The Structural Basis of an Innate Behavioural Pattern

‘… is it not possible that beneath all the variations of individual behavior there lies an inner structure of inherited behavior which characterizes all the members of a given species, genus or larger taxonomic group -just as the skeleton of a primordis ancestor characterizes the form and structure of all mammals today? Yes, it is possible!’

Konrad Lorenz (1958) 

Recent work, mainly on invertebrates, has established beyond doubt the reality of an inner structure for inherited behaviour. Ethological concepts like innate releasing mechanisms and fixed action patterns have been shown to be the products of specific neural networks, whose operations can account for many of the salient features of behaviour (e.g. Kandel, 1976). The explanation of behavioural concepts in terms of integrative neurophysiology requires, as an essential step in the analysis, the fractionation of exceedingly interactive networks in ways that are informative rather than merely destructive. There are two general approaches to fractionating nervous systems. One is to focus on a particular behaviour pattern (e.g. Kennedy, 1975). Focus on a particular pattern can be accomplished by using a stimulus that evokes a repeatable behavioural response; the neural systems that are activated by the stimulus can then be studied. The other approach is to focus on a particular neuroanatomical entity (e.g. Maynard, 1972). This is especially useful in segmented animals, where it is inviting to cut through the cables of pure axons that string neural ganglia together and then analyse the isolated portions. Discrete anatomical units containing from 10 to approximately 10⁵ neurones have been studied in this way.

Each approach has inherent limitations. Kennedy (1975) has advocated the former approach in the following terms:

‘… behaviors, not anatomical entities, are the elements on which natural selection works in shaping neuronal organization. Thus the system we must complete is the behavioral act, not the anatomy of the control center. Elements in the neuronal ensemble may participate in other behaviors as well; or the anatomical unit may contain behaviorally irrelevant elements. Neither should matter as long as our inventory of the neurons and connections that participate in the behavior is complete.’

We have found it useful to use both strategies by concentrating on those aspects of crayfish escape behaviour that can be studied in the isolated abdomen. In this review, I give a brief overview of the main features of the abdominal nervous system of the crayfish, Procambarus clarkii, as revealed by the work of numerous investigators, and then focus on recent advances in the analysis of the escape response, made primarily in F. B. Krasne’s laboratory and in my own. A key concept that now guides our work is that arthropod nervous systems and behaviour have evolved, in large part, by the gradual modification of individual body segments and that much can be learned by intersegmental comparisons. (For a brief summary of studies along these same lines see Miller, Hagiwara & Wine, 1984.)

Anatomically, the crayfish abdomen is simply the posterior portion of the body, and as such contains the posterior homologues of gut, axial muscles and appendages (Fig. 1). Functionally, however, the abdomen is analogous to an appendage, and for this reason is often referred to as the tail. Crayfish have two distinct modes of locomotion both of which involve the abdomen. During walking, which in adults constitutes more than 99% of the time spent in motion, the extended abdomen (Fig. 1A) is used for balance. It can also help steer the animal, by acting like a rudder, while the pitch and stroke frequency of the swimmerets can be varied to correct rolls and to aid in yawing movements (Davis, 1968). In backward walking, especially up an inclined plane, the curled tail is rhythmically applied to the substrate to help the animal move (Kovac, 1974). The abdomen has five similar segments and two terminal ones which have been modified to form the tailfan. Each anterior segment contains a massive set of axial phasic flexor and extensor muscles, which control rapid abdominal tailflips, and a superficial set of tonic muscles which control posture (Fig. 1B,C). Each anterior segment also contains a pair of appendages called swimmerets that are homologous with the walking legs (Fig. 1C,D). These are the exclusive motor structures in those ganglia.

In the 6th abdominal segment, the endopodite and exopodite of the swimmerets are expanded into the uropods; a 7th segment has become the telson. Together, these structures form the tailfan (Fig. IE). The alterations in the exoskeleton of the terminal segment have had important consequences for both the muscles and sensory structures. The greatly enlarged surface area of the tailfan is covered with sensory hairs (and presumably functions as a sensory organ), although it is not as specialized in this regard as are the cerci of some insects. The muscles have also been considerably modified. Comparative work on the telson muscles and their innervation is being carried out within (e.g. Dumont & Wine, 1983) and among species (Paul, 1981).

Each anterior segment is controlled by a single ganglion (Fig. 2A) which communicates with neighbouring ganglia via paired connectives and with the periphery via three paired nerves or roots. The most anterior (or 1st) is a mixed sensory/motor nerve that mainly innervates the swimmerets; the 2nd nerve is also mixed and innervates the extensors and the remaining surface of the segment. The most caudal, or 3rd nerve, is the purely motor nerve to the flexor muscles (not shown), which is cleanly split into tonic and phasic branches.

The supracellular organization of a middle abdominal ganglion was studied by Skinner (1984a,b) based on serial sections of plastic embedded material. She provides the following general picture (Fig. 2). All cell bodies are located in the ventral rind, dorsal to which are the neuropile, connective axon tracts and commissures. Twelve axon tracts and four commissures are discernible; all but three tracts and one commissure are named for similar and possibly homologous features seen in orthopteran insects (Gregory, 1974; Pipa, Cook & Richards, 1959). Five neuropile regions are also named. The horseshoe neuropile is largest and most ventral, and is named for its appearance in the horizontal plane; the open part of the horseshoe is anterior. The lateral neuropiles are next largest and form large bulges near the 1st and 2nd roots. The anterior and posterior midline neuropiles are very small, ventral regions; the remaining interstices between tracts and commissures are filled with tract neuropile. In the light microscope, each neuropile region (except the two midline neuropiles which are not differentiated from one another) displays a characteristic texture which can be resolved by electron microscopy to differences in the size and distribution of profiles, differences in concentration of dense core vesicles and certain inclusion bodies, and whether or not a glomerular organization is present.

Each anterior abdominal ganglion contains the cell bodies of approximately 300 pairs of neurones plus a small number of unpaired cells. Cobalt backfill studies (unpublished) show that approximately 100 of the neurones are efferents, 100 others project axons into the connectives, and the remaining 100 are local interneurones. In future work, it will be important to place individually identified neurones in the context of the newly defined, supracellular organization. That will be informative in two ways: it will indicate the extent to which tracts and neuropile regions have functional significance, and should provide insight into the types of synaptic interactions we can expect from a given neurone.

The last abdominal ganglion is a fusion product of two ganglia (Fig. 3C,D), with a cluster of neurones near the posterior intestinal nerve that innervate the gut. (The gut neurones have no obvious homologues in anterior ganglia, J. P. C. Dumont, personal communication.) Despite fusion, the terminal ganglion contains about the same number of neurones as a middle ganglion, perhaps because some descending interneurones and the efferents to one set of appendages have been lost. In addition, any remaining neurones that would have projected to the next anterior or posterior segment are classed here as intraganglionic interneurones.

Differences sometimes occur in the numbers of homologous neurones found in different ganglia (e.g. Mittenthal & Wine, 1978). It has been shown in grasshoppers that such differences can arise either because a homologue develops a sufficiently different structure to be no longer recognizable or because the cell is deleted (Bate, Goodman & Spitzer, 1981).

More than 600 neurones in the abdominal nervous system are identified, although in some cases the identifications are not unique but instead consign the cell to a grounds of several cells which cannot be distinguished except by their peripheral terminations; which are rarely mapped. Identified neurones are listed in Table 1. Only a few motor pools are completely mapped: about 90% of the cells in the abdominal CNS are unidentified; for interneurones of the anterior abdominal ganglia the figure is more than 95 % unidentified.

Composite soma maps of identified neurones in the 3rd and 6th ganglia are shown in Fig. 3. Because the identified cells tend to be the largest cells in the ganglion, the maps give an exaggerated impression of the proportion of neurones that have been identified. The problem of studying integrative mechanisms at the unit level in a ganglion of this size is greatly understated by counting only cells that originate in the ganglion. Integration occurs in the neuropile, which contains terminals from interneurones and afferents that outnumber the neurones originating in the ganglion by at least ten to one. Some projecting interneurones cause special complications since they have input and output sites in several ganglia (Hughes & Wiersma, 1960) and can participate in integration for weeks even though they lack their cell body (Krasne & Lee, 1977). Another potential complication is the claim that significant post-hatching cell addition occurs in crayfish central ganglia (Roth & Suppes, 1973). Since preliminary evidence suggests that all large efferents and interneurones are present shortly after hatching (e.g. Kennedy, 1974), cell addition may occur preferentially in local interneurones.

Neurotransmission within the abdominal CNS is beyond the scope of this review. However, some interesting new findings are that (1) proctolin elicits swimmeret beating when perfused through the ventral artery (B. Mulloney, personal communication); and (2) octopamine and serotonin are probably present (Kravitz et al. 1976; Eloffson, 1983; Beltz & Kravitz, 1983), and have opposite effects (serotonin inhibits and octopamine excites) on the lateral giant escape response (Glanzman & Krasne, 1983).

The cell bodies of all exteroceptors and most proprioceptors are located in the periphery. The only known mechanoreceptors with cell bodies in the abdominal central nervous system are the non-spiking stretch receptors described by Heitler (1982), although the cord stretch receptors (Wiersma & Hughes, 1961; Grobstein, 1973) also presumably have central cell bodies. The most common exteroceptive mechanoreceptors are the hair cells. These cells are typically dually innervated and some of them are directionally sensitive to waterborne stimuli (Wiese, 1976). Four types of hairs have been found on the telson, some types in reproducibly characteristic positions (Fig. 4A,B). Hair receptors on the telson are also frequency selective: receptors have been found which respond well to water movements as slow as 0·2 Hz and have a fairly flat frequency threshold curve, while others appear to be acceleration sensitive and have an optimal response at about 80 Hz (Plummer, Dumont & Wine, 1983, and Fig. 4C). The receptors which respond to very low frequencies are especially interesting because their spontaneous firing is suppressed by high frequency stimulation. This suppression could be achieved by efferent inhibition, mechanical uncoupling, or summation of hyperpolarizations (either endogenous or mediated by peripheral, interafferent inhibition) that might be produced by antiphase movements. Preliminary correlations of physiologically characterized receptors with sensory hair types reveal that long hairs and feathered hairs (Fig. 4B) are innervated by low-pass receptors.

The muscle receptor organs (MROs) are the most thoroughly studied proprioceptors and four central effects have been documented (Fig. 5), all of which act to silence the MROs (negative feedback). So far, it has been documented that the MROs excite their own peripheral inhibitors (Eckert, 1961), the peripheral inhibitors of phasic flexors (Wine, 1977b), and both phasic and tonic extensor motor neurones (Fields, 1966; Wine, 1977b). The entire central course of the muscle receptor axons has been worked out by Bastiani (1981). Tonic and phasic axons have an overlapping distribution and are indistinguishable on structural grounds. The MRO terminals in the 6th abdominal ganglion make characteristic ‘J’ structures and innervate numerous cells bilaterally (Fig. 5E; Bastiani, 1981).

All known receptors in the abdomen are listed in Table 1. Many types of receptors have almost certainly been missed; receptor hunting is somewhat out of fashion, and for the few active experts the specialized appendages offer richer grounds, since it seems unlikely that new types of receptors will be found in the abdomen. However, it will be useful if receptor types found elsewhere are also found in the abdomen, since it will be easier in the abdomen than elsewhere to trace their central connections. So far as is known, all receptors are cholinergic (Barker, Herbert, Hildebrand & Kravitz, 1972). Orthograde degeneration is rapid for all receptor axons tested (Bittner & Johnson, 1974) with the sole exception of the muscle receptors, whose isolated axons persist for many weeks (Bittner, 1977).

Almost all axial abdominal motor neurones have been identified (Table 1). Motor neurones have multiterminal endings on many muscle fibres, and the motor fields overlap so that each fibre is polyinnervated. All axial muscle fibres also receive peripheral inhibition. Neuromuscular synapses are typically stable or facilitating. An exception is the depression-prone neuromuscular synapse of the flexor motor giant (Bruner & Kennedy, 1970). The excitatory transmitter of the motor neurones is probably glutamate (Takeuchi & Takeuchi, 1974); the inhibitory transmitter (in lobsters) is gamma-aminobutyric acid (GABA) (Otsuka, Kravitz & Potter, 1967). Evidence is accumulating that the pentapeptide proctolin is co-released by three of the five tonic flexor motor neurones (Bishop, Wine & O’Shea, 1984). As in insects (Adams & O’Shea, 1983) proctolin appears not to affect the amplitude of synaptic potentials; unlike in insects it appears to enhance tension amplitude without affecting tension duration (C. A. Bishop & F. Nagy, personal communication).

Less is known about the innervation of the swimmerets and their homologues, the uropods. A cell count based on extracellular nerve recordings and intramuscular recordings has been made for the uropods (Larimer & Kennedy, 1969), which indicates that at least 27 motor neurones and up to 5 peripheral inhibitors are involved. Motor pools were visualized with cobalt backfills (Reichert, Plummer & Wine, 1983b); these revealed about 42 central cells with axons in the 2nd and 3rd roots of the 6th ganglion, but some of those might be afferents. Cell by cell identification of efferents to the appendages are now being carried out by W. J. Heitler, M. Hisada and B. Mulloney (personal comments). For an extensive review of crustacean neuromuscular systems see Govind & Atwood (1982).

Our understanding of the processing of sensory information by the crayfish abdominal nervous system is still at a preliminary stage. Although much is known about mechanosensory receptors and primary sensory interneurones, most (and perhaps all) behavioural systems require converging input from many sensory interneurones before they are activated, and this has made it difficult to trace information through the nervous system. I n the isolated abdomen, long sensory-motor pathways that pass through the rostral nervous system are eliminated, and the focus is on shorter pathways that engage the abdominal motor systems directly. Even so, only one behavioural system, the LG-mediated escape system, has been traced from receptors to muscles, and even that system is understood only in outline (Fig. 6). In an attempt to quantify the convergence required to activate motor systems, we have been using intracellular techniques to do a cell by cell study of interganglionic sensory interneurones, many of which Wiersma and his colleagues had previously catalogued with extracellular methods (Hughes & Wiersma, 1960; Wiersma & Hughes, 1961; Wiersma & Bush, 1963).

Projecting sensory interneurones have been studied most intensively in the 6th abdominal ganglion. This ganglion innervates the tailfan, which appears to be specialized for sensory reception, and most processed sensory information leaves the ganglion via a small set of paired interneurones with cell bodies in the ganglion. We used cobalt backfills to obtain an estimate of about 50 pairs of interneurones, and then began studying these cells individually, using Lucifer Yellow injections to identify each cell and intracellular recordings during natural stimulation to establish its physiological profile (Sigvardt, Hagiwara & Wine, 1982).

Twenty-eight sensory interneurones have been identified so far (Figs 7, 8). In addition, several other projecting interneurones have been found which do not respond reliably to sensory stimulation. These include the lateral giant, ‘corollary discharge’ interneurones that are fired by the giant axons (including some neurones that produce primary afferent depolarization, PAD) and unidentified, ‘higher order’ interneurones. Some of the latter, as well as some of the interneurones that we have classified as sensory, may function as ‘command’ neurones for the postural or swimmeret system (Wiersma & Ikeda, 1964; Kennedy, Evoy, Dane & Hanawalt, 1967; Miall & Larimer, 1982a; Larimer & Jellies, 1983). We have not searched diligently for pre-motor axons which might have axons in the dorsal cord, and our techniques lead us to under-sample neurones which have small axons and poor responses to sensory input. Therefore, many of the projecting interneurones in the terminal ganglion which remain unidentified are likely to be non-sensory.

In general terms, what we have learned so far is as follows (Fig. 7).

(1) Each identified sensory interneurone has a unique set of anatomical and physiological properties.

(2) Primary sensory interneurones (cells close to and highly responsive to primary afferent input) can be either conventional (i.e. having non-spiking dendrites and a single spike-initiating zone), or’pan-spiking’ (i.e. having highly electrogenic dendritic membrane and multiple spike-initiating zones). Pan-spiking neurones always have multi-segmental receptive fields (Sigvardt et al. 1982).

(3) We have scant evidence for hierarchical synaptic organization among the projecting interneurones in the last ganglion. Some projecting interneurones weakly excite others (Zucker, 1972), but this interaction is usually reciprocal. We have no evidence for strong coupling such as is found among descending interneurones from the brain (Glantz, 1978). At present we consider the projecting cells as mainly independent, parallel channels.

(4) Some projecting interneurones are influenced by local sensory interneurones. Only one such pathway has been identified (Reichert et al. 1983a,b), but it remains possible that the majority of polysynaptic inputs are via local interneurones.

(5) Post-synaptic inhibition evoked by escape commands has been detected in only two interneurones, and in one of these it has been shown, in part, to be caused by some of the same inhibitory neurones that cause presynaptic inhibition of the afferent-to-interneurone synapse (Kirk & Wine, 1983; M. D. Kirk, unpublished data).

(6) Interneurones sensitive to water movements can be differentiated by their responsiveness to frequencies between 0·5 Hz and 400 Hz into ‘low-pass’, ‘high-pass’ and ‘broad-band’ cells. Interneurones responsive to low frequencies are ‘inhibited’ an high frequencies: inhibition involves the peripheral suppression of firing in low frequency afferents mentioned earlier; whether this is abetted by postsynaptic inhibition of the interneurones has not yet been established (Plummer et al. 1983).

(7) So far, we have only modest evidence that the sensory interneurones at this level are abstracting information, rather than simply summing it. As a first approximation, we can identify three classes of exteroceptors: (a) low-pass headward, (b) low-pass tailward and (c) high-pass. We thus have examples of interneurones which receive information from each class selectively, from (a) and (b), and from all three. The only spatial interaction documented so far is lateral inhibition among low-pass, headward-sensitive units (see next section).

If the fluid surrounding the tailfan is moved gently at frequencies as low as 0-2 Hz, seven pairs of sensory interneurones in the last ganglion fire vigorously for as long as the oscillations continue. Three pairs of these non-adapting interneurones do not respond to touching of the tailfan and are actually inhibited by high frequency water movements (i.e. >200 Hz). If the oscillations are directed along the rostro-caudal axis of the horizontal plane, one interneurone fires vigorously only in response to headward movements, one exclusively to tailward movements, and the third is bidirectional. The excitatory receptive fields of these three interneurones are restricted to the ipsilateral uropods and telson. When sensory roots within their excitatory sensory field are stimulated electrically the cells receive mixed excitation and inhibition and usually fire only one or two impulses even to intense shocks. We call these three interneurones ‘slosh’ cells. Four other directional interneurones have been found which differ from pure ‘slosh’ cells in that they also respond to touch, adapt slightly, have either bilateral, contralateral or multisegmental receptive fields, and fire short bursts of impulses to sensory nerve shocks. We call these ‘slosh-touch’ interneurones. Prolonged searches of the cell body layer, the neuropile and the connectives have not revealed evidence for more than seven interneurones in the ‘slosh’ and ‘slosh-touch’ category (M. R. Plummer, G. Hagiwara, H. Reichert, K. A. Sigvardt, K. Wiese & J. J. Wine, in preparation). Each of these interneurones has an axon which runs the entire length of the abdomen and presumably terminates in the brain.

A local, non-spiking ‘slosh neurone’ has also been found (Reichert et al. 1983a,b). This neurone is excited by headward water movement across the half of the tailfan ipsilateral to its cell body. It projects across the midline and inhibits both the headward and the bidirectional slosh cells on the other side. Since a bilateral pair of such cells exist, the system displays feedforward, lateral inhibition, and as such the subsystem should display common-mode attenuation. The sensory subsystem as we now understand it is shown in Fig. 9.

It is now technically possible to record from all 14 axons while exposing the tailfan to various stimuli. Individual neurones display moderate directional selectivity to an oscillating probe (Wiese & Wollnick, 1983). However, without knowing either the destination of the 14 axons in this system, or the animal’s response to activity in the system, we cannot hope to discover what information is being extracted. We could, of course, record from all 14 axons simultaneously and identify patterns that are reliably evoked by various stimuli. But we would have no assurance that the animal looks at the output of these interneurones in the same way: it may compare only a subset of the cells, or it may integrate the output of this system with other information.

In a system as highly differentiated as the abdominal CNS, few generalizations about anatomical organization are likely to be absolute. Nevertheless, the system is highly ordered, and such order is of enormous practical significance to investigators, quite apart from any issues it might pose for developmental biologists. A brief summary of generalizations is as follows.

Afferents rarely cross the midline, rarely synapse on motor neurones, and do not directly inhibit their postsynaptic targets. All exteroceptive afferent cell bodies are located in the periphery, and their axons usually terminate in the ganglion where they enter, but some run anteriorly for several segments or posteriorly for one segment. Afferents continue to be added during postembryonic development and eventually outnumber efferents by about 100 to 1. Afferents are cholinergic, and show rapid, orthograde degeneration. The receptors of the muscle receptor organs are atypical in that their axons bifurcate and are distributed along the entire length of the CNS, their terminals cross the midline in the last abdominal ganglion, they synapse directly with efferents, they display very slow orthograde degeneration, they are fixed in number at hatching and they are postsynaptically inhibited in the periphery.

Efferents are fixed in number at approximately 100 per hemisegment, the majority of which innervate the appendage. Motor neurones to axial muscles may decussate, but appendage motor neurones rarely do. Efferents typically have single axons which exit via a root in the ganglion of origin or, rarely, via a root in the next anterior ganglion. Most have single sites for spike initiation. Orthograde degeneration is very slow (weeks to months); putative transmitters are glutamate, GABA, proctolin and acetylcholine.

Interneurones are so diverse that only limited generalizations can be made. They range from small, local, non-spiking neurones to ‘pan-spiking’ projecting interneurones which run the entire length of the nerve cord and can initiate spikes at many sites within each of many ganglia. Most interneurones which have axons have only one, which either ascends or descends in the connectives; about 18 % have bifurcating axons (Jones & Page, 1983). Commissural interneurones are intraganglionic interneurones which have a single axon that projects via a commissure to the other side of the ganglion (e.g. Reichert et al. 1983a,b). Projecting interneurones show slow orthograde degeneration (Wine, 1973). Beyond the putative use of GABA by some inhibitory interneurones (Ochi, 1969) the transmitters used by crayfish interneurones are entirely unknown: the only well-characterized excitatory intemeuronal synapses are electrical (e.g. Furshpan & Potter, 1959; Zucker, 1972; Roberts et al. 1982).

Integration begins with receptors which are rather narrowly tuned to specific stimulus features such as the direction and frequency of water movement or the bending of a single joint. These features are maintained and in some cases sharpened at the level of primary projecting interneurones. Receptor input is heavily modulated at the very earliest stages: both efferent control and presynaptic inhibition have been documented — the latter may be ubiquitous. Not all sensory interneurones are discriminitive; some respond to a very broad frequency range and have receptive fields that encompass half the animal’s body surface. It is believed that specific sets of sensory interneurones converge on higher order interneurones which in turn make divergent connections via premotor ‘driver’ neurones to motor neurones. In at least one system (the swimmerets) endogenous oscillations can be generated by the premotor and motor neurones; these are coordinated by interganglionic connections (some made by motor neurone branches) and are controlled by interganglionic ‘command’ neurones (Wiersma & Ikeda, 1964). Integration at the level of interacting behaviour patterns, details of how a sensory stimulus recruits a specific pattern, of how spatial and temporal patterns of movement are generated by cellular interactions, and of how an activated motor system acts back on the animal’s sensory systems have so far been worked out only for escape behaviour. The escape system is described briefly in the next section.

Crayfish escape the rapid strikes of predators like racoons and fish by rapid backward swimming. The shock wave produced by a forceful movement toward the crayfish is detected by hair receptors that converge via sensory interneurones onto either the lateral giant (LG) or medial giant (MG) interneurones whose giant axons (the largest by far in the CNS) run the entire length of the nerve cord and drive abdominal flexor motor neurones (see Fig. 6) and, at the same time, inhibit the receptors, motor neurones and muscles of the abdominal extensor system. Thus, the swimming powerstroke is produced at short latency by contractions of the massive dominal flexor muscles. The flexion powerstroke is completed within about 30 ms. Even before flexion is completed, virtually all of the elements in the flexion circuit, from receptors to muscles, are also inhibited by long-lasting inhibitory postsynaptic potentials (IPSPs) that are triggered by the command signal and persist for the duration of flexion and re-extension. Re-extension is a chain-reflex caused by sensory feedback from flexion (Reichert, Wine & Hagiwara, 1981). It is triggered by the coincidence of sensory influx and the termination of the short-duration IPSPs produced by the flexion command neurones in extensor motor neurones, extensor muscles and the extensor muscle stretch receptors. In summary, the first tailflip of a short-latency escape response involves five steps:

1. sensory triggering of the flexion command axons,

2. excitation of the flexor muscles,

3. short-lasting inhibition of the extensor system,

4. delayed, long-lasting inhibition of the flexor system and

5. delayed, feedback excitation of the extensors.

Usually, the initial tailflip is immediately followed by a series of tailflips occurring at a frequency of 10–20 eyeless⁻¹. This is escape swimming. Swimming is not a reiteration of the initial tailflip. It does not involve the giant axons, which are in fact inhibited during swimming, and the phase relations of extension and flexion are reversed so that extension leads flexion, which follows at short and relatively constant latency. Thus, when a crayfish swims slowly it keeps its abdomen tucked while coasting and then rapidly extends and flexes it again at the next cycle. This means that, during swimming, extension cannot be a flexion-induced reflex. In fact, swimming, like all cyclic behaviour patterns studied so far, is under the control of a central pattern generator (CPG), and the extension reflex is inhibited during swimming.

The swimming CPG is turned on by the same stimulus that causes the giant axons to fire. The CPG’s onset is always considerably delayed, so that the first tailflip occurs before the CPG becomes active (Reichert & Wine, 1982, 1983).

In summary, the sequence of events in a crayfish rapid escape response is shown in Fig. 10. It may be useful to clarify some of the predictions of this model: It should be possible to obtain either a single command-mediated tailflip in isolation or the CPG in isolation; direct stimulation of the command neurone should produce only a single flexion and extension; and interference with flexion should interfere with reextension. All of these predictions have been verified.

Orientation of the escape response is achieved in two ways. The initial response has only two variations: a rapid backward dart or a forward pitching movement. Each movement is commanded by a separate pair of giant axons with independent decision processes. The initial decisions ignore much detailed information about the stimulus, such as its laterality, and produce laterally symmetrical responses. The delayed activation of the CPG is accompanied by greater finesse in movement; the CPG codes at least the laterality as well as the rostro-caudal direction of the stimulus, giving a minimum of four distinct response modes.

Escape responses are only one aspect of a crayfish’s behaviour and must be integrated with everything else it does. This means escape responses influence and are influenced by other behaviour patterns, including past behaviour. Once triggered, escape has the highest priority of any behaviour pattern in the animal’s repertoire: it will inhibit or override any competing pattern. However, the excitability of the triggering process is modulated by a wide range of influences so that in some circumstances it is virtually impossible to elicit escape (Krasne & Wine, 1975; Wine, Krasne & Chen, 1975).

In the preceding overview, the details of transduction, excitation-contraction coupling, synaptic action and connectivity patterns among the receptors, interneurones, motor neurones and muscles were deliberately ignored to give a broad perspective of the behaviour pattern. To an ever increasing extent, the pattern can be explained in terms of such details and a comprehensive review was recently published (Wine & Krasne, 1982).

In this section, I review selected recent findings about just one aspect of the escape response: the neural connections that mediate the initial patterns of flexion when the giant axons fire. After summarizing the ‘basic’ giant axon-giant motor neurone flexion circuit, I review recent evidence for sets of parallel connections with non-giant neurones which appear useless or maladaptive. To explain these connections it has been hypothesized that the giant axon circuitry evolved from the more complex nongiant circuitry that still controls most tailflips, and it has been proposed that all connections which the giants once had with the non-giant circuits have been maintained unless selected against. Connections detrimental to the motor patterns produced by the giant axons have either been lost, weakened or inhibited (Wine & Krasne, 1982; Krasne & Wine, 1984). Because the effects of weakened or inhibited synapses can be overridden, at least in principle, it has gradually emerged that the giant axon motor circuits contain numerous mechanisms that might enable them to produce altered motor patterns to the same command (e.g. Kramer, Krasne & Bellman, 1981a; Miller et al. 1984). Whether and under what circumstances they might do so is a question that is now being investigated.

In each of the seven abdominal segments, a pair of giant flexor motor neurones innervate almost all of the fast flexor (FF) muscles in its segment via powerful but depression-prone synapses. The motor giants are in turn fired monosynaptically by rectifying electrical synapses from the giant interneurones. A striking difference exists in the pattern of connections between the MG and LG axons and the motor giants: the MG axons synapse with the motor giants in every segment, whereas the LG axons synapse with the motor giants in the anterior three segments but not in the posterior four segments. The behavioural consequences of that arrangement are shown in Figs 11 and 12A.

The first level of complexity added to that simple arrangement is as follows. In the 2nd and 3rd abdominal ganglia, the LGs and MGs fire a pair of large interneurones, termed 12 and 13, which descend to the terminal ganglion where they synapse on many neurones, including the motor giants to the telson flexor muscles (Kramer et al 1981a; Kramer, Krasne&Wine, 1981b; Fig. 12B). The summated EPSPs of 12 and 13 are sometimes sufficient to fire the motor giants (MoG) (Fig. 12D). This pathway is redundant with the basic connectivity pattern of the MGs, but for the LGs the effect would be disruptive and would cause its motor output pattern to resemble the MG one. However, the telson motor giants are strongly inhibited within a few ms of MG and LG firing (Kramer et al. 1981a,b; Dumont & Wine, 1983; Fig. 12E,F). This inhibition comes too late to interfere with the direct input from the first MG impulses, but is early enough to cancel the effect of the input from 12 and 13. As far as we can tell, the net effect of all of this additional wiring is to leave the basic motor pattern unchanged (Fig. 12C). All the elements shown in Fig. 12C have been identified (Fig. 13).

A second level of complexity is that, in each ganglion, the motor giants are in parallel with up to nine non-giant, fast flexor (FF) motor neurones which, in aggregate, co-innervate the same muscles as the motor giants (Fig. 14). In other words, each fast flexor muscle fibre receives input from both a motor giant and one or more non-giant motor neurones. In the 2nd and 3rd ganglia, the LGs and MGs make equivalent, suprathreshold connections to all the FF motor neurones via the intervening’driver’ interneurone called the segmental giant (Roberts et al. 1982; Fig. 13A). These connections reinforce the effects of the LG and MG commands. However, in the posterior four segments, firing of the FFs would disrupt the basic motor pattern of the LG-MoG network. The obvious solution to this problem would be to eliminate the synapses between the LGs and SGs in the posterior four segments, just as the synapses to the MoGs were eliminated. However, the crayfish has not adopted the obvious solution. Instead, perhaps because the SGs are important for exciting other neurones, the LGs and MGs maintain equivalent, suprathreshold connections to the SGs the whole length of the abdomen, but the LG pattern is preserved because in the posterior segments the strength of the synapses between the SGs and FFs is below threshold for firing the motor neurones (Miller et al. 1984; J. P. C. Dumont & J. J. Wine, in preparation; Fig. 14).

Even though the SG-to-FF synapse is usually subthreshold in posterior segments, single LG impulses sometimes fire FF motor neurones (Miller et al. 1984) and multiple impulses, which often occur with natural stimuli (Wine & Krasne, 1972), summate to fire the cells with much higher probability (Miller et al. 1984). This suggests that some activity in the posterior fast flexor muscles should occur during tailflips mediated by multiple LG impulses. Such activity would disrupt the forward pitch of the animal and cause the trajectory to resemble an MG-mediated, backward tailflip. One way to quantify the tendency of the animal to pitch forward is to measure the angle of the body, relative to the substrate, at the termination of initial flexion (Fig. 15). For 37 naturally evoked, LG-mediated tailflips this angle was 107 ± 12°, whereas for 10 naturally evoked, MG-mediated tailflips the angle was 26 ± 9°. When the LGs were fired directly via implanted electrodes, single shocks gave a pitch angle of76± 15° (N = 36), whereas triplets (at 200 Hz) reduced the angle to 64 ± 12° (N = 11) (Fig. 15; G. Hagiwara, L. Miller & J. J. Wine, unpublished). The decreased angles seen with, central stimulation may mean that in these instances, as in physiological preparations, some FF motor neurones in posterior ganglia are being recruited by the LG spikes. How then are we to explain the greater angle and relatively invariant form of naturally-evoked, LG-mediated tailflips?

It appears that another level of complexity is required to explain the spatial pattern of LG tailflips. We have circumstantial evidence that, during LG-mediated tailflips, the flexor muscles in the posterior segments, and especially in the telson, are inhibited by early and strong firing of the flexor peripheral inhibitors^* (FIs), while in anterior segments FI firing is delayed and serves mainly to limit the duration of the response. The evidence for this is as follows (Fig. 16). When a crayfish is tapped on the abdomen (an adequate stimulus for an LG tailflip) a powerful sensory barrage is produced in the peripheral inhibitory neurones of the fast flexor muscles (FI neurones). In fact, shocks to sensory nerves in the last ganglion typically fire at least one of the FIs to the telson flexor muscles well below the threshold for LG activation (J. P. C. Dumont & J. J. Wine, in preparation and Fig. 16C). The FIs are also excited, with a delay, by central pathways activated by the giant axons (Wine & Mistick, 1977). This delay is shortest in posterior ganglia (Uyama & Matsuyama, 1980; Fig. 16D) and becomes very short in the telson (Fig. 16E). Furthermore, although the MGs and LGs have equivalent effects on the FIs in anterior ganglia (Wine & Mistick, 1977), only the LGs strongly excite the FIs to the telson flexor muscles (J. P. C. Dumont & J. J. Wine, in preparation; Fig. 16E).

All these findings are consistent with early and strong inhibition of flexor muscles in the posterior segments and telson during LG-mediated tailflips. Peripheral inhibition in these segments might be very effective. Since the motor giants never fire in segments 4 and 5 during LG tailflips (because they receive no input from LGs in those segments) and rarely fire in the telson segments (because the polysynaptic pathway to them from the LGs is inhibited, Fig. 12), peripheral inhibition only needs to counteract the effects of the FFs. Because the FFs typically require summation to fire in these segments, they are probably firing at least several ms later than they fire in anterior segments, and hence are more vulnerable to peripheral inhibition. Thus, although it has not yet been demonstrated directly, we have considerable circumstantial evidence that peripheral inhibition of the flexor muscles helps maintain the spatial motor pattern of LG tailflips.

To summarize: an LG tailflip has its characteristic spatial pattern because the segments innervated by ganglia 4, 5 and 6 do not flex, whereas the segments innervated by the first three ganglia do (Fig. 11). We believe four different features jointly contribute to the production of the spatial pattern. At the most basic level, the differential activation of the flexor muscles in the anterior and posterior segments of the abdomen occurs because the motor giants are monosynaptically excited by the LGs only in the anterior segments. However, the LGs indirectly excite telson flexor MoGs and non-giant FFs in ganglia 4 to 6. The disruptive consequences of these potential flexor outputs in the posterior segments are prevented by central inhibition of the telson MoGs, by a weakened excitatory pathway to the FF motor neurones in the posterior segments, and by earlier and stronger peripheral inhibition of the fast flexor muscles in posterior segments. These features are summarized in Fig. 17.

We are attempting to uncover principles of neural integration by concentrating on a single invertebrate species. For the crayfish, we have found it necessary to focus still more narrowly: our anatomical and neurophysiological studies are confined to the abdominal nervous system, and our neurobehavioural analysis deals almost exclusively with escape behaviour. Within these narrow confines, our studies have provided neural underpinnings for several basic ethological concepts, such as innate releasing mechanisms and fixed action patterns. The neural networks we are constructing are more complicated than many people expected, and our progress is therefore slower. The complexity might be an artifact of our inadequate understanding, and we may eventually discover a much simpler way of expressing the circuitry. We doubt that. It seems more likely that our present picture is, in fact, oversimplified.

Why do simple behaviour patterns require complex neural networks? We do not yet know, but one possibility for the apparent mismatch is that conventional descriptions of behaviour may overlook many nuances that have evolved to ensure optimal performance in a wide range of conditions. This leads to simplified descriptions of behaviour, and causes us to underestimate the complexity of the neural circuits that control behaviour.

The timeliness of this review was made possible by the generosity of many colleagues who shared pre-publication manuscripts with me. I am deeply grateful to M. Bastiani, C. Bishop, J. Dumont, G. Hagiwara, M. Kirk, A. Kramer, F. Krasne, L. Miller, B. Mulloney, F. Nagy, M. Plummer and K. Skinner for permission to use previously unpublished data.