INTRODUCTION

TOP ABSTRACT INTRODUCTION THEORY AND METHODS RESULTS DISCUSSION REFERENCES

MEMORY AND PHASE-DEPENDENT gating of afferent inputs are hallmarks of the respiratory control system (reviewed in Refs. 18, 52, 57, 61). In particular, brief hypoxia or electrical stimulation of the carotid sinus nerve (CSN) in some mammalian species induced short-term potentiation (STP) of inspiratory drive, characterized by an afterdischarge of phrenic motor activity (79), and a short-term depression (STD) of respiratory rhythm, characterized by a poststimulus decrease of respiratory frequency (6, 10, 23). Recently, it has been shown that carotid STP manifests itself not only as afterdischarge of inspiratory activity but also as sustained shortening of inspiratory duration (TI) in the poststimulus period (59). Furthermore, both forms of STP resemble a biphasic leaky integrator (52, 59) with differing integration time constants in the induction phase and recovery phase during and after stimulation, respectively (59, 79). In contrast, recent studies showed that carotid STD of respiratory frequency stems mainly from a corresponding prolongation of expiratory duration (TE), with induction and recovery dynamics resembling those of a biphasic leaky differentiator (58). Both carotid STP and STD are abolished or braked by pharmacological blockade of N-methyl-D-aspartate (NMDA) receptor channels (7, 58, 59).

These phenomena, when taken together, point to several interesting features of carotid chemoafferent signaling for respiratory pattern generation. First, the distinctive effects of carotid chemoreflex, STP, and STD on inspiratory motor output and respiratory rhythm suggest a plurality of carotid chemoafferent pathways. Among them, a critical area in the ventrolateral pons has been shown to mediate the hypoxia-induced carotid STD of respiratory frequency (10). Although the loci of carotid STP and carotid chemoreflex pathways are not clearly understood (reviewed in Ref. 57), it has been suggested that stimulation of carotid chemoreceptors may activate parallel central pathways that contribute separately to the shortening of TI and increase of inspiratory motor output (42, 43, 71). Second, the differing effects of carotid STP and STD on TI and TE suggest that the corresponding pathways may be logically gated or locked to the inspiratory (I) and expiratory (E) phases of the respiratory rhythm, respectively. Third, such dynamic afferent modulation of the respiratory rhythm, with a short-term memory that persists beyond the primary stimulus, implies that the maintenance of carotid STP and STD may involve tonic secondary inputs in convergent pathways preceding the respiratory oscillator, in addition to the primary CSN stimulus itself (57-59). Finally, the complementarities of carotid STP and STD as neural integrator and differentiator, or low-pass and high-pass filters (57, 82), with similar dependence on NMDA receptors suggest that they may spring from similar mechanisms. These observations shed light into the intricate architecture of the neural network that regulates carotid chemoafferent traffic in the brain.

To elucidate the integrative behavior of this complex neural network en bloc, we first propose a "top-down" (84) working model that consolidates the aforementioned discrete hypotheses regarding the pathway and phase specificity of carotid chemoreflex, STP, and STD. Within this conceptual framework, we introduce an advanced system identification method called "stroboscopic interferometric filtering technique" (SIFT), an analytical approach (82) employing strobe stimuli that are phase locked to a base rhythm to resolve the spatiotemporal components of afferent signaling in parallel gated or ungated pathways. With the use of this analytical procedure, we characterized the dynamic logic responses of the respiratory-pattern generator to strobe CSN inputs in anesthetized, glomectomized, vagotomized, paralyzed, and ventilated rats. The results corroborated the working model and revealed some novel features that were obscured under stepwise CSN pulse-train stimulation or hypoxic stimulation typical of previous studies. The resultant model of carotid chemoafferent signaling unifies several reported neuronal mechanisms and offers a methodical guidepost for future elucidation of other postulated elemental structures and functions at the cellular level.

	THEORY AND METHODS

TOP ABSTRACT INTRODUCTION THEORY AND METHODS RESULTS DISCUSSION REFERENCES

Working Model of Carotid Chemoafferent Signaling

View larger version (22K): [in this window] [in a new window]   Fig. 1.   A: working model of carotid chemoafferent processing. The respiratory oscillator is depicted as biphasic with reciprocally inhibiting expiratory (E) and inspiratory (I) aggregate superneurons that respectively inhibit () and excite () inspiratory premotoneurons (I-PMN). Inputs from carotid sinus nerves (CSN) induce synaptic accommodation and short-term depression (STD) or long-term depression (LTD) (inner ) in the nucleus tractus solitarius (NTS), which signals the respiratory oscillator via parallel reflex and leaky integrator (L) pathways. Carotid STD is mediated by an integrator pathway that promotes the E phase, whereas carotid short-term potentiation (STP) is mediated by an inhibitory integrator pathway to the I phase and an excitatory integrator pathway to I-PMN. A reflex pathway (dotted line) may also directly excite I-PMN in cat (79) and rat (23) perhaps only with relatively high-CSN stimulus amplitudes, since lower stimulation amplitudes did not elicit this reflex in the rat (59). B: Monophasic integrator with step (top) or differentiator (bottom) may be produced in a primary, autogated pathway such as NTS or other ("?") sites through monosynaptic STP (inner ) or STD (inner ) with augmentation or accommodation of synaptic transmission, respectively. Open boxes show neural activity, and shaded boxes show synaptic activity. C: biphasic integrator is obtained if the primary input presynaptically induces STP (double  with ) in a secondary pathway (such as the pons) excited by a tonic input, which sustains an "afterdischarge" after the primary input is removed. The leaky integrators depicted in A are of this type. Note that synaptic STD produces inverted neural integrators. A biphasic differentiator is obtained (top) when a biphasic integrator is combined with antagonistic reflex via the primary pathway or a biphasic integrator-with-step (bottom) when combined with agonist reflex.

Respiratory oscillator. Previous models of carotid chemoreflex (1, 9, 22, 28) have assumed a black-box representation of the respiratory oscillator and its afferent and efferent processes. These system-level functional models adequately describe the ventilatory responses to hypoxic and hypercapnic inputs but not the corresponding changes in respiratory pattern. In contrast, recent models of the respiratory oscillator (14, 43, 65-67, 77) have postulated various neural network configurations for the sundry respiratory neurons that are thought to contribute to respiratory pattern generation. However, although these sophisticated models are capable of simulating the three-phase respiratory pattern, none of them has satisfactorily reproduced the effects of respiratory afferent inputs, such as the carotid chemoreflex response. Moreover, none of the proposed neural network configurations has been fully validated experimentally to the exclusion of other competing models, and the precise interneuronal and afferent-efferent connectivities of the respiratory oscillator remain largely uncertain.

NTS relay. The CSN-mediated inputs from carotid chemoreceptors are relayed centrally by glutamatergic neurons in the NTS (21, 69). These neurons are known to exhibit activity-dependent synaptic accommodation (33, 40, 41, 85, 86), characterized by a gradual decrease of synaptic strength at a rate and magnitude proportionate to the frequency of repetitive afferent stimulation, which persists as STD poststimulation (type I behavior). Some first-order excitatory synapses in the NTS also exhibit NMDA-receptor-dependent long-term depression (LTD) on repetitive low-frequency (60, 85) or high-frequency (54) afferent stimulation (type II behavior). Presently, it is not clear whether carotid chemoafferent relay neurons in NTS are type I or type II neurons. We hypothesize that carotid chemoafferent relay neurons in the NTS signal the respiratory oscillator via multiple parallel pathways that separately control TE and TI with distinct expressions of carotid STD and STP. For simplicity, the possible influence of other descending or ascending inputs convergent to NTS neurons (20, 46, 48, 68, 74, 78, 81) is excluded in this model.

Carotid signaling for E phase. Carotid STD is modeled as a leaky neural integrator that promotes the E phase (58), presumably by excitation of E-neuron (or inhibition of I-neuron) via the ventrolateral pontine pathway (6, 10). The putative carotid chemoreflex shortening of TE is tentatively assumed to be mediated by an inhibitory primary pathway (4, 5, 8), which combines with the parallel carotid STD pathway to form a biphasic differentiator that is gated to the E phase (58). These agonist-antagonist pathways with differing response dynamics may account for the absence of net changes in TE on sustained carotid chemoreceptor activation (72) and the presence of posthypoxic frequency depression (6, 10).

Carotid signaling for I phase. Carotid STP is modeled as leaky neural integrators for the I phase (59), which secondarily promote ventilatory output presumably through parallel excitation of I-PMN and inhibition of I-neuron (or excitation of E-neuron). In addition, direct excitation and inhibition of these neurons via corresponding primary pathways (31, 42) may account for the carotid chemoreflex augmentation of inspiratory motor output and shortening of TI, respectively. These parallel integrator and reflex pathways are assumed to be gated to the I phase independently of those for the E phase, which have distinct response dynamics (see Figs. 4 and 5).

Carotid signaling for I-PMN. The bulbospinal I-PMNs in dogs have been shown (11, 30, 39) to receive tonic and phasic excitatory inputs that are dependent on NMDA and -amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors, respectively, as well as phasic inhibitory inputs that are dependent on both type A -amino-butyric acid (GABA_A) and glycine receptors. These observations are captured in the present I-PMN model with a tonic CSN input featuring NMDA receptor-dependent STP (59) as well as phasic excitation and inhibition from I- and E-neuron, respectively. This model is supported by recent spike train cross-correlation analyses in cats and rats that suggested that I-PMN receives tonic excitation from carotid chemoreceptor inputs (42) as well as phasic excitation (42) and phasic inhibition (14) from I-related and E-related neurons in the VRG, respectively. Presumably, the tonic and phasic excitations determine the gain and the pattern of the inspiratory drive, respectively, whereas the phasic inhibition provides the gating.

Models of Neural Integrator and Differentiator

Monophasic integrator and differentiator. The neural integrator and differentiator characterizations of carotid STP and STD (52, 58, 59) conform to a class of neural networks capable of "brain calculus" operations (38, 57). The mechanisms of such sophisticated neural computations are presently unclear; for neural integrator characterization, the candidate hypotheses include reverberation in a recurrent network (18, 64) and activity-dependent phasic augmentation of excitatory synaptic transmission with poststimulus STP (59, 73, 79). However, the feasibility of a reverberating integrator network has been questioned on theoretical grounds due to its inherent complexity and nonrobustness to instability (82). Indeed, extensive experimental explorations have failed to confirm reverberation as a cause of respiratory afterdischarge (79), whose demonstrated dependence on NMDA receptor (59) suggests a possible involvement of glutamatergic excitatory synapses. These observations lend support for synaptic augmentation-STP as a possible cellular correlate of neural integrator (Fig. 1, B and C).

Biphasic integrator and differentiator. Another way to sustain poststimulus memory is through a tonic secondary input (possibly from central chemoreceptors) in a convergent pathway (Fig. 1C), with synaptic STP/STD being induced pre- or heterosynaptically in the secondary pathway or homosynaptically in the common pathway following the confluence of the primary and secondary pathways (58). In the heterosynaptic and homosynaptic configurations, the STP/STD effect is necessarily accompanied by a reflex response (a stepwise response at the beginning and end of the CSN input) mediated by the primary pathway (see Fig. 1B). In the presynaptic configuration, where the CSN input serves only a modulatory role, a reflex effect is absent unless there is a bypass pathway (Fig. 1C). Such agonistic and antagonistic combinations of reflex (primary) and integrator (secondary) pathways with resultant integrator-plus-reflex (integrator-with-step) and differentiator effects are hypothesized in the working model for carotid STP and STD (Fig. 1A).

Polarity of integrator in relation to signaling pathway. In Fig. 1A, the presumed polarity of an integrator is related to that of the corresponding pathway relative to the respiratory rhythm. For example, the pontine integrator may be either normal or inverted (with corresponding synaptic STP or STD) depending on whether the carotid STD pathway is excitatory or inhibitory to E-neuron (or alternatively, inhibitory or excitatory to I-neuron). Thus the carotid STD prolongation of TE may result from either increased excitation or decreased inhibition of E-neuron (alternatively, either increased inhibition or decreased excitation of I-neuron) by the pontine pathway. Such an interrelationship between the polarity of a neurotransmission pathway and of its neural plasticity is characteristic of feedforward neural networks organization (56).

Stroboscopic Interferometric Filtering Technique

To circumvent this difficulty, we introduce a novel system identification procedure (82) called SIFT by analogy to stroboscopy and interferometry, two established techniques for studying phasic phenomena. Here, phasic CSN stimuli are synchronized to either the respiratory E or I phase in a stroboscopic fashion. The resulting phase-contrast input-output relationship serves both as a "stroboscope" that distinguishes the phasic (reflex or accommodative) and memory effects and as an "interferometer" that identifies any phasic gating effects. The following stroboscopic and interferometric principles for strobe CSN inputs can help to "sift" the resultant respiratory response into corresponding reflex, memory, and gating components in varying carotid chemoafferent signaling pathways.

On-strobe filtering of synaptic transmission in NTS. The synaptic accommodation and STD/LTD components of NTS neurons (60, 85) may be discriminated stroboscopically as follows. Specifically, synaptic accommodation modulates the reflex response to a strobe CSN input in an identical manner from strobe to strobe (Fig. 2A). If the latter is 1:1 phase locked to the respiratory rhythm, then the reflex and accommodation components are intertwined in the resultant response and are not distinguishable. In contrast, the synaptic STD/LTD component may accumulate exponentially with repeated strobes, resulting in a monophasic inverted integrator (and a monophasic differentiator when combined with the reflex-accommodation component) at on-strobe intervals (Fig. 2B). This on-strobe filtering effect obviates the confounding influence of synaptic accommodation and tags the STD/LTD and other memory components in NTS for further analysis.

View larger version (29K): [in this window] [in a new window]   Fig. 2.   Stroboscopic filtering effects due to autogating in primary pathways. Stimulus strobes (inset at left; compare with Fig. 1) in the form of repetitive CSN pulse trains are applied periodically in synchrony with the respiratory phases. Open boxes show neural activity, and shaded boxes show synaptic activity. A: in the absence of synaptic STD or LTD, on-strobe filtering of synaptic accommodation ( inside triangle) in NTS leaves the primary pathway nonadaptive from strobe to strobe. B: off-strobe filtering of synaptic accommodation with STD or LTD (inner ) in NTS distinguishes the integrator effects in the primary pathway from those in the secondary pathways. This off-strobe filtering effect is specific to the primary pathway and is obscured by successive integration in the secondary pathway, which is not subject to autogating. In both examples, responses in primary or secondary pathways combine at output neuron to form a differentiator at on-strobe intervals and integrator at off-strobe intervals, with first-order dynamics in A and second-order dynamics in B.

Off-strobe filtering of NTS signaling. With strobe CSN inputs, the monophasic differentiator response of NTS is disabled at off-strobe intervals due to autogating. This off-strobe filtering effect (Fig. 2B) is specific to primary pathways and may be obscured by successive integration of the monophasic differentiator response in secondary pathways, which are not subject to autogating. Thus the off-strobe filtering effect is observable at the output only in the absence of secondary pathways.

Half-wave interference of respiratory rhythm. Because of autogating, a primary pathway may influence the respiratory rhythm only during the strobe period. The resultant response may be either a phase prolongation or phase shortening, depending on the respiratory phase relationship of the strobe input as well as the polarity (i.e., E or I promoting) of the primary pathway (Fig. 3A). Specifically, the primary input may excite or inhibit an oscillator neuron native to (i.e., with direct projection from) the primary pathway in or out of synchrony, eliciting, respectively, a direct response in the native neuron or crossover (inverse) response in the complementary oscillator neuron through reciprocal inhibition. A well-known example of such native vs. crossover response patterns is the Hering-Breuer inflation reflex, in which a sustained input from pulmonary slowly adapting stretch receptors may either prolong the E phase or shorten the I phase, depending on the respiratory phase in which the input is applied (82). The magnitudes of these facilitatory or inhibitory responses are modulated by synaptic plasticity as reflected by the memory in the primary pathway, subject to the on-strobe and off-strobe filtering effects (Fig. 2). These predicted half-wave interference patterns under strobe CSN inputs serve as physiological markers for the reflex and memory effects of the primary pathway.

View larger version (36K): [in this window] [in a new window]   Fig. 3.   Stroboscopic interference in a half-center neural oscillator. Stylized phase-transition diagrams showing oscillation between 2 reciprocally inhibiting neurons with alternating (indicated by wide curved arrows) active (dark gray) and inhibited (light gray) states. Constructive (+) and destructive interference () characterized by phase prolongation and shortening, respectively, are indicated by increases and decreases of neuronal sizes (denoted by bold circle with arrow) relative to control (dashed circle). A: half-wave interference elicited by excitatory stroboscopic input in the primary pathway (with NTS). The effect can be either + (left) or  (right) when the strobe is on and is null when strobe is off (autogating effect). B: full-wave interference by excitatory tonic input in the secondary pathway (with Pons) is characterized by alternating +/ half-cycles, provided that the secondary pathway is not phase-logic gated (left). This full-wave interference pattern may be rectified to half-wave if the secondary stimulus is logically gated to one phase (right). Such half-wave interference may be + (right) or  (not shown) depending on the phase relationship of gating. In A and B, opposite interference patterns result for inhibitory strobe or inhibitory tonic inputs (not shown).

Full-wave interference of respiratory rhythm. In contrast, a tonic input in a primary or secondary pathway may elicit full-wave interference of the respiratory rhythm, with alternating half-cycles of native and crossover responses (Fig. 3B, left). In SIFT, the use of strobe CSN inputs predisposes the primary pathways to half-wave interference, and thus any full-wave interference patterns may serve as physiological markers for the secondary pathways.

Phase-selective filtering of carotid afferent traffic. In addition to strobe-dependent autogating of the NTS relay, carotid chemoafferent pathways may also be subject to intrinsic logic gating that is phase locked to the respiratory rhythm. For example, the full-wave interference by a tonic secondary input (Fig. 3B, left) may be rectified to half wave if the secondary pathway is logically gated to only one phase (Fig. 3B, right). This phase-selective half-wave interference, however, is not contingent on the strobe CSN inputs, and thus the effects of phase-logic gating and autogating may be readily distinguished by using strobe inputs with differing phase relationships to the respiratory rhythm.

Experimental Design and Model Testing

View larger version (30K): [in this window] [in a new window]   Fig. 4.   Model predictions of changes in expiratory and inspiratory durations (TE and TI, respectively) in response to various CSN inputs. Left: special cases of the working model in Fig. 1A. Right: corresponding predicted responses to step (tonic) CSN input and inspiratory-strobe (SI) or expiratory-strobe (SE) CSN inputs, with or without phase-logic gating. Horizontal bars indicate the duration of tonic (solid bars), SI (hatched bars), or SE (crosshatched bars) CSN stimulation. Adaptations in NTS are not shown. A: if only primary pathways are present, step (tonic) CSN input should shorten both TE and TI independently, provided that each primary pathway is phase-logic gated to the corresponding native neuron. Otherwise, crossover responses (dotted lines) may counter or even reverse the native effects on TE and TI. SI and SE inputs produce only in-phase responses due to autogating, which may be further modulated by phase-logic gating (if present). B: if only secondary pathways are present, step or SI and SE CSN inputs should yield similar response patterns. For the gated case, all inputs produce progressive prolongation and shortening of TE and TI, respectively. For the ungated case, crossover effects increase the magnitudes of the corresponding gated responses. C: if both primary and secondary pathways are present, the response patterns are the sum of the corresponding responses in A and B.

In all cases, if the signaling pathways are not phasically gated, then a response in TE elicited by a tonic input to E-neuron may echo as a crossover (inverse) response in I-neuron, and vice versa, resulting in mixed (native and crossover) disturbances in TE and TI that are correlated with one another. In contrast, if all the pathways are logically gated to the E or I phase, then the TE and TI responses to these tonic inputs become uncorrelated. It is of interest to note that, for both gated and ungated cases, the interference effects mediated by the primary pathways are always distinct for tonic and strobe CSN inputs (Fig. 4A), whereas those via the secondary pathways are always qualitatively similar (Fig. 4B).

In preliminary model testing, we found that the reported effects of step CSN inputs on TE and TI (Fig. 5) and phrenic activity were satisfactorily reproduced by the working model, provided that the primary and secondary pathways were phase-logic gated (compare with Fig. 4C). Accordingly, SIFT analysis would predict that the TE and TI responses to SI and SE CSN inputs should display gated integrator-differentiator characteristics (Fig. 4C) on subtraction of the accommodation and STD/LTD effects of the NTS with on-strobe and off-strobe filtering (Fig. 2).

View larger version (18K): [in this window] [in a new window]   Fig. 5.   Changes in phrenic burst rhythm on stepwise (tonic) CSN pulse-train stimulation in anesthetized and vagotomized rats. Data are mean values () of breath-by-breath TE (n = 10) and TI (n = 10) normalized to control values. Solid lines are curve fits by the working model (compare with Fig. 4C). Note the differentiator effects for TE with distinct time constants (onset of 10.4 ± 1.9 s; offset of 39.2 ± 18.4 s) and integrator-with-step effects for TI (onset of 5.3 ± 1.7 s; offset of 23.7 ± 3.1 s). Insets: enlarged views of the sections enclosed by dashed lines showing that the "reflex" components for both TE and TI responses might have developed progressively in the first 2 breaths at the onset and offset of the CSN input rather than abruptly (solid lines) in the first breath as presumed previously. These fast transient responses (dotted lines) to step CSN inputs were fitted more closely by a refined model shown in Fig. 8, which was confirmed by the present experimental study using strobe inputs. This figure was adapted from data in Refs. 58 and 59.

On the other hand, close examination of the previous step CSN data revealed that the immediate response to step application or removal of CSN input might consist of fast integrator transients that developed progressively in one or two breaths rather than abruptly in the first breath in a stepwise fashion (Fig. 5). This observation suggests the possibility that even the "reflex" response to CSN inputs might be mediated indirectly by secondary pathways with fast integrator dynamics rather than directly by primary pathways. In this event, the predicted responses to SI and SE CSN inputs would share certain characteristics of both models illustrated in Fig. 4, B and C. The following experiments were designed expressly to test these model predictions.

Experimental Methods

A phrenic nerve (Phr) was dissected at the C₅ level via a ventral approach and mounted on custom-made silver-wire bipolar recording electrodes (OD = 0.0045 in.). Phrenic activity was amplified and low-pass filtered (AI 402 × 50, CyberAmp 380, Axon Instruments), and the resulting raw signal was time averaged with a leaky integrator (Paynter filter, time constant of ~15 ms). Both the raw Phr activity and integrated Phr activity (Phr) were monitored on a Tektronix digital oscilloscope and recorded (AT-MIO-16E-1, National Instruments) on a computer. The peak amplitude of Phr and the phrenic burst frequency (f) were calculated for each respiratory cycle. To calculate TI and TE, the start of TI (end of TE) was defined as the point at which Phr started to rise, whereas the end of TI (beginning of TE) was defined as the point where the slope of Phr was most negative.

Both CSNs were surgically isolated and cut distally, with the adjoining tissues of one CSN (~3 mm) being removed as much as possible and away from the stimulating electrode to minimize current spread. Brief 5-s CSN stimulations (20-25 Hz, 0.1-ms pulse duration) at varying current intensities were initially used to determine the threshold current required to induce appreciable increases in phrenic activity. A moderate stimulus intensity (1.5-2 × threshold, 20-90 µA) was then chosen to produce increases in phrenic activity while obviating response saturation, nerve fatigue, or activation of nonmyelinated C fibers (12). At the conclusion of the experiment, the proximal CSN end was crushed, and the CSN was again stimulated to check possible current spread to the glossopharyngeal nerve. None of the animals tested showed any inhibitory or apneic respiratory response that would indicate current spread.

Electrical CSN stimuli at the chosen settings were applied stroboscopically (Fig. 6) during either the I phase (SI stimuli) or E phase (SE stimuli) over a 4-min period, which ensured that near steady-state conditions were attained in each trial for all animals. The CSN input was triggered cycle-by-cycle by using a custom-designed data-acquisition routine (LabVIEW 4, National Instruments) that detected the rising or falling edges of the Phr signal. To ensure robust triggering, SI generally began within the first one-eighth to one-fourth of the rising phase, whereas SE generally began after one-half of the falling phase elapsed or after the inflection point of the declining waveform occurred. As a precaution, the SI train duration (0.15-0.25 s) and the SE train duration (0.4-0.7 s) were chosen to be always shorter than the stimulated TI and TE, respectively.

View larger version (21K): [in this window] [in a new window]   Fig. 6.   Effects of strobe CSN pulse-train stimulation on the phrenic neurogram. Stimulus strobes (hatched and crosshatched bars, positioned relative to the bold red traces) were delivered either in SI (left) or SE (right). A: oscilloscopic displays of transient phrenic neurogram illustrating the time-dependent effects of SI and SE stimuli in 1 rat. Bold traces show the average integrated phrenic nerve (Phr) waveform during 1-min control period (bold, dashed black) and for the last 10 s of a 4-min stimulation period (bold red). Thin traces show the cycle-by-cycle plots of Phr recorded over a 10-s time window at the beginning of stimulation (blue) and poststimulation (green). Arrows indicate corresponding bidirectional movements of expiratory-inspiratory off-switch under SI and SE. B: steady-state average phrenic waveforms (n = 5) are shown for the control period (dashed black) and for the last 10 s of the stimulated period either normalized in time and amplitude to the control waveshape (thin red) or nonnormalized (bold red). The averaged data show that the wave shape remains largely unchanged, except for increased amplitude and reduced TI.

The dynamic responses were analyzed using exponential least-squares regression, and the goodness-of-fits of different exponential curves were compared using the F test at the 5% significance level. Statistical significance of the parameter estimates was established using paired, two-tailed Student's t-test at the 5% level.

	RESULTS

TOP ABSTRACT INTRODUCTION THEORY AND METHODS RESULTS DISCUSSION REFERENCES

Experimental Results

Plasticity of phrenic wave shape. SI and SE CSN stimulations elicited similar time-dependent disturbances in the phrenic neurogram throughout the strobe-stimulation and recovery periods (Fig. 6A). In the steady state (Fig. 6B), SI and SE both resulted in an increase in the rate of rise of Phr activity characterized by a simultaneous increase in Phr and decrease in TI, without any appreciable changes in the waveshape of Phr when normalized.

View larger version (26K): [in this window] [in a new window]   Fig. 7.   Phase-contrast interference of respiratory rhythm by strobe CSN stimuli. Responses are expressed in terms of phrenic inter-burst interval (TE; A), phrenic burst interval (TI; B), and peak amplitude of Phr (C). Left: data for CSN inputs (means ± SE, n = 9) were normalized (Norm) with respect to corresponding mean values during the control baseline periods. Baseline values for CSN were as follows: TE = 0.99 ± 0.07 s and TI = 0.39 ± 0.02 s. These values were similar for both SI and SE stimulation protocols. Vertical dashed lines indicate beginning and end of CSN stimulation. Note SI and SE inputs for CSN inputs resulted in similar integrator and differentiator characteristics in the 3 respiratory parameters. Right: multiexponential curve fits (solid lines) plotted against the mean experimental data during and after stimulation. Curve fit values were calculated using least-squares regression and compared with competing models using the F test (P < 0.05). Model parameters are reported in Table 1.

Plasticity of TE. SI and SE CSN stimulations produced similar biphasic leaky differentiator-like responses in TE, with larger magnitudes for SE than for SI (Fig. 7A). The decrease of TE at the start of SI and SE CSN stimulations had a small but nonzero (₁ = 1.5-2.1 s) integrator time constant (), with TE reaching a nadir within the first few breaths before returning biexponentially (₂ = 6.5-7.8 s, ₃ = 65-80 s) to the baseline toward the end of the strobe-stimulation period. After removal of the stimulus, TE rebounded rapidly to a peak value above the baseline, again with a small but nonzero (₁ = 2.4-2.9 s) integrator time constant. Thereafter, this "aftercharge" in TE gradually returned to baseline with a single slow time constant (₂ = 132-207 s). These SIFT response patterns for TE are similar to those resulting from step CSN inputs (58) except a reflex component was absent and replaced by a fast biphasic integrator component, and a slow monophasic integrator component was also detected.

Plasticity of TI. SI and SE CSN stimulations both produced biphasic inverted leaky integrator-like responses in TI (Fig. 7B) with similar response time constants and steady-state magnitudes (Table 1). Poststimulation, TI gradually returned to the baseline in the form of a biexponential afterdischarge, unlike the monoexponential decrementing response at strobe-stimulation onset. Curve fitting revealed that both the onset and offset responses had a relatively fast time constant (₁ = 4.9-6.9 s), whereas only the offset response had an additional time constant slower by at least one order of magnitude (₂ = 70-142 s). These SIFT response patterns for TI are similar to those resulting from step CSN inputs (59) except that neither a reflex component nor a fast integrator component (compare with the TE response) was evident.

Plasticity of Phr. For both SI and SE CSN inputs, the response in Phr paralleled those for TI but in opposite directions. The increase in Phr in the steady state was significantly greater with SE than with SI CSN inputs (16.3 ± 5.9 vs. 8.8 ± 3.7%, P < 0.05).

Refined Functional Model of Carotid Chemoafferent Signaling

View larger version (19K): [in this window] [in a new window]   Fig. 8.   Functional model of carotid chemoafferent signaling compatible with both strobe CSN and step CSN data. Conventions are as in Fig. 1A. Animations of the model for expiratory and inspiratory strobe inputs are shown in the attached videoclips (http://jap.physiology.org/cgi/content/full/94/3/1213/DC1). All pathways are phase-logic gated, as indicated by the switches. The carotid STD component shown to promote E phase likely resides in the pons (6), whereas the other pathway locales remain uncertain. Pathways terminating on the respiratory oscillator could have opposite (excitatory vs. inhibitory) polarities if they project to the opposite side of the oscillator (E-neuron instead of I-neuron and vice versa) or if they contain inverted integrators instead. Parallel integrators are functionally equivalent to serial integrators in cascade with a bypass pathway for the slower integrator. Dotted line indicates possible carotid "chemoreflex" pathway to I-neuron time gated to the late-I phase. See text.

Carotid chemoafferent pathways were phase-logic gated. Although the responses in TE, TI, and Phr to SI and SE CSN inputs all exhibited integrator and differentiator patterns (Fig. 7), the integrator time constants for TE were significantly different from those for TI and Phr (P < 0.05, Table 1), with no evidence of crossover either way (compare with Fig. 4). Therefore, it appears that all the carotid chemoafferent pathways were logically gated to either the E phase or I phase, as illustrated in Fig. 8. This observation supports the phase-logic gating hypothesis of the working model (Figs. 1A and 4) based on previous findings with step CSN inputs (Fig. 5).

Carotid "chemoreflex" shortened TE through a fast integrator gated to E phase. The working model assumed that CSN input shortened TE through an E-gated reflex pathway inhibitory to the E phase (Fig. 1A). The present results, however, showed that the response in TE at the onset/offset of the SE and SI CSN inputs (Fig. 7A) developed progressively over several breaths like an integrator, rather than abruptly within the first breath. This behavior was most striking under the SI CSN input, whereby any reflex modulation of TE would be suppressed by off-strobe filtering, and yet the integrator response in TE remained. These surprising revelations suggested that the carotid "chemoreflex" modulation of TE might be mediated indirectly by an E-gated secondary integrator rather than directly via a primary pathway (Fig. 8). The time constant of this occult integrator was extremely short (₁ = 1.5-3.0 s, Table 1) under strobe CSN inputs and perhaps even shorter under step CSN inputs (Fig. 5), making it difficult to discern within the resolution of a single respiratory cycle (~1.5 s in vagotomized rats).

Carotid STD lengthened TE through a slow-recovery integrator gated to E phase. In agreement with the working model (Fig. 1A), the SIFT data (Fig. 7A) demonstrated an E-promoting, E-gated integrator that had relatively fast induction and slow recovery (₂ = 6.5-7.8 s and 130-210 s, respectively), evidencing significant dynamic hysteresis (52). The response characteristic and the depression effect on the respiratory frequency suggest that this integrator may be associated with the pontine-mediated posthypoxic frequency depression response (6, 10). This pontine pathway for carotid STD (Fig. 8) is assumed to be connected in parallel with the E-inhibiting fast integrator described above because these integrators are seen to operate independently in widely differing timescales (especially in the recovery phase) and have opposite effects on TE. Alternatively, these integrators could be connected in series, provided that there is a bypass for the slow integrator, which circumvents its low-pass filtering of the faster dynamics.

Lack of carotid chemoreflex in I phase under strobe CSN inputs. Remarkably, a reflex shortening of TI typically seen during step CSN pulse-train stimulation in rats (Fig. 5) was absent with SI CSN input (Fig. 7B). Because the SI stimulus train during the I phase ended before I-E phase switching (see Experimental Methods and Fig. 6), it may be inferred that a CSN stimulus may elicit reflex shortening of TI in rat provided it is applied late in the I phase. However, careful reexamination of the data in Fig. 5 revealed that this "chemoreflex" shortening and relengthening of TI probably developed progressively in the first two breaths at the onset and offset of CSN stimulation, respectively, rather than abruptly within the first breath. Therefore, it seems that this apparent reflex response in TI (if present) might also be mediated by a fast integrator in the I-related pathway as with the E-related pathway. This timing-dependent "chemoreflex" effect for TI is depicted in the proposed functional model (Fig. 8) as a possible time-gated pathway to the I-neuron. A corresponding reflex pathway for Phr motor output is not included in Fig. 8, as such a reported reflex effect on Phr in cats (16, 17, 79) was not indicated in the SIFT data (Fig. 7) or in previous studies with step CSN input at relatively low-stimulation currents in rats (59).

Carotid STP shortened TI through dual integrators gated to I phase. In agreement with the working model (Fig. 1A), both SI and SE CSN inputs revealed (Fig. 7B) an I-inhibiting, I-gated integrator with relatively fast induction and recovery time constants (₁ = 4-7 s for both). In addition, the SIFT data (Table 1) revealed a second integrator with a relatively slow recovery time constant (₂ = 70-140 s), although its induction dynamics was obscured by a counteracting monophasic integrator from the NTS (see below). These integrators are distinguished from those for TE by their distinct onset/offset time constants. Again, these dual integrators are assumed to be connected in parallel (Fig. 8) in light of their independent actions over widely different timescales, although a series connection is also possible provided there is a bypass pathway for the slow integrator that circumvents its low-pass filtering of the faster dynamics.

Carotid STP augmented Phr activity through dual integrators gated to I phase. Similarly, the SIFT data suggested that the carotid STP of Phr motor output may be ascribed to I-gated dual integrators (Fig. 7C and Table 1). However, these dual integrators are distinguished from those for TI in that they were Phr facilitating instead of I inhibiting. Furthermore, their response magnitudes were distinct for SI and SE CSN inputs, unlike the similar response magnitudes with TI. Therefore, these integrators appear to have separate excitatory effects on I-PMN independent of the corresponding inhibitory effects on I-neuron (Fig. 8). Furthermore, because the SI and SE CSN inputs did not alter the Phr waveshape when normalized for TI and amplitude of Phr (Fig. 6), these carotid STP pathways to I-neuron and I-PMN appear to modulate mainly the timing and gain rather than the waveshape of inspiratory motor outflow.

Synaptic STP/STD in NTS modulated carotid STP/STD. From the SIFT theory, synaptic STD/LTD in NTS may appear as a monophasic inverted integrator (Fig. 2B). Multiexponential analysis of the TE response (Fig. 7 and Table 1) indeed identified such an inverted integrator, which emerged slowly during induction (₃ = 65-79 s) and was obscured during recovery. This monophasic inverted integrator for TE was evident not only with SE but also SI CSN inputs, suggesting that the off-strobe filtering effect of the NTS integrator might be obscured by subsequent integration in the secondary pathways (Fig. 2B).

View larger version (17K): [in this window] [in a new window]   Fig. 9.   Simulations of carotid chemoafferent signaling from NTS to the respiratory oscillator via expiratory pathways (A) and inspiratory pathways (B). Results for the pathways to I-PMN were similar to B. Simulations were based on the model shown in Fig. 8 with parameter values from Table 1, for the case of SE CSN activation. Simulations for SI CSN activation and continuous CSN activation were qualitatively similar to those in the SE case and are not shown. The phasic and STD/LTD responses in NTS are integrated continuously by dual integrators in secondary pathways, with the outputs (dotted lines in boxes) gated to either the respiratory E phase or I phase. The predicted response in the output (solid line) closely matched the mean experimental data () for TE and TI . Note the synaptic STD/LTD effect in NTS is preserved by the dual integrators in A and annihilated in B, depending on the integrator time constants and polarity of corresponding signaling pathways. See text.

Model Simulations

Strobe CSN inputs. Figure 9A shows the model-simulated responses of the expiratory pathways to strobe CSN inputs. It can be seen that differential processing of the reflex-accommodation component of the NTS is provided by the excitatory-inhibitory actions of the dual integrators with differing response magnitudes (0.28 vs. 0.34) and induction time constants (6.5 s vs. 2.1 s). However, the pontine integrator has a recovery time constant (207 s) that is more than 10 times slower than its induction, and such pronounced dynamic hysteresis makes it much less responsive to the subsequent STD/LTD component compared with the fast integrator (recovery time constant of ~2.9 s). As a result, the monophasic inverse integrator effect of the NTS is transmitted to the respiratory oscillator via the fast integrator virtually free of antagonism, adding to the differentiator effect of the expiratory pathways.

Step CSN inputs. The simulation results for step CSN inputs were qualitatively similar to those for strobe inputs (Fig. 9) except for differing integrator time constants (see Fig. 5 legend). The refined model simulated more closely the initial development and decay of the fast integrator responses (Fig. 5, insets) than the working model in Fig. 1A, which assumed that these responses were primarily reflex mediated. Thus the refined model extends the working model by accounting for both the fast and slow integrator effects in the carotid chemoafferent signaling pathways.

	DISCUSSION

TOP ABSTRACT INTRODUCTION THEORY AND METHODS RESULTS DISCUSSION REFERENCES

The SIFT data corroborate the hypothesis that carotid chemoafferent signaling via the NTS is dynamically filtered downstream by a parallel bank of integrators and differentiators in varying time and frequency scales, which are logically gated to the E and I half-cycles. The aptness of the proposed functional model is contingent on the soundness of the underlying theory and methodology, and its validation may shed light on the complex organization of carotid chemoafferent pathways for respiratory pattern generation.

Critique of Methodology

Modeling perspectives. Previous studies of respiratory control have focused on the identification of discrete cellular mechanisms in reduced or anatomically isolated preparations, with a view to accumulating a complete empirical basis piecemeal for ultimate bottom-up (cellular-to-system) integrative modeling. In contrast, the present study drew on largely top-down (system-to-cellular) integrative modeling based on experimental observations at the system level in order to discern the structural and functional organization of carotid chemoafferent pathways en bloc consistent with current understanding of related discrete mechanisms at the cellular level, with minimal assumptions about the structure of the respiratory oscillator. Both modeling approaches have advantages and disadvantages (84). At this stage, bottom-up modeling of carotid chemoafferent signaling is precluded by the current dearth of available information at both the cellular and system levels, which undersold the true complexity of the problem.

SIFT as a probe for complex neural systems. Neural integrator, differentiator, and logic gating are novel physiological paradigms, and their elucidation calls for new experimental and theoretical methodologies. The intricate characteristics of these elemental processes and their manifold connectivity in a spatiotemporally extended network rendered carotid chemoafferent signaling with such complexity that defied experimental exploration with conventionally physiological, anatomic, and imaging approaches. Indeed, the seemingly mundane response characteristics under strobe CSN inputs (Fig. 7) belied the underlying peculiarities, which became transparent only under the magnifying glass of SIFT. In essence, SIFT helps to sift the data for insights beyond their face values (82). Without delving into its profound theoretical underpinning, the power of SIFT is due to two well-known techniques in phase-contrast imaging: stroboscopy (autogating effect) and interferometry (rhythmic interference effect). Systematic application of these basic principles to the SIFT data made it possible to noninvasively dissect the various reflex, memory, and gating effects of the carotid chemoafferent pathways and logically piece the jigsaw together, revealing their inner workings component-by-component and their overall architecture all at once.

Validity of CSN stimulation. Pursuant to the SIFT theory, we used electrical CSN stimulation instead of natural stimuli to activate carotid chemoafferents in specific phases of the respiratory rhythm in an all-or-none fashion. The possible influence of baroreceptive CSN fibers may be discounted for several reasons. First, pressor responses are negligible with the relatively low-stimulation currents being used (49, 59). Second, it has been shown that carotid baroreceptor stimuli (13, 70) or aortic nerve stimulation (2, 23) have only relatively minor and disparate (inhibitory) effects on respiration compared with carotid chemoreceptor stimuli or CSN stimulation. Third, both strobe and step CSN inputs (58, 59, 79) elicited similar respiratory responses as with natural stimuli (6, 23), consistent with carotid chemoafferent activation. It should be noted, however, that CSN stimulation reveals only the acute effects in central pathways and not any plasticity that might develop at the carotid chemoreceptors level (62) or over a longer term of hours and days (34, 61).

Organizing and Operating Principles of Carotid Chemoafferent Pathways

Dynamic signal conditioning. An interesting organizing principle demonstrated in the proposed model is that NTS signaling to the respiratory oscillator may be mediated indirectly by a bank of integrators via secondary pathways, with little or no direct reflex neurotransmission via primary pathways. Surprisingly, the carotid chemoreflex shortening of TE (and probably also TI) proved to be mediated by an occult fast integrator with tonic secondary excitation. Because of their high responsiveness, these fast integrators could be saturated rapidly within a normal respiratory cycle by a sustained strong stimulus, thus contributing to their elusiveness. In the present study, this occult integrator for TE was discerned unambiguously with the use of strobe pulse-train stimulations at moderate intensities.

Logic gating. Another important organizing principle embodied by the proposed model is that the carotid chemoafferent pathways may be phase-locked to the respiratory rhythm. Functionally, such phase-locked behavior works like a logic gate or temporal filter, passing or suppressing neural traffic at selected time intervals. This fundamental organizing principle implies that carotid chemoafferent traffic may be regulated centrally in a highly orderly stop-and-go manner, much like the traffic lights at a crossroads. The advantages of such phase-logic gating or temporal filtering are obvious. Ungated tonic inputs to the E and I neurons are bound to elicit counteracting crossover responses in TE and TI, thus compromising their net effects on the respiratory frequency. Logic gating allows the E and I phase to be regulated independently by carotid chemoafferent feedback in a manner that is tailored to each phase.

Parallel segregated processing. The carotid chemoafferent pathways appear to be organized in a parallel and segregated fashion in two respects. First, these pathways are segregated by their gating to distinct phases of the respiratory rhythm. Second, within each respiratory phase, there are dual integrators acting in parallel. Such parallel and segregated neural architecture allows multiple neural pathways to operate independently of one another with minimum cross talk (56). In contrast, in a series configuration, the response bandwidth would be necessarily limited by the integrator with the slowest dynamics unless it has a bypass pathway, which is inherent in homosynaptic or heterosynaptic forms of neural integrator (Fig. 1, B and C). For example, the relatively slow (homosynaptic) STD/LTD integrator in NTS is not limiting to the succeeding secondary integrators because it is secondary to direct reflex neurotransmission in NTS. Thus, functionally, it is as though the STD/LTD integrator in NTS acted in parallel with the secondary integrators even though structurally they are connected in series. In any event, some degree of segregation and parallelism seems necessary so that different integral-differential processing modes may operate independently of one another in multiple phase-specific spatial domains over a wide range of timescales.

Multivariate feedback compensation. The stability of closed-loop respiratory control is influenced importantly by changes in the gain and phase delay of the carotid chemoreflex response (28). It has been generally assumed that the dynamics of carotid chemoreceptor feedback is characterized by a linear compartment with a fast time constant ascribed to the carotid chemoreceptors (1). The present study showed that carotid chemoreceptor feedback is modulated centrally by nonlinear neurodynamic and neurological processes specific to various respiratory timing and amplitude components. Specifically, the combination of secondary integrators with varying polarities, response speeds, and phase-logic gates allows parallel integral-differential processing of the carotid chemoafferent signal, with resultant differentiator (high-pass filtering) effects for the E phase and integrator (low-pass filtering) effects for the I phase in varying frequency bands. From engineering systems theory (45), a high-pass filter affords graded phase-advance (phase-lead) compensation, whereas a low-pass filter introduces graded phase-lag compensation of the feedback signal, which could be suitably tuned to maximize closed-loop stability. In the respiratory system, such dynamic compensations of the gain and phase of multivariate afferent feedbacks are further modified by the dynamic hysteresis, stimulus threshold, and saturation effects (52) as well as self-organized redundancy (80) that are characteristic of carotid chemoafferent pathways. Such complex time-frequency filtering of carotid chemoafferent traffic with inherent activity- and phase-dependent nonlinearities may exert profound influence on respiratory stability during hypoxia and sleep (52, 83) and could contribute to the dynamic optimization of respiratory pattern and maintenance of homeostasis in health and in disease states (50, 51, 53, 55).