Glia fuel neurons with locally synthesized ketone bodies to sustain memory under starvation

Abstract

During starvation, mammalian brains can adapt their metabolism, switching from glucose to alternative peripheral fuel sources. In the Drosophila starved brain, memory formation is subject to adaptative plasticity, but whether this adaptive plasticity relies on metabolic adaptation remains unclear. Here we show that during starvation, neurons of the fly olfactory memory centre import and use ketone bodies (KBs) as an energy substrate to sustain aversive memory formation. We identify local providers within the brain, the cortex glia, that use their own lipid store to synthesize KBs before exporting them to neurons via monocarboxylate transporters. Finally, we show that the master energy sensor AMP-activated protein kinase regulates both lipid mobilization and KB export in cortex glia. Our data provide a general schema of the metabolic interactions within the brain to support memory when glucose is scarce.

Fig. 1. During starvation, mushroom body neurons rely on ketone body metabolism to sustain ketone body-dependent associative memory formation.

a, Inhibition of ACAT1 expression in adult MB neurons impaired memory after massed training in starved flies (F_2,42 = 9.49, P = 0.0004), but not in fed flies (F_2,42 = 0.74, P = 0.485). b, After massed training, memory was impaired in starved flies expressing a Sln RNAi in adult MB neurons (F_2,56 = 11.33, P < 0.0001), but not in fed flies (F_2,55 = 2.09, P = 0.133). c, Images of the Laconic FRET sensor expressed in MB neurons through the Tub-Gal80^ts;VT30559 driver, obtained by two-photon microscopy in the mTFP and Venus channels with the same conditions that were used for live recordings (scale bar, 20 µm). v, vertical lobe, corresponding to one of the axon fascicles of MB neurons; s, soma of MB neurons. d, In the fed condition, application of 10 mM of acetoacetate (red dashed line) resulted in a decreased Laconic ratio followed by a plateau in the MB soma of control flies, revealing lactate efflux from MB neuronal somas after acetoacetate bath application (mean trace ± s.e.m.). Quantification of the mean Laconic ratio at the plateau was performed on the last 100 s of recording (red line). Inhibition of Sln expression in adult MB neurons impaired lactate efflux evoked by acetoacetate application (t₁₉ = 3.355, P = 0.003). n represents either a group of 40–50 flies analysed together in a behavioural assay (a and b) or the response of a single recorded fly (d). Data are expressed as the mean ± s.e.m. with dots as individual values, and analysed by one-way analysis of variance (ANOVA) with post hoc testing by Newman–Keuls pairwise comparisons test (a and b) or by unpaired two-sided t-test (d). Asterisks refer to the least-significant P value of post hoc comparison between the genotype of interest and the genotypic controls (a and b), or to the P value of the unpaired t-test comparison (d). **P < 0.01, NS, not significant.

Fig. 2. During starvation, cortex glia mobilize their own fatty acid store to provide ketone bodies to sustain ketone body-dependent associative memory.

a, KB production pathway. Triacylglycerols stored in LDs are hydrolysed by the lipase Bmm into FAs and diacylglycerol. FAs are then activated by an acyl-CoA synthetase and imported as acyl-CoA into the mitochondria by the carnitine shuttle system whose CPT1 is a component. Then acyl-CoA enters the β-oxidation cycle to produce acetyl-CoA that will be used to generate acetoacetate by the successive actions of a thiolase, the HMGS and the HMG-CoA lyase. b–d, Downregulation in adult cortex glia of each of the three key enzymes of KB production, Bmm (b), CPT1 (c) and HMGS (d) impaired K-AM (Bmm: F_2,27 = 23.29, P < 0.0001; CPT1: F_2,39 = 7.304, P = 0.002; HMGS: F_2,32 = 12.66, P < 0.0001) but not ARM in fed flies (Bmm: F_2,29 = 0.58, P = 0.567; CPT1: F_2,31 = 0.825, P = 0.448, HMGS: F_2,33 = 0.21, P = 0.815). e–g, BODIPY LD staining and quantification in starved and fed flies expressing or not an RNAi targeting one of the three key enzymes of KB production in adult cortex glia. e, Inhibition of Bmm expression in adult cortex glia of fed flies did not change the mean area of LDs observed in the brain region where cortex glia enwrap MB neuronal soma (t₉ = 0.193, P = 0.851), whereas larger LDs were observed in starved Bmm^RNAi expressing flies compared to controls (t₁₁ = 5.085, P = 0.0004). f,g, Similarly, inhibition of either CPT1 or HMGS expression in adult cortex glia had no effect on LD mean area in the fed condition (CPT1: t₁₀ = 0.950, P = 0.364; HMGS: t₉ = 0.121, P = 0.907), whereas during starvation an increase in LD mean area was observed as compared to control flies (CPT1: t₁₆ = 3.792, P = 0.0016 ; HMGS: t₁₁ = 2.690, P = 0.021). n represents either a group of 40–50 flies analysed together in a behavioural assay (b–d) or one BODIPY-stained brain (e–g). Data are expressed as the mean ± s.e.m. with dots as individual values, and analysed by one-way ANOVA with post hoc testing by Newman–Keuls pairwise comparisons test (b–d) or by unpaired two-sided t-test (e–g). Asterisks refer to the least-significant P value of a post hoc comparison between the genotype of interest and the genotypic controls or to the P value of the unpaired t-test comparison. ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05. Scale bar, 20 µm.

Fig. 3. Chaski is the cortex glia monocarboxylate transporter required to provide ketone bodies to sustain ketone body-dependent associative memory.

a, Inhibition of Chk expression in adult cortex glia impaired starved K-AM (F_2,31 = 8.22, P = 0.001), while memory after massed training in fed flies was normal (F_2,33 = 0.43, P = 0.653). b, BODIPY LD staining and quantification in starved and fed flies expressing or not the Chk RNAi in adult cortex glia. Inhibition of Chk expression in adult cortex glia in fed flies did not change LD mean area (t₁₀ = 1.785, P = 0.105), whereas during starvation an increase in LD mean area was observed as compared to controls (t₁₂ = 3.181, P = 0.008). c, Images of the Laconic sensor expressed in cortex glia with the Tub-Gal80^ts;R54H02 driver, obtained by two-photon microscopy in the mTFP and Venus channels, under the same conditions as those used for live recordings. d, Application of 10 mM acetoacetate (red dashed line) resulted in a decreased Laconic ratio followed by a plateau in cortex glia of starved control flies, showing lactate efflux from cortex glia after acetoacetate bath application (mean trace ± s.e.m. of ten recordings per genotype). Quantification of the mean Laconic ratio at the plateau was performed during the last 100 s of recording (red line). Inhibition of Chk expression in adult cortex glia impaired this lactate efflux evoked by acetoacetate application (t₁₈ = 2.791, P = 0.012). n indicated within the graph represents a group of 40–50 flies analysed together in a behavioural assay (a), one BODIPY-stained brain (b) or the response of a single recorded fly (d). Data are expressed as the mean ± s.e.m. with dots as individual values, and analysed by one-way ANOVA with post hoc testing by Newman–Keuls pairwise comparisons test (a) or by unpaired two-sided t-test (b and d). Asterisks refer to the least-significant P value of a post hoc comparison between the genotype of interest and the genotypic controls (a) or to the P value of the unpaired t-test comparison (b and d). **P < 0.001, *P < 0.05. Scale bar, 20 µm (b and c).

Fig. 4. During starvation, AMPK is required in cortex glia for ketone body production and export to sustain ketone body-dependent associative memory in neurons.

a, Inhibition of AMPKα expression in adult cortex glia impaired K-AM (F_2,25 = 8.05, P = 0.002), while ARM was normal in fed flies (F_2,33 = 1.76, P = 0.189). b, BODIPY LD staining and quantification in starved and fed flies expressing or not an AMPKα RNAi in cortex glia. In fed flies, inhibition of AMPKα expression in adult cortex glia did not change the LD mean area (t₁₀ = 1.308, P = 0.220), whereas an increase in LD mean area was observed in starved flies as compared to controls (t₁₀ = 2.660, P = 0.0239). c, Starvation strongly increased Bmm and CPT1 mRNA levels (Bmm: t₆ = 4.25, P = 0.0054; CPT1: t₆ = 7.28, P = 0.0003). d, In starved flies expressing AMPK RNAi in glial cells, Bmm and CPT1 mRNA levels did not differ from those of fed flies (Bmm: t₅ = 1.34, P = 0.238; CPT1: t₅ = 0.76, P = 0.482), whereas in the genotypic control groups, starvation induced a significant increase in each gene’s mRNA level (Bmm: t₅ = 6.54, P = 0.001; CPT1: t₅ = 8.55, P = 0.0004). e, Application of 10 mM acetoacetate (red dashed line) resulted in a decreased Laconic ratio followed by a plateau in cortex glia of starved control flies, showing lactate efflux from cortex glia after acetoacetate bath application (mean trace ± s.e.m.). Quantification of the mean Laconic ratio at the plateau was performed during the last 100 s of recording (red line). Inhibition of AMPKα expression in adult cortex glia impaired this lactate efflux evoked by acetoacetate application (t₁₄ = 3.393, P = 0.004). n represents a group of 40–50 flies analysed together in a behavioural assay (a), one BODIPY-stained brain (b), mRNA extracted from a group of 50 flies (c and d) or the response of a single recorded fly (e). Data are expressed as the mean ± s.e.m. with dots as individual values, and analysed by one-way ANOVA with post hoc testing by Newman–Keuls pairwise comparisons test (a) or by unpaired two-sided t-test (b–e). Asterisks refer to the least-significant P value of a post hoc comparison between the genotype of interest and the genotypic controls (a) or the P value of the unpaired t-test comparison (b–e). ***P < 0.001, **P < 0.01, *P < 0.05. Scale bar, 20 µm (b).

Fig. 5. Model of metabolic coupling between glia and neurons during starvation.

The AMP/ATP ratio decreases during starvation (1), resulting in the activation of AMPK, the major cellular sensor of energy state. AMPK is required to sustain K-AM formation (2). AMPK is required in cortex glia to increase Bmm and CPT1 expression during starvation and to regulate KB transport by Chk (red arrows). During starvation, FAs are mobilized from the internal stores of cortex glia via the action of the lipase Bmm (3). FAs are then imported into the mitochondria via CPT1 and are subsequently oxidized to generate acetyl-CoA, which is used by HMGS for ketogenesis (4). Eventually, KBs are exported from cortex glia via Chk and taken up by neurons via Sln (5). In neurons, during starvation, KBs are used by ACAT1 to generate acetyl-CoA in the mitochondria for energy (6). TCA, tricarboxylic acid cycle; RC, respiratory chain; K, KB; β-Ox, β-oxidation. The metabolic pathways are symbolized by curved arrows, with the pathway position of the enzyme identified in this study appearing in bold.

Extended Data Fig. 1. Control experiments for KB oxidation and uptake by MB neurons to form memory during starvation.

(a) In fed flies, ACAT1 RNAi GD7132 expression did not affect memory formed after spaced training (F_2,33 = 1.215, P = 0.310). When ACAT1 RNAi GD7132 expression was not induced, memory after massed training in starved flies was normal (F_2,41 = 0.82, P = 0.446). (b) Inhibition of ACAT1 expression in adult MB neurons using a second non-overlapping RNAi impaired memory formed after massed training in starved flies (F_2,42 = 3.54, P = 0.038), whereas ARM or LTM in fed flies was not affected (massed training: F_2,38 = 3.38, P = 0.045, Newman-Keuls post-test is not significant; spaced training: F_2,24 = 0.93, P = 0.407). Non-induced starved flies showed no memory defect after massed training (F_2,41 = 7.376, P = 0.0018, Newman-Keuls post-test is not significant for comparison between Tub-Gal80^ts;VT30559/UAS-ACAT1 RNAi HMC03340 and the genotypic control UAS-ACAT1 RNAi HMC03340/+). (c) The memory formed after massed training in starved wild-type Canton S flies was similar between female and male flies (t₃₀ = 0.45, P = 0.653). (d) Immunohistochemistry of CRIMIC Sln-T2A-Gal4>UAS-mCD8::RFP brains showing Sln expression pattern in red (RFP) and either pan-neuronal anti-nc82 counterstaining in green or cortex glia anti-WRAPPER counterstaining (2 top panels: nc82 and lower panel: WRAPPER). The top row displays a global view of the posterior brain (40x objective acquisition) and the lower panel shows at higher magnification (100x objective acquisition) the MB calyx region surrounded by the cortex region with MB neuronal somas. Sln is expressed in a large proportion of neurons of the central brain, and a clear expression in MB neurons can be detected in the calyx (top and middle panel: nc82 co-staining). The Sln expression pattern only partially overlaps with cortex glia labeling (lower panel: WRAPPER co-staining). pb: protocerebral bridge of Central complex, s: soma of MB neurons, Ca: Calyx of MB. (e) Immunohistochemistry of Chk-Gal4^MI15450>UAS-mCD8::GFP brains showing the Chk expression pattern in red (RFP) and either pan-neuronal anti-nc82 counterstaining in green or cortex glia anti-WRAPPER counterstaining (2 top panels: nc82 and lower panel: WRAPPER). The top row displays a global view of the posterior brain (40x objective acquisition) and the lower panels show at higher magnification (100x objective acquisition) the MB calyx region surrounded by the cortex region with MB neuronal somas and cortex glia processes. Chk showed a diffuse expression in the neuropil of the central brain and a clear pattern of expression in cortex glia cells, as revealed by the honeycomb-like pattern. pb: protocerebral bridge of Central complex, s: soma of MB neurons, Ca: Calyx of MB. (f) Inhibition of Chk expression in adult MB neurons did not impair memory after massed training in starved flies (F_2,21 = 0.12, P = 0.886). (g) In fed flies, LTM was normal when Sln expression was downregulated in adult MB neurons (F_2,44 = 0.055, P = 0.946). When Sln RNAi GD1940 expression was not induced, memory after massed training in starved flies was normal (F_2,33 = 1.24, P = 0.302). (h) Inhibition of Sln expression in adult MB neurons using a second non-overlapping RNAi impaired memory formed after massed training in starved flies (F_2,63 = 5.72, P = 0.0052). By contrast, downregulation of Sln expression did not affect memory formation in fed flies after massed training (F_2,63 = 2.51, P = 0.089), or after spaced training (F_2,33 = 2.124, P = 0.13). When Sln RNAi KK104306 expression was not induced, memory formed after massed training in starved flies was normal. (F_2,39 = 1.36, P = 0.269). n represents a group of 40–50 flies analyzed together in a behavioral assay. Data are expressed as mean ± s.e.m. with dots as individual values, and analyzed by one-way ANOVA with post hoc testing by Newman-Keuls pairwise comparisons test. Asterisks refer to the least significant P-value of post hoc comparison between the genotype of interest and the genotypic controls. **P<0.01, *P<0.05, ns: not significant. Scale bar: 40 µm (d and e).

Extended Data Fig. 2. Control experiments for KB uptake by MB neurons.

(a) MCT switches to the other side of the membrane at a higher rate when a substrate is bound (MCT – L; k2) than when no substrate is bound (MCT; k1), with k1<k2. Trans-acceleration occurs because the conformational switch of MCT across the cell membrane is facilitated when an adequate substrate such as KB is bound. Lactate efflux can be monitored indirectly by an intracellular Laconic FRET sensor that is sensitive to lactate concentration. When KB is applied, the high concentration of KB in the extracellular media increases the rate of lactate transport in the opposite direction, corresponding here to lactate efflux. (b) In fed flies expressing the intracellular Laconic FRET sensor in MB neurons, application of 10 mM of L-lactate on the brain induced a strong increase in the Laconic FRET ratio showing that the intracellular lactate concentration subsequently increased in response to lactate uptake by MB neurons (t₅ = 2.627, P = 0.047). Thus, as previously shown by several other research groups, the Laconic FRET sensor can be used to monitor the intracellular lactate level. (c) To assess that Sln downregulation in MB neurons do not affect the lactate intracellular concentration to a level below the Laconic FRET sensor sensitivity threshold, thus preventing detection of any Laconic FRET ratio change in Fig. 2d, we applied a saturation treatment to reach a maximal plateau of Laconic FRET ratio (that is saturation of the sensor) that was used to normalize the different genotypic conditions. In fed flies expressing the Laconic FRET sensor in MB neurons, application of 5 mM NaAz, a strong inhibitor of mitochondrial respiration, induced a strong increase in the Laconic FRET ratio, reaching the saturation level of the sensor (mean trace ± s.e.m.). The last 100 s of recording (red bar), when the Laconic FRET sensor had reached saturation, were used to normalize; the initial Laconic ratio was measured during the 120-s baseline recording before NaAz application. In fed flies, the lactate basal concentration in MB neurons of flies expressing Sln RNAi was similar to the genotypic control (t₁₄ = 0.2091, p = 0.837). n represents the response of a single recorded fly. Data are expressed as mean ± s.e.m. with dots as individual values, and analyzed by one-sample two-sided t-test with theoretical mean = 0 (b) or unpaired two sided two-sample t-test (c). Asterisks refer to the P-value of the t-test comparison. *P<0.05, ns: not significant.

Extended Data Fig. 3. Control experiments for KB production by cortex glia to sustain K-AM formation in MB neurons.

(a) In fed flies when Bmm expression was downregulated in cortex glia LTM was normal (F_2,27 = 0.29, P = 0.749). When RNAi expression was not induced, K-AM was normal (F_2,27 = 0.27, P = 0.768). (b) A second non-overlapping RNAi targeting Bmm (Bmm RNAi GD5139) was used to confirm the specific K-AM defect. In order to increase RNAi efficiency, Dicer2 expression was induced together with the RNAi expression in adult cortex glia using the Tub-Gal80^ts; UAS-Dcr2, R54H02 line. Inhibition of Bmm expression in adult cortex glia using Bmm RNAi GD5139 impaired K-AM in starved flies (F_2,42 = 10.70, P = 0.0002), whereas ARM or LTM in fed flies was not affected (ARM: F_2,33 = 1.55, P = 0.228; LTM: F_2,45 = 0.02, P = 0.982). Non-induced starved flies showed no K-AM defect (F_2,27 = 0.003, P = 0.997). (c) Downregulation of CPT1 expression in adult cortex glia did not affect LTM (F_2,37 = 1.32, P = 0.279). When RNAi expression was not induced, K-AM was normal (F_2,36 = 2.12, P = 0.135). (d) A second non-overlapping RNAi targeting CPT1 (CPT1 RNAi KK100935) was used to confirm the specific K-AM defect. As for Bmm RNAi GD5139, we used the Tub-Gal80^ts; UAS-Dcr2, R54H02 line to increase RNAi efficiency. Inhibition of CPT1 expression in adult cortex glia using CPT1 RNAi KK100935 impaired K-AM (F_2,45 = 8.72, P = 0.0006), whereas ARM or LTM in fed flies was not affected (ARM: F_2,45 = 0.87, P = 0.424; LTM: F_2,43 = 0.06, P = 0.939). Non-induced starved flies showed no K-AM defect (F_2,45 = 0.66, P = 0.524). (e) When HMGS expression is downregulated in the cortex, LTM was normal (F_2,43 = 1.74, P = 0.187). When RNAi expression was not induced, K-AM was normal (F_2,38 = 0.43, P = 0.656). (f) A second non-overlapping RNAi targeting HMGS (HMGS RNAi HMC04928) was used to confirm the specific K-AM defect. Inhibition of HMGS expression in adult cortex glia using this RNAi impaired K-AM (F_2,31 = 4.53, P = 0.019), whereas ARM or LTM in fed flies was not affected (ARM: F_2,32 = 1.56, P = 0.227; LTM: F_2,42 = 0.72, P = 0.494). Non-induced starved flies showed no K-AM defect (F_2,45 = 1.58, P = 0.217). n represents a group of 40–50 flies analyzed together in a behavioral assay. Data are expressed as mean ± s.e.m. with dots as individual values, and analyzed by one-way ANOVA with post hoc testing by Newman-Keuls pairwise comparisons test. Asterisks refer to the least significant P-value of post hoc comparison between the genotype of interest and the genotypic controls. **P<0.01, *P<0.05, ns: not significant.

Extended Data Fig. 4. KB production is not required in other glial cells type or MB neurons to sustain K-AM.

(a) When HMGS expression was downregulated in adult MB neurons, K-AM was normal (F_2,27 = 1.068, P = 0.358). (b) When either CPT1 or HMGS expression was downregulated in adult astrocyte-like glia, K-AM was normal (CPT1: F_2,36 = 1.166, P = 0.323; HMGS: F_2,38 = 0.778, P = 0.466). (c) When either CPT1 or HMGS expression was downregulated in adult ensheathing glia, K-AM was normal (CPT1: nF_2,62 = 1.189, P = 0.311; HMGS: F_2,20 = 0.139, P = 0.871). (d) In starved flies, memory formed after spaced training is impaired when HMGS is downregulated in cortex glia (F_2,40 = 9.46, P = 0.0004). When RNAi expression was not induced, memory formed after spaced training in starved flies was normal (F_2,27 = 0.39, P = 0.678). (e) Full view of a fed wild-type Canton S fly central brain stained with BODIPY, as in Fig. 2e-g. The left panel shows a confocal plane in which the ROI (white square) for subsequent quantification of LDs has been selected. The right panel shows a confocal plane from the same brain but 4 µm deeper, showing the ROI position toward the MB calyx that is used as a landmark to define the ROI x-y axis position. n represents a group of 40–50 flies analyzed together in a behavioral assay. Data are expressed as mean ± s.e.m. with dots as individual values, and analyzed by one-way ANOVA with post hoc testing by Newman-Keuls pairwise comparisons test. Asterisks refer to the least significant P-value of post hoc comparison between the genotype of interest and the genotypic controls. **P<0.01, ns: not significant. Scale bar: 40 µm (e).

Extended Data Fig. 5. Control experiments for KB export by cortex glia to sustain K-AM formation in MB neurons.

(a) In fed flies, when Chk expression was downregulated in adult cortex glia, LTM was normal (F_2,37 = 0.96, P = 0.393). When Chk RNAi expression was not induced, K-AM was normal (F_2,32 = 0.40, P = 0.677). (b) Since starvation induced strong lethality in Chk^MB04207 homozygous flies, we used heterozygous Chk^MB04207/+ flies to confirm Chk function in K-AM. Furthermore, since both Chk^MB04207 flies (which are in a w¹¹¹⁸ background) and w¹¹¹⁸ flies have lower memory scores, we only counted heterozygous females (w¹¹¹⁸/+; Chk^MB04207/+) and not hemizygous males (w¹¹¹⁸/Y; Chk^MB04207/+), and counted only heterozygous w¹¹¹⁸/+ females as control flies. Heterozygous w¹¹¹⁸/+; Chk^MB04207/+ female flies had a strong K-AM defect as compared to control w¹¹¹⁸/+;+ female flies (t₂₇ = 3.24, P = 0.0032). Neither ARM (t₂₃ = 0.89, P = 0.379) nor LTM (t₂₆ = 0.10, P = 0.918) was affected in these heterozygous w¹¹¹⁸/+;Chk^MB04207/+ female flies. (c) Inhibition of Sln expression in adult cortex glia did not impair K-AM (F_2,36 = 0.23, P = 0.794). (d) In starved flies expressing the Laconic FRET sensor in cortex glia, application of 5 mM NaAz induced a strong increase in the Laconic FRET ratio, reaching the saturation level of the sensor (mean trace ± s.e.m.). The last 100 s of recording (red bar), when the Laconic FRET sensor had reached saturation, were used to normalize; the initial Laconic ratio was measured during the 2-min baseline recording before NaAz application. In starved flies, the lactate basal concentration in cortex glia expressing Chk RNAi was similar to the genotypic control (t₁₄ = 0.743, P = 0.469). n represents either a group of 40–50 flies analyzed together in a behavioral assay (a–c) or the response of a single recorded fly (d). Data are expressed as mean ± s.e.m. with dots as individual values, and analyzed by one-way ANOVA with post hoc testing by Newman-Keuls pairwise comparisons test (a and c) or by unpaired two-sided t-test (b and d). Asterisks refer to the least significant P-value of post hoc comparison between the genotype of interest and the genotypic controls (a and c) or to the P-value of the unpaired t-test comparison (b and d). **P<0.01, ns: not significant.

Extended Data Fig. 6. Control experiments for AMPK requirement in cortex glia to sustain K-AM formation in MB neurons.

(a) In fed flies, AMPKα expression in cortex glia is not required for LTM (F_2,52 = 0.33, P = 0.717). When AMPKα RNAi expression was not induced, K-AM was normal (F_2,21 = 1.02, P = 0.378). (b) Inhibition of AMPKα expression in adult cortex glia using a second non-overlapping RNAi impaired K-AM in starved flies (F_2,33 = 9.95, P = 0.0004), whereas memory formed in fed flies after either massed or spaced training was not affected (ARM: F_2,27 = 0.23, P = 0.799; LTM: F_2,27 = 0.187, P = 0.831). Non-induced flies showed no K-AM defect (F_2,27 = 0.16, P = 0.855). (c) In wild-type Canton S fly heads, starvation did not change the mRNA level of either AMPK, HMGS, Chk, ACAT1 or Sln (AMPK: t₄ = 0.382, p = 0.722, HMGS: t₄ = 1.40, p = 0.24; Chk t₆ = 0.033, p = 0.98; ACAT1 t₁₀ = 0.439, p = 0.669; Sln: t₆ = 0.389, p = 0.711). (d) In starved flies expressing the Laconic FRET sensor in cortex glia, application of 5 mM NaAz induced a strong increase in the Laconic FRET ratio, reaching the saturation level of the sensor (mean trace ± s.e.m.). The last 100 s of recording (red bar), when the Laconic FRET sensor had reached saturation, were used to normalize; the initial Laconic ratio was measured during the 2-min baseline recording before NaAz application. In starved flies, the lactate basal concentration in cortex glia expressing AMPKα RNAi was similar to the genotypic control (AMPK: t₁₅ = 0.279, P = 0.784). (e) Application of 10 mM acetoacetate (red dashed line) results in a decreased Laconic ratio followed by a plateau in cortex glia of starved control flies, showing lactate efflux from cortex glia after acetoacetate bath application (mean trace ± s.e.m.). Inhibition of HMGS expression did not change the lactate efflux evoked by acetoacetate application (t₁₅ = 0.004, P = 0.99). Quantification of the mean Laconic ratio at the plateau was performed during the last 100 s of recording (red line). (f) In starved flies, the lactate basal concentration in cortex glia expressing HMGS RNAi was similar to the genotypic control (HMGS: t₁₂ = 0.112, P = 0.913). n represents either a group of 40–50 flies analyzed together in a behavioral assay (a and b), mRNA extracted from a group of 50 flies (c) or the response of a single recorded fly (d-f). Data are expressed as mean ± s.e.m. with dots as individual values, and analyzed by one-way ANOVA with post hoc testing by Newman-Keuls pairwise comparisons test (a and b) or by unpaired two-sided t-test (c-f). Asterisks refer to the least significant P-value of post hoc comparison between the genotype of interest and the genotypic controls (a and b) or to the P-value of the unpaired t-test comparison (c-f). **P<0.01, ns: not significant.

Extended Data Fig. 7. Validation of RNAi efficiency of the RNAi lines used in this study.

(a–d) Using the pan-neuronal driver elav to drive constitutive expression of the specified RNAi, we observed a significant reduction of the targeted mRNA level in fly heads: ACAT1 RNAi GD7132 (a) t₄ = 3,182, p = 0.033; ACAT1 RNAi HMC03340 (b) t₄ = 2,790, p = 0.049); Sln RNAi GD1940 (c) t₈ = 2,600, p = 0.032 and Sln RNAi KK104306 (d) t₄ = 3,227, p = 0.032. (e-j) Using the pan-glial driver repo to drive constitutive expression of the specified RNAi, we observed a significant reduction of the targeted mRNA level in fly heads: Bmm RNAi JF0146 (e) t₄ = 4,329, p = 0.012; Bmm RNAi GD5139 (f) t₄ = 5,309, p = 0.006; CPT1 RNAi HMS00040 (g) t₄ = 2,832, p = 0.047; CPT1 RNAi KK100935 (h) t₄ = 2,784, p = 0.049; HMGS RNAi KK107372 (i) t₄ = 4,533, p = 0.011 or HMGS RNAi HMC04928 (j) t₄ = 5,295, p = 0.006. (k-m) Because constitutive expression of Chk RNAi or AMPK RNAi in glial cells was lethal, the RNAi efficiency was assessed by measuring the targeted mRNA level in fly heads expressing the RNAi in adult glial cells using the pan-glial inducible driver Tub-Gal80^ts; Repo-Gal4: Chk RNAi GD1829 (k) t₁₀ = 2.43, P = 0.035; AMPK RNAi JF01951 (l) t₄ = 2.84, P = 0.047 and AMPK RNAi HMC04979 (m) t₄ = 5.72, P = 0.0046. n represents mRNA extracted from heads of a group of 50 flies. Results are shown as ratios to the reference gene tubulin and data are expressed as mean ± s.e.m. with dots as individual values, and analyzed by unpaired two-sided t-test. Asterisks refer to the P-value of the unpaired t-test comparison. **P<0.01, *P<0.05.