2-hydroxyglutarate detection by magnetic resonance

内容需要下载文档才能查看 内容需要下载文档才能查看

technical reports

2-hydroxyglutarate detection by magnetic resonance spectroscopy in subjects with IDH-mutated gliomasChangho Choi1,2, Sandeep K Ganji1,2, Ralph J DeBerardinis3–5, Kimmo J Hatanpaa5–7, Dinesh Rakheja6,8, Zoltan Kovacs1, Xiao-Li Yang5,7,9, Tomoyuki Mashimo5,7,9, Jack M Raisanen5–7, Isaac Marin-Valencia3, Juan M Pascual3,10,11, Christopher J Madden5,7,12, Bruce E Mickey5,7,12, Craig R Malloy1,2,9,13, Robert M Bachoo5,7,9,10 & Elizabeth A Maher5,7,9,10

tumors2,6. Immunohistochemistry with a commercially available anti-body to the R132H mutation of IDH1 identifies approximately 93% of the mutations, but the remaining 7% of tumors carrying a different IDH1 or an IDH2 mutation require direct sequencing for detection7. As 2HG is produced by all known IDH-mutant enzymes, evaluation of 2HG abundance is an alternative indirect method for determining IDH status. The finding that 2HG is present at high levels in IDH-mutated gliomas has raised the possibility that this metabolite could be and prognostic information.(PRESS)8 and difference editing9Isocitrate dehydrogenase converts isocitrate to a-ketoglutarate (aKG) in and phantom analyses for detection of 2HG in the human brain and the cytosol (IDH1) and mitochondria (IDH2). The recent identification applied them to tumor masses in 30 adults with all grades of gliomas. of mutations in IDH1 and IDH2 among most humans with World Health Analysis of MRS data was blinded to IDH status. For each case in which Organization (WHO) grade 2 and 3 gliomas1,2 has directed attention 2HG was detected by MRS, we confirmed an IDH1 or IDH2 mutation to the role of abnormal metabolism in the pathogenesis and progres-in the tumor. Failure to detect 2HG by MRS was associated with the sion of these primary brain tumors. The mutations are confined to the detection of wild-type IDH1 and IDH2 in each case. The sensitivity and active site of the enzyme and result in a gain of function that generates specificity of the method described here and the ease with which it could 2HG (ref. 3) and induces DNA hypermethylation4,5. The abundance of be incorporated into standard magnetic resonance imaging suggests that this metabolite, normally present in vanishingly small quantities, can be 2HG detection by MRS may be an important biomarker in the clinical elevated by orders of magnitude in gliomas with IDH1 or IDH2 muta-management of these patients.tions. Intracellular concentrations on the order of several micromoles 3Mutations in isocitrate dehydrogenases 1 and 2 (IDH1 and IDH2) have been shown to be present in most World Health Organization grade 2 and grade 3 gliomas in adults. These mutations are associated with the accumulation of 2-hydroxyglutarate (2HG) in the tumor. Here we report the noninvasive detection of 2HG by proton magnetic resonance spectroscopy (MRS). We developed and optimized the pulse sequence with numerical and phantom analyses for 2HG detection, and we estimated the concentrations of 2HG using spectral fitting in the tumors of 30 subjects. Detection of 2HG correlated with mutations in IDH1 or IDH2 and with increased levels of d-2HG by mass spectrometry of the resected tumors. Noninvasive detection of 2HG may prove to be a valuable diagnostic and prognostic biomarker.Southwestern Medical Center, Dallas, Texas, USA. 3Department of Pediatrics, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 4McDermott Center for Human Growth and Development, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 5Harold C. Simmons Cancer Center, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 6Department of Pathology, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 7Annette Strauss Center for Neuro-Oncology, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 8Children’s Medical Center, Dallas, Texas, USA. 9Department of Internal Medicine, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 10Department of Neurology and Neurotherapeutics, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 11Department of Physiology, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 12Department of Neurological Surgery, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 13Veterans Affairs North Texas Health Care System, Dallas, Texas, USA. Correspondence should be addressed to C.C. (changho.choi@utsouthwestern.edu) or E.A.M. (elizabeth.maher@utsouthwestern.edu).

1Advanced Imaging Research Center, University of Texas Southwestern Medical Center, Dallas, Texas, USA. 2Department of Radiology, University of Texas p© 2012 Nature America, Inc. All rights reserved.

technical reports

Figure 1 Theoretical and

experimental spectra of 2HG.

(a) Quantum-mechanically calculated spectra of the 2HG H4 resonances, at 3 T, are plotted against TE1 and TE2 of PRESS (subecho times of the first slice– and second slice–selective 180° radiofrequency

pulses, respectively). (b) Calculated difference-edited multiplets of the 2HG H2 resonance are plotted against subecho times TE1 and TE2 of scalar difference editing. Shown for each TE1-TE2 pair are, top to bottom, E180-on (brown) and E180-off

(green) subspectra, and the difference between the two subspectra (blue). Here, E180 denotes editing 180° pulses tuned to 1.9 p.p.m. PRESS and edited spectra are all broadened to a singlet line width of 4 Hz. Spectra in a and b are scaled equally for direct comparison. Relaxation effects were not included in the calculations. (c) Calculated and phantom spectra of 2HG for PRESS and difference editing. The echo times were 97 ms and 106 ms for PRESS and editing. The concentrations of 2HG and glycine in the phantom were both 10 mM (pH = 7.0). Spectra are scaled with respect to the glycine singlet at 3.55 p.p.m.

a

1151059585TE2 (ms)

756555453525

20

30

40

50

60

70

80

2.52.0p.p.m.

c

CalculatedGlycine

PhantomGlycine

PRESS(TE1, TE2)(32,65) ms

H4H4?

H2

H3H3?

H2

H4H4?H3H3?

90100110

TE1 (ms)

Differenceediting(TE1, TE2)(26,80) ms

b

85

4.253.75p.p.m.

E180-on (A)E180-off (B)(A–B) / 2

E180-on(A)

80TE2 (ms)

E180-off(B)

© 2012 Nature America, Inc. All rights reserved.

75

70

Difference(A–B) / 2

4.03.53.02.52.0Chemical shift (p.p.m.)

4.03.53.02.52.0Chemical shift (p.p.m.)

65

25

35

45

55

65

75

85

95

TE1 (ms)

(H3 and H3?) (Supplementary Fig. 1). The 2HG resonances are all sca-lar coupled, and, consequently, the spectral pattern and signal strength vary with changing echo time of MRS sequence. A maximum 2HG sig-nal may be expected at ~2.25 p.p.m. where the H4 and H4? spins resonate proximately to each other. Because of its capability of full refocusing, in the present study we used a PRESS sequence as a major tool for 2HG measurement.

We conducted quantum mechanical simulations to search for opti-mal experimental parameters. The simulation indicated that the 2HG H4 resonances give rise to a maximum multiplet at total echo time of 90–100 ms, for which the first subecho time, TE1, is shorter than the second subecho time, TE2 (Fig. 1a). Given the large spectral distance of the H2 resonance from its weak coupling partners (H3 spins), we also measured the H2 resonance by means of difference editing. Selective 180° rotation of the H3 spins was switched on and off within a PRESS sequence in alternate scans to induce unequal H2 multiplets in sub-spectra. Subtraction between the spectra generated an edited 2HG H2 multiplet, canceling other resonances that were not affected by the edit-ing 180° pulses. The computer simulation indicated that a large edited H2 signal can be obtained using a short-echo-time set in which TE1 should be the shortest possible (Fig. 1b). We optimized the subecho times of the PRESS and difference editing sequences to (TE1, TE2) = (32, 65) ms and (26, 80) ms, respectively. We tested these optimized MRS sequences in an aqueous solution with 2HG that was synthesized in house. The spectral pattern and signal intensity of 2HG were consistent between calculation and experiment (Fig. 1c).

The optimized PRESS provided a 2HG multiplet at approximately 2.25 p.p.m. with maximum amplitude among echo times greater than 40 ms (Supplementary Fig. 2a,b). Moreover, the optimized echo time gave rise to narrowing of the multiplet and substantial reduction of 2HG signals at approximately 1.9 p.p.m. Similar signal modulation

nature medicine occurred in the glutamate multiplets, allowing 2HG to be measured with high selectivity against the background signals of adjacent reso-nances (Supplementary Fig. 2c,d). The optimized echo time is rela-tively long, so signal loss due to transverse relaxation effects may be considerable in vivo. However, given that 2HG does not have a well-defined spectral pattern at short echo times (for example, 30 ms), the 2HG signals can be better resolved at the optimized long echo time, benefiting from the suppressed complex baseline signals of mac-romolecules. The signal yield of 2HG in difference editing was low (38%) compared to that in PRESS (Fig. 1), but the editing provides a useful tool for proving 2HG elevation because the edited signal at 4.02 p.p.m. is uniquely generated via the coupling connections of 2HG. In vivo, because the difference editing uses spectral difference induced by selective 180° rotations tuned at approximately 1.9 p.p.m., the 4.15-p.p.m. resonance of the glutamate moiety of N-acetylaspartyl-glutamate11 is co-edited, but the resonance is relatively distant from the 2HG 4.02-p.p.m. resonance and thus does not interfere with 2HG editing (Supplementary Fig. 3). The lactate resonance at 4.1 p.p.m.12 is not co-edited because the coupling partners at 1.31 p.p.m. are not influenced by the editing 180° pulse.

For spectral fitting, in the present study we used model spectra that were calculated including the effects of the volume-localized radiofre-quency pulses used for in vivo measurements, allowing spectral fitting by signal patterns identical to those obtained by experiment. Calculation of spectra at numerous echo times for MRS sequence optimization was efficiently accomplished using the product operator-based transfor-mation-matrix algorithm in the quantum-mechanical simulations13–15 (Supplementary Methods). The spectral pattern of 2HG is pH depen-dent10, with large shifts noted for pH < 6 (Supplementary Fig. 4). We performed computer simulations and MRS sequence optimization for 2HG detection assuming pH ~7.0 in tumors16–18.

p

technical reports

Table 1 Correlation between 2HG detection by MRS PRESS and IDH1 and IDH2 mutational status

Histological diagnosis

Oligodendroglioma (WHO grade 2)Oligodendroglioma (WHO grade 2)Oligodendroglioma (WHO grade 2)Oligodendroglioma (WHO grade 2)Oligodendroglioma (WHO grade 2)Astrocytoma (WHO grade 2)Astrocytoma (WHO grade 3)Oligoastrocytoma (WHO grade 3)Oligodendroglioma (WHO grade 3)Oligoastrocytoma (WHO grade 3)Astrocytoma (WHO grade 3)Astrocytoma (WHO grade 3)Astrocytoma (WHO grade 3)Astrocytoma (WHO grade 3)Astrocytoma (WHO grade 3)

Secondary glioblastoma (WHO grade 4)

(mM (CRLB))2.7 (13%)3.3 (11%)2.6 (14%)1.7 (17%)3.3 (7%)4.2 (10%)2.1 (16%)3.9 (6%)8.9 (3%)3.4 (8%)2.7 (11%)5.3 (6%)2.5 (16%)2.2 (15%)ND2.1 (15%)NDNDNDNDNDNDNDNDNDNDNDNDNDND

IDH1 and IDH2 mutations by DNA sequencingIDH2 (R172K)IDH1 (R132H)IDH1 (R132C)IDH1 (R132H)IDH1 (R132C)IDH1 (R132H)IDH1 (R132H)IDH1 (R132H)IDH1 (R132H)IDH2 (R172W)IDH2 (R172G)IDH1 (R132H)IDH1 (R132C)IDH1 (R132C)None

IDH1 (R132H)NoneNoneNoneNoneNoneNoneNoneNoneNoneNoneNoneNoneNoneNone

immunohistochemistryNegativePositiveNegativePositiveNegativePositivePositivePositivePositiveNegativeNegativePositiveNegativeNegativeNegativePositiveNegativeNegativeNegativeNegativeNegativeNegativeNegativeNegativeNegativeNegativeNegativeNegativeNegativeNegative

© 2012 Nature America, Inc. All rights reserved.

Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)Glioblastoma (WHO grade 4)

The MRS measures labeled ‘ND’ (not detected) were 2HG estimates ≤0.08 mM with CRLB ≥85%. The MRS estimates of 2HG concentrations were significantly different between mutated and wild-type IDH (P = 6 × 10-8, unpaired t-test).

p2HG detected in MRS spectra from subjects with gliomas

We included 30 subjects with gliomas in the 2HG MRS analysis (Table 1). The optimized PRESS was applied to the tumor mass. A representative normal brain spectrum (Fig. 2a) showed the expected pattern of cho-line, creatine and N-acetylaspartate (NAA) without evidence of 2HG. In contrast, the classic pattern of elevated choline with decreased cre-atine and NAA was present in all glioma grades (Fig. 2b–f). A signal attributed to 2HG was discernible at 2.25 p.p.m. in the WHO grade 2 and 3 tumors (Fig. 2c–f), but not in the glioblastoma (Fig. 2b). We identified an IDH1 or IDH2 mutation in each of these cases. We analyzed the single-voxel–localized PRESS data with lin-ear combination of model (LCModel) software19, using spec-tra of 20 metabolites as basis sets, calculated incorporating the volume-localized pulses. We estimated the concentration of 2HG using the brain water signal from the voxel as reference and adjusted the relaxation effects on the observed metabolite sig-nals using published relaxation times of brain metabolites for 20–2233 T. With a 2-min scan on 2 × 2 × 2 cm areas of brain tissue, 2HG was measurable for concentrations >1.5 mM, with Cramér-Rao lower bound (CRLB) < 18% (Table 1). With the use of precisely calculated model spectra for spectral fitting, the LCModel fits reproduced the in vivo spectra closely, resulting in residuals at the noise levels that did not show considerable chemical-shift dependences.

g-Aminobutyric acid (GABA), glutamate and glutamine have reso-nances of 2.1–2.4 p.p.m. Thus, the signals are partially overlapped with the 2HG 2.25-p.p.m. signal in PRESS spectra and can interfere with 2HG estimation depending on their signal strengths. Spectral fitting with the calculated spectra enabled resolution of the metabolites with CRLB < 20% for concentrations above ~2 mM (Fig. 2). We validated the PRESS detection of 2HG using two methods. First, we compared spectral fitting outputs from a basis set with or without a 2HG signal. For spectra without measurable 2HG signals (Fig. 3a), the residuals were essentially identical between the two fitting methods. However, spectra with a noticeable signal at 2.25 p.p.m., when fitted using a basis set with-out 2HG, resulted in large residuals at 2.25 p.p.m. (Fig. 3a). For spectra with intermediate 2HG concentrations, the residuals were progressively larger with increasing 2HG estimates. This result shows that the signal at 2.25 p.p.m. is primarily attributable to 2HG without substantial interfer-ence from the neighboring resonances. Second, we used difference edit-ing to confirm the PRESS measurements of 2HG in seven subjects. When a signal at 2.25 p.p.m. was discernible in PRESS spectra, we detected an edited H2 signal at 4.02 p.p.m. (Fig. 3b). When 2HG was not mea-surable in PRESS spectra, there was no observable edited peak at 4.02 p.p.m. (Fig. 3b). This co-detection of the PRESS 2.25 p.p.m. peak and the edited 4.02 p.p.m. signal supports the idea that the signals are both attributable to 2HG. The similarity between the 2HG concentrations

nature medicine

technical reports

a

b

c

the presence of a 2HG peak in MRS is 100% correlated with the presence of a mutation in IDH1 or IDH2 and elevated concen-trations of d-2HG in the tumor. Moreover, the absence of a 2HG peak when MRS is performed in a tumor mass is 100% corre-lated with wild-type IDH1 and IDH2 and lack of accumulation of d-2HG in the tumor tissue. Thus, the ability to detect 2HG by MRS in a tumor mass is both highly sensitive and specific.

Normal brain

Grade 4 glioblastoma(wild-type IDH1 and IDH2)Grade 2 oligodendroglioma

(IDH1 mutated)

NAA

Cr

CrCho

In vivoFit

Residuals2HG (0)

GABA (1.1 ± 0.2)Glu (9.0 ± 0.2)Gln (2.4 ± 0.2)

4

3

2

1p.p.m.

4

3

2

Cho

Cr

NAA

Lac

Lip

ChoCr

NAA

Lac

2HG (0)

GABA (0.5 ± 0.2)Glu (4.6 ± 0.2)Gln (5.3 ± 0.2)

1p.p.m.

4

3

2

2HG (3.3 ± 0.2)GABA (0.4 ± 0.2)Glu (1.4 ± 0.2)Gln (2.9 ± 0.2)

1p.p.m.

d

Grade 2 oligodendroglioma

(IDH1 mutated)Cho

e

Grade 3 oligodendroglioma

(IDH1 mutated)Cho

f

Grade 3 astrocytoma(IDH1 mutated)

Lip

Development of multivoxel imaging of 2HG

We extended the optimized

ChoCrCrPRESS echo time method to GlyNAANAACrLacNAALac

multivoxel imaging of 2HG. The subject with grade 3 oligo-2HG (4.6 ± 0.3)2HG (9.1 ± 0.3)2HG (8.9 ± 0.3)dendroglioma, whose single-GABA (0.1 ± 0.1)GABA (0.2 ± 0.2)GABA (0)

voxel MRS data are shown in Glu (1.4 ± 0.2)Glu (0.5 ± 0.2)Glu (2.1 ± 0.2)

Gln (2.6 ± 0.2)Gln (2.0 ± 0.2)Gln (2.3 ± 0.2)

Figure 2e, was scanned with

4321p.p.m.4321p.p.m.4321p.p.m.21 × 1 cm resolution on a slice

1.5 cm thick that included the

Figure 2 In vivo 1H spectra and analysis. (a–f) In vivo single-voxel–localized PRESS spectra from normal brain (a) and

tumor mass (Fig. 4a). The tumors (b–f), at 3 T, are shown together with spectral fits (LCModel) and the components of 2HG, GABA, glutamate

patterns of the single voxel–and glutamine, as well as voxel positioning (2 × 2 × 2 cm3). Spectra are scaled with respect to the water signal

acquired spectra were repro-from the voxel. Vertical lines are drawn at 2.25 p.p.m. to indicate the H4 multiplet of 2HG. Shown in brackets is the estimated metabolite concentration (mM) ± s.d. Cho, choline; Cr, creatine; Glu, glutamate; Gln, glutamine; Gly, duced in spectra obtained with glycine; Lac, lactate; Lip, lipids. Scale bars, 1 cm.the multi-voxel MRS method.

The 2HG signal at 2.25 p.p.m.

estimated by PRESS and editing provides evidence that the PRESS mea-was clearly discernible in spectra from the tumor regions (Fig. 4b). surement of 2HG is valid, as the edited 2HG signal at 4.02 p.p.m. was Spectra from contralateral normal brain showed no 2HG signals at 2.25 generated without substantial interference from the scalar coupling con-p.p.m. (Fig. 4c). A map of 2HG concentrations (Fig. 4d) showed that nection between the 4.02 p.p.m. and ~1.9 p.p.m. resonances, which is a 2HG was concentrated at the center of the T2w-FLAIR hyperintensity unique feature of 2HG among known brain metabolites12.region. We estimated the 2HG concentrations using the normal brain

NAA concentration of 12 mM12 as reference, giving a 2HG concentra-Validation of MRS measures of 2HG by tissue analysistion approximately 9 mM at the center of the tumor mass, in agreement We analyzed each tumor for IDH gene status by immunohistochemistry with the 2HG estimates by single-voxel MRS. The spatial distribution for the IDH1 R132H mutation and by gene sequencing of IDH1 and pattern of choline was similar to that of 2HG in the region of T2w-IDH2 (Table 1). Of the 30 subjects studied, 15 had measurable 2HG by FLAIR hyperintensity but, as expected, was found throughout the brain, MRS, and in each case we confirmed an IDH1 (12 of 15 subjects) or an whereas 2HG showed rapid drop off in normal brain. The NAA con-IDH2 (3 of 15 subjects) mutation. The remaining 15 subjects did not centrations were low in the tumor mass, and the choline/NAA ratio in tumors was high relative to that of normal brain tissue. Because of their of IDH1 and IDH2 revealed no mutations. The MRS estimates of 2HG ability to detect 2HG in small volumes, the metabolic measures by the concentrations were significantly different between subjects with IDH multivoxel MRS method may contain reduced partial-volume effects mutations and wild-type IDH genes (unpaired t-test; P = 6 × http://wendang.chazidian.compared to the single-voxel MRS. As 2HG is unique to tumor cells, We validated these results further by measuring d-2HG and l-2HG the specificity of detection is a key advance in clinical MRS for IDH-concentrations in tumor samples by liquid chromatography–tandem mutated gliomas.mass spectrometry (LC-MS/MS) for the 13 subjects for whom suf-ficient frozen material from the initial tumor resection was available DISCUSSION

(Supplementary Table 1). Samples from the brain tissue adjacent to We have detected 2HG noninvasively by optimized MRS methods tumors were available for analysis from three subjects; thus, this tissue in subjects with gliomas and have shown concordance of 2HG levels served as relative normal controls (Supplementary Fig. 5). l-2HG and with mutations in IDH1 and IDH2, as well as accumulation of 2HG d-2HG were clearly differentiated in the spectra. Five wild-type IDH1 in tumor tissue. To our knowledge, this is currently the only direct and IDH2 glioblastoma tumors had similar concentrations of l-2HG metabolic consequence of a genetic mutation in a cancer cell that can and d-2HG. In all tumor samples, l-2HG was <1.0 nmol per mg protein. be identified through noninvasive imaging. The signal overlaps of 2HG In marked contrast, d-2HG levels in IDH1- and IDH2-mutated tumors with GABA, glutamate and glutamine, which occur in short-echo-time were 20-fold to 2,000-fold higher than those in wild-type IDH glioblas-standard data acquisitions, were overcome with multiplet narrowing tomas (Supplementary Fig. 6). Taken together, these data show that by MRS sequence optimization and spectral fitting using precisely

nature medicine p© 2012 Nature America, Inc. All rights reserved.

technical reports

external reference signal such as from a phantom24, referencing Fitting with 2HGFitting without 2HGResidualsPRESSDifference editing

(threefold magnified)(fourfold magnified)with respect to brain water sig-ChoNAANAA

nals may be a realistic means of CrWith 2HGWithout 2HGCr

ChoCho2HG2HG2HGestimating metabolite concen-NAA

trations in tumors25.

Cr

Two studies have previously reported in vivo detection of

2HG2HG

2HG in the brains of people (9.1 mM)(8.7 mM)

with 2-hydroxyglutaric acid-uria26,27. Large singlet-like signals at 2.5–2.6 p.p.m. were 2HG2HG

assigned to 2HG, although this (5.7 mM)(5.4 mM)

chemical shift assignment is not consistent with in vitro high-resolution magnetic resonance 2HG2HG

spectra of 2HG at neutral pH10. A 2HG signal at approximately 2.5 p.p.m., which is actually a

2HG2HGmultiplet, can occur only at low

pH (~2.5), as reported previ-432143212.52.02.52.043214321

ously10 and confirmed in this p.p.m.p.p.m.p.p.m.p.p.m.p.p.m.p.p.m.

work (Supplementary Fig. 4).

Figure 3 Validation of 2HG PRESS measurements. (a) LCModel fitting results (fits and residuals) of PRESS spectra

The pH measured noninvasively obtained with basis set with or without 2HG. Data are displayed in order of increasing 2HG estimates, above to below.

in a wide range of tumors ranges (b) PRESS and difference-edited spectra from four subjects are shown in pairs, together with LCModel fits and

between 6.8 and 7.2. Even using 2HG signal components. Vertical lines are drawn at 2.25 p.p.m. and 4.02 p.p.m. in the PRESS and edited spectra,

respectively.microelectrode studies, the low-est pH measured was ~6.0. An calculated basis spectra of metabolites. The methods presented here intracellular pH of ~7.0 has been reported in cancer cells16–18. The estimated metabolite concentrations using the brain water signal as chemical shifts and coupling constants, used for MRS data analysis reference in tumors, assuming an equal contribution of gray and white in the present study, were measured at pH 7.0 (ref. 10). As the pro-matter. The metabolite estimation may be valid only when the water ton NMR spectrum of 2HG is close to constant between pH 6.5–7.5 concentration is similar among regions of the brain and between nor-(Supplementary Fig. 4), the efficiency of detecting 2HG in gliomas mal brain and tumor tissues. The water concentration in tumors could should not be sensitive to tumor pH.be increased as a result of the effects of high cellularity or brain edema, A PRESS sequence used for 2HG measurement in the present study which would result in an underestimate of metabolite concentrations is commonly available in clinical magnetic resonance systems. The in the present study. Given a maximum possible water concentration field strength used here, 3 T, is becoming more commonly used in the of 55.6 M (bulk water), the true metabolite concentrations could be up academic and clinical magnetic resonance community, and the data-to 30% higher than our estimates, which were obtained using a water acquisition method could be implemented on standard hardware already concentration of 42.3 M, calculated from the published values for the in place in many magnetic resonance imaging centers. Without the need water concentrations in gray and white matter23. Although uncertain-for more specialized instrumentation or the production of expensive ties in metabolite estimates can theoretically be minimized by using an exogenous probes, the detection of 2HG by MRS is a method that could

Figure 4 Spectroscopic imaging of 2HG. (a) Multivoxel imaging spectra from a subject with a WHO grade 3 oligodendroglioma are overlaid on the T2w-FLAIR image. The grid size is 1 × 1 cm, with slice thickness 1.5 cm. The spectra are displayed between 4.1 p.p.m. and 1.8 p.p.m. (left to right). (b,c) Two representative spectra (one from the tumor and another from the

contralateral normal brain) are shown together with LCModel fits and

residuals. Mins, myoinositol. (d) The estimated concentrations of 2HG, choline and NAA in individual voxels were color coded for comparison. The NAA level in gray matter in normal brain was assumed to be 12 mM. Scale bars, 1 cm.

ab

p© 2012 Nature America, Inc. All rights reserved.

ab

Cr

Gly

Cho

d

NAA2HG

In vivoFitResiduals

[2HG] map

mM9630

Cr

[CHO] map

4

4.03.53.02.52.0p.p.m.

32

c

Cr

Mins

Cho

Cr

Glu

NAA10

[NAA] map

12

In vivoFitResiduals

840

4.03.53.02.52.0p.p.m.

nature medicine