Plasticity in the Functional Properties of NMDA Receptors Improves Network Stability during Severe Energy Stress

Abstract

Brain energy stress leads to neuronal hyperexcitability followed by a rapid loss of function and cell death. In contrast, the frog brainstem switches into a state of extreme metabolic resilience that allows them to maintain motor function during hypoxia as they emerge from hibernation. NMDA receptors (NMDARs) are Ca²⁺-permeable glutamate receptors that contribute to the loss of homeostasis during hypoxia. Therefore, we hypothesized that hibernation leads to plasticity that reduces the role of NMDARs within neural networks to improve function during hypoxia. To test this, we assessed a circuit with a large involvement of NMDAR synapses, the brainstem respiratory network of female bullfrogs, Lithobates catesbeianus. Contrary to our expectations, hibernation did not alter the role of NMDARs in generating network output, nor did it affect the amplitude, kinetics, and hypoxia sensitivity of NMDAR currents. Instead, hibernation strongly reduced NMDAR Ca²⁺ permeability and enhanced desensitization during repetitive stimulation. Under severe hypoxia, the normal NMDAR profile caused network hyperexcitability within minutes, which was mitigated by blocking NMDARs. After hibernation, the modified complement of NMDARs protected against hyperexcitability, as disordered output did not occur for at least one hour in hypoxia. These findings uncover state-dependence in the plasticity of NMDARs, whereby multiple changes to receptor function improve neural performance during metabolic stress without interfering with their normal role during healthy conditions.

Significance Statement

NMDA-glutamate receptors play a major role in the loss of brain homeostasis during metabolic stress. In contrast, frogs have a remarkable capacity to use plasticity that improves brainstem circuit function from minutes to hours during hypoxia, likely as an adaptation to survive emergence from hibernation. We found this occurs, in part, through modification of NMDA receptors that renders them less permeable to Ca²⁺ and more likely to desensitize during high activity states. These NMDA receptor modifications do not influence normal network function but protect against hyperexcitability caused by hypoxia. This work points to endogenous plasticity mechanisms that improve network function during energy stress without altering activity when the brain is well oxygenated.

Introduction

Brain function requires high rates of ATP synthesis through oxidative metabolism (Bordone et al., 2019). The sudden loss of oxygen, and subsequent decline in ATP levels, triggers hyperexcitability that leads to ion dysregulation and cell death, termed “excitotoxicity” (Buck and Pamenter, 2018). A key contributor to excitotoxicity involves the activation of NMDA-glutamate receptors (NMDARs), causing pathological Ca²⁺ influx and saturation of intracellular Ca²⁺ buffering systems (White and Reynolds, 1995; Szydlowska and Tymianski, 2010). As rates of ATP synthesis wane, active transport of Ca²⁺ out of the cell becomes increasingly difficult, which culminates in cell death through reactive oxygen species, prodeath signaling pathways, and mitochondrial damage (Wu and Tymianski, 2018). Therefore, hyperexcitability caused by overstimulation of NMDARs and subsequent Ca²⁺ influx represents a critical step in the loss of homeostasis that drives neurological issues during hypoxia.

Unlike most mammals, some species experience variable oxygen partial pressures in their environments and have evolved strategies to survive brain hypoxia (Larson et al., 2014). Survival strategies often involve entry into a hypometabolic state associated with arrested synaptic transmission to conserve energy (Buck and Pamenter, 2018). However, some animals must remain active during hypoxia (Larson and Park, 2009; Czech-Damal et al., 2014; Larson et al., 2014). This presents the challenge of maintaining circuit function while avoiding hyperexcitability and excitotoxicity. An extreme example of this problem is embodied by the respiratory network of American bullfrogs. This network generates rhythmic output through mechanisms that involve AMPA and NMDA-glutamate receptors (Kottick et al., 2013) and, therefore, requires ongoing aerobic metabolism due to the high cost of synaptic transmission (Adams et al., 2021). However, for several months each year, frogs hibernate in ice-covered ponds without breathing air, using only skin gas exchange. As a consequence, blood oxygen falls dramatically, which may be as low as 1–3 mmHg (Tattersall and Ultsch, 2008). Low oxygen in this environment does not pose an immediate threat due to reduced metabolic rates in the cold (Tattersall and Boutilier, 1997). However, during emergence at warm temperatures, this life-sustaining network must restart activity on the background of severe hypoxia. An additional problem lies in the fact that frogs are not generally considered to be strongly hypoxia-tolerant, and low O₂ levels present during emergence cannot power brainstem circuits (Adams et al., 2021). To overcome these challenges, hibernation induces metabolic plasticity at synapses that improve network function during severe hypoxia from a few minutes to several hours (Bueschke et al., 2021a; Hu and Santin, 2022; Amaral-Silva and Santin, 2023). Therefore, hibernation in frogs provides insight into plasticity that shifts a typically hypoxia-intolerant circuit into a state that functions remarkably well during severe metabolic stress.

As in mammals, hypoxia in the frog brain induces network hyperexcitability that disrupts patterned output, followed by a swift loss of function (Adams et al., 2021; Bueschke et al., 2021a). Thus, this network must engage mechanisms that constrain excitability to maintain activity with a limited energy supply upon emergence from hibernation. Many hypoxia-tolerant vertebrates have low levels of NMDARs or suppress NMDARs in hypoxia to conserve ATP and avoid excitotoxicity (Bickler et al., 2000; Bickler and Buck, 2007; Wilkie et al., 2008). Therefore, we hypothesized that shifting from “hypoxia intolerance” to “functional hypoxia tolerance” involves reduced NMDAR function within the network. To test this hypothesis, we assessed the NMDAR tone of the respiratory network, NMDAR currents using whole-cell voltage-clamp, and single-cell quantitative PCR. In contrast to our hypothesis, hibernation did not influence the role of NMDARs in the network, current amplitude, kinetics, and hypoxia sensitivity. We instead found that hibernation decreased the Ca²⁺-permeability of NMDARs and enhanced desensitization, serving to reduce Ca²⁺ influx and lower depolarizing drive during high activity states. These modifications oppose the loss of homeostasis driven by the normal profile of NMDARs during hypoxia. Overall, we identified NMDAR plasticity that improves network activity during energy stress without influencing their contribution in well-oxygenated conditions.

Materials and Methods

Animal husbandry and ethical approval

The use of animals was approved by the Institutional Animal Care and Use Committee (IACUC) at the University of North Carolina at Greensboro (Protocol #19-006) and the University of Missouri (Protocol #39264). Female American bullfrogs (Lithobates catesbeianus) were ordered from Phrog Farm (Twin Falls, ID, USA). Frogs were randomly assigned to one of two groups: control or hibernation and were placed in designated plastic tubs containing dechlorinated water bubbled with room air. Control frogs were kept at ambient room temperatures (20°C) with access to wet and dry areas and fed pellets provided by Phrog Pharm once a week. Plastic tubs containing frogs assigned for aquatic overwintering were placed into low-temperature incubators (Thermo Fisher Scientific) or a walk-in environmental chamber and gradually reduced from 20 to 4°C over the course of 7 d. Once at overwintering temperatures, screens were placed below water level to ensure frogs used cutaneous respiration as the primary source of gas exchange, and they were kept at 4°C for 30 d before use. Frogs exhibit metabolic suppression at these temperatures, consistent with hibernation (Tattersall and Ultsch, 2008). Therefore, we refer to this group as “hibernation.” All frogs were kept under a 12 h light/dark cycle.

Dissection of brainstem

Frogs were deeply anesthetized with isoflurane (1 ml) in a sealed 1 L container until loss of toe-pinch response. They were then rapidly decapitated, and heads were submerged for dissection in chilled artificial cerebrospinal fluid (aCSF; in mM, 104 NaCl, 4 KCl, 1.4 MgCl₂, 7.5 D-glucose, 1 NaH₂PO₄, 40 NaHCO₃, and 2.5 CaCl₂; Bueschke et al., 2021a), which was bubbled with 98.5% O₂ and 1.5% CO₂ for oxygenation, resulting in aCSF with a pH of 7.9 ± 0.1. These CO₂ and pH values are normal for frogs at room temperature (Howell et al., 1970). The brainstem–spinal cord was rapidly exposed, and the forebrain was crushed. Brainstem–spinal cord nerve roots were carefully trimmed, allowing its excision from the cranium, which was followed by the dura membrane removal. Following all dissections, the preparations were held at room temperature (22 ± 1°C), and all experiments were performed at this temperature.

Extracellular nerve root recordings

For experiments that assessed motor output from the intact network, we assessed extracellular motor output from the vagus nerve rootlets, which innervates the glottal dilator muscle to control airflow in and out of the lung. Dissected preparations were pinned ventral side up in 6 ml Petri dishes coated with Sylgard 184 (Dow Corning). Brainstem–spinal cords were continuously superfused with oxygenated aCSF using a peristaltic pump (Watson Marlow). Population activity from the vagal nerve root (CNX) was recorded using a suction electrode attached to a fire-polished borosilicate glass pulled from a horizontal pipette puller (Sutter Instruments). Extracellular signals were amplified (×1,000) and filtered (low-pass, 1,000 Hz; high-pass, 100 Hz) using an A-M Systems 1700 amplifier and then digitized using PowerLab 8/35 (ADInstruments). The raw signal was integrated and rectified (100 ms τ) using LabChart data acquisition system (ADInstruments). 4 h following decapitation, a stable signal was recorded for ∼30 min, and the preparation was challenged with hypoxic aCSF (98.5% N₂ balanced with CO₂) or NMDAR antagonist, D-AP5 (50 µM). All preparations included in this study produced rhythmic motor output associated with lung breathing.

Motoneuron labeling and slice preparation

For experiments that assessed NMDA receptor currents in motoneurons from brain slices, we isolated the fourth root of the glossopharyngeal–vagus complex and attached a fire-polished borosilicate glass pipette to it. Tetramethylrhodamine dextran dye was then added to the tip of the pipette (Invitrogen) in contact to the nerve root for at least 2 h to diffuse to the soma. We then sliced the brainstem–spinal cord preparations at 300 µM using a vibratome (Technical Products International series 1000). During dye loading and slicing, the tissue was maintained in regular aCSF (described above).

NMDAR currents

Slices were superfused with an extracellular solution (76.5 mM NaCl, 2.2 mM KCl, 7.5 mM D-glucose, 10 mM HEPES, 300 µM CdCl₂, 20 mM TEA-Cl, 250 nM TTX, 10 µM DNQX) containing two different Ca²⁺ concentrations. High Ca²⁺ had 10 mM of CaCl₂ and low Ca²⁺ 1 mM of CaCl₂. Sucrose was added to maintain consistent osmolarity (330 mOsm). Thus, 60 mM was added to the high Ca²⁺ solution, and 89 mM was added to the low Ca²⁺ solution. MgCl₂ was excluded to prevent Mg²⁺ block of NMDARs. Extracellular solutions were oxygenated by bubbling 98.5% O₂ and 1.5% CO₂.

Labeled vagal neurons were approached by glass pipettes (2–4 MΩ resistance) using a micromanipulator (MP-285/ MPC-200, Sutter Instruments) attached to a head stage (CV203BU, Molecular Devices). Positive pressure was applied to the pipette while approaching the cell and quickly removed, gentle negative pressure was used to form a >1 GΩ, and the whole-cell access was obtained by breaking the seal with rapid negative pressure. We did not observe any obvious differences in cell viability and ability to obtain patch-clamp recordings in high and low Ca²⁺ solutions. The solution filling the patch pipette (in mM, 76.5 K-gluconate, 10 D-glucose, 10 HEPES, 1 Na₂-ATP, 0.1 Na₂-GTP, 2 MgCl₂, 30 TEA-Cl) was designed with Na⁺ and K⁺ concentrations equal and opposite to the extracellular solution to generate a reversal (E_rev) potential of ∼0 mV without Ca²⁺ (assuming no difference in ion selectivity between Na⁺ and K⁺) as seen previously (Jatzke et al., 2002).

To evoke NMDAR currents, we focally applied NMDA and glycine to the cell body. For this, a borosilicate pipette pulled with a tip diameter of ∼5 µm was filled with NMDA (1 mM) and glycine (10 µM; both from Hello Bio) dissolved in the extracellular solution. The pipette was driven by a Picospritzer II, and the solution was applied during a 10–20 ms puff onto the soma of labeled vagal motoneurons (General Valve Corporation) to activate NMDARs. Glycine is a coagonist of the NMDAR but did not elicit a response when applied alone in control or hibernation neurons.

NMDAR E_rev was recorded by a voltage-clamp step protocol with a Δ+5 mV step between −20 and 40 mV. In three recordings, E_rev was more hyperpolarized, and those cells were stepped from −40 to 20 mV in 5 mV increments (control, n = 16; hibernation, n = 16). NMDAR desensitization was assessed by puffing at a rate of 0.5 Hz for a total of 10 pulses at −20 mV (control, n = 12; hibernation, n = 14). Hypoxia was applied to brain slices by bubbling extracellular solution with 98.5% of N₂ and 1.5% of CO₂ for ∼5 min before recording. All desensitization protocols and single activation amplitude and kinetics experiments were performed on neurons held at −20 mV. All data were acquired in pClamp 11 software using Axopatch 200B amplifier and Axon Digidata 1550B digitizer (Molecular Devices).

NMDAR subunit electrophysiology

Vagal motoneurons were recorded using patch clamp as described above. For this protocol, the pipette solution contained (in mM) 110 K-gluconate, 2 MgCl₂, 10 HEPES, 1 Na₂-ATP, 0.1 Na₂-GTP, and 2.5 EGTA. Once gaining whole-cell access, the solution bathing the slice was changed from regular aCSF to 104 mM NaCl, 4 mM KCl, 7.5 mM D-glucose, 1 mM NaH₂PO₄, 40 mM NaHCO₃, 2.5 mM CaCl₂, 250 nM TTX, 10 µM DNQX, 3 µM glycine, and 2 µM strychnine. NMDA currents were elicited once every minute by focal application of NMDA (1 mM) and glycine (3 μM) diluted in the aCSF in 500 ms pulses. The cell was continuously monitored in voltage clamp (−60 mV), and after observing a stable current in three consecutive NMDA puffs (∼10–15 min of recording), we applied a specific NMDA subunit inhibitor in aCSF.

The subunit GluN2A was inhibited using 1 µM TCN 201 (Tocris Bioscience; Bettini et al., 2010; Edman et al., 2012); GluN2B was inhibited by 2 µM RO 25–6981 (Tocris Bioscience; Abrahamsson et al., 2017; France et al., 2017); GluN2C/GluN2D was inhibited using 10 µM QZN 46 (Tocris Bioscience; Hansen and Traynelis, 2011); and GluN3 was blocked by 30 µM TK30 [4-(2,4-dichlorobenzoyl)-1H-pyrrole-2-carboxylic acid; Santa Cruz Biotechnology; Kvist et al., 2013; Christian et al., 2021]. A time control experiment was performed recording NMDAR currents while maintaining the slice in the aCSF described above with no inhibitor added.

Electrophysiology data analysis

Extracellular motor output

Burst amplitude and frequency were determined by averaging amplitude of fictive breaths selected in a 5 min window prior to D-AP5 application. Amplitude and frequency in the presence of NMDAR antagonist were analyzed for 5 min after burst amplitude stabilized. Peak amplitude was determined using the Peak Analysis extension in LabChart (ADInstruments). To characterize chaotic bursting behavior during hypoxia, we analyzed the peak of the largest “nonrespiratory” motor burst that clearly disrupted patterned network output associated with respiratory activity and normalized it to the background nerve signal value at the baseline to allow comparisons across groups.

NMDAR Ca^2± permeability

Relative Ca²⁺ permeability was determined by the degree of shift in the NMDAR reversal potential (E_rev) based on the mean data of the population in different Ca²⁺ concentrations in extracellular recording solution. The underlying premise is that receptors that are permeable to Ca²⁺ have a depolarizing shift in E_rev, while Ca²⁺-impermeable channels do not (Jatzke et al., 2002). We evoked NMDAR currents using low (1 mM) and high (10 mM) Ca²⁺ and measured the peak current evoked by NMDA-glycine at each voltage step. We then plotted the current–voltage relationship and interpolated E_rev through the x-intercept of the line of best fit from 3–5 plot points near the intercept where I = 0 pA. The currents elicited by NMDA-glycine were small near E_rev and were unlikely to be affected by voltage errors due to the series resistance (R_s). However, the holding current at depolarized voltages was often substantial, even in the presence of TEA to block outward currents. Therefore, we corrected the holding voltage for R_s errors. For this, we measured R_s at each voltage step based on the peak of the transient current and corrected the holding voltage by the voltage error caused by R_s. Currents were measured using the Peak Analysis extension in LabChart. Averaging/decimation at 0.05–0.1 ms was applied to the traces prior to peak analysis to filter out high-frequency spontaneous synaptic activity. This level of filtering was chosen as it did not alter the amplitude of the NMDAR current.

Single-cell NMDAR currents and desensitization

Single-cell NMDAR currents were measured using similar methods as the NMDAR Ca²⁺ permeability, where peak currents induced by NMDA and glycine were assessed at −20 mV. Decay time constant was determined between 90 and 10% of the peak height and calculated using the Peak Analysis tool in LabChart. Desensitization was assessed by measuring the peak current evoked at each step. The baseline for each current in the series was taken as the recovered current after proceeding puff.

NMDAR subunit inhibition: The current amplitude was analyzed using peak analysis in LabChart. The average of the last three currents before inhibitor application was compared with currents recorded after 10 min of exposure to the drug. A set of time control experiments was also performed, where we puffed NMDA + glycine with no drugs for 10 min and then compared that amplitude to the first three stable currents. The use of a time control was critical, as we observed a slight “run up” of the agonist-evoked current over the 10 min protocol in most cells.

Single-cell real-time quantitative PCR

We used single-cell quantitative PCR (qPCR) to determine mRNA expression for NMDAR subunits using the same cell harvesting, RNA extraction, cDNA synthesis, and preamplification procedures detailed in Pellizzari et al. (2023).

PCR primers for open reading frames of the genes that code for NMDA glutamate receptor subunits, Grin1, Grin2a, Grin2b, Grin2c, Grin2d, Grin3a, and Grin3b, were designed from sequences found in the coding DNA sequence for L. catesbeianus (Tables 1, 2). For this, we used annotated amino acid sequences for GluN subunits from Rana temporaria as a query in the L. catesbeianus amino acid database. This search revealed peptide sequences with high amino acid sequence conservation. We then performed a reciprocal BLAST against the entire nonredundant protein database using hits from L. catesbeianus to verify the identity of the target. Accession numbers were then used to identify the open reading frame in the CDS to design PCR primers. Only Grin1, Grin2a, and Grin3a were identified in the bullfrog CDS, likely due to low coverage and/or poor assembly of the L. catesbeianus genome (Hammond et al., 2017). For Grin2b, Grin2c, Grin2d, and Grin3b, we found the regions of the coding sequence in R. temporaria (a species closely related to L. catesbeianus) and Nanorana parkeri (more distantly related frog species) with high similarity to design PCR primers. Our rationale was that close sequence identity at the nucleotide level between these two species (∼98%) would allow us to design primers for use in L. catesbeianus.

Table 1.

Primer sequences for SYBR Green qPCR assays

Target	Forward	Reverse	Efficiency
Grin2b	GAGTGGGAGGACAGAACTGC	TGGCAGACTTCACAGCGGAT	81%
Grin2c	GGTGCTGCAAGTGGAAACTG	GGATGTAGCGGGCACTCTTT	100%
Grin2d	TGTTCGCATGGGAGCATCTT	GCCTTTGGATCCTCAGCACT	95%
Grin3b	CCAGCGATCTTCCTTGGTGT	TGTCCAACCAGCCAGTGAAA	84%
18S rRNA	CAGGCCGGTCGCCTGAATAC	GGCCCCAGTTCCGAAAACCA	101%

Table 2.

Primer sequences for probe-based qPCR assays

Target	Forward primer	Reverse primer	Efficiency
Grin1	TGGTGGCGGTAATGTTGTATC	TCAGTGCATCCTCCTCTTCTTC	91%
Grin2a	TGACCCATTGCCGAAGTTG	GCATTCGGTCTTCATCAATGTCA	91%
Grin3a	GGCCAGACCAAGCACAAAG	CACAAAGGGTGGCTCAATCAAG	96%
	Prober sequences	Probe-quencher
Grin1	CAGGTTCAGTCCTTTTGGCCGGTT	FAM-BHQ1
Grin2a	ACCCAGTTATACTGGTCACTACTCAACCA	ROX-BHQ2
Grin3a	CAGAAGTCATTTCCAACAACCACCTCGA	VIC-BHQ1

Primer sets were validated with a series of four fourfold dilutions of brainstem cDNA. All primer sets used here produced efficiencies >80% (Tables 1, 2) and a single peak in the melt curve in a SYBR Green assay, suggesting the amplification of a single PCR product. As we observed in Pellizzari et al. (2023), some primer sets that produced one peak in the melt curve using bulk brain cDNA as the input material showed multiple peaks in the melt curve after single-cell preamplification. When this occurred, these primer sets were redesigned.

All neurons in this experiment were assessed for the expression of each of the seven Grin subtypes. Grin2b, Grin2c, Grin2d, and Grin3b were run using SYBR Green assays, and Grin1, Grin2a, and Grin3a were run in one multiplex assay. For SYBR Green assays, quantitative PCR was run in 10 µl reaction volumes containing 2.5 µM forward and reverse primers and followed the instructions of the 2× SYBR Green Mastermix (Applied Biosystems, Thermo Fisher Scientific). Assays were run on 96-well plates on an Applied Biosystems QuantStudio 3 (Applied Biosystems, Thermo Fisher Scientific) using the following cycling conditions according to the SYBR Green instructions: 50°C-2 m, 95°C-10 m, 95°C-15 s, and 60°C-1 m. Following 40 cycles of PCR (95°C-15 s, 60°C-1 m), melt curves for all PCR products were acquired by increasing the temperature in increments of 0.3°C for 5 s from 60–95°C. For multiplex assays, we ran triplexed probed-based assays. For this, we used the same primer concentration as described for SYBR Green assays, 312.5 nM reporter probes, and followed the instructions of the 5× PerfeCTa qPCR ToughMix mastermix (Quantabio). A 18 s ribosomal RNA was run to ensure the quality of the sample and for normalization of copy number to account for the possibility of different amounts on cDNA input and efficiency in the cDNA synthesis reaction.

Absolute quantitation of transcript abundance was estimated through copy number standard curves as previously described (Santin and Schulz, 2019). We normalized absolute copy number by a normalization factor using 18 s Cq values to account for the possibility of different amounts of harvested cytoplasm and efficiencies of the cDNA synthesis reaction across samples (Garcia et al., 2018). We ran qPCR for all Grin subunits on 19 control neurons and 20 hibernation neurons. In some neurons, Grin subunits appeared to be absent, or to be very lowly expressed. However, it is also possible a lack of detection represents a false negative due to stochasticity in the cDNA synthesis reaction due low RNA input quantities associated with the single cell. Thus, we included Grin genes in analysis that had C_q values <29.5 after preamplification. The only exception to this was for Grin2D. The abundance was consistently low for most samples; therefore, we included all data points in the analysis for Grin2D.

Statistical analysis

Statistical analysis was performed using GraphPad Prism 9 (GraphPad Software). Comparisons between independent two groups were carried out with a two-tailed unpaired t test or a Welch’s test if standard deviations differed. If data did not follow a normal distribution, a Mann–Whitney U test was performed. With three or more groups, we used a one-way ANOVA. In experiments with a “before-after” design, we used a two-tailed paired t test. In experiments with two main effects, we used a two-way ANOVA. One-way and two-way ANOVAs were followed up with Holm–Sidak multiple-comparisons test when appropriate. Individual data points were presented in addition to mean ± SD or with box and whisker plots; on the latter, boxes represent the interquartile range, and whiskers indicate maximum and minimum values in the dataset. Significance was accepted when p < 0.05.

Results

Function of the brainstem respiratory network in severe hypoxia increases from minutes to hours after animals emerge from hibernation (Bueschke et al., 2021a; Amaral-Silva and Santin, 2023). To determine if reductions in NMDARs play a role in this response, we assessed the NMDAR tone in in vitro brainstem–spinal cord preparations. This preparation produces rhythmic output associated with breathing that can be recorded from the cranial nerve X rootlet (Fig. 1A). NMDARs are involved in transmitting synaptic input from the respiratory rhythm generator to the motor pools (Kottick et al., 2013). If the contribution of NMDAR was reduced following hibernation, we expected to observe a lower sensitivity to the block of NMDARs using D-AP5 relative to controls. D-AP5 reduced the amplitude of the motor output by ∼50%, corroborating previous results demonstrating that NMDAR transmission plays a large role in recruiting motoneurons to the population burst (Kottick et al., 2013). However, sensitivity of the motor amplitude to D-AP5 did not change after hibernation (Fig. 1B,C; t₍₁₃₎= 0.9325; p = 0.37; unpaired t test). In addition, sensitivity of respiratory burst frequency to D-AP5, which provides an assessment of rhythm generator function, did not change after hibernation (Fig. 1D; t₍₁₃₎= 1.699; p = 0.11; unpaired t test). These results show that the overall contribution of NMDARs to network function is unchanged following hibernation.

Figure 1.

Hibernation does not alter the NMDAR tone of respiratory motor output. A, Schematic of the in vitro brainstem preparation. Raw motor output associated with breathing can be recorded from the cut cranial nerve X rootlet (CNX). Raw signals are then rectified and integrated for analysis. B, Representative respiratory motor bursts in baseline conditions and after block of NMDA-glutamate receptors with 50 µM D-AP5. Control animals are shown on the top and hibernation is shown on the bottom. C, Box and whisker plot comparing burst amplitude in D-AP5 relative to control between controls (left, green; n = 8) and hibernation (right, blue; n = 7) brainstems, showing no significant change in D-AP5 sensitivity (unpaired t test). D, Box and whisker plots comparing burst frequency changes in response to D-AP5. There was no significant difference between the change in burst frequency induced by D-AP5 across groups. Dots represent each data point of individual experiments.

Although the contribution of NMDAR to network function did not change, we hypothesized that functional properties of the NMDAR may play a role in improving robustness in hypoxia. For example, plasticity in current amplitude or kinetics (i.e., smaller currents with fast deactivation times) may lower ionic fluxes through NMDARs and, therefore, play a role in energy conservation. Moreover, some hypoxia-tolerant species contain NMDARs that are inhibited by hypoxia, offsetting excitotoxicity during energetic stress (Bickler et al., 2000). Therefore, we measured NMDAR currents in identified vagal motoneurons from brain slices using whole-cell voltage clamp in response to focal application of the NMDAR coagonists, 1 mM NMDA and 10 µM glycine (Fig. 2A). Current amplitude and decay time constants at −20 mV did not change in response to hibernation (Fig. 2B–D; amplitude: t₍₂₄₎= 1.502, p = 0.1461, unpaired t test; deactivation time constant: t₍₂₄₎= 0.4516, p = 0.6556, unpaired t test). In addition, superfusion of aCSF bubbled with a hypoxic gas mixture (0% O₂) for 5 min did not consistently alter the amplitude or decay time in controls (Fig. 2E; amplitude: t₍₄₎= 1.455, p = 0.2193, paired t test; deactivation time constant: t₍₄₎ = 1.980, p = 0.1189, paired t test) or following hibernation (Fig. 2F; amplitude: t₍₃₎= 0.2362, p = 0.8285, paired t test; deactivation time constant: t₍₃₎= 0.1515, p = 0.8892, paired t test). Therefore, hibernation does not alter whole-cell NMDAR currents, as well as their decay kinetics and sensitivity to acute hypoxia.

Figure 2.

Hibernation does not alter the whole-cell NMDAR current amplitude, decay kinetics, and hypoxia sensitivity in identified respiratory motoneurons. A, Schematic illustrating the backfill labeling process to identify motoneurons and the setup to focally apply NMDAR agonists (1 mM NMDA and 10 µM glycine) to the cell body. B, Traces showing typical responses to focal application of NMDA and glycine. C, D, NMDAR current amplitude and decay time constant are unaltered between control (n = 11 neurons from N = 5 animals) and hibernation (n = 13 neurons from N = 4 animals) neurons following unpaired t tests. Boxes show interquartile range and whiskers represent min and max values with dots showing individual data points. E, F, Individual NMDAR amplitude and decay time constant values before and after ∼5 min of hypoxia (0% O₂). Paired t test results suggest hypoxia does not change either NMDAR variable in control (n = 5 neurons from N = 4 animals) or hibernation (n = 4 neurons from N = 4 animals) neurons. Line represents the “before” and “after” response for individual neurons in response to hypoxia.

Ca²⁺ influx through NMDARs plays a role in network dysfunction during hypoxia (Szydlowska and Tymianski, 2010). Thus, we assessed the Ca²⁺ permeability of NMDARs before and after hibernation, as Ca²⁺ selectivity may increase or decrease without obvious changes in the whole-cell current amplitude and kinetics (Skeberdis et al., 2006; Murphy et al., 2014). For this, we measured the reversal potential of the NMDAR current (E_rev) in low Ca²⁺ (1 mM) and high Ca²⁺ (10 mM). Changes in E_rev during exposure to high Ca²⁺ concentrations provide a way to assess the relative Ca²⁺ permeability of ligand-gated ion channels (Jatzke et al., 2002). Consistent with the canonical role of NMDAR as a Ca²⁺ permeable channel, raising extracellular Ca²⁺ depolarized E_rev of control neurons (Fig. 3A, top). Divalent cations can adhere to the outer edge of cell membranes and alter the surface charge, making the apparent voltage at the membrane more depolarized with increasing concentrations of divalent cations, termed “charge screening” (Hille, 2001). This is important to consider, given that we found a depolarizing shift in E_rev in response to high Ca²⁺. However, charge screening is unlikely to affect our interpretation since E_rev shows a <2 mV difference between 1 and 10 mM Ca²⁺ on Ca²⁺-impermeable kainate receptors (Jatzke et al., 2002). Thus, most of the ∼13 mV depolarization in E_rev we observe between low and high Ca²⁺ in control neurons likely reflects its true Ca²⁺ permeability and not changes in the surface charge. After hibernation, the change in E_rev during high Ca²⁺ was far smaller (Fig. 3A bottom). Indeed, there was a significant interaction between Ca²⁺ concentration and group on E_rev (Fig. 3C; F_(1,60)= 7.735; p = 0.0072; two-way ANOVA), with a strong increase in E_rev by high Ca²⁺ in the control group and no significant change in E_rev in high Ca²⁺ after hibernation. Therefore, hibernation leads to a reduction in the Ca²⁺ permeability of the NMDAR.

Figure 3.

Hibernation decreases Ca²⁺ permeability of NMDARs. A, B, Raw whole-cell voltage-clamp traces held at a range of voltages (−20 to +40 mV, Δ5 mV) responding to NMDAR activation via focal application of NMDAR agonists. Control (A) is shown on the top in green, and hibernation (B) is shown in blue. Reversal potentials (E_rev; voltage at which current equals zero) in these example traces are highlighted to visualize the typical reversal potentials in 1 mM and 10 mM Ca²⁺ from control and hibernation motoneurons. C, Results of NMDAR E_rev (n = 16 neurons, all groups) show significance in two-way ANOVA interaction (p = 0.0072), indicating that hibernation affects the response to Ca²⁺. Pairwise comparisons reveal a significant depolarization of E_rev by 10 mM Ca²⁺ in control neurons but not hibernation neurons. We used four animals for hibernation in low Ca²⁺ and five animals for all other groups. Dots represent each data point of individual cells in each group, with boxes representing interquartile range and whiskers displaying min and max E_rev values. E_rev values shown in this plot are corrected for calculated series resistance errors based on the holding current. ****signifies p < 0.0001.

We next investigated the dynamic properties of NMDAR currents. Desensitization describes the degree to which a receptor loses responsiveness to the agonist in response to continued exposure, which may reflect a mechanism to dynamically reduce NMDAR currents when the network is in a high activity state. For this, we simulated NMDAR receptor activation that likely occurs during large phasic motor activation that disrupts normal rhythmic output in severe hypoxia (Fig. 4A; Adams et al., 2021). Thus, we puffed NMDAR agonists every 2 s for a total of 10 pulses per neuron to assess desensitization in a physiologically relevant way (Fig. 4B). We observed a moderate degree of desensitization in controls, whereby the 10th pulse in the series produced a current that was ∼60% of the baseline amplitude. NMDARs from hibernators also desensitized, but the final puff elicited a current that was significantly smaller than controls, at ∼20% of the initial value (Fig. 4C; t₍₂₄₎= 2.946; p = 0.0071; unpaired t test). In addition, a two-way ANOVA for relative current amplitude across all puffs shows a main effect of group, indicating that currents from hibernators were more desensitized across the entire experimental protocol (Fig. 4D; F_(1,24)= 7.084; p = 0.0137; two-way ANOVA). Therefore, NMDAR currents not only become less permeable to Ca²⁺ after hibernation, but they also pass less current during repetitive activation that otherwise causes network output to lose homeostasis.

Figure 4.

Hibernation enhances NMDAR desensitization in response to repetitive activation. A, Example recording of population motor output from a control brainstem preparation ∼5 min into hypoxia. This illustrates the transition from normal patterned output to chaotic, higher-frequency output with large amplitude “non-respiratory” bursting activity that occurs during hypoxia. B, Representative traces of vagal motoneuron response to pulses of NMDAR/glycine onto the soma at 0.5 Hz. Controls under a moderate degree of desensitization (green, top), while hibernation neurons (bottom, blue) tended to have smaller current amplitudes by the end of the experiment. C, Final (10th) pulse current amplitude relative to initial pulse. Hibernation motoneurons (n = 14 neuron from N = 4 animals) had a significantly lower final amplitude than controls (n = 12 neurons, N = 5 animals), suggesting that NMDARs after hibernation are more susceptible to desensitization. D, Plot showing all data from the experimental series. Two-way ANOVA reveals a main effect of group and puff number. A significant group effect indicates that NMDAR currents throughout the experimental series were overall smaller after hibernation, consistent with enhanced desensitization. Dots represent each data point from individual cells, boxes represent the interquartile range, and lines display mean with min and max values.

NMDARs are heterotetramers composed of an obligatory GluN1 subunit and a combination of GluN2A-D and/or GluN3A-B protein subunits (Paoletti et al., 2013). To gain insight into the NMDAR subunits that correspond to these physiological changes, we used inhibitors for the subunits GluN2A, GluN2B, GluN2C/GluN2D, and GluN3 to uncover the physiological participation of NMDA subunits in neurons from controls and hibernators. In controls, we observed a significant inhibition of the total NMDAR current when the GluN2B and GluN2C/GluN2D subunits were inhibited compared to time control (where time control is exposure to NMDA + glycine without any inhibitors). This resulted in a decrease of 44% for GluN2B (t₍₁₇₎= 4.915; p = 0.0001; t test) and 29% for GluN2C/D (t₍₁₅₎= 2.916; p = 0.0106; t test). It is important to appreciate that most time controls underwent a slight increase in current amplitude during the experimental protocol (Fig. 5A); thus, the decreases induced by each drug represent an underestimation of the true contribution of the subunit to the total current. After hibernation, GluN2B maintained its participation in the NMDA current, showing a decrease relative to time control when inhibited (t₍₂₂₎= 3.103; p = 0.0052; t test), which did not differ from control neurons (t₍₂₀₎= 1.427; p = 0.1689; t test; Fig. 5C). However, unlike controls, GluN2C/GluN2D no longer contributed to the NMDA current after hibernation (t₍₁₉₎= 0.2833; p = 0.7800; t test compared with time control), showing a significant difference from the control group (t₍₁₅₎= 2.612; p = 0.0196; t test; Fig. 5D). The subunits GluN2A (Fig. 5B) and GluN3 (Fig. 5E) did not seem to have a consistent functional contribution to the NMDAR current in control or overwintered frogs. Therefore, the GluN2B subunit appears to be the main GluN2 subunit generating the puff-evoked NMDA current, while GluN2C/D plays a smaller role and is lost after hibernation.

Figure 5.

Physiological contributions from NMDAR subunits and changes after hibernation. Change in NMDAR current amplitude after 10 min of recording (A, time control), inhibition of GluN2A (B, 1 µM TCN 201), GluN2B (C, 2 µM RO 25-6981), GluN2C/GluN2D (D, 10 µM QZN 46), and GluN3 (E, 30 µM TK30). Left, Change in current amplitude (expressed as % of baseline) in control neurons compared with after hibernation. Right, Example traces of a baseline current overlapped with a current after 10 min of recording (A) or after 10 min of the inhibitor exposure (B–E) in neurons from frogs in control and hibernation conditions. The “after drug” or time controls traces are slightly offset from their respective baseline recordings to enhance visibility. Number of cells (n) and frogs (N) used in these experiments: Time control in control n = 10, N = 4, in hibernation n = 11, N = 7; GluN2A inhibition in control n = 9, N = 3, in hibernation n = 9, N = 6; GluN2B inhibition in control n = 9, N = 3, in hibernation n = 13, N = 6; GluN2C/GluN2D inhibition in control n = 7, N = 3, in hibernation n = 10, N = 6; GluN3 inhibition in control n = 7, N = 2, in hibernation n = 10, N = 9. Scale bars represent 200 pA and 5 s in each trace.

To understand the potential for transcriptional control of NMDAR subunits, we followed up physiology experiments with single-cell qPCR to measure the mRNA expression of all seven NMDAR subunits that encode the NMDAR (the Grin gene family). These data are summarized in Figure 6. Consistent with the dominant contribution of GluN2B to the total NMDAR current, Grin2B appeared to be the most abundant Grin transcript for both controls and hibernators, along with Grin1 that codes for the obligatory GluN1 subunit. Pharmacologically, QZN 46 is selective for both GluN2C and GluN2D. However, most neurons lacked mRNA expression of Grin2D (Fig. 6), indicating the GluN2C is the main functional subunit blocked by QZN 46 in control neurons. Consistent with our functional results, after hibernation we observed a significant decrease in the mean abundance Grin2C (p = 0.0003; Mann–Whitney U test). Surprisingly, although we did not observe a contribution from the GluN3 subunit to the NMDAR in either group, we observed mRNA expression of both Grin3A and Grin3B, and each of these subunits decreased after hibernation (Grin3A: t₍₃₄₎= 2.920, p = 0.0062, unpaired t test; Grin3B: t_(21.62)= 2.597, p = 0.0166, Welch’s t test). These data suggest that a transcriptional program influences the NMDAR subunit composition in concert with modification of whole-cell current properties.

Figure 6.

Single-cell qPCR shows that hibernation changes in the NMDAR subunit profile in motoneurons. Single-cell mRNA abundances for all 7 Grin family members. The dominant subunits were Grin1, which encodes the obligatory NMDAR subunit, GluN1, as well Grin2B and Grin3B. Hibernation led to decreases in the abundance of Grin2C, Grin3A, and Grin3B. Sample sizes for each gene are as follows for control and hibernation. Grin1, n = 19 control, n = 20 hibernation; Grin2A, n = 16 control, n = 16 hibernation; Grin2B, n = 18 control, n = 20 hibernation; Grin2C, n = 19 control, n = 20 hibernation; Grin2D, n = 19 control, n = 20 hibernation; Grin3A, n = 18 control, n = 18 hibernation; Grin3B, n = 19 control, n = 20 hibernation. Control cells came from N = 6 animals and hibernation cells came from N = 5 animals. *p < 0.05; **p < 0.01; ***p < 0.001.

What are the functional impacts of NMDAR modifications on the ability of the network to function during hypoxia? In control brainstems, the first several minutes of hypoxia leads to large, uncoordinated output that interferes with the normal patterned activity of the respiratory network (Fig. 7A₁). This type of chaotic activity is thought to arise, at least in part, due to motor hyperexcitability induced by energy stress, as it does not occur when the brainstem is well oxygenated (Adams et al., 2021; Bueschke et al., 2021a). To determine the extent to which NMDARs play a role in this response, we exposed a group of brainstem preparations to D-AP5 and then measured the degree of large amplitude output that disrupted normal activity shortly after the onset of hypoxia. In the presence of D-AP5 (Fig. 7A₂), the mean amplitude of this chaotic “non-respiratory” activity was strongly blunted relative to controls. In addition, the small degree of disruption was often not sufficient to perturb patterned output during hypoxia as shown in Fig. 7A₂. These results show that the normal profile of NMDARs (with a permeability to Ca²⁺ and relatively less desensitization) contribute to chaotic output that disrupts normal function of the network during hypoxia.

Figure 7.

Modified NMDAR profile after hibernation protects against network hyperexcitability during hypoxia. Extracellular recordings of motor nerve output from the brainstem–spinal cord preparation during hypoxia. A₁, Typical response to hypoxia in controls, whereby large amplitude bursting disrupts patterned output from the respiratory network. A₂, Motor output during hypoxia in a preparation that was pretreated with D-AP5 to block NMDARs. In the presence of D-AP5, large amplitude bursting is less severe and often does not disrupt the respiratory rhythm. This indicates the normal profile of NMDARs contributes to disruptive motor activity in hypoxia. B₁, Patterned motor output from a hibernation preparation over the same timescale as controls. Despite intact NMDARs, there is no observable disruptive motor output. B₂, A hibernation preparation pretreated with D-AP5, also showing disrupted motor output. C, Summary data of the degree of large amplitude nonrespiratory bursting from n = 11 controls, n = 9 control + D-AP5, n = 8 hibernation, and n = 7 hibernation + D-AP5 analyzed by a one-way ANOVA. Non-respiratory bursting was smaller on average after the application of D-AP5, and this was not statistically different from hibernation preparations with or without D-AP5.

As disruptive motor activity during hypoxia arose largely due to NMDARs, this type of activity provided a sensitive indicator of how the modified NMDAR profile influenced network excitability in hypoxia. Strikingly, the chaotic, large amplitude, non-respiratory bursting in hypoxia that was caused largely by NMDARs did not occur in any hibernation preparation (Fig. 7B₁). We also applied D-AP5 to a group of hibernation preparations. The addition of D-AP5 did not influence network function during hypoxia, as there was no disruptive activity to suppress (Fig. 7B₂). These results are summarized in Figure 7C by a one-way ANOVA for control, control + D-AP5, hibernation, and hibernation + D-AP5 (F_(3,31)= 15.25; p < 0.0001; one-way ANOVA). Post hoc tests reveal that blocking NMDARs in control preparations suppresses the large non-respiratory bursting in hypoxia (p = 0.0002; Holm–Sidak multiple-comparisons test). This is not different than the degree of large amplitude bursting during hypoxia from hibernation preparations without (p = 0.6824; Holm–Sidak multiple-comparisons test) or with D-AP5 (p = 0.6824; Holm–Sidak multiple-comparisons test). These results show that control preparations without functional NMDARs (because they have been blocked by D-AP5) behave similar to circuits with intact NMDARs after hibernation when they have become less permeable to Ca²⁺ and desensitize more strongly. Thus, modifications in NMDAR function act in a way to prevent hyperexcitable network states during metabolic stress while maintaining their role in network output under well-oxygenated conditions.

Discussion

During energy stress, NMDAR activation disrupts circuit output in a wide range of species, including amphibians (Fig. 7). To emerge from hibernation in an ice-covered pond, frogs must restart critical behaviors, including breathing, on the background of extraordinarily low oxygen levels (Ultsch et al., 2004; Tattersall and Ultsch, 2008). The challenge lies in the fact that most frogs are not generally considered to be a strongly “hypoxia-tolerant” species. Metabolically, a suite of plasticity mechanisms arise during hibernation that improve the fuel supply needed to operate brainstem synapses under anaerobic conditions (Bueschke et al., 2021a; Hu and Santin, 2022; Amaral-Silva and Santin, 2023). Here, we tested the hypothesis that these animals also use physiological mechanisms to avoid hyperexcitable network states induced by hypoxia. We identified two key modifications to NMDARs that offset hyperexcitable output in hypoxia: reduced Ca²⁺ permeability (Fig. 3) and enhanced desensitization (Fig. 4).

These modifications are consistent with plasticity that acts to maintain network excitability and reduce metabolic demands in hypoxia. First, enhancing desensitization reduces depolarizing drive during repetitive stimulation that normally occurs during hypoxia (Fig. 4). As NMDARs contribute to chaotic bursting that disrupts patterned network output (Figs. 4A, 7A), stronger desensitization during high activity states likely acts as a potent brake on excitability. Second, lowering Ca²⁺ permeability likely incurs a cost savings. Here, roughly 50% of the population motor output that drives breathing is generated through NMDARs (Fig. 1), which means for every breath the animal takes, increases in intracellular Ca²⁺ from NMDARs must be cleared to maintain ion homeostasis. Although the regulation of intracellular Ca²⁺ is multifaceted, the plasma membrane Ca²⁺ ATPase plays a large role in maintaining Ca²⁺ homeostasis in active neurons (Schmidt et al., 2017; Malci et al., 2022). This pump likely consumes more than three times the ATP as the Na⁺/K⁺ ATPase, as it extrudes 1 Ca²⁺ ion per ATP hydrolyzed and transports Ca²⁺ against a larger electrochemical gradient compared with Na⁺ and K⁺. Thus, in addition to minimizing pathways for Ca²⁺-induced excitotoxicity (Szydlowska and Tymianski, 2010), lowering the Ca²⁺ permeability of NMDAR may also reduce the metabolic burden of Ca²⁺ regulation, allowing neurons to allocate energy to other processes required to maintain homeostasis.

Interestingly, these modifications do not alter the normal role of the NMDARs in generating motor output under well-oxygenated conditions (Fig. 1). One reason for this may be that NMDAR currents in this study were measured at the cell body. Thus, the plasticity in NMDAR function we observe may be localized to extrasynaptic regions, which are well known to elicit the pathological actions of NMDARs during hypoxia and ischemia (Tu et al., 2010; Zhou et al., 2013). It is not yet known if hibernation influences NMDARs at respiratory-related synapses. However, if hibernation alters the properties of synaptic NMDARs, reduced Ca²⁺ permeability and enhanced desensitization likely have little influence on network output since these modifications do not alter the whole-cell current amplitude and kinetics over the timescale of individual networks population bursts (<1 s burst every ∼10 s; Fig. 1A). Altogether, NMDAR plasticity appears to contribute to a state of resilience during metabolic stress without influencing normal functioning of the network under healthy conditions.

What cellular mechanisms influence the functional changes we observe? NMDAR receptors are composed of an obligatory GluN1 subunit and GluN2A-D and GluN3A-B subunits, which determine receptor properties (Paoletti et al., 2013). Heterologous expression of GluN3 with GluN1 renders the receptor impermeable to Ca²⁺ (Chatterton et al., 2002). Although single-cell qPCR results show altered expression of mRNA that code for the GluN3 subunit, functional measurements indicate that it does not contribute to the NMDAR current. Thus, GluN3 is not responsible for decreases in Ca²⁺ permeability. Interestingly, Ca²⁺ permeability of NMDAR containing GluN2B is strongly increased by phosphorylation (Skeberdis et al., 2006; Murphy et al., 2014). Indeed, inhibiting protein kinase A decreases Ca²⁺ permeability without changing the amplitude and kinetics of the postsynaptic current carried by NMDARs (Skeberdis et al., 2006; Fig. 5), similar to our results showing similar current properties despite different Ca²⁺ permeabilities. As GluN2B had the greatest contribution (Figs. 5, 6), we speculate that alterations in the phosphorylation of GluN2B may regulate Ca²⁺ permeability. For desensitization, many factors influence this property, including subunit composition (Vicini et al., 1998), protein binding partners (Sornarajah et al., 2008), and intracellular signaling pathways (Alagarsamy et al., 1999). Most relevant to our results, GluN1/GluN2C heterodimers undergo weak desensitization (Dravid et al., 2008; Alsaloum et al., 2016). We observed a decrease in Grin2C mRNA and a reduced contribution of GluN2C to the whole-cell current after hibernation. These results suggest that transcriptional control of GluN2C may shift neurons to favor subunit combinations with greater degrees of desensitization, such as GluN1/GluN2B (Vicini et al., 1998). As a caveat, we acknowledge that the whole-cell NMDAR current may represent the sum of many potential combinations of subunits. Although we did not find significant activity of GluN3 and GluN2A, some individual cells did appear to respond to their inhibitors, suggesting that variable combinations of NMDAR subunits compose the total whole-cell current. In addition, receptor expression may vary depending on the location within the cell (e.g., synaptic vs extrasynaptic). This complicates the connection between single-cell mRNA abundances and NMDAR function measured at the cell body. Nevertheless, combining physiological and molecular approaches points to mechanisms that alter the functional properties of the NMDARs to improve function during energy stress.

The frog brainstem has a remarkable ability to improve its function during hypoxia and ischemia after hibernation (Bueschke et al., 2021a). This response involves enhancing anaerobic glucose metabolism at active synapses (Bueschke et al., 2021a; Amaral-Silva and Santin, 2023). Dominant hypotheses for neural circuit evolution posit that natural selection acts on the efficiency of synapses, optimizing the ratio of information transfer to ATP consumption (Harris et al., 2015; Quintela-López et al., 2022). However, synapses in this network can switch into a state that supports activity for hours with ∼1/15th the amount of ATP through adjustments in synaptic metabolism that maximizes glycolytic ATP production (Amaral-Silva and Santin, 2023). Results here suggest that physiological modifications help to lower the cost of network activity to match the dramatically reduced rate of energy production while maintaining seemingly normal output. Given that this circuit makes physiological and energetic modifications to run on very little ATP, these results demonstrate some neural circuits may reside far from “optimal” efficiency. This begs the question of why animals would suppress such a state until it is critical for survival (e.g., emerging from an ice-covered pond after months of hibernation).

Although our results show strong support for NMDAR plasticity as a potential energy-saving mechanism, they also hint at a potential cost of maintaining an “ultra-high” efficiency network state. NMDARs play a key role in plasticity, whereby Ca²⁺ influx through the receptor potentiates synaptic strength in a wide range of systems, including the respiratory motoneurons in this species (Bueschke et al., 2021b). NMDAR modifications we observe likely bring about energy savings but have features that appear to be incompatible with NMDAR-dependent plasticity: strongly reduced Ca²⁺ permeability and greatly enhanced desensitization which quickly reduces channel activity upon repetitive stimulation. Indeed, Ca²⁺ impermeable NMDARs cause memory impairments and reduce long-term potentiation in mice (Conde-Dusman et al., 2021; Hurley et al., 2022). Therefore, we suggest a trade-off exists between metabolic robustness and the capacity for plasticity within circuits. It seems sensible for an animal to favor an extreme degree of energetic resilience to restart critical behaviors in severely hypoxic conditions. However, when animals are not faced with energy limitations, as is the case for most vertebrates most of the time, it is clearly beneficial for the brain to adapt to changes in the environment through plasticity. In support of a trade-off between plasticity and metabolic resilience, the rodent hippocampus shows a dorsal–ventral gradient in damage by ischemia (Ashton et al., 1989), which mirrors the capacity for synaptic plasticity owing to the degree of Ca²⁺ permeability of the NMDARs (Hurley et al., 2022). Therefore, hypotheses for selective pressures shaping the evolution of neural circuits may need to incorporate how synapses have balanced the need for plasticity and robust function during energy limitations.

In conclusion, a failure to balance energy supply and physiological demands of neurons leads to disordered circuit activity. We identified a circuit that has the capacity to modify NMDARs to reduce Ca²⁺ influx and constrain excitability in hypoxia without altering their normal contribution to network function. These findings represent a state-dependent form of NMDAR plasticity that likely plays an adaptive role by promoting coordinated network activity in a circuit that cannot typically maintain homeostasis in low oxygen. These results provide insight into natural mechanisms that reduce neural circuit reliance on high rates of energy production. Yet, they also point to a potential trade-off between energetic robustness and plasticity that requires Ca²⁺ influx through NMDARs. Uncovering mechanisms that shift the balance between these two states may inform novel neuroprotection strategies with high clinical relevance.