A Novel CEST-Based Approach for Reliably Assessing Skeletal Muscle Oxidative Phosphorylation: OXCEST

Abstract

Purpose

To develop and validate a novel chemical exchange saturation transfer (CEST) MRI method to map skeletal muscle OXPHOS (Oxidative Phosphorylation CEST or OXCEST).

Theory and Methods

Our proposed OXCEST method acquirses creatine (Cr)-weighted CEST maps by applying RF saturation (B₁) at only two frequency offsets: +1.8 ppm (targeting the Cr amine resonance) and −1.8 ppm (to calculate MTR_asym at 1.8 ppm). The pre-exercise MTR_asym is modeled as a second-order polynomial function of B₀ (f(B₀)). Next, the post-exercise alteration in MTR_asym is hypothesized to be affected by both an exercise-induced increase in Cr and changes in B₀ inhomogeneity. By inputting post-exercise B₀ values into f, the change in MTR_asym due to B₀ variation alone was estimated, thus allowing for the quantification of only Cr-related post-exercise MTR_asym changes. OXCEST and ³¹P-MRS were performed in seven subjects across two sessions to compare the OXCEST-derived Cr recovery time constant (T_Cr) with the ground-truth phosphocreatine recovery time constant (T_PCr) derived by ³¹P-MRS.

Results

Function f reliably described the relationship between pre-exercise MTR_asym and B₀ (R² = 0.87 ± 0.07 in the lateral gastrocnemius (LG); R² = 0.98 ± 0.01 in the medial gastrocnemius (MG); R² = 0.96 ± 0.03 in the soleus). The mean pre-exercise MTR_asym was approximately 6%−7% for all muscle groups. Following exercise, MTR_asym increased by 11.4% ± 4.5% in LG and 8% ± 2.4% in MG, and showed mono-exponential recovery (R² > 0.97). The combined T_Cr of LG and MG was found to be significantly correlated with T_PCr (R² = 0.83, p = 0.005).

Conclusion

OXCEST enables reliable assessment of post-exercise Cr recovery and demonstrates significant correlation and good agreement with ³¹P-MRS.

1. Introduction

Mitochondrial oxidative phosphorylation (OXPHOS) is a central metabolic pathway responsible for generating adenosine triphosphate (ATP), the primary energy currency of the cell. OXPHOS capacity varies significantly across populations: endurance-trained athletes exhibit markedly higher mitochondrial activity than sedentary lean or overweight individuals [1]. Conversely, impaired mitochondrial OXPHOS is a hallmark of numerous pathological conditions, including neuromuscular disorders, primary mitochondrial diseases, and age-related decline [2–5].

Currently, invasive muscle biopsy remains the clinical gold-standard method for assessing OXPHOS capacity [6]. However, biopsies are inherently limited by patient discomfort, inability to capture dynamic metabolic changes, and poor feasibility for repeated longitudinal assessments. 31-Phosphorus magnetic resonance spectroscopy (³¹P-MRS) has traditionally been regarded as the MR-based gold standard technique to non-invasively quantify mitochondrial function in skeletal muscle [7–11]. During muscle contraction, phosphocreatine (PCr) is hydrolyzed to regenerate ATP from ADP, and during recovery, it is resynthesized via the creatine kinase (CK) reaction, which stoichiometrically couples PCr and creatine (Cr) [12, 13]. ³¹P-MRS has been extensively applied to study mitochondrial dysfunction in aging and mitochondrial diseases [8, 14–16]. Nevertheless, its wider application of ³¹P-MRS is challenged by relatively low intrinsic sensitivity, the requirement for specialized hardware, and lower spatial resolution compared to proton-based MRI techniques.

Over the past decade, imaging the post-exercise recovery kinetics of Cr and PCr has emerged as a ¹H-based alternative to ³¹P-MRS by acquiring a partial or full z-spectrum [17–22]. Currently, magnetization transfer ratio asymmetry (MTR_asym), defined as the difference between signals at the positive and negative frequency offsets relative to water, is the most widely used chemical exchange saturation transfer (CEST) quantification metric. Recently, Phillip Zhe Sun [23] introduced an expedient B₀ inhomogeneity correction algorithm for fast pH imaging that requires only two frequency offsets, based on the observation that MTR_asym can be modeled as a second-order polynomial function of B₀ inhomogeneity. Building on this principle, we aimed to evaluate the feasibility of a new technique, termed oxidative phosphorylation CEST (OXCEST), which integrates the fast B₀-correction method with acquisition at only two offsets: +1.8 and −1.8 ppm. Restricting CEST acquisition to two frequencies enables higher temporal resolution, comparable to that of ³¹P-MRS. We evaluated the feasibility of OXCEST in healthy subjects by directly comparing outcomes with those obtained from gold-standard ³¹P-MRS.

2. Theory

2.1. Cr Mapping in Skeletal Muscle

A typical z-spectrum is primarily influenced by the following CEST pools: (i) direct water proton saturation (DS), the largest water signal intensity drop occurring at 0 ppm, (ii) conventional semi-solid magnetization transfer (MT) arising from the “bound” pool of water molecules cross-relaxing with protons in the free water pool, (iii) amide proton transfer (APT) at 3.5 ppm, arising from the exchangeable amide protons in the backbone of mobile proteins and peptides, (iv) exchange-relayed nuclear Overhauser effect (rNOE) at −3.5 ppm, from aliphatic protons of mobile macromolecules, and (v) Cr CEST at 1.8 ppm, from the exchangeable amine protons of Cr.

The MTR_asym analysis of the z-spectrum can separate the symmetrical contribution of DS and MT (Equation 1).

$${\text{MTR}}_{\text{asym}}=\frac{S\left(-1.8ppm\right)-\text{S}\left(+1.8ppm\right)}{S\left(-1.8ppm\right)}\times 100$$ (1)

At high B₁ (˜3 μT), used in this study, the contributions of PCr, APT, and rNOE at ±1.8 ppm are negligible [17, 24–27]. Therefore, without considering B₀ inhomogeneity effects, RF irradiation at only two frequency offsets (±1.8 ppm) is sufficient to quantify the pre-exercise CEST contribution of Cr using MTR_asym (MTR_asymPre). However, as the resonance frequency in MRI is proportional to B₀, even a minor magnetic field inhomogeneity (ΔB₀) can cause substantial frequency shifts in the z-spectrum [28]. To correct for this B₀ inhomogeneity, additional frequency offsets are acquired [18, 20].

Following interventions such as mild plantar flexion exercise, pH in skeletal muscle remains relatively constant at effort below 60% of maximal voluntary contraction (MVC) during plantar flexion [29], and therefore does not significantly influence the CEST contrast. Consequently, exercise-induced change in Cr level, mediated by CK, is expected to be the primary contributor to the post-exercise MTR_asym (MTR_asymPost). Additional contribution to MTR_asymPost comes from exercise-induced motion, which perturbs B₀ [30], and causes post-exercise ΔB₀ (ΔB_0Post) to differ from the pre-exercise ΔB₀ (ΔB_0Pre). Given that PCr resynthesis is stoichiometrically coupled with a decrease in free Cr, a higher temporal resolution scan will enable capturing rapid changes in Cr levels. Here, we propose limiting the CEST acquisition to only two offsets (+1.8 and −1.8 ppm), thereby attaining a temporal resolution of 10 s (5 s per offset × 2 offsets).

2.2. Fast B₀ Correction in OXCEST

To account for pre- and post-exercise B₀ inhomogeneities, we modeled the relationship between MTR_asymPre and the corresponding ΔB_0Pre across all pixels within a muscle. A second-order polynomial function was fit using least-squares regression to obtain the coefficients c₀, c₁, c₂: [23]

$${\text{MTR}}_{\text{asymPre}}\left(i\right)=c_0+c_1\times {\text{ΔB}}_{0\mathrm{Pr}e}\left(i\right)+c_2{\left[{\text{ΔB}}_{0\mathrm{Pr}e}\left(i\right)\right]}^2$$ (2)

where i = 1, 2, …, N and N is the total number of pixels in a muscle ROI. This fitted function f (c₀, c₁, c₂) therefore characterizes, at a pixel level, how much the MTR_asymPre varies with ΔB_0Pre in a given muscle. ΔB_0Post and ΔB_0Pre are measured independently and hence were not assumed to be identical. Next, ΔB_0Pre values were replaced with ΔB_0Post in f to estimate the B₀-related contribution to the post-exercise MTR_asym (MTR_asymPost):

$${\text{MTR}}_{{\text{asymΔB}}_{\text{0}}}\left(i\right)=c_0+c_1\times {\text{ΔB}}_{0\mathrm{Pr}e}\left(i\right)-c_2\times {\left[{\text{ΔB}}_{0\mathrm{Pr}e}\left(i\right)\right]}^2$$ (3)

We hypothesized that the observed MTR_asymPost at each pixel reflects two additive effects: (a) a B₀-induced shift, MTR_asymΔB0, and (b) a physiologic change due to exercise-induced change in Cr concentration, MTR_asymΔCr.

Thus, the measured MTR_asymPost can be expressed as follows:

$${\text{MTR}}_{\text{asymPost}}\left(i\right)={\text{MTR}}_{\text{asymΔB0}}\left(i\right)+{\text{MTR}}_{\text{asymΔCr}}\left(i\right)$$ (4)

By subtracting the estimated B₀-induced component, we isolate the Cr-related contribution:

$${\text{MTR}}_{\text{asymΔCr}}\left(i\right)={\text{MTR}}_{\text{asymPost}}\left(i\right)-{\text{MTR}}_{\text{asymΔB0}}\left(i\right)$$ (5)

Finally, for each muscle, the net MTR_asymΔCr was computed as the mean value across all pixels within the segmented region.

3. Methods

3.1. Study Design

The protocol was approved by the Institutional Review Board, and written informed consent was obtained from all participants prior to imaging. Seven healthy adults (male: 5, mean age: 34.6 ± 6.1 years) were enrolled. Each participant underwent two MR sessions: one for OXCEST and one for ³¹P-MRS performed on different days using a 3T whole-body MR scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany). During each session, participants performed a mild plantar flexion exercise for 2 min at a constant rate of 45 flexions per minute guided by a metronome, using an MR-compatible ergometer (Trispect, Ergospect GmbH, Innsbruck, Austria). The pedal resistance was adjusted to correspond to 30%−40% of each participant’s MVC. The MRI protocols for OXCEST and ³¹P-MRS are summarized in Figure S1.

3.2. OXCEST MRI

OXCEST data were acquired with a Tx/Rx 1/15-channel ¹H knee coil (Siemens Healthcare, Erlangen, Germany). A localizer was first obtained on the right calf to position the imaging slice at the thickest section of the calf muscle. Pre-exercise acquisitions included: (i) one image without selective saturation, (ii) a B₁ map using a double angle method (30° and 60°) [31], (iii) a B₀ map using standard water-saturation shift-referencing (WASSR) [28], and (iv) four baseline CEST images [22]. Post-exercise acquisitions commenced immediately following the 2-min exercise and consisted of: (i) dynamic CEST imaging (60 sets of positive and negative offset images at 10 s temporal resolution), (ii) B₀ map, (iii) B₁ map, and (iv) one reference image without selective saturation. OXCEST imaging parameters were set as follows: TR/TE = 4.7/2.3 ms, slice thickness = 10 mm, flip angle = 10°, B_1rms = 3 μT, in-plane resolution = 1.25 × 1.25 mm², matrix size = 128 × 128, field of view = 160 × 160 mm². OXCEST images were acquired with saturation frequency offsets of ±1.8 ppm relative to the water resonance, achieving a temporal resolution of 10 s (5 s per offset × 2 offsets). Two dummy scans were included to achieve T1 steady state. For four participants, shimming was constrained to a voxel encompassing the three muscles of interest. Although MTR_asym was relatively stable around optimal RF power level [32], both pre- and post-exercise OXCEST images were corrected for minor B₁ variations using an inverse scaling model, assuming a linear relationship between B₁ amplitude and CEST signal intensity [31].

3.3. ³¹P-MRS

³¹P-MRS was performed with a ¹H/³¹P dual-tuned flex transmit/receive surface coil of 11 cm diameter (RAPID Biomedical GmbH, Würzburg-Rimpar, Germany) placed over the calf, using an unlocalized free induction decay (FID) sequence: measurements = 96, averages = 2, bandwidth = 5000 Hz, and TR = 5 s. The total data acquisition time was 15 min: 5 min of baseline, 2 min of exercise, and 8 min of post-exercise recovery. The temporal resolution of ³¹P-MRS was kept at 10 s to match the temporal resolution of OXCEST acquisitions.

3.4. Data Analysis

All images and data/statistical analyses were performed on MATLAB v2023b.

3.4.1. OXCEST

The lateral gastrocnemius (LG), medial gastrocnemius (MG), and soleus (Sol) muscles were manually segmented on the pre-exercise reference image and on each post-exercise CEST image. Because exercise can induce positional shifts and subtle anatomic changes, including transient hypertrophy post-exercise, regions of interest (ROIs) were re-segmented on each post-exercise image until muscle boundaries were stabilized across consecutive time points (typically within 2 min). The tibialis anterior (TA) was also manually segmented as an expected negative control for the plantar-flexion exercise. The B₀-corrected MTR_asymΔCr signals, reflecting exercise-induced changes in Cr, were computed for each muscle group as described in Section 2.2. The post-exercise Cr recovery time-constant (T_Cr) was estimated by fitting a mono-exponential decay model:

$${\text{MTR}}_{\text{asymΔCr}}\left(t\right)=a_0+a_1\ast e^{\left(-\frac{t}{T_{Cr}}\right)}$$ (6)

where a₀ is the baseline value and a₁ is the amplitude of exponential decay. Initial fitting parameters were set as a₀ = min(MTR_asymΔCr), a₁ = max(MTR_asymΔCr) − min(MTR_asymΔCr), and T_Cr = 60 s, with bound constraints a₁ > 0 and T_Cr > 0. The coefficient a₁ was used to quantify the exercise-induced increase in MTR_asym. Pre-exercise B₀-corrected MTR_asymPre values for each muscle were obtained from the coefficient c₀ in Equation (2).

To assess the effects of B₀-correction, uncorrected MTR_asym values were averaged across all pixels in each muscle at each time point. The exercise-induced change was then calculated as the difference between post-exercise MTR_asymPost at the first time point immediately after exercise and the pre-exercise baseline MTR_asymPre. Additionally, we retrospectively analyzed our previously published CrCEST MRI data [33]. These data were acquired at six frequency offsets (±1.5, ±1.8, and ±2.1 ppm), enabling assessment of Cr recovery kinetics using two approaches: (i) the conventional method, which evaluates CrCEST using all six offsets with a temporal resolution of 30 s, and (ii) our proposed method, which uses only two offsets (±1.8 ppm) with the same temporal resolution, hereafter referred to as low-temporal-resolution OXCEST (LTR-OXCEST). This design allowed direct comparison between the established and proposed approaches. Notably, when exponential fitting of Cr recovery data was performed using the conventional approach, the estimated T_Cr values in some cases exceeded physiologically reported ranges; therefore, for comparison purposes, T_Cr values were capped at 300 s.

3.4.2. ³¹P-MRS

The ³¹P-MRS spectra were phase- and baseline-corrected, then fitted with Lorentzian functions. The PCr peak area was quantified from each FID and normalized to the corresponding pre-exercise value. Post-exercise PCr levels were expressed as a percentage of the pre-exercise level. The PCr recovery time constant (T_PCr) was estimated by fitting the data to a mono-exponential growth function: [34]

$$\text{PCr}\left(t\right)=b_0+b_1\ast \left(1-e^{\left(-\frac{t}{T_{PCr}}\right)}\right)$$ (7)

where b₀ represents the normalized PCr level at the first post-exercise time point, and b₁ is the amplitude of recovery.

In addition, the exercise-induced pH shift was estimated from the chemical shift difference between PCr and inorganic phosphate (P_i) (δ) in ³¹P-MRS spectra using the adjusted Henderson-Hasselbalch equation: [35]

$$\text{pH}\text{=}p{K}_A+\log \left(\frac{\delta -{\delta }_{HA}}{{\delta }_A-\delta }\right)$$ (8)

where pK_A = 6.75 is the dissociation constant of P_i, δ_HA = 3.27 is the chemical shift of the protonated form of P_i, and δ_A = 5.63 is the chemical shift of the non-protonated form of P_i.

3.5. Statistical Analysis

Linear regression was performed to evaluate the relationship between T_Cr and T_PCr, and a two-sided p value was calculated to test the significance of the association. Lin’s concordance correlation coefficient (CCC) was also used to assess the agreement between the two measures. Since the majority of the ³¹P-MRS signal acquired using a surface coil originates from the gastrocnemius muscle [21], the mean T_Cr values from the LG and MG were used for statistical comparison with T_PCr. The coefficient of determination (R²) and an F-statistic were calculated from the least squares fit of a 2nd-order polynomial relating MTR_asymPre to ΔB_0Pre in LG, MG, and Sol. 95% confidence intervals (CI) were calculated for the predicted B₀-corrected MTR_asymPre and the exercise- induced increases in MTR_asym. Following mono-exponential fitting of Cr recovery in LG, MG, and Sol, the root mean squared error (RMSE), R², and standard deviation (SD) of residual noise (reported as a percentage of exercise-induced MTR_asym increase) were determined. A p < 0.05 was considered statistically significant.

4. Results

4.1. Demonstration of Fast B₀ Correction in OXCEST

The OXCEST approach is illustrated in Figure 1. First, the LG, MG, and Sol muscles were manually segmented on a reference anatomical image of the calf (Figure 1A). Pixel-wise MTR_asymPre values were then calculated from the corresponding CEST-weighted images (Figure 1B), and a ΔB_0Pre map was derived from the pre-exercise WASSR scan (Figure 1C). In the MG muscle, scatterplots of MTR_asymPre versus ΔB_0Pre demonstrated a strong quadratic relationship (Figure 1D), with the second-order polynomial fit described by the following equation: MTR_asymPre = 5.91 + 152 × ΔB_0Pre—191 × ΔB_0Pre2 (R² = 0.97, p < 0.001). The B₀-corrected baseline MTR_asymPre in MG was estimated by setting the ΔB_0Pre = 0, yielding a net value of 5.91% (95% CI: 5.74%−6.08%). Analogous analysis for LG (Figure S2A) and Sol (Figure S2B) yielded baseline B₀-corrected MTR_asymPre values of 6.76% (95% CI: 6.37%−7.15%) and 7.2% (95% CI: 7.01%−7.29%), respectively. Following exercise, the ROIs were re- segmented on the post-exercise CEST-weighted images to account for any positional changes (Figure 1E). Pixel-wise MTR_asymPost values were computed from OXCEST images (Figure 1F), and a ΔB_0Post map was obtained from a WASSR scan (Figure 1G). In MG, scatterplots of MTR_asymPost versus ΔB_0Post are shown in Figure 1H. The black crosses represent pixel-wise MTR_asymΔB0 values predicted using Equation (3). The vertical distance between each MTR_asymPost and its associated MTR_asymΔB0 represents the Cr-specific signal change (MTR_asymΔCr) per pixel. Similar exercise-related increases in Cr were observed in LG and Sol. The subject-specific fitted coefficients (c₀, c₁, c₂) describing the relationship between MTR_asymPre and B_0Pre for LG, MG, and Sol are rovided in Table S1. Among these, coefficient c₂ exhibited the greatest variability across subjects, followed by c₁, while c₀ was the most stable. This trend was consistent across all three muscles.

FIGURE 1.

Demonstration of the OXCEST approach in a 35-year-old female participant. (A) Segmentation of the lateral gastrocnemius (LG), medial gastrocnemius (MG), and soleus (Sol) muscles on a pre-exercise axial anatomical calf image. (B, C) Pre-exercise magnetization transfer ratio asymmetry (MT_RasymPre) map acquired at +1.8 ppm and the corresponding B₀ inhomogeneity (ΔB_0Pre) map derived from a WASSR scan, both overlaid on the segmented muscle masks. (D) Pixel-wise regression of MTR_asymPre against ΔB_0Pre in MG; the dotted line represents the 2nd-order polynomial fit. (E) Segmentation of the three muscles on a post-exercise axial anatomical calf image. (F, G) Post-exercise MTR_asym (MTR_asymPost) and ΔB₀ (ΔB_0Post) maps acquired following mild plantar-flexion exercise. (H) Scatterplot of MTR_asymPost versus ΔB_0Post in MG (circles). Black crosses indicate the expected value of MTR_asym due solely to B₀ changes (MTR_asymΔB0), assuming no change in creatine (Cr) concentration. Thus, the Cr-related change in Mak_sym is quantified as the vertical distance between each circle and its corresponding cross, representing the ΔB₀-corrected Cr effect.

4.2. Post-Exercise Cr Recovery Mapped by OXCEST

Figure 2 shows representative B₀-uncorrected MTR_asymPost maps and B₀-corrected MTR_asymΔCr maps, acquired with OXCEST over a 10-min recovery period. Pixel intensities reflect the Cr-specific MTR_asym signal, with red denoting elevated values and blue indicating baseline values. At 10 s post-exercise, all three muscle groups displayed a marked increase in free Cr levels. During the first minute of recovery, Cr levels decreased rapidly as they rephosphorylate to PCr via the CK reaction.

FIGURE 2.

Effect of OXCEST-based B₀ correction on post-exercise MTR_asym, acquired at 10 s temporal resolution over a 10-min recovery period. Shown are pixel-wise MTR_asymPost maps (in %) from a representative axial slice of the lower leg encompassing the soleus (Sol), lateral gastrocnemius (LG), and medial gastrocnemius (MG) muscles: (A) without B₀ correction, and (B) with B₀ correction. Without B₀ correction, MTRasym values frequently extended beyond physiologically plausible limits. Following B₀ correction, elevated MTR_asym observed immediately post-exercise reflects an increased Cr level, which decays exponentially over time. This spatiotemporally resolved mapping of Cr dynamics enables visualization of muscle-specific variations in mitochondrial OXPHOS capacity.

Table 1 summarizes muscle-wise B₀-corrrected MTR_asymPre, exercise-induced increases in MTR_asym, and the goodness-of-fit (R²) values for the polynomial function f. Across LG, MG, and Sol, R² values were consistently high (LG: 0.87 ± 0.07, p < 0.001; MG: 0.98 ± 0.01, p < 0.001; Sol: 0.96 ± 0.03, p < 0.001). Net MTR_asymPre values were 6.03% ± 1.7% in LG, 6.76% ± 0.5% in MG, and 6.87% ± 0.4% in Sol for all the subjects. Narrow 95% CIs for MTR_asymPre were observed in MG and Sol, as well as in LG of four subjects with constrained shimming. In contrast, in the three subjects without constrained shimming, B₀ inhomogeneity exceeded 0.1 ppm in the LG, resulting in wider 95% CIs. Pre- and post-exercise ΔB₀ values are reported in Table S2. The largest exercise-induced increase in MTR_asym was observed in LG (11.4% ± 4.5%), followed by MG (8% ± 2.4%) and Sol (4% ± 2%), all with narrow 95% CIs. Parameters from mono-exponential fitting of Cr recovery are presented in Table S3, with R² values > 0.97 in both LG and MG. Net MTR_asymPre values, together with post-exercise changes with and without B₀ correction, are presented in Table S4. Without B₀-correction, MTR_asymPre values ranged widely from −34% to +28.7%. Finally, post-exercise changes in MTR_asym and the corresponding T_Cr values for the TA muscle are summarized in Table S5. Both parameters demonstrated high inter-subject variability.

TABLE 1.

Metrics derived from OXCEST method.

ID	MTRasymPre versus B 0Pre (R 2)			B 0-corrected MTRasymPre			Exercise-induced increase in MTRasym
	LG	MG	Sol	LG	MG	Sol	LG	MG	Sol
1	0.833	0.98	0.912	4.49 (3.1, 5.9)	7.1 (7, 7.2)	6.9 (6.8, 7.1)	18.3 (17.6, 19)	8.9 (8.7, 9.2)	6 (5.8, 6.2)
2	0.962	0.992	0.967	6.8 (6.4, 7.2)	5.9 (5.7, 6.1)	7.2 (7, 7.3)	9.3 (9.1, 9.5)	9.3 (9.1, 9.5)	2.4 (2.1, 2.6)
3	0.74	0.982	0.955	5.7 (5.4, 5.9)	6.3 (5.6, 6.9)	7.1 (7, 7.2)	6.4 (6, 6.8)	5.9 (5.7, 6.1)	1.6 (1.1, 2)
4	0.841	0.975	0.927	8 (5, 10.9)	7.4 (7.2, 7.5)	6.8 (6.7, 6.9)	10.6 (10.4, 10.8)	12.1 (11.9, 12.3)	5.3 (5.1, 5.5)
5	0.881	0.981	0.982	8.7 (5.5, 11.8)	6.8 (6.6, 6.9)	7 (6.9, 7.2)	5.3 (5.1, 5.5)	4.1 (3.9, 4.3)	1.5 (1.4, 1.7)
6	0.952	0.996	0.993	4.5 (−3.8, 12.7)	6.4 (6.3, 6.6)	5.9 (5.8, 6)	15.8 (15.5, 16.2)	7.8 (7.7, 8)	6.3 (6.1, 6.5)
7	0.86	0.985	0.981	4 (0.5, 7.4)	7.4 (7.3, 7.5)	7.2 (7.1, 7.4)	14.1 (13.6, 14.6)	8.2 (7.9, 8.5)	5 (4.9, 5.2)

Note: Some values are reported as mean (95% confidence interval).

4.3. Comparison of OXCEST-Derived Cr and ³¹P-MRS-Derived PCr Recovery

Figure 3 shows the comparison between OXCEST-derived post-exercise Cr recovery kinetics and the ground truth post-exercise PCr recovery kinetics in the same subject. Both recovery profiles were well described by mono-exponential fits. The T_PCr measured by ³¹P-MRS was 32 s. Using OXCEST, T_Cr values were 27 s in LG, 25 s in MG, and 29 s in Sol. Figure 3C illustrates the fitted post-exercise PCr recovery curve alongside the fitted average post-exercise Cr recovery curve from LG and MG, demonstrating close agreement between OXCEST and ³¹P-MRS. T_PCr and T_Cr for all subjects in this study are summarized in Table 2. The average T_PCr measured was 51 ± 15 s, while the T_Cr values were 56 ± 19 s (LG), 58 ± 29 s (MG), and 52 ± 33 s (Sol). The combined T_Cr of LG and MG showed a significant correlation with T_PCr (R² = 0.83, p = 0.005) (Figure 4). A CCC value of 0.81 was found between T_Cr and T_PCr, suggesting good agreement [36]. Representative Sol recovery curves of subjects 5 and 7 with very low T_Cr values are presented in Figure S3. Additionally, exercise did not induce any appreciable changes in pH values in six of the seven subjects (Table S6), while subject 2 exhibited a pH change of 0.7.

FIGURE 3.

Comparison of post-exercise phosphocreatine (PCr) and creatine (Cr) recovery kinetics measured using 31P-MRS and OXCEST. (A) Non-localized 31P-MRS shows mono-exponential PCr recovery. (B) OXCEST-derived Cr signal demonstrates mono-exponential decay in lateral gastrocnemius (LG), medial gastrocnemius (MG), and soleus (Sol). (C) Mono-exponential fits of PCr recovery and mean Cr recovery (LG and MG) reveal closely matched time constants: T_PCr = 32 s and T_Cr = 26 s.

TABLE 2.

Recovery time constants derived from OXCEST and 31P-MRS (n = 7).

ID	31P-MRS TPCr (s)	OXCEST TCr (s)
		LG	MG	Sol
1	41.06	46.15	66.54	90.4
2	32.24	27.19	25.27	29.33
3	50.43	71.04	65.23	72.46
4	42.74	38.94	21.37	96.52
5	83.28	87.96	112.39	14.56
6	47.24	65.95	45.34	49.84
7	59.77	56.66	67.92	11.75

FIGURE 4.

Regression analysis comparing post-exercise creatine (Cr) recovery time constant (T_Cr) measured using OXCEST and phosphocreatine (P_Cr) recovery time constant (T_PCr) determined with ³¹P-MRS. A significant linear correlation was observed between T_Cr and T_PCr (R² = 0.83, p = 0.005), with a good agreement between the two measurements (Lin’s CCC = 0.81). The dashed ellipse denotes the 95% confidence interval.

Furthermore, retrospective analysis of Cr recovery rates in LG, MG, and Sol from conventionally acquired six-offset, 30-s temporal resolution CrCEST data demonstrated a strong correlation (R² = 0.77, p < 0.001) and good agreement (Lin’s CCC = 0.83) between T_Cr estimates derived from LTR-OXCEST and conventional CrCEST (Figure S4).

5. Discussion

In this study, we introduced OXCEST, a novel imaging method for mapping skeletal muscle OXPHOS by applying a fast B₀ correction approach for accurately estimating post-exercise Cr recovery kinetics. The OXCEST approach further demonstrated strong agreement with the gold-standard technique, ³¹P-MRS.

By sampling only two frequency offsets, the OXCEST approach generates B₀-corrected, Cr-specific MTR_asym images, enabling the capture of post-exercise Cr recovery dynamics at very high temporal resolution. This is critical, as the literature reports T_PCr values ranging from 20 to 60 s in healthy adults [11, 37], implying that approximately 63% of post-exercise PCr recovery occurs within the first T_PCr seconds. Given that PCr resynthesis is stoichiometrically coupled with a corresponding decrease in free Cr, capturing the rapid Cr dynamics in muscle demands high temporal resolution. In our study, OXCEST-derived T_Cr was significantly correlated with and showed good agreement with ³¹P-MRS-derived T_PCr, a marker of mitochondrial OXPHOS capacity. Additionally, the high R² values for the mono-exponential fits in both LG and MG indicate that the model explains over 97% of the variance in the Cr recovery data, supporting the appropriateness of the mono-exponential model for describing post-exercise Cr dynamics. Therefore, the OXCEST approach serves as a valuable complement to the ³¹P-MRS technique and offers a non-invasive tool for quantifying skeletal-muscle energetics in clinical settings.

We found that the 2nd-order polynomial relationship between MTR_asymPre and ΔB_0Pre closely matched previously reported findings in a pH imaging study in rats [23]. In our study, this fit yielded strong R² values within the observed ΔB_0Pre range of −0.25 to +0.25 ppm, indicating robustness across modest B₀ inhomogeneity. We postulate that the proposed B₀-correction approach in this study will be valid if the pre- and post-exercise B₀ are within moderate drifts (≤ 0.25 ppm) away from the label frequency of 1.8 ppm. Under these conditions, MTR_asym can still be reliably estimated using the proposed framework. However, in the absence of adequate shimming, large ΔB₀ deviations may introduce DS-related contamination of the CEST signal, thereby compromising the validity of the relationship. Further higher order polynomial forms of the fitting function f may be appropriate when the Lorentzian linewidth of the Cr amine proton CEST pool is narrow or Cr levels are low.

Next, we found the net B₀-corrected MTR_asymPre values for each muscle group to be approximately 6%−7%. Reliable estimation of this parameter requires the presence of regions within the muscle where ΔB₀ values are close to zero. When ΔB₀ values within a muscle are confined to a narrow range offset from 0 ppm, the estimate of the y-intercept c₀ in Equation (2) exhibits high variability, leading to wide 95% CIs. The MG and Sol consistently satisfied this criterion, with a fraction of their ROIs lying within the ΔB₀ ranges close to 0 ppm across all seven subjects. In contrast, the three LG cases that failed to meet this criterion were those in which global rather than local shimming was applied. Reported Cr-weighted CEST studies have demonstrated comparable MTR_asymPre values, supporting the validity of our OXCEST-based quantification [19, 21, 22, 27, 33, 38]. The exercise-induced increase in MTR_asymΔCr was found to be highest in LG, followed by MG and Sol, consistent with previous findings. The consistently narrow 95% CIs for these measurements further confirm that OXCEST provides robust MTR_asymPost maps and that initial shimming was not a limiting factor.

We found that the first minute post-exercise is particularly critical for accurately characterizing the mono-exponential dynamics of Cr recovery. In healthy skeletal muscle, T_PCr typically ranges from 20 to 60 s [11, 37]. Owing to its exponential nature, a muscle with T_PCr = 20 s recovers ˜95% of PCr within the first minute post-exercise, whereas a muscle with T_PCr = 60 s only recovers ˜63% over the same duration. This underscores the advantage of OXCEST’s expedient B₀-correction algorithm, which enables 10 s temporal resolution and thereby allows detailed sampling of the steep early recovery phase. The strong correlation between OXCEST-derived T_Cr and ³¹P-MRS-derived T_PCr further validates the accuracy of this method.

Post-exercise ΔMTR_asym and T_Cr capture complementary aspects of muscle energetics. ΔMTR_asym primarily reflects the magnitude of immediate muscle activation, that is, the extent of Cr accumulation during exercise. In contrast, T_Cr characterizes the rate of conversion of Cr back to PCr via CK and therefore serves as an indicator of mitochondrial OXPHOS capacity. Importantly, the degree of muscle activation does not necessarily predict recovery kinetics. For example, a localized ³¹P-MRS study [39] reported low T_PCr values (< 20 s) in the soleus of several subjects, even when exercise-induced PCr depletion was substantial (˜50%). LG, MG, and Sol also differ in structure and fiber-type composition. The soleus is enriched in oxidative (“slow-twitch”) fibers (80%; range: 64%−100%) compared with gastrocnemius (57%; range: 34%−82%) [40]. Oxidative fibers have slower contraction rates, higher mitochondrial content, greater reliance on OXPHOS, and higher resistance to fatigue [41]. This intrinsic variability in fiber type composition of the three muscles contributes to both interindividual and intermuscular differences in T_Cr.

Although the TA is generally regarded as a negative control during plantar flexion exercises, both T_Cr values and post-exercise increases in MTR_asym in our cohort exhibited substantial inter-subject variability. While one study reported less than a 1% increase in TA MTR_asym following mild exercise [22], other investigations, including unpublished data from our group, have demonstrated measurable TA activation during plantar flexion [33, 42]. This variability is likely attributable to inadvertent TA engagement in some individuals, potentially arising from differences in leg or ankle positioning on the pedal or uneven force application required to maintain the prescribed exercise load [43, 44].

Regarding the specificity of OXCEST towards Cr, Chen et al. [17] reported that PCr exhibits a CEST resonances near the Cr peak at +1.8 ppm downfield of water, with the extent of this overlap dependent on B₁. In this study, the PCr-related ΔZ effect at ˜1.95 ppm diminished at B₁ amplitudes ≥ 1 μT. In the present work, the OXCEST acquisitions were performed at a B₁ of 3 μT, a level at which prior studies have shown PCr contribution to be minimal [18]. Therefore, we consider any residual PCr overlap to have minimal impact on the specificity of Cr quantification in our approach.

Any factor that influences CEST contrast, such as T₁ and T₂ alterations, is expected to affect the accuracy of Cr quantification [45]. While strenuous exercise to the point of fatigue can alter T₁ and T₂* [46], there is no evidence to suggest that mild exercise, such as that performed in our study, induces significant changes in muscle T₁ and T₂. Indeed, a previous study found no appreciable T₂ alterations following mild exercise [22]. Although our participants also performed only mild exercise, we did not specifically evaluate the potential influence of T₂* on MTR_asym.

Exercise can induce shifts in muscle position as well as subtle anatomic changes, including transient post-exercise hypertrophy. To minimize potential shifts, the leg was stabilized by packing the coil with memory foam to provide sufficient restraint during exercise. To address geometric changes, we implemented a rigorous protocol in which muscle-specific ROIs were re-segmented on each post-exercise CEST-weighted image. This procedure was repeated iteratively until muscle boundary stability was confirmed across consecutive time points, thereby ensuring accurate and consistent tracking of muscle-specific recovery dynamics.

Although in this study, we acquired baseline MTR_asym prior to exercise to provide a reference and to aid in B₀ correction. However, as this study focused on post-exercise changes in OXCEST contrast, the necessity of pre-exercise scans is limited. In fact, acquiring such baseline data following the exercise paradigm may be more motion-robust and ensure consistency of slice positioning between baseline and post-exercise acquisitions. Future work will therefore evaluate whether post-exercise baseline scans provide a more practical and reliable strategy for B₀ correction and comparison.

This study has certain limitations. First, we used the WASSR approach to generate pre- and post-exercise ΔB₀ maps [28]. A foundational assumption of the OXCEST framework is that B₀ inhomogeneity remains stable post-exercise. However, gradual field drift may still occur due to the heating of scanner components during gradient-intensive MR sequences, as previously reported in NOE experiments [24, 47]. Similar local magnetic field drifts are also well recognized in functional MRI studies [48]. In addition, post-exercise changes such as muscle recovery or subtle calf movement can affect the B₀, potentially introducing errors in correction. To mitigate such effects, an interleaved gradient echo-based B₀ mapping sequence could be implemented alongside the CEST acquisition to regress out frequency drifts [49]. Acquiring dynamic ΔB₀ along with each frequency offset pair will enable accurate computation of MTR_asymΔB0, and subsequently MTR_asymΔCr and T_Cr, making OXCEST more robust to motion and the resulting B₀ fluctuations over time. Second, our study included only healthy volunteers. In subjects with neuromuscular diseases, lactate accumulation can significantly lower intracellular pH, altering exchange rates and leading to the underestimation of the CEST contrast [19]. Notably, pH-dependent effects have been addressed in recent work by Philip Zhe Sun through the inclusion of an additive correction term [23]. Third, while our assumption that APT and rNOE contributions at ±1.8 ppm are negligible at B₁ = 3 μT is supported by studies in brain tissue [24–26], no analogous investigations have been performed in skeletal muscle, where macromolecular content and exchange properties may differ. Finally, OXCEST and ³¹P-MRS recovery data were acquired in separate sessions on different days. Although exercise bouts were standardized, inter-session variability may have influenced direct comparisons. A larger-scale study with simultaneous monitoring of PCr and Cr recovery kinetics is needed to confirm reproducibility and robustness. Although the comparison between conventionally acquired CrCEST data used to assess conventional T_Cr and LTR-OXCEST demonstrated good agreement, a dedicated study with optimized acquisition parameters for both OXCEST and CrCEST will be necessary to accurately determine the advantages of higher temporal resolution in characterizing Cr recovery kinetics. Such a comparison would be valuable to further substantiate the accuracy and reliability of this novel technology.

6. Conclusion

In this study, we introduced OXCEST, an advanced technique for mapping skeletal muscle OXPHOS by quantifying Cr using dual frequency offset CEST acquisitions. We demonstrated its feasibility by capturing dynamic post-exercise metabolic changes in healthy participants and validated its performance against the gold standard ³¹P-MRS. These results highlight OXCEST’s potential as a non-invasive alternative for in vivo assessment of skeletal muscle mitochondrial OXPHOS. Moreover, the principles underlying OXCEST are broadly applicable to other CEST-based imaging approaches, particularly those requiring high temporal resolution to monitor physiological or pharmacological intervention.

Supplementary Material

Supporting Information

Additional supporting information can be found online in the Supporting Information section.

Figure S1: Schematic of the imaging protocol for Oxidative Phosphorylation Chemical Exchange Saturation Transfer (OXCEST) and 31 P-Magnetic Resonance Spectroscopy (³¹P-MRS) foreach subject. MRI scans were performed over two separate sessions on different days. During each session, participants performed mild plantar flexion exercise inside the scanner for 2 min at a rate of 45 flexions per minute.

Figure S2: Pixel-wise scatterplot of Oxidative Phosphorylation Chemical Exchange Saturation Transfer (OXCEST) magnetization transfer ratio asymmetry (MTR_asym) versus main magnetic field inhomogeneity (ΔB0) in the lateral gastrocnemius (panel A) and soleus (panel) of the same representative subject shown in Figure 1. Black crosses indicate pre-exercise MTR_asym (MTR_asymPre); colored hollow circles represent post-exercise MTR_asym (MTR_asymPost).

Figure S3: Representativefit curves for two subjects with unusually short creatine recovery time (TCr) of less than 15 s.

Figure S4: Comparison of creatine (Cr) recovery time constants (T_Cr) obtained using conventional CrCEST (six frequency offsets) and low-temporal-resolution OXCEST (LTR-OXCEST, two frequency offsets), both with a temporal resolution of 30 s. Each point represents data from lateral gastrocnemius (LG, red triangles), medial gastrocnemius (MG, green circles), or soleus (Sol, blue squares). Results demonstrated strong correlation (R² = 0.77, p < 0.001) and good agreement (Lin’s CCC = 0.83) between methods.

Table S1: Fitted coefficients of OXCEST that modeled MTR_asymPre against B_0Pre in lateral gastrocnemius (LG), medial gastrocnemius (MG), and soleus (Sol) muscles.

Table S2: mrm70133-sup-0001-Supinfo.docx. B₀ values (in ppm) pre-and post-exercise in lateral gastrocnemius (LG), medial gastrocnemius (MG), and soleus (Sol) muscles.

Table S3: Noise level and goodness-of-fit metrics for mono-exponential Creatine recovery in lateral gastrocnemius (LG), medial gastrocnemius (MG), and soleus (Sol) muscles.

Table S4: Effect of B0 correction on MTR_asym in lateral gastrocnemius (LG), medial gastrocnemius (MG), and soleus (Sol) muscles.

Table S5: OXCEST metrics in tibialis anterior (TA) across all subjects (n = 7).

Table S6: Exercise-induced pH changes derived from ³¹P-MRS spectra.

Data Availability Statement

Data will be made available upon request.