1H magnetic resonance spectroscopic imaging of deuterated glucose and of neurotransmitter metabolism at 7 T in the human brain

Impaired glucose metabolism in the brain has been linked to several neurological disorders. Positron emission tomography and carbon-13 magnetic resonance spectroscopic imaging (MRSI) can be used to quantify the metabolism of glucose, but these methods involve exposure to radiation, cannot quantify downstream metabolism, or have poor spatial resolution. Deuterium MRSI (2H-MRSI) is a non-invasive and safe alternative for the quantification of the metabolism of 2H-labelled substrates such as glucose and their downstream metabolic products, yet it can only measure a limited number of deuterated compounds and requires specialized hardware. Here we show that proton MRSI (1H-MRSI) at 7 T has higher sensitivity, chemical specificity and spatiotemporal resolution than 2H-MRSI. We used 1H-MRSI in five volunteers to differentiate glutamate, glutamine, γ-aminobutyric acid and glucose deuterated at specific molecular positions, and to simultaneously map deuterated and non-deuterated metabolites. 1H-MRSI, which is amenable to clinically available magnetic-resonance hardware, may facilitate the study of glucose metabolism in the brain and its potential roles in neurological disorders. The sensitivity, chemical specificity and spatiotemporal resolution of proton magnetic resonance spectroscopic imaging at 7 T allow for the discrimination of deuterated and non-deuterated neurotransmitters and glucose metabolites in the human brain.

Impaired glucose metabolism in the brain has been linked to several neurological disorders. Positron emission tomography and carbon-13 magnetic resonance spectroscopic imaging (MRSI) can be used to quantify the metabolism of glucose, but these methods involve exposure to radiation, cannot quantify downstream metabolism, or have poor spatial resolution. Deuterium MRSI ( 2 H-MRSI) is a non-invasive and safe alternative for the quantification of the metabolism of 2 H-labelled substrates such as glucose and their downstream metabolic products, yet it can only measure a limited number of deuterated compounds and requires specialized hardware. Here we show that proton MRSI ( 1 H-MRSI) at 7 T has higher sensitivity, chemical specificity and spatiotemporal resolution than 2 H-MRSI. We used 1 H-MRSI in five volunteers to differentiate glutamate, glutamine, γ-aminobutyric acid and glucose deuterated at specific molecular positions, and to simultaneously map deuterated and non-deuterated metabolites. 1 H-MRSI, which is amenable to clinically available magnetic-resonance hardware, may facilitate the study of glucose metabolism in the brain and its potential roles in neurological disorders.
Non-invasive, affordable and reliable mapping of glucose metabolism in the human brain is needed for clinical and neuroscientific studies. Current approaches, such as fluorodeoxyglucose ([ 18 F]FDG) positron emission tomography (PET), glucose chemical exchange saturation transfer 1 , carbon-13 ( 13 C) magnetic resonance spectroscopy ( 13 C-MRS; ref. 2) and direct deuterium magnetic resonance spectroscopic imaging ( 2 H-MRSI; ref. 3) of glucose metabolism, are technically challenging and require special hardware or expensive synthesis of intravenous Article https://doi.org/10.1038/s41551-023-01035-z for metabolic mapping in humans, we employed state-of-the-art methods of single-voxel (SV) 1 H-MRS 18 and echo-less three-dimensional (3D) 1 H-MRSI 19 , which have previously shown high sensitivity for the detection of small functional responses to physiological stimulation 20,21 . Our objective was to trace and map the decaying metabolite signals after non-invasive oral administration of deuterated Glc, by using widely available hardware. The experiments were conducted in two sessions with 2 H-Glc oral administration to determine between-session repeatability, and in another session after non-deuterated Glc (normal dextrose, 1 H-Glc) ingestion to mitigate the possible metabolic effects of the Glc load, particularly those of hyperglycaemia and insulin release 22 . We also acquired one session with direct 2 H-MRSI and a dedicated coil to directly compare metabolic effects of 2 H-Glc to those obtained indirectly with QELT.
Changes in the concentration of GABA and GABA 2 were not statistically significant, possibly due to higher variance in GABA and GABA 2 typical for non-edited 1 H-MRS methodology. However, GABA 2 changes in difference spectra were quantified, with mean within-session Cramer-Rao-Lower-Bounds (CRLB) <20%, which is generally acceptable for analysis.
To corroborate the analysis of the LCModel (linear combination model) of single-participant data, high signal-to-noise ratio (SNR) spectra were calculated by summing the spectra from all five participants. The summed spectra and their difference (Fig. 3) clearly demonstrated the effect of deuterium enrichment (Fig. 3a-d). While the quantification of single-participant time-courses did not reveal a change in GABA 2 , the quantification of difference spectra yielded acceptable Cramér-Rao lower bounds, that is, estimates of quantification error of 16% for GABA 2 . While the SNR of the 1 H-MRS difference spectrum (Fig. 3c) did not differ much from that of the 2 H-MRS spectrum, the 1 H-MRS difference spectrum shows substantially higher spectral resolution than the 2 H-MRS spectrum (Fig. 3e).
The exponential and linear fits of Glu 4 , Gln 4 and Glc 6 data clearly demonstrated decaying signals in the 2 H-Glc session (session 1; Fig. 4, left column). The slopes were reproduced in session 4 with direct measurement of deuterated compounds, that is, Glx 4 (Glu 4 + Gln 4 ) and Glc 6 ( Fig. 4, right column). While the slopes of directly measured 2 H-Glx 4 (0.03 ± 0.006) had a between-participant variance of 19%, similar to indirectly measured linearly fitted slopes of Glu 4 (−0.01 ± 0.003, coefficient of variation (c.v.) = 18%), Glc 6 had a higher between-participant variance in the slopes measured with both direct (0.009 ± 0.004, c.v. = 46%) and indirect (−0.007 ± 0.002, c.v. = 34%) methods. Figure 4 demonstrates similar concentration differences (first minus last time-point) of ~2 mM for Glx (Glu and Gln) and 0.6 mM for Glc with radioactive tracers. However, a recent animal study has shown the potential of the quantitative exchanged-label turnover (QELT) technique for the indirect detection of deuterated compounds via proton magnetic resonance spectroscopy ( 1 H-MRS) to quantify glucose metabolism in the rat brain 4 . Orally administered deuterium-labelled glucose is readily taken up by brain cells, and the deuterons are incorporated into downstream glucose metabolites 5 . Because deuterons ( 2 H) substitute protons ( 1 H) in the molecule, they do not contribute to the proton spectrum. Hence, an increase in deuterium-labelled metabolites is reflected by a decrease in metabolite signals in 1 H-MRS.
About 95% of the glucose in the healthy brain enters the tricarboxylic acid (TCA) cycle for glutamate cycling. The rest converts anaerobically to lactate (Lac) 6 . Glutamate (Glu) is then involved in the glutamate/glutamine cycle, ammonia detoxification and GABA (γ-aminobutyric acid) synthesis to supply excitatory and inhibitory neurotransmitters. The balance between aerobic and anaerobic glycolysis is critical for maintaining brain energy homoeostasis, and the shift from aerobic glucose (Glc) metabolism to anaerobic pathways (that is, the Warburg effect) characterizes pathological conditions seen in tumours and ischaemia 7 . Anaerobic metabolism also plays a crucial role in mitochondrial dysfunction, an early condition in the pathophysiology of severe neurological diseases such as Alzheimer's disease 8 , and neuropsychiatric disorders such as major depression and schizophrenia 9 .
The gold standard for the clinical examination of metabolismpositron emission tomography (PET), with the glucose analogue 18 F-fluorodeoxyglucose ([ 18 F]FDG)-indeed provides quantitative measurements of the cerebral metabolic rate of Glc (CMR Glc ) with excellent molecular sensitivity and test-retest variability 10,11 . However, [ 18 F]FDG cannot assess downstream metabolism, which is potentially relevant for diagnosis and treatment evaluation 12,13 . 13 C-MRS is a research tool that can quantify these compounds of the brain's metabolic cycles 13,14 ; however, it suffers from poor spatial localization. Hyperpolarized 13 C-MRSI is another emerging molecular MRI method for the rapid and pathway-specific investigation of dynamic metabolic and physiologic processes that is currently evaluated in clinical studies but not yet available in clinical routine 15 . The power of 13 C-MRSI lies in its ability to investigate the entire metabolic pathway, including downstream metabolites, which is not possible via [ 18 F]FDG-PET. However, currently, the widely used substance 13 C1-labelled pyruvate mainly measures the conversion rate of pyruvate to lactate and the entry into the TCA cycle with the formation of carbon dioxide, but it cannot provide information about glutamate/glutamine cycling and GABA synthesis. Both PET and hyperpolarized 13 C-MRSI require additional costly hardware and invasive intravenous substrate administration, with PET also requiring harmful radioactive and unstable tracers. However, parallel measurements on PET/MR hybrid systems with [ 18 F]FDG using functional PET and deuterium metabolic imaging (DMI) can provide an opportunity to evaluate different aspects of metabolism 16,17 .
Recently, direct 2 H-MRS detection (DMI) provided alternative measures to address the aerobic/anaerobic imbalance, but the technique also requires special hardware and estimates only a limited number of deuterated compounds, which diminish its clinical applicability. In contrast, the proposed QELT-MRS promises to overcome these limitations by using harmless and stable tracers and by allowing quantification of both cerebral metabolic rates of Glc (CMRGlc) as well as the turnover of downstream intracellular Glc metabolism 4 .
Because the deuterium method enables quantification of both the oxidative and anaerobic glucose utilization and assesses neurotransmitter synthesis, we aimed to show the capabilities of QELT-MRSI after peroral administration of 6,6'-2 H 2 -Glc ( 2 H-Glc) in the healthy human brain. Compared with direct 2 H-MRS detection, the indirect technique allows for the quantification of extended metabolic profiles while reliably reflecting downstream metabolism, as shown in a single animal study 4 . Motivated by the desire to establish an easy-to-apply approach Article https://doi.org/10.1038/s41551-023-01035-z both methods (QELT and DMI). The changes in Glu 4 and Gln 4 were also in quantitative agreement with a previous 13 C study conducted with orally administered labelled Glc at similar amounts as used in the current study (that is, fractional isotopic enrichments of ~1.5 mM for Glu and ~0.5 mM for Gln in the brain) 24 .
The regional difference in the Glu 4 drop between the GM and the WM during the 2 H-Glc session was reflected by the spectra shown in Fig. 5a. The signal attenuation around 2.34 ppm (that is, Glu 4 resonance) and the lack of other major signal changes in other LCModel-quantified spectral regions (1.9-4.2 ppm) corroborated the robust detection of ΔGlu 4 with single-participant spectral analysis during the 2 H-Glc experiment. The regional averages of the 2 H-MRS spectra measured in the same participant (the last time-point, session 4) are shown in Fig. 5b, bottom row. The 1 H-MR spectra from a single MRSI voxel acquired in three distinct brain regions are shown in Supplementary  Fig. 4 to demonstrate the spectral quality of single-participant data.
As the magnitude of the Glu 4 signal attenuation is influenced by the variable initiation and duration of the MRS data acquisition among scans, each single-participant temporal MRSI dataset was fitted with an exponential function to characterize the speed of signal decay-the time constant (tau)-which is well characterized in each scan session. The averaged fits are shown in Fig. 6a along with averaged fits of the 2 H-MRSI data (Fig. 6b). The exponential fit of Glu 4 concentrations yielded tau values (the rate constants) of 44 ± 22 min and 52 ± 23 min in GM and WM, respectively, in the 2 H-Glc sessions (session 1). While the between-participant c.v.s for the concentration differences in Glu 4 were 26% (GM) and 21% (WM), the respective c.v.s calculated from the rate constants (tau) were 50% (GM) and 45% (WM). The decay was 18% faster in GM than in WM, on average. On the other hand, the slopes of linearly fitted 2 H-MRSI yielded minimal difference between the GM and WM, probably due to the substantial partial volume effect given by the nominal spatial resolution of ~12.5 mm 3 vs ~5 mm 3 for 1 H-MRSI. The between-session c.v.s in the slopes between session 1 and session 2 were 18.3% ± 12.4% and 12.4% ± 7.5% in GM and WM, respectively (Supplementary Fig. 5).
The linear regression of the Glu 4 /tCr time-courses measured during the 1 H-Glc session 3 yielded non-significant slopes of 0.00002 ± 0.00006 and 0.0001 ± 0.0002 (P > 0.07), thus indicating no effect of glucose load on the Glu 4 and good stability of the measurements ( Supplementary Fig. 5).
In the GM and the WM, 372 ± 169 and 737 ± 280 voxels, respectively, fulfilled the quality assessment criteria and were thus used to calculate regional means for each time-point and their c.v.s per session (Supplementary Table 1). The LCModel quantified 9 metabolites referenced to tCr with within-session c.v.s below 7% (that is, Asp, myo-Ins, Tau, Glu 2+3 , Glu 4 , Gln, GABA, tCho and tNAA) as assessed on the data measured in the 1 H-Glc sessions. Glu 4 was consistently quantified with c.v.s below 2% in both GM and WM.
Finally, single-participant MRSI time-courses obtained with high time resolution were fitted with linear regression per voxel ( Fig. 7a,b). The time-course slopes (P < 0.05, r < −0.8) were steeper in the GM than in the WM by 19% for 1 H-MRSI and yielded a clear distinction between GM and WM on the slope map (Fig. 7a). The smaller WM/GM difference of 6% for 2 H-MRSI data can be ascribed to a greater partial volume effect than in 1 H-MRSI obtained with substantially higher spatial resolution.

Discussion
We have shown the potential of deuterium labelling to measure the turnover of metabolites involved in oxidative glucose metabolism in the human brain. Impaired glucose homoeostasis and mitochondrial dysfunction are key components in the pathophysiology of neurodegenerative diseases such as Alzheimer's and other forms of dementia, and also in metabolic disorders including obesity and insulin resistance 25 , moving the imaging of brain glucose metabolism to centre stage 26 . The development of non-invasive methods that enable objective, dynamic and longitudinal metabolic tracking in health, disease and aging are needed to track drug-treatment effects and to investigate emerging therapies.
In this work, we showed dynamic downstream Glu mapping after 2 H-Glc administration in humans using a multi-voxel 1 H-MRSI sequence, H-Glc ingestion. The spectra in a and b were linewidth-matched with exponential line-broadening and subtracted. The resulting difference spectrum in c represents the effect of metabolite 2 H enrichment. The metabolite components (d) were obtained via LCModel analysis using a basis set containing simulated spectra of neurochemicals that were undergoing deuteration (that is, Glu, Gln, GABA and Glc). The proton signals that originated from different carbon groups were separated. The summed 2 H-MRS spectrum from the last acquired time-point resembles the 1 H-MR spectrum except for the spectral resolution (e).
Article https://doi.org/10.1038/s41551-023-01035-z which was corroborated by the well-established single-voxel functional MRS methodology 20,27 , as well as direct measurements of deuterated compounds with 2 H-MRSI (DMI). Our work benefited from the combination of ultra-high-field (7 tesla (T)) and high natural abundance of protons in the human body. The 1 H-based QELT-MRS(I) approach used commercially available radio-frequency coils and optimized 1 H-MRS approaches to overcome the major drawbacks of direct techniques tuned to the deuterium frequency ( 2 H-MRS) that require dedicated coils and pulse sequences 28 . Compared with 2 H-MRS, the indirect QELT approach also enabled the detection of GABA (despite a non-significant GABA 2 signal drop) and the separation of Glu from Gln, which is not feasible using 2 H-MRS even at ultra-high MR fields (16.4 T) due to the limited spectral resolution of 2 H-spectra 29 . The orally administered deuterated Glc is safe, stable and affordable compared with radioactive PET tracers with short half-lives, which require technically challenging onsite preparation and expensive cyclotrons. We have also shown the ability of QELT-MRS to overcome challenges of 13 C-MRS, which was previously the only method for quantitative assessment of TCA cycle kinetics and Glu and Gln cycling. The modified basis sets with separated Glu and Gln components on the C4 position enabled the extraction of quantitative information about Glu and Gln turnover in a manner similar to that of the technically challenging 13 C-MRS. Although GABA changes have been tracked in individual participant data in a preclinical study at 9.4 T 4 , but not in our data, our results depicted signal changes in GABA 2 using difference spectra calculated from group-averaged data, with acceptable quantification errors   Article https://doi.org/10.1038/s41551-023-01035-z previously used to unravel metabolite consequences of chromatic and achromatic stimulation 27 , hence analysis of pooled spectral differences can be analogically utilized to compare cohorts of controls and patients. We expect a further boost to the quantification of J-coupled, relatively low-abundance metabolites such as GABA and Gln by optimization of the sequences for higher fields (that is, above 7 T), possibly enabling robust GABA detection even for low-field MR scanners 30 .
The deuteration of molecules (that is, replacing 1 H with 2 H nuclei) is a simple chemical procedure that allows the labelling of a broad range of substances with minimal influence on their in vivo kinetics within metabolite cycles 31 . Other deuterated tracers, such as deuterated choline, could be indirectly detected with 1 H-MRS in brain tumours 32 . Other potential candidates are deuterated ketone bodies, such as beta-hydroxybutyrate 33,34 , which serve as alternative brain fuels especially during fasting, with critical implications in AD 35 , some forms of epilepsy 36 and brain tumours. Acetate is preferentially utilized by astrocytes and could thus be used to probe energy metabolism in astrocytes 37 . While we did not detect any lactate changes due to minimal activity of anaerobic glycolysis in the healthy brain, 1 H-MRS is highly sensitive to lactate 20 . Indirect QELT-MRS(I) can thus be used to measure the Warburg effect, a shift toward anaerobic glycolysis due to mitochondrial failure in ischaemia and brain tumours, as shown in animal glioma models 4 . Increased lactate is also a hallmark of aging-related mitochondrial dysfunction 38 , and it is conceivable that lactate production accompanies deteriorated Glu production via the TCA cycle in neurodegeneration 39   Our study involves several methodological advancements with respect to the initial preclinical animal QELT at 9.4 T, these advancements being critical for future applications in humans. We achieved substantial signal gain due to the use of an echo-less 1 H-MRSI sequence with concentric ring trajectory readout, as well as high spatial and temporal resolution 3D imaging at 7 T 19,40 . Preliminary 7 T QELT-MRS(I) results in humans 41 that were obtained using a standard vendor-provided single-voxel PRESS MRS sequence suffer from poor spatial selection and echo-time (TE)/J-evolution related signal losses at 7 T 42 , along with a slow phase-encoded chemical shift imaging sequence with a 4 × 4 cm box within a single slice subdivided into 4 × 4 voxels of 1 cm³ volume in the centre of the brain in ~4 min per time frame. In contrast, the semi-LASER single-voxel sequence used in the current study provides accurate quantification of extended neurochemical profile even at 7 T 43 ; and our multi-voxel 1 H-MRSI sequence achieved 8-fold higher spatial resolution (that is, 0.12 cm³) in 3 min per time frame and almost 10,000-fold more voxels over a 3D brain volume using a fast spatial-spectral encoded echo-less free-induction-decay (FID)-MRSI alternative, which is recommended for 7 T 43 . FID-MRSI offered several-fold increased signal amplitudes for our target J-coupled metabolites compared with the TE of 28 ms. Also, our spatial resolution was 67-fold higher than in previous direct 2 H-MRSI studies at 4 T (0.12 ml vs 8 ml) 28 and 16-fold higher compared with our 2 H-MRSI scans at 7 T (0.12 ml vs 2 ml). Finally, our whole-brain 2 H-MRSI was obtained with a spatial resolution similar to a recent human study at 9.4 T (2.0 ml vs 3.0 ml) 3 . The differences in obtained spatial and temporal resolution can be explained by the different sensitivities of 2 H and 1 H detection, with SNR estimated and experimentally validated to be at least ~5-fold higher for 1 H 4 , when considering the ~6.5-fold higher gyromagnetic ratio, half the nuclear spin and 10-fold slower t 1 relaxation, neglecting differences in coil hardware (for example, ~3-fold higher for array coils). However, a realistic comparison must also consider the fact that 2 H-MRS detects only the signal of interest (that is, 2 H-labelled compounds), whereas 1 H-MRS also detects signals from >80% of non-labelled substances (which are not of primary interest).
The nearly 20% difference in the glutamate turnover in GM and WM in this study is lower compared with the current gold standard ([ 18 F]FDG-PET studies), which measured the differences in the CMRGlc (oxidative) at 0.18 and 0.24 µmol g −1 min −1 (refs. 44,45), that is, ~33% higher oxidative Glc consumption ascribed to the higher inhibitory and excitatory demands of the grey matter 46 . A 13 C study estimated the rate of the TCA cycle (V TCA ) at 0.88 ± 0.12 in the grey and 0.28 ± 0.13 in the white matter, that is, 68% difference between GM and WM 47 , and the outcomes were corroborated by other studies reviewed in ref. 14. The Glu 4 decays in our study do not directly reflect V TCA due to the compartmental nature of Glu 4 metabolism. Labelled Glu 4 is released to the synaptic cleft and taken up by glia (astrocytes), where Glu is converted to Gln and returned to neurons 48 . Thus, the Glu 4 decaying constants reflect both the V TCA and Glu/Gln cycle. The enrichment of the glial glutamate pool with neuronal Glu might explain the lower differences in Glu 4 turnover between GM and WM. While we observed differences in Glu 4 slopes between GM and WM, the between-participant differences (last minus first time-point) were similar in the PCC (ΔGlu 4 = 15% ± 3%, 80% GM/WM fraction, SV-MRS), GM (ΔGlu 4 = 13% ± 4%, MRSI) and WM (ΔGlu 4 = 14% ± 3%, MRSI). This indicates that similar relative differences between the first point and steady state are achieved at different speeds. The within-session differences are in agreement with a previous 13 C study, which reported peak fractional enrichment of Glu 4 of 16% ± 2% at 130 min after oral administration of 13 C-Glc 49 . Thus, the fractional enrichment in our data is very similar considering the shorter acquisition window of 98 ± 7 min on average, the 0.8 g kg −1 dose of 2 H-Glc vs the lower dose of 13 C-Glc (0.65 g kg −1 ) in ref. 49 and similar between-participant variance. The robust detection of within-session differences in Glu 4 was enabled by excellent c.v.s of 2% for both 1 H-MRS and 1 H-MRSI, which were well below the detected differences. While the between-participant variance in the concentration differences between the first point and the last point was 19% (PCC: SV-MRS), 26% (GM: MRSI) and 21% (WM: MRSI), the between-participant variance in Glu rates appeared similar for MRS (~29%) and higher for MRSI (50% for GM and 45% for WM). Thus, MRS fitting yielded a similar variance as the gold standard ([ 18 F]FDG-PET) measures of CMRGlc, reported in the range of 19%-29% (mean - median) between participants 16,50 . This can be mainly ascribed to high physiological variation in resting brain metabolism. The same applies to the between-session variance, which was 14% ± 8% for [ 18 F]FDG-PET and 11% ± 9% for Glu 4 in single-voxel 1 H-MRS data 51 . The current work is a preliminary study, and further technical improvements are expected to yield higher reproducibility and time/spatial resolution by integrating techniques for instability correction 52 , advanced B 0 shimming or k,t-undersampling 53 . Despite these methodological limitations, we clearly proved that observed metabolite changes are related to deuterium enrichment and the measured signal time-courses are suitable for quantitative modelling of metabolite kinetics. The repeated session after 1 H-Glc ingestion demonstrated that metabolite changes were not related to instability in the spectral quality or the metabolic effects of hyperglycaemia.
Increasing concentration of 2 H-Glc in the brain, reproducibly measured with both QELT and DMI, is probably caused by continuous substrate supply during mild hyperglycaemia, which resembles the glucose excursion of an oral glucose tolerance test. Our 2 H-Glc and Glc 6 brain time-courses differ from the time-courses measured previously with DMI due to different routes of administration of the 2 H-Glc, that is, oral vs intravenous bolus 29 . The negligible blood Glc enrichment with deuterium does not contribute to the 2 H-Glc delivery and transport to the brain, hence brain 2 H-Glc levels reflect only the conversion of the 2 H-Glc to the downstream metabolites. Thus, directly measured brain 2 H-Glc decreased shortly after initiation of the scans in the previous intravenous experiment, whereas indirectly and directly measured 2 H-Glc increased throughout our scans. The same effects of both intravenous and oral administration of 1-13 C-Glc on isotopic brain Glc levels were reported 49 . Despite MR methods being unable to distinguish compartmental origins of Glc signals, dominant contribution comes from the extracellular compartment since there is roughly three orders of magnitude lower concentration intracellularly due to rapid Glc metabolism within brain cells 54 . Even though the calculation of metabolic fluxes is rather complex due to the non-steady state of blood 2 H-Glc levels, this can be solved in future studies once the blood 2 H-Glc enrichment is known. Another study utilized 13 C-Glu and 13 C-Gln time-courses measured after oral administration of 13 C-Glc to calculate rates of TCA and Gln synthesis (V TCA and V gln ) 24 . Thus, future QELT studies will benefit from blood sampling to measure blood glucose deuterium enrichment as a function of time and quantitative compartmental modelling approaches 55 .
The practicality of non-invasive peroral tracer administration in our study, which is more comfortable for patients, highlights the clinical feasibility of our approach 24 . Despite the delay in blood enrichment with the labelled substrate, our between-participant variance in Glu 4 decay constants of 29% corroborates previous work in which oral administration of 13 C-Glc provided similar results compared to an intravenous clamp, albeit with lower precision 24 in estimating V TCA with between-participant c.v.s (n = 4) of 62% and 23% for oral and intravenous administration, respectively. The somatostatin-free protocol could contribute to a higher variance of the modelling outcomes compared with 2 H-Glc works with somatostatin infusion (frequently used in pancreatic clamps) 24 that blocks endogenous insulin/glucagon release. The use of clamp protocols with continuous tracer application in future studies could thus further improve reproducibility and/or shorten the acquisition window due to longer periods near the 'steady-state' 2 H-Glc Article https://doi.org/10.1038/s41551-023-01035-z blood enrichment and might, along with higher fields above 7 T, further improve detection of Gln 4 and GABA 2 .
While the measurements of deuterated Glc and Glx with DMI in an animal experiment allowed calculation of CMRGlc and V TCA 29 , QELT also concomitantly quantifies Glc and separates Glu and Gln. Thus, QELT can estimate Glu/Gln cycling in combination with measurements of Glc turnover, which is challenging with 13 C methods due to overlapping signals of Glc with residual water signal and high chemical shift between Glc and Glu/Gln 29 . A recent 13 C study demonstrated substantial improvement of reproducibility of Gln time-courses at individual-participant level by prolonged acquisitions of 3 h that minimize scatter of delayed Gln 4 increase 24 , while others suggested 'snapshot' acquisition during 30-60 and 120-150 min with a 1 h break 56 .
While 7 T MR scanners have been recently approved for clinical use, they are still not as widely available as lower-field 3 T scanners 18,57 . Advancements in 3 T SV-MRS that detected subtle 3% Glu and ~28% lactate responses to visual stimulation, 58 and in 1 H-MRSI that allowed whole-brain Glu mapping within 4 min 57 promise implementation of the indirect QELT-MRS(I) at 3 T despite the lower spectral and reduced spatial resolution.
Non-invasive quantification and imaging of metabolite changes relevant to Glc metabolism and neurotransmitter synthesis (Glu/Gln, GABA) are crucial for the understanding of many brain disorders.
Here we have shown the capabilities of an affordable QELT-MRS(I) approach for metabolic measurements in humans that uses widely available 1 H-MR scanner equipment and time-resolved whole-brain mapping, and that allows indirect detection of deuterated compounds also in a clinical setting. The negligible risk associated with deuterium administration compared with radioactive [ 18 F] FDG predetermines the methodology, especially for studies with a multi-session longitudinal design to track treatment effects or disease progression over time. In contrast to direct 2 H-MRS, QELT-MRS allows for the quantification of an extended neurochemical profile, including non-deuterated compounds. Thus, the current methodology may represent a step forward for future metabolic projects in the resting or activated human brain in disease, health and aging, and may offer relevant metabolic information using widely available hardware in a single MR session.

Study design
Five healthy right-handed volunteers (30 ± 4 yr old, 4 males and 1 female) were scanned on a 7 T whole-body MR scanner (Siemens Healthcare) utilizing a commercially available 32-channel receive-array coil (Nova Medical) for 1 H-MRS and a dual-tuned ( 2 H/ 1 H) quadrature transmit-receive birdcage coil specifically developed for the project (Stark Contrast MRI Coils Research) for 2 H-MRSI acquisitions. All participants were lean (body mass index: 22.6 ± 1.4 kg m −2 ) without a history of diabetes or other metabolic and severe diseases. Blood glucose measurements were performed with a standard strip glucometer in triplicates after capillary blood sampling from the leg of the volunteers. The study was approved by the Ethics Committee at the Medical University of Vienna. All participants signed informed consent. Each participant underwent four MR scans in total: The first session ('test' session) used an interleaved single-voxel 1 H-MRS/ 1 H-MRSI (QELT) protocol performed after oral administration of 2 H-Glc. The second session ('retest' session) was performed identically to the first session to assess test-retest repeatability.
The fourth session ('reproducibility' session) acquired direct 2 H-MRSI dynamically using the dual-tuned head coil following oral administration of 2 H-Glc to directly compare the results with those obtained in the first session (that is, QELT).
One participant was additionally scanned after 2 H-Glc administration with a 1 H-MRSI-only protocol to obtain the time-course with high temporal resolution. All sessions were conducted in the morning after an overnight fast. Both compounds ( 2 H-Glc and 1 H-Glc) were dissolved in ~300 ml of water and ingested in equal amounts (0.8 g kg −1 body weight) immediately before the scan was initiated. MRSI and MRS data were interleaved with navigator images obtained before each 1 H-MRSI/ 1 H-MRS or 2 H-MRSI block to ensure stable position of the localized volume. The navigator images were registered to the atlas space with Autoalign methodology implemented on the MR scanner 59 . The coordinates of the MRS volumes of interest (MRS-VOI) were later used in the second MR session for consistent MRS-VOI placement. The first MRS/MRSI block was acquired after calibrating MRS radio-frequency pulses and B 0 shimming within 30 min after 2 H-Glc/ 1 H-Glc administration. The following MRS/MRSI blocks were obtained after the acquisition of the T 1 -weighted MP2RAGE image with an isotropic resolution of 1.1 × 1.1 × 1.1 mm 3 : repetition time (TR), 3,900 ms; TE, 2.8 ms; flip angle, 4°/5°; and acquisition time, 3:52 min. Our goal was to cover a time window of 120 min following the ingestion of the tracer, assuming slower dynamics of deuterium enrichment in the human brain compared with that of rats 4 .

MRI and 1 H/ 2 H MRS acquisitions
Standard second-order B 0 shimming was performed via an imaging-based approach (two iterations) for multi-voxel (MRSI) and FASTMAP 60 for single-voxel (SV-MRS) acquisition. MRSI and SV-MRS acquisitions were interleaved except for one study session, where only the MRSI data were acquired with a high time resolution of 5 min after 2 H-Glc ingestion. 1 H-MRSI data were obtained via an FID-MRSI sequence 19 with an ultra-short acquisition delay of 1.3 ms, a short TR of 320 ms, ellipsoidal 3D k-space encoding using concentric ring trajectories, variable temporal interleaves, 36 × 36 × 26 matrix, 5 × 5 × 4.8 mm 3 voxel size, 2 min 58 s per block, 558 complex points, 34° (Ernst) excitation flip angle, 600 µs pulse duration and 7 kHz pulse bandwidth. The excited 40-mm-thick slab was centred around the posterior cingulate region. 2 H-MRSI data were acquired using a phase-encoded non-selective FID-MRSI sequence with an acquisition delay of 1.5 ms after a 500 µs block excitation pulse adjusted to an 86° (Ernst) flip angle and a TR of 290 ms. An ellipsoidal k-space with a matrix size of 16 × 16 × 14 was acquired with two weighted averages (and retrospective Hamming filtering), with 12.5 × 12.5 × 12.5 mm³ voxel size, a vector size of 256 and a spectral bandwidth of 500 Hz at 6 min 37 s per block.
Single-voxel MRS data were obtained from the PCC region using a semi-LASER sequence (TR 7 s, TE 28 ms, 3 min 43 s per block, 2,048 complex points, 90° asymmetric sinc pulse with a duration of 2.5 ms, FOCI pulse bandwidth and a duration of 4.2 ms) 61 . A 22 × 20 × 20 mm (anterior-posterior × left-right × superior-inferior) voxel was placed mid-sagittally on the basis of anatomical landmarks. The voxel was rotated in the sagittal plane by 30° such that it was aligned with the posterior border of the splenium. To mitigate possible effects of patient motion and chemical shift displacement, the voxel was backed away anteriorly from the splenium and caudally from the occipital-parietal fissure by 2 mm 62 . All spectra were collected with water suppression 63 and outer volume suppression (number of excitations, NEX = 32), along with unsuppressed water spectra used to remove residual eddy currents (NEX = 2) and as a reference from which to derive metabolite concentration estimates (NEX = 2).

Segmentation of MPRAGE scans
The T1w-MRI images were segmented in Freesurfer (v5.3) to obtain masks of the brain GM and WM. The masks were resampled to the MRSI space and used to obtain tissue-specific averages of metabolite levels. In addition, probabilistic maps of the GM, WM and cerebrospinal fluid Article https://doi.org/10.1038/s41551-023-01035-z (CSF) were derived by segmenting the T1w-MRI images using the SPM12 software package. The probabilistic tissue maps were thresholded with an in-house-written MATLAB script using the iterative method of threshold selection 64 to determine the within-PCC-VOI fraction of GM, WM and CSF.

Processing of MRSI
MRSI data were reconstructed offline with an in-house-developed software pipeline consisting of MATLAB (R2013a, MathWorks), BASH (v4.2.25, Free Software Foundation) and MINC (MINC tools v2.0, McConnell Brain Imaging Center). Data processing included an iMUSICAL coil combination 65,66 , water normalization 67 , k-space reconstruction with in-plane convolution gridding, spatial Hamming filtering, channel-wise noise decorrelation and off-resonance correction. 40 Due to the short TR, obtaining metabolite concentration estimates would require strong assumptions about t 1 relaxation. Therefore, the metabolite concentrations were instead quantified in institutional units using LCModel (v6.3-1, LCModel) with a basis set that included 17 simulated brain metabolites and a measured macromolecular background (detailed below) 68 . Data processing of 2 H-MRSI data was simpler, involving only spatial Hamming filtering before Fourier transform.

Processing of SV-MRS
Single shots were corrected for small frequency and phase fluctuations, and residual eddy currents and 32 single shots were summed per block. The frequency and phase of the spectra obtained for each block were aligned to each other within the scanning session and quantified in LCModel. The metabolite concentrations were corrected for variable water content in the grey and white matter, as well as for the within-voxel CSF fraction 69 .
Finally, the spectra measured in session 1 representing the first time-points (FIRST) and the last time-points (LAST) were pooled together and summed, resulting in two sums (FIRST and LAST). The FIRST and LAST from the 2 H-Glc session were subtracted, and a difference spectrum was calculated. The difference spectrum characterized metabolite changes following 2 H-Glc ingestion. The difference spectra were quantified in LCModel using the basis sets that contained only the metabolites enriched with deuterium (detailed below) to estimate quantification errors, that is, relative CRLB 27 .

Quantification of MR spectra
Basis sets previously routinely used to quantify 1 H-MRS/MRSI data were modified to reflect the fact that deuterium is incorporated at certain carbon positions in the molecules. Modifications to the basis sets were performed on the basis of previous animal experiments, theoretical predictions and preliminary analysis of the current data. Glu, Gln and GABA included the split of proton signals that originated from different carbon positions for C2, C3 and C4. Thus, the basis set included all molecular variants that occurred in the brain in a dynamically changing ratio and took into account the homonuclear ( 1 H-1 H) and heteronuclear ( 1 H-2 H) coupling constants. For instance, the basis set of glutamate included six components, that is, those present in the molecule with all positions occupied with protons (C2 1 H, C3 1 H 2 , C4 1 H 2 ), in the molecule with one deuteron on the C4 (C2 1 H, C3 1 H 2 , C4 1 H 2 H) and two deuterons on C4 (C2 1 H, C3 1 H 2 ). Thus, we distinguished three variants of the C3 resonance affected by homo-and heteronuclear coupling with 1 H and 2 H at C4. While the couplings between protons and deuterons on the C3 and C4 (Glu and Gln) and C2 and C3 (GABA) were simulated, the couplings between C2 and C4 were minimal and were neglected. Yet, the number of components had to be reduced for MRS and MRSI data to preserve the stability of the fits. This was performed by neglecting the heteronuclear and homonuclear coupling effects of the deuteration, assuming that the dominant signal change occurred on the position where proton(s) is/are replaced by deuteron(s), that is, C4 for Glu and Gln, and C2 for GABA. Thus, we used only two components per metabolite (Glu 2+3 , Glu 4 ; Gln 2+3 , Gln 4 ; and GABA 2 , GABA 3+4 ) to quantify SV-MRS data. The basis set was further simplified for MRSI, where we split only Glu 2+3 and Glu 4 resonances since the fitting with more components yielded a less stable quantification of neurochemical profiles. Finally, we used the components whose signals were expected to change due to progressive deuteration during the scans (that is, for Glu and Gln: C4 1 H 2 , C4 1 H 2 H and two variants of C3 1 H 2 present in the molecule with one and two deuterons; and for GABA: C2 1 H 2 , C2 1 H 2 H and two variants of C3 1 H 2 ) to quantify the difference spectra. As difference spectra do not contain the background of signals from the metabolites that were stable over the task, we could account for the subtle hetero-and homonuclear coupling effects. The

Quality control of MRS(I) data
Voxels with sufficient temporal stability and spectral quality were selected using masks. The masks included voxels with c.v.s below 12% for the three main metabolites that remained stable during the scan: tCr, tCho and tNAA. The criteria also used parameters provided by LCModel (full-width-at-half-maximum <0.1 ppm, SNR > 5, zero-order phase <40°) in line with expert recommendations 70 . These criteria were used to calculate regional GM and WM means in metabolite concentrations, as well as to select the spectra used for the calculation of high-SNR sums that represented either GM or WM. The metabolite concentrations quantified with all CRLBs were used for further analysis except those that could not be quantified, with CRLBs of 999%. This criterion avoided bias due to an arbitrarily set CRLB threshold, cutting off lower concentrations with higher relative CRLB 71 . The metabolites quantified with a CRLB of 999% in most of the time-points (MRS), or consistently in more than 10% of voxels (MRSI), were not analysed 72 .

Statistical analysis
Differences in metabolite concentrations were calculated between the first and last time-point, and their significance was tested with the standard paired t-test separately within both sessions ( 2 H-Glc, 1

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
The main data supporting the results in this study are available within the paper and its Supplementary Information. Source data for Figs. 1, 4 and 6 and for Supplementary Figs. 1-3 and 5 are provided with this paper. The raw data acquired in the study are too large to be publicly shared, yet they are available for research purposes from the corresponding authors on reasonable request. The data generated by post-processing methods (that is, metabolite maps, MR spectra and outcomes of their quantification in the LCModel) are available at https://doi.org/10.5281/ zenodo.5705959. The shared data are in the minc, niifti and MRSpa data formats. A priori information ('the basis sets') needed for MRS/MRSI data quantification in the LCModel is also available via the same link.

Code availability
The custom code for the time-course analysis using linear and exponential fits was performed using custom-made Python code Last updated by author(s): Oct 14, 2022 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend,

March 2021
Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy The main data supporting the results in this study are available within the paper and its Supplementary

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
Five healthy volunteers were enrolled. This was an exploratory study. Hence, no sample-size calculation was performed. The minimum number of enrolled participants was defined on the basis of previous studies of similar focus.
Data exclusions No data were excluded.

Replication
The experimental findings (dynamic signal drops in deuterium-labelled downstream metabolites of glucose) were replicated in 5 healthy volunteers using dynamic 3D MRS imaging as well as with a conceptually different single-voxel MRS approach, and were also in agreement with direct 2H-MRSI experiments in the same healthy volunteers.
Randomization The order of the actual study experiment (1H-MRS and 1H-MRSI following peroral administration of deuterium-labelled glucose) and that of the control experiment (the same experiment with non-labelled glucose) was pseudo-randomized. The additional retest-scans (to assess testretest repeatability for 1H-MRSI and single-voxel 1H-MRS) were performed afterwards without randomization; thus, three participants had deuterium first, whereas the other two volunteers had the glucose session first.

Blinding
The person processing the MRS(I) data was blinded to the acquisition condition (deuterated or non-deuterated glucose). The researchers responsible for the final analysis were not blinded. The researchers that performed the MRI data acquisition were also not blinded, because they had to prepare the appropriate amount of glucose solution (body-weight matched). However, because the entire process was highly automated, this knowledge could not affect the results and conclusions.
Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Recruitment
Lean and non-diabetic participants were selected randomly from the pool of volunteers ar the High Field MR Center to avoid biasing of brain responses by metabolic disorder (impaired glucose homeostasis, insulin resistance). There is no potential bias related to recruitment.

Ethics oversight
Ethical commission of the Medical University of Vienna. All participants signed informed consent.
Note that full information on the approval of the study protocol must also be provided in the manuscript.
Magnetic resonance imaging Field strength 7 Tesla Sequence & imaging parameters 1H-MRSI data were obtained via an FID-MRSI sequence with an ultra-short acquisition delay of 1.3 ms, a short TR of 320 ms, and ellipsoidal 3D k-space encoding using concentric ring trajectories (CRT), variable temporal interleaves, 36x36x26 matrix, 5 x 5 x 4.8 mm3 voxel size, 2:58 min per block, 558 complex points, 34° (Ernst) excitation flip angle, 600-μs pulse duration, and 7-kHz pulse bandwidth. The excited slab was centred around the posterior cingulate region.
Single-voxel 1H-MRS data were obtained from the posterior cingulate (PCC) region using a semi-LASER sequence (TR 7s, TE 28 ms, 3:43 min per block, 2048 complex points, 90° asymmetric sinc pulse with a duration of 2.5 ms, FOCI pulse bandwidth, and a duration of 4.2 ms, 45 kHz). A 22 × 20 × 20mm (AP×LR×SI) voxel was placed mid-sagittally, on the basis of anatomical landmarks. The voxel was rotated in the sagittal plane by 30° such that it was aligned with the posterior border of the splenium. To mitigate possible effects of patient motion and chemical-shift displacement, the voxel was backed away anteriorly from the splenium and caudally from the occipital-parietal fissure by 2 mm. All spectra were collected with water suppression and outer-volume suppression along with unsuppressed water spectra.
2H-MRSI data were acquired using a phase-encoded non-selective FID-MRSI sequence with an acquisition delay of 1.5 ms after a 500-μs block excitation pulse adjusted to a 86° (Ernst) flip angle and a TR of 290 ms. An ellipsoidal k-space with a matrix size of 16 × 16 × 14 was acquired with two weighted averages (and retrospective Hamming filtering), with 12.5 × 12.5 × 12.5 mm³ voxel size, a vector size of 256, and a spectral bandwidth of 500 Hz in 6:37 min per block.

Area of acquisition
For 1H-MRSI, the excited slab was centred around the posterior cingulate region, covering approximately the upper half of the brain.
For single-voxel 1H-MRS, volumes of interest were placed in the posterior cingulate (PCC) region.
For 2H-MRSI, no slab selection was possible, and was only limited by the sensitive volume of the used birdcage coil, which restricted the investigated volume to the entire cerebrum.
Diffusion MRI Used Not used Preprocessing Preprocessing software T1w images preprocessed with Freesurfer. MRS data were processed in MRSpa. MRSI data were processed with in-house developed MATLAB and Python scripts, which used freely available software (MINC, FSL, SPM).

Normalization
Data were not normalized. For single-voxel MRS, this is not possible, but a standardized positioning of the voxel was ensured. For MRSI, automated GM/WM segmentation of the entire MRSI slab using established software (Freesurfer) was performed, and positioning was standardized via Autoalign software to ensure the same positioning for all participants. Bringing all metabolic maps to the same model space would not have any added value.