Article Figures & Data
Figures
- Figure 1.
One- and two-neuron hypotheses. A , Event A is a complex spike typically observed in the MNTB, composed of A1 generated by synaptic transmission at the calyx of Held and A2, which is the action potential of the MNTB neuron. Event B is sometimes observed in the same recordings as event A and resembles A1 in shape. B , The one-neuron hypothesis proposes that events A and B originate from a single “complex unit” (i.e., a calyx of Held and MNTB neuron). In this case, events A1 and B should always be separated by a time equal to or greater than the RP. C , The two-neuron hypothesis proposes that events A and B originate from two different neurons: one complex unit giving rise to event A and a second structure in the vicinity of the recording electrode giving rise to event B. Here, events A1 and B can occur closer together than the refractory period and can overlap (“superposition”). The dashed lines indicate the trigger level use in the triggering algorithm. The long and short dead times used in the triggering algorithm are also illustrated: they are limited by the duration of events A and B, respectively.
- Figure 2.
Micropipette recordings from 39 axons in the trapezoid body, identified as globular bushy by a “primary-like with notch” poststimulus time histogram to short tones at CF and monaural responsiveness to the ipsilateral ear. The data shown were obtained at the SPL generating the shortest ISI for the given neuron. A , ISI histograms normalized to the maximum number of intervals: a single histogram is shown for every globular bushy cell. B , C , Minimum interspike interval ( B ) and spike rate ( C ) as a function of CF, at the SPL generating the smallest ISI. One symbol per neuron is shown. The dashed lines in A and B indicate the smallest ISI encountered in the entire sample (0.68 ms). The bin size of the ISI histograms was 0.1 ms.
- Figure 3.
Modeling: proof of principle. A , Rate curves for two separately recorded GB axons with overlapping, but different, frequency tuning. B , C , ISI histograms for all spike times of GB axons 1 and 2. D , ISI histogram for merged spike times of GB axons 1 and 2. E–G , Same analysis on synthetic neural waveform, obtained by assigning event A to all spike times of GB axon 1 and event B to all spike times of GB axon 2. Triggering and spike-sorting algorithms were applied to the synthetic waveform, from which spike times were derived for events A and B. E , F , ISI histograms of all spike times of event A and B. G , ISI histogram for merged spike times of events A and B.
- Figure 4.
ISI analysis reveals intervals incompatible with the single-neuron hypothesis. A , After triggering and spike sorting, all events A are grouped together. B , In this recording, a second event B was also identified, which resembles event A1 in shape. C , Spike rate against stimulus frequency for both events A and B. The inset shows the threshold curve with CF and SR, measured before spike sorting. D , E , ISI histograms for all events A1 and B. F , Two examples in which event B precedes event A1 by less than the RP, indicating that events A and B originate from two separate neurons. G , ISI histogram for combined events A1 and B.
- Figure 5.
Second example of the ISI analysis. For the explanation, see the corresponding panels in Figure 4.
- Figure 6.
The spike-sorting algorithm misses event B when it overlaps with event A. A – C , Event “B” (manually identified) occurs just before event A1 ( A ) or between events A1 and A2 ( B , C ). In all cases, the sequence of events is treated as a single complex spike.
- Figure 7.
Peak aligning and cross-correlation methods. A , One stimulus group of complex spikes. The spikes are triggered on the A1 peak, ensuring that the A1 peaks are time aligned. The time between the A1 and the A2 peaks (t A1A2) differs slightly between spikes. Therefore, the A2 peaks are not time aligned. B , Peak-aligned spikes. All spikes are resampled so that t A1A2 is equal to the mean t A1A2 of the non-peak-aligned spikes. C , The solid line shows mean spike of the peak-aligned spikes. The dashed line shows mean spike of the non-peak-aligned spikes. D , E , To improve peak timing, we upsampled and cross-correlated the A1 and A2 peaks of the mean spike (shown in C , redrawn as shaded profiles in D ) with the A1 and A2 peak of each upsampled, non-peak-aligned, spike.
- Figure 8.
Event B was not encountered in recordings with high SNR. SNR is plotted against the percentage of events B for each recording (176 recordings). The percentage of events B is calculated as the number of events B divided by the number of events A plus the number of events B. Recordings in which the combined ISI analysis showed that events A1 and B occurred closer together than the refractory period are shown as crosses; the others as circles. The dashed line is at an SNR of 20.
- Figure 9.
Variation in intraspike intervals. A , The thin line shows mean time (left ordinate) between peak A1 and A2 (t A1A2) and its SD (error bars). The thick line shows spike rate (right ordinate). Top left inset, Linear regression analysis was used to quantify the correlation between t A1A2 and spike rate. The solid line shows the line of best fit. Top right inset, Threshold curve showing CF and SR measured before spike sorting. B , All spikes per stimulus for a select number of stimulus frequencies. C , ISI plotted against t A1A2 for the same stimulus frequencies. D , Spike train recorded to one repetition of the 760 Hz stimulus, with indication of events A1 and A2. The stimulus is superimposed.
- Figure 10.
Second example of variation in intraspike intervals. A , The thin line shows mean time and SD (error bars) of t A1A2. The thick line shows spike rate. Top right inset, Linear regression analysis to quantify the correlation between t A1A2 and spike rate. The solid line shows the line of best fit. Top left inset, Threshold curve showing CF and SR measured before spike sorting. B , Mean spike per stimulus for a select number of stimulus frequencies.
- Figure 11.
At the population level, intraspike interval is positively correlated with spike rate. Linear regression was used to determine the correlation between spike rate and t A1A2 for each recording. The slope of this relationship is plotted against SNR. Each point represents the correlation analysis for one recording, across all stimulus conditions. Significant correlations (p > 0.05; r 2 > 0.25) are shown with a cross, and others are shown with a circle.
- Figure 12.
Variation in peak amplitude. A , The thin line (left ordinate) shows mean A2 peak amplitude and its SD (error bars). The thick line (right ordinate) shows spike rate. Top left inset, Linear regression analysis was used to quantify the correlation between A2 peak amplitude and spike rate. The solid line shows the line of best fit. Top middle inset, Mean spikes per stimulus aligned on the A1 peak. Top right inset, Threshold curve showing CF and SR measured before spike sorting. B , Mean spike per stimulus for each stimulus frequency. C , D , A2 peak amplitude and intraspike interval t A1A2 plotted against ISI for each stimulus frequency. E , Spike train recorded to a segment of the 11 kHz stimulus. The decrease, at short ISIs, of the amplitude of component A2 (but not A2) is clearly visible.
- Figure 13.
At the population level, A2 peak amplitude is negatively correlated with spike rate. Linear regression was to determine the correlation between spike rate and absolute A2-peak for each recording. The slope of this relationship is plotted against recording SNR. Each point represents the correlation analysis for one recording, across all stimulus conditions. Significant correlations (p > 0.05; r 2 > 0.25) are shown with a cross, and others are shown with a circle.
