Deuterium brain imaging at 7T during D2O dosing

To characterize the (2H) deuterium MR signal measured from human brain at 7T in participants loading with D2O to ˜1.5% enrichment over a six‐week period.


INTRODUCTION
The low natural abundance (∼0.015%) and gyromagnetic ratio (6.54 MHz/T) of deuterium ( 2 H) reduce the available NMR signal compared to 1 H. However, the shorter longitudinal relaxation times of 2 H allow faster signal averaging, partially compensating for the reduction in SNR associated with the reduced signal strength. The minimal equipment modifications required for implementing 2 H imaging and the simplicity of the pulse sequences that can be used, mean that 2 H imaging has significant potential for use in clinical applications. This has led to an increasing interest in the use of deuterium magnetic resonance in conjunction with injection or ingestion of 2 H-labeled compounds, as a means of monitoring cellular metabolism. [1][2][3] Deuterium metabolic imaging (DMI) involves using 2 H chemical shift imaging to map the distribution of the metabolic products of administered 2 H-labeled compounds. The majority of experiments in humans and animals have used [6,6 ′ -D 2 ] glucose. [2][3][4][5][6][7][8][9][10][11][12][13][14][15][16] In this case, lactate and glutamine/glutamate (Glx) are produced alongside deuterated water (HDO), and their relative concentrations reflect the cells' preference for glucose metabolism, i.e., aerobic, anaerobic glycolysis or oxidative phosphorylation. Spatially localized, elevated, lactate production has been observed using DMI in a patient with a glioblastoma, 2 in keeping with an increased glycolysis in neoplastic cells, known as the Warburg effect. 17 DMI can provide spatially resolved measurements of metabolite concentrations and pathway fluxes, 1,4 but this often requires knowledge of the local relaxation times of the 2 H signals from metabolites. The signal from naturally abundant HDO can be used in calculating absolute concentrations of other metabolites. Therefore, knowledge of the relaxation times of HDO in different tissues is important for quantitative measurements. Previous measurements of HDO relaxation times in human participants have used non-localized signals 2,3,11 which do not allow the variation of relaxation times across regions and compartments to be evaluated.
Oral intake of heavy water is commonly used for assessment of body composition 18 and is increasingly being applied in studies of triglyceride synthesis 19 and protein turnover. [20][21][22] These approaches generally involve analysis of body fluid samples or tissue biopsies. However, 2 H magnetic resonance allows direct, non-invasive, measurement of the concentration of HDO, deuterated lipids, and other metabolic products of 2 H labeled water and so could complement invasive measurements. The feasibility of imaging the distribution of HDO in the body following oral ingestion of D 2 O was demonstrated in animal experiments carried out in the 1980 s, 23-27 but has not yet been performed in humans.
Here, we implemented deuterium MRI at 7T and used it to characterize HDO signals from the human brain in four healthy participants who increased their deuterated water content to ∼1.5% for a six-week period by drinking D 2 O. The heavy water loading was carried out as part of a parallel study.

METHODS
Six healthy participants took part in this 2 Himaging sub-study which was approved by the local institutional ethics committee with the volunteers giving informed consent. Two participants were scanned during a set-up phase in which we established the feasibility of 2 H imaging and identified favorable imaging parameters. Here, we report data from four participants (A-D) who were subsequently scanned using the optimized imaging protocols on a 7T Achieva scanner (Philips Healthcare), operating at 45.  [28][29][30] with no adverse events reported, however some participants experienced a brief period of dizziness during the initial loading phase due to the rapid rise in body water enrichment. 28 Saliva samples were collected from participants at regular intervals during the study and analyzed using gas chromatography-mass spectrometry (GCMS). 31 Two participants (A and B) were scanned during the initial eight-hour loading period to monitor the time-course of deuterated water concentration changes in the brain. A scanning protocol of ∼15 minutes duration was performed before dosing and after 30, 90, 150, 210, 270, 360, 420, and 540 minutes. The protocol comprised a 1 H scout scan for planning, followed by acquisition of 2 H pulse-acquire spectra from the whole head and from a 2-cm-thick axial slice positioned over the lateral ventricles. Both used the following scan parameters: flip angle = 90 • , 2048 samples, bandwidth (BW) = 3000 Hz, repetition time TR = 1 s an 64 averages (scan time, T scan = 64 s).
We then acquired axial, 3D MEGE 2 H images (20 averages, T scan = 453 s, FOV = 288 × 288× 80 mm 3 , 6 × 6 × 10 mm 3 voxels, = 33 • , TR = 62 ms, five echoes, TE 1 = 8.9 ms and ΔTE = 8.4 ms). Axial 1 H 3D GE images (T scan = 232 s, 32 slices, FOV = 288 × 288 × 80 mm 3 , 3 × 3 × 2.5 mm 3 voxels, TE = 5.9 ms, TR = 39 ms) were also acquired. This scanning protocol was repeated 17 days after the initial loading to provide comparative data at steady-state enrichment. The spectroscopy measurements made before loading provided an estimate of the signal from naturally abundant deuterium in water: scaling subsequent measurements then allowed the absolute HDO concentration to be estimated at each time-point. The HDO concentration in the body was also estimated from the ratio of the total imbibed D 2 O volume to an estimate of total body water. 32 We used the image data to track the changes in 2 H signal from different tissue compartments. The 2 H images acquired at each time-point were summed over the five echoes, and regions of interest (ROI) were then formed for a background region, general brain tissue, the lateral ventricles and for a region of high signal intensity thought to arise from blood vessels and CSF in the superior cistern. The SNR in images acquired before loading was too low to make good estimates of natural abundance signals, so values were normalized to the signal measured in the superior cistern ROI at the last time point of the initial loading period. 2 H relaxation times for water in CSF, GM, and WM were calculated from data acquired during the six-week loading period using the dual-tuned 2 H/ 1 H coil. In each session, we acquired 2 H 3D sagittal MEGE images (voxels = 6 × 6 × 10 mm 3 , FOV = 288 × 288 × 240 mm 3 , slices = 24) at a range of TR values, along with 1 H 3D MEGE images (voxels = 3 ×3 × 5 mm 3 , FOV = 288 × 288 × 240 mm 3 , slices = 48, 15 echo times with TE 1 = 2.5 ms, ΔTE = 2.34 ms, and TR = 41 ms). 2 H MEGE data from Participants A and B were acquired with five echoes (TE 1 = 4.3 ms, ΔTE = 8.4 ms), = 60 • , and TR = 68, 136, 272, 544 ms, with 8, 4, 2, and 1 averages, so that T scan = 487 s per image. 2 H MEGE data from participants C and D were acquired with six echoes (TE 1 = 4.3 ms, ΔTE = 8.4 ms) and one additional TR-value (TR = 816 ms, one average, T scan = 730 s). The number of TE and TR values were increased to improve fitting quality. The 2 H scanning sessions were performed twice on Participants C and D.
We also acquired 1 H MPRAGE images (0.7 mm resolution) and 1 H 3D MEGE images (3 × 3 × 5 mm 3 voxels, 15 echo times, TE 1 = 2.5 ms, ΔTE = 2.57 ms, and TR = 41 ms) from each participant in a separate scanning session using the Nova coil. These images were used for image segmentation and estimation of the 1 H T 2 * values.
For calculation of maps of 2 H relaxation rate constants R 1 and R 2 * , we first estimated the variation of flip angle ( ) over the image volume by summing the images across TEs, at each TR, and fitting the data voxel-wise to a saturation recovery curve (i.e., fitting signal variation with TR for , R 1 , and signal amplitude). The resulting flip-angle maps were smoothed by averaging over 5 × 5 × 5 voxel neighborhood and the -values then used as fixed parameters in dual-fitting the variation in signal intensity S i,j across TR i and TE j values for R 1 , R 2 * and signal amplitude, A. This involved minimisation of (1) using the Matlab fmincon command. 1 H R 2 * maps were obtained by similar fitting to the exponential signal decay with TE in the 1 H MEGE data acquired using the Nova coil.
To evaluate the relaxation times in different compartments, we segmented the 1 H MPRAGE data (FSL FAST 33 ) and transformed the resulting GM, WM, and CSF masks to the space of the 2 H relaxation time maps. Following brain extraction (FSL BET 34 ) and bias field correction, an affine matrix was obtained from image co-registration (FSL FLIRT 35,36 ) that transformed the 1 H MEGE data acquired using the Rapid Biomedical coil to the space of the 1 H MEGE Nova Medical coil data, along with an affine matrix for the 1 H MEGE to MPRAGE transformation. The MEGE data were summed across echoes and repetition times before co-registration.
The brain-extracted MPRAGE image was segmented to create binary masks for GM, WM, and CSF using FSL FAST. 33 These masks were transformed to the 2 H space using the previously obtained affine matrices and the outer regions of the CSF mask were manually removed so that the majority of the mask comes from the lateral ventricles. The new masks were applied to the relaxation maps for calculation of mean relaxation times for CSF, GM, and WM. Figure 1 shows example 2 H image data obtained during the steady-state loading period. Figure 1A shows 3D sagittal image data produced by summing the MEGE data over TE and TR values. The resulting images clearly depict the brain anatomy and have a similar appearance to T 2 * -weighted, 1 H-images. The CSF in the ventricles and at the cortical surface appears hyperintense, while regions where there is little partial-voluming with CSF, such as the white matter in the corpus callosum, appear hypointense. Figure 1B shows the variation of image intensity with TE and TR in a central sagittal slice. The slower T 2 * decay of the CSF signal compared with that of the GM and WM signals is evident, along with the signal saturation at reduced TR, and the reduction of contrast at low TE and TR values. Figure 2 shows maps of the relaxation rate constants from two participants, with the dominant feature in the 2 H maps being the reduced R 2 * and R 1 values in the ventricles. Table 1 reports the average and SDs of the 2 H T 1 and T 2 * values, along with 1 H T 2 * values measured in GM, WM, and CSF in the four participants. Figure 3 reports example 2 H images acquired from Participants A and B during the loading process. The different regions of interest in which the signal changes were tracked are indicated on Figure 4A (for Participant A) and Figure 4B shows the time courses from the different ROIs, along with the steady-state values measured after 17 days of loading. Figure 4C plots the temporal variation of the absolute 2 H concentration estimated from the spectroscopy measurements. The concentration calculated from the cumulative D 2 O dose and estimated total body water volume is shown for comparison.

DISCUSSION
The results shown in Figures 1 and 3 indicate that 2 H images of 6 × 6 × 10 mm 3 voxel size with a useful SNR can be acquired in 7.5 minutes at 7T with a head-sized bird-cage coil, when participants have been deuterium-enriched to ∼1.5% concentration (∼100 times natural abundance). After summing over echo times these images ( Figure 3) showed SNR ∼16 in brain tissue in the steady-state condition (after 17 days of loading). The measured relaxation times were reasonably consistent across the six measurements (Table 1), with CSF having significantly higher T 1 and T 2 * values than GM or WM (p < 0.007 for two-sample t-test). The measured T 1 and T 2 * values were consistently higher in GM than in WM, but the differences did not reach statistical significance (T 1 : p = 0.21; T 2 * : p = 0.08). The relatively coarse resolution of the 2 H images made it difficult to avoid the effects of partial voluming, particularly of CSF and GM, and the limited range of TE (4.3-46.3 ms) and TR (68-816 ms) values reduced the accuracy of measurement of the long T 1 and T 2 * values in CSF, as is evident from the larger SDs of these measurements (Table 1). Longer echo trains with a duration that exceed the expected T 2 * value and measurements at longer TR values should be used in future experiments targeting a better characterization of 2 H HDO relaxation times in CSF. For example, simulations show that inclusion of an additional measurement with a TR of 1500 ms would halve the SD of the estimated T 1 relaxation time of CSF but would also require 20 minutes of additional scanning time: use of an inversion recovery sequence may therefore be a better option. The average values of the relaxation times are consistent with values reported from non-localized measurements of HDO signals in human, 2,3,11 cat 27 and rat 1,2 brain.
Focusing on human brain measurements, De Feyter et al. 2 reported HDO T 1 of 346 ± 5 ms at 4T, while Ruhm et al. 11 measured 362 ± 6 ms at 9.4T -values which lie between the values for CSF (510 ms) and GM/WM (320/290 ms) measured here at 7T. As expected, the measured 2 H T 1 -values are significantly shorter than the corresponding 1 H values at 7T, 32 due to the quadrupolar relaxation of 2 H. The long T 1 of HDO in CSF relative to GM/WM will lead to greater saturation in the CSF signal in short CSI measurements used for DMI (for example Ruhm et al. used TR = 155 ms 11 ) which needs to be considered when quantifying signals from other 2 H-labeled metabolites using natural abundance HDO signals. Bi-exponential T 2 decay was previously identified at 4T 2 and 7T 37 using non-localized spin echo measurements: at 7T large (small) pools were found to have relaxation times of 29 ± 1 (412 ± 40) ms, respectively, 37 consistent with our identification of short Note: These values were produced by averaging over segmented relaxation time maps, similar to those shown in Figure 2. Average values and SDs across participants are also shown. The TE-summed MEGE images in Figures 1A and 3 show contrast that is dominated by T 2 * -weighting, with the CSF appearing hyperintense relative to gray and white matter, as is the case in T 2 * -weighted 1 H images. A notable difference between the 1 H and 2 H R 2 * maps ( Figure 2) is that deep GM structures which appear with elevated R 2 * in 1 H maps due to their high iron content 38 do not appear hyperintense in the 2 H maps. This is a consequence of the dominance of quadrupolar, rather than dipolar, relaxation in the case of 2 H and the relatively short T 2 relaxation times that the quadrupolar interactions produce in tissue, along with the lower 2 H gyromagnetic ratio. Together these mean that the large, microscopic and macroscopic field inhomogeneities generated by the iron-rich inclusions in deep GM structures regions, which increase the 1 H R 2 * relaxation rate constants, do not have a significant effect on the measured 2 H R 2 * -values. The 1 H R 2 * maps also show larger regions of hyperintensity near the frontal sinuses due to the greater field-inhomogeneity-related intra-voxel dephasing resulting from the higher γ of 1 H. Signals from structures outside the brain (apart from the eyeball) are only evident in the 2 H images acquired with the shortest TE value ( Figure 1B) most likely because of the very short T 2 * of HDO in muscle. 39,40 Figure 4 shows that the changes in HDO concentration during the initial heavy water loading could be readily tracked with imaging and spectroscopy. The concentrations estimated from the 2 H spectra are in reasonably good agreement with the values calculated from the cumulative D 2 O dose and body mass ( Figure 4C.) The signal amplitudes measured from ROIs in the brain images all have similar time-courses and maintain relatively constant ratios, with values that are most likely dictated by differences in T 2 * -weighting and water fraction in the different brain regions. This implies that the dispersal kinetics following oral ingestion of D 2 O are rapid throughout the body on the timescale of the measurements. This is consistent with previous measurements based on blood sampling which indicate that the half-life of absorption into blood is ∼12 minutes, with similar time constants for dispersal into other body water compartments. 41,42 In our experiments the subject came out of the magnet bore between measurements, leading to the potential for changes in signal intensity due to variation of the slice position. Nevertheless the 2 H signals tracked the monotonically increasing dose and the values measured at maximum dose were similar to those measured 17 days later during the steady state loading period. Although both participants had approximately the same weight and target D 2 O dose, Participant B was only able to ingest 600 ml during the initial loading. The deuterium concentration measured from Participant A was consequently higher at the end of the loading period. The rest of participant B's loading was completed over the following 4 days, along with the daily 50 ml top-up and similar concentrations were measured from the two participants in the steady state ( Figure 4C). The GCMS measurements of deuterium concentrations in the saliva samples from Participant A and B were 1.51% ± 0.09%, and 1.53% ± 0.17%, respectively.

F I G U R E 3
Rapid increases in body water enrichment can lead to feelings of dizziness and nausea. These symptoms can occur at relatively low enrichments while equilibrium has not yet been achieved and are thought to result from temporary effects on the vestibular system due to density changes in the semi-circular canals of the inner ear. 43 Some participants experienced these effects and so the rate of D 2 O loading was slowed. The rapid loading was required for the parallel study, but a more gradual increase in heavy water uptake could be used for future MR-loading experiments to minimize these effects.

CONCLUSIONS
Deuterium MR measurements at 7T have been successfully used to track the increase in concentration of 2 H in brain during heavy water loading to 100 times natural abundance, in four human participants. Gradient echo images with an SNR of 16 and a voxel volume of 0.36 ml could be acquired in 7.5 minutes. 2 H T 1 and T 2 * relaxation times from water in GM, WM, and CSF have also been measured at 7T. These relaxation times can be applied in research protocols using the natural abundance 2 H signal from water for calibration. In future work we aim to track uptake from a single D 2 O dose on a shorter time scale, using faster, interleaved acquisition of 2 H images and spectra.