Selective homonuclear polarization transfer for spectroscopic imaging of GABA at 7T

Abstract

We develop and implement a selective homonuclear polarization transfer method for the detection of 3.0ppm C-4 GABA resonance by spectroscopic imaging in the human brain at 7T. This single shot method is demonstrated with simulations and phantoms, which achieves comparable efficiency of detection to that of J-difference editing. The macromolecule resonance that commonly co-edits with GABA is suppressed at 7T through use of a narrow band pre-acquisition suppression pulse. This technique is implemented in humans with an 8 channel transceiver array and high degree B₀ shimming to measure supplementary motor area and thalamic GABA in controls (n=8) and epilepsy patients (n=8 total). We find that the GABA/NAA (N-acetyl aspartate) ratio in the thalamus of control volunteers, well controlled and poorly controlled epilepsy patients are 0.053±0.012 (n=8), 0.090±0.012 (n=2), and 0.038±0.009 (n=6).

Introduction

As the primary inhibitory neurotransmitter in human brain, the robust in vivo detection of GABA remains of high interest. The concentration of GABA varies between white and gray matter, and is sensitive to a variety of neurologic and psychiatric disorders [1,2]. However, its detection is challenging for several reasons. Given its low concentration (~1mM) and high degree of spectral overlap with creatine, amino acids and macromolecule resonances, some form of spectral editing is needed for unambiguous detection. Furthermore, because of the white-gray matter variation of GABA and possible regional variation from subcortical gray matter regions (thalamus, basal ganglia [3,4]), its measurement would be preferably performed as a spectroscopic image to enable more accurate localization to specific tissues.

Most measurements of GABA have focused on selective detection of the C-4 3.0ppm protons which overlap directly with the creatine 3.0ppm methyl resonance (varying in concentration from 6 to 9mM, [5,6]). To eliminate this spectral overlap with creatine, J-difference editing using the C-3 1.9ppm coupled resonance is commonly used and has a 50% efficiency of detection [7–10]. Due to the similarity and chemical shifts of the macromolecule (MM) and GABA spin systems some additional form of numerical correction or spectral suppression is required to eliminate spectral overlap between GABA and MM [11–13]. However, because J-difference editing requires a scan-to-scan subtraction, these methods are sensitive to system and volunteer instability. Multiple quantum methods are substantially less sensitive to movement; however the detection efficiencies can be substantially lower [14–17], e.g., 25 to 39%. In this report we describe an alternative approach to detect GABA using a modification of our previously described J-refocused coherence transfer sequence [18]. This single shot “selective homonuclear polarization transfer” method eliminates magnetization at 3.0ppm, including creatine and the C-4 GABA resonance using a pre-acquisition inversion recovery suppression method. During the polarization transfer step, magnetization from the 1.9ppm C-3 GABA resonance is transferred to the 3.0ppm position. This approach was originally described by von Kienlin and has since been suggested by others [19–21], although to our knowledge, its implementation in humans has not been published. We used this approach to measure GABA in the SMA region and in the thalamus in control volunteers (n=8 each region). We also implemented this approach to evaluate thalamic and neocortical GABA levels in n=8 partial onset epilepsy patients, 6 of whom are poorly controlled, 2 of whom are seizure free. This study was motivated by the pivotal role GABA is thought to have in the hyperexcitability of epileptic brain [22–24], and the ongoing clinical evaluation of a thalamic stimulator for the treatment of intractable epilepsy [25].

Theory and Methods

The J-refocused coherence transfer sequence is a double spin echo acquisition that refocuses J- modulation through a homonuclear polarization transfer between coupled spins [26,27], Fig 1. In this approach, minimization of J-modulation for the I (and S spin) magnetization is achieved by virtue of bi-directional coherence transfer from S → I and I → S. We have previously used this approach at 7T to detect the cerebral amino acids at moderate echo times [18]. With the moderate echo time, this method decreases contributions from broad macromolecule resonances and thereby reduces uncertainties in the baseline of 7T spectra. It is evident from the J-refocused acquisition that the J-coupling between partner spins enables a single shot editing method for GABA: initial suppression of the C-4 3.0ppm resonance (and overlapping/adjacent resonances including creatine, choline and macromolecules) via an inversion recovery sequence followed by a homonuclear polarization transfer sequence, results in transferred magnetization from the C-3 1.9ppm GABA resonance to the C-4 3.0ppm. This magnetization is therefore detected without the overlapping and adjacent resonances.

Figure 1.

Pulse sequence showing use of two RF distributions to detect GABA. The homogeneous distribution is optimized over the entire (intracranial) slice and is used for all pulses except for the outer volume suppression which uses the ring distribution. The two 90° pulses (numerically optimized 5.12msec) were applied with 0. 78G/cm slice selection. The refocusing pulses were optimized 6 element broadband semi-selective pulses, total duration 4.82msec. Gradients were not used for specific coherence selection in this sequence, and for the spin echo were conventionally placed about the refocusing pulses (2.35G/cm, 5.12ms, two axes for each pair).

The TE dependence of this sequence is calculated using GAMMA simulations and is shown in Fig. 2 in comparison to a pulse acquire sequence and the base J-refocused transfer acquisition. The simulation used chemical shift values and coupling constants as reported by Govindaraju [28]. While the accuracy of the simulation is good for the C-4 resonance, it is noted that the C-2 (2.3ppm) shows some variation with the experimental data. This is likely due to some unavoidable B₁ inhomogeneity present with the 2liter phantom sphere and/or inaccuracies in the coupling constants used for simulation at 7T. The C-4 GABA 3.0ppm resonance exhibits minimal sensitivity for TEs between 34 and 45msec. At TE=50msec there is a significant decline in integrated area. Based on the simulated spectra at TE=40msec, integration of the 3.0ppm resonance area from its apparent baseline shows an efficiency of 50% in comparison to the pulse acquire acquisition. This efficiency is comparable to the J-edit difference approaches. In comparison to 7T, simulation of the same sequence at 3T assuming identical offsets and coupling constants (i.e., stronger coupling) at TE=40msec gave 85% of the detection efficiency.

Figure 2.

Simulation (right) and 20mM GABA (1ml/liter Gd-Magnevist) phantom data (left) for the selective homonuclear polarization transfer sequence. The echo times are listed in ms. A simulated pulse acquire acquisition is shown for amplitude comparison. The resonance at ~2.75ppm in the phantom is due to Gd-DTPA (38).

Because the coherence transfer sequence has a high B₁ dependence (>sin⁵θ), a key aspect of successful GABA detection is achieving sufficient B₁ amplitude and homogeneity for spectroscopic imaging. At 7T, conventional RF coils have been challenging for use in spectroscopic sequences due to problems with decreased B₁ homogeneity [28,29] and increased power requirements [30] when compared to lower fields. In this work we use an inductively decoupled transceiver array to overcome these problems. RF shimming was used to generate specific RF distributions for homogeneous excitation and outer volume suppression [31,32]. In an axial slice containing the thalamus, the homogeneous distribution with 1kHz B₁ and 10% standard deviation over the slice was typically achieved using 1.7–2.2kW. The global power deposition was typically 4 to 6Watts, i.e., for an average 3kg head giving 1.3 to 2.0W. Based on FDTD simulations of the human head [32], the maximum 10g average SAR was therefore ~4 to 6W/kg.

The second 90° transfer pulse also requires excellent phase accuracy to achieve high efficiency refocusing. For out-of-isocenter acquisitions where phase accrual may occur (which can be spectrometer-platform dependent), the theoretical optimum of 90° (relative to the excitation phase) needs to be measured. As discussed [18], we calibrate the transmit phase spectroscopically using the water resonance thereby enabling a rapid and accurate determination.

All data were acquired with an Agilent DirectDrive 7T MR system using an 8 coil (single row) transceiver array. The transceiver array was used with RF shimming to optimize multiple RF distributions to achieve 1kHz B₁ over large volumes sufficient for planar spectroscopic spin echo imaging with outer volume suppression. The pulse sequence used to detect GABA is shown in Fig. 1. To eliminate the transfer to the overlapping macromolecule resonance at 3.0ppm (1.7–3.0ppm coupling), a narrow band suppression pulse is applied at 1.6ppm. Accurate suppression of macromolecule requires excellent B₀ homogeneity. Under conditions of poor B₀ homogeneity, macromolecule resonances can be shifted outside the suppression bandwidth resulting in incomplete suppression and an increase in signal at the C-4 3.0ppm GABA position. To maximize the B₀ homogeneity 1^st–4^th degree shims were used. In this head-only MR system, 3^rd degree shims are integral to the gradient tube. The 4^th degree terms were implemented with a high degree shim insert (Resonance Research Inc., Billerica MA) [34]. Non-iterative B₀ shimming was performed using field mapping at 64x64 resolution [33]. The shimming performance was assessed by the standard deviation of B₀ over the ROI (σB₀^Global) and over individual spectroscopic imaging pixels (σB₀^Local) [34]. The two slice selective numerically optimized 90° pulses were matched and selected 10mm thick slices. The slice selection strengths were 0.7g/cm, resulting in a 1.1mm spatial mismatch between the transferring coherences (3.0, 1.9ppm). Spatial mis-match was not an issue for the broad-band semi-selective refocusing pulses. As the gradient crushers in this sequence are not performing specific coherence selection, they are conventionally placed in the pre-spin echo and around the refocusing pulses. Since both the first and second 90° excitation pulses can bring in water signal, all the spin echo crushers were executed at relatively high amplitude, 2.35gauss/cm (1000Hz/mm) along two axes, 5.12ms each.

Water suppression was achieved simultaneous to suppression of the 3.0ppm region by an inversion recovery sequence using the homogeneous RF distribution. The bandwidth of the water suppression pulse is ±375Hz, positioned at 3.8ppm. Outer volume suppression of skin/skull lipids was achieved by use of two inversion recovery pulses applied with the ring distribution. For evaluation of cortical gray and white matter, an axial plane through the supplementary motor area was studied. For the thalamic studies, a 1cm thick AC-PC angulated slice was used. In the thalamic study, the ROI used for shimming included the large majority of the entire brain within the slice with exclusion of an ovoid region over the frontal anterior ventricle. All studies were acquired as 16×16 spectroscopic images (TR 1.5s) using rectangular sampling. Each study was comprised of 3 GABA measurements (with macromolecule suppression) (~19min in total), a GABA+MM acquisition (no MM suppression pulses) (6.4min) and a double spin echo acquisition without the 2^nd 90 degree transfer pulse to detect the major singlets (N-acetyl aspartate NAA, creatine, choline). The entire duration of the study was typically 70min.

For each of the cortical gray-white matter (SMA) and thalamic regional studies, n=8 volunteers were studied. In these gray–white matter studies, the same number of voxels (4) was used from each volunteer for analysis and regression. In all studies, the spectroscopic images were processed with a Hanning filter in the spatial domain, and in the spectral time domain with 6Hz Gaussian broadening and 200Hz convolution difference. The fullwidth at half maximum (FWHM) after spatial filtering was 2.4cm. All resonances were analyzed as Gaussian lineshapes in the spectral domain using internally written software. Measurements of NAA, creatine and choline were determined from the double spin echo acquisitions without the second 90° transfer pulse. The concentration of the metabolites was estimated by assuming an N-acetyl aspartate (NAA) concentration of 10mM for both gray and white matter [5,6]. This assumption may not be valid in epilepsy patients if thalamic NAA concentrations are depressed; in this case these whole tissue GABA concentrations would be an over-estimate. However with concentrations of NAA being useful as a measure of neuronal integrity, GABA/NAA ratios can be particularly informative as a proportional measure of GABA influence on normal neuronal function. An alternative reference is tissue water; however the tissue water acquired here (for purposes of phase referencing) is highly saturated.

N=8 “localization-related” epilepsy patients (n=2, seizure free, n=6 not seizure free) were recruited from the Yale Comprehensive Epilepsy Center. These patients suffer from seizures due to a focal onset type of epilepsy and were all on anti-epileptic medications, with each patient taking anywhere from 1 to 4 different medication types.

Results

Fig. 2 shows the simulated GABA spectra for varying TE and data acquired from a phantom, showing good agreement. The acquisition at TE=40msec was chosen for the in vivo study for its moderate echo, minimal transverse relaxation losses and recovery of the transferred magnetization. Fig. 3 shows data from control and epilepsy subjects showing gray and white matter spectra from the SMA region. In the SMA, the shimming was excellent, with a mean σB₀^Global of 5.1±1.6Hz over the ROI (entire slice) for the 8 volunteers. A larger GABA resonance is seen from gray in comparison to white matter. It is also clear that the macromolecule suppression reduces the resonance area at 3.0ppm by ~50%. Thus a failure to include the MM suppression pulse would result in a factor of 2 over-estimation of the GABA concentration. To assess how GABA varies with creatine, we used the macromolecule suppressed GABA/NAA data in a regression with Cr/NAA finding a very significant relationship, R= +0.75, p<0.0001 (Figure 4). The regression is positive, consistent with the known higher concentrations of both creatine and GABA in gray matter than in white matter. In predominantly gray and white matter, GABA/NAA is 0.052±0.016 and 0.043±0.014 respectively. Scaling to 10mM NAA and assuming equivalent transverse relaxation losses, this gives GABA concentrations of 1.56±0.48mM and 1.29±0.52 in predominantly gray and white matter respectively.

Figure 3.

Data from a healthy control (I – IV) and epilepsy patient (V) are shown. From the control, spectra (I), B₁ (II), B₀ (III) maps and scout (IV) are shown. The spectra (A–D) and their corresponding locations are shown. For each spectrum are shown (bottom to top): J-refocused acquisition (TE=34msec), GABA detection including macromolecule, simple double spin echo without 90° coherence transfer, and macromolecule suppressed (TE=40msec) (both insets are 2× in amplitude). The lines shown on spectrum C identify the 3.0ppm and 3.75ppm GABA and alpha amino acid resonances. In (V) are shown spectroscopic imaging data from an epilepsy patient.

Figure 4.

Regression of GABA/NAA with Cr/NAA from the SMA region in controls (4 pixels per volunteer) is shown. The regression is significantly positive as shown, consistent with increases in both GABA and creatine occurring in gray compared to white matter.

Figure 5 shows data from the thalamus. To improve the consistency of shimming in this region, an ROI excluding the anterior frontal ventricles was used, achieving a mean and standard deviation of the σB₀^Global at 8.9±1.1Hz over the 8 subjects. In the control volunteers, similar to measurements from the SMA, macromolecules constitute ~50% of the 3.0ppm resonance in the non-MM suppressed data. In the thalamus, GABA/NAA is 0.053±0.012; putamen and pallidum 0.054±0.010, which scales to 1.59±0.36 and 1.62±0.30mM respectively. In the two well controlled epilepsy volunteers, the GABA/NAA is substantially larger in both the thalamus and in parasagittal gray matter at 0.090±0.012, 0.099±0.013 respectively. The n=6 poorly controlled epilepsy patients have lower GABA/NAA in the thalamus (0.038±0.009) while in the parasagittal predominantly gray matter, the GABA/NAA ratios are closer to control values (0.048±0.007, see Table 1).

Figure 5.

Data from a control (A) and two epilepsy patients (B, C). For each volunteer, scout and spectra (loci indicated) are shown. For the control, a B₀ map is also shown. At each locus indicated, the GABA (macromolecule suppressed) spectrum is shown with 2× magnification. The well controlled epilepsy patient shows a much larger thalamic GABA resonance than compared with control or the poorly controlled patient.

Table.

GABA ratios and concentrations in controls and epilepsy patients.

	GABA/NAA	GABA/Cr	NAA/Cr	GABA (mM)
Thalamus, Control (n=8)	0.053±0.012	0.071±0.015	1.38±0.22	1.59±0.04
Thalamus, well controlled epilepsy (n=2)	0.090±0.012	0.119±0.013	1.34±0.05	2.70±0.04
Thalamus, poorly controlled epilepsy (n=6)	0.038±0.009	0.054±0.013	1.43±0.13	1.14±0.3
Mixed GM, Control (n=8)	0.050±0.017	0.080±0.017	1.64±0.18	1.50±0.51
Mixed GM, well controlled epilepsy (n=2)	0.099±0.018	0.186±0.006	1.91 ±0.30	3.00±0.54
Mixed GM, poorly controlled epilepsy (n=6)	0.048±0.007	0.077±0.018	1.62±0.33	1.44±0.15

Quantification taken relative to NAA at 10mM in white and gray matter

Discussion

We have developed a selective homonuclear polarization transfer sequence to measure thalamic GABA in controls and epilepsy patients at 7T. The efficiency of GABA detection is comparable to that of J-difference editing, 50%. Not surprisingly, the spectra of Figures 3 and 5 show that this sequence also detects the alpha amino 3.75ppm resonance. However, as this resonance includes contributions from several amino acids (including glutamate, glutamine, alanine), it is non-optimal for measurement of these amino acids. As a single shot editing sequence for GABA, this method is relatively insensitive to small scan-to-scan variations but does require excellent RF performance and B₀ shimming. As can be seen from Figure 3, the accuracy of the pre-spin echo suppression of creatine (and choline) is excellent, as the double spin echo without transfer showed no significant residual creatine. Accurate spectral editing also requires suppression of the overlapping macromolecule signal, which can otherwise increase the detected 3.0ppm resonance by a factor of ~2. Given the 60Hz separation between the macromolecule and GABA coupled spin partners at 7T (1.9, 1.7ppm), and suppression bandwidth of ±60Hz applied at 1.6ppm, the variation in B₀ should ideally be less than 15Hz given their triplet structures. Assuming a Gaussian distribution in the B₀ values, a standard deviation of less than 7.5Hz (i.e., 2SD within 15Hz) in the B₀ is required to achieve >95% suppression of the MM resonance over all pixels. In the SMA for our n=8 volunteers, the mean and standard deviation of the variance in B₀ over all of the measured pixels is 1.6±0.3Hz. In the thalamic pixels only, the mean and standard deviation of B₀ variance is 2.0±1.7Hz. Thus for both sets of data, we should be able to unequivocally achieve accurate spectral editing at all pixels. It should be noted however that beyond the direct field value (B₀^Local) itself, the consistency of spectral editing between different pixels depends on the consistency of the σB₀^Local which was 1.9±0.20 and 5.7±1.5 for the SMA and thalamic data respectively.

In this study we determined GABA in reference to NAA because of NAA's role as a measure of neuronal integrity, making it a useful normalization parameter. This is particularly relevant for epilepsy, where there is considerable interest in understanding the relative inhibitory or excitatory function corrected for possible tissue atrophy. Given that there is no reason to believe that relaxation times for metabolites will be substantially different, we therefore believe that the GABA/NAA ratio is informative in itself. Nonetheless, alternative quantification may be clearly achieved through tissue water.

Our data from control volunteers show that GABA/NAA increases with Cr/NAA, consistent with the literature finding that GABA is higher in gray matter than white. It is also consistent with GABAergic studies of metabolic sensitivity (hypothesized as a consequence of its high tonic firing activity [35]), which may be the basis for why creatine supplementation has been found to be neuroprotective for GABAergic neurons [36]). In pure gray and white matter, the measurements of GABA/NAA are likely to be larger and smaller (respectively) due to the relatively large voxel sizes (in plane FWHM 2.4cm) used in this study. The thalamus and basal ganglia did not show substantially different values for GABA/NAA in comparison to mixed gray matter voxels in the controls. As a group, the epilepsy patients showed considerably more variability in the GABA/NAA measurements. The two well controlled patients displayed much higher GABA/NAA in all regions evaluated, while the six poorly controlled patients displayed the smallest GABA/NAA values in the thalamus. This is consistent with literature, finding that in mixed cortical (occipital) voxels, GABA levels are higher in well controlled epilepsy patients in comparison to poorly controlled patients [37].

The thalamus is commonly thought to be a site of cerebral integration for sensory, executive and associative integration. Not surprisingly for epilepsy, the thalamus is also believed to be a key locus for seizure propagation [22], with this perspective established from both basic rodent models of epilepsy and in clinical epilepsy. Bertram and others have studied several models of limbic epilepsy to find that GABA mediated inhibition in the midline thalamus is critical in the modulation of seizure propagation [23,24]. Clinically, there is an ongoing clinical trial of a deep brain thalamic stimulator for medically intractable epilepsy [25]. Our data are consistent with this view of the thalamus being an important locus of seizure regulation, given a significant decline in the thalamic GABA/NAA in the poorly controlled patients in comparison to mixed gray matter voxels. It is of note that the seizure free patients have similar GABA/NAA values in predominantly gray matter versus the thalamus. We believe that thalamic GABA measurements will be of significant interest in a variety of neurological, psychiatric and pain disorders (e.g., fibromyalgia), and specifically for epilepsy where such measurements may help identify which patients may benefit from devices such as the thalamic stimulator.