Using frequency labeled exchange transfer (FLEX) to separate out conventional magnetization transfer effects from exchange transfer effects when detecting paraCEST agents

Abstract

Paramagnetic chemical exchange saturation transfer (paraCEST) agents combine the benefits of a large chemical shift difference and a fast exchange rate for sensitive MRI detection. However, the in vivo detection of these agents is hampered by the need for high B₁ fields to allow sufficiently fast saturation before exchange occurs, thus causing interference of large magnetization transfer (MT) effects from semi-solid macromolecules. A recently developed approach named frequency labeled exchange transfer (FLEX) utilizes excitation pulses instead of saturation pulses for detecting the exchanging protons. Using solutions and gel phantoms containing the europium (III) complex of DOTA tetraglycinate (EuDOTA-(gly)^-₄), it is shown that FLEX allows the separation of chemical exchange effects and MT effects in the time domain, therefore allowing the study of the individual resonance of rapidly exchanging water molecules (k_ex >10⁴ s^-1) without interference from conventional broad-band MT.

Introduction

Chemical exchange saturation transfer (CEST) is a relatively new contrast mechanism in which low-concentration solutes are detected indirectly by MRI through the transfer of radiofrequency (RF)-induced saturation from labile solute protons to water protons (1-6). Currently, there is an increasing effort underway to design paramagnetic CEST (paraCEST) agents for biological target identification by MRI such as temperature, glucose, nitric oxide, pH, and enzyme activity (2-3,6-12). The main benefit of paraCEST agents over their diamagnetic counterparts is that they can be engineered to have large chemical shift differences (Δω > 10 ppm) between the exchanging proton pools and the free water protons. Consequently, saturation pulses can be applied well away from the water resonance, thus avoiding off-resonance saturation of bulk water. In addition, faster exchanging compounds can be used while still adhering to the slow exchange limit (k_ex < Δω) necessary for CEST. This allows for more saturated spins to be transferred within a given time frame, resulting in an increase of transfer efficiency as expressed by the proton transfer ratio (PTR). However, detection of paraCEST compounds with such rapid exchange rates (k_ex of about 10⁴ s^-1) requires fast magnetic labeling before the proton or water molecule has exchanged to water. When using RF saturation, this needs to be achieved by increasing the B₁ level of the RF irradiation. Unfortunately, this also comes with an increase in conventional MT effects in biological tissues, which, despite the large Δω, can still complicate the analysis in paraCEST studies (1,13-15). The MT component can in principle be separated from the CEST effect by either a multi-pool Bloch equation fit or a recently introduced Lorentzian-line-fit analysis (15), but the accuracy of separation for such analyses depends on the model and fitting parameters and requires high-frequency sampling of the Z-spectra.

Recently, a novel approach dubbed frequency-labeled exchange transfer (FLEX) was proposed, which differs fundamentally from CEST in that it detects exchanging protons using frequency transfer instead of saturation transfer (5,16). Unlike CEST, which is based on a drop in the water signal intensity due to saturation, FLEX detects the modulation of water signal based on the evolution of solute proton magnetization as a function of an evolution time (t_evol), which can be described by a free induction decay (FID). This signal modulation in the time domain provides an opportunity to remove rapidly decaying components with short T₂*, such as MT effects. In addition, the use of excitation pulses allows magnetic labeling of the exchangeable protons with a time scale on the order of microseconds, providing the potential of high labeling efficiency. The aims of this study were to evaluate the FLEX technique for detecting rapidly exchanging paraCEST agents (k_ex >10⁴ s^-1) and for its ability to separate out MT interference in an agarose phantom at physiological temperature.

Materials and Methods

The Eu³⁺-based paraCEST agent

Two phantoms containing 10 mM EuDOTA-(gly)^-₄ in tris buffer (pH = 7) were prepared. One contained just the solution, while the other contained the solution mixed with 4% agarose in order to generate MT effects similar to those observed in biological tissue. The details for synthesizing EuDOTA-(gly)^-₄ are described elsewhere (7,17). Previous work has shown the bound water protons of this paraCEST agent to resonate around 50 ppm from water (at 37 degrees) and that the water molecules exchange with a rate of about 19×10³ s^-1, based on the bound-water life time of 52 μs (18).

FLEX pulse sequence

The FLEX approach has been described in detail previously (5,16). Briefly, the pulse sequence (Fig. 1) contains a preparation time (t_prep) consisting of a series of n label-transfer modules (LTMs), followed by detection using imaging (here a spin echo sequence) or spectroscopy. When using spectroscopy on solutions, detection has to be performed by a readout gradient in order to avoid radiation damping. Within each LTM, exchangeable solute protons are frequency labeled and transferred to water. In the labeling part, a pair of selective 90_x/90_-x RF pulses is applied, in between which chemical shift evolution of the exchangeable protons occurs over a period, t_evol. The offset frequency (o1) of the labeling pulse is selected based on the combined requirement of minimal water excitation (phase 2π at water frequency) and keeping the selective pulses sufficiently short to avoid major signal losses due to the rapid exchange of the coordinating water molecules during the RF pulse. After storage of this frequency information in the form of longitudinal magnetization by the 90_-x pulse, a waiting period, t_exch, allows exchange transfer to the solvent. For long exchange times (needed for slower exchanging protons), this period needs a gradient to avoid radiation damping. However, for the rapidly exchanging paraCEST compound used here, this time was less than a millisecond and a gradient could not be inserted without affecting the time domain signal quality. In FLEX, signal amplification occurs because fresh z-magnetization is present for solute protons at the start of each LTM, as provided through exchange with unlabeled protons from the large water pool. The PTR can be derived under the assumption that RF excitation of the solute pool does not perturb the water pool. This is particularly true for paraCEST agents since the frequency of solute is far away from the water resonance frequency.

$${\mathrm{PTR}}_{\mathrm{s}}=x_{\mathrm{s}}\cdot {\mathrm{\lambda }}_{\mathrm{s}}\cdot (1-e^{-k_{\mathrm{ex},\mathrm{s}}\cdot t_{\mathrm{exch}}})\cdot \sum _{i=1}^n{e}^{\{-1+(i-1)/n\}{t}_{\mathrm{prep}}/T_{1\mathrm{w}}},$$ [1]

where x_s is the fractional concentration of the solute (s) protons and λ_s the labeling efficiency obtained by applying the RF pulse pair, which can be obtained by squaring the fractional amount of transverse magnetization created by single excitation pulse (5,16). The summation term reflects that magnetization transferred to water (w) in the first LTM will experience T_1w decay over the full t_prep, while that transferred in the n_th module will hardly relax. In contrast to CEST, which detects the decrease of water signal, FLEX measures modulation in the water signal as a function of t_evol in a manner depending on the frequency difference (Δω_so1) between the o1 of the 90_x excitation pulse and the solute resonance. In general, if there are multiple types of exchanging solute protons the total effect on the water signal can be described by a FID:

$$I_{\mathrm{w}}=\sum _s{\mathrm{PTR}}_{\mathrm{s}}\cdot e^{-(k_{\mathrm{ex},\mathrm{s}}+1/T_{2\mathrm{s}}^{\ast })\cdot t_{\mathrm{evol}}}\cdot \cos (\mathrm{\Delta }{\mathrm{\omega }}_{\mathrm{sol}}\cdot t_{\mathrm{evol}}).$$ [2]

If some free water protons are inadvertently excited, they will also contribute. Fortunately, similar to standard MRI, the availability of a time domain signal presents the possibility to differentiate components based on different resonance frequencies or different decay rates. For instance, the exponential signal decay term provides the opportunity to remove the water component as well as any rapidly decaying components, such as MT caused by semi-solids if the semi-solid 1/T₂* > (k_ex,s + 1/T_2s*).

Figure 1.

Schematic diagram of the spin-echo based FLEX pulse sequence consisting of a series of n label-transfer modules (LTMs) in which exchangeable solute protons are frequency labeled during evolution time, t_evol and subsequently transferred to water during exchange time, t_exch. Within each LTMs, a pair of rectangular 90_x/90_-x labeling pulses are used (duration: 19.9 μs; strength: 295 μT). All gradients are along z-direction (G_dephase= 9.7 G/cm, 2.2 ms; G_readout= 9.7 G/cm, 4.4 ms). G_dephase and G_readout gradients are applied to avoid radiation damping during the spin echo sequence.

NMR data acquisition

¹H NMR data were acquired on a vertical-bore (89 mm) Varian Unity INOVA 400 MHz spectrometer (Palo Alto, CA). CEST experiments used a 5 s presaturation pulse, with B₁ of 7 μT, applied at regular frequency intervals (2 ppm steps) over a ± 100 ppm range with respect to the free water proton frequency. FLEX experiments used a pair of rectangular 90_x/90_-x labeling pulses (duration: 19.9 μs; strength: 295 μT). The FLEX FID was acquired by varying t_evol from 1 μs to 148 μs using time increments of 3 μs (dwell time); o1 was set to 123 ppm to ensure that the off-resonance effective field generated by FLEX labeling pulses (19) minimized water excitation while still providing sufficient labeling efficiency (λ_s=0.48) for the paraCEST protons. Other FLEX parameters were: t_exch = 400 μs and the number of LTMs was 0, 100, 500, and 1000 for pure solution and 0, 500 for the agarose phantom. The FLEX-FID signal was normalized with respect to the water signal without LTMs. Both FLEX and CEST prepulses were followed by a spin-echo acquisition (TE= 4.7 ms) that included magnetic field gradients (strengths listed in Fig. 1) to avoid radiation damping and allowed acquisition of a projection of the sample. All FLEX and CEST experiments were performed at 37 °C with a recycle delay between subsequent scans of 17 s (≧5 × T_1w). The time taken to acquire CEST experiments was 37 minutes (101 frequency offsets), while FLEX experiments took 14.2 minutes (50 evolution time points). Note that the acquisition time relies on the length of the recycle delay and preparation time for both approaches which is mainly determined by the length of saturation for CEST and the length of the exchange time and number of LTMs for FLEX. In addition, total scan time depends on the number of frequency offsets for CEST and evolution time points for FLEX. Thus, the acquisition time can be substantially reduced for either approach by optimizing these acquisition parameters and the steady state for each particular application. This optimization was not done in the present study which was only meant to show the possibility of separating MT and exchange effects in the time domain and therefore no quantitative comparisons were made between FLEX and CEST. All data were processed in Matlab (MathWorks, Natica, MA).

Results

Figure 2a shows a CEST spectrum of 10 mM EuDOTA-(gly)^-₄ solution. After performing asymmetry analysis with respect to the free water frequency assigned to 0 ppm, a CEST effect (PTR) of about 6.3% is found at 50 ppm. The FLEX FID for 500 LTMs as a function of t_evol is shown in Fig. 2b, indicating a signal modulation with amplitude of about 2.3% of the water intensity at a frequency of Δω_so1 attributed to the paraCEST agent. The FID modulates with a period of about 34 μs, corresponding to the offset frequency of coordinated water (123 – 50 ppm = 29,200 Hz at 400 MHz). The FLEX spectrum in Fig. 2c, generated by Fourier-transform of the FLEX FID in Fig. 2b, shows a peak at 50 ppm that corresponds to the paraCEST peak in the CEST spectrum. Notice that the 50 ppm resonance in the FLEX spectrum is much larger than the free water signal (0 ppm), due to the limited excitation of bulk water (0.17% of the free water signal modulation). The FLEX signal can be amplified by increasing the number of LTMs as illustrated in Fig. 2c. The decay rate of the FLEX FID is caused by the combined effects of exchange and effective transverse relaxation (Eq. 2). Assuming a negligible contribution of T_2,s*, the exchange rate was fitted to be 19×10³ s^-1.

Figure 2.

CEST and FLEX spectra of a solution of 10 mM EuDOTA-(gly)^-₄ in tris buffer (pH = 7.0, T = 37° C). (a) CEST spectrum indicating a 6.3% CEST asymmetry effect (red line) at 50 ppm (b) FLEX FID as a function of evolution time (t_evol) obtained using a preparation time with 500 LTMs revealing a signal modulation of about 2.3% of the water signal and a period of about 34 μs, corresponding to the offset of the coordinated water protons with respect to the transmitter. (c) FLEX spectrum generated by Fourier transformation of the FLEX FID, indicating a paraCEST peak at 50 ppm. The signal is amplified by increasing the number of LTMs.

A Z-spectrum of 10 mM EuDOTA-(gly)^-₄ mixed with 4% agarose is shown in Fig. 3a. It can be seen that a strong MT profile obscures the paraCEST peak indicated by an arrow. Asymmetry analysis reveals a 3% CEST effect at the peak of 50 ppm (red line). Figure 3b shows the FLEX FID for 500 LTMs, indicating a rapid signal drop in the first 20-30 μs, followed by a more gradual decay. A FLEX spectrum obtained by a simple Fourier transform of the FID (Fig. 3c) shows that the agent peak is still obscured by a broad MT profile. To separate out the paraCEST signal from water and broad MT effect, the FLEX data were processed using time-domain analysis. Using prior knowledge about the frequencies of the bound solute water (29200 Hz) and free water (49200 Hz) components, the FLEX FID of the agarose phantom could be separated into three signal components. It is worth mentioning that prior knowledge of the resonance frequencies contributing to the FLEX signal is convenient for time domain fitting but not a prerequisite, since the frequencies can also be fitted from the time domain signal. The initial rapid signal drop with a fitted signal amplitude of 5.3% in the FLEX FID (green line, Fig. 3d) was assigned to the short T₂* component of the semi-solid MT effect, while the moderate decay (1.7%) was assigned to the paraCEST agent (red line, Fig. 3d). The signal of free water (0.2%) did not contribute much to the FLEX FID (blue line, Fig. 3d). The fitted FLEX spectrum shows that the agent’s resonance (red line, Fig. 3e) could be isolated from MT and water (Fig. 3e). In addition, this time-domain fit provided an exchange rate of 21×10³ s^-1. To appreciate the quality of the fit, the residuals for the FID and spectrum are shown in Figs. 3f and 3g, respectively. The small residual is clear in the time-domain (Fig. 3f), but difficult to distinguish in the frequency domain (Fig. 3g).

Figure 3.

CEST and FLEX spectra of 10 mM EuDOTA-(gly)^-₄ mixed with 4% agarose in tris buffer (pH = 7.0, T = 37° C). (a) CEST spectrum and the result of asymmetry analysis (red line) with respect to the water frequency indicating a 3% CEST asymmetry effect. In the Z-spectrum, the paraCEST peak located at 50 ppm (arrow) is obscured by a strong MT profile. (b) FLEX FID for 500 LTMs as a function of t_evol (c) FLEX spectrum obtained by Fourier transform of the FLEX FID. (d) Time domain analysis of the FLEX FID showing three signal components (paraCEST agent bound water (signal magnitude: 1.7%), red line; free water (0.2%), blue line; semi-solid bound water (5.3%), green line). (e) Corresponding FLEX spectra generated by Fourier transform of the individual components in (d). (f) The residual plot obtained by calculating the deviation of time-domain data points in (d) from fitted curve. (g) The residual spectrum corresponding to the Fourier transform of f.

Discussion

ParaCEST agents with large chemical differences between exchanging pools have great potential for designing responsive molecular imaging agents (4-5,7,20). Using conventional RF saturation transfer, a main obstacle in vivo is that high B₁ fields are necessary to sufficiently saturate such compounds with rapidly exchanging protons or water molecules, which leads to strong MT effects that interfere with CEST detection. We have shown that the frequency-label based FLEX technique provides an alternative way to measure paraCEST agents by exploiting the combined advantages of fast labeling by selective pulse excitation and separating out frequency components by time domain analysis. Similar to continuous saturation transfer, sensitivity enhancement can be achieved because the solute proton pool is refreshed by unlabeled protons from the free water pool. In the case of FLEX, the frequency transfer has to be repeated by applying a series of LTMs in which exchangeable solute protons are selectively labeled and subsequently transferred to water.

In order to simulate in vivo conditions, we prepared a solution of a paraCEST agent in a 4% agarose gel to generate a large MT effect. Our results showed that both CEST and FLEX experiments were sensitive to MT effects. The CEST approach still allowed measurement of the paraCEST effect (Fig. 3a) through the use of Z-spectrum asymmetry analysis, but it was reduced about two-fold. Also, in vivo detection may be more complicated due to inherent asymmetry in the MT effect. The FLEX spectrum of the paraCEST compound obtained by direct Fourier transform of the FLEX FID to the frequency domain also was obscured by MT effects (Figs. 3c). Fortunately, the initial rapid signal drop (short T₂*) associated with the MT component in the FLEX FID could be easily separated out by time domain analysis (Figs. 3d, 3e). Such separating out based on decay rate should not be affected by asymmetry and provides a fast alternative to current solutions employing Bloch or Lorentzian fitting of the Z-spectra (15).

Similar to the CEST results, a reduction in FLEX magnitude was observed between the pure solution and agarose phantoms, but this was only about 26%. This reduction is not trivial to interpret. FLEX experiments will induce pulsed MT effects because the use of hard pulses generates a transverse component that dephases rapidly. This can either cause a full signal loss corresponding to the size of this excited semi-solid pool when transferred to the free water pool before being flipped back to the z-axis or a 50% loss when being flipped back. This dephased component accumulates when applying multiple LTMs. We are currently working on the theory describing such effects, which is beyond the goal of this paper. The FLEX approach also allows for differentiating multiple exchanging protons since the modulation of FLEX FID is calculated from the frequency difference between the labeling pulses offset o1 and the resonance frequency of exchanging protons (Eq. 2). As such it has potential for imaging multiple biomarkers simultaneously. Finally, the FLEX method also provides an alternative and quick way to measure exchange rates by fitting FIDs for the individual components (Eq. 2). The k_ex values of the agent’s water molecules in solution and in agarose gel were measured to be 19×10³ and 21×10³ s^-1, respectively, which corresponds well to the previously published rate k_ex =19×10³ s^-1 (18).

In CEST experiments, determination of MTR_asym is complicated by variation of the B₀ field over the sample or subject. For the FLEX time domain analysis, a slight frequency shift caused by B₀ distortion will affect the modulation of FLEX signal, which may be an issue when studying diaCEST agents with resonance frequencies close to water, but not so much for the well-separated paraCEST agents. Both CEST and FLEX approaches will suffer from inhomogeneity in the B₁ field, leading to spatial variation in saturation or labeling efficiency. However, since the FLEX excitation can be applied over a larger frequency range, this can probably be addressed by pulse profiling. In the future, the availability of multichannel transmit hardware (21-22) should reduce this issue for both CEST and FLEX.

One potential disadvantage of FLEX compared to CEST is that a large series of excitation pulses is needed, which may hamper translation to human studies based on specific absorption rate (SAR) concerns. When calculating the SAR, we found that SAR for FLEX (LTM=500) was around 6 times higher than for CEST using the acquisition parameters applied in this study. However, SAR for FLEX can easily be reduced by lengthening the labeling pulse duration (SAR ∝B₁²t_sat) if the exchange loss during the excitation is not a big factor. This would be favorable for more slowly exchanging agents such as diaCEST agents, for which exchange loss would be minimal even with pulse durations on the order of a millisecond. In addition, specialized paraCEST agents with a longer life time (23-24) to limit exchange losses during the RF pulses may be helpful to reduce the B₁ power for frequency labeling

Similar to multidimensional experiments, the length of the FLEX scan time will depend on the number of points sampled for the FID and, within each repetition time, on the number of LTMs applied. For paraCEST agents, the exchange time in each module can be kept short (range of a millisecond or less) and T_1w losses will be limited.

Conclusion

The first results from FLEX experiments on a paraCEST agent at physiologic temperature and pH in 4% agarose were presented. The FLEX approach allowed separation of MT and exchange transfer effects in the time domain, providing easy detection of the bound water resonance of the paraCEST agent. These findings demonstrate potential for in vivo imaging of low-concentration paraCEST agents and a fast alternative to separate out MT contributions without a need for Bloch-equation or Lorentzian fitting of the Z-spectrum.