Homeostatic regulation of a motor circuit through temperature sensing rather than activity sensing

Summary

Homeostasis is a driving principle in physiology. To achieve homeostatic control of neural activity, neurons monitor their activity levels and then initiate corrective adjustments in excitability when activity strays from a set point. However, fluctuations in the brain microenvironment, such as temperature, pH, and other ions represent some of the most common perturbations to neural function in animals. Therefore, it is unclear if activity sensing is a universal strategy for different types of perturbations or if stability may arise by sensing specific environmental cues. Here we show the respiratory network of amphibians mounts a fast homeostatic response to restore motor function following inactivity caused by cooling over the physiological range. This response was not initiated by inactivity, but rather, by temperature. Compensation involved cold-activation of noradrenergic neurons via mechanisms that relied in part on inhibition of the Na⁺/K⁺ ATPase, causing β-adrenoceptor signaling that enhanced network excitability. Thus, acute cooling initiates a modulatory response that opposes inactivity and enhances network excitability. As the nervous system of all animals is subjected to changes in the microenvironment, some circuits may have selected regulatory systems tuned to environmental variables in place of, or in addition to, activity-dependent control mechanisms.

Graphical Abstract

eTOC Blurb:

Using the motor circuit that produces breathing in frogs, Cannon and Santin show that acute cooling initiates a feedback loop involving noradrenergic signaling to regulate activity.

Introduction

Negative feedback homeostasis is a core principle in physiology that explains how organisms maintain balance despite ongoing challenges in the internal and external environment¹. While feedback systems regulate critical body variables like blood pH, body temperature, and hormone production, neural circuits also control activity levels and other aspects of neural function such as synaptic strength through homeostatic mechanisms². Within this framework, neurons are thought to sense their own activity levels via cellular reflections of network activity, including changes in firing rate, intracellular Ca²⁺, or neurotransmitter receptor activation^3–6. During perturbations in activity, alterations in these processes serve as transduction steps that engage intracellular signaling pathways to adjust neuronal excitability in the opposite direction of the disturbance. Homeostatic processes are crucial to maintaining stable activity levels, playing a role in visual system development⁷, sleep⁸, emergence from hibernation⁹, and formation of memory specificity¹⁰.

Many animals inhabit environments with challenging abiotic stressors, such as fluctuations in temperature, pH, and salinity that can cause disturbances in neural function¹¹. In addition, mammals thought to have constant internal conditions also experience pronounced changes in the brain microenvironment^12–14. While these challenges have the potential to disrupt neuronal activity, most animals persist unscathed, at least within limits. Homeostatic regulation is a logical candidate for stability during these challenges, and studies often place these processes in the context of activity-dependent signaling^15–17. However, an underappreciated possibility is that the physical environment per se may directly initiate regulatory responses. In principle, temperature-sensitive ion channels could produce signaling events that correct disturbances by temperature^{18, 19}. The Na⁺/K⁺ ATPase and other electrogenic ion transporters have high temperature and pH dependencies and can influence network excitability through changes in the membrane potential and ion gradients²⁰. Finally, some G-protein coupled receptors are directly modulated by temperature, pH, and extracellular Na⁺ ions which could influence neural activity via metabotropic signaling^21–23. Therefore, in addition to activity-dependent signals commonly implicated in feedback control, physical variables of the neuronal microenvironment may act through parallel pathways to elicit regulation.

Here, we test the hypothesis that activity-dependent signals vs. those from the ambient environment elicit compensation that regulates motor behavior. For this, we used the amphibian respiratory network, as these animals experience a wide range of ambient temperatures and undergo rapid changes in temperature depending on their microhabitat. To meet their metabolic demands, brainstem circuits must produce rhythmic output across a range of temperatures. However, temperatures below ~15°C strongly depress output^{24, 25}, which could be catastrophic on the short term as breathing must continue, at least to some extent, for metabolic homeostasis²⁶. To overcome this challenge, we show here that inactivity caused by acute cooling induces a response consistent with homeostatic regulation over the timescale of tens of minutes to enhance excitability of the respiratory network. Rather than activity-dependent signals, we detail a regulatory strategy that controls network activity by sensing brain temperature.

Results

Respiratory-related motor activity can be recorded from the ex vivo brainstem-spinal cord via cranial nerve rootlets that innervate the lower jaw and glottal musculature. Activity from this preparation closely matches the respiratory motor pattern seen in freely behaving animals²⁷, allowing the study of the central respiratory network independent from peripheral inputs. As expected, cooling from 22°C to 10°C stopped respiratory motor output (Figure 1A). However, after periods of inactivity between ~10–20 minutes, 6/9 preparations resumed bursting. Following 30 minutes at 10°C, the temperature was warmed to the baseline temperature of 22°C. All preparations experienced an overshoot in burst frequency relative to baseline and then drifted back to control levels by approximately 30 minutes (Figure 1E & F). These responses are consistent with real-time homeostatic regulation of circuit output^28–31: Once cooling stops the network, adjustments in excitability attempt to restore activity. Upon return to the baseline temperature, compensation during the perturbation manifests as hyperexcitability, which then fades as activity drifts back near the initial value.

Figure 1. Cold temperature induces a homeostatic response that boosts network excitability without sensing activity.

A.-D. Example recordings of network inactivity induced by cooling (A). After ~10 minutes of inactivity by cooling activity recovers. Upon warming the brainstem, activity overshoots the baseline and then recovers in the next tens of minutes. Inactivity treatments at warm temperatures do not show this response (B; 2 mM glycine, C; hypoxia, D; 100 nM TTX). E. Change in instantaneous burst frequency from baseline over the following 30 minutes after cooling (n=9, paired t test), indicating enhanced network excitability compared to baseline following cooling. Time controls left at 22°C do not show this response (n=7, paired t test). F. Averaged data showing the compensatory increase in burst rate following rewarming (left) and no change in time controls (right). G. Summary data from glycine, hypoxia, and TTX, indicating that no compensatory overshoot was observed as seen after cooling. H. Summary data showing the activity (or lack of activity) at 10°C, glycine, and hypoxia (Kruskal-Wallis Test; p=0.0046; p values shown in the figure are Dunn’s multiple comparison tests). See also Figure S1, Figure S2, and Data S1.

We hypothesized that the compensatory response following inactivity by cooling may be driven by an activity-dependent signal, consistent with much of the literature in neuronal homeostatic regulation². To test this, we silenced the network at 22°C over the same time course as cooling experiments using 3 approaches: 2 mM glycine to enhance glycinergic synaptic inhibition (n=6), 100 nM tetrodotoxin (TTX; n=4) to block action potentials, and hypoxia (n=6) to stop activity by reducing ATP production, which is likely to also occur during the cold. Silencing preparations with glycine and hypoxia did not lead to compensatory bursting (activity could not occur in TTX) (Figure 1B–D,H). Washing each compound restored activity, but did not lead to an overshoot after activity returned (Figure 1B–D, G), further indicating that inactivity per se does not lead to a compensatory increase in excitability as was consistently observed following inactivity by cooling. Therefore, acute temperature changes, rather than the loss of activity, appear to be the variable that drives compensatory responses.

A key aspect of homeostatic regulation involves Ca²⁺ signaling that acts as a second-messenger³². Several ion channels may allow increases in intracellular Ca⁺ during cooling. In amphibians, TRPM8 and TRPV3 are cold-activated channels with Ca²⁺ permeability^{33, 34}. However, inhibitors of each channel did not influence the compensation response (Supplemental Figure 1A–B&E,F). In addition, cooling slows the inactivation/desensitization rates of L-type Ca²⁺ channels and NMDA-glutamate receptors by 10–50-fold over 10°C, causing channels to remain open for longer in the cold^{35, 36}. Inhibitors of L-type Ca²⁺ channels and NMDARs failed to oppose the compensatory overshoot in response to cooling (Supplemental Figure 1C–D&E,F). Surprisingly, the block of NMDARs led to a greater compensatory overshoot frequency compared to controls, which never returned to baseline over the 30-minute recovery period (Supplemental Figure 1D&E). In sum, TRPM8, TRPV3, and L-type Ca²⁺ channels are not involved in the compensation response, while NMDARs appear to function as a constraint to prevent too large of compensation responses, at least in the recovery period.

Neuromodulation is a candidate effector of homeostatic regulation³⁷. The respiratory rhythm of amphibians is thought to be generated by a group of neurons in the reticularis parvocellularis, but is influenced by modulatory systems, including serotonin, norepinephrine, orexin, and more³⁸. To assess the degree to which compensation involves the core rhythm generating and motor neurons vs. long-range modulatory input, we employed the “thick slice” preparation³⁹. This preparation contains the neurons needed for central pattern generation but produces activity that is ~10 times slower than normal and lacks much of the neuromodulation present in the intact brainstem. When we cooled thick-slices from 22°C to 10°C and back to 22°C (n=7), we did not observe compensation consistent with that seen in the intact brainstem (Supplemental Fig. 2A): Preparations reduced burst frequency upon rewarming, rather than increasing, indicative of a loss of the compensation response (Supplemental Figure 2B). This suggested that modulatory centers in the intact preparation play a role in enhancing excitability during acute temperature changes.

While many modulatory inputs exist, norepinephrine influences temperature responses of the respiratory network²⁴. Most neurons have some degree of temperature sensitivity, typically being slowed by cooling and enhanced by warming in a quasi-predictable manner⁴⁰. However, noradrenergic neurons of locus coeruleus (LC) in amphibians do not follow this trend: Instead, they are strongly activated by cooling and inhibited by warming via intrinsic mechanisms^{41, 42}. Given the paradoxical activation of LC neurons by cooling, we addressed the role of the noradrenergic system as a temperature sensor that boosts the excitability of the respiratory network following inactivity by acute cooling.

For this, we first used a chamber that separated isthmus structures containing the LC from the brainstem (Fig. 2A). Since LC neurons are activated by cold temperatures, we hypothesized that keeping the midbrain/isthmus segment warm would reduce the degree of compensation if activation of these neurons by cooling is involved. When clamping the midbrain/isthmus at the control temperature and cooling only the brainstem, we observed reduced compensation: Preparations that recovered activity in the cold did so to a lesser extent and did not overshoot baseline frequency upon return to 22°C (Fig. 2A–C). Therefore, neurons within the midbrain/ isthmus must be cooled to fully express homeostatic responses in the brainstem.

Figure 2. Involvement of isthmus/midbrain cooling and β-adrenoreceptor signaling in homeostatic responses.

A. Example recordings showing network responses to acute cooling and warming only the brainstem region while maintaining the midbrain/pons at 22°C. When only cooling the brainstem, no compensatory overshoot response is observed, demonstrating a requirement for midbrain/pontine cooling. D-F show mean data of the peak overshoot response up rewarming (two-way ANOVA followed by Holm-Sidak Multiple Comparisons test) and activity at 10°C (unpaired t test) with cooling only the brainstem (green) compared to cooling the entire brainstem (control; black). B-D. Example recordings of homeostatic responses to acute temperature changes after blocking β, α1, and α2-adrenergic receptors with 8 μM propranolol, 500 nM prazosin, and 2 μM yohimbine. G-H. show mean data of the peak overshoot response up rewarming compared to recovery (two-way ANOVA followed by Holm-Sidak Multiple Comparisons test) and activity at 10°C (unpaired t test) in the presence of each adrenoreceptor inhibitor. See also Figure S3 and Data S1.

We next determined the involvement of norepinephrine receptors in compensation. For this, we bath-applied antagonists of α1, α2, and β-adrenergic receptors to the intact preparation. α1 and α2 did not play a role, where typical compensatory responses were observed (Fig. 2 E–H). In contrast, the β-adrenoreceptor antagonist, propranolol, reduced the degree of recovery in the cold and strongly blunted the compensatory overshoot upon rewarming (Fig. 2 D, G–H). In addition, activation of βRs using the agonist isoproterenol at warm temperatures mimicked the response to cooling in two ways. First, isoproterenol enhanced network excitability, as demonstrated by an increase in burst frequency. Second, increased excitability had a slow washout time, where burst rate remained elevated after returning to baseline conditions for 30 minutes (Supplemental Fig. 3). Therefore, β-adrenergic signaling is both necessary and sufficient to produce aspects of the network responses observed following acute temperature changes.

Given the requirement for midbrain/ isthmus cooling and norepinephrine signaling, we addressed how neurons of the LC, which we have recently verified to be noradrenergic⁴¹, transduce cold into enhanced excitability. Corroborating previous results⁴³, LC neurons in slices increase their firing rates during acute cooling (Figure 3A&C), consistent with a requirement for midbrain/ isthmus cooling and noradrenergic signaling in the compensation response. To determine if these responses were also seen in neurons of the brainstem, we assessed motoneurons from the hypoglossal nucleus. Brainstem motoneurons neurons did not exhibit a firing response to cooling (Supplemental Figure. 4), suggesting specificity of cold-activation within the LC.

Figure 3. Cold activates neurons within the noradrenergic locus coeruleus, which involves inhibition of the electrogenic Na⁺/K⁺ ATPase.

A. Example recording of an LC neuron in a brain slice, demonstrating strong activation by cooling from 22°C to 10°C. B. Mean data showing increases in LC neuron firing rate during cooling (n = 9 cells from N = 4 animals; paired t test). C. Raw voltage clamp recording (v_hold= - 50 mV) during cooling before and after application of 3 μM ouabain to inhibit the Na⁺/K⁺ ATPase (n = 14 cells from N = 11 animals). Cooling elicits a reversible inward current (I_cold). Application of ouabain induces an inward current, but cooling in the presence of ouabain reduces the magnitude of I_cold. D. Mean data showing I_cold in aCSF vs. ouabain (paired t test). E. Comparison of I_cold vs. I_ouabain (paired t test). See also Figure S4 and Data S1.

For the molecular sensor of cold within LC neurons, we reasoned that TRPM8, TRPV3, NMDARs, and L-type Ca²⁺ channels are unlikely to contribute in large part because these processes did not account for the network homeostatic response. As ATPases are enzymes with potentially high temperature sensitivity²⁰, we focused on how cold may inhibit the tonic hyperpolarizing current from the electrogenic Na⁺/K⁺ ATPase (Na⁺ pump) to stimulate LC neurons. We tested this by voltage clamping the cold-induced inward current (I_cold) that depolarizes neurons from resting potential. Cooling LC neurons led to an inward I_cold, consistent with the potential for cooling to decrease hyperpolarizing drive, and therefore, depolarize neurons (Figure 3B&D). We then added an inhibitor of the Na⁺ pump, 3 μM ouabain. Ouabain elicited an inward current, indicating inhibition of the tonic outward/hyperpolarizing current from the Na⁺ pump. After reaching steady-state, subsequent cooling in the presence of ouabain produced I_cold with a reduced magnitude (Figure 3B–D). These results show that the inhibition of the Na⁺ pump contributes to I_cold because it was smaller when the Na⁺ was already inhibited. While ouabain also led to an inward current on average, it was significantly smaller than I_cold, which suggests that additional mechanisms are at play in generating I_cold along with inhibition of the Na⁺ pump (Figure 3E). Nevertheless, the inhibition of the electrogenic Na⁺ pump serves as a critical transduction step for midbrain/ isthmus cooling to stimulate LC neurons, which then elicits network compensation via β-adrenergic signaling.

Finally, we sought to link inhibition of the Na⁺ pump to network compensation that occurs during acute temperature changes. We expected that experimentally offsetting inhibition of the Na⁺ pump in the cold would reduce compensation. While there are no known selective pharmacological agonists of the Na⁺ pump, monensin is Na⁺-H⁺ exchanging ionophore, which increases intracellular Na⁺ to activate the Na⁺ pump. This approach has been used to activate the Na⁺ pump in motor systems of leeches, rats, and amphibians^44–46. To address this in our system, we applied 1 μM monensin to LC neurons in brain slices. Monensin initially hyperpolarized the membrane and reduced activity at 22°C, consistent with a greater electrogenic Na⁺ pump activity (firing rate: baseline- 2.1±2.2 Hz, monensin- 0.15±0.4 Hz; p=0.0172; paired t-test, membrane potential baseline −54.6±2.7 mV, monensin- −58.6±; p=0.0019, paired t-test). Cooling in the presence of monensin still depolarized neurons to a similar extent as controls. However, as membrane potential started from a more negative value, the absolute membrane potential and firing rates were lower than untreated controls in the cold (Figure 4A–C). Accordingly, monensin-exposed neurons in the cold had similar membrane potentials and firing rates to untreated controls at baseline (membrane potential- p=0.943, firing rate- p=0.540; unpaired t-tests). These results suggest that monensin increases Na⁺ pump activity to counteract at least some of the inhibition that normally occurs in the cold, maintaining activity around baseline levels (Fig. 4A–C).

Figure 4. Pharmacological manipulation that activates the Na⁺/K⁺ ATPase reduces cold-activated firing of LC neurons and opposes homeostatic network recovery.

A. Example recordings of LC neurons in aCSF (control) and in the presence of 1 μM monensin to activate the Na⁺/K⁺ ATPase. Control neurons depolarize and strongly increase firing rate. With monensin, membrane potential was more negative to start, consistent with activating of the Na⁺/K⁺ ATPase. There was a similar depolarization in response to cooling, but having started from a more negative membrane potential, the activity increase in response to cooling is blunted. B-C. Mean data comparing the cold-induced membrane depolarization and firing responses of LC neurons in aCSF and monensin (two-way ANOVA; control: n = 9 cells from N = 4 animals, monensin: n=11 cells from N=6 animals). For firing rate (B), there was a significant group effect (p=0.0015), temperature effect (p<0.0001), and interaction existed between group and temperature (p=0.0146). While cold activated neurons in both groups, the response was blunted by monensin (activity was less in monensin compared to control; Holm-Sidak multiple comparisons test). For membrane potential (C), there was a significant group effect (p<0.0001) and temperature effect (p=0.0172). While cold depolarized neurons in both groups, the membrane potential was more hyperpolarized in monensin at baseline and 10°C (Holm-Sidak multiple comparisons test). D-E. Example recordings of homeostatic responses during acute temperature changes in aCSF (control) and in the presence of 1 μM monensin. F-G show mean data of the peak overshoot response up rewarming compared to recovery (two-way ANOVA followed by Holm-Sidak Multiple Comparisons test) and activity at 10°C (unpaired t test) in controls and 1 μM monensin. See also Data S1.

Given these results showing that monensin opposes cold-activation of nonandrogenic neurons, we then assessed network compensation in the presence of monensin. Consistent with LC neuron results, network output at 22°C slowed slightly after intact preparations were exposed to monensin (baseline: 16.4±5.5 burst per minute, Monensin: 11.4±4.4 bursts per minute, n=8, p=0.0212; paired t-test). In response to acute temperature changes, monensin reduced the compensation response: Preparations that recovered activity in the cold did so to a lesser extent and had a less pronounced overshoot upon the return to warm temperatures (Fig. 4D–G). These results show that manipulating a key cellular mechanism that reduces cold-activation of noradrenergic neurons attenuates the compensatory network response during activity disturbances caused by acute temperature change. In sum, these results point to a feedback loop for the regulation of network activity, where instead of sensing activity, temperature sensitivity of modulatory neurons leads to compensation within a motor circuit.

Discussion

Over 30 years ago, an important set of studies suggested that neurons transduce their own activity levels, in turn, causing cellular adjustments aimed at controlling network output through activity-dependent homeostasis^{47, 48}. This set in motion a blueprint for explaining how the healthy nervous system maintains stable activity patterns, and how issues with these regulatory systems could disrupt circuit output in disease⁴⁹. A core interpretation of nearly all models of feedback homeostasis in the nervous system, from short to long time scales, is that some aspect of neural activity— be it intracellular Ca²⁺ dynamics, neurotransmitter receptor activation state, or membrane potential— is vital for regulation^{6, 28, 50–52}. Here, our results diverge from this theme and introduce that specific environmental cues, in this case temperature, trigger homeostatic responses without tracking activity of the circuit.

To achieve regulation in this way, our results demonstrate an important role for modulatory neurons as temperature sensors (Fig. 2A), β-adrenoreceptor signaling (Fig. 2D, G–H), and a built-in constraint by NMDA-glutamate receptors to prevent too much compensation from occurring (Supplemental Fig. 2D–E). The respiratory rhythm in adult amphibians is thought to be generated by mechanisms that involve reciprocal inhibition, as well as excitatory neurons that carry this rhythm to motor pools^{25, 53}. Therefore, excitatory metabotropic signaling from β-adrenoreceptors and constraint by NMDARs may act somewhere along this path. While rhythm-generating centers and/or motoneurons are likely candidates, all pharmacological tools used to draw these conclusions were presented to the intact preparation. Accordingly, it remains possible these mechanisms may act within modulatory neurons. Furthermore, β-adrenoreceptors and NMDARs are likely to be temperature sensitive themselves, shaping the compensation response. In sum, while we identify critical mechanisms that cause and control the regulatory response during acute cooling, future work is needed to understand the interaction between these components.

To activate β-adrenoreceptors to enhance excitability in the brainstem, a key step is midbrain/isthmus cooling, which involves the activation of noradrenergic neurons via inhibition of the Na⁺ pump. Indeed, inhibition of the Na⁺ pump occludes the current that activates LC neurons in the cold (Figure 3), and applying monensin to oppose Na⁺ pump inhibition that occurs in the cold prevents the strong activation of the LC neuron firing (Figure 4). While we show that midbrain/isthmus cooling is necessary for the compensation response (Figure 2A), these network experiments involving monensin must be interpreted cautiously as it was applied to the intact preparation, and therefore, may also act within the brainstem. In addition to cooling as a key stimulus, the Na⁺ pump is a target of neuromodulation^{54, 55}, where modulatory signaling may synergize with cooling to further inhibit the Na⁺ pump to activate noradrenergic neurons in the intact system. Another important point is that Na⁺ pumps are ion transporters in all neurons. Therefore, it may seem contradictory for it to serve a specialized role in the cold activation of modulatory neurons. However, the α subunit of the Na⁺ pump undergoes substantial RNA editing, which gives rise to a range of functionally diverse proteins⁵⁶. We speculate that RNA editing of the Na⁺ pump within LC neurons may contribute to cold sensitivity. Finally, while our work here identified loss of the hyperpolarizing pull from Na⁺ pump as a way to stimulate LC neurons in the cold, other mechanisms likely contribute. When inspecting the individual recordings, ~50% of cells presented with residual I_cold in the presence of ouabain. Indeed, we previously showed that cold increases input resistance in LC neurons, which points to the temperature-sensitive inhibition of K⁺ channels contributing to cold activation^{42, 43}. Regardless of other potential mechanisms beyond that which we identified here, our results show that equipping modulatory neurons with temperature sensors plays a role in defending activity homeostasis of a motor network.

Why use an environmental variable like temperature to regulate the activity of a motor circuit? We suggest that this strategy may build specificity for regulation in different environments and allow circuits to implement regulation across timescales. For specificity, while breathing is a critical rhythmic behavior, amphibians use peripheral sensory systems to inhibit ventilation, potentially, as a predator detection and avoidance strategy^{57, 58}. If sensors of global network activity induced compensation over the timeline we observed here, the homeostatic response that followed could be detrimental to the goal of predator avoidance by haphazardly restarting or increasing breathing. Concerning how these mechanisms may be adaptive in the cold, CO₂/pH within the brainstem is the primary stimulus for ventilation in air-breathing vertebrates. However, CO₂/pH chemosensors cannot respond to changes in CO₂ in the cold and do not influence ventilation at 10°C^{42, 59, 60}. This likely occurs because CO₂ excretion becomes more reliant on passive loss via the skin at this temperature⁶¹. However, lung O₂ uptake is likely to remain important⁶², and frogs take breaths of air at 1 and 5 °C when the skin can fulfill all gas exchange needs⁶³. Therefore, these results suggest that the central respiratory circuits switch from CO₂/pH to temperature per se as an important drive to breathe at cool temperatures. In this way, temperature sensitivity seems to allow the network to restore activity in conditions where it might be needed while leaving the ability to keep the network silent in other situations where it is adaptive to do so.

Using environmental signals such as temperature may also allow circuits to perform regulation across different timescales. A problem in neuronal regulation has been identified, where it is difficult to maintain homeostasis with multiple sensors operating over different time intervals that use the same activity signal (e.g., altered intracellular Ca²⁺)^{50, 64} [but see a recent study that offers a theoretical solution to this problem¹⁷]. Results in the amphibian respiratory network have begun to accumulate, indicating that separating regulation mechanisms by induction cue may offer a workaround to this problem. On short timescales (seconds to minute), respiratory motoneurons sense increases in activity through rises in intracellular Na⁺ and membrane depolarization via activation of an isoform of the Na⁺ pump with low Na⁺ affinity and Kv7 channels, respectively⁶⁵. During inactivity over weeks to months seen in hibernation, synaptic compensation also enhances motoneuron and network excitability using both inactivity and Ca²⁺- dependent signals¹⁵. Temperature sensing at modulatory neurons which activates metabotropic receptors as we show here may provide a way to restrict regulation to intermediate time scales (tens of minutes) without interfering with faster and slower processes. Therefore, using different induction cues for regulation seems to endow circuits with the capacity to meet different regulatory challenges of varying duration and type.

More broadly, these results point to organizational rules for regulation that may be applicable to other systems. For example, several circuits have been reported to lack compensation via activity-dependent signaling, at least over certain time scales^66–68. It is possible that regulation may be achieved instead by transducing select variables from the neuronal microenvironment. While the temperature perturbation used here is relevant to amphibians and other poikilothermic species (i.e., all animal groups except for most mammals and birds), certain mammalian brain regions such as the hippocampus and nucleus accumbens can vary by as much as 4°C Celsius depending on the level of metabolic activity¹², in addition to other microenvironment shifts such as pH and extracellular K^{+ 13, 69}. Beyond environmental sensors for feedback control, these processes likely work in tandem with network designs that build resilience to environmental perturbations to prevent activity disturbances before they occur^{25, 70, 71}. Overall, new insights into how neural circuits accomplish the feat of maintaining stability despite ever-changing internal and external environments will likely be found along a path that integrates activity-dependent, as well as environmentally-driven feedback processes.

Resource Availability

Lead Contact:

J.M. Santin, santinj@missouri.edu

Materials Availability:

This study did not generate any new or unique reagents.

Data and Code Availability:

• All raw data used to create the figures have are supplied as Data S1 A–J.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request

STAR Methods:

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animals

All protocols were approved by the Animal Care and Use Committee (ACUC) at the University of Missouri (protocol # 39–264). Adult female American Bullfrogs (n=119), Lithobates catesbeianus, purchased from Rana Ranch (Twin Falls, Idaho), were used for all experiments. Frogs were housed in 20-gallon plastic tanks filled with dechlorinated tap water that was perpetually bubbled with room air, with access to both wet and dry areas. Water was cleaned daily and changed as needed. Frogs were fed pellets once a week and were kept on a 12hr:12hr light:dark cycle, with all experiments taking place during the light cycle.

METHOD DETAILS

Brainstem-spinal cord preparation

Frogs were deeply anesthetized using approximately 1 ml of isoflurane in a closed 1 L chamber until response to toe pinch was no longer observed. Rapid decapitation with a guillotine was utilized to achieve euthanasia. Brainstem dissections were performed in ice-cold artificial cerebrospinal fluid (aCSF). The forebrain was pithed immediately, and the brainstem-spinal cord was carefully extracted while preserving the nerve roots. The remains of the forebrain were then separated, and the brainstem was cut caudal to spinal nerve III. Following the dissection, the brainstem was pinned to a ~6mL Sylgard-coated dish and superfused with oxygenated aCSF using a peristaltic pump. In a subset of experiments, the brainstem was pinned in a “split bath” dish containing walls protruding from either side, creating two chambers superfused by two separate pumps with two separate temperature controls. In this dish, the brainstem was positioned between the two walls, with the midbrain/isthmus region sitting on one side, and the brainstem on the other. An agar block was placed between the two walls and positioned above the brainstem preparation close enough to effectively modulate each chamber’s temperature, but far enough to allow sufficient perfusion to that region. For extracellular nerve recordings, near-synchronous respiratory-related activity associated with lung ventilation can be observed on cranial nerves V, VII, X, and hypoglossal. Either cranial nerve X (vagus) or hypoglossal activity was recorded using a glass suction-electrode, and AC amplified (1000x, A-M Systems Model 1700, A-M Systems, Carlsborg, WA, USA), filtered (10Hz-5kHz), and digitized (PowerLab 8/35 ADInstruments, Colorado Springs, CO, USA).

Drugs

Several pharmacological tools were used in this study: 2 mM Glycine (dissolved in DI water), 10 μM verapamil (dissolved in DI water), 20 μM D-APV (dissolved in DI water), 3 μM ouabain (dissolved in ethanol), 2 μM monensin (dissolved in ethanol), 2 μM Trpvicin (dissolved in DMSO), 100 nM TTX (diluted from 10mM; dissolved in DI water), and 1μM (−)-Bicuculline methiodide (dissolved in DI water) were from Hello Bio (Princeton, NJ, USA). 8 μM propranolol (diluted from 50mM; dissolved in DI water), 500 nM prazosin (dissolved in DI water), and 2 μM yohimbine (dissolved in DI water) were from TCI America (Portland, OR, USA). 3 μM strychnine hydrochloride (dissolved in DI water) was from Sigma-Aldrich (St Louis, MO, USA). 500 nM AMG 333 (dissolved in ethanol) was from Tocris.

Experimental protocols

Nerve output was allowed to stabilize for approximately 1 hour at room temperature before drug application or temperature change. In control experiments (n=9), the temperature of the aCSF was decreased from 22°C to 10°C using a bipolar in-line temperature controller (Warner Instruments Model CL-100, Warner Instruments, Hamden, CT, USA), and the temperature of the solution was monitored with a thermocouple positioned close to the brainstem. The temperature was held at 10°C for 30 minutes before warming back up to the baseline temperature of 22°C.

A subset of experiments was performed to induce network inactivity in control temperatures. In one group (n=6), 2mM glycine was added to the perfusate and allowed to superfuse for 30 minutes before getting washed out. A different group (n=4) was exposed to 100nM tetrodotoxin for ~60 min prior to washing out. TTX was used for 60 minutes because 100 nM took ~20–30 minutes to fully stop activity, which allowed us to more closely match the duration of inactivity near 30 minutes. A final group (n=6) was exposed to anoxia for 25 min prior to washing out. We selected 25 minutes because most preparations took ~5 minutes to restart upon the return of O₂, which again allowed us to match the duration of inactivity. Hypoxia was achieved via superfusion of aCSF that was bubbled with a gas mixture that contained 0% oxygen, 98.5% nitrogen, and 1.5% carbon dioxide. The temperature was kept at 22°C for the duration of each of these experiments. For all experiments that used various inhibitors, each was applied for 10–15 minutes before cooling and then followed the normal protocol to cool and rewarm the tissue.

Thick slice preparation

To address the capacity for a minimal unit of the respiratory network to undergo the compensation response, we used a recently developed thick-slice preparation (Saunders and Santin, 2023). This preparation exhibits features of the intact brainstem preparation (burst pattern and stereotyped sensitivity to modulators as seen in the intact brainstem such as suppression by opioid agonist and stimulation by serotonin) but produces activity at ~1/10 of the normal rate. Following dissection of the intact brainstem, vagus nerve output from the intact brainstem was briefly observed preceding transection to ensure preparations were viable. Using small spring scissors, the brainstem was transected rostral and caudal to the vagus nerve root. Shortly following transection, 3μM strychnine + 1μM bicuculine was added to the aCSF perfusate which is needed to trigger burst activity (Saunders and Santin, 2023). Vagal motor output was given >30 min to stabilize prior to decreasing the temperature from 22°C to 10°C. Following a 30-minute period at 10°C, the temperature was increased back to baseline and activity was recorded for another 30–40 minutes.

Whole-cell patch clamp electrophysiology

Following dissection, brainstem preparations were adhered to an agar block mounted on a vibratome tissue slicer and submerged in cold, bubbled aCSF. 300 μM transverse slices were obtained from regions containing either the locus coeruleus (LC) or hypoglossal motor nucleus and left to recover in room-temperature, bubbled with aCSF. Upon transfer to the recording chamber, slices were immobilized with a nylon grid and kept with constant oxygenated aCSF super-perfusion. Neurons were visualized with a fixed-stage microscope (FN1, Nikon Instruments Inc., Melville, NY, USA). Glass patch pipettes were pulled using a P1000 horizontal pipette puller (Sutter Instruments, Novato, CA, USA) and backfilled with artificial-intracellular solution composed of the following (in mM): 110 K-gluconate, 2 MgCl₂, 10 HEPES, 1 Na₂-ATP, 0.1 Na₂-GTP, 2.5 EGTA. Pipettes were positioned near the neurons of interest using a micromanipulator (Sutter Instruments, Novato, CA, USA) and whole-cell recordings were acquired using an Axopatch 200B amplifier and Axon Digidata 1550 digitizer (Molecular Devices, San Jose, CA, USA).

LC neuron experiments

Neurons found in the area thought to be homologous to the locus coeruleus of mammals (González et al., 1994; Marin et al., 1996), which also express dopamine beta hydroxylase for norepinephrine synthesis (Amaral-Silva and Santin, 2023), have previously been found to increase firing rates during acute cooling (Santin et al., 2013). To establish the response of LC neurons to cold temperatures, a subset of cells (n=9) was recorded in current-clamp mode (I = 0) and allowed to establish a baseline firing-rate (~5 min.). In cells that did not spontaneously fire, a small positive current was introduced and increased until the cell fired and remained firing at a consistent rate. Once baseline was established, the temperature of the chamber was decreased from 22°C to close to 10°C (temperature varied 1–2° between cells depending on the probe’s position in the chamber) using a bipolar in-line temperature controller (Warner Instruments). Temperature remained around 10°C for 10 minutes before the chamber was brought back to the baseline temperature.

In a different set of experiments, LC neurons (n=14) were recorded in voltage-clamp mode (holding voltage = −50mV) and allowed to establish a baseline current. After baseline was established, the temperature of the chamber was decreased from 22°C to around 10°C for a brief period prior to returning to the baseline temperature. The cell was given several minutes to stabilize at 22°C before the slice was exposed to 3μM ouabain in aCSF. Ouabain was continuously supplied into the chamber for long enough to observe an inward current in most cells. When this inward current saturated, a second cooling was performed in the presence of ouabain. There are two dominant Na⁺-K⁺ ATPase isoforms expressed in neurons: those containing the α1 subunit and those containing an α3 subunit. α3-containing isoforms have ouabain IC₅₀ values in the 30–60 nM range, while ouabain inhibits α1-containing isoforms with IC₅₀s ranging from ~20–60 μM⁷². Therefore 3 μM ouabain likely inhibits α3-containing isoforms maximally, while it will only partially block α1-containing isoforms. While inhibition of the Na⁺-K⁺ ATPase is likely to underlie the inward current (loss of an outward current) in response to 3 μM ouabain, we do not yet know which of the two isoforms are expressed in LC neurons, their relative proportions, or details of their kinetic activities (Na⁺ and ATP sensitivity).

In a separate subset of current-clamp recordings, LC neurons (n=11) were exposed to 1μM monensin following an initial stabilization period (>5 min.). After several minutes (>5 min.) of monensin exposure, the temperature of the chamber was decreased from 22°C to 10°C, where it remained for a duration of 10 min. and then returned to baseline temperature for an additional 10-minute period.

Hypoglossal motoneuron experiments

A series of hypoglossal motoneuron current-clamp recordings (n=9) were obtained to compare to the temperature response of cells in the locus coeruleus. Neurons were given time to establish a baseline firing rate (5 min.) prior to being cooled from 22°C to 10°C for a 30-minute period. The temperature was then brought up to baseline for an additional 15 minutes.

QUANTIFICATION AND STATISTICAL ANALYSIS

Data Analysis

Brainstem preparations

Burst frequency for 5 minutes at baseline, in the last 10 minutes of cold, the peak 5 minutes during the overshoot, and then burst frequency at 30 minutes were analyzed for all network level experiments. In experiments that used various pharmacological tools, these were applied for 10–15 minutes before cooling. The effect of each drug before cooling is shown in Figure S5. To quantify the overshoot we analyzed the change in burst frequency compared to baseline. While the starting frequency across preparations differed, the change from baseline during the overshoot period did not correlate with starting frequency (i.e., we observed similar absolute changes regardless of the baseline activity level) (Figure S6). Thus, we performed analyses on the absolute change from baseline. While nearly all preparations exhibited an overshoot following cooling suggestive of compensation in the cold (e.g., all groups except monensin and propranolol), a minority of these preparations did not show activity during cold temperatures. Therefore, when comparing burst activity in the cold under different conditions, we only used preparations that recovered activity. If groups did not have at least n=5 preparations that recovered in the data set, no analysis during the cold was performed.

Whole-cell patch clamp

For control LC current-clamp experiments, we analyzed the spike frequency for 30 seconds during the baseline period and the period in the cold where frequency was at its peak. The same was done in current-clamp experiments in which LC neurons were exposed to monensin with the addition of a 30 second period after monensin exposure, but before the temperature was reduced.

In voltage-clamp experiments, the change in membrane current between baseline and the first period of cold (I_cold) was analyzed. Then, after an inward current produced by ouabain was observed, the change in membrane current induced by ouabain (between the recovery and immediately prior to the second period of cooling) was taken as “I_ouabain”. All quantitative data used to generate the figures are presented in Data S1.

Statistics

Activity during the cold was compared with an unpaired t test. To compare changes in burst activity during the overshoot and recovery to baseline, we used repeated measures two-way ANOVA and Holm-Sidak Multiple Comparisons tests. For comparisons of only one group with multiple time points (isoproterenol experiments), we used a one-way repeated measure ANOVA or Kruskal-Wallis test, followed by Holm-Sidak Multiple Comparisons tests or Dunn’s multiple comparisons tests. Firing rate data before and after cooling, and I_cold before and after ouabain, were compared using a paired t test. Comparisons of I_cold vs. I_ouabain and LC, hypoglossal response to cooling, the change in firing in control vs. monensin used unpaired t tests. Significance was accepted with p<0.05.

Supplementary Material

2

Data S1 All raw data used to generate the figures. Related to Figures 1–4 and Figures S1–6.

A) Description of Data S1A. All raw data used to generate Figure 1.

B) Description of Data S1B. All raw data used to generate Figure S1.

C) Description of Data S1C. All raw data used to generate Figure S2.

D) Description of Data S1D. All raw data used to generate Figure 2.

E) Description of Data S1E. All raw data used to generate Figure S3.

F) Description of Data S1F. All raw data used to generate Figure 3.

G) Description of Data S1G. All raw data used to generate Figure S4.

H) Description of Data S1H. All raw data used to generate Figure 4.

I) Description of Data S1I. All raw data used to generate Figure S5.

J) Description of Data S1J. All raw data used to generate Figure S6.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
Glycine	Hello Bio	Cat#HB0299
Verapamil hydrochloride	Hello Bio	Cat#HB1237
D-APV	Hello Bio	Cat#HB0225
Ouabain	Hello Bio	Cat#HB1140
Monensin	Hello Bio	Cat#HB4882
Trpvicin	Hello Bio	Cat#HB9908
Tetrodotoxin citrate	Hello Bio	Cat#HB1035
(−)-Bicuculline methiodide	Hello Bio	Cat#HB0893
Propranolol hydrochloride	TCI America	Cat#P0995
Prazosin hydrochloride	TCI America	Cat#P0938
Yohimbine hydrochloride	TCI America	Cat#Y0002
Strychnine hydrochloride	Sigma-Aldrich	Cat#S8753
AMG333	Tocris	Cat#6874
Experimental models: Organisms/strains
Lithobates catesbeianus	Rana Ranch (Twin Falls, ID)
Other
PowerLab 8/35	ADInstruments
A-M Systems Model 1700	A-M Systems
Warner Instruments Model CL-100	Warner Instruments
FN1 Fixed-stage Microscope	Nikon Instruments Inc.
P1000 horizontal pipette puller	Sutter Instruments
ROE-200 Micromanipulator	Sutter Instruments
Axopatch 200B amplifier	Molecular Devices
Axon Digidata 1550 digitizer	Molecular Devices

Highlights.

• Rhythmic motor activity stops during acute cooling but then recovers minutes later

• Recovery responses were driven by temperature rather than inactivity.

• Temperature sensing involved noradrenergic neurons and β-adrenoceptor signaling

• Environmental variables may regulate circuits with activity-dependent mechanisms