Synopsis

Keywords: Quantitative Imaging, Metabolism, Human Brain NAD Imaging

Nicotinamide adenine dinucleotide (NAD) is an important molecule critical for cellular metabolism and signaling. The cerebral NAD concentration is only at submillimolar level, and thus, difficult to measure and imaging in vivo. In this study, we utilize 3D ³¹P MRSI-based NAD assay in combination with a SPICE-based signal processing method to evaluate the feasibility for non-invasively mapping the NAD contents in healthy human brains at 7T. The results suggest that by taking advantage of ultrahigh field scanner and advanced post-processing, it is possible to obtain whole-brain NAD maps with NAD measurements consistent with the literature reported values.

INTRODUCTION

Nicotinamide adenine dinucleotide (NAD) can be found in all living cells in oxidized (NAD⁺) and/or reduced (NADH) form. It acts as a coenzyme in various cellular redox reactions, and also serves as a substrate for different NAD⁺-consuming enzymes; thus, NAD plays important roles in cellular metabolism and signaling ^1-3. The NAD concentration of human brains is at submillimolar level and difficult to measure in-situ. Recently developed in vivo ³¹P MRS-based NAD assay allows non-invasive measurement of intracellular NAD contents and redox ratio in human visual cortex ^4,5. However, intracellular NAD level in various brain regions has not been systematically investigated due to technical challenges. In this study, we evaluated the feasibility of mapping NAD in the entire human brain by leveraging ultrahigh field 3D ³¹P MRS imaging (MRSI) in combination with an advanced SPICE-based post-processing approach to greatly enhance the sensitivity of ³¹P MRSI-based NAD assay at 7T.

METHODS

In vivo ³¹P MRSI acquisition: Four healthy volunteers participated in this study using a study protocol approved by University of Minnesota IRB. In vivo ³¹P MRSI data were acquired using a 90-cm bore 7T Magnex magnet with a Varian INOVA console. A ¹H/³¹P double-tuned TEM volume coil with quadrature driving was used. Three dimensional (3D) ³¹P chemical shift imaging (CSI) was acquired using a Fourier series window imaging technique ⁶ with spectral bandwidth=5000Hz, flip angle=45°, FOV=20×20×22cm³, matrix size=15×15×13, TR=0.5s, cylindrical voxel with 2.4cc nominal voxel size and total acquisition time of 51 minutes.
SPICE-based data processing and analysis: The data processing pipeline includes: (a) correction of frequency variations; (b) denoising using a subspace-based method; and (c) model-based spectral quantification. First, to remove the frequency variations in the ³¹P-MRSI signals caused by field inhomogeneity, the resonance frequency of PCr in each voxel was estimated using HSVD and aligned to 0 ppm ⁷. Then the ³¹P-MRSI signals were denoised using subspace modeling as proposed in the SPICE method ^8-10. More specifically, based on the partial separability model ⁸, the ³¹P-MRSI signals (denoted as $$$\rho(x,t)$$$) can be represented using a low-rank subspace model as:
$$\rho(x,t)=\sum^{L}_{l=1}u_l(x)v_l(t)$$
where $$$\{v_l(t)\}, \{u_l(x)\}, L$$$ denote the basis functions, corresponding spatial coefficients and the model order, respectively. The basis functions could be estimated and learned from training data ¹⁰, but they were estimated from the data itself to preserve signals features. Denoising was performed by solving the following optimization problem:
$$\hat{U}=\textrm{arg }\textrm{min}_U{\parallel}\rho_r-UV{\parallel}+\lambda{\parallel}WU{\parallel}_2^2$$
where $$$U$$$ and $$$V$$$ are the matrix forms of $$$\{u_l(x)\}$$$ and $$$\{v_l(t)\}$$$, respectively. $$$\rho_r$$$ is the original MRSI signals, and $$$W$$$ is the total variation operator. With the estimated $$$\hat{U}$$$, the denoised MRSI signals were generated as $$$\rho_d=\hat{U}V$$$. Spectral quantification was performed using a time-domain fitting method using a basis set including resonance structures of all the measurable ³¹P metabolites ^11,12. After spectral fitting, the concentrations of metabolites were calculated using γ-ATP as an internal reference:
$$C_m=\frac{C_{{\gamma}ATP}}{N_m}\frac{E_{{\gamma}ATP}}{E_m}\frac{S_m}{S_{{\gamma}ATP}}$$
where $$$C,S,E,N$$$ are the concentration, fitted signals intensity, T₁ saturation factor, and number of ³¹P spins in each metabolite, respectively. $$$C_{{\gamma}ATP}$$$ was set as 2.8 mM ⁵.

RESULTS

Figure 1A shows the typical ³¹P spectra of a single voxel extracted from a 3D ³¹P-CSI before and after applying the SPICE-based processing method that achieves several-fold (≥3×) noise reduction and preserves the true signal of phosphorous metabolites for spectral fitting and quantification. Figure 1B displays representative metabolite maps of phosphocreatine (PCr), γ- and α-ATP, as well as total NAD before and after SPICE processing, indicating that the quality of the maps is greatly improved, especially for low-concentration metabolites such as NAD. Figure 2 shows the concentration maps of NAD covering the entire human brain of a representative subject, suggesting that intracellular NAD concentrations appear to have a relatively homogeneous distribution, with a slightly higher level in midbrain or thalamic regions. Similar observations in other three subjects are shown in Fig. 3A. Figure 3B plots the actual NAD concentration levels across subjects (≈ 0.41mM), which were consistent among all four subjects studied to date.

DISCUSSION and CONCLUSION

Using a double-tuned ¹H/³¹P head volume coil at 7T, we were able to obtain 3D ³¹P MRSI covering entire human brain with reasonably high sensitivity and spatial resolution. However, for small metabolite signals such as NAD, it is still very challenging to image and quantify NAD due to limited signal-to-noise ratio (SNR) in individual CSI voxels. In this study, we applied the subspace-based approach to process the original 7T ³¹P MRSI data and achieved over 3-fold noise reduction, which made it possible for the first time to quantify and map low-level NAD contents of the entire human brain. We found that the brain NAD concentration and its spatial distribution were consistent among different subjects, and the averaged whole-brain NAD levels were in good agreement with the reported values obtained from human occipital region with sufficiently high ³¹P SNR and without applying any denoising method ⁵.
In conclusion, by combining the ultrahigh-field 3D ³¹P MRSI with an advanced post-processing method, we demonstrate the feasibility of whole-brain NAD mapping in healthy human at 7T. This approach would allow non-invasive monitoring of cerebral NAD and its changes in different brain regions under various brain conditions.

Acknowledgements

NIH Grants: U01EB026978, R01 CA240953, NS118330, P41 EB027061.