Abstract
Ten years ago, Luc Pellerin and Pierre Magistretti in Lausanne (Switzerland) made the first observation that glutamate stimulates, in a dose-dependent manner, glucose uptake into cultured astrocytes. The second observation was that glucose was mainly metabolized to lactate, indicating net aerobic glycolysis. The last set of data defined the signaling pathway: glutamate was acting via its transporter not its receptors (Pellerin and Magistretti, 1994). As such, these data do not appear so revolutionary. So what are the reasons for the long-standing and often emotional debate on this issue (Gjedde et al, 2002; Chih and Roberts, 2003; Dienel and Cruz, 2004; Hertz, 2004)?
Undoubtedly, the main reason arises from the hypothesis that was put forward by Pellerin and Magistretti: ‘glutamate uptake-induced aerobic glycolysis into astrocytes is the cellular mechanism coupling neuronal activity to glucose utilization‘. What a gap it seems from data obtained on cultured astrocytes to a well-known physiologic response of the brain! In a timely manner, their work came at a moment of increasing interest in the general cell biology underlying the brain imaging signals and they should be recognized for proposing the first cellular hypothesis about neurometabolic coupling. It should be emphasized that until now, and despite the intense debate of their hypothesis, there is no alternative signaling pathway that links neuronal glutamatergic activity to glucose use that would fit with what has been observed using in vivo brain imaging. We still do not know with certainty which signaling pathway enables a tight adjustment of glucose use to meet the increasing needs of the glutamatergic synapse but the great advantage of the aforementioned hypothesis is that it paved the way for a large body of fundamental research on this topic.
A Look Back
A comprehensive and authoritative historical perspective of brain imaging has already been published (Raichle, 1998) and we will only underline selected data that are relevant to this viewpoint.
Neuronal activity requires energy that is provided almost exclusively by oxidation of glucose. Because the brain has very little energy reserve, a continuous vascular supply of glucose and oxygen is mandatory to sustain neuronal activity. This supply is regulated locally and dynamically to meet the increased energetic demand of functional activation. For decades, these principles have driven the study of brain energy metabolism and its relation to brain activity. They also provide the foundation for brain imaging methods such as positron emission tomography (PET). Using 2-[18F]fluoro-2-deoxy-
Although an increased lactate concentration had been reported during seizures in the early 1970s (Bolwig and Quistorff, 1973), many would have predicted that increases in local blood flow would be reflected in local increases in oxidative metabolism of glucose during physiologic activation. In the mid-1980s, evidence from brain imaging studies with PET has indicated otherwise. Fox and his colleagues showed that in normal, awake adult humans, visual or somatosensory stimulation results in dramatic increases in blood flow and glucose utilization but minimal increases in oxygen consumption. These results supported the ‘heretical’ notion that glucose consumption during increased neuronal activity might be nonoxidative (Fox and Raichle, 1986; Fox et al, 1988). Other groups using magnetic resonance spectroscopy (MRS) have confirmed and extended these paradoxical results by reporting lactate concentration increases during visual stimulation (Prichard et al, 1991; Sappey-Marinier et al, 1992). In fact, the most convincing support to these data can be found in the discovery that in vivo increases in blood oxygenation could be detected by magnetic resonance imaging (MRI) (Ogawa et al, 1990), a signal known as the blood-oxygen-level-dependent (BOLD) signal. It is very important to keep in mind that it is only because cerebral blood flow increase exceeds changes in oxygen consumption during neuronal activation that we are able to detect BOLD signal. If this was not the case, then the decrease in the amount of deoxyhemoglobin because of the large fractional increase in blood flow will not occur, so that no BOLD signal would be detected. Therefore, in the mid-1990s, a consensus was achieved, mainly based on in vivo observations, that at least part of the glucose used during neuronal activation could be consumed in a nonoxidative manner. Where do we go from here? Two major questions remained unanswered at that time: (1) Why do neurons produce lactate during activation and as a corollary, how could we reconcile this on a strict energetic level? (2) What is (are) the triggering signal(s)? Experiments performed on cultured astrocytes gave the first answers. (1) It is the astrocytes and not the neurons that produce lactate and lactate can be oxidized by neurons to produce sufficient ATP. (2) The triggering signal is glutamate uptake (more precisely, it is the elevation of intracellular sodium associated with glutamate uptake) into astrocytes.
During the last 10 years, a bulk of data has enriched these pioneering results (see Pellerin and Magistretti, 2004 for a review). However, two recent reviews have critically examined this astrocyte—neuron lactate shuttle hypothesis (Chih and Roberts, 2003; Hertz, 2004). While some arguments presented are certainly valid, they were mainly based on theoretical or in vitro data (including enzyme kinetics, thermodynamics, substrate availability, etc.) and it was concluded that the proposed shuttle process is either of little functional significance or uncertain. Although criticisms from a theoretical point of view could be interesting to stimulate the discussion, it is a great pity that these authors did not provide any alternative model to support the neurometabolic coupling. More disturbing, they tried to disprove recent experiments performed in vivo arguing that they cannot fit into their theoretical argumentation, a scientific approach that we do not support. While data obtained using modeling or cultured cells have provided invaluable contribution to dissect mechanisms and elaborate new working hypothesis, as models they have their own limitations. Obviously, the relevance of these results needs, at least at one point, to be tested and verified in vivo. It should be noted that almost all recent experiments performed ex vivo or in vivo, while being all open to criticisms, have provided data that rather support the lactate shuttle.
In Vivo Steps to the New Lactate Shuttle
What should be mentioned is that the astrocyte—neuron lactate shuttle hypothesis has been contemporaneous of the discovery that these star-like cells play an active role in regulating glutamatergic transmission among other key physiologic functions reported until now (Nedergaard et al, 2003). The importance of astrocytes for glutamatergic transmission comes first from the observation that neurons lack the enzyme pyruvate carboxylase so that they cannot perform net synthesis of glutamate from glucose, a metabolic pathway which takes place in astrocytes (see Hertz and Zielke, 2004 for references). The majority of the neuronal information is conveyed via the rapid excitatory glutamatergic system, which accounts for 80% to 90% of cortical synapses. Therefore, an important energy load is created by the operation of glutamatergic synapses. A growing body of experimental data indicates that their activity is tightly regulated by dynamic interactions between astrocytes and neurons (Haydon, 2001). The different steps in glutamatergic transmission include glutamate release, glutamate uptake and recycling by astrocytes (glutamate/glutamine cycling), and excitatory pre- and post-synaptic activity. Which among these processes are those participating to the signaling chain that links glutamate activity to glucose uptake?
Numerous (but not all, see Choi et al, 2002) in vivo MRS studies indeed suggest that the energy requirements of glutamatergic neurons potentially explain a large fraction of the cortical glucose consumption, and that this demand is coupled to glutamate release and recycling (Shulman et al, 2004). In brief, 13C MRS studies in the rat performed over a range of cortical activity showed that the coupling between glutamate/glutamine cycling and neuronal glucose oxidation in the cerebral cortex is close to 1 (Sibson et al, 1998). These findings suggest that the energy requirements of pre- and postsynaptic events related to synaptic transmission may be met by the oxidative metabolism of one mole of glucose per glutamate molecule released synaptically. Further, the basal energetic rate unrelated to neuronal activity (measured at isoelectricity) was only ~15% of the awake, resting rate, indicating that some 85% of cortical energy consumption is coupled to neuronal activity. It was then hypothesized that glycolytic energy supports the very rapid process of glutamate uptake by astrocytes, while oxidative metabolism might provide most of the energy required by processes involved in synaptic transmission at the level of the tri-partite glutamatergic synapse. Until now, no clear demonstration for such a dissociated process was available. A recent two-photon fluorescence imaging study provides illuminating evidence to support this view (Kasischke et al, 2004). These authors took advantage of the subcellular resolution achieved by multiphoton microscopy to monitor the intrinsic fluorescence of β-nicotinamide adenine dinucleotide (NADH), a sensitive indicator of both oxidative and glycolytic energy metabolism. Using hippocampal slices, they were able to provide the following results: (1) the presence of elevated cytoplasmic NADH in astrocytes compared with neurons under resting conditions indicates higher glycolytic capabilities of the former; (2) activation of Schaffer collateral pathway induced a biphasic metabolic response with an early NADH decrease (the so-called ‘dip’) followed by an increase (the so-called ‘overshoot’), as already observed in vivo; (3) these two responses are spatially anticolocalized, the dip being restricted to dendrites of neurons while the overshoot was of astrocytic origin. They concluded that during activation, a transient and rapid oxidative metabolism (NADH is first consumed by the respiratory chain and then produced during activation of the tricarboxylic cycle) takes place in dendrites and is followed by a late nonoxidative glycolytic metabolism (net production of NADH before conversion to NAD+) in astrocytes. This very elegant work provides strong evidence that in an integrated preparation (slice), energy metabolism is compartmentalized between astrocytes and neurons especially during increased neuronal activity. Because of the high temporal resolution achieved by this technique, it becomes possible to reconcile many of the previous contradictory results and to envisage a revised version (in terms of the sequence of events) of the astrocyte—neuron lactate shuttle hypothesis (Pellerin and Magistretti, 2004). The proposed new sequence of events (neuronal oxidative then astrocytic nonoxidative glycolysis) was already suggested based on in vivo high temporal measurements of lactate concentrations in the hippocampus during electrical activation of the perforant pathway (Hu and Wilson, 1997). However, one important issue that should be taken into account is the duration and intensity of the activation. Indeed, both components of NAD(P)H fluorescence (dip and overshoot) were previously shown to be linked to postsynaptic activity during a minor and short-lasting neuronal activity (Shuttleworth et al, 2003) while the late NADH overshoot is triggered by presynaptic activity (of astrocytic origin) during an intense and prolonged activation (Kasischke et al, 2004). This question should await new methodological tools with high spatial and temporal resolution to get a definitive answer. Finally, this new version of the model also implies that FDG-PET studies will measure glucose uptake into astrocytes and not neurons, because neurons oxidize lactate during activation (Schurr et al, 1999). Indeed, these authors provided evidence that in hippocampal slices in which neuronal lactate use was inhibited by the lactate transporter inhibitor α-cyano-4-hydroxycinnamate (4-CIN), activation by glutamate elicited a permanent loss of neuronal function, with a two- to three-fold increase in tissue lactate content. They concluded that lactate is a crucial aerobic energy substrate that enables neurons to endure activation.
According to the seminal observation on cultured astrocytes (Pellerin and Magistretti, 1994), glutamate uptake by specific glial glutamate transporters (GLAST and GLT-1) is the triggering signal for glucose uptake. Our in vivo experiments using [14C]2-deoxy-D-glucose and performed in living mice strongly support this notion. We indeed observed that the metabolic response to whisker stimulation was markedly reduced in the somatosensory cortex of 10-day-old (P10) knockout mutant (KO) mice for either GLAST or GLT-1, the two main astrocytic glutamate transporters (Voutsinos-Porche et al, 2003a). Mice were first studied at that age because the two transporters exhibit different developmental expressions (Danbolt, 2001), so that GLAST and GLT-1 are similarly expressed in the cortex during the second postnatal week while GLT-1 is the predominant transporter in adulthood. In contrast to what was asserted recently (Hertz, 2004), we indeed performed experiment in adult mutant mice too and reported that the inhibition of glucose uptake during whisker activation was only significant in GLT-1 KO mice (Voutsinos-Porche et al, 2003b). This result has been strongly questioned on the basis that these animals might present an altered glutamatergic transmission (Hertz, 2004). Indeed, it was originally published that homozygous GLT-1 KO mice gained weight more slowly than did wild type and tended to die prematurely (50% survival after 6 weeks) mainly during spontaneous convulsive seizures (Tanaka et al, 1997). While criticisms are often constructive, nothing is better than doing experiments to prove their relevance. At least in the young animals, glutamate transport was only partly affected (because the other transporter was present) and all our biochemical analyses of various glutamatergic markers have failed to show any alteration. Furthermore, we have found no significant decrease in glucose uptake after blockade of AMPA/KA and NMDA receptors in these P10 mice (Voutsinos-Porche et al, 2003a). Finally, concerning the adult GLT-1 KO mice (which were studied at the age of 4 to 6 weeks), the basal [14C]2-deoxy-
Enriching the Model with the Waves and the Flow
Besides neurometabolic coupling, neurovascular coupling is the other physiologic response of the brain that is relevant to functional brain imaging. Because astrocytes are linked to both phenomena (Leybaert, 2005), we will briefly discuss the latest results that are pertinent to our viewpoint. One of the most important features of astrocytes functioning is that these cells communicate with each other by the propagation of calcium elevation (Araque et al, 2001). In primary cell culture and brain slices, astrocytes have been shown to respond to glutamate activation or neuronal activity by rhythmic elevations of intracellular calcium that can spread to adjacent astrocytes as waves (Schipke and Kettenmann, 2004). The signaling mechanisms responsible for their propagation are the release of ATP by astrocytes and the gap junction connectivity (Charles and Giaume, 2002). The best-characterized consequence of such calcium elevations is the release of glutamate by regulated exocytosis (Bezzi et al, 2004). Recent findings have proposed two major consequences of these calcium waves that support the emerging view that astrocytes function as a network to ensure coordinated metabolic and vascular responses to neuronal activity. Using both sodium and calcium imaging in astrocytes, it was observed that these cells also exhibit sodium waves in parallel to calcium waves. Glutamate released by a calcium wave-dependent mechanism participates to the regenerative propagation of cytosolic sodium waves. Furthermore, these sodium waves give rise to a spatially correlated increase in glucose uptake (Bernardinelli et al, 2004).
The second set of observations is linked to the well-known and strategic position of astrocytes close to the vasculature (Kacem et al, 1998). While one of the first hypotheses on the putative role of astrocytes in directly regulating cerebral blood flow was elaborated in 1998 (Harder et al, 1998), more direct evidence has been provided only recently in acute cortical slices (Zonta et al, 2003) indicating that astrocytes do regulate microarteriole tone through a mechanism involving a pulsatile release of prostaglandins evoked by glutamate-induced calcium oscillations. Measurements of calcium dynamics using either two-photon imaging or laser scanning confocal microscopy have enriched the ‘vasoactive repertoire’ of astrocytes because these cells can induce vasoconstriction through the release of other derivatives of AA (Mulligan and MacVicar, 2004) and can also suppress calcium oscillations and vasomotion of intraparenchymal arterioles during glutamatergic activation (Filosa et al, 2004). Clearly, more experiments are needed to determine the role of these astrocytic networks in coordinating both vascular and metabolic responses to an increased neuronal activity.
Unanswered Questions
As usual, new data give rise to more questions than answers. As already mentioned, the identity of the glutamate transporter(s) involved is still not known with certainty and whether they do play a role in humans will be an important issue regarding functional brain imaging. A second and important question which deserves to be clarified is the following: does the astrocyte—neuron lactate shuttle apply to all levels of activity or only during activation? Indeed, it is well-known that neurons are endowed with glucose transporters (such as Glut3) and are able to oxidize glucose to produce ATP. Because neurons ‘swim’ in an extracellular pool of lactate (Hu and Wilson, 1997), they probably also use a certain proportion of lactate in parallel with glucose (for references about the use of lactate by neurons, see Pellerin, 2003). However, functional imaging studies provide significant metabolic/vascular signals only during the phasic changes in neural activity. One possibility would be that during activation (for a definition of a control or baseline against which the condition of interest can be compared, see Gusnard and Raichle, 2001), neurons would use more lactate than glucose. Why? We still do not know but it is noteworthy that the first step of conversion from lactate into pyruvate does not require ATP while glycolysis starts with the phosphorylation of glucose.
Conclusion
The entrance of astrocytes in the scene of brain energy metabolism and now blood flow has been invaluable to elaborate new and testable hypotheses to provide the biological bases of brain imaging. Without their active contribution, it would have been very difficult to identify solid signaling pathways that could account for the well-known metabolic and vascular responses that occur during functional activation. Up to now, the astrocyte—neuron lactate shuttle hypothesis still promises to help unravel important cellular and molecular aspects of the neurometabolic coupling.