|Home | About | Journals | Submit | Contact Us | Français|
Endogenous burster neurons (EBs) have been found at the level of the facial nucleus (VIIn), and 500 μm caudally, within the preBötzinger Complex (preBötC). They have been proposed as causal to, or playing no role in, respiratory rhythmogenesis. Little is known about their broader distribution in ventrolateral medulla. Here, a Ca2+ indicator was used to record respiratory network activity in ventrolateral medulla, and, following synaptic blockade, to identify EBs active at perfusate K+ concentrations ([K+]o) of 3, 6, and 9 mM. Recordings were made along the respiratory column, extending 300 μm rostrally, and 1100 μm caudally from the caudal pole of VIIn (VIIc), in the in vitro tilted sagittal slab preparation, isolated from neonate male and female Sprague-Dawley rats. Activity under matching [K+]o concentrations in the intact respiratory network was subsequently investigated. Respiratory neurons (n=401) formed statistically significant clusters at the VIIc, within the preBötC, and 100 μm caudal to the preBötC. EBs (n=693) formed statistically significant clusters that overlapped with respiratory clusters at the VIIc, and preBötC. EB activity increased significantly as [K+]o was increased, as did neurons that remained coupled following synaptic blockade. The overlap between respiratory and EB clusters in regions of ventrolateral medulla identified as rhythmogenic supports the hypothesis that EBs are constituents of rhythmogenic networks. In addition, the observation of truncated inspiratory bursts and ectopic bursting in respiratory neurons was observed when [K+]o was elevated in the intact network is consistent with a causal role for EBs in respiratory rhythmogenesis.
In mammals, circuits essential for respiratory rhythmogenesis are distributed bilaterally as columns in ventrolateral medulla. These networks remain spontaneously active in a variety of neonate rodent in vitro medullary preparations (Suzue, 1984; Smith and Feldman, 1987; Smith et al., 1991). Two anatomical regions have been shown to have rhythmogenic function in vitro: the retrotrapezoid nucleus / parafacial respiratory group (RTN/pFRG), ventral and caudal to the facial nucleus (VIIn) (Onimaru et al., 2009), and the pre-Bötzinger complex (preBötC), 500 μm caudal to the caudal pole of the VIIn (VIIc) (Ruangkittisakul et al., 2006; Ruangkittisakul et al., 2008). Both these regions are rich in endogenous bursters (EBs), which display periodic bursting in the absence of synaptic input (Johnson et al., 1994; Onimaru et al., 1995). EBs have been proposed to play a critical role in respiratory rhythm generation, elaborated using cellular (Butera et al., 1999a) and network (Butera et al., 1999b) models, which have obtained experimental support (Del Negro et al., 2001). More recently, it has been shown that EBs are not necessary for respiratory rhythmogenesis (Del Negro et al., 2002a), consistent with the conjecture that respiratory rhythm is generated by excitatory networks of neurons whose bursting properties are irrelevant to their rhythmogenic function (Rekling and Feldman, 1998). This hypothesis has gained empirical support (Del Negro et al., 2002a; Pace et al., 2007; Pace and Del Negro, 2008), and has been elaborated in models (Kosmidis et al., 2004; Rubin et al., 2009). Thus, the role of EBs in respiratory rhythmogenesis continues to be debated.
Although others have characterized EBs within identified respiratory rhythmogenic networks using optical methods (Koshiya and Smith, 1999; Thoby-Brisson et al., 2009), typically, EBs have been studied using single-unit recordings. Because definitive EB identification is obtained following blockade of synaptic transmission only one burster can typically be identified per experiment. Thus, estimates of the number and distribution of EBs along the respiratory column are lacking. Based on these necessarily sparse data, the functional role of EBs has been difficult to assess.
Here, the overlap between respiratory networks and EBs was characterized. Optical recordings were made from the surface of the sagittal slab preparation (Barnes et al., 2007), along the respiratory column under control conditions, and following synaptic blockade at physiological (3 mM) and elevated (6 mM, 9 mM) [K+]o. Because this approach allowed us to sample activity from the network at the exposed surface of the slice, the incidence, distribution, and extent of overlap between respiratory networks and EBs could be assessed directly. We report three main findings: i. EBs form clusters along ventrolateral medulla; ii. clusters of respiratory neurons and clusters of EBs only overlap at the VIIc and in the rostral preBötC, suggesting that EB properties reflect a functional specialization that contributes to the generation and regulation of respiratory rhythm; iii. relatively low EB numbers at physiological [K+]o, suggest that under conditions of low respiratory drive, EBs that are subthreshold to bursting may amplify synaptic inputs via the same the nonlinear membrane properties that generate bursting at more depolarized levels (Ramirez et al., 2004; Feldman and Del Negro, 2006), while active EBs may directly contribute to rhythmogenesis.
In accordance with methods approved by the Institutional Animal Care and Use Committee, neonate male and female rat pups (postnatal day 0-3) were anesthetized with isoflurane. The brainstem (transected just rostral to VIIn) and spinal cord were isolated using standard methods (Smith and Feldman, 1987) in a dissection chamber perfused with chilled (5° C) artificial cerebrospinal fluid (aCSF) containing (in mM): 128.0 NaCl, 3.0 KCl, 1.5 CaCl2, 1.0 MgSO4, 21.0 NaHCO3, 0.5 NaH2PO4, and 30.0 glucose, equilibrated with 95%O2-5% CO2 (carbogen).
A tilted sagittal slab preparation was cut at a compound angle of 2.2° rostrocaudal, and 12° ventrodorsal tilt, using a device developed for this purpose (Mellen, 2008). Level of section was determined as a ratio of brainstem width: respiratory networks were exposed by cutting at 0.34 of the distance from the midline to the lateral edge of the preparation; the blind side was cut at 0.7-0.8 of the full width of the brainstem. Although small variations between preparations are unavoidable, both because of experimenter error, and biological variability, this variation must be small for the preparation to remain viable: preparations cut too medially fail to generate a stable rhythm, while preparations cut too laterally fail to generate optical signals.
The preparation was then mounted on the narrow edge of 2 mm thick semi-circular inert elastomer slab (Sylgard, Dow-Corning, Midland MI) and incubated for 1 h in an carbogenated solution containing the low affinity (KD=1.86 μM) cell-permeant Ca2+ indicator fluo-8L AM (ABDBioquest, Sunnyvale CA; 50 μg), 10 μl of a 20% solution of the surfactant pluronic F-127 (Invitrogen, Carlsbad CA), dissolved in DMSO, to which 750 μl aCSF was added for a final concentration of 60 μM. After dye-loading, the dye-loaded preparation was transferred to a recording chamber (~2 ml volume; JG 23 W/HP, Warner Instruments, Hamden CT), mounted on an upright microscope (Axioscop 2, Zeiss Instruments), perfused (4 ml/min) with carbogenated aCSF maintained at 27° C.
A suction electrode connected to a differential amplifier (EXT-01C/DPA 2F; NPI Electronics, Tamm, Germany) was placed on ventral root C1, and the population recording was digitized at 1 kHz (PCI-MIO-16XE-10, National Instruments, Austin TX). Fluorescence images were obtained using a cooled (-80 °C) CCD camera (EM-CCD 9100-13, Hamamatsu Corp, Bridgewater NJ), connected to a frame-grabber (IMAQ PCI-1422, National Instruments, Austin TX), and illuminated using a shuttered xenon arc lamp (Lambda DG-4, Sutter Instruments, Novato CA). Low magnification (10X) images of the preparation were obtained to detect regions of optical activity, and to obtain reference images for the alignment of high magnification (20X) image series used during experimental protocols. In experiments in which respiratory networks were recorded before and during synaptic blockade, exposure times were minimized by recording at 20X in 2X2 binning mode (pixels=1.6×1.6 μm), which permitted brief exposures (≤ 20 ms), sampled at 3 Hz. Thus, over the course of an experiment, total light exposure did not exceed 12 s, thereby minimizing photobleaching and phototoxicity. In network-intact [K+]o manipulation experiments, the sampling rate was increased to 10 Hz so as to be able to resolve ectopic bursting; because these experiments involved a total of 300 s of recordings, total exposure time was 20 s. Because lipophilic indicators only label cells within 40 μm of the slice surface (Funke et al. 2007), only superficial cells were recorded from.
Overlapping low magnification (10X) image series were obtained at 3 Hz for 40-60 s to detect regions of respiration-modulated neuronal activity. Identified respiratory networks were then recorded at high magnification (20X) for 40-60 s at 3 Hz in 2X2 binning mode. EB activity was detected by recording from the same regions from which respiratory network activity had been sampled, following blockade of fast synaptic transmission using a cocktail consisting of the AMPA receptor blocker 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX; 20 μM), the NMDA receptor blocker 2-amino-5-phosphopentanoic acid (APV; 20 μM), the GABAA receptor blocker bicucculine (10 μM), and the glycine receptor blocker strychnine (1 μM; all drugs from Sigma-Aldrich, St. Louis MO). This cocktail was dissolved in aCSF containing 3, 6, and 9 mM [K+]o. The presentation order of each [K+]o solution was randomized to minimize the bias associated with slow-timecourse changes in recording conditions, such as preparation run-down, photobleaching and dye sequestration. In each experiment, optical recordings were carried out from up to two locations with little or no overlap. A digital translation stage (MT-2000, Sutter Instruments, Novato CA) that provided recording location coordinates permitted reproducible repositioning from one location to the other. Optical recordings were initiated 5 min after cessation of motor output, and 3 min following changes in [K+]o under continued synaptic blockade; in subsequent experiments in which synaptically coupled networks were recorded under 3, 6, and 9 mM [K+]o, equilibration time was extended to 10 min. Because recordings were made from up to two locations, under 4 conditions, with delays associated with wash-in and repositioning, a balance had to be struck between recording duration, and slow time-course changes associated with and dye sequestration (Thomas et al., 2000). In light of these constraints, recording epochs under synaptic blockade conditions lasted no less than 133s and no more than 167 s (400-500 frames).
Ca2+ transients associated with neuronal activity were extracted from each image series using semi-automated methods implemented in LabView (National Instruments, Austin TX) carried out in 3 stages: region of interest (ROI) generation, signal extraction and manual screening (Mellen and Tuong, 2009). Single-wavelength Ca2+ indicators such as fluo-8L used here provide information about relative rather than absolute changes in [Ca2+]i; in addition, although fluo-8L has relatively low Ca2+ affinity (Kd=1.86 μM), signal decay does not only reflect neuronal activity, but also dye-Ca2+ binding kinetics. As a consequence, although qualitative changes in Ca2+ signals may be interpretable, generally, only event times associated with optical trace peaks are extracted for subsequent analysis.
Traces were subjected to high-pass filtering: each trace was subjected to a 14 s moving average low-pass filtering to eliminate physiologically interesting transients from the signal. High-pass filtering was then obtained by subtracting the low-pass filtered signal from the raw signal, and the average of this low-pass filtered signal was used as our estimator of static fluorescence F, which was used to scale changes in relative intensity ΔF. Finally, to facilitate accurate peak detection, jitter associated with fluctuations in arc-lamp intensity was reduced by applying a two-frame moving average to the high-pass filtered trace. Peaks were then extracted using trough-to-peak amplitude. A quadratic fit algorithm was then applied to extract peak times, and dots corresponding to peak locations were superimposed on optical traces to permit validation of peak selection parameters. Because the moving average did not uniformly filter data at the beginning and end of datasets, spurious peaks associated with unfiltered data were eliminated from the beginning and end of each trace.
Identification of respiration-modulated neurons was achieved by inspecting traces superimposed on root activity, and based on the burst-triggered average of the optical trace, which in the case of respiration-modulated neurons showed a well-defined peak or trough in relation to the averaged inspiratory burst. EBs were selected in two stages: first all traces obtained under conditions of synaptic blockade showing time-varying fluctuations in brightness were included. Thereafter, only traces whose peak-to-peak periods had a coefficient of variation (CV) ≤ 0.50 were retained.
For each experiment, high-magnification images and their associated ROIs were aligned to the low-magnification images obtained at the experiment’s outset (Photoshop; Adobe Systems, San Jose, CA) based on anatomical and morphological matches. From each aligned set of images, line drawings were extracted, consisting of: i. outlines (drawn by hand) of the brainstem’s ventral surface and the VIIn, which could be identified by banding and larger somata within the VIIn, and by densely packed, smaller cells along its border; ii. ROIs generated under control conditions, and following synaptic blockade, under each [K+]o condition. These line-drawings were exported to Illustrator (Adobe Systems, San Jose, CA). All sequentially recorded high-magnification recordings were brought into register with each other based on anatomical and morphological matches, so that neurons active under more than one condition could be identified based on the overlap of their ROIs, and their associated traces could be indexed as sharing a common source. In order to collate data across experiments, an Illustrator file consisting of 5 layers was generated: the first contained the anatomical outlines of the ventral surface and the VIIn, subsequent layers contained dots aligned to ROIs color-coded according to recording condition (control, 3, 6, and 9 mM [K+]o). Data across experiments were then aligned using the VIIc and the ventral surface within a common coordinate frame that spanned all recording locations (dashed box, Figure 1). In-house software was then used to extract dot coordinates from each dataset’s aligned dot diagram, based on dot color, for analysis using spatial statistical methods. Because the VIIc shows little mediolateral taper, the rostrocaudal location of VIIc used here to align datasets is a reliable landmark.
Because of the low sampling rates used here, cross-correlation could not be used to characterize coupling between respiratory network constituents in the intact network. Under conditions of synaptic blockade however, cross-correlation was used to identify neurons that remained coupled despite blockade of fast synaptic transmission. Coupling was treated as a binary variable, with the mean of cross-correlation coefficients obtained from all pair-wise comparisons between respiration-modulated neurons (R2=0.67) as the threshold for classifying EBs as coupled. Coupling following fast synaptic transmission blockade was tested for between all EBs in every dataset.
To test whether EBs and respiratory neurons show complete spatial randomness (CSR) in their distribution, the χ2 test based on 100 μm × 100 μm quadrat counts was used. In addition, we tested for the presence of statistically significant clusters by using the dirichlet (or voronoi) tessellation method, a measure to detect spatial clustering, in which for a given point p in the pattern X, a tile is drawn that contains the region of space closer to p than to any other point of X (Okabe et al., 1992; Duyckaerts and Godefroy, 2000). Thus, for points in clusters, the associated tiles are small. Statistically significant clusters were identified by comparing dirichlet tile sizes from the actual data, to tile areas obtained from surrogate datasets in which point coordinates from the original dataset were randomized using the quadratureresample command, which creates a resampled point pattern by dividing the bounding rectangle enclosing all points in the dot diagram (dotted box, Figure 1) into rectangular quadrats and randomly resampling the list of quadrats, followed by the rjitter command, which randomly displaced each point in the dataset. Because the logarithm of tile sizes from surrogate datasets approximated a normal distribution, an estimator of the 95% confidence interval (CI) for log tile sizes from randomized distributions were obtained from 10 surrogate datasets, and points associated with contiguous dirichlet tiles obtained from the actual datapoints whose logarithmically transformed tile size was smaller than the 95% CI of the surrogate dataset tile sizes were considered constituents of statistically significant clusters (Supplemental figure 1). All analyses were carried out using the spatstat module of the open-source statistics package R (Baddeley and Turner, 2005).
Descriptive statistics and tests for statistical significance of cell counts and respiratory periods were carried out using R and Origin (OriginLabs; Northampton MA); data are reported as mean ± SEM, except in cases when data are skewed, then the median is reported.
In 23 experiments, a total of 35 locations were sampled over consecutive epochs, both under control conditions, and following synaptic blockade, at 3, 6, and 9 mM [K+]o, in randomized presentation order. In 12/23 experiments, recordings at 2 locations were carried out; in the remaining 11/23 postnatal day 0 to postnatal day 3 preparations, sequential recordings were made from only 1 location. Recordings extended 300 μm rostrally, and 1100 μm caudally from the VIIc along ventrolateral medulla, covering a region including, but not limited to, the preBötC and the RTN/pFRG (Figure 1). In total, 693 EBs were recorded from, as well as 401 respiratory neurons of which 24% (97/401) were also EBs. Although the low sampling rates used here preclude detailed classification of respiratory neurons based on firing phase, tonically active expiratory neurons, silent only during inspiration could unambiguously be identified, and constituted 16 % of the respiratory neuron sample (64/401), double the frequency reported in the embryonic mouse en bloc preparation (Eugenin et al., 2006). Subsequently, another 7 experiments from synaptically coupled networks at 3, 6, and 9 mM [K+]o were carried out from respiratory networks at VIIc (n=3) and at the preBötC (n=4); a total of 405 respiratory neurons were recorded, of which 37% (151/405) were active at more than one [K+]o.
In a representative experiment, respiratory networks (Figure 2 A, white ROIs) were recorded at the VIIc (4 cells), and 500 μm caudally, at the level of the preBötC (7 cells). Because respiratory neurons could robustly be identified by comparison to motor output traces recorded from C1, control activity was only sampled for 40-60 s (Figure 2 B). Respiratory neurons had high CVs (to the left of traces, Figure 2 B), both because of skipped cycles (shaded box, Figure 2 B), and ectopic bursts (i.e., neuronal activity unaccompanied by inspiratory motor output; vertical arrows, Figure 2 B). In this dataset, EBs were found at 3 mM only at the level of the preBötC (Figure 2 A, blue ROIs), but at 6 mM (Figure 2 A, green ROIs), and 9 mM (Figure 2 A, red ROIs) EBs were found in both recording locations. In this experiment, 3 of the respiration-modulated neurons at the VIIc were found to be EBs following synaptic blockade; of these, one was active at both 3 and 6 mM [K+]o. In addition, one non-respiratory EB was active at both 6 mM and 9 mM [K+]o (traces associated with a single unit’s activity are linked by fine lines in Figure 2 B).
Because EBs were recorded optically, and thus sampled in parallel, coupling that persisted despite fast synaptic transmission blockade could be detected. Because many bursters had similar frequencies, even if uncoupled, they appear to align because of slowly drifting phase relations. Coupling could be identified by cross-correlation analysis. In this dataset, only two pairs of EBs met or exceeded the threshold obtained from cross-correlation analysis of respiration-modulated neurons in the coupled networks (heavier traces, Figure 2 B), one pair at the level of the preBötC under 9 mM [K+]o (R2=0.78), the other at VIIc under 6 mM [K+]o (R2=0.86). In these pairs, the phase relationship between optical traces is constant, but in addition, aperiodic fluctuations in burst envelope and bursting frequency are reproduced in both traces of each pair, consistent with coupling. Because the ROIs from which these these traces were obtained are at a distance from one another (linked by dotted lines with arrowheads, Figure 2 A), it is unlikely that these closely matched traces were generated by a common neuron.
Despite the low sampling rates used here, optical recordings revealed considerable EB heterogeneity. In some cells, bursts shared the steep rise and exponential roll-off common to respiration-modulated neurons, while others had symmetric saw-tooth or rounded burst envelopes; further, some cells showed burst-to-burst variability in amplitude, whereas others had highly stereotypic burst amplitudes and envelopes.
Respiratory neuron periods (median = 6.8 s) measured prior to synaptic blockade (Figure 3 A.i) were significantly shorter (p<0.01, Wilcoxon ranked means test) than EB periods (median =11.7 s); this difference is apparent in the relatively small overlap between the histograms of respiratory periods (Figure 3 B, solid grey) and EB periods (Figure 3 B, black line). By inspection, respiratory neurons appear to have more variable periods than EBs. On one hand, respiratory neurons were identifiable by comparison with motor output, thus cells that failed to fire in a given respiratory cycle, and/or that fired in the absence of motor output could nonetheless be identified as respiratory neurons despite having variable periods. On the other hand, because all neurons with a CV>0.5 were excluded from this study, the variability of the sample included in this study reflects the inclusion criteria used.
Mean respiratory and non-respiratory EB periods at 3 mM (13.0 s; 13.3 s) were not significantly slower than EB period at 6 mM (12.3 s; 13.0 s) or 9 mM (12.0; 12.6 s; p>0.2; Figure 3 A.i-iv). At each [K+]o, mean period reflects the contribution of neurons active only in that recording epoch, as well as the contribution of neurons active in more than one recording epoch. Neurons active in multiple recording epochs were 26% (181/693) of EBs recorded from, but their contribution to mean EB period was greater than their numbers, since their activity at each [K+]o generated a datapoint. Thus, for each [K+]o, roughly half of the EBs recorded from were active at more than one [K+]o (Figure 3 C, grey bars). The lack of significant differences in period at different [K+]o is consistent with the observation that slightly less than half of the neurons active at multiple [K+]o showed a decrease in period at a more elevated [K+]o. Thus, the lack of significant differences between EB periods obtained at different [K+]o reflects both the lack of consistent voltage-dependence among EBs active at more than one [K+]o, and the homogeneity of bursting frequencies among EBs active at only one [K+]o (Onimaru and Homma, 2008).
Of the 97 respiration modulated EBs recorded from, 24 were active at 3 mM, 33 at 6 mM, and 40 at 9 mM [K+]o. The average number of respiration-modulated EBs per experiment was 2.8 ±2 (dark grey bar Figure 3 C). In addition, 41% of these respiration-modulated EBs (40/97) were active at more than one [K+]o.
The average number of EBs per recording location increased with [K+]o, from 3 mM (4.0±0.7 cells; total n=91), to 6 mM (12.1±2.1 cells; total n=291), and 9 mM (20.0±2.2 cells total n=450); these differences were statistically significant (p<0.01, with Bonferroni correction; Figure 3 C). This increase is also apparent in EB totals at each [K+]o, which increase monotonically from 3 mM (n=91) to 6 mM (n=291) to 9 mM (n=450); Although this increase is plausibly due to the transition from quiescence at low [K+]o to bursting at higher [K+]o, it also likely overstates EB numbers, due to increased coupling detected at higher [K+]o. We found that the number of coupled bursters (as a percentage of all bursters) increased from 1.5±0.1% at 3 mM [K+]o to 7.1±2.5% at 6 mM, and 11.4±2.6% at 9 mM (Figure 3 D). Although only the difference between 3 mM and 9 mM [K+]o was statistically significant (p< 0.05, with Bonferroni correction, Figure 3 D), the observed coupling between cells following synaptic blockade at higher [K+]o suggests that at higher [K+]o, at least some of the rhythmically active cells follow, rather than spontaneously generate rhythmic activity. This observation suggests that at elevated [K+]o, with the synaptic isolation methods used here, the number of EBs are likely overestimated.
A goal of this study was to characterize the anatomical distribution of EBs in relation to respiratory networks. To this end, datasets were aligned using the VIIc and the ventral surface as aligning landmarks (fine gray lines Figure 4 A). Based on functional-anatomical studies (Smith et al., 1991; Ruangkittisakul et al., 2008), the boundaries of the preBötC span 200 μm, centered 500 μm caudal to the VIIc; these anatomical boundaries are shown in Figure 4 as heavy broken lines.
Consistent with many earlier published reports, respiratory neurons (hollow black circles, Figure 4 A) were found all along the rostrocaudal extent of ventrolateral medulla, while EBs (hollow red circles, Figure 4 A) were relatively sparse caudal to the preBötC (n=150). Respiration modulated neurons that were later identified as EBs (filled black circles) were mainly found at or rostral to the preBötC. Although the overall proportion of respiration-modulated EBs to respiration-modulated neurons was 24% across all [K+]o, within the boundaries of the preBötC, this proportion was 55% (28/51); in a 200 μm window centered on VIIc, this proportion was 53% (23/43).
The χ2 test rejected the null hypothesis that the distribution of either respiratory neurons or EBs conformed to complete spatial randomness (CSR; p<0.001). To qualitatively describe clustering, density plots were generated using a 100 μm Gaussian smoothing kernel. These plots revealed that points associated with respiratory neurons formed 4 clusters, within the VIIn and at the VIIc, two adjacent clusters within the preBötC, and a fourth cluster caudal and slightly dorsal to the preBötC (Figure 4 B). By contrast, EBs formed only 3 clusters, two partially merged clusters at the VIIc and extending dorsal and caudally, and a third cluster at the rostral margin of the preBötC (Figure 4 C).
Because the shape and number of peaks in the density plot were sensitive to choice of smoothing kernel size, we used dirichlet tessellations to identify statistically significant clusters. Statistically significant clusters (tiles and points outlined in black in Figure 4 B and in red in Figure 4 C) superimpose onto peaks in the density plot. Applied to respiratory neurons, these methods identified three statistically significant clusters of more than 8 cells within, but at the caudal margin of the VIIn; two clusters of at least 16 contiguous cells at the rostral and caudal margins of the preBötC; finally, a third region containing 2 clusters of more than 22 cells extended caudally from approximately 100 μm caudal to the preBötC.
Statistically significant clusters of EBs were larger and fewer than those associated with respiratory neurons. The first was found at or just rostral to the VIIc, the second about 50 μm caudal to the VIIc and slightly dorsal to the first cluster, each containing at least 21 cells; a third statistically significant cluster containing 43 cells was identified at the rostral margin of the preBötC (Figure 4 C). Although other dirichlet tiles associated with both respiratory neurons and EBs that were significantly smaller than tiles generated by shuffled surrogate data were distributed along the neuraxis, because they were spatially isolated, they are not considered markers of statistically significant clusters.
In order to directly compare the distribution of respiratory neurons and EBs, outlines of the large statistically significant clusters in Figure 4 B and C are superimposed (Figure 4 D). Strikingly, respiratory neurons (blue outlines, Figure 4 D) and EBs (red outlines, Figure 4 D) only overlapped at the VIIc and at the rostral margin of the preBötC. These regions of overlap between statistically significant clusters of respiratory neurons and EBs, closely matched statistically significant clusters of respiration-modulated EBs (purple outlines, Figure 4 D, associated density plot not shown). In addition to these overlapping clusters, both EBs (between the VIIc and the preBötC) and respiratory neurons (caudal to the preBötC) form clusters with little or no overlap.
Subsets of EBs were active at more than one [K+]o (Figure 3 C). Because elevating [K+]o depolarizes neurons (Del Negro et al., 2001), EBs whose bursting mechanism is voltage-dependent would be expected to increase their bursting frequency with increasing [K+]o. As a consequence, EBs active at more than one [K+]o were classified as voltage-dependent EBs (VDEBs) or voltage-independent EBs (VIEBs), depending on whether burst frequency increased with [K+]o or not. Traces from a VIEB (Figure 5 A.i) and a VDEB (Figure 5 A.ii) both recorded slightly caudal to the VIIn (Figure 5 B) reveal that while in this VIEB, bursting frequency is unaffected by changes in [K+]o, bursting pattern is transformed from sharply rising and steeply decrementing at 3 mM [K+]o to rounded in shape (Figure 5 A.i, top trace vs bottom traces), consistent with near-tonic spiking. By contrast, in the VDEB shown here, while frequency increases with [K+]o, burst envelopes remain steeply rising and decrementing (Figure 5 A.ii). The overall number of VIEBs and VDEBs were roughly equal, both as a fraction of total EBs, and as a fraction of respiration-modulated EBs (Figure 5 C). The proportion of respiration-modulated EBs active at multiple [K+]o (41%; 40/97), was almost double that of all EBs (26%; 181/693).
By combining data across experiments in a dot diagram (Figure 5 D), a qualitative feature that emerges is that there are almost no VIEBs (n=86) caudal to the preBötC, and few VDEBs (n=95) rostral to VIIc. Using dirichlet tiling to detect statistically significant clustering of VDEBs and VIEBs (tiles outlined in black and red, respectively, Figure 5 D), clusters associated with these two cell classes show little overlap: VIEBs cluster as a band along dorsal VIIn running caudally past VIIc, while VDEBs form one cluster just caudal to the VIEB cluster at the level of the VIIn, and another caudal to the preBötC. In addition to these non-overlapping clusters, smaller (n> 7) VIEB and VDEB clusters overlap at the rostral margin of the preBötC.
Because the number of EBs increased significantly with [K+]o, and subsets of them have been found to be respiration-modulated and voltage-dependent, the presence of EBs within respiratory rhythmogenic networks is predicted to give rise to ectopic bursting among respiratory network constituents at elevated [K+]o. To test this conjecture, 7 experiments were carried out in which the synaptically coupled network was recorded at 3 mM, 6 mM, and 9 mM [K+]o. Increases in [K+]o modified both respiratory network activity, and motor output. Respiratory network constituents increased between 3 mM and 9 mM [K+]o, from 20 ± 9 cells at 3 mM, to 23 ± 7 cells at 6 mM, and 35 ± 9 cells at 9 mM (Figure 6 A); repeated measures ANOVA revealed statistically significant differences between 3 mM and 9 mM, and 6 mM and 9 mM (p<0.05). In addition, the period of respiratory motor output decreased with [K+]o (Figure 6 B top, hollow black squares). Period variability within each recording epoch, estimated using the mean of the CVs also increased with [K+]o (Figure 6 bottom, hollow black squares).
Across experiments, a total of 44 cells were active at all 3 [K+]o levels; the activity of these cells was analyzed in greater detail to quantify change in ectopic activity. Using repeated-measures ANOVA, mean cell periods at 6 mM were found to be significantly shorter than at 3 mM and 9 mM (p<0.01; Figure 6 B top, hollow red circles). Period variability (estimated using the mean of CVs under each [K+]o) increased with [K+]o, and the CV at 9 mM was significantly greater than at 3 mM (p=0.02; Figure 6 bottom, hollow red circles). Accurate estimate of respiratory neuron period was complicated by increasing peak height variability with [K+]o, apparent both in the motor output (attenuated bursts indicated by hollow arrows Figure 6 D), and in the associated optical traces (Figure 6 D) obtained from one set of recordings from VIIn (Figure 6 C). As can be seen from the optical traces, although ectopic bursts (solid red arrow-heads Figure 6 D) occur at 3 mM [K+]o, their frequency increases with increasing [K+]o. Burst triggered averages (Figure 6 E) show that the phase of respiratory neuron activity at 3 mM [K+]o (black lines) can be substantially altered at 6 mM (blue lines, Figure 6 E) and at 9 mM (red lines, Figure 6 E); phase of activity can become so variable as to flatten the burst-triggered average (cells 1 and 3), suggesting that in this preparation, 9 mM [K+]o disrupts respiratory rhythmogenic network function. Ectopic bursting at the cellular level is the expected effect of elevated [K+]o on VDEBs, and the increased variability of motor output amplitude and frequency is consistent with these neurons playing a rhythmogenic role (Onimaru and Homma, 2008).
In this study, optical recordings characterized the overlap between respiration-modulated neurons and EBs in ventrolateral medulla. The sample size provides robust support for two main findings: both respiratory networks and EBs form clusters that overlap in regions identified as rhythmogenic; EB numbers show a steep dependence on [K+]o, with relatively few active at physiological (3 mM) [K+]o.
The anatomical distribution of respiratory neurons is consistent with earlier studies. Respiratory neurons were found along the caudal, ventral, and dorsal margins of VIIn. These neurons formed a significant cluster at the VIIc, which overlaps with the putatively rhythmogenic RTN/pFRG. This structure includes neurons developmentally specified by the transcription factor phox2b (Onimaru et al., 2009; Thoby-Brisson et al., 2009), that are distributed along the ventrolateral margin of the VIIn, curving medially at the VIIc, and that have also been shown to play a key role in central chemosensation (Lazarenko et al., 2009; Guyenet and Mulkey, 2010). Likewise, the preBötC has been identified as a 200 μm region centered 500 μm caudal to the VIIc (Ruangkittisakul et al., 2008), which in our data coincides with a statistically significant cluster of respiration-modulated neurons. Finally, a significant cluster of respiration-modulated neurons lying caudal and dorsal to the preBötC was found, consistent with relay neurons in the ventral respiratory group (Dobbins and Feldman, 1994). Although in any given experiment, only neurons at the surface were recorded from, across experiments, the respiratory column was transected at different mediolateral levels, due to experimenter error and/or biological variability. Thus these data aggregate respiratory network activity from a range of mediolateral levels. Further, because putatively rhythmogenic glutamatergic neurons coexpressing the neurokinin-1 receptor and somatostatin within the preBötC are confined to a narrow band of cells in the mediolateral axis (Stornetta et al. 2003), available data suggest that, at least at the level of the preBötC, rhythmogenic networks show little mediolateral dispersion.
Less is known about the distribution of EBs in ventrolateral medulla. Based on single-unit (Onimaru et al., 1989; Johnson et al., 1994; Onimaru et al., 1995; Del Negro et al., 2002a) and optical recordings (Koshiya and Smith, 1999; Koizumi et al., 2008), both the RTN/pFRG and the preBötC have been shown to be rich in EBs. These studies left unresolved the distribution of EBs along ventrolateral medulla. If EBs were found to be uniformly distributed, or if the overlap between EBs and respiratory networks was small, this would undermine the hypothesis that EBs represented functionally specialized respiratory network constituents. The novel, and -- because of the relatively large sample size -- robust finding of this study is that, while both respiration-modulated neurons and EBs are distributed along the length of ventrolateral medulla, statistically significant clusters of these two classes of neurons only overlap at the VIIc, and within the preBötC, both identified loci of respiratory rhythmogenesis. This overlap is also apparent in the high proportion of respiration-modulated EBs to respiration-modulated neurons within the preBötC and in VIIc, consistent with an earlier study (Koizumi et al., 2008). Another feature of respiration-modulated EBs was that the proportion of EBs active at more than one [K+]o was near twice that for EBs as a group. These features are consistent with a functional role for EBs within rhythmogenic networks.
In the transverse slice, respiratory rhythm persists after pharmacological disruption of the biophysical mechanism for bursting prevalent in preBötC neurons in perinatal rodents (Del Negro et al., 2002a; discrepant with Koizumi and Smith, 2008). Based on these findings, the obligatory role of EBs in respiratory rhythmogenesis has been challenged (Feldman and Del Negro, 2006; Del Negro et al., 2008). Evidence that two distinct functional anatomical regions are rhythmogenic (Mellen et al., 2003; Dubreuil et al., 2009; Onimaru et al., 2009) establishes that respiratory rhythm can arise via more than one mechanism. Thus, generically, the observation that a given structure or mechanism is not obligatory for a behavior does not support the inference that the structure or mechanism plays no role in that behavior. A similar point has been made about the role of EBs in respiratory rhythmogenesis (Ramirez et al., 2004). The appearance of ectopic peaks in respiratory neuron traces, together with increased variability in inspiratory burst amplitude and respiratory period at elevated [K+]o under network-intact conditions is consistent with EBs contributing to respiratory rhythmogenesis.
Although the anatomical distribution of EBs is consistent with rhythmogenic function, the observed steep dependence of EB numbers on [K+]o suggests that under physiological conditions, the contribution of EBs is dependent on the level of excitatory drive. The relatively low number of EBs active at physiological [K+]o does not rule out rhythmogenic function however, since network models have shown that EB-driven network bursting can be obtained in a network in which as few as 10% of its constituents are rhythmically active (Butera et al., 1999b). In addition to a possible direct role, the presence of quiescent conditional bursters in respiratory network may contribute to rhythmogenesis at the network level, via nonlinear amplification of synaptic drive via the same biophysical mechanisms that produce bursting at more depolarized membrane potentials (Feldman and Del Negro, 2006), as has been shown in modeling studies (Butera et al., 1999b; Purvis et al., 2007). A corollary to this is that mechanisms for rhythmogenesis may be qualitatively different at physiological [K+]o (3 mM) than at the elevated [K+]o (9 mM) routinely used in vitro, since at elevated [K+]o more conditional bursters will enter into a bursting regime, potentially transforming the contribution of EBs from primarily amplifiers of synaptic drive to sources of rhythmic activation of the networks in which they are embedded.
Despite the overlap between EB and respiratory neuron clusters, most EBs were not respiration-modulated under control conditions. Because descending and afferent inputs have been removed in vitro, and cells at the sliced surface have lost processes and synaptic inputs, active in vitro networks likely represent a fraction of those active in vivo. The significant increase in respiration-modulated neurons as [K+]o was elevated from 3 mM to 9 mM in synaptically coupled networks supports this conjecture. As a consequence, the observation that most EBs lacked respiratory modulation provides only weak support for the inference that they are non-respiratory. More accurate quantification of non-respiratory EBs awaits characterization of respiratory network phenotypes, so that more robust exclusion criteria can be applied.
In the experiments carried out here, [K+]o was manipulated to shift membrane potential, and hence manipulate neuronal excitability. In a single-unit recording study taking a similar approach, EB bursting frequency was found to be steeply dependent on [K+]o, and no VIEBs were described (Del Negro et al., 2001). In this study however, a robust effect of [K+]o on frequency was obtained under conditions of excitatory transmission blockade only. In another study in which both fast excitatory and inhibitory transmission were blocked, EB Vm was found to be insensitive to [K+]o manipulation because of the presence of a background Na+ conductance (Tryba et al., 2003), thus, neurons matching the VDEBs in our study were not encountered. These studies, as well as our own data, suggest that EBs have heterogeneous biophysical properties, a feature which has been shown to extend the dynamic range of rhythmogenic networks (Butera et al., 1999b).
Optical recordings revealed that subsets of EBs remained coupled following synaptic blockade. Such coupling would be expected to persist between gap-junction coupled neurons, and gap junction coupling between respiratory network constituents has both been observed (Rekling et al., 2000; Thoby-Brisson et al., 2009), and would be expected in tissue harvested from neonates (Kandler and Katz, 1995). More likely though, given the large number of metabotropic and/or ionotropic conductances left unblocked using our cocktail, coupling is at least in part mediated by chemical synapses. As with other studies in which synaptic blockade is implemented via by blocking fast excitatory (Koshiya and Smith, 1999; Del Negro et al., 2001; Thoby-Brisson and Ramirez, 2001; Del Negro et al., 2002b; Koizumi and Smith, 2008; Koizumi et al., 2008), or fast excitatory and inhibitory (Tryba et al., 2003; Pena et al., 2004; Del Negro et al., 2005; Zavala-Tecuapetla et al., 2008) transmission, synaptic transmission is more accurately described as attenuated than blocked. Other approaches such as low [Ca2+]o (Onimaru et al., 1989; Johnson et al., 1994; Onimaru et al., 1995; Del Negro et al., 2001; Del Negro et al., 2002a), Cd2+ blockade of Ca2+ channels (Del Negro et al., 2002b; Koizumi and Smith, 2008; Koizumi et al., 2008), or Na+ channel blockade via tetrodotoxin are each problematic because they may disrupt the biophysical mechanisms that give rise to bursting. Another approach, depletion of vesicular stores via vacuolar proton ATPase blockade (Cavelier and Attwell, 2007), effectively blocks synaptic release with less impact on biophysics, but is too slow to be practicable in most cases. Inosofar as non-bursters are included in our EB sample however, the robust chemical coupling that yokes them to EBs may render them functionally equivalent to the EBs that drive them.
Here, evidence has been presented suggesting that EBs are constituents of respiratory rhythmogenic networks at the level of the RTN/pFRG, and preBötC. Characterizing the role or roles of EBs in respiratory rhythmogenesis is challenging, because anatomically distributed heterogenous networks in ventrolateral medulla interact to generate respiratory rhythm and regulate it from breath to breath. Because optical recording methods offer access to portions of these networks in parallel, they provide a powerful tool for elucidating cellular and synaptic mechanisms underpinning this rich and complex behavior.
This work was supported by National Institutes of Health grant HL06800. Thanks to Dr. Consuelo Morgado-Valle for comments on this manuscript.