Neural Activity Triggers Neuronal Oxidative Metabolism Followed by Astrocytic Glycolysis

See allHide authors and affiliations

Science  02 Jul 2004:
Vol. 305, Issue 5680, pp. 99-103
DOI: 10.1126/science.1096485


We have found that two-photon fluorescence imaging of nicotinamide adenine dinucleotide (NADH) provides the sensitivity and spatial three-dimensional resolution to resolve metabolic signatures in processes of astrocytes and neurons deep in highly scattering brain tissue slices. This functional imaging reveals spatiotemporal partitioning of glycolytic and oxidative metabolism between astrocytes and neurons during focal neural activity that establishes a unifying hypothesis for neurometabolic coupling in which early oxidative metabolism in neurons is eventually sustained by late activation of the astrocyte-neuron lactate shuttle. Our model integrates existing views of brain energy metabolism and is in accord with known macroscopic physiological changes in vivo.

Functional neuroimaging techniques such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) have provided valuable insights into the working brain and are widely used in both neuroscience and clinical neurology. However, fundamental questions related to the cellular and molecular aspects of neurometabolic and neurovascular coupling are still unresolved. For example, a PET study has demonstrated nonoxidative glucose consumption during focal neural activity (1), whereas other in vivo studies have reported oxidative metabolism (2, 3). Many of the discordant results might be related to methodological issues, differing stimulation protocols, and the restricted measurement of either oxidative or glycolytic metabolic signatures.

The in vitro observation that glutamate uptake stimulates astrocytic lactate secretion to feed neurons provides a mechanism for the coupling of neuronal activity to glucose utilization (4) and is supported by ex vivo and in vivo studies (57). However, the astrocyte-neuron lactate shuttle model has been challenged (8, 9), and the ability to resolve astrocytic from neuronal metabolism is needed to verify this hypothesis (6, 8, 10). Multiphoton microscopy (11) provides subcellular metabolic imaging (12, 13) by utilizing the intrinsic fluorescence of β-nicotinamide adenine dinucleotide (NADH) as a sensitive indicator of both oxidative (14) and glycolytic (15, 16) energy metabolism.

Our measurements were done on the well-known hippocampal slice preparation. The emission spectrum (Fig. 1A) and dynamics of the intrinsic tissue fluorescence seen upon pharmacological manipulation of the respiratory chain activity (Fig. 1B) show that the reduced form of the protein-bound pyridine nucleotides NADH and nicotinamide adenine dinucleotide phosphate (NADPH) (17) are the dominant fluorophores in the hippocampal neuropil. Taking into account analytical measurements of [NADH] and [NADPH] in rodent brain under resting (18) and under activated conditions (19, 20), we conclude that both the resting fluorescence and the fluorescence fluctuations are dominated by NADH. Although the two-photon action cross section of NADH in solution (Fig. 1A) is 1/100 to 1/1000 the magnitude of conventional fluorophores, direct two-photon excitation of intrinsic NADH fluorescence is still feasible. The concentration of NADH in vivo can be in the millimolar range and the intracellular environment substantially enhances its fluorescence. For example, we have measured >10-fold increases in the action cross section for NADH when bound to malatedehydrogenase and when in viscous medium.

Fig. 1.

Intrinsic fluorescence in the neuropil originates from NADH. (A) The two-photon (2P)-action cross section of NADH (open squares, top x axis, right y axis) overlaps with the 1P-excitation-spectrum of NADH (solid line, bottom x axis, left y axis) when plotted at twice the wavelength. The 2P-emission spectrum of intrinsic fluorescence from the neuropil of CA1 (dotted line, bottom x axis, left y axis) is blue-shifted by ∼20 nm compared with the spectrum of aqueous NADH (solid line); this indicates protein-binding of NADH (17). (B) Cyanide was applied to acute slices in concentrations of 0, 0.1, 0.25, 0.75, 1.0, and 2.0 mM. The dose-dependent response proves that this is NADH (each group n = 5 to 6, average ± SEM). (Inset) Dynamics of the NADH fluorescence response under cyanide perfusion for 5 min.

Pyramidal neurons and the neuropil exhibited a granular fluorescence pattern that colocalized with Rhodamine 123, which implies a mitochondrial origin for NADH. The remaining homogeneous fluorescence in the neuronal somas and nuclei was dim. Brightly fluorescent star-shaped cells were disseminated throughout the hippocampal neuropil (Fig. 2A). We identified these cells as protoplasmic astrocytes by glial fibrillary acidic protein (GFAP) immunolabeling and by using transgenic mice expressing green fluorescent protein (GFP) under the control of the GFAP promoter (Fig. 2B). The 1:1 correlation of the bright intrinsic fluorescence from the star-shaped cells with GFP demonstrates that the intrinsic signal alone is sufficient for the identification of astrocytes and their processes (Fig. 2C).

Fig. 2.

Astrocytes exhibit a characteristic NADH fluorescence pattern. (A to C) Astrocytes can be identified by means of their characteristic NADH fluorescence pattern as demonstrated by the 1:1 correlation of the disseminated, intrinsically bright fluorescent cells in the neuropil of CA1 (stratum radiatum) with GFP-expressing astrocytes. (D to F) A GFP-expressing astrocyte shown at high magnification with confluent cytoplasmic NADH fluorescence and granular mitochondrial NADH fluorescence (arrowheads) along its processes. (G) Distribution of mitochondria (cytochrome oxidase subunit IV immunolabeling) in the neuropil of CA1 (stratum radiatum). (H) Additional GFAP-labeling of the same section shows that the concentration of mitochondria in astrocytes is not elevated in comparison with the surrounding neuronal neuropil. (I) Astrocytes (GFAP immunolabeling) and dendrites (MAP-2 immunolabeling) are the two dominant compartments in the neuropil of the stratum radiatum of CA1. Nuclei were counterstained with DAPI in (G to I). Scale bars, 20 μm.

Astrocytes responded to brief hypoxic episodes (5 min) with a concurrent transient elevation of their intrinsic fluorescence, indicating NADH as the source. We could not detect any significant difference in onset, amplitude, and recovery of the response to hypoxia between neighboring astrocytes and neurons at a temporal resolution of 30 s. The fluorescence texture in astrocytes was uniform, with elevated levels in the cytoplasm and nucleus relative to neurons, implying a substantial cytoplasmic NADH fraction. Large and bright mitochondria were apparent in astrocytes and along their processes (Fig. 2, D to F). Using the GFP-GFAP fluorescence as an image-processing mask, we determined that the level of the astrocytic NADH fluorescence exceeded that of the surrounding neuronal neuropil in a given optical section by ∼60%. Immunolabeling of cytochrome oxidase and GFAP confirmed the presence of large mitochondria in astrocytes and demonstrated that the confluent intrinsic fluorescence pattern in astrocytes cannot be accounted for by the mitochondrial NADH (Fig. 2, G and H). The presence of elevated cytoplasmic NADH in an astrocyte compared with a neuron under resting conditions indicates higher glycolytic capacities in the astrocytes. Comparing immunolabeling of microtubule-associated protein-2 (MAP-2) and GFAP also showed that dendrites and astrocytes are the dominant compartments in the stratum radiatum of CA1 (Fig. 2I).

Our ability to identify astrocytes within the neuropil solely by means of their characteristic intrinsic fluorescence pattern provides the opportunity to investigate activity-dependent metabolism in glial and neuronal compartments. Activation of the Schaffer collateral pathway by extracellular stimulation (32 Hz for 20 s) induced a robust biphasic activity-dependent metabolic response that has been observed in numerous ex vivo and in vivo studies (2123) with an early 3 to 4% decrease in NADH fluorescence with a minimum after ∼10 s (the “NADH dip”) and a subsequent ∼3% increase (the “NADH overshoot”) peaking at ∼60 s (Fig. 3A). The time course of both peaks was independent of the stimulus duration (5, 20, 60 or 240 s). However, with longer periods of stimulation (60 and 240 s) the overshoot was sustained in both amplitude and duration (Fig. 3A, inset). Biphasic changes in NADH fluorescence (above 12%) were observed in the stratum radiatum of CA1, which indicated that the response is a consequence of synaptic activity in the Schaffer collateral pathway (Fig. 3, B and C). Surprisingly, the NADH dip and the NADH overshoot were highly anticolocalized (Fig. 3D) with a linear correlation coefficient (r) of –0.84 and a coefficient of determination (r2) of 0.71, which suggested that the biphasic NADH response is due to different cell types and even their compartments.

Fig. 3.

The early and late phases of activity-dependent metabolism are anticolocalized. (A) Activation of the Schaffer collateral pathway (32 Hz for 20 s) induces a biphasic NADH fluorescence response in the neuropil (n = 8, average ± SEM). The inset shows the response after stimuli were applied for longer times (60 s and 240 s). (B and C) Percent-change maps were superimposed on the original raw images (see the legend on the right). Both the NADH dip (B) and the NADH overshoot (C) are strongest in the stratum radiatum of CA1. Labels: Stratum pyramidale (SP), stratum radiatum (SR), stratum moleculare-lacunosum (SML), dentate gyrus (DG), tip of the stimulation electrode (tip). (D) The binary image of the NADH-dip (green) and the NADH-overshoot (red) in the neuropil of CA1 shows that the early and late phases of the biphasic NADH response are highly anticolocalized with a linear correlation coefficient of –0.84. Scale bars, 50 μm.

In order to obtain an independent measure of neural activity, we recorded local field potentials in the stratum radiatum of CA1. We estimated the induced neural activity by multiplying the frequency of each evoked field potential by its amplitude. During stimulation (32 Hz) of the glutamatergic Schaffer collateral pathway, we measured a ∼200 fold increase in neural activity compared with baseline and recovery (Fig. 4A).

Fig. 4.

Early oxidative metabolism in dendrites is eventually sustained by glycolysis in astrocytes. (A) Evoked field potentials (inset) were used as an independent measure of neural activity. CNQX (25 μM) abolished the postsynaptic potentials and reduced the induced neural activity by a factor of >100 (purple plot). (B) The NADH dip is dendritic and the consequence of postsynaptic glutamate receptor activation as demonstrated by its almost complete annihilation after application of CNQX. The astrocytic NADH overshoot is substantially reduced but remains statistically different with respect to the controls (all groups n = 8, average ± SEM), which suggests an astrocytic origin of the late response. (C) Percent-changes superimposed on the raw image (contrast enhanced) show that the NADH overshoot often colocalizes with astrocytes and their processes. Scale bars, 20 μm. (D) The biphasic NADH response is the sum of two spatially and tempo-rally distinct monophasic metabolic responses, which indicated early oxidative recovery metabolism in dendrites followed by late activation of glycolysis in astrocytes (n = 8, average ± SEM).

To assess the contribution of the dendritic compartment to the overall metabolic response, we applied the glutamate receptor antagonist CNQX (6-cyano-7-nitroquinoxaline-2,3-dione) and waited until the evoked responses had disappeared before starting the brief tetanic stimulation. CNQX blockade of postsynaptic kainate/AMPA receptors reduced the induced activity during stimulation by a factor of ∼100 (Fig. 4A). In all trials, CNQX almost completely eliminated the NADH dip, but not the NADH overshoot (Fig. 4B). This result establishes the dendritic origin of the NADH dip. The dip residue (∼15% of the original response) was statistically not distinct with respect to the controls and may be attributed to the axonal compartment. The overshoot was attenuated, but remained with an ∼2% elevation distinct from controls (P < 0.01). We conclude that the NADH overshoot can be pharmacologically dissected from the dendritic NADH dip and is triggered by presynaptic activity and not by postsynaptic receptor activation, which suggest an astrocytic origin of the overshoot.

The superposition of the dip and the overshoot onto raw images at higher magnifications revealed that the dip originated from small, often circular, areas with an average intensity ∼10% higher than the average tissue intensity (fig. S1). Furthermore, these dip regions were never found in areas that could be morphologically identified as astrocytes. The intensity, shape, and location of the dip regions are consistent with the interpretation that these are neuronal mitochondria (Fig. 2, F and H). In contrast, overshoot regions were significantly larger than the dip regions, and they colocalized with astrocytes and their processes (Fig. 4C).

The inherent spatial three-dimensional (3D) resolution of multiphoton microscopy enabled us to optically dissect the biphasic NADH response between its early dendritic and late astrocytic components. We confined the analysis to regions where the NADH fluorescence change following stimulation exceeded twice the standard deviation of its spontaneous fluctuation before stimulation. This represents a robust threshold for statistically significant responses. The analysis revealed that the biphasic response actually is a superposition of two temporally and spatially distinct monophasic metabolic responses (Fig. 4D). The early dendritic response (green plot in Fig. 4D) is a transient NADH oxidation that closely matches the NADH dip. The NADH level reaches a minimum after ∼10 s with a half-time of ∼3 s. The recovery of the first response to baseline levels is much slower, with a half-time of ∼10 s. The late, astrocytic response (red plot) is a transient NADH production that resembles the NADH overshoot in the second half of the total fluorescence response (black plot). The astrocytic response starts at the nadir of the early response ∼10 s after stimulation onset and reaches a maximum after ∼30 s, with a half time of ∼19 s. The integrated signal intensity in the area of astrocytic NADH production exceeded the integrated neuronal NADH consumption on average by ∼57% after a 20 s stimulus. This net NADH production was further enhanced when longer stimuli were applied (Fig. 3A).

The early, dendritic response with fast NADH oxidation and slow recovery to baseline indicates oxidative recovery metabolism. The curve is the sum of NADH oxidation by the respiratory chain and subsequent NADH production by the Krebs cycle dehydrogenases (23, 24). The late astrocytic response with a monophasic NADH increase is the consequence of a temporary net production of NADH during nonoxidative glycolysis, where the reduction of NAD+ by glyceraldehyde-3-phosphate dehydrogenase temporarily exceeds the subsequent oxidation of NADH by the conversion of pyruvate to lactate (15, 16). The glycolytic nature of the NADH overshoot is corroborated by analytical measurements reporting about a threefold elevation of cytoplasmic NADH peaking ∼1 to 2 min after onset of seizure activity with a concurrent but longer-lasting lactate production (19). The cytoplasmic NADPH did not show significant fluctuations (20). We explain the attenuation (∼39%) of the astrocytic NADH production upon CNQX by the closure of the dendritic lactate sink with subsequent product inhibition of nonoxidative glycolysis. Coincidently, CNQX increases the extracellular glutamate availability and should unmask any significant contribution of mitochondrial NADH production by astrocytic glutamate oxidation to α-ketoglutarate, which we did not observe.

Our data demonstrate the spatial and temporal separation of early oxidative and late glycolytic metabolism between the neuronal and glial compartments during focal neural activity and provide direct experimental evidence for the existence of the astrocyte-neuron lactate shuttle. However, the sequence of the observed metabolic events differs from the initial proposal of this model (4), which predicted an overproduction of lactate from the very onset of focal activity. Our results appear closer to their more recent interpretation (25).

A transient increase in deoxyhemoglobin (the “initial dip”) of the cerebral microcirculation was attributed to a focal increase in cerebral oxygen metabolism (2). This interpretation was contested (26) but recently corroborated by direct measurements of early neuronal oxygen (27) and cerebral lactate (9) consumption. According to our model, the transient mitochondrial NADH oxidation is the direct consequence of respiratory chain activation as the physiological event underlying the macroscopic signs of oxidative metabolism, all of which occur within the first seconds after onset of induced activity. The spatial confinement of the NADH dip to dendrites further supports our view, since the preferred localization of cytochrome oxidase is in postsynaptic dendritic areas (28).

The kinetics of the late astrocyte NADH response as an indicator of enhanced glycolysis are in accord with reports that demonstrated nonoxidative glucose consumption during induced brain activity (1, 29). Note that the temporal resolution of PET or NMR spectroscopy would not have permitted the detection of early oxidative metabolism in these studies. However, one study (30) using a microsensor with significantly higher temporal resolution described a biphasic change in extracellular lactate with an early decrease (duration 10 to 12 s), followed by a long-lasting overshoot (peak after ∼60 s). This result is in full agreement with our model, which is based on our ample resolution in both space and time.

Our results confirm that early neuronal oxidative metabolism is the default response to increased neural activity. Only after a significant period (∼10 s) with depletion of substrates for oxidative metabolism (9, 30) is astrocytic glycolysis activated. Thereby, extracellular lactate might serve as a buffer preventing activation of the astrocyte-neuron lactate shuttle during minor or short-lasting neural activity. The observation that the transient NADH production as an indicator of nonoxidative glycolysis in astrocytes exceeds neuronal NADH consumption and further increases with longer stimulation strengthens this interpretation.

Our model integrates existing views of the organization of brain energy metabolism during focal neural activity (13, 5) and has direct implications for the design and interpretation of functional neuroimaging studies: The discovery that early oxidative metabolism is entirely neuronal strengthens the motivation for the current search for the initial dip in BOLD-fMRI, and the confinement of glycolysis to astrocytes implies that 18F-fluorodeoxyglucose–PET studies measure glucose uptake into the glial and not the neuronal compartment during focal neural activity.

Supporting Online Material

Materials and Methods

SOM Text

Fig. S1


References and Notes

Stay Connected to Science

Navigate This Article