Surgery and recording
All experiments were conducted in accordance with local and federal guidelines for the care and use of laboratory animals. Adult male cats (2.6–4.2 kg) were anaesthetized with pentobarbitone sodium (35 mg kg−1i.p.). Anaesthetic depth was judged from the absence of limb withdrawal reflexes and from mean arterial blood pressure, which was monitored via a cannula in the carotid artery. Standard Ringer solution and supplementary pentobarbitone were administered via a second cannula in the external jugular vein. A cannula was placed in the trachea to facilitate artificial ventilation. Throughout surgery and subsequent recording, blood pressure was maintained > 60 mmHg by administering fluid i.v., body temperature was maintained between 35 and 37°C using radiant heat, and end-tidal PCO2 was maintained around 3.5 % by adjusting tidal volume. At the end of the recording session, the animal was killed using an overdose of pentobarbitone.
A laminectomy was performed from segments L7 to L4, exposing spinal roots carrying axons supplying the triceps surae. The left hindlimb was completely denervated with the exception of the nerve supplying the medial gastrocnemius muscle. The medial gastrocnemius was freed of its insertion and surrounding tissues, and a cord was tied to its tendon. The animal was mounted in a rigid recording apparatus, and skin flaps were fashioned to hold mineral oil pools bathing the spinal cord and left leg. The medial gastrocnemius (MG) nerve was mounted on a hook electrode to permit stimulation.
Dorsal root filaments from segments L7 and S1 were tested for the presence of stretch-sensitive afferents by elevating them on bipolar hook electrodes and monitoring electrical activity in them while gently pulling on a cord attached to the medial gastrocnemius tendon. Once a filament with a significant number of afferents was found, all other dorsal root filaments from S1 to L6 were cut in order to minimize synaptic noise in motoneurones. For most recordings, the animal was paralysed with gallamine triethiodide (Flaxedil), and artificially ventilated. During the epochs of paralysis, adequacy of anaesthesia was assured by monitoring end-tidal PCO2 and blood pressure; anaesthetic level was adjusted so that mean levels of these measures were kept near pre-paralysis values, and transient changes in response to paw pinch or electrical stimulation of muscle nerves were absent.
Two microelectrodes (filled with 2 m potassium acetate) were employed. The first (10–20 MΩ) was driven into medial gastrocnemius group I a afferent axons (identified by orthodromic stimulation latency and high sensitivity to stretch) in the dorsal root filament. This electrode both recorded spontaneous action potentials and delivered depolarizing pulse trains that elicited action potentials. The second electrode (5–10 MΩ) was driven into antidromically-identified motoneurones (supplying the medial gastrocnemius or synergists) in the spinal cord. This electrode was used both to record membrane potential at rest (for measurement of I a-triggered EPSPs) and to pass steady depolarizing current sufficient to cause the motoneurone to fire at a steady average rate, usually 10–15 spikes s−1 Data were accepted for further analysis when the amplitude of action potentials in the motoneurone exceeded 65 mV.
Data acquisition and analysis
Once a I a afferent and a motoneurone were simultaneously isolated, they were tested for a synaptic connection by spike-triggered averaging (STA) the motoneurone membrane potential (sampled at 100–133 kHz). If the averaged EPSP was ca. 100 μV or greater, steady depolarizing current was injected into the motoneurone to induce it to fire. Trains of brief (0.5 ms) depolarizing pulses were injected into the I a axon at stimulation frequencies (20–60 or 50–100 pulses s−1) below the range where temporal summation was expected to be significant (Honig et al. 1983). At least 2000 stimuli were delivered to the I a axon, and 400–2500 action potentials were recorded from the motoneurone for each frequency. For seven cases in which it was permitted by stable conditions, additional STA EPSPs were recorded after the period of motoneurone firing to test for stationarity. All trials involving motoneurone activation were recorded on magnetic tape (DC, −11 kHz) for subsequent analysis.
From real-time or tape-recorded trials, the times of motoneurone spikes as well as of spontaneous or driven I a spikes were recorded after band-pass filtering the signals and presenting them to Schmidt triggers coupled to TTL pulse generators. The times of onset of the TTL pulses were recorded to the nearest 10 μs.
For each bout of I a activation, a histogram (cross-correlogram or peristimulus time histogram (PSTH), depending on whether the I a activity was spontaneous or driven) describing the times of firing of the motoneurone with respect to the I a was constructed using 100 μs bins, as described in Cope et al. (1987) (cf. Moore et al. 1966). The cusum (Ellaway, 1978) was constructed in order to facilitate identification of the limits of the primary correlogram peak, which is associated with the increase in firing probability caused by the I a EPSP (Moore et al. 1970; Kirkwood, 1979). The mean percentage increase (MPI) of the primary correlogram peak above the pre-trigger baseline was computed as described by Cope et al. (1987).
Following Cope et al. (1987) and Garnett & Stephens (1980), the significance of the primary correlogram peak was first assessed from the z-statistic (Cox & Lewis, 1966) for comparison of two Poisson processes:
where n1 and n2 are the numbers of events in two histogram bins and t1 and t2 are the associated bin durations. In our application, the two bins correspond to the duration of the correlogram peak and a 5 ms pre-trigger baseline interval. For two independent Poisson processes, z follows a standard normal distribution (Cox & Lewis, 1966), so the primary correlogram peak could be considered significant at the 5 % level if |z| > 1.96 (two-tailed test).
In addition, because neither the trains of stimuli nor the trains of motoneurone action potentials fit the assumptions of a Poisson process (Perkel et al. 1967), significance of the correlogram peaks was assessed using the non-parametric bootstrap technique (Efron & Tibshirani, 1993), which sidesteps assumptions about the statistical distributions involved. Using the BCα method of Efron & Tibshirani, 95 % confidence intervals for primary correlogram peaks were computed from the distributions of 1000 bootstrap estimates drawn from the samples of first cross-intervals between I a and motoneurone firing times. A peak was considered significant if its 95 % confidence interval did not span 0 impulses s−1. Confidence intervals estimated from the bootstrap agreed well with those derived from the z-statistic; the size of the lower confidence interval showed only a 1 % to 12 % difference (either positive or negative) for the two methods, and enlarging the bootstrap sample to 10 000 for a subset of the trials decreased that discrepancy to a maximum of 8 %. For the relatively strong I a-motoneurone synapses evaluated in this study, the two methods led to the same conclusions as to the significance of all twenty-six correlograms tested. Thus it appears that the normal approximation given by Cox & Lewis (1966) would be adequate, despite the regularity of the spike trains studied.
The bootstrap was also used in the analysis of MPI statistics. First, BCα 95 % confidence intervals were constructed for each MPI from the sample of waiting times. An MPI was considered significant if its confidence interval did not span 0 %. Second, MPIs from a single synapse stimulated at different rates were compared by evaluating the confidence interval for the difference between MPI values at successive test frequencies, again derived from a bootstrapped sample.
Frequency-dependent changes in EPSPs were estimated from tape records of the membrane potentials of motoneurones firing repetitively. As described by Cope et al. (1987), we performed STA of motoneurone membrane potential using as triggers only those I a spikes that occurred within a fixed interval (in this case, 10–14 ms) prior to a motoneurone spike (the ‘ramp EPSPs’ of Cope et al. 1987). These triggers were selected using an analog delay line (CWE ADL-832; DC −10 kHz) and custom circuitry that generated TTL-level trigger pulses when the timing criteria were met. Motoneurone potential was also delayed (10 ms more than the I a triggering train) prior to sampling (12-bit, 100 kHz) to permit examination of the pre-trigger ramp. In most cases, at least fifty sweeps were averaged to estimate the ramp EPSP. Ramp EPSP amplitude was measured after subtracting a straight line fit to 3–5 ms of the pre-trigger baseline of each average.