The Segmental Giant Neurone of the Hermit Crab Eupagurus Bernhardus

An enigmatic feature to emerge from recent studies on the neural circuitry of the crayfish escape tail-flip is the segmental giant neurone (SG) (Kramer, Krasne & Wine, 1981; Roberts et al. 1982; Heitler & Darrig, 1986). This neurone receives powerful input from the through-conducting giant fibres (GFs), and makes outputs to the fast flexor excitor motor neurones (FFs) of the abdomen. Both input and output are made via electrical synapses. Only the motor giant (MoG), a highly specialized FF, receives its major input directly from the GFs. Thus, centrally the SG acts as a 1:1 iterneuronal spike relay between GFs and FFs. However, anatomically, the SG has distinct similarities to a return-stroke motor neurone of the swimmeret system, although its axon is blind-ending and has no peripheral output (Heitler, Cobb & Fraser, 1985; Heitler & Darrig, 1986).

Hermit crabs also have an escape system which is used for withdrawal into the shell (Wiersma, 1961; Chapple, 1966). This is mediated by a single pair of GFs which appear to be homologues of the crayfish medial giant neurone (Chapple & Hearney, 1974; Stephens, 1985). The FF population of the hermit crab is largely homologous to that of the crayfish (Mittenthal & Wine, 1978) and there is one large flexor neurone which receives rectifying electrical input from the GF, and which is probably homologous to the crayfish MoG (Umbach & Lang, 1981).

The aim of this report is to re-examine the system of giant neurones in the hermit crab, and study the properties of any neurone that may be homologous to the crayfish SG. This is of interest because hermit crabs differ from crayfish in two major respects. First, the abdomen is massively modified, having lost most of its hard cuticle and become asymmetrically twisted in order to fit into the mollusc shell. Second, the swimmerets on the right side of the animal are vestigial. Thus adult hermit crabs differ from crayfish in the two systems in which there is evidence for SG involvement.

Hermit crabs (Eupagurus bemhardus) were collected locally from St Andrews bay and kept in tanks of circulating sea water. Prior to dissection a crab was removed from its mollusc shell by carefully crushing the shell in a vice. The abdomen was cut from the anterior body, and the abdominal ventral nerve cord was dissected free from ganglia 3-5. The nerve cord was pinned, dorsal surface upwards, on a Sylgard platform, and submerged in saline (Atwood & Dorai-Raj, 1966; but with Tris substituted for NaHCO₃). The saline was kept chilled to 8°C.

Hook electrodes were used to record extracellularly and to stimulate the anterior and posterior connectives. Various combinations of hook, pin and suction electrodes were used to record extracellularly and stimulate the first and third roots of the fourth abdominal ganglion. Intracellular recordings were made with glass microelectrodes (resistance 20–40 MΩ) filled either with potassium acetate or 5% Lucifer Yellow dissolved in lmoll^-1 lithium chloride. Penetrations were all made from the dorsal aspect of the ganglion into axonal or neuropile processes. Prior to penetration, the ganglionic sheath was softened by removing the saline and then applying a few specks of protease (Sigma Type XIV) for about 30 s, before washing off with saline. Sometimes a small tear was made in the ganglionic sheath to further facilitate penetration. In some experiments a computer-based signal averager was used to reduce noise (mainly spontaneous activity) in the response to stimulation. Neurones were injected with Lucifer Yellow using 0·5s negative current pulses of 10-100 nA delivered at 1 Hz for 30- to 180-min periods. Tissue was fixed in 4 % formaldehyde in saline, dehydrated in alcohol and cleared in methyl salicylate. Preparations were examined in whole mount with epi-illumination (Zeiss filter set 06), and drawn using a camera lucida attachment.

Initial experiments were performed using extracellular stimulation and recording techniques to define the major elements present in the system. The fourth ganglion (G4) was isolated (Fig. 1), and electrodes were each placed on the left and right anterior connectives (1AC, rAC), left and right posterior connectives (1PC, rPC), left and right first roots (1RI, rRI ; these innervate the swimmerets, which are vestigial on the right side) and left and right third roots (1R3, rR3 ; these innervate the abdominal slow and fast flexor muscles). These various nerves were stimulated in turn with gradually increasing intensity, while recording from the other nerves (Figs 2, 3).

There was some variability between preparations, and the following descriptions represent the responses observed most frequently in 54 experiments. Units are named after their apparent crayfish homologues for convenience of reference. The justification for this is given in later sections, in which the results of experiments where neurones are recorded intracellularly and stained with Lucifer Yellow are described.

A low threshold unit which was conducted through the ganglion, and which did not initiate any activity in Rl or R3, was often found in both connectives (Fig. 2A). As the intensity of stimulation was raised, the somewhat larger spikes of the GF were recruited at a precise threshold (Fig. 2B,C). This unit was conducted through the ganglion, and recruited a variable number of corollary discharge interneurones (CDIs) in both the ipsilateral and contralateral connectives (Fig. 3A,B). At this threshold a number of units was also recruited in the peripheral nerves. The SG was observed in the ipsilateral Rl, and was reliably recruited at exactly the GF threshold (Figs 2B,C, 3A,B). On one occasion the contralateral SG was also recruited. In the ipsilateral R3, a complex potential containing several FF units and the MoG was recruited at the GF threshold. On the contralateral side the MoG was also usually recruited at this threshold, occasionally accompanied by one or two FF units (Fig. 2B,C). With increasing stimulus intensity beyond the GF threshold, complex potentials were recruited in all nerves (data not shown).

At a low and precise stimulus threshold a complex potential was recruited in the ipsilateral R3 (Fig. 3C,D). This consisted of a variable number of FF motor neurones, and often closely resembled the response initiated by the ipsilateral GF, but without the unit potential of the MoG (compare Fig. 3A bottom trace with Fig. 3C second trace, and Fig. 3B fourth trace with Fig. 3D top trace). The complex potential usually declined in amplitude as the preparation aged, and often disappeared completely by the end of the experiment. Occasionally one or two contralateral FF units were recruited at this threshold. If the stimulus intensity was increased, considerable activity could be recruited in all nerves (data not shown).

The GFs usually had spike amplitudes between 60 and 100 mV, and stained to reveal a large through-conducting axon, with little or no arborization within G4 (Figs 4J, 7D). In a preparation in which two microelectrodes were inserted into rGF (one for recording voltage, one for injecting current; Fig. 4), AC stimulation first initiated several PC spikes including the IGF and some CDIs, and then, at a slightly higher threshold, the rGF spiked. When rGF spiked, a prominent rRl unit, the rSG (see below for identification) was recruited. A hyperpolarizing current pulse which blocked the through-conducting rGF spike invariably blocked the rSG (Fig. 4B). A depolarizing current pulse which initiated an rGF spike also initiated one or more CDI units in both PC (Fig. 4D) and AC (Fig. 4F), and usually initiated the rSG. The only exception was when the rGF spike coincided with a burst of CDI activity resulting from a previous GF spike. Under these circumstances the rSG unit was sometimes not initiated (data not shown). The rGF is thus necessary and usually sufficient to elicit an rSG spike.

In rR3 the large MoG unit was initiated below rGF threshold, probably at IGF threshold (Fig. 4G). When the AC stimulus was increased to recruit the rGF, and consequently the rSG, at least one rFF unit followed the rMoG spike at short latency (Fig. 4H). When the rGF spike was blocked with hyperpolarizing current, the short latency FF unit was also blocked. A longer latency rR3 unit, possibly the flexor inhibitor (FI, see below), was unaffected (Fig. 4L).

In seven out of eight experiments (five right side, three left side) the GFs received no input from R1 at the threshold which recruited ipsilateral R3 activity (SG threshold), and no input from R3. One aberrant preparation in which the rGF received input from rRl is described below.

The SG is identified physiologically as a neurone that has a spike that is normally activated both by connective stimulation at exactly the ipsilateral GF threshold (Fig. 5B,C,E) and by stimulation of the ipsilateral R1 at the threshold that recruits the ipsilateral R3 units (Fig. 5D). In some preparations the GF-initiated SG spike could be seen to arise from a pre-potential which was probably an electrical EPSP (Fig. 5E, see below). Spike amplitude varied widely between preparations (20–90mV). This may be partly due to differences in the quality of penetration, although low-amplitude potentials were sometimes obtained in penetrations with resting potentials in excess of —70 mV which were stable for more than 1 h. Variations may also have been due to differences in the site of electrode penetration. Sometimes a small PSP was apparent upon stimulation of the contralateral GF or R1 (Fig. 5A,G,H).

Neurones with these properties were penetrated eight times on the right side of G4 and 13 times on the left side, and in each case where adequate staining was achieved (14 times) a highly characteristic morphology was revealed (Fig. 6). There is a cell body in the anterior ipsilateral ventral quadrant of the ganglion, which is joined by a very thin neurite to the proximal end of an axon which runs in the ipsilateral Rl. Central to this region the neurone expands into a massive integrating segment, with a prominent anterior-posterior orientation. Finger-like dendrites line up along the course of the ipsilateral GF (Figs 6C, 7D). There is considerable variability in the morphology of the dendritic branching, but no consistent differences in overall size, extent of dendritic arborization or axon diameter were observed between the left and right SGs. The hermit crab SG thus has a similar morphology to that of the crayfish (if the differences in overall ganglionic shape are allowed for), except that it has little or no contralateral arborization.

The SG was reliably driven to spike by the ipsilateral GF. The spikes followed GF stimulation at frequencies up to at least 100 Hz, and when they failed they revealed an underlying EPSP which followed further increases in frequency until the GF spikes themselves failed. Simultaneous penetrations of the SG and ipsilateral GF (Fig. 7) suggest that this interaction is via a rectifying electrical synapse. Hyperpolarizing current injected into the GF blocked the GF spike induced by extracellular stimulation of the connectives, and there was a simultaneous failure of the SG spike, but the hyperpolarizing current did not spread from the GF to the SG. In contrast, hyperpolarizing current injected into the SG did spread antidromically to the GF, and could also block a spike in that neurone. Depolarizing current injected into either neurone did not spread to the other, but it should be noted that in these experiments presynaptic voltage was not monitored, and with the Lucifer-filled electrodes insufficient depolarizing current could be injected to initiate a spike before the electrode blocked.

In only one preparation was a small EPSP observed in the GF on stimulating the ipsilateral R1. In this preparation simultaneous penetration of the ipsilateral SG and GF showed that current of either polarity passed in both directions between the two neurones (data not shown). The GF-SG connection in this preparation was thus non-rectifying, for some unknown reason.

In the crayfish one of the strange features of the SG is that its axon diminishes in diameter as it progresses along Rl, and eventually terminates in a blind ending (Heitler & Darrig, 1986; Heitler et al. 1985). It has not been possible to trace Lucifer Yellow to the termination of the hermit crab SG because Rl is at least 16 mm in length in the hermit crab G4, and in general our Lucifer stains of the neurones of marine crustaceans have for unknown reasons not been as intense as our stains of freshwater crayfish neurones. However, extracellular recordings of the hermit crab SG spike show that it has maximum amplitude at about 4 mm distance from the ganglion, and then gets progressively smaller as the hook electrodes are moved distally along Rl. About 8-10mm from the ganglion the spike becomes indistinguishable from baseline noise, and at this point stimulation of Rl fails to elicit R3 activity. A similar diminution in SG spike amplitude is reported for crayfish (Kramer et al. 1981), suggesting that the hermit crab SG axon terminates within Rl in a similar manner to that of the crayfish.

Motor neurones innervating the abdominal flexor muscles are identified physiologically as neurones which receive an antidromic spike on extracellular stimulation of R3, or anatomically as neurones with an axon in that root. Four classes of such neurones exist in crayfish. The single MoG is directly driven 1:1 by the GFs via a powerful rectifying electrical synapse located at the base of R3. It has a very large cell body contralateral to its axon, and no neuropile arborization within the ganglion. The single flexor inhibitor (FI) does not receive unitary short-latency input at GF threshold, and has a large, contralateral cell body. Slow flexor (SF) motor neurones innervate the superficial flexor muscles, have small cell bodies and do not receive unitary excitatory input at GF threshold. The remaining neurones are the FF excitatory motor neurones, which receive unitary input via the SG at GF threshold, and which have cell bodies which are larger than the SFs, but smaller than the MoG or FI. In crayfish, of the FFs with posteriorly directed axons, four have ipsilateral cell bodies (FPI group) and two have contralateral cell bodies (FMC group). Cobalt backfills of R3 of the hermit crab Pagurus pollicaris (Mittenthal & Wine, 1978) and the anomuran squat lobster Galathea strigosa (Sillar & Heitler, 1985) reveal similar groupings, except that in Galathea there is no MoG, but rather the homologous neurone appears as an unspecialized FF.

In our experiments on hermit crabs we have only examined abdominal flexor motor neurones with axons in the posterior R3. Such neurones have been recorded from and/or stained intracellularly in a total of 21 preparations in which the extracellular records indicated normal functioning of the GF and SG. Of these, four were identified as the MoG, 14 were FFs, two were the FI, and the remaining one was an SF. The justification for this identification, and the characteristic responses to GF and SG activation, are described next.

The MoG has a large axon which is connected to a large contralateral cell body by a relatively large neurite. The only dendritic arborization consists of a tuft of small dendrites towards the posterior part of the ganglion. In this region the MoG comes into close apposition with the ipsilateral SG and GF (Fig. 8). This anatomy allows unambiguous identification of the MoG.

The MoG is reliably driven to spike 1:1 upon activation of the ipsilateral GF (Figs 2, 3, 9, 10; Umbach & Lang, 1981; Stephens, 1985) and, like the SG, follows GF spikes up to very high frequency. The MoG also appears to receive input from the contralateral GF. The evidence for this comes from several experiments in which the connectives were separated so that the GFs could be stimulated independently (Figs 2, 3) and from experiments in which the connectives were stimulated jointly, but the GFs were activated at different thresholds. In most such experiments the MoG spiked at the lowest threshold of connective stimulation which caused activation of either GF. A typical example is shown in Fig. 9A. With increasing AC stimulation an IMoG spike was recruited at precisely the threshold which recruited an rSG spike recorded extracellularly in rRl (trace not shown), which was thus almost certainly the rGF threshold, but which was below the 1SG (and hence IGF) threshold. Thus, if the activation was not in fact caused by the contralateral GF (either directly or indirectly), there must be another connective neurone with a very similar threshold which drives the MoG. However, stimulation below the threshold of both GFs never recruited the MoG, and on the few occasions when stimulating a low-threshold contralateral GF failed to induce an MoG spike, increasing the intensity of stimulation did not recruit the MoG until ipsilateral GF threshold was reached. These observations argue against the possibility of a second neurone. In contrast, a pathway parallel to the GFs which provokes bilateral activity in the MoGs has been reported in the thoracic-abdominal ganglion of Pagurus pollicaris (unpublished observations quoted in Stephens, 1985). Dual microelectrode penetrations of the MoG and contralateral GF in G4 of Eupagurus will be needed to settle this question.

The MoG spike was usually followed by a 5-10 mV depolarizing IPSP, of variable characteristics. Sometimes it was clearly a multi-component potential, as shown by the step increase in amplitude which occurred as extracellular stimulation of the AC recruited first one and then the other GF (Fig. 9C,D), but sometimes the IPSP was only recruited by one of the GFs. The IPSP could block a subsequent GF spike induced by extracellular stimulation and timed to coincide with it. This inhibition closely resembled that impinging on the crayfish MoG with similar stimulation (Wine, 1977). A large antidromic spike was recorded in the MoG upon stimulation of R3 when the microelectrode was placed towards the posterior end of the ganglion (Fig. 9E).

Extracellular stimulation of the ipsilateral Rl caused a short-latency PSP in the MoG which was recruited at exactly the threshold of the ipsilateral SG (Fig. 9F-K). This could not be abolished or reversed with depolarizing current (unlike the longer- latency IPSP described above) but could be attenuated or abolished by superimposing it on the IPSP. Superimposing the Rl-mediated PSP on a GF-induced spike did not block or reduce the amplitude of the latter. It is therefore likely that the potential is an EPSP. However, on no occasion could sufficient current be injected into the MoG to induce it to spike, even when superimposed on this EPSP.

Dual penetrations of the SG and MoG showed that the EPSP is in fact mediated by the SG. A pulse of hyperpolarizing current injected into the SG so as to bracket the moment of extracellular stimulation of Rl abolished the SG spike and the MoG EPSP (Fig. 10A-D). The MoG also received an EPSP upon stimulation of the contralateral Rl, of which the first component was recruited at the threshold which recruited contralateral R3 units, suggesting that it was mediated by the contralateral SG. However, dual penetrations have not been made to confirm this.

Stimulating the ipsilateral Rl sometimes caused a large depolarizing IPSP in the MoG following the SG-induced EPSP, similar to the IPSP following GF activation (Fig. 9D). However, this IPSP was not abolished by preventing the SG from spiking with hyperpolarizing current (Fig. 10C,D), and the IPSP was not always recruited exactly at the SG threshold. This suggests that it is not mediated by the SG itself. In one preparation, neither GF nor Rl stimulation induced the IPSP on its own, but when the two stimuli were brought close together temporarily, the IPSP was recruited (Fig. 10E-H). The GF-induced spike could be abolished by collision with the Rl-induced IPSP (Fig. 10H). It seems likely that the IPSP is induced by a homologue of the crayfish motor giant inhibitor (MoGI; Wine, 1977) interneurone. This can be activated by a number of sources, including the GFs, R3 and Rl units. The activation is often suprathreshold when derived from a single source, but when it is subthreshold, the sources can summate.

Stimulating the contralateral Rl rarely activated the large depolarizing IPSP in MoG, but several smaller PSPs occurred. These smaller potentials were common to the SG, indicating that there is some source which makes input to both neurones (Fig. 9K).

Eight IFFs were stained, seven from the FPI group, and one from the FMC group. Four rFFs were stained, all from the FPI group. There was some variation within groups, but it was not possible to identify individuals with certainty (Fig. 11). Two more IFFs were identified physiologically, but not stained.

Most FFs showed essentially the same response to activation of the ipsilateral GF as to activation of the SG (Fig. 12). They received a relatively large unitary EPSP upon activation of the GF. They also received a unitary EPSP upon stimulation of the ipsilateral R1 at a fixed threshold (Table 1). In some preparations the appearance of the EPSP from these two sources was virtually identical, although in others the GF-induced EPSP had additional components which were not matched in the Rl-induced response (Fig. 12A,B). The EPSPs were often suprathreshold early in a preparation, but usually became subthreshold as the preparation aged. This could be correlated with a decline in the number of non-MoG R3 units activated by GF stimulation (Fig. 12D-F).

Two FFs (right FPI group) were encountered that received bilateral input. In one preparation stimulating the PC first activated the IGF (the rMoG but not the rSG spiked) and induced a small EPSP in the FF. As the intensity of stimulation was increased the rGF was recruited (as evidenced by the appearance of an rSG spike). When this happened the FF EPSP showed a step increase in amplitude (Fig. 12H.I). Stimulating both IRl and rRl induced unitary EPSPs in the FF (Fig. 12J,K).

Simultaneous penetrations of the SG and ipsilateral FFs (two preparations) strongly suggested that the EPSP received in response to GF activation was mediated by the SG (Fig. 13). The IGF was induced to spike by stimulating the AC. This caused a spike in the 1SG and a short-latency EPSP in the IFF (Fig. 13A). A very similar spike and EPSP were caused by stimulating IRl (Fig. 13B). Stimulating 1R3 caused an antidromic spike in the FF, but little or no potential change in the 1SG (Fig. 13C). The SG spike induced by stimulating 1R1 could be abolished by injecting a pulse of hyperpolarizing current. When this happened, the EPSP in the FF was also abolished (Fig. 13D-F). There was little or no sign of a residual EPSP at the time that the EPSP had previously been apparent, showing that virtually all the input the FFs received on stimulation of Rl is a result of SG activation, as opposed to activation of sensory axons.

In two preparations abdominal flexor motor neurones were stained which had large contralateral cell bodies and a relatively extensive neuropile arborization (Fig. 14D,E). This anatomy is similar to that of the crayfish FI (Wine, 1977) and the FI of the squat lobster (Sillar & Heitler, 1985). These hermit crab neurones did not receive unitary EPSPs with GF activation, but rather a complex late potential (Fig. 14A,B). This response is again similar to that of the crayfish FI. Thus we tentatively identify these neurones as the hermit crab FL Stimulation of R1 caused considerable synaptic input, but this occurred before the SG was recruited, and thus was probably derived from the stimulation of sensory axons.

Other than the FIs described above, only one abdominal flexor motor neurone was encountered that did not receive input from the GF and/or SG. This neurone had the smallest cell body of any abdominal flexor motor neurone stained in our experiments, and was the only motor neurone which could be induced to spike tonically by injecting less than 10 nA depolarizing current. These factors suggest that it is one of the SF motor neurones. Data are not shown.

In this paper we describe the characteristics and interconnections of various neurones involved in controlling abdominal flexion of the hermit crab. We have concentrated upon the segmental giant neurone, but also provide information about the through-conducting GF interneurones, the MoG flexor motor neurone and the FF motor neurones. These names reflect our belief that the neurones we describe are homologues of similarly named neurones in the crayfish. The hermit crab GFs and MoGs have been described before, although in less detail, but we describe the hermit crab SG for the first time. Since this neurone is a crucial but puzzling part of the circuitry, the first point to establish is the basis of the claims of homology with the crayfish.

The term ‘homology’ is used in this discussion to refer to structures in divergent species of animal which are thought to derive from a single structure present in the ancestral animal common to both species (rather than meaning equivalent structures within divergently specialized segments of a single animal). Thus a swimmeret of the hermit crab is homologous to a swimmeret of the crayfish, although very different in appearance. Obtaining definite proof of such homology is often difficult. For skeletal structures it is sometimes possible to obtain a complete sequence of fossil intermediary types, but this is clearly not possible for neurones. Instead, one must rely on ‘common features’. These include the general anatomical characteristics (soma position and size, axon location, morphology of major neuropile arborization, etc.), physiological characteristics (spiking/non-spiking, input impedance, etc.) and circuit characteristics (sources of input, targets of output, modulation, etc.). It is these common features, as described in the results above, that indicate homology of the hermit crab and crayfish neurones.

The circuit driving abdominal flexion in G4 of the hermit crab can be summarized as follows. Each GF drives the ipsilateral SG directly in a 1:1 manner. Dual mieroelectrode penetrations suggest that the connection is via an electrical synapse which is usually, but not always (one exception was found), rectifying. Each GF also drives the ipsilateral MoG in a 1:1 manner, and previous work (Umbach & Lang, 1981) suggests that this connection too is via a rectifying electrical synapse. The MoG and the SG both receive PSPs from sources other than the GFs, and some of this input is common to the two neurones. The MoG also receives indirect input from the GFs in the form of depolarizing IPSPs.

Each SG makes connections to all ipsilateral FFs and some contralateral FFs. The resulting FF input is sometimes suprathreshold in the early stages of an experiment, but more usually it is subthreshold. The connection has a short latency, but its nature has not otherwise been determined. If it is via an electrical synapse (as in crayfish), the synapse must be rectifying since hyperpolarizing current does not spread from the SG to the FF, even when the current is sufficient to block the SG spike. For some FFs the short-latency input received upon stimulation of the GF is almost identical to that received from the SG, and it is thus likely that virtually all this input is mediated via the latter neurone. Other FFs receive a somewhat larger amplitude input with GF activation, and these may receive input from both the SG and the GFs, as is the case in crayfish (Roberts et al. 1982). Each SG also makes input to the ipsilateral, and possibly the contralateral, MoG. This input is small, probably excitatory, but always subthreshold.

The basic similarities of this circuit to that of the crayfish (see Wine & Krasne, 1982, for a review) are striking, but there are some differences in detail. Thus in the hermit crab the SG usually receives input from only the ipsilateral GF, while in the crayfish the SG receives input from both ipsi- and contralateral medial GFs (the homologues of the hermit crab GF), although the ipsilateral input is stronger. It is not clear in the crayfish whether any FFs receive input from the contralateral SG, but they definitely do in the hermit crab. Again, in the hermit crab, common synaptic input impinges on the SG and MoG, but it is not clear whether this also occurs in the crayfish. Finally, the MoG in the hermit crab receives input from the SG, while this does not occur in the crayfish.

The present results show conclusively that a neurone homologous to the SG of crayfish exists in hermit crabs. We may now ask whether any of the properties of the hermit crab SG shed light on the function of the SG or the origin of the giant fibre system as a whole.

There are at least two possible evolutionary paths by which the SG could have become involved in the circuitry of abdominal flexion (Fig. 15), and they have rather different implications for the function of the SG. Both possibilities start from the assumption that the proto-SG was a limb motor neurone, probably a swimmeret return-stroke motor neurone, as suggested by several anatomical and physiological aspects of the SG (Roberts et al. 1982; Heitler & Darrig, 1986).

The first path, put forward as a ‘best guess’ by Wine & Krasne (1982), assumes that three basic changes occurred: (1) the GFs evolved from pre-flexor interneurones, perhaps those involved in backwards swimming; (2) the MoGs evolved from FF motor neurones to become the primary efferents of the giant system; (3) the SGs evolved as the primary ‘central driver’ of the system. Thus the GFs would initially have driven only the FF/MoG pool. The precise schedule of evolution from this state is not proposed in detail, but presumably the proto-SG then evolved to receive input from the GFs, so that the GFs were activating the FFs/MoG and proto-SG in parallel. The critical point in the scheme is that the FFs must then have lost the major part of their direct connection to the GFs, and instead became connected to the SG.

This scheme thus has two stages of neural change which must have occurred in response to two sorts of selective pressure. In the first stage the proto-SG would have evolved to receive input from the GFs. The selective advantage might have been either that swimmeret protraction aided swimming, or simply that it was useful to bend the swimmerets out of the way of the flexing abdomen. The second stage would have been that the FFs (but not the MoG) switched their input source from the GFs to the SG. The selective advantage of this is by no means clear, but one possibility is that the GFs were suffering occasional spike failure due to excessive electrical loading, and the SG was incorporated as a buffer amplifier (Roberts et al. 1982).

Alternatively, if the SG was capable of being ‘switched off’ in its function as a swimmeret motor neurone, this might have been a useful pre-adaptation as a site for modulation of the GF-FF pathway. In any case, the circuit function of the SG according to this evolutionary scheme is to satisfy whatever selective pressure it was that brought about this switch in FF input source.

The second possibility for the evolution of the system starts from a different position. It assumes that the GFs started as pre-swimmeret inter neurones, driving an escape behaviour mediated by swimmeret protraction. The critical point in this scheme is that the FFs would have ‘added on’ to the circuit, evolving to receive input from the SG. The selective advantage of this would have been that tail flexion aided swimmeret protraction in driving the animal backwards. As tail flexion became the dominant propulsive force in the escape behaviour, the largest and most powerful FF would have bypassed the SG to receive input directly from the GF, and evolved to become the MoG. Again, there must have been an intermediate stage in which at least one FF (the proto-MoG), and possibly several FFs, received input from both SG and GF neurones. The attraction of this scheme is that a swimmeret-driven escape behaviour forms a plausible extension of the locomotory function which swimmerets are thought to have acquired in the early evolutionary history of the Crustacea.

If this second scheme is correct, it becomes unnecessary to propose any specific function for the SG. The SG exists as a consequence of the path through which the neural circuitry evolved, rather than because of any specific selective advantage its presence conveys to the animal. The SG may well have been secondarily modified to perform particular functions, but the primary reason for its existence lies in the evolutionary history of the behaviour.

Do any data presented here help decide between these schemes? The hermit crab MoG has anatomical features which place it intermediary between a crayfish FF and MoG motor neurone, in that the site of interaction with the GFs is located within the ganglion rather than outside it at the base of R3. Our results show that the hermit crab MoG also has intermediary physiological features. In particular, the MoG is driven by the GFs, but unlike the crayfish MoG, it also receives input from the SG. This finding adds weight to the theory that the MoG evolved from an unspecialized FF, but is compatible with either evolutionary scheme. Scheme two would suggest that the SG-MoG connection is the remnant of an originally generalized SG-FF connection which the MoG largely lost as it diverged from the FFs and developed the specialized GF-MoG connection. Scheme one would suggest that the GF-MoG connection represents the originally generalized GF-FF connection, and that the small SG-MoG connection is a result of the MoG not diverging from the FF pool until the FFs had already started to switch their inputs from GF to SG. In a similar manner the weak GF-FF connection of crayfish is compatible with either scheme. Scheme one interprets this as a remnant of the originally generalized GF—FF connection, scheme two interprets it as indicating that originally the whole FF motor pool started to bypass the SG and connect to the GFs directly, but as the MoGs evolved to become the major efferent path the selective pressure was concentrated on them, and the remaining FFs stuck in the primitive condition.

Either scheme requires that new neural connections should have been made, and old connections broken during the course of evolution, and both schemes allow function to be maintained during the changes. More comparative data may allow a decision to be made between them, but we must reluctantly conclude that while the information on the hermit crab SG that we provide suggests that the SG has a long-established role in crustacean tail flexion, it does not yet allow definite conclusions to be made about the evolutionary path by which this role was acquired.

This work was supported by grants from the SERC and Royal Society to WJH. We thank Professor W. Stewart for his gift of Lucifer Yellow.