Hippocampal Single-Voxel MR Spectroscopy with Long Echo Time at 3 Tesla using sLASER Sequence

Abstract

The hippocampus is one of the most challenging brain regions for proton MR spectroscopy (MRS) applications. Moreover, quantification of J-coupled species such as myo-inositol (m-Ins) and glutamate + glutamine (Glx) is affected by the presence of macromolecular background. While long echo time (TE) MRS eliminates the macromolecules, it also decreases the m-Ins and Glx signal, and as a result, these metabolites are studied mainly with short TE. Here we investigate the feasibility of reproducibly measuring their concentrations at long TE of 120 ms, using sLASER sequence, as this sequence was recently recommended as a standard for clinical MRS. Gradient offset-independent adiabatic refocusing pulses were implemented, and an optimal long TE for the detection of m-Ins and Glx was determined using T₂ relaxation times of macromolecules. Metabolite concentrations and their coefficients of variation (CV) were obtained for a 3.4 mL voxel centered on the left hippocampus on 3 Tesla MR systems at two different sites with 3 healthy subjects (32.5 ± 10.2 years (mean ± standard deviation)) per site, each subject scanned in two sessions, and each session comprising of three scans. Concentrations of m-Ins, choline, creatine, Glx and N-acetyl-aspartate were 5.4 ± 1.5, 1.7 ± 0.2, 5.8 ± 0.3, 11.6 ± 1.2, and 5.9 ± 0.4 millimolar (mean ± standard deviation), respectively. Their respective mean within-session CVs were 14.5 ± 5.9%, 6.5 ± 5.3%, 6.0 ± 3.4%, 10.6 ± 6.2%, and 3.5 ± 1.4%, and their mean within-subject CVs were 17.8 ± 18.2%, 7.5 ± 6.3%, 7.4 ± 6.4%, 12.4 ± 5.3%, and 4.8 ± 3.0%. The between-subject CVs were 25.0%, 12.3%, 5.3%, 10.7%, and 6.4%, respectively. Hippocampal long-TE sLASER SVS can provide macromolecule-independent assessment of all major metabolites including Glx and m-Ins.

INTRODUCTION

Despite its classification as a cortical structure¹, the hippocampus is fundamentally different from the rest of the cortex. It has unique neuronal cytoarchitecture and organization, capacity for neurogenesis, and high degree of synaptic plasticity. These properties underlie its central role in learning and memory^2,3, but unfortunately also render it particularly susceptible to the effects of aging and disease^3,4. Indeed, the hippocampus is affected in most neurological disorders (e.g. multiple sclerosis⁵, traumatic brain injury⁶, human immunodeficiency virus⁷ and schizophrenia⁸). This widespread hippocampal involvement across neurological disorders, especially Alzheimer’s disease (AD) and its preclinical and prodromal states⁹, whose prevalence is increasing¹⁰, dictates a need to advance non-invasive methods to quantify and track hippocampal pathology in order to enable earlier diagnosis, prediction of disease progression, and response to therapy. As shown in multiple studies of the hippocampus in the above-mentioned disorders^8,11–13, in vivo proton MR spectroscopy (MRS) is well positioned to fulfill such roles. It measures levels of metabolites, e.g. N-acetyl-aspartate (NAA), creatine (Cr), choline compounds (Cho), myo-inositol (m-Ins), glutamate and glutamine (Glu and Gln, respectively), which are surrogate markers for different functions or cellular states¹⁴.

Unfortunately, the hippocampus is one of the most challenging brain regions for obtaining reliable MRS data. Its small size necessitates small voxels, which suffer from low signal-to-noise ratio (SNR), and its location is prone to magnetic susceptibility effects, which degrade spectral resolution^15,16. Therefore, to enable high quality clinical studies, appropriate approaches need to be developed and validated for feasibility and precision. Short echo time (TE, TE < 50 ms) MRS is generally preferred due to increased SNR and limited J-evolution (i.e. signal degrading) of metabolites¹⁷. However, resolving spectral overlap at short TE remains a challenge, and the optimal way to report systematic error sources such as the macromolecular (MM) background is still under debate¹⁸.

Long TEs (TE > 100 ms) benefit from easier quantification due to the lack of MM background, but signal decay due to T₂ relaxation and J-coupling effects limits most studies to the metabolites exhibiting major singlets, i.e. NAA, Cr and Cho^19,20. The question whether there is a TE which offers adequate SNR for quantification of other metabolites, yet without the confounding effect of a MM background has not been answered fully, partly due to lack of knowledge of macromolecular concentrations and T₁ and T₂ relaxation times. Fortunately, effective T₂ relaxation times of 10 different macromolecular resonances were recently measured at 3 T, which enables the selection of the shortest possible TE at which the MM background has decayed to the noise level²¹. Simple extrapolation from this report shows that TE of 120 ms provides a simplification of decomposition of signals and quantification of metabolites (Figure 1).

Figure 1:

Simplified simulations of T₂ relaxation of normalized signals for macromolecules (MM) and metabolites. The T₂ range of macromolecules shows shortest and longest relaxation times measured at 3 T (M2.04 and M1.21 respectively)²¹. The average T₂ relaxation time for macromolecules (21.9 ms) was calculated from relaxation times of all MM signals measured at 3 T²¹. Similarly, the T₂ range of metabolites shows short and fast relaxing metabolites, glutamine (Gln)²⁴ and N-acetyl-aspartate (NAA)¹⁶, respectively. The average T₂ relaxation time for metabolites (195.0 ms) was calculated from T₂ relaxation times of NAA, creatine and choline measured in hippocampus at 3 T¹⁶, and myo-inositol, glutamate, glutamine and lactate measured in occipital cortex at 3 T²⁴. Note that at TE = 120 ms, most MM signal is gone, while on average 60% of the metabolite signal remains. This demonstrates that relatively little metabolite signal was lost while the specific choice of TE = 120 ms enabled easily quantifiable Glx and m-Ins resonances.

To build on this knowledge, the two objectives of this work were: (1) to minimize chemical shift displacement errors (CSDE) and minimize macromolecular background by implementing gradient offset-independent adiabatic (GOIA) refocusing pulses using WURST modulation for RF and gradient waveforms (GOIA-W)²² in a semi-adiabatic localization by adiabatic selective refocusing (sLASER) sequence²³ with TE of 120 ms; and (2) to establish the signal variability of the resulting long-TE sLASER approach in terms of coefficients of variation (CVs) of metabolic concentrations: within-session CVs (from three back-to-back scans within the same subject and session), within-subject CVs (first block from two separate sessions within the same subject), and multi-center variability in terms of between-subject CVs (across two scanners from the same vendor at different institutions with different subjects).

METHODS

Theoretical and Experimental Investigations of Long TE

T₂ relaxation times of brain metabolites²⁴ are longer than those of macromolecules²¹. This fact can be exploited by using TE sufficiently long to let MM background decay to noise but still detect longer relaxing metabolites (Figure 1). Based on average T₂ relaxation time of MM (21.9 ms), TE of 120 ms allows the MM background to decay to ≤ 0.5% of its original signal. The longest T₂ relaxation time measured in MM at 3 T was reported for M1.21 (1.21 ppm) which was 38.7 ms²¹. However, this signal had also second lowest concentration and its resonance frequency did not overlap any of the major metabolites. Therefore, its contribution in spectrum measured with TE of 120 ms was neglected.

J-coupling interaction with TE (J-modulation) can have an impact on intensities of spectral peaks. Coupled metabolites such as m-Ins, Glu and Gln (Glu + Gln = Glx) are especially susceptible to J-modulation because of their relatively lower intensity and larger overlap in the spectrum. Therefore, the spectral lineshapes were investigated via the density matrix formalism with Magnetic Resonance Spectrum Simulator (MARSS)²⁵ and in vivo measurements at TEs of 90, 100, 110, 120 and 130 ms. More details can be found in Supplementary Information, S1.

Subjects and Scanner Hardware

Six healthy adult subjects (32.5 ± 10.2 years (mean ± standard deviation), 3 men / 3 women). They were scanned as follows: three (1 man / 2 women) at the Center for Biomedical Imaging (CBI) at New York University Langone Health, and the other three (2 men / 1 woman) at Jerome L. Greene Science Center at Columbia University, part of the Columbia MR Research Center (CMRRC). Exclusion criteria were any MRI contraindications, neurological disorders, virus infections, previous head trauma and alcohol or drug abuse. All subjects provided informed consent, and all studies were approved by the Institutional Review Boards of the two sites.

All subjects were scanned in Siemens Prisma 3 T systems (Siemens Healthineers, Erlangen, Germany) with a maximal gradient amplitude of 34 mT/m per direction (or a maximum combined gradient amplitude of 80 mT/m) at a slew rate of 200 T/m/s using a standard clinical 20-channel head coil (Siemens).

MRI, Segmentation and Shimming

Hippocampi were localized employing a 3D high-resolution sagittal magnetization-prepared rapid gradient-echo (MP-RAGE) sequence (TR = 2400 ms, TE = 2.24 ms, TI = 1060 ms, resolution 0.8 mm isotropic, flip angle = 8°, slices per slab = 208, FOV = 256 × 240 mm², matrix size = 320×200, GRAPPA acceleration = 2, total acquisition time = 6:38 min). The brain tissue within the VOI was segmented from the MP-RAGE images with SPM12 software package²⁶, yielding gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) masks. Their volume of interest (VOI) fraction (f_gm, f_wm, f_csf) were calculated with in-house written MATLAB R2018b (MathWorks, Natick, MA, USA) scripts.

B₀ shimming was performed first on the whole brain volume with the standard “GRE Brain” option implemented in the scanner (Siemens) and then by manually adjusting the shim currents to a water linewidth below 30 Hz in magnitude mode. Next, the adjustment volume was changed to that of the VOI, and interactive shimming of the linear shim currents was performed to achieve water linewidth below 20 Hz in magnitude mode. This sequential shimming volume approach (whole head, followed by VOI) allowed for satisfactory B₀ homogeneity in the hippocampus and prevented the generation of spurious echoes from insufficient water suppression²⁷. Similar approach was used for hippocampal MRS at 7 T²⁸.

Hippocampal MRS

Metabolites were measured with a sLASER sequence with an optimized sinc RF excitation pulse with bandwidth of 4.4 kHz (inferior-superior, IS) with duration of 2 ms and four GOIA refocusing RF pulses based on WURST modulation (GOIA-W(16,4))²⁹ with bandwidths of 15 kHz (left-right, LR, 2 pulses) and 20 kHz (anterior-posterior, AP, 2 pulses) and duration of 3.5 ms. Additionally, a 32-step phase cycling was implemented³⁰. Water signal was suppressed with a WET scheme³¹. All spectra were acquired as 2048 complex points, with a spectral bandwidth of 2000 Hz (or 15.6 ppm). A 3.4 mL VOI (26×10×13 mm³, AP×IS×LR), was placed in the left hippocampus avoiding the amygdala and inferior white matter. The VOI was typically inclined by 20–35° in the axial plane. Its position was cross-checked with resliced MP-RAGE images in all three anatomical planes. The left hippocampus was chosen because of AD literature showing faster gray matter loss in the left hemisphere³², and left hippocampal atrophy preceding atrophy of the right hippocampus^12,33. Metabolite spectra were acquired with TE of 120 ms, TR of 1.5 s and number of acquisitions of 256 with 4 preparation acquisitions, i.e. dummy scans. Water signal was acquired as an internal reference for metabolite quantification with the same VOI and sequence with TE of 25 ms, TR of 10 s and 16 acquisitions. The short TE and long TR were chosen to minimize relaxation effects on water signal. The total acquisition time for metabolite measurements was 6:30 min and for water measurements 2:40 min.

Data Processing and Quantification

For all experiments the raw data from all individual acquisitions and receivers were stored separately. Since the individual spectral acquisitions have very low SNR, 32 acquisitions were combined, and averaged over one phase cycle. These sub-averages with higher SNR were then phase- and frequency-aligned over the region from 1.0 to 4.2 ppm using a script written in MATLAB and FID-A toolkit³⁴. The water reference was used for sensitivity-weighted summation of the receive channels and eddy current phase correction³⁵. The water residual was removed from averaged acquisitions with singular value decomposition³⁶ in FID-A.

Further spectral processing and analysis was done with the freeware MRS software package INSPECTOR^37,38. The complex data were truncated up to 1024 points and zero filled to 4096 points. No line-broadening was performed. The baseline was kept flat and without any corrections. Basis set for linear combination modeling was calculated with MARSS. The basis set contained 9 metabolites: m-Ins, s-Ins, Cho, Cr, Gln, Glu, NAA and lactate (Lac). Since several amino acids resonate close to the signal of lactate³⁹, detected signal from this frequency range is termed Lac+. Cr signal was modeled with two basis functions, Cr (CH₂) and Cr (CH₃). Signal uncertainties were estimated as Cramér-Rao lower bounds (CRLB)⁴⁰. Absolute metabolite concentrations were calculated using the measured water signal⁴¹, taking into account published water and metabolic relaxation times (more details in Supporting Information, Table S2).

Variability of Metabolite Concentrations and Signal Quality

Variability within sessions, within subjects, and between subjects were assessed with CVs, defined as the ratio of the standard deviation to the mean. “Within-session” refers to variability among three consecutive scans in one session of the same subject. “Within-subject” was defined as variability between the first scan from both distinct sessions in the same subject. “Between-subject” was defined as variability between all subject scans, where subject scan was defined as first scan from the first session. Overview of the experimental setup and illustration of calculation of the three variances is shown in Figure 2. Metabolite concentrations for all subjects were reported from session #1 and scan #1. SNR was measured as a ratio of the NAA peak amplitude at 2.02 ppm and the root mean square of the noise amplitude from 8 to 12 ppm of the spectrum in question. The homogeneity of the B₀ field was assessed with full-width at half-maximum (FWHM) of the NAA peak at 2.01 ppm.

Figure 2:

(A) All experimental measurements for one subject and illustrations of data used for calculation of within-session and within-subject variability. (B) Illustration of data used for calculation of between-subject variability.

RESULTS

Investigations of Long TE

Simulated intensities and shapes of m-Ins and Glx peaks were influenced by long TEs as illustrated in Figure 3A. The FWHM values and CRLBs measured in INSPECTOR for the J-coupled systems of m-Ins and Glx at each TE, are shown in Table S1. The FWHM of m-Ins decreased with TE, whereas the FWHM of Glx increased with TE. The lowest CRLBs of both m-Ins and Glx were measured at TE of 120 and 130 ms. This was consistent with the observation that at these TEs both metabolites showed distinguishable peaks in positive phase, with high SNR for m-Ins and the pseudo-singlet of Glx. All five simulated spectra were in agreement with the in vivo measurements in the hippocampal region and showed minimal presence of the MM background (Figure 3B).

Figure 3:

Investigating spectral peaks measured with long TEs with sLASER. (A) Simulation of metabolite signal evolution as a function of increasing TEs, with basis functions of m-Ins and Glx shown separately. (B) In vivo verification of the simulations in a voxel centered on the left hippocampus. The spectra were line broadened (Lorentzian 0.5 Hz, Gaussian 0.5 Hz²) for better visualization. The linewidths and concentrations of the simulated spectral peaks were adjusted to in vivo spectra. Note the similarities between the simulated and in vivo datasets, demonstrated by the fidelity of the fit as well as by the minimal residuals. All metabolites include m-Ins, s-Ins, Cho, Cr, Gln, Glu, NAA and Lac.

Segmentation and Data Quantification

Voxel placement for the left hippocampus is shown in Figure 4A. Across all six subjects, the mean ± standard deviation f_csf, f_gm, f_wm were 0.06±0.04%, 0.64±0.03% and 0.30±0.03%. The mean VOI fraction and standard deviation of gray matter in all sessions was 64.4 ± 2.9%, and showed relatively stable localization of hippocampal tissue. The median f_gm and f_wm CVs were 3.3% and 3.6%, respectively, showing high reproducibility of the voxel placement. A hippocampal spectrum acquired at TE of 120 ms is shown with its fitted function in Figure 4B. The basis set used for the quantification is shown in Figure 4C. All simulated hippocampal metabolites were quantified in every spectrum. Mean metabolite concentrations (reported from session #1 and scan #1) from both sites and with their corresponding mean CRBLs are summarized in Table 1.

Figure 4:

(A) Placement of the 3.4 mL voxel (26 × 10 × 13 mm³) in the left hippocampus (29 years, woman), containing mainly its body and tail. (B) Spectrum from the hippocampus fitted with linear combination modelling. Note, no macromolecular background in the residual signal. (C) Basis set used for quantification of hippocampal metabolites.

Table 1:

Metabolite concentrations, standard deviations and Cramér Rao Lower Bounds of all subjects from Session #1 and Scan #1, measured at two different sites.

Site			m-Ins	s-Ins	Cho	Cr	Gln	Glu	Glx	NAA	Lac+
CBI	Concentration	Mean	5.0	0.6	1.6	5.7	2.0	9.2	11.2	5.9	2.6
		S.D.	2.0	0.2	0.2	0.4	1.3	2.4	1.8	0.5	0.6
	CRLB	Mean	16.4	39.6	6.4	3.8	32.1	11.4	-	2.2	15.7
		S.D.	6.1	24.2	1.7	0.6	27.2	1.8	-	0.3	3.5
CMRRC	Concentration	Mean	5.8	0.4	1.8	5.9	2.7	9.2	11.9	5.9	2.4
		S.D.	0.5	0.4	0.3	0.3	0.8	1.1	0.4	0.3	0.9
	CRLB	Mean	24.8	79.8	8.2	4.9	22.9	12.5	-	2.6	30.9
		S.D.	8.4	35.0	3.8	0.3	12.1	1.2	-	0.3	26.2
All	Concentration	Mean	5.4	0.5	1.7	5.8	2.4	9.2	11.6	5.9	2.5
		S.D.	1.4	0.3	0.2	0.3	1.0	1.7	1.2	0.4	0.7
	CRLB	Mean	20.6	59.7	7.3	4.3	27.5	12.0	-	2.4	23.3
		S.D.	8.0	34.8	2.8	0.8	19.5	1.5	-	0.4	18.7

Metabolite concentrations are in millimolar (mM); CBI - Center for Biomedical Imaging; CMRRC - Columbia MR Research Center; CRLB - Cramér Rao Lower Bounds in %; S.D – standard deviation.

Variability of Metabolite Concentration

The mean SNR and FWHM from all sessions were 15.3 ± 3.2 and 8.7 ± 1.4 Hz respectively. The tissue fractions and mean NAA SNR and FWHM per subject and session are shown in Table 2. The within-session CVs ranged from low values for NAA (across all six subjects: session one 3.5 ± 1.4%; session two 2.2 ± 1.6) to high values for s-Ins (62.3 ± 59.8%; 56.6 ± 59.8%, respectively). The within-session CVs of all basis set metabolites are shown in Table 3. Within-subject CVs ranged from a minimum of 0.6% for NAA in subject #6 to a maximum of 141.4% for s-Ins in the same subject. The within-subject CVs of all volume fractions and metabolites from scan 1 of each session are summarized in Table 4. Between-subject CVs of volume fractions and metabolite concentrations from session 1 and scan 1 are summarized in Table 5. Independent t-testing demonstrated no significant site effect for any of these values (p>0.05). Between-subject CVs in metabolite concentrations pooled across both sites ranged from a low for Cr (5.3%) to a high for s-Ins (62.7%). Glutamate (18.0%) and m-Ins (25.0%) demonstrated between-subject CVs of intermediate magnitude. Comparison of spectral quality and concentrations of 1 subject in within-session, 1 subject in within-subject and all 6 subjects in between-subjects is illustrated in Figure 5.

Table 2:

Tissue fractions measured in all subjects and sessions. Mean values of SNR and FWHM per subject were calculated from all three scans in a session. The first three subjects were scanned at CBI and the other three were scanned at CMRRC.

		fcsf [%]	fwm [%]	fgm [%]	SNR	FWHM [Hz]
Sub. #1	Sess. #1	6.4	30.4	63.1	13.3 ± 0.5	7.7 ± 0.2
	Sess. #2	8.5	31.5	59.7	20.3 ± 1.1	7.1 ± 0.6
Sub. #2	Sess. #1	4.4	29.5	66.0	15.3 ± 1.5	8.9 ± 0.1
	Sess. #2	2.9	34.0	63.0	20.3 ± 1.1	6.9 ± 0.4
Sub. #3	Sess. #1	3.4	30.2	66.3	16.6 ± 0.5	8.2 ± 0.6
	Sess. #2	1.8	31.9	66.3	17.3 ± 0.5	7.9 ± 0.7
Sub. #4	Sess. #1	9.3	21.6	69.1	14.3 ± 1.5	9.6 ± 0.7
	Sess. #2	14.7	26.2	59.2	17.0 ± 1.0	7.3 ± 0.2
Sub. #5	Sess. #1	5.1	31.5	63.4	12.6 ± 2.3	8.9 ± 0.6
	Sess. #2	5.1	30.0	64.8	12.3 ± 0.5	10.4 ± 1.2
Sub. #6	Sess. #1	2.5	30.2	67.3	10.6 ± 1.1	9.5 ± 0.6
	Sess. #2	4.4	31.2	64.4	12.6 ± 0.5	11.0 ± 0.4
Mean	Sess. #1	5.2	28.9	65.9	13.8	8.8
S.D.		2.4	3.6	2.3	2.1	0.7
Mean	Sess. #2	6.2	30.8	62.9	16.6	8.4
S.D.		4.7	2.6	2.9	3.5	1.8

CBI - Center for Biomedical Imaging at New York University Langone Health; CMRRC - Columbia MR Research Center at Columbia University; FWHM - full-width at half-maximum; S.D. - standard deviation; Sess – session; SNR – signal-to-noise ratio; Sub – subject;

Table 3:

Within-session coefficients of variation [%] calculated per subject and session. The first three subjects were scanned at CBI and the other three were scanned at CMRRC.

		m-Ins	s-Ins	Cho	Cr	Gln	Glu	Glx	NAA	Lac+
Sub. #1	Sess. #1	13.8	39.6	2.1	6.4	30.7	13.6	4.4	1.5	7.7
	Sess. #2	4.9	38.4	4.4	4.4	24.0	2.0	4.2	1.5	14.9
Sub. #2	Sess. #1	6.2	18.9	9.9	4.9	28.9	22.1	19.6	4.3	33.1
	Sess. #2	18.9	64.1	2.3	9.4	66.1	9.7	17.9	4.2	47.6
Sub. #3	Sess. #1	19.3	10.1	2.1	2.7	6.9	16.2	12.9	2.9	20.2
	Sess. #2	33.7	31.6	8.3	2.8	33.1	12.8	14.0	0.7	16.1
Sub. #4	Sess. #1	9.8	87.3	5.6	4.4	12.7	3.2	3.0	4.5	33.4
	Sess. #2	22.7	18.0	2.0	2.9	21.2	4.5	6.2	1.5	19.9
Sub. #5	Sess. #1	16.0	46.4	15.4	5.4	4.5	15.6	13.9	5.3	7.5
	Sess. #2	26.3	14.5	6.0	11.6	20.8	13.6	9.8	4.1	6.8
Sub. #6	Sess. #1	22.0	171.3	3.5	12.4	14.1	9.3	9.5	2.4	33.2
	Sess. #2	25.7	173.2	8.0	11.5	30.5	7.3	10.8	0.9	57.9
Mean	Sess. #1	14.5	62.3	6.5	6.0	16.3	13.3	10.6	3.5	22.5
S.D.		5.9	59.8	5.3	3.4	11.1	6.5	6.2	1.4	12.6
Mean	Sess. #2	22.0	56.6	5.2	7.1	32.6	8.3	10.5	2.2	27.2
S.D.		9.7	59.8	2.7	4.2	17.1	4.6	5.0	1.6	20.5

CBI - Center for Biomedical Imaging at New York University Langone Health; CMRRC - Columbia MR Research Center at Columbia University; S.D. - standard deviation; Sub – subject; Sess – session.

Table 4:

Within-subject coefficients of variation [%] calculated per subject (Sub.). The first three subjects were scanned at CBI and the other three were scanned at CMRRC.

	fcsf	fwm	fgm	m-Ins	s-Ins	Cho	Cr	Gln	Glu	Glx	NAA	Lac+
Sub. #1	20.2	2.7	3.9	13.6	9.3	3.0	6.4	2.3	19.7	15.5	6.0	2.5
Sub. #2	29.2	10.2	3.4	4.7	40.7	2.6	4.2	37.4	16.2	17.3	4.3	7.2
Sub. #3	45.1	3.9	0.0	7.4	19.1	9.4	8.5	2.4	6.3	5.1	3.0	22.8
Sub. #4	31.5	13.5	10.9	28.4	75.6	1.2	4.7	33.2	1.7	6.4	5.8	12.4
Sub. #5	0.1	3.4	1.6	3.1	30.2	10.8	19.6	15.9	15.2	15.3	9.3	21.3
Sub. #6	39.5	2.3	3.1	50.0	141.4	17.7	1.1	7.2	18.0	15.0	0.6	83.1
Mean	27.6	6.0	3.8	17.8	52.7	7.5	7.4	16.4	12.8	12.4	4.8	24.9
S.D.	16.0	4.7	3.7	18.2	49.1	6.3	6.4	15.5	7.2	5.3	3.0	29.6

CBI - Center for Biomedical Imaging at New York University Langone Health; CMRRC - Columbia MR Research Center at Columbia University;

Table 5:

Between-subject coefficients of variation [%] calculated from session #1 and scan #1.

	fcsf	fwm	fgm	m-Ins	s-Ins	Cho	Cr	Gln	Glu	Glx	NAA	Lac+
CBI	32.3	1.6	2.7	39.0	24.6	10.5	6.7	63.5	25.8	16.3	8.7	23.8
CMRRC	60.9	19.4	4.4	9.2	111.6	14.7	4.4	29.8	11.9	3.5	5.0	40.0
All	46.8	12.5	3.5	25.0	62.7	12.3	5.3	43.6	18.0	10.7	6.4	29.2

CBI - Center for Biomedical Imaging at New York University Langone Health; CMRRC - Columbia MR Research Center at Columbia University;

Figure 5:

(A) Average and standard deviation of three spectra measured in one subject in the first session for calculation of within-session variability with boxplots of corresponding metabolite concentrations. (B) Average and standard deviation of two spectra measured in one subject in both sessions for calculation of within-subject variability with boxplots of corresponding metabolite concentrations. (C) Average and standard deviation of six spectra measured in all subjects in the first session for calculation of between-subject variability with boxplots of corresponding metabolite concentrations. All spectra were normalized to NAA peak (2.01 ppm).

DISCUSSION

This work explores challenges in long TE hippocampal SVS at a clinical field of 3 T and provides an outlook for studies that have limited acquisition time to obtain high-quality spectra without macromolecular contamination.

Long TE

The use of short TE comes with the well-known advantages of higher SNR, the ability to quantify a larger number of metabolites, and less sensitivity to T₂ relaxation effects. The major drawback is the fact that the spectral fitting routine needs to account for the macromolecular background. Although it is possible to collect its signal on a per-subject basis⁴², due to time constraints and the need for more involved post-processing, this approach has not been adopted neither in the clinic nor in research applications. As a result, it is not known how much error is introduced by macromolecular background changes in disease⁴³, which could be problematic since the overall contribution to the spectrum due to macromolecules is large. The typical solution is to use regularized optimization algorithms that can simultaneously fit metabolites and a baseline spline function. The spline function, however, has no physical basis and, as a result, can produce line shapes which are not representative of macromolecules, thereby leading to potential inaccuracy in the assessment of metabolite concentrations^18,44. Most long-TE studies did not report metabolites other than the singlets, especially in the hippocampus⁴⁵. It was shown, however, that long TE benefits glutamate detection without macromolecular contamination⁴⁶. Based on T₂ relaxation times of macromolecules²¹ and simulations of metabolites, we used a TE of 120 ms at which the Glx multiplet is coalesced into a pseudo-singlet, quantifiable with high accuracy due to its prominent signal and the flatter or non-existent macromolecular background present at long TEs. Long TE of 120 ms enabled robust detection and analysis of m-Ins and Glx, which are metabolites of interest in diseases affecting the hippocampus

Segmentation

Low variability in localization and voxel composition are crucial for hippocampal MRS. Our voxel contained relatively high gray matter content (high f_gm), which was reproducibly obtained between sessions in the same subject (low within-subject f_gm CV). Voxel brain tissue composition was comparable to values reported in the literature^15,16, however because of the age differences between the cohorts, further comparisons of tissue and CSF fractions with previous studies are unwarranted.

Sequence and Signal Quality

As recommended by the recent methodological MRS consensus articles^47,48, sLASER localization was chosen for its order-of-magnitude smaller CSDE compared to the current SVS standard of PRESS⁴⁹. sLASER provides numerous benefits over PRESS, besides the smaller CSDE⁵⁰: it offers improved spatial profile¹⁸, reduced sensitivity to B₁ inhomogeneities⁵¹, and reduced anomalous J-modulation³⁶, which is particularly relevant at long TE. The use of GOIA pulses is also recommended by the same expert body⁴⁸, with GOIA-W identified as “optimal” for sLASER^17,52. Incurring a large CSDE is particularly undesirable in the hippocampus, whose small size already poses a problem for MRS. Specifically, a spectrum from a sequence with a large CSDE, such as PRESS, will be affected by the heterogeneous tissue and CSF composition around the hippocampus which can introduce line broadening and uncertainty in signal origination, both of which decrease the accuracy and reproducibility of the experiment. This circumstance is especially undesirable considering the high scientific relevance of the hippocampus in the study of many neurological disorders.

We presented sLASER sequence with adiabatic GOIA-W refocusing pulses with bandwidth of 15 and 20 kHz which largely minimize the CSDE. These pulses have also relatively low requirements for specific absorption rate (SAR) enabling use of short repetition times, in our case 1.5 s. In two cases (subject #1 and #2), SNR improved dramatically between sessions. This can be attributed to better shimmed B₀ or better positioned hippocampus with respect to the head coil. Since subject #1 has very similar FWHMs in both sessions, the latter explanation is more likely. We encourage to invest time and resources for best possible fixation of the head within the RF coil, since this is crucial for hippocampal MRS.

Metabolite Concentrations and Their Variability

An early study showed that the NAA concentrations measured with MRS in the hippocampus (NAA = 7.6 ± 0.9 mM)⁵³ was lower than in other brain structures. More recent data from 3 T are not in agreement with each other: NAA ≈ 7 mM¹⁵ vs. NAA = 10.0 mM¹⁶, although both studies used the same software pipeline. Our data showed even lower concentration of NAA (5.9 ± 0.4 mM). Further investigations using localization techniques with minimal signal contamination and absolute quantification are needed to confirm that hippocampal NAA concentrations are different from other GM. This is a plausible hypothesis, given the unique neuronal cytoarchitecture and organization of the hippocampus, and the fact that the WM fibers within the hippocampus are less myelinated than WM in the forebrain, potentially contributing to a different metabolic signature^54,55. Reduced contribution of N-acetyl-aspartyl-glutamate (NAAG, 2.04 ppm) in the spectra due to its relatively short T₂ relaxation time²⁴ could be part of the explanation of lower concentration of NAA, although NAAG is expected to have much lower concentration than NAA⁵⁶.

Next, we discuss the source of variation within each set of reported CVs. The within-session CVs are affected by noise and patient motion. Therefore, as expected, they were lower than the within-subject CVs acquired across-sessions, which in addition contain variations from different voxel placement and different B₀ shim. In turn, for most metabolites the latter were lower than the between-subject CVs, which integrate all of the above variations, as well as different voxel tissue fraction (due to hippocampal anatomy, not different positioning) and biological differences. The between-subject CVs allowed the assessment of variations between the two different sites. The low CVs in case of f_gm and major metabolites (Cho, Cr and NAA) suggested well-localized hippocampal VOI and feasible sLASER sequence implementation across two centers. The CVs calculated for m-Ins and Glx from subjects measured at CBI showed higher values. In one subject, the water suppression pulses were most likely not calibrated properly and their frequency and bandwidth influenced the m-Ins signal, which in turn resulted in concentration of 2.8 mM and contributed to relatively high CV for this site. The difference in CVs between the sites in case of Glx could be caused by motion or unknown physiological phenomenon in the beginning of the first session at CBI. Subjects at CBI were measured at various times during the day, however all subjects at CMRRC were measured around the same time in the evening. Circadian variations in Glx⁵⁷, and in other metabolites, are speculative at this point, but considering the high sensitivity of MRS, they should be taken into account in the future studies.

Sample size estimations for future studies are dependent on knowledge of the technique’s CVs, as reported here, and the magnitude of difference between the cohorts. The following confounding factors, however, need to be taken into account in practice. First is that data from disorders affecting the hippocampus is seldom reported in absolute concentrations (mM); second, unknown disease-caused variations in relaxation times can vary both the magnitude of these differences and the technique’s CVs in the patient cohort. These changes are potentially modulated by disease stage and subtype, in an interplay with normal ageing⁵⁸. Third, smaller hippocampal volumes would decrease the SNR, hence yield higher CVs. All of the above are relevant to one of the main applications of ¹H MRS in the hippocampus, i.e. monitoring progression to AD, which involves very different disease states (preclinical to AD with dementia) across a large age range of elderly subjects with hippocampal volumes smaller than those sampled in this study. These limitations are important to consider in planning any ¹H MRS study, and especially when using sequences incurring T₂ and T₁ weighting, as presented here. However, as initial aid in planning a hippocampal study in schizophrenia, mild cognitive impairment or AD, we refer the reader to a recent literature review⁵⁹. It lists in absolute amounts, metabolite concentrations of patients and controls for all studies which report statistically significant differences (for estimating the magnitude of the expected difference); and any published metabolite (and water) relaxation times in controls and these diseases (for estimating possible errors).

Prior to this work, variability in metabolite concentrations as measured by MRS in the hippocampus at 3 T has been investigated in two other works^15,16. Compared to our study, Allaili et al., who used shorter TE and smaller VOI (TE = 65 ms, VOI = 2.4 mL)¹⁶, showed similar median CVs at 256 acquisitions per scan, except for lower median Glx CV. In comparison with Bednarik et al. who used short TE and slightly larger VOI (TE = 28 ms, VOI = 4.1 mL)¹⁵, our data shows higher CVs for all metabolites. Both cited studies used a previously measured macromolecular background from the occipital cortex. Our approach is based on the observation that, to date, macromolecules in the hippocampus have not been measured in healthy subjects or in disease; therefore, it may be prudent to remove any quantification ambiguities introduced by this uncertainty by using a long TE acquisition, which still provides all key metabolites. We note that such application is not limited to the hippocampus, since the observation cited above applies to most brain regions and diseases. Indeed, the current validation of feasibility in the hippocampus, which is one of the most challenging regions to study with MRS, renders this approach applicable in any other brain region without the need for further development.

Limitations

This study had the following limitations. First, the number of subjects was small and the within-subject and between-subject variability can likely be decreased with a larger cohort and a narrower age range.Second, the same subjects should be measured at different sites for a true measure of between-subject reproducibility. However, conducting a medium-to-large scale multi-center reproducibility study was not the purpose of this work. Third, larger than expected variations in within-session metabolite concentrations point to possible body movements throughout the three scans. Since the VOI is very small, even slight head movements can alter the localization, which can result in variations of metabolite concentrations. Fourth, since we used long TE and relatively short TR, the T₂ and T₁ weighting of metabolites is not negligible. We used reported relaxation times from different brain regions, since only the major metabolites’ T₂ have been measured in the hippocampus³⁹. Therefore, the T₁ and T₂ values were not matched to the exact sequence and region, hence, the resulting variation can impact the quantified concentrations. Finally, simulations and all in vivo measurements are performed with a specific parameters and RF pulses used in the sLASER sequence. Therefore, generalizations of the findings presented here may not be relevant to all measurement set-ups.

Conclusion

The presented long-TE MRS protocol with sLASER sequence employing GOIA-W adiabatic refocusing pulses at 3 T enabled acquisition of all major hippocampal metabolites. Spectra measured with echo time of 120 ms showed no macromolecular background and robust detection of m-Ins and Glx.

ABBREVIATIONS

AD

Alzheimer’s disease

CBI

Center for Biomedical Imaging

CSF

Cerebrospinal fluid

f_csf

Cerebrospinal fluid fraction

CSDE

Chemical shift displacement errors

CV

Coefficients of variation

CMRRC

Columbia MR Research Center

CRLB

Cramér-Rao lower bounds

Cr

Creatine

Cho

Choline

FWHM

Full-width at half-maximum

Glu

Glutamate

Glx

Glutamate + glutamine

Gln

Glutamine

GOIA

Gradient offset-independent adiabaticity

GM

Gray matter

f_gm

Gray matter fraction

Lac

Lactate

MM

Macromolecules

MARSS

Magnetic Resonance Spectrum Simulator

MP-RAGE

Magnetization-prepared rapid gradient-echo

m-Ins

Myo-inositol

NAA

N-acetyl-aspartate

s-Ins

Scyllo-inositol

sLASER

Semi-adiabatic localization by adiabatic selective refocusing

SNR

Signal-to-noise ratio

VOI

Volume of interest

WM

White matter

f_wm

White matter fraction

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.