Interactions between respiratory oscillators in adult rats

Breathing in mammals is hypothesized to result from the interaction of two distinct oscillators: the preBötzinger Complex (preBötC) driving inspiration and the lateral parafacial region (pFL) driving active expiration. To understand the interactions between these oscillators, we independently altered their excitability in spontaneously breathing vagotomized urethane-anesthetized adult rats. Hyperpolarizing preBötC neurons decreased inspiratory activity and initiated active expiration, ultimately progressing to apnea, i.e., cessation of both inspiration and active expiration. Depolarizing pFL neurons produced active expiration at rest, but not when inspiratory activity was suppressed by hyperpolarizing preBötC neurons. We conclude that in anesthetized adult rats active expiration is driven by the pFL but requires an additional form of network excitation, i.e., ongoing rhythmic preBötC activity sufficient to drive inspiratory motor output or increased chemosensory drive. The organization of this coupled oscillator system, which is essential for life, may have implications for other neural networks that contain multiple rhythm/pattern generators. DOI: http://dx.doi.org/10.7554/eLife.14203.001


Introduction
Coupled oscillator neural networks driving behavior are widespread, e.g., for swimming (Grillner, 2003), and locomotion (Goulding, 2009;Talpalar et al., 2013). Amongst complex and vital behaviors in mammals, breathing, an exceptionally reliable and continuous behavior throughout postnatal life, is one that we may be closest to understanding (Feldman and Kam, 2015). Not only do we know the location of the neural microcircuits that generate respiratory rhythm, but we also have direct, accurate and reliable behavioral measures of the output, i.e., breathing. We hypothesize that the respiratory rhythm central pattern generator (CPG) in mammals is comprised of two oscillators : inspiratory rhythm originates in the preBö tzinger Complex (preBö tC) in the ventrolateral medulla (Smith et al., 1991) and active expiratory rhythm originates in the rostral medulla ventrolaterally adjacent to the facial nucleus (parafacial lateral region; pF L ) (Pagliardini et al., 2011;Huckstepp et al., 2015).
In mammals at rest, during wakefulness and sleep, when active breathing movements are primarily inspiratory, generation of the underlying rhythm appears driven by the preBö tC. As metabolic demand increases, e.g., during exercise, the pF L appears to turn on to produce active expiration. Thus, while breathing is a unified act of inspiratory and expiratory airflow, we postulate this behavior results from the coordinated interaction of two anatomically and functionally distinct oscillators (Mellen et al., 2003;Janczewski and Feldman, 2006;Pagliardini et al., 2011;Huckstepp et al., 2015). To understand the generation and control of respiration we need to determine how these two oscillators interact. For example, in adult rats the preBö tC can generate inspiratory rhythm while the pF L is quiescent (Pagliardini et al., 2011;Huckstepp et al., 2015). Is the converse true?
To investigate their independent and interactive functions, we used a pharmacogenetic approach to selectively inhibit the preBö tC and/or activate the normally quiescent pF L . We bilaterally transfected the preBö tC with the G i/o -coupled allatostatin receptors (AlstR), which when activated by allatostatin (Alst) silences transfected preBö tC neurons (Tan et al., 2008). In the same rats, we bilaterally transfected the pF L with the G q -coupled HM 3 D DREADD receptor (HM 3 DR) that when activated by clozapine-N-oxide (CNO) depolarizes (Armbruster et al., 2007) transfected pF L neurons. We slowed respiration in a controlled manner by titrating the dose of Alst applied to the AlstRtransfected preBö tC, allowing us to examine the dynamics of this presumptive coupled oscillator system as we shifted the balance of activity from the preBö tC to the pF L . Depressing preBö tC activity resulted in quantal slowing of breathing, similar to slowing of breathing following opiate depression of preBö tC activity in vitro and in juvenile rats (Mellen et al., 2003;Janczewski and Feldman, 2006). As preBö tC activity waned, burstlets appeared on inspiratory muscle electromyograms (EMGs) and in airflow, consistent with our postulate that burstlets, not bursts, in the preBö tC are rhythmogenic (Kam et al., 2013). In complementary experiments, we applied CNO to the HM 3 DR transfected pF L to activate this normally quiescent oscillator. We confirmed that pF L activation initiates active expiration (Pagliardini et al., 2011;Huckstepp et al., 2015). By combining these protocols, i.e., silencing the preBö tC while simultaneously driving the pF L , we removed some confounding factors from these initial experiments. Importantly, we found that active expiration could not be induced when preBö tC inspiratory driven motor activity was suppressed and chemosensory drive was absent, indicating that in adult rats active expiration is driven by the pF L but requires an additional source of network excitation such as ongoing preBö tC activity or chemosensory drive. The eLife digest Mammals breathe air into and out of their lungs to absorb oxygen into the body and to remove carbon dioxide. The rhythm of breathing is most likely controlled by two groups of neurons in a part of the brain called the brain stem. One group called the preBö tzinger Complex drives breathing in (inspiration), and normally, breathing out (expiration) occurs when the muscles responsible for inspiration relax. The other group of neurons -known as the lateral parafacial region -controls extra muscles that allow us to increase our breathing when we need to, such as during exercise.
Huckstepp et al. set out to determine how these two groups of neurons interact with one another in anesthetized rats to produce a reliable and efficient pattern of breathing. The experiments provide further evidence that inspiration is mainly driven by the preBö tzinger Complex. Whilst activity from the lateral parafacial region is needed to cause the rats to breathe out more forcefully than normal, a second low level of activity from another source is also required. This source could either be the preBö tzinger Complex, or some unknown neurons that change their activity in response to the levels of oxygen and carbon dioxide in the blood or fluid of the brain.
Further investigation is required to identify how these interactions go awry in diseases that affect breathing, such as sleep apneas.

Results
Viral targeting of the preBö tC and pF L We made two pairs of viral injections in each adult rat, i.e., bilateral injections into the preBö tC ( Figure 1A,B) and into the pF L ( Figure 1C,D). In histological sections the preBö tC is defined as the neurokinin-1 receptor (NK1R) dense area ventral to the semi-compact nucleus ambiguous ( Figure 1A,B) and the pF L is defined as the area ventral to the lateral edge of the facial nucleus, juxtaposed to the spinal trigeminal tract ( Figure 1C,D) (Huckstepp et al., 2015). In representative 40 mm sections: from preBö tC injection sites, 161 ± 42, representing 82 ± 6%, neurons expressed GFP (n=3); from pF L injection sites, 112 ± 31, representing 81 ± 3%, neurons expressed mCitrine (n=3). Transfection sites ranged from~350-600 mm in diameter. As the responses to Alst and CNO did not differ between the largest and smallest injections sites, the effects of receptor activation were due to silencing or driving of the preBö tC and pF L respectively and not to spread of virus to neighboring regions. We found no labeling of neurons in regions of the medulla other than within the injection sites (data not shown). In particular, we found no fluorescent reporters, i.e., eGFP or mCitrine, in the Bö tzinger complex (Bö tC; Figure 1E,F).

Silencing the preBö tC during active expiration leads to apnea
The preBö tC can generate inspiratory rhythm while the pF L is quiescent (Pagliardini et al., 2011). Can the pF L generate a respiratory rhythm when the preBö tC is quiescent? If active expiration is independent of inspiratory activity, then active expiration driven by activation of HM 3 DRs in pF L should persist after cessation of inspiratory activity resulting from injection of Alst in the preBö tC. In anesthetized adult rats with an active expiratory breathing pattern induced by CNO, injection of Alst into the preBö tC led to a significant decrease in respiratory activity (n=8; p=0.008 for all variables; Figure 7), ultimately resulting in apnea, with no rhythmicity in V T , Ð Dia EMG , Ð GG EMG , and Ð Abd EMG . After rhythmic inspiratory activity ceased, Ð Abd EMG continued tonically for a short duration (15 IQR 12 s; Figure 7A,Bii) before disappearing, which was not different from the duration of tonic Ð Abd EMG following the onset of apnea in the absence of CNO (Figure 2Biii; p=0.7). When inspiratory activity stopped, rats were severely hypercapnic (ETCO 2 64.6 IQR 14.5 mm Hg; Figure 7Bii), this was not different from the hypercapnia at the onset of apnea in the absence of CNO (ETCO 2 57.8 Activity following bilateral Alst injection. (C) Comparison between ventilation in rats at rest (Rst) and following Alst: (Ci) before Alst had taken full effect ( » Bii). (Cii) After Alst had taken full effect ( » Biii). Lines connect data from individual experiments, box and whisker plots show combined data. Data are normalized to highest parameter, i.e., f, T I, T E , V T , Ð GG EMG , Ð Dia EMG , or Ð Abd EMG , value regardless of whether it belonged to control or Alst group. ffrequency, T I -inspiratory period T E , -expiratory period, V T -tidal volume, Ð GG EMG -integrated genioglossus electromyogram, Ð Dia EMG -integrated diaphragm electromyogram, Ð Abd EMG -integrated abdominal electromyogram. DOI: 10.7554/eLife.14203.004 IQR 22.3 mm Hg; p=0.8). Rats were mechanically ventilated to restore CO 2 and O 2 levels to within the normal range ( Figure 7). To assess the state of the respiratory CPG the mechanical ventilator was intermittently turned off. Shortly after allatostatin injections into the preBö tC, no respiratory activity was seen following cessation of mechanical ventilation, and rats were placed back on the ventilator. As the time from allatostatin injection in the preBö tC increased, removal from the ventilator led to brief periods of spontaneous breathing. When spontaneous breathing occurred, the return of inspiratory activity always preceded the return of expiratory abdominal activity; active expiration did not return until V T reached 3.2 IQR 1.7 mL, Ð Dia EMG reached 13.1 IQR 8.2 a.u., and Ð GG EMG reached 5.2 IQR 4.2 a.u. (n=8; Figure 8). That is, at low levels of reinitiated inspiratory activity there was no active expiration, which appeared only after inspiratory activity reached a higher value. If rats were able to spontaneously breathe without ventilation, they were allowed to do so. However, most early periods of spontaneous breathing deteriorated, and when rats became apneic again they were re-ventilated; in 5 of 6 rats spontaneous breathing occurred during ventilation, and in the remaining rat, spontaneous breathing following removal from the ventilator was sustained. When spontaneous breathing occurred during ventilation the return of inspiratory activity always preceded the return of expiratory abdominal activity. Therefore, following silencing of the preBö tC with Alst in the presence of CNO to drive the pF L , at no time during the re-initiation of spontaneous breathing, either with or

Technical consideration
Our experimental design required long lasting activation of receptors at 4 separate loci, i.e., pre-Bö tC bilaterally and pF L bilaterally. Optogenetics can have millisecond resolution, making them a valuable tool for studying breathing on a breath by breath basis, e.g., (Pagliardini et al., 2011). However opsins can rapidly (>seconds) desensitize and their activation requires placement of optic fibers to illuminate each transfected area, a particularly challenging problem for four regions in the brainstem. Therefore, we chose a pharmacogenetic approach to excite the pF L by activation of HM 3 DRs and inhibit the preBö tC by activation of AlstRs. This way we were able to switch neurons in these regions on or off over a period of minutes by application of CNO to the ventral surface of the medulla, and careful titration of the dose and infusion rate of Alst injected into the preBö tC.
We note AAV2/5 can retrogradely transfect of some types of afferent neurons, e.g., injections of AAV2/5 into the entorhinal cortex retrogradely labels a subset of dentate gyrus neurons, but not the dense afferent projections from other nuclei (Aschauer et al., 2013). We assert that our results are unlikely to be confounded by retrograde transfection of distant neurons: i) Other than the injection site, we found no labeling of medullary neurons following transfection of the pF L with AAV2/5 viruses that coexpress mCitrine with a DREADD receptor, in this ( Figure 1E,F) or our previous (Huckstepp et al., 2015) study; ii) CNO was applied directly to the ventral medullary surface, and the effective concentration for receptor activation would be limited to the first few hundred microns beneath the ventral surface, and; iii) we find essentially the same effects of activation of neurons in the pF L as optogenetic photoactivation of the same neurons transfected with lentivirus (Pagliardini et al., 2011), which is not retrogradely transported.
Here we discuss the output of the preBö tC and pF L , which ultimately form the final output of the respiratory network. Though we do not refer to other nuclei, we do not rule out their contribution to the control of respiration; for example following silencing of the preBö tC, the increase in chemosensory drive to the respiratory oscillators, may come from increased drive from other respiratoryrelated nuclei, such as the pontine nuclei, i.e., locus coreleus, parabrachial nucleus, and the Kolliker Fuse nucleus, the ventral respiratory group or other medullary nuclei, i.e., RTN/pF V , medullary raphe, nucleus tractus solitarii, or even astrocytes. In addition, due to the necessity to perform injections into the medulla, we were unable to record directly from the preBö tC or pF L . Therefore, the activity of the preBö tC and pF L, was assessed by motor output recorded from respiratory muscles, which may not always reflect the activity of these oscillators, as they may have subthreshold activity, e.g., (Kam et al., 2013).

Discussion
Many rhythmic behaviors are driven by neural networks that (presumably) contain coupled oscillators (Grillner, 2003;Goulding, 2009). Here, we examined the coupled oscillator microcircuit that controls breathing, a motor behavior that is presently unique insofar as the locations of its oscillators, i.e., pre-Bö tC and pF L , are known (Smith et al., 1991;Pagliardini et al., 2011;Feldman et al., 2013;Feldman and Kam, 2015;Huckstepp et al., 2015). In rats, while the function of the preBö tC as the critical site for generation of inspiratory rhythm is consistent across all relevant developmental stages, from the third trimester in utero through adulthood, and state, e.g., sleep-wake, rest-exercise (Smith et al., 1991;Janczewski and Feldman, 2006;Tan et al., 2008;Kam et al., 2013), the presumptive function of the pF L and its coupling with the preBö tC appears to change developmentally and with state ( Figure 9) (Onimaru and Homma, 2003;Iizuka and Fregosi, 2007;Oku et al., 2007;Onimaru et al., 2008;Thoby-Brisson et al., 2009;Pagliardini et al., 2011;Huckstepp et al., 2015). To understand the context in which we interpret our data, we first summarize this disparate literature with reference to Figure 9.
From third trimester in utero to adulthood, NK1R-expressing neurons lying under the lateral edge of the facial nucleus and extending laterally out to the spinal trigeminal tract have respiratory-rhythmic burst activity (Onimaru et al., 2008;Thoby-Brisson et al., 2009;Pagliardini et al., 2011;Huckstepp et al., 2015). In E14.5 mice, these neurons are designated as comprising the embryonic parafacial nucleus (e-pF) (Thoby-Brisson et al., 2009); in postnatal rats, these neurons are designated as comprising the parafacial respiratory group (pFRG) (Onimaru and Homma, 2003), and; in the adult rat, these neurons are designated as comprising the pF L (Pagliardini et al., 2011;Huckstepp et al., 2015). While 3 distinctly different groups of NK1R neurons expressing respiratoryrhythmic bursting could exist in sequence within this homologous anatomical location at distinctly different developmental time points, the most parsimonious interpretation is of one population of neurons studied at different developmental time points; we postulate that this is the case. For the balance of the DISCUSSION, we will refer to this location as the pF L .
NK1R-expressing neurons in the ventral respiratory column (including the parafacial region) are almost exclusively glutamatergic (Guyenet et al., 2002). In embryonic and postnatal rodents, the Figure 9. Developmental and state-dependent changes in coupling between pF L and preBö tC. Functional connections of undetermined connectivity are indicated as broken lines. As pF L neurons are excitatory (Onimaru et al., 2008;Thoby-Brisson et al., 2009) and lack inhibitory markers (Ellenberger, 1999;Tanaka et al., 2003), inhibitory connections from pF L to preBö tC are indirect (see Figure 7 in Huckstepp et al., 2015).  (Smith et al., 1991), the preBö tC can oscillate in the absence of the pF L in transverse slices, and (Bii) the pF L can oscillate independently following suppression of preBö tC rhythm by bath application of opioid agonists (Takeda et al., 2001;Janczewski et al., 2002). Biii) Immediately following birth, respiratory rhythm is driven by pF L (Onimaru and Homma, 2003;Oku et al., 2007). (Biv) Shortly after birth (>1 day), the breathing CPG becomes driven by the preBö tC (Oku et al., 2007). (C) Juvenile stage. (Ci) Expiration and inspiration alternate and are reciprocally coupled. (Cii) PreBö tC and pF L are differentially affected by fentanyl, which shifts breathing to an expiratory-dominant pattern. (Cii + iii) preBö tC and pF L can be independently suppressed by activation of Breuer-Hering deflation reflex (BHDR; Cii) or inflation reflex (BHIR; Ciii) (Janczewski and Feldman, 2006). (D) Adult Stage: (Di) breathing is inspiratory driven by preBö tC while pF L activity is normally suppressed at rest (also see Pagliardini et al., 2011 andHuckstepp et al., 2015); (Dii) activation of HM 4 DR transfected pF L neurons by CNO (see Figure 6) or optogenetic activation (Pagliardini et al., 2011), or suppression of AlstR transfected preBö tC neurons with Alst (see Figure 2, 7) can induce active expiration; (Dii) as preBö tC neurons project to the pF V but do not appear to project to the pF L (Tan et al., 2010), excitatory drive from the preBö tC to the pF L is most likely through an intermediate excitatory relay, such as the pF V . (Diii) Depression of inspiration by Alst, in presence or absence of CNO, leads to tonic expiratory activity during hypercapnia (see Figures 2, 7) or (Div) apnea during hypoxia (see Figures 2, 7). (Dv) As breathing returns, abdominal activity remains absent until inspiratory activity is near normal levels (see Figure 8), implicating an indirect involvement of preBö tC excitatory neurons in expiration either through its excitatory projections throughout breathing CPG (Tan et al., 2010), including pF V that contributes to expiratory activity (Huckstepp et al., 2015), or through mechanosensory feedback that can provides expiratory drive (Remmers, 1973;Davies and Roumy, 1986;Janczewski and Feldman, 2006). DOI: 10.7554/eLife.14203.011 pF L is comprised of glutamatergic NK1R-expressing neurons (Onimaru et al., 2008;Thoby-Brisson et al., 2009). In the adult rodent, the presumptive pF L expiratory oscillator is comprised (mostly) of NK1R-expressing neurons ( Figure 1D, see also Huckstepp et al., 2015). Since the pF L and pF V do not appear to contain any inhibitory neurons (Ellenberger, 1999;Stornetta and Guyenet, 1999;Tanaka et al., 2003;Fortuna et al., 2008;Abbott et al., 2009), we conclude that any resultant inhibitory action associated with their activity occurs through an intermediate relay of inhibitory neurons, perhaps located in the preBö tC (Morgado-Valle et al., 2010) or Bö tC (Schreihofer et al., 1999).
In E14.5 mice, the pF L is rhythmic before the preBö tC appears to form (Figure 9Ai; see also Thoby-Brisson et al., 2009). At E15.5, the preBö tC forms (Thoby-Brisson et al., 2005) and it becomes rhythmically coupled to the pF L ( Thoby-Brisson et al., 2009). At this time: i) the preBö tC and pF L can oscillate independently following a transverse section caudal to facial nucleus ( Figure 9Aiii); ii) the pF L can oscillate in the absence of preBö tC rhythm following bath application of the glutamatergic antagonist CNQX (Figure 9Aiv), and; iii) the preBö tC can oscillate in the absence of the pF L following administration of the sodium channel blocker riluzole (Figure 9Av) (Thoby-Brisson et al., 2009). Thus embryonically, the preBö tC and pF L are independent, coupled oscillators, and rhythmic output of the network is a convolution of the faster rhythm of the pF L and the slower rhythm of the preBö tC (Thoby-Brisson et al., 2009); the pF L ( Thoby-Brisson et al., 2009) provides excitatory drive to the preBö tC, while bursting in the preBö tC, comprised of inhibitory (Rahman et al., 2015) and excitatory neuronal activity (Gray et al., 2010), inhibits and excites pF L neurons (Thoby-Brisson et al., 2009). Whether the projections between the preBö tC and pF L are direct or indirect (Figure 9Aii) is yet to be determined (Thoby-Brisson et al., 2009).
In postnatal rats, respiratory rhythm appears similar to the embryo; the glutamatergic pF L provides excitatory drive to the preBö tC (Onimaru and Homma, 2003;Onimaru et al., 2012), while bursting of the preBö tC, comprised of inhibitory (Morgado-Valle et al., 2010) and excitatory neurons (Gray et al., 2010), inhibits and excites different subsets of pF L neurons (Takeda et al., 2001;Onimaru et al., 2007). At this time: i) the preBö tC can oscillate in transverse slices, i.e., in the absence of the pF L (Figure 9Bi) (Smith et al., 1991) and; ii) the pF L can oscillate following suppression of preBö tC rhythm by bath application of opioid agonists (Figure 9Bii) (Takeda et al., 2001;Janczewski et al., 2002). Thus postnatally, the preBö tC and pF L are also independent, coupled oscillators. Immediately following birth, respiratory rhythm appears driven by the pF L (Figure 9Biii) (Onimaru and Homma, 2003;Oku et al., 2007), perhaps to protect against the opioid surge in the fetal brain during birth (Janczewski et al., 2002) that would act to suppress preBö tC activity (Takeda et al., 2001). Shortly after birth (~P2-P4), the breathing central pattern generator (CPG) matures, and becomes driven by the preBö tC (Oku et al., 2007;Mckay et al., 2009;Kennedy, 2015) (Figure 9Biv), and remains so throughout life.
Here, we used a novel pharmacogenetic strategy to investigate how the preBö tC and pF L interact to produce the appropriate breathing pattern in adult spontaneously breathing vagotomized rats ( Figure 9D). The preBö tC is the dominant oscillator, active at rest to drive inspiratory movements (Figure 9Di, see also Pagliardini et al., 2011 andHuckstepp et al., 2015). The pF L is the subsidiary conditional oscillator, quiescent at rest (Figure 9Di); in the presence of concurrent, or reduced, pre-Bö tC activity (Figure 2), and appropriate inputs, such as an increase in expiratory drive through activation of exogenous receptors (Figure 6, see also Pagliardini et al., 2011) or altered blood gases (Huckstepp et al., 2015), the pF L drives active expiration (Figure 9Dii). In the absence of preBö tC activity, chemosensory drive increases network excitability and drives active expiration, though abdominal activity is tonic due to a lack of phasic inhibition from the preBö tC (Figure 9Diii). However, the pF L appears incapable of independently driving respiratory movements in the adult rat when preBö tC activity is silenced by exogenous receptors (Figures 2, 7, 9Div) or when preBö tC activity is low and blood gases are normal (Figures 8, 9Dv). As preBö tC neurons project to the pF V but do not appear to project to the pF L (Tan et al., 2010), excitatory drive from the preBö tC to the pF L is most likely through an intermediate excitatory relay, such as the pF V (Figure 9Dii).
Silencing the preBö tC leads to quantal slowing of breathing and transmission of burstlets to motoneuron pools Hyperpolarizing preBö tC neurons initially decreased f, V T , and Ð Dia EMG (Figures 2, 3, see also Tan et al., 2008). Breathing slowed in a quantal manner (Mellen et al., 2003) with missed inspiratory bursts, rather than through a gradual increase in period (Figures 2Bii, 3A-C). Concurrently, as drive within the preBö tC diminished and chemosensory drive increased, presumptive burstlets in the pre-Bö tC (Kam et al., 2013) appeared to be transmitted to motor pools, leading to low level inspiratory motor activity, i.e., Ð Dia EMG and Ð GG EMG , resulting in minimal inspiratory airflow ( Figure 3A-B). Here, lowered drive in the preBö tC from activation of the allatostatin receptor creates burstlets in the preBö tC; under normal conditions these burstlet signals are not strong enough to drive motor activity. However, increased excitability in the premotor and motor network caused by increased chemosensory drive (from decreased ventilation) causes these normally subthreshold events to become suprathreshold, and thus these burstlets are transmitted to the motor output. During quantal slowing of breathing, inspiratory activity on the Dia EMG and GG EMG was interspersed with Abd EMG activity (Figures 2A-B, 7A-B). Interestingly, even when expiratory activity was at its highest, low level preBö tC activity, as seen as burstlets on the Dia EMG and GG EMG was still able to inhibit Abd EMG activity (Figure 7Bii). These observations are consistent with our hypothesis that burstlets originate in the preBö tC and, under atypical conditions can be transmitted to motoneuron pools to be seen as small events in muscle EMGs (Kam et al., 2013).
Silencing the preBö tC leads first to active expiration then to apnea Following hyperpolarization of preBö tC neurons by activation of AlstRs, active expiration appeared (Figures 2, 3, 9Dii). Hypoxia and hypercapnia can themselves induce active expiration (Iizuka and Fregosi, 2007;Huckstepp et al., 2015), whether changes in blood gases are sensed by pF L neurons, or by neurons driving the pF L , remains to be determined. In either case, active expiration could have been due to increased hypoxic and/or hypercapnic drive to the pF L as inspiratory movements waned (Figures 2A-B, 7A-B, 9Dii). Alternatively, active expiration could be due to disinhibition of a conditional expiratory oscillator resulting from a loss of (presumptive) inhibitory preBö tC drive (Kuwana et al., 2006;Morgado-Valle et al., 2010) to the pF L (Figure 9Dii). Subsequently, when inspiratory motor outflow disappeared and hypercapnia and hypoxia inexorably increased, active expiration ceased (Figures 2, 9Div, see also Tan et al., 2008), perhaps due to the inhibitory effect of severe hypoxia on expiratory motor output (Sears et al., 1982;Fregosi et al., 1987). At normal blood gas levels, active expiration may be seen in the absence of inspiratory motor activity resulting from lung inflation, e.g., in juvenile rat (Janczewski and Feldman, 2006) ( Figure 9C). Nonetheless, from the data presented here in the adult rat, we conclude a necessary role of preBö tC excitatory drive in generating active expiration in normal breathing (discussed further below).
Activating the pF L can induce active expiration but not during to apnea Depolarizing pF L neurons by activation of HM 3 DRs ( Figure 6) changes breathing in a manner similar to their disinhibition or optogenetic photostimulation (Pagliardini et al., 2011). In the presence of inspiratory motor output, depolarizing pF L neurons led to substantial active expiration, but when inspiration ceased so did active expiration (Figures 7, 9Div). Tupal et al observed the absence of expiratory rhythm in perinatal mice lacking Dbx1 neurons and concluded that Dbx1-derived parafacial neurons are an essential component of an expiratory oscillator (Tupal et al., 2014). Our data suggests an alternative interpretation, that suppressing preBö tC neuron activity, either acutely as done here or perhaps genetically via Dbx1 deletion, is sufficient to prevent active expiration, without any need to invoke explicit perturbations of the pF L .
Loss of inhibitory output from the preBö tC leads to a loss of phasic expiratory activity As rats went from eupnea to apnea following suppression of preBö tC activity, and after initiation of active expiration, Ð Abd EMG transitioned from phasic to tonic: Ð Abd EMG oscillated with a normal antiphase relationship to inspiration when preBö tC drive and hypercapnia were moderate, but became tonic during apnea accompanied by severe hypercapnia (Figures 2A-B, 7A-B, 9Diii). Similarly, in anesthetized cats as ventilation moves in the opposite direction from hypocapnic apnea to eupnea with a resultant increase in CO 2 , expiratory motor activity transitions from tonic to phasic (Sears et al., 1982). We suggest, like Sears et al, that near the transition from eupnea to apnea, expiratory activity is phasic due to periodic inhibition of tonic activity during inspiration, and becomes tonic once the phasic inspiratory inhibition is lost (Figures 2Bii-iii, 7Bii, 9Diii). We suggest the loss of phasic inhibition is the result of silencing of inhibitory preBö tC inspiratory neurons by Alst (Kuwana et al., 2006;Morgado-Valle et al., 2010), implying a significant role for inhibitory preBö tC neurons in shaping expiratory output.
Respiratory network activity is a requirement for active expiration Interestingly, tonic Ð ABD EMG activity was not seen during the transition from apnea to eupnea (Figures 8Aii, 9Dv) in contrast to its presence during the reverse transition from eupnea to apnea (Figures 7Bii, 9Diii). Following apnea during ventilation to maintain blood gases, when pF L neurons were excited by activation of HM 3 DRs there were three phases to the re-initiation of breathing. Initially, soon after silencing the preBö tC, no inspiratory or expiratory activity was seen, even in the absence of mechanical ventilation. Next, presumably as the effect of Alst on preBö tC neurons was waning, no inspiratory or expiratory activity was seen during ventilation, but spontaneous breathing was present upon removal from the ventilator. Here, spontaneous breathing was likely due to increased chemosensory drive to the preBö tC overcoming the waning hyperpolarizing effect of the Alst on preBö tC neurons. Ultimately, as spontaneous breathing continued and chemosensory drive diminished, rats would once again become apneic and require mechanical ventilation ( Figure 8A-C). During this phase, upon removal from the ventilator, Ð Abd EMG did not return until inspiratory motor, and presumably preBö tC, activity reached a threshold level ( Figure 8A-C). Finally, spontaneous breathing occurred during ventilation, once the effect of Alst on preBö tC neurons had worn off ( Figure 8D). During this phase, inspiratory activity on the GG EMG and Dia EMG always returned before Abd EMG ( Figure 8D). At no time during mechanical ventilation, or during the periods where rats were briefly removed from the ventilator, did active expiration occur in the absence of inspiration. As blood gases were normal (Figure 8), the loss of expiratory activity is unlikely to be due to excessive hypoxia or hypercapnia. Thus there appears to be a minimum level of respiratory network excitability required for active expiration. In the absence of preBö tC activity, this network excitability may be provided by chemosensory drive, accounting for the Ð Abd EMG activity seen shortly before the onset of apnea. However following recovery from apnea the increase in network excitability is provided by the preBö tC either directly through its extensive excitatory projections throughout the respiratory network (Tan et al., 2010), e.g., the pF V that can modulate expiratory activity (Marina et al., 2010;Huckstepp et al., 2015), or indirectly when V T became large enough to induce mechanosensory feedback (that can provide expiratory drive, e.g., Remmers, 1973;Davies and Roumy, 1986;Janczewski and Feldman, 2006).

Summary
We have developed a new strategy using complementary pharmacogenetics for studying coupled oscillator systems. By independently altering the excitability of two anatomically and functionally separate respiratory oscillators, we have uncovered a fundamental interaction and further delineated their role within the breathing CPG. We conclude that though respiration results from the interaction of two distinct oscillators, the preBö tC (inspiration) and the pF L (expiration) are not organized as symmetrical half centers. Instead in the adult rat, the preBö tC is the dominant oscillator and the pF L is the subsidiary conditional oscillator that is normally suppressed at rest; whereas the preBö tC can drive breathing alone, the pF L is unable to drive breathing (including active expiration) in the absence of an additional form of network excitation, i.e., ongoing rhythmic preBö tC activity sufficient to drive inspiratory motor output or increased chemosensory drive when changes in blood gases are below the threshold where hypoxia is sufficient to inhibit abdominal muscle recruitment. This hierarchy is established by the sensitivity of the system to each oscillator, even when preBö tC drive is low, it is able to drive inspiratory bursts and inhibit expiration, whereas even when pF L drive is high, it requires a certain amount of network excitability from other sources to drive expiratory activity. This asymmetrical organization may be relevant to other neural networks that contain hierarchically organized coupled oscillators for pattern generation, such as may underlie the asymmetrical acts of flexion and extension in locomotion (Grillner, 2003;Talpalar et al., 2013). The interactions of the preBö tC and pF L change with development and maturation (Figure 9), and with state, represent another layer of complexity in understanding the neural control of breathing.

Ventral approach
Anesthesia was induced with isofluorane and maintained with urethane (1.2-1.7 g/kg; Sigma, St Louis, MO) in sterile saline via a femoral catheter. Rats were placed supine in a stereotaxic apparatus on a heating pad to maintain body temperature at 37 ± 0.5˚C. The trachea was cannulated, and respiratory flow was monitored via a flow head (GM Instruments, UK). A capnograph (Type 340: Harvard Apparatus, Holliston, MA) was connected to the tracheal tube to monitor expired CO 2 , as a proxy of blood gases homeostasis. Paired electromyographic (EMG) wires (Cooner Wire Co, Chatsworth, CA) were inserted into genioglossal (GG), diaphragmatic (Dia), and oblique abdominal muscles (Abd). Anterior neck muscles were removed, a basiooccipital craniotomy exposed the ventral medullary surface, and the dura was resected. A bilateral vagotomy was performed to remove confounding factors such as feedback from lung stretch receptors that can drive abdominal activity (Janczewski and Feldman, 2006), after which exposed tissue around the neck and mylohyoid muscle were covered with dental putty (Reprosil; Dentsply Caulk, Milford, DE) to prevent drying. As micturition is inhibited under anesthesia, rats bladders were expressed preceding and during the experiment to remove any risk of autonomic dysreflexia from bladder distension; to maintain fluid balance rats were given an IP injection of saline every time the bladder was expressed. Rats were left for 30 min for breathing to stabilize. At rest, spontaneous breathing consisted of alternating active inspiration and passive expiration. From a ventral approach allatostatin (Alst; 10 mM; 100-200 nL; Antagene Inc, Sunnyvale, CA) in sterile saline was injected bilaterally into the preBö tC to hyperpolarize neurons transfected by AlstRs. Coordinates were (lateral from the basilar artery, caudal from the rostral hypoglossal nerve rootlet, dorsal from the ventral surface in mm): preBö tC; (2.0, 0.6, 0.7). Small adjustments were made to avoid puncturing blood vessels. Rats were ventilated for the duration of ensuing apnea. After breathing stabilized, clozapine-N-oxide (CNO; 90 mM; Santa Cruz Biotechnology, Dallas, TX) in sterile saline was applied to the ventral medullary surface to depolarize pF L neurons transfected with HM 3 DRs. Once breathing stabilized, rats received a second set of bilateral injections of Alst. Though rats were ventilated with room air for the duration of ensuing apnea, they were intermittently removed from the ventilator to assess spontaneous breathing, and drives to inspiration and expiration. Ventilation depths and speed were chosen to match end-tidal CO 2 to that when the rat was spontaneously breathing room air. Once rats were spontaneously breathing and no longer required ventilation, CNO was removed from the ventral surface of the medulla and the medulla was washed in PBS. All 8 rats underwent the entire procedure, i.e., there were 8 biological replicates, and were only exposed to each condition once, i.e., there were no technical replicates. In age-matched rats not transfected with AlstR-or HM 3 DR-expressing AAVs, we injected Alst into the preBö tC or applied CNO to the medullary surface to see if these protocols produced non-specific effects All 8 rats underwent the entire procedure, i.e., there were 8 biological replicates, and were only exposed to each condition once, i.e., there were no technical replicates.

Data analysis and statistics
Sample sizes were calculated using Gpower 3 v3.1.9.2 (http://www.ats.ucla.edu/stat/gpower/); using a 'means: Wilcoxon signed-rank test (matched pairs)' test, with a desired power of 90%, at a 5% significance level, and an effect size of 1.15 (calculated from the initial effect of Alst on respiratory frequency). Data were only included from animals where the preBö tC and pF L were successfully targeted bilaterally, and no data were excluded from these animals. All statistical analysis was performed in Igor Pro (WaveMetrics, Lake Oswego, OR).
EMG signals and airflow measurements were collected using preamplifiers (P5; Grass technologies, Rockland, MA) connected to a Powerlab AD board (ADInstruments, Australia) in a computer running LabChart software (ADInstruments), and were sampled at 400 Hz/channel. High pass filtered (>0.1 Hz) flow head measurements were used to calculate: tidal volume (V T , peak amplitude of the integrated airflow signal during inspiration, converted to mL by comparison to calibration with a 3 mL syringe), inspiratory duration (T I , beginning of inspiration until peak V T ), expiratory duration (T E , peak V T to the beginning of the next inspiration), and f (1/[T I +T E ]). EMG data, expressed in arbitrary units (A.U.), were integrated (t=0.05 s; Ð Dia EMG , Ð GG EMG , and Ð Abd EMG ) and peak amplitude of each signal was computed for each cycle. To obtain control values, all parameters except end-tidal CO 2 (ETCO 2 ), were averaged over 20 consecutive cycles preceding each experimental manipulation (X control ). After Alst, measurements were taken at 2 time points: i) 20 cycles were averaged where only partial effects were seen, and; ii) 20 points were averaged following apnea. After CNO, 20 cycles were averaged after breathing had stabilized. After Alst in the presence of CNO, 20 points were averaged following apnea. In the presence or absence of CNO, capnograph peaks were averaged for 10 cycles preceding Alst, and for 5-10 cycles preceding apnea. Following removal from ventilation, the amplitude of the inspiratory bursts and Ð Dia EMG and Ð GG EMG activity immediately preceding the first Ð Abd EMG burst (Figure 8, blue lines dashed lines) were recorded to calculate inspiratory parameters at which active expiration returned.
For each rat we obtained X control , and the average of 20 cycles during the stimulus (X stimulus ). X control values and their associated X stimulus values for each parameter in each rat were combined into a single data set. To facilitate graphical comparisons data were normalized to the highest value in the data set regardless of whether it belonged to X control or X stimulus group (C in Figure 2, 4-7). Therefore the highest value in the data set, whether it be X control or X stimulus , was 1.0; except for measurements of V T , Ð Dia EMG , and Ð GG EMG following ventilation, which are displayed as absolute values. Recorded data were not normally distributed, and were therefore analyzed using nonparametric statistical test, and reported as median and interquartile range (IQR). Statistical tests performed in Igor Pro (WaveMetrics), are 2-sided Wilcoxon signed-rank tests with a significance level of p 0.05. Data are displayed as box and whisker plots for comparison of groups, and as line graphs for individual experiments. Kernel density estimations (Parzen, 1962;Epanechnikov, 1969), were used to determine the distribution of respiratory periods. After calculating the optimal bandwidth, i. e., bin size (Park and Marron, 1990;Sheather and Jones, 1991), the data was smoothed (Cao et al., 1994) and plotted. The modality of kernel density plots, were used to assess baseline respiratory periods and whether breathing slowed by quantal integers of that baseline. Bandwidth selection, data smoothing, and kernel density plots were performed in Microsoft excel (Microsoft Corporation, Redmond, WA) using an add-in written by the royal society of chemistry (http://www. rsc.org/Membership/Networking/InterestGroups/Analytical/AMC/Software/kerneldensities.asp).

Funder
Grant reference number Author National Institutes of Health NIH HL 70029 Jack L Feldman The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Author contributions RTRH, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article; LEH, KPC, Acquisition of data, Drafting or revising the article; JLF, Conception and design, Analysis and interpretation of data, Drafting or revising the article