.19–0.75)	0.83 (0.52–0.98)	0.58	0.60
rCBV 90th pc	1561 ± 1065	2087 ± 1165	2453	0.50 (0.19–0.75)	0.92 (0.62–1.00)	0.59	0.56
ADC mean	1161 ± 176	1087 ± 195	955	0.25 (0.07–0.52)	0.91 (0.48–0.98)	0.59	0.51
ADC 10th pc	779 ± 114	787 ± 115	884	0.13 (0.02–0.40)	0.82 (0.48–0.98)	0.52	0.90

Receiver operating characteristic (ROC) area under the curve (AUC) analyses performed for NOE mean, NOE 10^th percentile (NOE¹⁰), APT mean, APT 90^th percentile (APT₉₀), dns-APT mean, and dns-APT 90^th percentile (dns-APT₉₀) in order to assess the contrasts’ ability to predict (1) WHO tumor grade (grades III‒IV versus grade II), (2) IDH mutation status (IDH1-R132H wildtype versus mutation), and (3) MGMT promoter methylation status (methylated versus unmethylated) in untreated glioma patients. Same analyses were performed for ADC (mean, 10^th percentile) and rCBV (mean, 90^th percentile) obtained from DWI and DSC MRI at 3T. Best sensitivity and specificity with 95% CI shown for each contrast and category with stronger weighting of specificity (condition: specificity >75%). Mean signal intensity and standard deviation for each contrast and category over all patients additionally provided (CEST MRI [%], ADC [µm²/s], rCBV [dimensionless])

Discussion

In this study we investigated novel relaxation-compensated multipool CEST MRI contrasts at 7.0T and demonstrated the non-invasive predictability of IDH mutation status and WHO tumor grade in untreated glioma patients, with the best test performance using the dns-APT metric. None of the investigated CEST contrasts and none of the advanced MRI methods at 3.0T allowed significant differentiation between glioma patients with methylated and unmethylated MGMT promoter.

Prediction of IDH Mutation Status

Relaxation-compensated multipool CEST MRI, particularly APT and dns-APT imaging, performed markedly better in predicting IDH mutation status than WHO tumor grade in the investigated study cohort. Therefore, we infer that genetic status, by means of IDH mutation status, is more reflective than the histologic class, regarding APT-mediated CEST effects.

The latest WHO classification of CNS tumors termed gliomas of WHO grades I‒III together as “lower grade gliomas” with a great majority of IDH-mut and a wide range of overall survival within this group.^36,37 Glioblastomas were divided into GBM IDH-wt, most frequently de novo tumors (~90% of cases), and GBM IDH-mut (~10% of cases), closely corresponding to so-called secondary glioblastoma.³⁶ Consequently, the 2016 WHO CNS definition allows for dynamic classification based on both phenotype and genotype, taking into account the prognostic importance of IDH mutation status.³⁶ Generally, histologically low-grade diffuse IDH-wt tumors are known to clinically often resemble aggressive grade IV GBMs.³⁸ Mutations in IDH gene-encoded enzymes are expected to cause widespread disturbances of cellular metabolism, including alteration of amino acid concentrations and global downregulation of protein expression.²¹ Hence, APT-mediated CEST signal is expected to be significantly lower in IDH-mut tumors, which is in agreement with our study findings. Alterations of tissue microenvironment (eg, pH, lactate acid production), particularly in IDH-wt tumors, may additionally affect the amide CEST signal. Therefore, CEST MRI is an alternative indirect method for determining IDH mutation status, significantly extending the existing repertoire of non-invasive imaging biomarkers used in the diagnostic workup of brain tumor patients.

Best test performance of CEST MRI based preoperative prediction of IDH mutation status was achieved using APT and dns-APT metrics with highest sensitivity/specificity (95%/100%, AUC = 0.98) for the dns-APT₉₀ contrast. This result outperforms the findings of Jiang et al, who reported a sensitivity and specificity of 57% and 100%, respectively (AUC = 0.89), using APTw MR imaging at 3T.²¹ However, Jiang et al investigated APTw CEST metrics in LGG patients, thereby limiting the comparability of the study cohorts. NOE-mediated CEST imaging yielded inferior performance in IDH status prediction (sensitivity, 61%; specificity, 83%) compared with all investigated APT CEST metrics in our study.

Compared with rCBV and ADC ROC analyses, the amide CEST metrics, particularly dns-APT, yielded clearly better results in IDH mutation status prediction. However, results of rCBV ROC analysis were significant too, which is in agreement with previous studies reporting significantly increased rCBV in IDH-wt tumors of glioma patients.^39,40 ADC values tended to be lower in IDH-wt tumors, but differences were not statistically significant.

Prediction of WHO Tumor Grade

Preoperative assessment of glioma employing APTw CEST MRI based on MTR_asym at clinical field strength (3.0T) has previously been shown in several studies to positively correlate with the tumors’ histopathological grade.^24,41–43 Moreover, the ability of APTw imaging to distinguish non-enhancing high-grade lesions from LGG was demonstrated.⁴⁴ Our study confirms these findings and additional proofs that these previous results particularly hold true for isolated APT effects, employing relaxation-compensated dns-APT MRI. For conventional (non-dns) APT effects at 7T, a positive trend but no significant differences according to WHO tumor grade were observed, which is in agreement with previous results obtained in smaller patient cohorts (n = 10).³⁴ Regarding NOE-mediated CEST effects at 7T, Heo et al reported significantly decreased NOE signals in HGGs compared with LGGs (grade IV vs grades I–III, grade III vs grades I–II),³⁴ which could not be confirmed in our study cohort. In this work, NOE effects were not significantly different according to tumor grade. Perfusion imaging by means of rCBV did not enable significant prediction of LGG versus HGG but revealed a clear trend of increased rCBV in HGG tumors, while no trend was observed for the investigated ADC metrics.

MGMT Promoter Methylation Status Prediction

None of the investigated CEST contrasts and none of the advanced MRI methods at 3T allowed for prediction of MGMT promoter methylation status in glioma patients. This result is in contrast to the findings of Jiang et al, who recently reported discriminability of MGMT promoter methylation status in patients with GBM using APTw MRI metrics based on MTR_asym.²² However, a trend toward higher signal intensity in tumors with unmethylated MGMT promoter was also observed in our data. Moreover, Jiang et al employed the APTw sequence at 3.0T with imaging parameters different from our multipool CEST approach at 7.0T.

Generally, prediction of histopathological parameters using MRI parameters has previously been assessed employing radiomics analysis. Radiomics approaches combine information obtained from multiple conventional MR sequences and have so far been tested in relatively small study cohorts with promising results for prediction of IDH mutation status and MGMT promoter methylation.^45–48 CEST, however, is fundamentally different, since the new MRI contrasts provide additional information on a subcellular protein level. Therefore, both approaches are promising and may be combined in the future to most accurately predict histologic features non-invasively. While MRI techniques can currently not replace histopathological analysis, the presented approach may be of major value in surgery planning when performing biopsy and in the follow-up setting for therapy monitoring, as serial tissue biopsies are usually not feasible.

Limitations

Some limitations of this study need to be acknowledged.

• (1) Although this study included the so far largest glioma patient cohort undergoing relaxation-compensated multipool CEST MRI at 7T, the results need to be validated in independent study cohorts. Moreover, the demonstrated relationship of CEST MRI and IDH mutation status in glioma needs to be confirmed in HGG patient cohorts of the same WHO grade. The significant results of this work can justify and motivate such studies.

• (2) The employed multipool CEST MRI approach used a single-slice readout sequence. For routine clinical application we suggest the implementation of three-dimensional readout sequences, which can be combined with the multipool CEST preparation module.

• (3) MR units at 7T are currently not widely available in most clinical settings, which therefore limits the immediate applicability of the presented approach. However, vendors are pushing toward the increase of 7T units on the market with the goal to include them in the clinical routine. Moreover, several studies reported separation of different CEST pools also at clinical field strength (3.0T),^49,50 and the knowledge gained in this study can thus be used as prior knowledge for the preparation and evaluation of those CEST approaches.

Conclusions

In this prospective study we investigated NOE- and APT-mediated CEST effects at 7T and their non-invasive predictability of histopathological parameters in newly diagnosed untreated glioma. Relaxation-compensated multipool CEST MRI allowed for prediction of IDH mutation status and differentiation of LGG versus HGG tumors. Our results suggest that the genetic subtype, by means of IDH mutation status, has a greater effect on the amide CEST signal than the histologic grade of the tumor, particularly at dns-APT imaging. Therefore dns-APT CEST may be considered an imaging biomarker to non-invasively assess genetic features in the preoperative workup and to monitor treatment response of glioma patients.

Funding

This study did not receive funding from third parties.

Conflict of interest statement. None declared.