Abstract
For years, the field of brain energy metabolism has been dominated by the concept that glucose was the sole energy substrate used by adult brain cells to sustain neural activity (Sokoloff, 1989). Moreover, it was determined from oxygen consumption and carbon dioxide production rates that the respiratory quotient of the brain is close to 1, indicating that oxidative phosphorylation was by far the predominant process to generate energy in the central nervous system. Because the ratio of O2/glucose use is near 6.0, almost complete oxidation of glucose takes place to provide most of the energy needed by the brain, leaving relatively little importance to other pathways. With time, these statements have been raised to the status of central dogma of brain energy metabolism and have remained almost undisputed ever since, despite episodic reports of putative exceptions. Part of this situation could be explained by the stunning success of the 2-deoxyglucose autoradiography technique that was later adapted for human brain imaging with the advent of positron emission tomography (PET). Application of this new tool to explore brain functions also led to a relative disinterest in the fundamental features of brain energy metabolism, with the assumption that the previously established principles safely form the unquestionable basis upon which these newly emerging brain imaging techniques can rely.
A first set of studies that questioned these relationships among brain metabolic parameters came from the imaging field rather than from neurochemists and, more specifically, from the group of Marcus Raichle at Washington University (Fox and Raichle, 1986; Fox et al., 1988). From parallel measurements of blood flow changes, glucose use and oxygen consumption by PET in humans, Raichle and colleagues observed an uncoupling between oxygen consumption on one hand and glucose use or blood flow changes on the other. These observations were interpreted as evidence for a predominant enhancement of glycolysis in activated areas that would provide the energy necessary for neuronal activity. The occurrence of a transient elevation in lactate levels within the activated area was later evidenced in humans by magnetic resonance spectroscopy (Prichard et al., 1991; Sappey-Marinier et al., 1992), suggesting the possibility of using the mismatch between blood flow changes and low oxygen consumption enhancement to image localized increases in neuronal activity on the basis of the predicted changes in oxy/deoxyhemoglobin ratio. This proposal directly led to the development of functional magnetic resonance imaging (fMRI) (Kwong et al., 1992; Ogawa et al., 1992).
Using a completely different approach, our group came to realize that a particular type of glial cells, the astrocytes, might play a central role in regulating brain energy metabolism. Of particular interest was the observation that astrocytes responded to a prominent index of neuronal activity, the excitatory neurotransmitter glutamate, by enhancing both glucose use and lactate production (Pellerin and Magistretti, 1994). Our original proposal was that this phenomenon could contribute to the 2-deoxyglucose uptake response observed in brain imaging. A logical corollary to this hypothesis was that if some glucose is glycolytically used by astrocytes (with the production of lactate), neurons may receive from astrocytes a metabolic intermediate to complement their own energy substrate use (presumably fulfilled entirely by glucose otherwise). Lactate provided an obvious candidate for this role and the existence of a transfer process between astrocytes and neurons was proposed, a concept that came to be known as the astrocyte-neuron lactate shuttle (Bittar et al., 1996; Pellerin et al., 1998). This hypothesis, which was also based upon a number of observations from different in vivo and in vitro preparations, led to several new experiments that turned out to provide findings consistent with it (e.g., see Magistretti and Pellerin, 1999). However, much remains to be done to explore the full extent and possible variations of the model proposed. For example, the contribution of lactate under basal conditions remains to be evaluated, since in the model we had proposed initially, glucose is the exclusive source under such basal conditions (see, in particular, Pellerin and Magistretti, 1994). The proposed model is, we think, rather useful as it forces us to challenge and reevaluate established concepts, an essential step in scientific progress.
In this issue, Chih and Roberts (2003) present their own critical appraisal of the most common data supporting the lactate shuttle hypothesis. The important point here is not so much to decide, based upon the actual pieces of evidence, whether an hypothesis is right or wrong but rather to point out what is heuristically valid in it, what have we learned, what remains to be assessed, what new hypotheses can be proposed, and which experiments are critical for this, a task to which our group is fully committed. To take an example, the question still remains open as to why astrocytes process glucose glycolytically with lactate production, in face of the fact that they possess mitochondria and express an active oxidative phosphorylation. As a starting point, it is of critical importance to clearly define what the concept of the astrocyte-neuron lactate shuttle really represents and what it does not.
Definition of the Astrocyte-Neuron Lactate Shuttle
In the past few years, a number of different misleading interpretations have been made of what constitutes the astrocyte-neuron lactate shuttle. Although it could be admitted that such a concept evolves in time with new evidence, the time has come to clarify its present status. The first component of the shuttle is constituted by the metabolic response of astrocytes to glutamate at excitatory synapses. An enhancement of aerobic glycolysis, defined as preferential glucose use and lactate production despite sufficient oxygen levels to support oxidative phosphorylation, occurs in astrocytes in response to neuronal activation at glutamatergic synapses. It has never been inferred that all glucose must be converted to lactate by astrocytes before reaching neurons. This is neither true for the basal state nor during an activation. According to the astrocyte-neuron lactate shuttle hypothesis, glucose is directly used by neurons at rest. This view has been an integral part of the hypothesis since its initial formulation (see for reference Fig. 5 of Pellerin and Magistretti, 1994). It is in fact not excluded that an enhancement of glucose use in neurons could also take place upon activation. However, what recent in vivo data convincingly support (see below) is the contribution of astrocytes both to the enhancement in glucose use and lactate formation. Lactate is released in the extracellular space where it adds to a preexisting extracellular pool. At the moment, there is no direct evidence but also no requirement for lactate released by astrocytes to be directly funneled to neurons in a very tight spatial relationship. It is probably not desirable because the existence of a large extracellular pool would effectively dampen rapid fluctuations and ensure a more steady supply.
The second part of the shuttle concerns lactate use by neurons. Although clear evidence has been provided for lactate use by neurons, this does not preclude glucose use by these cells, neither at rest nor upon activation as mentioned above. Nevertheless, concomitant lactate use by neurons, both at rest and upon activation, is most likely in view of the present evidence (see below). What needs to be determined is the exact proportion of each used under different conditions as well as the subcellular location where it occurs. The astrocyte-neuron lactate shuttle concept does not exclude that some processes might prefer to use energy generated from glucose through glycolysis (with or without lactate production), but considering both the importance of oxidative metabolism in providing the energy sufficient to cover the majority of neuronal energy needs as well as the preferential use of lactate as an oxidative substrate by neurons (see below), it is still not clear on which basis, supported by solid experimental evidence, lactate use by neurons upon activation can be so categorically excluded (Chih and Roberts, 2003).
The astrocyte-neuron lactate shuttle concept does not state and does not require a tight spatial and temporal coupling between its two component, that is, enhancement of aerobic glycolysis in astrocytes and enhancement of oxidative metabolism (with glucose and lactate as substrates) in neurons. This characteristic offers a possible explanation for the different mismatches or uncoupling observed in functional imaging and spectroscopy. For example, if they do not occur simultaneously, but depending upon which one will precede the other, either a transient extracellular lactate peak or a decrease in lactate concentration will be observed. In most cases, the net result will be lactate transfer, but again, it does not exclude that other aspects could contribute to these phenomena, such as the recent proposal that glycogen resynthesis accounts for at least part of the glucose taken up that is not oxidized in certain activation paradigms (Dienel et al., 2002).
Glucose as the sole energy substrate for the brain
The concept that glucose is the major energy substrate provided by the circulation to the adult brain seems to hold true, even if under a number of circumstances, other substrates can contribute significantly to cover at least part of brain energy needs. For example, it was recently suggested that the fatty acid octanoate could contribute up to 20% of total brain oxidative energy production (Ebert et al., 2003). Lactate and ketone bodies have been also demonstrated to provide a significant energy source for the brain during development but also in abnormal situations such as diabetes, prolonged starvation, or hypoglycemia. Interestingly, it was suggested that lactate produced by muscles during strenuous exercise could also be used by the brain as an oxidative substrate, based on measured arteriovenous differences (Ide et al., 2000). The concept that peripheral lactate can be used by the brain under physiologic conditions and influence brain energetics has found a stunning confirmation recently. Amiel and collaborators have demonstrated by FDG-PET in humans that raising plasma lactate to levels in the range of those reached during exercise leads to a significant reduction of glucose utilization by the brain (Smith et al., 2003). This observation also provides strong support for the notion that lactate is a useful energy substrate for neurons (see point below). Nevertheless, it seems quite indisputable that glucose is an essential substrate to sustain brain activity, and it remains doubtful that glucose can be entirely replaced satisfactorily by other substrates.
Lactate as an energy substrate for neurons
This debated issue has been reviewed recently (Bouzier-Sore et al., 2002; Pellerin, 2003) such that it is unnecessary to present it all again here. Suffice it to say that evidence accumulated over decades both in vitro and in vivo leaves no doubt that lactate is oxidized and used as an energy substrate by neurons. Moreover, it was clearly demonstrated recently that lactate is a preferential oxidative substrate for neurons but not for astrocytes, as compared with glucose (Bouzier-Sore et al., 2003; Itoh et al., 2003). Because the concentration of lactate in the extracellular space was found to be at least similar if not superior to glucose concentration under resting conditions (Abi-Saab et al., 2002), it is likely that a substantial part of neuronal energy needs will be covered by lactate. What remains to be determined is the proportion of glucose and lactate used by neurons at rest and whether it varies as a function of the level of activity, especially if one considers that extracellular lactate concentration tends to rise during activation.
Monocarboxylate transporters: What, where, and why?
The discovery of specific monocarboxylate transporter isoform expression on parenchymal cells in the adult brain was predicted by the postulate of a net lactate transfer between astrocytes and neurons. The recent confirmation of MCT2 being predominantly expressed in neurons whereas MCT1 and MCT4 are found instead in astrocytes is consistent with distinct roles in these two cell types (Bergersen et al., 2002; Debernardi et al., 2003; Pierre et al., 2002). Of course, it can not be concluded from the simple distribution of these isoforms that lactate is shuttled from one cell type to the other. Nevertheless, such a distribution begs for an explanation. Chih and Roberts propose that MCT2 expression in postsynaptic terminals could be used for lactate efflux and would indicate rather a predominance of glycolysis at this location. They support their contention by the observation that some glycolytic enzymes are also found at synapses. Although this possibility is interesting and can not be excluded, it remains speculative for the moment. Moreover, the presence of glycolytic enzymes could be unrelated to their enzymatic activity but rather play a structural role as was demonstrated previously in postsynaptic densities (Rogalski-Wilk and Cohen, 1997). It should be also reminded that MCT2 expression in neurons is not restricted to dendritic spines but is found upon the surface of several neuronal processes in the neuropil including dendrites. Because dendritic shaft contains numerous mitochondria and also forms synapses, it seems very likely that lactate will be used at least at this location, and could even fulfill dendritic spine needs, as mitochondrial ATP could migrate into spines (Sorra and Harris, 2000). Thus the presence of MCT2 remains mostly consistent with lactate use in neurons and correlates with oxidative metabolism. Moreover, in peripheral tissues such as testes, in which lactate has been recognized as an important energy substrate, MCT2 expression was found to predominate in cells using lactate as an energy source (Boussouar et al., 2003). The example of testes is quite instructive because, as hypothesized for the brain, one cell type, the Sertoli cell, produces lactate to metabolically support another one, the spermatid. Another interesting aspect also observed in testes is related to the control of MCT expression. It was recently demonstrated that MCT2 expression is regulated via specific neurotransmitter action in cultured cortical neurons (Pierre et al., 2003). This observation suggests that it could be possible to adjust monocarboxylate uptake capacity to match a certain level of neuronal activity. Such perspective deserves to be further explored.
Glial cells as detectors of synaptic activity and providers of energy substrates for neurons
The question of whether glial cells, and in particular astrocytes, can metabolically respond to neuronal activity and release energy substrates has been further addressed recently both in ex vivo preparations and in vivo. It was demonstrated both in knockout animals for glia-specific glutamate transporters or after an antisense treatment against one of these transporters that the enhanced glucose use response normally obtained upon neuronal activation was dependent upon the level of expression of such astrocyte-specific glutamate transporters (Cholet et al., 2001; Voutsinos-Porche et al., 2003). Combined with previous data showing that the elevation in lactate levels observed after neuronal activation is prevented by the use of glutamate transporter inhibitors (Demestre et al., 1997), these observations clearly establish not only that astrocytes respond to synaptic activity and participate in glucose use changes observed upon stimulation but also that they represent an important source of lactate production within the brain. Parallel work in ex vivo preparations has provided further insights into this particular role of glial cells. Using a particularly elegant approach, Coles and colleagues have determined the uptake and distribution of locally applied deoxyglucose, glucose, and lactate between the axon and associated Schwann cells (Vega et al., 2003). They came to the unexpected conclusion that a major part of glucose use (78%) occurs in glial cells. Considering the fact that energy needs of glial cells have been estimated to be only 5% of total energy expenditure, whereas neurons account for the rest, a likely explanation for this paradox is the release of metabolic intermediates by glial cells for use by neurons. Their own data point at lactate as the most likely candidate. In a different set of experiments performed in an optic nerve preparation, Ransom and colleagues demonstrated that glycogen contained in glial cells was mobilized during high frequency stimulation, and this was necessary to sustain action potentials in axons (Brown et al., 2003). Moreover, they concluded that a monocarboxylate such as lactate was most likely formed by glial cells from glycogen and transferred to axons to be used as an energy substrate.
A selection of interesting and unresolved questions
Despite the success of the lactate shuttle hypothesis to explain certain aspects related to brain energy metabolism, it has also raised a number of issues that remain unresolved for the moment. Among them is the question related to the demand of excitatory (glutamatergic) neurons versus neurons containing other neurotransmitters, especially inhibitory ones like GABA. This topic, and more specifically the contribution of inhibitory activity to imaging signals, has been the subject of debate recently (Arthur and Boniface, 2002; Raichle, 2001). Our most recent data seem to indicate that a mechanism similar to the one demonstrated for glutamate in astrocytes does not take place for GABA, thus suggesting that GABAergic activity does not contribute significantly to signals linked to energy metabolism (Chatton et al., 2003). Another critical issue is the contribution of nonoxidative versus oxidative glucose metabolism during brain activation. It seems now that a predominant increase of glycolysis over oxidative metabolism during activation greatly depends upon the type of stimulation. Although in several cases, a commensurate increase in oxygen consumption is eventually measured in parallel with an increase in glucose uptake, their temporal evolution are distinct, thus suggesting that glycolysis and oxidative metabolism could occur in separate phases. Regarding their cellular occurrence, current evidence points to astrocytes as the major source of lactate during activation. It is however not excluded that neurons could undergo a glycolytic phase (with lactate production) in parallel with enhanced oxidative metabolism or alternatively switch from one to the other. This scenario is not inconsistent with the lactate shuttle hypothesis, as long as lactate produced both by astrocytes and neurons is eventually oxidized by neurons. Indeed, the lactate shuttle hypothesis that we have proposed is fully consistent with the in vivo observations made over several decades that glucose is the almost exclusive energy source for the brain and that its full oxidation accounts for the energy production necessary to face the needs of neuronal activity. The lactate shuttle hypothesis introduces a transient glycolytic step at the earliest stage of activation with lactate production, followed by almost complete oxidation of lactate. Experimental evidence is consistent with the view that an initial glycolytic step occurs in astrocytes in response to synaptically released glutamate and that lactate oxidation occurs predominantly in neurons. This contrasts with the conventional hypothesis that proposes that lactate produced aerobically either by neurons, astrocytes, or both is washed out of the brain. Obviously, more data are needed to distinguish these different possibilities and better define under which circumstances they occur.
CONCLUSIONS
Clarifying the molecular and cellular mechanisms linking neuronal activity to different metabolic processes represented by glucose and oxygen consumption as well as blood flow changes is of fundamental importance, not only for basic knowledge of brain energy metabolism but also because it constitutes the basis upon which a number of functional brain imaging techniques in use both in fundamental and clinical investigations rely. Thus a better knowledge of these features bears the promise of improving the use of these techniques and the interpretation of the signals generated by them. Moreover, considering the suspected implication of disturbances in energy metabolism in a number of neurologic disorders (e.g., Alzheimer's disease), it could be of prime importance to improve our understanding in this field if we are to suggest possible avenues for treatment. Understanding the complexity underlying such processes represents, however, a daunting task. Modelization offers a means through which it could be possible to evaluate the interplay between several different parameters and determine their importance or participation as a function of various conditions, as was recently demonstrated (Aubert and Costalat, 2002). Notwithstanding, and as experience has proved in the past, we still need experimentalists who propose and test ideas on the basis of experimental observations.