Brain-derived ketone bodies can replace glucose to power neural function

Abstract

The brain is sensitive to disruptions in glucose metabolism, requiring constant delivery to support neural activity. Here, we discovered a vertebrate with the surprising capacity to abandon glucose metabolism and replace it with ketone bodies produced entirely within the brain. In frogs—animals with seemingly typical glucose demands—hibernation shifts brain bioenergetics to allow ketone bodies made within the brain to sustain neural activity without ATP from glucose metabolism. This involves, in part, the upregulation of fatty acid catabolism, ketone body synthesis, and transport from astrocytes to neurons to maintain synaptic transmission. Brain-derived ketone bodies also prevent decrements in activity that otherwise occur during hypoxia. These results provide insight into how frogs restart brain circuits following months of underwater hibernation when facing severe hypoxia and hypoglycemia that otherwise impair neural performance. Overall, these results reveal a capacity for the vertebrate brain to temporarily abandon glucose while maintaining costly functions using locally sourced ketone bodies independent from body energy stores.

Introduction

Brain activity is energetically expensive, and failure to produce ATP through aerobic respiration swiftly impairs neural function. As oxidative metabolism of glucose accounts for most of the ATP production in the brain¹, it is not surprising that defects in glucose metabolism contribute to multiple neurological issues seen in stroke, Alzheimer’s disease, Parkinson’s disease, ALS, aging, and many more². While alternative fuels can support the brain, such as lactate and pyruvate, which are made from glucose, and ketone bodies from fatty acids, neural activity cannot exist without a large fraction of its energy originating from glucose or its derivatives^3–8. Therefore, without consistent glucose delivery, there is not enough energy to power costly physiological processes that sustain healthy neuronal activity⁹.

Although most animals obey this rule, several species, such as painted turtles, naked mole-rats, ground squirrels, and some birds evolved to survive glucose and/or oxygen limitations encountered as a part of their life histories^10–13. Most amphibians are not generally counted among those strongly tolerant of brain energy stress, with neural activity having oxygen and glucose needs of an average vertebrate^14,15. Yet, there is more than meets the eye. During aquatic hibernation, frogs have low metabolic rates and respire only through their skin while submerged in cold water. The combined effects of low metabolic rate, lack of food, and skin gas exchange cause oxygen and wholebody glycogen stores to fall¹⁶. This is not an immediate threat due to low energy requirements during hibernation, but to emerge in the spring, vital neural circuits must accelerate activity in the face of severe energy limitations, where neural activity should be difficult if not impossible¹⁵. To overcome these constraints, hibernation transforms brainstem circuits to produce activity with little energy, running for at least 3 hours during tissue anoxia and 2 hours during oxygen and glucose deprivation¹⁷. This large improvement occurs by shifting to brain glycogen as a primary fuel source¹⁷, bolstering synaptic glycolysis¹⁸, and maintaining synaptic transmission while reducing its energetic cost¹⁹. Therefore, a vertebrate with seemingly typical metabolic requirements can switch to local brain energy reserves to support behavior in times of stress¹⁴.

Past studies illuminated how energy sourced from brain glycogen can prolong neural activity without oxygen and glucose delivery. However, glycogen is a depletable energy source, which suggests that other local fuels may be critical to overcome energy limitations. Although the brain is not generally considered to maintain large onboard fuel stores, along with glycogen²⁰, lipid droplets are also present in astrocytes and can be converted to ketone bodies²¹, a well-known alternative fuel that supports brain homeostasis during impaired glucose metabolism⁴. Moreover, ketone bodies like beta-hydroxybutyrate (BHB) produce ~30% more ATP than glucose at the same oxygen consumption, serving as a more efficient fuel, with potential benefits in hypoxia²². Therefore, we hypothesized that internal energy sources unrelated to glucose may maintain neural function during energy stress after emergence from hibernation. Here we present several lines of evidence that hibernation triggers a metabolic plasticity response that allows motor circuits to function without glycolysis using ketone bodies produced within the brain.

Results

Neural activity drops its glucose requirement after hibernation

We assessed the function of brainstem circuits that generate breathing, as it is a rhythmically active motor circuit critical to survival after hibernation¹⁴. In addition, this network and some of its cellular components can be studied ex vivo, producing physiological motor patterns of breathing that resemble in vivo activity²³. Ex vivo tissue preparations also allow isolation of metabolic disturbances to the brain without the confounding influence from body energy stores and cerebral blood flow. To assess neural activity’s requirement for glucose metabolism, we replaced D-glucose with 2-deoxy-D-glucose (2DG-0G) under aerobic conditions with no additional fuels supplemented. With this treatment, aerobic metabolism by mitochondria remains intact, but there is little-to-no capacity for a contribution from glycolysis because external glucose is removed and 2DG leads to the production of 2-deoxy-glucose-6-phosphate, which competitively inhibits hexokinase. Additionally, fuels derived from glucose metabolism, such as pyruvate and lactate, are also reduced in 2DG-0G, as their production requires flux through glycolysis. As expected¹⁵, network activity in controls was strongly disrupted by 2DG-0G (Figure 1 A, top panel, Figure 1B). When performing the same experiment on hibernators under the same conditions (Figure 1A, bottom panel), activity persisted for significantly longer (Figure 1B), maintaining baseline frequency at 60 minutes. While activity slowed on average by the end of 90 minutes, all preparations produced activity that appeared outwardly normal. Exposure to 0% O₂ to stop aerobic respiration quickly silenced all hibernator circuits running without glycolysis (Figure 1A, bottom panel). Therefore, activity depended on aerobic respiration that was not supported by ATP production from any possible remaining flux through glycolysis.

Figure 1. Neural activity can abandon glucose metabolism after hibernation while maintaining tissue beta hydroxybutyrate homeostasis.

A) Example extracellular recordings of cranial nerve X output driven by rhythmic respiratory circuits upon stopping glucose metabolism with 0 glucose and 5 mM 2-DG (2DG-0G) in a control (top) and hibernator in the presence of O₂. After 90 minutes without glycolysis, O₂ was removed to confirm hibernators were running without glycolysis using aerobic metabolism. Inset illustrates the brainstem preparation.

B) Mean data of network burst rate over 90 minutes without glycolysis. ▪ indicates differences across groups at each time point without glycolysis, and * represent differences from baseline within each group. Repeated Measures Two-way ANOVA results: Time, p<0.0001, Group, p=0.0034, Time x Group, p=0.0145. **** indicates p<0.0001, ▪ p<0.05, and ▪▪ <0.01 in Holm-Sidak Multiple Comparisons tests.

C) Example recordings of rhythmically-evoked AMPA-glutamate synaptic transmission onto hypoglossal motoneurons delivered for 1s at 20 Hz every 5 s following 30 minutes without glycolysis (2DG-0 glucose) in controls and hibernators. Gray tracings are baseline evoked rhythmic synaptic transmission. Colors (green=control, blue=hibernator) show transmission after 30 minutes without glycolysis. In hibernators, it is difficult to discern the baseline because transmission was the same after 30 minutes without glycolysis, while in controls, activity was nearly abolished.

D) Mean data of synaptic transmission (area under the curve; AUC of the transmission envelope upon rhythmic stimulation) during and after recovery from 2DG-0G. ▪ indicates differences across groups at each time point without glycolysis, and * represent differences from baseline within each group (control, n=14 neurons from N=6 animals, hibernators, n=13 neurons from N=6 animals). Repeated Measures Two-way ANOVA results: Time, p<0.0001, Group, p<0.0001, Time x Group, p<0.0001. ** indicates p<0.01, ****p<0.0001, ▪▪▪▪ p<0.0001, and ▪▪ p<0.01 from Holm-Sidak Multiple Comparisons tests.

E) Tissue beta hydroxybutyrate (BHB) concentration changes in brainstem half exposed to 2DG-0G for 60 minutes relative to the control half maintained in normal conditions. Controls decreased BHB, while hibernators maintained tissue BHB. * indicates p<0.05 from Mann-Whitney U test.

We then assessed the two most energetically-demanding cellular processes needed to produce circuit output: excitatory synaptic transmission and action potential firing²⁴. For this, we measured AMPA-glutamatergic transmission (Figure Supplement 1) and firing activity in hypoglossal motoneurons from brainstem slices, as these neurons are part of the respiratory network²⁵. Rhythmically evoked synaptic transmission to generate patterned input similar to that observed in the intact network was reversibly suppressed by 2DG-0G, declining steadily over 30 minutes (Figure 1C top panel, Figure 1D). However, hibernators maintained synaptic activity at baseline levels without glycolysis (Figure 1C bottom panel, Figure 1D). In contrast, firing rates and intrinsic membrane properties in controls and hibernators were maintained without glycolysis (Figure Supplement 2), supporting previous work indicating that these processes are energetically cheap compared to synapses^18,26. These results demonstrate that ongoing synaptic transmission, thought to critically rely on glucose metabolism²⁷, can switch to a non-glucose, aerobic energy source generated within the brain.

Maintenance of ketone body homeostasis without glycolysis

While there are several potential alternative fuels that do not involve glucose or its derivatives, ketone bodies are well-known to maintain neuronal homeostasis when glucose is limited. Ketone bodies are thought to arise in the brain mainly from fatty acid metabolism in the liver, which then travel through the circulation to be transported across blood-brain barrier to support neuronal metabolism²². As a lone fuel source, however, ketone bodies cannot sustain ongoing neural activity after the early neonatal period^4,6, and while brain-derived ketone bodies have been identified²¹, their relevance to neural function is not clear.

To address the potential for ketone bodies produced within the isolated brainstem without glycolysis present, we measured tissue BHB concentration after 1 hr of 2DG-0G. Without glycolysis, BHB content decreased in controls, indicating the presence of ketone bodies in isolated brainstems and its consumption when glycolysis is blocked (Figure 1E). However, as network activity and synaptic transmission fell strongly over this time (Figure 1A &1B), any consumption of BHB is not sufficient to maintain activity. In contrast, hibernators maintained BHB concentrations during 2DG-0G (Figure 1E) while also maintaining neural activity and synaptic function (Fig 1A & 1B). Together, these results suggest that the isolated frog brainstem consumes BHB when glucose metabolism is stopped, but that hibernators may synthesize it to prevent net decreases to provide fuel for ongoing activity.

Brain-derived ketone bodies power neural activity without glycolysis

We then tested the possibility that ketone body production within the brain supports activity when running without glucose metabolism. For this, we measured synaptic transmission after blocking ketone body synthesis using 2.5 μM hymeglusin, a compound that selectively inhibits HMG-CoA synthase, a critical enzyme in the production ketone bodies^28–32. Applying hymeglusin in controls did not influence the requirement for glycolysis, showing a similar degree of reduced transmission by 30 minutes of 2DG-0G (Figure 2A, 2B, 2G). However, hymeglusin opposed much of the improvement without glycolysis in hibernators, showing weakened evoked transmission after 30 minutes (Figure 2C, 2D, 2G). To verify this outcome occurred through interfering with ketone body synthesis that provided energy to synapses, we performed a rescue experiment (Figure 2E–F). After 30 minutes of synaptic activity in hymeglusin+2DG-0G in hibernators, providing 5 mM BHB over the next 30 minutes increased transmission (Figure 2E), while transmission continued to decline in hymeglusin+2DG-0G without BHB over the 60 minute period (Figure 2F, 2H). The requirement for ketone body synthesis to maintain synaptic activity was mirrored in intact rhythmic networks from hibernators (Figure Supplement 3). Hymeglusin did not affect baseline synaptic transmission when glucose metabolism was intact (Figure Supplement 5 A), suggesting the loss of glucose metabolism initiates ketone body production for the preservation of synaptic transmission. Interestingly, when assessing neuronal excitability, hymeglusin exposure with glucose present increased firing rates only in hibernators with no effect on membrane potential (Figure Supplement 5 B–D), pointing to influences of HMG-CoA synthase activity on intrinsic excitability at baseline in addition to providing fuel during the loss of glycolysis. Overall, these results indicate active synthesis of ketone bodies within the slice sustains ongoing transmission without input from glycolysis after hibernation.

Figure 2. Brain ketone body synthesis maintains synaptic transmission without glucose metabolism after hibernation.

A-D) Example recordings of rhythmically-evoked synaptic transmission in brainstem slices at baseline and after 30 minutes without glycolysis in the presence or absence of 2.5 μM hymeglusin, an inhibitor of HMG-CoA synthase to produce BHB (colors). In controls, hymeglusin had no influence on the failure of synaptic transmission without glycolysis. After hibernation, robust synaptic transmission without glycolysis was opposed in the presence of hymeglusin, as indicated by smaller evoked synaptic currents during the stimulus train (light purple).

E-F) Example recordings of zoomed in rhythmically-evoked synaptic transmission in hibernator brainstem slices at baseline, after 30 minutes in hymeglusin without glycolysis, and then either 60 minutes in hymeglusin without glycolysis (E) or 60 minutes in hymeglusin without glycolysis but with 5 mM BHB added at 30 minutes (F). By 60 minutes without glycolysis and BHB synthesis, transmission falls to near zero, while adding BHB at 30 minutes recovers toward baseline levels.

G) Mean data of the change in synaptic transmission (area under the curve; AUC of the transmission envelope upon rhythmic stimulation) after minutes of 2DG-0G with and without hymeglusin (HG). One-way ANOVA, p<0.0001. **** indicates p<0.0001 and **p<0.01 from Holm-Sidak Multiple Comparisons Test. Control, n=14 neurons from N=6 animals. Control+HG, n=15 neurons from N=5 animals. Hibernator, n=13 neurons from N=6 animals. Hibernator+HG, n=18 neurons from 8 animals.

H) Mean data showing the effect of 60 minutes without glycolysis in the presence of hymeglusin with BHB (dark, +BHB) or without BHB (light, -BHB) added at 30 minutes. Data points are normalized to the initial baseline of each cell. Repeated Measures two-way ANOVA results: Group x time interaction, p<0.0001, group, p=0.0165, time, p=0.6350. p values are from Holm-Sidak Multiple Comparisons Test, where transmission fell between 30 and 60 minutes without BHB while it increased when BHB was supplemented.

While these results demonstrate ketone bodies can replace glucose, it is not clear if hibernation leads only to a capacity to produce ketone bodies, or if it also bolsters the capacity for synapses to import and use ketone bodies for fuel. To address these possibilities, we replaced glucose with BHB in controls. As control brains have a low biosynthetic capacity (Fig 1E), this address their ability to import and use ketone bodies provided externally as fuel. Providing BHB while blocking glycolysis did not sustain synaptic transmission (Figure Supplement 4A), and replacing glucose with BHB led to failure of rhythmic activity in intact rhythmic networks while hibernators continued unscathed (Figure Supplement 4B). Collectively, these results show that control frogs neither meaningfully produce nor consume BHB to support network activity. Therefore, hibernators gain both the ability to internally generate ketone bodies and use them as fuel to maintain activity without glucose metabolism.

A role for astrocytes

Brain Ketone body synthesis is thought to involve fatty acid oxidation and subsequent production in astrocytes, as these cells have a metabolic profile comparable to liver hepatocytes²¹. To test for the role of astrocytes in maintaining synaptic transmission without glycolysis, we incubated slices in 1 mM L-α-aminoadpic acid (AAA) for 2 hours, as this treatment induces astrocytic stress that disrupts their function in various species, including frogs^33–35. Supporting previous work indicating that similar incubations with AAA induces stress and death in astrocytes but not neurons, acute exposure to 1 mM AAA did not impair synaptic transmission (Figure Supplement 6A), and the 2 hr incubation did not affect neuronal membrane properties (Figure Supplement 6B). We then compared synaptic transmission after 30 minutes in 2DG-0G in slices that were maintained in aCSF or those with astrocytes disrupted. During the inhibition of glycolysis with 2DG-0G, synaptic activity in controls was reduced to a similar extent with or without AAA (Figure 3A, 3B, 3E), suggesting that astrocytes do not influence the typical failure of synaptic activity without glycolysis. In contrast, robust transmission seen in hibernators without glycolysis was opposed by AAA treatment, where synaptic activity fell to ~44% of baseline by 30 minutes (Figure 3C, 3D, 3E). This demonstrates that improvements in synaptic activity without glycolysis involves metabolic support by astrocytes, the primary site of BHB synthesis in the brain²¹.

Figure 3. Astrocytes contribute to synaptic activity without glycolysis and is associated with changes in gene expression for fatty acid metabolism and BHB transport.

A-D) Example recordings of rhythmically-evoked synaptic transmission in brainstem slices in controls and hibernators with and without exposure to 1 mM L-α-aminoadpic acid (AAA) to disrupt astrocytes. Transmission in controls fell from baseline (black) without glycolysis (2DG-0G) with or without intact astrocytes (A and B). Hibernators maintained transmission without glycolysis (2DG-0G) with astrocytes intact (C), but had reduced transmission when astrocytes were disrupted by AAA (D).

E) Mean data of the change in synaptic transmission (area under the curve; AUC of the transmission envelope upon rhythmic stimulation) after 30 minutes of 2DG-0G with and without astrocyte disruption by AAA. One-way ANOVA, p<0.0001. **** indicates p<0.0001 and **p<0.01 from Holm-Sidak Multiple Comparisons Test. Control, n=14 neurons from X=6 animals. Control+AAA, n=13 neurons from N=7 animals. Hibernator, n=13 neurons from N=X animals. Hibernator+AAA, n=14 neurons from 6 animals.

F-I) Increased mRNA expression for candidate genes involved in ketone body production and transport in brainstem tissue. CPT-1 (carnitine palmitoyl transferase 1), ACAA1 (acetyl-CoA acyltransferase 1), MCT1 (monocarboxylase transporter 1), MCT4 (monocarboxylase transporter 4). p values in plots are from two-tailed unpaired t test. q values are from the two-stage linear step up procedure of Benjamini, Krieger,and Yekutieli to control for the false discovery rate. N=11 per group for each gene.

An enhanced capacity of the brain to generate and use ketone bodies for fuel could relate to alternations in fatty acid breakdown, ketone body synthesis, ketone body catabolism, and transport. Therefore, we assessed mRNA expression of several candidate genes involved in these processes using quantitative real-time PCR in brainstem tissue homogenates. Brainstems from hibernators increased the mRNA expression of genes involved fatty acid metabolism that are thought to exist largely in astrocytes: Carnitine palmitoyl transferase I (CPT-1)³⁶ and peroxisomal thiolase that perform the last step of beta oxidation (ACAA2: enzyme that converts acyl-CoA into acetoacetyl-CoA)³⁷ (Figure 3 F–G). In addition, monocarboxylate transporters 1 and 4, which transfer BHB across the cell membrane, were upregulated, where MCT4 is primarily expressed in astrocytes³⁸ (Figure 3 H–I). Candidate genes involved in ketone body synthesis and catabolism were unaltered, along with several others for fatty acid metabolism and neuron-specific MCT2 (Figure Supplement 7). While these results do not exclude the possibility of post-transcriptional regulation on these cellular processes, and do not explicitly isolate gene expression to specific cell types or sub-cellular compartments, they point to brainstem-wide transcriptional program favoring fatty acid breakdown and BHB transport out of astrocytes, consistent with their role in improving synaptic function without ongoing glycolysis after hibernation.

Brain-derived ketone bodies promote activity in hypoxia

Finally, brainstem activity also becomes hypoxia-tolerant when frogs emerge from hibernation, where they maintain activity during tissue hypoxia and anoxia^17,18. This results from enhanced glycolytic support and reduced aerobic requirements for neural activity^17,18. However, BHB is a more efficient fuel than glucose, increasing ATP production by ~30% compared to glucose at similar rates of oxygen consumption³⁹. Therefore, the production of BHB during hypoxia could enhance energy status beyond using only glucose. To address this possibility, we measured network activity while reducing tissue pO₂. Corroborating previous reports, controls reduced network activity when tissue pO₂ dropped below 30 mmHg, whereas hibernators maintained activity near control levels down to anoxia (Figure 4A–C)⁴⁰. To determine if ketone body synthesis contributed to the shallow relationship between activity and pO₂ in hibernators, we used hymeglusin to block ketone body synthesis and BHB replacement to rescue. In the presence of hymeglusin, activity fell from baseline at pO₂ values ≤30 mmHg (Figure 4D, 4E, 4G). When we applied BHB in the presence of hymeglusin, the hypoxia-tolerant phenotype of the hibernators was restored, where activity did not significantly change from baseline down to anoxia (Figure 4F, 4G). Finally, hymeglusin did not affect the activity vs. pO₂ relationship in controls (Figure Supplement 8A), nor did it affect the apparent O₂ consumption (acute change in tissue pO₂ relating to O₂ consumption)⁴⁰ in controls or hibernators at baseline (Figure Supplement 8B). Therefore, not only can brain-derived ketones power activity in hypoglycemic conditions, but they also contribute to the maintenance of activity in hypoxia after hibernation.

Figure 4. Ketone body production contributes to improvements in network activity during hypoxia when glucose metabolism is intact.

A-B) Example recordings of tissue pO₂ (top), network burst rate (middle), and raw nerve activity (bottom) during graded tissue hypoxia in a control (A) and hibernator (B). In controls, as tissue pO₂ progressively decreases (top panel) network burst rate also falls (middle panel). Hibernators maintain network activity as O₂ falls. Inset illustrates the experimental set up used to measure network activity and tissue pO₂ 350 μM below the ventral surface within a component of the respiratory network.

C) Mean data showing the oxygen sensitivity (change in burst rate vs. tissue pO₂) in controls and hibernators. ▪ indicates differences across groups at each pO₂, and * represent differences from baseline within each group. Repeated Measures Two-way ANOVA results: pO₂, p<0.0001, Group, p=0.0006, pO₂ x Group, p=0.0005. ** indicates p<0.01, *p<0.05, ▪▪▪ p<0.001, ▪▪p<0.01, and ▪ p<0.05 from Holm-Sidak Multiple Comparisons tests.

D-F) Example recordings of tissue pO₂ (top) and network burst rate (bottom) during graded tissue hypoxia in hibernators, hibernators in the presence of hymeglusin (HG), and hibernators in the presence of HG and 5 mM BHB.

G) Mean data showing the oxygen sensitivity (change in burst rate vs. tissue pO₂) in hibernators with no drug, 2.5 μM hymeglusin, and 2.5 μM hymeglusin + 5mM BHB. * represent differences from baseline within the hymeglusin group. Repeated Measures Two-way ANOVA results: pO₂, p<0.0001, Group, p=0.0538, pO₂ x Group, p=0.0182. ** indicates p<0.01, and *p<0.05, from Holm-Sidak Multiple Comparisons tests.

Discussion

The maintenance of neural circuit activity is widely believed to require the metabolism of glucose. Even acute impairments in its delivery or metabolism rapidly disrupt the physiological processes that generate neural activity⁹. In contrast, we demonstrate that neural activity’s requirement for glucose is not fixed, but instead, can be remodeled and replaced with ketone bodies produced within the brain itself. Given that adult synapses have sizable glucose demands that cannot usually be fully replaced by alternative fuels^4,5, the capacity to abandon glucose and replace it with brain-derived ketone bodies, indicates a greater capacity for metabolic flexibility than is currently appreciated for the active brain.

The experiments here support that frogs typically have high glucose demands and cannot maintain neural function solely using internally-derived or externally-provided ketone bodies, much like adult mammals^4,5. However, hibernation reprograms metabolic needs such that neural activity can abandon glucose and operate on ketone bodies produced within the brain, with likely importance for restarting neural circuits as animals emerge from hibernation when oxygen and body glycogen stores are limited^14,41. Key components of this transformation involve a gained ability to: i. sense the loss of glucose metabolism, ii. increase ketone body production from products of beta-oxidation likely within astrocytes, iii. improve the capacity for release, uptake and/or metabolism of ketone bodies, and iv. fuel synaptic transmission while achieving network homeostasis. Mechanistically, astrocytic AMP-dependent protein kinase may be activated by glucose deprivation, leading to the inhibition Acetyl-CoA Carboxylase to facilitate fatty acid import into the mitochondria through CPT-1 for ketone body synthesis²¹, which we show to be strongly upregulated in hibernators (Figure 3F).

Our results support the role of brain-derived ketone bodies as a large contributor to running without glycolysis. We used two manipulations that disrupt their production, hymeglusin, which selectively inhibits HMG-CoA synthase involved in the synthesis of ketone bodies (acetoacetate and BHB), and AAA to lesion astrocytes, the principle source of ketone bodies in the brain. While our results support the role of brain-derived ketone bodies, a few possibilities may underlie the sub-cellular organization of their synthesis. In mammals, ketone body synthesis occurs primarily in mitochondria and uses HMG-CoA Synthase 2⁴². Lower vertebrates contain only the gene that codes for HMG-CoA synthase 1, which is canonically cytoplasmic rather than mitochondrial. However, HMG-CoA synthase 1 can exist in the cytosol or peroxisomes with a likely role in ketone body synthesis, as these compartments contain the full enzyme complement required to produce either BHB or acetoacetate^43–46. Furthermore, HMG-CoA synthase 1 in fish and human cancer cells can paradoxically show mitochondrial compartmentalization, activity, and a contribution to respiration^32,47,48. Any of these routes, likely within astrocytes, have the potential to contribute to ketone bodies within the hibernator brain to operate without glycolysis. Our results also do not exclude the possibility of other internal energy sources, as manipulations that impaired ketone body production still resulted in greater synaptic transmission than controls (Figures 2&3). Recent evidence suggests that neurons may burn triglycerides during glucose deprivation⁴⁹ and learning⁵⁰. Fatty acids within oligodendrocytes can also undergo beta-oxidation when glucose is limited⁵¹, which in principle could liberate ketone bodies for consumption by neurons. Additionally, microglia increase fructose synthesis in neurological disorders that occur with reduced glucose metabolism ⁵², and fructose can improve the recovery of synaptic transmission in naked mole-rats following glucose deprivation¹¹. Overall, while the production steps of ketone bodies and other endogenous energy sources require further exploration, our results demonstrate a wholesale shift in the capacity for ketone bodies synthesized within brain to power neural activity without glucose metabolism.

In addition to a large improvement in neural activity without glycolysis, brain-derived ketone bodies also contributed to the improvement in neural activity during hypoxia. Improved motor function in hypoxia/anoxia after hibernation was previously interpreted to result from enhanced glycolytic capacity and a lower energetic cost of activity^14,18. However, results here suggest that BHB synthesis helps to uphold activity during hypoxia. As BHB provides more ATP per unit of oxygen consumed than glucose²², it is tempting to speculate that brain-derived BHB provides an additional, more efficient fuel at low tissue pO₂ to support ATP synthesis through aerobic respiration. However, BHB also has diverse roles in cell signaling which may contribute to this response, where it can bind to surface receptors and acutely influence neurotransmitter and neuromodulatory systems⁵³. Indeed, hymeglusin, and presumably a reduction in endogenously produced ketone bodies, increased neuronal excitability when glucose metabolism was intact only in hibernators (Figure Supplement 5D). This is consistent with the ability for BHB to stimulate G-protein coupled receptors to influence neuronal excitability independent from its roles in energy metabolism^54,55. Regardless of the specific energetic or signaling pathways, we demonstrate that circuits can exploit locally-sourced ketone bodies to promote activity during hypoxia in addition to serving as an important energy source when glucose metabolism is absent.

The findings have general implications for understanding energetic plasticity in neural circuits. While brain function is thought to be constrained by glucose metabolism, it appears that some circuits are equipped with latent states that afford greater flexibility to run on internal, non-glucose fuels than is presently appreciated. Part of this resilience involves events that reprogram neural circuits to generate, transport, and metabolize ketone bodies as a fuel for synaptic function during glucose deprivation. A previously unknown capacity to stop glucose metabolism and replace it with locally sourced ketone bodies, and likely other endogenous fuels, is especially timely, as many neurological disorders are caused by or associated with impairments in glucose metabolism, often with ketone body metabolism left largely intact^1,2. While there are idiosyncrasies across various animal models and humans, at a fundamental level, many of the energetic rules are conserved^56,57. Thus, these results provide a path to understanding how to transition active neural circuits away from glucose and replace them with locally-derived alternatives without severely compromising brain function. Understanding the regulatory processes that control these state transitions have the potential to inform new strategies that protect neural activity in a wide-range of disease states associated with compromised glycolysis.

Material and Methods

Animals

All experiments performed were approved by the Animal Care and Use Committee (ACUC) at the University of Missouri (protocols #39264 and #65623). Adult American Bullfrogs, Aquarana catesbeiana, of undetermined sex were purchased from Rana Ranch (Twin Falls, Idaho, USA) and Niles Biological (Sacramento, CA, USA). Controls and cold-acclimated hibernators were maintained as previously described ^19,58. We refer to the cold-acclimated group here as “hibernators,” as cold-submergence in ranid frogs leads to metabolic suppression beyond that which can be explained by reduced body temperature alone, characteristic of hibernation in vertebrates⁴¹. Briefly, animals were kept in plastic tubs with dechlorinated tap water at room temperature bubbled with room air. Control frogs were acclimated to these conditions for at least a week after arrival at the laboratory before experiments and had access to wet and dry areas in the tanks. Pellet food was provided once per week and was eaten ad libitum. Hibernated frogs were kept in plastic tanks in temperature control incubators for > 1 week before temperature was lowered in stepwise manner to 4°C over 7 days in a walk-in temperature-controlled environmental chamber. Once water temperature reached 4°C, air access was blocked using a plastic screen placed in the tank. After at least 3 weeks of submergence, experiments commenced.

Tissue Preparations for Physiology Experiments

Brainstem-spinal cord preparation and data processing

Rhythmically active brainstem-spinal cords were generated as previously described^19,58 In summary, frogs were anesthetized with isoflurane (1 ml per liter, v/v) until loss of the toe pinch reflex and then rapidly decapitated with a guillotine. The brainstem and rostral portion of the spinal cord were then dissected from the head while submerged in cold artificial cerebrospinal fluid (aCSF; concentrations in [mM]: 104 NaCl, 4 KCl, 1.4 MgCl₂, 7.5 glucose, 40 NaHCO₃, 2.5 CaCl₂ and 1 NaH₂PO₄, bubbled with 98.5% O₂, 1.5% CO₂; pH ≈ 7.85). Once exposed, the forebrain was pithed. The brainstem-spinal cord was removed with the cranial nerve roots intact and placed ventral side up in a 6 ml chamber superfused with gased aCSF. All experiments in controls and hibernators were performed at room temperature (20-22 °C). Flow rate of the superfusion was maintained at 6 ml per minute with a peristaltic pump (Ranin Rabbit 4 channel head, Mettler-Toledo Rainin, Oakland, CA, USA).

To monitor output from the respiratory motor network, we recorded activity of cranial nerve X with custom made glass suction electrodes. Nerve activity was amplified 1000X (A–M Systems Model 1700, A–M Systems, Carlsborg, WA, USA), filtered (10 Hz – 5 kHz), and digitized with a Powerlab 8/35 (ADInstruments, Colorado Springs, CO, USA). For quantification, amplified nerve output was integrated and rectified online with a T of 100 ms using the LabChart software (ADInstruments, Colorado Springs, CO, USA). Preparations were then given 2-3 hours to stabilize before experiments. For experiments addressing the requirement of glycolysis for activity of the preparation (controls, N=8, hibernators, N=7), glucose was removed from the aCSF, and 5 mM 2-deoxy-D-glucose was added (2DG-0G) and left for 90 minutes. To demonstrate the requirement for aerobic respiration in the maintenance of activity in 2DG-0G, after 90 minutes the aCSF was bubbled vigorously with 98.5% N₂/1.5% CO₂, a treatment that causes tissue anoxia under the conditions of our experiments⁴⁰. For experiments that assessed the ability of the network to run on beta hydroxybutyrate (BHB) with no glucose (control, N=8; hibernator, N=7), glucose was removed and substituted with equimolar BHB. In experiments that address the influence of hymeglusin on longevity of the hibernator neural activity without during 2DG-0G, the protocol was as follows. After the baseline period in aCSF, 2.5 μM hymeglusin was added to the aCSF and superfused the preparation for 40 minutes (n=6). In naïve hibernators (n=6), no drug was added. Then 2DG-0G was applied for 90 minutes. At the end of the 90 minute period of 2DG-0G+hymeglusin, 5 mM BHB was added to the aCSF. Finally, we performed a set of experiments where oxygen in the bathing solution was gradually decreased while measuring the tissue partial pressure of O₂ (pO₂). For this, we measured tissue pO₂ as described recently by our group⁴⁰. To standardize the initial pO₂ across experiments, the electrochemical sensor was inserted into brainstem until 100 mmHg was reached. This protocol was first done in controls (n=6) and hibernators (n=6). In addition, 2 addition groups of hibernators were included: 2.5 μM hymeglusin (n=6) and 2.5 μM hymeglusin+ 5 mM BHB (n=5). Finally, hymeglusin was applied to control preparations, as well (n=6). Hymeglusin is also likely to inhibit HMG-CoA synthase involved in cholesterol synthesis; however, we are not aware of mechanisms involving the acute loss of cholesterol synthesis that would explain rapid disruption of neuronal homeostasis only in hibernators upon energy stress and rescue by ketone bodies.

For processing of nerve activity data, burst rate of activity related to breathing was counted over ten minute sampling periods and presented as bursts per minute. Breathing-related neural activity has a characteristic shape of “smooth” rising and falling phases that occur over ~1s and is the default activity mode of this preparation (Supplemental Figure 3, inset). In cases where burst amplitude of breathing-related activity was measured, the Peak Analysis module of LabChart Analysis software was used to measure the height, from the baseline to the burst peak (ADInstruments, Colorado Springs, CO, USA). Non-respiratory bursts, which are indicative of a loss of homeostasis during energy stress¹⁹, were identified by their long and/or large amplitude relative to breathing-related bursts and longer duration (ranging from 5 to 20 seconds) (Supplemental Figure 3, inset) and presented as bursts in 10 minutes. Finally, sensitivity of the respiratory burst activity to decreases in tissue pO₂ were presented as % change from baseline as a function of tissue pO₂. Changes in pO₂ during fixed bath oxygen relate to changes in tissue oxygen consumption⁴⁰. To assess the potential for changes in oxygen consumption by hymeglusin application, we measured tissue oxygen after hymeglusin was applied.

Hypoglossal motoneuron-containing slice preparation, patch clamp electrophysiology, and data processing

Tissue slices containing hypoglossal motoneurons for patch clamp electrophysiology experiments were generated following previously established protocols ^59,60. Briefly, the brainstem-spinal cord was dissected as described for the brainstem-spinal cord preparation but instead of recording nerve activity, was sectioned using a Campden Vibrating Microtome (7000smz, Campden Instruments; Lafayette, IN, USA). For this, the brainstem-spinal cord was superglued to an agar block and then cut into 300 μM-thick sections. Slices containing the hypoglossal motor pool were given at least 1 hr to recover from the slicing procedure, transferred to a 0.5 ml chamber, and fixed to the chamber with a nylon grid for patch clamp electrophysiology experiments. The chamber was situation in a fixed stage microscope for imaging of the slice (FN1, Nikon Instruments Inc., Melville, NY, USA). Oxygenated aCSF was fed through the chamber with a gravity-fed superfusion system at ~1.5 ml per minute.

Hypoglossal motoneurons in the rostral part of the motor pool were identified anatomically based on their dorso-medial position relative to the 4^th ventricle and large cell body size (>20 μM diameter along the shortest axis) as previously described⁶⁰. Slices were imaged using a Hamamatsu ORCA Flash 4.0LT sCMOS camera (Hamamatsu Photonics, Hamamatsu City, Japan) interfaced with NIS elements software (Nikon). Hypoglossal motoneurons in the rostral region were targeted as they are part of the buccal force pump for ventilation in anuran amphibians^25,61 and therefore play a role in breathing and other orofacial behaviors that are critical to survival^62,63. Once identified, motoneurons were approached by applying positive pressure through the glass microelectrode for patch clamp recording. Electrophysiological signals were amplified by an Axon Instruments 200B patch clamp amplifier and digitized with a Digidata 1550 digitizer (Molecular Devices). Microelectrodes were fabricated with a Sutter Instrument P-1000 micropipette puller (Novato, CA, USA) and had resistances of 5–6 MΩ when filled with the pipette filling solution (in mmol): 110 potassium gluconate, 2 MgCl₂, 10 Hepes, 1 Na₂-ATP, 0.1 Na₂-GTP and 2.5 EGTA, pH ~ 7.2 with KOH. As the motoneuron was approached, debris was cleared by the positive pressure and when an indentation was observed in the center of the cell body, positive pressure was removed, increasing the resistance between the microelectrode and the cell membrane. Negative pressure was then applied until a seal >1G was observed and was then broken using rapid negative pressure by mouth to obtain whole-cell access.

After obtaining whole-cell access, neurons were maintained in voltage-clamp mode at a holding voltage of −80 mV. Excitatory postsynaptic currents (EPSCs) carried by AMPA-glutamate receptors via stimulation of solitary tract (ST) axons⁶⁴ were maximally evoked using an AM-Systems isolated pulse stimulator (Model 2100, , A–M Systems, Carlsborg, WA, USA) with bipolar tungsten steroelectrodes (WE3ST0.1B10, Microprobes for Life Sciences, Gaithersburg, MD, USA) AMPA-glutamatergic identity of these synapses was determined by applying 20 μM DNQX, which eliminated evoked tranmission. Maximal synchronous EPSC amplitudes were determined by increasing the stimulus current until asynchronous vesicle release was observed (EPSCs that trailed after the synchronous EPSC and were larger than spontaneous EPSCs). The stimulus intensity was then reduced to maintain the largest possible synchronous EPSC amplitude without asynchronous activity. For experiments in this study, the main protocol was as follows.

The amplifier was switched to current clamp mode to measure membrane properties (membrane potential, firing rate during stepwise current injection, and input resistance). The amplifier was switched back to voltage clamp with a holding potential of −80 mV and axons were stimulated rhythmically at maximum intensity (20 Hz, 1s duration, every 5 s) to simulate the ongoing rhythmic input to these neurons. After 5 minutes of baseline stimulation, 2DG-0G was applied for 30 minutes during rhythmic stimulation. Every 10 minutes, rhythmic stimulation was paused, and the amplifier was switched back to current clamp mode to assess membrane properties. After assessing membrane properties, the amplifier was switched back to voltage clamp mode and rhythmic synaptic stimulation resumed for the next 10 minute interval.

In some experiments (control, n=14; hibernator, n=13), glucose metabolism was restored for the next 30 minutes during rhythmic synaptic stimulation. In a different set of experiments, controls (n=15) and hibernator (n=18), this protocol was performed in the presence of 2.5 μM hymeglusin. For this, after entering whole-cell mode, baseline membrane properties were determined in current clamp, and then measured rhythmic trains of EPSCs in voltage clamp for 5 minutes. Hymeglusin was applied for 10 minutes during rhythmic synaptic stimulation, and then amplifier was changed back to current clamp to assess membrane properties in the presence of hymeglusin. After returning to voltage clamp and restarting rhythmic stimulation, 2DG-0G was applied in the presence of hymeglusin for the next 30 minutes. In a subset of hibernator cells in the presence of hymeglusin, n=7 continued to be rhythmically stimulated with 2DG-0G for the next 30 minutes (60 minutes in total) or n=8 had 5 mM BHB added during the next 30 minutes while being stimulated in the presence of 2DG-0G (60 minutes in total, first 30 minutes 2DG-0G+hymeglusin and the last 30 minutes 2DG-0G+hymeglusin+BHB). In a different set of experiments, control neurons (n=15) were exposed to the rhythmic stimulation protocol in the presence of 2DG-0G with 5 mM BHB.

Finally, to assess the involvement of astrocytes, we incubated slices for 2 hr in 1 mM L-α-aminoadpic acid (AAA), as this treatment has been demonstrated to induced death in astrocytes while having minimal effects on neurons^33–35. While we are not aware of studies tracing the developmental lineage of astrocytes in frogs, the central nervous system of frogs contains glial cells consistent with astrocytes, with glycogen granules⁶⁵, Glial Fibrillary Acidic Protein (GFAP) expression, and an ohmic K⁺ selective membrane⁶⁶. Following incubation of slices, the same protocol was performed as before with 2DG-0G during rhythm stimulation of EPSCs. In n=3 cells, AAA was applied acutely for 30 minutes and did not influence synaptic transmission. In all experiments, negative pressure was applied to the microelectrode to maintain series resistance through the experimental period. If series resistance changed by 20% and could not be reduced via negative pressure, the experiment was abandoned. One slice was used per exposure to 2DG-0G. If the recording was lost in the middle of 2DG-0G, the slice was discarded.

All patch clamp electrophysiology data were analyzed using the Easy Electrophysiology data analysis program (Easy Electrophysiology Ltd., London, UK). Synaptic transmission was assessed as the area under the curve (AUC) during 1 minute of stimulation at the end of each sampling period. Firing rate was taken as the firing rate during 500 ms current injections and is presented as either maximally firing rate or the entire frequency vs. current relationship. Input resistance is taken as the change in voltage change from resting membrane potential during a 100 pA step.

Drugs

DNQX was from Hello Bio (Princeton, NJ, USA). Hymeglusin was from Cayman Chemical (Ann Arbor, MI, USA). L-α-aminoadpic acid (AAA) was from Thermo Scientific Chemicals (Waltham, MA, USA). R)-(-)-3-Hydroxybutyric acid sodium salt (BHB) was Santa Cruz Biotechnology Inc. (Dallas, TX, USA).

Tissue BHB determination

To determine the influence of inhibiting glycolysis on tissue beta hydroxybutyrate (BHB) we used an experimental approach that allows each brain to serve as its own control. Following dissection of the brainstem-spinal cord in controls and hibernators as described above, the preparation was transected down the midline, separating the two hemispheres. One half was left in normal aCSF (control half), and the other was exposed to 2DG-0G to inhibit glycolysis (2DG-0G half). After one hour in 2DG-0G, 40 mg of brainstem from each half was quickly dissected, dried of excess liquid, weighed, minced into small pieces, and then added to separate 1.5 mL tubes containing 229 μL of Tris buffer. The contents were then homogenized with a mechanical tissue homogenizer. To measure tissue BHB, we used a beta hydroxybutyrate colorimetric assay kit from Cayman Chemical following the manufactures instructions (kit #700190, Ann Arbor, MI, USA). Briefly, the homogenate was centrifuged at 1,000 x g for 10 minutes at 4°C. The supernatant was removed and then added to a microcentrifuge tube and centrifuged at 10,000 x g for 10 minutes at 4°C. The supernatant was removed and the pellet was resuspended in 150 μl of assay buffer. Resuspended pellet was then stored in a −80 freezer until all samples were collected. BHB concentrations were assayed in all samples run in duplicate on the same plate using a Varioskan LUX plate reader (Thermo, Waltham, MA, USA) and quantified by constructing a standard curve the following manufacturer’s instructions. We then normalized BHB concentrations to total protein from each sample in duplicate using a Bradford Assay. The BHB concentration for each 2DG-0G half was divided by the concentration from the control half to determine the change in BHB in response to 2DG-0G within the same brain.

RNA isolation and real-time quantitative PCR

All methods for sample collection, RNA isolation, cDNA synthesis of 1 μg of cDNA, and real-time quantitative PCR (qPCR) using 1/5 diluted cDNA template were performed as described previously by our lab⁶⁷. This study, we identified candidate genes involved in steps of beta oxidation, ketogenesis, ketolysis, and BHB transport (Table 1). We first sought to identify annotated genes in the Aquarana catesbeiana genome. Only 1 candidate gene was annotated. We then used sequences from Rana temporaria or Xenopus laevis, frog species with well-annotated genomes, to query the coding sequence of non-annotated “hypothetical proteins” in the Aquarana catesbeiana genome⁶⁸. Once a hypothetical protein was identified as likely to be homologous to that of Rana temporaria or Xenopus laevis, the animo acid sequence was BLASTed against the entire non-redundant protein data base (BLAST, NCBI, USA). In the case of 6 candidate genes, queries returned hits of high sequence identity with the excepted results. For example, a protein BLAST of Aquarana catesbeiana HMGCS1 returned a list of hits containing HMGCS1 from a variety of other species. We then used the coding DNA sequence (CDS) to design qPCR primers for these candidates. If we were unable to identify annotated or hypothetical proteins with relevant hits in the reBLAST of the sequence from Aquarana catesbeiana, we used CDS sequences from Rana temporaria to design qPCR primers, as this species is closely-related containing a genome with a chromosome-level assmebly⁶⁹. All qPCR primers were designed using PrimerBlast and validated in-house. For this, all primers were tested using a 4-fold dilution series and produced a single peak during melt curve analysis. If these criteria were not met, forward and reverse primers were redesigned. Changes in gene expression were determined using the 2⁻ΔΔct method⁷⁰. 18s ribosomal RNA (rRNA) from stock cDNA (original 1 μg diluted 5-fold) was further diluted by 50-fold and used as the control gene in analysis. C_t values of 18S rRNA for controls (20.4±0.4) and hibernators (20.3±0.5) were not different (p=0.93, two-tailed unpaired t test).

Table 1.

Primer information for candidate genes

Targets	Forward Primer Sequence	Reverse Primer Sequence	PCR Efficiency (%)	NCBI accession number Aquarana catesbeiana	NCBI accession number Rana temporaria
CPT-1L	GGGCCATCTCTGGGAGTACA	CAGTCTGCTTGGAGGAGGGA	91.6	PIO37846.1	■
CPT-1M	AGCTGGGGACCGAGTTTCTT	GCATCAGGGTCCAGGCAAAC	96.4	PIO31695.1
CPT2	GGTGCAGCTAGGAGGCTTTG	GCAACCAATCCAGTCGTCGT	94.1	■	XM_040360458.1
LCAD	GACAGGTGAGCCGGGAATTG	CACCAATGCCACCATGCTCT	94.2	■	XM_040357444.1
VLCAD	TCCGGCTGTTTGTTGCTCTC	GCTCAGTGGATTGGTGACGG	90.5	■	XM_040334301.1
MCAD	ACGCGATGGAGAGGAAGACA	CCCAGGCTGCTCTCTGGTAT	90.4	■	XM_040360675.1
ACAA1	GGACGTCAACCTGAAGCCTG	GGCAGCTCTAGCAACCACAG	94.7	■	XM_040353382.1
ACAA2	CACTGCTGAGAACCTGGCTG	GTCCTGAGCGGCTTTCCATC	90.9	■	XM_040336875.1
HMGCS1	CCTCAGGGCTCAGTGGATGA	GCCTCCAGGGGTCCATCATT	98.6	PIO24663.1	■
HMGCL	ACCTGTGCGAGGGTATGTGT	TGTGTGCCACCTCTGCTACT	90.0	■	XM_040343136.1
BDH1	CCAAGGCCTGTCTCCCTCTT	GGGGGAACGAGCTGGATTTG	98.9	PIO23142.1	■
BDH2	GACACGAGTGCTGGATGTGA	GGATCCGTGGTGGACAAAGC	85.0	ACO51978.1
ACAT2	TTGTGGTTGCCGGTGGTATG	TGGGACATCCCCAGCTTTCA	94.1	■	XM_040340584.1
OXCT1	AGCGTCATTCTTCTCCAGCG	TTGACACCTGCATTGCTCCC	97.6	■	XM_040338845.1
AACS	GGGAGTGGAAGCATGGTGTG	TGTGTGGGCTGAGATGGGAG	97.5	PIO41121.1	■
MCT1	TGGCTTTGCTTTTGGCTGGC	GGCCCAAGAAGAACAGGGCA	90.0	■	XM_040337478.1
MCT2	TGCCAGGCCGTTAACTGGAA	GACAAGGCCCGCGTAGGAAG	93.4	PIO23679.1	■
MCT4	CCCCGTCTTCCTCTGTTGCC	ATAAGAAAGCCACCGCGCCA	88.0	■	040330784.1
18S rRNA	CAGGCCGGTCGCCTGAATAC	GGCCCCAGTTCCGAAAACCA	101.068	KV934453.1	■

Statistics and data presentation

Data are presented as mean ± standard deviation or box and whisker plots, with single data points from individual experiments, where it is most appropriate for visualization. For physiology assessments and measurements of tissue BHB, when data sets were approximately normally distributed, parametric statistical tests were used. When two groups were compared in before-after/time course experiments (i.e., with dependent samples), a two-tailed paired t-test or repeated measures two-way ANOVA were used. When groups contained independent samples, a two-tailed unpaired t-test or one-way ANOVA was used. One-way and two-way ANOVAs were followed with Holm-Sidak multiple comparisons tests. In cases where data sets were not normally distributed, non-parametric statistical tests were used. In these cases, when analyzing two independent groups, a Mann-Whitney tests was used, and when three with three or more groups, a Kruskal-Wallis test was used. For mRNA expression, all comparisons for each of the 18 candidate genes between controls and hibernators were compared two-stage linear step up procedure of Benjamini, Krieger,and Yekutieli to control for the false discovery rate. Significance was accepted when p< 0.05. All analyses were performed using GraphPad Prism (v9.4.1, San Diego, CA, USA).