Advantages of Chemical Exchange-Sensitive Spin-Lock (CESL) Over Saturation Transfer (CEST) for Hydroxyl- and Amine-Water Proton Exchange Studies

Abstract

The chemical exchange (CE) rate of endogenous hydroxyl and amine protons with water is often comparable to the difference in their chemical shifts. These intermediate exchange (IMEX) processes have been imaged by the CE saturation transfer (CEST) approach with low-power and long-duration irradiation. However, its sensitivity is not optimal, and more importantly, the signal is contaminated by slow magnetization transfer processes. Here, the property of CEST signals is compared to a CE-sensitive spin-locking (CESL) technique irradiating at the labile proton frequency. Firstly, using a higher power and shorter irradiation in CE-MRI yields i) increasing selectivity to faster chemical exchange rates by higher sensitivity to faster exchanges and less sensitivity to slower CE and magnetization transfer processes, and ii) decreasing in vivo asymmetric magnetization transfer contrast measured at ±15 ppm. The sensitivity gain of CESL over CEST is higher for a higher-power and shorter irradiation. Unlike CESL, CEST signals oscillate at a very high power and short irradiation. Secondly, time-dependent CEST and CESL signals are well modeled by analytical solutions of CE-MRI with asymmetric population approximation (CEAPA), which can be used for quantitative CE-MRI, and validated by simulations of Bloch-McConnell equations and phantom experiments. Lastly, in vivo amine-water proton exchange contrast measured at 2.5 ppm with ω₁ of 500 Hz is 18% higher in sensitivity for CESL than CEST at 9.4 T. Overall, CESL provides better exchange rate selectivity and sensitivity than CEST; therefore, CESL is more suitable for CE-MRI of IMEX protons.

Introduction

The chemical exchange (CE) process provides novel MRI contrasts to probe tissue pH, mobile proteins, and many metabolites and has gained increasing interest in preclinical and clinical studies, such as stroke, tumor, and cartilage degeneration (1–9). The CE-MRI contrast is generally studied by a chemical exchange saturation transfer (CEST) technique (10), in which biomolecules are indirectly detected by the reduction of water magnetization after their selectively irradiated labile protons exchange with water. In most previous CEST studies, the CE falls into the slow-exchange regime; i.e., the exchange rate (k) ≪ the resonance frequency separation (δ) between water and labile protons, such as studies of the endogenous amide-water proton transfer (APT) effect (1,3,6), as well as many paramagnetic CEST agent applications (11–15). Recently, endogenous CEST has been extended to studies of other proton exchanges, such as the hydroxyl-water exchange from glycogen, glycosaminoglycan, glucose, and myo-inositol (16–23) or the amine-water proton exchange from amino acids, creatine, and mobile protein side chains (24–27). Under commonly used magnetic field strengths (3 T to 11.6 T), many of these exchange processes fall into the intermediate-exchange (IMEX) regime (roughly, 1/3 < k/(2πδ) < 3, with k and δ in s⁻¹ and Hertz units, respectively).

In CEST, effective saturation of faster-exchanging labile protons requires higher irradiation power. However, because the resonance frequencies of endogenous hydroxyl and amine protons are close to water (being ~1–3 ppm), the applicable irradiation power is limited by direct water saturation (DWS) (10). Thus, IMEX-CEST studies have adopted relatively low-power and long-duration off-resonance irradiation (16,18,20,26), which is known to be optimal for slow exchange applications. As such, the sensitivity of IMEX-CEST is not optimized. More importantly, the IMEX signal can not be effectively separated from confounding CEST signals with overlapping frequencies and slow exchange rates, such as the APT signal (1). Additionally, the IMEX signal will be contaminated by the nuclear Overhauser enhancement (NOE) signal from aliphatic protons of mobile protein/lipids (7,17,28–32) and the in vivo asymmetric magnetization transfer contrast (MTC) from semisolid macromolecules when conventional asymmetric analysis is applied. An alternative chemical exchange-sensitive spin-locking (CESL) technique can suppress the DWS (33,34), yielding much wider freedom in the selection of irradiation parameters. Specifically, the choice of much-higher-power and shorter-duration irradiation becomes possible, which can be exploited to improve the sensitivity and/or exchange rate selectivity of IMEX signals. While some promising results have been reported regarding these aspects (25,33), a more systematic comparison of the two techniques and experimental validation are necessary.

Although the CE process can be well described by the Bloch-McConnell equations, an analytical description of the IMEX signal can provide more insight to understand the signal properties, to optimize imaging contrast, and to determine quantitative CE parameters, such as the exchange rate and the biomolecule concentrations. Note that the IMEX signal can not be described by current CEST theoretical models, which were developed in the slow exchange regime (7,24,35–38). To take full advantage of the freedom in the selection of irradiation parameters, it is preferable to have an analytical description of IMEX signals, which may be achieved by the recent model for CE-sensitive MRI under asymmetric population approximation (CEAPA) (33,39). We have applied the steady-state CEAPA solutions to both CEST and CESL MRI for characterizing Z-spectra, hence determining the exchange rate and metabolite concentration (33). Although we have also derived time-dependent CEAPA analytical solutions (40), these solutions were not experimentally confirmed for IMEX applications.

In this work, irradiation time-dependent CEST and CESL signals were simulated by the Bloch-McConnell equations and measured in phantoms with multiple irradiation powers for IMEX protons and compared with analytical CEAPA solutions. Then, the sensitivity and exchange rate selectivity of CEST and CESL were determined, and the accuracy of the CEAPA solutions was examined. The sensitivity advantage of CESL over CEST and the capability of minimizing in vivo asymmetric MTC with high-power and short-duration irradiation were validated by in vivo experiments in normal and ischemic rat brain. Part of the results have been reported in a recent meeting (19).

Theoretical background

Assuming that the population of exchanging pools is highly asymmetric (i.e., the water pool is much larger than others), Trott and Palmer reported the longitudinal relaxation rate in the rotating frame for a two-pool chemical exchange (39) and then extended it to multi-pool exchange (41). We have applied this relaxation model and derived time-dependent analytical solutions for both CEST and CESL MRI signals (25,33,40). Similar solutions have also been obtained by other groups (34,42).

Exchange-mediated relaxation

Chemical exchange between labile protons and water protons can be sensitized by off-resonance continuous-wave irradiation with a radiofrequency (RF) offset of Ω from water, a Rabi frequency of ω₁ (= γB₁/2π), and duration of t_irrad. The relaxation of water proton magnetization under $B_{1,\text{eff}}(=2\pi \sqrt{{\mathrm{\Omega }}^2+{\omega }_1^2}/\gamma )$ can be decomposed into two components: one parallel to the B_1,eff direction is the longitudinal relaxation rate in the rotating frame (R_1ρ), and one perpendicular to the B_{1, eff} direction is the transverse relaxation rate in the rotating frame (R_2ρ). Due to inhomogeneities in B₀ and B₁, in practice, the apparent R_2ρ relaxation rate is much faster than R_1ρ. In multiple-pool exchange systems, provided that all non-water exchanging pools are small and their exchange is predominantly with water, R_1ρ can be expressed as (41):

$$R_{1\rho }=R_1·{cos}^2\theta +(R_2+\sum _i{R}_{ex,i})·{sin}^2\theta ,$$ [1]

where θ = arctan(ω₁/Ω) is the angle between B_1,eff and B₀, R₁ and R₂ are the longitudinal and transverse relaxation rates of water protons excluding the CE effect, and R_{ex, i} is the exchange-mediated relaxation rate between water and the i-th labile proton pool. Assuming that the labile proton concentration is much smaller than water and their R₁ and R₂ relaxation rates are much slower than the exchange rates, R_ex from each individual labile proton pool can be expressed as (41):

$$R_{ex}\approx \frac{4{\pi }^2·P_S·k·{\delta }^2}{4{\pi }^2{(\delta -\mathrm{\Omega })}^2+4{\pi }^2{\omega }_1^2+k^2},$$ [2]

where δ is the separation of labile proton resonance from water proton frequency and P_S is the ratio of labile proton to water proton concentration. The factors of 4π² in Eq. [2] take into account that the units of Ω, δ, and ω₁ are commonly used as Hz (= 2π·rad/s), while the unit of relaxation rates and k is s⁻¹.

CE-sensitive MRI with saturation transfer vs. spin-lock

In the simplest case, a CEST or CESL preparation has a continuous-wave off-resonance RF irradiation that induces an effective B_1,eff, tilted at an angle θ from the B₀ direction (33). With CESL, there are two additional on-resonance hard pulses: one flips the water magnetization from the Z-axis to B_1,eff direction before the off-resonance irradiation for the spin-lock (SL), and the other flips the water magnetization back to the Z-direction after SL for imaging. When the magnetization component perpendicular to the B_{1, eff} dephases instantaneously (apparent R_2ρ →∞), the normalized magnetization at the RF offset of Ω has been derived as (25,40,42):

$$S_{\mathit{irrad}}(\mathrm{\Omega })/S_0=\left\{\left[cos(\lambda \theta )-S_{SS}\right]·e^{-R_{1\rho }·t_{\mathit{irrad}}}+S_{SS}\right\}·cos(\lambda \theta )$$ [3]

where irrad = CEST or CESL irradiation, and S_SS = (R₁·cosθ/R_1ρ) is the normalized steady-state signal along the B_1,eff direction. λ represents the relative error of the SL, which equals 0 for ideal SL and 1 for CEST, because CEST can be considered an approximation of CESL without SL flip pulses (33,40). The λ term is equivalent to (1 − φ/θ), where φ is the actual flip angle of the on-resonance flip pulses (40). Such error can be due to imperfectness in B₁ and/or B₀. The signal intensity of CEST is lower than that of CESL by a projection factor of cosθ at the steady state when t_irrad · R_1ρ ≫ 1 and by cos²θ at very short irradiation (i.e., t_irrad · R_1ρ ≪ 1). The CE-sensitive images are typically obtained by so-called asymmetry analysis, from the difference in the normalized signal intensities of two images at the label (Ω = +δ) and reference (Ω = −δ) frequencies, which is referred to as the asymmetry of the magnetization transfer ratio (MTR_asym) for CEST and the spin-lock ratio (SLR_asym) for CESL. The ratio of SLR_asym to MTR_asym approaches 1/cos²θ at very short t_irrad and 1/cosθ at the steady state, similar to the ratio of S_CESL to S_CEST.

Materials and Methods

Simulations

Numerical solutions of Bloch-McConnell equations and the analytical CEAPA equations (Eqs. [1]–[3]) were performed using Matlab 7.0^® (MathWorks, Natick, MA) Two-pool exchange was simulated to compare the sensitivity of CESL versus CEST and to validate analytical CEAPA solutions. The default parameters assumed in the simulations were δ = 1 ppm (400 Hz or 2510 rad/s at 9.4 T), k = 1500 s⁻¹, R₁ = 0.35 s⁻¹, and P_S = 0.001 if not specified otherwise, and the R₁ and R₂ values of the labile proton were assumed to be the same as water. S_CEST and S_CESL at Ω = ±δ were simulated with the CEAPA solutions for five ω₁ values from 50 to 400 Hz for R₂ of 3 and 25 s⁻¹, and MTR_asym and SLR_asym were calculated as [S(−δ)− S(+δ)]/S₀. R₂ of 3 s⁻¹ is closer to that of aqueous phantom, and R₂ of 25 s⁻¹ is due to in vivo T₂ of 40 ms at 9.4 T (43). A power of SL flip pulse was chosen as 500 Hz, and the duration was calculated so that the flip angle equals θ (= arctan(ω₁/Ω)).

To mimic actual experimental conditions, two additional situations were simulated. First, to evaluate CEAPA solutions for CEST, the RF offset of a labile proton was assumed to be equally distributed between Ω = 390 Hz to 410 Hz for the condition in the presence of small B₀ inhomogeneity of 20 Hz. Specifically, water magnetization was divided into 101 parts with a B₀ shift of −10 to 10 Hz in 0.1-Hz steps. The time evolution of the magnetization vector of each part was calculated by Bloch equations and added together. Secondly, to estimate the effect from the imperfect SL due to B₁ and/or B₀ inhomogeneity, an error of SL λ was assumed, so that the magnetization would be flipped to the cone with an angle of λθ with B_{1, eff} direction by the first flip pulse and flipped back after SL to the cone with an angle of λθ with the Z-axis. λ was simulated for selected values between 0.0 (ideal CESL) to 1.0 (ideal CEST).

To compare the dependence of IMEX versus confounding slow exchange CE-MRI signals on irradiation parameters, four-pool exchanges were simulated assuming the magnetization of water proton exchanges with amine, amide, and aliphatic proton magnetization. While the in vivo aliphatic NOE signals are upfield from water (i.e., negative RF offset), they affect the IMEX CE-MRI signal through the asymmetry analysis. To generate a similar magnitude observed in vivo at 9.4 T, we assumed that R₁ = 0.5 s⁻¹ and R₂ = 25 s⁻¹ as water relaxation rates; k_amine = 7500 s⁻¹, δ_amine = 1200 Hz (3 ppm), and P_amine = 0.0012 for the amine pool; k_amide = 30 s⁻¹, δ_amide = 1400 Hz, and P_amide = 0.0012 for the amide pool (44); and k_aliphatic = 20 s⁻¹, δ_aliphatic = −1200 Hz, and P_aliphatic = 0.003 for the aliphatic pool. SLR_asym was calculated with the CEAPA solutions as a function of irradiation power and duration at Ω = ±1200 Hz. The selectivity of amine signals was evaluated by an index of contaminations from slow-exchanging protons, which was calculated as ε_i = |SLR_asym(i)|/SLR_asym(amine), where i represents amide and/or aliphatic protons. The selectivity is highest when the contamination index approaches 0.

MR experiments and data analysis

All phantom and animal MRI experiments were performed on a 9.4-T horizontal-bore magnet. A 38-mm inner diameter volume coil (Rapid Biomedical, Ohio) was used for phantom experiments, while a detunable volume excitation and surface receiver coil combination (Nova Medical, MA, USA) was used for animal studies. Before the CEST and CESL experiments, T₁ map was obtained using an inversion-recovery sequence, B₁ field was mapped for calibration of the transmit power (13), and B₀ map was obtained using the water saturation shift referencing scheme to evaluate the field homogeneity (45). The pulse sequence for CEST and CESL MRI has been described previously (9), where the CE contrast was first generated by a CEST or CESL preparation, and then, images were acquired with a spin-echo echo-planar imaging (EPI) technique. To normalize CE-sensitive signals, S₀ images were acquired at an offset frequency of 300 ppm.

Phantom experiments

Phantoms were imaged with a field of view = 4 × 4 cm², matrix size = 64 × 64, slice thickness = 5 mm, echo time (TE) = 30 ms, and repetition time (TR) = 16 s. Four sets of experiments were performed at room temperature.

1. To evaluate the difference in saturation transfer (ST) and SL signal sensitivity due to DWS, 2% (by weight) agar phantom was used, where there is no CE effect. ST and SL images were acquired for irradiation times from 0 to 4 s, with a fine step of 5 ms in the initial 80 ms, at a frequency offset of 400 Hz and three ω₁ values of 100, 200, and 400 Hz. Two different field inhomogeneity conditions were compared: a highly homogeneous condition with a water line width of 6 Hz and a more inhomogeneous condition, where the shimming was modified to give a water line width of 46 Hz.

2. To compare the CEST and CESL signals in the intermediate exchange regime, 50 mM myo-inositol (Ins, cyclohexane-1,2,3,4,5,6-hexol, C₆H₁₂O₆) containing hydroxyl groups was dissolved in phosphate-buffered saline (PBS, pH = 7.4) containing 10 mM of phosphate buffer. MTR_asym and SLR_asym were measured at 1 ppm with six ω₁ values of 70, 100, 141, 200, 282, and 400 Hz. For each ω₁, irradiation times were varied from 0 up to 6 s. In addition, on-resonance R_1ρ dispersion was measured for 11 ω₁ values from 125 to 2828 Hz. By fitting the on-resonance R_1ρ dispersion data to Eqs. [1] and [2], P_S, R₂, δ, and k values were obtained. Together with the measured R₁ value, S_CEST and S_CESL, as a function of t_irrad and ω₁, were simulated for Ω = +1 and −1 ppm using Eq. [3], respectively, from which MTR_asym and SLR_asym were calculated as [S(−1 ppm) − S(+1 ppm)]/S₀ and compared with the experimental data.

3. To examine whether the CEAPA solution also works for conditions closer to tissue, 50 mM Ins in PBS was also mixed in 0.7%, 1%, 2%, and 3% agar gel. SLR_asym at 1 ppm was measured with ω₁ = 160 Hz and 21 t_irrad values, ranging from 0 to 2 s in a 0.1-s step. In Eq. [1], the addition of agar increases the water R₂ and also introduces a relaxation term, R_MT, due to the magnetization transfer effect. The effective water R₂ (R_{2, eff} = R₂ + R_MT) values of these mixtures were measured by on-resonance SL experiments with a SL frequency of 4000 Hz to suppress the CE effect (40). Additionally with the P_S, δ, and k values previously obtained from Phantom Expt (ii), MTR_asym and SLR_asym were calculated and compared with the experimental data.

4. To demonstrate that higher-power SL irradiation improves the selectivity of IMEX versus slow CE signals, four phantoms in PBS with a pH of 7.4 and 0.15 mM MnCl₂ were PBS-only, and three metabolites: 30 mM glutamate (Glu) containing a primary amine group with a chemical shift difference of 3 ppm from water, 200 mM nicotinamide (Nic) containing two amide groups with 3.4- and 2.6-ppm offsets from water, and 30 mM Glu together with 200 mM Nic. CESL Z-spectrum was measured with ω₁ = 120 Hz and 4 s, and SLR_asym at 3 ppm was measured with ω₁ = 240 Hz for t_irrad values up to 2 s and with ω₁ = 750 Hz for t_irrad values up to 0.4 s.

In vivo experiments

Thirteen male Sprague-Dawley rats were studied with approval by the Institutional Animal Care and Use Committee at the University of Pittsburgh. Six animals were for in vivo asymmetric MTC studies, 3 were for time-dependent CESL and CEST behavior, and 4 were for the CEST versus CESL comparison of ischemic contrast. The animals were anesthetized with isoflurane (5% for induction and 2% during surgery) in a mixture of O₂ and air gases, with the O₂ concentration kept at 30% throughout the experiment. The right femoral vein and artery were catheterized to deliver maintenance fluid and to monitor the arterial blood pressure, respectively. In four animals, a middle cerebral arterial occlusion was carried out to induce pH-reducing permanent ischemia in the left hemisphere (46). After surgery, the isoflurane level was reduced to 1.3%–1.5% during MRI scans. The dynamic blood pressure and end-tidal CO₂ were monitored. End-tidal CO₂ level was kept within 3.0% and 4.0%, and the rectal temperature was maintained at 37.0 ± 0.5°C using a feedback-controlled heating pad.

All in vivo CE-MRI images were acquired by the spin-echo EPI sequence with a field of view = 3.2 × 3.2 cm², matrix size = 64 × 64, TE = 28 ms, slice thickness = 2 mm, and four consecutive slices without gap. Three in vivo experiments were performed.

1. To evaluate the dependence of in vivo asymmetric MTC on irradiation parameters, MTR_asym was measured in rat brains at ±15 ppm (n = 6) with TR = 10 s and calculated as [S(−15 ppm) − S(+15 ppm)]/S₀. Because asymmetric MTC extends to a wide frequency range (47), a large offset of 15 ppm was chosen to minimize contaminations from direct water saturation, as well as CE and NOE effects. Four irradiation power and duration pairs, which have been recently applied in hydroxl- and amine-water exchange studies, were compared: ω₁ = 63 Hz for 4 s (20,23), 150 Hz for 2 s (48), 250 Hz for 1 s (49), and 500 Hz for 0.15 s (25,50). To compare the quantitative data in gray and white matter, regions of interest (ROIs) were selected in the cortex and the corpus callosum area.

2. Amine-water proton exchange (APEX)-sensitive CEST and CESL measurements were performed at 2.5 and −2.5 ppm with continuous-wave ω₁ of 500 Hz, t_irrad between 0 and 500 ms (25), and TR = 7 s. This offset was chosen, because for the fast amine-water proton exchange, the frequency of specificity/selectivity reduces, and the CE signal can be detected with similar sensitivity by adjusting ω₁ and t_irrad (25). To examine whether the CESL results are sensitive to inaccurate spin-locking due to B₁ and B₀ inhomogeneities, CESL data of normal animals (n = 3) were also acquired with a 30% error in the flip angle of on-resonance flip pulses (i.e., λ = 0.3), without changing the long off-resonance irradiation pulse.

3. The same APEX parameters shown above in In Vivo Expt (ii) were used to compare the ischemic contrast of CEST versus CESL. To identify ischemic regions in stroke animals (n = 4), ADC maps were obtained using a spin-echo EPI sequence without diffusion-weighting and with a diffusion-weighting b value of 1200 s/mm² along six different directions. In the ischemic animals, MTR_asym and SLR_asym maps were obtained. An ischemic ROI, with about 2 × 2 mm² in one slice, was selected from the ADC map, and a control ROI was chosen in the same brain region contralateral to the ischemic ROI. Then, MTR_asym and SLR_asym were compared as a function of irradiation time from 0 to 500 ms.

Results

Sensitivity of CEST vs. CESL MRI: Simulations and Phantom Studies

Time-dependent magnetization was simulated using Bloch-McConnell equations with 150 Hz irradiation applied at 1 ppm (Fig. 1). The CESL signal is dependent on the error of the flip pulses, λ (Fig. 1A). For λ = 1, the simulated CESL signal is the same as the CEST signal, while λ of 0 is for the ideal CESL condition. At a short t_irrad, CEST shows a significant signal oscillation, as expected (33,42,51,52). When λ decreases from 1.0 to 0.0 (e.g., on-resonance flip angle increases from 0 to θ), the signal intensity increases with reduced oscillations. For λ ≤ 0.5, the CESL signal is already fairly close to that of the ideal CESL. With the presence of field inhomogeneity, the apparent R_2ρ relaxation increases, so that the oscillation quickly disappears (Fig. 1B), and the Bloch-McConnell simulation results match well with the CEAPA solutions for both CEST and CESL. To experimentally confirm the simulation data, saturation transfer and spin-locking MRI of 2% agar phantoms were measured at 1 ppm with ω₁ = 100, 200, and 400 Hz for a water spectral line width of 6 Hz and 46 Hz. Consistent with the simulation data (Fig. 1A and B), the t_irrad-dependent ST MRI signal oscillation is clear when the field homogeneity is high (Fig. 1C) but disappears quickly with large field inhomogeneity (Fig. 1D).

Fig. 1. Simulated and experimental signal intensities as a function of off-resonance irradiation duration.

(A–B) Time-dependent signals were simulated using the Bloch-McConnell equations during off-resonance irradiation with pulse power ω₁= 150 Hz and frequency Ω = 400 Hz (i.e., 1 ppm at 9.4 T). Other parameters used were R₁ = 0.35 s⁻¹, R₂ = 3 s⁻¹, and P_S = 0. In reality, the flip of water magnetization in the SL preparation is not likely to be ideal due to imperfect B₁ and/or B₀. Thus, the impact of flip angle error (λ) to the SL signals was simulated for four different λ values (A). λ = 1 and 0 corresponds to CEST and ideal CESL, respectively. λ of 0.5 significantly reduces the signal oscillation and improves the sensitivity compared to CEST. When there is a small B₀ inhomogeneity with ±10 Hz distribution, the oscillation component of CEST vanishes quickly (gray line in B), and the simulation results of the Bloch-McConnell equations match very well with those obtained from the CE-sensitive relaxation model with asymmetric population approximation (CEAPA) solutions (circles in B). (C–D) Time-dependent saturation transfer (ST) and spin-lock (SL) data of 2% agar phantom were obtained at Ω = 400 Hz under two field shimming conditions: water spectral line width (FWHM) = 6 Hz (C) and 46 Hz (D). The ST MRI signal oscillation at short irradiation times (<100 ms) is reduced when there is larger field inhomogeneity. Inset: the ROI was overlaid on the ω₁ map. S_irrad: the signal intensity with off-resonance irradiation at 1 ppm; t_irad: off-resonance irradiation time.

Fig. 2 compares the optimal ω₁ and t_irrad values for CESL with R₂ of 25 s⁻¹ for three exchange rates: k = 100 s⁻¹ (A), 1500 s⁻¹ (B), and 5000 s⁻¹ (C). Overall, CE-MRI is much more sensitive in the slow exchange regime, and the maximal SLR_asym is reduced for k = 5000 s⁻¹ (k/δ = 2) than k = 1500 s⁻¹ (k/δ = 0.6). With increasing k, the optimal CE sensitivity occurs at higher ω₁ and shorter t_irrad. The ratio of SLR_asym to MTR_asym is nearly independent of k value and is shown in Fig. 2D, which is larger at higher ω₁ and shorter t_irrad.

Fig. 2. Exchange rate-dependent CESL sensitivity as a function of off-resonance irradiation power and duration.

SLR_asym as a function of ω₁ and t_irrad was simulated from CEAPA solutions for exchange rate k = 100 s⁻¹ (A), 1500 s⁻¹ (B), and 5000 s⁻¹ (C), where the color bar is in units of % of S₀. The ratio of SLR_asym to MTR_asym (D) is nearly identical for all three exchange rates. Other parameters used were R₁ = 0.35 s⁻¹, R₂ = 25 s ⁻¹, P_S = 0.001, and offset frequency δ = 400 Hz. Faster CE-sensitive contrast is optimal at a higher irradiation power and shorter duration (red region in C). ω₁: off-resonance irradiation power; t_irad: off-resonance irradiation time.

To further compare CESL and CEST sensitivity in the IMEX regime, MTR_asym (dashed line) and SLR_asym (solid line) were simulated using CEAPA analytical solutions for k = 1500 s⁻¹ with R₂ of 3 s⁻¹ (Fig. 3A and 3B) and 25 s⁻¹ (Fig. 3C and 3D). Three observations were noted. i) Compared to R₂ of 3 s⁻¹, the CE signal with a larger R₂ of 25 s⁻¹, measured by both MTR_asym and SLR_asym, is reduced due to the increased R_1ρ (but the same R_ex, see Eq. [1]), and the CE signal peaks at lower ω₁ (Fig. 3C vs. 3A). ii) At ω₁ of 50 Hz, MTR_asym and SLR_asym are nearly identical and increase with t_irrad. With increasing ω₁, the peaks of MTR_asym and SLR_asym shift to smaller t_irrad values, because the CE contrast is additionally affected by the increase of ω₁ (or θ) due to the cosθ term (see Eq. [3]). iii) With the same irradiation parameters, SLR_asym is always higher than MTR_asym, and the difference increases with ω₁. The ratio of SLR_asym to MTR_asym approaches 1/cos²θ at very short t_irrad and 1/cosθ at the steady state (Figs. 3C and 3D).

Fig. 3. Simulated and experimental CEST versus CESL contrast.

(A–D) MTR_asym and SLR_asym, as a function of irradiation duration t_irrad, were simulated with CEAPA solutions for selected ω₁ values. Two different R₂ values were used: 3 s⁻¹ (A–B) and 25 s⁻¹ (C–D). SLR_asym (solid lines in A and C) is always larger than MTR_asym (dashed lines), but the difference is more significant at higher ω₁ and shorter t_irrad values. The ratio of SLR_asym to MTR_asym (B and D) ranges between 1/cos²θ at very short t_irrad and 1/cosθ at long t_irrad, approaching the steady state. Other parameters used were R₁ = 0.35 s⁻¹, P_S = 0.001, and δ = 400 Hz. (E–F) MTR_asym (open symbols) and SLR_asym (filled symbols) of 50 mM myo-inositol (Ins) in PBS were measured at 1 ppm with ω₁ = 100, 200, and 400 Hz at varying irradiation durations (E), and their ratios were obtained (F). For the highest ω₁ of 400 Hz, we only acquired t_irrad up to 2 s to minimize RF heating in the samples. MTR_asym and SLR_asym peaks shift to shorter t_irrad at higher ω₁. The ratio between SLR_asym and MTR_asym increases with ω₁ and is larger at shorter irradiation times (F). To verify the experimental data with CEAPA models, on-resonance spin-locking experiments as a function of ω₁ were performed, and chemical exchange parameters were obtained (see texts). Simulated MTR_asym and SLR_asym data (lines in E) using these chemical exchange parameters match well with the experimental data (symbols in E). t_irad: off-resonance irradiation time.

To confirm our CEAPA simulation data with experimental data, MTR_asym and SLR_asym of 50 mM Ins in PBS were measured at six ω₁ levels and are shown in Fig. 3E–F for three different ω_1. Peaks of MTR_asym and SLR_asym shifted to shorter t_irrad with increasing ω₁. For the same ω₁, SLR_asym is always higher than MTR_asym, and the SLR_asym/MTR_asym ratio increases with increasing ω₁ and shorter t_irrad (Fig. 3F), similar to the prediction from the CEAPA solutions (see Fig. 3B).

CEAPA Solutions vs. Tissue-Mimicking Phantom Data

To examine whether CEAPA solutions can explain the experimental irradiation time-dependent MTR_asym and SLR_asym data in the Ins phantom (Fig. 3E), P_S = 0.003, δ = 370 Hz (2323 rad/s), k = 1250 s⁻¹, and R₂ = 0.4 s⁻¹ were obtained by fitting on-resonance R_1ρ dispersion data with Eqs. [1] and [2] (19). Using these fitted values of Ins (except for an R₂ of 0.5 s⁻¹, which gives a better match than 0.4 s⁻¹) and a measured R₁ of 0.35 s⁻¹, irradiation time-dependent MTR_asym and SLR_asym were simulated (Fig. 3E, lines). Both MTR_asym and SLR_asym simulation data (lines) matched well with the experimental data (symbols).

To further examine whether the CEAPA solutions apply to conditions closer to in vivo, we measured the SLR_asym (symbols in Fig. 4), R₁, and R₂ of four Ins samples with different agar concentrations. With increasing agar concentration, SLR_asym, as well as the peak t_irrad, decreased. The water R_{2, eff} value is nearly proportional to the agar concentration, being 4.02, 5.97, 12.50, and 18.08 s⁻¹ for 0.7%, 1%, 2%, and 3% agar samples, respectively. In contrast, the water R₁ value only very weakly increases with agar concentration, being 0.35, 0.36, 0.37, and 0.38 s⁻¹ for these samples. Using the aforementioned fitted results of Ins and the measured R₁ and R_{2, eff} values, SLR_asym can be predicted from Eq. [3] (solid lines in Fig. 4), which match well with the experimental data for all four samples (symbols in Fig. 4). Thus, these results indicate that CEST and CESL data can be well explained by the analytical CEAPA solutions for both aqueous and tissue-mimicking systems.

Fig. 4. Experimental vs. simulated CESL signal of tissue-mimicking phantoms.

SLR_asym data of 50 mM myo-inositol (Ins) in 0.7%, 1%, 2%, and 3% agar were obtained with a ω₁ of 160 Hz. SLR_asym data were also simulated by the CEAPA models with chemical exchange parameters (see texts). The measured SLR_asym data of Ins phantoms (symbols) can be well explained by the simulated data by the CEAPA models (solid lines), indicating that analytical CEAPA can be applied for quantitative models of CE-MRI. t_irad: off-resonance irradiation time.

Exchange Rate Selectivity of CE-MRI to IMEX Protons

Due to their overlapping frequencies, the in vivo IMEX MTR_asym or SLR_asym signal contains contaminations from both amide and aliphatic protons (28,31,32,53). Besides the sensitivity enhancement, another major benefit of using higher irradiation power is sensitizing faster exchange processes, as can be seen in Fig. 2A–C. To demonstrate this property, SLR_asym at 3 ppm was simulated as a function of irradiation power and duration for an amine pool, an amide pool, an aliphatic pool, and the combination of all three pools (Fig. 5A–D). The amide signal is slightly positive, whereas the aliphatic proton signal is largely negative. With low-power and long-duration irradiation, the negative SLR_asym is dominated by the contribution from aliphatic protons, whereas with high-power and short-duration irradiation, the positive SLR_asym is dominated by the contribution from amine protons (Fig. 5D). This is very similar to our observations in vivo (25).

Fig. 5. Selectivity of amine CESL MRI increases with high irradiation power: simulations.

SLR_asym as a function of ω₁ and t_irrad was simulated at an off-resonance frequency 3 ppm (i.e., 1200 Hz) from CEAPA solutions for amine pool only with population P_amine = 0.0012, exchange rate k_amine = 7500 s⁻¹, and chemical resonance frequency δ_amine = 1200 Hz (A); amide pool only with P_amide = 0.0012, k_amide = 30 s⁻¹, and δ_amide = 1400 Hz (B); aliphatic pool only with P_aliphatic = 0.003, k_aliphatic = 20 s⁻¹, and δ_aliphatic = −1200 Hz (C); and all three pools together (D), where the color bar is in units of % of S₀. The contamination of slow exchanging protons to the amine signal — i.e., ε = |SLR_asym(non-amine)|/SLR_asym(amine) — was shown as a percentage in (E) for amide pool only, (F) for aliphatic pool only, and (G) for the combination of amide and aliphatic pools. ω₁: off-resonance irradiation power; t_irad: off-resonance irradiation time.

The contamination of amide and aliphatic protons to the amine CE signal was shown in Fig. 5E–G. If ε (=|SLR_asym(non-amine)|/SLR_asym(amine)) = 0, then there is no contamination. At a high irradiation power, such as >400 Hz, the contamination of amide and hydroxyl protons is minimal. Compared to SLR_asym which is sensitive to both ω₁ and t_irrad (Fig. 2A–C and 5A–D), the index of contamination ε is mainly determined by ω₁ and is only weakly dependent on t_irrad. This indicates that the exchange rate selectivity is mainly determined by the irradiation power, whereas the selection of t_irrad is important for the optimization of sensitivity.

The selectivity of IMEX versus slow exchange by ω₁ was demonstrated in phantoms. Amide proton was chosen for this purpose, given the difficulty of constructing phantoms with the aliphatic NOE effect only. In Fig. 6A, the Z-spectra and the SLR_asym of Glu and Nic phantoms showed that the CESL signals of amine and amide protons are overlapping. With a relatively low ω₁ of 120 Hz, the SLR_asym signal from Nic with amide protons is higher than that of Glu with amine protons at 3 ppm — the offset frequency of amine protons. SLR_asym at 3 ppm of the Glu and Nic phantoms was compared for two higher ω₁ values. With a medium ω₁ of 240 Hz, Glu and Nic have similar CE sensitivity, with peaks at t_irrad > 1 s (Fig. 6B). When ω₁ was increased to 750 Hz, the peak of SLR_asym shifted to 0.2 s or less (Fig. 6C), where the CE sensitivity of Glu is much larger than that of Nic. SLR_asym maps, measured at 3 ppm, are shown in Fig. 6D; both Glu and Nic contribute equally for a ω₁ of 240 Hz and t_irrad of 1 s, but the Glu signal is predominant for a ω₁ of 750 Hz and t_irrad of 0.1 s. At a large ω_1, an increase in θ (and the R_1ρ) decreases the water magnetization and, hence, the amide-CE signal (Eq. [1]). But the CE signal from Glu does not drop, because the higher ω₁ improves the saturation efficiency.

Fig. 6. Selectivity of amine CESL MRI increases with high irradiation power: phantom experiments at room temperature.

CESL experiments were performed at various experimental parameters for 30 mM glutamate (Glu), 200 mM nicotinamide (Nic), 30 mM Glu plus 200 mM Nic, and PBS phantoms. (A) Z-spectra (thick) and 3-times amplified SLR_asym (thin) measured with a low power of ω₁ = 120 Hz and t_irrad = 4 s show that amine-CESL signals are overlapping with that of amide. (B–C) SLR_asym at 3 ppm was calculated as a function of t_irrad for a medium power ω₁ = 240 Hz (B) and a high power of 750 Hz (C). With 240 Hz, Glu (black) and Nic (green) signals are similar and reach a peak at t_irrad > 1 s. With 750 Hz, the Glu signal reaches a peak at t_irrad <0.2 s, where the Nic SLR_asym is minimized. The enhancement of amine selectivity is also shown in the SLR_asym map (D) obtained with ω₁ = 750 Hz and t_irrad = 0.1 s. ω₁: off-resonance irradiation power; t_irad: off-resonance irradiation time

Overall, with the coexistence of slow- and intermediate-exchange species, a higher ω₁ and shorter t_irrad pulse can be exploited to enhance the selectivity and sensitivity to IMEX CE-MRI signals and minimize the contamination from slow-exchange protons. Note that CEST has similar selectivity properties, but a very short irradiation time can not be used because of DWS-induced signal oscillation, as shown in Figs. 1A and 1C.

Minimizing in vivo MTC asymmetry with high irradiation power

There is a concern that a higher ω₁ may induce larger MTC asymmetry from semisolid macromolecules, resulting in a larger error of MTR_asym for CE-MRI. To examine this issue, MTR_asym at 15 ppm was measured in rat brains (Fig. 7), which is dominated by the MTC asymmetry. Similar to the MTC effect, this MTC asymmetry is significantly higher in magnitude in white matter than gray matter. In general, MTC asymmetries in both the cortex (blue pixels) and corpus callosum (green pixels) decrease with higher power and shorter duration. With 500 Hz and 0.15 s, MTC asymmetry is suppressed, indicating that a high-power and short-saturation pulse is able to reduce the MTC asymmetry and increase the signal selectivity to IMEX protons.

Fig. 7. Dependence of in vivo asymmetric MTC effect on irradiation parameters.

In vivo MTR_asym of rat brains was measured at an RF offset of 15 ppm with four different irradiation power and duration pairs. Maps of one representative animal are shown (A), and the averaged MTR_asym (n = 6 rats) was calculated from the cortex (blue pixels in A) and corpus callosum (CC, green pixels) ROIs (B). The asymmetric MTC is higher in the white matter than in gray matter and decreases with a higher irradiation power and shorter duration condition.

In vivo CESL vs. CEST in the IMEX domain

Fig. 8 shows in vivo CEST and CESL signals measured at an RF offset of ±2.5 ppm and an ω₁ of 500 Hz, which is sensitive to amine-water proton exchange (25). In a single pixel in normal rat brain cortex (the red pixel in Fig. 8A, inset), a large oscillation was observed at t_irrad < 100 ms for the reference CEST images acquired at −2.5 ppm and for MTR_asym, but not for CESL (Fig. 8A–B). Consistent with the simulation results (Fig. 1A), a flip angle error of <0.3 is sufficient to minimize the CE-MRI signal oscillation, leading to results close to the ideal CESL (Figs. 8A and 8B). Similar signal behavior can be observed in other pixels and all animals (n = 3). Since the peak response of MTR_asym (and SLR_asym) occurs at about 100 to 200 ms (Fig. 8B), the initial fluctuation may interfere with the proper selection of irradiation time for maximizing MTR_asym. For t_irrad = 150 ms, the whole brain-averaged SLR_asym is 18% higher than MTR_asym and is 15% higher with λ = 0.3 (see Fig. 8B). While CESL significantly improves the sensitivity to MTR_asym, this enhancement is not due to the additional asymmetric MTC caused by the two flip pulses. With t_irrad = 0, SLR_asym is only −0.07 ± 0.15% (n = 3).

Fig. 8. In vivo amine-water proton exchange (APEX) signals measured with CEST and CESL.

CEST and CESL MRI of rat brains were measured with RF offsets of ±2.5 ppm and ω₁ of 500 Hz. (A–B) Data of a single pixel (inset in A) were examined with different flip angle error λ (=1 for CEST and 0 for CESL). The CEST signal intensity at −2.5 ppm and MTR_asym (black lines) oscillate for short t_irrad. Similar to the simulation and phantom studies, the in vivo results with flip angle error λ = 0.3 are already close to those of ideal spin-lock with λ = 0 and have higher sensitivity than saturation transfer with λ = 1. (C) In ischemic rats, the MTR_asym and SLR_asym maps with t_irrad = 0.15 s show similar ischemic contrast, but the latter has about 20% higher sensitivity. While SLR_asym maps are calculated from the difference of two normalized maps of S/S₀ at ± 2.5 ppm, this ischemic contrast mostly originates from the APEX-weighted signal at 2.5 ppm. Note the difference gray scales of the maps. (D) The averaged MTR_asym and SLR_asym (n = 4) as a function of t_irrad were obtained from the 2×2 mm² ROIs in the ischemic ipsilateral (blue) and normal contralateral sides (red box). The CESL SLR_asym signal (solid) has higher sensitivity and higher ischemic contrast (red vs. blue) than CEST MTR_asym (dashed).

In ischemic rats, the ischemic lesion area can be observed easily in both the SLR_asym and MTR_asym maps obtained with a ω₁ of 500 Hz and t_irrad = 150 ms (Fig. 8C), but the SLR_asym intensity was larger than MTR_asym. This ischemic contrast is mainly due to the APEX-weighted signal at the 2.5 ppm, because a negative contrast can be detected at this RF offset but not at the reference frequency of −2.5 ppm. To examine t_irrad-dependent CE-MRI signals, two ROIs were chosen: an ipsilaterial ischemic region selected from the ADC map and a contralateral normal region. Both MTR_asym and SLR_asym peaked at a t_irrad of 150 ms for the contralateral ROI and 200–250 ms for the ipsilateral ROI (Fig. 8D). The shift in peak t_irrad in the ischemic ROI is likely due to a longer T₂ induced by edema and, consequently, an increased optimal irradiation duration (25). The maximum contrast between ipsi- and contra-lateral ROI occurred at t_irrad = 300 ms and is 4.2 ± 1.2% for MTR_asym and 5.0 ± 1.3% for SLR_asym (n = 4).

Discussions

Recent studies have broadened the spectrum of CE-MRI to a wide range of biomolecules with hydroxyl or amine groups. These hydroxyl- and amine-water exchange rates are usually between 1000–10,000 s⁻¹ (2,7,9), and the chemical shifts from water are ~1–3 ppm, which is 800–2500 rad/s at 3 T and 2500–7500 rad/s at 9.4 T, respectively. Therefore, many of these proton exchanges fall into the IMEX regime. While most of these IMEX studies adopted the conventional CEST approach of irradiation with a long duration (>1s) and relatively low power (~40 to ~150 Hz) (16–18,20,22,23,26,49), our results show that such a choice of irradiation parameters is not optimal. A simple addition of two on-resonance flip pulses to convert CEST to CESL acquisition can suppress the DWS and provide important benefits. Namely, with the CESL technique, irradiation parameters can be optimized in a much wider range, so that sensitivity and/or selectivity of IMEX CE-MRI signals can be improved.

Improve the IMEX Sensitivity: CEST vs. CESL

The sensitivity of CE-MRI is closely dependent on the rate of chemical exchange (slow vs. intermediate exchange) as well as imaging parameters. In CEST, the sensitivity is approximately proportional to the saturation efficiency, defined as $\alpha =\frac{{\omega }_1^2}{({\omega }_1^2+k^2)}$ (2,54,55). For slow-exchange applications, such as the well-studied amide proton transfer effect (56), CEST and CESL are almost identical, and the MTR_asym is optimal at the steady state (33). For CEST studies of hydroxyl- or amine-water proton exchange with an exchange rate of ~5000 s⁻¹ (26,33), the saturation efficiency is only 2.5% for a typical ω₁ of around 125 Hz (800 rad/s) but can be increased to 28% with a ω₁ of 500 Hz. However, because amine- and hydroxyl CE-MRI is often obtained at 2–3 ppm and ~1 ppm, respectively, an increase of the saturation efficiency induces a larger DWS effect.

For IMEX applications, a high ω₁ and short irradiation often provides optimal CE sensitivity (Fig. 2) (25,26), in which case CESL will have a higher sensitivity than CEST. In practical applications, the B₀ and B₁ inhomogeneity would lead to imperfect SL and reduce the sensitivity gain of CESL over CEST; therefore, schemes with improved SL accuracy, such as adiabatic pulses, would be beneficial. Importantly, the requirement of the accuracy of the two flip pulses in CESL is not stringent, and the sensitivity enhancement can be achieved, even with 50% error in the flip pulses. Our in vivo APEX studies show an 18% sensitivity gain when CESL is used and a 15% gain when the B₁ of the flip pulse is intentionally reduced by 30%. Generally, the sensitivity gain of CESL over CEST would be larger for systems for a higher ω₁ and shorter irradiation duration (Fig. 2D), which will optimize the sensitivity for a faster chemical exchange rate. Similarly, the sensitivity gain of CESL over CEST would be larger for smaller chemical shifts from water (i.e., a larger DWS effect). Therefore, the sensitivity enhancement of CESL should also be higher at lower magnetic fields, such as a clinical field of 3 T.

In addition to irradiation power and duration, the RF offset can also be adjusted in CESL, although it is only considered at the labile proton frequency in this work. From Eq. [2], the full width at a half-maximum of R_ex as a function of off-resonance frequency equals $2\times \sqrt{{\omega }_1^2+k^2/{(2\pi )}^2}$ (33). Thus, for an exchange rate k comparable to or faster than the chemical shift δ, the CE contrast can be detected in a wide range of RF offsets (see Fig. 8D in (33)). Specifically, CE-sensitive contrast can even be obtained by CESL applied on the water resonance frequency (i.e., on-resonance SL), which is not possible with the CEST approach. Although the R_1ρ of on-resonance SL has contributions from other relaxation pathways besides CE and poor chemical specificity of a certain type of labile proton, it can be applicable in a dynamic study with administered CE-sensitive agents, where the difference of R_1ρ before and after the administration can be mostly attributed to the change of CE-sensitive agent (e.g., glucose (57)).

Enhance the exchange rate selectivity: CESL with high power and short pulse

Various endogenous labile protons have similar resonance frequencies, which are close to the water resonance frequency. The amine CE-MRI has contamination from amide protons with close resonance frequencies, which is more significant for lower irradiation power (Figs. 5 and 6). Recent CEST studies have also shown large NOE signals of up to ~10% from aliphatic protons of mobile protein/lipids (28,31,32). These aliphatic protons cover a wide resonance frequency range of 0 to −5 ppm from water and therefore will affect amine as well as hydroxyl CE-MRI signals when the conventional asymmetry analysis is adopted. Another contamination to CE-MRI is the large asymmetric MTC from semi-solid macromolecules, which is very broad in a Z-spectrum due to its extremely short proton T₂, with a center of spectrum of around −2.5 ppm. Recent brain studies have shown that the magnitude of the asymmetric MTC effect is around −2% and −4% at 3 T and 9.4 T (25,47), respectively, for an irradiation power of 1–3 μT, often used for hydroxyl and amine CEST studies. While these confounding signals cannot be separated from that of the IMEX signal, one distinct property to be exploited is their large difference in exchange rates.

With the existence of multiple exchange rates, the exchange rate selectivity can be achieved by tuning the irradiation frequency ω₁ to the targeted exchange rate (13,50). Also, long irradiation duration is preferred for slow exchange effects, which accumulate slowly. Thus, a combination of higher irradiation power and shorter duration effectively reduces the signal contributions from slow exchange processes and increases the selectivity to faster exchange processes, as shown in our phantom (Fig. 6) and in vivo results (Fig. 7). Due to DWS-induced oscillation in CEST at short irradiation times, the CESL approach would be a better choice. It should be noted that in the slow exchange regime, exchange rate filtering can also be achieved by a train of pulses with varied delays or varied rotation angles (58,59), in which the CE contrast may be tuned to labile protons with exchange rates of up to several hundred times per second.

Analytical model and quantification of IMEX effects

Even though R_1ρ appears to be specific only to spin-locking experiments at first glance, it represents the longitudinal relaxation rate in the rotating frame during any on- or off-resonance RF pulses. As described in Eqs. [1] and [2], the CE-MRI signal is sensitive to both CE and non-CE-related relaxation rates. The derivation of the CE-mediated relaxation term R_ex (Eq. [2]) requires an assumption that the labile proton concentration is much smaller than water, which is valid for in vivo studies where water is the dominant pool, as compared to other biomolecules. Another assumption is that the R₁ and R₂ relaxation rates of the labile proton are much smaller than the chemical exchange rate k, which has been extended by Zaiss et al. to a more general case where R₂ can be much faster than k (42). The analytical solution (Eq. [3]) was validated by the simulation of Bloch-McConnell equations and by the close match with experimental data. These irradiation time-dependent solutions can be applied to a whole range of chemical exchange studies with slow, intermediate, or fast exchange rates and allow us to characterize CE-sensitive MR data better and optimize imaging parameters, such as irradiation power and duration, easily.

R_1ρ is sensitive to water R₁ and R₂ of protons, as seen in Eqs. [1] and [2]. For slow exchange applications, θ is negligible and the CE effect is mostly competing with water R₁, while for IMEX studies, θ is relatively large; so, the R₂ contribution to R_1ρ becomes significant, affecting CE-sensitive MR signals (see Fig. 3 with R₂ of 3 s⁻¹ vs. 25 s⁻¹). Therefore, to quantify IMEX effects and obtain information, such as metabolite concentration or exchange rates, it is necessary to remove these non-CE components. To this end, R_1ρ can be measured at the labile proton frequency and the reference frequency, and the non-CE relaxation rates should be cancelled or minimized by taking the asymmetry of R_1ρ (R_{1ρ, asym}). We have recently proposed an acquisition approach, dubbed irradiation with toggling on-resonance inversion preparation, to measure R_1ρ efficiently (40).

Conclusions

We have demonstrated that the signals of CEST and CESL MRI can be well described by analytical CEAPA solutions and that CESL MRI provides higher sensitivity and exchange rate selectivity than CEST. Therefore, CESL would be especially valuable for hydroxyl- or amine-water proton exchange applications, which are close to the intermediate exchange regime.

Abbreviations used

APA

asymmetric population approximation

APEX

amine-water proton exchange

CE

chemical exchange

CEAPA

chemical exchange model with asymmetric population approximation

CESL

chemical exchange spin-lock

CEST

chemical exchange saturation transfer

DWS

direct water saturation

FWHM

full width at half maximum

IMEX

intermediate exchange

MT

magnetization transfer

MTC

magnetization transfer contrast

MTR_asym

asymmetry of magnetization transfer ratio

NOE

nuclear Overhauser enhancement

PBS

phosphate-buffered saline

R_1ρ

spin-lattice relaxation rate in the rotating frame

RF

radiofrequency

ROI

region of interest

SL

spin-lock

SLR_asym

asymmetry of spin-lock ratio