Presynaptic Mechanisms and KCNQ Potassium Channels Modulate Opioid Depression of Respiratory Drive

Opioid-induced respiratory depression (OIRD) is the major cause of death associated with opioid analgesics and drugs of abuse, but the underlying cellular and molecular mechanisms remain poorly understood. We investigated opioid action in vivo in unanesthetized mice and in in vitro medullary slices containing the preBötzinger Complex (preBötC), a locus critical for breathing and inspiratory rhythm generation. Although hypothesized as a primary mechanism, we found that mu-opioid receptor (MOR1)-mediated GIRK activation contributed only modestly to OIRD. Instead, mEPSC recordings from genetically identified Dbx1-derived interneurons, essential for rhythmogenesis, revealed a prevalent presynaptic mode of action for OIRD. Consistent with MOR1-mediated suppression of presynaptic release as a major component of OIRD, Cacna1a KO slices lacking P/Q-type Ca2+ channels enhanced OIRD. Furthermore, OIRD was mimicked and reversed by KCNQ potassium channel activators and blockers, respectively. In vivo whole-body plethysmography combined with systemic delivery of GIRK- and KCNQ-specific potassium channel drugs largely recapitulated these in vitro results, and revealed state-dependent modulation of OIRD. We propose that respiratory failure from OIRD results from a general reduction of synaptic efficacy, leading to a state-dependent collapse of rhythmic network activity.


INTRODUCTION
Respiratory depression is the primary cause of death associated with opioid-based analgesics and drugs of abuse. In 2015, the opioid epidemic claimed 33,091 deaths in the US, accounting for 16.3 deaths per 100,000 (Rudd et al., 2016). Despite extensive investigations, central mechanisms of opioid action on respiratory drive remain incompletely resolved (Shook et al., 1990;Dahan et al., 2009). Identifying cellular and molecular mechanisms underlying Opioid-Induced Respiratory Depression (OIRD) may provide insights for reversing OIRD and controlling in vivo variability.
We combined pharmacological and genetic approaches in mice, employing both in vitro electrophysiological recordings from rhythmic preBötC slices, and in vivo plethysmography without anesthesia.
Unexpectedly, in vitro OIRD produced by bath-applied DAMGO (a MOR1-specific agonist), was not reversed by TertiapinQ, a GIRK-specific blocker, suggesting a minimal role for increased GIRK conductance resulting from MOR1 activation. In addition, application of ML297, a GIRK-specific activator failed to mimic DAMGO-mediated in vitro OIRD. By contrast, genetic removal of Ca V 2.1 (P/Q-type calcium channel) in Cacna1a KO mice, sensitized in vitro preBötC rhythms to DAMGO depression, consistent with OIRD acting through MOR1-mediated inhibition of presynaptic calcium channels. We similarly identified KCNQ potassium channels as a novel modulator of OIRD, but independent of coupling to MOR1 signaling. KCNQ-specific blockers (Chromanol 293B, XE991) reversed, and activators (retigabine, ICA 69673) mimicked in vitro OIRD. To test for a presynaptic site of action, miniature excitatory post-synaptic currents (mEPSCs) were recorded in TTX from genetically identified Dbx1-expressing inspiratory interneurons (Bouvier et al., 2010). DAMGO suppressed mEPSC frequency significantly, and subsequent XE991 application restored mEPSC frequencies in ∼70% of these neurons. In vivo plethysmography corroborated these in vitro findings: systemic XE991 reversed morphineinduced OIRD in an apparent state-dependent fashion, whereas ML297 (GIRK1-specific activator) failed to suppress. Moreover, retigabine (KCNQ-specific activator) dramatically depressed respiratory frequencies. These results support a presynaptic site of modulatory action by KCNQ potassium channels, acting in concert with MOR1-activated suppression of voltage-gated calcium channels as the predominant mechanism underlying OIRD in the preBötC.

Animal Welfare
All animal procedures were approved by IACUC, Seattle Children's Research Institute (SCRI).
This reaction generates products of 1500 bp for copies of the Cacna1a deletion and 930 bp for WT Cacna1a.

Extracellular Recordings
Fictive inspiratory bursts were recorded as integrated multiunit recordings from large bore glass pipettes filled with ACSF (∼0.1 M ) placed over the preBötC region, on the rostral face of respiratory slices. Slices were maintained under rapidly recirculating bath perfusion (13 mL/min) with oxygenated ACSF at 30 • . Extracellular signals were acquired with an AC differential amplifier (Model 1700; A-M Systems, Sequim, WA), bandpass filtered between 0.1 and 10 KHz, and amplified x10K. Raw extracellular signals were acquired at 10 KHz with pClamp10.4. In parallel, extracellular signals were recorded after on-line processing through a custom analog signal integrator (Univ. Chicago Electronic Lab), which rectified and integrated transients using a τ = 60 ms. Output from this integrator is dependent upon both the frequency of clustered transients and the amplitude of individual events, and enhances the detection of high frequency multi-unit spike bursts (>17 Hz) under these conditions. ACSF was supplemented to 8 mM K + over 30 min to evoke fictive inspiratory rhythms. Drugs were bath applied in 10 min intervals, and burst frequency measurements made during the last 2 min of drug application. All experiments employing sequential application of drugs were performed as a series on individual slices, then normalized to baseline frequencies for analysis. Bath exchange time was estimated to be 2-3 min, under our experimental conditions.

Intracellular Recordings
Whole-cell voltage-clamp recordings were made from preBötC inspiratory neurons under visual control using video-enhanced Dodt-IR optics and fluorescence for either ZsGreen (Ai6) or tdTomato (Ai14) (Koch et al., 2013 -13) and adult (P25-60) animals of mixed genders were used. Recording sessions consisted of 10 min of baseline, followed by 30-60 min of recording following drug delivery. Drugs were delivered intraperotineally (IP) with 31 gauge insulin syringes using minimal volumes (<30 µL for neonates; <250 µL for adults). Drug series and dosages included: (1) ML297 (50 mg/kg), (2) Retigabine (10 mg/kg), (3) Morphine (10 mg/kg for neonates; 150 mg/kg for adults) followed by XE991 (3 mg/kg) (4) DMSO vehicle control. All experiments employing sequential application of drugs were performed as a series on individual animals, then normalized to baseline frequencies for analysis. Drugs were dissolved in 100% DMSO due to poor solubility in saline, to limit injection volumes. Plethysmography signals were acquired as differential pressure signals, relative to equally pressurized control chambers, using pClamp 10.4 and analyzed with Clampfit10 (Molecular Devices, Sunnyvale, CA, United States). Calibration pressure pulses were generated with a manual pipettor (P20, Eppendorf AG, Enfield, CT, United States). For analysis of respiratory frequency, a large number of breaths were curated (500-1000 events) per animal for each measurement, manually excluding large amplitude irregular movement artifacts and rapid high amplitude episodes (>5 Hz) representing active whisking/sniffing (Moore et al., 2013). The distributions of these events were fitted to single Guassian distributions in Origin 8 (OriginLabs, Northhampton, MA, United States), and further statistically analyzed and plotted in Prism 5.04 (GraphPad Software, San Diego, CA, United States).

RT-PCR of preBötC Islands
RT-PCR followed standard procedures. Briefly, total RNA was extracted from preBötC "islands, " micro-dissected and pooled from freshly cut preBötC slices with RNAzol RT (Molecular Research Center, Inc, Cincinnati, OH). First-strand synthesis of single-stranded cDNA (sscDNA) was generated with 1.0-5.0 µg total RNA, primed with random hexamers, and reverse transcriptase following the vendor's protocol (SuperScript IV, Thermo Fisher Scientific). Control reactions without reverse transcriptase were generated in parallel, as a control for residual genomic DNA. 2.0 µL (1:100 dilution) of sscDNA was used for each RT-PCR with gene-specific primer sets and 35-40 cycles of amplification, using GoTaq Hot Start Master Mix (Promega, Madison, WI, United States). See Table 1 for RT-PCR primer sequences.

Statistical Analysis
Datasets were tested for normality by D'Agostino and Pearson omnibus tests. Normal datasets were tested for statistical significance by unpaired t-tests (unless otherwise noted) assuming equal variance, with p-values determined by two-tails at the 95% confidence level. Non-parametric datasets were tested for statistical significance by Mann-Whitney tests with p-values determined by two-tails at the 95% confidence level. Miniature excitatory post-synaptic current (mEPSC) inter-event interval datasets were plotted as cumulative fractional distributions and tested for statistical significances by paired Wilcox-Signed Rank comparisons (modified Kolmogorov-Smirnov test), with p-values determined at the 95% confidence level (OriginLabs, Northampton, MA, United States). Normal datasets are plotted as mean with standard error (SE). Non-parametric datasets are plotted as median with interquartile range (IQR). Statistical calculations were performed using Prism 5.04 (GraphPad Software, San Diego, CA, United States) and Origin8 (OriginLabs, Northampton, MA, United States).

DAMGO Suppresses in vitro Rhythmic Inspiratory Bursts in prebötC Slices
Rhythmic extracellular inspiratory bursts were reliably recorded from transverse slices containing the preBötC derived from neonatal C57BL/6J mice (P7-P13) in 8 mM extracellular K + , using blunt glass electrodes placed over the preBötC region. Inspiratory bursts were recognized as large amplitude integrated signals with a frequency of ∼0.02-0.3 Hz, confirmed in some cases by observing synchronous bilateral bursts from a second electrode placed over the contralateral preBötC. DAMGO ([D-Ala 2 , N-MePhe 4 , Gly-ol]-enkephalin), a MOR1-specific agonist, was bath applied to mimic the effect of central opioid action. Burst frequency was monitored as a highly sensitive measure of the integrity of the respiratory network, based on previous studies (Hayes et al., 2012;Wang et al., 2014;Guerrier et al., 2015;Song et al., 2015). Integrated burst amplitudes were largely unaffected by DAMGO, and provided a reliable means to monitor burst frequency, as an index for in vitro respiratory depression. Bath application of 30 nM DAMGO depressed in vitro inspiratory burst frequencies with an IC 50 of ∼30 nM, consistent with the EC 50 (45 nM) of DAMGO activation of MOR1 measured by [ 35 S]-GTPγS accumulation (Alt et al., 2002). Inspiratory burst frequencies were suppressed to 10% of initial frequencies by 100 nM DAMGO, and restored by washout (Figure 1). Transverse slices thus provided an effective in vitro model system to study opioid action on the preBötC inspiratory network (Greer et al., 1995;Mellen et al., 2003;Ballanyi et al., 2009;Lorier et al., 2010).

Activators of KCNQ Potassium Channels
Depress Inspiratory Rhythmic Activity in vitro, Mimicking the Respiratory Effect of DAMGO KCNQ (Kv7) potassium channels are highly conserved voltagegated potassium channels characterized by partial activation at subthreshold membrane potentials, exceptionally slow gating kinetics, and promiscuous coupling to diverse GPCRs (Brown and Adams, 1980;Wang et al., 1998;Jentsch, 2000;Wei et al., 2005;Sun and MacKinnon, 2017). These features make FIGURE 1 | DAMGO, a mu-opioid receptor specific agonist, suppresses fictive inspiratory bursts recorded from in vitro preBötC medullary slices. (A) Integrated extracellular bursts recorded from an isolated preBötC, in response to bath application of increasing titers of DAMGO (30, 100 nM), followed by wash. Two minute segments shown at expanded time scale for baseline (black), 30 nM DAMGO (blue), 100 nM DAMGO (red), and wash (green). (B) Summary of burst frequency suppression by DAMGO, yielding an IC 50 of ∼30 nM (N = 19 slices; * * * p < 0.0001; paired t-test). Wash restores burst frequency. Mean and SE plotted, with individual replicant values.
this class of potassium channels particularly well-suited to regulating basal membrane excitability, which in turn may critically influence neuronal network dynamics (Brown and Passmore, 2009;Honigsperger et al., 2015). We took advantage of two pharmacological activators specific for KCNQ potassium channels, ICA 69673 and retigabine to test the involvement of this class of potassium channels on preBötC inspiratory rhythms. These openers operate on different structural domains of KCNQ subunits, lending an additional test for specificity. ICA 69673 binds to residues in S3 in the voltage sensor domain of KCNQ subunits (Padilla et al., 2009;Wang et al., 2017), whereas retigabine binds to a hydrophobic pocket near the cytosolic base of the ion conduction pathway formed by S5 and S6 residues, including a critically conserved S5 tryptophan (KCNQ3 W265) (Schenzer et al., 2005;Wuttke et al., 2005;Lange et al., 2009;Kim et al., 2015). Both activators potently suppressed in vitro inspiratory burst frequency with IC 50 s (∼1.0 µM for ICA 69673; ∼0.5 µM for retigabine) comparable to their EC 50 s for activation of KCNQ channels from heterologous expression studies (0.7 µM for ICA 69673 and 0.4 µM for retigabine, both assayed with KCNQ2/3 heteromeric channels) (Figures 2A,B, 3A, left). This similarity of action by two activators which act on different portions of the channel subunit, and the congruence of the dose-response profiles for in vitro respiratory rhythm depression and activation of recombinant KCNQ channels, provides strong pharmacological evidence for involvement of KCNQ potassium channels in the modulation or generation of preBötC inspiratory rhythms.
We next tested the role for KCNQ potassium channels in MOR1-mediated depression of preBötC inspiratory rhythms. Depression of inspiratory rhythms by activated KCNQ channels could occur either through a general suppression of neural excitability, or KCNQ channels could serve a more specific modulatory role in OIRD of the preBötC inspiratory circuit. To test these possibilities, KCNQ-specific blockers were applied to preBötC slices after suppression of inspiratory rhythmic activity with DAMGO (100 nM). We reasoned that recovery of inspiratory rhythms in the presence of DAMGO suppression FIGURE 3 | Pharmacological interrogations of in vitro preBötC slices suggests a modulatory role for KCNQ potassium channels in opioid-induced respiratory depression (OIRD), and a minor role for GIRK potassium channels. (A, left) Dose-response curves of inspiratory burst frequencies in response to increasing titers of DAMGO, and two activators of KCNQ potassium channels (ICA 69673, Retigabine) which act on different structural domains of the channel subunit. Both activators inhibit burst frequencies with an IC 50 = ∼0.7-1.0 µM, comparable to their EC 50 for activation of KCNQ channels. (A, right) Dose-response curves of inspiratory burst frequencies in response to increasing titers of ML297, a GIRK1 subunit specific activator. Modest depression of burst frequencies, at concentrations 10-100-fold higher than the EC 50 s (0.16-0.9 µM) for GIRK1 containing heteromeric channels. (B, left) Dose-response curves of inspiratory burst frequency in response to increasing titers of the KCNQ blockers XE991 and Chromanol 293B (293B), applied in the presence of DAMGO (100 nM) to suppress respiratory rhythms. (B, right) Dose-response curves of inspiratory burst frequency in response to increasing titers of TertiapinQ (TPQ), a GIRK-specific blocker, applied in the presence of DAMGO (100 nM). No recovery of respiratory burst frequency was observed at the highest concentration of TPQ (100 nM). By contrast, both KCNQ blockers partially rescued respiratory rhythms suppressed by DAMGO, at relatively high concentrations (20-100 µM). Shown above plots are the EC 50 s and IC 50 s of each compound for specific molecular species of homo-and hetero-tetrameric GIRK and KCNQ channels reported in the literature (see text and Table 2  would provide evidence for a specific role of KCNQ channels in the inspiratory network. Two KCNQ open channel blockers, XE991 and Chromanol 293B, were applied after suppression of inspiratory bursts by DAMGO. Both KCNQ blockers were found to restore inspiratory burst frequencies in the presence of 100 nM DAMGO, a concentration that completely suppressed inspiratory rhythmic activity (Figures 2C,D, 3B, left). Control application of these blockers alone had no effect on baseline inspiratory frequency (Figure 4). These two KCNQ blockers are known to exhibit different specificities for different molecular species of KCNQ channels formed as homo-and heteromeric tetrameric combinations of the 5 vertebrate KCNQ α-subunits (Jentsch, 2000). Chromanol 293B shows greater specificity for KCNQ1/KCNE1 heteromeric channels found in cardiac myocytes and epithelial cells (IC 50 = ∼11 µM), than for neuronal KCNQ channels assembled from KCNQ2-4 (IC 50 > 500 µM).
KCNQ, GIRK, and MOR1 Transcripts Are Expressed in preBötC, However, KCNQ Is Not Directly Coupled to MOR1 Intracellular Signaling To verify the presence of Kcnq, Girk, and Mor1 transcripts in preBötC, RT-PCR was performed on isolated preBötC "islands, " micro-dissected from transverse slices used for in vitro recordings ( Figure 5A). RT-PCR confirmed the expression of all neuronal Kcnq subunits (Kcnq2-5) in preBötC, as well as the presence of Kcnq1 transcripts. Kcnq1 expression is enriched in cardiac, epithelial and vascular smooth muscle cells (Sanguinetti et al., 1996;Chouabe et al., 1997;Brueggemann et al., 2011;Stott et al., 2013). A positive Kcnq1 RT-PCR signal from preBötC tissue may thus reflect expression from residual vasculature left in our tissue sample rather than specific neuronal expression, although this possibility is unresolvable at our level of analysis. RT-PCR also confirmed the presence of Mor1 and Girk subunits, particularly Girk1, Girk2, and Girk3 which are known to express widely in neurons of the brain, where GIRK1/2 heteromeric channels are the predominant species (Rifkin et al., 2017). Although Girk4 expression is known to be predominantly cardiac, limited neuronal expression has been found in the brain, particularly in inferior olive cells (Wickman et al., 1997(Wickman et al., , 1998. Our positive Girk4 RT-PCR signal may reflect contamination from neighboring inferior olive cells in our isolated preBötC islands. Nonetheless, RT-PCR showed strong signals for known preBötC molecular markers including, tachykinin receptor (Tacr1), somatostatin FIGURE 5 | RT-PCRs of micro-dissected preBötC "islands" reveal transcripts for Kcnq1-5, Girk1-4, and Mor1. However, heterologous co-expression of MOR1 with KCNQ4/3 or KCNQ5/3 in Xenopus oocytes provide no evidence for direct coupling downstream of activated MOR1. (A) Transcripts for all Kcnq1-5 (Q1-Q5) subunits were detected, along with mu-opioid receptor (Mor1) and Girk1-4 subunits. Known preBötC molecular markers also robustly detected, including Substance P receptor (Tacr1), gastrin-release peptide receptor (Grpr1), somatostatin (Sst). Weaker reactions detected for somatostatin receptor (Sstr2), and Substance P (Tac1). For each primer set, paired RT-PCRs were performed with first-strand single-stranded cDNAs generated with (+) and without (-) reverse transcriptase. (Sst), and gastrin-releasing peptide receptor (Grpr1) related to sighs (Gray et al., 2001(Gray et al., , 2010Li P. et al., 2016). Weaker signals for somatostatin receptor (Sstr2) and substance P (Tac1) were also observed. Thus, transcripts for Kcnq and Girk potassium channel subunits, along with the mu-opioid receptor (Mor1) were all found to be present in micro-dissected preBötC islands.
In vivo KCNQ activation could suppress inspiratory rhythms as molecular effectors coupled to intracellular signaling pathways downstream of MOR1 activation, or through an independent mechanism. Direct coupling to MOR1 as a possibility was suggested by prior studies which reported modulation of native M-currents by somatostatin and dynorphins (Moore et al., 1988;Madamba et al., 1999;Qiu et al., 2008). Although all KCNQ channels are known to couple to Gα q -GPCRs, these previous studies suggested coupling between KCNQ channels with Gα i/o -and Gα s -GPCRs, including somatostatin, opioid, and β-adrenergic receptors, in neurons and smooth muscle cells (Moore et al., 1988;Sims et al., 1988;Madamba et al., 1999;Qiu et al., 2008). Therefore, heterologous expression studies were undertaken to test this possibility using recombinant KCNQ potassium channels and MOR1 coexpressed in Xenopus oocytes. Control oocytes injected with cRNAs encoding MOR1 and GIRK1(F137S) carrying a missense mutation in the P-loop which permits functional expression of homomeric channels (Chan et al., 1996), expressed inward rectifying currents that augmented twofold with bath-applied DAMGO (200 nM). However, similar experiments co-expressing MOR1 with either KCNQ5/3 or KCNQ4/3 heteromeric channels showed no effect following bath applied DAMGO (Figures 5B,C). Therefore, we observed no evidence for direct coupling between MOR1 receptors and KCNQ channels from these heterologous reconstitution experiments. Thus, we hypothesized that depression of inspiratory rhythmic activity by activated KCNQ channels likely occurs through a separate mechanism, independent of direct coupling to MOR1 downstream of G-protein signaling.

Genetic Loss of Cacna1a (P/Q-Type) Calcium Channels Increases Sensitivity to OIRD
Suppression of presynaptic voltage-gated calcium channels (P/ Q-, N-, and R-type) by Gα i/o -coupled GPCRs such as MOR1, via membrane-delimited signaling through Gβ/γ released by ligand binding is a well-established phenomenon (Dunlap and Fischbach, 1978;Herlitze et al., 1996;Ikeda, 1996;Colecraft et al., 2001;Agler et al., 2005;Zamponi and Currie, 2012), although its potential role in OIRD has not been examined, to our knowledge. We took advantage of a mouse KO strain defective for Cacna1a (Jun et al., 1999) which encodes the P/Q-type Ca v 2.1 calcium channel to decrease presynaptic calcium channel function, in order to genetically mimic this effect on OIRD sensitivity. We previously demonstrated that loss of Cacna1a function in this strain suppresses sighs and reduces evoked EPSPs in inspiratory interneurons in preBötC slices (Koch et al., 2013).
Sensitivity to OIRD was assessed in preBötC slices obtained from mutant Ca V 2.1 KO and WT animals, by recording inspiratory burst frequency as a function of increasing concentrations of bath applied DAMGO. As an example, continuous plots of instantaneous inspiratory burst frequency as a function of time and drug application are shown in Figure 6, for heterozygous Ca v 2.1 KO/+ and WT preBötC slices derived from the same genetic background (C3H). Strikingly, inspiratory burst frequency declined precipitously in heterozygous Ca V 2.1 KO/+ slices at DAMGO concentrations exceeding 6 nM, and were completely suppressed at 10 nM ( Figure 6A). DAMGO suppression was reversed by the addition of naloxone (1 µM), a MOR1-specific competitive antagonist demonstrating reversibility. By contrast, WT (C3H) burst frequency was largely unaffected by DAMGO concentrations at 6-10 nM, with an IC 50 of ∼10 nM, similar to C57BL/6J ( Figure 6B).
Median dose-response profiles for multiple WT (C3H) and heterozygous Ca V 2.1 KO/+ slices are shown in Figures 6C,D. Although initial inspiratory burst frequencies were indistinguishable between Ca V 2.1 KO/+ and WT slices, the loss of one copy of Cacna1a in heterozygous slices resulted in a ∼10-fold greater sensitivity to suppression by DAMGO (Ca V 2.1 KO/+, IC 50 = ∼1 nM; WT, IC 50 = ∼10 nM). Inspiratory bursts from homozygous Ca V 2.1 KO/KO slices were less frequent and more irregular than heterozygous KO/+ or WT slices (Koch et al., 2013), and thus precluded quantitative analysis.
Taken together, these results are consistent with a major role for presynaptic voltage-gated calcium channels in OIRD. We hypothesized that compromised synaptic transmission due to acute MOR1-mediated suppression of presynaptic calcium channels may serve as a major mechanistic component underlying OIRD in the preBötC.

mEPSC Recordings From Dbx1 + Inspiratory Interneurons Support a Presynaptic Site of Action for OIRD and Modulation by KCNQ K + Channels
To obtain evidence for a presynaptic site of action for OIRD and KCNQ channels, miniature excitatory post-synaptic currents (mEPSCs) were recorded from genetically-identified Dbx1 + inspiratory interneurons (Bouvier et al., 2010;Picardo et al., 2013) in the presence of TTX (1 µM) to block spontaneous action potentials. We took advantage of the fortuitous optical clarity provided by sparse labeling of Dbx1 + cells in Cre-loxP crosses carrying two copies of Dbx1-CreERT2 with Rosa26 CAG-floxed-stop reporters expressing either ZsGreen (Ai6) or tdTomato (Ai14) observed in the absence of tamoxifen injections, usually administered to activate conditionally expressed CreERT2 recombinase (Figure 7). Sparse labeling presumably occurred by "leaky" expression of Dbx1-CreERT2, since the overall expression pattern without tamoxifen closely resembled that observed with tamoxifen injections at embryonic day 10.5, but at a much lower density of labeled cells. This allowed us to easily distinguish Dbx1 + neurons from Dbx1 + astrocytes by morphology for visually-guided patch clamp recordings (Kottick et al., 2017).
Dbx1 + inspiratory interneurons were identified as ZsGreen or tdTomato expressing neurons within the preBötC region which received spontaneous excitatory inspiratory synaptic bursts, synchronous with integrated extracellular population bursts from the contralateral preBötC. TTX (1 µM) selectively blocked large amplitude excitatory synaptic bursts and left a baseline frequency of spontaneous mEPSCs (Figures 8A,B). Addition of DAMGO (100 nM) significantly reduced spontaneous mEPSC frequency (increased inter-event intervals) in 70% (5/7) of recorded Dbx1 + neurons (Figures 8C,D), without significantly altering the distribution of mEPSC amplitudes (Figures 8E,F). Subsequent addition of XE991 (20 µM) increased mEPSC frequency (increased cumulative accumulation of short inter-event intervals) in 86% (6/7) of recorded neurons (Figures 8C,D). Individual neurons were observed to be responsive to both DAMGO and XE991 (57%, 4/7), only DAMGO (14%, 1/7), or only XE991 (28%, 2/7), indicating heterogeneity and independence of presynaptic modulatory mechanisms at individual synapses. Taken together, these results suggest that suppression of inspiratory rhythm generation in the preBötC primarily occurs through a reduction of synaptic transmission at excitatory glutamatergic synapses by a presynaptic mechanism. Blockade of presynaptic KCNQ channels may sufficiently compensate for MOR1mediated reduction of synaptic efficacy to restore rhythmic network function.

In vivo OIRD Is Mimicked Strongly by Systemic KCNQ Activator and Weakly by GIRK-Specific Activator, in an Age-Dependent Fashion
To investigate the relevance of our in vitro findings to live mice, in vivo plethysmography recordings were obtained from unanesthetized neonatal mouse pups (P7-13) and adult mice (P25-P60), by whole-body plethysmography. Plethysmography traces were analyzed by measuring inter-event intervals for a large number of inspirations (>500), excluding segments containing movement artifacts. Mean respiratory frequencies were derived by fitting individual datasets to single Gaussian distributions.
The effect on respiration by systemic activation of GIRK potassium channels was tested after intraperitoneal (IP) injection of ML297 (Figures 9Ai,ii), a small molecular weight activator of heteromeric GIRK1 channels (GIRK1/2, GIRK1/3, GIRK1/4) FIGURE 7 | Sparse labeling of Dbx1-derived (Dbx1 + ) preBötC neurons using Dbx1 Cre−ET2 mouse line, without tamoxifen treatment. (A) Embryonic tamoxifen injection (e10.5) labels a dense column of Dbx1 + cells with tdTomato extending radially from the dorsal ventricular surface through the ventrally-located preBötC, viewed by Dodt-IR optics (Ai) and RFP fluorescence (Aii). This slice from a P7 animal heterozygous for Dbx1 Cre−ERT2 and Ai27 (Dbx1 Cre−ERT2/+ ; Rosa26 Ai27/+ ). (B) Similar slice without tamoxifen sparsely labels individual Dbx1 + cells in a similar pattern as in 7A. This slice from an P7 animal homozygous for Dbx1 Cre−ERT2 and Ai27 (Dbx1 Cre−ERT2/Cre−ERT2 ; Rosa26 Ai27/Ai27 ). Viewed by Dodt-IR (Bi) and RFP fluorescence (Bii). (C) Sparsely-labeled fluorescent neurons and astrocytes (D) from the same slice in 7B at higher magnification. (Kaufmann et al., 2013;Wydeven et al., 2014). A relatively high dose of ML297 (50 mg/kg) was chosen because IP injections at this dose exert overt behavioral effects in adult mice, including decreased spontaneous locomotion and increased anxiolytic behavior, which provided an independent indicator of drug access into the CNS (Wydeven et al., 2014). ML297 rapidly caused decreased spontaneous locomotion in both pups and adults. In pups, this was followed by the appearance of repetitive whole-body twitches which could be recorded as large amplitude transient pressure artifacts in plethysmography recordings. Despite these gross behavioral changes, respiratory frequency was only marginally decreased after ML297 injection (for pups from 4.5 to 3.9 Hz; for adults 3.1-2.8 Hz). Thus, systemic activation of GIRK potassium channels by ML297 suppressed respiratory frequency only modestly (∼10%) in both neonatal pups and adults, and seems unlikely to contribute significantly to the mechanism of OIRD in unanesthetized mice which can approach a 50% reduction of respiratory rate with morphine at the doses we examined (10 mg/kg for neonates; 150 mg/kg for adults).
We next examined the effect of systemic retigabine, an FDA-approved specific activator of KCNQ potassium channels (Amabile and Vasudevan, 2013 ; Figures 9Bi,ii). A moderate dose of retigabine was chosen (10 mg/kg), reported to control seizures and prevent tinnitus in rodent models (Li et al., 2013;Kalappa et al., 2015;Ihara et al., 2016). Similar to ML297, IP injections of retigabine rapidly decreased spontaneous locomotion and in pups induced repetitive whole-body twitches. However, unlike ML297, respiratory rates were profoundly suppressed by retigabine in neonatal pups, and significantly reduced in adults ( Figure 9C). In neonatal pups, mean baseline respiratory rate of 3.9 Hz was reduced by ∼50% to 2.2 Hz after retigabine injection compared to vehicle controls (p = 4.5 e-8; t-test). In adults, retigabine reduced respiratory rate by ∼20%, approximately double the reduction observed with ML297 (p = 0.012; t-test). These results indicate a dominant role for KCNQ potassium channels in the control of in vivo eupnic basal respiratory frequency under the unanesthetized state. Systemic activation of KCNQ channels by retigabine significantly depressed respiratory frequency, particularly in neonatal pups (54% of baseline frequency) equal to the depressive effect of systemic morphine (56% of baseline frequency). Importantly, these results suggest that depression of respiratory drive may potentially be an unrecognized adverse side-effect of retigabine at moderate doses, particularly at early ages.

State-Dependent Reversal of in vivo OIRD by Systemic KCNQ Blocker
We then examined the ability of systemic XE991 to reverse OIRD due to morphine delivered IP, in both neonatal and adult animals. In neonatal animals, morphine injected at 10 mg/kg resulted in a 36% suppression of mean respiratory frequency from 4.5 to 2.9 Hz (Figure 10), but with considerable variance among individual animals. Higher doses of morphine in neonates induced spontaneous locomotion as previously described (Jacquet et al., 1976;Koek et al., 2012), which precluded plethysmography measurements under unrestrained conditions. Adults exhibited a lower mean basal respiratory rate (3.3 Hz), and a dose of 150 mg/kg morphine achieved a similar degree of acute respiratory suppression as neonates (39% reduction of mean respiratory rate).
Responses to subsequent injection with XE991 (3 mg/kg) were found to be dependent upon the breathing rate measured after depression by morphine (Figures 10B-E). When morphine depressed respiratory frequency below a "threshold" frequency (2.9 Hz in neonates and 2.0 Hz in adults), subsequent XE991 increased and partially restored respiratory frequency (from a mean of 2.5 to 3.1 Hz in neonates, a 24% increase; from 1.8 to 2.4 Hz in adults, a 33% increase). By contrast, when morphine FIGURE 8 | mEPSCs recorded from Dbx1-derived (Dbx1 + ) preBötC neurons support a presynaptic site of action for DAMGO and KCNQ potassium channels. (A) Dbx1 + preBötC neuron labeled with tdTomato, without tamoxifen, imaged with Dodt-IR (left) and RFP fluorescence (right). Slice homozygous for Dbx1 Cre−ERT2 and Ai14 (Dbx1 Cre−ERT2/Cre−ERT2 ; Rosa26 Ai14/Ai14 ). (B) Representative mEPSCs recorded from a identified Dbx1 + inspiratory neuron in response to sequential bath application of TTX (1 µM), DAMGO (100 nM), and XE991 (20 µM). Currents measured at a holding potential of -60 mV. (C) Representative cumulative fractional distribution plot of mEPSC inter-event intervals (IEIs) recorded from an inspiratory Dbx1 + neuron, in response to TTX (black), DAMGO,TTX (red), and XE991,DAMGO,TTX (blue). DAMGO,TTX (red) distribution is significantly shifted toward longer IEIs relative to either TTX (black) or XE991,DAMGO,TTX (blue) (Continued) failed to depress respiratory rate below these "threshold" frequencies in either neonates or adults, subsequent XE991 failed to increase respiratory rate. We designated these two populations "responders" and "non-responders, " defined by the ability ("responders") or failure ("non-responders") to increase respiratory rate depressed under morphine by more than 5%, after subsequent injection with XE991. These two populations exhibited statistically distinct response characteristics, comparing morphine-depressed respiratory rates between "responder" and "non-responder" populations (p = 0.025 for neonates, p = 0.0043 for adults; Mann-Whitney). Similar non-parametric comparisons between "responder" and "non-responder" populations for baseline and morphine+XE991 respiratory rates were statistically insignificant, with the exception of a significant difference between adult morphine+XE991 responder and adult nonresponder populations (p = 0.0063; Mann-Whitney). These results suggest a "state-dependence" in the ability of XE991 to restore respiratory rate, dependent upon the degree of individual responsiveness to respiratory depression by morphine.
These results are reminiscent of other studies which suggest that the modulatory effects of respiratory neuromodulators are dependent on baseline respiratory frequency (Doi and Ramirez, 2010), and the modulatory state of the animal (Langer et al., 1985). The ability of XE991 to reverse OIRD may thus be similarly dependent upon the internal modulatory state of the animal, defined by basal levels of endogenous respiratory-related neuromodulators (Langer et al., 1985;Gray et al., 1999;Doi and Ramirez, 2010;Li P. et al., 2016;Yackle et al., 2017).

DISCUSSION
Our study suggests that opioids act presynaptically and are modulated by presynaptic KCNQ channels. As shown in Figure 11, we hypothesize that presynaptic terminals which provide excitatory input for the essential kernel of glutamatergic respiratory interneurons, contain both MOR1 opioid receptors and KCNQ potassium channels, in addition to the normal complement of presynaptic calcium channel (P/ Q-, N-, and/or R-type) and vesicular release molecular machinery. Opioids are known to activate presynaptic MOR1 at many synapses (Vaughan et al., 1997;Tao and Auerbach, 2005;Zhu and Pan, 2005;Hjelmstad et al., 2013), which signal to reduce voltage-dependent calcium influx by a well-described Gα i/o signaling pathway. Downstream of receptor activation, presynaptic calcium channels are inhibited by membrane-delimited binding between Gβ/γ released from activated MOR1, and the second intracellular loop between homology domains I and II of P/Q-and N-type calcium channels (Herlitze et al., 1996;Ikeda, 1996;Zamponi et al., 1997;Agler et al., 2005). Although our results demonstrate a specific role for P/Q-type calcium channels in preBötC rhythm generation consistent with MOR1-mediated OIRD, a similar role for N-type calcium channels likely operates within the preBötC (Lieske and Ramirez, 2006;Koch et al., 2013). Suppression of calcium channel function occurs in part by slowing activation kinetics and shifting intrinsic voltagedependence toward depolarized potentials, thus compromising evoked synaptic transmission (Zamponi and Currie, 2012). Based on our data, we hypothesize that independently, partially activated presynaptic KCNQ potassium channels contribute to setting basal presynaptic membrane potential, as observed in hippocampal culture and slices (Vervaeke et al., 2006;Peretz et al., 2007) and the glutamatergic Calyx of Held (Huang and Trussell, 2011). Thus, blockade of KCNQ channels with XE991 may depolarize presynaptic terminals sufficiently to compensate for the reduced function of calcium channels inhibited by bound Gβ/γ. The inhibitory action of Gβ/γ on calcium channels is voltage-dependent, and reversible by strong depolarizations (Colecraft et al., 2000;Zamponi and Currie, 2012). However, how much calcium channel activity may be restored by the likely modest depolarization resulting from blockade of presynaptic KCNQ channels is unclear, and other mechanisms may explain our results. Direct measurements of Ca 2+ influx by genetic approaches expressing genetically-encoded calcium sensors may provide a direct test of this hypothesis (Dreosti et al., 2009;Hoppa et al., 2014).
This model for OIRD is consistent with an essential role for glutamatergic transmission in preBötC rhythmogenesis (Funk et al., 1997;Del Negro et al., 2010), and a determinative influence of glutamatergic synaptic dynamics in shaping network rhythmicity (Guerrier et al., 2015;Kottick and Del Negro, 2015). Our findings are also consistent with previous demonstrations that ampakines, which act as enhancers of glutamatergic transmission, effectively rescue OIRD and other forms of apnea (Ren et al., 2006(Ren et al., , 2015Lorier et al., 2010;Ogier et al., 2007). 5HT4a receptor agonists (Manzke et al., 2003) and other treatments that elevate intracellular cAMP (Ballanyi et al., 1997;Ruangkittisakul and Ballanyi, 2010) have also been reported to rescue OIRD, suggesting additional mechanisms mediated by PKA. We demonstrate that excitatory input onto Dbx1-derived neurons are presynaptically inhibited by DAMGO. There is growing evidence that Dbx1-derived neurons are necessary and FIGURE 9 | Whole-body plethysmography from awake, unanesthetized neonatal and adult mice, in response to systemic delivery of GIRK1 activator (ML297) or KCNQ activator (retigabine; RTG). Profound suppression of respiratory frequency by KCNQ activator, but only modest suppression by GIRK1 activator. (A) Modest reduction of respiratory frequency with ML297 (50 mg/kg, IP) from a neonatal pup. Representative plethysmography recordings before (Ai, top), and after drug injection (Ai, bottom). (Aii) Distributions of instantaneous breath frequencies from the same animal, fitted to single Gaussian distributions. (B) Large reduction of respiratory frequency with retigabine (10 mg/kg, IP) from a neonatal pup. Plethysmography records before (Bi, top) and after drug injection (Bi, bottom). (Bii) Distributions of instantaneous breath frequencies from the same animal, fitted to single Gaussian distributions. (C) Summary of plethysmography data for neonates (P7-13) and adults (P25-60), normalized to baseline respiratory frequencies. Both neonates and adults exhibited modest and similar reductions (∼10%) in respiratory frequency with ML297. However, retigabine caused large reductions in respiratory frequency in both neonates (46% reduction) and adults (20% reduction). Adult respiratory suppression by retigabine was significantly stronger than ML297 (p = 0.012; t-test). Results from vehicle controls (DMSO; IP) with neonates and adults were not significantly different and pooled. Population mean and SE plotted, along with individual responses derived from means of fitted Gaussian curves, for each condition, for neonatal retigabine (N = 11), neonatal ML297 (N = 12), adult retigabine (N = 9), adult ML297 (N = 6), DMSO control (N = 8). Statistical significance: * * * p < 0.0001, * * p = 0.0003, by un-paired t-test. sufficient for respiratory rhythmogenesis (Hayes et al., 2012;Wang et al., 2014), suggesting that these depressive effects in preBötC may contribute to OIRD. However, we cannot exclude the possibility that DAMGO may also critically affect other neurons within the brainstem respiratory network.
An additional caveat is that our in vitro slice studies were conducted in ACSF elevated from 3 to 8 mM, experimentally necessitated to evoke consistent inspiratory rhythms. In vitro transverse preBotC slices typically fail to generate reliable respiratory bursts in 3 mM K + . Raising external potassium from 3 to 8 mM would be expected to shift the potassium reversal potential from -98.5 to -73.8 mV, assuming an internal K + concentration of 150 mM, based on a Nernst equilibrium potential. For non-voltage gated potassium channels such as GIRK, this might be expected to reduce its relative repolarization influence compared to voltage-gated KCNQ FIGURE 10 | State-dependent reversal of respiratory suppression by morphine with XE991. Whole-body plethysmography from unanesthetized neonatal and adult mice, in response to systemic IP injections with morphine and XE991, a KCNQ-specific blocker. (A) Representative plethysmography recordings from an adult "responder" mouse during baseline breathing (Ai, blue), after morphine injection (150 mg/kg) (Ai, red), and subsequent injection with XE991 (3 mg/kg) (Ai, green). (Aii) Distributions of instantaneous breath frequencies from the same animal under each condition, fitted to single Gaussian distributions. XE991 rescue of respiratory suppression by morphine was variable, and dependent upon the respiratory frequency observed after morphine injection. (B) Neonatal "responders" exhibited strong suppression of respiratory frequency to morphine, below a 2.9 Hz "threshold" frequency. All animals partially recovered respiratory frequency with XE991 (N = 14, p < 0.0001; paired t-test). (C) Adult "responders" similarly exhibited strong respiratory suppression to morphine, below a 2.0 Hz "threshold" frequency. All animals partially recovered respiratory frequency with XE991 (N = 8, p = 0.0003; paired t-test). (D) Neonatal "non-responders," by contrast failed to exhibit morphine suppression of respiratory frequency below a 2.9 Hz "threshold." None of these animals increased respiratory frequency more XE991 (N = 5, ns; paired t-test). (E) Adult "non-responders" similarly failed to exhibit morphine suppression of respiratory frequency below a 2.0 Hz "threshold." None of these animals responded to XE991 with an increased respiratory rate (N = 5, ns; paired t-test). * * * p < 0.0001, * * p = 0.0003, ns, non-significant. potassium channels, due to a combination of reduced K + driving force, and fixed intrinsic outward conductance for GIRK channels versus non-linear voltage-dependent increases in conductance for KCNQ channels. Other determinative factors include the relative expression levels of these channel types. Whether these considerations may significantly alter the relative FIGURE 11 | Model for presynaptic site of action by opioids underlying opioid-induced respiratory depression (OIRD). We hypothesize that glutamatergic presynaptic terminals onto Dbx1-derived (Dbx1 + ) inspiratory interneurons within the preBötC compartmentalize mu-opioid receptors (MOR1) and KCNQ potassium channels, in addition to the normal complement of presynaptic calcium channels associated with vesicular release, including P/Q-type (Ca V 2.1) and N-type (Ca V 2.2) calcium channels. Activation of MOR1 leads to suppression of presynaptic calcium channel function, via direct binding with Gβ/γ generated by receptor activation. This leads to reduced synaptic transmission across the excitatory inspiratory circuit, defined by an essential kernel of Dbx1 + glutamatergic neurons. Because rhythmogenesis within this circuit is largely network-driven, reduced synaptic efficiency leads to collapse of rhythmogenic capacity, and opioid-induced apnea. We further propose that active KCNQ potassium channels contribute to maintaining presynaptic resting membrane potentials. Blocking these channels may depolarize presynaptic terminals, mitigating the effects of opioid-mediated suppression of calcium channel function. We emphasize that our model proposes that presynaptic KCNQ channels act independent of direct signaling downstream from MOR1 activation, but function in parallel with MOR1-mediated inhibition of presynaptic calcium channels by virtue of co-compartmentalization to the presynaptic terminal. Sufficient restoration of synaptic transmission may thus revive rhythmogenesis, in spite of ongoing signaling by activated MOR1. Our model of OIRD suggests a relatively minor role for MOR1-coupled activation of GIRK potassium channels, however somatic GIRK currents have been reported by other studies.
roles of these two potassium channels on burst generation in preBotC circuits may require computational studies in realistic network models, with biophysically accurate GIRK and KCNQ conductances, along with further experimental studies. However, the combination and concordance of our in vitro and in vivo experimental results suggests a relatively minor role for GIRK activation compared to KCNQ activation in suppressing respiratory bursts.
Alternative scenarios could also contribute. Gβ/γ suppression of synaptic transmission is described at some synapses, which bypasses direct involvement of calcium channels or altered influx of Ca 2+ (Blackmer et al., 2001Gerachshenko et al., 2005). By this mechanism, Gβ/γ interferes with the SNAREs that comprise the vesicular release mechanism, by binding to SNAP-25, one of the three key proteins which forms the ternary SNARE complex. This prevents Ca 2+ -induced vesicular fusion, perhaps by sterically interfering with synaptotagmin/Ca 2+ action (Zurawski et al., 2017). Another possible mechanism is based on the recently revealed voltage-dependence of many GPCRs, including MOR1. Many GPCRs respond to membrane depolarization with reduced affinity for agonist binding, and a gating current indicative of an intrinsic voltage sensor (Ben-Chaim et al., 2006;Ohana et al., 2006;Parnas and Parnas, 2007). The basis for this voltage-dependence derives from a Na + ion that occupies a site within the ligand binding pocket, which is required for high affinity ligand binding (Livingston and Traynor, 2014;Vickery et al., 2016). Depolarization electrostatically displaces Na + from this pocket and lowers ligand affinity. Conceivably, depolarizations caused by blocking KCNQ channels may reduce the ligand binding affinity of presynaptic MOR1 directly at the presynaptic terminal, or by promoting repetitive depolarizations in intrinsic burster neurons (Ramirez et al., 2011(Ramirez et al., , 2012, sufficiently to decrease MOR1 signaling. Whether these possible mechanisms operate within preBötC respiratory neurons to contribute to OIRD await future studies. Rescue of in vivo OIRD by KCNQ blockers was more variable than that observed with in vitro slices. Several reasons may explain increased in vivo variability, an issue of clinical importance since this variability makes OIRD potentially unpredictable and dangerous. First, opioids may act at multiple brainstem sites affecting respiration (Stucke et al., 2008(Stucke et al., , 2015, including the Kolliker-Fuse nucleus which provides excitatory drive onto the respiratory network, counteracting apneusis (Molkov et al., 2013;Smith et al., 2013;Levitt et al., 2015). Inhibition of pontine structures exaggerates OIRD (Stucke et al., 2008(Stucke et al., , 2015Prkic et al., 2012). Secondly, glutamatergic transmission within the preBötC is affected by multiple convergent neuromodulators, including norepinephrine, serotonin and substance P (Doi and Ramirez, 2008;Funk et al., 2011;Ramirez et al., 2012). Conceivably, multiple Gα q -coupled neuromodulators may act on the same pool of presynaptic terminals via decreased PIP2 levels to regulate basal transmitter release by modulating KCNQ potassium channels (Suh and Hille, 2002;Hoshi et al., 2003;Suh et al., 2006;Sun and MacKinnon, 2017) and other presynaptic effectors. Thirdly, opioids may lead to compensatory release of other neuromodulators (Muere et al., 2014;Langer et al., 2017), or disinhibition of excitatory circuits, leading paradoxically to increased agitation and hyperactivity (Hnasko et al., 2005;Mustapic et al., 2009;Koganezawa et al., 2011). In addition, differing degrees of systemic hypoxia between "responder" and "non-responder" individuals may contribute to state-dependence of recovery by KCNQ blockers. All these considerations may interact to result in in vivo OIRD state-dependence, and variable efficacy of pharmacological interventions.
In contrast to our findings, OIRD has been proposed to be caused by MOR1-mediated post-synaptic activation of GIRK potassium channels (Montandon et al., 2016a,b). Moreover, previous in vitro studies have reported upregulation of a potassium current (Gray et al., 1999) and small hyperpolarizing shifts in basal membrane potentials in inspiratory preBötC neurons (Montandon et al., 2011) in response to MOR1 activation. However, the upregulated potassium current reported in Gray et al. (1999) exhibited outward rectification, rather than inward rectification which would be expected for a GIRK conductance (Rifkin et al., 2017). Reports of baseline hyperpolarization by opioids are also limited and inconsistent between different laboratories (Takeda et al., 2001;Ballanyi et al., 2009Ballanyi et al., , 2010Montandon et al., 2011). In balance, in our judgment these studies provide inconclusive evidence for widespread MOR1 activation of post-synaptic GIRK channels, despite our results which show the presence of Girk transcripts in the preBötC. What elements of this model of GIRK-mediated suppression of excitability may contribute to OIRD remains an open question.
Our studies support a modest role for GIRK channels in OIRD, either by the failure to rescue in vitro slice models of OIRD with a GIRK-specific blocker (TertiapinQ), or to mimic OIRD by systemic delivery of a GIRK1-specific activator (ML297) in vivo, under unanesthetized conditions. Under these conditions, ML297 likely activates all GIRK1-containing tetrameric channel species, including the predominant GIRK1/2 heteromeric channel species found in brain (Rifkin et al., 2017). However, our pharmacological results cannot exclude the possibility that a component of OIRD may be mediated by homo-or heteromeric GIRK2/GIRK3 channels (GIRK2/3, GIRK2, GIRK3) not activated by ML297, even though these are considered relatively minor GIRK species. Several major experimental and interpretative differences could underlie the discrepancies between our conclusions from those proposed by Montandon et al. (2016a,b). In these studies, drug delivery and in vivo experimental measurements were made under anesthesia with 1.5-2.5% isoflurane. Isoflurane suppresses respiratory drive, and even at low concentrations may potentiate the depressive effects of other opioid-related respiratory suppressants. The use of isoflurane thus complicates the interpretation of in vivo results. By contrast, urethane is the anesthetic of choice for in vivo respiratory studies, since it yields effective analgesia with minimal cardiorespiratory suppression (Doi and Ramirez, 2010;Pagliardini et al., 2012). Alternatively, systemic delivery of drugs without anesthetics as in our study, may provide a more accurate measure of opioid action on in vivo respiratory drive. Another significant interpretative error comes from the stated use of flupirtine as a GIRK-specific activator (Montandon et al., 2016b). Flupirtine is an early analog of retigabine. Both are generally recognized as specific activators of KCNQ, not GIRK potassium channels. Despite limited off-target effects including potentiation of δ-subunit containing GABA_A receptors (Klinger et al., 2015;Treven et al., 2015) and block of Kv2.1 potassium channels (Stas et al., 2016), to our knowledge there are no reports of either flupirtine or retigabine activating GIRK channels. However, offtarget potentiation of δ-subunit containing GABA_A receptors by the high concentrations of flupirtine used (200-300 µM) in the in vivo microperfusion studies of Montandon et al. (2016aMontandon et al. ( ,b, 2017 combined with KCNQ activation, rather than specific GIRK activation, may account for the respiratory depression observed in these studies. These studies also report the loss of respiratory depression by both flupirtine and DAMGO, as well as a reduced augmentative effect by the positive neuromodulator GR73632 (an NK-1 receptor agonist) in homozygous Girk2 KO mice, compared to WT controls (Montandon et al., 2016a,b). Loss of Girk2 results in seizure-prone mice and the dysregulation of GIRK1 subunits trafficked to the plasma membrane (Signorini et al., 1997;Ma et al., 2001). Conceivably, compensatory mis-regulation of KCNQ and other ion channels or receptors due to chronic overexcitation in constitutive Girk2 KO mice may result in DAMGO-dependent OIRD effects different from acute pharmacological manipulations in the WT background. We believe these choices in experimental design and potential errors in data interpretation present important challenges to the prevailing model of GIRK-mediated OIRD (Montandon et al., 2016a,b).
Thus, we conclude that opioids act primarily by broadly disrupting the efficacy of excitatory transmission within the preBötC inspiratory network, leading ultimately to the precipitous collapse of the rhythmogenic capacity of the network. This may occur in a manner functionally similar to the physical ablation of neurons from the network (Hayes et al., 2012;Wang et al., 2014). Consistent with our conclusion, manipulations designed to augment opioid-compromised synaptic transmission, for example by either pharmacologically depolarizing presynaptic terminals with specific presynaptic potassium channel blockers, or by allosterically potentiating the open-time kinetics of post-synaptic AMPA receptors with ampakines ), may provide a useful framework and strategy for reversing OIRD. Glutamatergic synaptic transmission is the target of a variety of excitatory neuromodulators (Schneggenburger and Forsythe, 2006;Doi and Ramirez, 2008). The inhibitory action of opioids on glutamatergic synapses, acting in concert with endogenous neuromodulators released in a state-dependent manner, may thus influence the efficacy of in vivo respiratory depression exerted by opioids. Unraveling the interplay between endogenous neuromodulators and exogenous opioids acting on synaptic transmission between rhythmogenic neurons will be a critical next step for understanding and preventing mortality associated with OIRD.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.

ETHICS STATEMENT
The animal study was reviewed and approved by IACUC at Seattle Children's Research Institute.

AUTHOR CONTRIBUTIONS
AW designed research, performed research, and analyzed data. AW and J-MR wrote the manuscript.

FUNDING
This study was funded by grants from the US National Institutes of Health (P01HL090554, R01HL126523-05A1, and R01HL144801-01 to J-MR) and Seed Funds from Seattle Children's Research Institute.