Multimodal neuroimaging provides a highly consistent picture of energy metabolism, validating ³¹P MRS for measuring brain ATP synthesis

Abstract

Neuroimaging methods have considerably developed over the last decades and offer various noninvasive approaches for measuring cerebral metabolic fluxes connected to energy metabolism, including PET and magnetic resonance spectroscopy (MRS). Among these methods, ³¹P MRS has the particularity and advantage to directly measure cerebral ATP synthesis without injection of labeled precursor. However, this approach is methodologically challenging, and further validation studies are required to establish ³¹P MRS as a robust method to measure brain energy synthesis. In the present study, we performed a multimodal imaging study based on the combination of 3 neuroimaging techniques, which allowed us to obtain an integrated picture of brain energy metabolism and, at the same time, to validate the saturation transfer ³¹P MRS method as a quantitative measurement of brain ATP synthesis. A total of 29 imaging sessions were conducted to measure glucose consumption (CMRglc), TCA cycle flux (V_TCA), and the rate of ATP synthesis (V_ATP) in primate monkeys by using ¹⁸F-FDG PET scan, indirect ¹³C MRS, and saturation transfer ³¹P MRS, respectively. These 3 complementary measurements were performed within the exact same area of the brain under identical physiological conditions, leading to: CMRglc = 0.27 ± 0.07 μmol·g⁻¹·min⁻¹, V_TCA = 0.63 ± 0.12 μmol·g⁻¹·min⁻¹, and V_ATP = 7.8 ± 2.3 μmol·g⁻¹·min⁻¹. The consistency of these 3 fluxes with literature and, more interestingly, one with each other, demonstrates the robustness of saturation transfer ³¹P MRS for directly evaluating ATP synthesis in the living brain.

Numerous brain disorders, like neurodegenerative diseases, are associated with impairment in energy metabolism. This observation has been driving considerable technological developments in medical imaging, aiming at measuring brain energy metabolism. PET combined with ¹⁸F-2-fluoro-2-deoxy-D-glucose (¹⁸F-FDG) injection has been used in research on normal and pathological brain for ≈30 years (1–3). More recently, magnetic resonance spectroscopy (MRS) has brought new tools for imaging cerebral energy fluxes (4–16).

However, methodological developments are still needed to answer the clinical need for earlier diagnosis and follow-up of neurodegenerative pathologies. Although PET detection of ¹⁸F-FDG has proven efficient to map cerebral glucose consumption (CMRglc), this technique does not directly reflect energy storage and utilization that is mainly derived from glucose oxidation. Also, PET is unlikely to become widely accessible, due to its invasiveness and cost. However, magnetic resonance is of widespread use for anatomical imaging, and clinical scanners can be equipped for metabolic imaging at limited cost. Indeed, MRS has proven powerful to measure brain energy metabolism by detecting ¹³C, ¹⁷O, or ³¹P nuclei, which only require dedicated radiofrequency components. Quantitative measurement of cerebral oxidative metabolism by ¹³C (4–11) or ¹⁷O (14–16) MRS has been demonstrated. However, these approaches require i.v. injection of expensive precursors followed by time-consuming acquisition. In contrast, ³¹P MRS has the potential to measure the cerebral rate of ATP synthesis V_ATP without any injection, using the magnetization transfer technique as originally demonstrated by Brown et al. (17). Up to now, only 2 measurements of cerebral V_ATP have been reported in rodents (12, 18), and 1 in humans (13). MRS-derived V_ATP reported in these pioneer studies appeared consistent with literature values and coupled with brain activity over a large range of anesthesia (18). However, MRS-derived V_ATP may be contaminated by a near-equilibrium exchange reaction between inorganic phosphate (P_i) and ATP catalyzed by the glycolytic enzymes GAPDH and phosphoglycerate kinase (PGK). Such a contamination, leading to an overestimation of the NMR-detected V_ATP, was inconsistently reported on yeast suspensions (19), with a possible dependence on the growth medium. MRS-derived V_ATP was also shown to be contaminated by glycolytic enzymes ex vivo on perfused peripheral organs (20, 21), on rodent skeletal muscle (22), and human skeletal muscle (23). Given these elements, further studies are required to establish ³¹P magnetization transfer as a reliable quantitative method for measuring the cerebral rate of energy synthesis V_ATP.

In this context, our purpose has been to validate the ³¹P MRS measurement of brain V_ATP, by comparing the rate of ATP synthesis measured in primates with CMRglc and TCA cycle rate (V_TCA) measured by PET and ¹³C MRS, respectively. Under normal physiological conditions, glucose breakdown (through glycolysis and TCA cycle) is stoechiometry coupled to ATP synthesis, so that theoretical V_ATP can be calculated from the measured CMRglc and V_TCA, providing a direct way to validate MRS-measured V_ATP.

Results show that CMRglc and V_TCA measured in the primate brain are highly consistent, showing the expected 1:2 stoechiometry between glycolysis and TCA cycle (1 glucose giving rise to 2 pyruvate molecules). Also, the theoretical value of V_ATP derived from CMRglc and V_TCA is very close to that directly measured using ³¹P MRS and magnetization transfer. These results demonstrate that ³¹P MRS is a reliable method to measure brain ATP synthesis in vivo.

Results

High-resolution scout MRI of the monkey brain is presented on Fig. 1, showing the 8-mL volume of interest (VOI) where the 3 metabolic fluxes were measured. Anatomical landmarks on coronal, axial, and sagital images made it possible to reposition the VOI identically for all imaging sessions.

Fig. 1.

Three-dimensional MRI images acquired in 1 monkey. The position of the VOI in which CMRglc, V_TCA, and V_ATP were measured is presented on coronal (A), axial (B), and sagital images (C).

A typical ¹⁸F-FDG PET image (axial view; acquisition time, 20 min) is presented on Fig. 2A. The corresponding ¹⁸F time activity curve in the selected VOI and the experimental ¹⁸F arterial time activity curve are plotted in Fig. 2B. Kinetic analysis of ¹⁸F-FDG uptake produced an estimation of ¹⁸F-FDG-6-P and ¹⁸F-FDG contributions to total activity (Fig. 2B), leading to CMRglc = 0.27 ± 0.07 μmol·g⁻¹·min⁻¹.

Fig. 2.

CMRglc measurement by ¹⁸F PET. (A) VOI from which the total ¹⁸F activity was extracted, (B) corresponding ¹⁸F-FDG PET time-activity curve (♦) and best fit by the 2-tissue compartment model (bold solid line). Arterial function measured in the same session is presented (thin solid line). Modeled contributions of ¹⁸F-FDG and ¹⁸F-FDG-6-P are shown (gray lines).

Fig. 3A shows a stacked plot of difference ¹H-{¹³C} spectra (averaged over the 8 ¹³C NMR sessions) acquired during an i.v. infusion of [U-¹³C₆] glucose. Corresponding ¹³C-enrichment time courses of glutamate C3 and C4 are presented in Fig. 3B. The continuous lines represent the best fit to experimental data according to the metabolic model. This analysis leads to V_TCA = 0.63 ± 0.12 μmol·g⁻¹·min⁻¹.

Fig. 3.

V_TCA measurement by indirect ¹³C NMR. (A) Stack plot of ¹H-{¹³C} difference spectra in glutamate C3 and C4 region. Glutamate ¹³C enrichment appears at the frequency of C3 and C4 mother resonances on our ¹H difference spectra. Spectra were post processed by a 5 Hz lorentzian line broadening. (B) ¹³C enrichments time courses (averaged over 8 sessions) measured for glutamate C3 (▵), glutamate C4 (□), and best fit to the data (solid lines).

The P_i region of ³¹P spectra acquired with γ-ATP saturation for 4 different saturation times (0.5, 1.0, 1.5, and 2 s) is presented in Fig. 4A (averaged over the 15 ³¹P NMR sessions). It shows that P_i amplitude is decreased while the saturation time is increased. Note that P_i decrease on this figure is due to both saturation transfer effect and RF bleed over effect, the later explaining the decrease of resonances surrounding P_i (mostly phosphodiesters and phosphomonoesters ≈3 and 7 ppm, respectively); ³¹P spectra were corrected for RF bleed over by subtracting γ-ATP saturated spectra from control spectra where saturation was applied symmetrically to γ-ATP. Inversion-recovery experiments yielded a mean value of T1^mix = 2.05 s. Consequently, the averaged P_i attenuation vs. t_sat was fitted fixing the following parameters: T1^mix = 2.05 s, θ = 60°, TR = 2.95 s. The best fit is shown on Fig. 4B. The estimation of k_f and T1^int from averaged M_S(t_sat)/M_C(t_sat) fitting and Monte Carlo simulation yielded k_f = 0.10 ± 0.03 s⁻¹ and T1^int = 2.1 ± 0.4 s. Cerebral P_i concentration was fixed to [P_i] = 1.3 mM, according to literature values (24–29). By using a brain density of 1g/mL, we obtained the ATP synthesis rate V_ATP = 7.8 ± 2.3 μmol·g⁻¹·min⁻¹.

Fig. 4.

V_ATP measurement by ³¹P saturation transfer NMR. (A) P_i region of ³¹P spectra acquired on γ-ATP saturation for 4 different saturation times (0.5, 1.0, 1.5, and 2 s). P_i amplitude decreases as saturation time is increased. (B) P_i attenuation [M_S(t_sat)/M_C(t_sat)] vs. saturation time t_sat. The solid line is the best fit to experimental P_i attenuation (♦). The dotted lines indicate the lower and upper limits of the fit based on k_f and T1^int SD.

The 3 metabolic fluxes measured in this study and their metabolic couplings are presented on Fig. 5.

Fig. 5.

Brain energy metabolism as measured by multimodal imaging: ¹⁸F-FDG was used to quantify CMRglc, ¹³C MRS was used to measure V_TCA, and ³¹P MRS was used to measure V_ATP. The 3 fluxes are highly consistent, validating ³¹P MRS as a robust tool for quantifying brain ATP synthesis.

Discussion

Consistency of Each Measured Flux with Literature Values.

Taken independently, the CMRglc and V_TCA fluxes measured in this study are consistent with literature values of brain energy metabolism for anesthetized animals (30–33), and particularly with the values reported by Boumezbeur et al. (30) in the monkey brain (CMRglc = 0.23 ± 0.03 μmol·g⁻¹·min⁻¹; V_TCA = 0.53 ± 0.13 μmol·g⁻¹·min⁻¹). The small differences between these values and CMRglc and V_TCA measured in this study could be explained by differences in VOI position, and consequently by different contributions of gray and white matter to the detected fluxes: in our study, gray matter accounts for 58 ± 2% of the detected brain tissue, as measured on high resolution T1-weighted images (data not shown). This percentage is likely to be higher than gray matter content in the voxel detected by Boumezbeur et al. (30), which was located deeper in the brain and included significantly less cortical areas. Given the strong dependence of glucose metabolism on gray matter content (7), a slight difference in gray matter fraction might explain our higher CMRglc and V_TCA values (0.27 and 0.63 μmol·g⁻¹·min⁻¹, respectively). PET measurements of CMRglc have been reported recently in the monkey brain, and found to be close to 0.3 μmol·g⁻¹·min⁻¹ in brain areas similar to our VOI (31). This value is very close to our measurement (0.27 μmol·g⁻¹·min⁻¹).

Our V_TCA values (0.63 ± 0.12 μmol·g⁻¹·min⁻¹) are in agreement with reported human values that range from 0.57 to 0.77 μmol·g⁻¹·min⁻¹ (6, 8, 9, 11).

The measurement of V_ATP reported in this article is, to our knowledge, the first one performed in the monkey brain. Compared with the only one reported measurement of cerebral V_ATP that was performed in humans (13), the V_ATP value reported in this study appears slightly lower (7.8 ± 2.3 μmol·g⁻¹·min⁻¹ vs. 12.1 ± 2.8 μmol·g⁻¹·min⁻¹). However, the SD ranges of both measurements overlap so that no significant difference can be inferred.

Consistency Between CMRglc and V_TCA.

In this study, a [V_TCA/CMRglc] ratio of 2.25 is measured, reflecting the metabolic coupling between glycolysis and TCA cycle. Considering that glucose is the unique metabolic fuel under normal physiological conditions, a [V_TCA/CMRglc] ratio of 2 would be expected, because glycolysis produces 2 molecules of pyruvate per glucose. The measured [V_TCA/CMRglc] ratio is 12.5% higher than the expected theoretical value, leaving room for a possible contribution of fatty acids to the TCA cycle (CMR_FA = V_TCA − 2 CMRglc). However, given the fact that the SD on each flux (V_TCA and CMRglc) is ≈15%, the [V_TCA/CMRglc] ratio is not high enough to claim a significant contribution of the fatty acid pathway. The 12.5% excess over theoretical ratio could be partly ascribed to experimental inaccuracy.

Potential Effect of Glycemia on the Measured Fluxes.

Blood glucose was measured during the course of PET and ¹³C-NMR sessions. Glycemia measured at the beginning of ¹³C-NMR sessions (before ¹³C glucose injection) and during the course of PET measurements were all in the 3–5 mmol·L⁻¹ normoglycemic range; ¹³C measurement of V_TCA was performed under hyperglycemia (9–15 mmol·L⁻¹ in our study), because a several fold increase in glycemia is required to bring ¹³C enrichment of blood glucose up to >50%. In this context, one may wonder whether CMRglc and V_TCA are affected by blood glucose content. CMRglc dependence on glycemia has been studied by several groups, who concluded that brain glucose uptake is not affected by acute hyperglycemia. Evidence for this result was demonstrated in rodents as well as in humans (34, 35). Unfortunately, V_TCA dependence on glycemia has not been studied as thoroughly as CMRglc, due to the lack of method (besides ¹³C NMR) for noninvasive measurement of the TCA cycle flux. It must be kept in mind that V_TCA measured using ¹³C-labeled glucose includes the contributions of all possible substrates of acetyl-CoA oxidation. Therefore, NMR-measured V_TCA is directly coupled to CMRO₂ (14), as shown ex vivo on brain slices (36). Since several studies have reported that CMRO₂ remains independent from glycemia in mammals (37, 38), the absence of glycemic dependence can be reasonably extended to V_TCA. Note that glycemia was not measured during ³¹P-NMR sessions, because V_ATP measurement does not require vascular catheterisation. Due to the similarity of experimental conditions with PET and preinfusion ¹³C-NMR sessions, it can be reasonably considered that V_ATP was measured under normoglycemia. Given these elements, one can assume that the metabolic fluxes measured in our study were not noticeably affected by glycemic differences.

Validation of V_ATP as Measured by NMR.

Given the values of CMRglc and V_TCA, the corresponding ATP synthesis rate V_ATP can be theoretically expressed by establishing stoechiometry between molecules of ATP and reducing equivalents (i) generated by the TCA cycle and oxidative phosphorylation, (ii) generated by glycolysis, and (iii) consumed by the fatty acid pathway. It is well known that the degradation of 1 acetyl-CoA in the TCA cycle leads to the production of 1 ATP, 3 NADH,H⁺, and 1 FADH₂, and that glycolysis leads to the production of 2 pyruvate, 2 ATP, and 2 NADH,H⁺, a posteriori transformed in 2 FADH₂ by crossing the mitochondrial membrane. Also, the conversion of pyruvate into acetyl-CoA in the mitochondrial matrix leads to the production of 2 NADH,H⁺. Under the hypothesis of all pyruvates produced through glycolysis being converted into acetyl-CoA, the rate of this reaction can be assumed to be equal to CMRglc. Then, the corresponding V_ATP can be expressed as:

where (P/O)_NADH,H+ and (P/O)_FADH2 are the number of ATP molecules per atom of oxygen produced by the degradation of 1 NADH,H⁺ and 1 FADH₂, respectively, in the respiratory channel. Based on established values for the P/O ratios [(P/O)_NADH,H+ = 2.3 and (P/O)_FADH2 = 1.4; see ref. 39], the theoretical rate of ATP synthesis expected from the rate of CMRglc and V_TCA we measured in the present study was: V_ATP^TH = 8.3 ± 1.8 μmol·g⁻¹·min⁻¹. This value is very close to the NMR-measured V_ATP (7.8 ± 2.3 μmol·g⁻¹·min⁻¹). The slightly lower measured value could be explained by a partial uncoupling between the generation of the proton gradient and the use of this gradient for synthesis of ATP from ADP and P_i. Possible explanation for partial uncoupling might involve specific uncoupling proteins like the brain-specific UCP4 (40, 41). However, the theoretical expected value V_ATP^TH is well within the measurement error of our ³¹P-NMR measured V_ATP. Therefore, our study does not allow us to conclude on significant uncoupling process in the mammal brain.

To our knowledge, only one previous study was performed in humans to assess the coupling between V_TCA and V_ATP, measured by ¹³C NMR and ³¹P NMR, respectively (23). This study, performed in skeletal muscle, reported the following values: V_TCA = 0.06 μmol·g⁻¹·min⁻¹ and V_ATP = ≈5 μmol·g⁻¹·min⁻¹. According to Eq. 1, the theoretical value of V_ATP expected from the measured V_TCA ranges from V_ATP^TH = 0.5 μmol·g⁻¹·min⁻¹ (assuming that FA is the sole substrate for muscle, i.e., CMRglc = 0 and CMR_FA = ½V_TCA) to V_ATP^TH = 0.8 μmol·g⁻¹· min⁻¹ (assuming that glucose is the sole substrate, i.e., CMRglc = ½V_TCA and CMR_FA = 0). Thus, the range of expected V_ATP^TH values is 10 to 6 times as low as ³¹P-NMR measured V_ATP (≈5 μmol·g⁻¹·min⁻¹), which demonstrates that the GADPH shunting is dominant in human skeletal muscle. In contrast, the present study gives a cerebral V_ATP slightly lower than the theoretical value expected from CMRglc and V_TCA; thus, ruling out the possibility of a contribution of reversible synthesis by GADPH/PGK at the glycolytic level in the brain. Our data would be consistent with the fact that P_i is mostly present in neurons while a strong glycolytic component takes place in astrocytes as postulated by Lei et al. (13). This explanation relies on the hypothesis that lactate (provided by astrocytes through an astrocyte-neuron lactate shuttle) is the major substrate for neuronal oxidative metabolism. Although experimental evidences argue in favor of lactate shuttle having a key role in glutamatergic activation (42, 43), the significance of this shuttle remains controversial (44, 45). More importantly, the extent of physiological conditions under which the lactate shuttle dominates direct neuronal glucose uptake remains to be explored. It must be kept in mind that the animal and human metabolic fluxes discussed here were measured under light anesthesia or alpha state (peaceful awakefulness). Further studies will be required to establish that lactate shuttle remains predominant under those physiological conditions.

In conclusion, this study is, to our knowledge, the first report of cerebral CMRglc, V_TCA, and V_ATP measurements in the same animals under identical physiological conditions. This unique multimodal approach allows a cross-validation of the described ¹⁸F-FDG, ¹³C MRS, and ³¹P MRS techniques. In particular, this is a direct in vivo validation of the ³¹P saturation transfer method as a reliable measurement of the cerebral ATP synthesis rate. Implementation of the 3 methodological approaches brings a unique integrated picture of brain energy metabolism, from glucose phosphorylation to mitochondrial ATP synthesis. This multimodal approach will help to better understand mitochondrial energy defects that are thought to have a key role in neurodegenerative illnesses (46, 47).

Materials and Methods

The study was conducted on 3 healthy male monkeys (macaca fascicularis, body weight ≈5 kg) after they were fasted overnight. All experimental procedures were performed in strict accordance with the recommendations of the European Community (86/609) and the French National Committee (87/848) for care and use of laboratory animals. For both PET and NMR sessions, animals were identically anesthetized by a single ketamine-xylazine intramuscular injection followed by an i.v. infusion of propofol (≈200 μg/kg/min), intubated, and ventilated. The head was placed in the Sphinx position by using a home made stereotaxic frame with bite-bar and ear rods. During NMR and PET sessions, physiological parameters were monitored and remained stable within normal ranges: 35–37 °C for body temperature, 50–60 mm Hg for noninvasive blood pressure as measured with an air-cuff placed around the arm, 90–110 min for cardiac frequency, 18–23 min for respiratory frequency, and 35–40 mm Hg for expired CO₂ saturation.

Experimental Design.

Neuroimaging sessions.

In this study, the experimental plan aimed at measuring the 3 following metabolic fluxes in a same 2 × 2 × 2 cm³ cerebral VOI on the group of monkeys: (i) CMRglc using PET acquisitions after i.v. injection of ¹⁸F-FDG (1-h acquisition time, 6 sessions: 2 per monkey); (ii) V_TCA using indirect ¹³C-NMR spectroscopy during an i.v. infusion of [U-¹³C₆] glucose (2-h acquisition time, 8 sessions: 2 to 3 per monkey); and (iii) V_ATP using ³¹P-NMR spectroscopic acquisition by saturation transfer (1.40-h acquisition time, 15 sessions: 4 to 6 per monkey). Given the long acquisition time of NMR measurements, animal welfare motivated our decision not to perform ¹³C-NMR and ³¹P-NMR during the same session, and to allow at least 2 weeks recovery between 2 subsequent imaging sessions for each monkey.

Each imaging modality has its own sensitivity to the measured metabolic flux: under our experimental conditions, V_TCA determination by ¹³C NMR was less accurate than CMRglc determination by PET, likely due to the intrinsic low sensitivity of NMR spectroscopy. V_ATP determination by ³¹P NMR was even less accurate the V_TCA determination, mostly due to the high number of measured parameters required to assess V_ATP. Our purpose being to compare the 3 fluxes, the number of imaging sessions was empirically set to compensate for sensitivity differences between modalities. This empirical design allowed assessing the 3 fluxes with similar accuracies.

VOI positioning and multimodal registration.

Proper comparison of the 3 metabolic fluxes requires reproducible positioning of the VOI in the brain for all imaging sessions. For ¹³C and ³¹P NMR sessions, high-resolution 3D MRI images of the brain (gradient echo sequence, matrix 128 × 128 × 128, resolution 1 × 1 × 1 mm³) were acquired to position the 8-mL VOI. Accurate repositioning was easily achieved for each monkey by using anatomical landmarks on coronal, axial, and sagital images (Fig. 1), so that NMR acquisitions were performed in comparable VOIs. Since PET provided images of the whole brain, it was necessary to localize the 8-mL VOI on PET images to assess CMRglc from this volume only. We registered 3D reconstructed PET images with 3D MRI by using robust and fully automated rigid registration method (48). Registration allowed one to accurately localize the NMR detected VOI within PET images and to extract ¹⁸F time activity from this VOI.

CMRglc Measurement by ¹⁸F PET.

PET sessions.

PET experiments were performed on an ECAT EXACT HR+ tomograph (Siemens-CTI). After completing a transmission scan with a ⁶⁸Ga-⁶⁸Ge source for attenuation correction, 24 emission scans (63 slices, 4.5-mm isotropic intrinsic resolution) were collected for 60 min after ¹⁸F-FDG i.v. bolus injection (≈2.5 mCi). To obtain arterial blood function, blood samples were withdrawn every 15 s during the first 2 min of acquisition; then, at 2.30, 3, 5, 7, 10, 15, 20, 30, 40, and 50 min. Arterial blood radioactivity was measured in a cross-calibrated γ-counter (Cobra Quantum D5003; Perkin-Elmer). For control of physiological stability, measurements of glycemia, blood pH, pO₂, pCO₂, sO₂ were performed at the beginning, after 20 min and at the end of the acquisition protocol.

Flux quantification.

The 8-mL VOI was extracted from the 24 PET images as described above, and the corresponding time activity curve was generated with regional activity calculated for each frame and plotted vs. time. The time activity curve was then fitted by using a 2-tissue compartment model in PMod (PMOD Technologies) (49), under the hypothesis of irreversible phosphorylation (k₄ = 0). The kinetic constants k₁, k₂, and k₃ describing the exchanges between the pools of plasmatic ¹⁸F-FDG, cytoplasmic ¹⁸F-FDG, and cytoplasmic phosphorylated ¹⁸F-FDG-P were derived from this adjustment. The lumped constant (LC) was fixed to 0.42, according to previous studies (30, 31). Last, CMRglc was calculated from the values of glycemia, LC, k₁, k₂, and k₃.

V_TCA Measurement by Indirect ¹³C NMR.

NMR setup and voxel positioning.

MR experiments were performed on a whole-body 3 Tesla system (Bruker) equipped with a surface coil placed on top of the head (double-tunable ¹H-³¹P, Ø ≈ 4.5 cm). V_TCA sessions started with the acquisition of the 3D MRI and the positioning the VOI. VOI shimming was performed down to ≈8 Hz by means of the FASTMAP algorithm (50), for first- and second-order shim coils.

Indirect ¹³C NMR spectroscopy.

Spectra were collected by using a ¹H STEAM sequence (TE/TM/TR = 21/110/2500, 256 transients). Additional localization was achieved by a B₁-InSensitive TRain to Obliterate signal (BISTRO) outer volume suppression (OVS) pulse train (51), consisting of 15 modules repeated at increasing RF power levels. Each module was formed by 3 double-band hyperbolic secant pulses selectively saturating slabs outside the 2 × 2 × 2 cm³ VOI along the X, Z, and Y directions (52). OVS was combined with VAPOR water suppression (53). At the beginning of the experiment, a baseline ¹H STEAM spectrum was acquired within the VOI. Then, ¹H STEAM spectra were collected during a 2-step infusion protocol of [U-¹³C₆] glucose. Infusion started with a 3-min bolus of [U-¹³C₆] glucose (≈0.3 mL/kg/min, 20% wt/vol), leading quickly to a 3-fold increase in glycemia: blood glucose concentration before bolus infusion was typically 3–5 mmol·L⁻¹ and reached 9–15 mmol·L⁻¹ after the bolus. Then, a continuous 2-h i.v. infusion of a 2:1 mixture of [U-¹³C₆] and unlabeled glucose was performed at lower rate (≈0.01 mL/kg/min, 20% wt/vol). This 2-step infusion leads to a stabilization of ¹³C fractional enrichment (FE) of plasma glucose ≈55 to 60% after 5 min and until the end of infusion (52). Control measurements of glycemia were performed by using a Onetouch glucose meter (Lifescan).

Spectra quantification and V_TCA measurement.

Measurement of ¹³C-glutamate enrichment from the ¹H spectra was based on a previously described method (30, 52). Briefly, the subtraction of ¹H spectra acquired during the [U-¹³C₆] glucose infusion from the baseline ¹H spectrum results in difference spectra exhibiting only labeled nuclei: mainly glutamate C3 and C4 in the brain; ¹³C enrichment has 2 effects on the ¹H spectrum: ¹H coupled to ¹³C (satellite resonance) appear as a doublet, whereas ¹H bound to ¹²C (mother resonance) decrease. Subtracting a ¹³C-enriched spectrum from a baseline spectrum provides a difference spectrum where mother resonances exhibit positive intensities (at 2.11 ppm for GluC3 and 2.35 ppm for GluC4), whereas ¹³C satellite resonances appear antiphased (at 2.85 and 1.85 ppm for GluC4 and at 1.65 and 2.65 ppm for GluC3); ¹H difference spectra collected during ¹³C-glucose infusion are dominated by changes in mother resonances, as shown on Fig. 3A; ¹H difference spectra were quantified by using Java-based MR user interface (jMRUI; see ref. 54), which performs time domain analysis of free-induction decays. An original basis set of simulated ¹H difference spectra of glutamate C4 and C3 ¹³C-enrichment was implemented for the quantitation based on quantum estimation algorithm (QUEST; see ref. 55) within jMRUI, by using literature values of resonance frequencies and J-coupling constants (56). The time courses of C4 and C3 enrichment were first estimated. To assess the value of the TCA cycle flux V_TCA, a single compartment model describing the incorporation of ¹³C from blood glucose into brain glutamate was implemented on Matlab (The MathWorks Inc.) (5, 33). The following assumptions were made: (i) glucose transport through the blood-brain barrier follows a reversible Michaelis-Menten kinetic, (ii) the exchange rates V_X between glutamate and α-ketoglutarate and between oxaloacetate and aspartate are equal, (iii) the rate of the glutamate/glutamine cycling is equal to 0.46 × V_TCA (5). The concentrations of glutamate, glutamine, aspartate, lactate, oxaloacetate, and α-ketoglutarate, as well as the glucose transport parameters, were taken from a previous monkey study (52).

V_ATP Measurement by ³¹P Saturation Transfer NMR.

NMR setup and voxel positioning.

MR experiments were performed on the previously described 3T system equipped with the same surface coil. VOI positioning and shimming were performed the same way as for V_TCA sessions.

Theory.

Consider the chemical equilibrium between inorganic phosphate P_i and ATP:

where k_f and k_r are the unidirectional rate constants of ATP synthesis and hydrolysis, respectively. To derive the rate of ATP synthesis V_ATP = k_f[P_i], k_f is measured by progressively saturating the magnetization of γ-ATP and observing the effect on the magnetization of the exchange partner P_i. Then, k_f is derived from the Bloch equations modified for chemical exchange. This method can be applied to ATP synthesis because the lifetime of P_i (1/k_f) is in the same order as the intrinsic longitudinal relaxation time of P_i T1^int. This observation explains why the saturation transfer method is specifically sensitive to ATP synthesis, although P_i is involved in several biochemical reactions (12, 13, 17, 18). However, it must be noted that the P_i to ATP conversion catalyzed by the GAPDH/PGK enzymes at the glycolytic level is fast enough to potentially contaminate ³¹P-NMR measured P_i attenuation (20–23).

Saturation Transfer ³¹P MRS Experiment.

The ³¹P spectra were collected from the VOI by using an OVS-localized saturation transfer sequence (57). A saturation pulse (variable length t_sat) was followed with a BISTRO OVS pulse train (total OVS train length t_OVS, 310 ms). A 100-μs broadpulse was placed immediately after the OVS module for nonselective excitation. The saturation frequency was first set to γ-ATP frequency (−7.35 ppm relative to P_i), and spectra were collected for 4 different values of t_sat (0.5, 1.0, 1.5, and 2 s) by using a 2.95-s TR and 512 transients for each t_sat. Then, control spectra without γ-ATP saturation were collected. To correct for the RF bleed over effect, control spectra were acquired with a saturation frequency set to +7.35 ppm relative to P_i (control saturation symmetric to γ-ATP) for the same 4 values of t_sat. The total acquisition time including shimming for each t_sat was ≈100 min.

Inversion recovery experiment.

The T1 of P_i in the presence of chemical exchange T1^mix was measured by using an inversion recovery sequence: a 180° adiabatic hyperbolic secant pulse (length, 4 ms) set on P_i frequency and a gradient crusher (length, 2 ms) were placed before the OVS module at variable inversion time TI (TI, time between inversion pulse and acquisition). Note that for the TI = 0 experiment, the inversion pulse was placed between the OVS module and the acquisition pulse. Repetition time was fixed to 6.4 s to allow full relaxation of P_i magnetization (T1^mix is expected ≈2 s at 3 Tesla); ³¹P spectra were acquired for 7 values of TI (128 transients).

Spectra quantification.

All ³¹P spectra were zero-filled to 2,048 points and analyzed by using an Advanced Method for Spectral Fitting (AMARES) within jMRUI (54, 58); 13 ³¹P multiplets were included (24), assuming lorentzian line shapes. Shimming variations between experiments were corrected relative to the estimated linewidth of the dominant resonance PCR. Baseline and 2-order phase corrections were performed. For the saturation transfer experiment, the averaged P_i attenuation ratio was calculated as the average ratio of P_i magnetization on γ-ATP saturation M_S over P_i control magnetization M_C for each monkey and each saturation time.

Metabolic modeling and V_ATP measurement.

The time evolution of P_i longitudinal magnetization can be modeled by using the Bloch equation modified for chemical exchange between P_i and γ-ATP:

where M_zPi and M_zγATP are the longitudinal magnetizations of P_i and ATP, respectively. Mz_Pi⁰ is the fully relaxed longitudinal magnetization of P_i. P_i attenuation M_S(t_sat)/M_C(t_sat) depends on the user-fixed sequence parameters (delays t_OVS, t_sat, and TR, excitation angle θ), and on the following unknown parameters: P_i relaxation time in presence of chemical exchange T1^mix (59), unidirectional rate constant of ATP synthesis k_f, and P_i intrinsic relaxation time in the absence of chemical exchange T1^int (57). T1^mix was estimated by the inversion recovery experiment as previously described. The M_S(t_sat)/M_C(t_sat) vs. t_sat curve was fitted by using a nonlinear least squares algorithm leading to the estimation of the unknown parameters k_f and T1^int. Monte Carlo simulation was also performed on this dataset to assess the accuracy on the 2 fitted parameters. The flux of ATP synthesis V_ATP was derived from the equation V_ATP = k_f[P_i], where the cerebral P_i concentration [P_i] was estimated from the literature (24–29).