T2 relaxation times of macromolecules and metabolites in the human brain at 9.4 T

Relaxation times can contribute to spectral assignment. In this study, effective T2 relaxation times ( T2eff ) of macromolecules are reported for gray and white matter–rich voxels in the human brain at 9.4 T. The T2eff of macromolecules are helpful to understand their behavior and the effect they have on metabolite quantification. Additionally, for absolute quantification of metabolites with magnetic resonance spectroscopy, appropriate T2 values of metabolites must be considered. The T2 relaxation times of metabolites are calculated after accounting for TE/sequence‐specific macromolecular baselines.


| INTRODUCTION
Single-voxel proton MRS, a noninvasive technique, has complemented MRI by providing a means to detect and quantify concentrations of metabolites in the human brain. This has proven clinically useful, as shown in the review paper by Öz et al, 1 in establishing biomarkers for pathologies in the brain.
Advantages of MRS using ultrahigh-field scanners (≥7 T) include higher SNR and increased spectral resolution. 2 Hence, the macromolecular background signal (MMB) lying underneath the metabolites has to be characterized more precisely. Especially for short TE sequences, reliable metabolite quantification is more challenging without accounting for the MMB. Therefore, a measured macromolecular (MM) spectrum should be included in the fitting model. 3 In addition, understanding macromolecules may help to identify valuable biomarkers for pathologies and several diseases of clinical relevance. [3][4][5][6] The T 2 relaxation times of MM peaks have been reported in previous studies in rat brain [7][8][9][10] and for the M 0.92 peak in human brain at 2.1 T by Behar et al. 11 However, T 2 relaxation times of multiple individual MM peaks in human brain have not been reported to the best of our knowledge. Estimating the T 2 relaxation times of individual macromolecules at 9.4 T could help in understanding and modeling their behavior. 12 To derive absolute concentrations of metabolites in MRS, several different approaches exist. [13][14][15] In one method, the unsuppressed internal water signal is used as a reference. 16 However, the calculation of absolute concentrations from the apparent concentrations output by a quantification software requires a correction factor, which includes the T 2 relaxation times of the metabolites of interest. 16 In addition, to record 2D-MRS data, it is useful to have a rough estimate of T 2 relaxation times to optimize TE range, such that one obtains the highest possible SNR. 17,18 Additionally, altered T 2 relaxation times might provide information about evolving pathological or physiological states. 19 Several studies have reported the T 2 relaxation time for metabolites in different regions of the human brain for a range of magnetic field strengths. [20][21][22] The T 2 relaxation times were measured at 9.4 T for singlets, but not for J-coupled metabolites by Deelchand et al 23 ; however, this study did not consider the influence of T 2 -dependent MM spectra, which affects the T 2 relaxation estimation of the singlets.
The primary goal of this study was to measure the effective T 2 relaxation times (T eff 2 ) of MM peaks, which includes both relaxation and J-evolution effects, in a gray matter (GM)-rich and a white matter (WM)-rich voxel at 9.4 T in human brain. In addition, T 2 relaxation times of singlets and J-coupled metabolites in the GM-rich voxel were calculated after correcting for TE-specific MMBs. Furthermore, absolute concentrations of metabolites are reported after correcting for the corresponding T 2 and T 1 relaxation times. 24 The FWHM (Δ 1∕2 ) of MM peaks and metabolites were analyzed quantitatively with respect to T 2 relaxation times, in addition to microsusceptibility and macrosusceptibility effects.

| Technical description and subjects
All measurements were performed on a 9.4 T Magnetom whole-body MRI scanner (Siemens Healthineers, Erlangen, Germany) using a home-built proton coil with 8 transmit and 16 receive channels. 25 For single-voxel spectroscopy experiments, 3 channels at the bottom of the coil were driven using an unbalanced three-way Wilkinson splitter as previously described. 26 Eleven healthy volunteers (8 males, 3 females, age: 26.3 ± 2.8 years) participated in this study for data acquisition in the GM-rich voxel. Data for the WM-rich voxels was acquired from 5 healthy volunteers (3 males, 2 females, age: 27.8 ± 1.9 years). The study was approved by the local ethics board, and written informed consent was given by all subjects before the examination.

| Data acquisition
Gradient-echo images were acquired using a 2D-FLASH sequence (in-plane resolution: 0.7 × 0.7 mm 2 , 3.5-mm slice thickness, flip angle: 25º) along axial and sagittal orientations to facilitate placement of the spectroscopy voxel. A GM-rich voxel with the dimensions of 2 × 2 × 2 cm 3 was chosen in the occipital lobe for T 2 measurement of metabolites, whereas in addition to the GM-rich voxel, a WM-rich voxel of the same size was chosen in the occipital-parietal transition for T 2 measurement of MM peaks. First-order and second-order Conclusion: The T 2 relaxation times of all macromolecular and metabolite peaks at 9.4 T in vivo are reported for the first time. Metabolite relaxation times were used to calculate the absolute metabolite concentrations.

K E Y W O R D S
absolute quantification, macromolecules, MR spectroscopy, T 2 relaxation time, ultrahigh magnetic field B 0 shimming was performed using FAST(EST)MAP, 27 and then voxel-based power calibration 28,29 was executed.
Double inversion recovery (DIR) metabolite-cycled semi-LASER 30 and metabolite-cycled (MC) semi-LASER 26 spectra were acquired in the same 11 healthy volunteers. The TR was set to 10 seconds in the case of DIR-MC semi-LASER, and to 6 seconds for the MC semi-LASER, respectively, to ensure complete T 1 recovery of MM resonances and metabolites. 31 A 16-step phase-cycling scheme 32 was implemented for both spectroscopy sequences.
A series of DIR-MC semi-LASER spectra at different nonlinearly spaced TEs (TE = 24 ms, 32 ms, 40 ms, 52 ms, and 60 ms; TI 1 /TI 2 = 2360/625 ms; number of excitations = 32; transmit reference frequency ( ref ) = 2.4 ppm) was acquired to estimate the T To calculate the T 2 relaxation times of metabolites, a series of MC semi-LASER spectra at different nonlinearly spaced TEs (TE = 24 ms, 32 ms, 40 ms, 52 ms, and 60 ms; number of excitations = 96; ref = 7.0 ppm) was acquired, originally measured to characterize the T 2 relaxation times of downfield metabolites/resonances in the same healthy volunteers. 33 Setting ref to 7.0 ppm led to maximum chemicalshift displacement effects of +11.5% for water and +25% for NAA(CH 3 ).
In this study, for absolute quantification of metabolites, MC semi-LASER spectra (TE/TR: 24/6000 ms; number of excitations = 32) were acquired, with ref set to 2.4 ppm to minimize the chemical-shift displacement effect. To avoid any influence of MC pulses on quantification based on water, water-reference signals (number of excitations = 16) were acquired with semi-LASER (TE = 24 ms; ref = 4.7 ppm) without metabolite cycling.
Finally, magnetization-prepared two rapid gradient-echo 34 images were acquired using the same coil with RF transmission via all eight channels to calculate tissue-volume fractions for absolute quantification.

| Data preprocessing
Raw data were analyzed with in-house-written software in MATLAB (version 2016a; MathWorks, Natick, MA). The metabolite and MM MRS data were processed as described previously. 26,30 The following steps were used in processing the raw data: (1) truncation of FIDs at 250 ms for both metabolite and MM data; (2) frequency and phase alignment; (3) MC subtraction; (4) averaging; (5) zero-order phase and eddy current correction using the phase information from the MC water signal; (6) coil channel combination using a singular value decomposition method; (7) peak alignment in the frequency domain to 3.028 ppm and 3.925 ppm for the metabolite spectra and MM data, respectively; (8) residual water removal using a Hankel singular value decomposition method; and (9) truncation of FIDs at 150 ms for the MM data.

| Macromolecule fitting
An MM basis set was created in LCModel V6.3-1L 35 using simulated Voigt peaks, which is the best possible approximation used in literature 7,8,30,36 and does not account for J-evolution of these MM peaks. The chemical shifts of the Voigt lines were varied systematically based on previously reported values, 7,8,30 in which the ones reported by Pfeuffer et al 8 were found to be the best; therefore, the reported chemical shifts were chosen for the MM peaks from 0.9 to 2.3 ppm and from 2.9 to 3.5 ppm. The values of M 2.56 and M 2.70 were first reported in a human brain study at 9.4 T. 30 The chemical shifts of these two peaks were adapted to fit the shifts reported by the 17.2 T rat study. 7 The chemical shifts for MM peaks between 3.5 ppm and 4.2 ppm were chosen based on a previously mentioned study, 7 as the peaks were resolved more clearly in Lopez et al 7 compared with the other two studies. 8,30 All of these chemical shifts are summarized in Table 1 (4.17 ppm). The same basis set was used throughout the TE series, with the M 2.70 peak being simulated with a negative amplitude for TE = 52 and 60 ms, as this peak was observed to be fully inverted due to J-evolution. A much narrower Voigt line was added to fit the residual creatine at 3.925 ppm. The chemical shifts, Δ 1∕2 s, and amplitudes were constrained in the LCModel as described in Supporting Information Annex B.
Each individual MM data set and the sum of all data sets were fitted to the simulated MM basis set while enforcing a flat baseline by setting the LCModel parameter DKNTMN to 99.

| Metabolite fitting
The metabolite basis set was simulated in Vespa (ver. 0.9.3) 37 using full quantum mechanical density matrix calculations for the semi-LASER sequence, 38 including the actual complex excitation and adiabatic RF pulse shapes for all TEs specified in section 2.2. The following 17 metabolites were simulated: NAA, NAA glutamate (NAAG), γ-aminobutyric acid (GABA), aspartate (Asp), creatine (Cr), glutamate (Glu), glutamine (Gln), glutathione, glycerophosphocholine, glycine (Glyc), myo-inositol (mI), scyllo-inositol (Scy), lactate (Lac), phosphocreatine (PCr), phosphocholine (PCho), phosphoethanolamine (PE), and taurine (Tau). Their chemical shifts and coupling constants were taken from Govindaraju et al, 39 Because of the strong overlap and ill-posed problem of fitting PCho and glycerophosphocholine separately, these were combined to a total choline metabolite (tCho) with corresponding volume fractions of 0.6 mM PCho and 1 mM glycerophosphocholine based on the mean concentration values from de Graaf. 42 Similarly, PCho, glycerophosphocholine, and PE were combined with tCho+ (corresponding to volume fractions of 0.6 mM PCho, 1.0 mM glycerophosphocholine, and 1.5 mM PE 42 ), to obtain a robust metabolite fit for the T 2 calculations across the TE series.
Then, all individual metabolite spectra were fitted using the LCModel with the simulated basis set, including the respective MMB summed across subjects for each TE. The metabolite spectra summed across subjects were fitted similarly. Because the summed MMB spectra were included in the fit, DKNTMN was set to 0.25, resulting in a stiffer LCModel baseline compared with the default LCModel value of 0.15. 43 To further improve the fitting procedure, two soft constraints were introduced into the echo series fit, namely, NAA(CH 3 )/NAA(CH 2 ) = 1.2 ± 0.02 (to account for possible faster decay of NAA(CH 2 ) and mI/Glyc = 5 ± 0.5 (value extracted from the TE = 24 ms fit) (Supporting Information Annex B). No soft constraints were used for GABA or other metabolites.

| T 2 relaxation calculations
The concentrations of MM peaks and metabolite peaks from the individual data and the summed spectra were fit to a mono-exponential decay across the TE series to yield T 2 relaxation-time estimates. The mean coefficient of determination (R 2 ) value was calculated for the exponential fit across the individual subject data and was used to evaluate the quality of the exponential fits. Relaxation times with a mean R 2 smaller than 0.5 were discarded. The T 2 times for Lac, Tau, and Scy are not reported, as they did not satisfy the R 2 criterion.

| Linewidth calculations
After quantification and extracting the fitted lineshapes of MM peaks and metabolite singlets from the .coord files of LCModel quantification, the Δ 1∕2 were measured. The contribution of T 2 relaxation to Δ 1∕2 was calculated according to T eff 2 −1 , using the calculated T 2 values.

The residual linewidth was defined as
The B 0 components were calculated from NAA(CH 3 ) and tCr(CH 2 ) as respectively. The values of Δ micro , Δ macro represent the microsusceptibility and macrosusceptibility, respectively.
The unsuppressed water-reference signal from the voxel was used as the internal concentration reference for absolute quantification calculated in millimolal values (millimoles per kilogram of solvent [mmol/kg]). The concentrations of the metabolites were absolutely quantified using the formula (Supporting Information Annex A) given by Gasparovic et al 16 for TE = 24 ms, including corrections for the T 1 and T 2 relaxation times of water in different compartments and the metabolite relaxation times. More specifically, the T 1 and T 2 relaxation times of water at 9.4 T in GM (T GM 1 = 2120 ms; T GM 2 = 37 ms) and WM (T WM 1 = 1400 ms; T WM 2 = 30 ms) were taken from Hagberg et al. 34 Relaxation times for CSF were calculated based on data from the same work (T CSF 1 = 4800 ms; T CSF 2 = 181 ms). The T 1 relaxation times of metabolites were taken from Wright et al. 24 The T 2 relaxation times of the metabolite peaks calculated in this study were used. For the metabolites, for which T 2 relaxation times could not be estimated, the mean T 2 relaxation time of all other metabolites was used.

| RESULTS
Spectra without major artifacts were obtained for all subjects: The NAA(CH 3 ) SNR ranged from 526 ± 130 to 334 ± 82 for TE = 24-60 ms, respectively. However, between 0.9 and 1.8 ppm, some spectra, especially MM spectra from WM voxels (Figure 1 and Supporting Information Figure S1), presented outer-volume lipid impurities for some subjects, as lipid suppression techniques such as outer-volume saturation were not used. The value of Δ 1∕2 of the unsuppressed water signal was 17.6 ± 1.3 Hz. The TE series of MM and metabolite spectra from the GM-rich voxel are shown in Figure 1, where the shaded area depicts the SD, illustrating the reproducibility of the data quality. Supporting Information Figure S1 shows the TE series of WM-MM spectra. No data sets were excluded.

| Macromolecule fitting
The spectrum from the GM-rich voxel summed across subjects, together with all the fitted Voigt lines for the MM peaks, is shown in Figure 2 for TE = 24 ms and in Supporting Information Figure S2 for the other TEs. Supporting Information Figures S3 and S4 show the fit of the WM-TE series spectra summed across subjects. The fit residual is minimum without structured noise, indicating a high fit quality. Similar fit quality was achieved for all data sets across all TEs. The M 2.70 peak is observed to undergo a full inversion due to J-evolution over the TE steps. More precisely, the full-inversion M 2.70 occurred (TE = 24 ms) is shown with the fit using simulated Voigt lines. The residual total creatine 3.9 singlet [tCr(CH 2 )] peak in the spectra is modeled with a significantly narrower linewidth between TE = 52 and TE = 60 ms, and was simulated as a negative peak. The signal of the MM and the metabolite peaks decreased with increasing TE, as expected. The residual CH 2 resonance of total creatine [tCr(CH 2 )] was modeled by fitting a sharper Voigt line at 3.925 ppm with a measured Δ 1∕2 of around 0.035 ppm (14 Hz) across subjects. Next, the residual Cr was extracted from the LCModel fit, and was subtracted from the MM spectra to yield a more appropriate MMB for the metabolite fits ( Figure 3). The resulting MMB was included in the basis set to fit the metabolite spectra for T 2 relaxation-time calculations as well as for absolute quantification.

| Metabolite fitting
The fit of the metabolites to the summed metabolite spectra is shown in Figure 4. The fit residual is small, indicating a high fit quality. For TE = 32, 40, 52 and 60 ms, the basis set modeled the J-evolution of mI, NAA(CH 2 ), Asp, Glu, and Tau well (Supporting Information Figure S5).
Adding ascorbic acid or glucose to the simulated basis set did not improve the fit in terms of Cramer-Rao lower bounds, T 2 -fit result confidences, and residual artifacts. These resonances were not found by LCModel in most cases. Hence, these metabolites were not included in the final basis set.

| T 2 results
The calculated T eff 2 and T 2 relaxation times ( Figure 5) for MM peaks and metabolites, respectively, were overall in good agreement for the fits of individual subject data and the summed spectra. Box plots of the resulting T 2 relaxation times are shown in decreasing order for metabolites and MM peaks in Figure 5. The T 2 relaxation time of the residual tCr(CH 2 ) peak in the MM spectra was also in agreement with the relaxation time of the tCr(CH 2 ) peak in the metabolite spectra.
The T 2 relaxation times of metabolites and MM peaks from the summed spectra and individual data, together with the mean and SD of R 2 of the corresponding exponential fits, are listed in Table 2.
The T 2 relaxation times of metabolites were found to lie between 55 ms and 105 ms, except for NAAG with approximately 40 ms. In contrast, the values of T eff 2 for all MM resonances were between 13 ms and 40 ms in WM and between 13 ms and 37 ms in GM.
The R 2 for the metabolite exponential decay fits were all above 0.70, except for Asp and Glyc. The mean R 2 of the exponential decay fits was above 0.70 for all MM peaks as well, except for M 4.03 , which was subjected to water residuals and noise at longer TEs, hence the lower R 2 value. Furthermore, the uncertainties in the T eff 2 relaxation times of the M 1.21 and M 1.39 were influenced by lipid contaminations.

| Linewidth calculations
The value of Δ 1∕2 of the MM resonances was measured to vary between 35 Hz and 85 Hz across all TEs (Table 1), whereas T eff 2 −1 was calculated to be between 4 Hz and 30 Hz (Figure 6). In the case of metabolite resonances, and for the residual tCr(CH 2 ) in the MM spectra, a Δ 1∕2 of 11 Hz to 20 Hz was found, whereas T eff 2 −1 ranged between 2 Hz F I G U R E 3 Echo time series of the GM-MM spectra summed across subjects with the tCr(CH 2 ) residual subtraction shown. The spectra after the residual tCr(CH 2 ) subtraction were used as the MM baseline in the basis set for the metabolite fitting. The red line shows the residual tCr(CH 2 ), and the black line shows the MM spectra after the residual subtraction. Residual tCr(CH 2 ) is extracted from the LCModel fit as shown in Figure 2 and 5 Hz. This led to a Δ singlet value of approximately 12 Hz for the metabolite singlets.
Supporting Information Figure S7 shows the Δ residual values of metabolites and MM peaks, which were calculated using the Δ singlet of NAA(CH 3 ) and tCr(CH 2 ) for the metabolite and MM spectra, respectively. The values of Δ residual of metabolites were around zero after applying the correction. In contrast, MM peaks had Δ residual values ranging between 10 Hz and 60 Hz.

| Concentrations
Absolute concentrations in millimoles per kilogram with and without T 2 correction are shown in Figure 7. The values for the metabolites are reported in Table 3 with and without the T 2 correction factor. In Supporting Information Figure S8 and Supporting Information Table S1, absolute concentrations are given in millimoles per tissue volume in a liter (mmol/L).

MM spectra
The DIR technique sufficiently nulled all of the metabolites except tCr(CH 2 ) for the chosen TI 1 and TI 2 , as the T 1 relaxation time of this resonance is the shortest among the singlets, as reported by Deelchand et al. 23 Indeed, the difference between the T 1 relaxation times of some of the MM peaks 44 and the CH 2 group of Cr is not large; thus, there is a residual peak present in the MM spectra. The residual tCr(CH 2 ) was subtracted using the fit of the Cr singlet fit from the LCModel. The resulting spectra (Figure 3) after the residual subtraction visually showed no leftover Cr.
For the fitting of the MM spectra, the chemical shifts and Δ 1∕2 were systematically varied to achieve the lowest SD of the T 2 results among subjects, to maximize the R 2 values, and to minimize the mean Cramer-Rao lower bounds. These systematic variations supported that the values chosen by Lopez et al 7 were among the most suitable choices, and these values were also well justified by peak characteristics seen at 17.2 T. from GM with fitted metabolites and measured macromolecular background (MMB). The basis set configuration shown here was used to estimate the T 2 relaxation times and includes the tCr split into its moieties [tCr(CH 3 ) and tCr(CH 2 )], NAA split into NAA(CH 2 ) and NAA(CH 3 ) moiety, and total choline (tCho) and phosphoethanolamine (PE) combined into tCho+. For each TE, a corresponding basis set was simulated with Vespa. For absolute quantification, the NAA moieties were combined, whereas creatine (Cr), phosphocreatine (PCr), tCho, and PE were fitted as independent metabolites. Abbreviations: Asp, aspartate; tCr(CH 2 ), GABA, γ-aminobutyric acid; Gln, glutamine; Glu, glutamate; Glyc, glycine; GSH, glutathione; Lac, lactate; mI, myo-inositol; NAA(CH 2 ), NAA-aspartyl moiety; NAA(CH 3 ), NAA-acetyl moiety; NAAG, NAA glutamate; Scy, scylloinositol; Tau, taurine; tCho+, combined phosphocholine, glycerophosphocholine, and phosphoethanolamine molecules; tCr(CH 2 ), total creatine 3.9 singlet; tCr(CH 3 ), total creatine 3.0 singlet Hence, these values were chosen, with minor deviations as indicated in Table 1.
For the first time, the J-evolution of the M 2.7 resonance (Figures 1 and 3, Supporting Information Figure S1) was investigated in this study. This resonance was preliminarily assigned to β-methylene protons of aspartyl groups. 30 The Biological Magnetic Resonance Bank amino acid database 45 lists the following coupling constants for the β-methylene protons (δ2.7 ppm) of aspartate amino acids: approximately 5 and 8 Hz between α-methylene and β-methylene protons, and 17.5 Hz between β-methylene protons. These coupling constants are comparable to those of Asp and the aspartate moieties of NAA and NAAG, 40 which experience full inversion between TEs of 52 and 60 ms. All observations support the preliminary attribution of the M 2.7 resonance to the aspartyl groups. Nevertheless, the possibility cannot be excluded that the MM spectra also included NAA(CH 2 ) residuals, provided that this moiety has a short enough T 1 relaxation time.
In particular, at longer TEs (52 and 60 ms), the contribution of NAA(CH 2 ) could be more significant, as metabolites have longer T 2 relaxation times than macromolecules.

| Metabolite fitting
Previous work at 9.4 T reported concentrations of 18 metabolites in the human brain using an MC semi-LASER sequence. 26 However, due to the complexity of the adiabatic pulses and their spin locking effect, 46 the simulated basis set was approximated using a spin-echo sequence with TE = 6.5 ms. 30 In this study, the basis set for MC-semi-LASER was simulated using actual adiabatic RF pulse shapes. In addition, the same TEs were used in the acquisition of both metabolite and MM spectra. This allowed to individually fit the metabolite spectra with the matching simulated basis set and the corresponding MMB for each TE. The fit results ( Figure 4 and Supporting Information Figure S5) show that the simulated basis set represents the J-evolution patterns of the acquired spectra well.

| T 2 results
This study reports the T eff 2 of individual MM peaks as well as T 2 relaxation times of both singlets and J-coupled metabolites at 9.4 T in human brain. The values of T eff 2 of MM peaks are in agreement with previous work. Behar et al reporting 44-ms T 2 at 2.1 T for M 0.92 in human brain, 11 which is in agreement with a slow decay across field strengths, as shown in the rat brain studies. [7][8][9][10] The T 2 of metabolite singlets, such as NAA(CH 3 ), tCr(CH 2 ), tCr(CH 3 ) and tCho, are higher than those previously reported at 9.4 T. 23 The higher values in this work can be attributed to the fact that a TE-specific experimentally measured MMB was included, which is a faster-decaying component. The reported T 2 relaxation times of the metabolites appear to follow the same trend compared with previous results, 7,20,21,47 with the longest relaxation times found for the singlets NAA(CH 3 ), tCho and tCr(CH 3 ) [the T 2 of tCr(CH 3 ) larger than for tCr(CH 2 ), difference decreasing with increasing field strength], 7,20,21,23,47 and significantly shorter T 2 for NAAG than for NAA moieties. 20 All reported T 2 relaxation times show the expected negative correlation with increasing field strength. Glutamate T 2 is higher than that of Gln, in agreement with previous studies. 7,20 However, the difference is unexpected, considering the similar molecular weight of the two metabolites and the similar distribution within the brain 48 ; hence, most likely the T 2 of Gln is underestimated. Nevertheless, differences could arise from the presence in different compartments, or bindings to different transporters or enzymes.
The summed and individual fits are in good agreement for T 2 relaxation times for both MM resonances and metabolites ), the per-subject fits (mean and SD) with the confidence of the exponential fit (R 2 ). Some exponential decay fits are shown in Supporting Information Figure S6. The T 2 relaxation times of GABA, Lac, Scy, and PE are not included, as they did not satisfy the imposed R 2 > 0.5 criterion.
( Table 2). It can also be noted that the relaxation time of the tCr(CH 2 ) in the metabolite spectra and the residual present in the MM spectra are in good agreement. The differences between T eff 2 relaxation times of MM peaks in WM-rich and GM-rich voxels were never investigated previously, nor were the relaxation times calculated for the peaks individually. Therefore, an attempt was made to calculate T eff 2 for both tissue compositions in this work. The T eff 2 relaxation times found in this study are comparable between these voxels.
However, it is known that the region of interest in the human brain is crucial when applying T 2 correction for absolute concentrations of metabolites. 20 Hence, no attempt was made to report the T 2 relaxation times of metabolites separately for GM and WM, as the selected voxel was neither  purely WM nor GM. This difference in T 2 relaxation times is highly influenced by the iron concentrations across the human brain, as shown by Hasan et al. 49 Hence, the T 2 relaxation times of metabolites reported in this work are specific to the region of the GM-rich occipital lobe.
The TE values of 24, 32, 40, 52, and 60 ms were chosen for the calculation of T 2 relaxation times, while keeping the shorter MM T eff 2 in mind. The MM signals were almost entirely decayed at TE = 60 ms. Having a corresponding MMB for the metabolite spectra improves the quantification of metabolite concentrations, hence the choice of identical TEs. The chosen TEs were sufficient to estimate the T 2 relaxation times of MM peaks and metabolites with shorter T 2 times. However, including some longer TE values would improve the accuracy of the T 2 relaxation times of metabolites with longer T 2 s.

| Linewidth calculations
For any given voxel, B 0 inhomogeneities originate from a mixture of effects of Δ micro , Δ macro as well as tissue compartment effects. 23 This B 0 effect experienced by spins is identical, whereas T 2 relaxation time is metabolite/resonance-specific. The Δ 1∕2 of metabolite singlets, both in MC-semiLASER and tCr(CH 2 ) in DIR MC-semiLASER, clearly show the two components of the linewidth ( Figure 6). The Δ 1∕2 of tCr(CH 2 ) is in line with the ones previously reported. 23 The large differences between the Δ 1∕2 s of MM peaks and T eff 2 −1 ( Figure 6, Table 1) indicate that these resonances are potentially composed of unresolved multiplets and/or different protons resonating at similar chemical shifts, which are strongly overlapping. Also, Supporting Information Figure S7 shows that the values of Δ residual for MM peaks are between 10 Hz and 60 Hz, whereas for metabolites they are closer to 0 Hz. The Δ residual value is consistent between GM and WM MM peaks, indicating that the magnitude of potential overlap and/or J-evolution 11 component for each MM peak is similar between different tissue types. These peaks could originate from amino acids 45 which, depending on the larger protein structure they belong to, can have different chemical shifts, but are distributed around a main resonance frequency for the bulk of protein peaks. Table 3 provides a consolidated comparison of absolute concentrations (millimoles per kilogram) of metabolites from literature 21,23,29,50,51 and this work.

| Concentrations
Metabolite concentrations from this study are reported with and without T 2 correction for a fair comparison between the other studies and this work, as most of the other studies did not include T 2 correction due to the use of ultrashort TEs (<10 ms). The concentrations of NAA, NAAG, tCho, PE, Tau, Glu, GABA, and mI with T 2 correction are overall in good agreement with the literature. 21,23,29,50,51 Concentrations of Asp, Gln, Glyc, and Glu are significantly higher after including the T 2 correction factor. However, their concentration values match those from the literature when considered without T 2 correction. Potential overestimations in this study could also arise due to an underestimation of the T 2 relaxation times of Gln, Glu and Asp, which exhibit strong J-evolution effects at TE = 52 ms and TE = 60 ms. In particular, the Gln concentration is likely overestimated, as the spline baseline exhibits negative behaviour in that area, which could not be compensated in the LCModel fits. Improved spectral resolution at ultrahigh field between Glu and Gln also influenced Gln concentrations in the current study. Furthermore, the loss of magnetization for individual MM components because of T 1 differences in DIR semi-LASER compared with semi-LASER could also influence our results. All referenced literature used single inversion recovery sequences to measure the MMB component. Similarly, Cr and PCr concentrations match the literature without T 2 correction, but are somewhat higher (~25%) when the T 2 correction is applied. This deviation could indicate that including the T 2 correction factor in the absolute quantification yields higher concentrations compared with the literature, as in most studies, no correction for the corresponding T 2 relaxation times of the metabolites was applied, or the Cr + PCr concentration was set to 8 mmol/kg. The shorter TEs used in the aforementioned studies, however, should not generally have a significant impact on metabolite concentrations.
Concentration of Scy, on the other hand, is lower in this study compared with other studies. However, it remains unclear whether care was taken to use all six carbon atoms, written as 1-6 CH in Govindaraju et al, 40 each having a proton resonating at the same frequency, thus contributing to their Scy basis set simulation.
Absolute concentrations in millimoles per liter, given in Supporting Information Figure S8 and Supporting Information Table S1, are in excellent agreement with the concentrations that corrected for metabolite relaxations given by Penner et al 52 and with the literature comparison presented in the same article.

| CONCLUSIONS
In this study, for the first time, T 2 relaxation times of 14 individual macromolecule peaks ranging from 13 ms to 45 ms are measured in both GM-rich and WM-rich voxels.
In addition, in vivo transverse relaxation times of 12 metabolites and metabolite moieties in a GM-rich voxel in the occipital lobe at 9.4 T are reported. The T 2 relaxation values ranged from 40 ms to 110 ms and were used as a correction factor for the absolute quantification of metabolites. The T eff 2 and T 2 values for MM peaks and metabolites, respectively, confirm the decreasing trend of transverse relaxation times with increasing static magnetic field. Finally, this work quantitatively shows the contribution of T 2 relaxation times and B 0 components to the linewidth of MM peaks. The residual linewidth includes not only components of J-coupling, but also chemical-shift distributions of amino acid proton groups.