Address correspondence and reprint requests to Dr. O.A.C. Petroff at Department of Neurology, Yale University, 333 Cedar Street, New Haven, CT 06520-8018, U.S.A. E-mail: email@example.com
Summary: Purpose: Several findings suggest that energy metabolism and the glutamate–glutamine cycle may be impaired in epilepsy. Positron emission tomography often shows interictal hypometabolism of the epileptogenic hippocampus. In vivo microdialysis studies show that seizure-associated glutamate release is doubled, and clearance is slowed. We hypothesized that the glutamate–glutamine cycle between neurons and glia may be decreased in the epileptic human hippocampus.
Methods: A 20% solution of 2-13C-glucose was infused before resection of the epileptogenic hippocampus. Blood glucose isotopic fractions were measured every 30 min. Blood and brain specimens were frozen quickly; perchloric acid extracts of the small metabolites were prepared and analyzed by proton and carbon magnetic resonance spectroscopy (MRS) at 11.75 Tesla.
Results: Standard histology showed 12 with hippocampal sclerosis and five with minimal neuron loss. The relative rates of glutamate–glutamine cycling with respect to glutamate synthesis were decreased in biopsies affected by hippocampal sclerosis (mean, 0.08; 95% confidence interval, 0.04–0.12) compared with those with minimal neuron loss (0.52; 95% CI, 0.30–0.75). Mean cellular glutamate concentrations were higher in minimal neuron loss (8.9 mM; 95% CI, 7.4–10.4) than hippocampal sclerosis (7.3 mM; 95% CI, 5.9–8.7). Cellular glutamine concentrations (mean, 2.8 mM; 95% CI, 2.4–3.2; n = 17) were the same in all groups.
Conclusions: The epileptogenic, gliotic human hippocampus appears to be characterized metabolically by slow rates of glutamate–glutamine cycling, decreased glutamine content, and a relative increase in glutamate content. We hypothesize that the low rate of glutamate–glutamine cycling that results from a failure of glial glutamate detoxification could account for slow glutamate clearance from synapses and continuing low-grade excitotoxicity.
The majority of excitatory neurons in the human neocortex release glutamate as their primary neurotransmitter. Several factors may cause hyperexcitability of glutamatergic neurons including alterations to receptors, ion channels, glutamate transporters, and the strength of inhibitory synapses with γ-aminobutyric acid (GABA)ergic neurons. Recently there has been renewed appreciation for the potential role of impaired energetics in altering neuronal function (1,2). Much of the evidence for a key role of energetics has come from the application of magnetic resonance spectroscopy (MRS) to study brain glutamate metabolism (3–7). These studies have shown that contrary to some previous views, glutamatergic function is energetically expensive (2,8,9). Approximately 70–80% of total energy consumption in the normal cerebral cortex is used by glutamatergic neurons and their associated glia, which contain the majority of the large glutamate pool previously identified by 14C-labeling studies (10,11). The high fraction of energy associated with glutamatergic neurons is not surprising, given that they compose a large fraction of cortical neurons and synapses.
A large fraction of the glutamate released by neurons is taken up by surrounding astrocytes and converted to glutamine (12–15). Glutamine is then returned to glutamatergic neurons, which lack the enzymatic capacity to resynthesize glutamate lost through neurotransmission from glucose without depleting mitochondrial stores of tricarboxylic acid (TCA) cycle intermediates (15–17). The complete pathway is called the glutamate–glutamine cycle. Studies in the rat cerebral cortex have shown that the rate of the glutamate–glutamine cycle is coupled in a close to 1:1 ratio to neuronal (primarily glutamatergic) glucose oxidation above the rate measured with an isoelectric electroencephalogram (EEG) (4). There is a highly significant association between electrical activity measured by using the EEG, the rate of glucose consumption, and glutamate–glutamine cycling from deep pentobarbital anesthesia through nicotine-induced and evoked-potential–induced cortical activation (2,4,8,9). In vivo measurements in the human cerebral cortex have found a ratio consistent with the findings in the animal model (6,7). The relation measured in vivo is consistent with cellular studies that have shown that the major energy-requiring processes for glutamatergic function (pyramidal cell action potentials, before and after synaptic ion pumping, and glial glutamate uptake) are to a first order proportional to both the glutamatergic pyramidal cell firing rate and resultant glutamate release (18).
Several findings suggest that energy metabolism and the glutamate–glutamine cycle may be impaired in epilepsy. Measurements using positron emission tomography (PET) of patients with anterior mesial temporal lobe epilepsy (MTLE) often show interictal hypometabolism centered in the seizure focus (19–21). In vivo microdialysis studies show that seizure-associated glutamate release is doubled in the epileptogenic human hippocampus (22). Postictal glutamate clearance is slowed nearly threefold. These findings suggest that glutamate stores released by seizure activity are increased, and postictal clearance mechanisms (e.g., the glutamate–glutamine cycle) are less effective in the epileptogenic region. Defects in the glutamate–glutamine cycle could reflect mitochondrial dysfunction in neuron or glia or both. In this study we measured the rate of glutamate–glutamine cycling with respect to the neuronal rate of energy metabolism in the TCA cycle in the epileptogenic human hippocampus. Our hypotheses were that energy metabolism and the glutamate–glutamine cycle may be impaired in epilepsy. Some of the data were reported in an abstract (23).
Seventeen patients (five men) with TLE selected for resection of the hippocampus were invited to participate in this project. All tissues were resected for therapeutic reasons, not experimental ones. The median age was 35 years (interquartile, 22–41; range, 18–53 years). Patient characteristics (Table 1) and pathology were obtained from the medical record. The experimental protocol was approved by the Yale University Human Investigations Committee. Informed consent was obtained before surgery from all subjects.
Left hippocampal atrophy, increased signal in left hippocampus
CBZ, LTG, PHT
Right medial temporal sclerosis
Bilateral hippocampal atrophy
Left hippocampal atrophy
Left hippocampal atrophy
Right hippocampal atrophy, encephalomalacia in right frontal lobe
CBZ, LTG, PHT
Increased signal in right hippocampus, some cortical atrophy; prominent ventricles
Right hippocampal atrophy, increased signal in right hippocampus
Right hippocampal atrophy
Left hippocampal atrophy, increased signal in left hippocampal body
CBZ, LTG, TPM
Left posterior inferior temporal lobe tumor
No significant neuronal loss
CBZ, TPM, VPA
Small enhancing foci in right inferior frontal and temporal lobe regions
No significant neuronal loss
GBP, PHT, TPM
No significant neuronal loss
Right hippocampal atrophy; increased signal
No significant neuronal loss
Right cystic lesion in amygdala
No significant neuronal loss
A 20% solution of 2-13C-glucose was infused at a rate of 4 g/h, before the resection of the epileptogenic hippocampus. Blood glucose isotopic fractions were measured every 30 min. Blood glucose increased from 4.5 mM (interquartile, 4.1–5.4) to 6.8 mM (interquartile, 5.8–7.2; n = 17) after 2.2 h of infusion and remained stable thereafter. Hippocampal biopsies were obtained 4.3 h (interquartile, 3.8–4.9) after the start of the infusion of 2-13C-glucose with a steady-state isotope enrichment of 26% (interquartile, 19–29%). With the blood supply intact, a sample of the pes of the epileptogenic hippocampus was removed and frozen in carbon dioxide snow. Median weight of the hippocampal sample was 0.6 g (interquartile, 0.3–1.1; n = 17).
Samples were extracted with 3 ml cold 15% (1.5 M) perchloric acid and centrifuged for 15 min at 3,000 g (29,419 newtons) at 4°C. The supernatants were brought to neutral pH with a solution (1–1.5 ml) containing 100 mM K2HPO4, 500 mM KCl, and 9 M KOH. Centrifuged for 10 min at 3,000 g, the neutral supernatants were treated with Chelex, filtered, and lyophilized. The dried powders were dissolved in neutral 50 mM deuterated phosphate in D2O.
Brain metabolites and isotopomer concentrations were measured using high-field analytical MRS (24,25). Proton-spectroscopy was performed on a Bruker AM 500 spectrometer at 500 MHz and 25°C using 30-degree pulse width (5 μs), 6-s repetition time, 6,024-Hz sweep width, 32 K digital resolution, with and without broadband 13C-decoupling (WALTZ-16). Absolute metabolite concentrations were measured by comparing the intensity of the identified resonances with that of a known amount dichloracetic acid and isopropanol.
13 C-spectroscopy was performed on a Bruker AM 500 spectrometer at 125.77 MHz at 25°C by using 90-degree pulse width (7 μs), 2-s repetition time, 35,714-Hz sweep width, 32 K time-domain sampling, and dual power level, broadband 1 H-decoupling (WALTZ-16). Metabolite concentrations were measured by comparing the intensity of the identified resonances with the resonances of isopropanol, with appropriate corrections for differences in T 1 and nuclear Overhauser effect (NOE). There is good agreement with the correction factors calculated from T 1 and NOE measurements and the empiric correction factor ratios measured in model solutions (standards) (24,25) . Representative high-resolution proton and carbon spectra of the perchloric acid extracts of the hippocampal biopsies are shown in Fig. 1 .
In our previous studies in humans and rats, we focused on measuring rates of brain metabolism from the time course of 13C labeling of the C4 positions of glutamate and glutamine during a 1-13C-glucose infusion (4,6,26,27). In this study, the labeling from only one time point may be obtained. To determine whether glutamate–glutamine cycling is impaired in MTLE, we developed a strategy for determining the ratio of the glutamate–glutamine cycle (Vcycle) to the rate of the total glutamate synthesis rate, TCA cycle (Vtca), and total glutamine synthesis (Vgs) from steady-state 13C-labeling patterns in glutamate and glutamine. The strategy takes advantage of the label from 2-13C-glucose being incorporated into the internal positions of glutamate and glutamine only through the glia (5). Label from 2-13C-glucose, which enters the TCA cycle through pyruvate dehydrogenase (PDH), is incorporated into 5-13C-glutamate and 1-13C-glutamate. It does not label the internal positions of glutamate or glutamine. In contrast, 13C-label entering the TCA cycle from the anaplerotic pathway through pyruvate carboxylase will label glial glutamine C2 and C3 initially because both pyruvate carboxylase (PC) and glutamine synthetase (GS) are found exclusively in glia. Subsequently labeled glutamine will be taken up by the neurons to replenish released glutamate by the glutamate–glutamine cycle. The labeling in neuronal C3 glutamate is diluted relative to the precursor C3 glutamine because of unlabeled carbons entering through the neuronal TCA cycle (Vunlabeled). From the ratio of C3 glutamine to C3 glutamate labeling at isotopic steady state, the ratio of the rate of glutamate–glutamine cycle to neuronal TCA cycle (Vcycle/Vtca) may be calculated by using the equations given later (5). We have validated this strategy in the rat cerebral cortex by comparison of the rates calculated from the steady state 2-13C-glucose experiment with those calculated from a dynamic 1-13C- and 2-13C-glucose study (5). Furthermore, similar results have been obtained by using 14C-labeled carbon dioxide, which also incorporates label through PC (28).
Although in principle, the steady-state labeling from a C1 glucose experiment could be used to determine the same information, the analysis is complicated by both the flux through PDH and PC bringing label into the internal positions of glutamate and glutamine. Therefore even though the total labeling in the internal positions of glutamate and glutamine is larger, the sensitivity to PC and, as a consequence, Vcycle is reduced.
The diagram in Fig. 2 illustrates the metabolic model used to analyze the labeling data. The model has been tested extensively by using 13C and 15N MRS in the rat (4,5,27,29). The model is a mathematical description of the glutamate–glutamine cycle with the imposition of constraints obtained from carbon and nitrogen mass balance requirements. The isotopic enrichments at the C1, C2, C3, C4, and C5 positions of glutamate and glutamine were measured at close to steady-state conditions in surgically excised tissue. The 13C-label initially flows into neuronal C3 glutamate from C3 glutamine only via the glutamate–glutamine cycle. Label leaves the glutamate pool via both the neuronal TCA cycle (due to rapid exchange with α-ketoglutarate, the glutamate pool acts isotopically as if it were an intermediate in the TCA cycle) and the glutamate–glutamine cycle. Later, an additional label source will be 3-13C-glutamate that recycles in the TCA cycle. In addition, the pentose shunt will scramble label from 2-13C-glucose into 3-13C-pyruvate, which will introduce label into 4-13C-glutamate via PDH. At isotopic steady state, the label inflows and effluxes from the glutamate and glutamine pools must be equal, and the isotopic enrichment of 3-13C-glutamate is determined by the following isotopic balance equation.
where GN3 is the isotopic enrichment of the C3 position of glutamine, and GT3 is the isotopic enrichment of C3 glutamate. Rearrangement of this expression and solving for Vcycle/Vtca yields
Because tissue glutamate concentrations primarily reflect the glutamate content of glutamatergic neurons, the variable Vtca monitors the mitochondrial activity of the glutamatergic neuron.
Our choice of the 2-13C-glucose isotope is motivated by its applicability to steady-state analysis. Isotopic label entering through PC appears in 3-13C-glutamine and 3-13C-glutamate, whereas label entering through PDH appears in 5-13C-glutamate and 5-13C-glutamine (5,30,31). The relative metabolic rates of these two pathways labeling the glutamate content of the brain reflect the de novo synthesis of glutamate from glucose and, conversely, the fraction of the glutamate content that has been oxidized or otherwise lost (16,17,28,31). This ratio may be expressed as
This formulation measures the relative contributions of PC and PDH to the synthesis of the entire glutamate pool. This ratio monitors primarily the metabolic activity of neurons under physiologic conditions, because the rate of glutamate synthesis is dominated by glutamatergic neurons, with lesser contributions from GABAergic neurons and glia.
A similar ratio may be calculated for the entire glutamine pool, which should reflect at steady-state the relative contribution of PC and PDH to the glutamate pool that forms the substrate for glutamine synthetase. Only glia contain glutamine synthetase; therefore this ratio monitors the metabolic activity of glial glutamate.
Two-sample t-tests assuming unequal variances were used to assess significance. Scattergrams and product–moment or rank-order correlation coefficients were used to test for associations. Two-tailed parametric statistics were used to determine 95% confidence intervals (95%CIs) of mean values and interquartiles for median values (32).
As shown in Fig. 3, the relative ratio of the glutamate–glutamine cycle to the neuronal TCA cycle was decreased in biopsies affected by hippocampal sclerosis (mean, 0.08; 95%CI, 0.04–0.12; n = 12; p < 0.005) compared with those with minimal neuron loss (0.52; 95%CI, 0.30–0.75; n = 5). No associations with antiepileptic drugs (AEDs) in use at the time of surgery, duration of the epilepsy, gender, or age were seen that did not reflect the pathology. There were no significant associations between glutamate–glutamine cycle and the neuronal TCA cycle ratio and hippocampal glutamate or glutamine content.
The fraction of the total cellular glutamate pool synthesized by way of glial PC with respect to pyruvate dehydrogenase (glutamate–PC/PDH) was higher in hippocampal tissue with minimal neuron loss (mean, 0.14; 95%CI, 0.06–0.23; n = 5; p < 0.04) than in hippocampal sclerosis (0.05; 95%CI, 0.02–0.09; n = 12). Similarly, the fraction of cellular glutamine synthesized by way of glial PC with respect to PDH (glutamine–PC/PDH) was the same in hippocampal tissue with minimal neuron loss (mean, 0.28; 95%CI, 0.17–0.39; n = 5; p ∼ 0.9) than in hippocampal sclerosis (0.27; 95%CI, 0.16–0.38; n = 12). There were no associations with cellular glutamate or glutamine for either ratio.
Mean cellular glutamate concentrations were marginally higher (p ∼ 0.08) in biopsies showing minimal neuronal loss (8.9 mM; 95%CI, 7.4–10.5; n = 5) than in those with hippocampal sclerosis (7.3 mM; 95%CI, 6.0–9.7; n = 12). Mean cellular glutamine concentrations were the same (p ∼ 0.92) in biopsies showing minimal neuronal loss (2.7 mM; 95%CI, 1.4–4.1; n = 5) and those with hippocampal sclerosis (2.8 mM; 95%CI, 2.4–3.2; n = 12). There was no association between cellular glutamate and glutamine concentrations. No associations between cellular glutamate or glutamine concentrations and AEDs in use at the time of surgery, duration of the epilepsy, gender, or age were seen that did not reflect the pathology.
Overall our values for cellular glutamate are higher (mean, 7.8 M; 95%CI, 6.7–8.8; n = 17) than values reported from a similar epilepsy-surgery series (6.5 μmol/g wet weight in histopathologically unremarkable hippocampus and 4.0 in hippocampal sclerosis), but show a similar pattern, lower in the gliotic hippocampus, reflecting the loss of glutamatergic neurons (33). Mean cellular glutamate concentrations of hippocampus obtained at autopsy from nondemented subjects ranged from 5.3 to 8.2 mM (mean, 7.0; 95%CI, 5.6–8.5; n = 5) (34–38). Compared with the autopsy series mean, cellular glutamate is elevated in epileptic hippocampi with minimal neuron loss (p < 0.04). Surprisingly, cellular glutamate concentrations were the same in our specimens with hippocampal sclerosis as in histologically normal-appearing hippocampi obtained at autopsy (p ∼ 0.75). Brain glutamate content is heavily weighted toward the cellular concentration in glutamatergic neurons; glial glutamate content is normally very low (11,39–41). Cellular glutamate decreases in proportion to neuronal loss or simplification (loss of neuronal volume through shrinkage of dendrite, synapses, and other processes). The neuron loss, particularly of large glutamatergic neurons, and glial proliferation should decrease glutamate concentrations in the sclerotic hippocampus. Our findings suggest that there appears to be a relative increase in cellular glutamate content in the epileptogenic human hippocampus.
Values for tissue glutamine concentrations from epileptic human hippocampus do not appear to have been published. Our cellular glutamine values (mean, 2.8 mM; 95%CI, 2.4–3.2; n = 17; p < 0.03) appear to be low compared with the cellular glutamine concentration (3.8 mM; 95%CI, 2.8–4.8; n = 7) of the hippocampus obtained at autopsy from nondemented subjects (34). Brain tissue glutamine content reflects primarily glial concentrations (39,41,42). The lack of association between cellular glutamine and pathology, reflecting the degree of glial proliferation and neuron loss, suggests that glial glutamine levels are low in most cases of hippocampal sclerosis.
Measurements of the glutamate–glutamine cycle to the neuronal TCA cycle ratio in the normal human hippocampus are not available. The closest comparison with our results is recent measurements using in vivo 13C-MRS in the human occipital–parietal lobe and the awake and lightly anesthetized rat forebrain (4,6,43). Under low-light, unstimulated conditions, values ranged between 0.4 to 0.5. The relative ratio of the glutamate–glutamine cycle to glutamate synthesis is decreased in epileptic hippocampi that show sclerosis. The association with histopathology is striking. Neuron–glia cycling is not low in all epileptic hippocampi; it is lowest in those with significant loss of neurons and glial proliferation. Because this ratio is calculated from the relative labeling of the glutamate (primarily neuronal) and glutamine (synthesized in glia) pools, as opposed to absolute flux rates, the low glutamate–glutamine cycle to the neuronal TCA cycle ratio is not simply due to reduced cellular density or generalized hypometabolism. If the remaining neurons and glia were functioning normally, the ratio of glutamate–glutamine cycle to the neuronal TCA cycle would be independent of neuronal loss.
Although the glutamate–glutamine cycle model is explicitly described here, as discussed in detail in a recent article by Sibson et al. (5), the ratio of glutamate–glutamine cycle to the neuronal TCA cycle calculated from steady-state labeling is independent of the pathway by which glutamate precursors enter the neuron from the astrocyte (28). Because steady-state labeling analysis is used, the results are to a first order independent of whether label exchange between the mitochondrial and cytosolic glutamate pools is rapid relative to the TCA cycle, in contrast to the analysis of 13C and 14C time-course data (4–7). The main assumptions in the analysis of labeling results are that pyruvate carboxylase is localized to the astrocyte and that the majority of glutamate is in the neuron (5,28,43). To the extent that the gliotic hippocampi have elevated relative amounts of astrocytic glutamate, the degree of impairment of glial neuronal glutamate cycling will be greater than that calculated from Eq. 2.
These observations are confirmed by the significantly lower glutamate–PC/PDH ratio seen in hippocampal sclerosis. With rapid glutamate–glutamine cycling, the PC/PDH ratio for glutamate and glutamine should become similar. As glutamatergic neurons are lost and glia proliferate, the glutamate–PC/PDH ratio becomes more heavily weighted by the contribution from glia. The glutamate–PC/PDH ratio should approach the value for the glutamine–PC/PDH ratio as neurons are lost. The glutamine–PC/PDH ratio is the same for minimal neuron loss and hippocampal sclerosis, but the glutamate–PC/PDH ratio is far lower in hippocampal sclerosis. Therefore, neuron–glia cycling must be reduced in hippocampal sclerosis.
Elevated cellular glutamate and impaired clearance by the glia could contribute to increased cerebral excitability and ongoing excitotoxicity (44–46). The high cellular glutamate would result in enhanced glutamate release from the epileptic hippocampus during a seizure (22). Our observations are in line with studies of the epileptic human neocortex, showing elevated cellular glutamate in regions that show the greatest interictal activity (spiking) (45,46). Studies using microdialysis reveal enhanced elevation of extracellular glutamate during a seizure and impaired glutamate clearance in the epileptogenic hippocampus in mesial temporal sclerosis that may contribute strongly to increased cerebral excitability and ongoing excitotoxicity (13,22,44–46). These findings may be explained by impaired clearance of released glutamate, or enhanced release, possibly due to alterations in glutamate homeostasis.
Our primary observation is that in the epileptogenic human hippocampus resected at surgery, the rate of glutamate–glutamine cycling is very low, despite the normal hippocampal glutamate content, and below normal cellular glutamine. The lowest rates of glutamate–glutamine cycling are measured in hippocampal sclerosis. We hypothesize that the low rates of neuron–glia cycling are caused by glial dysfunction, downregulating glutamine synthetase (Fig. 4). Partial failure of glutamine synthesis would result in increased glial glutamate content, which could contribute to slowed glutamate uptake and enhanced glutamate transporter reversal. In the epileptogenic human hippocampus, glutamate transporter expression appears to be increased (47). It is possible that these glutamate transporters are not fully functional, contributing to slow glutamate–glutamine cycling. However, our observations of unexpectedly high glutamate and below-normal glutamine would not be explained completely by glutamate transporter dysfunction alone. Similar objections apply to hypotheses invoking glutamine transporter dysfunction or dysregulation of phosphate-activated glutaminase (PAG) or glutamate dehydrogenase (GDH). Such problems would be expected reduce tissue glutamate content and possibly increase glutamine content (16,48,49).
Inhibitors of glutamine synthesis (e.g., methionine sulfoximine) result in convulsions (50). In the ferric chloride model of focal epilepsy, glutamine synthetase activity is reduced in the seizure focus (51). Inhibition of glutamine synthesis would be expected to slow glutamate–glutamine cycling markedly, decrease tissue glutamine content, and release and increase glial glutamate content and efflux (45,50). During glial depolarization, enhanced glutamate efflux would be expected to foster spread of the seizure. Low rates of glutamate–glutamine cycling resulting from dysfunction of glutamine synthesis could contribute to the epileptic state, even under conditions of severe neuron loss and gliosis.
This study demonstrates that useful metabolic information may be obtained from the epileptogenic hippocampus at surgery. We conclude that the epileptic state with hippocampal sclerosis is characterized by slow glutamate–glutamine cycling and a relative increase in hippocampal glutamate content. The slow glutamate cycling is characteristic of low rates of glutamatergic neurotransmission that, in this situation, may contribute to focal epileptic encephalopathy (2,4,5,27). We hypothesize that the low rate of glutamate–glutamine cycling could promote epileptogenicity by a failure to clear extracellular glutamate rapidly (22,40,45,46,52). We further suggest that the key defect lies in glia (39,52,53). The low hippocampal glutamine content and slow glutamate–glutamine cycling point to dysregulation of glutamine synthesis as a likely possibility.
Acknowledgment: Grant support was provided by National Institutes of Health (NINDS) PO1-NS39092. We thank the anesthesia staff, especially Dr. Keith J. Ruskin, for their help with the infusion and blood sampling.