Osteoarthritis is one of the debilitating joint diseases of the musculoskeletal system. It affects more than 10% of adults and 70% of the population over the age of 65 years and has a significant negative impact on the quality of life of elderly individuals (1–3). Although osteoarthritis is now increasingly viewed as a metabolically active joint disorder of diverse etiologies, cartilage tissue degeneration is primarily implicated. It is generally believed that the initiating event of osteoarthritis is predominantly due to loss of proteoglycans from the tissue (4). Proteoglycans (PG) are complex macromolecules that consist of proteins and polysaccharides. Aggrecan is the most common of these PG and accounts for ∼80–90% of the total PG. It consists of a protein core with a long extended domain to which many glycosaminoglycan (GAG) side chains are attached. Chondroitin sulfate is the predominant GAG molecule found in cartilage. In order for appropriate therapeutic intervention in osteoarthritis, there is a critical need for diagnostic methods that quantify the early molecular changes in cartilage before the manifestation of morphological changes.
As conventional magnetic resonance imaging (MRI) is neither proven sensitive nor accurate for the detection of early biochemical changes associated with the loss of PG, there have been several sophisticated MRI methods proposed to quantify these changes in vivo. These include sodium MRI, T1rho (T1ρ) MRI, and delayed gadolinium-enhanced MRI contrast (5–7). Although sodium MRI is highly specific to PG, it requires special coils and hardware (multinuclear option). It is also associated with low signal-to-noise ratio and requires ultrahigh fields. On the other hand, delayed gadolinium-enhanced MRI contrast is a promising method that can be performed on standard clinical scanners. However, it has logistical issues such as long waiting periods following the injection of contrast agent. T1ρ MRI is a novel imaging method that has the ability to generate endogenous contrast that is sensitive to in vivo PG and collagen content (6). Recently, it has been shown that chemical exchange saturation transfer (CEST) of labile OH protons on GAG with bulk water leads to a significant reduction of bulk water magnetization creating “gagCEST” (8). Using this approach, significant gagCEST was reported from both ex vivo and in vivo cartilage without any systematic analysis of static magnetic field (B0) inhomogeneity (8).
It is well known that the B0 inhomogeneity would significantly affect the accuracy of the computed CEST values (9–11). B0 correction requires an estimate of local B0 variation. For B0 estimation, one can either use gradient echo MRI methods (12) or off-resonance saturation-based methods (10, 11, 13). Recently published water saturation shift referencing (WASSR) approach (10) is also an off-resonance saturation-based method with optimized saturation pulses to provide only direct water saturation. For B0 correction, one can use either analytical expression in steady-state saturation conditions (9) or interpolated (or fitted) Z-spectral data (10, 11, 13). Interpolated WASSR data were used for B0 estimation.
In this study, the feasibility of performing gagCEST on human cartilage in vivo was evaluated at 3 and 7 T field strengths. Human cartilage gagCEST maps were computed before and after the correction of B0 inhomogeneity. Saturation pulse parameters were optimized to obtain maximal gagCEST in human cartilage, and numerical simulations were performed to examine the effects of direct saturation (14) of water as well as GAG OH protons on the observed gagCEST.
MATERIALS AND METHODS
In CEST experiments, frequency selective saturation of solute spins that are in exchange with solvent spins (e.g., water) leads to the transfer of saturated magnetization to the solvent thus decreasing the signal intensity of the solvent spins. Subsequently, longitudinal relaxation returns each nuclear spin system to its equilibrium values and eventually the system reaches a steady state. The steady-state magnetization is given as:
where Msat8 is the steady-state amplitude of the water proton magnetization during the irradiation of exchangeable solute spins; M0 is the amplitude of the water proton magnetization in the absence of saturation, k1 is the pseudo-first-order exchange rate constant, and T1w is the longitudinal relaxation time of water protons (14–16). This magnetization is then imaged to detect the CEST effect from solute nuclear spins. In order for the CEST effect to be efficiently observed, the slow to intermediate exchange condition (Δω > k) must be fulfilled, where Δω is the chemical shift (or offset frequency) of the exchanging spins and k is the exchange rate. In general, the CEST effect of the solute spins is computed using following equation:
where M0 is the water equilibrium magnetization, Msat (±Δω) are the water magnetizations obtained with saturation at a “+” or “−” Δω offset of the water resonance. In interpreting the CEST effect, other factors that play role are the amplitude and duration of the saturation pulse. These effects can be incorporated into a general solution obtainable from a theoretical analysis of a two-site exchange model in the presence of solute saturation. An analytical expression for the CEST effect can be derived (17–20) as:
where k is the exchange rate (s−1 or Hz), α is an efficiency factor with α = 1 describing complete saturation (obtained with a sufficiently high amplitude saturation pulse), f is the fraction of exchangeable protons with respect to the total number of protons including water, R1w (=1/T1w) is the longitudinal relaxation rate of water protons, and tsat is the length of the saturation pulse.
Although Eq. 3 is useful in understanding the general CEST principles, its applicability is limited to solute selective steady-state saturation only at +Δω offset frequency with no other saturation contaminations. In practice, there is a contamination from the direct saturation (DS) of water to both Msat (±Δω) and from DS of solute protons while saturating at −Δω to Msat(−Δω). DS of water reduces the available bulk water protons for chemical exchange and reduces CEST contrast. DS of solute protons while saturating at −Δω additionally reduces water magnetization due to chemical exchange with the solute protons. In such cases to fully understand CEST and DS effects, one needs to resort to numerical simulations of Bloch–McConnell equations (17).
The effects of B0 and B1 variations on the observed CEST values are rather complex. As the CEST asymmetry is based on the subtraction of images Msat (±Δω), any asymmetry created with local B0 variation will contaminate the observed CEST asymmetry. Hence, a very good estimate and correction of local B0 inhomogeneity is imperative to get accurate CEST asymmetry. B1 variations affect both the CEST effect (Eq. 3 provides a hint for this through the empirical factor α) and the amount of DS contamination.
CEST MR Sequence
The pulse sequence used in this study consists of a frequency selective saturation pulse train (user selected saturation offset frequency [Δω], saturation duration, and B1rms) followed by a chemical shift selective fat saturation pulse and a segmented radiofrequency spoiled gradient echo readout acquisition with centric phase encoding order. At the end of the gradient echo acquisition segments, a variable delay has been added to provide T1 recovery and keep the sequence under system radiofrequency safety limits. This sequence is flexible enough to be used for both the WASSR data acquisition and the CEST imaging.
The saturation pulse train is composed of Hanning windowed rectangular pulses and delays between them. At 3 T, a 48-ms pulse with a 2-ms delay is used, whereas at 7 T, a 99.8-ms pulse with a 0.2-ms delay is used. The number of pulses in the train can be adjusted to provide variable saturation duration. The Hanning window shape and pulse duration were chosen based on MRI scanner hardware limits and minimal artifacts in phantom tests. The peak B1 of the Hanning windowed pulse is set to provide the required B1rms value. The saturation pulse excitation bandwidth (50%) is 10 Hz with 1% bandwidth of 40 Hz (∼0.28 ppm at 3 T) for saturation train duration of 0.5 s. For longer saturation durations, these bandwidths are narrower.
The study was conducted under an approved Institutional Review Board protocol of the University of Pennsylvania. Five subjects were taken from a normal population in the age range of 28–40 years. Informed consent from each volunteer was obtained after explaining the study protocol. CEST imaging and Z-spectrum acquisitions on the human knee were performed at 3 T using an 18-cm diameter, eight-channel transmit-receive phased-array knee coil on a Siemens clinical scanner (Magnetom Tim Trio, Siemens Medical Solutions, Malvern, PA) and at 7 T using the standard circularly polarized head coil on a Siemens 7 T research scanner (Siemens Medical Solutions, Malvern, PA).
The actual study protocol consisted of the following steps: a localizer, WASSR, Z-spectral or CEST acquisitions, and B1 data collection. For WASSR acquisitions, Δω range of −1 to +1 ppm with step size of 0.05 ppm was used. For Z-spectrum acquisitions, Δω range of −5 to +5 ppm with step size of 0.1 ppm was used. For CEST acquisitions, a limited Δω range required for B0 correction was used. This range was based on a quick inspection of dark regions in raw WASSR images (on scanner) at different WASSR saturations Δω. Typical Δω ranges used for CEST acquisitions were −1.7 to −0.3 ppm and 0.3 to 1.7 ppm with 0.1 ppm steps for a total of 30 images.
Knee imaging parameters were slice thickness = 5 mm, flip angle = 10°, readout pulse repetition time = 5.6 ms, echo time = 2.7 ms, field of view = 140 × 140 mm2, and matrix size = 192 × 192 with a segment size of 96. The whole sequence was repeated every 6 s at 3 T and every 8 s at 7 T for each Δω.
For WASSR acquisitions, a 0.2-s saturation pulse with B1rms of 0.13 μT was used in all cases. For complete Z-spectral acquisitions, a 0.5-s saturation pulse with a B1rms of 2.2 μT was used with the same volunteer (n = 2) at 3 and 7 T. For investigating the effects of saturation parameters, multiple CEST images were collected on two volunteers at both 3 T and 7 T using saturation pulses with B1rms of 18.5 Hz (0.4 μT) and 31 Hz (0.7 μT) over a duration range of 0.1–2.0 s, 62 Hz (1.4 μT) and 93 Hz (2.2 μT) over a duration range of 0.1–1.0 s, and 124 Hz (2.9 μT) over a duration range of 0.1–0.5 s. At higher B1rms values, the signal–to-noise ratios of CEST images using longer duration pulses were too poor to measure reliable gagCEST.
CEST imaging was performed with four volunteers on both 3 T and 7 T using the imaging protocol as described earlier with a saturation duration of 0.5 s and B1rms of 2.2 μT.
All image processing and data analysis were performed using in-house programs written in MATLAB (version 7.5, R2007b). The cartilage section was manually segmented from the anatomical image, and all data processing was performed only on this section. Acquired CEST data (at Δω = ±1.0 ppm) or Z-spectral data (typically −5.0 to +5.0 ppm) were directly used to generate gagCEST maps or Z-spectral asymmetry curves using Eq. 2 to get data without B0 correction. The mean and standard deviation of gagCEST or CEST asymmetry values were calculated over the small region of interest drawn on cartilage region in the image. The gagCEST map was overlaid on one of the original anatomical images.
B0 and B1 Corrections
WASSR data acquired over the Δω range of +1.0 to −1.0 ppm at steps of 0.05 ppm at each voxel are smoothed and interpolated using a cubic spline to generate data with a step size of 0.01 ppm. The Δω corresponding to the minimum of the interpolated data was used as the B0 value (δω) at each voxel (resolution = 0.01 ppm). Acquired CEST data (at offset frequencies, typically +0.3 to +1.7 ppm and −0.3 to −1.7 ppm) or Z-spectral data (typically −5.0 to +5.0 ppm) were smoothed and interpolated using a cubic spline to generate data with a step size of 0.01 ppm. For B0 inhomogeneity correction, each voxel data value at Δω ppm was replaced by the interpolated data value from (Δω − δω) ppm. Either Z-spectral asymmetry curves or CEST maps (based on the data from ±1.0 ppm) were generated using Eq. 2.
B1 field maps were obtained using a 2D single slice fast spin echo readout sequence with echo time = 12 ms pulse repetition time = 6 s, 128 × 128 image matrix. Two images were obtained using preparation square pulses with flip angles 30° and 60° (pulse duration = 0.3 ms). The 30° flip angle radiofrequency pulse amplitude was used as the reference B1 or B1ref. Flip angle (θ) maps were generated by solving the following equation:
where S(θ) and S(2θ) denote voxel signals in an image with a preparation flip angle of θ and 2θ, respectively. From the flip angle map, a B1 field map can be obtained using the relation, B1 = θ/(360τ) The coefficient B1/B1ref can be used if needed for B1 scaling of CEST values (21).
Bloch–McConnell equation solvers with two exchanging components (water and GAG) were written in MATLAB for analyzing the effects of CEST and DS at both 3 T and 7 T with the saturation pulse trains used in the experiments (14). Both the residual water magnetization affected only by DS without any CEST effects and the CEST asymmetry values contaminated with DS at different offset frequencies were calculated for the different saturation pulse durations and B1rms values used in the experiments.
Figure 1 shows Z-spectra (a and c) and CEST asymmetry (b and d) plots of a small region of interest from human knee cartilage before and after B0 inhomogeneity corrections from 3 and 7 T. Without any corrections for B0 inhomogeneity, a clear shift (∼0.6 ppm) in the Z-spectrum was observed in this region of interest. This shift in the data is removed after correction for the B0 inhomogeneity. gagCEST calculated from the asymmetry plots generated without B0 correction show large effects (>20%), whereas after B0 correction the calculated gagCEST was negligible at 3 T and ∼6% at 7 T. The error bars shown in Fig. 1 represent the standard deviation of the gagCEST values at each ppm over the region of interest. A large number of voxels in corrected gagCEST map at 3 T showed both positive and negative values at different Δω. Hence, the gagCEST asymmetry derived through integration over an offset range around 1 ppm is also negligible. The effect of B1 inhomogeneity in the cartilage region was minor (<10%) in this study at both 3 T and 7 T, and hence, no correction was necessary.
The top row of Fig. 2 shows a fat suppressed anatomical human knee image (a) and gagCEST maps (b and c) without and with B0 correction at 3 T. Without any correction for B0 inhomogeneity, a >20% gagCEST was observed in cartilage (Fig. 2b), whereas with B0 inhomogeneity corrections, negligible gagCEST was observed. The corresponding images and maps from the same volunteer at 7 T are shown in bottom row. After B0 correction ∼6% gagCEST was observed.
Figure 3 shows plots of knee cartilage gagCEST at varying saturation B1rms and durations obtained at both 3 T (a) and 7 T (b). Observed gagCEST was negligible at 3 T for all B1rms and durations, whereas at 7 T the maximum gagCEST was observed at saturation B1rms of 2.2 μT and duration of 0.5 s.
Figure 4 depicts corrected gagCEST images from the four healthy volunteers at 3 T (a) and 7 T (b). Again, the observed gagCEST at 3 T is negligible, whereas at 7 T it is ∼6%.
Figure 5 shows our simulation results at 3 and 7 T: (a) The effect of water DS for a 0.5 s duration saturation pulse at different B1rms values is shown. (b) the effect of DS of the GAG pool while saturating at −1 ppm for a 0.5 s duration saturation pulse at various B1rms values is shown. The reductions in water and GAG magnetizations reduce the gagCEST sensitivity at 3 T. (c) Simulated CEST asymmetry spectra for a saturation B1rms of 2.2 μT and duration of 0.5 s show that the theoretical gagCEST expected at 3 T is 0.5% at 1.0 ppm, whereas at 7 T it is 5.8%. This is in line with the experimental results reported earlier.
Given the geometry of knee and cartilage distribution, as shown in this study, there may be substantial B0 field variations within and around the cartilage. B0 inhomogeneity leads to a shift in the Z-spectra and affects the magnitude of gagCEST observed at 1.0 ppm (Fig. 1). This suggests that the significant gagCEST (>20%) reported earlier at 3 T (8) is mainly due to the presence of B0 field inhomogeneity in the human cartilage. After B0 corrections, calculated gagCEST values were negligible at 3 T and ∼6% at 7 T (Figs. 2 and 4).
The saturation pulse amplitude (B1) dependency of gagCEST has been evaluated in in vivo cartilage (Fig. 3). These experimental results show that gagCEST stayed negligible, when B1rms was varied between 0.4 and 2.9 μT at 3 T and peaks at saturation B1rms of ∼2.2 μT and duration of ∼0.5 s with a value of ∼6%. It is interesting to note that at lower B1rms values, we seem to be getting a small negative gagCEST. This is consistent with a nuclear Overhauser effect induced water signal loss from GAG CH protons while saturating at −1.0 ppm. At higher B1rms values, this effect is suppressed. This has also been previously shown with in vitro data (8).
As seen from our simulation results (Fig. 5), one of the main reasons for the reduced efficiency of gagCEST at 3 T seems to be DS effects leading to reduced water magnetization (when saturating at ±1 ppm) as well as chemical exchange with the reduced GAG magnetization (when saturating at −1 ppm). Another potential major cause for the gagCEST efficiency difference between 3 and 7 T is that the reported OH protons exchange rate kex of 1000 s−1 (22) is in fast exchange regime at 3 T (kex > Δω [∼800 rad s−1]), whereas it is in slow exchange regime at 7 T (kex ≲Δω [∼1800 rad s−1]).
It is worthwhile to report the recent gagCEST study (23) in human knee cartilage performed at 7 T has shown a peak value ∼3% at 1.2 ppm. As there is not enough information about the experimental parameters used in this study, it is difficult to compare our results with this study.
In summary, numerical simulations as well as experimental results demonstrate that the DS effects of water and GAG are substantial contributors for negligible gagCEST observed after B0 correction in cartilage at 3 T. As GAG loss from cartilage is expected to result in a further reduction in gagCEST, this method is not expected to lead to accurate quantification of GAG content in healthy or degenerated cartilage at 3 T. Given its magnitude (∼6%) gagCEST at high fields such as 7 T holds promise as a clinically viable technique.
This work was performed at an NIH-NCRR supported Biomedical Technology Research Center.