Neuromodulation or energy failure? Metabolic limitations silence network output in the hypoxic amphibian brainstem

Abstract

Hypoxia tolerance in the vertebrate brain often involves chemical modulators that arrest neuronal activity to conserve energy. However, in intact networks, it can be difficult to determine whether hypoxia triggers modulators to stop activity in a protective manner or whether activity stops because rates of ATP synthesis are insufficient to support network function. Here, we assessed the extent to which neuromodulation or metabolic limitations arrest activity in the respiratory network of bullfrogs—a circuit that survives moderate periods of oxygen deprivation, presumably, by activating an inhibitory noradrenergic pathway. We confirmed that hypoxia and norepinephrine (NE) reduce network output, consistent with the view that hypoxia may cause the release of NE to inhibit activity. However, these responses differed qualitatively; hypoxia, but not NE, elicited a large motor burst and silenced the network. The stereotyped response to hypoxia persisted in the presence of both NE and an adrenergic receptor blocker that eliminates sensitivity to NE, indicating that noradrenergic signaling does not cause the arrest. Pharmacological inhibition of glycolysis and mitochondrial respiration recapitulated all features of hypoxia on network activity, implying that reduced ATP synthesis underlies the effects of hypoxia. Finally, activating modulatory mechanisms that dampen neuronal excitability when ATP levels fall, K_ATP channels and AMP-dependent protein kinase, did not resemble the hypoxic response. These results suggest that energy failure—rather than inhibitory modulation—silences the respiratory network during hypoxia and emphasize the need to account for metabolic limitations before concluding that modulators arrest activity as an adaptation for energy conservation in the nervous system.

INTRODUCTION

The nervous system of most animals requires a constant supply of O₂ to function properly. Many species, however, face a shortage of O₂ in their tissues owing to environmental or physiological constraints. Accordingly, animals use a host of mechanisms to ensure the survival of neurons when O₂ falls below a level that supports aerobic metabolism. One well-studied strategy to survive O₂ deprivation involves engaging inhibitory modulators or neurotransmitters that arrest brain activity. This serves to offset the high cost of neural function and provides several benefits such as improving neuron survival, maintaining brain [ATP] and [ADP], and allowing neurons to recover activity once O₂ returns (1, 2). Thus, when neural systems appear to arrest activity in a protective manner, these responses tend to be interpreted as adaptions for energy conservation and hypoxia tolerance in the nervous system (3).

Although inhibitory modulators provide a way to conserve brain energy by reducing neural activity, it can be challenging to assess the contribution of these processes to the inhibition of complex circuits. This difficulty arises because an arrest may reflect a regulated, modulatory response in some systems, but in others, unmatched metabolic demands during hypoxia may reduce network activity. For example, turtles engage a suite of modulator-triggered mechanisms that suppress brain activity during anoxia (1). In direct contrast, electric fish species with large brains have a reduced ability to swim and discharge the electric organ in hypoxic water, a negative outcome thought to stem from the high metabolic cost of a large brain (4). Further complicating matters, energetic failure and modulatory systems may interact to produce a stereotyped response to energy stress. For example, the anoxic coma in insects has been described as a “vulnerability” because some of the underlying mechanisms resemble metabolic failure seen in ischemic stroke, but also as “advantageous by conserving energy,” since modulatory mechanisms are at play (5). Thus, an activity shutdown during oxygen deprivation may involve inhibitory modulators, energetic failure, or some combination of the two. The specific mechanisms that underlie a shutdown have implications for understanding adaptations for hypoxia tolerance in the nervous system.

We used the respiratory network in the adult American bullfrog, Lithobates catesbieanus, to address the extent to which metabolic failure versus modulatory mechanisms cause an activity arrest during severe hypoxia. During severe hypoxia, rhythmic output from this network falls silent within a few minutes (6–8), a response with potential physiological relevance when blood oxygen drops to low levels during diving (9) and breath holding (10). This neural response is thought to arise from the hypoxia-sensitive release of norepinephrine (NE), which then inhibits the network though indirect GABA_A/glycine pathways (6, 11). The depression of neural output from the respiratory network has been deemed an adaptive, energy-saving arrest because it may help conserve energy at the organismal level, as well as improve neuron survival and recovery after O₂ is restored (6, 8, 11). This interpretation is reasonable; however, several arguments could be made to suggest that energetic limitations contribute to a reversible arrest of activity independent of noradrenergic modulation. First, ATP must be continuously synthesized to ensure communication at excitatory synapses, commanding a large fraction of a neuron’s energy budget (12–14). As rhythmic output from the bullfrog respiratory network relies on glutamatergic transmission (15–17), coordinated network activity may be especially prone to metabolic failure. Next, glycolytic ATP production can maintain ion homeostasis in silent neurons during anoxia (18). Thus, if network function collapses because of insufficient rates of ATP production during severe hypoxia, glycolysis may still be able to fuel essential housekeeping functions until O₂ returns, a response more consistent with energetic failure than a programmed downregulation of energetic demands. Finally, neural function and behavior can recover following hours of anoxia even after ATP reduction by 97% in Drosophila (19), demonstrating remarkable robustness of some systems to dramatic energy depletion. Collectively, these examples demonstrate that neural circuits may not always need to rely on modulatory mechanisms to survive and recover from bouts of oxygen deprivation.

On this background, we tested the extent to which energetic limitations versus noradrenergic mechanisms contribute to the reversible activity shutdown during severe hypoxia in the respiratory network of bullfrogs in vitro. To address this, we first compared neural respiratory responses between severe hypoxia and NE. We then tested whether pretreatment of preparations with NE occludes, and application of an adrenergic receptor inhibitors blocks, the hypoxic response as would be expected if severe hypoxia triggers the release of NE to silence the network. Next, we applied inhibitors of different metabolic pathways in the presence of oxygen to assess if metabolic failure without hypoxia recapitulates the hypoxic shutdown. Finally, we applied activators of cellular mechanisms that sense energy status, K_ATP channels and AMP-dependent protein kinase, to assess whether energy sensors may contribute to the hypoxic response.

METHODS

Ethical Approval

All experiments have been approved by the Institutional Animal Care and Use Committee at The University of North Carolina at Greensboro.

Animals

Forty-eight adult female American bullfrogs (L. catesbieanus) were purchased from Rana Ranch (Twin Falls, ID). Bullfrogs (∼100 g) were housed in plastic tanks with dechlorinated, aerated tap water at 22°C and fed pellets provided by Rana Ranch twice per week. Frogs had access to both wet and dry areas. Experiments for series 1, 3, and 4 were performed between May and September of 2019, and those for series 2 were performed in August of 2020.

Brainstem-Spinal Cord Preparation

The brainstem-spinal cord preparation was generated as described previously (17). Briefly, frogs were deeply anesthetized with ∼1 mL of isoflurane in a 1 L container until loss of the toe-pinch reflex and then rapidly decapitated. The head was immersed in cold (2–4°C) artificial cerebrospinal fluid (aCSF) containing (in mM) 104 NaCl, 4 KCl, 1.4 MgCl₂, 7.5, d-glucose, 40 NaHCO₃, 2.5 CaCl₂, and 1 NaH₂PO₄ bubbled with 1.5% CO₂-98.5% O₂ (pH = 7.85). The skull covering the forebrain and brainstem was quickly removed, the animal was decerebrated, and then the brainstem-spinal cord was carefully removed. The dura covering the brainstem-spinal cord was removed. The brainstem was placed in room-temperature aCSF bubbled with 1.5% CO₂-98.5% O₂ circulated with a peristaltic pump at ∼8 mL/min (Mini Pump Variable Flow, Fisher Scientific, Hampton, NH) for 3–4 h to recover from the dissection. All preparations used in this study had stable rhythms for the entire period following the dissection and throughout the experimental protocols.

Extracellular Nerve Recordings

Respiratory-related extracellular nerve activity was recorded from cranial nerve roots using suction electrodes. Glass electrodes were produced from Borosilicate glass pulled to a fine tip using a horizonal pipette puller (P87; Sutter Instruments, Novato, CA), broken to fit snuggly around the nerve roots, and then fire-polished. Cranial Nerves V (CN V) and X (CN X) were pulled into glass electrodes. Extracellular activity was amplified (×1,000) and filtered (low pass, 1,000; high pass, 100) using with an AM-Systems 1700 amplifier (Sequim, WA), digitized with a Powerlab 8/35 (ADInstruments, Sydney, Australia), and rectified and integrated (100 ms τ) using the LabChart data acquisition system (ADInstruments). Throughout the results, we show raw traces from CNV; however, treatments had similar effects on both nerves.

Experimental Procedures

Series 1 experiments: comparing responses to hypoxia with norepinephrine.

In these experiments, preparations were exposed to either severe hypoxia (n = 9, 1.5% CO₂-98.5 N₂) or 5 µM NE (n = 6) for 25 min. Previous work by Winmill et al. (8) measured tissue PO₂ under similar experimental conditions as we used here and found PO₂ to be close to 0 mmHg within the tissue; thus, we refer to this manipulation as severe hypoxia. Five micromolar NE was chosen because it produces a saturating response but does not appear to desensitize adrenergic receptors in the bullfrog brainstem-spinal cord preparation (21). Preparations treated with NE were then exposed to severe hypoxia for an additional 25 min to determine the response to hypoxia on the background of activated adrenergic receptors.

Series 2 experiments: response to severe hypoxia in the presence of an adrenergic antagonist.

Preparations (n = 5) were exposed to 500 nM prazosine to block α1 adrenergic receptors, the receptor that mediates the depressive actions of NE in this network (21). In these experiments, we bath-applied prazosin for 20 min to block α1 receptors and then exposed the preparation to severe hypoxia (0% O₂). After 25 min of severe hypoxia, we allowed each preparation to recover baseline burst frequency in oxygenated aCSF for ∼1 h. To verify whether prazosin blocks the response of the network to NE, we then exposed the same preparation to 5 µM NE.

Series 3 experiments: comparing responses to hypoxia with chemical inhibition of ATP synthesis.

Preparations were exposed to inhibitors of glycolysis [2-deoxy d-glucose substituted for d-glucose (7.5 mM 2-DG; n = 5) and 1 mM iodoacetate (IAA; n = 5)] and mitochondrial respiration [1 mM cyanide (CN; n = 5)] for 25 min. All preparations exposed to 2-DG and CN recovered burst frequency to baseline values upon washout; however, IAA is well established as an irreversible inhibitor of glyceraldehyde-6-phosphate dehydrogenase ([G6APD(H)] (23), and preparations did not recover during washout. IAA inhibits G6APD(H) with an apparent IC₅₀ of ∼100 µM in cultured cells (23); however, IAA is commonly used at ∼1–2 mM in in vitro tissue preparations to ensure complete block of glycolysis (18, 24). Each chemical is water-soluble and was dissolved in aCSF.

Series 4 experiments: K_ATP channel and AMP-dependent protein kinase activation.

Preparations were exposed to the K_ATP channel activator, diazoxide (n = 6; 600 μM), and AMPK activator, [1 mM AICAR; n = 7) for 25 min. AICAR is water-soluble and was solved in aCSF; however, diazoxide is not water soluble and was stored at 75 mM stocks in dimethyl sulfoxide (DMSO). The final concentration of DMSO in the experimental solution was 0.85%. In these experiments, preparations were first exposed to DMSO at 0.85% and allowed to stabilize for 30 min. We then applied another solution containing diazoxide to ensure any effect was caused by diazoxide, not DMSO.

Drugs

The following drugs were used in this study: iodoacetic acid (IAA; Acros Organics, Morris Plains, NJ), norepinephrine bitartrate (NE; Tocris Bioscience, Bristol, UK), prazosin hydrochloride (Alfa Aesar, Harverhill, MA), sodium cyanide (CN; Acros Organics, Morris Plains, NJ), 2-deoxy d-glucose (2-DG; Alfa Aesar, Tewksbury, MA), 5-aminoimidazole-4-carboxamide 1-β-d-ribofuranoside (AICAR; TCI America, Portland, OR), and diazoxide (TCI America).

Data Analysis

Frequency of the motor activity associated with lung breathing was quantified for a 5-min baseline period before each treatment. In series 1, 2, and 3 experiments, effects of treatments on lung burst frequency was quantified during a 3-min sampling period between 13 and 16 min (noted as “15 min”) and 22–25 min (noted as “25 min”). Time until the large motor burst was also quantified as the time between application of the treatment and time until initiation of the large motor burst. For series 4 experiments, the last 5 min of the exposure to the drug was analyzed. Lung bursts were identified by near-synchronous, large amplitude extracellular activity on both CN V and CN X as has been done previously (6). Large-amplitude bursts always occurred on both nerves simultaneously in brainstem-spinal cord preparation and were approximately 1 s in duration. Smaller amplitude bursts typically occurred on CNV but occasionally were present on CNX. These small bursts are referred to as “buccal bursts” and were qualitatively assessed throughout this study for presence or absence of treatment effects. Preparations exhibited nonrespiratory bursting (broad bursts that are presumably not of respiratory origin). Except for series 4, these bursts were not analyzed, as they do not reflect activity of the respiratory network.

Statistics

When comparing responses to severe hypoxia and metabolic inhibition across treatments, we chose to compare normalized responses (treatment frequency divided by the baseline frequency) to account for variability in the baseline burst frequency. These responses were analyzed with two-way ANOVA, with time and treatment as the main effects. Because each preparation had its activity normalized to baseline, the baseline value for each preparation was 1; thus, only normalized values during hypoxia and metabolic inhibitors were used in the ANOVA. When two-way ANOVA yielded a significant P-value for main effects or the interaction, a Holm–Sidak multiple-comparisons test was performed. The time till the large motor burst was analyzed using an unpaired t test when comparing two groups and one-way ANOVA when comparing three or more groups. Lung burst frequency responses to diazoxide and AICAR were performed using a paired t test. Proportions of occurrences between groups were compared using Fisher’s exact test. All analyses were performed using GraphPad Prism (GraphPad Prism 6.01; GraphPad Software, San Diego, CA). Significance was accepted when P < 0.05. Error bars represent means ± SEM.

RESULTS

The isolated brainstem-spinal cord preparation produces spontaneous rhythmic output created by central pattern-generating networks that drive activity of cranial motor nerves innervating respiratory muscles (Fig. 1A). We extracellularly recorded activity from the CNV and CNX roots and then rectified and integrated the raw signals for analysis. Large-amplitude, low-frequency bursts that occurred near-synchronously on both nerve roots represent neural activity that leads to airflow into or out of the lungs. We refer to this activity as a “lung burst.” Low-amplitude, high-frequency bursts that most occurred on CNV represent neuronal activity associated with ventilation of the buccal cavity, and we refer to such output as “buccal bursts” (25).

Figure 1.

Stereotyped motor response of the bullfrog respiratory network to severe hypoxia (0% O₂). A: simplified cartoon representation of the bullfrog brainstem-spinal cord preparation. In vitro brainstem-spinal cord preparations produce spontaneous rhythmic motor output associated with respiratory movements. Two types of motor patterns were present in most preparations: low-amplitude, high-frequency bursts associated with ventilation of the buccal cavity (buccal bursts) and high-amplitude, low-frequency bursts associated with ventilation of the lungs (lung bursts). Raw signals from the trigeminal (CN V) and vagus (CNX) nerves were rectified and integrated for analysis. B: example of the characteristic response to bath application of 0% O₂. During severe hypoxia, a large motor burst occurred, followed by cessation of both lung and buccal motor patterns. Preparations recovered when O₂ was restored.

Figure 1B shows an example of the stereotyped response to severe hypoxia (0% O₂) which, in our hands, involves three consistent features: 1) a large increase in the baseline motor activity upon exposure to severe hypoxia; we refer to this response as the “large motor burst,” 2) cessation of both lung and buccal activity, and 3) recovery of rhythmic motor output following reoxygenation.

We first compared responses of the respiratory network to severe hypoxia (0% O₂; n = 9) with that to NE (n = 6). We expected that if NE is a key driver of the response to severe hypoxia, the responses would be qualitatively similar. Figure 2A shows the response to 25 min of severe hypoxia. The large motor burst was followed by silence of lung activity. In eight of nine preparations, we detected buccal activity at baseline, and after the large motor burst during severe hypoxia, buccal activity stopped in each preparation. When we exposed the network to 5 μM NE, a concentration that produces a maximal effect but does not desensitize NE receptors (21), lung burst activity slowed and buccal activity always persisted (Fig. 2B). A large motor burst never occurred during exposure to NE, and both lung and buccal activities never stopped. Summary data of lung burst frequency are presented in Fig. 2C, showing that severe hypoxia and NE both reduced frequency of network output, but 0% O₂ produces a significantly larger decrease (two-way ANOVA, significant treatment effect, P = 0.0009; no time or interaction significance; Holm–Sidak multiple-comparisons test, P = 0.0015 at 15 min and P = 0.0006 at 25 min). Furthermore, the proportion of preparations exhibiting buccal activity during the last 5 min of severe hypoxia (0/8) or NE (6/6) were significantly different (P = 0.0003, Fisher’s exact test). Thus, severe hypoxia and NE cause qualitatively different responses.

Figure 2.

Norepinephrine and severe hypoxia elicit distinct motor responses. A: an example of the response to severe hypoxia, where a large motor burst is followed by silence of network output. B: illustrates the typical response to 5 μM norepinephrine (NE). In contrast to severe hypoxia, NE does not stop network output and does not produce a large motor burst. C: mean data showing that severe hypoxia (n = 9) and NE (n = 6) both reduce output from the respiratory network, but severe hypoxia produces a larger decrease. **P < 0.01 and ***P < 0.001 indicate statistical significance between severe hypoxia and NE at each time point (Holm–Sidak multiple comparisons test). Data are plotted as means, and error bars represent SE.

To address the possibility that increases in NE during hypoxia contribute to this stereotyped response, we performed saturation and blocker experiments. In the saturation experiment, we pretreated preparations with saturating concentrations of NE and then exposed them to 0% O₂. Figure 3A shows an example recording of a preparation pretreated with NE and then exposed to severe hypoxia. In the presence of NE, the salient features of the response to severe hypoxia persisted; a large motor burst occurred, lung activity was abolished, and buccal activity stopped. These trends are captured in the mean data, whereby lung activity decreased despite starting from an absolutely lower baseline in the presence of NE, a response similar to that observed without NE (Fig. 3B). In addition, the time until the large motor burst in severe hypoxia did not differ between preparations pretreated with NE or untreated control preparations (Fig. 3C; P = 0.9070). These data show that pretreatment with NE does not stop severe hypoxia from further reducing network output nor does it prevent the large motor burst. Next, we exposed preparations to hypoxia in the presence of an α1 adrenergic receptor antagonist, 500 nM prazosin, that has been shown to account for the depression of output by NE in this preparation (21). In the presence of prazosin, the hypoxic response persisted unabated (Fig. 4, A and B). To verify that prazosin blocks the response to NE, each preparation was allowed to recover in oxygenated aCSF and was then treated with 5 μM NE. Indeed, prazosin fully prevented the depression of lung output by NE (Fig. 4, A and C). These results show that NE and adrenergic signaling are neither necessary nor sufficient to account for the stereotyped response to severe hypoxia.

Figure 3.

Norepinephrine (NE) does not occlude the stereotyped response to severe hypoxia. A: a trace showing the experimental design and typical response to severe hypoxia in the presence of NE. NE reduced the frequency of the motor output and then severe hypoxia was applied to the preparation. In the presence of NE, severe hypoxia still produced a large motor burst and stopped lung and buccal motor output. B: mean data showing that severe hypoxia produces a similar frequency response of the network with (n = 6) or without NE (n = 9) (two-way ANOVA). C: mean data showing the time until the large motor burst during severe hypoxia with and without NE. NE does not significantly affect the time until hypoxia elicits a large motor burst (two-tailed unpaired t-test). Individual data points are shown for each group. Data in B are plotted as means, with error bars representing SEM. Raw values in C are shown as circles and represent responses for individual preparations.

Figure 4.

Block of adrenergic receptors that prevents depression by NE does not interfere with the response to severe hypoxia. A: a data trace from a single preparation exposed to 0% O₂ and then NE in the presence of 500 nM prazosin, an α1 adrenergic receptor antagonist. The left side of (A) shows the network response to 0% O₂ in the presence of 500 nM prazosin. Prazosin did not affect the depression caused by severe hypoxia. After severe hypoxia, this preparation was reoxygenated, allowed to recover, and then exposed to 5 µM NE to verify that prazosin reduces the response to NE. The right side of (A) shows that NE does not depress lung burst frequency in the presence of prazosin, confirming that NE depresses network output through the activation of α1 adrenergic receptors. B and C: mean data from n = 5 preparations that underwent the same experimental protocol, whereby (B) plots mean data for responses to severe hypoxia with and without prazosin and (C) shows responses to NE with and without prazosin. Data are plotted as means ± SE.

Because NE signaling failed to explain the response to severe hypoxia, we next tested the hypothesis that generalized energy failure causes these features. To test this hypothesis, we applied inhibitors of glycolysis and mitochondrial respiration in the presence of baseline O₂ tensions. To inhibit glycolysis, we used two different chemicals: 2-deoxy d-glucose (2-DG; n = 5) substituted for d-glucose to reversibly inhibit glycolysis and iodoacetic acid (IAA, 1 mM; n = 5) to irreversibly inhibit glycolysis (24). To chemically block mitochondrial respiration, we used a reversible complex IV inhibitor, cyanide (CN, 1 mM; n = 5). As observed in Fig. 5A1–4, chemical inhibition of different metabolic process mimicked the qualitative actions of severe hypoxia; each blocker caused a large motor burst and silenced both lung and buccal activities, like that observed during severe hypoxia. Mean data are shown in Fig. 5B. No significant differences occurred between lung burst frequency during exposure to any form of metabolic inhibition, and time until the large motor burst did not significantly differ among treatments (Fig. 5C). After washout of reversible metabolic inhibitors (2-DG, CN, and severe hypoxia), burst activity returned to baseline, and recovery values did not differ significantly from each other (2-DG, 118.1 ± 3.7% of baseline, CN, 106.1 ± 35.1% of baseline, and hypoxia, 146.5 ± 14.2% of baseline; P = 0.4997, one-way ANOVA). Activity never returned after washout of IAA, consistent with its action as an irreversible inhibitor of glycolysis in tissue preparations (24) and isolated nerves (26). These results show that four different mechanisms leading to reduced ATP synthesis each produce similar qualitative actions, suggesting that generalized energy failure accounts for the response of the network to severe hypoxia. Further, these results show that reduced O₂ tension, as previously hypothesized (6, 8), is not required to produce the characteristic response to severe hypoxia because 2-DG, IAA, and CN were applied under baseline oxygen conditions, implying that reduced ATP synthesis per se underpins the stereotyped response to severe hypoxia.

Figure 5.

Pharmacological inhibition of glycolysis and mitochondrial respiration closely mimics the motor response to severe hypoxia. A: typical responses of the in vitro preparation to severe hypoxia (A₁), inhibition of glycolysis (A₂, with 2-deoxy d-glucose, 2-DG and A₃, with iodoacetate, IAA), and mitochondrial respiration (A₄, sodium cyanide). B: the frequency of lung burst output did not differ significantly among different forms of metabolic inhibition (two-way ANOVA). C: the time until the large motor burst also did not differ significantly among each form of metabolic inhibition (one-way ANOVA). Individual data points are shown for each group. Data in B are plotted as means, with error bars representing SEM. Raw values in C are shown as circles and represent responses for individual preparations.

The previous experiments suggested that decreased ATP synthesis caused the stereotyped response to severe hypoxia. How might reduced ATP synthesis stop network output? We hypothesized that cellular energy sensors, K_ATP channels and AMP-dependent protein kinase (AMPK), could play a role. K_ATP channels are normally closed when the intracellular concentration of ATP is high and open when ATP falls, causing hyperpolarization of the membrane (27). Thus, activating K_ATP channels could silence the network during hypoxia. We exposed preparations to an activator of K_ATP channels, 600 μM diazoxide, to test whether activating K_ATP channels mimics the response to severe hypoxia and metabolic inhibition. Diazoxide did not significantly change lung burst frequency after 25 min of exposure (P = 0.1708, paired t test), suggesting that reduced ATP synthesis does not silence network output by increasing an ATP-sensitive K⁺ conductance (Fig. 6A1–2, and B). Issues with drug efficacy are unlikely to explain the lack of an effect because 600 μM reversibly increased the frequency of episodic large amplitude non-respiratory bursting in each preparation (Fig 6,C1–2, and D; P = 0.0093, repeated-measures ANOVA).

Figure 6.

Pharmacological activation of K_ATP channels does not resemble the response to severe hypoxia. A: example recordings of CN V output before (left, A₁) and the same preparation after a 25-min exposure to the K_ATP channel activator, 600 μM diazoxide (right, A₂). B: diazoxide did not significantly influence lung burst frequency (n = 6) (paired t-test). In contrast to lung burst frequency, diazoxide reversibly increased the frequency of episodic nonrespiratory bursts. Example recordings are shown in C_1–2 (arrows pointing to nonrespiratory bursts), and population data are presented in (D) (repeated-measures one-way ANOVA). **P < 0.01 and *P < 0.05. Data are plotted from individual preparations, where connected dots represent values before and after experimental treatments.

AMPK is a well-conserved energy sensor with variable effects on neuronal and synaptic processes in response to reduced ATP synthesis, one response being a reduction in neuronal firing rates (28). To test whether AMPK activation recapitulates network silence caused by severe hypoxia, we exposed preparations to an activator of AMPK, 1 mM AICAR. On average, AICAR reduced the frequency of lung burst activity by 25% (P = 0.0423; paired t test). Six of the seven preparations decreased activity during exposure to AICAR, but one increased (Fig. 7). Although activating AMPK reduced lung burst frequency, it does not account for the large motor burst and subsequent network silence caused by each form of metabolic inhibition. Collectively, our results suggest that a generalized reduction in ATP synthesis causes the characteristic response to severe hypoxia of this network in vitro.

Figure 7.

Pharmacological activation of AMP-dependent protein kinase (AMPK) does not resemble the response to severe hypoxia. A: example recordings of CN V output before (left, A₁) and the same preparation after a 25-min exposure to the AMPK activator, 1 mM AICAR (right, A₂). AICAR significantly reduced lung burst frequency (paired t test) but did not silence output and did not produce a large motor burst. *P < 0.05; n = 7. Data are plotted from individual preparations, where connected dots represent values before and after exposure to AICAR.

DISCUSSION

A depression of neuronal activity is often interpreted as a protective response to conserve brain energy status. Yet, neurons cannot operate normally when ATP synthesis fails to meet demands of circuit performance. Thus, the degree to which passive energetic limitations (i.e., failure) versus active modulatory mechanisms (i.e., protection) explain an arrest during hypoxia can be difficult to assess for a circuit or behavior. Here, we evaluated how well each possibility explains a shutdown of rhythmic output in the respiratory network of the severely hypoxic frog brainstem, a response thought to arise from increases in the inhibitory noradrenergic tone. Our results suggest that a mismatch between energy supply and demand causes the arrest, with little-to-no contribution from noradrenergic signaling and well-conserved energy sensors, AMPK and K_ATP channels. Taken together, our results highlight the need to consider energetic limitations on circuit performance before concluding that a neuromodulator shuts down network activity as an adaptation for energy conservation in the nervous system.

Severe Hypoxia and Norepinephrine Depress Respiratory Neural Activity through Different Mechanisms

Previous work concluded that severe hypoxia decreases output of the frog respiratory network through noradrenergic signaling. This interpretation was based on experiments that either blocked adrenergic receptors (6) or physically removed noradrenergic neurons from the preparation (11). In this study, we came to a different conclusion—that increases in the noradrenergic tone do not silence motor output during severe hypoxia. We make this claim for the following reasons: First, severe hypoxia and NE produced qualitatively different responses. After ∼10 min of severe hypoxia, a large motor burst was followed by silence of both lung and buccal network activities (Figs. 1 and 2A). In contrast, when we applied NE at a concentration that produces a maximal response (21), lung activity slowed and buccal activity continued (Fig. 2B). Next, we found that NE signaling was neither necessary nor sufficient to account for the hypoxic response in this preparation. If severe hypoxia caused the release of NE to reduce output from the network, we first expected pretreatment with NE would prevent further decreases by 0% O₂. This result did not occur (Fig. 3). This type of approach is referred to as an “occlusion” or “saturation” experiment and is often used to identify if a molecule or pathway plays a role in a given physiological response. For example, activation of GABA and adenosine receptors decreases neuronal excitability, which prevents further decreases by anoxia in turtle neurons (29, 30). In addition, in our hands, blocking receptors that mediate the response to NE did not affect the response to severe hypoxia (Fig. 4). The failure of NE to explain the response to severe hypoxia is consistent with some of the data from previous work that showed severe hypoxia and NE may reduce network output through distinct mechanisms. Specifically, GABA_A/glycine receptor blockers prevent the depression of lung output by NE but do not influence the response to severe hypoxia when assessed using aCSF with a similar composition to that used in this study (6). Thus, the sum of the data supports the notion that severe hypoxia stops, and NE slows, the network through distinct mechanisms.

Reducing ATP Synthesis Mimics the Shutdown during Severe Hypoxia

Our finding that NE signaling did not account for the response to severe hypoxia suggested that other factors cause the depression. Given that active circuits are costly to maintain, we hypothesized a failure of ATP synthesis during severe hypoxia is a logical cause of network silence. Our data support this hypothesis because inhibition of either glycolysis or mitochondrial respiration produced qualitatively similar outcomes as 0% O₂. Specifically, 2-DG, IAA, and CN, all in the presence of baseline O₂ tensions, produced a large motor burst followed by silence of lung and buccal activity like that observed during severe hypoxia (Fig. 5). Because different inhibitors of different parts of metabolism all produce the same outcome as severe hypoxia, these results argue strongly for a general energetic limitation on circuit performance as the cause of the arrest. This result was not entirely expected, as byproducts of glycolysis can modulate firing rates independent from aerobic metabolism (24), and reactive oxygen species that influence neural activity can vary based on the O₂ tension (22). Indeed, 0% O₂ arrests NMDA glutamate receptors in turtle neurons, but cyanide elicits both increases and decreases in NMDA receptor currents, demonstrating distinct outcomes with block of aerobic metabolism using different methods (31). Because four types of metabolic inhibition with different mechanisms of action all produced similar neural responses, our data imply that an acute reduction of ATP synthesis leads to the stereotyped response to severe hypoxia that we report here.

When ATP falls in the cell, energy sensors that reduce excitability could account for network depression. AMPK and K_ATP channels are two widely studied energy sensors that could serve this role. Contrary to this possibility, we found that activation of K_ATP channels and AMPK did not mimic the network response to metabolic failure. Specifically, activation of K_ATP channels did not affect the frequency of respiratory bursting, and AMPK activation reduced, but did not stop, network output. It is important to note that diazoxide used at the concentration here (600 µM) activates not only plasma membrane K_ATP channels (32, 33) but also mitochondrial K_ATP channels (mitoK_ATP) channels. Activation of mitoK_ATP channels is part of the pathway that actively arrests function of AMPA and NMDA glutamate receptors in cortical neurons from turtles during anoxia (34, 35). Thus, our findings that diazoxide did not reduce neuronal activity provides evidence that 1) plasma membrane K_ATP channels are not activated during anoxia to shut down the network and 2) a synaptic arrest process involving mitoK_ATP channels is probably not involved. Collectively, these results are consistent with the hypothesis that generalized energy failure, but not an active modulatory response, accounts for the response to 0% O₂ in the respiratory network of frogs.

What Limits Network Output during Energetic Stress?

If conserved energy sensors, noradrenergic signaling, and mechanisms used to arrest synapses in anoxia tolerant animals do not likely explain the network shutdown, then what mechanisms may explain this response? Other inhibitory modulators, such as adenosine and GABA, have been proposed to reduce neuronal activity during anoxia (29, 30). However, GABA and adenosine accumulate over hours in the anoxic frog brain (36), much longer than the ∼10 min it takes to stop network activity in our study (Figs. 1–4). Furthermore, GABA rises to a maximum of <10 µM after ∼6 h of anoxia (36), yet concentrations of 500 µM and 5 mM are required to decrease and silence lung bursts in this preparation, respectively (38). In addition, mechanisms that lead to cell death (e.g., anoxic depolarization, lethal Ca²⁺ influx, apoptosis, etc.) are unlikely to play a role because activity recovers within several minutes after ATP synthesis resumes. Because inhibition of ATP synthesis quickly and reversibly stops activity, we suggest that costly circuit functions requiring an ongoing supply of ATP may rapidly collapse to cause network failure. For example, blocking metabolism in rodent hippocampal neurons causes glutamatergic synaptic function to fail over tens of seconds because presynaptic vesicle cycling requires activity-dependent ATP synthesis (39, 40). In these studies, inhibition of ATP synthesis minimally affects the intracellular ATP concentration of presynaptic neurons at baseline and does not lead to apparent cell death, but rather triggers a large-scale synaptic dysfunction because the high demands of synaptic function cannot be met by concurrent increases in ATP production. Such a response is consistent with fast network failure observed here, as well as previous work in anoxic frogs that demonstrated a slow decrease in whole-brain [ATP] (36). In addition, high-frequency axonal firing requires constant synthesis of ATP and falls silent ∼5 min after inhibiting oxidative metabolism (20). Therefore, action potentials may fail to propagate when neurons can no longer maintain Na⁺ and K⁺ gradients through energy-consuming Na⁺-K⁻ ATPases. However, recent work estimates that action potential firing presents a relatively small energetic burden compared to synaptic transmission in many neuronal cell types (12, 14, 39). Thus, glutamatergic synapses may be prone to failure during short-term energetic stress in this rhythmically active network (15–17). While we recognize that less obvious modulatory processes that are sensitive to reduced ATP concentrations may contribute to stopping the network, our data support the idea that acute energetic limitations likely play a major role.

Perspectives and Significance

We conclude that a failure of ATP synthesis to match demands of circuit function drives an activity arrest in the respiratory network of American bullfrogs. This finding has two important implications, first, for interpreting plasticity in the effects of hypoxia on this network and, second, for understanding hypoxia tolerance of nervous systems in general. For the amphibian respiratory network, responsiveness to hypoxia changes in several contexts including metamorphosis (6), overwintering (7), and exposure to teratogens like nicotine and ethanol (41). Interpreting how hypoxia responsiveness changed mainly involved discussions of plasticity in hypoxia-sensitivity of the noradrenergic system. However, our data suggest this is unlikely to be the case in mature animals and imply that changes in ATP synthesis, availability, or use within the network have the potential to play a major role. Next, our results support a view that modulatory mechanisms involved in arresting activity are not an absolute requirement for vertebrate neurons to survive at least short bouts of severe hypoxia. Tolerating hypoxia in the adult vertebrate brain has long been proposed to involve a suite of modulatory and synaptic mechanisms that actively arrest spiking with the apparent goal of reducing energy expenditure (1, 42). However, plausible neuromodulatory and energy-sensing mechanisms do not explain neural responses to severe hypoxia here. Instead, blocking the main routes of ATP synthesis mimics the stereotyped response to severe hypoxia, and activity always recovers shortly after ATP synthesis resumed. We propose the most parsimonious explanation for these data is that high-cost neuronal functions simply fail because of insufficient energy supply; network output then stops, lowering the demands of housekeeping functions to permit survival. Thus, active circuits may fail owing to energetic limitations on performance but still survive moderate periods of severe hypoxia without engaging modulatory systems typically viewed as a hallmark of hypoxia tolerance in the vertebrate brain (1, 37). Future work must, therefore, consider the relative contributions of active modulatory inhibition and passive metabolic failure when interpreting how an intact circuit shuts down and then survives hypoxia or anoxia.

GRANTS

This work was supported by the National Institutes of Health Grant R15NS112920 and laboratory startup funds by University of North Carolina at Greensboro (to J. M. Santin); by the National Institutes of Health Maximizing Access to Research Careers (MARC) Undergraduate Student Training in Academic Research (U-STAR) program award to Daniel Herr (T34GM113860) (provided funds to S. Adams).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

J.M.S. conceived and designed research; S.A., T.Z., N.B., and J.M.S. performed experiments; S.A. and J.M.S. analyzed data; J.M.S. interpreted results of experiments; J.M.S. prepared figures; J.M.S. drafted manuscript; S.A. and J.M.S. edited and revised manuscript; S.A., T.Z., N.B., and J.M.S. approved final version of manuscript.