Mapping glutamate in subcortical brain structures using high-resolution GluCEST MRI

Abstract

In vivo measurement of glutamate (Glu) in brain subcortex can elucidate the role these structures play in cognition and neuropsychiatric disorders. However, accurate quantification of Glu in subcortical regions is challenging. Recently, a novel MRI method based on the Glu chemical exchange saturation transfer (GluCEST) effect has been developed for detecting brain Glu in millimolar concentrations. Here, we use GluCEST to map Glu distributions in subcortical structures of the human brain (e.g. amgydala, hippocampus). Overall, GluCEST was ~40% higher in gray matter than in white matter. Within the subcortical gray matters, amygdala showed the highest GluCEST contrast. Utilizing MR spectroscopic data, In vivo GluCEST detection sensitivity (~0.8% per mM) in subcortical gray matters was evaluated and consistent to the previously reported values. In general, the GluCEST map approximates the Glu receptor distribution reported in previous Positron Emission Tomography (PET) studies. These findings suggest that high-resolution GluCEST MRI of subcortical brain structures may prove to be a useful tool in diagnosis brain disorders or treatment responses.

Introduction

Glutamatergic dysfunction in the brain is implicated in cognitive deficits (1-3) and neuropsychiatric disorders including schizophrenia, Alzheimer’s disease, drug abuse addiction and in Autism depression (4). Thus, the demand for techniques that quantify and localize changes in brain glutamate (Glu) levels is high. This is of particular interest in subcortical brain regions, including the amygdala, hippocampus, caudate and putamen, as dysfunction in these regions affect cognition (5,6), and emotion processing (7) in various disorders.

Proton magnetic resonance spectroscopy (1H-MRS) is one of the techniques used to assess potential disruptions in neuronal integrity and associated neurochemical dysregulations. While standard 1H-MRS at 3 T allows for sufficient sensitivity to the neurochemistry in many regions of the brain, direct quantification of specific neurotransmitters such as Glu within small subcortical regions remains limited at this field strength (8,9). MRS at ultra-high field 7 T (10) offers enhanced sensitivity due to higher Larmor frequency and the improved chemical shift dispersion, which enables better characterization of individual resonances (11).

Chemical exchange saturation transfer (CEST) MRI is a novel molecular imaging method that detects metabolites in millimolar (mM) concentration. For example, this technique has been used to quantify macromolecules based on the amide protons transfer (APT) effect (12,13), free calcium (14), starch hydrolysis and glucose mutarotation (15), liver glycogen (16), cartilage proteoglycan (17), and brain Glu with exchangeable amine protons (18). During CEST MRI, exchangeable protons are selectively irradiated by the application of a radio-frequency (RF) pulse. The exchange of saturated magnetization with bulk water protons results in reduced bulk water signal in a concentration and pH dependent manner. The CEST technique offers improved spatial resolution as compared to MR spectroscopic techniques due to the prolonged saturation process and the accumulation of an exchanging effect (18).

Previously employed APT methods (19,20) differentiate protein amide protons in gray and white matters, by showing higher APT effect in the white matter than the gray matter. For example, Dula et al. (20) demonstrated that white matter brain structures, such as splenium, have higher (~75%) APT contrasts than the gray-matter structures including caudate and putamen. Recently, we have demonstrated the feasibility of mapping Glu in healthy brain using the Glutamate amine proton CEST (GluCEST) effect (18). The notable difference between APT and GluCEST is the experimental conditions. In APT imaging, due to the low exchange rates of amide protons (~30 s⁻¹) associated with labile macromolecules and peptides, a lower saturation pulse amplitude (B₁ root mean square or B_1rms = 50 Hz (1.2 μT)) and longer duration (~6 s) were used (13) compared to GluCEST MRI (B_1rms = 155 Hz (3.6 μT) and 1 s). Using high saturation B₁, faster exchange (~2000 Hz) amine protons associated with Glu can be mapped using GluCEST method. GluCEST has been demonstrated to differentiate Glu in gray and white matters by showing higher contrast in gray matter than in white matter (18).

This study is to further investigate whether GluCEST MRI can be used to map glutamate distributions in small subcortical gray-matter structures. In addition, we compare GluCEST contrast to the ¹H-MRS data to estimate the in vivo GluCEST detection sensitivity.

Methods

All experiments were performed on a Siemens 7 T whole-body scanner with a vendor supplied 32-channel volume RF coil. The study was conducted under an approved Institutional Review Board (IRB) protocol.

GluCEST MRI

GluCEST MRI was acquired from six healthy male subjects (18-67 years old) using previously reported methods (18). T₁-weigthed anatomical 3D Magnetization Prepared RApid Gradient Echo (MPRAGE) images of whole brain (160 axial slices) were collected using parameters: repetition time (TR)/echo time (TE) = 3.2/2110 ms, inversion time (TI) = 1.5 s, in-plan resolution = 0.9 × 0.9 mm² and slice thickness = 1 mm. The 3D reformatting of the MPRAGE images was used for selecting one axial and one sagittal slice for GluCEST imaging. The GluCEST imaging parameters were: number of averages = 2, slice thickness = 5 mm, field of view = 200 × 200 mm², matrix size = 192 × 192, TR = 16 s. The saturation pulse train with 10 Hanning-windowed pulses (100 ms each, 99.8% duty cycle, B_1rms of 155 Hz or 3.6 μ;T) was followed by Fast Low-Angle SHot imaging (FLASH) readout (flip angle = 10°, readout TR/TE = 5.5/2.6 ms) of entire K-space. Raw CEST images were acquired at varying saturation offset frequencies ranging from ±2 to ±4 ppm (relative to water resonance) with a 0.25 ppm increment. To remove field inhomogeneities induced artifacts in GluCEST maps, images for B₁ and ΔB₀ field mapping were acquired and used for GluCEST reconstruction as previously described (17,18,21). In brief, CEST images obtained from ±2 to ±4 p.p.m were interpolated using the cubic spline method to generate images with a fine step size of 0.01 ppm. B₀ corrected CEST images at ±3 parts per million (ppm) were generated from the interpolated CEST images by picking signals according to the frequency shift in the ΔB₀ map. The B₀ corrected ± 3.0 ppm images were then used for computing the percentage GluCEST contrast, which is equal to 100 × (M_−3ppm − M_+3ppm)/M_−3ppm, where M_−3ppm and M_+3ppm are B₀ corrected images at −3ppm and +3ppm respectively. B₁ inhomogeneity artifacts in GluCEST maps were removed using B₁ calibration curves as previously reported (21). The B₀ and B₁ corrected GluCEST contrasts were then averaged within expertly drawn regions-of-interest (ROIs). Overall, the acquisition of CEST images from one slice takes about 10 min. The acquisition of additional images required for mapping B₁ and B₀ field variation takes about 5 min.

MRS of Glutamate

MRS data were collected from six healthy male subjects in a separated study. Localized manual shimming of the B₀ field was performed in left amygdala and hippocampus (Voxel size: 10 × 10 × 10 mm³) separately to obtain localized water line width of ~0.15 ppm or less. Voxel placement was guided based on the 3D reformatting of high-resolution whole-brain MPRAGE images collected in the same parameters as the above. Water suppressed short echo single voxel spectra (SVS) were obtained with a custom-modified Point-RESolved Spectroscopy (PRESS) sequence (22,23) to quantify Glu using parameters: spectral width = 4 kHz, number of points = 2048, averages = 128, TE = 20ms and TR = 3s. By changing the gradient ramp time and spoiler area, we were able to use short TE of 20ms. Water suppression was achieved with a variable power RF pulses with optimized relaxation delays (VAPOR) (24), which was played out before the SVS acquisition. Acquisition time was ~6 min per spectrum. Water reference spectra were also acquired using the same sequence with water suppression turned “off” and 16 averages. Total acquisition time for spectra from both voxels, structure scans and shimming was approximately 40 min.

We used custom-built spectroscopy processing packages developed in MATLAB (R2007b, MathWorks, Natick, Massachusetts, USA) for spectral processing, as recently reported (10). In brief, SVS spectra were processed from the raw free induction decay data by exponential apodization of 10 Hz, Fourier transformation, phase correction, baseline removal (25) and followed by least-squared curve fitting. For Glu quantification, water suppressed SVS were fitted as a sum of Gaussian peaks initially positioned at locations corresponding to brain metabolites. The integration of the 2.35 ppm Glu peak was normalized by water reference signal for Glu quantification.

Statistical Analysis

Two-tailed paired Student’s t-tests were used to compare GluCEST contrasts and Glu quantifications using ¹H-MRS in different brain structures. Difference is considered to be significant for p<0.05.

Results

A high resolution GluCEST map of an axial brain slice corrected from B₀ and B₁ inhomogeneities is shown in Figure 1B. ΔB₀ map (Figure 1C) of this slice generally shows ~0.5ppm variations and the B₁ map (Figure 1D) <50% of variance compared to the reference B₁ (B_1ref). GluCEST map (Figure 1B) shows higher GluCEST in subcortical structures, such as putamen, caudate and thalamus, than cortex and occipital gray matters. White matter has much lower contrast than all gray-matter tissues and negligible contrast is seen from the cerebrospinal fluid (CSF). The spatially smoothed GluCEST map shown in Figure 2B approximates the PET volume distribution map of glutamate receptor (26) (Figure 2D), despite differences in brain shape (Figure 2A vs. 2C), resolution and sensitivity, demonstrating the ability of GluCEST MRI to map regional distribution of Glu in brain.

Figure 1.

A representative GluCEST map from an axial brain slice. (A) Proton anatomic image. B) ΔB₀ map; C) B₁ map (The reference B₁ or B_1ref = 155 Hz or 3.6 μ;T); D). GluCEST map. Slight differences of CEST contrast in corresponding tissues across hemispheres may be due to brain positioning asymmetry.

Figure 2.

Comparison of GluCEST map with Glu receptor PET image. Proton anatomic image (A) and the corresponding GluCEST map (B). The recolored GluCEST map is reprocessed from Figure 1D by smoothing (2 by 2), rescaling and applying a colormap similar to the PET image, the combination of Black, “Winter” and “Hot” colormaps in MATLAB. Anatomic (T₁-weighted, C) and corresponding PET maps of distribution volumes (DVs) of metabotropic Glu receptor (Reprinted by permission of the Society of Nuclear Medicine: J Nucl Med ³¹, Figure 2, copyright (2007)). Although the GluCEST map and the PET image show similar pattern, their differences may due to differences in brain shapes, resolutions and slice thicknesses.

High resolution GluCEST imaging was also performed for a sagittal brain slice containing amygdala and hippocampus structures (Figure 3). ΔB₀ map (Figure 3B) of this slice generally shows variation within interval (−1, 1) ppm and B₁ map (Figure 3C) shows <50% variation in most regions of the brain despite the area around cerebellum and a small portion of the frontal lobe. Due to the large B₁ inhomogeneity, B₁ inhomogeneity induced artifact in GluCEST map (Figure 3D) cannot be fully corrected in these two regions (masked with ovals). The magnified GluCEST map (Figure 3F) from the amygdala-hippocampus region shows higher contrast from amygdala than hippocampus.

Figure 3.

A representative GluCEST map from a sagittal brain slice. A representative MRI data from sagittal brain slice. (A) Proton anatomic image showing amygdala and hippocampus in the white box. B) ΔB₀ map; C) B₁ map; D). GluCEST map of the same slice. Two regions marked with ovals have <50% B₁ that cannot be fully corrected for B₁ inhomogeneity. E-F). A 15 times magnification of the proton image and CEST map from the amygdala and hippocampus region (white square in A).

The summarized mean GluCEST contrast of the subcortical structures is shown in Figure 4. Amygdala shows the highest contrast (15±1%) in all the gray-matter structures and is significantly higher than hippocampus (12±1%). Consistent with previous reports (18), white matter tissue has about 40% less contrast than most gray-matter tissues. The GluCEST regional distribution approximates to glutamate spatial variations quantified by ¹H-MRS (27).

Figure 4.

Regional variation of average (n = 6) GluCEST contrast from different brain structures. The observed GluCEST is about 40% higher in gray matter, such as putamen and occipital lobe than in white matter regions, such as internal capsule and callosum splenium. GluCEST in amygdala is significantly higher than that in occipital gray matter (*p<0.05), which is significantly higher white matter regions (e.g. internal capsule).

A modified PRESS based SVS sequence using short TE down to 20ms was successfully implemented in the sub-cortex at 7 T. Using a 32-channel head coil, we were able to obtain water suppressed spectra from amygdala and hippocampus with 1cc voxel size and adequate signal to noise ratio (Figure 5-6). The Glu peak located at 2.35ppm is clearly visible and well separated from N-acetylaspartate (NAA) at 2ppm. Glu concentration quantified from MRS was higher in amygdala as compared to hippocampus (Figure 5-6). Mean Glu concentration (Figure 5B) was 17±3 mM in the amygdala and 14±3 mM in the hippocampus. These values are comparable to a recent study at 3 T (8). The difference of GluCEST contrast and Glu concentration quantified by MRS between amygdala and hippocampus are 2.8±1.3% a n d 3 .4±2.0mM respectively, giving GluCEST detection sensitivity of ~0.8% per mM in vivo, which is close to the previously reported 0.77% per mM in gray matter (18).

Figure 5.

A). Anatomic images in axial (top) and coronal (bottom) axis used for prescribing amygdala and hippocampus voxels (white squares) for MRS. B). Representative MR spectra from amygdala and hippocampus. The Glu peak at 2.35ppm is well defined. C). Averaged (+/−SD) concentrations of Glu obtained using ¹H-MRS in amygdala and hippocampus.

Figure 6.

Stack plots of all the MR spectra collected from amygdala (Top) and hippocampus (Bottom) at 7T. The black line was used for lining up all the spectra and the red dotted lines were placed approximately on the averaged peak height of glutamate peak at 2.35 ppm.

Discussion

Subcortical Glu was measured using 7 T ¹H-MRS and CEST MRI, a novel alternative molecular imaging method that has been recently utilized for detecting metabolites in millimolar concentration. The Glu peak at 2.35ppm was clearly resolvable using ¹H-MRS and the concentration of Glu was higher in the amygdala than the hippocampus. Consistently, the observed GluCEST is about 40% higher in gray matter than in white matter. Within gray-matter structures, amygdala shows the highest GluCEST contrast, which is consistent with previous findings that amygdala has the highest ratio of relative regional cerebral blood flow to glucose metabolism in the brain (28). GluCEST maps also show a similar distribution pattern as compared to PET maps of the Glu receptor (26). Comparing GluCEST regional variation to glutamate concentrations quantified using ¹H-MRS, we estimate the in vivo GluCEST detection sensitivity of 0.8% per mM. This is consistent with the previous estimation based on numerical simulations of full Bloch-McConnell equations (18). Taken together, these data provide additional evidence that Glu in subcortical structures is measurable using GluCEST MRI.

In our previous study (18), a superficial brain slice in the axial view was scanned to demonstration the feasibility of GluCEST MRI for differentiating glutamate levels in gray and white matter. Here, we extend upon these findings to map glutamate distribution in brain subcortex. Importantly, neurochemical dysfunction within subcortical structures is often implicated in neuropsychiatric and neurodegenerative disorders. For example, the amygdala facilitates perception of salient stimuli and is associated with social/emotional processing and emotional learning, which is affected in schizophrenia (29). Moreover, schizophrenia is associated with persistent dysfunction of Glu transmission (30). Given the importance of the subcortical structures in emotion, memory and other behavioral domains, accurate quantification of the neurochemicals and neurotransmitters via GluCEST may prove useful in improving diagnosis specificity or quantifying treatment response.

However, a few issues must be considered. Precise quantification of Glu from short-echo MRS remains a challenge due to overlapping signal from macromolecules (31,32). While, GluCEST MRI may be more suitable for molecular imaging of brain glutamate, GluCEST detection sensitivity is field dependent and varies with tissue pH and imaging parameters, including RF saturation power and duration.

Direct conversion of GluCEST contrast into absolute concentration remains a challenge because GluCEST contrast as 3ppm may have contributions from creatine, Gamma-Aminobutyric acid (GABA), amide protons and intrinsic magnetic transfer asymmetry as discussed previously (18). Furthermore, using in vitro experiments of denatured protein solutions to estimate in-vivo conditions (e.g. exchange rates) may not be analogous. Based upon the numerical simulations of full Bloch-McConnell equations (18), we estimate that glutamate is the major contributor to the observed GluCEST. However, more work is needed to precisely determine the amount of contaminations from other factors.

Finally, successful implementation of GluCEST in some brain regions remains challenging as CEST MRI is vulnerable to B₀ and B₁ field inhomogeneities and proper correction (17,18,21) is required to precisely determine GluCEST contrast. In this study, the use of larger saturation offset range than the actual B₀ variation ensures that GluCEST can be fully corrected from B₀ inhomogeneity. However, there are substantial B₁ inhomogeneities (up to <50%) in the sagittal acquisition that cannot be fully corrected using calibration based methods (21). Hence, caution should be taken when interpreting the data from regions with large B₁ variation.

In conclusion, given the consistency of subcortical GluCEST contrasts with PET and MRS studies, GluCEST mapping may serve as biomarker of brain Glu distribution, in vivo. High-resolution GluCEST MRI could help elucidate normal brain function and improve diagnosis of neural disorders associated with dysfunction of subcortical structures.

Abbreviations

APT

amide proton transfer

B_1rms

B₁ root mean square

CEST

Chemical exchange saturation transfer

Glu

glutamate

Glx

glutamate/glutamine

GluCEST

Glutamate chemical exchange saturation transfer

IRB

Institutional Review Board

mM

millimolar

NAA

N-acetylaspartate

ppm

parts per million

PRESS

Point-RESolved Spectroscopy

PET

Positron Emission Tomography

¹H-MRS

Proton magnetic resonance spectroscopy

RF

radio-frequency

ROI

region of interest

SVS

single voxel spectrum

T

Tesla

VAPOR

variable power RF pulses with optimized relaxation delays