How does the nervous system produce behavior? The desire to answer this question has led investigators to examine the biophysical properties of individual neurons and their synaptic connectivity during the production of specific behaviors (i.e. motor neuron impulse patterns). In order to accomplish this task 'simple' preparations were necessary where the activity of identified neurons could be directly correlated with behavior. Such preparations needed to be amenable to physiological recording techniques (primarily intracellular recording) while still producing a recognizable motor pattern. These requirements necessitated the use of reduced preparations: semi-intact and isolated nervous systems. The relative ease with which neuronal activity can be monitored in neuronal cell bodies, the simplicity of both the nervous system and the musculature, and a small repertoire of behaviors made the medicinal leech an ideal preparation to address how the nervous system produces behavior in terms of synaptic interactions between identified neurons. In this article, we will describe the organization of the leech nervous system, the experimental approach employed in characterizing identified neurons in the leech, and the extent to which the goal of understanding the neuronal basis of behavior has been accomplished with respect to leech swimming. We will also discuss how the initial focus of this research has expanded to take into account the role that neurotransmitters play in controlling swimming, how the decision to swim is processed and the behavioral specificity of identified neurons.

Leech Neuroanatomy

Medicinal leeches, Hirudo medicinalis, belong to the phylum Annelida which comprises the segmented worms. The most distinguishing characteristic of Annelida is body segmentation along the anterioposterior axis, which is clearly reflected in the organization of their central nervous system (CNS) (Barnes, 1980). The CNS of the medicinal leech is composed of 34 ganglia: six fused ganglia form the head ganglion, two of which form the supraesophageal ganglion and four form the subesophageal ganglion; seven fused ganglia form the tail ganglion and a chain of 21 ganglia, with one per mid-body segment, form the segmental nerve cord (Fig. 4.1a). The four fused ganglia comprising the subesophageal ganglion, the segmental ganglia and the six fused ganglia of tail ganglion arise developmentally from the same neuroblasts, while the two fused ganglia comprising the supraesophageal ganglion have a different developmental origin (Weisblat et al., 1980). Each segmental ganglion and each of the four fused ganglia of the subesophageal ganglion contain approximately 200 pairs of neurons with unipolar cell bodies that range in diameter from 105 to 1005 and are located in a monolayer surrounding a central neuropile (Macagno, 1980). All synaptic connections between neurons occur in a central neuropile. Both impulse and synaptic activity can be monitored directly from the cell bodies of individual neurons with fine microelectrodes (Nicholls and Baylor, 1968). Over 50 pairs of segmental neurons have been identified and can be reliably located in the segmental ganglia based on their position within the ganglion, size, shape and physiological properties, while only 10 - 15 pairs of neurons have been identified in the subesophageal ganglion (Sawyer, 1986 and Brodfuehrer, personal observation). The segmental ganglia are joined by connectives which are composed of two large lateral bundles of axons and a thin, medial bundle called Faivre's nerve. A pair of nerve roots arises from each segmental ganglion to innervate the body wall musculature surrounding that segment. Three distinct muscle layers: circular, oblique and longitudinal, form the tubular hydroskeleton (i.e. body wall) of the leech (Fig. 4.1b) These three layers, together with the dorsoventral muscles are largely responsible for the relatively simple repertoire of movements in leeches (Rowlerson and Blackshaw, 1991).

'Early' Identified Neurons in the Leech Nervous System

In the late 1800s it became apparent that neurons in the leech CNS could be readily recognized from animal to animal based solely on anantomical characteristics: the size and position of neuronal cell bodies within a segmental ganglion. The first 'identified neurons' in the leech were a pair of giant nerve cells (ganglion balls, i.e. Leydig cells) and a pair of 'colossal cells' (i.e. Retzius cells) described by Leydig and Retzius, respectively (Kandel, 1976). Today, all the neuronal cell bodies on the dorsal and ventral surfaces of the segmental ganglia have been identified to some extent. They've all been given letter or number designation based primarily on position within the segmental ganglion, while the structure - function relationship of almost half of these neurons is known (Muller, et al., 1981).

The first extensive examination of neuronal identity was performed by Nicholls and Baylor (1968) on mechanosensory neurons which are sensory neurons that respond to cutaneous stimuli. Mechanosensory neurons were well suited for this study because their somata lie within the segmental ganglion and their axons extend out the nerve roots where they terminate directly in the tegument. No peripheral synapses are involved in processing cutaneous stimuli.

How does one establish the structure - function relationship of neurons in a segmental ganglion? Nicholls and Baylor (1968) addressed this question using a preparation consisting of a flap of body wall that was several segments in length and connected to a segmental ganglion by the nerve roots (Fig. 4.2a). Using this preparation, Nicholls and Baylor (1968) found that each segmental ganglion contains three classes of mechanosensory neurons that respond specifically, and selectively, to various degrees of cutaneous stimulation: touch (T) cells (three pairs per ganglion) respond to light touch of the body wall; pressure (P) cells (two pairs per ganglion) respond following marked deformation of the body wall and nociceptive (N) cells (two pairs per ganglion) respond to noxious stimuli (Figs. 4.2B - 4.2e). The size, shape, and position of the mechanosensory neurons within each segmental ganglion were remarkably stereotyped, as were the size, shape and position of their receptive fields on each body wall segment. Similarly, the physiological properties of each class of mechanosensory neurons were distinct from one another but consistent from one segmental ganglion to the next. Thus each class of mechanosensory neurons had a unique structure - function relationship. More importantly, the results of Nicholls and Baylor (1968) suggested that all 200 pairs of segmental neurons could be identified and their functional role in leech behavior discerned. One of the first behaviors systematically studied at the level of identified neurons was leech swimming.

Swim Motor Pattern

A leech swims by undulating its extended and flattened body in the dorsoventral plane producing a wave that travels rearward along the animal. The crests and troughs of the undulatory wave are produced by antiphasic contraction of dorsal and ventral longitudinal muscle in each body wall segment, while an intersegmental delay of this contractile cycle in posterior segments causes its rearward progression. The forces exerted against the water by the rearward traveling body wall propel the leech forward (Friesen and Pearce, 1993 and Stent et al, 1978). The period of the segmental contractile cycle varies from about 400 ms to 2000 ms. A constant relationship between contractile cycle period and intersegmental delay time ensures that at all swimming speeds the body wall of intact leech maintains a waveform equal to approximately one wavelength (Kristan et al. 1974).

Two early studies of leech swimming movements provided the ground work for further investigations into the neural basis of swimming. First,ll (1905) showed that disconnecting the head and tail ganglia from the segmental nerve cord did not abolish leech swimming. Hence swimming movements are generated within segmental ganglia. Second, Gray et al. (1938) demonstrated that the neural activity responsible for coordinating swimming occurred via the intersegmental connectives and not through the body wall. A result of these two studies was the development of a semi-intact preparation, where several mid-body segmental ganglia were exposed for intracellular and extracellular recording, while leaving more anterior and posterior body wall regions intact (Kristan et al., 1974). Using semi-intact preparations, and later using isolated nerve cord preparations, the role of specific neurons in the segmental ganglia and subesophageal ganglion, with respect to swimming, was determined (Kristan et al., 1974 and Kristan and Calabrese, 1976). Experiments were designed to identify how peripheral sensory information propagated, neuron-to-neuron, through the nervous system to generate the antiphasic contractions of the dorsal and ventral longitudinal muscles of a swimming leech.

Swim-Generating Network

The culmination of approximately 30 years of research has been the identification of five functional classes of neurons (mechanosensory, motor, oscillator, gating and trigger) that transduce mechanosensory stimulation into the swim motor program (Fig. 4.3a).

Body wall movements during swimming are controlled by four groups of segmental motor neurons: the ventral excitors (VEs) that excite the ventral longitudinal muscles; ventral inhibitors (VIs) that inhibit the ventral longitudinal muscles and the VEs; dorsal excitors (DEs) that excite the dorsal longitudinal muscles; and dorsal inhibitor (DIs) that inhibit the dorsal longitudinal muscles and the DEs (Ort et al., 1974). The membrane potential of each motor neuron oscillates and fire bursts of action potentials during their depolarized phase (Fig. 4.3b).

At the center of the swim-generating network is an ensemble of neurons that comprise the central pattern generator or oscillator which provides the appropriate intra- and intersegmental phasic input to segmental motor neurons that produce the undulatory body wall movements characteristic of swimming leeches. Three physiological criteria were used to determine if a given neuron was a member of the swim oscillator. First, the membrane potential of the neuron had to oscillate in phase with the swim motor pattern. Second, the neuron had to be synaptically connected to other members of the oscillator and lastly, perturbations of membrane potential oscillations in the putative oscillator neuron had to cause a transient phase-shift of the swim motor pattern (Friesen et al., 1978 and Poon et al., 1978). To date, six paired (cells 115, 33, 28, 27, 123, and 60) and one unpaired interneuron (cell 208), and fours pairs of motor neurons (cells 1, 2, 102 and 119) meet these criteria (Friesen, 1989; Fig. 4.4a).

With the identification of a central swim oscillatory network and its output connections to motor neurons, attention turned to understanding how the swim oscillatory network was 'turned-on'. Weeks and Kristan (1978) identified an unpaired intersegmental interneuron, cell 204, located in segmental ganglia 10 - 16 that had 'command-like' properties for the initiation of swimming. In isolated cords of segmental ganglia, depolarization of a single cell 204 such that the cell fires at an average frequency of 20 - 50 Hz, initiated swimming with swimming activity generally lasting as long as a suprathreshold firing frequency was maintained in cell 204 (Fig. 4.4b). Because of this latter property cells 204 were classified as swim-gating interneurons (Friesen, 1989). Cells 204 activate the swim oscillator via direct connections to three oscillator neurons, cells 28, 115 and 208 (Nusbaum et al., 1987). What was still lacking was an understanding of how the swim-gating interneurons were themselves driven. Cutaneous stimuli which elicited swimming only indirectly excited cells 204. A neuron or groups of neurons had to be interposed between the mechanosensory neurons and the swim-gating interneurons.

The identification of cells Tr1 provided, in part, the missing link between the mechanosensory input and the activation of swim-gating interneurons. Cells Tr1 are paired interneurons that have their somata in the subesophageal ganglion and extend their axons the length of the ventral nerve cord (Brodfuehrer and Friesen, 1986a,b). They receive direct excitatory input from P and N cells located in the subesophageal ganglion and in the first segmental ganglion, and provide output directly onto all segmental swim-gating interneurons (Brodfuehrer and Friesen, 1986b). Furthermore, brief (approximately 1 s), high frequency (30 to 50 Hz) stimulation of cell Tr1 can elicit a swim episode (Fig. 4.4c), with the length of elicited swim episode being independent of cell Tr1 stimulation intensity. These physiological properties along with the fact that cells Tr1 are silent during swimming led to their classification as swim trigger neurons.

With the identification of cells Tr1 and their input - output connections, the goal of understanding the neuronal basis of leech swimming has been accomplished at one level. That is, at the level of knowing how specific mechanosensory inputs propagate through the leech CNS to activate the swim oscillator and produce swimming movements. Although significant this connectionist model of the leech swim-generating network does not adequately explain the neuronal mechanism governing swimming nor does it explain the behavioral variability observed in the ability of a given input to initiate swimming. A complete understanding of the leech swim-generating network requires a thorough characterization of the biophysical (ionic currents and synaptic transfer functions) and the biochemical (neurotransmitter and receptor phenotypes) properties that govern the synaptic interactions between neurons, and must incorporate behavioral variability into the model. In the next two sections we will discuss recent observations that extend our connectionist understanding of leech swimming to include the roles that the neurotransmitter glutamate and the 'internal state' of the nervous system play in the initiation of swimming.

Glutamate and the Initiation of Swimming

The latency for swim initiation following stimulation of trigger neurons ranges from 1 to several seconds. During this latency period there is a gradual increase in the membrane potential and firing frequency of swim-gating interneurons which are necessary for the initiation of swimming (Brodfuehrer and Friesen, 1986c). Insights into the mechanism underlying this process occurred when it was shown that pressure ejection of L-glutamate, kainate or quisqualate onto the soma of cell 204 produced sustained excitation in cell 204 that closely mimicked cell 204's activity pattern following stimulation of cell Tr1 (Brodfuehrer and Cohen, 1990). Further analyses using several antagonists to non-NMDA receptors (DNQX, kynurenic acid and joro spider toxin) demonstrated that cells 204 possess non-NMDA receptors and that their activation by glutamate, which is most likely released by cell Tr1, leads to prolonged excitation of cells 204 (Thorogood and Brodfuehrer, 1995; Fig. 4.5a). Furthermore, immunoreactivity to GluR 5/6/7 localizes onto processes of cell 204 (Thorogood et al., 1996; Fig. 4.6). Experiments by Dierkes et al. (1996) demonstrated that the influx of both Na+ and Ca2+ underly glutamate-induced excitation of leech neurons and that increases in the intracellular free Ca2+ concentration occurs through voltage-sensitive ion channels and not through ionotropic glutamate receptors. A possible scenario for cell Tr1-induced prolonged excitation of cell 204 is that glutamate released by cell Tr1 binds to non-NMDA receptors on cell 204 causing an influx of Na+. The Na+ influx initially depolarizes cell 204 enough to open voltage-sensitive Ca2+ elevating intracellular free Ca2+ levels which, through a yet identified pathway, produces sustained membrane depolarization of cell 204.

L-Glutamate and non-NMDA receptors also appear to mediate the synaptic connections from P cells to cell Tr1, and from cell 204 to swim oscillator neurons - cells 208, 28 and 115 (Thorogood and Brodfuehrer, 1995 and Thorogood et al., 1996; Figs. 5B and C). However, the time course of these synaptic events are different than that between cell Tr1 and cell 204. Activation of neither cell Tr1 nor oscillator neurons greatly outlasts the stimulus duration in cell P or cell 204, respectively. This suggests that there exists a functional distribution of glutamate receptors in leech neurons (Dierkes et al., 1996). It is also interesting that L-glutamate plays a role in controlling the initiation of swimming only with respect to the input pathway to the swim oscillator (see Fig. 4.3a). On the output side, the synaptic connections from oscillator neuron cell 208 to DE-3, a dorsal excitor motor neuron, and from DE-3 to the dorsal longitudinal muscles are not blocked by the same non-NMDA receptor antagonists that block glutamatergic connections on the input side (Thorogood et al., 1996).

Distributed Processing of the Decision to Swim

As is the case with leech swimming, the control of rhythmic motor patterns has primarily been described in terms of neurons and pathways that activate oscillatory networks (Pearson, 1993). Moreover, little attention is paid to the fact that stimulation of these neurons or pathways, even in reduced preparations, does not guarantee a fixed motor output; rather the motor response is highly variable. For example, stimulation of cell Tr1 can lead to swimming on one trial, but might not on the next trial (Brodfuehrer & Friesen, 1986c; Fig. 4.4c). The variability observed in cell Tr1's swim-initiating ability is inconsistent with the control of swimming being dictated simply by the activation of swim-gating and oscillator interneurons. It is also doubtful that neuromodulators or prior experience are critical factors in regulating whether cell Tr1 triggers swimming since variability in swim responsiveness occurs immediately after isolating the nerve cord from the leech, when inter-stimulus intervals are short (less than 20 s) and fluctuate within a given preparation over the time course of an experiment. Furthermore, application of serotonin, the most potent neuromodulator of leech swimming, affects swimming only after bathing nerve cords for 10 - 20 min (Willard, 1981; Brodfuehrer and Friesen, 1986b and Hashemzaadeh-Gargari and Friesen, 1989). Thus, it is likely that the short-term variations in cell Tr1's swim-initiating ability are associated with changes in the 'internal state' of the leech nervous system and are not induced by the action of known neuromodulators of leech swimming.

Recent attempts to correlate the activity of individual neurons with the likelihood that a given stimulus will initiate swimming has led to the hypothesis that the control of swimming involves two parallel systems originating in the head ganglion that have opposite effects on the segmental swim-generating network: a swim-activating system that excites the segmental swim-generating network and a swim-inactivating system that inhibits or suppresses it (Brodfuehrer and Burns, 1995). In order for a given stimulus to initiate swimming the swim-activating system must be "turned on" and the inactivating system "turned off". Evidence supporting this dual control mechanism is based primarily on the activity patterns observed in cells 204 and cell SIN1, an identified interneuron in the subesophageal ganglion (Brodfuehrer and Burns, 1995).

In quiescent preparations (i.e. when swimming is not occurring) cells SIN1 are normally tonically active. When swimming occurs, spiking activity in cells SIN1 ceases and their membrane potentials hyperpolarize approximately 0.5s to 1.5 s prior to the onset of the first swim cycle (Fig. 4.7a). In addition, depolarization of cell SIN1 during swimming generally terminates the swim episode. Although suppression of cell SIN1's activity is necessary for swimming, it is not sufficient to initiate swimming since hyperpolarization of cells SIN1 alone does not initiate swimming. A concurrent requirement is the activation of the segmental swim-generating network. Before swimming starts, spiking activity ceases in cell SIN1 while increasing in cell 204 (Fig. 4.7b).

To date, only a few putative members of the swim-activating and inactivating systems have been identified (Brodfuehrer et al., 1995a). A potential candidate for the swim-activating system are cells SE1, a pair of interneurons in the subesophageal ganglion (Brodfuehrer et al., 1995b). Cells SE1 are generally spontaneously active, and receive feedback from the oscillator network during swimming. Their inclusion in the swim-activating system is based on the following observations: 1) Cell 204 receives direct excitatory input from cell SE1. 2) The level of excitation in cell 204 is positively correlated with the firing frequency of cell SE1. In fact, spiking activity in cell SE1 regulates the spontaneous level of activity in cell 204 to such an extent that when the swim motor program is not active, elimination of spiking activity in cell SE1 abolishes almost all EPSPs in cell 204 (Fig. 4.7c). 3) The level of excitation in both cell 115 and DE-3 is positively correlated with the activity of cell SE1. 4) Cell SE1 directly excites cells 28, 208, and DE-5. Thus cells SE1 have a profound excitatory influence on the swim-generating network.

On the otherhand, no strong candidtes for the swim-inactivating system have been identified. Cell SIN1 is most likely only a minor component of the swim-inactivating system since its activity level is controlled by other yet identified neurons, some of which are undoubtedly associated with leech behaviors that are incompatible with swimming such as whole-body shortening. Members of the swim-activating and -inactivating systems may also be part of a dynamic network that defines the 'internal state' of the leech nervous system, which is itself variable and modifiable, and influences the behavioral responsiveness of a leech to constant, repetitive stimuli (Grobstein, 1994).

Two observations document the existence of intrinsic variability within the leech nervous system, and that it affects the ability of a given input to elicit swimming. First, the motor output of isolated nerve cords intact from the head ganglion to the tail ganglion (H-T preparation) varies continuously in the absence of variations in input, but occassionally produces swimming without external input (Fig. 4.8a). Second, in an H-T preparation identical peripheral (dorsal posterior, DP) nerve stimulation sometimes, but not always, triggers swimming (Fig. 4.8b). There is no threshold stimulus voltage that consistently elicits swimming (Fig. 4.8c). In constrast, in a preparation consisting of a chain of ganglia from segmental ganglion 3 to the tail ganglion there is a clear threshold stimulus voltage where DP nerve stimulation reliably elicits swimming (Fig. 4.8c). Thus a property 'intrinsic' to the nervous system affects the behavioral responsiveness of leech preparations and is modifiable by changing the number of ganglia comprising the preparation; specifically by eliminating the influence of the head ganglion on the rest of the nervous system. This observation suggests that the ability of cell Tr1 stimulation to trigger swimming may depend upon the internal state of the nervous system.

Multifunctional Interneurons

The neuronal basis for several leech behaviors are known to varying degrees. As more and more neurons are identified as functional components controlling a particular behavior it has become apparent that not all interneurons are behaviorally specific. Some interneurons are multifunctional and play a role in controlling more than one behavioral. Cell 115 was originally classified as a member of the swim oscillator network based on the phase of its membrane potential oscillations during swimming and its synaptic connections with other oscillatory neurons: cell 115 receives direct excitatory input from oscillator neuron cell 208 and forms reciprocal inhibitory connections with three other oscillator neurons, cells 28, 102 and 1 (Nusbaum et al., 1987 and Friesen, 1989). Recently, cell 115 has been shown to contribute significantly to the the production of at least two other behaviors is cell 115, local bending and whole body shortening (Lockery and Kristan, 1990 and Wittenberg and Kristan, 1992). Similarly the swim-gating interneurons, cells 204, are excited during the extension but not contraction phase of crawling (Kristan et al., 1988). Choice experiments have also shown that both trigger neurons and swim excitor neurons are excited to the same extent by electrical stimulation of the body wall independent of whether whole-body shortening or swimming occurs (Shaw and Kristan, 1995). These observations suggest that identified neurons need not be dedicated to specific behaviors, and corresponds with our increased understanding that motor control is largely a distributed process of the nervous system rather than a dedicated process of a relatively small subset of the nervous system (Morton and Chiel, 1994). With respect to the the concept of identified neurons, these observations suggest that identity of a neurons may depend on structure, function and behavioral context.


Since the experiments of Nicholls and Baylor, the initial characterization of identified neurons has provided significant insight into the circuitry transforming mechanosensory input into the motor output of swimming. From physiological characterization of only a small percentage of cells within the leech CNS, we have gained important information about how the decision to swim is processed and how the rhythmic motor pattern is generated. While many of the synaptic connections in the swim-generating circuit have been identified, the elucidation of the biophysical and biochemical mechanisms underlying these connections has only recently begun. The observation that constant input can result in variable motor output suggests that, in addition to describing a cell's identity in terms of structure and function, factors such as behavioral context and the 'internal state' of the nervous system must also be considered. As circuits controlling other behaviors become known, one can examine the interactions between these networks to understand issues of behavioral choice at the level of identified neurons. The leech CNS has expanded our understanding of how the nervous system produces behavior and continues to serve as an excellent model in this endeavor.