Introduction

Neuroimaging techniques are essential to identifying metabolic biomarkers of brain function and disease, paving the way for personalised approaches to neuropsychiatric and neurodegenerative disorders¹. While fluorodeoxyglucose positron emission tomography (FDG-PET) remains the gold standard for clinical assessment of cerebral glucose metabolism^2,3, its focus on glucose uptake provides limited insight into downstream metabolic pathways that directly support neuronal signalling, including the glutamate-GABA cycle that underlies excitatory-inhibitory balance. Here, we introduce proton-observed proton-edited carbon-13 magnetic resonance spectroscopy (POPE¹³C-MRS), the first clinically compatible method enabling simultaneous indirect detection of glutamatergic and GABAergic ¹³C-labelling using standard ¹H hardware. POPE¹³C-MRS uses targeted proton editing on standard proton (¹H) MRI hardware to indirectly measure neurotransmitter-specific glucose metabolism by detecting ¹H-¹³C satellite resonances, thereby enabling ¹³C labelling readout without broadband ¹³C excitation or the associated radiofrequency (RF) power deposition.

X-nuclear Magnetic Resonance Spectroscopy, which uses MRI-visible isotopes such as carbon-13 (¹³C) or deuterium (²H) to track labelled metabolic substrates, offers a powerful means to study neurometabolic pathways in vivo^4,5. In particular, ¹³C-MRS following administration of ¹³C-labelled glucose allows real-time measurement of neurotransmission and energy metabolism via the tracking of label incorporation into the downstream metabolites^6,7. However, despite this unique metabolic specificity, the implementation of ¹³C-MRS in human studies has been limited by three major technical challenges^8–12: its inherently low sensitivity (due to the lower gyromagnetic ratio of ¹³C compared to ¹H), the requirement for high-power broadband radiofrequency (RF) excitation, and the need for specialised and costly RF hardware¹³.

Several strategies have been developed to address these limitations. Indirect ¹³C-MRS, such as proton-observed carbon-edited (POCE) ¹³C-MRS, improve sensitivity by detecting the protons attached to ¹³C-labeled metabolites, utilising the higher intrinsic sensitivity of protons^9,14,15. However, these approaches rely on broadband ¹³C editing and decoupling, which substantially increases RF power deposition and SAR limiting their feasibility in human studies. Hyperpolarisation techniques provide dramatic gains in sensitivity by increasing nuclear spin polarisation far beyond thermal equilibrium levels^16,17, enabling real-time metabolic imaging with unprecedented signal enhancement¹⁸. Yet this approach requires specialised equipment and infrastructure and is restricted to observing rapid metabolic processes due to the short (tens of seconds) lifetime of the hyperpolarised signal. Deuterium MRS (²H-MRS) has recently emerged as another promising strategy, enabling dynamic metabolic imaging (DMI) by exploiting the short longitudinal relaxation time of deuterium nuclei, allowing for rapid data sampling and higher signal-to-noise ratio¹⁹. However, it requires dedicated hardware and cannot resolve key neurotransmitter signals, as it does not distinguish glutamate from glutamine or detect GABA, metabolites that play central roles in excitatory and inhibitory neurotransmission through the glutamate-glutamine cycle between neurons and astrocytes^20,21. More recently, indirect ‘dynamic’ labelling methods^22,23, such as quantitative exchanged-label turnover (QUELT-MRS)²⁴, have been proposed. These infer metabolite labelling from the decay of ¹H resonances, when replaced by ²H or ¹³C, but rely on baseline scans and assumptions of steady-state metabolite pools following the labelled substrate administration, conditions that may not always hold during metabolic perturbations, such as those likely present in clinical populations.

Despite these advances, a straightforward and clinically deployable method for detecting ¹³C-labelling of glutamate and GABA in the human brain has remained elusive. Here, we develop and validate POPE¹³C-MRS, a method that tracks the incorporation of ¹³C from a labelled substrate (e.g. uniformly labelled [U-¹³C₆]-glucose, which contains 6 ¹³C atoms) into downstream metabolites via measurable changes in proton MRS spectra, enabling non-invasive probing of excitatory, inhibitory, and glycolytic pathways in the brain (Fig.1). POPE¹³C-MRS uses a proton-edited sequence which is tailored to detect ¹H signals from lactate H3 (LacH3), glutamate H4 (GluH4), GABA H4 and glutamate+glutamine H2 (GlxH2). Signals from these protons are only visible when they are bound to ¹²C, not ¹³C, meaning that these signals decrease as protons initially bound to ¹²C become coupled to ¹³C, as the [U-¹³C₆]-glucose is metabolised, with the transferred signal appearing as characteristic ¹³C satellite resonances for selected metabolites.

Figure 1:

We first establish and validate the method in pre-clinical models and then demonstrate its feasibility in the human brain. By enabling simultaneous and targeted measurement of excitatory and inhibitory metabolic pathways, POPE¹³C-MRS provides a method for investigating neurometabolic coupling in vivo, bridging a key gap between glucose uptake measurements and neurotransmitter-specific metabolism.

Results

We established the theoretical foundation and optimised POPE¹³C-MRS in mice (Study 1), validated neurotransmitter labelling detection (Study 2), and assessed feasibility in the human brain (Study 3).

Study 1: POPE¹³C-MRS general optimization

To test whether indirect detection of ¹³C-labelling with sensitivity to GABA and lactate could be achieved using proton-edited MRS, we administered [U-¹³C₆]-glucose in mice while continuously acquiring MEGA-sLASER spectra. The results clearly demonstrate POPE¹³C-MRS can reliably detect the progressive incorporation of ¹³C into neurotransmitter pools through consistent changes in ¹H-¹²C signal attenuation and emergence of ¹H-¹³C satellite resonances (Fig.2). Because only a single echo time (TE) can be used for each edited sequence, we needed to optimize our TE for a single labelled resonance per spectrum. In the GABA-edited experiment, we selected GlxH2 as the optimization target because its ¹³C satellites were (i) most clearly resolved, (ii) minimally confounded by spectral overlap, and (iii) suitable as an internal reference for normalizing metabolite labelling. We then empirically determined the corresponding heteronuclear J-coupling constants (J_CH, the distance between the two satellite peaks) by sampling TE values around the expected theoretical optimum (with TE=N.1/2J, with N the number of J-revolutions). The J_CH constants are given by the distance between the two resonances of the doublet (in Hz), which was measured using a range of TEs, starting around established optimal values for GABA²⁵ (68ms) and lactate (100ms, tested in phantom) (Supplementary Fig.2). We determined that a TE=69.5 ms (J_CH(GlxH2)=143.9±2.7 Hz) for GABA-edited and TE=106 ms (J_CH(LacH3)=122.7±1.0 Hz) for lactate-edited POPE¹³C-MRS should maximize the satellite detection. The J of GlxH2 was found to be slightly more variable, most likely because it is a combination of both J_CH(GluH2) and J_CH(GlnH2), which may not be identical, thus adding variability. Notably, the agreement between signal attenuation (GlxH2 and LacH3) and satellite appearance supports the internal consistency of the POPE¹³C-MRS readout (Fig.2c).

Figure 2:

Study 2: POPE¹³C-MRS detects ¹³C labelling of GABA, glutamate, glutamine and lactate after¹³C-glucose administration in mice

We then tested this protocol in Study 2 to assess the detectability and quantifiability of the satellites after a one-hour ¹³C-glucose infusion. Our data indicated that the ¹³C satellites of GlxH2 (¹³C-GlxH2) was consistently detectable across animals (Fig.3a) and could be quantified despite low labelling concentration (CRLB(GlxH2_tf)=76%±27%, mean±s.d.; supplementary Fig.3). Measuring satellite signals are useful to control for change of total metabolite concentration throughout the infusion, a process that cannot be assessed by solely inferring the metabolite labelling from the drop of the initial ¹H-¹²C signal. This was particularly relevant for Glx and lactate as we observed a 22±14% increase in total Glx and 13±5% in total lactate concentrations following one-hour infusion of ¹³C-Glc compared to the baseline scan in mice. Notably, the fitting of GluH4 and GlnH4 led to some residual signal likely due to GABAH2 resonance that could not be fitted and was thus not included in the basis set. Nonetheless, their CRLB values were in an acceptable range (GluH4_t0 = 9%±4% and GlnH4_t0 = 24%±33%). Interestingly, no lactate ¹³C-labelling was detected during the one-hour timeframe of the infusion, which was likely due to the use of medetomidine in these experiments, known to prevent lactate build up and labelling²⁶ compared to isoflurane²⁷, which was used in Study 1 (Fig.3b). While high lactate concentration and lactate ¹³C-satellites can be seen when using isoflurane, the use of medetomidine led to a complete absence of ¹³C-incorporation in the lactate pool (Supplementary Fig.4).

Figure 3:

Figure 4:

Within one hour of s.c. ¹³C-glucose infusion, the isotopic fractional enrichment (FE) of GlxH2 reached 0.21±0.06, while it was 0.36±0.09 for GABAH4 and 0.13±0.05 for GluH4. The comparatively high FE of GABA may reflect a combination of factors, including the relative metabolite pool sizes (Table 1; uncorrected concentrations: GABA: 4.40±1.07 [mM] vs. Glu: 3.99±0.71 [mM]) or differences in GABA metabolic turnover rates within the brain structures covered by the voxel (hippocampus and striatum). The amount of labelling of each metabolite pool is dependent on the ¹³C-glucose infusion time, thus metabolite labelling may be more comparable across groups or across experiments when reported relative to each other, such as by using the ¹³C-GluH4/¹³C-GABAH4 ratio. Alternatively, as the FE of GlxH2 (Glx_FE) can be assessed with reference to its ¹³C-satellites, it can be used as a normalizing factor to report labelling of glutamate or GABA. The ¹³C-GluH4/¹³C-GABAH4 ratio here was thus 0.37±0.19, while the ¹³C-GluH4/Glx_FE was 2.82±1.42 and the ¹³C-GABAH4/Glx_FE was 8.32±3.77 (table 1). While coefficients of variation (CV) between animals were between ∼0.1-0.2 for basal metabolite concentrations, the CV were higher for the ¹³C-labelling concentrations (0.3-0.4), which is likely due to the variability in infusion efficiency across animals. Notably, using ¹³C-labelling ratios did not reduce inter-animal variability, consistent with the strong dependence of ratio stability on the arterial input function (AIF) and labelling dynamics, which we investigated below. Overall, these results confirm the feasibility of detecting neurotransmitter ¹³C-labelling within one hour of infusion in mice, while highlighting the critical influence of tracer delivery and labelling dynamics on quantification variability.

Table 1:

Study 3: POPE¹³C-MRS detects brain neurotransmitter labelling in human brain

To evaluate the feasibility of POPE¹³C-MRS in humans, we tested the acquisition protocol in two participants to determine whether ¹³C labelling of brain GABA, glutamate, glutamine and lactate could be detected. Participants either received ¹²C-glucose (unlabelled) or ¹³C-glucose (labelled) intravenously. Each session included a baseline POPE¹³C-MRS scan in the dorsal anterior cingulate cortex (dACC), followed by the administration of 0.23 g/kg glucose over 30 minutes, and a post-infusion scan 1 hour later to allow for metabolic incorporation of ¹³C (Fig5.a,b). Following ¹³C-glucose administration, POPE¹³C-MRS detected reproducible ¹³Clabelled GlxH2 satellites resonances across acquisitions (Fig.5,c,d). The J-coupling constant of GlxH2 was measured and found to be similar to that of the mouse (Humans: J_CH(GlxH2)≈146 Hz; Mouse: 144Hz). Quantification of the spectra indicated a 21% decrease in GABAH4 and 37% decrease in GluH4, after the ¹³C-glucose infusion (Table 2). Notably, GlxH2 signal dropped by 33%, but the GlxH2-satellites signals only grew by 19%. This discrepancy suggests suboptimal sensitivity of the current acquisition to the GlxH2 satellites in humans, likely reflecting a combination of imperfect TE optimization for human J-coupling values, low signal-to-noise (SNR), underestimated editing pulse profile, or physiological motion. Interestingly, no lactate ¹³C-labelling was observed (Fig.5e), probably due to the low conversion from glucose to lactate in human brain. Overall, these changes reflect incorporation of glucose-derived carbon into excitatory and inhibitory neurotransmitter pools, demonstrating the ability of POPE¹³C-MRS to probe neurotransmitter-specific metabolism in vivo.

Figure 5:

Table 2:

The ¹²C-glucose session served as a control, allowing us to assess the stability and repeatability of POPE¹³C-MRS measurements in the absence of ¹³C labelling. By comparing the pre- and post-infusion scans, we could not identify changes in total metabolite levels due to sole administration of the 5% glucose bolus (Supplementary Fig.5), confirming that the drop in Glu and GABA observed with ¹³C-Glc are due to labelling and not changes in total metabolite pool concentration.

Estimation of an optimal acquisition window for ¹³C-labelled metabolite ratios in POPE¹³C-MRS

Compared with time-resolved quantification^10,28, metabolite labelling ratios offer a more practical approach for assessing excitatory–inhibitory metabolic balance. They improve SNR and allow for interruptions in acquisition – critical for the long scans required to resolve GABA labelling – but their stability, and hence comparability, is highly dependent on the ¹³C-labelling profile of the input tracer (supplementary Fig.6). We used a combination of experimental data and metabolic modelling to validate the stability and interpretability of metabolic labelling ratios accessible with POPE¹³C-MRS. Using ¹³C-MRS data in mouse brain under continuous ¹³C-glucose infusion, we sought to define the optimal acquisition window that maximizes the stability of metabolite ¹³C-labelling ratios and thereby minimizes sensitivity to variability in infusion protocols. We first evaluated ratio stability (Supplementary Fig.7) using theoretical labelling curves derived from a simplified pseudo–3-compartment model of brain metabolism (Supplementary Fig.7.a,d), compared to its associated original data (Supplementary Fig.7,e; data reused from previous work in Cherix et al.²⁸), and then assessed temporal stability in the POPE¹³C-MRS data acquired here (Supplementary Fig.7).

Across all three ratios accessible with POPE¹³C-MRS (¹³C-GluH4/Glx_FE, ¹³C-GABAH4/Glx_FE and ¹³C-GluH4/¹³C-GABAH4), stability increased with infusion duration and was maximal once labelling approached a plateau, whereas substantial variability was observed during the initial phase of the infusion (approximately the first hour). These findings highlight the importance of maintaining a stable, ideally near-linear arterial ¹³C-glucose input to promote constant metabolite ratios over time. Although theoretical modelling predicted the ¹³C-GluH4/¹³C-GABAH4 ratio to be the most stable across acquisition windows, this advantage was not fully realized in practice owing to the lower SNR of the GABA resonance, which introduced additional variability and could compromise glutamate labelling estimates. Together these results indicate that, when using continuous ¹³C-glucose infusion, labelling ratios are more reliable during later acquisition windows after labelling has stabilized, especially when GABA turnover is slow.

We next modelled the labelling profile expected from our human infusion protocol. The protocol comprised a 30-min bolus, a regime that is in theory suboptimal for maintaining stable metabolite labelling ratios (supplementary Fig.6e,f). Using the corresponding modelled arterial input function (Fig.5a), we predicted the resulting labelling dynamics of GluH4 and GABAH4 (Fig.5b) and compared these predictions with the empirically observed labelling profiles (Fig.5c). Notably, FE of Glu and Glx remained relatively stable across the acquisition window (80-100min post infusion), whereas GABA labelling increased substantially midway through the scan. This discrepancy between the model and the empirical data suggest that post-infusion glucose clearance and brain tracer availability may be slower or more prolonged in vivo than assumed by our model, likely reflecting physiological complexity not captured by the simplified AIF. Collectively, these findings indicate that, under our current human protocol, metabolite labelling curves approach a quasi-steady state approximately 2h after infusion onset, at which point labelling ratios may become reliable.

Discussion

We introduce and validate POPE¹³C-MRS, a neuroimaging approach that enables simultaneous detection of glutamatergic and GABAergic metabolic labelling using standard MRI hardware. By combining molecular specificity with clinical compatibility, POPE¹³C-MRS addresses key practical constraints associated with indirect detection of neurotransmitter-specific ¹³C labelling in vivo. Using a cross-species validation approach, we demonstrate consistent detection and quantification of neurotransmitter-specific ¹³C labelling across mouse and human brain. This provides access to excitatory-inhibitory metabolic balance in vivo, a key feature of brain function that has been difficult to assess non-invasively. While in vivo GABA labelling in the human brain has been reported previously^10,29–31, this is, to our knowledge, the first demonstration of detectable GABAergic labelling using an indirect ¹³C-MRS approach compatible with standard MRI hardware. The key advance lies in enabling simultaneous and internally consistent detection of excitatory and inhibitory metabolic labelling using a simple and widely accessible protocol. By enabling concurrent measurement of glutamatergic and GABAergic labelling, POPE¹³C-MRS provides direct access to excitatory-inhibitory metabolic balance, a central feature of brain function that has previously been difficult to assess non-invasively in humans.

Current clinical assessments of cerebral metabolism rely primarily on glucose-analogue PET imaging, which provides sensitive and quantitative measures of glucose uptake but offers no insight into downstream neurotransmitter-related metabolic pathways, including glutamate-glutamine cycling and GABA synthesis, which are altered in several neurological and psychiatric disorders. Emerging molecular MRI approaches, including hyperpolarized ¹³C methods³² and deuterium metabolic imaging³³, allow pathway-specific interrogation of metabolic fluxes, yet their clinical adoption remains constrained by their requirement for specialized hardware and acquisition sequences or by their restricted metabolic coverage. Dynamic indirect ¹³C - and ²H-MRS offer a complementary strategy to probe downstream neuroenergetic and neurotransmitter-related processes while remaining compatible with standard MRI platforms^22–24,31. POPE¹³C-MRS builds on this line of methodological developments but is specifically tailored to enable reliable access to GABAergic metabolism, which has remained technically challenging with existing indirect approaches. By refining widely available ¹H-MRS techniques and leveraging standard radiofrequency hardware, POPE¹³C-MRS addresses key limitations of direct detection at ²H or ¹³C frequencies, which typically require dedicated coils and tailored pulse sequences. The method is based on a standard proton editing sequence (MEGA-sLASER), making it readily deployable in clinical research environments. Similar to recent indirect ¹³C approaches (e.g. selPOCE) that prioritize targeted detection of selected resonances to infer neurotransmission cycling^9,34, POPE¹³C-MRS deliberately trades breadth of metabolic coverage for robustness and specificity, focusing on metabolites most informative for excitatory–inhibitory metabolism. In this context, POPE¹³C-MRS is not intended as a general-purpose ¹³C-MRS replacement, but as a targeted tool optimized for probing GABAergic metabolism alongside glutamatergic and glycolytic pathways. Using this approach, we reliably detected labelling of GABAH4, GluH4 and GlxH2 in both mouse and human brain.

Isotopic labelling of GABA and lactate in the human brain has been reported previously^29–31; however, detection of GABA has been severely constrained by its slow metabolic turnover and strong spectral overlap with neighbouring resonances. As a result, reported FE of GABA in human brain have typically been low (<12%), often approaching the noise floor of conventional acquisition protocols. In contrast, POPE¹³C-MRS enabled reliable detection of GABAH4 labelling within 100 minutes after infusion onset, reaching a mean FE of ∼21% over the average acquisition window. The use of editing-based ¹H-MRS not only improves measurement precision for GABA and lactate, but also enhances applicability at lower field strengths, where specific absorption rate constraints associated with broadband editing pulses can be less restrictive. This opens a pathway for translating POPE¹³C-MRS to more common 3 Tesla systems, potentially broadening applicability while preserving sufficient SNR for edited metabolite detection. While animal studies benefit from a broader range of ¹³C-MRS methodologies that offer higher metabolic coverage and are less constrained by SAR limitations, the simplicity and clinical compatibility of POPE¹³C-MRS may also facilitate translational pipelines by harmonizing protocols across species.

The ability to quantify GABA, glutamate and lactate labelling with improved precision provides biologically meaningful access to key components of brain metabolism. GABA is the principal inhibitory neurotransmitter in the mammalian brain, and imbalances between excitatory and inhibitory metabolic activity are implicated in a wide range of neurological and psychiatric disorders^35,36. Importantly, GABA is particularly abundant in deep brain structures such as the thalamus or striatum^37–39, which are difficult to probe with conventional MRS approaches. By enabling indirect ¹³C measurements in these regions, POPE¹³C-MRS may extend metabolic imaging beyond cortical targets that have dominated prior ¹³C-MRS work¹⁰. Lactate, as a major end-product of glycolysis, reflects aerobic glycolysis and astrocytic metabolic activity, processes that remain incompletely understood in the human brain⁴⁰. In our study, however, ¹³C-labelling of lactate was not detectable in human, consistent with previous reports of low conversion of glucose into lactate under normal physiological conditions^11,41. Notably, higher levels of lactate labelling from glucose have been reported in several pathologies^42–44, suggesting that POPE¹³C-MRS may provide a sensitive tool to detect such metabolic alterations. Because lactate labelling depends not only on glucose uptake but also on astrocytic glycogen turnover⁴⁵, which may causes an apparent dilution of ¹³C-lactate signal, the POPE¹³C-MRS may provide indirect sensitivity to astroglial metabolism and glycogen shunting, thereby offering new opportunities to study neurometabolic coupling in vivo. Consequently, POPE¹³C-MRS may be particularly well suited for clinical investigations of disorders characterised by dysregulated cerebral metabolism, including conditions associated with brain insulin resistance and altered excitatory–inhibitory balance.

Despite these advances, several technical challenges and limitations remain. While sensitivity and prolonged acquisitions required to resolve GABA labelling currently limit full metabolic fluxes modelling, the ability to derive stable metabolic labelling ratios provide a practical and biologically meaningful proxy for excitatory-inhibitory metabolic balance in vivo. However, such ratios are inherently dependent on the tracer administration protocol, arterial input function, and the temporal dynamics of metabolite labelling. Because labelling curves typically follow exponential saturation rather than linear trajectories, ratio stability can only be expected within specific temporal windows. We therefore identified acquisition windows in which labelling ratios are sufficiently stable to provide reliable and interpretable readouts, emphasizing the need for protocol harmonization to enable meaningful comparisons across studies. FE of lactate has previously been used as AIF in animal studies^46,47; however, this was not feasible under light anaesthesia or in human herein, necessitating alternative normalization strategies to improve the comparability of labelling measures. We therefore corrected metabolite labelling by the FE of ¹³C-GlxH2, which was expected to be less sensitive to specific excitatory or inhibitory flux changes than ¹³C-GABAH4 or ¹³C-GluH4. Alternatively, the ¹³C-GluH4/¹³C-GABAH4 ratio provided a comparatively stable metric over prolonged infusion periods, reducing sensitivity to infusion timing. An additional source of variability may arise from the contamination of the GABAH4 signal by co-edited macromolecule (MM) resonances, such that the edited signal reflects GABA+ rather than GABA alone⁴⁸. This macromolecular contribution can vary across subjects and physiological states and may therefore contribute to variability in apparent labelling across studies.

A further methodological consideration concerns the quantification of ¹³C satellite resonances, which are inherently challenging without appropriate calibration. This difficulty arises from the fast heteronuclear J-evolution (with J_CH-coupling constants typically ranging from 130 to 160 Hz), which can cause signal dephasing, and from the reduced SNR due to the splitting of the original signal into two components. Many indirect ¹³C-MRS approaches rely on baseline scans acquired prior to tracer administration, implicitly assuming stable total metabolite pool sizes. While this assumption held in our human control experiments with unlabelled glucose, we observed changes in total Glx and lactate pools in mice following glucose administration, indicating that this assumption may not hold in all physiological or pathological contexts. Although ¹³C satellite detection can, in principle, provide access to changes in total metabolite pools, their reliable measurement requires careful calibration of echo times and editing pulse profiles. In humans, we observed a small discrepancy between GlxH2 labelling estimated from peak disappearance and from satellite resonances appearance, likely reflecting imperfect sequence calibration and subtle differences in effective J-coupling constants between species. Notably, the use of sLASER rather than PRESS likely proved critical for robust satellite detection, owing to improved refocusing homogeneity of fast-evolving satellite resonances with adiabatic pulses⁴⁹.

Future work should focus on improving sensitivity, optimising tracer delivery, and validating reproducibility in larger human cohorts. These developments will be essential to establish POPE¹³C-MRS as a robust tool for investigating neurometabolic coupling and excitatory–inhibitory balance in clinical populations.

Materials and Methods

POPE¹³C-MRS experiments were conducted in mice and humans to evaluate sequence optimisation, metabolite labelling detectability, and cross-species feasibility.

Study 1 and 2 (mouse)

Animals

Male and Female C57BL6/J mice were used at the age of 3 months. Animals were housed in standard conditions (12h day-light cycle, 20-24°C, 46-65% humidity). Animals had ad libitum access to standard rodent chow diet and water. All procedures were approved by the University of Oxford ethical review committee and conducted under UK Home Office regulations (Animal [Scientific Procedures] Act 1986), and in compliance with the ARRIVE (Animal Research: Reporting in vivo Experiments) guidelines.

Anaesthesia and monitoring

Anaesthesia was induced with a bolus of 4% isoflurane mixed with air after which a single bolus of medetomidine in saline (0.1 mg/ml) was administered (0.2 mg/kg, s.c.) followed by a continuous infusion (0.4 mg/kg/h, s.c.) with minimal isoflurane administration (0.2% in air containing 35% O₂). Some animals were only kept under 1-1.5% isoflurane following the bolus for the protocol optimization. The breathing rate and rectal temperature were monitored during the entire scans using a small animal monitoring system (SA Instruments Inc., New York, USA).

¹³C-glucose Infusion

After the baseline POPE¹³C-MRS scan (see below), animals were administered a (20% w/v) solution of (99%) uniformly ¹³C-labelled glucose ([U-¹³C₆]-Glc) solution (CK Isotopes Ltd.) via a sub-cutaneous (s.c.) cannula connected with an infusion pump (PHD 2000, Harvard Apparatus) which was inserted in the animal’s flank. A subcutaneous infusion was used to achieve slower absorption kinetics and approximate linear increase in blood glucose ¹³C-labelling, which consisted of a 5 min bolus (2.83 mL/kg, 99% FE) followed by continuous flow (10 mL/kg/h, 99% FE).

MRI/MRS acquisition

POPE¹³C-MRS was acquired on a 7T (70/20) BioSpec MRI scanner (Bruker, Ettlingen, DE) equipped with an 86 mm transmit volume coil and 2×2 receive ¹H-cryoprobe and using Paravision 360.1.1. Anatomical T₂-weighted MRI images were acquired for the MRS voxel placement with a localizer (FLASH, TE/TR= 3ms/15ms, 2 averages). The MRS voxel (6.5 × 3.4 × 3.2 mm³) encompassed the dorsal hippocampal area and striatum at the centre-top of the brain. Shimming (MAPSHIM) was performed in the voxel to reach a water Full Width at Half Maximum (FWHM) below 15Hz. POPE¹³C-MRS data were acquired using an in-house implementation of a MEscher-GArwood semi-Localization by Adiabatic Selective Refocusing (MEGA-sLASER)⁵⁰ sequence, configured to selectively detect GABA/glutamate or lactate. Editing pulse offsets and bandwidths were optimised for each metabolite (see Supplementary Methods).

Data processing and modelling

Spectra were processed using jMRUI and quantified with QUEST⁵¹ (see Supplementary Methods). ON and OFF transients were phase- and frequency-corrected and averaged separately before subtraction (ON-OFF). Baseline (t₀) and post-infusion (t_f) spectra were then aligned, apodised (2Hz), removal of the NAA signal using Hankel-Lanczos Singular Value Decomposition (HLSVD)⁵², and analysed using predefined basis set (Supplementary Fig.1). Absolute concentrations were estimated using the unsuppressed water signal as an internal reference (80% brain water content ⁵³).

To assess the reliability and consistency of metabolite ¹³C-labelling ratios, we used raw data from Cherix et. al.²⁸ acquired on a horizontal 14.1 T scanner in mouse brain using indirect ¹³C-MRS (¹H[¹³C]-MRS) as previously described^54,55. The labelling curves fitted from this dataset using a pseudo 3-compartment model was then used to estimate how the theoretical labelling ratios (e.g. ¹³C-GluC4/¹³C-GABAC4) fluctuate throughout this experiment under a continuous ¹³C-Glc infusion. A similar approach was then used for the data acquired herein with POPE¹³C-MRS in mice. To determine how ¹³C-labelling ratios are affected by the acquisition window, i.e. the timing between the start and end of the acquisition relative to the start of the ¹³C-Glc infusion protocol, the average ratio was computed for variable windows and was reported relative to its final ratio value.

Study 3 (human)

Experiments in humans conformed to the Declaration of Helsinki, apart from pre-registration, and were approved by the Medical Sciences Interdivisional Research Ethics Committee of the University of Oxford (MS IDREC-R94883/RE001). Two healthy male participants (27 and 33 years old) were recruited and provided written informed consent. Neither participant had a history of diabetes, hyperglycaemia, neurological or psychiatric conditions and both met local MRI safety criteria. Participants had a one-hour scan, followed by a ¹³C-glucose infusion and a second one-hour scan.

Infusion protocol

¹³C-labelled D-glucose ([U-¹³C₆]Glc, 99% ¹³C enrichment) was purchased (CK Isotopes Limited) and processed for intravenous (i.v.) infusion in accordance with EU cGMP principles. Participants were asked to fast for 8 hours prior to the experiment. Participants received either ¹³C-glucose or unlabelled (¹²C) glucose (0.23 g/kg, 5% w/v)² via intravenous infusion over 30 min. The remaining isotope was flushed through the infusion line using isotonic saline.

MRI/MRS acquisition

Participants were scanned in a Siemens MAGNETOM 7T system (Siemens Healthineers, Erlangen, Germany) with a Nova Medical 8Tx32Rx head coil (Nova Medical Inc, Wilmington, MA, USA). An anatomical T1w MPRAGE image was acquired for placing the MRS voxel (TR=2.6 s, TE=3.31 ms, isotropic resolution 0.7 x 0.7 x 0.7 mm, 256 slices). The MRS voxel was positioned in the dorsal anterior cingulate cortex (dACC, 16 x 16 x 35 mm). First- and second-order shims were adjusted first by vendor-implemented gradient-echo shimming (“GRE Brain”)⁵⁶, then fine adjustment of shims were done using FAST(EST)MAP⁵⁷. A MEGA-sLASER sequence (CMRR Spectro Package, University of Minnesota⁵⁸) was used (TR = 4s, 2048 points, acquisition BW = 3kHz, number of transients =32 x 2 editing conditions, editing bandwidth = 160Hz) for GABA (TE = 69.5 ms, editing offset 1.90 ppm and 7.50 ppm) and lactate (TE = 106 ms, editing offset 4.10 ppm and 5.10 ppm). Water suppression was achieved with VAPOR⁵⁹ (150Hz Bandwidth) and outer volume saturation (OVS) was enabled on the anterior-posterior axis. The MRS acquisition consisted of five GABA-POPE¹³C-MRS and five Lactate-POPE¹³C-MRS segments acquired in interleaved sequence, with an unsuppressed water reference signal acquired at the beginning and the end of the MRS protocol. An identical protocol was used at baseline (t0; before the ¹³C-Glc infusion) and final (tf; after the ¹³C-Glc infusion) scans.

Data processing and modelling

MRS data was processed as described above for the mouse data, i.e. using QUEST routine⁶⁰ in jMRUIv6.0⁵¹ using a combination of basis sets with identical prior knowledge.

Plasma [U-¹³C₆]glucose enrichment following a 30 min constant intravenous infusion of 5% glucose (99% ¹³C; 3.2 mmol min⁻¹) in our participant was modelled using a finite-duration input function with first-order clearance, during infusion and mono-exponential washout thereafter . The apparent distribution volume of glucose was estimated⁶¹ to be V_d = 0.26 L kg^-1 and a first-order fractional clearance rate of plasma glucose was estimated⁶² to be k = 0.029 min^-1. Plasma fractional enrichment AIF(t) was computed assuming clamped total glucose (baseline 5 mmol L⁻¹) and 99% tracer purity. Brain metabolite labelling was modelled using first-order driven kinetics to estimate the labelling dynamics of GABAH4 and GluH4 within a simplified representation of cerebral metabolism (Equations 1 and 2). Equation 1 describes glutamate labelling dynamics, and Equation 2 describes GABA labelling dynamics, where effective influx and efflux constants represent composite metabolic fluxes derived from literature values.

Effective influx and efflux constants V_Glu,in = 0.86 µmol g^-1min^-1, V_Glu,out = 0.95 µmol g^-1min^-1and V_GABA,in = V_GABA,out = 0.086 µmol g^-1min^-1 were defined as composite fluxes including literature-based values for human brain. Glutamatergic influx V_glu,in was defined as the sum of transmitochondrial (V_x) and excitatory neurotransmission (V_NT^e) fluxes, while glutamatergic efflux also included glutamate decarboxylation (V_GAD), i.e. V_glu,in = V_x + V_NT^e and V_glu,out = V_x + V_NT^e + V_GAD, with V_x = 0.57 µmol g^-1min^-1,⁶³ V_NT^e accounting for 90% of total neurotransmission cycling^29,64 V_NT = 0.32 µmol g^-1min^-1and V_GAD accounting for 10% of k_Glu,in (V_GAD = 0.086 µmol g^-1min^-1).^15,29 GABAergic influx was set equal to its efflux, with glutamate decarboxylation being equivalent to the sum of GABAergic neurotransmission (V_NTⁱ = 0.1 · V_NT = 0.032 µmol g^-1min^-1),) and GABA recycling (V_s𝑕untⁱ = 0.054 µmol g^-1min^-1), i.e. V_gaba,in = V_GAD and ^V_gaba,out ^{= (V}_NT^i+V_s𝑕unt^{i) = V}_GAD.

Statistics

Statistics were all performed with GraphPad Prism (GraphPad Software, San Diego, CA, USA). All values are given as mean± standard deviation. unless stated otherwise. The ¹³C-labelled metabolite ratio analysis and human ¹³C-labelling dynamics were performed using MATLAB (R2020a).

Data availability

Source data underlying the figures are provided as a Source Data file. Additional data supporting the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgements

This study was supported financially by the Swiss National Science Foundation (P500PM_203208 to AC) and the University of Oxford Medical Sciences Internal Fund (Pump-Priming, 0014920 to AC). The Wellcome Centre for Integrative Neuroimaging (WIN, now OxCIN) was supported by core funding from the Wellcome Trust (203139/Z/16/Z and 203139/A/16/Z). This work was supported by the NIHR Oxford Health Biomedical Research Centre (NIHR203316). The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care. We also thank the UK BBSRC (grant number BB/W019582/1) for support. WTC is funded by Wellcome [225924/Z/22/Z]. CJS holds a Wellcome Trust Senior Research Fellowship [224430/Z/21/Z].

The MRS package was developed by Edward J. Auerbach and Małgorzata Marjańska and provided by the University of Minnesota under a C2P agreement. We thank Damian Tyler (OCMR Oxford) for advice and feedback on the manuscript. We thank Sean Smart (OxCIN core staff) for support with preclinical equipment. We thank Roswell Shelhamer (Cambridge Isotope Laboratories, Inc.) and Ben Pepper (CK isotopes Ltd.) for their helpful discussions and support in sourcing the isotope. We acknowledge Rolf Gruetter for constructive feedback on the manuscript and for supporting the reuse of previously acquired data.

This research was funded in whole, or in part, by the Wellcome Trust [Grant number 203139/Z/16/Z, 203139/A/16/Z, 224430/Z/21/Z, and 225924/Z/22/Z]. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.

Additional information

Authors’ contributions

AC designed the study. AC, MT, WC and JC optimized the protocol. AC, JC, MT and OH acquired the data. AC analysed and interpreted the data. AC drafted the manuscript. All the authors assisted in revising the manuscript and approved the final version.

Funding

Swiss National Science Foundation (P500PM_203208)

• Antoine Cherix

Wellcome Trust (WT)

https://doi.org/10.35802/224430

• Charlotte J Stagg

Wellcome Trust (WT)

https://doi.org/10.35802/225924

• William T Clarke

Wellcome Trust (WT)

https://doi.org/10.35802/203139

• Oliver Haermson

• Antoine Cherix

• William T Clarke

• Jason P Lerch

• Charlotte J Stagg

• Mohamed Tachrount

• Jon Campbell

UKRI | Biotechnology and Biological Sciences Research Council (AFRC) (BB/W019582/1)

• Oliver Haermson

• William T Clarke

• Mohamed Tachrount

• Jon Campbell

• Jason P Lerch

• Antoine Cherix

• Charlotte J Stagg

Additional files

Supplementary Figure legends and Supplementary material

Source data