Deuterium labeling enables non-invasive 3D proton MR imaging of glucose and neurotransmitter metabolism in the human brain

28 Impaired brain glucose metabolism characterizes most severe brain diseases. Recent 29 studies have proposed deuterium ( 2 H)-Magnetic Resonance Spectroscopic Imaging (MRSI) 30 as a reliable, non-invasive, and safe method to quantify the human metabolism of 2 H-labeled 31 substrates such as glucose and their downstream metabolism (e.g., aerobic/anaerobic 32 glucose utilization and neurotransmitter synthesis) and address the major drawbacks of 33 positron emission tomography (PET) or carbon ( 13 C)-MRS. Here, for the first time, we show 34 an indirect dynamic proton ( 1 H)-MRSI technique in humans, which overcomes four critical 35 2 H-MRSI limitations. Our innovative approach provides higher sensitivity with improved 36 spatial/temporal resolution and higher chemical specificity to


Study design 117 118
Five healthy, right-handed volunteers (30±4 y.o., 5 males) were scanned on a 7T whole-body 119 MR scanner (Siemens Healthcare, Erlangen, Germany) utilizing a 32-channel receive-array 120 coil (Nova Medical, Wilmington, MA, USA). All participants were lean (BMI = 22.6±1.4 kg/m2) 121 without a history of diabetes or other metabolic and severe diseases. The study was 122 approved by the Ethical Commission at the Medical University of Vienna. All participants 123 signed informed consent. Each participant underwent two MR scans, one after 6,6'-2 H2-Glc 124 (Deu-Glc), the other after non-deuterated D-Glc (normal dextrose, nonDeu-Glc) 125 administration; the scans were 92±49 days apart. All these scans used an interleaved single-126 voxel MRS/MRSI protocol. One subject was additionally scanned after Deu-Glc 127 administration with MRSI-only protocol to obtain the time-course with high temporal 128 resolution. All sessions were conducted in the morning after an overnight fast. Both 129 compounds were dissolved in ~300 mL of water and ingested in equal amounts (0.8 g/kg 130 body weight) immediately before the scan was initiated. MRSI and MRS data were where only the MRSI data were acquired with a high time resolution of five minutes after Deu-149 Glc ingestion. 150 MRSI data were obtained via an FID-MRSI sequence 11 with an ultra-short acquisition delay 151 of 1.3 ms, a short TR of 320 ms, and ellipsoidal 3D k-space encoding using concentric ring 152 trajectories (CRT), variable temporal interleaves, 36x36x26 matrix, 5x5x4.8 mm 3 voxel size, 153 2:58 min per block, 558 complex points, 34° (Ernst) excitation flip angle, 600 µs pulse 154 duration, and 7 kHz pulse bandwidth. The excited slab was centered around the posterior 155 cingulate region. 156 Single-voxel MRS data were obtained from the posterior cingulate (PCC) region using a semi-157 LASER sequence (TR 7 s, TE 28 ms, 3:43 min per block, 2048 complex points, 90° 158 asymmetric sinc pulse with a duration of 2.5 ms, FOCI pulse bandwidth, and a duration of 159 4.2 ms, 45 kHz). 17 A 22×20×20 mm (AP×LR×SI) voxel was placed mid-sagittally, based on 160 anatomical landmarks. The voxel was rotated in the sagittal plane by 30° such that it was 161 aligned with the posterior border of the splenium. To mitigate possible effects of patient 162 motion and chemical shift displacement, the voxel was backed away anteriorly from the 163 splenium and caudally from the occipital-parietal fissure by 2 mm. 18 All spectra were collected 164 with water suppression 19 and outer volume suppression (number of excitations, NEX = 32) 165 along with unsuppressed water spectra utilized to remove residual eddy currents (NEX = 2) 166 and as a reference from which to derive metabolite concentration estimates (NEX = 2). 167 168

Segmentation of MPRAGE scans 169 170
The T1w-MRI images were segmented in Freesurfer (v.5.3) to obtain masks of the brain gray 171 matter (GM) and white matter (WM). The masks were resampled to the MRSI space and 172 used to obtain tissue-specific averages of metabolite levels. In addition, probabilistic maps of 173 the GM, WM, and cerebrospinal fluid (CSF) were derived by segmenting the T1w-MRI images 174 using the SPM12 software package. The probabilistic tissue maps were thresholded with an 175 in-house-written MATLAB script using the iterative method of threshold selection 20 to 176 determine the within-PCC-VOI fraction of GM, WM, and CSF.  Finally, the spectra representing the first time-points (FIRST) and the last time-points (LAST) 200 were pooled together and summed, resulting in two sums (FIRST and LAST) for each session 201 (Deu-Glc and nonDeu-Glc). The FIRST and LAST from the Deu-Glc session were subtracted, 202 and a difference spectrum was calculated. The difference spectrum characterized metabolite 203 changes following Deu-Glc ingestion. In addition, the LAST points from the two sessions 204  Modifications to the basis sets were performed based on the previous animal experiments, 215 theoretical predictions, and preliminary analysis of the current data. Glu, Gln, and GABA 216 included the split of proton signals that originated from different carbon positions for C2, C3, 217 and C4. Thus, the basis set included all molecular variants that occurred in the brain in a 218 ( 1 H-2 H) coupling constants. For instance, the basis set of glutamate included six components, 220 i.e., those present in the molecule with all positions occupied with protons (C2 1 H, C3 1 H2, 221 C4 1 H2), in the molecule with one deuteron on the C4 (C2 1 H, C3 1 H2, C4 1 H 2 H), and two 222 deuterons on C4 (C2 1 H, C3 1 H2). Thus, we distinguished three variants of the C3 resonance 223 affected by homo-and heteronuclear coupling with 1 H and 2 H at C4. While the couplings 224 between protons and deuterons on the C3 and C4 (Glu and Gln) and C2 and C3 (GABA) 225 were simulated, the couplings between C2 and C4 were minimal and were neglected. Yet, 226 the number of components had to be reduced for MRS and MRSI data to preserve the stability 227 of the fits. This was performed by neglecting the heteronuclear and homonuclear coupling 228 effects of the deuteration, assuming that the dominant signal change occurred on the position 229 where proton(s) is/are replaced by deuteron(s), i.e., C4 for Glu and Gln and C2 for GABA. 230 Thus, we used only two components per metabolite (Glu2+3, Glu4; Gln2+3, Gln4; and GABA2, 231 GABA3+4) to quantify SV-MRS data. The basis set was further simplified for MRSI, where we 232 split only Glu2+3 and Glu4 resonances since the fitting with more components yielded a less 233 stable quantification of neurochemical profiles. Finally, we used the components whose 234 signals were expected to change due to progressive deuteration during the scans (i.e., for 235 Glu and Gln: C4 1 H2, C4 1 H 2 H and two variants of C3 1 H2 present in the molecule with and two 236 deuterons; and for GABA: C2 1 H2, C2 1 H 2 H, and two variants C3 1 H2) to quantify the difference 237 spectra. As difference spectra do not contain the background of signals from the metabolites 238 that were stable over the task, we could account for the subtle hetero-and homonuclear 239 coupling effects. Voxels with sufficient temporal stability and spectral quality were selected using masks. The 244 masks included voxels with CVs below 12% for the three main metabolites that remained 245 stable during the scan, i.e., tCr, tCho, and tNAA. The criteria also utilized parameters 246 provided by LCModel (FWHM<0.1 ppm, SNR>5, zero-order phase <40°) in line with expert 247 recommendations. 29 These criteria were utilized to calculate regional GM and WM means in 248 metabolite concentrations, as well as to select the spectra used for the calculation of high-249 SNR sums that represented either GM or WM. The metabolite concentrations quantified with 250 all CRLBs were used for further analysis except those that could not be quantified with CRLBs 251 of 999%. This criterion avoided bias due to an arbitrarily set CRLB threshold, cutting off lower 252 concentrations with higher relative CRLB. 30 The metabolites quantified with a CRLB of 999% 253 in most of the time-points (MRS), or consistently in more than 10% of voxels (MRSI), were 254 Gln (Gln4) by 14.3% ± 2.0% (p = 0.00014) and 12.3% ± 4.5% (p=0.0019), respectively, 294 following Deu-Glc ingestion in accord with previous animal data, 1 Glu4 and Gln4 were stable 295 during the control experiment using the administration of nonDeu-Glc (Table 1 Bar diagrams demonstrate means (errors bars represent standard error of the mean) of concentration comparisons quantified from the first and last time-point spectra following Deu-Glc and nonDeu-Glc administration. Peaks that originated from the specific carbon position undergoing deuteration, namely, the 4 th carbon position for Glu (Glu4) and Gln (Gln4) and the 2 nd carbon position for GABA (GABA2), were separated. Concentrations were compared with a standard, two-tailed, paired t-test between the first and last time-point within the Deu-Glc and nonDeu-Glc sessions.
Glc scan than during the nonDeu-Glc session, reflecting the 1 H-invisible signal from the 306 deuterated component of Deu-Glc. The Glc time-course gradually increased during the first 307 half of the experiment and returned to baseline values subsequently. As our experiment 308 resembled an oral glucose tolerance test, the time-course approximately reflects the glycemic 309 levels typical for healthy subjects. Changes in the signal of GABA and GABA2 were not 310 statistically significant, possibly due to higher variance in GABA and GABA2 typical for non-311 edited 1 H-MRS methodology. The robust signal change is visually discernible on individual 312 subject spectra (Fig.2) 313 Table 1. Metabolite concentration estimates, stability, and temporal changes with and without deuterated glucose. Data are calculated from concentrations quantified in µmol/g for single-voxel MRS and referenced to total creatine (tCr) for 3D-MRSI. The regional means from the gray and white matter voxels were used to calculate between-subject averages in metabolite concentrations. The average concentrations and their respective coefficients of variations measured immediately after nonDeu-Glc administration (the first timepoint). The within-session differences were calculated by subtracting the concentrations of the first and last data point, and their statistical significance was assessed via a standard, two-tailed, paired t-test (separately for Deu and nonDeu scans). The asterisks indicate statistical significance after correction for multiple comparisons with the false discovery rate method, which limited the likelihood of false positives to 3%. conc.

Deu-Glc
Dconc. nonDeu-Glc   .3. Difference spectra and their quantification. Summed spectra from all subjects (N=5) represent the first and last time-points after Deu-Glc and nonDeu-Glc ingestion. The spectra in panel A and B were linewidth-matched with exponential line-broadening and subtracted. While the resulting difference spectrum in panel A represents the effect of metabolite Deu enrichment, the similar difference spectrum in panel B corroborates the fact that metabolite changes were not related to the nonDeu-Glc administration, but were, indeed, a consequence of the deuterium enrichment. The difference spectra reflect oscillations of the Glc signal after Deu/nonDeu-Glc administration. The metabolite components were obtained via LCModel analysis using a basis set containing simulated spectra of neurochemicals that were undergoing deuteration (i.e., Glu, Gln, and GABA). The proton signals that originated from different carbon groups were separated. calculated by summing the spectra from all subjects (N = 5). The summed spectra and their 318 differences displayed in Fig. 3 clearly demonstrate the effect of deuterium enrichment (Fig.  319 3A) and rule out hyperglycemia and/or acquisition instabilities as possible confounders (Fig,  320   3B). While the quantification of single-subject time-courses did not reveal a change in GABA2, 321 the quantification of difference spectra yielded Cramér-Rao lower bounds, i.e., the estimates 322 of quantification error, of 16% for GABA2. Glu4 and Gln4 concentrations during the session (Fig. 4). The exponential rate constants (tau) 327 were obtained per subject from Glu4 with a coefficient of variation of 26%. The decaying 328 slopes of Gln4 were obtained with a between-subject variance of 50%. 329 Overall, concentrations of 14 metabolites were quantified with average within-session CVs 330 below 5% (MM, Cr, GPC, myo-Ins, PCr, PE, NAA, Tau, Glu2+3, Glu4, Gln4, tCho, tCr, tNAA), 331

Fig. 4. Fitting of time courses obtained by quantification of single-voxel MR spectra in LCModel.
Decay in the concentrations of glutamate (Glu4) and glutamine (Gln4) were fitted using the exponential function Y=Y0 (-t/tau) +c (Glu4), and linear regression Y = a + bX (Gln4). The time-courses and their fits following Deu-Glc ingestion contrast with those with stable Glu4 and Gln4 fits in nonDeu-Glc sessions.
20% as assessed from concentrations obtained from the nonDeu-Glc scan (Table 1) Table 1). For 339 all other reliably quantified metabolites, no significant changes were found during the Deu-340 Glc scan (Table 1). As expected, all metabolites were stable during the nonDeu-Glc session 341 (tCho, tNAA, GABA, Gln, Glu4, Glu2+3, myo-Ins). The regional difference in the Glu4 drop 342 between GM and WM during the Deu-Glc session was reflected by the spectra shown in Effect of Deu-Glc on the spectra obtained from the gray and white matter with 3D multivoxel MRS. Difference spectra were calculated by subtraction of spectra obtained from the first and last timepoint after deuterium ingestion in one healthy volunteer. The spectra were selected using a quality control mask and segmented gray or white matter masks. The signal loss at 2.34 ppm reflects the exchange of protons and deuterons at the 4 th carbon position in the glutamate molecule and is displayed as a positive peak in the difference spectra. The signal decay is convincingly found in both gray and white matter. Lack of unwanted signals in the quantified range 1.9-4.2 ppm verifies good spectral quality and stability during the acquisition. are shown in Fig.6. The exponential fit of Glu4 concentrations yielded tau values (the rate 352 constants) of 44±22 minutes and 52±23 minutes in GM and WM, respectively, in the Deu-Glc 353 sessions. The decay was 18% faster in GM than in WM, on average. The linear regression 354 of the Glu4/tCr time-courses measured during the nonDeu-Glc session (exponential fits) 355 yielded non-significant slopes of 0.00002±0.00006 and 0.0001±0.0002 (p>0.07), thus 356 indicating no effect of glucose load on the Glu4 and good stability of the measurements. While 357 the between-subject CVs for the concentration differences in Glu4 were 26% (GM) and 21% 358 (WM), the respective CVs calculated from the rate constants (tau) were 50% (GM) and 45% 359 (WM). In GM and WM, 372±169 and 737±280 voxels, respectively, fulfilled the quality assessment 374 criteria and were, thus, used to calculate regional means for each time-point and their CVs 375 per session (Table 1). LCModel quantified 10 metabolites referenced to tCr with within-376 session CVs below 7%, i.e., Asp, myo-Ins, Tau, Glu2+3, Glu4, Gln, GABA, tCho, and tNAA as 377 assessed on the data measured in the nonDeu-Glc session. Namely, Glu4 was consistently 378 quantified with CVs below 2% in both GM and WM. 379 380 Finally, single-subject MRSI time-courses obtained with high time resolution (Fig. 7A, 7B) 381 were fitted with linear regression per voxel (Fig. 7C). The resulting slopes (p < 0.05, r < -0.8) 382 were steeper in the GM than in the WM by 19% and yielded a contrast between gray and 383 white matter on the slope map (Fig. 7C). The linear regression was found more appropriate 384 Fig 6. Fitting of time courses from averaged regional MRSI maps: Exponential averaged decays of Glu4/tCr within-session time-courses after Deu-Glc ingestion (A) are in contrast with stable linear fits for Glu4/tCr after nonDeu-Glc (B). The exponential fits showed 18% faster decay (smaller constant of the decay -tau, M=M0 (-t/tau) +c) in the gray (blue) than in the white matter (red). than using an exponential function to fit time-courses obtained from each voxel with a lower 385 signal-to-noise ratio than time-courses that used regional means, as shown in Fig. 7B. 386  ) show fractional decay of regional Glu4 averages in the gray and white matter regions. Voxel-wise linear regression was applied to Glu4 time-courses (panel C). Respective slopes tend to be higher in the gray matter, which suggests a higher glutamate turnover in the gray than in the white matter.

Discussion (2008 words) 387 388
The current work reveals the tremendous potential of deuterium labeling to measure the 389 turnover of metabolites involved in oxidative glucose metabolism in the human brain. Glu4 changes captured in individual data. The fitting of difference spectra constructed by 420 subtracting spectra pooled from multiple subjects indeed benefits from a high signal-to-noise 421 ratio and elimination of signals from static metabolites, including the molecular background. 422 We expect a further boost of quantification of J-coupled relatively low-abundant metabolites 423 such as GABA and Gln by optimization of the sequences for higher fields (i.e., above 7T) 424 and/or by spectral editing techniques, possibly enabling robust GABA detection even at low-425 field MR scanners. 37,38 . 426

427
The fact that the resting and the stimuli-activated brain both utilize the same mitochondrial Our study demonstrated major methodical advancements that are critical for future 448 applications. We reduced the voxel volume, minimized the partial volume effect, and 449 achieved substantial signal gain due to the implementation of a concentric ring trajectory 450 readout. 11, 25 We have already demonstrated that MRSI can be further accelerated via k-451 space undersampling to extend the coverage of 2 H-to-1 H-MRSI over the whole brain. 23 The 452 spatial resolution in our project was almost two magnitudes higher than for previous direct 453 subjects. 51,52 This can be mainly ascribed to high physiological variation in the resting brain 471 metabolism. Also, this is a preliminary study, and further technical improvements are 472 expected to yield even higher reproducibility also on a voxel-by-voxel level. The variance can 473 be further lowered by using only a single, e.g., only 2 H-to-1 H-MRSI technique, thereby 474 improving the time resolution and the exponential fits ( Fig 7B) and yielding ~3x more time 475 points than the current study. The stability of the measurements might be further increased 476 by prospective motion correction methods based on tracking with an optical camera or 477 navigators, 53 further boosting the stability of quantification while refining the assessment of 478 smaller brain regions. The drawbacks of the current 1 H-MRSI methods can be overcome by 479 advanced B0 shimming approaches, such as higher-order shims above 2 nd order, 480 complementary matrix B0 shims, and dynamic shim updates to mitigate temporal B0 481 instabilities during dynamic acquisitions. These advancements will make challenging brain 482 regions (hippocampus) and more spectral components accessible with higher reproducibility. 483 Finally, a major developmental aspect will be the incorporation of (k,t)-undersampling 484 methods, which are routinely used in clinical, dynamic, contrast-enhanced MRI 54 and for 485 which >10-fold improvements in temporal resolution or alternatively higher spatial resolution 486 at similar temporal resolution are currently commonly achieved without compromising spatial 487 resolution. 55,56 Despite these methodological limitations, we clearly proved that observed The primary hormone responsible for the regulation of peripheral and brain glucose 512 metabolism is insulin, which is released from the endocrine pancreas in response to rising 513 blood glucose levels. Insulin is transported across the blood-brain barrier to the CNS via 514 saturable transporters, affecting both brain function and peripheral metabolism. 14 While 515 peripheral hyperinsulinemia does, indeed, lead to brain hyperinsulinemia, their extent and 516 kinetics are distinct since the transport of insulin to the brain is saturable, and the insulin has 517 a longer half-life in the brain than in the periphery 58 . There is no clear consensus about 518 whether and how systemic hyperinsulinemia affects brain metabolism. 59-61 Although we did 519 not observe significant changes in metabolite levels after dextrose administration, in 520 agreement with the previous study, 62 we cannot rule out that hyperglycemia and 521 hyperinsulinemia affect the speed of deuterium enrichment that is not accompanied by 522 changes in metabolite concentrations. In another study, however, insulin infusion did not 523 significantly impact glucose transport kinetics with plasma insulin concentrations up to 100 524 pmol/L. 60 Therefore, our measurements of metabolite decays with postprandial (moderate) 525 without insulin intolerance. Also, in another study, the circulating insulin did not influence the 527 brain glucose uptake, but the insulin resistance status significantly affected brain glucose 528 kinetics. 63 Hence, while peripheral insulin does not play a significant role in the brain glucose 529 metabolism, the metabolic effects of central insulin action and resistance could be more 530 pronounced. Obesity and type 2 diabetes are associated with alterations in glucose 531 metabolism and insulin resistance, including central insulin signaling defects, 64 which also 532 play an essential role in AD development. 65 In addition, the transport of insulin across the 533 blood-brain barrier is affected by insulin resistance and obesity. compounds. Thus, the current methodology provides a critical step forward for future 556 metabolic projects in the resting or activated brain in disease, health, and aging, which will 557 offer relevant metabolic information using widely available hardware in a single MR session. 558