Modeling synaptic interactions between mammalian breathing and swallowing central pattern generators

Experimental preparation All procedures were approved by the Animal Ethics Committee of the Florey Department of Neuroscience and Mental Health, University of Melbourne, and complied with institutional guidelines. Experiments were performed in juvenile Sprague–Dawley rats of either sex (postnatal days 17–21) using the in situ arterially perfused brainstem preparation. Animals were anesthetized with 2–5% isoflurane to a surgical plane, transected below the diaphragm, and transferred to ice-cold artificial cerebrospinal fluid (aCSF) for decerebration. The aCSF contained (in mM): 125 sodium chloride, 3 potassium chloride, 1.25 potassium dihydrogen phosphate, 1.25 magnesium sulfate, 25 sodium bicarbonate, 2.5 calcium chloride, 10 D-glucose, and 1.25% Ficoll. The heart and lungs were removed, the phrenic nerve was isolated for recording, and the descending aorta was prepared for cannulation. The cerebellum was removed, and the vagus and superior laryngeal nerves (SLN) were isolated for recording and stimulation. The preparation was placed in a recording chamber and reperfused via the aorta with carbogenated (95% oxygen, 5% carbon dioxide) aCSF at 31 °C using a peristaltic pump. Phrenic and vagal nerves were recorded with suction electrodes. Signals were amplified 400 times, band-pass filtered between 1 and 7500 Hz, digitized at 30 kHz using an Intan headstage, and acquired with Open Ephys. Respiratory activity resumed within minutes. A bolus of sodium cyanide (0.1 mL, 0.1% w/v) was applied to convert apneustic activity into a stable three-phase eupneic rhythm. Perfusion flow was adjusted to maintain stable rhythmic activity. To evoke swallows, the SLN was stimulated. Swallowing threshold was determined using 10-second stimulus trains (20 Hz, 0.1 ms pulse width) at increasing currents (10–80 microamperes). Sequential swallowing was assessed at twice threshold. For phase-resetting experiments, single swallows (n = 100) were evoked at random intervals of 20–25 seconds using 20 Hz stimulation (250 ms burst duration, 0.1 ms pulse width, twice threshold), ensuring coverage of all respiratory phases. Data analysis Data processing was performed in Python using SciPy. Signals were mean-subtracted and band-pass filtered with a second-order Butterworth filter (300–3000 Hz), applied forward and backward to avoid phase distortion. SLN stimulus artifacts were removed by linear interpolation around stimulus times. Signals were rectified and low-pass filtered (12 Hz) to obtain smooth respiratory envelopes. The processed signals were downsampled by a factor of 100 using successive decimation steps, resulting in a final sampling rate of 300 Hz. To compute the phase response curve (PRC), continuous recordings were segmented into trials containing a single SLN stimulus. The phrenic nerve activity was further low-pass filtered at 1 Hz to retain the fundamental respiratory rhythm (typically 0.2–1 Hz). The Hilbert transform was applied to obtain the analytic signal and an initial phase estimate (protophase). After subtracting the mean of the analytic signal, a refined phase estimate was computed following standard phase-reconstruction procedures. The baseline respiratory frequency was estimated from a linear fit to the phase before stimulation. Linear regression was performed separately on phase before and after the stimulus while keeping the baseline frequency fixed. The phase shift was defined as the difference between the intercepts of these two fits. This yielded the phase response curve describing the dependence of respiratory phase shift on the timing of SLN stimulation.