Introduction

Phosphorus (31P) magnetic resonance (MR) is a unique non-invasive tool for studying muscle physiology and has been used extensively for studying diseases that adversely affect muscle function, such as heart failure1, stroke, congenital myopathies2 and ischemic heart disease3. A unique feature of saturation transfer (ST) 31P-MR techniques is their ability to measure the kinetics of important metabolic reactions that involve phosphorus containing metabolites4. High-energy phosphate is reversibly transferred between adenosine triphosphate (ATP) and phosphocreatine (PCr) through the creatine kinase (CK) reaction. Disturbances in the kinetics of the CK reaction are found in many disease that affect myocardial energy metabolism5,6. However, in the skeletal muscle the physiological significance of alterations on the CK exchange rate are less understood.

The kinetics of the CK reaction have been studied extensively using surface radiofrequency (RF) coils and, either unlocalized7 or localized methods with limited tissue coverage8. However, such methods lack the ability to measure muscle-specific energy metabolism and therefore cannot identify potential heterogeneneities in muscle function in health and disease. The development of relatively high-resolution and large tissue coverage imaging approaches for identifying spatial heterogeneities in the kinetics of the CK reaction may enable a better understanding of the CK role in the muscle.

A major roadblock for the development of high-resolution 31P-MR methods with sufficient tissue coverage is the low MR sensitivity of the 31P nucleus and the relatively low concentration of 31P containing metabolites in the human tissue. The increased availability of ultra-high field (UHF) whole body magnets (≥7.0 T) has allowed the development of time efficient 31P-MR methods that provide adequate tissue coverage and significantly increased spatial resolution9,10,11. However, only a few tens of UHF systems are currently operational worldwide12, while most of the clinical MRI work is performed almost exclusively in fields of 3.0 T or lower. Therefore, it is equally important to develop 31P-MR tools for the diagnosis of complications in the skeletal muscle in fields of 3.0 T or lower.

In this study, we aimed at developing and implementing a three-dimensional saturation transfer (ST) 31P-MRI method capable of mapping the CK reactions kinetics in the entire muscle of the lower leg on a clinically approved 3.0 T whole body magnet. Our method maps the unidirectional conversion rate of PCr-to-ATP at relatively high spatial resolution (i.e. 0.5 mL voxel size), in a total acquisition time of 40 min and can potentially be used in a routine clinical setting.

Results

In order to estimate the pseudo first-order forward rate constant (kf) of the CK reaction, we need to measure the phosphocreatine (PCr) signal while we saturate the γ-adenosine triphosphate (γ-ATP) resonance for different durations (i.e. the progressive ST experiment)13. Assuming a two-pool exchange system between PCr and γ-ATP and complete saturation of γ-ATP, we can estimate the exchange rate between the two metabolites by solving the modified, for chemical exchange, Bloch equation14 (see methods). To confirm the efficiency of our saturation pulses, we acquired unlocalized 31P spectra in the entire volume of the lower leg muscles in all of our volunteers. Typical 31P spectra can be seen in Fig. 1, where the full width at half maximum of the PCr resonance peak was 12.3 ± 4.8 Hz (mean ± SD) across all subjects. As we show in the same figure, the ST module saturates γ-ATP to levels not measurable above the noise.

Figure 1
figure 1

Unlocalized ST-31P-MRS of the lower leg.

A) Fully relaxed 31P spectrum without ST preparation (tsat = 0 s) (left) and 31P-ST Spectrum with complete saturation (arrow) of the γ-ATP resonance (tsat = 6.84 s) (right). B) Series of ST spectra at different tsat. C) PCr signal intensity fitted to Eq.2, (Pearson's product moment correlation = 0.9992).

Fat infiltration, especially in high body mass index (BMI) subjects, can reduce the volume fraction of lean muscle in a certain volume of tissue affecting quantification of PCr concentration. In order to account for the fat content in the muscle, the Hierarchical IDEAL method15 was applied across all subjects. The fat volume fractions in the TA were 0.035 ± 0.006 across all subjects, significantly lower than in the GL (0.056 ± 0.019, P = 0.012), the GM (0.051 ± 0.020, P < 0.001) and the S (0.063 ± 0.020, P = 0.002) muscles. Figure 2 shows a case of a higher BMI (30.4) subject. The water/fat separation method shows fat infiltration in the leg muscles. The fat volume fraction in the GM was 0.11 and 0.04 in the TA, which, if not accounted for (especially the GM), can lead to under-estimation of the PCr concentration.

Figure 2
figure 2

Water/fat fractions in the lower leg.

The water (left) and fat (right) images of a subject (BMI = 30.4) are shown. The fat fraction (right) shows fat infiltration in the leg muscles.

In the turbo spin echo (TSE) imaging method, the echo amplitude is modulated as a function of the echo position in k-space, resulting in blurring16. We estimated the point spread function (PSF) of our imaging method as the result of the echo modulation to be at 1.6 pixels in the anterior to posterior direction, where the echo train is sampled.

Figure 3A shows an anatomical cross-section of the lower leg muscles of a lean subject (BMI = 19.5) and a cross-section of the PCr concentration map (Fig. 3B). PCr in the same cross-section decreases with increasing saturation time (tsat) as shown in Fig. 3C. By segmenting signals in the PCr images and fitting data to Eq.2 (methods), we estimated kf and the unidirectional flux of PCr to form ATP (Vf) for four muscle groups of the leg [Gastrocnemius Lateralis (GL) and Medialis (GM), Soleus (S) and Tibialis Anterior (TA)]. We estimated the intrinsic spin lattice relaxation (T1') in the presence of saturating irradiation17 from Eq.3. The results are summarized in Table 1. In the TA, kf was significantly lower than the GL (P = 0.002), the GM (P = 0.029) and the S (P = 0.037). The metabolic fluxes, Vf, in the TA were lower than both the GL (P < 0.001) and the GM (P = 0.043). We did not find any statistically significant differences among T1' values in the four muscle groups.

Table 1 Results of the ST experiments in four major leg muscles. Asterisks depict statistically significant differences between muscle groups (P < 0.05)
Figure 3
figure 3

Saturation transfer 31P-MRI.

A) Anatomical proton image where four major muscle groups are identified. B) PCr concentration map in a cross-section of the lower leg. C) ST-31P-MRI of the same cross-section at different saturation times. PCr signal intensity decreases as the system is allowed to exchange for longer time.

Discussion

Our study focused on the development and implementation of a ST- 31P-MRI method capable of obtaining muscle specific measurements of the CK reaction and metabolic fluxes of PCr to ATP in the entire volume of the muscles of the lower leg using a clinical high-field (3.0 T) scanner. Our method enables such measurements with relatively high spatial resolution (i.e. 0.5 mL voxel size) and large muscle coverage within an acquisition time of 40 min.

Our imaging approach (Fig. 4B) is based on a fast 3D-TSE sequence, which excites and acquires a single resonance of the 31P-MR spectrum (i.e. PCr). By trading spectral information for imaging speed, we are able to accelerate acquisition time and increase spatial resolution compared to 31P-MRS18,19 approaches, while obtaining full coverage of the muscle. As shown in Eq.2, estimation of kf only requires measurement of PCr signal at different saturation times.

Figure 4
figure 4

Saturation transfer spectroscopy (ST-31P-MRS) and imaging (ST-31P-MRI) pulse sequences.

A saturation transfer module consists of a train of Gaussian pulses, which saturates the γ-ATP resonance. Spoiler gradients are used between two consecutive pulses to destroy any remaining transverse magnetization. The number of Gaussian pulses defines tsat in each experiment. A) Unlocalized 31PMRS data were acquired by sampling the free induction decay (FID) after implementing the ST module. B) Identical ST preparation module as in 31P-MRS, followed by a spectrally selective 31P-MRI imaging sequence.

Our results are in close agreement with previous 31P-MR studies. Valkovic et al.7, measured a mean kf in the Gastrocnemius muscle at 0.31 s−1 both at 3.0 T and 7.0 T, using unlocalized 31P-MRS. Our mean measured kf in the GL was 0.32 s−1 (Table 1) and 0.31 s−1 in the GM. In an earlier study, Bottomley et al.20 reported a mean kf of 0.27 s−1 using a localized MRS method at 1.5 T. Our measurements of PCr concentration are well within the range reported in several studies reviewed by Kemp et al.21. As seen in Table 1, our measurements showed significant lower kf in the TA compared to the other three muscle groups (i.e. GL, GM, S). These results agree with the values we reported at UHF22, using 31P-MRI. To the best of our knowledge, no ST studies using spectroscopy have been reported for that particular muscle. In terms of our T1' measurements, Schar et al.8 reported mean values of 2.27 s in the calf muscle at 3.0 T, which is very close to the mean reported values in our work for the GL (2.11 s) and the GM (2.15 s).

The main limitation of our spectrally selective imaging approach, compared to 31P-MRS methods, is the lack of localized measurements of ATP, which may have a two-fold effect in quantifying kf and Vf. First, in the ST experiments, there could be remaining γ-ATP signal that is not fully suppressed. If that is the case, there could be local errors in the estimation of kf that are not accounted for by Eq.2. However, unlocalized 31P-MRS measurements in the entire volume of the lower leg muscles (Fig. 1) did not show any evidence of unsuppressed γ-ATP signal, hence we do not expect that this would affect quantification in our study. Second, 31P-MRS studies assume constant ATP concentration in the skeletal muscle21 and estimate PCr concentration from the ratio of the PCr signal to that of ATP. Our study requires the use of external phantoms and in some cases (i.e. high BMI subjects) water-content corrections that complicate the process of quantifying resting PCr concentration.

Of note, our imaging method could also be translated for studying metabolic activity in the brain. The CK reaction rate has been studied with one-dimensional spectroscopy23 at 3.0 T, while our method could provide full-coverage of the brain in similar acquisition time.

Compared to UHF, 31P MRI at 3.0 T suffers from a close to three-fold decreased signal to noise ratio24. In our implementation at UHF22, we used a constant saturation time prior to imaging and measured T1' in a separate experiment. However, at 3.0 T, we found that mapping T1' was particularly challenging, with the images acquired at different inversion times not yielding sufficient SNR in order to have accurate fittings of the recovery curve. In the work presented here, we used the progressive saturation approach, where we saturate γ-ATP for different durations. Through a three-parameter fitting of Eq.2 to the data, we recovered T1 and kf for each muscle which was then used in Eq.3 to estimate T1'. In this approach, the PCr signal for all tsat is between 40–100% of the reference signal (Fig. 1), allowing for accurate fitting of Eq.2 to the data.

In summary, 31P-MR is unique in its ability to non-invasively probe human metabolism in vivo. High-field (3.0 T) clinical magnets provide sufficient sensitivity in order to perform imaging experiments with large tissue coverage and relatively high spatial resolution for measuring muscle-specific energy metabolism in the skeletal muscle and potential disruptions in local energy metabolism caused by disease.

Methods

Theory

The CK reaction can be written as:

with kf and kr the pseudo first-order forward and reverse rate constants respectively4. One approach for measuring kf is to saturate the γ-ATP resonance for different durations (the progressive ST experiment)13. The saturated magnetization of γ-ATP is transferred to PCr (Eq.1), resulting in a net decrease of the PCr signal. The forward rate kf can be calculated from the relative decrease of the PCr signal as a function of saturation time (tsat). The CK rate constant kf multiplied by the PCr concentration yields the unidirectional flux of PCr to form ATP. Under fully-relaxed conditions, assuming complete saturation of the γ-ATP resonance, the magnetization of PCr as a function of tsat can be derived from the solution of the Bloch equations, modified for chemical exchange (assuming a two-pool exchange system)14:

Where M(tsat) the magnitude of the PCr signal measured at different times (tsat), c a parameter accounting for direct spill-over effects (i.e. direct saturation of PCr by saturating irradiation on γ-ATP) and T1 the spin-lattice relaxation time of PCr. By measuring the PCr signal for several tsat, kf can be estimated through a three-parameter (i.e. c, kf and T1) fit of the data to Eq.2.

The intrinsic spin lattice relaxation (T1') in the presence of saturating irradiation is17:

Human Subjects

The study was fully Health Insurance Portability and Accountability Act (HIPAA)–compliant and approved by the NYU Institutional Review Board. We obtained written informed consent from all participants in this study. The methods were carried out in accordance with Food and Drugs Administration (FDA) guidelines. We recruited ten non-smoking healthy volunteers (seven men, three women, 32.0 ± 3.5 years of age, range 29–41, BMI 22.6 ± 3.9, range 18.5–30.4) without any medical history of disease affecting muscle function or blood flow. We imaged them on a 3.0 T MRI system (Tim Trio, Siemens Medical Solutions, Erlangen, Germany) using a dual-tuned (31P/1H) transmit-receive quadrature birdcage knee coil (Rapid MRI, Ohio) (18 cm inner diameter).

31P-ST Spectroscopy and Imaging

We performed shimming on the entire volume of the lower leg muscles, using the proton channel of the dual-tuned coil and an iterative shimming algorithm provided by the manufacturer. We acquired ST spectroscopy and imaging data using the pulse sequences shown in Fig. 4.

Equation 2 holds for complete saturation of γ-ATP14. To confirm the efficiency of our ST module to saturate γ-ATP, we tested it using an unlocalized pulse acquire sequence (Fig. 4A). Acquisition parameters of the sequence are summarized in Table 2. We zero-filled data to 8-k data points prior to reconstruction and applied baseline correction for each spectrum separately.

Table 2 Acquisition parameters of the reference (tsat = 0) and ST 31P-MRS and 31P-MRI data

We acquired ST imaging data using the progressive saturation 31P-MRI sequence shown in Fig. 4B25. Acquisition parameters of the sequence are summarized in Table 2. In both the ST spectroscopy and imaging acquisitions, we used a ST module consisting of a train of Gaussian pulses (each 50 ms long, with 56 Hz bandwidth and 0.85 μT strength at peak), which saturates the γ-ATP resonance. Between two consecutive pulses a 7 ms delay allows for the use of spoiler gradients (magnitude 20 mT m−1; duration 5 ms) to destroy any remaining transverse magnetization. The number of Gaussian pulses defines tsat in each experiment.

The imaging sequence was a centric ordered three-dimensional turbo spin echo (3D-TSE) with a frequency selective 90° pulse (16 ms duration, 125 Hz bandwidth) that excited only the PCr resonance25. The effective echo was 26 ms, which resulted in minimum contamination from ATP signal26. A train of 24 non-selective 180° pulses was applied after each excitation, using equal area crusher gradients to remove the free inductions decays (FIDs) that were produced by imperfections of the refocusing pulses. During the ST imaging experiments, the SAR levels never exceeded 5% of the maximum allowed under normal operating mode.

This k-space sampling scheme resulted in increased blurring in the imaging direction where the ETL was sampled. We simulated the effect of signal modulation during the ETL acquisition in order to predict the point-spread-function (PSF) along that direction, using an average T2 value of PCr of 365 ms based on previous measurements11 which were in close agreement with T2 values at 3.0 T reported in the literature24.

For absolute quantification of PCr in the muscle we imaged a phantom with comparable to the in vivo coil loading, using the same 31P-MRI sequence and parameters (with the ST module turned off), with the exception of repetition time (TR) of 60 s. We used the phantom to derive the calibration curve of PCr as a function of signal intensity. The phantom consisted of three sealed cylindrical tubes containing different concentrations (25, 50 and 75 mM) of inorganic phosphate (Pi).

1H Water-Fat Imaging

Fat infiltration, especially in high BMI subjects, can reduce the volume fraction of lean muscle in a certain volume of tissue affecting quantification of PCr concentration. In order to account for the fat content in the muscle, we used a three-dimensional gradient echo sequence with three echo-times (TE: 2.1, 2.8 and 3.7 ms; flip angle: 3°; TR: 12 ms, total acquisition time: 3 min), with the same FOV and frame of reference as the PCr images. The matrix size was 128 × 128 × 40. We reconstructed water and fat fraction images using the Hierarchical IDEAL method15 on all ten healthy volunteers. These fractions represent the proton density fat and water fractions and can be converted to volume fat and water fractions as shown previously27.

Data Analysis

For each muscle group, we estimated kf from Eq.2 by fitting the mean signal intensity of the segmented PCr datasets at different tsat (excluding slices 1 and 8 at the edge of the coil). We estimated Vf from the product of kf and PCr concentration in the same volume of the muscle. We compared kf and Vf values among four different muscle groups [Gastrocnemius Lateralis (GL) and Medialis (GM), Soleus (S) and Tibialis Anterior (TA)], with paired t-tests with a 5% significance level.