MEGA-PRESS of GABA+: Influences of acquisition parameters

Abstract

γ-aminobutyric acid (GABA) was the first molecule that was edited with MEGA-PRESS. GABA edited spectroscopy is challenged by limited selectivity of editing pulses. Coediting of resonances from macromolecules (MM) is the greatest single limitation of GABA edited spectroscopy. In this contribution, relative signal contributions from GABA, MM and homocarnosine to the total MEGA-PRESS edited signal at ~3 ppm, i.e., GABA+, are simulated at 3 tesla using several acquisition schemes. The base scheme is modeled after those currently supplied by vendors: it uses typical pulse shapes and lengths, it minimizes the first echo time (TE), and the delay between the editing pulses is kept at TE/2. Edited spectra are simulated for imperfect acquisition parameters such as incorrect frequency, larger chemical shift displacement, incorrect transmit B₁-field calibration for localization and editing pulses, and longer TE. An alternative timing scheme and longer editing pulses are also considered. Additional simulations are performed for symmetric editing around the MM frequency to suppress the MM signal. The relative influences of these acquisition parameters on the constituents of GABA+ are examined from the perspective of modern experimental designs for investigating brain GABA concentration differences in healthy and diseased humans. Other factors that influence signal contributions, such as T₁ and T₂ relaxation times are also considered.

Graphical abstract

The influences of acquisition parameters on edited GABA signal at 3.02 ppm were investigated using simulation of MEGA-PRESS sequence at 3T. Signal contributions from macromolecule and homocarnosine to the edited signal were also considered.

1. INTRODUCTION

γ-aminobutyric acid (GABA) editing sequences are hampered by coediting of macromolecules (MM) and homocarnosine (Hcar) due to the limited selectivity of the editing pulses. In the original work on GABA editing, MM contribution was characterized alongside that of GABA^1,2, and the measured signal was designated as total signal as opposed to pure GABA signal. Rothman et al. estimated the MM contribution to GABA* to be 60% at 2.1 T¹. In the next several applications^2,3 at 4 T, MM signal was found to be approximately 50% of the total signal, which was abbreviated GABA+_. Fifty percent became the accepted approximation of MM signal contribution⁴ to the combined edited GABA, MM and Hcar resonances, which continues to be abbreviated as GABA+. Several early approaches reduced³ or eliminated^5,6 the MM contribution to GABA edited spectra by increasing the frequency selectivity of the editing pulse (i.e., by optimizing pulse shape³, using a long pulse¹, or scanning at higher field) or by introducing a symmetric pulsing scheme⁵ to suppress MM contribution. Use of lysine (Lys) was introduced to mimic MM for development purposes⁵, since Lys residues are probably the predominant contributors to the J-coupled MM signals at 3.0 ppm and 1.7 ppm in the brain⁷. An inversion recovery approach was introduced to measure MM contamination in vivo⁶. Hcar was known from an early stage to coedit with GABA and to have a concentration of ~0.5 mM in the occipital lobe⁸, compared to a GABA concentration of ~1 mM^1–6.

The earliest GABA edited experiments were attentive to diligent setting of acquisition parameters. Pulse frequencies were set carefully, including the chemical shift of the editing pulse^1,5 and the center frequency of the localization pulses. Frequency drift was watched over, since a misset frequency would cause a notable change in signal^6,9. Short, high bandwidth localization pulses were used to minimize chemical shift displacement error (CSDE), but required high transmit B₁¹, which was generally achieved via surface radiofrequency (RF) coils^1–3,6 . Short localization pulses also left sufficient time in the optimal 1/2J echo time (TE) (68 ms) for long editing pulses with high frequency selectivity. It was acknowledged that the estimated GABA concentration was dependent on the T₁ and T₂ relaxation times of GABA and the reference signal^1–3,6.

In spite of widespread use of MEGA-PRESS for GABA editing, the effect of imperfect acquisition parameters on GABA+ signal and MM co-editing is often overlooked. Acquisition parameters are often set using approaches that have been adapted from imaging protocols, and those approaches are not optimal for spectroscopy. Vendor-specific differences in acquisition parameters (i.e., differences in pulse sequence timing, pulse shapes and lengths, and MM suppression schemes) are the current focus of optimizing test-retest repeatability and minimizing site-to-site variance^10–12. However, the following aspects of optimization are frequently not reported. 1) The procedure used to calculate and set the frequency of the editing pulse is frequently not reported. For example, it could be calculated from the water frequency or from the frequency of a strong singlet in a non-edited spectrum. 2) At what time the frequency of the editing pulse is set in the protocol, and whether the frequency is checked and/or updated during acquisition of the edited spectrum is frequently not reported. This becomes important if there is drift during the acquisition. 3) The length and bandwidth of localization pulses are frequently not reported. Maximum system B₁ limits how short localization pulses can be and still achieve excitation, which limits the ability to minimize CSDE. Lengthening the pulses makes the bandwidth smaller and the CSDE greater. The procedure used to calibrate RF power is often not described, perhaps because RF power is not calibrated. 4) T₁ and T₂ relaxation times of GABA^10–20 are often not considered. 5) MM coediting is usually mentioned whereas that of Hcar is not.

The goal of this paper is to provide comparative data on the influences of acquisition parameters on edited GABA and MM. The extent to which Hcar contributes to the GABA+ resonance is also considered. Reported measures of GABA+ in healthy and diseased humans are cast In light of influences of acquisition parameters and coedited compounds.

2. EXPERIMENTAL DETAILS

2.1. Simulation methods

Metabolite spectra were simulated in MATLAB (The MathWorks Inc, Natick, MA) based on density matrix formalism²¹. All simulations were performed at 3 T using a spectral width of 6 kHz and 4096 complex points during signal acquisition.

The PRESS sequence was simulated by taking into account the RF shapes, durations and interpulse timings. 2D localization²² consisting of 40×40 spatial points²³ was achieved by sweeping the frequency of each pair of refocusing pulses along each spatial dimension from -FWHM to FWHM, where FWHM is the bandwidth in Hz of the refocusing pulses as previously described²³. The carrier frequency was set to 3.0 ppm. Resulting spectra are presented as if all compounds have the same concentration.

The edited signals for GABA, Lys and Hcar were integrated within a ±25 Hz region around 3.01 ppm after applying an exponential decay function to match their linewidths to in vivo condition of 8 Hz, performing three times zero-filling and baseline correction.

2.2. Selection of acquisition schemes

Figure 1 illustrates the MEGA-PRESS pulse sequence used to simulate edited GABA, Lys, and Hcar resonances under the experimental conditions of interest at 3 T. The base sequence was designed as a compromise among those currently available from the major vendors¹⁸. Non-proprietary and accessible pulse shapes and lengths were used, including a 3.6 ms Hamming sinc for excitation (B₁ = 14.3 μT, bandwidth = 2.47 kHz measured at FWHM), 5.8 ms Mao pulses for refocusing (B₁ = 22 μT, bandwidth = 0.99 kHz at FWHM), and 15 ms sinc-gauss pulses (bandwidth = 81 Hz at FWHM) set at 1.90 ppm for edit ON and at 7.46 ppm for edit OFF conditions. No RF pulse used more than 24 μT for B₁ maximum, in accordance with capabilities of currently available hardware on clinical systems. The spacing between the middle of the editing pulses (τ₃ + τ₄) was kept at TE/2 (TE = 68 ms) and the first spin echo (τ₁) of PRESS was kept as short as most of the vendors are able to achieve¹², i.e., 14.7 ms. τ₂ and τ₅ were set to maintain the desired TE with τ₂ = τ₅. Several acquisition conditions were investigated using the following acquisition parameters that were different from those mentioned above:

Figure 1:

Diagram of the MEGA-PRESS sequence timing used in simulations. Pulse shapes for: PRESS excitation, Hamming sinc; PRESS refocusing, Mao; and MEGA editing, sinc-gauss. The spacing between the middle of the editing pulses (τ₃ + τ₄) was kept at TE/2. Symmetry was kept with τ₃ = τ₄ and τ₂ = τ₅. Details for the pulse lengths and powers, timing and settings of chemical shift used to simulate several acquisition scenarios are detailed in the text.

1. Frequency misset and drift were simulated for GABA and Lys by shifting the editing pulse carrier frequency to ±2, ±4, ±10, and ±20 Hz from 1.9 ppm. This was based on our experience that respiration can cause frequency to vary by about 4 Hz, but it is not exceptional for us to see frequency shifts as high as 10 Hz, and cooling after sequences with high gradient demand can cause up to 20 Hz drift¹⁰ over a typical MEGA-PRESS edited acquisition.

2. Larger chemical shift displacement was simulated for GABA by increasing the length of the refocusing pulses from 5.8 ms to 9.45 ms (and decreasing B₁ from 22 μT to 13.5 μT) while keeping the delay between the editing pulses at TE/2 i.e., 34 ms.

3. The effects of miscalibrated B₁ for PRESS localization (excitation and refocusing pulses) and editing pulses were simulated by using 25% lower B₁ than the optimal B₁.

4. The echo time was increased to TE = 80 ms while keeping the spacing between the middle of the editing pulses (τ₃ + τ₄) at TE/2 = 40 ms and the first spin echo (τ₁) of PRESS at 14.7 ms.

5. An alternative timing (t_alt) in which the two editing pulses are separated by less than TE/2 as used by some vendors was simulated. In our simulation, we used τ₂ = τ₃ = τ₄ = τ₅ = 13.325 ms to simulate GABA at TE = 68 ms (the delay between editing pulses was 26.65 ms). GABA was simulated at this t_alt using editing pulse lengths of 15 ms and 20 ms.

6. The MM-suppressed editing sequence, i.e., placing the “off” pulse symmetric about the problematic coupling partner (as opposed to symmetric about water) was simulated for GABA and Lys. Since Lys was used to imitate MM, the symmetric pulse was placed at 1.52 ppm in this case, i.e., symmetric about the 1.71 ppm of Lys ⁵CH₂ relative to the 1.90 ppm that was used for GABA editing.

2.3. Selection of coupling systems

The GABA system is yet to be fully worked out²⁴, though our pulse-acquire simulations suggest that the missing information is inconsequential at in vivo line widths. We used the system specified in Govind et al.²⁵ because it used phantom (i.e., pure) spectra that were stated to have been measured at physiologic pH and temperature, and it included geminal couplings.

The MM coupling system relevant to editing GABA ²CH₂ at 3.01 ppm via coupling to ³CH₂ at 1.89 ppm and specified tentatively in the literature was used^7,26, i.e., Lys ⁶CH₂ at 3.02 ppm via coupling to Lys ⁵CH₂ at 1.71 ppm. Robin de Graaf provided the Lys coupling system in Table 1, which he derived from human plasma samples scanned at 25°C and pH of 7.2²⁷. Since the ⁴CH₂ - ⁵CH₂ - ⁶CH₂ moiety is most pertinent to MEGA editing of the 3.01 ppm resonance of GABA, we only simulated these protons. Note that it would take undue computational time to simulate the whole molecule with localization since Lys has nine protons. We used pulse-acquire simulation at short TE and non-localized PRESS at long TE (i.e., 68 ms) to affirm that differences in the Lys resonance at 3.02 ppm between our abbreviated 6-proton simulation and the full 9-proton simulation were inconsequential at in vivo line widths, which is in agreement with the literature²⁸.

Table 1.

Lys coupling system used for simulations. Derived for ref²⁷ from human plasma samples at 25°C and pH 7.2.

Spin	Chemical shift (ppm)	J-couplings(Hz)	Connectivity
2CH	3.746	6.09	2–3
		6.09	2–3’
3CH2	1.881	−15.27	3–3’
		5.67	3–4
		10.51	3–4’
3’CH2	1.898	10.13	3’−4
		6.1	3’−4’
4CH2	1.431	−12	4–4’
		6	4–5
		8	4–5’
4’CH2	1.4929	10	4’−5
		6	4’−5’
5CH2	1.712	−12	5–5’
		6	5–6
		8	5–6’
5’CH2	1.713	10	5’−6
		6.5	5’−6’
6CH2	3.018	−10	6–6’
6’CH2	3.0198

Parameters for the Hcar system were taken from de Graaf²⁹. The spectrum used to derive this system was measured at 25°C and phosphate-buffered (50 mM, pH 7.0).

3. RESULTS

Figure 2 and Table 2 show that the edited GABA signal is impacted 8% or less over the frequency misset range from −4 to +10 Hz. However, the impact on GABA signal at a ±20 Hz frequency misset is on the order of 50%. Editing for Lys at 0 Hz from 1.90 ppm is suboptimal as expected, given that the coupling partner for Lys is at 1.71 ppm. The edited Lys signal is impacted as much as 38% over the GABA-defined frequency misset range of ±10 Hz.

Figure 2.

Impact of frequency misset on MEGA-PRESS edited GABA and Lys resonances near 3.01 ppm. GABA (top) and Lys (bottom) were simulated with the base sequence with editing pulses placed at 1.9 ppm (i.e., 0 Hz frequency offset) and shifted ±2, ±4, ±10, and ±20 Hz from 1.9 ppm.

Table 2.

Integrals of edited GABA and Lys 3.01 ppm resonances under ideal (0 Hz shift) and the misset frequencies illustrated in Figure 2. GABA and Lys values are normalized to the area under the GABA resonance measured with the base sequence. Lys values are additionally presented normalized to placement of the editing pulse at 1.9 ppm (i.e., 0 Hz frequency offset).

Frequency shift from 1.9 ppm (Hz)	GABA	Lys	Normalized Lys
−20	0.46	0.70	0.91
−10	0.82	0.88	1.13
−4	0.96	0.85	1.09
−2	0.98	0.82	1.05
0	1.00	0.77	1.00
2	1.00	0.72	0.94
4	1.00	0.67	0.86
10	0.92	0.48	0.62
20	0.62	0.23	0.29

Figure 3 and Table 3 demonstrate that chemical shift displacement errors and utilization of miscalibrated B₁ for PRESS localization or editing pulses impact the GABA signal in the range 7% to 26% whereas typically encountered drift (4 Hz) has less than 4% impact on GABA signal (but as much as 14% on Lys signal). Going to the longer TE (80 ms) leads to a loss of 8% of signal (without taking T₂ losses into account). Different sequence timing (alternative timing) has a 12% impact, i.e., a smaller impact than miscalibration of editing pulse power and comparable to miscalibration of PRESS pulse power. Using the longer, more selective editing pulses leads to a loss of only 3% of GABA signal beyond the loss due to the alternative timing.

Figure 3.

Relative impacts on MEGA-PRESS edited ²CH₂ GABA of frequency misset (+4 Hz offset), large CSDE, suboptimal powers of localization (25% lower localized B₁) and editing (25% lower editing B₁), longer echo time (TE = 80 ms), and alternative timings for TE = 68 ms with 15 and 20 ms editing pulses. The dotted line is a point of reference to compare to the base sequence. The normalized integral of GABA resonance at 3.03ppm relative to the base sequence is given in parenthesis.

Table 3.

Integrals of edited GABA resonances under acquisition circumstances of Figure 3. All values were normalized to the area under the GABA resonance measured with the base sequence. t_alt = alternative timings.

Simulation	Normalized Integral
Base	1.00
4 Hz shift	1.00
CSDE	0.91
Miscalibrated (lower) PRESS 90° and 180° B1	1.07
Miscalibrated (lower) editing pulse B1	0.74
TE = 80 ms	0.92
talt	0.88
talt, 20 ms edit pulse	0.85

Figure 4 illustrates how under-powering the localization pulses leads to a counterintuitive 7% increase (table 3) in edited GABA signal. The sub-optimally powered pulse has a different FWHM and excites more signal outside the nominal volume-of-interest (VOI). Sub-optimal power has influences on both the edit-on and edit-off experiments, leading to an increase in the area under the edited GABA resonance outside the nominal VOI.

Figure 4.

Spatial 2-dimensional representation of simulation of MEGA-PRESS edited ²CH₂ GABA with the base sequence at optimal and suboptimal PRESS localization pulse powers (75% of the power needed for full inversion). Warm to cool colors represent the area under the edited GABA resonance. Black squares are the same for both panels and illustrate the intended nominal localization boundaries.

Figure 5 shows the well-known reduction in editing efficiency that occurs when using symmetric pulsing to eliminate MM contamination. Importantly, Figure 5 shows that even under optimal frequency setting, a noteworthy amount of Lys signal remains when the MM-suppressed scheme is used with MEGA-PRESS and 15 ms editing pulses at 3 T, as previously shown²⁸.

Figure 5.

GABA and Lys 3.01 ppm resonances when MEGA-PRESS editing is simulated with the symmetric pulsing scheme (editing pulses at 1.9 and 1.52 ppm, blue) for elimination of MM contribution relative to the base scheme (editing pulses at 1.9 and 7.46 ppm, black). For symmetric pulsing, instead of setting the “off” pulse of the editing symmetric about water, they were set symmetric about the problematic coupling partner. Since Lys was used to imitate MM, the symmetric pulse was placed at 1.52 ppm, i.e., symmetric about the 1.71 ppm chemical shift of Lys ⁵CH₂ relative to the 1.9 ppm used for GABA editing.

Figure 6 illustrates the MEGA-PRESS simulation of Hcar coedited with GABA (edit ON pulse at 1.9 ppm). The edited GABA resonance is shown for comparison. The simulations were done with equal concentrations of Hcar and GABA. The area under the Hcar resonance is 91% that of GABA.

Figure 6.

Coediting of Hcar resonance with GABA simulated with the base sequence and setting the “on” editing pulses at 1.9 ppm. The black line shows the edited GABA resonance and the gray line shows the coedited Hcar resonance. GABA and Hcar were simulated at equal concentration.

4. DISCUSSION

Whereas most of the influences on GABA+ signal examined herein have already been reported, the goal of this study was to provide a systematic characterization of the influence of acquisition parameters. The reported outcomes agree with the literature^10,12,28,30. Whereas many previous reports estimated MM contribution by using a frequency shifted GABA spectrum, we used Lys as a surrogate.

Recent publications report GABA or GABA+ signal differences in the range of 3% to 30% between healthy and diseased humans^31–35. In comparison, frequency missets within 10 Hz, which are likely to occur in participants who move during the scan or if time elapses between setting the frequency and running MEGA-PRESS acquisition, have little influence on GABA signal but up to 38% influence on MM signal (Table 2). Assuming 1:1 GABA:MM signal, a 10 Hz frequency misset could cause a 21% change in GABA+ signal.

Another source of uncertainty in GABA quantification is possible variability in MM concentration with participant characteristics such as disease and age. Human brain MM concentration has been reported to vary in disease as much as 100%^36,37. A recent study on human brain MM reported an ~10% higher MM level in older than young adults³⁸. Assuming 1:1 GABA:MM signal, 10% and 100% higher MM would cause 5% and 50% higher GABA+ signal, respectively.

Our findings (Figure 5) agree with the literature that using the MM-suppressed symmetric pulsing scheme with MEGA-PRESS at 3 T reduces the GABA signal and does not eliminate all of the MM signal (mimicked through Lys signal). Imperfect suppression of MM is likely due to the asymmetry or strong coupling in Lys as previously shown²⁸. The effectiveness of the MM-suppressing symmetric pulsing scheme improves with longer editing pulses with better frequency selectivity⁵, but this is difficult to achieve with MEGA-PRESS. The effectiveness of the symmetric pulsing scheme also improves at higher fields (e.g., at 7 T there is no loss in GABA-editing efficiency and MM are almost completely suppressed even with 15 ms editing pulses²⁸).

The current findings agree with recent literature¹⁰ in that vendor-specific differences in MEGA-PRESS implementation cause editing efficiency for GABA to vary by up to 10%. However, miscalibrated transmit B₁ can cause differences of similar size or greater (Table 3). Also, if the maximum transmit B₁ is insufficient for a given individual, and localization pulses are lengthened to accommodate lower maximum B₁ (as in the Siemens implementation), signal can change by ~10% due to increased CSDE. Our simulation of the two-dimensional editing efficiency profile (Figure 4) is in agreement with literature³⁰.

Changes in relaxation times could also affect GABA signal. Scant literature on GABA suggests that T₁ is in the vicinity of 1.3 s³⁹ and T₂ of 88 ms⁴⁰ at 3 T. Commonly used repetition time (TR) and TE are 2000 and 68 ms, respectively¹¹. If T₁ were to become 20% longer or T₂ 20% shorter⁴¹, then GABA signal would change by ~20%.

The normal concentration of Hcar has been reported to range from 0.3 to 0.6 mM⁴² in postmortem brain and at 0.47 mM in in vivo brain using Hcar resonances that are partially resolved at 7.05 and 8.02 ppm in short TE spectra at 2.1 T when macromolecule resonances are removed by subtraction⁸. The 3.0 ppm resonances of GABA and Hcar cannot be separated using editing based on the coupling partners at ~1.90 ppm, even at high field, due the close proximity of chemical shifts. Normal, medication and disease-associated human brain Hcar levels have been reported to vary in the range of 2 to 4 times^8,42–44. Based on reported in vivo contributions and relative editing yields (Figure 6), Hcar signal in GABA+ is not negligible.

In addition to acquisition parameters examined in this study, robust normalization and quantification must be considered to achieve practical utility. Spectral width and number of discrete points, which were held constant in this work, can influence aliasing and peak definition and thus quantification. Frequency and phase corrections of individual shots and surveillance for absence of subtraction artifact are other important aspects of quantification that were not considered in the present simulated data. Other acquisition parameters that are important for in vivo application but were not simulated in this project are B₀ shims, eddy current correction, and consistency of VOI placement.

To acquire the most physiologically relevant data possible, some acquisition parameters have obvious optimal settings. Because each human participant loads the RF coil to a different extent, B₁ calibration of all pulses for each experiment is warranted. It is also helpful to keep PRESS localization pulses short to minimize CSDE⁴⁵. Frequency can be set immediately before each acquisition and locked at best, or tracked and reset when drift approaches 10 Hz if locking is not possible. Choice of other acquisition parameters depends on circumstances. It is crucial that users understand that MM coediting is present and that an informed decision on whether to use a MM suppressing approach (and if so, which one) needs to be made based on the study hypotheses (i.e., whether MM are likely to confound) and limitations (i.e., SNR and scan time). It is also crucial for users to understand that frequency-associated errors in relative GABA and MM signal acquisition cannot be corrected with the frequency and phase correction step of post-processing. In addition to those presented herein, other MM-reduction approaches based on increasing the length of the editing pulses have been proposed^4,46,47. Scanning at longer TE, generally to accommodate longer, more selective editing pulses, makes the measure more susceptible to T₂ variance, in addition to the loss of GABA signal to T₂ outright. While scanning at the short TR of 2 s increases SNR per unit time, it also makes the measure more susceptible to condition-specific variance in T₁ than if TR were longer.

One way to remove MM contribution with substantially less loss of MEGA-PRESS edited GABA signal (TE=68 ms) is to scan at ultra-high field⁶. However, difficulty achieving sufficient B₁ to limit CSDE currently confines this approach to the periphery of the brain. Another possibility is to forgo editing and rely on meticulous fitting of ultra-short TE ultra-high field spectra, which have the additional advantage of reduced sensitivity to T₂. Such meticulous fitting involves robust characterization of the MM contribution.

5. CONCLUSIONS

Acquisition parameters that are not usually mentioned in recent literature have impact on GABA and GABA+ signal that is comparable to those that are current areas of focus for development toward clinical trials. Given that individual parameter missets cause GABA+ signal errors of magnitude comparable to physiologically interesting differences, combinations of missets could obscure interpretation.

Several matters could be addressed to advance the field of MEGA-PRESS editing of GABA+. There is need for full specification of the coupling systems of GABA, Lys and Hcar at physiologic temperature and pH. There is also need to measure human brain GABA T₁ and T₂ in healthy and diseased humans. Knowledge on MM T₁ and T₂, as well as the extent to which MM and Hcar contributions vary is needed. Better understanding of why symmetric pulsing does not fully remove coedited Lys signal could lead to optimized MM-suppression schemes at 3 T. It might be possible to characterize GABA in non-edited spectra at 3 T, although baseline contributions and many realistic experimental conditions⁴⁸ would make this challenging. New pulse sequences could be developed that allow longer editing pulses. Spectral variability and the effects of frequency drift could be minimized with real-time motion and shim correction and frequency navigators⁴⁹.

List of abbreviations

CSDE

chemical shift displacement error

FWHM

full-width-half-maximum

GABA

γ-aminobutyric acid

GABA+

total edited signal at 3 ppm

Hcar

homocarnosine

Lys

lysine

MM

macromolecules

RF

radiofrequency

t_alt

alternative timing

TE

echo time

TR

repetition time

VOI

volume-of-interest