Water‐suppression cycling 3‐T cardiac 1H‐MRS detects altered creatine and choline in patients with aortic or mitral stenosis

Cardiac proton spectroscopy (1H‐MRS) is widely used to quantify lipids. Other metabolites (e.g. creatine and choline) are clinically relevant but more challenging to quantify because of their low concentrations (approximately 10 mmol/L) and because of cardiac motion. To quantify cardiac creatine and choline, we added water‐suppression cycling (WSC) to two single‐voxel spectroscopy sequences (STEAM and PRESS). WSC introduces controlled residual water signals that alternate between positive and negative phases from transient to transient, enabling robust phase and frequency correction. Moreover, a particular weighted sum of transients eliminates residual water signals without baseline distortion. We compared WSC and the vendor's standard ‘WET’ water suppression in phantoms. Next, we tested repeatability in 10 volunteers (seven males, three females; age 29.3 ± 4.0 years; body mass index [BMI] 23.7 ± 4.1 kg/m2). Fat fraction, creatine concentration and choline concentration when quantified by STEAM‐WET were 0.30% ± 0.11%, 29.6 ± 7.0 μmol/g and 7.9 ± 6.7 μmol/g, respectively; and when quantified by PRESS‐WSC they were 0.30% ± 0.15%, 31.5 ± 3.1 μmol/g and 8.3 ± 4.4 μmol/g, respectively. Compared with STEAM‐WET, PRESS‐WSC gave spectra whose fitting quality expressed by Cramér‐Rao lower bounds improved by 26% for creatine and 32% for choline. Repeatability of metabolite concentration measurements improved by 72% for creatine and 40% for choline. We also compared STEAM‐WET and PRESS‐WSC in 13 patients with severe symptomatic aortic or mitral stenosis indicated for valve replacement surgery (10 males, three females; age 75.9 ± 6.3 years; BMI 27.4 ± 4.3 kg/m2). Spectra were of analysable quality in eight patients for STEAM‐WET, and in nine for PRESS‐WSC. We observed comparable lipid concentrations with those in healthy volunteers, significantly reduced creatine concentrations, and a trend towards decreased choline concentrations. We conclude that PRESS‐WSC offers improved performance and reproducibility for the quantification of cardiac lipids, creatine and choline concentrations in healthy volunteers at 3 T. It also offers improved performance compared with STEAM‐WET for detecting altered creatine and choline concentrations in patients with valve disease.

concentrations. We conclude that PRESS-WSC offers improved performance and reproducibility for the quantification of cardiac lipids, creatine and choline concentrations in healthy volunteers at 3 T. It also offers improved performance compared with STEAM-WET for detecting altered creatine and choline concentrations in patients with valve disease. 3 T, cardiac, heart, human, PRESS, STEAM, water-suppression cycling 1 | INTRODUCTION Magnetic resonance spectroscopy ( 1 H-MRS) is an established method to assess metabolite concentrations in the human heart. In diseases such as obesity and type 2 diabetes, the heart's uptake and oxidation of fatty acids are not balanced, resulting in an accumulation of triglycerides. 1 The accumulation of triglycerides leads to metabolic derangement, which has been proposed as a contributory and/or perpetuating factor in nonischaemic cardiomyopathies. 1,2 Therefore, most published cardiac 1 H-MRS studies have focused on quantifying lipid concentrations. [3][4][5] with sufficient data quality, 1 H-MRS can also assess lower concentration metabolites such as creatine (Cr) and choline* (Cho) that are of high clinical interest. [6][7][8] Cr is an important component in the creatine kinase (CK) enzyme system. 9 CK catalyses the transfer of a high-energy phosphate group from adenosine triphosphate (ATP) onto Cr to form phosphocreatine (PCr) and adenosine diphosphate (ADP) in the mitochondria: ATP + free Cr ⇌ ADP + PCr: In the myofibrils, the reverse reaction provides a very rapid supply of ATP to meet demand during bursts of effort such as sustaining contraction throughout systole. 10,11 These ATP-delivery processes become deranged in heart failure, according to the energy-starvation hypothesis. 12 In failing human and animal hearts, free Cr and PCr concentrations drop substantially soon after disease onset 13 and before failure of ATP homeostasis. 9,14 Note that, in vivo, the free Cr and PCr 1 H-MRS peaks are overlapping. It is therefore common to report 'total creatine' (tCr), defined as the sum of free Cr and PCr. Cr concentrations obtained via 1 H-MRS can be used to calculate the heart's energy reserve (jΔG ATP j), 15,16 which in turn can predict isovolumic contractile reserve 17 and peak cardiac work. 18,19 Hence, a decrease in jΔG ATP j limits the heart's maximum mechanical work and, eventually, a reduced jΔG ATP j could lead to heart failure. On top of that, jΔG ATP j is a potential biomarker that could be used to evaluate novel 'energy-sparing' treatments for heart failure, for example, drugs that modulate fatty acid oxidation, such as etomoxir 20 and perhexiline. 21 Cholines are a class of quaternary ammonium salts that serve several physiological functions such as maintaining structural integrity in cell membranes and facilitating cell membrane signalling. 22 The Cho derivative acetylcholine is an important neurotransmitter in the autonomic nervous system, which regulates heart rate and blood pressure, 23 and in the brain, where defective cholinergic signalling is implicated in Alzheimer's dementia. 24 Cholines have also been identified as a potential biomarker for identifying active tumours, for example, in the brain, 25 breast 26 and liver. 27 However, whether myocardial choline levels change in disease is still poorly understood in both animal models and humans.
To date, most in vivo cardiac 1 H-MRS studies have used single-voxel spectroscopy (SVS) techniques such as STimulated Echo Acquisition Mode (STEAM) and Point RESolved Spectroscopy (PRESS) together with a pulse sequence module that suppresses water signals. 28 Patient motion, including respiration and cardiac motion, mean that there can be phase and frequency shifts between each transient recorded by the scanner. This can cause incoherent averaging during postprocessing, and hence reduce signal-to-noise ratio (SNR). A common approach for both cardiac and brain spectroscopy involves phasing and frequency shifting individual transients using the largest peak (residual water or a metabolite). 3,[28][29][30][31][32] To ensure that there is a sufficiently large reference peak for frequency and phase correction, many studies only partially suppress water (so-called 'weak water suppression'). 28,29 However, with weak water suppression, the residual water peak can still cause problems during spectral fitting. 33 In this work, we propose instead to use a water-suppression cycling (WSC) method, 34 originally developed for brain 1 H-MRS, to improve the quality of cardiac single-voxel 1 H-MRS at 3 T. The WSC technique inserts a weak water suppression module and adds an inversion pulse in alternate transients. The weak water suppression property of this module provides a high SNR residual water peak for frequency and phase correction, while inverting the residual water signal every other scan allows for practically complete elimination of the residual water peak during postprocessing (as we show below), which improves spectral fitting (e.g. by avoiding errors from baseline distortion). WSC is similar conceptually to the method known as 'metabolite cycling', 7,35 except that during metabolite cycling it is the metabolite peaks that are inverted, whereas WSC only inverts the residual water peak.
Here, we aimed to test the feasibility of the WSC approach to improve quantification of low-concentration metabolites such as Cr and Cho (as well as lipids) in the heart at 3 T; and to investigate potential changes in cardiac Cr and Cho levels in patients with valvular stenosis. This is a particularly exciting, but as of yet underinvestigated group of patients, in which 15% of patients undergoing surgical aortic valve replacement have reduced left ventricular ejection fraction (LVEF; <50%). [36][37][38] However, at present it is not possible to anticipate or explain why the myocardium in some patients, but not others, exhibits contractile decline in the face of sustained severe pressure overload. Nonetheless, otherwise unexplained contractile decline into the borderline (LVEF 50%-59%) range worsens prognosis [38][39][40] and, in the absence of good understanding regarding the underlying mechanism, there is currently no specific medical therapy other than to refer patients for timely valve replacement once the valve disease has progressed into the severe range. The metabolic hypothesis of heart failure posits that reduced metabolic reserve is both a permissive and a causative factor at play, and while there are many animal models that support this, 41,42 the evidence in human pressure overload hypertrophy is sparser 13,43 because of the difficulty in obtaining human myocardium, and also because there are few established techniques for noninvasive myocardial metabolic phenotyping. Despite the known association between reduced PCr/ATP ratio and pathological left ventricular hypertrophy, there is mixed evidence regarding whether tCr itself is reduced in this setting, [43][44][45][46][47] yet this would be a key variable to establish if one is to estimate the concentration of free ADP from the CK equilibrium expression and thereby come to an estimate of the free energy of ATP hydrolysis.

| Water-suppression cycling
The water-suppression cycling (WSC) module was first implemented by Ernst and Li for 1 H-MRS in the brain. 34 This technique alternates between positive and negative residual water signal every other transient. The residual water signal can be used to correct for phase and frequency shifts in single shots and its alternating phase allows for the elimination of the water peak upon postprocessing.
The implementation of the pulse sequence and the postprocessing of each transient is presented in Figure 1. WSC is an interleaved method.
The WSC module comprises four chemical shift selective RF pulses spaced 50 ms apart for partial water suppression prior to a standard singlevoxel pulse sequence such as PRESS or STEAM. Each RF pulse is a Gaussian with a time-bandwidth product of 0.896 centred on water, and the spoiler gradients between pulses suppress residual transverse magnetisation (M xy ). 34 The third pulse in the module, a 180 chemical shift selective inversion pulse, is activated in odd-numbered acquisitions and disabled in even-numbered acquisitions, yielding two different spectra, S odd and S even . This alternation results in the residual water and metabolite peaks having the same phases in S odd and opposite phases in S even .
Three corrections were applied before the odd-and even-numbered acquisitions were averaged together. First, frequency correction was performed to correct for minor frequency differences between the water signals in each transient. Second, a zero-order phase correction was performed based on the phase of the residual water peak to make it positive real in even transients and negative real in odd transients. Third, a weighting factor w, which compensates for T 1 relaxation of the water signal and imperfections in the inversion pulses, was determined by least-squares fitting, such that spectra summed according to Equation 2 have minimum residual water signal. Finally, the odd and even spectra were averaged using Equation 2, where the denominator ensures that metabolite amplitudes remain unchanged (see Figure S1 for simulation results): In many ways, the WSC approach is similar to the method of metabolite cycling. 35 However, WSC has potential advantages, such as the residual water peak is eliminated before spectral fitting, and that because the metabolites are not inverted, WSC may be less sensitive to the T 1 values of metabolites and the efficiency of the inversion pulse, as shown in Ernst and Li 0 s original study. 34

| METHODS
All data were collected using a whole-body clinical 3 T Prisma MRI scanner (Siemens Healthineers, Erlangen, Germany) with an 18-channel cardiac receive array coil (Siemens Healthineers) and a 32-channel spine receive array coil (Siemens Healthineers).

| 1 H MRS sequence
We compared the vendor's standard 'water suppression enhanced through T 1 effects' (WET) 48 water-suppression module against the WSC module described above. This gave four different protocols: To minimise the loss of signal due to T 2 decay, we also modified the gradient slew rates and amplitudes to achieve shorter echo times for all four sequences (26 ms for PRESS and 10 ms for STEAM). On top of that, to improve performance of the vendor's WET water suppression for in vivo scans, we followed the procedure of Rial et al. 3 and used an additional breath-held prescan to manually optimise the water-suppression flip angles.

| Phantom study
A phantom study was conducted to validate the ability to quantify Cr concentrations for each sequence together with its postprocessing steps in the absence of motion. A six-compartment phantom was prepared in a silicone ice-cube tray with varying concentrations of Cr ( Figure S2). The
For each session, the volunteers were positioned head-first supine in the scanner with ECG gating (Siemens). Cine images were used to obtain horizontal long-axis and short-axis views of the heart. The MRS voxel was placed in the midinterventricular septum, to minimise the chance of contamination from epicardial fat. The trigger delay was adjusted for mid-diastole, where cardiac motion is minimised, 49 Table S1 for SNR, Cramér-Rao lower bounds [CRLB] and linewidths for the preliminary study). Thus, in the main study, three sequences were run on the volunteers: STEAM-WET, STEAM-WSC and PRESS-WSC. Parameters for all three sequences matched the phantom experiment, with the following changes: 2 s TR for water-suppressed acquisitions and the water-suppression bandwidth for the WSC module was set at 65 Hz. For STEAM-WET, 150 water-suppressed spectra were acquired over 30 breath-holds and three nonwater-suppressed spectra were acquired over one breath-hold to give a high-quality 'reference' dataset.
For STEAM-WSC and PRESS-WSC, 60 water-suppression cycled spectra (i.e. 30 pairs) were acquired over 10 breath-holds and three waternonsuppressed spectra were acquired over one breath-hold. Across all sequences, each breath-hold was 11-13 s long, and a 13-20 s pause was given between each breath-hold. Each session took around 1 h in total.
To obtain repeatability data, each volunteer was scanned in two different sessions on the same day. They were taken out of the scanner between sessions, given a 5-min break, and then the scan protocol was repeated. The total scan time for each volunteer was just over 2 h including the repeat session. However, the large number of breath-holds and long scan times were challenging for some volunteers. Based on an interim analysis of data from the first session in the first three volunteers, the STEAM-WSC was not repeated in the second session for the remaining volunteers.

| In vivo study in patients
Thirteen patients (10 males, three females; age = 75.9 ± 6.3 [range  years; BMI = 27.4 ± 4.3 kg/m 2 ) with severe symptomatic aortic or mitral stenosis were recruited as part of a project approved by regional (South Central Oxford C, 16/SC/0323) and local ethics and governance panels. Out of the group, eight patients had aortic stenosis with preserved LVEF, three had aortic stenosis with reduced ejection fraction and one had mitral stenosis. The inclusion criteria were age 18-85 years, with severe aortic stenosis and due to undergo clinically indicated aortic valve replacement, or no pressure-or volume-loading valve disease (i.e. no hypertrophy) but due to undergo cardiac surgery and willing to donate cardiac biopsy. The exclusion criteria were prior myocardial infarction, flow-limiting coronary disease, more than mild bystander valve disease, renal impairment (estimated glomerular filtration rate < 30 mL/min), or contraindication to MRI scanning. A table summarising the patient demographics is included in the supporting information (Table S2).

| Data postprocessing
Signals from individual coil elements for each shot were combined and averaged using MATLAB (Mathworks, Natick, MA, USA). Weighting and phase-correction factors for each coil were calculated from the water peak in the nonwater-suppressed spectra and applied to the watersuppressed spectra. 3,58 The spectrum from each transient was phase-and frequency-corrected based on either the lipid peak (for STEAM-WET and PRESS-WET) or the residual water peak (STEAM-WSC and PRESS-WSC) before averaging. A more detailed explanation of the postprocessing steps for WSC sequences can be found in the Theory section and in Figure 1. All spectra were analysed using the OXSA toolbox. 59 For the Cr-fat phantom, five Lorentzian peaks were fitted: residual water around 4.7 ppm, Cr CH 2 peak around 3.9 ppm, Cr CH 3 peak around 3.0 ppm and two lipid peaks around 1.28 and 0.84 ppm. For in vivo WET water-suppressed spectra, residual water signals were removed by the Hankel-Lanczos singular value decomposition (HLSVD) method. 60,61 The resulting spectra were fitted with seven peaks: residual water, two peaks for the methyl and methylene group of tCr at 3.0 and 3.9 ppm; a peak for the methyl group of Cho derivatives at 3.2 ppm; and three peaks for lipids chains at 1.99 ppm (allylic), 1.28 ppm (methylene) and 0.84 ppm (methyl). 4 Note that although we fit two peaks for Cr, we used only the Cr CH 3 signal at 3.0 ppm for analysis of tCr. In this paper, we use 'choline' to refer to any trimethylammonium (TMA) compound, as used by Gillinder et al. 6 ; however, the peak is likely to include contributions from other nitrogen-containing compounds such as carnitine or taurine. Each raw transient, intermediate outputs from the postprocessing pipeline, and the final fittings, were all visually inspected for potential errors.
Across all spectra, the SNR of each peak was estimated by taking the ratio of the peak height in a 10-Hz exponentially apodised spectrum (matched filtering) to the SD of the noise extracted from the Fourier Transform of the last 100 points of the free induction decay of the filtered spectrum. 62 Note that care must be taken when comparing spectroscopy SNR values between studies because there are many slightly different definitions of SNR. The accuracy of the spectral fitting was estimated by calculating the CRLB. 63 Due to time constraints, we were unable to run 150 breath-holds for all methods. Thus, to enable a fair comparison, we also randomly selected 12 breath-holds from the STEAM-WET-150 data to give a 60-measurement subset. For clarity, we refer to these subsampled data as 'STEAM-WET-60'. This subsampling was performed for STEAM-WET-150 data in both healthy volunteers and patients.

| Metabolite quantification
Fitted areas for each metabolite A were corrected for T 1 and T 2 decay and proton density according to: where S A refers to the uncorrected signal peak integral, S A * refers to the corrected signal peak integral, F 1, A and F 2, A are T 1 and T 2 correction factors and N A is the proton density.
For the phantom, the T 1 values were determined to be 0.991 s for the Cr CH 3  For in vivo acquisitions, Cr and Cho concentrations were calculated in μmol per gram of wet tissue (μmol/g) according to: The concentration of pure water was taken to be 55.5 mol/L and the myocardial tissue was assumed to contain 72.7% water by weight. 51 Myocardial lipid content is represented by the corrected ratio of the lipid methylene peak (1.28 ppm) area to the unsuppressed water peak area, expressed as a percentage. 3

| Statistical analysis
Sequence performance was assessed by comparing SNR, CRLB and linewidths of the three main metabolite peaks: lipid methylene peak, Cr CH 3 peak and Cho peak. The values were described using medians and interquartile ranges, and paired Wilcoxon signed-rank tests (two-tailed) were conducted across all possible pairings for SNR and CRLB in healthy volunteers. Metabolite concentrations in both healthy volunteers and patients were described using means and SDs with outliers identified as points more than three scaled median absolute deviations (MAD) away from the median and were excluded from the analysis.
Intersubject coefficients of variation (σ i 2 ) in metabolites were calculated for healthy volunteers by dividing the SD of the metabolite concentrations by the overall mean. Repeatability of STEAM-WET and PRESS-WSC were compared by calculating the intrasubject coefficients of variation (CV) and coefficients of repeatability (CR) for Cr, Cho and lipid quantification. The intrasubject CV can be calculated by taking the ratio of the SD of within-subject measurements (i.e. the SD of the difference between measurements 1 and 2, also known as σ w 2 ) to the overall mean. 66 The CR is the value below which the absolute difference of the two measurements would lie with 95% probability and is calculated by multiplying σ w 2 by 2.77 ( ffiffiffi 2 p × 1:96). 67 Power calculations were also performed using repeatability data from healthy volunteers with G*Power (version 3.1.9.7; Heinrich Heine Universitat, Dusseldorf, Germany). 68,69 The two-tailed a priori power calculation was performed with the following parameters: test family: 't Lastly, unpaired Student's t-tests (two-tailed, equal variance assumed) were conducted to analyse the differences in metabolite concentrations between healthy volunteers and patients. The metabolite concentrations in healthy volunteers were averaged across both sessions.

| Phantom study
The correlation between the measured and actual Cr concentration in the phantom for the four sequences is shown in Figure 3.  Table S2.  Table 1, while SNR, CRLB and linewidths are presented in Table S3. Lastly, randomisation of the choice of breath-holds for STEAM-WET-60 did not have any significant effects on SNR, CRLB or linewidths ( Figure S4).

| Metabolite concentrations in healthy volunteers
Myocardial metabolite concentrations in both healthy volunteers and the patient group are shown in Figure 6.

| Repeatability in healthy volunteers
Repeatability results are shown in Table 2 and Bland-Altman plots in Figure 7. As an illustration of the power of 3-T 1 H-MRS to assess Cr and/or Cho in clinical studies, Table 3 compares the calculated sample sizes needed to detect a 20% change in metabolite levels for paired and independent studies for both sequences. Sample sizes for independent studies are reported as the numbers of participants per group, assuming that only 60% of subjects have quantifiable spectra.

| Demonstration in patient cohort
All patients completed the examination. In one patient, the number of measurements per breath-hold had to be reduced to four to shorten the length of the breath-hold. Five patient datasets were discarded as they either had fitted linewidths of less than 2 Hz for any of the metabolites, For the current study, data points more than 1.5 interquartile ranges above the upper quartile or below the lower quartile are treated as outliers (denoted in black). The black cross and the attached whiskers represent the mean and the standard deviations across all healthy volunteers. The p-values from the unpaired Student's t-tests are also provided. AS, aortic stenosis; CHF, chronic heart failure which indicates that there was only noise in the relevant spectral region, or because of a CRLB value for the lipid peak of more than 500% in either of the acquisitions. All five had poor STEAM-WET data; four out of five also had poor PRESS-WSC data.
The remaining eight datasets consisted of seven patients with aortic stenosis with preserved systolic function and one patient suffering from aortic stenosis with impaired systolic function. Figure 8 shows spectra from a patient compared with those from a healthy volunteer.
T A B L E 2 Intersubject coefficient of variation (σ i 2 ), intrasubject coefficient of variation (σ w 2 ) and coefficient of repeatability for lipids, creatine and choline quantification in healthy volunteers (n = 10)   F I G U R E 8 Spectra from a healthy volunteer (left, both A and C) compared with that of a patient (right, both B and D) suffering from aortic stenosis. The spectra were acquired with PRESS-WSC in a healthy volunteer (A) and a patient (B), and with STEAM-WET-150 in a healthy volunteer (C) and a patient (D). All spectra are scaled by the corresponding water peak amplitude. A decrease in creatine (Cr) is seen between these pairs 5 | DISCUSSION 5.1 | Sequence performance

| Phantom
Our phantom results show that all four sequences have the potential to quantify Cr levels in the heart in vivo. In the main in vivo study, PRESS-WSC showed increased fitting accuracy (i.e. lower CRLB values) and improved repeatability compared with the longer 'gold standard' STEAM-WET-150 approach, despite having less than half the number of measurements. These improvements translate to decreases in the number of subjects that would be needed to detect a specified biological effect on Cr or Cho metabolism, as shown by the power calculations in Table 3. They also indicate that PRESS-WSC may be a shorter (and hence more patient-friendly) alternative to STEAM-WET for cardiac 1 H-MRS.

| Healthy volunteers
PRESS-WSC also outperformed both STEAM-WET and STEAM-WSC in terms of the raw SNR for matched scan durations. It is tempting to attribute the improvement in SNR seen in PRESS-WSC to the difference in signal intensity between PRESS and STEAM. PRESS acquires a spinecho signal that inherently has twice the signal intensity compared with a stimulated echo signal acquired by STEAM and this is clearly illustrated in the phantom dataset. However, perhaps surprisingly, PRESS-WET showed inferior performance in both SNR and CRLB compared with STEAM-WET in the preliminary in vivo study ( Figure S3). This indicates that SNR is controlled by other factors in vivo, for example, differences in sensitivity to motion and diffusion effects between the two SVS sequences.
In terms of linewidths, both WSC sequences showed decreases in metabolite linewidths compared with STEAM-WET-150, which could be due to more accurate frequency and phase corrections based on the larger, narrower residual water peak, rather than a smaller, broader lipid peak.
However, the linewidths achieved using STEAM-WSC sequence are on average lower compared with PRESS-WSC. This could be considered as surprising because PRESS-WSC was regarded as superior to STEAM-WSC when assessed with other measures of spectral quality (e.g. SNR, CRLB and repeatability). We observed a similar trend between STEAM-WET and PRESS-WET in our preliminary experiment, thus, we hypothesise that the difference in linewidth could be due to the differences in the STEAM and PRESS sequences. For example, PRESS had a longer TE, which would increase sensitivity to small motion effects, and PRESS has 180 pulses that may show more slice edge effects than the 90 pulses in STEAM (with a consequent increase in true voxel size, and hence in linewidth). There are also more complex factors in vivo that might have contributed to the final differences, such as bulk motion and diffusion effects, which will need to be further investigated to fully understand this surprising trend.
Across all in vivo acquisitions, the bandwidth of water-suppression pulses in WSC was larger (65 Hz or 0.5 ppm) than that for WET (50 Hz or 0.4 ppm). This slightly larger bandwidth was determined through empirical observation in a series of pilot experiments. We suspect that the need for increased water-suppression bandwidth arises because the WSC module needs to either invert or not invert the water. If the water peak drifts in some transients out of the water-suppression bandwidth, then this can lead to distorted lineshapes that are not fully cancelled during the WSC postprocessing.
This increased water-suppression bandwidth means that care should be taken when quantifying metabolites with peaks close to water (e.g. the methylene peak of Cr at 3.9 ppm). It is possible that those peaks might be partially suppressed in the presence of severe shot-toshot frequency drifts. In the current study, to avoid this potential issue, we only used the peak amplitude of the Cr methyl group at 3.0 ppm for cardiac Cr quantification, and not the Cr methylene peak at 3.9 ppm. Furthermore, we note that there is no significant difference in the closer Cho signal at 3.2 ppm between PRESS-WSC and STEAM-WET in our study, suggesting that this effect is not of concern here. Another possible factor that affects metabolite quantification is the lack of preparation scans in our protocol. Although preparation scans ensure that a steady state is reached before signal acquisition, they also lengthen the duration of each breath-hold. A quick calculation using literature T 1 values and nominal flip angles shows that the error in quantification due to the slightly higher signal in the first transient is very small, accounting for percentage differences smaller than 0.75% for Cr CH 3 and 2.2% for Cho. These potential errors are modest compared with other sources of error in the fitted amplitudes and can be neglected. respect with one more patient having analysable data quality despite the much reduced scan duration. Nevertheless, we are continuing to explore ways to achieve full robustness in patients as well as in healthy volunteers. We believe the underlying issues are not sequence-specific but include the familiar issues of heart rate variability, body habitus, compliance with breath holding, age and MR scan familiarity, etc.

| Repeatability
We showed that in less than half the scan duration time (60 vs 150 measurements, or 30 vs 10 breath-holds), PRESS-WSC greatly improves intrasubject repeatability for the quantification of Cr and Cho, and that it maintains repeatability for lipids ( Figure 7) compared with STEAM-WET.
For the repeatability of lipid quantification, Rial et al. reported an intrasubject CV of 19% using a STEAM sequence with breath-holding at 3 T. 3 There have also been a number of previous studies published at 1 From both the Bland-Altman plots ( Figure 7) and the CR, it can be seen that the lipid repeatability is differently influenced compared with Cr and Cho. Possible explanations for this include that the lipid peak is influenced by J-coupling. The difference in J-coupling evolution with different TEs could account for some of the difference in the repeatability changes. Alternatively, because the heart is surrounded by a layer of epicardial fat, inconsistent diaphragm positions between breath-holds and patient motion could perhaps result in some spectral transients being contaminated by epicardial fat.

| Simplicity of implementation
We note in passing that many of the prior studies in cardiac Cr and Cho quantitation have used relatively complicated motion compensation strategies. Our approach has a benefit that it is straightforward to implement because the only sequence modifications are small alterations to the water suppression module. Other changes are implemented in off-scanner reconstruction tools, which are simpler to alter. Therefore, even where our repeatability is comparable with some prior studies, this simplicity of implementation is a pragmatic benefit of the WSC approach.

| Metabolite quantification and changes in patient cohort
In both healthy volunteers and patients, the measured fat fractions and metabolite concentrations did not differ significantly between STEAM-WET and PRESS-WSC. Both sequences also gave absolute myocardial Cr and Cho concentrations and fat fractions that were similar to values from the other sequences and from previous studies. 4,6,50,51,71 There was no significant difference in myocardial lipid concentrations between patients and controls. The results were 0.30% ± 0.11% in healthy controls and 0.48% ± 0.51% in patients when measured with STEAM-WET-150 (p = 0.26), and 0.30% ± 0.15% in healthy controls and 0.25% ± 0.19% in patients when measured with PRESS-WSC (p = 0.94). This was unexpected, 73 however, Marfella et al. 74 reported that lipid contents are significantly higher in aortic stenosis patients with metabolic syndromes than in those without metabolic syndromes. Although none of our patients had diabetes mellitus, insulin resistance is a known correlate of heart failure, 75 so it is likely that our cohort consisted of a mixture of the two groups, which would be expected to attenuate differences in observed lipid concentration vs healthy volunteers. with the reduction, suggesting that the PRESS-WSC approach is better suited to assessing Cr depletion. In our patient cohort, we observed a significant decrease in Cr concentrations in patients vs volunteers with an average reduction of 21.9 μmol/g across the two sequences. Despite the small cohort scanned in this study, these findings suggest that PRESS-WSC, with its improved repeatability, is a better choice for assessing biologically relevant changes in Cr concentration than STEAM-WET.
In a study conducted at 1.5 T by Gillinder et al., the myocardial Cho water ratio was 0.24% ± 0.28% before any corrections. 6  Recently, a number of studies have linked dietary Cho intake to cardiovascular diseases in animal models. 77 Human studies to investigate this are more complex due to the significant influence of the intestinal microflora 78 ; there is no general agreement yet on the influence of dietary Cho intake on the incidence of cardiovascular diseases. [79][80][81][82][83] However, preclinical evidence suggests that Cho given as an injection has cardioprotective effects in rat models, including reducing postischaemia reperfusion injuries 84 and slowing the progression of hypertension, 85 and the whole-blood Cho level has been identified as a promising biomarker in acute coronary syndrome. 86,87 However, most of these studies focused on dietary Cho intake and/or supplements 79 without measuring any direct changes in Cho concentration in the heart. With emerging research aiming to modulate the cholinergic system and thereby target heart disease, 88,89 1 H-MRS is a valuable tool for direct measurement of myocardial Cho in vivo.

| Future works
In our analysis, we used relaxation times measured in skeletal muscles for quantifying cardiac metabolites. Although a previous imaging study has shown that T 1 and T 2 values for heart and skeletal muscles are very similar at both 1.5 and 3 T, 90 we believe that the accuracy in metabolite quantification could be improved by using relaxation values measured directly in the human heart at 3 T.
We also noted that the performance gains for PRESS-WSC were more modest in our patient cohort than in volunteers. In a future study, we intend to further optimise the WSC module. For example, it may be advantageous to simply cycle the water signal by 180 on alternate transients without adding the other water-suppression pulses in what would be a purely 'water-cycled' MRS method analogous to 'metabolite-cycled' methods. It could also be possible to reduce the spacing between water-suppression pulses or to remove the first 90 pulse in the WSC module to make the PRESS-WSC acquisition fit more easily into the shorter diastole seen in high heart-rate, arrhythmic patients, like those we scanned in the current study.
We observed decreases in both Cr and Cho levels in the patient group compared with healthy volunteers. However, we only had a limited study size to provide pathophysiology-related conclusions. Future studies including more patients with aortic or mitral stenosis are required to gain a better insight into the biochemical changes in the myocardium associated with this diseased state.

| CONCLUSIONS
We have shown that a water-suppression cycled PRESS sequence improves performance and repeatability in measuring cardiac metabolites (lipids, Cr, Cho) relative to the vendor's standard STEAM and PRESS sequences at 3 T. We quantified cardiac Cr and Cho in a 10-breath-hold protocol, which was robust in healthy volunteers. Lastly, in patients with severe symptomatic valvular heart disease listed for clinically indicated valve replacement surgery, we were able to detect changes in cardiac metabolites (Cr, Cho, lipids) in a third of the time using PRESS-WSC compared with the vendor's STEAM-WET sequence.