Introduction

Alternations in glutamate level have been observed in various neurological disorders^1,2. Cai et al³ demonstrated an advanced and non-invasive MRI technique based on chemical exchange saturation transfer (CEST) for in vivo molecular mapping of glutamate with higher resolution compared to ¹H MRS and PET. The computation of glutamate-weighted CEST (GluCEST) contrast using CEST asymmetry^3,4 analysis is challenging due to confounding effects such as magnetization transfer (MT)⁵ from semi-solid macromolecules. The contribution of MT underestimates GluCEST contrast and there is no systematic study reported for removal of confounding effects to accurately measure GluCEST. Several studies^6,7 reported fitting of MT spectrum at far off-resonance frequencies using different lineshapes. The primary objective of the study was estimation of GluCEST contrast after removal of MT contribution from Z-spectra. This study also focused on investigation of the appropriate lineshape to model MT pool for the given set of saturation pulse parameters used for GluCEST imaging.

Methods

Human study: Six asymptomatic human volunteers (M:F=5:1;35±14.5years) were scanned at 7T MRI scanner (Siemens) at the University of Pennsylvania. The study protocol consisted of the following steps: three-plane localizer, WASSR⁸, CEST and B₁ map data collection using pulse sequence reported previously³. 1H MRS was also carried out for three out of six healthy human volunteers on the same slice location as used in CEST data and spectra was acquired for voxels positioned on gray-matter (GM) and white-matter (WM) tissues. The acquired spectra were analyzed using LC model⁹ to compute the concentration of glutamate in GM and WM. For GluCEST imaging, Z-spectral data used in the current study were acquired from human brain at B_1rms=2.9µT and duration=800ms. Partial Z-spectra at offset-frequencies (±100ppm to ±14ppm) were fitted to model semi-solid MT component with Lorentzian, Gaussian, super-Lorentzian and 6^th degree polynomial function lineshapes. Average residual errors per pixel (σ) were calculated. The fitted MT component was interpolated over the whole offset frequency range and was removed from Z-spectra. GluCEST contrast was computed from modified Z-spectra (without MT component) at B_1rms=2.9µT using equations (i),(ii)

$$$ GluCEST_{M0} = \frac{M_{sat(-{\triangle}w)} - M_{sat(+{\triangle}w)}} { M_{0} }$$$ ..(i)
$$$GluCEST_{Neg} = \frac{M_{sat(-{\triangle}w)} - M_{sat(+{\triangle}w)}} { M_{sat(-{\triangle}w)} }$$$ .. (ii)
M_sat(+∆ω) and Msat(-∆ω) are the signal intensities (SI) with (∆ω=3ppm) downfield and upfield with respect to water resonance. M₀ is the SI without RF saturation. The difference between GluCEST maps before and after MT removal was compared using T-test (p<0.05 is significant). The Lorentzian, Gaussian and super-Lorentzian lineshapes⁶ are described in equation (iii),(iv),(v)

$$$g_{i}(2\pi(w-w_{0})) =A \times \frac{1} { 1+(2\pi(w-w_{0})\times T_{2m})^{2} }$$$ ..(iii)

$$$g_{i}(2\pi(w-w_{0})) =A \times e^{\frac{-(2\pi(w-w_{0})\times T_{2m})^{2}}{2}} $$$ ..(iv)

$$$g_{i}(2\pi(w-w_{0})) =A\times \int_{0}^{\frac{\pi}{2}}\frac{\sin\theta}{|3(\cos\theta)^{2}-1|} \times e^{\frac{-2\times (2\pi(w-w_{0})\times T_{2m})^{2}}{|3(\cos\theta)^{2}-1|^{2}}} d\theta $$$ ..(v)

w=offset-frequency from water-resonance, w₀=offset-frequency of MT pool, $$$\theta$$$=dipolar Hamiltonian-angle, A=scaling-factor. T_2m characterizes the linewidth of MT pool⁶. T_2m, w₀, A are parameters of fit.
Animal study: Preliminary study was performed at University of Pennsylvania. Rat gliosarcoma cells (9L) were injected on a Syngeneic female Fisher rat (F344/NCR, four-six weeks old). After 5 weeks of injection, the rat was scanned on 9.4T animal scanner (Varian) at B_1rms=5.9µT, duration=1s. CEST images were collected at multiple frequencies (ppm) ±100, ±50, ±25, ±12 to ±6 with step-size of 2, ±5 to ±0 with step-size of 0.25ppm.

Results

Better accuracy of fitting far-off resonance Z-spectra was achieved with super-Lorentzian (σ=0.0009) and Lorentzian (σ=0.0017) compared to other lineshapes (Fig.1). After MT removal, there was significant (p<0.01) increase in GluCEST_M0 (Fig.2) and decrease in GluCEST_Neg contrast (Fig.3). Spectroscopic results (Fig.4) showed that the glutamate concentration ratio in GM/WM was 1.6±0.13. After MT removal, GluCEST_Neg and GluCEST_M0 contrast-ratio in GM/WM was 1.47±0.35 and 1.52±0.26 respectively using Lorentzian lineshape, which is closest to spectroscopic result compared to other linehshapes. Before MT removal, the GluCEST_M0 contrast in the tumor region was 8.59% compared to contra-lateral region (4.71%) (Fig.5b). After MT removal using Lorentzian lineshape, GluCEST_M0 contrast was 10.72% the tumor region compared to contra-lateral region (7.19%) (Fig.5c).

Discussion

MT effect produces inherent asymmetry¹⁰ in the Z-spectra causing a systematic contamination in the quantification of GluCEST. At the saturation parameters of GluCEST imaging, contribution of CEST effects from amide, creatine, glucose, myoinositol, etc is significantly less^11,12. Thus the major confounding effect to GluCEST contrast is asymmetry in MT and is quite important to remove from Z-spectra while computing GluCEST. Though super-Lorentzian fits better to model MT component, the quantity of the MT removed from Z-spectra is appropriate using Lorentzian due to preservation of GluCEST contrast-ratio. The amount of MT removed from Z-spectra is overestimated using super-Lorentzian and underestimated using Gaussian and polynomial. The pre-clinical application showed importance of removal of MT component. MT uncorrected GluCEST_M0 contrast is 82% higher in tumor than contra-lateral normal region. Due to MT, GluCEST_M0 map shows false elevation of contrast in tumor region. After MT removal, accurate GluCEST_M0 contrast was estimated and was 50% higher in tumor region compared with normal region.

Conclusion

The improved estimation of GluCEST contrast is feasible after removal of MT component. This increases the specificity of GluCEST to glutamate concentration which might have potential in diagnosis and treatment monitoring of diseases

Acknowledgements

This study was supported by Indian Institute of Technology Delhi, Fortis hospital and University of Pennsylvania. This study was partially supported by funding support from MATRICS, SERB-DST Grant number: MTR_2017_001021. The authors acknowledge Dr. Puneet Bagga for technical discussions and Deepa Thakuri for data acquisition.