Spectroscopic imaging of D-2-hydroxyglutarate and other metabolites in pre-surgical patients with IDH-mutant lower-grade gliomas

Purpose Prognostically favorable IDH-mutant gliomas are known to produce oncometabolite D-2-hydroxyglutarate (2HG). In this study, we investigated metabolite-based features of patients with grade 2 and 3 glioma using 2HG-specific in vivo MR spectroscopy, to determine their relationship with image-guided tissue pathology and predictive role in progression-free survival (PFS). Methods Forty-five patients received pre-operative MRIs that included 3-D spectroscopy optimized for 2HG detection. Spectral data were reconstructed and quantified to compare metabolite levels according to molecular pathology (IDH1R132H, 1p/19q, and p53); glioma grade; histological subtype; and T2 lesion versus normal-appearing white matter (NAWM) ROIs. Levels of 2HG were correlated with other metabolites and pathological parameters (cellularity, MIB-1) from image-guided tissue samples using Pearson’s correlation test. Metabolites predictive of PFS were evaluated with Cox proportional hazards models. Results Quantifiable levels of 2HG in 39/42 (93%) IDH+ and 1/3 (33%) IDH– patients indicated a 91.1% apparent detection accuracy. Myo-inositol/total choline (tCho) showed reduced values in astrocytic (1p/19q-wildtype), p53-mutant, and grade 3 (vs. 2) IDH-mutant gliomas (p < 0.05), all of which exhibited higher proportions of astrocytomas. Compared to NAWM, T2 lesions displayed elevated 2HG+ γ-aminobutyric acid (GABA)/total creatine (tCr) (p < 0.001); reduced glutamate/tCr (p < 0.001); increased myo-inositol/tCr (p < 0.001); and higher tCho/tCr (p < 0.001). Levels of 2HG at sampled tissue locations were significantly associated with tCho (R = 0.62; p = 0.002), total NAA (R = − 0.61; p = 0.002) and cellularity (R = 0.37; p = 0.04) but not MIB-1. Increasing levels of 2HG/tCr (p = 0.0007, HR 5.594) and thresholding (≥ 0.905, median value; p = 0.02) predicted adverse PFS. Conclusion In vivo 2HG detection can reasonably be achieved on clinical scanners and increased levels may signal adverse PFS. Supplementary Information The online version contains supplementary material available at 10.1007/s11060-022-04042-3.

In this study, we investigated metabolomic characteristics of patients with IDH-mutant LrGG prior to surgery using a 2HG-targeted, long-TE [14] implementation of 3-D MR spectroscopic imaging (MRSI) that enabled whole-brain coverage. The goals were to evaluate metabolite features of molecular-based classifications of LrGG, correlate them with image-guided tissue pathology, and assess their predictive role on progression-free survival (PFS).

Patient population
Forty-five patients diagnosed with new or recurrent LrGG and presenting for surgical resection were recruited to this IRB-approved study following informed consent. A summary of patient population characteristics is provided in Table 1. The median age at recruitment was 34 years, with a range of 19-72 (60% male; 40% female).

Anatomical segmentation and ROIs
The T2 lesion (T2L) and contrast-enhancing lesion (CEL) were segmented using 3D Slicer [27]. Normal-appearing white matter (NAWM) ROIs were generated by segmenting pre-contrast T1-weighted images using the FSL FAST algorithm [28] and subtracting the T2L. Volumes of each ROI were determined along with the spectroscopically-defined region of CNI T2L > 2 in the T2L. MRSI voxels containing ≥ 50% T2L or ≥ 75% NAWM volumes were classified according to these respective ROIs.

Image-guided tissue sampling
Based on pre-surgical MRI parameters, suspected tumor within the T2 lesion was designated for intra-operative tissue sampling. Regions with low apparent diffusion coefficients (ADC), high CNI (> 2; metabolic lesion), and/or elevated cerebral blood volume (CBV) were defined as 5-mm spherical targets using BrainLab (BrainLab Inc., Munich, Germany) surgical navigation software [29]. Additionally, targets had been chosen according to surgical accessibility and heterogeneity criteria that stipulated spatially distributed sampling. Surgeons located and removed these imageguided targets during surgery at their discretion, and the LPS-coordinates for the actual site of tissue removal were recorded. Tissue was acquired early in the surgery to minimize potential tissue shift that would impact registration with imaging. Upon excision, samples were immediately fixed in 10% zinc formalin, dehydrated by graded ethanols, and embedded in Paraplast Plus wax (McCormick Scientific LLC, St. Louis, MO) using standardized immunohistochemistry protocols.

IHC parameters
A single pathologist performed immunohistochemical analysis of the image-guided tissue samples. While CNS WHO grading [7] characterized the overall tumor, each image-guided tissue sample was assessed according to tumor cellularity, measured by the mean number of cells within a 200 × field-of-view (0.776 mM 2 ), and proliferation (MIB-1 index), as determined by immunostaining with Ki-67 (clone MIB-1; Ventana, 790-4286). IDH mutation (IDH+) was assessed primarily via immunostaining with an anti-IDH1 R132H antibody (Dianova, H09), and in some instances by sequencing targeted to IDH1/2 mutational variants, allowing IDH-wildtype tumor to also be identified. Fluorescence in situ hybridization (FISH) analysis was used to assess 1p/19q co-deletion and immunostaining was performed for p53 mutation. Strong nuclear staining for p53 in ≥ 10% of tumor cells was used as a surrogate for TP53 mutation [30].

Statistical analysis
Wilcoxon rank sum tests in R were used to compare MRSI and volumetric parameters from IDH+ patients, according to ROIs, glioma grade, histological subtype, and status of pathological mutation (p53). Correlations were performed in R among continuous metabolite and pathology parameters (Pearson), and ordinal pathology parameters (Kendall-Tau). The influence of metabolite parameters on progression-free survival (PFS) was evaluated with Cox proportional hazards models, and PFS was compared between subgroups using the log-rank test. Due to the exploratory nature of this study, correction for multiple comparisons was not made.

WHO 2021 tumor characterization
A total of 108 image-guided tissue samples were acquired from 45 patients undergoing tumor resection as part of standard of care at UCSF. With respect to radiological presentation, the IDH+ patients displayed mostly non-enhancing tumors (36/42; 86%) on anatomical MRI, with mean ± standard deviation T2L volumes of 48.3 ± 36.9 cm 3 and CEL volumes (n = 6) ranging 0.14-2.58 cm 3 .

In vivo brain metabolism
Example pre-surgical MRSI data from a patient presenting with newly diagnosed IDH+ grade 2 astrocytoma are shown in Fig. 1. A CNI map overlaid on a FLAIR image highlights the metabolic abnormality within the T2 lesion (Fig. 1A), as illustrated by the corresponding spectra (Fig. 1B). Data reconstructed at the location of surgical tissue sampling indicated by the black ring show the 2HG resonance alongside other common brain metabolites (Fig. 1C). Based on LCModel quantification, the fitted contributions of individual metabolites were determined in the region of 2HG, with the red-dotted line depicting the specific chemical shift of 2HG at 2.26 ppm (Fig. 1D). The map of 2HG/tCr (Fig. 1E) demonstrates the localization of the oncometabolite within the T2L; whereas MI/tCho (Fig. 1F) increases outside of the lesion.

Spectroscopic and pathologic correlations
Tumor was verified by histopathology in 75/108 imageguided samples, 28 of which displayed reconstructed 2HG spectra at the site of sampling with CRLB < 25 (17 patients).

PFS analysis
The median PFS among all IDH+ patients was 3.8 years (95% CI 2.6-4.9 years; 17 censored), and 4.0 years (95% CI 2.7-5.3 years) in the newly diagnosed cohort (Supplementary Fig. 2A; 15 censored). Tumor grade, histological subtype and recurrence status were not significantly correlated with PFS. Cox proportional hazard models indicated that higher levels of 2HG/tCr (p = 0.0007, HR 5.594) and Glu/tCr (p = 0.006, HR 32.567), along with larger volumes of T2L (p = 0.02, HR 1.014) and non-enhancing lesion (p = 0.03, HR 1.014), were predictive of worse PFS in newly diagnosed patients (Supplementary Table 1). Higher levels of 2HG/tCr (≥ 0.905, median value; p = 0.02; Fig. S2B) and higher levels of Glu/tCr (≥ 0.945, median value; log-rank test, p = 0.02; Fig. S2C) were associated with a significant reduction in PFS (Table S1). Based on Cox regression analysis, there was no significant difference between populations empirically thresholded by 2HG/tCr or Glu/tCr levels with respect to extent of resection or treatment post surgery (Supplementary  Tables 2 and 3).  1 2HG-specific MRSI. Spectral data acquired prior to surgery from a patient with newly diagnosed IDH-mutant grade 2 astrocytoma. A CNI map overlaid on a FLAIR image (A) and corresponding spectra (B) illustrate elevated levels of choline species associated with cell turnover in the lesion; the black ring in the T2 lesion indicates where tissue was surgically sampled. Spectra reconstructed (C) and quantified (D) at the site of tissue sampling highlight the 2HG resonance among other common brain metabolites. The map of 2HG/ tCr (E) shows the localization of 2HG within the T2L, while the map of MI/tCho (F) is conversely elevated in normal-appearing tissue

Discussion
This MRSI study demonstrated the detection and quantification of 2HG in patients with LrGG, who underwent image-guided tissue sampling during surgery to assess regional histopathology. Among IDH+ LrGG patients, the full complement of metabolite data derived from 3-D MRSI helped to noninvasively differentiate tumor with respect to pathological grade, histological subtype (1p/19q codeletion), p53 expression, and NAWM. Additionally, levels of 2HG quantified from data that was reconstructed at tissue sampling locations correlated with other metabolites and cellularity. Based on empirical thresholding of lesion-wide 2HG levels, it was possible to distinguish a subpopulation of IDH+ patients that manifested adverse PFS associated with elevations of the oncometabolite.
Using a clinically feasible MRS strategy for detecting 2HG with enhanced 3-D coverage and 1-cm 3 resolution (t acq = 8.2 min.), this study achieved 91% apparent accuracy in predicting IDH mutational status, which is in line with prior literature detailing technical reliability [16,17]. However, this assessment of diagnostic accuracy must be contextualized by the considerable class imbalance created by the 3 IDH− patients relative to the larger overall population and the absence of a validation cohort to enrich the number of IDH-patients.
While a majority (93%) of patients expressing mutant IDH displayed 2HG-positive quantification, 1/3(33%) patients who lacked immunohistochemical and sequencing evidence of IDH mutation nevertheless indicated 2HG presence. This discordant case was evaluated by IHC for the most common IDH1 R132H -mutant enzyme, which accounts for approximately 90% of identified IDH1 mutations [1], and also sequenced for the range of IDH1 R132/R100 and rarer IDH2 R172/R140 mutational variants [31]. As a matter of technical limitation, such false identification of 2HG can arise with LCModel quantification of background metabolites GABA, Glu and Gln, as well as lipids from necrotic lesions [32,33], resonating in the H4,H4'-proton region of 2HG around 2.26 ppm. Other challenges to absolute detection can stem from partial volume effects in small lesions [20] and relatedly low 2HG concentrations (< 1 mM) [18] (here indicated by relatively higher tNAA levels), coupled with the relatively high spectroscopic spatial resolution that limited 2HG signal most noticeably from data reconstructed at tissue sampling sites (28/75). Because the study population was overwhelmingly comprised of patients with IDH+ LrGG, it was difficult to assess the broader metabolic implications of IDH mutations [10] relative to IDH-lesions. Earlier investigations have demonstrated glutamate reduction in IDH-mutagenized cell lines [34], surgically-derived IDH+ tumor samples [35,36], and high-grade IDH+ lesions (7 T MRS) [37] compared to nonmutated counterparts, perhaps as a consequence of conversion to 2HG-precursor alpha-ketoglutarate via glutamate dehydrogenase [36,38]. Some of the same studies additionally reported increased glutamine levels in IDH+ versus IDH-lesions, which may represent heightened metabolic dependency in the mutated phenotype [34,36,37]. Although  Table 2 primarily treating IDH+ lesions here, the characteristically reduced glutamate and elevated glutamine levels in tumor versus NAWM follow similar trends.
Besides 2HG production and apparent alterations to glutamine/glutamate metabolism, the IDH+ T2Ls displayed a variety of features distinguishing them from NAWM. The elevation in choline species, together with reduced levels of tNAA, was consistent with established literature characterizing increased cell turnover and reduced neuronal function in glioma, respectively [39,40]. Several studies have suggested that increased MI within the T2L versus NAWM can be understood as the result of heightened glial cellularity, particularly in astrocytomas [41]. As a known marker of malignant tumor, the elevation in Gly, which was fit in combination with MI due to spectral overlap, represents an expected finding [42]. The ratio of MI/tCho, previously evaluated as a marker of gliosis versus recurrent tumor [43], was shown to be reduced in the lesion owing to the excess choline content.
Despite lacking practical utility in discerning tumor lesions from NAWM, the ratio of MI/tCho was shown to function as a broad indicator of inter-tumoral malignancy when performing pairwise comparisons of molecular/pathological glioma classifications. A major caveat to this finding, however, was that aggressive tumors, which demonstrated an inverse relationship with MI/tCho, tended to be astrocytic (1p/19q-wildtype) and of higher grade. Higher values of MI/ tCho, for example, provided the best means of discriminating grade 2 versus 3 IDH+ glioma, with prior research showing effective stratification of grades 2-4 [44]. This ratio was likewise found to be higher in the least aggressive forms of IDH+ glioma across histological and molecular sub-types: e.g., oligodendroglial (1p/19q co-deleted) and p53-wildtype. Volumetric trends showing smaller T2L and CNI T2L > 2 regions in p53-wildtype and oligodendroglial tumors also distinguished less aggressive gliomas.
2HG correlations within surgically sampled regions supported a conventional understanding of the tumor microenvironment. While showing threshold-level significance, the relationship observed between 2HG and cellularity comports with the notion that tumor cell density influences 2HG productive capacity, as evidenced by other pre-surgical studies [12,13,18,20] and those employing ADC as a surrogate for cellularity [20,45]. Image-guided samples of tumors displayed a correlation between pathological measurement of cellular proliferation via MIB-1 and the spectroscopic marker of CNI, consistent with prior research [46]. Although 2HG was not found to be associated with the MIB-1 index, metabolite correlations with 2HG highlighted the abovementioned elevation of total choline and concomitant reduction in tNAA, which have historically been used as non-invasive measures of tumor malignancy. Despite the challenges that image-guided studies pose in terms of tissue shift and image-tissue misregistration, the technical implementation of 3-D MRSI coverage demonstrated in this study increased the opportunity for evaluating in vivo correlates of sampled tissue.
Of particular importance in this study were the findings that adverse PFS could most reliably be predicted by increasing levels of 2HG and glutamate, the latter having previously been associated with malignant progression [35]. These results suggest that spectroscopic measures of 2HG and other metabolites provide information that extends beyond the mere identification of IDH mutation by IHC and genomic analyses. Moreover, empirically thresholding levels of 2HG in IDH+ patients enabled a practical means of discriminating subpopulations with disparate PFS, showing the potential utility of 2HG as a prognostic marker. With the current challenges in monitoring response to treatment across a range of standard-of-care and experimental therapies, several promising studies have already sought to leverage this utility by capturing longitudinal changes in 2HG associated with disease progression and radiochemotherapy using MRS [18][19][20]. The collective advances being made in 2HG imaging lend support to the emergent clinical value of in vivo spectroscopy and its potential to enhance existing MR protocols for patients with LrGG.

Conclusion
In vivo 2HG detection can be achieved on clinical scanners and may have prognostic value based on the relationship between quantified levels and PFS. Spectroscopic strategies for resolving 2HG benefit from the technical implementation of 3-D MRSI that extends volumetric coverage.