Research ArticleBrain Imaging

Detection of 2-Hydroxyglutarate in IDH-Mutated Glioma Patients by In Vivo Spectral-Editing and 2D Correlation Magnetic Resonance Spectroscopy

See allHide authors and affiliations

Science Translational Medicine  11 Jan 2012:
Vol. 4, Issue 116, pp. 116ra4
DOI: 10.1126/scitranslmed.3002693

Abstract

Mutations in the gene isocitrate dehydrogenase 1 (IDH1) are present in up to 86% of grade II and III gliomas and secondary glioblastoma. Arginine 132 (R132) mutations in the enzyme IDH1 result in excess production of the metabolite 2-hydroxyglutarate (2HG), which could be used as a biomarker for this subset of gliomas. Here, we use optimized in vivo spectral-editing and two-dimensional (2D) correlation magnetic resonance spectroscopy (MRS) methods to unambiguously detect 2HG noninvasively in glioma patients with IDH1 mutations. By comparison, fitting of conventional 1D MR spectra can provide false-positive readouts owing to spectral overlap of 2HG and chemically similar brain metabolites, such as glutamate and glutamine. 2HG was also detected using 2D high-resolution magic angle spinning MRS performed ex vivo on a separate set of glioma biopsy samples. 2HG detection by in vivo or ex vivo MRS enabled detailed molecular characterization of a clinically important subset of human gliomas. This has implications for diagnosis as well as monitoring of treatments targeting mutated IDH1.

Introduction

Isocitrate dehydrogenase 1 (IDH1) is an intracellular enzyme that catalyzes the oxidative decarboxylation of isocitrate to α-ketoglutarate in the cytoplasm and in peroxisomes. Recent genomic studies have identified heterozygous point mutations in arginine 132 (R132) of the IDH1 enzyme (1, 2). These mutations result in a neomorphic activity leading to overproduction and accumulation of the R (also d) enantiomer of the metabolite 2-hydroxyglutarate (2HG) in 68 to 86% of grade II to III astrocytic and oligodendroglial tumors, as well as grade IV secondary glioblastoma, having higher frequency in young patients (35). Glioma patients with mutations in the gene IDH1 have a greater 5-year survival rate than patients with wild-type IDH1 gliomas (93% versus 51%) when correcting for age (3), suggesting that IDH1 mutations represent a clinically distinct subset of patients. In addition to glioma, mutations in IDH1 have also been found in patients with acute myelogenous leukemia and various other tumors, but at lower frequency than in glioma (6).

The full impact of the R132 mutation is not yet fully understood, but a major consequence of mutating this residue in IDH1 is a gain-of-function enzymatic activity favoring reduction of α-ketoglutarate to 2HG (7). This neomorphic activity leads to the accumulation of 2HG, a metabolite usually present in low levels in vivo as an error product of normal metabolism. Analogous mutations in the mitochondrial IDH2 isoform also result in 2HG production, but IDH2 mutations are found less frequently than IDH1 in various tumors, including gliomas (4).

2HG is a small biomolecule that has been shown ex vivo to identify IDH1/2-mutant tumors in humans (8). In transfected U87MG glioblastoma cell cultures, the intracellular concentration of 2HG can increase more than 100-fold (7), up to 5 to 35 μmol/g (or 5 to 35 mM, assuming tissue density of 1.05 g/ml), after introduction of R132-mutated IDH1 in the U87MG genome. Accumulation of 2HG to similar levels as in U87MG cell cultures was measured in human glioma biopsy samples with IDH1R132 mutations (7, 9). This high concentration of 2HG (5 to 35 mM) is suitable for detection by in vivo magnetic resonance spectroscopy (MRS). Because the sensitivity threshold of in vivo MRS is roughly 1 mM, 2HG is not expected to be visible under normal conditions, but 2HG might become measurable upon local accumulation owing to IDH mutation. Thus, the presence or absence of 2HG in the MR spectrum of glioma patients could effectively genotype tumors as being positive or negative for IDH1 or IDH2 mutations.

The presence of the S (also l) enantiomer of 2HG (l-2HG) has been suggested using MRS in vivo in patients with hydroxyglutaric aciduria (10, 11). The detection challenge arises from the fact that the 2HG spectrum is largely overlapping with glutamate (Glu) and glutamine (Gln)—both abundant brain metabolites that have a similar five-spin system. Peaks in the region of 2.6 to 2.4 ppm (parts per million) that were previously indicated (10, 11) for l-2HG are shared with Glu, Gln, and also by N-acetyl-l-aspartate (NAA). With the limited in vivo spectral resolution (0.1 ppm) present in most clinical settings, these overlapping species are not easily resolved using conventional one-dimensional (1D) MRS, especially if spectral fitting was not used, as in these earlier reports (10, 11). Spectral fitting programs (12, 13) try to model the in vivo MR spectrum as a combination of individual spectra (basis set) from all detectable metabolites. This approach might fail for some metabolites at clinically available fields when there is severe overlap, as it is known for γ-aminobutyric acid (GABA) (14), or as we show here for 2HG.

2D correlation spectroscopy (COSY) (15) can potentially differentiate the overlapping metabolite spectra, because correlating two chemical shifts of coupled spins creates specific patterns of signals (cross peaks) for each metabolite that are better separated in the plane of the 2D spectrum than single spectral lines in a 1D spectrum. The 2D COSY exploits the idea that there is less likelihood for two metabolites to have two identical shifts, even if they might share a common chemical shift in the 1D spectrum. In particular, the cross peaks involving Hα protons of 2HG appear in a region of the 2D COSY spectra where no other metabolite is found in healthy tissue or tumors without IDH mutations. Hence, although in 1D spectra the signals of 2HG appear in a region where other metabolites normally contribute, in 2D COSY spectra the cross peaks involving Hα protons of 2HG can be uniquely identified. Alternatively, spectral editing of 1D MRS, such as J-difference spectroscopy (14), can be tuned to detect a specific metabolite by removing the contribution of unwanted overlapping metabolites. The spectral-editing experiment can be easier to run on clinical scanners but offers limited metabolite information, whereas, on the other hand, the 2D COSY retains the full spectral information at the expense of complexity of the experiment.

Here, we show that 2HG can be detected in glioma patients using an optimized in vivo adiabatic 2D COSY method, developed previously for studying brain metabolism (16), or by spectral-editing MRS. We also find that fitting conventional 1D spectra might provide false-positive results. In vivo measurements were compared with ex vivo high-resolution magic angle spinning (HR-MAS) 2D MRS and liquid chromatography–mass spectrometry (LC-MS) of glioma biopsy samples. Results from brain phantoms, two glioma patients harboring the IDH1R132 mutation, and eight control cases, including primary glioblastoma (n = 4) and healthy volunteers (n = 4) with wild-type IDH1, demonstrate that noninvasive detection of 2HG using 1D spectral-editing and 2D correlation MRS is feasible and may allow stratification of patients on the basis of IDH1 mutation.

Results

Spectroscopic detection of 2HG in phantoms

We performed phantom experiments at 3 T on clinical scanners to establish that 2HG can be distinguished from other metabolites by localized 2D correlation MRS as well as localized spectral-editing 1D MRS. 2HG was added to a phantom containing a mixture of brain metabolites, and a recently developed 2D LASER-COSY sequence (16) with improved in vivo performance based on localized adiabatic selective refocusing (LASER) was used as described in Materials and Methods. An adiabatic spectral-editing sequence (MEGA-LASER) was newly designed here specifically for the purpose of 2HG detection (Materials and Methods and figs. S1 and S3). A series of phantoms with a range of 2HG concentrations expected to be present in IDH1-mutant tumors were also investigated to test the sensitivity limit of MRS. Assignments of 2HG (17) and other metabolites, such as myo-inositol (Myo), choline (Cho), NAA, Glu, Gln, and GABA, were made (Fig. 1) according to published literature values (18).

Fig. 1

2D LASER-COSY and 1D MEGA-LASER spectra from brain phantoms at 3 T, with 3 × 3 × 3 cm3 voxels used in all measurements. (A) Overlay of 2D LASER-COSY spectra from a phantom containing a mixture of normal brain metabolites (red contours) and a phantom where 2HG was added to the mixture of normal brain metabolites (blue contours). The Hα-Hβ cross peak of 2HG is at 4.02/1.91 (δ21) ppm. (B) Overlay of 1D MEGA-LASER from the same phantoms. The position of the Hα peak of 2HG at 4.02 ppm lines with the cross peak in the 2D spectrum above (dashed line). a.u., arbitrary units. (C) Intensity of 2HG signal in 2D LASER-COSY and 1D MEGA-LASER at different 2HG concentrations. Error bars represent 1 SD of two independent measures, with signal intensity normalized (Inorm) to the maximum intensity. Other metabolites shown: choline (Cho), γ-aminobutyric acid (GABA), glutamate (Glu), lactate (Lac), myo-inositol (Myo), and N-acetyl-l-aspartate (NAA).

Figure 1A shows the overlay of 2D LASER-COSY spectra recorded in a phantom containing a mixture of 2HG and brain metabolites (blue contours), and a phantom that contains only normal brain metabolites (red contours). The normal brain metabolites are at physiological concentrations in both phantoms. The Hα-Hβ cross peak of 2HG located at 4.02/1.91 (δ21) ppm is well separated from other metabolites, including the chemically similar metabolite Glu, with Hα-Hβ correlation located at 3.75/2.12 (δ21) ppm. Detailed information of the overlap between 2HG and other metabolites can be gleaned from spectra simulations (fig. S2). The strongly coupled five-spin system of 2HG, Glu, and Gln is very similar, and a large overlap is observed in the 2.6-to 2.0-ppm region for Hβ and Hγ protons. Additionally, GABA overlaps 2HG between 2.0 to 1.8 ppm and 2.4 to 2.2 ppm. As expected from the chemical structure, the largest separation between 2HG, Glu, and Gln is noticed for Hα protons owing to attached hydroxyl and amino moieties at Cα on 2HG and Glu/Gln, respectively. This Hα separation can be exploited in spectral-editing MRS.

Figure 1B shows the edited 1D spectra obtained with the MEGA-LASER sequence on the same phantom as Fig. 1A. The Hα multiplet signal of 2HG at 4.02 ppm (Fig. 1B, blue) is aligned with the 2HG cross peak from the 2D LASER-COSY spectrum. The signal at 4.02 ppm is missing in the brain phantom that does not contain 2HG (Fig. 1B, red). In addition to 2HG, the multiplet signals of Glu at 3.75 ppm and GABA at 3.01 ppm are co-edited, and their multiplets can be better observed in the inset of Fig. 1B. By comparison, in conventional 1D spectra obtained with LASER, the Hα proton of 2HG is largely overlapped by the strong Hβ peak of myo-inositol at 4.05 ppm (fig. S2). For in vivo spectroscopy, which typically has lower spectral resolution owing to susceptibility anisotropy of tissues, the Hα proton of 2HG might be obscured more even by the neighboring peaks of lactate (4.09 ppm) and both creatine and phosphocreatine (3.91 ppm).

MEGA-LASER showed excellent localization when compared to MEGA-PRESS in fig. S3, with no contamination of lipid signal from outside the voxel. The echo time (TE) of MEGA-LASER was optimized around the value of 1/2J (J, scalar coupling) for maximizing Hα signal of 2HG. The maximum was found for TE = 75 ms.

Calibration measurements were made for a series of 2HG phantoms with concentrations in the range of 0 to 16 mM (Fig. 1C). A strong correlation (R = 0.992) was found between 2D LASER-COSY cross peak volume and 2HG concentration. A sensitivity limit of 2 mM 2HG was calculated for 2D LASER-COSY, with a voxel of 27 cm3, measurement time of 12.8 min, and a minimum signal-to-noise ratio (SNR) of 5, for reliable identification of the Hα cross peaks. Similar strong correlation (R = 0.995) was found between 2HG concentration and the area of 2HG peak in spectral-editing MEGA-LASER (Fig. 1C). The SNR of 5 for 2 mM 2HG concentration and 27 cm3 voxel can be reached by MEGA-LASER with a shorter acquisition time of 5 min.

2HG detection in intact brain biopsies

To ensure that 2HG measurements were possible in human tissue, we measured brain biopsy samples (n = 10) before in vivo experiments (Table 1). Biopsies were used because they contain the full set of metabolites and tumor metabolic profiles that are hard to replicate in phantoms. 2D spectra were obtained with HR-MAS conditions at 14 T from brain biopsies representing varied pathologies and IDH1 mutation status (Fig. 2). The 1H-1H 2D TOBSY [total through-bond spectroscopy (18)] spectra of an IDH1-mutated anaplastic astrocytoma contained well-resolved and separated 2HG cross peaks involving the correlations of Hα with Hβ (4.02/1.91 ppm) and Hγ (4.02/2.24 ppm) (Fig. 2A). The 2HG cross peaks from biopsy spectrum overlaid entirely with the corresponding 2HG cross peaks of the phantom spectrum (Fig. 2A). Projections along δ1 and δ2 spectral dimensions through the 2HG cross peaks of the anaplastic astrocytoma biopsy are shown along axes of the 2D TOBSY. 2HG signals are not present in the 2D TOBSY spectra from both primary glioblastoma (Fig. 2B) and nontumor control (Fig. 2C) tissues, which were both wild-type IDH1. Notably, our 2HG findings from HR-MAS measurements are based on the single biopsy that had mutant IDH1.

Table 1

Brain biopsies from tumor or epileptic foci (n = 10) analyzed ex vivo with HR-MAS and LC-MS. Figure numbers are given for representative subjects. wt, wild-type.

View this table:
Fig. 2

HR-MAS spectra recorded at 14 T ex vivo on biopsy tissue from patients with and without IDH1 mutation. 1H-1H 2D TOBSY spectra are shown for all biopsies (the minimum contour levels were set five times the noise level). (A) For anaplastic astrocytoma biopsy tissue with IDH1R132 mutation (n = 1), the spectra are shown in green-blue contours. The phantom is shown in red-yellow. Projections along δ1 and δ2 show the 2HG cross peaks, outlined by a red rectangle. (B and C) Spectra for wild-type IDH1 patients: primary glioblastoma (B) (n = 1) and nontumor (C) (n = 1). The region where 2HG cross peaks would be expected is outlined by a red rectangle. For all 2D TOBSY brain spectra, several other metabolites can be identified. Amino acids: alanine (Ala), aspartate (Asp), histidine (His), isoleucine (Ile), leucine (Leu), lysine (Lys), proline (Pro), serine (Ser), and threonine (Thr). Membrane phospholipid–related compounds: ethanolamine (Etn), glycerol (Glr), glycerophosphocholine (GPC), glycerophosphoethanolamine (GPE), phosphocholine (PC), and phosphoethanolamine (PE). Sugars: l-fucose (lFuc) and β-glucose (bGlc). Miscellaneous: glutathione (GSH), lipids (Lip), and taurine (Tau).

In addition to 2HG, large qualitative and quantitative differences are easily observed among different biopsy samples, most notably the presence of lipids, l-fucose, and β-glucose, as well as the absence of glutathione (GSH) in glioblastoma; the increased GPC (glycerophosphocholine)–to–PC (phosphocholine) ratio in anaplastic astrocytoma compared to nontumor control biopsy; and a decreased GPC–to–PC ratio in glioblastoma compared to nontumor control biopsy (Fig. 2). Similar findings have been previously reported regarding increased lipids (19) and the presence of l-fucose in glioblastoma (20), and increased GPC in low-grade glioma versus increased PC in high-grade glioma (2123). For comparison, 1D HR-MAS spectra acquired on the anaplastic astrocytoma biopsy are shown in fig. S4. Because no tissue is destroyed during HR-MAS measurements, further assays are possible, such as histology, genomics, or LC-MS, to characterize the tumors. LC-MS was performed on the same biopsies and 2HG levels were measured to be 151.58 ng of 2HG per milligram of tissue (wet weight) for IDH1R132H anaplastic astrocytoma, 2.39 ng/mg for wild-type IDH1 glioblastoma, and 1.79 ng/mg for wild-type IDH1 nontumor control. The 2HG level in IDH1R132H anaplastic astrocytoma was 1.02 μmol/g, which is an order of magnitude above the lower sensitivity limit (0.1 μmol/g) of HR-MAS (24), whereas the wild-type IDH1 tissues had almost 100-fold less 2HG (0.01 to 0.015 μmol/g), concordant with previous results (7). The 2HG levels in wild-type IDH1 biopsies are <0.1 μmol/g (detection threshold of HR-MAS) and hence not visible in Fig. 2, B and C.

In vivo 2HG detection by spectral-editing and 2D correlation MRS

After confirming MRS detection of 2HG in biopsy tissue ex vivo, we performed MRS spectroscopy in vivo in a separate set of human subjects (n = 10) (Table 2). The results obtained on biopsies were important to identify the 2HG peaks that have the best chances to be detected in vivo and helped us select the appropriate in vivo methods. Two glioma patients (n = 2) with known IDH1 mutations (R132C and R132H), as well as two control groups, including primary glioblastoma patients without IDH1 mutations (n = 4) and healthy, nontumor volunteers (n = 4), were investigated (Table 2). Single voxels were prescribed on the basis of fluid attenuation inversion recovery (FLAIR) image abnormalities in tumor patients and magnetization prepared by rapid acquisition of gradient echo (MEMPRAGE) images in volunteers. 2D LASER-COSY, 1D MEGA-LASER, and 1D LASER spectra were acquired from the tumor patients and volunteers.

Table 2

Human subjects (n = 10) scanned with in vivo MRS. Figure numbers are given for representative subjects. wt, wild-type.

View this table:

2D LASER-COSY results from one patient with anaplastic astrocytoma—confirmed by tumor DNA sequencing to have R132C mutation of IDH1—are shown in Fig. 3A. A 27-cm3 voxel (3 × 3 × 3 cm3) was placed on the FLAIR images to include most of the solid tumor located in the splenium of the corpus callosum and the tail of the left hippocampus. The Hα-Hβ cross peak of 2HG was present in the 2D LASER-COSY spectrum at 4.02/1.91 ppm (δ21), with δ2 and δ1 projections well above the baseline noise level (Fig. 3A). Similar 2HG projections can be observed in the phantom spectrum (fig. S2A). Cross peaks of several other metabolites can be identified (Fig. 3A). Results from LCModel fitting of 1D LASER spectra are shown in fig. S5. Considering a basis set of spectra composed of the 20 most abundant metabolites in addition to 2HG, the fitting algorithm estimated the contribution of each metabolite so that the computed spectrum overlaps as best as possible with the measured spectrum (fig. S5).

Fig. 3

2D LASER-COSY spectra in vivo in human subjects at 3 T. (A) An anaplastic astrocytoma patient with IDH1R132C. The 2D LASER-COSY shows at 4.02/1.91 ppm the Hα-Hβ cross peak of 2HG. Projections along both spectral dimensions through 2HG cross peak indicate the SNR and spectral quality. The single voxel (3 × 3 × 3 cm3, red rectangle) was placed on the FLAIR images to include most of the tumor abnormality. (B) A primary glioblastoma patient (wt-IDH1). The 2D LASER-COSY does not contain any 2HG cross peak in the Hα-Hβ region outlined by the green rectangle. Projections through Glu + Gln cross peak indicate spectral quality. The single voxel (3.5 × 3.5 × 3.5 cm3, red rectangle) was placed on the FLAIR images to include most of the tumor abnormality. (C) A healthy volunteer (wt-IDH1). 2HG is not found in the Hα-Hβ region of 2D LASER-COSY outlined by the green rectangle. Projections through Glu + Gln indicate spectral quality. The single voxel (3 × 3 × 3 cm3, red rectangle) was placed on the MEMPRAGE images in the white matter of the occipital lobe, in a region similar to tumor locations from patients in (A) and (B).

An example of 2D LASER-COSY from a primary glioblastoma patient (wild-type IDH1 by tumor DNA sequencing) is shown (Fig. 3B). A slightly bigger voxel (3.5 × 3.5 × 3.5 cm3) was chosen owing to the extension of the tumor into the left occipital lobe. The 2D LASER-COSY does not contain any 2HG cross peak in the Hα region outlined by the green rectangle. Fitting methods applied to conventional 1D MRS (fig. S6) erroneously suggest the large presence of 2HG within confidence limits for goodness of fit (16% Cramer-Rao lower bounds), and that the level of 2HG is higher than NAA or GPC. These latter metabolites are both present in the 2D LASER-COSY spectrum; therefore, if the computed 1D MRS results in fig. S6 were true, the 2HG should be also visible in the 2D spectrum in Fig. 3B. This contradiction suggests that, in this case, the fitted 1D MRS result represents a false positive. Fitting programs, such as LCModel, assume that the composite spectrum can be obtained by a unique combination of individual metabolite spectra. However, this is known to fail in vivo for some metabolites because of adverse combination of lower resolution and severe overlap of weaker metabolite signals by stronger metabolite signals. The most known example is erroneous GABA measurement by fitting conventional 1D spectra (25). On the other hand, 2D LASER-COSY is more in line with the genetic analysis that showed no IDH1 mutations in this patient.

Figure 3C shows data from a healthy volunteer with wild-type IDH1. A 27-cm3 voxel was placed in the white matter of the left occipital lobe similar to the patient tumor positions. 2HG is absent, as expected, from the 2D LASER-COSY spectrum with the Hα cross peak region outlined in green. 2D spectral quality is indicated by projections through the Glu and Gln cross peak [Glu + Gln, 3.75/2.12 (δ21) ppm]. The fitting of the 1D MRS from the healthy volunteer is shown in fig. S7. The Cramer-Rao lower bound (23%) for 2HG fit is only slightly above the accepted limit (20%) for goodness of fit. However, 2HG was not expected to be found in a healthy control (see also fig. S9 where the LCModel fits 2HG with 17% Cramer-Rao lower bounds in the healthy contralateral hemisphere of the glioblastoma patient).

Results obtained with the spectral-editing 1D MEGA-LASER sequence are presented in Fig. 4. Overlay of spectra acquired from tumor and the healthy contralateral side in a secondary glioblastoma patient with IDH1R132H mutation is shown in Fig. 4A. The Hα multiplet signal of 2HG around 4.02 ppm is found only in the tumor spectrum and not on the healthy side. The multiplets of Glu and Gln (Glu + Gln), and GABA [+ macromolecules (MM)] are present in both voxels. LCModel fitting of the corresponding nonedited 1D LASER spectra showed 2HG in tumor (Cramer-Rao lower bounds 15%) (fig. S8) and also in the healthy voxel (Cramer-Rao lower bounds 17%) (fig. S9), contrary to expectations.

Fig. 4

1D MEGA-LASER spectra in vivo in human subjects at 3 T. In all subjects, two voxels (3 × 3 × 3 cm3 each) were placed in both brain hemispheres, symmetrically from the middle line. (A) A secondary glioblastoma patient with IDH1R132H mutation. (B and C) The spectra from subjects with wt-IDH1: primary glioblastoma (B) and healthy volunteer (C). MM denotes contamination of GABA signal with macromolecule signal.

Spectra from the control primary glioblastoma patient (wild-type IDH1) (Fig. 4B) and from the healthy volunteer (wild-type IDH1) (Fig. 4C) do not contain any 2HG, but they do show Glu + Gln and GABA + MM peaks. The Glu + Gln peaks in tumor voxels seem to be shifted slightly (0.01 ppm) upfield compared to healthy side spectra (Fig. 4, A and B). The shift can be caused by different pH conditions in tumors compared to healthy brain tissue, and by different Glu and Gln relative contributions. There is no shift for Glu + Gln peaks between right and left sides in the healthy volunteer (Fig. 4C). No shift was observed for the GABA + MM peaks in any subject (Fig. 4).

Quantification of 2HG from in vivo spectral-editing and 2D correlation MRS

Quantitative analysis of 2D LASER-COSY, 1D MEGA-LASER, and 1D LASER spectra was performed using the ratio of 2HG to the sum of Glu + Gln for reasons outlined in Materials and Methods. Volumes of cross peaks at 4.02/1.91 ppm for 2HG and 3.75/2.12 ppm for Glu + Gln were used to calculate the 2HG/(Glu + Gln) ratio from 2D LASER-COSY (Fig. 5A). Areas of peaks at 4.02 ppm for 2HG and 3.75 ppm for Glu + Gln were used to estimate the 2HG/(Gln + Glu) ratio from 1D MEGA-LASER (Fig. 5B). For 1D LASER, the values fitted by LCModel (Fig. 5, A and B) were used to calculate the ratio.

Fig. 5

Signal intensity ratios of 2HG to the sum of glutamate and glutamine (Glu + Gln). (A and B) Ratios are shown for all phantom and in vivo human spectra: 2D correlation MRS (LASER-COSY) (A), 1D spectral-edited MRS (MEGA-LASER) (B), and 1D conventional MRS (LASER) (A and B). Ratios are given as averages ± 1 SD (n = 2 for phantoms and IDH1R132 patients; n = 4 for wt-IDH1 subjects).

The 2HG/(Glu + Gln) ratios are plotted for phantoms (n = 2), mutant IDH1R132H glioma patients (n = 2), wild-type IDH1 glioblastoma patients (n = 4), and healthy volunteers (n = 4) (Fig. 5). In the case of phantoms, there is very good agreement between 2D correlation (LASER-COSY) MRS, 1D spectral-editing (MEGA-LASER) MRS, and 1D conventional (LASER) MRS for the 2HG/(Glu + Gln) ratio, which was close to 0.4, as expected from their respective concentrations: 2HG (3 mM), Glu (7.5 mM), and Gln (0 mM). Because of similar spin systems, the ratio of cross peak volumes was determined by their concentrations (assuming similar T2 and T1 times), without the need to correct for number of protons and buildup rates.

In the case of mutant IDH1R132 glioma patients, estimation of 2HG/(Glu + Gln) ratio showed slight differences, which, however, are not statistically significant (P = 0.28). 2D LASER-COSY and 1D MEGA-LASER found an average ratio of 1.27, whereas LCModel fitting estimated an average ratio of 1.11 (Fig. 5). In the wild-type IDH1 primary glioblastoma and nontumor controls, there was a significant difference (P = 0.03) between the 2HG/(Glu + Gln) ratios obtained by LCModel fitting (ratios 0.33 and 0.57), and 2D LASER-COSY (ratio 0.04) and 1D MEGA-LASER (ratio 0.03), respectively.

Discussion

The discovery that mutated IDH1/2 in gliomas can be correlated with survival benefit (3) has generated interest in using this mutation for diagnostic and prognostic purposes. 2HG is a metabolite that accumulates in human gliomas that harbor IDH1 mutations. Here, we preliminarily show that 2HG can be detected unambiguously and noninvasively by localized 2D correlation and 1D spectral-editing MRS in patients with mutated IDH1.

In vivo MRS detection of 2HG in gliomas has been suggested previously (7). Our results show that 2D LASER-COSY and 1D MEGA-LASER can reliably identify 2HG. The sensitivity of 2D LASER-COSY was about 2 mM (or 10 mg) for a 3 × 3 × 3 cm3 voxel, using a 13-min in vivo acquisition time and a minimum SNR of 5. The same sensitivity can be achieved by 1D MEGA-LASER in 5-min scan for the same voxel size. This is sufficient for the range (5 to 35 mM) of 2HG concentrations reported in IDH1-mutant tumors. Although the voxels used seem to be pretty large, several aspects besides maximizing sensitivity may justify this choice. First, gliomas are very infiltrative tumors with ill-defined margins, and active tumor exceeds the contours of the T1-weighted postcontrast images, which are mostly used to report tumor diameters or volumes. Second, IDH1 mutations seem to be uniformly expressed in tumors when present (26), so a large tumor volume could be included in the voxel. Finally, our method can separate or remove the contribution of normal metabolite; hence, the inclusion of healthy tissue, which we showed does not contain 2HG, does not alter 2HG estimation. Further improvements in spatial resolution and multivoxel acquisitions of 2D LASER-COSY (27) or spectral-editing MRS (28) are possible. In addition, the sensitivity of the ex vivo HR-MAS approach is 1 μM and may therefore be used as a nondestructive method for more detailed metabolite profiling in tumor samples.

Relative quantification of in vivo MRS data indicated that a 2HG/(Glu + Gln) ratio of >1 could be specific for IDH1 mutations. Moreover, existing data indicate that 2HG and Glu might be inversely proportional: a slight decrease of Glu together with a large increase of 2HG in IDH1R132H (9) compared with a slight increase of Glu (20) with virtually undetectable 2HG (7) in wild-type IDH1 gliomas. These results suggest that the 2HG/(Glu + Gln) ratio might have increased dynamic range for detecting IDH1 mutations compared with either metabolite alone.

Comparing 2D correlation and 1D spectral-editing MRS, each method has its own strengths and limitations. For example, all metabolites are preserved and identified by two well-defined chemical shifts in 2D COSY, whereas in 1D spectral editing, there is only one well-defined chemical shift in addition to a range of possible chemical shifts given by the bandwidth of the selective pulse (fig. S1). Conversely, spectral-editing experiments are easier to run and may require shorter scan times or smaller voxel sizes to detect the same concentration.

In addition to using 2HG as a biomarker, there is mounting interest in deciphering the biological mechanisms that link IDH mutations, 2HG production, and tumorigenesis. 2HG might act as an oncometabolite by competitive inhibition of α-ketoglutarate–dependent dioxygenases (29). This includes inhibition of histone demethylases and 5-methlycytosine hydroxylases, leading to genome-wide alterations in histone and DNA methylation, as well as inhibition of hydroxylases resulting in up-regulation of hypoxia-inducible factor 1 (HIF-1) (30). Hence, a large interest exists from pharmacological companies and research groups to develop inhibitors of mutated IDH1R132. The ability to objectively and noninvasively follow the effects of these compounds in animals and patients is a prerequisite for successful drug development.

The importance of a reliable method for a novel noninvasive method of detecting 2HG in vivo is underscored by the fact that no report exists about increased d-2HG in the blood, cerebrospinal fluid, or urine of glioma patients with IDH1 mutations. This situation is different from hydroxyglutaric aciduria metabolic disorders, which show high levels of l-2HG in body fluids. Therefore, other than tumor biopsy, no assay currently exists to probe 2HG in IDH-mutated gliomas. Although a biopsy might be necessary for the initial diagnosis, multiple serial brain biopsies are generally not feasible. Moreover, MRS methods could map 2HG distribution and identify 2HG hot spots, guiding biopsy procedures to increase the chances for correct typing of the tumor, because it is known that biopsy based on conventional computed tomography (CT) or magnetic resonance imaging (MRI) is suboptimal and subject to undergrading (26).

The in vivo 2D correlation and 1D spectral-editing MRS methods that we demonstrated can be repeated noninvasively, without harmful effects to patients, and might facilitate preclinical or clinical studies of new therapies, as well as assist with initial diagnostic workup. With further validation in humans, this approach could even allow molecular typing of IDH-mutant tumor using MRI investigations, which are already included in most patients’ diagnostic workup, resulting in cost-effective and rapid genotyping of IDH mutations.

Materials and Methods

Biopsy collection for HR-MAS and LC-MS

Biopsies (n = 10, Table 1) were collected at the time of surgery, snap-frozen in liquid nitrogen, and subsequently stored in a freezer at −80°C until HR-MAS measurement. Informed consent was obtained before surgery for biopsy collection. Biopsies were obtained from seven glioma patients: five primary glioblastoma (wild-type IDH1) and two anaplastic astrocytoma (one patient with IDH1R132H and one patient with wild-type IDH1). In addition, nontumor healthy control biopsies were obtained from three patients who had been surgically treated for epilepsy.

Selection of human subjects

Patients and healthy volunteers listed in Table 2 were scanned with informed consent approved by the Internal Review Board at our institution. Patients were diagnosed as part of standard diagnostic practice by neuropathologists who examined formalin-fixed, paraffin-embedded samples from the subjects that had been stained with hematoxylin and eosin (H&E). In total, 10 subjects (2 mutant IDH1R132 glioma patients, 4 wild-type IDH1 glioma patients, and 4 wild-type IDH1 healthy volunteers) were scanned with in vivo MRS.

Acquisition of in vivo MR spectroscopy

All in vivo MR scans were performed on 3 T Tim Trio scanners (Siemens) with a head 32-channel phased array for receive and body radio-frequency coil for transmit. Single voxel spectroscopy was performed with recently optimized 1D LASER (31) and the 2D LASER-COSY (16) sequences. In addition, a newly designed 1D MEGA-LASER was used for spectral editing (fig. S1). The same LASER module was used for localization in all sequences because of sharp excitation margins, minimal chemical shift displacement error, reduced lineshape modulation, insensitivity to B1 inhomogeneity or flip-angle errors. Low-power, gradient offset independent adiabaticity wurst modulated [GOIA-W(16,4) (31)] pulses were used with 3.5-ms duration, 20-kHz bandwidth, and 0.817-kHz maximum B1 field amplitude. Typical voxel sizes were 27 cm3 (3 × 3 × 3 cm3) or 42.8 cm3 (3.5 × 3.5 × 3.5 cm3), in the case of large tumors. A repetition time (TR) of 1.5 s was used for all acquisitions.

For 1D LASER and 2D LASER-COSY, a TE of 45 ms was used. 1D LASER spectra were collected with 128 averages (acquisition time of 3.2 min), and the 2D LASER-COSY spectra were acquired with 64 t1 increments (10-ppm δ1 spectral window), 8 averages per t1 increment, and 4 dummy scans for the first t1 (acquisition time of 12.8 min). The δ2 directly acquired spectral dimension was set to 1.25 kHz (~10 ppm), and the free induction decay (FID) had 512 points in all experiments.

The 1D MEGA-LASER spectra were acquired with TE of 75 ms, and 200 averages were collected (acquisition time of 5 min). In all sequences, water suppression was performed with WET scheme (32). Automatic shimming of the single voxels was performed with FASTESTMAP (33) to ensure linewidths of 6 to 12 Hz in human subjects. Anatomical MR images were collected to guide the position of MRS voxels. For patients, the preferred modality was axial fluid-attenuated inversion recovery (FLAIR) acquired with 10-s TR, 70-ms TE, 5-mm slice thickness (1-mm gap), 0.6 × 0.45 mm2 in-plane resolution, 23 slices, and 384 × 512 matrix (imaging time, 3.03 min).

For healthy volunteers, a multiecho MEMPRAGE (34) volumetric acquisition was performed, with 1-mm isotropic voxels, TR = 2.53 s, TE1/TE2/TE2/TE3/TE4 = 1.64/3.5/5.36/7.22 ms, inversion time TI = 1.2 s (imaging time, 6.1 min). Voxels on healthy volunteers were placed in similar regions as observed on patients to match coil sensitivity profile and regional metabolic differences.

Processing, analysis, and quantification of in vivo MR spectroscopy

Raw data were exported from the Siemens scanners for subsequent processing and analysis. The 1D LASER data (FID) were processed and quantified with LCModel (12) using a GAMMA-simulated basis set for LASER (Supplementary Methods). Fourier transform (FT), phase correction, and baseline correction were performed as part of the LCModel processing. For 1D MEGA-LASER data, the FT, phase correction, baseline correction, and line fitting were done in jMRUI (13). For 2D LASER-COSY, the FIDs of all 64 t1 increments were imported in Matlab (The MathWorks). Processing steps included (i) FT along t2, (ii) linear prediction forward to 128 points in t1 using the ITMPM method (35), (iii) FT along t1, and (iv) square-sine window function in both δ1 and δ2 dimensions to improve cross peaks and reduce diagonal ringing and baseline distortion. The 2D spectra were displayed as contour levels in magnitude mode, with the first contour level chosen five times the floor noise level as estimated from SD of noise floor in a signal-free spectral region (0.5 to 0 ppm/0.5 to 0 ppm, δ12).

A minimum SNR of 5 was considered for reliable identification of cross peaks from the noise. This was decided on the basis of the series of 2HG phantoms. At 1 mM, the Hα cross peaks had an SNR of ~2.5, which was considered insufficient to distinguish them from noise. Metabolites were assigned on the basis of the literature (17, 18, 36) values for their nuclear magnetic resonance (NMR) parameters, and cross peak volumes were integrated in Matlab. For quantification, the 2HG/(Glu + Gln) ratio was chosen for the following reasons: (i) 2HG, Glu, and Gln have a similar five-spin system; hence, the buildup of their COSY cross peaks and spectral-edited peaks is similar; (ii) Glu and Gln are largely present in both tumors and healthy brain, yielding clearly resolved cross peaks and spectral-edited peaks; (iii) the absolute quantification based on internal water cannot be used in tumors where water content varies largely; and (iv) 2HG quantification relative to Glu and Gln is preferred over quantification relative to creatine (37), because creatine does not have cross peaks or spectral-edited peaks, and may vary with disease. Thus, 2HG/(Glu + Gln) ratio enabled direct comparison of the three in vivo MRS methods. Additional details are given in the Supplementary Material.

Supplementary Material

www.sciencetranslationalmedicine.org/cgi/content/full/4/116/116ra4/DC1

Materials and Methods

Fig. S1. Pulse sequence diagram for the 1D MEGA-LASER spectral editing experiment.

Fig. S2. Phantom experiments and simulations for 2D LASER-COSY and 1D LASER at 3 T.

Fig. S3. Optimization of the 1D MEGA-LASER spectral editing on phantoms at 3 T.

Fig. S4. 1D HR-MAS spectra recorded at 14 T and 3-kHz MAS on a biopsy sample from one patient with R132H IDH1 anaplastic astrocytoma.

Fig. S5. LCModel fitting of the 1D LASER spectrum from the R132C IDH1 anaplastic astrocytoma patient.

Fig. S6. LCModel fitting of the 1D LASER spectrum from a wild-type IDH1 primary glioblastoma patient.

Fig. S7. LCModel fitting of the 1D LASER spectrum from a wild-type IDH1 healthy volunteer.

Fig. S8. LCModel fitting of the 1D LASER spectrum from the tumor voxel of the R132H IDH1 secondary glioblastoma patient.

Fig. S9. LCModel fitting of the 1D LASER spectrum from the healthy side voxel of the R132H IDH1 secondary glioblastoma patient.

References

References and Notes

  1. Acknowledgments: We would like to acknowledge J. A. Iafrate and D. N. Louis from the Department of Pathology of Massachusetts General Hospital for assistance with SNaPshot analysis for IDH mutation and useful comments on our results. We also acknowledge P. M. Black from the Department of Neurosurgery of Brigham and Women’s Hospital for access to biopsies that were analyzed by ex vivo HR-MAS spectroscopy. We thank M. Malgorzata and M. Garwood from Center for Magnetic Resonance Research at University of Minnesota for help with LCModel basis set for LASER excitation. Funding: This work was funded by grants from NIH (R01 1200-206456, S10RR013026, S10RR021110, and S10RR023401). O.C.A. was also supported by a KL2 Medical Research Investigator Training (MeRIT) award from Harvard Catalyst, The Harvard Clinical and Translational Science Center (NIH Award #KL2 RR 025757). Author contributions: O.C.A. provided conceptual design, obtained measurements, analyzed the data, and drafted the manuscript. G.S.K. obtained experimental measurements. E.G. and T.B. recruited patients and provided clinical guidance and manuscript review. A.A.T. provided support for biopsy measurements and manuscript review. V.R.F. performed LC-MS measurements, literature review, and manuscript editing. M.G.V.H. provided expertise on IDH1 mutations and 2HG, literature review, and manuscript editing. A.G.S. initiated the project, funding support, and manuscript review. Competing interests: The content is solely the responsibility of the authors and does not necessarily represent the official views of Harvard Catalyst, Harvard University and its affiliated academic health care centers, the National Center for Research Resources, or the NIH. O.C.A. and A.G.S. have applied for a patent for the 2D COSY-LASER method that is used in the paper, U.S. Patent Application Serial No. 13/237,799.
View Abstract

Navigate This Article