Single-Voxel ¹H MR spectroscopy of cerebral nicotinamide adenine dinucleotide (NAD⁺) in humans at 7T using a 32-channel volume coil

Abstract

Purpose:

Reliable monitoring of tissue nicotinamide adenine dinucleotide (NAD⁺) concentration may provide insights on its roles in normal and pathological aging. In the present study, we report a ¹H MRS pulse sequence for the in vivo, localized ¹H MRS detection of NAD⁺ from the human brain.

Methods:

Studies were performed on a 7T Siemens MRI scanner using a 32-channel product volume coil. The pulse sequence consisted of a spectrally selective low bandwidth E-BURP-1 90° pulse. PRESS localization was achieved using optimized Shinnar-Le Roux 180° pulses and overlapping gradients were used to minimize the echo time. The reproducibility of NAD⁺ quantification was measured in 11 healthy volunteers. The association of cerebral NAD⁺ with age was assessed in 16 healthy subjects 26–78 years old.

Results:

Spectra acquired from a voxel placed in subjects’ occipital lobe consisted of downfield peaks from the H2, H4 and H6 protons of the nicotinamide moiety of NAD⁺ between 8.9–9.35 ppm. The mean ± SD within-session and between-session coefficients of variation were found to be 6.14 ± 2.03 and 6.09 ± 3.20 % respectively. In healthy volunteers, an age-dependent decline of the NAD⁺ levels in the brain was also observed (β = −1.24 μM/yr, SE=0.21, p<0.001).

Conclusion:

We demonstrated the feasibility and robustness of a newly developed ¹H MRS technique to measure localized cerebral NAD⁺ at 7T MRI using a commercially available RF head coil. This technique may be further applied to detect and quantify NAD⁺ from different regions of the brain as well as from other tissues.

Introduction

Nicotinamide adenine dinucleotide (NAD⁺) is present in all living cells and it plays a vital role in cellular metabolism as a coenzyme for reduction-oxidation (redox) reactions, including those required for mitochondrial energy production (1). It also serves as a cofactor for other enzymes such as sirtuins, Poly (ADP-ribose) polymerases (PARPs) and ADP-ribosyl cyclases that are involved in cell survival and metabolism (2–4). The nicotinamide moiety of NAD⁺ serves as the site for redox reactions, allowing the reduced form of NAD⁺ (NADH) to function as an energy-transfer intermediate between various metabolic pathways. The ratio and concentration of NAD⁺ and NADH influence the rate of reactive oxygen species production in the mitochondria (5,6). An increasing number of studies suggest roles for NAD⁺ in mediating key cellular processes including gene expression, aging, neurodegeneration and cell death (7–9). The roles of NAD⁺ dependent enzymes such as sirtuins, PARPs and CD38 are also well documented in many reports (10,11). Reduced NAD⁺ levels can be restored by treatment with nicotinamide mononucleotide (NMN), an intermediate in the major NAD⁺ biosynthesis pathway (12,13), or by its precursor, nicotinamide riboside (14). Thus, cellular NAD⁺ may serve as a potential therapeutic target for metabolic and age-related disorders.

While NAD⁺ is most commonly quantified using invasive biochemical or auto-fluorescence methods (15,16), non-invasive quantification of tissue NAD⁺ is highly desirable for in vivo studies, particularly in organs not amenable to biopsy such as brain. Non-invasive in vivo detection and quantification of NAD species (NAD⁺/NADH) has been performed using ³¹P magnetic resonance spectroscopy (MRS) with ultra-high magnetic field MR scanners. NAD⁺ and NADH concentrations have been quantified by modeling the partially overlapping NAD⁺, NADH, and α-ATP resonances in the ³¹P MR spectrum (17,18). One study using ³¹P MRS showed a decline of NAD⁺ with aging in the human brain (19).

Recently, a non-water suppressed ¹H MRS technique with surface coil localization was used to detect protons of the nicotinamide moiety in NAD⁺ downfield of water in both the rodent and human brain (20,21). Due to cross-relaxation of the NAD⁺ protons with water, traditional water suppressed ¹H MRS methods cannot be used to detect the NAD⁺ protons (20). This ¹H MRS method has an advantage of being free from contamination from other metabolite peaks and offers ~15-fold higher sensitivity compared to ³¹P MRS.

Here, we demonstrate the development and validation of a spectroscopy (SVS) ¹H MRS pulse sequence to detect spatially-localized NAD⁺ signal from the human brain at 7T with a volume radiofrequency (RF) coil. NAD⁺ signals were quantified using water signal as an internal concentration reference. In a small cohort of healthy volunteers, we found an age dependent decline in the NAD⁺ concentration in the occipital cortex.

Methods

Sequence design

The pulse sequence uses a spectrally selective self-refocused Band-selective Uniform-Response Pure-phase (E-BURP-1) 90° pulse (22,23) and 3 narrow band spatially selective refocusing Shinnar-Le Roux (SLR) 180° pulses (24–27) for localization (Figure 1). The details of the excitation and refocusing pulses are given below.

Figure 1.

Pulse sequence for NAD⁺ spectroscopy. A spectrally selective self-refocused E-BURP-1 90° pulse (2 ppm bandwidth) followed by three narrow band spatially selective refocusing 180° SLR pulses for localization providing selective excitation and refocusing of NAD⁺ signal from the MRS voxel

The 90° pulse is 7 ms long to achieve 2 ppm (600 Hz) spectral excitation at 7T. The pulse shape and Bloch simulation M_Y profile for the E-BURP-1 pulse are shown in Supporting Information Figure S1. The excitation pulse used in the human scans was centered at 9.1 ppm with a range of 8.1 – 10.1 ppm so as to excite all three of the nicotinamide moiety protons of the NAD⁺. The spatially selective refocusing pulses for localization are based on the SLR pulse design and are 3.2 ms long with a bandwidth of 800 Hz (Fig. S1). The crusher gradient waveform timings were optimized to reduce the overall echo time (TE) to 19.3 ms, which is close to that reported in the previous study using a surface coil (21). The spectrally selective 90° pulse and three 180° pulses are centered at 9.1 ppm for NAD⁺. For NAD⁺ quantification and phase estimation, water spectra were acquired using the same SVS localization by moving the center frequency of E-BURP-1 and SLR pulses to 4.7 ppm.

Human studies

All human studies were conducted under an approved Institutional Review Board (IRB) protocol of the University of Pennsylvania on 7T Siemens scanner using a Siemens volume coil transmit/32-channel receive proton phased-array head coil. A total of 16 healthy subjects (12 males and 4 females) with age ranging from 26–78 years (43.8 ± 16.5 yrs) were scanned twice per session in one or two sessions.

MR scanning procedure

Data from 11 subjects with two scanning sessions were used to determine the reproducibility of the acquisition and data processing. After positioning each subject in the scanner, two successive NAD⁺ MRS scans were performed (first session). Following this, the subject was taken out of the scanner, given a 15–30 minutes break, then re-positioned in the scanner and two more NAD⁺ MRS scans were performed in succession (second session). For each session, NAD⁺ levels were measured in a 64 ml cubic voxel in occipital cortex, and the [NAD⁺] obtained from the two scans were used to estimate within-session coefficient of variation (CV) for each subject. Within-session CVs were averaged within subjects and then across subjects to obtain mean ± SD within-session CV. For each subject, between-session CV was determined separately for the first and second scans, and then these values were averaged first within and then across subjects to obtain mean ± SD between-session CV.

¹H MRS acquisition

T1-weighted anatomical three-dimensional magnetization-prepared rapid gradient echo (MPRAGE) images of whole brain were acquired for voxel localization using the following parameters: 176 axial slices, repetition time/echo time/inversion time (TR/TE/TI) = 2800/4.4/1500 ms, Flip angle = 7°, Resolution 0.8×0.8×0.8 mm³, iPAT=2, scan time 7:43 min. A voxel of 40×40×40 mm³ was placed in the occipital lobe of the head as shown in Figure 2.

Figure 2.

Acquisition and processing of ¹H MRS data. The top row describes the three acquisitions. The left column describes the steps for processing the water reference scan to calculate the final water integral. Channel selection, eddy current phase estimation, signal averaging, phase correction, and curve fitting was performed on the reference water scan. The middle column describes the use of a water eddy current scan for eddy current phase estimation. The right column describes NAD⁺ spectra processing by channel wise eddy current correction, channel combination, Fourier transform, and frequency alignment of 16 blocks containing 32 averages each. Spectra from all the blocks were then added to obtain the final NAD⁺ spectrum. The NAD⁺ spectrum was phase corrected as fitted to determine the integral of NAD⁺ H2, H4 and H6 peaks

The ¹H MRS protocol consisted of 3 scans, (i) Water Reference acquisition (TR = 8 s, excitation and refocusing pulses set at 4.7 ppm, 8 averages), (ii) Water Eddy Current acquisition (TR = 1 s, excitation and refocusing pulses set at 4.7 ppm, 8 averages), and (iii) NAD⁺ acquisition (TR = 1 s, excitation and refocusing pulses set at 9.1 ppm, 512 averages). The total acquisition time for NAD⁺ and water ¹H MRS was 10 minutes.

Data Processing

A flow chart illustrating the data acquisition and processing steps is depicted in Figure 2. Raw multi-channel time domain data from the scanner were used for post-processing using an in-house MATLAB script. The water reference scan was used to select channels with at least 5% of maximum signal. For the selected channels, channel-wise eddy current phase estimation as a function of time (28,29) was performed using water eddy current estimation scan. The selected channels were combined to obtain the final complex water reference spectrum. The real part of reference water spectrum was phase corrected and curve fitted to derive the water integral. Next, eddy current correction was performed for the NAD⁺ scan using the eddy current estimation derived for water. To generate the NAD⁺ spectrum, block averaging of 32 acquisitions was used, and each block spectrum was eddy current corrected and channel combined, Fourier transformed, spectral frequency aligned using the broad NAA amide/unassigned purine nucleotide signal between 7.8 and 8.6 ppm as frequency reference, and phase corrected. The final spectrum was produced as the sum of all the individual blocks. For NAD⁺ level quantification, the spectrum in the range of 8.8 to 9.6 ppm was fitted with 3 Lorentzian peaks and the integral of the fit at 9.35 ppm was used. NAD⁺ concentration was calculated using the water integral and estimated tissue water concentration. We applied a T1, T2 correction factor using equation 1:

$$\left[NA{D}^{+}\right]=\frac{I_{NAD}\times \left[Water\right]\times \left[\left(ex{p}^{\left(-\frac{TE}{T{2}_W}\right)}\right)\left(1-ex{p}^{\left(-\frac{TR}{T{1}_W}\right)}\right)\right]\times C{F}_{NADvis}}{I_w\times \left[\left(ex{p}^{\left(-\frac{TE}{T{2}_{NAD}}\right)}\right)\left(1-ex{p}^{\left(-\frac{TR}{T{1}_{NAD}}\right)}\right)\right]}$$ Equation 1

where I_NAD and I_water are the integrals of the NAD-H2 and water peaks, respectively, T1_W and T2_W are the average water T1 and T2 of human brain at 7T (30,31). T1_NAD and T2_NAD values were taken from de Graaf et al (20). Further, since de Graaf et al (21) reported a 50% visibility of NAD⁺ using ¹H MRS, an NMR visibility correction factor CF_NADvis was included in the calculation of the final NAD⁺ concentration.

Calculation of contribution of CSF, gray and white matter in the ¹H MRS voxel

The tissue composition of the SVS region-of-interest was determined using the following procedure: First, a binary mask of the location of the SVS voxel was created in the T1 structural image space using the header information contained in the dicom image files (both SVS and T1) produced by the scanner. The software used to perform this step is available in the ImScribe package described previously (https://www.med.upenn.edu/cmroi/imscribe.html) (32,33). Next, the T1 structural image was segmented into maps of predominantly gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) voxels using the FAST tool from the FMRIB software library, FSL image analysis package (34). Finally, the SVS spatial mask was multiplied with the tissue segmentation maps to determine the relative contribution from GM, WM and CSF voxels within the SVS region-of-interest. The total water content in the voxel was calculated as a weighted contribution of GM (80% or 44 M water), WM (70% or 38.5 M water) and CSF (100% or 55 M water) (35,36).

Reproducible placement of MRS voxel in brain

In order to enhance overlap of SVS voxel positioning across subjects, we maintained the consistency in voxel placement for multiple acquisitions of ¹H MRS from the same voxel using ImScribe. ImScribe uses high resolution T1-weighted images and the voxel information from the spectroscopy file acquired from the first scan as target template for subsequent scans and performs affine co-registration to derive the information needed for a new voxel placement as illustrated in Fig. 2.

Association of NAD⁺ levels with age

To determine the variability in [NAD⁺] explained by age, data from all 16 subjects were used in a linear mixed model, including a random subject intercept to account for repeated within-subject measures and age as a fixed-effect covariate. The statistical significance of the age fixed effect was assessed using a Kenward-Roger’s F-test. Additionally, we calculated the Pearson correlation of the within-subject mean [NAD⁺] with age. Analyses were conducted using R version 3.5.3 (37) with the packages lme4 (38) and pbkrtest (39).

¹H MRS acquisition in parietal lobe

To demonstrate the feasibility of this method for detection of NAD⁺ in brain regions other than the occipital lobe, we performed ¹H MRS using a smaller voxel (30×30×30 mm³) positioned in the frontoparietal cortex of two subjects. All other parameters were the same as previously described.

Results

¹H MR spectrum of NAD⁺ from occipital lobe

The downfield region of a ¹H MR spectrum acquired from a voxel placed in the occipital lobe of the human brain in vivo at 7T is shown in Figure 3. The downfield region displays three peaks of NAD⁺ at 8.9, 9.15 and 9.35 ppm as well as a broad signal between 7.8–8.6 ppm containing NAA amide-bound proton signal at 7.84 ppm and multiple unassigned signals from ATP and other purine nucleotides between 8.0 and 8.6 ppm. The well-resolved H2 resonance at 9.35 ppm was used for the quantification.

Figure 3.

NAD⁺ MRS in human brain. Left, Chemical structure of NAD⁺ showing the H2, H4 and H6 protons present on the nicotinamide moiety which are the target of downfield spectroscopy. Inset, Sagittal section from the anatomical images of the brain showing the position of MR voxel. Right, MR spectra acquired from the healthy human brain showing the three peaks arising from NAD⁺ at 8.9, 9.1 and 9.35 ppm

Motion related frequency drift during the scans may affect the signal-to-noise ratio of the spectra generated after adding blocks with high number of averages. This was addressed by taking spectra from each average and adding them individually. However, the spectrum from one average may not have enough signal to determine the purine peak between 8–8.5 ppm, which is used as a reference peak to align all the spectra from individual blocks in this study. To find the optimal number of spectra to be added per block, we performed a summation of blocks with 1,8, 32 and 256 averages per block on the spectra acquired from one young volunteer (Supporting Information, Figure S2). We found that using 32 averages per block produced spectra with well resolved peaks of NAD⁺ H2, H4 and H6 protons with acceptable reproducibility.

Reproducibility of NAD^{+ 1}H MR spectroscopy

Figure 4 depicts the spectra acquired from two scans in three healthy subjects, which show high reproducibility of the NAD⁺ peaks downfield of the unassigned NAA/amine peaks between 8–8.6 ppm. The mean ± SD within-session coefficient of variation for 11 subjects was found to be 6.14 ± 2.03 %. Additionally, the mean ± SD between-session coefficient of variation was 6.09 ± 3.20 %, showing satisfactory reproducibility in the MRS scans for the quantification of NAD⁺.

Figure 4.

Demonstration of ¹H MRS reproducibility. NAD⁺ spectra acquired from three subjects in two scans acquired in the same session showing the intra-scan reproducibility of the pulse sequence

Decline of cerebral NAD⁺ with normal aging

The linear mixed model for [NAD⁺] resulted in an estimated coefficient (SE) for age of −1.24 (0.21) (Kenward-Roger’s F_1,15.62 = 33.37, p < 0.001), corresponding to a mean decrease in NAD⁺ concentration of 1.24 μM per year. The standard error was calculated as within-subject standard deviation divided by the square root of the number of measurements for that subject. The negative association between measured [NAD⁺] and age is plotted in Figure 5. Additionally, the estimated variance of the subject-level random effect was 391.3 in the model without age, and 31.5 in the model with age, indicating that age explains approximately 91.9% of the between-subject variability in these data. Within-subject mean [NAD⁺] had a Pearson correlation with age of r = −0.86.

Figure 5.

Decline of cerebral NAD⁺ with aging. Individual NAD⁺ levels measured from the occipital cortex of healthy volunteers plotted against the age showing a linear decrease of NAD⁺ level with healthy aging. Filled circles represent within-subject means and error bars depict ± one standard error. Standard error was calculated as the within-subject standard deviation divided by the square root of the number of measurements for that subject

Detection of NAD⁺ in frontoparietal cortex

Figure 6 shows a spectrum acquired from frontoparietal cortex of a subject. The three distinct peaks of the nicotinamide moiety H2, H4 and H6 protons of NAD⁺ are well-resolved downfield of the broad NAA+ATP adenosine resonances.

Figure 6.

¹H MRS from deeper brain region. ¹H MR spectrum from a voxel positioned in the frontoparietal cortex of a 33-year-old volunteer showing well-resolved NAD⁺ H4, H6 and H2 proton resonances at 8.9, 9.15 and 9.35 ppm, respectively

Discussion

In this study, we demonstrate the non-invasive detection of NAD⁺ in the brains of healthy individuals using a spectrally- and spatially-selective ¹H MRS pulse sequence with a product 32-channel volume coil at 7T. The pulse sequence allows the use of a standard volume coil to acquire spectra from a voxel placed in arbitrary regions of the brain with a reasonable acquisition time. We also tested the reproducibility of NAD⁺ measured using this method in a sub-group of healthy volunteers by evaluating the within-session and between-session coefficients of variation of the method. Finally, we demonstrate that [NAD⁺] in the occipital lobe of the brain declines with age. Importantly, this method can be applied obtain the spectrum from a voxel placed anywhere in the brain using a volume coil.

Although ³¹P MRS provides more comprehensive information regarding phosphorylated biomolecules (40,41), there is a substantial overlap between the NAD⁺/NADH and ATP peaks necessitating fitting and simulation to quantify NAD⁺/NADH (17). This limitation was recently addressed by using unlocalized ¹H MRS (20,21) wherein the H₂, H₄ and H₆ protons of the nicotinamide moiety in NAD⁺ were reported to be easily detectable due to the absence of any spectral overlap at the downfield region of water. However, the use of surface coil in these studies limits its application to superficial regions of the brain and complicates the incorporation of NAD⁺ measurement in multimodal imaging protocols.

The use of a self-refocused E-BURP-1 excitation pulse with a low bandwidth (2 ppm in present study) centered at 9.1 ppm provides a 10³-10⁴-fold reduction in the water signal (Supporting Information Figure S3) comparable to the standard water suppression methods (42–45) in addition to producing a pure absorption spectrum in the passband region of the 90° pulse. This is required since traditional water suppression methods eliminate the NAD⁺ signal due to cross-relaxation between the nicotinamide and water protons (20,21). The sequence additionally used three SLR refocusing pulses centered at 9.1 ppm for voxel localization. However, the phase response of the E-BURP-1 pulse inside the passband (8.1–10.1 ppm) is constant, while outside (>10.1 and <8.1 ppm) it is non-linear (Supporting Information Figure S3). This explains the observed phase distortions of the broad combined peak between 7.8–8.6 ppm and other upfield peaks, making peak fitting for the broad peak with a combination of unspecified singlets hard to achieve. Accordingly, in this study, the fitting range was limited to the H2, H4, and H6 peaks of NAD⁺. And as the H2 peak at 9.35 ppm is least influenced by the broad resonance, it was selected for NAD⁺ quantification.

In vivo, NAD⁺ was shown by de Graaf et al (21) to have 86% and 49% NMR visibility with ³¹P and ¹H MRS, respectively. We used these data to correct for the ¹H MRS visibility of NAD⁺ in the present work. While de Graaf et al (21) used NAA and creatine as a concentration reference to measure [NAD⁺], we used the water reference signal obtained with the same SVS localization. Due to the use of low bandwidth E-BURP-1 pulse (2 ppm) for the spectrally selective excitation, we did not observe the complete NAA peak resonance at 7.9 ppm, which may also be affected by a large chemical shift displacement contribution. Further, since creatine resonates at close proximity to water at 3 ppm, the partial excitation of water signal can lead to large water signal leading to the necessity for water suppression. The use of low bandwidth pulses positioned at 4.4 ppm downfield to water in the current method allows the suppression of water signal without the requirement of traditional water suppression techniques.

We performed NAD⁺ SVS in the occipital lobe for comparison with previously published reports employing ³¹P (19) and ¹H MRS (21). In addition, to assess the potential of the technique in other brain regions, ¹H MR spectra were also acquired from voxels placed in the parietal lobes of two subjects. The measured cerebral NAD⁺ concentrations were found to be similar to those previously reported in the literature (4,17,21). Since detection and quantification of NAD⁺ using this pulse sequence relies on accurate reference voltage settings for the E-BURP-1 excitation pulse, B₁ variations currently limit the application of this method to a single voxel.

Cellular NAD⁺ is reduced and converted into NADH during glycolysis, the Krebs cycle, and a host of other reactions. NADH is then recycled back to NAD⁺ by donating the electrons to the mitochondrial electron transport chain or to support reactions such as the conversion of pyruvate to lactate by lactate dehydrogenase. All of these processes are directly linked with cerebral glucose metabolism (CMR_Glc) and neurotransmitter cycling. It has been shown recently that CMR_Glc steadily reduces with healthy aging by ¹⁸F-FDG PET (46) and ¹³C MRS studies (47). A recent study of mice gastrocnemius muscle has shown by biochemical analysis that NAD⁺ levels of 24 and 30-month old mice were reduced by a factor of 2 compared to 6-month old young adults (13). The authors also showed recovery of reduced NAD⁺ levels of 22-month old mice to the same level as the 6-month old mice following treatment with NMN (13). Our results also support a reduction in the level of cerebral NAD⁺ with age, though this should be replicated in a larger sample. Future work can leverage ultra-high field ¹H MRS of cerebral NAD⁺ to study the effects of NMN and other supplements that enhance NAD⁺ levels in older humans. This technique may further be useful in probing the role of NAD⁺ changes in neurodegenerative diseases by studying subjects with disease and age matched controls.

Conclusions

We demonstrated a new pulse sequence capable of spatially localized measurement of brain NAD⁺ concentrations using ¹H MRS at 7T with a volume head coil and showed a decline in the cerebral NAD⁺ in with healthy aging. This non-invasive technique provides a biomarker of brain redox state and enables the evaluation of potential therapies targeting the enhancement of NAD⁺ concentration in humans.

Supplementary Material

Supp figS1-3

Supporting Information Figure S1. Pulse shapes and the Bloch simulation M_Y profile of 90° and 180° pulses. a,b) pulse shape and simulation profile of E-BURP-1 90° pulse. c,d) pulse shape and simulation profile of SLR 180° pulse

Supporting Information Figure S2. Effect of block averaging of the NAD⁺ spectra. Number of blocks used was a) 256, b) 32, c) 8, and d) 1, on a dataset acquired from one volunteer

Supporting Information Figure S3. Water and NAD^{+ 1}H MR spectra. The full spectra from water and NAD⁺ scans are depicted in a and b respectively showing the selective excitation of water and downfield region using the pulse sequence. The spectral bandwidth for all scans was 4 kHz centered at 4.7 ppm