Motoneuron firing patterns underlying fast oscillations in phrenic nerve discharge in the rat

Abstract

Fast oscillations are ubiquitous throughout the mammalian central nervous system and are especially prominent in respiratory motor outputs, including the phrenic nerves (PhNs). Some investigators have argued for an epiphenomenological basis for PhN high-frequency oscillations because phrenic motoneurons (PhMNs) firing at these same frequencies have never been recorded, although their existence has never been tested systematically. Experiments were performed on 18 paralyzed, unanesthetized, decerebrate adult rats in which whole PhN and individual PhMN activity were recorded. A novel method for evaluating unit-nerve time-frequency coherence was applied to PhMN and PhN recordings. PhMNs were classified according to their maximal firing rate as high, medium, and low frequency, corresponding to the analogous bands in PhN spectra. For the first time, we report the existence of PhMNs firing at rates corresponding to high-frequency oscillations during eupneic motor output. The majority of PhMNs fired only during inspiration, but a small subpopulation possessed tonic activity throughout all phases of respiration. Significant time-varying PhMN-PhN coherence was observed for all PhMN classes. High-frequency, early-recruited units had significantly more consistent onset times than low-frequency, early/middle-recruited and medium-frequency, middle/late-recruited PhMNs. High- and medium-frequency PhMNs had significantly more consistent offset times than low-frequency units. This suggests that startup and termination of PhMNs with higher firing rates are more precisely controlled, which may contribute to the greater PhMN-PhN coherence at the beginning and end of inspiration. Our findings provide evidence that near-synchronous discharge of PhMNs firing at high rates may underlie fast oscillations in PhN discharge.

the generation of synchronous activity within motoneuron pools has been a subject of interest for many years. During the inspiratory phase of normal eupneic breathing (Richter et al. 1986; Richter and Ballantyne 1983), the respiratory rhythm generator/controller produces oscillations in the output of phrenic (PhN), hypoglossal, and other respiratory motor nerves that are well above the frequency of the primary breathing rhythm (Cohen et al. 1997). In the PhN, these oscillations normally exist in two dominant bands, known as high (HFO)- and medium-frequency oscillations (MFO) (Cohen et al. 1987, 1997; Davies et al. 1985; Dittler and Garten 1912; Huang et al. 1996; Marchenko et al. 2002; Marchenko and Rogers 2006a; Richardson and Mitchell 1982; Wyss 1939). HFO and MFO are prominent in the first and second half of the inspiratory burst, respectively (Cohen et al. 1987; Marchenko et al. 2002; Marchenko and Rogers 2006a; Richardson and Mitchell 1982). Based on analyses of firing synchrony among, and spinal projections of, bulbospinal neurons (Davies et al. 1985; Duffin and Li 2006; Duffin and van Alphen 1995; Tian and Duffin 1997) and on coherence between homologous and heterologous respiratory nerve activity (Cohen et al. 1987; Marchenko and Rogers 2006b), HFO and MFO are proposed to originate mainly supraspinally and intraspinally, respectively. Although a one-to-one relationship between phrenic motoneuron (PhMN) firing and PhN oscillations has only been demonstrated in the low-frequency band in the neonatal rat in vitro (Parkis et al. 2003), these oscillations may result from near-synchronous discharge of PhMNs firing at least once per cycle, driven by inspiratory bulbospinal units that fire at these rates, including those corresponding to HFO (Davies et al. 1985; Duffin and van Alphen 1995; Huang et al. 1996; Tian and Duffin 1997).

The hypothesis that fast oscillations, particularly HFO, are produced by synchronized PhMN discharge at the same frequencies has two major shortcomings. First, and most importantly, it has never been demonstrated that PhMNs fire at rates corresponding to HFO in any species during eupneic breathing (Christakos et al. 1991; Hayashi and Fukuda 1995; Kong and Berger 1986; Nail et al. 1972; St John 1979), when HFO are present. Because of the lack of evidence that PhMNs fire at HFO-related rates, other explanations have been offered to explain the fast oscillations in these, and other, respiratory motor outputs (e.g., van Brederode and Berger 2008). Second, simultaneous recordings of multiple PhMNs have neither been performed nor related to PhN oscillations (nor to each other), and therefore there is no evidence that their synchronous discharge underlies this phenomenon.

Fast respiratory rhythms (especially HFO) are very sensitive to anesthesia (Richardson and Mitchell 1982). For this reason, we use unanesthetized, decerebrate rat preparations when studying fast oscillation phenomena (Marchenko et al. 2002; Marchenko and Rogers 2009, 2006a, 2006b, 2007). In the present study, we test the hypothesis that fast respiratory rhythms (HFO in particular) can be produced directly via the firing of individual PhMNs at those rates. This was accomplished by utilizing a novel quantitative approach in which the discrete spike train is transformed into a continuous waveform, and the time-frequency coherence between this waveform and PhN activity was estimated. In doing so, we evaluate the firing properties of PhMNs and their relationship to fast oscillations in the PhN in the unanesthetized, decerebrate rat.

METHODS

Animal Preparation

General surgical preparation.

All procedures were approved by the Drexel University Institutional Animal Care and Use Committee, which oversees Drexel University's AAALAC International-accredited animal program. Eighteen spontaneously breathing, adult male Sprague-Dawley rats (340–380 g) were anesthetized with isoflurane vaporized in O₂ (Matrix; 4–5% induction, 1.85–2.15% maintenance) via a snout mask. Anesthetic depth was maintained at a level such that withdrawal reflexes and changes in heart rate and blood pressure in response to pinches of the distal hind limbs were absent. After tracheotomy with an atraumatic glass tube, animals were artificially ventilated with the same gas mixture (∼60 cycles/min, 2.3–3.0 ml tidal volume; Columbus Apparatus). Electrocardiogram (EKG) was measured via three small subcutaneous electrodes using conventional amplification and filtering (Neurolog; Digitimer, Welwyn Garden City, UK) and monitored using an audio amplifier (model AM10; Grass Instruments) and oscilloscope (Tektronix). One femoral artery and vein were cannulated for measurement of arterial pressure and infusion of drugs/saline, respectively. During all surgical procedures, rectal temperature was maintained at 37.0 ± 0.1°C via a servocontrolled heating blanket coupled to a rectal thermometer (Harvard Apparatus). By using a ventral approach, the PhNs were dissected free from the surrounding tissue, transected, and desheathed. Both internal carotid arteries were ligated just below the pterygopalatine artery to prevent bleeding after decerebration.

Decerebration and corpectomy.

After the initial surgical preparation, rats were placed prone in a stereotaxic device. Arterial and tracheal cannulae were connected to pressure transducers (CDXII; Argon Medical) for monitoring arterial blood pressure and lung inflation pressure, respectively, using conventional amplifiers (Gould Statham). Biparietal craniotomies were performed using a variable-speed surgical drill (Foredom Electric), the superior sagittal and straight sinuses were ligated using suture, and the neuraxis was carefully transected using a microspatula at the rostral border of the superior colliculus. Brain tissue rostral to the transection was removed by suction, and residual bleeding was arrested by applying small pieces of gelfoam (USP; Pharmacia) soaked with cold thrombin solution (50 U/ml USP, dissolved in physiological saline) to the transected surface.

After decerebration, animals were repositioned in a supine orientation. Infrathyroid portions of the trachea and esophagus were removed. The C₃–C₅ vertebrae were exposed by using a ventral approach by removal of the rectus capiti, superior oblique, and ventral portions of the longus colli muscles. The ventral surface of the C₃–C₅ spinal cord was exposed by gradual grinding of the vertebral bodies using a variable-speed drill and was sealed with bone wax. The dura was then opened using iridectomy scissors.

Before neuronal recording, bilateral pneumothoracotomies were performed to eliminate lung inflation-related movement artifacts and chest wall mechanoreceptor feedback. Fifteen to twenty minutes after corpectomy, anesthesia was slowly withdrawn and the animals were paralyzed by an intravenous bolus injection (2 mg/kg), followed by continuous infusion (3–4 mg·kg⁻¹·h⁻¹), of vecuronium bromide (Abbott Laboratories) dissolved in Ringer-Locke solution. A positive end-expiratory pressure of 1.0 cmH₂O was maintained to prevent atelectasis. End-tidal CO₂ was maintained between 4.0 and 4.5% (Capstar CWE) by adjusting minute volume. PhN recordings were not initiated until ≥1 h after decerebration. If necessary, animals were continuously infused with 0.9% saline (1.0–1.25% body wt or 10.0–12.5 ml·kg⁻¹·h⁻¹) to maintain a stable mean arterial pressure of 85–90 mmHg. The central ends of the PhN were placed on bipolar silver electrodes and immersed in a mineral oil pool formed by skin flaps.

Recording.

Monophasic recordings (10–5,000 Hz; Neurolog, Digitimer) of efferent activity were obtained after the peripheral ends of each PhN were crushed between the electrodes. The activity of spinal respiratory-related neurons was recorded (200–3,000 Hz; Neurolog) using glass microelectrodes with a tip outer diameter of 0.5–0.75 μm (17–20 MΩ) pulled from borosilicate glass capillaries (catalog no. 1B120-F4, WPI) and filled with 0.5 M NaCl and either 2.5% Neurobiotin or 2% pontamine sky blue. The microelectrode was held in a stepper motor assembly (T-NA08A50; Zaber Technologies), attached to the rail of the stereotaxic frame via a micromanipulator, and advanced in steps of 1.5–2.0 μm. Cells were labeled juxtacellularly by iontophoresis of Neurobiotin (+5–10 nA, 20 Hz, 20 ms; Axoclamp 2A) for 20–25 min or pontamine sky blue (−5 μA, 10 min). The electrical activity of the two phrenic nerves, extracellular discharges of respiratory neurons in phrenic nucleus, expiratory CO₂ level, arterial blood pressure, and lung inflation pressure were recorded onto the hard disk of a personal computer at 10,000 samples/s each, using a 16-bit analog-to-digital converter with visualization software (AD Instruments).

At the end of each experiment, animals were transcardially perfused with 400 ml of normal saline (20°C, pH 7.4) with heparin (1,000 U/l) followed by 500 ml of 4% paraformaldehyde in 0.1 M phosphate-buffered saline (PBS). The spinal cords were removed and postfixed in the same fixative for 48 h at 4°C. Before transverse sectioning, spinal cords were cryoprotected by incubation in 30% sucrose in PBS for 24 h. Sections (60 μm) were cut the following day to determine the position of juxtacellularly labeled units. In the case of Neurobiotin-labeled neurons, sections were then processed with Vectastain ABC kits (Vector Labs). Labeled cells were mapped onto standardized transverse sections from a stereotaxic atlas (Paxinos and Watson 2007).

Data Analysis

General PhN output variables including respiratory rate (RR), inspiratory (T_i) and expiratory duration (T_e), inspiratory burst amplitude, and single-unit activity were determined using ≥50 respiratory cycles. Spike2 (version 5; Cambridge Electronic Design), MATLAB (version R2011a; The MathWorks), IBM SPSS Statistics 19, and custom-written scripts for measurement of parameters from PhN and single-unit activity were used for data analysis. Data distribution was evaluated by applying the Lilliefors test for normality using MATLAB. Depending on how well the data conformed to a normal distribution, either parametric (t-test) or nonparametric (Mann-Whitney U) tests were applied to compare two groups of data (e.g., see Fig. 7). For comparison among multiple results (e.g., see Fig. 7), we used parametric (1-way and repeated-measures ANOVA) and nonparametric (Kruskal-Wallis and Friedman) tests. All values are reported as means ± SD or means ± SE, as needed. SE measures are given when groups of standard deviations are compared (e.g., when comparing burst onset and offset variabilities for groups of PhMNs; see Fig. 3). PhMN onset and offset times are given in percentages relative to normalized inspiration, where the individual inspiratory duration is set to 100% (i.e., in each cycle). Positive percentage values refer to average onset and offset after the corresponding normalized PhN burst initiation and cessation, respectively (see Fig. 3 and Table 1). Differences were considered significant at the 95% confidence level (P < 0.05).

Fig. 7.

Comparison of unit-PhN coherence values among PhMN classes. Ipsilateral (Ipsi; open bars) and contralateral (Contra; shaded bars) PhMN-PhN coherence (means ± SD) at LFO-MFO bands for LF, MF, and HF PhMNs and at HFO band for HF and HF+BG PhMNs. *P ≤ 0.05.

Fig. 3.

Comparison of maximal firing frequency (F_max), onset, and offset parameters among classes of PhMNs. A: F_max (means ± SD, Hz) in HF (open bars), MF (shaded bars), and LF (solid bars) PhMNs. B: onset times (means ± SD, %) relative to the beginning of PhN activity (inspiration was normalized to 100% in duration) for HF, MF, and LF PhMNs. C: differences in SD of onset times (means ± SE, %) for the same classes of PhMNs. D: offset times relative to the end of PhN activity. E: differences in SD of offset times (means ± SE, %). *P ≤ 0.05.

Table 1.

Frequency and onset/offset characteristics of purely inspiratory PhMNs

PhMN Type	Fmax, Hz [min, max]	Onset Time, % [min, max]	Offset Time, % [min, max]	SD for Onset Time, % [min, max]	SD for Offset Time, % [min, max]
HF	180.54 ± 26.13 [140, 224.7]	6.83 ± 10.93 [−6.63, 26.39]	1.88 ± 9.37 [−2.78, 28.32]	2.63 ± 1.84 [0.59, 5.82]	3.17 ± 2.90 [1.46, 11.20]
MF	89.55 ± 12.91 [73.38, 109.1]	9.82 ± 11.28 [−6.76, 26.99]	0.54 ± 4.60 [−5.44, 9.32]	5.89 ± 4.24 [2.51, 12.09]	3.89 ± 1.45 [1.88, 5.44]
LF	37.70 ± 5.04 [32.10, 45.73]	35.27 ± 17.64 [6.24, 59.42]	−3.26 ± 6.58 [−10.88, 9.44]	8.56 ± 6.74 [2.88, 19.10]	6.50 ± 2.65 [−10.88, 9.44]

Values are means ± SE, with minimum and maximum values in brackets, of parameters for low-frequency (LF), medium-frequency (MF), and high-frequency (HF) phrenic motoneurons (PhMNs). F_max, maximal firing frequency.

Event markers for single-neuron action potentials and for onset and offset of integrated (τ = 50 ms) PhN activity were derived from the raw recordings. Only inspiratory epochs were analyzed. Spike patterns of single units were analyzed by creating PhN onset-triggered histograms (i.e., cycle-triggered histograms; CTHs), as well as normalized composite CTHs for groups of PhMNs using 15 bins to span inspiration. K-means cluster analysis was used to classify PhMNs on the basis of their maximal firing frequencies (F_max). PhMNs were positively identified by the presence of a unitary waveform in the spike-triggered average (STA) of the ipsilateral PhN after the spike recorded in the ventral horn and by the absence of this waveform in a one-respiratory cycle-shifted STA of the same (Fig. 1, A–C, dark traces).

Fig. 1.

Identification and coherence analysis of phrenic motoneuron (PhMN) activity. A: low-frequency (LF) PhMN spike-triggered average (STA) of phrenic nerve (PhN) activity. B: same as A, but for a middle-frequency (MF) PhMN. C: same as A, but for a high-frequency (HF) PhMN. Vertical cursors show the lag of unitary waveform in ms. D: waveforms used to analyze unit-nerve coherence. Traces, from top, show ipsilateral PhN activity, semicosine + noise waveform, and extracellular PhMN recording. The semicosine wave was generated by convolving a positive half-cosine wave with each PhMN spike and adding band-limited white noise (see methods). Gray vertical bars highlight correspondence between PhMN spikes, noised semicosine wave, and PhN waves.

A novel method was employed to represent PhMN action potential events as a continuous signal to facilitate spike train-PhN coherence estimation. Positive semicosine waves (of amplitude 1.0, period equal to interspike interval) derived from the spike-event (digital) channels were used to generate continuous waveforms that allowed for PhMN-PhN coherence estimation (Fig. 1). White noise (5% of maximal amplitude) was added to avoid divide-by-zero errors in coherence estimates. This method allows achievement of good matching between spike and PhN channels (see Fig. 1D) and a more accurate representation of instantaneous frequencies than that provided by other methods (Christakos et al. 1994; Nawrot et al. 1999). To investigate the contribution of different classes of PhMNs to fast oscillations in PhN activity, time-frequency coherences were estimated for each unit and averaged over the set of neurons of a particular type (e.g., high-frequency PhMNs; see Fig. 6, A–D) using the same time and frequency resolutions described below. Finally, population time-frequency coherence estimates were segmented and time-averaged into first and second halves of the inspiratory period to assess the early vs. late contributions of different classes of PhMNs to the corresponding frequency bands in PhN oscillations. Ipsilateral and contralateral relationships were analyzed independently. Parameters were normalized (to either 100% or 1.0, e.g., of inspiratory duration) and averaged across the experimental data sets of individual animals.

Fig. 6.

Dynamic PhMN-PhN coherence. A1–D1: smoothed pseudo-Wigner-Ville distribution (SPWVD) time-frequency representation (TFR) population-averaged PhMN-ipsilateral PhN coherence. A2–D2: population-averaged PhMN-contralateral PhN coherence. A3–D3: time-averaged (reconstructed from SPWVD TFR) PhMN-PhN coherence for ipsilateral (red) and contralateral (blue) sides. A4–D4: representative individual PhMN-PhN (ipsilateral) coherence. A1–A4: LF PhMNs. B1–B4: MF PhMNs. C1–C4: HF PhMNs. D1–D4: HF+BG PhMNs. E1 and E2: normalized (%) TFR autospectra of population-averaged right (E1) and left (E2) PhNs. E3: reconstructed autospectra for right (cyan) and left (blue) PhNs. E4: normalized (%) autospectrum of an individual PhN. F1: TFR coherence between population-averaged left and right PhNs. F2: control, showing coherence between right PhN and band-limited (0–5,000 Hz) white noise. F3: reconstructed PhNR-PhNL (right-left; red) and PhNR-noise (black) coherence. F4: TFR coherence between individual pair of left and right PhNs (same animal). Gray horizontal lines in A3–D3 and F3 indicate the top 95% confidence threshold for coherence.

Data were down-sampled from 10,000 to 2,000 Hz and bandpass filtered (30–600 Hz, 3-db cutoff). For time-frequency representation (TFR) spectra and coherence estimations, the smoothed pseudo-Wigner-Ville distribution (SPWVD) was applied (O'Neal et al. 2005). The SPWVD at a given frequency, f (τ =2·f⁻¹), and time, t, is a sliding version of the Wigner distribution:

$${\tilde{W}}_x\left(t,\hspace{0.17em}f\right)={\displaystyle \int q\left(u-t\right)}{\displaystyle \int h\left(\tau \right)x\left(u-\frac{\tau }{2}\right)x*\left(u-\frac{\tau }{2}\right)\times e^{-i2\pi f\tau }\text{d}\tau \text{d}u}$$

where the time windowing, h, acts as a smoothing function in the frequency domain and the low-pass function, q, acts in the time domain (Goncalves and Baraniuk 1998), and x(t) is the Hilbert transform of the analyzed signal. The frequency resolution was set as 1.95 Hz/bin. To normalize TFR results across respiratory cycles of different lengths, inspiratory epochs were divided into 15 bins. Coherence was performed between the ipsilateral PhN and a randomly generated band-limited (0–5,000 Hz) white noise signal as control.

RESULTS

Classification and Firing Patterns

Thirty-three inspiratory PhMNs were recorded from the ventral horn of C₃–C₅ in 18 decerebrate adult male rats (Fig. 2 and Table 1). Some (n = 24) PhMNs fired only during inspiration, whereas others displayed background activity (BG) during other respiratory phases (n = 9). The K-means cluster analysis categorized all inspiratory PhMNs (on the basis of their F_max value) into 3 classes with final centers at 40, 90 and 181 Hz, which correspond to the means of maximal frequencies of PhMNs as well as to the ranges of PhN rhythm bands (LFO, MFO, and HFO; Table 2). As shown in Fig. 3A and Table 1, there is a significant (P < 0.05) difference between the F_max value of all groups of inspiratory neurons. PhMNs with exclusively inspiratory activity were classified according to their F_max value as high-frequency (HF; 111–225 Hz, n = 10), medium-frequency (MF; 56–110 Hz, n = 7), and low-frequency (LF; 10–55 Hz, n = 7) units (Fig. 4, pseudocolored plots). By the same metric, those containing background activity during expiration were identified and classified as HF+BG (n = 6), MF+BG (n = 2), and LF+BG (n = 1) units. All recorded inspiratory units exhibited an augmenting pattern except unit 32 (Fig. 4; MF+BG), which possessed decrementing firing dynamics. The averaged CTHs of inspiratory PhMN groups (LF, MF, HF, and HF+BG) reveal augmenting patterns (see Fig. 4, bottom). The firing patterns of individual PhMNs are shown in Fig. 5.

Fig. 2.

Location of recorded PhMNs. Top: photomicrograph of a PhMN (arrow) juxtacellularly labeled with Neurobiotin following recording. Bottom: locations of PhMN somata, reconstructed using standard ventral horn sections at C₃, C₄, and C₅. Red squares, HF PhMNs (n = 10); yellow squares, MF PhMNs (n = 7); black squares, LF PhMNs (n = 7); blue circles, HF+BG PhMN (n = 6); and green circles, MF+BG and LF+BG PhMNs (n = 3); where BG indicates PhMNs containing background activity. Numerals indicate archetypical gray matter regions. Arrow in C₄ section shows location of neuron labeled in photomicrograph.

Table 2.

Unit-PhN coherence values between inspiratory PhMNs and ipsilateral or contralateral PhN during entire, first half, or second half of inspiration

		Ipsilateral Coherence [Peak Hz]			Contralateral Coherence [Peak Hz]
PhMN Type	Frequency Band, Hz	Inspiration	1st Half	2nd Half	Inspiration	1st Half	2nd Half
LF	LFO, 10–60	0.125 ± 0.031 [48.83]	0.110 ± 0.017 [52.73]	0.142 ± 0.018 [46.88]	0.104 ± 0.024 [58.59]	0.087 ± 0.016 [52.73]	0.129 ± 0.019 [60.45]
MF	MFO, 61–110	0.288 ± 0.064 [74.22]	0.278 ± 0.063 [70.31]	0.286 ± 0.06 [74.22]	0.280 ± 0.062 [72.27]	0.287 ± 0.065 [72.27]	0.302 ± 0.065 [78.13]
HF	MFO, 61–110	0.151 ± 0.048 [99.61]	0.200 ± 0.027 [74.22]	0.159 ± 0.022 [111.33]	0.154 ± 0.051 [83.98]	0.170 ± 0.023 [89.84]	0.150 ± 0.021 [111.33]
	HFO, 111–220	0.113 ± 0.032 [152.35]	0.062 ± 0.019 [152.35]	0.157 ± 0.038 [154.3]	0.104 ± 0.025 [152.35]	0.071 ± 0.021 [154.3]	0.143 ± 0.034 [154.3]
HF+BG	HFO, 111–220	0.119 ± 0.036 [167.97]	129 ± 0.042 [171.88]	0.110 ± 0.038 [166.02]	0.099 ± 0.03 [166.02]	0.107 ± 0.035 [171.88]	0.090 ± 0.03 [166.02]

Values are means ± SD of coherence between inspiratory PhMNs and ipsilateral or contralateral phrenic nerve (PhN) for PhMNs in with oscillations in LF (LFO), MF (MFO), and HF (HFO) bands. Peak frequency is shown in brackets. HF+BG, HF PhMNs with background activity.

Fig. 4.

Firing patterns of PhMNs. Top: firing frequency dynamics of HF PhMNs (units 1–10), MF PhMNs (units 11–17), LF PhMNs (units 18–24), HF+BG PhMNs (units 25–30), MF+BG PhMNs (units 31–32) and an LF+BG PhMN (unit 33). Vertical color bar (right) indicates frequency range in Hz and applies to all units. Black vertical lines at “0” and “1” delineate normalized onset and offset of inspiration, respectively. Black boxes show the SD of unit activity onset and offset. Letters A–F indicate units selected for corresponding panels in Fig. 5. Dashed red lines divide different classes of PhMNs. Bottom: population-averaged cycle-triggered histograms (CTHs) of PhMN frequency dynamics within a given type. Green, LF PhMNs; blue, MF PhMNs; red, HF PhMNs; magenta, HF+BG PhMNs during inspiration (between dashed gray lines).

Fig. 5.

Representative PhMN firing patterns. A: LF PhMN (unit 22). B: MF PhMN (unit 17). C: HF PhMN (unit 3). D: LF+BG PhMN (unit 33). E: MF+BG PhMN (unit 31). F: HF+BG PhMN (unit 25). Traces, from top, show instantaneous firing frequency (25-ms resolution), extracellular neurogram, and PhN activity. Time scale bars are 0.5 s for all panels.

The differences between SDs (black boxes in Fig. 4) for onset and offset times of individual units were also analyzed. HF PhMNs are mostly early recruited and have more consistent firing onset times, characterized by smaller SDs (narrower dark horizontal boxes in Fig. 4), compared with MF and LF PhMNs (Fig. 3). LF PhMNs begin firing in the middle to late stages of inspiration, with relatively inconsistent onset and offset times. HF and MF PhMN onset times were not significantly different from each other, and both preceded (P < 0.05) onset times of LF units (see Table 1 and Fig. 3B). There were no significant differences (P > 0.05) between offset times of all classes of inspiratory PhMNs (see Table 1 and Fig. 3D). However, significant (P < 0.05) differences in offset time consistency (SDs) of both HF and MF vs. LF PhMNs (but not between HF and MF units) were uncovered (see Table 1 and Fig. 3, C and E). Because of inconsistencies in background firing rates (i.e., from different recording epochs), onset and offset times for PhMNs with background activity were not analyzed.

Time-Frequency Coherence

To investigate the contribution of different classes of PhMNs to the formation of fast oscillations in PhN activity, PhMN-PhN time-frequency coherence was estimated. Figure 6 shows group PhMN-PhN time-frequency coherence (column 1, ipsilateral PhN; column 2, contralateral PhN), time-averaged coherence (column 3), and individual PhMN-PhN time-frequency coherence (column 4, ipsilateral). As shown in Fig. 6, statistically significant coherence between MF and HF PhMN firing patterns (Fig. 6, B1–D1, B2–D2) and PhN oscillations is present at the onset of inspiration. LF units showed high coherence at the end of inspiration but failed to demonstrate this degree of phase constancy at the onset of firing (Fig. 6, A1 and A2). No significant differences in unit-PhN coherences were observed between ipsilateral and contralateral PhN when it was time-averaged over inspiration (Fig. 6, A3–D3; see Fig. 7 and Table 2), although differences may exist in peak values at particular time-frequency coordinates (Fig. 6, A1–D1, A2–D2). Because exclusively inspiratory HF PhMN-PhN coherence contained two significant frequency bands, an early MFO and a late HFO (Fig. 6, C1–C4), they were analyzed separately for each component (see Fig. 7 and Table 2). HF+BG units showed a striking augmenting pattern in HFO coherence, akin to their firing pattern (Fig. 4, bottom, magenta; Fig. 5F; Fig. 6, D1 and D2). As demonstrated in Fig. 6, C1 and C2, HF units start firing at the MFO frequency band but eventually show good matching during the second half of inspiration, where HFOs are sparsely distributed with respect to their firing frequencies. Early high-frequency coherence of HF+BG PhMNs (Fig. 6, D1 and D2) shows good temporal matching with early HFO in PhN autospectra (Fig. 6, E1 and E2) and bilateral PhN coherence (Fig. 6F1). The MFO band in PhN autospectra and left-right coherence (Fig. 6F1) has an initial start-up component that matches the onset time and frequency of unit-PhN coherence for exclusively inspiratory HF PhMNs (Fig. 6, C1 and C2). Also, population time-averaged coherence for MF PhMNs (Fig. 6B3) was significantly higher than for any other units (Fig. 6, A3–D3; see Fig. 7). Importantly, individual unit-PhN time-frequency coherence representations (Fig. 6, A4–D4) are reflective of the population-averaged values for each class of PhMN analyzed. Also, individual PhN autospectra and PhN left-right coherences (Fig. 6, E4 and F4) are consistent with averaged population activity (Fig. 6, A4–D4).

Finally, the contribution of the four classes of PhMNs to PhN oscillations was also determined by estimating the unit-PhN coherence in ipsilateral vs. contralateral sides during the first vs. second halves of inspiration. In LF PhMNs, the ipsilateral coherence during the second half of inspiration is significantly higher than the corresponding ipsilateral and contralateral coherences during the first half (Fig. 8A; see Fig. 7 and Table 2). MF PhMNs showed no difference between the first and second halves of inspiration for ipsilateral vs. contralateral sides (Fig. 8B). Additional specific significant differences in PhMN-PhN coherences are shown in Fig. 7 (asterisks). In the HFO band (see Table 2 for frequencies), HF PhMN-PhN coherence was significantly higher during the second part of inspiration than during the first (Fig. 8C; see Fig. 7 and Table 2). In the same neuron class, PhMN-PhN coherence in the MFO band exhibited during the first half of inspiration was significantly larger in the ipsilateral PhN (Fig. 8C, red vs. green). Significantly high coherence values for HF+BG PhMNs during the first half of inspiration, and low values during the second half (Fig. 8D and Table 2), were also revealed by TFR analysis (Fig. 6, D1–D3).

Fig. 8.

PhMN-PhN coherence for the first and second halves of inspiration (group data). A: LF PhMN-PhN coherence. B: MF PhMN-PhN coherence. C: HF PhMN-PhN coherence. D: HF+BG PhMN-PhN coherence. Pairs of colored bars along the frequency scale show inspiratory phase locations of statistically significant differences (P ≤ 0.05) between corresponding coherence values. Gray horizontal lines in all panels indicate the 95% confidence threshold for coherence.

DISCUSSION

There are four principal findings of this study. First, we report the discovery of a subset of PhMNs that fire action potentials at rates corresponding to frequencies categorized as high-frequency oscillations in the rat PhN. Second, nonparametric, time-frequency coherence analysis indicates that firing of different classes of PhMNs (classified by K-means cluster analysis according to F_max) is phase-locked with PhN oscillations over various epochs of inspiration. Third, certain classes of PhMNs display coherence with PhN oscillations in multiple frequency bands and may contribute to different bands during different parts of the inspiratory burst. Fourth, for PhMNs that did not exhibit tonic background firing during all phases of respiration, the consistency of firing initiation and cessation during each burst differed depending on the class of PhMN.

Methodological Considerations

Results from our previous studies (Marchenko et al. 2002; Marchenko and Rogers 2006a, 2006b, 2009) and other studies (Bruce 1988; Kocsis and Gyimesi-Pelczer 1997; Richardson and Mitchell 1982) have demonstrated the presence of fast rhythms (approximately 50–200 Hz) in the rat PhN inspiratory discharge during eupneic output. The discovery of PhMNs that fire at HFO-related frequencies, and whose activity is (statistically significantly) coherent with PhN discharge, promotes near-synchronous firing of PhMNs as an HFO-generating mechanism from the realm of speculative to possible. We believe that methodological techniques contributed to our ability to record from these neurons with regularity. Our experience suggests that these HF PhMNs are smaller neurons whose activity may be lost in the background when low-impedance microelectrodes are used to record extracellularly in the ventral horn. Our use of higher impedance microelectrodes (see methods) in these experiments enabled isolation of single-unit activity of HF PhMNs.

Before the present study was conducted, there was almost no evidence of PhMNs firing at frequencies corresponding to HFO in phrenic spectra in rats (∼140–210 Hz) or cats (70–100 Hz) during normocapnic eupnea (Christakos et al. 1991; Hayashi and Fukuda 1995; Jodkowski et al. 1988; Kong and Berger 1986; Nail et al. 1972; St.-John and Bartlett 1979). Richardson and Mitchell (1982) analyzed power spectra and interspike intervals (ISIs) of fibers teased from the cat PhN. For one PhMN, these authors showed at least four ISI peaks, one of which corresponds to HFO and the others to MFO and LFO. No neurograms of single fibers were shown, and the multiple-peak ISI may reflect either multifiber activity, cycle skipping in a single unit, or both. Another study (Christakos et al. 1991) reported high coherence between PhMN and ipsilateral PhN activity but mistakenly claimed that the ISI peak of that unit, which was actually in the MFO range, corresponded to HFO (see Fig. 4 of that report). No records of PhMNs were shown in the aforementioned report, and we assert that the presence of few or no ISIs with inverses corresponding to HFO disproves direct contribution to HFO by any individual PhMN.

We developed a novel method of estimating the coherence between individual PhMN spike trains (a discrete signal) and the unipolar PhN neurogram (a continuous signal). Previous studies (e.g., Christakos et al. 1991, 1994) have detailed approaches to this problem, as well. The common feature of all methods, including ours, is that a variable-length waveform, be it a half-cosine or a sin(x)/x, is convolved with a delta function located at the time of each spike. However, the technical details of forming a continuous waveform from a series of discrete events are less critical than the interpretation of results gleaned from them. The power spectrum of a spike train will invariably contain frequencies, typically harmonics of the actual firing frequencies, at which the neuron never fires. If actual firing rates are not reported, there is legitimate question regarding the functional implications of such findings. Investigators have routinely attributed “coherence” at frequencies to which the neuron could have contributed, at most, to subharmonics of those frequencies in the whole nerve activity (e.g., Christakos et al. 1991, Fig. 4). It is for this reason that we are particularly cautious when suggesting a causal link between a given PhMN's firing and PhN activity, and we impose a criterion that the neuron actually fires at those instantaneous rates. In the present study, we validated our claims on the basis of actual firing rates and specified causal effects further by using nonparametric time-frequency coherence analysis. Even when these criteria are met, causality is still constrained by network architecture (e.g., it is highly unlikely that HF PhMNs with significant coherence with the contralateral PhN cause HFO therein).

Nonparametric time-frequency analysis enables the specification of various “roles” played by different PhMN classes in generating fast oscillations in PhN activity. Our present results demonstrate the existence of HF PhMNs, which we separated into two classes depending on whether or not they had tonic background activity. One class, with tonic activity during all phases of respiration, contributed to the HFO band during the early part (first half) of PhN inspiratory activity (Fig. 6D and Table 1). As a group, these units have an incrementing discharge pattern, with a penultimate firing rate of 180.54 Hz (Table 1). The HF PhMNs that fired exclusively during inspiration contributed to the formation of HFO during middle/late inspiration (Figs. 4 and 6). This latter group also contributes to both the burst onset (“startup”; see Cohen 1981; Marchenko and Rogers 2006a) and to early inspiratory MFO in the PhN (Fig. 6C).

We also categorized slowly incrementing MF and LF PhMNs (Fig. 6, A and B). Exclusively inspiratory HF and MF units represent a class of early-recruited PhMNs, whereas LF units tend to be late-recruited (Figs. 4 and 5D). Statistical analysis revealed that the SD of onset times for MF and LF PhMNs tended toward significantly higher values than those observed for HF units (Fig. 3 and Table 1). In addition, SDs of offset times for LF units were significantly higher than those of MF and HF units, suggesting that control of HF PhMN startup and termination is more tightly regulated than those of MF and LF, and perhaps involve different underlying mechanisms. These may include differences in subnetwork connectivity, synaptic mechanisms (neurotransmitter, magnitude and spatial distribution) and intrinsic membrane properties such as action potential duration, time/space constants, and active conductances of PhMNs (Berger 1979; Dick et al. 1987; Funk and Parkis 2002; Iscoe et al. 1976; Jodkowski et al. 1987; Purpura and Chatfield 1952). Furthermore, it has been noted that PhMNs in rats, compared with those in cats, have a shorter afterhyperpolarization duration, a smaller action potential half-width, a larger input membrane resistance, a smaller rheobase, and a shorter minimum paired-pulse interval to provoke the second spike (Hayashi and Fukuda 1995). The last feature may partially explain the capability of rat PhMNs to produce the higher firing rates that likely generate PhN MFO and HFO bands.

Our results indicated a much stronger unit-PhN coherence for MF PhMNs than for other classes (Figs. 6, B1–B3, 7, and 8B; Table 2). This result indicates a more consistent phase-locking of MF PhMN firing to PhN oscillations at that frequency band, suggesting that the bulbospinal transfer function may be consistent across the population of those PhMNs. Historically, it has been argued that HFO “generation” originates at the medullary level, based on high coherence between heterologous nerves (Huang et al. 1996; Schmid et al. 1990), lesion studies (Romaniuk and Bruce 1991), and the existence of bulbospinal neurons that fire at these rates (Duffin and van Alphen 1995; Funk and Parkis 2002). The lower unit-PhN coherence for HF PhMNs compared with MF PhMNs may be explained by nonlinear features of the transfer function between supraspinal synaptic inputs and spinal motoneuron firing output (Hultborn et al. 2003). The same absolute jitter in spike timing (i.e., in ms) causes a larger phase angle error for high frequencies than for lower frequencies, thereby making HF PhMN-PhN coherence more susceptible to degradation than MF PhMN-PhN coherence. The relatively low unit-PhN coherence for LF PhMNs may be due to the high variability of, among other factors, unit firing onset (Figs. 3C and 4; Table 1). Coincident discharge between just a fraction of PhMNs could account for the HFO peak in the unit-PhN coherence (Richardson and Mitchell 1982). We found an early inspiratory unit-PhN coherence at HFO for PhMNs with background activity, whereas other investigators who studied PhMN-PhN relationships analyzed only exclusively inspiratory PhMNs (e.g., Christakos et al. 1991). Moreover, the generation and formation of fast inspiratory rhythms in rat can be substantially different from those of the cat and other mammalian species. As noted previously (Kocsis and Gyimesi-Pelczer 1997), the HFO band in PhN autospectra and in left-right coherence is weaker in rats than in cats. Inconsistency of PhN autospectra MFO and HFO peaks across animals in the rat was also reported by Marchenko et al. (2002). All of these factors may contribute to differences reported in cat vs. rat studies.

In conclusion, the present study supports the model of a heterogeneous PhMN population, a conclusion reached by others (e.g., Berger 1979). The present results suggest that respiratory motoneuron synchronization within a pool, without cycle skipping (i.e., firing at subharmonic frequencies) and “filling in” by different PhMNs in the population (Funk and Parkis 2002; Richardson and Mitchell 1982; van Brederode and Berger 2008), may underlie the fast oscillations observed in whole nerve activity. Nevertheless, our results do not exclude the possibility that HFO are produced by this epiphenomenological/population mechanism, only that they can be produced by synchronized firing at high rates. The contribution of network, synaptic, and intrinsic membrane mechanisms to, and the mathematical form of, motoneuron synchrony remains unclear. Further investigation is required to uncover the functional underpinnings of this issue.

GRANTS

This work was supported in part by National Heart, Lung, and Blood Institute Grant HL68143 (to R. Rogers).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

V.M. and R.F.R. conception and design of research; V.M. and M.G.Z.G. performed experiments; V.M. and M.G.Z.G. analyzed data; V.M. and R.F.R. interpreted results of experiments; V.M., M.G.Z.G., and R.F.R. prepared figures; V.M., M.G.Z.G., and R.F.R. drafted manuscript; V.M., M.G.Z.G., and R.F.R. edited and revised manuscript; V.M., M.G.Z.G., and R.F.R. approved final version of manuscript.