The human heart requires a continuous and adequate supply of energy to guarantee myocardial contractility that is required to support blood circulation. Myocardial energy homeostasis is maintained primarily by oxidative phosphorylation of adenosine diphosphate (ADP) in cardiomyocyte mitochondria. A disruption of myocardial energy homeostasis may impair mechanical function of the heart (Tewari et al., 2016b). Indeed, impaired mitochondrial function can lead to a life-threatening state of heart failure (Neubauer, 2007; Brown et al., 2016). Therefore, homeostasis of myocardial energy metabolism and its (mal-)adaptation in heart disease has been an important area of cardiovascular research (Taegtmeyer et al., 2016).

Phosphorus-31 MRS (³¹P-MRS) is a non-invasive and non-ionizing imaging modality that is uniquely capable of probing in vivo myocardial high-energy phosphate metabolism. This technique can quantify the steady-state myocardial phosphocreatine (PCr) over ATP concentration ratio (Bottomley, 2007), which has been commonly used to characterize the in vivo myocardial energy status. The PCr/ATP ratio is assumed to correlate with the cytosolic Gibbs free energy of ATP hydrolysis (ΔG_P), the energy available to cardiomyocytes to do work. However, this assumption is valid, if and only if, the myocardial creatine content is either known or can be assumed to be unchanged compared with healthy hearts (Wu et al., 2009). It has long been known that the myocardial creatine content can be reduced in the diseased heart (Cowan, 1934; Herrmann and Decherd, 1939), thus complicating a straightforward interpretation of measured PCr/ATP ratios in patients. Furthermore, the measured myocardial PCr/ATP ratio only reports on the balance between ATP turnover rate and ATP synthesis at a specific steady-state. The underlying cause of any observed difference between the PCr/ATP ratio in heart disease and in the healthy heart cannot be identified without additional measurements. Indeed, measurements of the myocardial PCr/ATP ratio at multiple steady-states or during transition between steady-states of cardiac work may unmask underlying energy deficits in heart disease (Dass et al., 2015). Furthermore, such measurements would allow for a meaningful characterization of the (patho-)physiology of in vivo myocardial energy homeostasis guided by computational modeling and simulations of cardiomyocyte energy homeostasis (Balaban, 2006; Beard and Kushmerick, 2009), which ultimately may facilitate diagnosis and risk stratification in patients.

Obtaining reliable dynamic PCr/ATP ratios from the human heart during transitions between cardiac work rates is unrealistic (van Beek et al., 1998). Instead, ³¹P-MRS measurements of the myocardial PCr/ATP ratio at multiple steady-states of cardiac work rates are feasible. Pioneered almost three decades ago (Conway et al., 1988, 1991; Weiss et al., 1990; Kuno et al., 1994), ³¹P-MRS measurements of the in vivo human heart during exercise have recently regained interest (Hudsmith et al., 2009; Betim Paes Leme et al., 2013; Dass et al., 2015; Levelt et al., 2016). These studies typically consisted of steady-state ³¹P-MRS data acquisition at rest and at one additional steady-state of low-intensity exercise (heart rates of 60–100 beats min⁻¹) or pharmacologically induced stress (Figure 1). Multiple measurements over a broader physiological range of cardiac work rates would be of major benefit for characterizing myocardial energy homeostasis. Indeed, successful implementation of more strenuous exercise regimens in clinical cardiac ¹H-MRI protocols has recently been reported (La Gerche et al., 2013; Pflugi et al., 2015; Roberts et al., 2015; Barber et al., 2016), with maximal heart rates during supine in-magnet bicycle exercise exceeding 160 beats min⁻¹ (La Gerche et al., 2013). However, ³¹P-MRS suffers from low sensitivity and poor measurement repeatability compared to ¹H-MRI, compromising a quantitative evaluation of potentially subtle changes in myocardial energy homeostasis.

Here, we investigated whether the current standard of ³¹P-MRS methodology to measure in vivo myocardial PCr/ATP ratios typically implemented on clinical research MR systems is sufficient to discriminate between exercise-induced changes in steady-state myocardial energy metabolism in health and disease. Hereto, we determined and compared the precision in terms of measurement variability and repeatability of commonly used localized ³¹P-MRS methods to quantify the myocardial PCr/ATP ratio in healthy subjects at rest. To compare the precision of cardiac ³¹P-MRS measurements with another in vivo application of ³¹P-MRS, we also determined the precision of ³¹P-MRS measurements of the PCr/ATP ratio in stationary calf muscle. The results were then used in computational model simulations of the healthy heart and of hypertrophic heart disease, to estimate the magnitude of change that may be expected for the myocardial PCr/ATP ratio over a broad physiological range of cardiac work rates. Our findings show that improvements of the ³¹P-MRS measurement precision combined with in-magnet exercise at high intensities will be required for such investigations to become of diagnostic merit.

We acquired ³¹P-MR spectra of the human heart using three approaches for localized signal acquisition (Figure 2). All spectra feature the distinct resonance peak of the high-energy phosphate PCr (chemical shift reference at 0.00 ppm). The three phosphate groups in ATP (α-, β-, and γ-ATP) are reflected by three multiplets at different chemical shifts upfield of PCr. Phosphodiesters (PDE) give rise to the peak at 3 ppm. Two peaks associated with 2,3-diphosphoglycerate (2,3-DPG) in erythrocytes in the ventricular blood appear further downfield of PCr. These peaks overlap with P_i resonating at a pH-dependent chemical shift of ~5 ppm relative to PCr. Contamination of the spectra with signal from 2,3-DPG in the blood prevented estimations of myocardial pH using the P_i-PCr chemical shift difference. Non-localized saturation recovery experiments of the chest yielded T₁ relaxation time constants for high-energy phosphate metabolites at 3 Tesla, and were 4.9 s for PCr, 1.9 s for γ-ATP, 2.7 s for α-ATP, and 3.1 s for β-ATP. These values were used to correct the observed PCr/γ-ATP ratios for partial saturation effects at the heart rate-dependent TR of localized ³¹P-MRS acquisitions.

The non-invasive nature of localized ³¹P-MRS makes this technique a candidate modality for measurements of in vivo human myocardial energy metabolism during cardiac stress tests. Despite this promising potential, practical and technological challenges have prevented ³¹P-MRS from becoming a widespread diagnostic imaging modality in the clinical workflow. Here, we determined the resting-state measurement variability and repeatability for three commonly used approaches for localized ³¹P-MRS of the human heart, and employed computational modeling to estimate their suitability for assessments of myocardial energy homeostasis over a broad range of cardiac work rates. Our results show that with the level of precision achieved by current methodology, altered energy homeostasis under pathophysiologic conditions such as decreased mitochondrial capacity or inefficient sarcomere energy utilization may not be detectable with cardiac ³¹P-MRS stress testing.

For the healthy human myocardium at rest, the literature mean value of the PCr/ATP ratio is ~1.7 ± 0.3 (Bottomley, 2007). However, normal myocardial PCr/ATP values reported for healthy subjects range from 0.9 ± 0.3 up to 2.5 ± 0.5, demonstrating a large variability among research sites that use different ³¹P-MRS methods for quantification of the human myocardial energy status (Figure 1, also comprehensively reviewed in Bottomley, 2007). Our results were corrected for heart rate-dependent partial saturation effects that could modulate the PCr/ATP ratio for our measurements at relatively short TR (i.e., TR < 5 × T₁). Nonetheless, for the three localization approaches applied in the same subjects, normal PCr/ATP ratios in the present work ranged from 1.57 ± 0.17 for single-voxel 3D ISIS up to 2.78 ± 0.50 for multi-voxel 1D CSI with 1D ISIS slice selection perpendicular to the surface coil. Such differences between methods may be attributed to different degrees of signal contamination. Indeed, single-voxel 3D ISIS generally benefits from a well-defined voxel shape (de Graaf, 2008), whereas CSI may suffer from Fourier bleeding that introduces signal contamination originating from tissue in voxels outside the region of interest (Keevil, 2006). Particularly in 1D CSI oriented perpendicular to the surface coil, Fourier bleeding of signal from high PCr levels present in superficial chest skeletal muscle may contribute to an overestimation of the actual myocardial PCr/ATP ratio, compromise measurement precision, and ultimately hamper the detection of changes in myocardial PCr levels. Moreover, experimental variation in surface coil placement and voxel positioning negatively affects measurement repeatability, compromising the applicability of current ³¹P-MRS methodology for diagnostic cardiac stress testing.

Only few laboratories have reported on method repeatability (Bland and Altman, 1986) of human cardiac ³¹P-MRS in test-retest study designs. Lamb et al. compared several signal acquisition localization schemes at 1.5 Tesla (Lamb et al., 1996), and found that the inter-examination repeatability coefficient for the PCr/ATP ratio was rather large: >45% for 1D CSI, 1D CSI with 2D ISIS, as well as for 3D ISIS. This was predominantly attributed to differences in coil placement and other practicalities between examinations rather than true physiological changes in the myocardium (e.g., of nutritional origin). Using 1D CSI, Schaefer et al. reported a test-retest repeatability coefficient of 22% for measurements of the human myocardial PCr/ATP ratio at 1.5 Tesla (Schaefer et al., 1992). The use of magnetic field strengths >1.5 T holds promise in terms of improved signal to noise ratios and/or shorter acquisition times, which is theoretically beneficial for signal quantification and therewith measurement repeatability. A repeatability coefficient of 53% was reported for ³¹P-MRS measurements of the PCr/ATP ratio with 31 min of acquisition time using 3D CSI at 3 Tesla (Tyler et al., 2009). Later, this protocol was adjusted by Dass et al. to achieve a clinically acceptable acquisition time of 8 min by lowering the 3D CSI spatial resolution and omitting cardiac triggering (Dass et al., 2010), but with unreported measurement repeatability. Similar to these reports in the literature, we found rather large repeatability coefficients of more than 40% for the myocardial PCr/ATP ratios obtained within 7 min of acquisition time. Clearly, these data suggest that with the strategies currently used for cardiac ³¹P-MRS, only large changes in the PCr/ATP ratio may be detected in the human myocardium. As such, ³¹P-MRS measurement of in vivo myocardial PCr/ATP ratio in humans has not evolved beyond its use as a research tool to study myocardial energy homeostasis in groups of patients with phenotypic cardiomyopathy (Lamb et al., 1999; Dass et al., 2015; Levelt et al., 2016).

Our simulations showed that a 50% reduction of the mitochondrial capacity to produce ATP results in only a small decrease of the PCr/ATP ratio at increased cardiac work rates. Due to a lack of nutrients and oxygen, mitochondrial ATP production may become marginal in ischemic conditions, leading to a more pronounced decrease of the PCr/ATP ratio during exercise. Indeed, in some cases, cardiac ³¹P-MRS exercise stress testing has provided encouraging results. Particularly, a transient exercise-induced decrease in the myocardial PCr/ATP ratio was observed in patients with coronary artery disease, which could not be detected in patients with non-ischemic heart disease (Weiss et al., 1990). This response improved after successful revascularization, suggesting clinical potential for this methodology in myocardial ischemia. Similarly, cardiac ³¹P-MRS exercise stress testing has been proposed as a means to noninvasively test therapeutic strategies in Chagas disease, where a reduction in the PCr/ATP ratio may be indicative of microvascular disease caused by a Trypanosoma cruzi parasite infection (Betim Paes Leme et al., 2013).

Furthermore, our simulations indicate that the normal myocardial P_i concentration is tightly regulated and maintained within a submillimolar range over the entire physiological range of cardiac work rates. In contrast, for both models of HCM, cytosolic P_i was predicted to increase to millimolar concentration levels approaching 2 mM at high cardiac work rates. Indeed, a rise in P_i has been reported for patients with hypertensive heart disease after pharmacologically induced stress (Lamb et al., 1999). Moreover, the HCM models predicted that myocardial creatine depletion (Cowan, 1934; Herrmann and Decherd, 1939) progressively aggravates the loss of cytosolic P_i concentration homeostasis. Combined, our simulations support the quantitative hypothesis of cytosolic P_i interference with myocardial mechanical function as proposed by Tewari et al. (2016a), which may explain progressive heart failure in human HCM. Notably, the predicted rise of cytosolic P_i concentrations into the millimolar range for higher cardiac work rates in HCM makes this metabolite an alternative target for diagnostic in vivo detection with ³¹P-MRS. This will, however, require major methodological improvements in terms of signal localization to prevent any contaminating signal from 2,3-DPG in the blood overlapping with P_i.

Finally, our model predictions showed that ³¹P-MRS measurements of the steady-state PCr/ATP ratio at high-intensity cardiac work rates are more sensitive to pathophysiological derangements in myocardial energy homeostasis than resting-state measurements. Protocols to perform such strenuous physical exercise inside a clinical MR scanner as an alternative to pharmacologically induced stressors have recently been developed (Jeneson et al., 2010; Gusso et al., 2012) and applied to study heart function (La Gerche et al., 2013) and perfusion (Pflugi et al., 2015) at heart rates > 160 beats min⁻¹. These protocols utilize a supine cycling exercise regime rather than isometric hand grip exercise (Weiss et al., 1990) or prone flexion of the legs (Hudsmith et al., 2009), and therefore facilitate a broad range of cardiac work rates. However, cycling motion of the legs combined with higher respiratory and heart rates may aggravate motion artifacts and signal contamination in localized ³¹P-MRS, deteriorate the ECG signal typically used for synchronizing measurements with the beating heart, and introduce magnetic field inhomogeneities that can compromise data quality. These aspects make obtaining quantitative results with ³¹P-MRS during high-intensity exercise even more challenging than at resting-state conditions, and are obviously detrimental to measurement precision. Moreover, strenuous exercise cannot be maintained at a steady-state level for a prolonged period of time, particularly in case of myocardial ischemia or other pathophysiological conditions, which imposes practical constraints on data acquisition time. Ongoing developments in coil design for radiofrequency transmission and signal reception (El-Sharkawy et al., 2009; Rodgers and Robson, 2015; Schaller et al., 2015; Löring et al., 2016), MR sequence design (Robson et al., 2005), and subsequent data processing (Zhang et al., 2013) for ³¹P-MRS may alleviate these issues by increasing sensitivity, spatial localization, and decreasing acquisition time. In addition, clinical MR scanners with a magnetic field strength of 7 Tesla are becoming more widely available and promise higher signal to noise ratios and a potential for higher spectral resolution (Stoll et al., 2016). Indeed, spectra with a signal to noise ratio similar to those acquired in 30 min at 3 Tesla were acquired in only 6 min at 7 Tesla, aided by the shorter longitudinal T₁ relaxation times for high-energy phosphate metabolites at 7 Tesla (Rodgers et al., 2014). On the other hand, higher magnetic field strengths require even more demanding solutions for minimizing magnetic field inhomogeneities and motion-induced artifacts that could diminish the theoretical gain in sensitivity and/or spectral, spatial, and temporal resolution.

The current work emphasizes the need for technological and methodological advancements of cardiac ³¹P-MRS. In addition, improvements are required for an experimental validation of computational model predictions of human myocardial energy homeostasis and its (mal-)adaptation in disease. Currently, such validations have been limited to in vitro assays and in vivo studies with animal models (Wu et al., 2008, 2009). Further developments in ³¹P-MRS methodology may lead to opportunities for in vivo model validation, and ultimately for cardiac ³¹P-MRS exercise stress testing to become of any diagnostic merit.

Simulations of human myocardial energy homeostasis over a broad range of cardiac work rates predict only moderate changes in the PCr/ATP ratio, even for hypertrophic cardiomyopathy at high-intensity work rates. The present study shows that current measurement precision of commonly used localized ³¹P-MRS methods for quantification of the myocardial PCr/ATP ratio precludes the detection of such changes. This prevents using ³¹P-MRS for diagnostic testing and risk stratification in the clinic. As such, these results warrant further developments in ³¹P-MRS methodology combined with more strenuous exercise stress testing protocols to facilitate in vivo cardiac ³¹P-MRS exercise stress testing of myocardial energy metabolism in patients.

AB, JB, DB, and JJ contributed to the design of the study. AB acquired and analyzed the cardiac ³¹P-MRS data. JB provided the computational model simulations. All authors contributed to the interpretation of the data. AB, JB, and JJ drafted the manuscript. AN, GS, SB, DB, and JJ critically revised the work for important intellectual content. All authors approved the final version of the manuscript, and agree to be accountable for the content of the work.