Introduction

Nicotinamide adenine dinucleotide (NAD) is a crucial cofactor in brain metabolism^1,2,3. Its detection and quantification is challenging due to not only its low concentration but also a reported polarization-exchange with water leading to strong signal suppression with standard MRS sequences. To avoid saturation-transfer from water, studies on humans at 7T^3,4 have used frequency-selective excitation with slice-selective refocusing that leaves the longitudinal magnetization of water intact.
Here, a comparison of three techniques to assess the potential of NAD⁺ quantification on clinical scanners is presented. They include standard semiLASER^5,6 with VAPOR water-suppression (WS), semiLASER with metabolite-cycling (MC)^7,8 as a technique avoiding water presaturation, and ISIS-localization with chemical-shift-selective excitation (I-CSE)⁹, a type of non-water-excitation (nWE) technique.

Methods

10 healthy volunteers for each technique (12 in total), supraventricular VOI¹⁰: 50x75x20cm³, 3T MR Scanner (Prisma, Siemens), 64-channel receive head coil. Three localization techniques: semiLASER⁵, MC-semiLASER^7,8, 2D I-CSE⁹. Transmit frequency at 7.3 ppm for semiLASER and MC-semiLASER; at 9.7 ppm for I-CSE. VOI-optimized B₁⁺. Data processing with jmrui¹¹ and MATLAB.
WS-sLASER: VAPOR with 135Hz bandwidth and optimized flip angles; partial outer-volume suppression (OVS); 1280 scans; TE=35ms; TR=2500ms.
MC-sLASER: MC pulse of 22ms; no OVS; 768 scans; TE=35ms; TR=2500ms. 2D I-CSE: full OVS; TE=10ms; TR=3000ms; 640 scans; adapted raw data post-processing.
2D I-CSE: full OVS; TE=10ms; TR=3000ms; 640 scans; adapted raw data post-processing.
Fitting in FiTAID¹² with identical Voigt models including prior-knowledge restraints and a simulated partial spectrum of NAD⁺ for all techniques but differing in downfield background modeling by a set of tightly-spaced Voigt lines. NAD⁺ Voigt line with fixed Lorentzian part (80ms T₂). For semiLASER, the upfield spectrum was fitted first. NAD⁺ frequency was fixed relative to the main downfield peaks. For I-CSE, residual water and upfield peaks were removed by HLSVD and only the downfield part fitted (model extended by 2 Voigt lines representing background signal from the amide-signal tail).
Quantification via unsuppressed water from MC- semiLASER and estimated tissue water content based on tissue fractions. Differences in VOI profile for the techniques accounted for based on phantom measurements.

Results and Discussion

Typical spectra from a single representative subject for all techniques are presented in Figure-1. The semiLASER techniques provide both, the upfield and downfield spectrum at very high SNR, while the I-CSE sequence is limited to the downfield part. Signals from exchangeable protons are more prominent with MC than WS (~8.2 and 8.5 ppm), but I-CSE with much shorter TE and very limited periods of magnetization exchange clearly impresses with very large amide signal contributions at 8-8.5 ppm.
In the up-scaled cohort spectra in Figure-2 a peak at 9.33 ppm attributable to the H2 proton of NAD⁺ is clearly visible for all methods. Furthermore, the techniques without WS show signal contributions from H6 and H4 of NAD⁺ (doublets at 9.14 and 8.83 ppm). With MC-semiLASER, those peaks are best represented with less overlap with the large amide signal than in I-CSE, which becomes even clearer when inspecting the results from model fitting in Figure-3. The model fits fairly well for all three techniques but results are visually most convincing for MC-semiLASER. For I-CSE the residues do not appear to contain noise only—probably due to amide baseline interference in the fit. In addition, linewidths and intensities for the NAD⁺ peaks were enforced by prior knowledge, not allowing for potential differences in T₂ and MR-visibility.
Table-1 contains the results of the quantification based on non-suppressed water. Raw tissue contents (i.e., without relaxation correction) show that I-CSE provides most NAD signal, followed by MC-semiLASER. After T₂-correction, estimated tissue contents are similar for MC-semiLASER and I-CSE, while a lower value for WS-semiLASER confirms the limited visibility associated with saturation transfer from water. For MC-semiLASER the cohort SD was very similar to the CRLB while for I-CSE this value is about double the CRLB. For WS-semiLASER, the CRLB are smallest in absolute terms, but largest as a percentage of the estimated value. The estimated tissue content of ~190 μ­molar suggests visibility of roughly 66% when using the values from ³¹P-MRS as reference⁴. This is higher than reported earlier for human brain⁴, but similar to the value for rat brain¹³.

Conclusions

Three MRS techniques with different approaches to deal with the water resonance have been evaluated.
Standard localization with water-presaturation with a long train of RF pulses does show the NAD⁺ resonances in an optimized setting with large SNR, though truly convincing only in a group average spectrum⁶.
Localization using semiLASER with moderately short TE but metabolite-cycling instead of WS and thus limited periods for saturation-exchange to occur is well-suited for NAD⁺ quantification, also for single subjects, given a relative CRLB of 14% and a cohort SD of 11%. A large VOI and 32 min scan time are needed for this precision.
Optimal acquisition conditions are achieved with I-CSE that features shorter TE and minimal time for saturation-exchange. However, at 3T the large amide signals yield an ill-defined background signal that interferes with robust and precise quantification of NAD⁺.
Comparing with tissue levels measured invasively or with ³¹P-MRS, the presented ¹H-MRS methods show a MR visibility of ~2/3 for both methods without presaturation, somewhat higher than reported previously for higher fields^3,4.

Acknowledgements

This work is supported by the Swiss National Science Foundation (SNSF #320030‐175984).