Synaptic Activity Drives a Genomic Program That Promotes a Neuronal Warburg Effect*

Synaptic activity drives changes in gene expression to promote long lasting adaptations of neuronal structure and function. One example of such an adaptive response is the buildup of acquired neuroprotection, a synaptic activity- and gene transcription-mediated increase in the resistance of neurons against harmful conditions. A hallmark of acquired neuroprotection is the stabilization of mitochondrial structure and function. We therefore re-examined previously identified sets of synaptic activity-regulated genes to identify genes that are directly linked to mitochondrial function. In mouse and rat primary hippocampal cultures, synaptic activity caused an up-regulation of glycolytic genes and a concomitant down-regulation of genes required for oxidative phosphorylation, mitochondrial biogenesis, and maintenance. Changes in metabolic gene expression were induced by action potential bursting, but not by glutamate bath application activating extrasynaptic NMDA receptors. The specific and coordinate pattern of gene expression changes suggested that synaptic activity promotes a shift of neuronal energy metabolism from oxidative phosphorylation toward aerobic glycolysis, also known as the Warburg effect. The ability of neurons to up-regulate glycolysis has, however, been debated. We therefore used FACS sorting to show that, in mixed neuron glia co-cultures, activity-dependent regulation of metabolic gene expression occurred in neurons. Changes in gene expression were accompanied by changes in the phosphorylation-dependent regulation of the key metabolic enzyme, pyruvate dehydrogenase. Finally, increased synaptic activity caused an increase in the ratio of l-lactate production to oxygen consumption in primary hippocampal cultures. Based on these data we suggest the existence of a synaptic activity-mediated neuronal Warburg effect that may promote mitochondrial homeostasis and neuroprotection.

Synaptic activity can drive changes in gene expression to promote long lasting adaptations of neuronal structure and function (1,2). Examples of such gene transcription-dependent adaptive responses are the formation of long term memories, activity-dependent dendritic remodeling, and the buildup of acquired neuroprotection. The latter is defined as a synaptic activity-and gene transcription-mediated increase in the resistance of neurons against harmful conditions (3). We have previously identified a number of so-called activity-regulated inhibitor of death genes that are able to confer such neuroprotection (4,5). Although the exact mechanisms through which those genes confer neuroprotection are not yet fully understood, the functional changes mediated by activity-regulated inhibitor of death genes all seem to result in the protection of mitochondria (5)(6)(7). In the present study, we re-examined the synaptic activity-regulated gene pool to search for additional genes that are more directly linked to mitochondrial function. We found activity-dependent changes in the expression of several genes that encode for mitochondrial maintenance factors, metabolic regulatory enzymes, and nutrient transporters. This prompted us to test the long term effect of synaptic activitymediated gene transcription on energy metabolism. In general, cells use two major pathways to generate ATP from glucose. Under conditions of sufficient oxygen supply, cells use glycolysis to break down glucose to pyruvate, which is then oxidized to CO 2 in mitochondria via the TCA cycle. Reducing equivalents that are generated in the TCA cycle feed electrons into the mitochondrial electron transfer chain to generate a proton gradient that drives ATP generation via ATP synthase (oxidative phosphorylation). Under conditions of insufficient oxygen supply, the cells can switch to fermentation. In this case, pyruvate that is generated through glycolysis cannot be oxidized in mitochondria and is therefore converted to L-lactate, which is finally exported from cells. Generation of L-lactate serves to recover NAD ϩ , which is required for a continuous flux of glucose through glycolysis. Under certain conditions, some cells generate ATP via lactic acid fermentation in the presence of nonlimiting oxygen availability. This phenomenon is a well known feature of cancer cells and has been termed aerobic glycolysis or the Warburg effect (8,9). In this study we analyzed oxygen consumption and L-lactate production in primary hippocampal cultures and found that synaptic activity shifts the mode of neuronal ATP generation from oxidative phosphorylation toward aerobic glycolysis. This mode of neuronal energy metabolism presumably reduces mitochondrial activity and the generation of mitochondrial reactive oxygen species. Thus, as has been proposed before (2), these findings identify a novel component of the mechanisms underlying acquired neuroprotection.

Results
Action Potential Bursting Induces Changes in the Expression of Metabolism-related Genes-In a previous study, we have characterized the pool of action potential (AP) 3 bursting-regulated genes (4). We then asked, which, if any, of these genes encode for mitochondrial proteins. To answer this question, we compared our previous gene list to the MitoCarta1.0 database of genes that encode for mitochondrially localized proteins (10). This analysis revealed that of the 877 previously identified synaptic activity-regulated genes (4), 27 encode for mitochondrial proteins. Of these genes, 16 were down-regulated, and 11 were up-regulated within 2-6 h of AP bursting ( Table 1).
The list of down-regulated genes included, among others, mitochondrial transcription factors, mitochondrial ribosomal proteins, and an assembly factor of the mitochondrial ATP synthase complex. Examples of up-regulated genes are the solute carriers Slc25a3 and Slc16a1 (monocarboxylate transporter, Mct1), and the key metabolic regulatory enzyme, pyruvate dehydrogenase kinase 3 (Pdk3). Activity-dependent regulation of all genes listed in Table 1 (except Sod1 and Rps15a) was confirmed by RNA-Seq data obtained in an independent set of experiments (supplemental Table S1). Complete RNA-Seq data including raw data and detailed methods have been deposited in the NCBI Gene Expression Omnibus and are accessible through series accession number GSE92275. Whole transcriptome data for activity-dependent gene expression changes are listed in supplemental Table S2. When we analyzed the newly identified list of activity-regulated mitochondrial genes, we noticed that several of them affect energy metabolism. Indeed, the concomitant up-and down-regulation of several functionally related genes suggested that AP bursting induces a change in neuronal energy metabolism. To further investigate this idea, we screened the gene chip and RNA-Seq data sets to identify additional activity-regulated genes with a role in energy metabolism. These analyses revealed five additional candidate genes, the plasma membrane localized glucose transporters Glut1 (Slc2a1) and Glut3 (Slc2a3), the monocar- 3 The abbreviations used are: AP, action potential; PDK, pyruvate dehydrogenase kinase; qRT-PCR, quantitative reverse transcription PCR; rAAV, recombinant adeno-associated virus; HIF, hypoxia-inducible factor; EGFP, enhanced GFP; PDH, pyruvate dehydrogenase; OCR, oxygen consumption rate; TTX, tetrodotoxin; LPR, L-lactate production rate; DIV, day(s) in vitro; ANLS, astrocyte-neuron-lactate-shuttle; NALS, neuron-astrocyte-lactate shuttle; KA, kainic acid; TM, transfection medium; ActD, actinomycin D; ANOVA, analysis of variance; CI, confidence interval.

TABLE 1
Genome-wide transcriptome analysis reveals nuclear encoded mitochondrial genes that are regulated by synaptic activity Transcriptome data from a previous study from our group were compared with the MitoCarta database of mitochondrially localized proteins. The table lists genes with evidence for mitochondrial localization whose expression is changed more than 1.5-fold by bicuculline-induced AP bursting. For each gene, the official symbol and full name as well as fold change compared with control after 2, 4, and 6 h of treatment with 50 M bicuculline and after 2 and 4 h of bath application of 20 M glutamate are listed. Additional activity-regulated genes with a role in energy metabolism were identified from transcriptome analyses. Transcriptome analyses were performed with Affymetrix GeneChips (4; n ϭ 3) and RNA-Seq (n ϭ 2). NA, not applicable.  (Table 1). We next validated the gene expression changes identified in the transcriptome data sets. We subjected primary mouse hippocampal cultures to 4 h of AP bursting and then analyzed changes in the expression of 13 genes from Table 1 by qRT-PCR (Fig. 1, A and B). To test whether activity-dependent changes in gene expression are mediated by nuclear calcium, an important signal in excitation-transcription coupling (2), we infected the cultures with a recombinant adeno-associated virus (rAAV) that expresses mCherry-tagged CaMBP4, a nuclear-localized peptide that inhibits calcium/calmodulin signaling (5,11). Control cultures were infected with a rAAV that expresses nuclear-localized mCherry. Our qRT-PCR analyses confirmed the previously observed up-or downregulation of all tested genes except Sod1. Furthermore, we found that the changes in expression of several genes (8 of 12) were sensitive to inhibition of nuclear calcium/calmodulin signaling. The expression of many glycolytic genes can be controlled by the transcription factor, hypoxia-inducible factor (HIF), the activity of which is controlled via constitutive degradation versus stimulus-induced stabilization of its subunit HIF-1␣. We used immunoblotting to analyze HIF-1␣ levels in control cultures and in cultures that were subjected to AP bursting for 4 h (Fig. 2). HIF-1␣ expression levels were very low under basal conditions and decreased even further after stimulation, indicating that AP bursting- induced changes in metabolic gene expression are not mediated by HIF signaling.
To test whether changes in the expression of nuclear encoded mitochondrial genes were accompanied by changes in mitochondrial abundance, we measured mitochondrial DNA (Fig. 1C) and TOM20 protein (Fig. 1D) throughout 16 h of AP bursting. Mitochondrial DNA levels did not change, whereas TOM20 protein levels were significantly reduced after prolonged AP bursting, which is in line with the decreased expression of mitochondrial transcription and maintenance factors. The expression of Mct1, Glut3, Ucp1, and Pdk3 was specifically changed by survival promoting synaptic activity (AP bursting) but not by bath application of glutamate, which activates extrasynaptic NMDA receptors and promotes cell death (3, 12) (Fig. 1E).
We next analyzed activity-dependent regulation of metabolic genes in vivo. We used qRT-PCR to measure expression of 15 genes listed in Table 1 in the hippocampus of control mice and mice that had been treated for 4 h with kainic acid, which is commonly used to induce epileptic seizures involving strong global synaptic activation in the hippocampus and cortex. The results established the up-regulation of all 6 genes tested (Fig.  1F) and the down-regulation of 4 of 9 genes tested (Fig. 1G). We also analyzed the expression of two key metabolic genes, Pdk3 and Glut3, in the hippocampus of kainic acid-treated rats, the results of which confirmed their activity-dependent regulation in vivo (Fig. 1H).
Together, this coordinate pattern of gene expression changes suggests a form of neuronal metabolic plasticity, in which synaptic activity activates a genomic program that shifts metabolism from oxidative phosphorylation toward aerobic glycolysis (see Fig. 7A). Central to this neuronal Warburg effect are increased levels of Pdk3, which attenuate the activity of pyruvate dehydrogenase and thus divert pyruvate metabolism away from the TCA cycle. Concomitantly, up-regulation of the mitochondrial uncoupling protein Ucp1 and down-regulation of genes required for oxidative phosphorylation (Atpaf1) and mitochondrial maintenance (Tfam, Tfb2m, Dguok, Mrpl19, and Mrpl44) serve to attenuate mitochondrial activity. Finally, up-regulation of the glucose transporters Glut1 and Glut3 serve to meet the increased glucose demand of aerobic glycolysis, whereas increased Mct1 and Mct4 expression ensures efficient export of L-lactate that is produced during this process.
Activity-mediated Changes in Metabolic Gene Expression Occur in Neurons-The primary mouse and rat hippocampal cultures used in this study consist of ϳ25% astrocytes and 75% neurons (see Fig. 6C). For the interpretation of our results, it is important to consider the glial cell content of the cultures, because the ability of neurons to up-regulate glycolysis has been highly controversial (13,14). Moreover, some authors claim that MCT1, which we found robustly induced upon stimulation of synaptic activity (Fig. 1, A and F), is an astrocyte-specific protein that serves to shuttle L-lactate from astrocytes to neurons (15)(16)(17). Thus, we had to consider the possibility that the increased expression of glycolysis-related genes in our mixed cultures is due to their up-regulation in astrocytes. To restrict our gene expression analysis to neurons, we infected rat hippocampal cultures with a rAAV that expresses EGFP from a neuron-specific CaMKII promoter and used FACS to isolate EGFP-positive cells. We used qRT-PCR to compare gene expression between mixed cultures and sorted neurons obtained from sister cultures. Analysis of Gfap and Aqp4 mRNA expression confirmed that FACS-sorted cells were highly depleted of astrocytes (Fig. 3B). Based on the 0.04-fold average expression of glial markers in sorted cells and the presence of 25% glia cells in mixed cultures, we calculated that the FACS-sorted population contained less than 1% glia cells. We found that Mct1 was expressed 0.94 Ϯ 0.14-fold (mean Ϯ S.D.) in purified neurons compared with mixed cultures. This is consistent with a recent in vivo transcriptome analysis that revealed similar Mct1 expression levels in astrocytes and neurons versus a several hundred-fold higher expression level in brain endothelial cells (18). We also observed that AP bursting-mediated changes in the expression levels of metabolic genes in mixed cultures were similar to the changes observed in the neurononly population (Fig. 3C). Indeed, up-regulation of Mct1, as well as down-regulation of Atpaf1 and Tfb2m, were even more pronounced when analyzed in purified neurons compared with mixed cultures, indicating that within the mixed cultures activity-dependent gene regulation occurs predominantly in neurons.
To further corroborate these findings, we analyzed activity regulated expression of metabolic genes in "glia-free" primary mouse cortical cultures that contain less than 0.2% astrocytes (see "Experimental Procedures" for details). We again found robust up-and down-regulation of gene expression after 4 h of AP bursting (Fig. 3, D and E). Notably, using this pure neuronal preparation, we detected an activity-mediated decrease in the expression of Mct2, a monocarboxylate transporter that is considered to be required for uptake of L-lactate by neurons. Thus, AP bursting promotes an increased expression of L-lactate exporters Mct1 and Mct4 with a concomitant decrease in expression of the L-lactate importer Mct2. This reciprocal reg- ulation is consistent with the above proposed metabolic shift toward aerobic glycolysis.
AP Bursting Mediated Increase in Pdk3 Expression Attenuates PDH Activity-Pyruvate dehydrogenase (PDH) is a key enzyme that links glycolysis to mitochondrial respiration by converting pyruvate to acetyl-CoA, which is subsequently oxidized in the TCA cycle. The activity of PDH is mainly controlled by its phosphorylation status. In excitable cells, calcium influx stimulates the calcium-dependent pyruvate dehydrogenase phosphatase, which activates PDH via its dephosphorylation. This mechanism serves to activate mitochondrial respiration under conditions of acute energy demand (19). In contrast, PDK-mediated phosphorylation serves to attenuate PDH activity and thus diverts pyruvate away from oxidative phosphorylation. Based on the activity-dependent changes in Pdk3 expression, we examined the short and long term effects of AP bursting on PDH phosphorylation. As expected, when we ana-lyzed PDH phosphorylation at Ser-293 shortly after the induction of AP bursting in primary rat hippocampal cultures, we found pronounced PDH dephosphorylation (Fig. 4, A and B). In contrast, when we kept the neurons bursting for 8 h, i.e. for a sufficient time to express the Warburg gene program, we observed a recovery of PDH phosphorylation despite ongoing calcium-dependent pyruvate dehydrogenase phosphatase activation (Fig. 4, A and B). This indicates that sustained AP bursting has a biphasic effect on PDH activity, i.e. a rapid calciumdependent activation that is followed by a slow and gene transcription-dependent inactivation. We note that we cannot rule out the possibility that PDH rephosphorylation is mediated by another kinase than Pdk3. However, given that Pdk4 is expressed at very low levels in our cultures and expression of Pdk1 and Pdk2 are not affected by AP bursting (supplemental Table S2), Pdk3 is by far the most likely candidate.

Calcium-dependent Gene Transcription Causes a Change in the Mode of Energy Metabolism in Primary Hippocampal
Cultures-Our findings suggested that synaptic activity drives a gene program that shifts the mechanism of neuronal ATP generation from oxidative phosphorylation toward aerobic glycolysis, which is characterized by increased L-lactate production and reduced oxygen consumption. We thus measured both parameters and used their ratio to calculate a "Warburg Index." This index is independent of cell number and energy consumption rate and thus directly reflects the mode of ATP generation (oxidative phosphorylation versus aerobic glycolysis). To validate this approach, we determined the index of two different cell lines, HEK-293 cells and Hep G2 carcinoma cells. As outlined above, aerobic glycolysis is a prominent feature of cancer cells, and we therefore expected the Warburg Index of Hep G2 cells to be higher than that of non-cancerous HEK-293 cells. We indeed measured a 2.05 Ϯ 0.35-fold higher (mean Ϯ S.D., p ϭ 0.035 determined by paired two-tailed t test, n ϭ 2 independent experiments) Warburg Index in Hep G2 cancer cells compared with HEK-293 cells. To validate the MitoXtra fluo-  rescence based oxygen consumption assay in primary neuronal cultures, we next measured the oxygen consumption rate (OCR) in cultures that had been acutely treated with bicuculline or tetrodotoxin (TTX). Compared with untreated cells, AP bursting increased OCR, whereas silencing of electrical activity decreased OCR (Fig. 5A). This finding is in line with the abovementioned calcium-dependent activation of mitochondrial respiration (19,20) that is required to sustain acute, high frequency synaptic activity (21). We finally measured the effect of the Warburg gene program on neuronal energy metabolism. To avoid measuring metabolism under conditions of extremely high or extremely low energy demand (i.e. during AP bursting or complete silencing, respectively), we established an alternative stimulation paradigm. We elicited neuronal calcium transients by application of three brief pulses of 25 mM extracellular KCl (see "Experimental Procedures" for details). This treatment was sufficient to induce the expression of activity-dependent genes (Fig. 5C) but did not elicit sustained AP bursting. Compared with control cells, the Warburg Index was markedly increased 6 h after application of KCl pulses (Fig. 5B), in line with a change in neuronal metabolism.

Synaptic Activity Shifts Neuronal Energy Metabolism toward
Aerobic Glycolysis-We next set up an additional paradigm to measure the long term effects of activity-regulated gene expression on neuronal energy metabolism under conditions of basal synaptic activity. For this we took advantage of the well characterized course of synapse development in primary hippocampal cultures (22,23). We reasoned that in immature primary cultures with few synaptic contacts and low basal synaptic activity, expression levels of activity-dependent genes would be low. In contrast, in mature cultures that contain many functional synapses, expression levels of activity-dependent genes including the Warburg genes would be higher, and thus energy metabolism should be shifted toward aerobic glycolysis. To test this, we first compared the expression levels of the well characterized neuronal activity regulated gene Atf3 (5,24,25) and the metabolic genes Pdk3 and Glut3 in primary rat hippocampal cultures on DIV5 versus DIV12. We found that expression of all three genes increased during neuronal maturation (Fig. 6A). We then measured the OCR and L-lactate production rate (LPR) in primary hippocampal cultures on DIV5 versus DIV12. We found that the Warburg Index increased ϳ2-fold between DIV5 and DIV12 (Fig. 6B), consistent with the idea of a synaptic activity-dependent change in neuronal energy metabolism. The number of astrocytes and the surface area covered by them did not change between DIV5 and DIV12 (Fig. 6, C-E), indicating that changes in energy metabolism are due to changes in neuronal and synaptic maturation rather than changes in the abundance of astrocytes. Of note, some authors have argued that neurons have a very limited glycolytic capacity because of too low protein expression of a key glycolytic regulatory enzyme, PFKFB3 (26), which seems incompatible with the here described increased glycolysis in mature neurons. To address this issue, we analyzed PFKFB3 protein expression in our primary rat hippocampal cultures. Similar to previous reports (26), expression of PFKFB3 was low in immature cultures (DIV5). However, PFKFB3 expression increased ϳ2-fold during neuronal maturation, consistent with an increase in aerobic glycolysis (Fig. 6F).

Discussion
In this study we have identified a set of synaptic activityregulated genes that promote a shift of neuronal energy metabolism from oxidative phosphorylation toward aerobic glycolysis. This synaptic activity-dependent form of metabolic plasticity adds to the list of well established activity-and gene transcription-dependent neuroadaptations.
Aerobic glycolysis, i.e. lactic acid fermentation in the presence of non-limiting amounts of oxygen, has originally been described as a prominent feature of cancer cells (8,9). Extensive research over the last decades has shown that this so-called Warburg effect affects biosynthetic pathways and redox homeostasis and thereby provides a substantial growth and survival advantage to cancer cells (27,28). Aerobic glycolysis has also been observed in the resting and activated brain (29 -31), but it has remained controversial whether it occurs in neurons, glia, or both. In neurons, aerobic glycolysis may promote neuroprotection by maintaining mitochondrial homeostasis. Although acute activation of mitochondrial respiration appears to be The p values (compared with basal OCR) are indicated above the bar graphs and were determined by two-tailed paired t test (n ϭ 5 independent experiments, df ϭ 4). B, quantification of a Warburg Index determined by the ratio of LPR to OCR in primary rat hippocampal cultures 6 h after application of vehicle (control, Ctrl) or application of three brief pulses of 25 mM KCl to depolarize the cells and induce calcium-dependent gene transcription (see "Experimental Procedures" for details). The p value was determined by two-tailed paired t test (n ϭ 4 independent experiments, df ϭ 3). C, qRT-PCR analysis of the expression of activity-regulated genes in primary rat hippocampal cultures 3 h after application of three brief pulses of 25 mM KCl. As expected, fold changes after three brief KCl pulses are smaller than after 4 h of continuous AP bursting (cf. Fig. 1). All graphs show mean, individual values, and 95% CI. required to support high frequency synaptic activity (21), it has been shown that, in the long run, high rates of mitochondrial activity cause oxidative stress and neurodegeneration (32). On the other hand, several studies have demonstrated a beneficial effect of reduced mitochondrial activity. For example, in a recent study, the cellular hypoxia response was activated in cell lines, in zebrafish, and in mice via pharmacological and genetic manipulations or via exposure to low oxygen levels (33). The hypoxia response caused a shift to glycolytic energy metabolism and provided neuroprotection in a mouse model of Leigh syndrome, a genetic defect of respiratory chain function. In another study, the natural compound geissoschizine methyl ether was used to inhibit mitochondrial respiration in cultured neurons (34). This treatment promoted aerobic glycolysis and protected against glutamate-mediated oxidative damage. Finally, overexpression of Pdk1 or lactate dehydrogenase A in the B12 neuronal cell line was shown to cause a decrease of mitochondrial respiration and to provide protection against amyloid ␤ toxicity (35,36). Together, these and other studies have shown that genetic or pharmacological manipulation of energy metabolism can protect cells against oxidative damage and cell death. Our data suggest that neurons possess an intrinsic activity-regulated genomic program that uses this mechanism to confer resilience against harmful conditions (Fig. 7B).
An important caveat of our study is the fact that most experiments were performed in primary hippocampal cultures that contain a mix of neurons and astrocytes. Several studies suggested that aerobic glycolysis and L-lactate production are a predominant feature of astrocytes. According to the so-called astrocyte-neuron-lactate shuttle (ANLS) hypothesis, astrocytederived L-lactate is taken up by neurons, which convert it back to pyruvate to fuel mitochondrial respiration (14,37). The ANLS hypothesis apparently argues against the existence of a neuronal Warburg effect that involves increased neuronal glucose uptake and L-lactate release. The ANLS hypothesis is, however, highly controversial (13, 38 -40). Several authors suggest that neurons in fact release L-lactate, which is taken up by astrocytes, dispersed via gap junctions throughout the astrocytic syncytium, and finally transferred from astrocytes to endothelial cells to clear L-lactate from the brain. This route of metabolite shuttling has been termed the neuron-astrocytelactate shuttle (NALS). In support of the NALS, it was shown that in vivo L-lactate is released from the activated brain, which is consistent with the high expression level of MCT1 in endothelial cells (18), and at the same time argues against the use of L-lactate as a major energy substrate for neurons. Similarly, L-lactate as the sole energy substrate is not sufficient to support strong neuronal activity during gamma  4 independent experiments, df ϭ 3). B, quantification of a Warburg Index determined by the ratio of LPR to OCR in primary rat hippocampal cultures on DIV5 and DIV12. The Warburg Index is independent of cell density and energy consumption rate and thus directly reflects the mode of energy metabolism, i.e. aerobic glycolysis versus oxidative phosphorylation. The p value was determined by two-tailed paired t test (n ϭ 5 independent experiments, df ϭ 4). C and D, quantification of GFAP immunofluorescence to determine the number of astrocytes (C, percentage of total cells) and the surface area covered by astrocytes (D, percentage of total area) in primary rat hippocampal cultures on DIV5 and DIV12. The p values were determined by two-tailed paired t test (n ϭ 3 independent experiments, df ϭ 2). E, representative immunofluorescence staining of primary rat hippocampal cultures on DIV5 and DIV12. GFAP labeling is indicated in grayscale, and Hoechst 33258-labeled nuclei are indicated in blue. Scale bar, 100 m. F, quantification of PFKFB3 protein expression by immunoblot analysis in primary rat hippocampal cultures on DIV5 and DIV12. A representative immunoblot is shown in the inset. The p value was determined by two-tailed paired t test (n ϭ 3 independent experiments, df ϭ 2). All graphs show mean, individual values, and 95% CI.
oscillations (41). Moreover, synaptic activity causes an increase in neuronal surface expression of the glucose transporter GLUT3, which has been shown to protect neurons from excitotoxicity (42,43).
Many studies that support the ANLS assume astrocytespecific and neuronal isoforms of certain transporters and enzymes. For example, similar to our results, the expression of glycolytic genes in the dorsal hippocampus of adult mice has been shown to change in an activity-dependent manner (17). Among the genes with increased expression were Glut1, Glut3, and Mct1. Based on the assumption that Mct1 is an astrocyte-specific L-lactate transporter, the authors interpreted their findings as support for the ANLS hypothesis. However, our FACS experiments as well as recent quantitative in vivo transcriptome analyses (18) have shown that neurons do express Mct1 at levels similar to those measured in astrocytes. Moreover, expression of Mct1 in neurons is even slightly higher than the neuronal expression of the presumed neuron-specific gene Mct2 (18). Together, these findings suggest that assigning individual Mct isoforms to individual cell types maybe an oversimplification and that cell typespecific analyses are highly warranted when studying mixed cultures and tissues.
In this study, we used FACS to show that within mixed cultures activity-dependent changes in the expression of glycolytic genes occur in neurons. It is therefore reasonable to assume that the increase in aerobic glycolysis in our mixed primary cultures predominantly occurs in neurons as well. Importantly, our data do not suggest that neurons exclusively use either glycolysis or respiration to generate ATP under any given condi-tion. Instead, our data suggest that the neuronal Warburg gene program affects the relative flux of glucose through each of these two simultaneously used pathways.
The aim of this study was to investigate activity-dependent changes in neuronal gene expression and their effect on mitochondrial function and energy metabolism. The experiments were therefore not specifically designed to provide evidence for either the ANLS or the NALS hypotheses. Regardless of whether acutely activated neurons use glucose-or lactate-derived pyruvate to drive strong mitochondrial respiration, aerobic glycolysis would be a safer mode of ATP generation during intermittent episodes. We therefore suggest that the newly identified neuronal Warburg gene program serves to maintain mitochondrial homeostasis throughout the ups and downs of neuronal energy demand.

Experimental Procedures
Animals-C57BL/6NCrl mice (Charles River) and Crl:SD Sprague-Dawley rats (Charles River) were used in this study. The animals were group-housed on a 12-h light/dark cycle and had ad libitum access to water and food. All procedures were done in accordance with German guidelines for the care and use of laboratory animals and with the European Community Council Directive 86/609/EEC. The experiments were approved by local authorities.
Kainate Treatment-To study activity-dependent gene expression in vivo, 10 -12-week-old male mice were administered with kainic acid (KA; Biotrend, 20 mg/kg i.p., dissolved in 0.9% saline, n ϭ 6) or vehicle (PBS, n ϭ 5). Six-week-old male rats were administered with kainic acid (10 mg/kg i.p., n ϭ 8) or FIGURE 7. Schematic representation of the main findings of this study. A, schematic illustration of the major steps of glucose flux through glycolysis and the TCA cycle. Genes that were found in this study to be up-regulated by synaptic activity are indicated in bold blue. They are located at strategic positions within the metabolic network to control the mode of ATP generation. B, proposed model of a synaptic activity-mediated neuronal Warburg effect. Synaptic activity causes an influx of Ca 2ϩ that leads to enhanced mitochondrial respiration via Ca 2ϩ -dependent activation of mitochondrial dehydrogenases. Excitation transcription coupling causes a delayed increase in expression of the Warburg gene program that results in enhanced aerobic glycolysis and reduced mitochondrial respiration. The neuronal Warburg effect thus keeps mitochondrial activity in a sustainable range and ensures sufficient spare respiratory capacity to allow neurons to quickly respond to further increases in energy demand. Both mechanisms are proposed to protect neurons from oxidative damage and metabolic failure and thus provide a neuroprotective effect.
vehicle (PBS, n ϭ 8). After injection, a trained observer monitored the severity of epileptic seizures for 4 h to categorize according to the following criteria: level 1, immobility; level 2, forelimb and tail extension, rigid posture; level 3, repetitive movements, head bobbing; level 4, rearing and falling; level 5, continuous rearing and falling; level 6, severe tonic-clonic seizure; and level 7, death. Only animals that exhibited levels 4 -6 of epileptic seizure behavior were included in the analysis. 4 h after administration of KA or vehicle animals were killed by cervical dislocation. The brains were removed quickly, and hippocampi were dissected in ice-cold dissection medium (44) containing 1 mM kynurenic acid (Sigma) and 10 mM MgCl 2 . Individual hippocampi were homogenized in 700 l of Qiazol reagent (Qiagen), and total RNA was isolated as described below.
Primary Cultures and Treatments-Primary dissociated cultures of hippocampal or cortical neurons were prepared from newborn C57BL/6NCrl mice (Charles River) or newborn Sprague-Dawley rats (Charles River) and maintained as described (4,45). In brief, neurons were grown in Neurobasal-A medium (Life Technologies) supplemented with B27 (Life Technologies), 0.5 mM glutamine, and 1% rat serum (Biowest). To prevent the proliferation of glial cells, cytosine ␤-D-arabinofuranoside (Sigma-Aldrich, 2.8 M) was added on DIV3. To obtain glia-free cortical cultures, cytosine ␤-D-arabinofuranoside was added on the day of plating (DIV0). On DIV11, gliafree cortical cultures contained 0.18 Ϯ 0.002% astrocytes (mean Ϯ S.D., n ϭ 4 independent cell preparations). On DIV8 growth medium was exchanged with transfection medium (TM) consisting of a mixture of buffered salt-glucose-glycine solution (10 mM Hepes, pH 7.4, 114 mM NaCl, 26.1 mM NaHCO 3 , 5.3 mM KCl, 1 mM MgCl 2 , 2 mM CaCl 2 , 30 mM glucose, 1 mM glycine, 0.5 mM sodium pyruvate, and 0.001% phenol red) and phosphate-free Eagle's minimum essential medium (9:1 v/v), supplemented with insulin (7.5 g/ml), transferrin (7.5 g/ml), sodium selenite (7.5 ng/ml) (ITS supplement, Sigma-Aldrich). Unless indicated otherwise, experiments were performed after a culturing period of 10 -12 DIV during which neurons express functional glutamate receptors and develop an extensive network of synaptic contacts. Cultures were treated with 50 M bicuculline (Sigma), 20 M glutamate (Sigma), or 1 M TTX (Tocris Bioscience). Control cells were treated with vehicle (0.1% DMSO; Merck). To obtain a positive control for detection of HIF-1␣ (Fig. 2), oxygen supply was blocked by sealing the cell culture wells with HS mineral oil (MitoXpress-Xtra HS assay kit; Luxcel Biosciences). KCl stimulation was performed by replacing the standard TM with TM containing 25 mM KCl and 125 M APV (AppliChem). The NMDA receptor (NMDAR) antagonist APV was included during depolarization to prevent depolarization-induced glutamate toxicity (20 h after stimulation, cell death rate as assessed by nuclear morphology was not different between control and stimulated cultures (control: 13.8 Ϯ 12.8% dead cells (mean Ϯ S.D.) versus KCl: 12.0 Ϯ 10.9%; p ϭ 0.3, t ϭ 1.383 determined by paired t test, n ϭ 3 independent experiments)) and to prevent continuous action potential bursting after KCl washout. The cells were treated with three 30-s KCl pulses separated by 10-min inter-vals. Control cells were treated with three 30-s pulses of standard TM.
Recombinant Adeno-associated Viruses-For expression of CaMBP4-Flag-mCherry, mCherry, or EGFP the AAV expression cassette contained a CaMKII promoter, woodchuck hepatitis virus post-translational regulatory element, and a bovine growth hormone poly(A) signal (6). rAAV vectors were generated by standard molecular biology techniques and verified by sequencing. Viral particles were produced as described (5). Neurons were infected with 10 9 -10 10 particles/ml on DIV4 to DIV5, yielding a typical infection rate of 80 -90%.
RNA-Seq-Mouse hippocampal neurons were left untreated or stimulated with 50 M bicuculline on DIV10. The cells were harvested after 4 h of AP bursting, and total RNA was extracted using a Qiagen RNeasy mini kit with on-column DNase I digestion according to the manufacturer's instructions. Library preparation and mRNA-Seq were performed at the EMBL Genomics Core Facility (Heidelberg, Germany). Raw data, details of the methods, and complete results of gene expression analysis are available through Gene Expression Omnibus Series accession number GSE92275.
Real Time PCR-To determine changes in mRNA expression, total RNA was extracted from primary cultures and hippocampal tissue using an RNeasy mini kit (Qiagen) with additional on-column DNase I digestion. cDNA was synthesized from 1 g of total RNA using a high capacity cDNA reverse transcription kit (Applied Biosystems). qRT-PCR was done on an ABI7300 thermal cycler using universal qRT-PCR master mix (Applied Biosystems) with the following TaqMan gene expression assays (Applied Biosystems): (Mm01620397_s1), and Ucp1 (Mm01244861_m1). Expression of target genes was normalized against Gusb and/or 18s rRNA as endogenous control genes. In experiments that compared different cell populations (Fig. 3) or cells at different developmental stages (Fig. 6), expression of target genes was normalized against the geometric mean of 18s rRNA, Actb, and Gapdh. To determine total mtDNA levels (Fig. 1C), genomic DNA was isolated with a Qiagen DNeasy kit. For real time PCR, 3 ng genomic DNA was used to measure nuclear DNA and mtDNA (46). Mitochondrial COX1 DNA concentration (Mm04225243_ g1) was normalized against actin-like 9 genomic DNA concentration (Actl9, Mm00809079_s1).
Fluorescence-activated Cell Sorting-Primary rat hippocampal cultures in 35-mm dishes were treated with bicuculline or vehicle (DMSO) for 4 h. The medium was then replaced with 2 ml of papain solution (74 mM Na 2 SO 2 , 27 mM K 2 SO 4 , 15.9 mM MgCl 2 , 0.23 mM CaCl 2 , 1.4 mM Hepes, 18 mM glucose, 0.25% phenol red, 1 mM kynurenic acid, 3.7 mM L-cysteine, 10 units/ml papain (CellSystems, St. Katharinen, Germany), 10 g/ml ActD), and the cells were kept in the incubator at 37°C for 20 min. The dishes were then placed on ice, and the cells were carefully dislodged from the dishes by pipetting with a 1-ml pipette tip and transferred to a centrifugation tube. Cells from four 35-mm dishes were pooled at this step. After centrifugation (5 min, 133 ϫ g, 4°C), the cell pellet was kept on ice, carefully resuspended in 1 ml of FACS buffer (10 mM Hepes, pH 7.4, 114 mM NaCl, 5.3 mM KCl, 1 mM MgCl 2 , 2 mM CaCl 2 , 30 mM glucose, 0.5 mM sodium pyruvate, 1% fetal calf serum, 0.1% phenol red, 10 g/ml ActD), triturated with a 1-ml pipette, triturated with a flame polished Pasteur pipette, and passed through a 35-m cell strainer (Corning catalog no. 352235). Papain solution and FACS buffer contained ActD (Biotrend), and the cells were kept on ice whenever possible to avoid changes in mRNA levels during cell preparation and FACS sorting. The cells were sorted on a FACSAria cytometer (Becton Dickinson), and total RNA was isolated from sorted cells with an RNeasy Micro Kit (Qiagen).
Oxygen Consumption Assay-Oxygen consumption was measured with the MitoXpress-Xtra HS assay (Luxcel Biosciences) according to the manufacturer's instructions. In brief, the cells were grown in 24-well glass-bottomed plates or 96-well clearbottomed plates and treated as indicated. Time-resolved fluorescence of the oxygen probe was measured at 37°C for 40 min on a Tecan Safire 2 plate reader (excitation, 370 -390 nm; emission, 640 -660 nm; lag time, 30 s; and integration time, 100 s). Wells with cells but without probe and wells with probe but without cells were used as background and signal controls, respectively. OCR was determined in arbitrary units based on the slope of fluorescence increase.
Measurement of the Warburg Index-The cells were grown in 5 replicate wells/condition in 96-well clear-bottomed plates. The neurons were plated into the same 96-well plate in consec-utive weeks so that 5-and 12-day-old neurons could be measured within one plate. On the day of measurement standard medium (TM or DMEM) was replaced with fresh medium that contained 2 mM glucose (50 l/well). After 8 h of incubation (or 6 h for KCl pulse-treated neurons), 20 l of medium was collected from each well. Replicate samples were pooled, cleared by centrifugation at 16,000 ϫ g for 5 min, and snap frozen in liquid nitrogen. Remaining medium in the wells was then replaced by fresh medium that contained 2 mM glucose and 60 nM MitoXpress-Xtra probe, and oxygen consumption was measured as described above. L-Lactate concentration in the cell culture supernatants was measured with an EnzyFluo TM L-lactate assay kit (BioAssaySystems) according to the manufacturer's instructions. The Warburg Index was calculated as the ratio of LPR (change in concentration during 6 -8 h of incubation) to OCR (slope of oxygen probe fluorescence increase). Because the ratio of these two measurements is independent of cell density and energy consumption rate, it directly reflects the mode of energy metabolism (oxidative phosphorylation versus aerobic glycolysis).
Immunocytochemistry and Quantification of Glial Content in Primary Cultures-The cells were fixed with 4% paraformaldehyde, 4% sucrose in PBS for 15 min, washed with PBS, and permeabilized with 0.3% Triton X-100 in PBS. The cells were blocked with 10% normal goat serum, 2% BSA and incubated with mouse anti-GFAP antibody (Sigma G3893, lot no. 078K4830, RRID:AB_477010, 1:1,000) or mouse anti-NeuN antibody (Millipore MAB377, lot no. LV1519148, RRID: AB_177621, 1:200) in 0.1% Triton X-100, 2% BSA at 4°C overnight. The cells were then incubated with Alexa 488-or Alexa 594-conjugated goat anti-mouse IgG secondary antibody (Thermo Fisher Scientific A-11001 RRID:AB_2534069), counterstained with Hoechst 33258 (1 g/ml), and mounted in Mowiol 4-88 (Calbiochem). For robust quantification of glial content in cortical cultures (Fig. 3), 12 evenly distributed fields of view (Ͼ2,500 cells) were examined on one coverslip/condition in four independent cell preparations. Images were recorded on a Leica DMIRBE inverted microscope equipped with a 10ϫ objective (NA 0.3) and a SPOT Insight 14-bit CCD camera (Visitron Systems) with VisiView imaging software (Visitron Systems). For quantification of glial content in primary hippocampal cultures (Fig. 6), sister coverslips from the same cell preparation were fixed and stained on DIV5 and DIV12. In total, 5 randomly selected fields of view were examined with a 20ϫ objective (NA 0.5) on one coverslip per condition in three independent cell preparations. The percentage of GFAP-positive cells and the percentage area covered by GFAPpositive processes were determined with ImageJ.
Cell Death Analysis-The cells were fixed with 4% paraformaldehyde, 4% sucrose in PBS for 15 min, washed with PBS, and counterstained with Hoechst 33258 (1 g/ml) for 10 min. The cells were mounted in Mowiol 4-88 (Calbiochem) and examined by fluorescence microscopy. The dead neurons were identified by amorphous or shrunken nuclei as described before (47).