Astrocytic PI3Kα controls synaptic plasticity and cognitive function via serine metabolism

Abstract

Astrocytes modulate neuronal activity by gliotransmission and through metabolic regulation, yet the relationship between these functions remains poorly understood. Here we show that phosphatidylinositol 3-kinase (PI3K) signaling in astrocytes controls synaptic plasticity and memory by regulating cellular metabolism. Deletion of the p110α isoform of PI3K in hippocampal astrocytes impairs NMDA receptor activation during the induction of hippocampal long-term potentiation (LTP). This defect is rescued by either d-serine (co-agonist of synaptic NMDA receptors) or its precursor l-serine, indicating that PI3K regulates synaptic plasticity through serine availability. Indeed, deletion of p110α rewires astrocyte metabolism, reducing glycolytic flux while enhancing mitochondrial respiration. This, in turn, would limit l-serine biosynthesis through the phosphorylated pathway. Consistently, mice lacking astrocytic p110α display memory deficits that are rescued by in vivo l-serine administration. These findings identify astrocytic PI3K p110α as a key link between cellular metabolism, synaptic plasticity and cognition.

The authors find that the PI3Kα pathway links metabolism to hippocampal synaptic plasticity by sustaining astrocyte glycolysis and ensuring L-serine/D-serine supply to neurons. Loss of astrocytic PI3Kα impairs memory in mice, an effect rescued by serine supplementation.

Introduction

Neurons are able to modify their synapses in response to the activity they receive, altering the strength of their connections, the efficacy of synaptic transmission, and eventually, the function of neural circuits. This phenomenon is called synaptic plasticity and likely underlies several cognitive processes, including learning and memory^1,2. However, neurons are not the only cells contributing to synaptic plasticity. In fact, it is now well established that astrocytes are able to respond to neuronal activity and extracellular signals and, in turn, modulate neuronal and synaptic functions. This occurs through the release of active molecules (gliotransmitters)^3–6, such as glutamate^7–13 or ATP^14–16, directly acting on neuronal receptors. In addition, astrocytes can also modulate neuronal function and brain activity via their role in metabolic support^17–19. Indeed, the exchange of metabolites between neurons and astrocytes is being revealed as an important contributor to neuronal function. This is best exemplified with the astrocyte–neuron lactate shuttle (ANLS)^20,21 or with the supply of L- and D-serine from the astrocyte for the activation of neuronal NMDA receptors (NMDARs)^22–24. These astrocyte–neuron metabolic interactions have been shown to have functional consequences for cognitive performance^25,26 and are probably linked to human disease^27,28. However, the intracellular mechanisms controlling this crosstalk between metabolic and synaptic regulation remain poorly understood.

In order to fill this gap, we have turned our attention to the phosphatidylinositol 3-kinase (PI3K) signaling pathway. The PI3K/mTOR pathway is well-known to mediate synaptic plasticity mechanisms^29–32. At the same time, this is a critical hub for the regulation of cellular metabolism^33,34, particularly controlling glucose homeostasis^35–38 via the intracellular trafficking of glucose transporters^39,40 and by modulating the activity of glycolytic enzymes^41,42. Therefore, the PI3K pathway is ideally suited to link metabolic regulation and synaptic plasticity. However, PI3Ks are a diverse family of signaling enzymes. Class IA PI3Ks are heterodimers composed of a p85 regulatory subunit and a p110 catalytic subunit, which catalyzes the production of phosphatidylinositol (3,4,5)-trisphosphate (PIP₃)⁴³. The p110α, p110β, and p110δ paralogs of the catalytic subunit, commonly referred to as isoforms, are encoded by different genes and have been shown to play distinct roles in cellular physiology and growth^34,44,45. Our group has recently described that neuronal p110α and p110β have separate functions in synaptic plasticity and neuronal architecture³². However, there is little information on regulatory mechanisms of metabolism in astrocytes, and certainly nothing is known about the potential role of astrocytic PI3K signaling in these cells and its impact on neuronal function.

In order to address this question, we have used a mouse model bearing a cell type- and brain region-restricted deletion of the p110α catalytic isoform of PI3K, combined with behavioral, electrophysiological, imaging, and bioenergetic assays. In light of these experiments, we now report that the lack of astrocytic p110α impairs NMDAR-dependent LTP and cognitive functions due to a decrease in the availability of D- and L-serine, which can be rescued in vitro and in vivo. Moreover, we show that L-serine deficiency is caused by a metabolic switch within the astrocyte, which decreases its glycolytic flux and enhances mitochondrial respiration in the absence of p110α.

Results

p110α knock-out in astrocytes alters dendritic spines in hippocampal neurons

To study the role of the p110α catalytic isoform of PI3K in astrocytes, we used a gene-targeted approach using Cre-lox recombination. Given the embryonic lethality of the complete knock-out of the PIK3CA gene⁴⁶, we performed bilateral stereotaxic injections with an adeno-associated virus (AAV) expressing the Cre recombinase under the glial fibrillary acidic protein (GFAP) promoter (AAV5-GFAP-Cre-GFP) in the hippocampus of adult p110α^flox/flox mice (Fig. 1a). Saline-injected animals were used as controls. With this experimental paradigm, after one month of infection, Cre was expressed in around 70–80% of the astrocytes in the stratum radiatum across the whole hippocampus (Supplementary Fig. 1A). The specificity of the infection for astrocytes was confirmed from the colocalization of Cre-GFP with a GFAP antibody, and not with anti-NeuN immunostaining (Fig. 1a and Supplementary Fig. 1B). This approach produced a significant reduction in p110α expression levels in infected animals (referred to as p110α^aKO, aKO stands for astrocytic knock-out) as compared to controls, without detectable changes in another catalytic subunit isoform, p110β (Supplementary Fig. 1C, D). Of important note, the knock-out efficiency was not expected to be complete, as p110α is ubiquitously expressed and is also present in neurons and cerebral vasculature⁴⁷. This decrease in p110α levels was accompanied by a significant reduction in the phosphorylation of the S473 residue of AKT and a slight but significant increase in the total AKT expression levels. The change in phosphorylation of another downstream target, GSK3, was not statistically significant (Supplementary Fig. 1E, F). Importantly, astrocytic viability (judged from the density of GFAP-positive cells in infected animals; Fig. 1b), morphology of the GFAP-positive intermediate filament cytoskeleton (Fig. 1c–f), or GFAP expression (Fig. 1g) did not appear to be altered by p110α depletion.

Fig. 1. Cre-LoxP-mediated conditional knock-out of p110α in astrocytes of the hippocampus and neuronal effects.

a Left, experimental scheme (created in BioRender. Fernández, A. (2026) https://BioRender.com/f95ibit) showing the injection sites in the hippocampus for the ablation of p110α from p110α^flox/flox mice to generate the p110α^aKO. Center, merged representative image of GFAP-Cre expression (shown in green) in the hippocampus after one month of AAV infection in combination with the nuclear marker DAPI (shown in blue). The CA1, SR (stratum radiatum), and DG (dentate gyrus) regions are depicted. Scale bar: 200 µm. Right, high magnification images of SR (up) and stratum pyramidale (down) subregions of CA1. AAV-infected astrocytes are labeled by GFP (shown in green), all the astrocytes by GFAP (shown in magenta), the layer of pyramidal neurons in CA1 by NeuN (shown in gray), and the nuclei by DAPI (shown in blue). Scale bar: 30 µm. b Quantification of the number of GFAP-positive astrocytes per mm² in the p110α^aKO and control conditions. Bars represent mean ± SEM. Control n = 14, p110α^aKO n = 21 slices from 4 to 9 mice. c Left: Representative images of GFAP immunostaining of Cre-mCherry-expressing and control astrocytes from SR of p110α^flox/flox mice. Right: Representative examples of astrocyte traces for the Sholl analysis. Scale bars: 20 µm. d Sholl analysis of astrocyte morphology for control and p110α^aKO slices. Statistical significance was calculated by two-way ANOVA repeated-measures. e Total length for astrocytic processes. Control n = 12, p110α^aKO n = 15 astrocytes from 2 mice per condition. f Number of astrocytic processes from control and p110α^aKO mice. Control n = 12, p110α^aKO n = 15 astrocytes from 2 mice per condition. g Integrated density values for GFAP immunostaining of astrocytes from control and p110α^aKO mice. Control n = 12, p110α^aKO n = 16 astrocytes from 2 mice per condition. h Left, representative confocal images of GFP-expressing dendrites and spines for each condition. Scale bars: 2 µm. Right, quantification of spine density from dendrites of CA1 neurons from control and p110α^aKO mice. Control n = 18, p110α^aKO n = 32 dendrites from 4 to 5 mice per condition. i Spine length from GFP neurons (cumulative distributions on the right) for p110α^aKO and control mice. Control n = 660, p110α^aKO n = 1009 spines from 4 to 5 mice per condition. j Quantification of mean spine head area from GFP neurons of p110α^aKO and control mice. Control n = 701, p110α^aKO n = 1070 spines from 4 to 5 mice per condition. Data are displayed as means ± SEM. Statistical analysis was carried out by two-sided unpaired t-test (b, e, f) or two-sided Mann–Whitney test (g–j) (**p < 0.01, ****p < 0.0001, ns: not significant). Source data are provided as a Source Data file.

To study neuronal morphology in the absence of astrocytic p110α, p110α^aKO mice were sparsely infected with a GFP-expressing Sindbis virus for 24 h to label individual CA1 pyramidal neurons (Supplementary Fig. 1G). The lack of p110α in astrocytes did not affect dendritic arborization or total dendritic length (Supplementary Fig. 1H, I). However, there was a significant decrease in spine density (Fig. 1h) that was accompanied by a small but significant increase in spine length (Fig. 1i), without changes in spine head area (Fig. 1j). To note, this phenotype is clearly distinct from the one observed upon deletion of neuronal p110α, which produced a strong dendritic atrophy and an increase in spine head area³².

Astrocytic deletion of p110α causes spatial and object recognition memory impairment

Previous studies from our laboratory dissected specific behavioral deficits upon deletion of neuronal p110α³². Therefore, we wanted to explore the role of the astrocytic p110α isoform in cognitive function and, in particular, in learning and memory. After one month of hippocampal infection, p110α^aKO and controls were evaluated with different behavioral tests.

First, general locomotor activity when mice are exposed to a novel environment was evaluated in the open field test (Fig. 2a). No abnormal locomotor behavior regarding speed, distance traveled, or preference for different areas of the arena was observed in the p110α^aKO mice compared to their controls (Fig. 2a, b). To assess possible deficiencies in working memory, we studied spontaneous alternation performed in the Y maze. The spontaneous alternation percentage was statistically higher than chance for both controls and knock-out animals, without significant differences between them (Fig. 2c).

Fig. 2. Astrocytic p110α contributes to spatial memory and object recognition.

Mice were tested in the open field to study locomotor behavior. Mice were placed individually in the center of an open field arena and allowed to freely explore for 15 min. a Left, representative picture of the open field arena and the three different areas in which it was divided, from outside to inside: periphery, inner, and center. Image created in BioRender. Fernández, A. (2026) https://BioRender.com/bi7jbk4. Right, average speed for the entire duration of the test. b Quantification of the traveled distance for the entire duration of the test, separated by the peripheral and the inner + center zones of the arena for the p110α^aKO mice and controls. The total distance traveled is also shown (shaded background). Two-way ANOVA and Bonferroni post-hoc tests were used to assess statistical differences between p110α^aKO and their controls (ns: not significant). Control n = 16, p110α^aKO n = 13–15. c Left, experimental scheme of the spontaneous alternation test performed in a Y-shaped maze. Image created in BioRender. Fernández, A. (2026) https://BioRender.com/3nz380x. Right, spontaneous alternation percentages from p110α^aKO and control animals. Increases in the percentage of alternation over chance were evaluated with the Wilcoxon signed-rank test (####p < 0.0001). Control n = 16, p110α^aKO n = 16. d, e Left, experimental diagram of the novel object location test (d; image created in BioRender. Fernández, A. (2026) https://BioRender.com/m8p9hu9) and the novel recognition test (e; image created in BioRender. Fernández, A. (2026) https://BioRender.com/69icgop). Right, discrimination indexes for the novel versus the familiar location (d) or the novel versus the familiar object (e) in p110α^aKO mice. Control n = 17–19, p110α^aKO n = 17–18. f, g Left, schematic representation of the three-chamber social interaction task for the sociability (f; image created in BioRender. Fernández, A. (2026) https://BioRender.com/ck73j1p) and for the social novelty (g; image created in BioRender. Fernández, A. (2026) https://BioRender.com/5xsjxu3) tests. Right, discrimination indexes for the social target over the inanimate object (f) and for the novel subject over the known acquaintance (g) for p110α^aKO mice. Control n = 14, p110α^aKO n = 14. Bars represent mean ± SEM, and individual values for each mouse are plotted as aligned dot-plots. To assess whether the DIs were different from zero, we performed the two-sided Wilcoxon signed-rank test (#p < 0.05, ##p < 0.01, ### p < 0.001) and two-sided Mann–Whitney test was used to assess statistical differences between p110α^aKO and their controls (*p < 0.05, **p < 0.01, ns: not significant). For all panels, individual values for each mouse and condition are represented along with the mean ± SEM as aligned dot-plots. Source data are provided as a Source Data file.

Next, we used the novel object location (NOL) and the novel object recognition (NOR) tests that have been widely used to study various aspects of learning and memory. At the NOL test, while the controls were able to significantly discriminate the displaced object, p110α^aKO mice failed to do so (Fig. 2d). Similarly, at the NOR test, we observed that p110α^aKO mice were unable to discriminate between the familiar and the novel object compared to their controls (Fig. 2e). Therefore, these results indicate that p110α in hippocampal astrocytes is required for spatial memory and object recognition memory.

Finally, we tested sociability and preference for social novelty consecutively, using a three-chamber arena. Using these tests, we observed that both controls and p110α^aKO mice significantly preferred social exploration (Fig. 2f) and the novel subject rather than the familiar one (Fig. 2g). These results indicate that astrocytic p110α is not required for sociability or for social memory.

The lack of p110α in astrocytes specifically impairs long-term potentiation

The behavioral deficits we have observed in p110α^aKO mice suggest that astrocytic p110α may also participate in synaptic plasticity paradigms related to learning and memory. To test this hypothesis, we evaluated synaptic transmission at Schaffer collateral-CA1 synapses from acute hippocampal slices using extracellular field recordings of excitatory postsynaptic potentials (fEPSPs).

Long-term potentiation (LTP) elicited by theta burst stimulation (TBS) was strongly reduced in p110α^aKO mice, as compared to control animals, even though it was not completely abolished (Fig. 3a). A similar result was observed with a different Cre-expressing AAV with a more stringent specificity for astrocytes (AAV5-GfaABC1D-Cre-4x6T⁴⁸; Supplementary Fig. 2A). Importantly, this impairment in LTP was not observed when astrocytic p110β was deleted using a similar Cre strategy on p110β^flox/flox mice (Supplementary Fig. 2B, C). This result indicates that astrocytic p110α is required for LTP. In contrast, long-term depression (LTD) elicited by low frequency stimulation was indistinguishable between p110α^aKO and control mice (Fig. 3b). Furthermore, basal synaptic transmission was unaltered in p110α^aKO hippocampal slices compared to their corresponding controls, as assessed by input/output curves of fEPSPs at different stimulation intensities under standard conditions (mostly mediated by AMPARs; Fig. 3c) or in the presence of CNQX and at low (0.2 mM) extracellular Mg²⁺ concentration to isolate NMDAR responses (Fig. 3d; see Supplementary Fig. 3A to ascertain NMDAR-specificity).

Fig. 3. p110α^aKO mice have an impairment in NMDAR-LTP.

a Left, time course of NMDAR-dependent LTP induced by TBS. Right, bar plot representing the mean of the fEPSP slope for each slice on the last 5 min of recording. b Left, time course of NMDAR-dependent LTD induced by LFS. Right, bar plot representing the mean of the fEPSP slope for each slice on the last 5 min of recording. a, b Representative traces for each condition are depicted in the upper part of the graph, the lighter color corresponding to the average trace of the baseline and the darker one to the post-induction. c Input–output curve of fEPSPs evoked at different stimulation intensities. Representative traces from one experiment for each condition are shown over the graph. Scale bar: 0.5 mV, 10 ms. d Input–output curve of NMDAR fEPSPs evoked at different stimulation intensities. Representative traces from one experiment for each condition are shown over the graph. Scale bar: 0.2 mV, 40 ms. e Pairs of pulses separated by 50, 100, 200, and 400 ms are delivered to the Schaffer collaterals for measuring the paired-pulse ratio (PPF). Control n = 8, p110α^aKO n = 16 from 7 to 8 mice. Representative traces for the 50 ms inter-stimulus interval are shown over the graph. Scale bar: 0.1 mV, 50 ms. Bars represent mean ± SEM, and individual values for each condition are plotted as aligned dot-plots. Statistical differences were assessed by two-way repeated-measures ANOVA with Bonferroni post-hoc test (c–e) and two-sided unpaired t-test (a, b) (***p < 0.001, ns: not significant). Two-sided Wilcoxon signed-rank test was performed to assess statistically significant changes from baseline (a, b) (##p < 0.01, ###p < 0.001). n, number of slices per condition from n > 6 mice per condition. Source data are provided as a Source Data file.

We also evaluated presynaptic function in acute hippocampal slices by measuring paired pulse facilitation. This was unaffected too at all inter-stimulus intervals (Fig. 3e), suggesting that the absence of astrocytic p110α does not significantly alter presynaptic glutamate release at CA3-to-CA1 synapses.

The LTP impairment of p110α^aKO mice is accompanied by a deficient NMDAR activation and can be rescued by D- and L-serine

Given the specific deficit in LTP caused by the absence of astrocytic p110α, we asked whether LTP induction may be altered in p110α^aKO animals. To address this point, we analyzed the fEPSP response during the four pulses at 100 Hz of the TBS protocol used for LTP induction. Specifically, we measured the fEPSP amplitude at 30 ms after the last pulse, which would correspond mostly to the NMDAR response (black bracket in Fig. 4a). Using this analysis, we observed that this slow fEPSP gradually increased in control animals across the four rounds of TBS of the LTP induction protocol (Fig. 4b, gray symbols). In contrast, this response was consistently smaller, and its increase was significantly attenuated in p110α^aKO animals (Fig. 4b–d). To note, similar experiments carried out in the presence of D, L-2-amino-5-phosphonovaleric acid (APV; NMDAR antagonist) in control slices showed a marked reduction in this fEPSP response, indicating that it indeed corresponds mostly to NMDAR activation (Fig. 4b, orange symbols). Given the normal synaptic transmission mediated by AMPARs and NMDARs in p110α^aKO animals (Fig. 3c, d), these results strongly suggest that the absence of astrocytic p110α produces a deficient NMDAR response specifically during the high-demanding conditions of LTP induction. To note, the AMPAR component of the fEPSP during LTP induction (peak of the response after the first pulse, indicated with an arrow in Fig. 4a) and the ratio of the AMPAR to the NMDAR component of the response also increased across the induction rounds (Supplementary Fig. 3B, C), probably reflecting the ongoing potentiation during the TBS protocol.

Fig. 4. Serine addition rescues NMDAR-dependent LTP deficiency in p110α astrocytic knock-out slices.

a Average trace of 10 consecutive trains of four pulses for the last induction round for the control (black) and the p110α^aKO (blue) slices. The stimulation artifact for the four pulses can be seen before every peak of response. The vertical arrow points to the peak of the first response, which corresponds mostly to the AMPAR fEPSP. The bracket at 60 ms represents the position where the NMDAR fEPSP amplitude is measured. b NMDAR responses for the four induction rounds for p110α^aKO, control and control APV-treated slices. Control n = 15, p110α^aKO n = 23, APV n = 11 slices from 2 to 15 mice. c NMDAR responses corresponding to the first (I1) and the fourth (I4) induction rounds in control and KO conditions. d Increment of NMDAR response from the first to the fourth round of LTP induction (I₄–I₁). e Left, time course of NMDAR fEPSP amplitude before and after the addition of 50 µM D-serine in the bath. Right, bar plot representing the mean of the fEPSP amplitude for each slice from the baseline and from the last 5 min of recording. f, g Left, time course of NMDAR-dependent LTP induced by TBS with and without treatment with 10 µM D-serine (f) or 50 µM L-serine (g). Right, bar plot representing the mean of the fEPSP slope for each slice on the last 5 min of recording. Representative traces for each condition are depicted in the upper part of the graph, the lighter color corresponding to the average trace of the baseline and the darker one to the post-induction. Bars represent mean ± SEM, and individual values for each condition are plotted as aligned dot-plots. Two-way ANOVA for matched samples and Bonferroni post-hoc were used in b and c to assess statistical differences between the induction rounds of each condition (#) and between p110α^aKO and the control (*). Two-sided Mann–Whitney and Wilcoxon signed-rank tests were used in d. One-way ANOVA and Bonferroni post-hoc were used in e (*p < 0.05, ***p < 0.001, ##p < 0.01, ###p < 0.001, ####p < 0.0001, ns: not significant). Two-sided unpaired t-test was carried out for f and g (ns: not significant). n, number of slices per condition from n > 4 mice per condition. Source data are provided as a Source Data file.

We then reasoned that the impairment of the NMDAR response during LTP induction may be related to the availability of D-serine, which would lead to a deficient NMDAR activation. Furthermore, D-serine is the preferred co-agonist in addition to glutamate for the activation of synaptic NMDARs during LTP⁴⁹. In fact, astrocytes are good candidates to control D-serine availability, either from the conversion of L-serine into D-serine by the serine racemase (although this point is still controversial; see refs. ^50–54) or from the biosynthesis of L-serine, which in the brain specifically occurs in glial cells^22,23,55,56.

To evaluate the D-serine dependence of NMDARs in p110α^aKO and control animals, we isolated the NMDAR fEPSPs (as in Fig. 3d) and monitored their sensitivity to exogenous D-serine. The rationale of this experiment is that a lower availability of endogenous D-serine will lead to a larger enhancement of NMDAR responses upon addition of exogenous D-serine, due to an increased d-serine occupancy of the D-serine-binding site in the NMDAR⁵⁷. As shown in Fig. 4e, NMDAR responses were significantly increased after the addition of 50 µM D-serine in both p110α^aKO and control slices, but this enhancement was significantly stronger in astrocytic p110α-deficient animals (Fig. 4e, right panel). This result strongly suggests that D-serine availability is deficient in the absence of astrocytic p110α. To note, this result is not contradictory with the normal NMDAR transmission we had observed in p110α^aKO animals (Fig. 3d), as this lower D-serine availability may be sufficient to sustain basal, low-frequency NMDAR responses.

To test whether a deficit in D-serine availability is responsible for the LTP impairment in the absence of astrocytic p110α, we carried out LTP experiments in the presence of 10 µM D-serine in the perfusion during the recordings. Under these conditions, the slices from p110α^aKO mice had normal LTP, comparable to controls, also in the presence of D-serine (Fig. 4f). Also, the deficient NMDAR activation during LTP induction in p110α^aKO slices was rescued in the presence of D-serine (Supplementary Fig. 4A–C). Then, to address whether this recovery of LTP was reflecting a specific deficit in D-serine production or a deficient biosynthesis of its precursor, L-serine, we tried these same rescue experiments with 50 µM L-serine in the perfusion during the recordings. Remarkably, the slices from p110α^aKO mice in the presence of L-serine showed levels of LTP comparable to those seen in the treated controls (Fig. 4g). To note, the extent of LTP expression in control slices was similar with or without D- or L-serine (compare Fig. 3a with Fig. 4f, g), arguing against a general enhancing effect of these metabolites on LTP. Finally, neither D-serine nor L-serine altered basal synaptic transmission in p110α^aKO animals as compared to controls (Supplementary Fig. 4D, E). Therefore, these combined results strongly suggest that the impairment in NMDAR activation during LTP induction in the absence of astrocytic p110α is due to a deficient biosynthesis of L-serine in astrocytes.

L-serine supplement in vivo rescues the behavioral deficits in p110α^aKO mice

Having observed that l-serine treatment in vitro was enough to rescue the LTP impairment of p110α^aKO mice, we decided to test whether the cognitive deficits of these animals could be rescued in vivo with a similar approach. To this end, we supplemented the mouse diet with 5 g/L of L-serine dissolved in the drinking water for 3 weeks, with the treatment starting after 2 weeks of infection and lasting until all behavioral assays were completed. After the behavioral assays, animals were sacrificed and D-serine was measured from total hippocampal extracts. Indeed, the treatment with L-serine significantly increased the levels of D-serine in p110α^aKO-treated animals, which were now similar to those of control animals (Fig. 5a). These results indicate that the in vivo L-serine treatment is effective in increasing D-serine concentration in p110α^aKO animals.

Fig. 5. L-Serine treatment in vivo rescues cognitive performance in p110α astrocytic knock-out mice.

a D-serine levels in total hippocampal extracts from control and L-serine-treated animals. Control n = 12, p110α^aKO n = 10, control + L-serine n = 13, p110α^aKO + L-serine n = 16 mice. Discrimination index for the novel location (b) or for the novel object recognition (c). The untreated controls and untreated p110α^aKO mice DIs are replotted from Fig. 2d, e, as reference. Individual values for each mouse and condition are represented along with the mean ± SEM as aligned dot-plots. To assess whether the DIs were different from zero, we performed the Wilcoxon signed-rank test (# p < 0.05, ## p < 0.001, ns: not significant). One-way ANOVA was used to assess statistical differences between groups. Pairwise comparisons of DIs between p110α^aKO vs. control + L-Ser or p110α^aKO vs. p110α^aKO + L-Ser were carried out with a two-tailed unpaired t-test, p110α^aKO vs. control + L-Ser p = 0.0072, p110α^aKO vs. p110α^aKO + L-Ser p = 0.0170 (b); p110α^aKO vs. control + L-Ser p = 0.0064, p110α^aKO vs. p110α^aKO + L-Ser p = 0.0058 (c). Control n = 17–19, p110α^aKO n = 17–18, Control+L-ser n = 7–12, p110α^aKO + L-Ser n = 10–16 mice. Source data are provided as a Source Data file.

We then tested the behavioral assays in which we had found p110α^aKO animals to be deficient, namely NOR and NOL tests (Fig. 2d, e). For the NOL test, treated p110α^aKO animals displayed significant discrimination for the object in the new location, which was now comparable to the discrimination of control animals with or without treatment (Fig. 5b). The results for the NOR test were similar: the p110α^aKO mice that had received the L-serine supplement were able to discriminate the new object over the familiar one significantly better than the untreated knockout animals, and were comparable to treated or untreated control animals (Fig. 5c).

These results indicate that a supplement of L-serine in the drinking water is enough to recover the cognitive deficits of mice lacking the p110α catalytic subunit of the PI3K in astrocytes. Also, they reinforce the link between LTP processes and learning, as well as the role of l-serine from the astrocyte for the correct functioning of neuronal circuits.

The lack of p110α decreases glycolytic flux and increases mitochondrial respiration in astrocytes

As mentioned above, it is well established that the PI3K pathway constitutes a central hub for the regulation of glycolysis^33,37,41,42. In addition, one of the major routes for L-serine biosynthesis is the phosphorylated pathway, which starts from the glycolytic intermediate 3-phosphoglycerate^22,58. Therefore, it is reasonable to hypothesize that the absence of p110α in astrocytes leads to a reduced L-serine biosynthesis because of a lower glycolytic flux. To test this hypothesis, we evaluated glycolytic function at two points: glucose uptake and lactate production.

First, we assessed glucose uptake from the incorporation of the glucose analog 2-deoxyglucose (2-DG) in pure astrocytic cultures from p110α^flox/flox mice infected with an adenovirus expressing Cre-mCherry, an empty vector (EV-mCherry), or an uninfected condition (Supplementary Fig. 5A, B). As shown in Fig. 6a, p110α^aKO astrocytes displayed a significant reduction of glucose uptake, compared to both EV-infected and uninfected astrocytes. Then, we analyzed lactate production as a readout of glycolytic flux by measuring the lactate concentration in the culture media. Consistent with the reduction in glucose uptake, we observed a significant decrease in lactate levels in the media of p110α^aKO astrocytes (Fig. 6b).

Fig. 6. The lack of p110α in astrocytes decreases glucose uptake and lactate production in the astrocyte, while increasing mitochondrial size and function.

a 2-DG uptake from astrocyte cultures infected with the adenovirus expressing Cre, with the empty vector (EV), or uninfected. All results are normalized to the 2-DG uptake in uninfected astrocytes. Uninfected n = 9, empty vector n = 13, and p110α^aKO n = 12 samples. b Extracellular lactate concentration levels from astrocytic culture media normalized to the lactate concentration in the uninfected condition. Uninfected n = 15, empty vector n = 18 and p110α^aKO n = 20 samples. c Representative images from the iGlucoSnFR sensor expressed in astrocytes in acute slices, with 0 and 20 mM glucose in the perfusion. Arrowheads indicate astrocytic somas (encircled), as examples for quantification. Scale bar: 20 μm. d Time course depicting the increase in fluorescence in astrocytes upon addition of 20 mM glucose in the perfusion. e Fluorescence increase measured from the last minute of the recording. Control n = 36, p110α^aKO n = 56 astrocytes. Statistical differences were assessed by Kruskal–Wallis and Dunn’s multiple comparisons tests (a, b) or by unpaired t-test and one-sample t-test (e). f Electron microscopy representative images of astrocytic mitochondria from control (left) and p110α^aKO (right) mice from the stratum radiatum of acute hippocampal slices. Scale bar: 0.25 µm. g Mitochondrial area. Control n = 181, p110α^aKO n = 186 mitochondria. h The aspect ratio corresponds to the relation between the major and the minor axis of the mitochondria. Control n = 201, p110α^aKO n = 186 mitochondria. i Number of mitochondria per area. Control n = 23, p110α^aKO n = 18 sections. g–i n = 3 mice per condition. Bars represent mean ± SEM and individual values from each sample are represented as a dot-plot overlay. Statistical differences were assessed by the Mann–Whitney test. j Merged representative image of the coinfection with the AAV5-GFAP-Cre-GFP under and the AAV1-mitoTag in the stratum radiatum of mouse organotypic slices. Cre-positive nuclei are shown in green and anti-HA labeled mitochondria in magenta. Scale bar: 25 µm. k OCR values normalized to protein levels from control and p110α^aKO. Control n = 4, p110α^aKO n = 5 mice. Statistical differences were assessed by two-way ANOVA matched-values and Bonferroni post-hoc tests (*p < 0.05, **p < 0.01, ***p < 0.001). PMG pyruvate, malate, glutamate; CCCP carbonyl cyanide 3-chlorophenyl. Source data are provided as a Source Data file.

Subsequently, to evaluate astrocytic glucose uptake under a more physiological setting, we employed AAV5 vectors to express the glucose biosensor iGlucoSnFR⁵⁹ under the GFAP promoter in the hippocampus of adult p110α^flox/flox mice. In this case, animals were co-infected with AAV5-GFAP-Cre-mCherry to simultaneously remove p110α in astrocytes (animals without Cre-mCherry infection, but expressing iGlucoSnFR, served as wild-type control; see representative images in Supplementary Fig. 5C). After 1 month of expression, hippocampal acute slices were prepared and the fluorescent signal of iGlucoSnFR in astrocytes was monitored in response to a glucose challenge from 0 to 20 mM (see representative image in Fig. 6c). As shown in Fig. 6d, e, control astrocytes displayed an increase in iGlucoSnFR fluorescence upon glucose addition. In contrast, the increase in fluorescence was significantly smaller and not sustained in p110α-deficient astrocytes. To note, this reduction in glycolytic flux occurred without detectable changes in the expression of the serine metabolism enzymes PHGDH or SHMT1, using total hippocampal extracts of control and p110α^aKO animals (Supplementary Fig. 5D, E). Intriguingly, the astrocytic glucose transporter GLUT1 and the glycolytic enzyme PKM2 were increased in the absence of astrocytic p110α (Supplementary Fig. 5D–G), perhaps as a compensatory effect.

Therefore, these combined results suggest that p110α is an important driver of glycolysis in the astrocyte, evidenced both by a lower uptake of glucose and by a reduced lactate yield.

To further test the role of p110α in the regulation of metabolism in astrocytes, we decided to check the morphology and function of the mitochondria in p110α^aKO animals. For that purpose, we used the p110α^flox/flox mice injected with the AAV5-GFAP-Cre-GFP and their genotype-paired controls. After one month of expression, we prepared hippocampal sections and processed them for electron microscopy. We specifically imaged the CA1 stratum radiatum and identified astrocytes from their clearer cytoplasm and glycogen granules. As shown in Fig. 6f, g, astrocytes from mice lacking p110α had slightly bigger mitochondria than controls, without changes in their shape (aspect ratio values; Fig. 6h) or density (Fig. 6i). Also, mitochondria cristae were slightly longer and thinner in p110α^aKO mitochondria, without differences in number (Supplementary Fig. 6). To evaluate mitochondrial function in the absence of astrocytic p110α, we measured the oxygen consumption rate from isolated mitochondria of astrocytes. For these experiments, we took advantage of the Cre-dependent mitoTag adeno-associated viral vector (AAV1-mitoTag) that expresses a recombinant TOMM20 protein fused to three hemagglutinin A (HA) in a Cre-dependent manner⁶⁰. We coinfected p110α^flox/flox mice with AAV5-GFAP-Cre-GFP and AAV1-mitoTag. In this manner, the Cre recombinase will drive the deletion of p110α from astrocytes and will tag their mitochondria through the HA-TOMM20 expression (see Fig. 6j for a representative image of coinfected astrocytes expressing Cre-GFP—green channel- and showing HA-labeled mitochondria—magenta channel). For these experiments, the controls were wild-type mice coinfected with the same AAVs, so that the Cre expression would drive astrocytic mitochondrial tagging, but would not induce the p110α knock-out. The mitochondrial isolation was performed using whole hippocampal extracts incubated with anti-HA-coated magnetic microbeads. Using these preparations, we observed that astrocytic mitochondria from p110α^aKO mice had a marked enhancement of oxygen consumption rate compared to the wild-type controls. This enhancement was statistically significant after the addition of complex II substrate succinate and for maximal respiration driven by the mitochondrial uncoupler CCCP (Fig. 6k).

All together, these results point to an altered metabolism of astrocytes in the absence of p110α, which results in a defective glycolytic flux and an increase in mitochondrial respiration, perhaps as a compensatory mechanism for the glycolytic deficiency.

Discussion

Astrocytes play a pivotal role in neuronal function not only by gliotransmission and modulation of synaptic responses, but also through energetic and metabolic regulation. This work sheds light on this matter, unraveling the distinct contributions of astrocytic PI3K signaling, and specifically the p110α isoform, to synaptic plasticity and cognitive function, highlighting its connection with metabolic regulation. In addition to the knowledge on basic mechanisms of synaptic plasticity, we believe this study provides further insight into brain disorders related to the PI3K pathway^31,61 and/or D- or L-serine deficiencies^27,62.

Perhaps one of the major contributions of this study is the dissection of specific mechanisms by which complex signaling pathways, with pleiotropic outcomes, may act on highly specialized cellular functions, such as synaptic plasticity and neuronal communication. The case of PI3K signaling is particularly paradigmatic. This pathway is long-known to be important for LTP^63–66. Within neurons, its role is related to AMPAR synaptic delivery and anchoring^29,32,65. However, we now show that this signaling pathway also contributes to LTP from a different cellular compartment, the astrocyte, and employing a completely different cellular function: regulation of glycolytic flux with the concomitant modulation of L-serine biosynthesis, which in turn determines D-serine availability for NMDAR activation during LTP. Furthermore, our results suggest that p110α plays a constitutive role in astrocytic glycolysis, since glucose uptake in p110α-deficient astrocytes is impaired under basal conditions in hippocampal slices or even in astrocytic primary cultures. It is also interesting that interfering with LTP by ablating p110α in neurons or in astrocytes produced almost opposite behavioral outcomes with respect to NOR, NOL, and Y-maze (this study versus ref. ³²). It then appears as if the locus of the impairment (astrocyte versus neuron) and the mechanism involved (NMDAR activation versus AMPAR recruitment) are more important for the behavioral outcome than their convergence on LTP impairment. Of course, these distinct roles at neurons and astrocytes are relevant for pharmacological studies, where PI3K may be blocked broadly across the tissue (for instance, in some early studies^63–65), but also when considering genetic mutations associated to human disease (for example, related to intellectual disabilities³¹), where the same mutation in different cell types (neurons, astrocytes) may contribute in distinct manners to the pathology.

In this sense, it is interesting that such a broad alteration in astrocytes (reduced glycolytic flux) may produce a precise phenotype in neurons: impaired long-term potentiation. Indeed, we have observed that other parameters of synaptic communication, such as basal synaptic transmission (mediated by AMPARs and NMDARs) or long-term depression, were not altered. This is likely reflecting the strong dependence on D-serine for LTP^49,50,54. In fact, we have observed that NMDAR responses increase across the consecutive rounds of LTP induction, most likely because of an activity-dependent enhancement of D-Ser availability. According to our results, this enhancement cannot be sustained in the context of a reduced glycolytic flux in the astrocyte, as caused by the absence of p110α. In a sense, this scenario is similar to the specific impairment in LTD we have observed when we manipulated another important signaling pathway in the astrocyte: p38α MAPK⁷. In this case, we found that deletion of astrocytic p38α produced a deficient release of glutamate from astrocytes, which again impaired the activation of neuronal NMDARs, but specifically during LTD induction⁷. Thus, we believe these studies are contributing to better define the intracellular signaling framework in the astrocyte that impinges on specific neuronal synaptic plasticity events.

We also observed that deletion of astrocytic p110α produced a reduction in neuronal spine density with a concomitant increase in their length. These effects may reflect the presence of more immature spines, perhaps as a consequence of the failure to be potentiated and stabilized^67,68. In turn, the reduced number of spines and their immature nature may affect their capacity to be potentiated and would ultimately impact learning and memory. Moreover, these effects could also be a consequence of the decreased lactate production in the p110α^aKO astrocytes, as it has been shown that lactate is important for LTP and spine dynamics^25,69.

On the other hand, we acknowledge that the specific alteration responsible for the reduced glycolysis in astrocytes lacking p110α remains to be determined. The PI3K/AKT pathway has been shown to regulate glycolytic flux by several mechanisms in other tissues^33,41,70,71. Interestingly, the knock-out for Akt2, the AKT isoform present in astrocytes⁷², is a common model for a type 2 diabetes-like syndrome, which shows an impairment in LTP and spatial memory^73,74. Moreover, insulin resistance, a hallmark of type 2 diabetes, is linked to memory deficits and synaptic plasticity dysregulation in neurodegenerative diseases, aging, or Alzheimer’s disease^75–77, and astrocytic insulin receptor knock-out mice showed blunted responses to insulin and reduced brain glucose uptake^78,79. This background suggests that the p110α^aKO might exhibit a diabetes-related phenotype, leading to a glycolytic deficiency, coupled with a decrease in glucose uptake and a decrease in lactate production, which finally causes an impairment in long-term potentiation and memory.

It is also notable that astrocytes with reduced glycolytic flux in the absence of p110α displayed enhanced mitochondrial respiration. While astrocytes are considered mainly glycolytic cells, the role of mitochondria in astrocytes should not be overlooked, taking into account the differentially regulated enzymes and the distinct assembly of electron transport chain supercomplexes between neurons and astrocytes^80,81. In fact, previous p110α knock-out studies in muscle and adipose tissue have shown enlarged mitochondria with enhanced oxidative capacity^82,83, in line with our observations in astrocytes. These results suggest a possible functional reorganization and adaptation of these knock-out astrocytes towards a more oxidative profile to sustain their own metabolic demands, possibly via fatty acid oxidation^80,84. On the other hand, these broader changes in astrocytic metabolism in response to p110α removal should also serve as a cautionary note for potential indirect effects at the synaptic and behavioral levels. Thus, even though synaptic and behavioral deficits were rescued by supplementation with D-serine (ex vivo) and L-serine (in vivo and ex vivo), we cannot rule out other metabolic changes caused by p110α removal in astrocytes that may also be contributing to the phenotypes we are describing here.

Finally, the fact that the cognitive defect of p110α^aKO mice can be rescued in vivo by L-serine supplementation opens the possibility of dietary interventions with therapeutic potential. Indeed, L-serine is used as a routine food additive and dietary supplement, and it is well tolerated even at high doses for therapeutic applications^62,85.

In summary, the results presented in this work highlight the dual role of p110α in orchestrating synaptic plasticity and metabolism in astrocytes in the hippocampus, with functional consequences for cognitive performance. Also, the application of an in vivo treatment to rescue the behavioral deficits paves the way for possible therapeutic strategies concerning neurological traits based on NMDAR hypofunction.

Methods

Animals and ethics statement

All biosafety procedures and animal experimental protocols were approved by the Ethical Committee of the Consejo Superior de Investigaciones Científicas (CSIC), in strict accordance with Spanish (RD 53/2013, 32/2007) and EU guidelines set out in the European Community Council Directives (2010/63/EU, 86/609/EEC; PROEX 160.1/21). Animal’s health and welfare were monitored by a designated veterinarian. Male and female animals were used without distinction in this study. All mice used were housed in standard cages (a maximum of 5 animals per cage) in humidity and temperature-controlled rooms. Autoventilated racks contained individually ventilated cages with absolute filters. Irradiated standard safe diet and autoclaved water were available ad libitum.

p110α^flox/flox is a transgenic mouse line with the 18^th and 19^th exons of the PIK3CA gene flanked by two LoxP sequences. These mice were generously donated by Bart Vanhaesebroeck (University College London, UK). Transgenesis procedures to generate these mice were previously described⁴⁷. Their genetic background is C57BL/6J. Genotyping was carried out by Polymerase Chain Reaction (PCR) using as primers 5′-GGATGCGGTCTTTATTGTC and 5′-TGGCATGCTGCCGAATTG (obtained from ThermoFisher). For the oxygen consumption rate experiments, C57BL/6J-OlaHsd wild-type mice were used as controls.

Stereotaxic in vivo microinjections of adeno-associated (AAVs) and Sindbis viruses

AAV5-GFAP-Cre-GFP, AAV5-GFAP-Cre-mCherry (serotype 5; Gene Therapy Vector Core at the University of North Carolina), AAV5-GfaABC1D-Cre-4x6T (Addgene plasmid #196410) and AAV5-GFAP104-mCherry (Addgene plasmid #58909) were delivered to the hippocampus of 2–3 months old mice through two bilateral injections (600 nl each) in the dorsal (in mm from bregma, anteroposterior (AP): −1.8, mediolateral (ML): ±1.3, dorsoventral (DV): −1.7) and ventral (AP: −2.8, ML: ±2.4, DV: −2) regions. Some animals were also injected with AAV1/2-GFAP-(Cyto)iGlucoSnFR for glucose imaging (Addgene plasmid #164513). Animals were placed in the stereotaxic frame (Harvard Apparatus) over a heating pad and immobilized using blunt ear holders. Anesthesia by inhalation of isoflurane was initiated at 2.5–3% in oxygen and maintained at 1.5% during the rest of the procedure. Before the start of the procedure, the animals received analgesia by subcutaneous injection of Meloxicam (5 mg/kg dose) and ocular protection with lipolac ophthalmic gel 2 mg/g (Bausch & Lomb, #764118). Viral particles were infused using a Hamilton Neuros syringe (Hamilton, #65460-02) inserted through small holes drilled into the skull and coupled to a micropump (Quintessential stereotaxic injector, Stoelting, #53311) at an infusion rate of 120 nl/min. After the infusion, the syringe was kept in place for 5 min, and slowly withdrawn. The virus was allowed to express for 1 month, and the accuracy of the infection was confirmed by monitoring the expression of the fluorescent reporter. For neuronal morphology experiments, injected mice were subjected to a second round of in vivo injections using the previous parameters to inject Sindbis viruses expressing GFP, and animals were sacrificed 24 h later.

For mitochondrial isolation from genetically modified astrocytes, the AAV5-Cre was coinfected with an AAV expressing a Cre-inducible mitoTag under the EF1α promoter (serotype 1; 500 nl per injection). As previously described⁶⁰, this AAV1-mitoTag expresses a Tomm20 gene fused to three HA sequences that allow the immunoisolation of mitochondria from Cre-expressing cells.

L-serine treatment

According to previous publications, L-serine can be safely administered with the diet through the food or the drinking water^27,86,87. For the serine-treated animals in this study, a concentration of 5 g/L of L-serine was dissolved in the drinking water, and the treatment was refreshed every 2 days for 3 weeks. A group of untreated control animals was kept in parallel.

Behavioral assays

Behavioral tests were conducted with 3–4-month-old mice of both sexes. After handling the animals for a week, tests were performed on consecutive days. Test cages and objects were cleaned thoroughly with 70% ethanol between subjects and phases to eliminate any olfactory cues. For all behavioral assays, animals were recorded from a camera located above the testing arena, and analysis was performed manually using the AnyMaze software (Stoelting). All behavioral experiments were carried out and analyzed blindly with respect to genotype.

Open field test

This test was used to evaluate general locomotor activity and as a habituation phase to novel object location and recognition tests. Mice were placed individually in the center of an acrylic box (40 × 40 × 40 cm) with white opaque walls at ground level. The animals were allowed to explore the arena for 15 min while being recorded from above the test field.

Y maze test

The Y maze test was performed to quantify spontaneous alternation and assess deficits or changes in spatial working memory. The apparatus consists of a Y-shaped maze with three identical enclosed arms. To perform the test, we placed each mouse in the center of the maze and recorded a single trial of 5 min from above. The sequence of arm entrances was quantified manually, and an alternation is considered when the animal enters consecutively in the three arms. The spontaneous alternation percentage was calculated as: % spontaneous alternation = (number of alternations)×100/(total number of arm entrances–2).

Novel object location (NOL) and recognition (NOR) tests

These tests were performed in two consecutive days, 24 h after the open field test, which was used as the first habituation phase. For the next phases, the same open field arena was used, but in the case of the NOL, the four walls were different in order to give the mice spatial cues. In the familiarization phase of the NOR, two identical objects in terms of color, texture, and shape were placed on two adjacent corners of the arena. Mice were allowed to explore the objects for 7 min while being recorded, and after 30 min one of the objects was replaced with another one with a different color and texture but similar dimensions. Mice were allowed to explore the objects for 7 min. The objects used were a pair of drinking bottles and a pair of cylindrical glass bottles. Both groups of objects were randomly assigned as familiar or novel among the subjects of study to ensure that they showed no preference for any of them. For the NOL, the objects were a pair of culture flasks identical in shape, dimensions, and texture. During the familiarization phase, both objects were placed in two adjacent corners of the arena, and after 30 min the position of one of them was changed so that both objects were placed on the diagonal axis. Exploration of the objects was defined as close contact sniffing (with the head directed towards the object). We then calculated a discrimination index (DI) that is the time spent exploring the relocated or the novel object minus the time exploring the familiar object or the object in the familiar position, divided by the total exploration time.

Social interaction and social memory tests

This task is used to evaluate mouse preference to establish social contact and their social interaction skills, and they also rely on the mice’s innate preference for novelty. For this test, we used a three-chamber acrylic box with clearly divided walls with a hole to give the testing mouse access to each chamber. The test consisted of three phases. In the first phase, wire cages were placed on the two outermost chambers, and the testing mice were allowed to explore the arena and the cages for 5 min. These cages allowed visual, auditory, olfactory, and minimal touch interaction. The second phase corresponds to the social interaction task, and it was performed 24 h later. In this test, an unfamiliar mouse was placed under the wire cage of one of the chambers (sex and age-paired), and an inanimate object was placed in the opposite cage in an identical configuration. The unfamiliar mice had been habituated to the cages for several days in time intervals of 15 min. The subject mouse was placed in the center chamber and allowed to freely explore all three chambers for 7 min. The test mouse was removed, and the third phase was immediately started after cleaning the arena and the wire cages thoroughly with 70% ethanol to eliminate any olfactory cues. The third phase corresponds to the social memory task, in which the inanimate object was replaced by another unfamiliar mouse, and the test was performed identically as in the second phase. Behavior was recorded from a camera located above the test box, and analysis was performed manually. We measured the time spent on each chamber and exploration, which was defined as close-contact sniffing of the wire cage (with the head directed towards the cage). Sitting or standing time on the wire cages was not quantified as exploration. We also calculated the discrimination index as the time exploring the subject minus the time exploring the object divided by the total exploration time, or the time exploring the new subject minus the time exploring the familiar one divided by the total exploration time.

Reagents

2-chloroadenosine, picrotoxin, and DL-APV were purchased from Sigma-Aldrich. D-serine and L-serine were purchased from Thermo Scientific. The antibody against p110α was generated in Mariona Graupera’s laboratory. Other primary antibodies were p110β (Abcam, #ab151549), NeuN (Abcam, #ab177487), Akt (Cell Signaling Technologies, #2920), phospho-Akt [T308] (Cell Signaling Technologies, #2965), phospho-Akt [S473] (Cell Signaling Technologies, #4060), phospho-GSK3α/β [S21/S9] (Cell Signaling Technologies, #9331), α-Tubulin (Sigma, #T6199), GFAP (Sigma, #G3893), GFP (ThermoFisher Scientific, #A-6455), mCherry (GeneTex, #GTX59788), GSK3α/β (Invitrogen, #44-610), GLUT1 (Santa Cruz Biotechnology, #sc-7903, for WB, and Merck, #CBL242, for IHC), PKM2 (Cell Signaling Technologies, #3198), PHGDH (Proteintech, #14719-1-AP), SHMT1 (Proteintech, #14149-1-AP) and hemagglutinin-A (HA, Covance/Biolegend, #MMS-101P).

Primary astrocytic cultures and viral delivery

Primary cultures of hippocampal astrocytes were obtained from 1- to 3-day-old p110α^flox/flox mice. Pups were decapitated, brains were extracted, whole hippocampi were dissected, and tissue was mechanically disintegrated. Then, the tissue was enzymatically digested with Accutase for 5 min at 37 °C in phenol red-free DMEM F12 (Gibco). Cells were centrifuged and plated in 0.1 mg/ml poly-L-Lysine-coated P100 plates in phenol red-free DMEM F12 medium supplemented with 10% (v/v) fetal bovine serum (FBS), 1 mM sodium pyruvate, 2 mM L-glutamine, 1% non-essential amino acids, and 1% penicillin/streptomycin. Cells were maintained at 37 °C and 5% CO₂. The medium was changed the day after and subsequently every 2–3 days until cells reached confluence. Then, cells were isolated by trypsinization and plated again, so a pure preparation of astrocytes was obtained.

For viral delivery, 7 days in vitro, astrocytes were seeded in P35 plates and infected with an adenovirus CMV-Cre-IRES-mCherry or an empty vector CMV-mCherry (Viral Vector Production Unit, UPV, Universitat Autònoma de Barcelona) for 5 days before performing glucose uptake or extracellular lactate measurements.

Acute hippocampal slices

Acute hippocampal slices were obtained from 3 to 4-month-old mice of both sexes. Animals were anesthetized by isoflurane inhalation and quickly decapitated once they were unresponsive to tail and foot pinches. The brains were rapidly extracted and submerged in Ca²⁺-free ice-cold dissection solution (10 mM D-glucose, 4 mM KCl, 26 mM NaHCO₃, 233.7 mM sucrose, 5 mM MgCl₂, and 0.001% (w/v) phenol red as a pH indicator) previously saturated with carbogen (5–10% CO₂, 90–95% O₂). Coronal slices (300 µm) were obtained by cutting the brain in the same solution with a vibratome (Leica, VT1200S) and left in carbogen-gassed artificial cerebrospinal fluid (aCSF: 119 mM NaCl, 2.5 mM KCl, 1 mM NaH₂PO₄, 26 mM NaHCO₃, 11 mM glucose, 1.2 mM MgCl₂, 2.5 mM CaCl₂, and osmolarity adjusted to 290 mOsm) for 1 h at 32 °C for recovery. After that time, the slices were maintained at 25 °C, and the recordings were performed in aCSF.

Electrophysiology

Field excitatory postsynaptic potentials (fEPSPs) were recorded from CA1 pyramidal neurons with recording glass electrodes while stimulating the Schaffer collateral fibers to evoke synaptic responses. During the recordings, the slices were placed in an immersion chamber constantly perfused with aCSF gassed with carbogen (5–10% CO₂, 90–95% O₂) and its temperature closely monitored at 25 °C. To test the D-serine occupancy of NMDARs, NMDAR fEPSPs were recorded in an aCSF with low Mg²⁺ (0.2 mM) and supplemented with 10 µM CNQX (AMPAR antagonist). For all experiments, the aCSF was supplemented with 100 µM picrotoxin (GABA_A receptor antagonist). For some recordings, 100 μM DL-APV was added to the aCSF to block NMDARs. For l-Serine and l-Serine rescue experiments, 10 and 50 µM were added to the aCSF, respectively.

Glass pipettes were filled with the aCSF used as an extracellular solution and placed in the CA1 stratum radiatum.

fEPSPs were recorded at different stimulation intensities (50 µs and 20–250 µA) on each slice to generate an input–output curve, and stimulation intensity was adjusted to 30% (for NMDAR-LTP and PPF) or 50% (for NMDAR-LTD) of the maximum response. At least 20 min of stable baseline was recorded (0.067 Hz stimulation) prior to plasticity induction or to D-serine application (50 µM) in the case of D-serine NMDAR occupancy experiments. NMDAR-dependent LTP was induced with a theta burst stimulation (TBS) protocol composed of 10 trains of bursts (4 pulses at 100 Hz with a 200 ms interval), and it was repeated for 4 cycles with a 20 s inter-cycle interval. NMDAR-dependent LTD was induced using a low-frequency stimulation protocol (LFS) that delivers 900 pulses at 1 Hz. Paired-pulse facilitation (PPF) experiments were performed with pairs of pulses of varying inter-stimulus intervals (50, 100, 200, and 400 ms). Data acquisition was achieved using Multiclamp 700A/B amplifiers and pCLAMP 9/10 software (Molecular Devices), and data analysis was performed with custom-made Excel (Microsoft) macros and with Clampfit software (Molecular Devices). All electrophysiology experiments were carried out and analyzed blindly with respect to genotype.

Immunohistochemistry and confocal microscopy

Mice were anesthetized by isoflurane inhalation and intracardially perfused with 0.1 M pH 7.4 phosphate buffer (PB) to wash the vascular system prior to injecting the fixative (4% paraformaldehyde in PB, pH 7.4). Brains were post-fixed overnight, immersed in the same fixative solution at 4 °C, and then sequentially dehydrated in solutions containing 15% and 30% sucrose in PB at 4 °C until they sank. Brains were cut in 50–75 µm coronal slices in cold phosphate-buffered saline (PBS) with a vibratome and then blocked with 5% bovine serum albumin (BSA) (Sigma, #A4503) and 0.25% Triton X-100 (BioRad, #1610407) in PBS for 2 h at room temperature. Once blocked, slices were incubated with the primary antibody in blocking solution at 4 °C overnight. After several washes, slices were incubated with the secondary antibody, being Alexa 488, 555, or 647 (ThermoFisher), depending on the experiment. Finally, they were stained with 4′,6-diamidino-2-phenylindole (DAPI; 1 μg/ml) for 5 min and mounted on adherent microscope slides (Thermo Scientific, #15438060) with Prolong Glass Antifade (ThermoFisher, #P-36982). Fluorescence images were acquired as z-stacks with a confocal inverted microscope (LSM800, Zeiss) using a ×20 NA 0.8 Plan-Apochromat or ×40 NA 1.3 oil Plan-Apochromat objectives and 488, 555, and 647 nm lasers in combination with ZenBlue 3.3 software. Z-stacks were reconstructed (maximum intensity projection) using Fiji v1.54f.

For animals infected with AAV5-GFAP-Cre-GFP, the percentage of infection was quantified as the ratio of GFP-positive cells over the total GFAP-positive astrocytes.

For neuronal and astrocytic morphology studies, neuronal dendritic trees or astrocytic processes were two-dimensionally traced with the simple neurite tracer plug-in. Sholl analysis was performed, tracing the number of intersections in 10 µm intervals from the soma for neurons, and the total basal and apical dendrite length was also measured. In the case of the astrocytic morphology, the tracing intersections were performed in 4 µm intervals. For spine morphology, images were acquired using a ×63 NA 0.8 Plan-Apochromat M27 oil immersion objective and a 488 nm laser. Spine density was defined as the number of spines divided by the dendritic length.

For the astrocytic GLUT1 expression analysis, z-stacks from 647-nm laser images were reconstructed. Diffuse signal from the stratum radiatum was analyzed as an indicator of GLUT1 astrocytic expression as previously described⁸⁸, using regions of interest (ROI) on stacked corrected images from each hippocampal slice. Mean intensity fluorescence on each ROI was normalized to control mice values after subtracting the background signal, obtained from the slices incubated with no primary antibody.

Glucose live imaging

For live imaging experiments with the glucose biosensor iGlucoSnFR⁵⁹, acute hippocampal slices from previously injected mice were obtained as described above. After the recovery period, they were incubated in zero-glucose carbogen-gassed aCSF (119 mM NaCl, 2.5 mM KCl, 1 mM NaH₂PO₄, 26 mM NaHCO₃, 0 mM glucose, 10 mM sucrose, 1.2 mM MgCl₂, 2.5 mM CaCl₂, and osmolarity adjusted to 290 mOsm) for 40 min at room temperature. Glucose was replaced with sucrose to maintain osmolarity.

Slices were placed in an immersion chamber constantly perfused with either zero-glucose aCSF or high-glucose aCSF (119 mM NaCl, 2.5 mM KCl, 1 mM NaH₂PO₄, 26 mM NaHCO₃, 20 mM glucose, 1.2 mM MgCl₂, 2.5 mM CaCl₂, and osmolarity adjusted to 290 mOsm) gassed with carbogen and maintained at 33 °C using an in-line solution heater (SH-27B, Warner Instruments). Confocal fluorescence images were acquired as Z-stacks at 1 μm step size with a confocal inverted microscope (LSM800, Zeiss) using a ×20 NA 0.8 Plan-Apochromat objective and 488 and 533-nm excitation lasers in combination with an emission range of 410–546 and 565–700 nm, respectively, with ZenBlue 3.3 software. Images were acquired every 30 s with an average perfusion flow of 2.5 ml/min.

For image analysis and quantification, the software Fiji v1.54f (ImageJ; National Institutes of Health) was used. Z-stacks with maximum intensity projections from 488-nm laser images were reconstructed and motion-corrected using the StackReg plug-in. Maximum intensity projections allow us to track individual cells even if there are small vertical changes in the focal plane (due to perfusion flow) during the time course. ROI of the somatic area of the astrocytes expressing the sensor in each field across time frames was analyzed. Examples of ROI delimitating the soma of individual astrocytes expressing the sensor are shown in Fig. 6c. Mean intensity fluorescence on each ROI was normalized to its baseline at 0 mM extracellular glucose. Cells that showed an increase higher than 7% of mean intensity fluorescence were excluded from the analysis. Values from the last five frames on the recording were taken for comparison between conditions. Each condition was compared to its normalized baseline. Representative image insets were prepared for figure presentation by applying brightness and contrast uniformly, pseudocolored, and smoothed with a Gaussian blur filtering.

Electron microscopy

Mice were anesthetized by isoflurane inhalation and intracardially perfused with PB prior to injecting the fixative (4% paraformaldehyde, 2% glutaraldehyde in 0.1 M PB, pH 7.4). Brains were left for 2 h at room temperature and overnight at 4 °C, immersed in the same fixative solution. Coronal slices of 200 µm were obtained with a vibratome and post-fixed in a 2% solution of osmium tetroxide in 0.1 M PB (1.5 h, RT). The slices were then washed and stained with 2% uranyl acetate in water (1 h, RT in darkness), washed, dehydrated in increasing concentrations of ethanol, and embedded in Epon resin. A series of ultrathin sections of the CA1 SR region of the hippocampus was collected and mounted on single oval slot copper/palladium grids. The sections were imaged on a JEM-1400 Flash transmission electron microscope (JEOL) coupled to a CMOS Oneview camera (Gatan). The SR was imaged with a ×6000–×8000 magnification to quantify mitochondrial density and a ×20,000–×30,000 magnification for mitochondrial morphology, measured using Fiji v1.54f as previously described⁸⁹. We measured the mitochondrial area, the aspect ratio (major/minor axis length), and the length and width of the cristae.

Western blotting

Whole hippocampi were homogenized in ice-cold lysis buffer containing 150 mM NaCl, 50 mM Tris–HCl (pH 7.5), 1 mM EDTA, 1% Triton X-100, Complete EDTA-free, and PhosSTOP and centrifuged at 13,000×g for 5 min at 4 °C. The resultant supernatant was processed by SDS–polyacrylamide gel electrophoresis and transferred to polyvinylidene fluoride (PVDF) membranes (Immobilon-P, Merck Millipore). Membranes were blocked for 1 h in 5% BSA in Tris-buffered saline with 0.1% Tween-20, then incubated with primary antibodies overnight at 4 °C in blocking solution. Membranes were washed and incubated for 1 h at room temperature in the corresponding horseradish peroxidase-conjugated (HRP) secondary antibodies (Jackson ImmunoResearch) prepared in blocking solution. Immunodetection was carried out by chemiluminescence (Immobilon Western, Millipore, WBKLS0500) using ImageQuant™ LAS 4000 mini biomolecular imager (GE Healthcare Life Sciences). The ImageJ software was used to analyze and quantify signal intensities obtained from digital images.

Oxygen consumption rate experiments

Isolation of mitochondria by immunoaffinity purification

Mitochondria were isolated from freshly collected hippocampi from AAV1-mitoTag and AAV5-Cre coinfected mice. Tissue homogenates were prepared using the Mitochondria Extraction Kit-Tissue (Miltenyi Biotec, #130-097-340) according to the manufacturer’s instructions. Briefly, following decapitation, brains were extracted and hippocampi dissected, minced with dissection scissors, and digested with the manufacturer’s extraction buffer for 30 min on ice. After centrifugation at 300×g for 5 min at 4 °C, pellets were resuspended in Protease Inhibition Buffer supplemented with EDTA-free protease inhibitor cocktail (Roche, #05892791001) and dissociated with a GentleMACS tissue dissociator (Miltenyi Biotec, #130-093-235). Homogenates were centrifuged at 500×g for 5 min at 4 °C, and the supernatants containing mitochondria were purified as directed in the Mitochondria Isolation Kit, mouse tissue, and the µMACS HA Isolation Kit (Miltenyi Biotec, #130-096-946 and #130-091-122). For magnetic labeling and isolation of the mitochondria, supernatants diluted in Separation Buffer were incubated with 50 μl of anti-HA Microbeads on a rotation wheel for 2 h at 4 °C. Subsequently, LS columns (Miltenyi Biotec, #130-042-401) were placed in a magnetic QuadroMACS^TM Separator (Miltenyi Biotec, #130-090-976) and equilibrated with 2 ml Separation Buffer. Microbead-coated mitochondria were applied onto the LS column, followed by three 3-ml washing steps with Separation Buffer. Columns were removed from the separator, and mitochondria were gently flushed out in 1.5 ml of Separation Buffer with a plunger. Mitochondria were pelleted by centrifugation for 2 min at 13,000×g and resuspended in cold Isolation Buffer.

Oxygen consumption rate measurements

Oxygen consumption rate (OCR) experiments were carried out in an Oroboros oxygraph-2k (Oroboros Instruments) in the MiR05 media (0.5 mM EGTA, 3 mM MgCl₂, 60 mM Lactobionic acid, 10 mM KH₂PO₄, 20 mM HEPES, 110 mM Sucrose, 1 g/L BSA, fatty acid free, pH 7.1). Freshly isolated mitochondria (30 µl) were injected into each chamber, and a mix of 2 mM malate, 5 mM pyruvate, and 10 mM glutamate was used as starting substrates for complex I, followed by 1.25 mM ADP + Mg²⁺. Subsequently, 10 mM succinate was added as complex II substrate. Then, different inhibitors of the mitochondrial complexes were added: 2.5 µM oligomycin (complex IV), 0.5 µM rotenone (complex I), and 2.5 µM antimycin A (complex III). The uncoupler carbonyl cyanide 3-chlorophenyl (CCCP) was used in 1 µM steps until no respiration increase was observed. In some experiments 10 µM cytochrome c was used as a mitochondrial viability control. The analysis was performed using the Oroboros DatLab software version 7.3.0.3 (Oroboros Instruments), measuring the respiration from one-minute stable regions.

Metabolite measurements

Glucose uptake assay

Glucose uptake from astrocytic cultures was measured using the Glucose Uptake Assay Kit (Abcam, #ab136955). In brief, astrocytes were starved in a serum-free medium for 6 h and then kept in a glucose free Krebs–Ringer–Phosphate–HEPES buffer (20 mM HEPES, 5 mM KH₂PO₄, 1 mM MgSO₄, 1 mM CaCl₂, 136 mM NaCl, 4.7 mM KCl, pH 7.4) and 2% BSA. After 40 min, astrocytes were treated with 10 mM 2-deoxyglucose (2-DG) for 10 min. Then, whole-cell extracts were prepared, and 2-DG quantification was carried out as described by the manufacturer’s instructions.

Extracellular lactate

Culture medium was deproteinized with 6% perchloric acid (HClO₄) and kept on ice for 1 h. Samples were centrifuged for 5 min at 4 °C and the pH adjusted to 6–7 with 20% KOH. Samples were centrifuged again to remove the precipitate, and the supernatant was mixed with lactate assay medium (1 M glycine, 0.4 M hydrazine-SO₄, 1.3 mM EDTA, pH 9.5) and 7.4 mM NAD⁺ freshly prepared. The absorbance was measured in a spectrophotometer at OD 340 nm (to detect basal NADH) (E₀), and then 10 µg/ml of lactate dehydrogenase (Roche, #10127876001) was added to convert all the lactate in the media to pyruvate with the production of NADH + H⁺. After 40 min of incubation at room temperature, the absorbance was read again at 340 nm (E₁). To calculate the lactate concentration in the media, we used the following equation: c = (E₁–E₀)/(ε*l), being c the lactate concentration (M), ε the extinction coefficient for NADH (6.22*10⁻³ M⁻¹ cm⁻¹) and l the length of the cuvette (cm).

D-serine in hippocampal extracts

D-Serine was measured with the DL-Serine Assay Kit (Abcam, #ab241027) following the manufacturer’s instructions. Briefly, animals were anesthetized by isoflurane inhalation and quickly decapitated once they were unresponsive to tail and foot pinches. The brains were rapidly extracted and the hippocampi dissected in ice-cold aCSF previously saturated with carbogen (5–10% CO₂, 90–95% O₂). The hippocampi were homogenized, deproteinized, and processed according to the manufacturer’s instructions. D-Serine concentration was calculated from a fluorescence signal (Excitation/Emission = 535/587 nm) against a standard curve of known D-serine concentrations.

Statistical analysis

Results were calculated as means ± standard error of the mean (SEM). Statistical differences between groups were determined via different parametric or non-parametric tests, and the cut-off value for statistical significance was set at p-value < 0.05. For pairwise comparisons, Mann–Whitney test was used for unpaired data and the Wilcoxon test for paired data. For multiple comparisons, we used two-way ANOVA (if indicated repeated-measures), followed by Bonferroni’s post-hoc analysis; or Kruskal–Wallis, followed by Dunn’s multiple comparisons test. For outlier identification, ROUT analysis was used. The values for n, p, and the specific statistical test performed for each experiment are described in the corresponding figures and legends. All statistical analyses and graphic representations were performed using GraphPad Prism software v8.3.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization and experimental design: A.F.-R., M.I.C., and J.A.E. Most experimental work and data analysis: A.F.-R. Mitotag stereotaxic injections and oxygen consumption rate experiments: A.F.-R., A.E.-P, and G.M. Assistance with some of the stereotaxic injections and behavioral assays: C.S.-C and C.G.-V. Assistance with some of the biochemical analyses: E.L.-M., C.G.-V., and E.M.-C. Assistance with astrocytic cultures: C.G.-V. Glucose uptake assay: C.G.-V. and S.G.-E. Neuronal morphology analysis: I.B.M. Electrophysiology on AAV5-GfaABC1D-Cre-4x6T- and AAV5-GFAP104-mCherry-injected animals: M.D.G.M. Extracellular lactate measurements: S.G.-E and C.B.-E. iGlucoSnFR imaging experiments: C.G.-V. and I.B.M., with provision of the iGlucoSnFR viral vector by M.S. Mouse genotyping: S.G.-E. Provision of p110α^flox/flox: M.G. Provision of mitotag: A.Q. Writing original draft and editing: A.F.-R., M.I.C. and J.A.E.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the Spanish Ministry of Science, Innovation and Universities grants PID2020-117651RB, PDC2021-120815-I00, and PID2023-149056OB-I00 to J.A.E., PID2020-114977RB-I00 and PID2023-151947OB-I00 to A.Q., PID2021-128737NB-I00 to M.S., and RYC2022-037937-I to M.I.C. This work was also supported by Inserm, the European Research Council (Micabra, ERC-2017-AdG-786467), Fondation pour la Recherche Medicale (DRM20101220445), the Human Frontiers Science Program, Region Aquitaine, the French State/Agence Nationale de la Recherche (ERA-Net Neuron CanShank ANR-21-NEU2-0001-04, CaMeLS ANR-23-CE16-0022-01, Hippobese ANR-23-CE14-0004-03) and the French government in the framework of the University of Bordeaux’s IdEx “Investments for the Future - France 2030” program/GPR BRAIN_2030 to G.M. The Spanish Ministry of Science and Innovation supported A.F.-R., E.L.-M., C.G.-V., C.S.-C., and M.D.G.M. with predoctoral contracts, and I.B.M with a “Juan de la Cierva” postdoctoral contract. A.E.-P. was supported by Région Nouvelle Aquitaine: Convention AAPR2022-2021-16762910. A.F.-R. was the recipient of an EMBO Scientific Exchange Grant (#9808). The Centro de Biología Molecular Severo Ochoa is the recipient of a ‘Severo Ochoa Excellence’ grant (CEX2021-001154-S) funded by MICIU/AEI (10.13039/501100011033) and receives institutional support from the Ramón Areces Foundation.

Data availability

Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77319-0.