INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

A major goal of neuroscience, and in particular of system neuroscience, is the understanding of how the nervous system processes sensory information and translates sensory inputs into actions or a behavior. As widely recognized (Bialek and Rieke 1992; Gerstner et al. 1997; Mainen and Sejnowski 1995; Shadlen and Newsome 1998; Stevens and Zador 1998), the issue of reproducibility is essential in understanding the core of the nervous code. In other words, it is essential to identify which features of the neural activity, underlying a given reproducible action or behavior, are reproducible. These features could be the timing of a specific action potential or the spike train of a given neuron or some average quantity of the population activity. An adequate experimental analysis of this issue requires the simultaneous measurement of behavior, possibly in a quantitative way, and of the electrical activity of neurons underlying this behavior. An exhaustive analysis cannot be obtained in mammalian nervous systems where it is possible to monitor the electrical activity of only an infinitesimal fraction of neurons involved in any significant reaction or behavior.

Simple nervous systems, such as those of invertebrates, may be more suitable for such an analysis. A thorough experimental analysis of a given behavior and of the underlying electrical activity can be obtained in simpler nervous system, such as the Aplysia (Byrne et al. 1974; Castellucci and Kandel 1974; Frost and Kandell 1995; Tsau et al. 1994) and the leech (Baader 1997; Kristan 1982; Nicholls and Baylor 1968; Stent et al. 1978; Stuart 1970; Wittenberg and Kristan 1992a). These nervous systems are "solvable" in the sense that it is possible to quantify the behavior while recording the electrical activity of a significant fraction of neurons involved in the behavior. As neurons, synapses, and sensory receptors in mammalian and invertebrate nervous systems have very similar properties, the understanding of information and parallel processing in these simple nervous systems provides a solid basis for unraveling how the brain of higher animals works.

The leech exhibits a limited set of repeatable behaviors: in particular, when a strong mechanical stimulus is delivered to its head, it withdraws and rapidly shortens (Kristan and Nusbaum 1982; Magni and Pellegrino 1978; Shaw and Kristan 1995, 1997, 1999; Wittenberg and Kristan 1992a,b) to escape from potential danger. The shortening reaction causes the simultaneous contraction of all or most of its body, involving the entire CNS. The nervous system of the leech Hirudo medicinalis is composed of a chain of 21 highly stereotyped ganglia, consisting of about 400 neurons each (Macagno 1981; Muller et al. 1981; Nicholls and Baylor 1968; Yau 1976). Ganglia are linked by the connective nerve fibers, the largest axon bundles of the body, which transmit electric signals along the chain (Muller et al. 1981). From each leech ganglion two pairson the right and on the leftof axon bundles emerge, usually referred to as the anterior and posterior roots. Axons of mechanosensory neurons and of motoneurons form these roots, and it is possible to obtain clear extracellular recordings of their action potentials from bifurcations of these roots (Ort et al. 1974; Pinato et al. 2000). The anterior root bifurcates into an anterior anterior (AA) root and a medial anterior root (MA), while the posterior root bifurcates into a posterior posterior (PP) root and a dorsal posterior (DP) root. The largest extracellular voltage signals recorded from the AA root are usually produced by action potentials of the two longitudinal motoneurons, 107 and 108 (Ort et al. 1974), while those from the MA root are produced by action potentials of the dorsoventral motoneuron 109 and the circular ventral motoneuron (CV) (Baader 1997; Ort et al. 1974). Action potentials of the longitudinal motoneurons L and 3 can be detected on extracellular recordings from the DP root, while those of longitudinal motoneurons 4, 5, and 8 are detected on recordings from the PP root.

The shortening reaction is mediated by the simultaneous activation of excitatory motoneurons innervating longitudinal muscles. The following motoneurons, representative of the different classes of longitudinal muscle excitors, have been tested during whole-body shortening and have been found to contribute to the reflex: motoneuron 3 (a dorsal excitator), motoneurons 4 and 108 (ventral excitators), and motoneuron L (excitator of all the longitudinal muscles in a half segment) (Shaw and Kristan 1995; Wittenberg and Kristan 1992a). The longitudinal motoneurons are expected to act as coherent functional groups in the production of a behavior [for example, this is the case found in swimming (Ort et al. 1974)]. For this reason, many other dorsal and ventral longitudinal excitors are expected to contribute to the whole-body shortening, but until now their involvement in the reflex has not been proofed and their concerted firing activity has never been investigated by simultaneous parallel recordings. On the other side, those motoneurons that are innervating antagonistic fibers of longitudinal muscles, are not supposed to be activated during whole-body shortening. For instance, the CV motoneuron is known to be silent during this reaction (Baader 1997).

The longitudinal motoneurons are activated through the S cell network and another parallel interneuronal pathway running through the connective fibers (Shaw and Kristan 1999). The S cell network consists of a chain of electrically coupled neurons, providing a fast signaling pathway; this network, however, is neither sufficient nor necessary for the activation of whole-body shortening (Shaw and Kristan 1999), which is triggered by chemical polysynaptic pathways.

The purpose of this work is to determine the pattern of firing of action potentials of motoneurons activated during whole-body shortening. In particular we want to answer these questions: how many motoneurons are activated during whole-body shortening in the leech; how concerted is the electrical activity of these motoneurons; does the underlying pattern or sequence of neuronal activity have a given structure; and what is the origin of the reliability of this essential and vital behavior? Some of these issues have been successfully addressed in the Aplysia by the Cohen group (Tsau et al. 1994; Wu et al. 1994).

By using simultaneous extracellular recordings from fine bifurcations of the roots, we monitored the firing pattern of a significant fraction of all leech motoneurons, in particular, those underlying whole-body shortening. By using video microscopy, we obtained a quantitative characterization of this simple behavior and quantified its reproducibility. Therefore our manuscript provides a simultaneous measurement of the electrical activity of at least one-third of all motoneurons involved in that behavior and a quantitative comparison of the reproducibility of firing of a single motoneuron and of the motoneuron population and then relates these results to the reproducibility of the behavior. These results indicate that whole-body shortening is mediated by the coactivation of a significant fraction of all leech motoneurons, thus indicating that this escape reaction is a widely distributed process. Our results also show that the great majority of coactivated motoneurons fire action potentials with a very low pairwise correlation. Indeed they fire action potentials in an almost statistically independent way. Our hypothesis is that the behavioral response becomes smooth and reproducible as a consequence of statistical independence, that these properties are a common feature of coactivated neurons, and that they are a typical pattern of parallel processing in the nervous system.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Preparation

Leeches H. medicinalis were obtained from Ricarimpex (Eysines, France) and kept at 5°C in tap water dechlorinated by aeration for 24 h. A semi-intact preparation was used consisting of a whole leech with one or three ganglia (usually the 10th-12th ganglia) isolated from the corresponding body wall (see Fig. 1A). The central part of the animal was held fixed and pinned down, but the head and tail were left free to move. In other experiments, a different preparation was used to monitor simultaneously whole-body shortening and deformations of a piece of skin innervated by only one ganglion: a piece of skin corresponding to a whole segment (see Fig. 1, B and C) or a half segment was isolated from the rest of the body but kept innervated by its ganglion. This piece of skin was fixed to the bottom of the recording chamber at the edges only so as to allow it to deform during muscle contraction.

View larger version (108K): [in this window] [in a new window]   Fig. 1. The preparation and the recording electrodes. A: 8 suction pipettes surrounding the 10th ganglion of a semi-intact leech were used to record the electrical activity from 8 fine branches emerging from the ganglion. B and C: in each of these panels, a picture of a semi-intact leech (on the right) and of a detail of its skin (on the left) are shown before (B) and at the end (C) of whole-body shortening. The semi-intact leech was imaged with a 1st miniaturized CCD camera viewing the recording apparatus (right image in each panel). The head and tail of the leech were kept intact and the 9th and 10th ganglia were isolated, while the skin of the 11th body segment was flattened to be monitored by another CCD camera mounted on a dissecting microscope viewing the skin (left image in each panel). The 2 CCD cameras were synchronized and their images were combined into a single image (see METHODS), as shown in B and C. The little yellow rectangle on the right of B indicates the flattened segment shown at higher magnification on the left of the same panel (surrounded by a big yellow rectangle) and on the left of C. The 2 white wires connected to the head are used to provide the electrical stimulation (see METHODS). The yellow dotted circle on the left of B indicates the point of the skin whose displacement, indicated by the yellow arrow on the left of C, has been analyzed in Fig. 2, B and D. The red dotted circle on the right of B indicates the head of the leech whose displacement, indicated by the red arrow on the right of C, has been analyzed in Fig. 2, A and C.

The roots emerging from ganglia were cleaned for recording with suction pipettes. The preparation was pinned in a silicone elastomer (Sylgard)-coated dish at room temperature (20-24°C). The tail of the leech was left intact and free to move. The head was either free to move or removed, or the head sucker was closed with a surgical thread. During dissection, the preparation was bathed in a Ringer solution with the following composition (in mM): 115.0 NaCl, 1.8 CaCl₂, 4.0 KCl, 12.0 glucose, and 10 mM Tris maleate buffered to pH 7.4 with NaOH (Muller et al. 1981).

Imaging

Two CCD cameras were used for monitoring whole-body shortening. One standard CCD camera mounted on a dissecting microscope was used to measure deformations of a piece of skin, induced by muscle contraction during whole body shortening or by direct stimulation of motoneurons impaled with a sharp intracellular microelectrode. The second miniaturized camera (Teli CS3500C) was mounted on a micromanipulator at a distance of about 10 cm from the entire leech. This camera was used to quantify whole-body shortening. Images from the two cameras were synchronized and combined in a single image by a Video Screen Splitter Model 613 (Colorado Video). Images were acquired at 8.3 Hz and stored on a PC using a frame grabber DT3155 (Data Translation) and the Axon Imaging Workbench 2.2 acquisition software (Axon Instruments). Displacements (arrows in Fig. 1C) of selected features (dotted circle in Fig. 1B) in images were obtained by correlation based algorithms (Aggarwal and Nandhakumar 1988).

Electrical recordings

Eight suction pipettes were used for extracellular recordings. Four pipettes were used to record the electrical activity from the right or left roots. In some experiments, four pipettes were used to record the electrical activity from the left and right AA and MA roots and from the left and right bifurcations (DP:B1 and DP:B2) of the dorsal posterior roots (Baader 1997; Ort et al. 1974) of the same ganglion. In some experiments, eight suction pipettes were used to record the electrical activity of the left and right AA and MA roots (or the DP:B1 and DP:B2 roots) of two adjacent ganglia. Often one suction pipette was used for recording en passant from the anterior connective entering into the ganglion under investigation. In other experiments, the left and right PP roots were sucked into suction pipettes and their electrical activity recorded. The electrical activity of motoneurons was monitored by intracellular recordings with sharp electrodes (input resistance, 30 M filled with 4 M potassium acetate) using Axoclamp-2b amplifiers (Axon Instruments, Foster City, CA). The extracellular voltage signals were recorded with standard analog amplifiers with a gain of 2 × 10⁴ and a bandwidth of 200-3,000 Hz. Voltage recordings were digitized at 10 kHz, stored on a personal computer, and analyzed with the program Clampex8 (Axon Instruments). Axon Imaging Workbench and Clampex 8 can run simultaneously on the same computer, allowing the synchronized acquisition of images and electrical signals. Voltage signals, either intracellular or extracellular, were also stored on an eight-channel digital audio recorder (DA-88 TASCAM).

Whole-body shortening was initiated by an electrical pulse of 0.8-1.2 mA, delivered to a platinum wire sutured tightly on the skin of the animal near the third segment as described in Shaw and Kristan (1995). The pulse lasted 1 s and was delivered at least every 3 min. We used the lowest possible stimulus amplitude able to elicit the shortening contraction. Similar results were obtained when trains of brief voltage pulses lasting 1 s were used. (In experiments on leeches with intact head and tail, the animal occasionally attached to the bottom of the perfusing chamber with its head sucker and in these cases electric stimulation often failed to elicit whole-body shortening. The behavioral response was more reproducible when the head sucker was removed or sealed with a surgical thread.)

Neuron identification

For each motoneuron, action potentials were evoked by passing a depolarizing current pulse through an intracellular microelectrode or by the injury discharge caused by cell penetration. With this procedure, it was possible to obtain a clear signature of extracellular voltage signals evoked by action potentials of many motoneurons. For example, an action potential in motoneuron L is associated with two large extracellular voltage signals on both the DP:B1 and DP:B2 roots. An action potential of motoneuron 3 causes one large extracellular signal on the DP:B2 root but occasionally also a smaller signal on the DP:B1 (Ort et al. 1974; Shaw and Kristan 1995). An action potential of motoneuron 109 is associated with a large extracellular voltage signal on the MA root with a much smaller signal on the AA root (Baader 1997; Ort et al. 1974). The firing of motoneurons 107 and 108 can be recognized by large voltage signals on the AA root (Baader 1997; Ort et al. 1974). Assignment of the largest extracellular voltage signal on the AA root to motoneurons 107 or 108 was achieved by intracellular recording. Action potentials from motoneurons 4, 5, and 8 were identified in extracellular recordings from the PP roots (Ort et al. 1974). Action potentials of motoneuron 8 always had the largest extracellular voltage signals, followed by motoneurons 5 and 4. These criteria for neuron identification were confirmed by simultaneous electrical recordings from the roots and by videomicroscopy of the skin contraction. This was done in preparations consisting of an isolated ganglion connected to the skin of half of its body segment. This preparation was also used to check the relative effects of the activation of identified motoneurons on muscle contraction during whole-body shortening. Muscle contraction was monitored by analyzing displacements of selected points on the skin.

When a tight seal was obtained, very large extracellular voltage signals, up to 1 mV, could be recorded from these fine roots. Different shapes of action potentials were classified as described in Pinato et al. (2000), and pairs or triplets of extracellular action potentials were detected. On the basis of the intracellular recordings made on the same preparation and of the appearance of extracellular signals on given roots, the different shapes of extracellular voltage signals were classified as originating from the left and right motoneurons L, 3, 4, 5, 8, 107, 108, and 109. This procedure has been applied to identify motoneurons in all the preparations used to collect the data reported in the present work.

Data analysis

The variability of action potentials identified as originating from the same neuron N was characterized by computing the coefficient of variation CV of the firing rate of the neuron. It is defined as CV = /average firing rate (AFR), where  is the standard deviation of the firing rate and AFR is the average firing rate of the neuron, over the number of trials of a given stimulation, in a given time window t. When the electrical activity of several neurons i = 1, ... , n was analyzed, the CV of the random vector X = (x₁, x₂, ... , x_n), with x_i being the firing rate of neuron N_i, was considered. In this case, the covariance matrix  replaced the unidimensional variance ² and the CV was defined as CV_X(t) =  where AFR_i is the average firing rate of neuron N_i. When the random variables x_i are statistically independent, the covariance matrix  is diagonal and the CV reduces to CV_X(t) = /_iAFR_i(t). In this case (Pinato et al. 1999), the CV of random vector X is equal to the CV of the random variable X = x₁ + x₂ + ··· + x_n. The CV of the ensemble of motoneurons in Fig. 10 was computed as the CV of the random variable X = x₁ + x₂ + ··· + x_n. To quantify the degree of correlation of two neurons, the cross-correlogram was computed, reporting in an histogram the differences between their spiking times as described in Brivanlou et al. (1998).

Whole-body shortening was quantified by computing the relative shortening S, defined as the ratio between the shortening of the anterior part of the leech and its initial length, both measured on the image plane. The variability of the evoked behavioral response was characterized by computing the coefficient of variation CV_S of the relative shortening S. It is defined as CV_S(t) = _S(t)/M_S(t), where _S is the standard deviation and M_S is the average value of S over the number of trials of a given stimulation, in a time window t, equal to the sampling period of image acquisition (see preceding text). The dependence of the CV on the time window t is analyzed in Fig. 5.

A similar approach was used to quantify the deformation of a piece of the leech skin and its reproducibility over the series of trials. In this case, the total displacement D of a selected landmark on the leech skin was measured on the image plan, and its coefficient of variation CV_D(t) = _D(t)/M_D(t) was evaluated. Here _D is the standard deviation and M_D is the average value of D over the number of trials of a given stimulation, in a given time window t, equal to the sampling period of the image acquisition.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Whole-body shortening was studied in a semi-intact preparation where the behavioral response was monitored with a CCD camera and the electrical activity measured with either suction or sharp intracellular electrodes (see Fig. 1A). To monitor also the local muscle contraction, a piece of the central body segment was isolated, flattened and kept innervated by a single ganglion (see Fig. 1, B and C), and its skin deformation was observed with another CCD camera.

Whole-body shortening

Whole-body shortening in a semi-intact leech (see Fig. 1, B and C) was elicited with a current pulse of 1 mA delivered to the head, lasting 1 s. As a consequence of the electric stimulation, the leech shortened its whole body. Whole-body shortening disappeared within 1 or 2 min after the cessation of the electric stimulation and the animal returned to its normal length. In some experiments, immediately after the execution of the shortening reflex, the animal started a twisting or writhing movement, a behavioral response known to be induced by noxious stimulations (Kristan et al. 1982) and sometime paired to whole-body shortening (Weston et al. 1984). Maximal shortening was achieved with voltage pulses lasting at least 0.5 s. The reproducibility of this escape behavior was quantified in different trials by analyzing the relative shortening S of the anterior part of the leech (see Fig. 2A) and the total displacement D (see Fig. 2B) of a piece of the skin. The shortening S and the displacement D are shown in Fig. 1C by a red and a yellow arrow, respectively. Shortening started within approximately 400 ms from the onset of the voltage pulse, and within 1 or 2 s the head and tail reached their final position rather reproducibly. The average shortening was about 35% (see continuous line in Fig. 2C) and the average contraction of the analyzed piece of skin was 13%, corresponding to about 30 pixels (see continuous line in Fig. 2D). The head, tail, and central piece of skin moved simultaneously within 120 ms, i.e., the image acquisition time. The skin of the isolated central body segment contracted longitudinally in an almost uniform way. After an initial increase, the CV of whole-body shortening and of skin displacement became less than 0.3 at maximal whole-body shortening. The low values of the CV indicate that this escape reaction is significantly reproducible.

View larger version (27K): [in this window] [in a new window]   Fig. 2. Reproducibility of whole-body shortening. A: shortening of the anterior part of the leech, relative to the initial length (see text), as a function of time for 8 different trials. B: total displacement in pixels of the point indicated by the dotted circle in Fig. 1B, for the same 8 trials as in A. C: average relative shortening (continuous line) and its coefficient of variation (dotted line) as a function of time (computed by the data shown in A). D: average total displacement of the point analyzed in B (continuous line) and its coefficient of variation (dotted line) as a function of time (computed by the data shown in B). Images were acquired at 8.3 Hz. The top trace in A and B indicates the artifact of electrical shock, representing the electrical stimulation.

Electrical activity during whole-body shortening

Figure 3A illustrates extracellular recordings obtained with eight suction pipettes from the left and right MA, AA, DP:B1, and DP:B2 of the 10th ganglion of a semi-intact leech during the electric stimulation (see top trace). This stimulation induced a vigorous electrical discharge with a delay of about 70 ms well before the onset of the behavioral response (see inset).

View larger version (28K): [in this window] [in a new window]   Fig. 3. Electrical recordings during whole-body shortening. A: electrical recordings obtained with 8 suction pipettes from the roots of the 10th ganglion of a semi-intact leech as shown in Fig. 1A. Extracellular recordings from the medial anterior (MA), the anterior anterior (AA), the B1 branch of the dorsal posterior (DP:B1), and the B2 branch of the dorsal posterior (DP:B2) [data collected from left (L) and right (R) roots]. Top trace: the stimulus artifact. Inset: the average relative shortening already reported in Fig. 2C (the time base is the same as in all the other traces of A, but just the first 5 s of the displacement are shown). The stimulus was a current pulse of 1 mA lasting 1 s and was delivered through 2 wires inserted in the skin between the 3rd and 4th segment (see Fig. 1B). B: recordings at higher sweep speed with identified neurons labeled as described in METHODS. Arrowed segments indicate extracellular signals associated with the left motoneuron 109.

During whole-body shortening, trains of action potentials lasting for several seconds were detected in all roots. Action potentials produced by specific motoneurons were identified, as shown on the expanded traces in Fig. 3B by analyzing their shapes and sizes in specific roots and by intracellular recordings (see METHODS). For instance, motoneuron LL was immediately identified because of the simultaneous presence of two extracellular voltage signals on the DP:B1 and DP:B2 roots. Each action potential from an identified motoneuron, such as motoneurons 3, 107, 108, 109, and L on the left (L) and right (R) side of the 10th ganglion, is labeled in Fig. 3B (see Ort et al. 1974). All these motoneurons were activated during whole-body shortening and fired action potential at frequencies varying between 10 and 20 Hz. The average latency of the first action potential evoked during whole-body shortening, measured from the onset of the stimulus, was 66.4 ± 43.6, 86.3 ± 33.0, 58.3 ± 7.3, 149.4 ± 196.1, and 96.2 ± 14.1 ms (n = 11) for motoneurons 3, 107, 108, 109, and L, respectively. This was in agreement with a previous analysis (Shaw and Kristan 1995).

In contrast to the behavioral reaction, spike trains of individual motoneurons were poorly reproducible from trial to trial and did not have any obvious regularity as shown in Fig. 4.

View larger version (20K): [in this window] [in a new window]   Fig. 4. Poor reproducibility of the firing pattern of motoneurons 107 and 108 during whole-body shortening. Electrophysiological data and behavioral response of the animal are recorded simultaneously in the semi-intact preparation shown in Fig. 1B. A-D each reproduce, in a raster plot, the occurrence times of spikes in motoneuron 107L (top) and in motoneuron 108L (middle), recorded from the AA root of the 10th ganglion in 4 different trials of the same stimulation. Bottom: the time course of the displacement (in pixels) of a selected feature on the skin of the 11th segment, flattened and pinned on the bottom of the recording chamber as described in Fig. 1B. A and C, top: the artifact of the electrical stimulation.

Every panel in Fig. 4 (A-D) reproduces the occurrence of spikes in motoneurons 107 (top), 108 (middle), and the corresponding whole body shortening (bottom) observed during the same trial. The firing pattern in these two motoneurons was highly variable and in some occasions almost absent (see Fig. 4D). On the contrary, the extent of the whole-body shortening was fairly reproducible, even during the trial when motoneuron 107 failed and motoneuron 108 almost failed to respond to the electrical stimulation.

Characterization of electrical variability

The characterization of the firing variability of a neuron using the CV depends on the value of the time window t used for computing the average and standard deviation of the firing rate. The choice of the value of t is dictated by statistical criteria and physiological considerations.

Figure 5A reproduces the CV of the firing of recorded motoneurons at the peak of the induced electrical activity as a function of the time window t. A general trend was observed in most motoneurons: the CV had a minimum for a value of t between 100 and 400 ms. In some experiments, the CV of specific motoneurons did not change significantly when t was varied from 50 to 800 ms. Therefore a choice of t between 100 and 400 ms will in general provide the lowest value for the CV even if, in a few experiments, there was an increase of the CV from 100 to 400 ms (see cells 8_L and 5_L in Fig. 5A).

View larger version (25K): [in this window] [in a new window]   Fig. 5. Choice of the time window t suitable to compute 1st-order properties of motoneuron firing activity. A: CV of the firing of 7 recorded motoneurons at the peak of their evoked electrical activity as a function of the time window t chosen to compute the CV and the average firing rate (AFR). , cell LL; , cell 107R; , cell 108R; ×, cell 109R; *, cell 8L; , cell 5L; and +, cell 4L. Some of these neurons are the same whose firing is shown in Fig. 3, Fig. 6, and Fig. 7. B: average number of evoked spikes for the same motoneurons of A at the peak of their evoked electrical activity as a function of the time window t used to compute the CV and the AFR. Symbols as in A. C: displacement induced on a patch of leech skin (skin preparation) stimulating intracellularly individual motoneurons with a depolarizing current step. Spike discharges at different frequencies are evoked in different motoneurons. , cell 3 (~50 Hz); , cell 3 (~30 Hz); , cell 8 (~25 Hz); *, cell L (~30 Hz); and , cell 108 (~30 Hz).

Figure 5B illustrates the average number of evoked spikes for the same motoneurons in a time window t starting from the first spike evoked by the electrical stimulation. All of these motoneurons fired on average between 1 and 10 spikes during the first 400 ms, which is the time window corresponding to the lowest CV (see Fig. 5A). This time window also corresponds to the delay between the electrical stimulation and the onset of the muscle contraction (see Fig. 2). Figure 5C reproduces the detected displacement of the skin, induced by passing a depolarizing current in single motoneurons, as a function of the number of spikes evoked in the neurons. In motoneurons 3, 8, L, and 108 the threshold for the detection of a skin deformation of at least minimum 10 µm occurred when at least four to six spikes were evoked in that motoneurons.

For several reasons, these results indicate that a time window between 200 and 400 ms is appropriate for computing the CV. First, this is the time window for which the CV is minimal and therefore maximizes reproducibility. Second, this time window corresponds to the delay between the onset of the stimulation and the initiation of the behavioral reaction. And finally, during this time window motoneurons fire enough spikes to produce a noticeable muscle contraction.

The poor reproducibility of the firing of motoneurons (see Fig. 4) could be caused by a variability in effective stimulation delivered by the metal wires sutured on the skin or by occasional failures of the neural signal to propagate from the head through the connective fibers down to the 10th ganglion. A way to test these possibilities is to record the very large signals produced by the S cells from the anterior connective entering into the 10th ganglion. The S cells network is responsible for a fast signaling pathway along the ganglia chain and contributes to whole-body shortening (Shaw and Kristan 1995, 1999). Figure 6A shows the AFR and the CV of the S action potentials during whole-body shortening obtained with an en passant suction pipette between the 9th and the 10th ganglion. The first action potential of the S cell appeared after 57.3 ms (n = 18) from the onset of the electrical stimulation (Shaw and Kristan 1995) and had a jitter of less than 4 ms in most preparations. As shown in Fig. 6A, during the electrical stimulation the CV of the S cell firing reached a value lower than 0.3, even when a binwidth of 50 ms was used. The CV of the S cell was usually below 0.3 for binwidth of 200 ms and always below 0.35 for values of t up to 400 ms. These results indicate that the neural signal traveling in at least one of the fast signaling pathways along the ganglia chain reached the ganglion reproducibly during each trial.

View larger version (24K): [in this window] [in a new window]   Fig. 6. First-order properties of the S cell and identified motoneurons during whole-body shortening. A: AFR and CV of the S cell connecting the 9th and 10th ganglion during whole-body shortening obtained from 30 different trials over a binwidth of 50 ms. B-F: AFR (bottom) and CV (top) of the left motoneurons 109, 107, 108, 3, and L of the 10th ganglion. Data were obtained from 11 different trials of the experiment illustrated in Fig. 3 over a binwidth of 200 ms.

The reproducibility of the neural response of individual motoneurons was quantified by analyzing the first-order statistics of the firing of identified motoneurons. Figure 6, B-F, shows the AFR and the CV for the identified motoneurons on the left side from 11 different trials of the experiment shown in Fig. 6 with a binwidth of 200 ms. The AFR of these motoneurons increased during electrical stimulation. After the termination of the electric stimulation, the electrical activity of some motoneurons was transiently depressed (cells 3 and 107), but that of other motoneurons (109_L and 109_R, this last not shown in the figure) remained high for several seconds. In particular, the AFR of both motoneurons 109 shows two distinct peaks, the first about 100 ms after the application of the stimulus and the other one after the end of the contraction. This second big and wide peak is probably correlated to the beginning of the writhing movement, observed in some trials after the completion of the whole-body shortening, as mentioned in the preceding text.

The CV of all motoneurons was usually high, about 2 or greater. Only during the initial activation did the CV approach a value smaller than 1. For motoneurons 3 and 107 the CV increased significantly immediately after the cessation of the electrical stimulation. Figure 6 does not report the trend of CV of the identified motoneurons for a bin of 50 ms, but the general trend was that the CV computed using a 50-ms bin was higher than the CV computed using a 200-ms bin, as suggested by Fig. 5A. Comparing the CV of the S cell with the CV of all identified motoneurons computed with the same binwidth of 50 ms, identified motoneurons showed a larger CV than the S cell. These results suggest that the origin of the variability of motoneuron firing may be within the segmental ganglion or in the other fast signaling pathways which coordinate the shortening reflex and run parallel to the S cell network.

In another series of experiments, whole-body shortening was studied while recording extracellular voltage signals from PP roots (see Fig. 7A). Also in this case, an evident increase of the occurrence of action potentials was observed. As shown in Fig. 7B, it was possible to identify extracellular voltage signals from at least three longitudinal 4, 5, and 8 motoneurons whose action potentials produced clear and large extracellular voltage signals.

View larger version (23K): [in this window] [in a new window]   Fig. 7. Electrical recordings from the PP root during whole-body shortening. A: suction pipette recording from the left PP root of the 10th ganglion during whole-body shortening. Top: the stimulus artifact, which was the same as in Fig. 3. B: recordings at higher sweep speed with identified neurons labeled as described in METHODS. C-E: AFR (bottom) and CV (top) of motoneurons 8, 5, and 4, respectively.

The AFR and CV of motoneurons 8, 5, and 4 are shown in Fig. 7, C-E, respectively. The firing of these longitudinal motoneurons was clearly increased during whole-body shortening. For motoneuron 4, the firing was transiently depressed after cessation of the electric stimulation. Similarly to motoneurons analyzed in Figs. 3 and 4, the firing of action potentials in motoneurons 4, 5, and 8 was characterized by a significant variability.

Second-order statistics of coactivated motoneurons

When different shapes of action potentials are extracted from analog recordings and assigned to identified motoneurons (see METHODS), it is possible to analyze their simultaneous firing. Figure 8 illustrates the occurrence of action potentials of 10 motoneurons during an episode of whole-body shortening. The total length of the episode illustrated in Fig. 8 is 2.5 s, during which the electrical stimulation was delivered for 1 s (see top trace).

View larger version (28K): [in this window] [in a new window]   Fig. 8. Firing of identified motoneurons during whole-body shortening. Each trace represents the firing of an identified motoneuron during whole-body shortening as a binary trace where an upper deflection indicates the occurrence of an action potential. Top: the timing of the current pulse of 1 mA delivered to the skin inducing whole-body shortening. From top to bottom, each trace represents the firing of the left and right motoneurons L, 3, 109, 107, and 108.

From visual inspection, it is evident that the firing of the left and right motoneurons L is highly correlated (see arrows joining almost coincident action potentials). Indeed during the time sweep considered in Fig. 8 only once (see black dot) did motoneuron LR fire an action potential not coinciding with any action potential of motoneuron LL. This correlated firing is a consequence of the electrical coupling between LR and LL (Nicholls and Purves 1970; Stuart 1970).

The second-order properties of the firing of coactivated motoneurons during whole-body shortening were analyzed by computing the cross-correlogram for pairs of motoneurons during the time of electric stimulation (1 s). Cross-correlograms were computed over 32 different trials for a total time of 32 s. As shown in Fig. 9A, the cross-correlograms of the right and left motoneurons L were highly peaked, indicating strong correlation of their electrical activity. In contrast, the cross-correlograms of all other pairs of motoneurons in the same ganglion (Fig. 9, B-G) were flat, indicating poor correlation of their firing. Poor correlation was also observed among motoneurons of two adjacent ganglia, for example between the left motoneurons 107 and 108 of the 10th and 11th ganglion (see Fig. 9H), respectively.

View larger version (33K): [in this window] [in a new window]   Fig. 9. Second-order statistics of motoneurons during global shortening. Cross-correlation of identified motoneurons during whole-body shortening. Data were collected during the application of a current pulse of 1 mA lasting for 1 s and over 35 different trials repeated every 3 min. Each panel represents the cross-correlogram of the firing of motoneurons indicated in the top right corner. All motoneurons in A-G are from the 10th ganglion. The data in H refer to 2 motoneurons from the 10th and 11th ganglia.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The data presented here describe new properties of the firing of action potentials of motoneurons coactivated during an escape reaction caused by electrical stimulation. This fundamental reaction was studied in the leech because all of its motoneurons have been identified and extensively studied (Ort et al. 1974; Sawada et al. 1976; Stent et al. 1978; Stuart 1970). The results presented here are similar to those obtained in the Aplysia when the siphon was touched (Tsau et al. 1994; Wu et al. 1994) and to the escape reaction in the cockroach (Camhi 1988) and are reminiscent of motoneuron firing patterns often observed in the mammalian spinal cord (Windhorst 1990). Therefore they may help in understanding basic properties of distributed processing and neural computation.

Whole-body shortening is a distributed process

Whole-body shortening started within 300-400 ms from the onset of the electric stimulation applied to the animal head (Shaw and Kristan 1995) (see also Fig. 2) and produced a change of about one-third in the total length of the animal. The posterior end of the animal started moving about 50-100 ms after the head, which is a delay comparable with the time taken by the parallel pathways of interneurons that contribute to the reflex by conducting an impulse along the entire length of the body (Shaw and Kristan 1995). Mechanical feedback may also play a role in the transmission of excitation, given the high sensitivity of stretch receptors (Blackshaw 1993). The relaxation phase of the shortening reaction was much longer and less reproducible than the contraction.

The present analysis shows that this escape reaction is mediated by the coactivation of several motoneurons innervating longitudinal muscles, proving that not only cells L, 3, 4, and 108 are involved, as already reported by Shaw and Kristan (1995), but also other longitudinal motoneurons such as 5, 8, and 107. In addition there are probably other cells that have not been identified in the present analysis. During whole-body shortening (see Fig. 3), these motoneurons fire action potentials at a rate not exceeding 20 Hz. Because of this they produce at most four or five action potentials before the onset of the behavioral response. Under these conditions, as shown in Fig. 5, individual motoneurons induce a slightly detectable muscle contraction. For this reason, the rapid and large muscle contraction underlying whole-body shortening can only be achieved by the recruitment and coactivation of many motoneurons, as the present analysis indicates. In each leech ganglion there are about 20 known pairs of motoneurons (Muller et al. 1981; Ort et al. 1974; Sawada et al. 1976; Stent et al. 1978; Stuart 1970) and therefore during whole-body shortening a significant fraction of all of the leech muscles are activated. Therefore whole-body shortening is a distributed process, mediated by a relatively large neuronal population, and involving probably all motoneurons innervating longitudinal muscles. This observation is reminiscent of similar findings in the Aplysia (Hickie et al. 1997; Tsau et al. 1994) when the siphon receives a mechanical stimulation. Also in this case hundreds of neurons are coactivated and the behavioral reaction is mediated by large distributed neuronal events. A similar distributed processing underlies the escape reaction in the cockroach, a more complex invertebrate (Camhi 1988).

All motoneurons here identified as coactivated during whole-body shortening are longitudinal motoneurons with the exception of motoneuron 109 innervating dorsoventral muscles (Stuart 1970). This motoneuron, as shown in Fig. 6, is activated to some extent during the electrical stimulation inducing whole-body shortening but becomes more activated after a few seconds when more complex movements of the leech occur, such as twisting and/or writhing. Therefore motoneuron 109 is certainly activated by electrical stimulation, but its involvement in whole-body shortening is not obvious.

Correlated firing and motor control in humans and mammals

The simultaneous firing of coactivated neurons involved in motor control has been investigated in mammalian and human motor system primarily using cross-correlation analysis. In particular, this technique has been extensively applied to analyze electromyographic (EMG) recordings from human motor units. Neighboring motor units from the same muscle usually exhibit a strongly correlated firing, referred to as short-term motor unit synchronization (Kirkwood and Sears 1978; Sears and Stagg 1976). This short-term synchronization was deduced from the narrow peak, lasting just a few milliseconds, centered near time 0 in the cross-correlogram constructed between the spike trains of pairs of single motor units. This synchronization is thought to arise from a common last-order branched-axon presynaptic inputs (Kirkwood and Sears 1978). Short-term synchronization between motor units has been detected in a variety of muscles in man. A list of the main studies in this field can be found in the recent review of Farmer et al. (1997). A number of these studies report that short-term correlation also exists between pairs of motor units recorded in different synergic muscles, for example corresponding muscles acting on adjacent or distal fingers of the hand (Bremner et al. 1991a). This may reflect the distribution of divergent common last order presynaptic inputs from corticospinal axons on different motoneurons pools (Farmer et al. 1997). Usually in these experiments the strength of the synchronization decreases as recordings are made from more distant muscle pairs, for example from more distant fingers (Bremner et al. 1991b) or from more distal muscle fibers in the same leg muscle (Nielsen and Kagamihara 1994). This looser synchronization, termed presynaptic synchronization (Kirkwood et al. 1982, 1984), is indicated by a broader peak in the cross-correlatiogram and has been detected in antagonist muscles coactivated during the same global movement (Neilsen and Kagamihara 1994).

A significant degree of correlation or of synchronization between distinct motor units has been associated to pathological processes, such as stroke or spinal cord disease (Datta et al. 1991; Farmer et al. 1993) and dystonia (Farmer et al. 1998), or in association with physiologic and voluntary tremor (Logigian et al. 1988). In addition, an abnormal degree of synchronization among the firing of neurons in the globus pallidus has been associated to Parkinson's disease and to motor control pathologies (Bergman et al. 1998; Nini et al. 1995).

Summarizing motor control in humans and mammals seems to require the coexistence of motoneurons firing with a different degree of correlation or of synchronization. This different degree of correlation is usually ascribed to variation in presynaptic coherence and may be optimal for efficient and reliable motor control.

Functional role of statistical independence in leech motor control

Our manuscript also shows that in the leech nervous system the majority of motoneurons, including those in the same ganglion, activated by the same stimulation, fire action potentials in a decorrelated way. As shown in Figs. 8 and 9, the global firing of coactivated motoneurons does not have any clear and statistically significant pattern. Indeed the pairwise cross-correlograms of the electrical activity of these motoneurons are flat (see Fig. 9) except the one between the left and right motoneurons L. These motoneurons are tightly coupled by electrical junctions (Nicholls and Purves 1970; Stuart 1970) in each leech ganglion, and electrical coupling is well known to be a powerful synchronizing mechanism (Grattarola and Torre 1977; Mann-Metzer and Yarom 1999). Motoneurons coactivated during whole-body shortening are primarily activated by the S cell network (Burrell et al. 1999) and by another parallel pathway of as-yet unidentified interneurons (Shaw and Kristan 1999). These multiple synapses are likely to have some degree of unreliability like most chemical synapses (Allen and Stevens 1994; Goda and Sudhof 1997; Larkman et al. 1997; Lisman 1997; Markram et al. 1997; Zador 1998), which could result in an almost complete loss of correlated activity (Pinato et al. 2000). Electrical coupling has been found among a variety of leech motoneurons (Ort et al. 1974). The strength of this coupling is low and a high-frequency spike train evoked in a motoneuron may induce some spikes in another motoneuron that is coupled to the first. But, this does not occur in a one to one relationship (Ort et al. 1974) and the coupling is not strong enough to synchronize their discharge. The flats cross-correlation histograms of Fig. 9 confirm this rule (the strong coupling between the right and left L cells being the exception).

Escape reactions are vital for animal survival and must be reliable and efficient. Our hypothesis is that there is at least one mechanism in which statistical independence among coactivated neurons can be useful. It is not difficult to understand how unreliable and uncorrelated spike trains, such as those illustrated in Figs. 4 and 8, could lead to a reproducible global behavior. Whole-body shortening is mediated by the electrical activity of a large ensemble of interneurons and motoneurons, and it is well known (Pinato et al. 1999, 2000) that averaging and/or pooling uncorrelated random variables reduces variability. Indeed when the electrical activity of an ensemble of motoneurons underlying whole-body shortening is considered, the CV of the total electrical activity is significantly smaller. As shown in Fig. 10, the CV of the ensemble activity (thick line) is quite small. It is always less than 0.5 during the contraction phase and is significantly smaller than the CV of individual motoneurons (thin lines). The CV of the pooled electrical activity is consistent with the CV of the behavioral response shown in Fig. 2.

View larger version (33K): [in this window] [in a new window]   Fig. 10. CV of an ensemble of motoneurons underlying whole-body shortening. The thick line is the CV of the ensemble of left and right motoneurons L, 3, 107, 108, 109 from the 10th ganglion. The thin lines, which are the same as those shown in Fig. 6, are the CV of some individual motoneurons.

Very similar results were obtained in several other cases of distributed neuronal processing. A very small correlation between neurons sharing the same stimulation has also been observed among the Aplysia neurons when the siphon was touched (Tsau et al. 1994). In this case, only 0.3% of neuron pairs exhibited a significant degree of correlation on a time scale of tens of milliseconds, while on a time scale of seconds, some degree of correlation is usually present (Tsau et al. 1994; Arisi and Torre, unpublished data). Statistical independence between coactivated neurons has also been observed in neuronal cultures composed by cortical neurons from neonatal rats (Pinato et al. 1999). In the vertebrate retina, sharp cross-correlograms of the firing of coactivated neurons were observed only between electrically coupled neurons, while pairs of neurons connected by chemical synapses had a much smaller degree of correlation (Brivanlou et al. 1998).

Conclusions

The results presented in this manuscript and previous behavioral analysis discusses some basic properties of motor control in the nervous system of the leech H. medicinalis. Leech motoneurons can fire up to 50 and 100 Hz when stimulated with extrinsic depolarizing current, but in the presence of physiological synaptic inputs fire at most at 20-30 Hz (see Fig. 6 and Fig. 7) (Mason and Kristan 1982). As whole-body shortening occurs within some hundreds milliseconds from the stimulus initiation, each motoneuron fires at most five or so spikes, which in turn produce a rather small contraction (see Fig. 5). Therefore a substantial whole-body shortening can only be achieved by the simultaneous recruitment of almost every longitudinal motoneuron. For this reason, the underlying neural process is largely distributed and involves the excitation of many neurons and motoneurons. Because of the presence of unreliable synapses and other biophysical mechanisms causing irregular firing patterns (Harsch and Robinson 2000; Mainen and Seinowski 1995), spike trains in individual motoneurons are poorly reproducible. And due to statistical independence among coactivated motoneurons, the population firing becomes reproducible (see Fig. 10). These considerations, drawn for the analysis of whole-body shortening, depend on basic biophysical properties of leech muscles, neurons, and synapses and are likely to be relevant for all motor behaviors of the leech.

The statistical independence among coactivated neurons was initially a surprising and unexpected result, which turned out to be shared by many other neuronal networks and is likely to be a basic feature of parallel processing of neuronal networks with unreliable synapses. Indeed our results shed a new light on the possible functional role of the statistical independence in a network of neurons devoted to drive a defense reflex that must be reproducible and reliable. Even if the activity of single motoneurons is intrinsically variable, statistical independence of the single units enables them, as a whole, to produce a highly reproducible and reliable firing pattern.

These results suggest a new and basic property of distributed processing, i.e., when a computation or an action is mediated by a large ensemble of neurons, statistical independence among coactivated neurons must be expected. This may be an important and beneficial feature of neural computation.