A cholinergic spinal pathway for the adaptive control of breathing

SUMMARY

The ability to amplify motor neuron (MN) output is essential for generating high-intensity motor actions. This is critical for breathing that must be rapidly adjusted to accommodate changing metabolic demands. While brainstem circuits generate the breathing rhythm, the pathways that directly augment respiratory MN output are not well understood. Here, we map first-order inputs to phrenic motor neurons (PMNs), a key respiratory MN population that initiates diaphragm contraction to drive breathing. We identify a predominant spinal input from a distinct subset of genetically defined V0_C cholinergic interneurons. We find that these interneurons receive phasic excitation from brainstem respiratory centers, augment phrenic output through M2 muscarinic receptors, and are highly activated under a hypercapnia challenge. Specifically silencing cholinergic interneuron neurotransmission impairs the breathing response to hypercapnia. Collectively, our findings identify a spinal pathway that amplifies breathing, presenting a potential target for promoting recovery of breathing following spinal cord injury.

In brief

Lin et al. identify a population of Pitx2+ cholinergic respiratory interneurons in the spinal cord that directly project to phrenic motor neurons and modulate their output. These interneurons are integrated into respiratory circuits and become activated to increase breathing amplitude in response to hypercapnia.

Graphical Abstract

INTRODUCTION

Motor circuits of the nervous system must operate over a dynamic range to adapt their output and generate behaviors of varying intensities in response to environmental challenges. This is especially critical for breathing, a motor behavior essential for maintaining blood gases required to sustain the metabolism of vital organs, including the brain, heart, and kidneys.¹ To meet metabolic demands during physiological challenges such as exercise or the changing environmental conditions associated with altitude, the neuronal network that controls breathing must operate over a large dynamic range. Specialized circuits have evolved to support robust, yet adaptable breathing in terrestrial vertebrates. While a network located in the brainstem is largely responsible for generating the rhythm and pattern of breathing,^2–4 motor neurons (MNs) projecting to muscles in the periphery are the final output of respiratory circuits. In mammals, phrenic motor neurons (PMNs) located in the cervical spinal cord innervate the diaphragm, the major inspiratory muscle that drives airflow into the lungs. The strength of diaphragmatic contraction determines the volume of each breath. Thus, precise regulation of PMN activity is vital for aligning breathing with metabolic needs and ensuring appropriate responses to environmental challenges.

Both central and peripheral chemoreflexes safeguard CO₂/pH homeostasis by mediating rapid ventilatory responses to atmospheric gas fluctuations and changes in metabolic demands.⁵ Inputs from diverse sources, including central and peripheral CO₂ and O₂ chemoreceptors, are integrated by brainstem networks to generate multidimensional changes in ventilation.^6–13 Elevated brain CO₂ (hypercapnia), for example, leads to a significant increase in both the respiratory frequency and amplitude of diaphragm contractions. While increases in frequency are mediated by projections from chemoreceptor neurons to key rhythmogenic regions confined to the brainstem,^9,14 mechanisms for increasing breathing intensity likely involve direct modulation of PMN activity and increased drive to PMNs from distinct premotor populations.

While the brainstem circuits that underlie the generation and modulation of the breathing rhythm have been well defined,^{2–4,15–18} much less is known about the topography, molecular identity, and function of downstream neurons that directly project to PMNs. Antidromic stimulation and retrograde tracing has revealed monosynaptic connections to PMNs, primarily from excitatory rostral ventral respiratory group (rVRG) neurons in the brainstem, which drive PMN activation during inspiration.^19–26 Although direct PMN inputs from spinal interneurons have been identified,^{20,23,27–30} the degree of connectivity between PMNs and local interneurons is unclear. Recent rabies virus tracing experiments in neonatal mice suggested that the number of direct PMN inputs that originate in the spinal cord is minor²⁵; however, tracing experiments with polysynaptic pseudorabies viruses (PRVs) in adult rats, cats, and ferrets reveal more extensive spinal respiratory circuits.^20,21,23 Multielectrode array recordings from the cervical spinal cord also identified interneurons with respiratory-related activity that are synaptically coupled to PMNs,^31,32 and excitatory interneurons at cervical and thoracic levels are able to sustain breathing after spinal cord injury (SCI).^33–35 Despite their roles in promoting respiratory recovery following SCI, we know little about the functions of spinal interneurons, including genetically defined subsets, in the control of breathing.^36–39 One possibility is that local spinal circuits may be important for controlling breathing intensity through direct modulation of respiratory MN output. We therefore hypothesized that distinct classes of spinal interneurons may regulate PMN activity, providing a hierarchically arranged gain control system for breathing that is spatially segregated from the brainstem-derived rhythm generator.

Here, we combined a genetic strategy with rabies virus-mediated tracing to label neurons with monosynaptic inputs to PMNs. We identified a morphologically and topographically distinct population of Pitx2+ V0_C interneurons, largely localized in the cervical spinal cord, with substantial direct projections to PMNs. We find that these interneurons are functionally integrated into respiratory circuits, can modify PMN output through M2 muscarinic receptors, and are activated in response to CO₂ exposure. Finally, we show that inhibiting cholinergic neurotransmission from Pitx2+ interneurons impairs the response to hypercapnia. We propose that spinal cholinergic interneurons provide a node for breathing gain control that is spatially segregated from the brainstem and can be recruited to amplify breathing under high metabolic demands and intense respiratory challenges. These interneurons represent an accessible therapeutic target for conditions in which respiratory function is compromised, such as SCI or amyotrophic lateral sclerosis (ALS).

RESULTS

PMNs receive substantial monosynaptic inputs from the spinal cord

To investigate the contribution of spinal networks to PMN activity, we combined a genetic strategy with rabies-based monosynaptic retrograde tracing to map inputs to PMNs. We utilized a modified glycoprotein (G protein)-deleted mCherry-tagged rabies virus (RabiesΔG-mCherry). Typically, the rabies virus requires G protein for transsynaptic transport, which is not endogenously expressed in mammalian cells. Therefore, G protein deletion from the rabies virus ensures that transsynaptic transport is not possible from the virus itself in wild-type mammalian neurons. To activate the transsynaptic transport mechanism and enable monosynaptic labeling specifically from MNs, we crossed RphiGT mice, which express G protein after Cre-mediated recombination, to Choline acetyltransferase (ChAT)::Cre mice (ChAT::Cre; RphiGT) to induce G protein expression only in cholinergic neurons, which include MNs (Figure S1A).^40,41 We validated the expression of G protein in ChAT::Cre; RphiGT mice by in situ hybridization at postnatal day 4 (P4). Notably, despite the existence of cholinergic interneurons in the spinal cord, the expression of G protein was only detectable in MNs (Figure S1B). We injected RabiesΔG-mCherry unilaterally into the diaphragm, which is solely innervated by PMNs, of ChAT::Cre; RphiGT mice at P4 to label PMNs and trace their synaptic inputs (Figure 1A). Our viral injections specifically targeted PMNs, as seen by the absence of labeled ventral roots and MNs at thoracic and lumbar levels of the spinal cord (Figure S1C). Each injection labeled 1–5 starter PMNs (Figures S2A and S2B).

Figure 1. Distribution of monosynaptic inputs to PMNs.

(A) Tracing strategy for mapping monosynaptic PMN inputs.

(B) Examples of mCherry-labeled brainstem (rVRG) and spinal cord neurons projecting to PMNs. moXII, hypoglossal motor nucleus; NA, nucleus ambiguus; cc, central canal. Scale bar, 200 μm.

(C) Distribution of PMN inputs in the brainstem and spinal cord (n = 7).

(D) Quantification of the rostrocaudal distribution of spinal inputs to PMNs.

(E) Spinal cord PMN inputs are largely localized to the ipsilateral side.

(F) Dorsoventral distribution of PMN inputs in the spinal cord.

(G) Cholinergic interneurons (ChAT+ INs) around the cc in the cervical spinal cord directly project to PMNs. Scale bars, 200 μm (top) and 50 μm (bottom).

(H) Rostrocaudal distribution of ChAT+ INs projecting to PMNs (ChAT+ INs → PMNs).

(I) ChAT+ INs comprise ~10% of total PMN inputs.

(J) Quantification of contralateral and ipsilateral ChAT+ INs projecting to PMNs.

(K) Quantification of the rostrocaudal distribution of ChAT+ INs projecting to PMNs.

(L) ChAT+ INs from the brachial and thoracic spinal cord directly project to PMNs. Scale bars, 200 μm (top) and 50 μm (bottom).

White arrowheads indicate mCherry+ cholinergic interneurons. See also Figures S1 and S2 and Video S1.

Next, we quantified the distribution of direct PMN inputs. All mCherry+ cells were located in either the brainstem or spinal cord and no mCherry+ cells were found in the cortex or cerebellum (data not shown). In agreement with a previous study,²⁵ we found that the majority (~60%) of inputs to PMNs originated from the brainstem, mainly from the rVRG, consistent with rVRG being the major driver of PMN activation (Figures 1B and 1C). In the brainstem, inputs were evenly distributed across the ipsilateral (to the injection site, 49.6%) and contralateral (50.4%) sides (Figure S2C). In addition, we also observed substantial (~40%) PMN inputs originating from the spinal cord (Figures 1B, 1C, 1G, 1L, S2F, and S2G; Video S1). Among spinal cord inputs, the large majority (~80%) were located at cervical levels (C1-C5). We also observed a small number of ascending interneurons from the brachial (C6-C8, ~10%) and thoracic (~10%) spinal cord (Figure 1D). Unlike brainstem inputs, spinal cord input neurons were found mostly ipsilateral to the injection site (80%, Figure 1E). In addition, most spinal cord neurons projecting to PMNs were localized either in the ventral (42.1%, Figure S2F) or intermediate (48.5%) spinal cord, although monosynaptic inputs from the dorsal spinal cord were also observed (9.4%, Figures 1F and S2G). We calculated the PMN connectivity index (number of mCherry+ neurons/starter cell) and found that a single PMN can receive input from tens of neurons, ranging from a dozen to over a hundred. However, regardless of the number of total inputs one PMN receives, input neurons show a similar distribution throughout the brainstem and spinal cord (Figure S2D). Overall, we found that spinal interneurons are a major source of monosynaptic inputs to PMNs.

To further investigate the identity of spinal monosynaptic inputs to PMNs, we examined the neurotransmitter profile of mCherry+ interneurons. We found that cholinergic interneurons (ChAT+ INs), usually located around the central canal of the intermediate spinal cord (Figures 1G and 1L), contributed around 10% of total inputs (25% of spinal inputs) to PMNs (Figure 1I), and the connectivity index for ChAT+ INs for individual injections was invariably ~1/10 of the total input (Figures S2D and S2E). While ChAT+ INs accounted for ~50% of inputs from the intermediate spinal cord, inputs from both the dorsal and ventral spinal cord were derived exclusively from ChAT− INs (Figures S2F–S2H). Similar to the distribution of other spinal inputs, 77% of PMN-projecting ChAT+ INs were from the ipsilateral side (Figure 1J), while 77.6% were located in the cervical C1-C5 spinal cord, largely overlapping with the rostrocaudal distribution of PMNs, and consistent with the overall distribution of total spinal inputs (Figures 1H and 1K). We also observed a small subset of ascending ChAT+ INs at brachial and thoracic levels (Figures 1H, 1K, and 1L), suggesting that these interneurons may mediate communication with other respiratory and non-respiratory MNs. Along the rostrocaudal extent of the spinal cord, ChAT+ INs accounted for 25%, 5%, and 33% of inputs from the cervical, brachial, and thoracic spinal cord, respectively (Figure S2I). Taken together, our rabies tracing experiments demonstrate that spinal ChAT+ INs provide significant input to PMNs.

PMN-projecting spinal ChAT+ INs are morphologically and topographically distinct

Our tracing experiments revealed that ~10% of PMN inputs correspond to a subset of ChAT+ INs in the cervical spinal cord. This contrasts limb-innervating MNs (LMNs), which receive extensive input from multiple populations of excitatory and inhibitory spinal interneurons, with only about 2% of their inputs originating from ChAT+ INs.^42,43 This biased connectivity suggests that ChAT+ INs likely have important modulatory roles in respiratory behaviors. To investigate whether distinct spinal ChAT+ INs project to different MN subtypes, we injected RabiesΔG-mCherry virus into a representative limb muscle, the biceps, to label ChAT+ INs that project to LMNs and compared their distribution and morphology with PMN-projecting ChAT+ INs (Figures 2A–2D). To analyze the topographical distribution of ChAT+ INs, each mCherry+ ChAT+ IN was assigned a Cartesian coordinate, with the midpoint of the spinal cord midline defined as (0,0). Interestingly, we found that PMN-projecting ChAT+ INs were on average located closer to the central canal than LMN-projecting ChAT+ INs (Figures 2E and 2F).

Figure 2. ChAT+ INs that project to PMNs are morphologically and topographically distinct.

(A–D) Transsynaptic retrograde labeling of ChAT+ INs projecting to PMNs (ChAT+ INs → PMNs) and limb (biceps)-innervating MNs (ChAT+ INs → LMNs). Representative images of contralateral (B) and ipsilateral (C) ChAT+ INs → PMNs and ChAT+ INs → LMNs (D). Scale bar, 100 μm. White arrowheads indicate mCherry+ cholinergic interneurons.

(E) Topographical distribution of ChAT+ INs → PMNs (n = 86 cells from 7 mice) and ChAT+ INs → LMNs (n = 69 cells from 5 mice). Rectangular region is enlarged to the right.

(F) Quantification of ChAT+ INs → PMNs and ChAT+ INs → LMNs horizontal distance to the central canal.

(G and H) Reconstruction of ChAT+ INs → PMNs (G) and ChAT+ INs → LMNs (H) morphology. Scale bar, 50 μm.

(I) Sholl analysis of ChAT+ INs → PMNs (n = 36 cells) and ChAT+ INs → LMNs (n = 35 cells). Solid lines indicate the mean of the group.

(J–N) ChAT+ INs → PMNs have higher maximum Sholl intersections (J), greater overall dendritic length (K), cover a larger area (L), and have higher maximum branch level (M) and greater maximum branch depth (N) compared with ChAT+ INs → LMNs. Solid lines indicate the mean and dashed lines indicate the 25th and 75th percentiles. **p < 0.01, ****p < 0.0001. See also Figure S3.

Next, we used Imaris software to reconstruct and examine the dendritic morphology of pre-motor ChAT+ INs by filament analysis (Figures S3A, 2G, and 2H). First, to investigate whether there is diversity within PMN-projecting ChAT+ INs, we traced both contralateral and ipsilateral populations and compared their dendritic morphology. While all PMN-projecting ChAT+ INs had comparable dendritic length, area, and maximum branch level and depth, contralateral ones branched more at proximal dendritic levels and had a higher number of maximum Sholl intersections (Figures S3B–S3G). In addition, diverse dendritic orientation patterns were observed within PMN-projecting ChAT+ INs (Figure 2G), suggesting that some morphological diversity exists even among ChAT+ INs targeting the same MN population.

We next compared the morphologies of ChAT+ INs projecting either to PMNs or LMNs. We found that PMN-projecting ChAT+ INs had more Sholl intersections, especially at proximal dendrites, a higher number of maximum Sholl intersections, higher maximum branch level and depth, greater filament length (i.e., overall dendritic length), and covered a larger area (Figures 2I–2N), indicating that they have a more complex dendritic morphology than ChAT+ INs targeting LMNs. Collectively, our findings demonstrate that ChAT+ INs that project to PMNs are topographically and morphologically distinct.

Pitx2+ V0_C neurons are the source of cholinergic synapses on PMNs

Large cholinergic synapses localized on the cell body and proximal dendrites of MNs are known as C boutons.⁴⁴ To define whether cholinergic inputs on PMNs are analogous to the C boutons observed on other MN subtypes, we investigated the distribution of cholinergic synapses on PMNs. Individual PMNs were traced by unilateral injection of RabiesΔG-mCherry virus into the diaphragm of control mice which lack G protein expression, and therefore allowed RabiesΔG-mCherry virus to function as a retrograde tracer of only PMNs (i.e., no transsynaptic labeling). Cholinergic inputs on a single PMN were identified by immunostaining for vesicular acetylcholine transporter (VAChT). We found that the majority of cholinergic synapses were located on the PMN soma (~30%) and proximal dendrites (~60%), with <10% found on PMN distal dendrites (Figure S4), similar to the distribution of C boutons on other MNs and consistent with ultrastructural studies.⁴⁵ The number of putative C boutons on PMN cell bodies increased over time during early postnatal stages, but remained fairly unchanged from P4 to adulthood, indicating that PMNs receive consistent cholinergic input (Figure S5C).

Cholinergic V0 interneurons (V0_C) that are derived from the Dbx1+ p0 progenitor domain and express Pitx2 post-mitotically are the sole source of C boutons on MNs.⁴⁶ To investigate whether cholinergic inputs to PMNs are also derived from V0_C interneurons, we genetically labeled Pitx2+ interneurons using Pitx2::Cre; ROSA26Sor^{tm9(CAG-tdTomato)} (Pitx2^tdTom)^47,48 mice. We found that over 90% of Pitx2-expressing cells were tdTomato+ at e16.5, indicating robust recombination in Pitx2^tdTom mice (Figures S5A and S5B). We labeled PMNs by intrapleural injection of cholera toxin subunit B in adult Pitx2^tdTom mice (Figures 3A and 3B)⁴⁹ and found that VAChT+ C boutons on individual PMNs were colocalized with tdTomato (Figures 3C and 3D), indicating that Pitx2+ V0_C interneurons are the source of cholinergic synapses on PMNs. To test whether all PMN C bouton inputs are derived from Pitx2+ interneurons, we quantified the overlap between VAChT and tdTomato in Pitx2^tdTom mice. We found that by P4 over 90% of C boutons on PMN cell bodies were tdTomato+, and this was maintained into adulthood (P60) (Figures 3E–3N, and S5D). Given that the recombination efficiency in Pitx2^tdTom mice is ~95%, we conclude that Pitx2+ V0_C neurons are the only source of PMN C boutons.

Figure 3. Pitx2-derived cholinergic synapses on PMNs.

(A) Transverse section of the cervical spinal cord showing retrograde cholera toxin subunit B (CTB) labeling (green) in PMNs (squared region). Scale bar, 100 μm.

(B) Enlargement of CTB-labeled PMNs (green) shown in (A). Square region indicates a PMN shown in (C). Scale bar, 20 μm.

(C) Enlargement of a single CTB-labeled PMN (CTB, green) shown in (B) from a Pitx2^tdTom adult mouse with synapses derived from Pitx2^Tdtom+ interneurons (red) and cholinergic synapses (VAChT, blue). Numbers indicate labeled synapses on the PMN soma. Scale bar, 5 μm.

(D) Magnification of numbered synapses from (C) showing colocalization of tdTomato and VAChT. Scale bar, 1 μm.

(E and F) Representative images of VAChT+ puncta on PMNs in P4 Pitx2^tdTom mice. Square region in (E) is enlarged in (F). Scale bars, 100 μm (E) and 20 μm (F).

(G) Enlargement of the square region in (F). Numbers indicate examples of synapses on the PMN soma. Scale bar, 5 μm.

(H) Magnification of numbered synapses from (G). Scale bar, 1 μm.

(I) Percentage of VAChT+ puncta that are Pitx2^tdTom+ on PMN somas at P4.

(J–N) As in (E–I) but at P60. See also Figures S4 and S5.

A subset of cervical Pitx2+ interneurons are integrated within respiratory circuits

Having revealed that a subset of Pitx2+ V0_C neurons near the central canal is anatomically connected to PMNs, we next addressed whether Pitx2+ interneurons are functionally integrated within respiratory circuits. This was first investigated by assessing whether their synaptic inputs and action potential output are correlated with respiratory network activity. We performed whole-cell patch-clamp recordings from Pitx2+ interneurons located near the central canal, identified by tdTomato expression in Pitx2^tdTom mice, in combination with extracellular recordings of C3/C4 ventral roots in mid-sagittal-hemisected brainstem spinal cord preparations obtained from neonatal (P3-P4) Pitx2^tdTom mice (n = 5 preparations; Figure 4A). Analysis of synaptic inputs recorded in voltage-clamp mode demonstrated that a subset of Pitx2+ interneurons within C3-C4 (50%, n = 8, Figure 4B) receive synaptic inputs (frequency = 43 ± 8 Hz, amplitude = 122 ± 50 pA) that are phase locked with the respiratory motor output recorded from ventral roots (Figure 4C; phase = 2.25° ± 0.3°). Current-clamp recordings revealed that respiratory-related Pitx2+ interneurons exhibited either a depolarization of the membrane potential (n = 2, amplitude = 7.14 mV) or bursts of action potential firing (n = 2, frequency = 9.7 Hz) that were also phase locked with respiratory-related ventral root output (Figure 4D; phase = 2.6° ± 0.5°). These results indicate that at least a proportion of Pitx2+ interneurons located within the C3-C4 cervical segments receive respiratory-related inputs and produce respiratory-related output. Interestingly, we did not find any respiratory-related Pitx2+ interneurons in more caudal cervical segments (C5-C7, n = 10, Figure 4B), suggesting that Pitx2+ interneurons in the region of PMNs are more likely to be integrated within respiratory circuitry, consistent with the distribution of PMN-projecting ChAT+ INs (Figure 1H). Overall, these results confirm that spinal Pitx2+ interneurons are not only anatomically connected to PMNs but also functionally integrated within respiratory circuits.

Figure 4. A subset of cervical Pitx2+ interneurons are integrated within respiratory circuits.

(A) Schematic of the experimental setup showing extracellular ventral root recordings and intracellular whole-cell patch-clamp recordings from individual Pitx2^tdTom+ interneurons (red) in hemisected brainstem-spinal cord preparations from Pitx2^tdTom neonatal mice.

(B) Pie charts showing the relative proportion of respiratory-related (red) and non-respiratory-related (gray) Pitx2+ interneurons within and below the C3–4 spinal segments.

(C) Example trace of voltage-clamp recording from a respiratory-related Pitx2+ interneuron (top) during ongoing respiratory burst (bottom). Red dotted box showing zoomed-in traces during a respiratory burst.

(D) As in (C) but regarding current-clamp recording.

Spinal ChAT+ INs modulate respiratory motor output

We next assessed the role of cervical Pitx2+ interneurons in modulating respiratory-related PMN output. This was achieved by pharmacologically blocking transmission at C bouton synapses and measuring any subsequent effects on respiratory network output recorded from the C3/C4 ventral roots of isolated brainstem spinal cord preparations obtained from neonatal (P2-P4) mice (Figure 5A). We utilized the M2 muscarinic receptor antagonist, methoctramine, because M2 receptors are the primary postsynaptic target for acetylcholine released at C bouton synapses.⁵⁰ In line with C boutons playing a role in facilitating motor output, we found that blocking M2 receptors with methoctramine (10 μM; n = 7 preparations; Figure 5B) reversibly reduced the amplitude of respiratory-related activity (19% ± 3% reduction, p = 0.0100; Figure 5C). This reduction in amplitude was paralleled by an increase in the frequency of bursting (55% ± 13% increase, p = 0.0107; Figure 5D).

Figure 5. Effect of methoctramine on the respiratory motor output.

(A) Brainstem-spinal cord preparation from neonatal mice.

(B) Raw (top) and integrated/rectified (bottom) traces from the C4 ventral root during baseline, methoctramine (10 μM), and washout. Boxes indicate 40 s of recording at the end of each condition (black, baseline; red, methoctramine; gray, washout), expanded at the bottom.

(C) Average respiratory-burst amplitude over the last 10 min during baseline (black), 10 μM methoctramine (red), and washout (gray); black lines show mean and SEM (n = 7).

(D) As in (C) but showing respiratory-burst frequency.

(E) Experimental design to block M2 receptors at spinal levels only in the brainstem-spinal cord preparation from neonatal mice.

(F) As in (B) but showing the effect of methoctramine at spinal level only.

(G and H) As in (C) and (D) but showing the effect of methoctramine at spinal level only (n = 10).

(I) Working heart-brainstem preparation from adult rats.

(J) As (B) and (F) but showing the phrenic neurogram trace from the working heart-brainstem preparation. Square boxes indicate 7 s of recording at the end of each condition (black, baseline; red, methoctramine; gray, washout), expanded at the bottom.

(K and L) As in (C) and (D) but showing the effects of methoctramine in the adult preparation. Data analyzed with mixed-effect model and Holm-Šídák’s multiple comparisons test. *p < 0.05, **p < 0.001. See also Figure S6.

Since our experiments involved blockade of M2 receptors throughout the brainstem and spinal cord, we next wanted to delineate the specific contribution of spinal cord interneurons to cholinergic modulation of respiratory motor output. This was achieved by using a “split bath” preparation in which the brainstem and spinal cord compartments were perfused separately (Figure 5E). This enabled us to block M2 receptors in the spinal cord only. We hypothesized that the reduction in the amplitude of respiratory-related output when methoctramine was applied to whole preparations could be explained, at least in part, by blockade of M2 receptors at C bouton synapses on PMNs. In line with our hypothesis, application of methoctramine (10 μM; n = 10 preparations; Figure 5F) to the spinal compartment of split bath preparations only led to a reversible reduction in the amplitude of respiratory-related output recorded from ventral roots (20% ± 6% reduction, p = 0.0120; Figure 5G), with no change in the frequency of bursting (12% ± 4% increase, p = 0.1221; Figure 5H).

Given that our in vitro experiments relied on isolated brainstem spinal cord preparations that are only viable when obtained from neonatal animals, we next extended our analysis to the working heart-brainstem preparation, which can be obtained from adult rodents (Figure 5I). In preparations obtained from adult rats, we recorded respiratory-related output from the phrenic nerve with extracellular electrodes while methoctramine (10 μM; n = 9 preparations; Figure 5J) was applied to the perfusate. Consistent with our recordings from neonatal tissue, we again found that methoctramine caused a reversible reduction in the amplitude of respiratory-related output recorded from the phrenic nerve (57% ± 5% reduction, p = 0.0294; Figure 5K). We did not observe any change in the frequency of phrenic nerve discharge upon methoctramine administration (2% ± 5% increase, p = 0.6432; Figure 5L). Interestingly, we found that methoctramine also elicited a reduction in the amplitude of respiratory-related activity recorded from external intercostal muscles via electromyography (44% ± 6% reduction, p = 0.007, Figure S6).

Taken together, these results confirm the existence of a spinal cholinergic pathway, acting via M2 receptors, that modulates the amplitude of PMN output in both neonatal and adult rodents. These data are consistent with modulation of respiratory motor output by local Pitx2+ interneurons and their C bouton contacts with PMNs.

Spinal cholinergic interneurons are activated under hypercapnia

Given the functional integration of Pitx2+ interneurons into respiratory circuits and their ability to alter PMN output in both in vitro neonatal isolated brainstem-spinal cord preparations and in situ adult working heart-brainstem preparations, we next investigated whether Pitx2+ V0_C neuron activity can modulate respiratory behaviors in vivo. In order to visualize V0_C activation, we utilized ChAT::eGFP mice, in which both ChAT+ INs and PMNs are labeled by GFP but can be distinguished by their locations. While PMNs are clustered in the ventral horn of the cervical spinal cord, ChAT+ INs are scattered throughout the spinal cord, with the distinct PMN-projecting Pitx2+ V0_C subset localized near the central canal (Figure 6A).

Figure 6. ChAT+ INs are activated under a hypercapnic gas challenge.

(A–C) (A) ChAT+ MNs and INs are labeled by GFP in ChAT::eGFP mice. Regions including PMNs and ChAT+ INs around the central canal are shown in (B) and (C), respectively. (B and C) Both PMNs (B) and ChAT+ INs (C) are activated, as indicated by high c-Fos expression (red), after exposure to 10% CO₂. White arrowheads indicate c-Fos+ ChAT+ INs. Scale bar, 40 μm.

(D) PMN activation is positively correlated to ChAT+ IN activation.

(E, G, and I) Percentage of c-Fos+ PMNs after 10% CO₂ for 1 h (E), 10% CO₂ for 15 min (G), and 5% CO₂ for 15 min (I).

(F, H, and J) Number of c-Fos+ ChAT+ INs after 10% CO₂ for 1 h (F), 10% CO₂ for 15 min (H), and 5% CO₂ for 15 min (J). *p < 0.05, **p < 0.01, ***p < 0.001. Data are represented as mean ± SEM. See also Figure S7.

In locomotor circuits, Pitx2+ V0_C interneurons increase MN excitability to ensure that a sufficient motor output is generated during demanding tasks, such as swimming.^46,51 We therefore hypothesized that cervical V0_C interneurons may contribute to increasing respiratory output during environmental or behavioral conditions that are associated with increased metabolic demand. To measure neuronal activation under a respiratory challenge, we exposed ChAT::eGFP mice to hypercapnic conditions in a plethysmography chamber. Under atmospheric air conditions (79% N₂, 21% O₂), only 30% of PMNs were activated, as indicated by the expression of the early immediate gene c-Fos. In contrast, more than 60% of PMNs were activated under either intense (10% CO₂ for either 1 h or 15 min) or moderate (5% CO₂ for 15 min) hypercapnic conditions (Figures 6B, 6E, 6G, and 6I). Moreover, the c-Fos mean intensity in PMNs was significantly increased after exposure to 10% CO₂ (Figures S7B and S7C), indicating that both the number of recruited PMNs and single PMN activity are increased under an intense hypercapnic challenge, as previously described.^52,53 CO₂ exposure does not lead to broad non-specific activation of spinal cord neurons, as we found that the number of c-Fos+ cells at cervical levels of the spinal cord, excluding MNs, was comparable with baseline levels (i.e., air exposure) after a 10% CO₂ hypercapnic challenge for 1 h (Figure S7A).

Since hypercapnia highly activates PMNs, it can serve as a powerful paradigm to examine whether spinal interneurons modulate PMN activation under an environmental challenge. We found that ChAT+ INs were highly activated under an intense (15 min or 1 h of 10% CO₂; Figures 6C, 6F, and 6H), but not under a moderate (15 min of 5% CO₂) hypercapnic challenge (Figure 6J), suggesting that ChAT+ IN activation and recruitment is dependent on the intensity level of the stimulus. c-Fos+ ChAT+ INs were located in close proximity to the central canal at cervical levels of the spinal cord, and thus are likely to directly project to PMNs and correspond to the Pitx2+ interneurons with respiratory-related activity in our in vitro recordings. While the number of activated ChAT+ INs recruited under an intense hypercapnic challenge increased, the c-Fos mean intensity in ChAT+ INs, unlike in PMNs, did not significantly change (Figures S7B–S7E), suggesting that recruitment of additional ChAT+ INs may contribute to increased PMN activation. Consistent with this idea, we found that the number of activated ChAT+ INs was positively correlated to the percentage of activated PMNs (Figure 6D).

Cholinergic interneuron silencing impairs the response to hypercapnia

Next, we set out to determine the contribution of ChAT+ INs to the hypercapnia response. We hypothesized that ChAT+ INs may promote PMN activation to drive increases in breathing amplitude during hypercapnia. To test this, we utilized a two-chamber whole-body plethysmography system to measure the breathing patterns in mice in which cholinergic neurotransmission has been removed from spinal ChAT+ INs using Cre/lox genetic strategies (Figure S8A). We initially utilized ChAT^flox/flox; Dbx1::Cre (Dbx1^ΔChAT) mice to ensure early and efficient ChAT deletion, as Dbx1::Cre targets p0 progenitors that give rise to Pitx2+ V0_C interneurons. We validated Dbx1^ΔChAT mice by examining the expression of ChAT in VAChT+ puncta on PMNs in adult (P120) mice. We found that over 95% of VAChT+ terminals on PMNs were devoid of ChAT expression in Dbx1^ΔChAT mice, indicating that cholinergic transmission from V0_C neurons to PMNs was largely eliminated in these mice (Figures S8B–S8D). We did not observe any changes in the number of VAChT+ synapses on PMN somas in adult Dbx1^ΔChAT mice (Figure S8E).

We then performed plethysmography experiments in adult Dbx1^ΔChAT mice and their paired control littermates (ChAT^flox/flox or ChAT^flox/+). Each pair of mice was subjected to 45 min of normal air (79% N₂, 21% O₂) followed by 15 min of either 10% CO₂ (10% CO₂, 69% N₂, 21% O₂) or 5% CO₂ (5% CO₂, 74% N₂, 21% O₂) (Figure 7A). When exposed to a hypercapnia challenge, control mice increase both their respiratory frequency and depth. Consistent with this, we found a significant increase in both frequency and tidal volume (the amount of air inhaled during a normal breath), resulting in a ~3-fold increase in minute ventilation (the volume of air inhaled per minute) in control mice after exposure to a 10% CO₂ hypercapnic challenge (Figures S9A–S9H). When comparing Dbx1^ΔChAT mice and their control littermates, we found that they exhibited similar breathing behaviors under normal air conditions. All breathing parameters, including frequency, minute ventilation, tidal volume, and peak inspiratory and expiratory flow showed similar distribution in Dbx1^ΔChAT and control mice (Figures S9C, S9E, S9G, S9I, and S9K). However, after exposure to 10% CO₂, the distribution peaks for all respiratory parameters, excluding frequency, were shifted to a lower value in Dbx1^ΔChAT mice, indicating that these mice had a compromised response to hypercapnia (Figures S9D, S9F, S9H, S9J, and S9L). After normalization to their paired control littermates, Dbx1^ΔChAT mice had significantly decreased tidal volume (~11% reduction, p = 0.0291), minute ventilation (~13% reduction, p = 0.0150), peak inspiratory flow (PIF) (~15% reduction, p = 0.0035), and peak expiratory flow (PEF) (~11% reduction, p = 0.0281) compared with control mice during the hypercapnia challenge (Figures S9F, S9H, S9J, and S9L), consistent with the altered distribution of respiratory parameters we observed.

Figure 7. Cholinergic interneuron silencing impairs the response to hypercapnia.

(A) Experimental setup for whole-body plethysmography and schematic of a representative breath. A Pitx2^ΔChAT or Dbx1^ΔChAT mouse was paired with a sex-matched control littermate and the mice were exposed to normal air conditions (21% O₂, 79% N₂) for 45 min, followed by 15 min of hypercapnia (5% CO₂, 21% O₂, 74% N₂ or 10% CO₂, 21% O₂, 69% N₂).

(B and C) Examples of breath traces under normal air and 10% CO₂ in Pitx2^ΔChAT mice and their control littermates. A single breath was enlarged in (C).

(D, H, L, and P) Breath frequency (D), minute ventilation (H), tidal volume (L), and PIF (P) distribution under normal air in Pitx2^ΔChAT and control mice (n = 14–17 per group).

(E, I, M, and Q) Mean and normalized frequency (E), minute ventilation (I), tidal volume (M), and PIF (Q) under normal air in Pitx2^ΔChAT and control mice.

(F, J, N, and R) Breath frequency (F), minute ventilation (J), tidal volume (N), and PIF (R) distribution under 10% CO₂ in Pitx2^ΔChAT and control mice (n = 14–17 per group).

(G, K, O, and S) Mean and normalized frequency (G), minute ventilation (K), tidal volume (O), and PIF (S) under 10% CO₂ in Pitx2^ΔChAT and control mice. See also Figures S8–S10.

Since Dbx1 is broadly expressed at early embryonic stages, we wanted to further restrict our manipulations to V0_C postmitotic neurons. Therefore, we repeated our plethysmography experiments in ChAT^flox/flox; Pitx2::Cre (Pitx2^ΔChAT) mice, which restricted ChAT deletion to Pitx2+ interneurons. Similar to Dbx1^ΔChAT mice, over 95% of VAChT+ terminals on PMNs did not express ChAT in adult Pitx2^ΔChAT mice, indicating efficient ChAT deletion from V0_C neurons (Figures S8F–S8H). We did not observe a decrease in VAChT+ terminals on PMNs in these mice (Figure S8I). We exposed Pitx2^ΔChAT mice and their control littermates to both normal air and hypercapnic conditions (Figure 7A). Since we found that the activation of ChAT+ INs was intensity dependent (Figure 6), we first investigated how blocking cholinergic neurotransmission from V0_C interneurons might impact breathing under a moderate hypercapnic challenge. Interestingly, when we exposed Pitx2^ΔChAT mice to either normal air or 5% CO₂, their breathing patterns were indistinguishable from their control littermates, suggesting that cholinergic modulation of PMNs might be increasingly important under more intense hypercapnic challenges (Figures 7D, 7E, 7H, 7I, 7L, 7M, 7P–7Q, and S10).

Consistent with our observations in Dbx1^ΔChAT mice, the tidal volume, minute ventilation, PIF, and PEF distributions were all shifted toward lower values (i.e., more breaths having a lower value) under 10% CO₂ hypercapnia, but not under normal air conditions, in Pitx2^ΔChAT mice (Figures 7B–7D, 7F, 7H, 7J, 7L, 7N, 7P, 7R, S9M, and S9N). In addition, the tidal volume (~8% reduction, p = 0.0033), minute ventilation (~8% reduction, p = 0.0011), PIF (~9% reduction, p = 0.0007), and PEF (~7% reduction, p = 0.0104), but not the breathing frequency (<1% reduction, p = 0.6246), were significantly decreased in Pitx2^ΔChAT mice under 10% CO₂ (Figures 7E, 7G, 7I, 7K, 7M, 7O, 7Q, 7S, and S9N). Taken together, our findings indicate that cholinergic modulation of PMNs, through direct C bouton contacts from spinal Pitx2+ V0_C interneurons, increases PMN output under a hypercapnia challenge, revealing a function for spinal cholinergic interneurons in the adaptive control of breathing.

DISCUSSION

PMNs generate the final output of respiratory motor circuits and have traditionally been thought to eschew inputs from the spinal cord and act as executioners of brainstem motor commands. Here, we combined mouse genetics, rabies-mediated viral tracing, electrophysiology, and behavioral experiments to demonstrate the role of a subset of spinal cholinergic interneurons in the facilitation of PMN output and increase in tidal volume in response to hypercapnia. Our data suggest that, far from being a static relay station for brainstem motor commands, PMNs integrate a range of modulatory inputs to match motor output to environmental or metabolic demands. Below, we discuss our findings in the context of PMN modulation, respiratory circuits, spinal interneuron diversity, and potential roles in promoting recovery following SCI.

Spinal cord interneurons and the regulation of breathing

While spinal interneurons with projections to PMNs have been anatomically and electrophysiologically described in multiple species, the contribution of genetically defined classes of spinal interneurons to distinct aspects of breathing remains unclear. Previous mapping of PMN inputs through transsynaptic viral approaches has revealed varying amounts of spinal interneuron inputs, ranging from substantial in adult rats to negligible in neonatal mice.^20,23,25,27 In addition to species differences, other potential sources of variability may be the tropism of the different viruses used (e.g., PRV vs. rabies) or the age of the injected animals (neonatal vs. adult), suggesting temporally dynamic inputs to PMNs that may be developmentally gained or lost. In addition to anatomical studies, microelectrode and multielectrode array recordings from the cervical spinal cord have identified a number of interneurons with both inspiratory- and expiratory-related activity, indicating that complex spinal circuits may be involved in PMN modulation downstream of brainstem circuits.^{31,32,54–58} However, unlike well-described functions for cardinal interneuron classes in locomotion, mapping respiratory functions to these populations has been elusive. For example, Renshaw cells that respond to PMN stimulation and are spontaneously active during inspiration have been identified but appear to be rare.^59–61 Ablation or inhibition of V2a neurons changes the breathing frequency, but these effects are mediated primarily through the brainstem rhythm-generating pre-Botzinger complex and accessory respiratory muscles rather than PMN modulation.^62,63 We find that Pitx2+ V0_C interneurons directly project to PMNs, produce respiratory-related output, and modulate breathing amplitude under an environmental challenge, providing both anatomical and functional evidence implicating a genetically defined spinal cord interneuron population in PMN modulation and the adaptive control of breathing in healthy animals.

Integration of cholinergic interneurons into respiratory circuits

What are the inputs to spinal respiratory cholinergic interneurons, and how do they fit into the broad respiratory network? Putative limb MN-projecting cholinergic interneurons in the lumbar spinal cord increase the excitability of MNs involved in locomotion to ensure robust firing and sufficient MN output during demanding tasks such as swimming.^46,51,64 These interneurons receive input from corticospinal neurons, locomotor central pattern generator circuits, descending serotonergic inputs, and polysynaptic inputs from sensory afferents to adjust their activity and increase muscle contraction amplitude.^46,65 While respiratory V0_C neurons may also receive synaptic inputs from these sources, their activation by elevated CO₂ levels in the absence of locomotor activity suggests alternative or additional inputs. One possibility is that they receive inputs from chemosensitive areas in the brainstem such as the retrotrapezoid nucleus or serotonergic neurons that stimulate breathing in response to elevated CO₂ levels^5,66; in fact, we do observe serotonergic axons in close proximity to mCherry+ ChAT+ INs in our rabies tracing experiments. In addition, respiratory V0_C neurons may receive inputs from pre-phrenic respiratory areas such as the rVRG; since cholinergic interneurons augment PMN output, the two populations might share common inputs. Previous studies have detected rVRG axons around cervical pre-phrenic interneurons,^23,67 and our recordings, which show synaptic currents phase locked with respiratory output, support this hypothesis. In addition to Pitx2+ V0_C interneurons, we also identified a number of non-cholinergic PMN-projecting interneurons (Figures S2F and S2G) mostly localized in the cervical spinal cord, although it is worth noting that these are much less numerous than spinal cord interneurons projecting to limb MNs (compare Figures 2B and 2C with 2D).^42,43,68 Altogether, our data suggest a more prominent contribution of spinal cord interneurons to the regulation of breathing than previously suggested.

While c-Fos experiments have limitations in detecting activation patterns with fine temporal resolution, we find that a hypercapnia challenge increases both the number of c-Fos expressing PMNs and the intensity of c-Fos expression, presumably corresponding to increases in both MN recruitment and firing rate. Conversely, only the number of c-Fos expressing ChAT+ INs changed, suggesting that V0_C recruitment is the predominant modality for increasing PMN gain and diaphragm output, similar to the recruitment of V2a interneuron subtypes to generate high-speed movements.^69,70 Both rabies virus tracing and C bouton quantification at PMN cell bodies reveal that there may be some variability in the number of V0_C inputs to each PMN. Although we did not attempt to correlate C bouton number to cell body size, one possibility is that large, fast-fatigable PMNs receive more C bouton inputs. Fast MNs have been shown to have a higher density of C boutons and we have previously found greater effects of muscarine on fast MN output.^64,71,72 In addition to the hypercapnia response, V0_C recruitment and PMN modulation might become increasingly important for augmenting PMN output during expulsive behaviors, such as coughing and sneezing, and exercise, when even greater PMN activation is required.⁷³

Cholinergic interneuron diversity

While comprising a relatively small subset of interneurons within the spinal cord, V0_C interneurons exhibit remarkable molecular and anatomical diversity, and extensively innervate spinal MNs. Despite their extensive projections to MNs, there seems to be some selectivity in their targeting. For example, ocular and cremaster MNs lack cholinergic inputs, while MNs innervating large proximal muscles receive greater numbers of inputs than those innervating small distal muscles.^71,72,74,75 In addition, postsynaptic clustering of certain ion channels differs among cholinergic synapses on different MN subtypes.⁷⁶ Within locomotor circuits, there are at least two populations of V0_C neurons, one of which projects exclusively to ipsilateral targets while another projects bilaterally, indicating that distinct programs underlie their connectivity.^42,46 Recent single-nucleus RNA-seq analysis of cholinergic interneurons in adult mice identified eight transcriptionally distinct clusters, two of which expressed Pitx2, further supporting the idea that there is significant diversity within this population, despite their common developmental origin from Dbx1-expressing progenitors.⁷⁷ Therefore, it is likely that distinct cholinergic interneurons innervate specific MN subtypes to mediate unique functions. We now show that V0_C interneurons participate in diverse behavioral responses, modulating MN output to control limb movements as well as breathing.

While our data clearly point to the existence of a subset of respiratory-related V0_C neurons, we cannot definitively determine whether these interneurons exclusively project to PMNs or also target limb MNs as dually projecting cholinergic interneurons have been observed in adult rats.⁷⁸ Interestingly, we found that, in adult perfused preparations, methoctramine also elicited a reduction in the amplitude of respiratory-related activity in external intercostal muscles, although we did not identify the source of this modulation. Therefore, it is likely that cholinergic modulation of multiple respiratory MN populations contributes to breathing regulation. One possibility is that V0_C neurons at thoracic levels may project to both phrenic and intercostal MNs to coordinately increase motor output under a respiratory challenge such as hypercapnia.

Implications for SCI and respiratory plasticity

While the role of spinal interneurons in the canonical control of breathing has remained somewhat elusive, recent studies have highlighted their critical function in respiratory plasticity and recovery after SCI. For instance, animal studies have demonstrated that changes in the function or connectivity of propriospinal neurons are crucial for improving breathing function during both acute and chronic stages of injury.^{33–35,38,39} Excitatory interneurons are particularly important for recovery, as evidenced by models of non-traumatic cervical myelopathy³⁴ and C2 hemisection SCI.³³ Specifically, in C2 hemisection SCI, increased connectivity between V2a interneurons and PMNs has been linked to the spontaneous recovery of breathing.^79,80 Activation of V2a interneurons can restore function in the paralyzed hemidiaphragm following a C2 hemisection injury, while silencing these neurons significantly hinders recovery.³⁵ Although the role of ChAT+ INs in recovering breathing function after SCI remains to be elucidated, their importance in neurodegenerative conditions is suggested by studies of V0_C neurons in locomotor circuits in ALS.^81–83 Given our findings supporting a role for V0_C interneurons in facilitating diaphragm activation during hypercapnia, they represent a promising neuronal population that can be coopted to counteract the reduced hypercapnic drive response seen in patients with cervical SCI⁸⁴ and enhance recovery after injury.

Limitations of the study

Our study identified a spinal interneuron population that amplifies breathing under hypercapnia. However, we did not elucidate the mechanisms that underlie the response of these interneurons to elevated CO₂ and whether they receive inputs from chemosensory neurons or are intrinsically chemosensitive. Cholinergic neuron inactivation did not abolish the hypercapnia response, suggesting that this pathway acts in parallel to previously characterized chemosensory pathways^9,85 and consistent with a modulatory function. Since our plethysmography experiments involved chronic inactivation of cholinergic neurotransmission in V0c neurons, it is possible that there was some level of compensation over time, underestimating the role of this population in the hypercapnia response. Acute inhibition or activation of cholinergic interneurons may reveal a more substantial contribution to breathing adaptations. Finally, we did not investigate the role of this interneuron population in recovering respiratory function after SCI.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to the lead contact, Polyxeni Philippidou (pxp282@case.edu).

Materials availability

This study did not generate unique reagents.

Data and code availability

• The data generated in this study are available from the lead contact upon request.

• The code used for plethysmography analysis is available at GitHub (Zenodo: https://doi.org/10.5281/zenodo.15732816).

• 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

RphiGT (JAX# 024708),⁴⁰ ROSA26Sor^{tm9(CAG-tdTomato)} (Ai9, JAX# 007909),⁴⁷ ChAT::eGFP (JAX# 007902),⁸⁷ ChAT^flox/flox (JAX# 016920),⁸⁸ ChAT::Cre,⁴¹ Dbx1::Cre,⁸⁹ and Pitx2::Cre⁴⁸ lines were generated as previously described and maintained on a mixed background. No more than five adult mice were housed in a microisolator cage at one time under a 12-h light/dark cycle. Procedures and mouse maintenance performed in the United States were executed in accordance with protocols approved by the Institutional Animal Care Use Committee of Case Western Reserve University (assurance number: A-3145–01, protocol number: 2015–0180). Procedures performed in the United Kingdom were conducted in accordance with the UK Animals (Scientific Procedures) Act 1986, were approved by the University of St Andrews Animal Welfare Ethics Committee and were covered under Project Licences (P6F7B721E and PP8253850) approved by the Home Office. Experiments carried out in Canada were approved by the Animal Care Committee at the University of Calgary (AC19–0037). Both male and female mice were used in this study and we did not observe an influence of sex on the results.

METHOD DETAILS

Rabies-based monosynaptic tracing

RabiesΔG-mCherry virus production and monosynaptic tracing were performed as previously described.^86,91 Briefly, rabies injection solution was made by mixing RabiesΔG-mCherry virus (titer of around 1¹⁰ TU/mL) with silk fibroin (Sigma-Aldrich# 5154)⁹² at a 2:1 ratio. 1–1.5 μL of rabies injection solution was unilaterally injected into the diaphragm or biceps muscle of ChAT::Cre; RphiGT mice at P4 using a nano-injector (Drummond). Mice were sacrificed 7 days post-injection (P11). Specificity of the unilateral injection of the diaphragm/PMN infection was confirmed by checking the mCherry fluorescent signal in the diaphragm and spinal cord (from cervical to lumbar levels, Figure S1C). Fluorescent signal at the ventral roots was used as an indicator of starter MN labeling. No signal at ventral roots outside the C3-C5 spinal cord was detected in animals with phrenic-specific labeling. 100 μm-thick consecutive sections from the cerebrum to the spinal cord were harvested by Leica VT1000S vibratome. Sectioning was stopped if no mCherry+ cells were observed in more than 20 consecutive sections (2 mm). The connectivity index was calculated by dividing the number of mCherry+ labeled cells by the number of starter PMNs in individual animals.

CTB retrograde labeling

PMNs in the cervical spinal cord were retrogradely labeled via trans-thoracic intrapleural injections of cholera toxin B subunit (CTB).^49,93 Animals were anesthetized with 4% isoflurane and depth of anesthesia was confirmed through the absence of withdrawal reflex in response to toe pinch. Mice were positioned sideways with their left side up on a heating pad. The target area on the chest was shaved and disinfected with a warm 10% betadine solution. The injection site was identified as the 5th– 7th intercostal space (at the intersection between an imaginary line drawn from the mouse axilla just behind its elbow, and a line drawn half-way up the chest wall). 10 μL of a 2% (w/v) CTB solution were injected with a 29G needle syringe at the injection site, at a maximum depth of 5 mm from the skin. Mice were allowed to recover in a heated recovery chamber and monitored closely for any sign of respiratory compromise such as that following unintentional pneumothorax. Once recovered, mice were put back in their original cages and monitored daily for 72 h. CTB-injected mice were then perfused and their tissue extracted and used for immunohistochemical analyses.

Immunohistochemistry, confocal microscopy, and image processing

Mice aged older than P12 were anesthetized with either a Ketamine/Xylazine cocktail or Pentobarbital and underwent transcardial perfusion with ice-cold phosphate buffered saline (PBS; pH 7.4 without Ca²⁺ or Mg²⁺) to remove the blood, followed by ice-cold 4% paraformaldehyde (PFA). Neonatal mice were dissected acutely after being anesthetized with Ketamine/Xylazine cocktail. The spinal cord was dissected, incubated in PFA overnight at 4°C, washed in PBS, incubated in 30% sucrose overnight and embedded in Optimal Cutting Temperature (OCT) compound and frozen at −80°C for cryosectioning.

Transverse cryosections (16 or 30 μm) of the cervical spinal cord were obtained using a CM3050S Leica cryostat. For identification and characterization of monosynaptic cholinergic interneurons, 100 μm thick sections were harvested with a Leica VT1000S vibratome. Sections were incubated in PBS containing 1% bovine serum albumin (BSA) and 0.1–0.5% Triton X-100 for 2 h at room temperature. After blocking/permeabilization, tissue was incubated with primary antibodies for overnight to 72 h. After primary incubation, slides were washed 3 times with PBS, followed by incubation with secondary antibodies for 2 h to overnight at room temperature. Finally, slides were washed a further 3 times with PBS, mounted sequentially on Superfrost plus gold glass slides (Thermo Scientific) and let dry before applying the Vectashield Vibrance mounting medium (Vector Laboratories) and cover glasses (VWR).

The following primary antibodies were used in this study: goat anti-ChAT (Sigma, RRID: AB_2079751, 1:300), goat anti-VAChT (Millipore, RRID: AB_2630394, 1:1000), rabbit anti-VAChT (Synaptic Systems, RRID: AB_10893979, 1:500), chicken anti-RFP (Rockland, RRID: AB_10704808, 1:500), rabbit anti-DsRed (Takara Bio, RRID: AB_10013483, 1:1000), rabbit anti-c-Fos (Synaptic Systems, RRID: AB_2905595, 1:1000), goat anti-Scip (Santa Cruz Biotechnology, RRID: AB_2268536, 1:5000), rabbit anti-Pitx2 (1:16000)⁴⁶ and rabbit anti-CTB (Novus Biologicals, RRID: AB_962919, 1:500). Fluorophore-coupled secondary antibodies used were: donkey anti-chicken Alexa Fluor 594 (Sigma, SAB4600094), donkey anti-goat Alexa Fluor 405 (Invitrogen, AB_2890272), donkey anti-goat Alexa Fluor 488 (Invitrogen, A11055), donkey anti-goat Alexa Fluor 647 (Jackson ImmunoResearch, AB_2340437), donkey anti-rabbit Alexa Fluor 488 (Abcam, ab150073), and donkey anti-rabbit Alexa Fluor 555 (Invitrogen, A31572).

Confocal microscopy images were captured using a Zeiss LSM800 laser scanning confocal microscope. ZEN (blue edition) software was used for image acquisition. Images of PMNs, identified using CTB, were captured from serial Z-stacks with 0.2 μm interval and XYZ voxel dimensions of 312 × 312 × 600 nm, from which resultant 2D images were produced by summating the intensity across 2 μm thick z-stacks passing from the center of the neurons. Images were visualized and processed using FIJI.⁹⁴

In situ hybridization

In situ hybridization was performed as previously described.^95,96 Briefly, a T7 polymerase promoter sequence was added to the 5′ end of the reverse primer of the PCR primers for rabies virus G protein (Forward: AAAGCATTTCCGCCCAACAC, Reverse: TAATACGACTCACTATAGGGCCTCGTCACCGTCCTTGAAA) and DNA was amplified from a plasmid expressing rabies virus G protein (pAAV-EF1a-FLEX-GTB).⁹⁰ pAAV-EF1a-FLEX-GTB was a gift from Edward Callaway (Addgene plasmid# 26197, RRID: Addgene_26197). Next, RNA probe was generated using T7 polymerase and digoxigenin (DIG) labeling mix. 16 μm-thick cryostat sections from P4 ChAT::Cre; RphiGT mice were used for hybridization. Anti-DIG antibody conjugated to alkaline phosphatase (AP, Roche, RRID: AB_514497) was applied to visualize the signal from the RNA probe.

Three-dimensional monosynaptic mapping reconstruction

Images of sections from the brainstem to the spinal cord were organized in sequential order in a folder and imported in Imaris (Oxford Instruments), which automatically generates a three-dimensional (3D) brainstem-spinal cord model based on imported sequential images. The contour of the brainstem to spinal cord was outlined by the ‘Surface’ function in Imaris. mCherry+ monosynaptic inputs (magenta) and starter PMNs (turquoise) were identified by the ‘Spot’ function in Imaris and their colors were assigned based on their identities. Starter PMNs are located within clustered ChAT+ MNs in the ventral spinal cord (Video S1).

Filament and VAChT+ synapse analysis

Dendritic morphology of ChAT+ INs was reconstructed and analyzed using the ‘Filament’ function in Imaris. Definitions of the statistics used for filament analysis can be found in the Imaris V 6.3.1 Reference Manual (http://www.bitplane.com/download/manuals/ReferenceManual6_3_1.pdf).

For VAChT+ puncta quantification on retrogradely traced mCherry+ PMNs, VAChT+ puncta were counted by the ‘Spot’ function in Imaris. Traced mCherry+ PMNs were reconstructed by the ‘Filament’ function in Imaris. Only the VAChT+ puncta that were on the PMNs were included (filtered by the intensity of the mCherry signal). To categorize the subcellular location of the VAChT+ puncta, the cell body of the PMNs was delineated by the ‘Surface’ function in Imaris. VAChT+ puncta that were outside a 100μm radius of the center of the cell body were considered to be on the distal dendrites.

For quantification of puncta colocalization, both VAChT+ and Pitx2^tdTom+ puncta on PMNs were identified by the ‘Spot’ function in Imaris. VAChT+ puncta on PMNs were then classified as Pitx2^tdTom+ or Pitx2^tdTom-based on the intensity of the tdTomato signal in the ‘Spot’ function. The total number of PMNs was counted to calculate the average number of VAChT+ puncta per PMN.

Topographical analysis

To plot the topographical distribution of mCherry+ ChAT+ INs, the X and Y coordinates of ChAT+ INs, the spinal cord outline, the central canal and MNs were identified in Imaris using the ‘Spot’ function.^91,97,98 XY coordinates were rotated and standardized to a spinal cord coordinate map where the central canal is the origin (0, 0). The sizes of the standardized coordinate map were determined by averaging the size of the spinal cord. Since the shape of the spinal cord changes along the rostrocaudal axis and LMNs are located slightly caudal to PMNs, the standardized dimensions of the spinal cord for the ChAT+ INs → PMNs are 2000 μm in the horizontal and 1500 μm in the vertical direction (±1000 for X axis and ±750 for Y axis), while the standardized dimensions of the spinal cord for the ChAT+ INs → LMNs are 2000 μm (horizontal) and 1250 μm (vertical) (±1000 for X axis and ±625 for Y axis). The absolute values of the X coordinates of ChAT+ INs were used as their distance to the central canal.

For rostral to caudal distribution of ChAT+ INs that project to PMNs, sections from different animals were aligned based on the spinal cord atlas⁹⁹ and each section was assigned a rostrocaudal position ID. Their XY positions in the spinal cord were determined as described above.

In vitro isolated brainstem spinal cord preparation

17 C57/BJ6 neonatal mice (P2-P4) of both sexes were used for in vitro electrophysiology experiments. Animals were deeply anesthetized with 4% isoflurane in 100% oxygen before being decerebrated, eviscerated and pinned ventral side down in a dissecting chamber lined with Sylgard silicone elastomer (Dow) that was filled with carbogen-bubbled (95% oxygen, 5% carbon dioxide) artificial cerebrospinal fluid (aCSF, containing 120 mM NaCl, 3 mM KCl, 1.25 mM NaH₂PO₄, 1 mM CaCl₂, 2 mM MgSO₄, 26 mM NaHCO₃ and 20 mM D-glucose) at 4°C. The brainstem and spinal cord were exposed and dissected as previously described,¹⁰⁰ and the brain-stem was transected at the pontomedullary junction. Preparations were then transferred to a recording chamber perfused with recirculating, recording aCSF warmed to 25°C–28°C and given 1-h recovery time prior to the initiation of baseline measurements. Extracellular neurograms were obtained using tight fitting suction electrodes attached to the ventral root of the third or fourth cervical spinal segment (C3/4). For split-bath experiments, the recording chamber was divided into two compartments with the use of a plastic wall and by applying Vaseline around the preparation. To confirm successful splitting of the compartments, food coloring was applied sequentially in both compartments at the end of each experiment. Experiments in which any leak of dye was observed from one compartment to the other were excluded from analyses. Signals were amplified 1000 times, and bandpass filtered (10–1000 Hz) using a differential AC amplifier (Model 1700, A-M Systems), digitized at 5 kHz using a Digidata 1440 (Molecular Devices), acquired using Axoscope software (Molecular Devices) and stored on a computer for offline analysis. Signals were analyzed using the Dataview software (courtesy of Dr. W.J. Heitler, University of St Andrews).

Whole-cell patch clamp electrophysiology

Whole-cell patch clamp recordings were obtained from 22 tdTomato+ interneurons on brainstem-spinal cord preparations obtained from 5 Pitx2^tdTom neonatal mice (P3-P4) of both sexes. Access was gained to Pitx2+ interneurons located near the central canal by performing a mid-sagittal hemisection of the spinal cord using an insect pin. Preparations were stabilized in a recording chamber by pinning them to an agar block and visualized with a 40× objective using infrared illumination and differential interference contrast (DIC) microscopy. Cells were visualized and whole-cell recordings obtained under DIC using pipettes (L: 100 mm, OD: 1.5 mm, ID: 0.84 mm; World Precision Instruments) pulled on a Flaming Brown micropipette puller (Sutter Instruments Model P97) to a resistance of 2.5–3.5 MΩ. Pipettes were back-filled with intracellular solution (containing 140 mM KMeSO4, 10 mM NaCl, 1 mM CaCl2, 10 mM HEPES, 1 mM EGTA, 3 mM Mg-ATP and 0.4 mM GTP-Na2; pH 7.2–7.3, adjusted with KOH). Signals were amplified and filtered (6 kHz low pass Bessel filter) with a Multiclamp 700B amplifier, acquired at 20 kHz using a Digidata 1440A digitizer with pClamp Version 10.7 software (Molecular Devices) and stored on a computer for offline analysis.

Adult perfused preparation

9 prepubescent male Sprague-Dawley rats (age: 4–6 weeks; 80–180g) were deeply anesthetized with 5% isoflurane in air before being bathed in ice-chilled physiologic saline solution (115 mM NaCl, 4 mM KCl, 1 mM NaHCO3, 1.25 mM NaH2PO4, 2 mM CaCl2, 10 mM D-glucose, and 12 mM sucrose), decerebrated at approximately the mid collicular level, and spinally transected near the thoracolumbar junction. Rats were eviscerated and vagotomised, and then perfused via the descending aorta with physiologic saline solution equilibrated to 40 mmHg PCO2 and balance oxygen, pressure held above 90 mmHg, and the temperature at 32°C–33°C. Extracellular neurograms were obtained from the phrenic nerve using silver wire hook electrodes and electromyograms recorded from muscles of the 5th intercostal muscles. Signals were acquired at 5 kHz, amplified 1000 times, and bandpass filtered from 0.1 to 1 kHz (Axoscope 9.0). Pharmacological manipulations of M2 receptors were performed by delivering 10 μM methoctramine (Sigma-Aldrich #M105) through the perfusate.

Whole-body plethysmography

Freely moving mice were placed in a chamber for whole body plethysmography (emka, Figure 7A). The air flow was maintained at 0.75 L/min per chamber for all gas mixtures. Breathing measurements were obtained from pairs of adult mice (P60–120), with each pair consisting of one control and its sex-matched mutant littermate. Normal air was given for 45 min, followed by 5% or 10% CO₂ for 15 min (Figure 7A). All breaths were collected initially and plotted for overview. Since breathing patterns can greatly vary under normal air depending on the activity level of the animal, we selected breaths during which the animal was resting (not moving, sniffing or grooming, accompanied by a consistent pattern of low frequency breaths) to represent breathing under normal air conditions. When switching from normal air to hypercapnic conditions, a response curve was observed due to the gas exchange in the chamber being a gradual process. We selected breaths after maximum breathing frequency was reached after the switch to represent breaths under hypercapnic conditions. For each animal, at least 3 trials were performed, and all the trials were included in the analysis. For Dbx1^ΔChAT mice and their control littermates, 8697–30757 breaths under air and 11793–68143 breaths under 10% CO₂ for each animal were included for analysis. For Pitx2^ΔChAT mice and their control littermates, 997–28228 breaths under air, 6727–48128 breaths under 10% CO₂ and 2787–22118 breaths under 5% CO₂ for each animal were included for analysis. All qualifying breaths were used to characterize their distribution. The mean values from all qualifying breaths collected were used to represent individual animals for group comparisons. Normalization was presented as fold control, where the control was the matched littermate that was recorded at the same time.

Hypercapnic challenge and c-Fos expression analysis

P20 ChAT::eGFP mice were placed in whole body plethysmography chambers for normal air or hypercapnic gas challenge. Either normal air (79% N₂, 21% O₂), 5% CO₂ (with 74% N₂ and 21% O₂), or 10% CO₂ (with 69% N₂ and 21% O₂) were given for either 1 h or for 15 min. The 15-min trials were followed by 45 min of normal air. Mice were euthanized by intraperitoneal injection of a ketamine/xylazine cocktail solution and dissected immediately after the gas challenge. Continuous spinal cord sections were mounted on 10 individual slides (10 replicates). One set of the spinal cord sections was used for c-Fos immunostaining and quantification. The numbers shown in Figures 6 and S7 are total numbers from one set of spinal cord sections and correspond to one-tenth of the overall PMN/ChAT+ IN numbers in one animal. c-Fos+ and ChAT+ cells were identified separately by using the ‘Spot’ function in Imaris. c-Fos+ cells in the white matter of the spinal cord were rare and were not included in our counts. Based on their location in the spinal cord, neurons were divided into MN and non-MN groups. A spot-to-spot distance measuring filter in the ‘Spot’ function was applied to define c-Fos/ChAT colocalization and count c-Fos+ ChAT+ INs.

QUANTIFICATION AND STATISTICAL ANALYSIS

Data were reported as mean ± SEM in the results section. Data were presented as violin plots, boxplots or bar plots with each dot representing data from an individual mouse. For violin plots, solid lines indicate the mean and dashed lines indicate 25^th and 75^th percentile. Appropriate and equivalent nonparametric tests (Mann-Whitney or Kruskal-Wallis) were conducted when data failed tests of normality or equal variance with Shapiro Wilk and Brown-Forsythe tests, respectively. For the data in Figure 5, a repeated measures mixed-effects model along with Holm-Šídák’s multiple comparisons test was used. Unpaired t-tests were performed on data with two variables. One sample t test (hypothetical value = 1) was used for data after normalization. Statistical analyses were performed using Graph Pad Version 9.0 (Prism, San Diego, CA, USA). p < 0.05 was considered to be statistically significant, where *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. Statistical details of experiments can be found in the Results and Figure Legends.

Supplementary Material

SUPPLEMENTAL INFORMATION

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.116078.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
goat anti-ChAT	Sigma	RRID: AB_2079751
goat anti-VAChT	Millipore	RRID: AB_2630394
rabbit anti-VAChT	Synaptic Systems	RRID: AB_10893979
chicken anti-RFP	Rockland	RRID: AB_10704808
rabbit anti-DsRed	Takara Bio	RRID: AB_10013483
rabbit anti-c-Fos	Synaptic Systems	RRID: AB_2905595
goat anti-Scip	Santa Cruz Biotechnology	RRID:AB_2268536
rabbit anti-Pitx2	Zagoraiou et al.46	N/A
rabbit anti-CTB	Novus Biologicals	RRID: AB_962919
sheep anti-DIG-AP-conjugated	Roche	RRID: AB_514497
Bacterial and virus strains
RabiesΔG-mCherry	Wickersham et al.86	N/A
Chemicals, peptides, and recombinant proteins
Silk fibroin	Sigma-Aldrich	Cat# 5154
Methoctramine	Sigma-Aldrich	Cat# M105
Potassium Chloride	Sigma-Aldrich	Cat# P9541
Sodium Chloride	Sigma-Aldrich	Cat# S7653
Magnesium Chloride, Hexahydrate	Sigma-Aldrich	Cat# M2670
Magnesium Sulfate	Alfa-Aesar	Cat#33337
Calcium Chloride	Sigma-Aldrich	Cat# C7902
Sodium phosphate, dibasic, anhydrous	Sigma-Aldrich	Cat# S71504
Sodium bicarbonate	Sigma-Aldrich	Cat# S5761
D-Glucose, anhydrous	Sigma-Aldrich	G8270
Potassium Methyl Sulfonate	Sigma-Aldrich	Cat# 83000
HEPES	Tocris	Cat# 3173
EGTA	Sigma-Aldrich	Cat# 324626
Sodium Guanosine Triphosphate	Sigma-Aldrich	Cat# G8877
Sucrose	Sigma-Aldrich	Cat# S9378
Magnesium Adenosine Triphosphate	Sigma-Aldrich	Cat# A9187
Cholera Toxin Subunit B	Sigma-Aldrich	Cat# C9903
Experimental models: organisms/strains
Mouse: B6; 129P2-Gt(ROSA) 26Sortm1(CAG-RABVgp4,-TVA)Arenk/J (RphiGT)	Takatoh et al.40	JAX# 024708
Mouse: B6.Cg-Gt(ROSA) 26Sortm9(CAG-tdTomato)Hze/J (Ai9)	Madisen et al.47	JAX# 007909
Mouse: B6.Cg-Tg(RP23-268L19-EGFP) 2Mik/J (ChAT:eGFP)	Tallini et al.87	JAX# 007902
Mouse: B6; 129-Chattm1Jrs/J (ChATflox)	Misgeld et al.88	JAX# 016920
Mouse: B6.129S-Chattm1(cre)Lowl/MwarJ (ChAT::Cre)	Rossi et al.41	JAX# 031661
Mouse: B6; 129P2-Dbx1tm2(cre)Apie/Orl (Dbx1:Cre)	Bielle et al.89	EM# 01924
Mouse: Pitx2:Cre	Liu et al.48	N/A
Oligonucleotides
Primer G protein Forward: AAAGCATTTCCGCCCAACAC	This paper	N/A
Primer G protein Reverse: TAATACGACTCACTATAGGGCCTCGTCACCGTCCTTGAAA	This paper	N/A
Recombinant DNA
pAAV-EF1a-FLEX-GTB	Haubensak et al.90	RRID: Addgene_26197
Software and algorithms
Zen (blue edition)	Zeiss	N/A
ImageJ/Fiji	https://imagej.net/ij/	N/A
Imaris	Oxford Instruments	N/A
Axoscope	Molecular Devices	N/A
pClamp Version 10.7	Molecular Devices	N/A
Multi Clamp Commander	Molecular Devices	N/A
Dataview	Bill Heitler, University of St Andrews	N/A
Prism	GraphPad Software Inc	N/A
Code for plethysmography analysis	This paper	Zenodo: https://doi.org/10.5281/zenodo.15732816

Highlights.

• Pitx2+ V0_C spinal cord interneurons project to phrenic motor neurons (PMNs)

• V0_C interneurons modulate PMN output and are activated in response to hypercapnia

• V0_C interneuron silencing impairs the ventilatory response to hypercapnia