Central Circuitry in the Jellyfish Aglantha Digitale: III. the Rootlet and Pacemaker Systems

Aglantha digitale probably has the most complicated nervous system among known Cnidaria in terms of circuitry, having pathways for both escape and non-escape locomotion (Mackie and Meech, 1995a,b); other medusae that have been studied in detail, such as Polyorchis penicillatus (Spencer, 1978, 1979; Spencer and Arkett, 1984), can swim in only one way. The complexity of the wiring in these nervous systems is surprising in a group of animals considered to be primitive, but it is matched in the area of neuroendocrinology where recent studies on the sea pansy Renilla köllikeri have also revealed unexpected complexity (Anctil, 1987, 1989; Anctil et al., 1991; Pani et al., 1995). Since the time of Pantin (1952), and as knowledge of these nervous systems has increased, it has become increasingly clear that cnidarian nerves and synapses operate in much the same way as those of higher animals, and the lack of a brain is now seen as an adaptation to radial symmetry rather than an indication of primitiveness (Satterlie and Spencer, 1987). In the case of A. digitale, the bundles of nerves running around the margin quite clearly fulfil the role of a central ganglion (Mackie and Meech, 1995a,b). Here, we describe a new and behaviourally important conduction system and its interactions with those described previously.

In A. digitale, slow, rhythmic swimming occurs when the animal is fishing for food (Mackie, 1980) and originates in the output of pacemaker neurones in the marginal nerve rings, as in other hydromedusae (Satterlie and Spencer, 1983). Escape swimming follows tactile or vibrational stimulation of the tentacles and outer surfaces of the margin and involves the conduction of action potentials around the margin in the ring giant axon (Roberts and Mackie, 1980). In both types of swimming, the subumbrellar swimming muscles are excited by impulses propagated in the eight motor giant axons that run up the inside of the bell from their origins in the inner nerve ring. However, during slow swimming, the motor giants conduct Ca²⁺ spikes, while in escape swimming they conduct Na⁺ spikes (Mackie and Meech, 1985). The muscle contractions seen in slow swimming are much weaker than those seen in escape swimming (Donaldson et al., 1980), and the two responses can easily be distinguished by the naked eye. Tentacle contractions accompany swimming in both cases, but are much more sudden and violent in the escape response, during which excitation of the ring giant axon is always accompanied by excitation in the giant axons that run down the tentacles, causing twitch responses. Analysis of the underlying circuitry has revealed two sets of interneurones, the relay and carrier systems, that function in the transfer of information from the pacemaker neurones to the tentacle action systems during slow swimming (Mackie and Meech, 1995a,b), bringing about the slower, graded contractions.

The production of two different sorts of spike in the motor giants has been investigated by intracellular recordings and voltage-clamp and patch-clamp analysis of the membrane channels (Meech and Mackie, 1993a,b, 1995). Na⁺ spikes are generated by fast-rising excitatory postsynaptic potentials (PSPs) representing input from the ring giant (RG) axon, while Ca²⁺ spikes arise from slow PSPs representing input from the pacemaker (P) system. These will be referred to here as RG-PSPs and P-PSPs respectively.

We now find that there is a third type of excitatory synaptic input into the motor giants which, like the RG-PSP, generates Na⁺ spikes but which does not come from the ring giant. The evidence presented in this paper makes it clear that the new type of PSP is due to input from adjacent motor giants transmitted directly between the rootlet interneurones. This rootlet system was first described by Weber et al. (1982), whose summary drawing, based on Lucifer Yellow injections, is reproduced here (Fig. 1). The rootlet neurones, like the lateral neurones and basal plexus that distribute excitation to the muscle sheet (Kerfoot et al., 1985), were found to be dye-coupled to the motor giant and to each other. Each motor giant makes contact with processes of these rootlet neurones in the inner nerve ring. These taper away to a few slender processes in the zone midway between the motor giants, where they appear to overlap with those associated with the neighbouring axon. Unlike Lucifer Yellow, horseradish peroxidase (HRP) injected into motor giants did not penetrate the rootlet system, showing that there was no syncytial continuity between the two, although the motor giants themselves are syncytial structures (Weber et al., 1982). Gap junctions were revealed by electron microscopy at interfaces between motor giants and certain neurones in the inner nerve ring. The latter were thought to be rootlet neurones, so it was concluded that the rootlet system is connected to the motor giants by membrane partitions containing gap junctions. The failure of HRP to enter the rootlet neurones was consistent with this interpretation. In the present paper, we confirm the general picture of the rootlet system presented by these workers.

Specimens of Aglantha digitale Müller caught off the dock at the Friday Harbor Laboratories, University of Washington, USA, were kept in glass containers at 7 °C until used. They were dissected in sea water containing 115 mmol l⁻¹ Mg²⁺ to immobilize them. The top of the bell was cut off, leaving a cylinder of body wall that was cut vertically and pinned out flat, subumbrella up, in a Sylgard-lined Petri dish using cactus spines as pins. Preparations were fixed for 3–4 h in 4 % paraformaldehyde in 0.1 mol l⁻¹ phosphate-buffered saline (PBS) at pH 7.3 followed by several rinses in 0.1 mol l⁻¹ PBS containing 0.3 % Triton X-100 and 0.03 % sodium azide (PTA). To visualize general nerve structure and layout, a mouse anti-tubulin antibody (Amersham, N356) diluted 1:50 in PTA was used; to demonstrate possible synaptic areas, a rabbit anti-syntaxin 1 antibody (Calbiochem, 574784) diluted 1:25 or 1:50 was used. Goat serum was added to the incubating solutions at 3–5 %. After 8–12 h, preparations were rinsed in PTA and treated with the appropriate secondary antibodies for a further 8–12 h before washing and mounting in 50 % glycerol containing 0.3 % n-propyl gallate. FITC-coupled goat anti-mouse secondary antibody was used to display tubulin and CY5-coupled goat anti-rabbit secondary to display syntaxin 1. In all these experiments, controls were run in which the primary antibody was omitted. Mounted preparations were studied by laser scanning confocal microscopy using a BioRad MRC-600 microscope at Friday Harbor and a Zeiss LSM 410 microscope at the University of Victoria.

Samples were prepared as above but were pinned out on thin, flexible strips of Sylgard in sea water containing 115 mmol l⁻¹ Mg²⁺. The tips of thin-walled glass electrodes were filled with 5 % carboxyfluorescein filtered through 0.2 μm cellulose acetate membrane filters. The shaft was filled with 2 mol l⁻¹ potassium acetate. Dye was injected into giant motor giant axons by ionophoresis, using 1 nA, 2 ms hyperpolarizing current pulses, at 2 pulses s⁻¹. Preparations were left for 5 min to allow the dye to penetrate the rootlet system. The strip of Sylgard bearing the preparation was then inverted and pressed down lightly against a coverglass sealed into a 1 cm diameter hole in the bottom of a small Petri dish with the preparation centred over the hole. This arrangement allowed filled axons to be brought into focus with substage objectives. Observations were made using the BioRad MRC-600 confocal microscope.

Observations on the behaviour of whole animals were made in finger bowls under a dissecting microscope. For recordings, preparations were dissected and pinned out as described above for carboxyfluorescein fills. Extracellular recordings were made with polyethylene suction electrodes pulled out to a 30–60 μm internal tip diameter. Intracellular recordings were made with glass microelectrodes filled with filtered 3 mol l⁻¹ KCl and having a resistance of 50–60 MΩ. Amplified signals were displayed on an oscilloscope and stored on tape with an instrumentation tape recorder for later study. Stimuli were applied externally through small coaxial bipolar metal electrodes. Temperature was maintained at 10–12 °C using a thermoelectric cooling stage. Further details of these procedures are given in Mackie and Meech (1995a).

A fairly complete picture of the major components of the nervous system was obtained with anti-tubulin fluorescence microscopy. The motor giant axons, lateral neurones, basal plexus and nerve ring elements were all well shown (Fig. 2A), confirming the general layout of the subumbrellar motor components described from Lucifer Yellow injections by Weber et al. (1982) (see also Fig. 1). However, because the great mass of neurones comprising the inner nerve ring was also stained, it was impossible to follow the rootlet neurones for any distance in these preparations. There was little background staining of the striated muscle sheet in which these elements lie. Nerves running out across the velum were clearly shown. In preparations labelled with anti-syntaxin 1, a low level of immunoreactivity was visible thoughout the cytoplasm of the motor giant axons and nerve ring elements, but there were also numerous, strongly immunoreactive spots within the nerve rings (Fig. 2B). These were restricted to the rootlet neurones and other elements within the nerve rings and were not seen in outlying parts of the nervous system, such as the lateral neurones, basal plexus elements and velar neurites, nor along the main extent of the motor giant axons. In controls omitting the primary antibody, no fluorescence was observed. While these findings are not conclusive, they suggest that the anti-syntaxin 1 immunoreactive spots may be synaptic sites.

Carboxyfluorescein fills enabled us to follow the rootlet neurones for considerable distances within the inner nerve ring from the point where they connect with the motor giant. In Fig. 3, the two motor giants were 2.18 mm apart at their bases, and the rootlet neurones in Fig. 3A extended past the midpoint between the two axons by at least 230 μm, suggesting a 460 μm zone of overlap, similar to the distance shown in Fig. 1B. The dye became hard to see as it entered the narrow terminal neurites of the rootlet neurones, and the actual overlap zone may therefore be considerably larger. The overlapping rootlet neurones do not end separately, as shown by Weber et al. (1982), but associate very closely so that they can no longer be separately distinguished. This finding is important because it provides a structural basis for the direct interaction between adjacent motor giants via the rootlet system suggested by the physiological evidence. Neither the mass of neurites forming the bulk of the inner nerve ring nor the velar nerves were filled.

Stimulation of the inside of the bell in the vicinity of the motor giants with a fine probe, taking care not to touch or cause vibrations in the margin or velum, was found to evoke an escape swimming response of the type associated with Na⁺ spikes in the motor giant axons (Mackie and Meech, 1985). As with escape swimming evoked by pinching the tentacles or prodding or vibrating the margin and velum (Roberts and Mackie, 1980), the response consisted of one or two very strong contractions each of which propelled the animal a distance equivalent to several body lengths. The two responses differed, however, in that the response to stimulation of the outer parts invariably included an immediate, strong twitch contraction of the all the tentacles, while subumbrellar stimulation led to graded, incremental tentacular contractions. As it was known from previous work that the twitch contractions of the tentacles seen in the former case are conducted round the margin by the ring giant axon, it seemed likely that the ring giant was not involved in the response that follows stimulation of the subumbrella.

If, as the behavioural observations implied, the ring giant was not involved in conducting this escape response around the margin, we had to find another pathway, and none of those so far investigated had the necessary properties. Suspicion therefore fell upon the rootlet interneurones, known from the work of Weber et al. (1982).

The motor giant axons run up the inside of the bell in the strips of ectodermal tissue (radial strands) overlying the endodermal radial canals. Direct stimulation of a motor giant axon by means of a small, coaxial, bipolar electrode placed directly over a radial strand elicited a Na⁺ spike that spread to all the other motor giant axons on both sides of the one stimulated. Fig. 4A shows spikes recorded intracellularly (top trace) and extracellularly (bottom trace) from one motor giant axon following stimulation of the adjacent motor giant axon (S1) and a motor giant axon three strands away (S2). The absence from the external trace, in these and many other recordings, of ring giant axon spikes confirms that the ring giant plays no part in conduction of the response. In sea water containing 97–105 mmol l⁻¹ Mg²⁺, transmission between motor giant axons was usually blocked, although the stimulated axon still fired normally, suggesting that chemical synapses are involved in the pathway linking the axons.

When the axon was punctured with the tip of a microelectrode between the stimulating point and the margin, it withered locally and transmission to other axons was blocked, although stimulating the distal, intact part of the axon still evoked action potentials and local muscle responses. Cuts made through the subumbrellar muscle sheet on either side of the motor giant and adjacent to the inner nerve ring confirmed that the transmission route to the other motor giant axons was along the margin. Conduction velocity around the margin in the experiment shown in Fig. 4A was 50 cm s⁻¹. Transection of the inner nerve ring (which includes the rootlet neurones) blocked the spread of the response, while transection of the outer nerve ring did not. These findings, taken in conjunction with the histological evidence presented above, strongly suggest that the rootlet interneurones interconnect to form a conduction pathway involved in the transmission of Na⁺ spikes between the motor giant axons.

A second consequence of stimulating over a radial strand was the triggering of impulses in the pacemaker (P) system, which generates rhythmic slow swimming. A single shock was found to trigger one or a series of pacemaker potentials, readily recorded by a suction electrode placed over the inner nerve ring (see Fig. 4B). Depending on the size of the animal, P potentials are conducted around the margin at 40–60 cm s⁻¹ (Meech and Mackie, 1995), probably by the group of large (approximately10 μm in diameter) neurones running in the inner nerve ring described by Weber et al. (1982). These large axons do not extend into or send branches up the radial strands, and P potentials were never recorded from the radial strands, so stimulating a strand could not excite the pacemaker system directly. Creating a lesion in the motor giant between the stimulating point and the margin blocked the production of P potentials in the margin at the same time as it blocked spike propagation in the motor giant. Triggering of P potentials was also blocked in 105 mmol l⁻¹ Mg²⁺. We conclude from these observations that the motor giant axons and/or their rootlet neurones synapse with pacemaker axons in the margin, exciting them whenever they conduct a Na⁺ spike. The P system normally shows a regular pattern of spontaneous bursts (fictive slow swimming) and, if excited by input from a motor giant around the time when it is due to produce a spontaneous burst, the timing may be advanced and a burst may result (Fig. 4B).

Each P event recorded extracellularly was associated with a depolarization in the adjacent motor giant axon (Fig. 4B,C,D).

These depolarizations are the ‘slow postsynaptic potentials’ of Meech and Mackie (1995). In sea water, each slow, or P-PSP, generated a Ca²⁺ spike in the motor giant (Fig. 4C), but in water containing more than 83 mmol l⁻¹ Mg²⁺ only the PSP was seen. P-PSPs are slowly rising events rarely exceeding 18 mV amplitude, with rounded tops and a long declining phase (τ=60 ms, where τ is the time constant of decay). In some preparations in which [Mg²⁺] was greater than 87 mmol l⁻¹, the P system fired twice in rapid succession and the correponding PSP was a two-step event (Fig. 4D).

A third consequence of stimulating over a radial strand was the appearance of a small electrical event in the extracellular recording that propagated down to the margin and around it in the outer nerve ring and up the radial strand to the manubrium. We refer to these signals as impulses in the manubrial (M) system, which is involved in feeding behaviour. The response is blocked by cutting the radial strands. M pulses are almost certainly carried by the bundles of small FMRFamide-immunoreactive neurites that run up the radial strands close to the motor giants, connecting sensory elements in the margin with a nerve plexus overlying the muscles in the manubrium (Singla, 1978; Mackie et al., 1985; G. O. Mackie, unpublished physiological data). There is no evidence that this system is involved in locomotion and it will be ignored in the present account, but M pulses are labelled in some of the figures.

Recordings made from one motor giant axon following stimulation of another were invariably complicated by the fact that the P system was triggered at the same time. When, as in Fig. 5A, an immediately adjacent axon was stimulated, the first event recorded by the extracellular electrode (lower trace) was the correlate of the Na⁺ spike seen in the intracellular recording (upper trace), but this was followed by one or more impulses in the P system, followed in turn by an impulse in the relay (R) system and the accompanying slow wave (W). P–R–W sequences are a normal feature of the preparation. R and W components follow spontaneous P events as well as those evoked by direct stimulation of the inner nerve ring and those, as here, triggered indirectly by excitation of motor giant axons. These events are part of a circuit that functions to cause the tentacles to contract during pacemaker-evoked swimming (Mackie and Meech, 1995a). The P system provides direct input to the motor giants in the form of slow or P-PSPs (Meech and Mackie, 1995). These are represented in Fig. 5A by a hump in the spike after-potential (arrowhead). Addition of 5 mmol l⁻¹ Mn²⁺, which depresses synaptic transmission, rapidly blocked the P events and abolished the hump (Fig. 5B), leaving the spike after-potential in its ‘pure’ form, a smoothly decaying curve. After a further period in the same solution, during which the preparation was stimulated repeatedly, the spike failed, leaving an underlying excitatory postsynaptic potential attributable to input transmitted via the rootlet neurone pathway (rootlet PSP) (Fig. 5C).

Rootlet PSPs are fast-rising events closely resembling the RG-PSPs that follow stimulation of the ring giant axon, which likewise generate Na⁺ spikes in the motor giants (Meech and Mackie, 1995). Rootlet PSPs are readily distinguished from P-PSPs because they rise rapidly to a sharp peak, reach larger amplitudes and decay more rapidly (τ=10 ms in Fig. 5C). At their maximum amplitude (Fig. 5D), rootlet PSPs surpass the threshold for Na⁺ spike generation, which lies at around −32 mV (Meech and Mackie, 1995). In the preparation shown in Fig. 5E, the stimulated axon occasionally fired twice in response to single shocks, but only the first action potential produced a spike at the recording site. The second produced a rootlet PSP that failed to reach spike threshold. In several other preparations in 3–4 mmol l⁻¹ Mn²⁺ or 105 mmol l⁻¹ Mg²⁺, similar abortive rootlet PSPs were obtained by giving two shocks 20–40 s apart.

While spikes conducted in the P system were easily recognizable in extracellular recordings, those conducted in the rootlet neurones were invisible. The small blip in the extracellular recording of Fig. 5C did not precede the PSP but coincided with its rising phase and cannot, therefore, represent the presynaptic rootlet event. Presumably, the rootlet pathway involves relatively few, small units, insufficient to generate a detectable level of external action current. This is consistent with the histological picture. In practice, however, our inability to record rootlet spikes extracellularly meant that their presence could only be deduced from intracellular recordings and, because the rootlet and P events were propagated at similar conduction velocities and tended to arrive close to one another in time, anomalous recordings in which a motor giant spike appeared to arise from the top of a P-PSP were often seen (see Fig. 6A, upper trace). When such events were observed in 93 mmol l⁻¹ Mg²⁺, the spike sometimes failed, revealing an underlying rootlet PSP surmounting the P-PSP (Fig. 6B, upper trace). On other occasions, the P- and rootlet PSPs were generated simultaneously and fused to form a composite depolarization that could give rise to spikes. Study of numerous recordings suggests that spikes are never generated by P-PSPs but always by rootlet PSPs, even where the two are ‘fused’.

Where the stimulated axon was immediately adjacent to the one from which the recordings were made, the rootlet event nearly always arrived before the P event or events (Fig. 5A). Where the stimulated axon was two or more axons away, the P event arrived at the same time as the rootlet event or slightly ahead of it (Fig. 6A). Reasons for this are suggested in the Discussion.

A number of recordings were made in which an axon was stimulated at its distal end and the recording electrodes were placed proximally, where the axon joined the margin (Fig. 6C,D). The preparation shown in Fig. 6C was in 90 mmol l⁻¹ Mg²⁺, which depresses synaptic transmission, and the giant axon spike is seen in the intracellular recording in its ‘pure’ form with no humps in the after-potential. The spike is represented extracellularly as a smaller deflection than usual because the external electrode had been placed very slightly to one side of the junction to reduce the size of the spike and so to visualize P events better. There was no indication of triggered P events. In the same preparation, after a period without stimulation, a stimulus produced a giant axon spike that triggered a ripple of three P events, each of which fed back into the motor giant, producing P-PSPs as humps in the decay curve (Fig. 6D). The first of these P-PSPs appeared approximately 2 ms after the peak of the motor giant spike. While accurate measurements were not possible, this delay is approximately right for a two-synapse loop, assuming synaptic delays of 0.8 ms (Kerfoot et al., 1985) and allowing for some conduction time in the local circuit.

A final point clarified by these recordings concerns the ‘piggyback’ effect, the acceleration of impulses conducted in one system due to depolarizing input from a faster system running in parallel with it (Mackie, 1976). Potentials representing the relay system (R) are typically seen to follow spontaneous and evoked P impulses after 25–40 ms, but the conduction velocity of these events depends on the immediate past history of excitation in the R system and on whether other parallel systems are active at the same time (Mackie and Meech, 1995a). These authors found that R potentials propagate slowly on their own (<12 cm s⁻¹), but when triggered and preceded by pacemaker impulses, their conduction velocity can double, and when they follow a burst of ring giant impulses, they can conduct at over 40 cm s⁻¹. In Fig. 6E,F, stimuli were delivered to a motor giant axon three radial strands away from the recording site. The rootlet pathway sometimes failed to conduct all the way through to the recording point, leaving the P–R–W sequence in which the conduction velocity of the relay system was approximately 15 cm s⁻¹ (Fig. 6E). On an occasion when the rootlet pathway did not fail and a spike was generated in the motor giant at the recording point (Fig. 6F), the relay system conducted at 33.5 cm s⁻¹. Acceleration of the conduction velocity of the relay system under similar circumstances is also shown in Fig. 6A,B. We conclude that excitation propagated in the rootlet neurones can exert a direct ‘piggybacking’ influence on the conduction of events in the relay system. It should be noted that in elevated [Mg²⁺], the rootlet pathway can fail at any point along the margin, so partial degrees of piggybacking are frequently seen.

All the evidence presented here points to the conclusion that the motor giant axons communicate directly with one another via rootlet interneurones, which synapse with one another in a zone of overlap. The rootlet neurones are dye-coupled and, presumably, electrically coupled to the motor giant by gap junctions, allowing synaptic currents to flow into the base of the motor giant where we record them. Na⁺ spikes arriving at rootlet synapses generate large postsynaptic potentials (the rootlet PSPs), sufficient to evoke postsynaptic Na⁺ spikes. Depression of synaptic transmission by the addition of divalent cations allows the rootlet PSP to be seen without the spike. Given the large size and presumably high capacitance of the motor giant, it seems remarkable that such a large PSP can be generated in the motor giant when the synapses are made between a small number of fine neurites several hundred micrometres away. Unfortunately, we were not able to determine the numbers and locations of synapses between rootlet processes because of the difficulty of distinguishing the rootlet neurones from other fine neurites in the immunofluorescence preparations. The results of labelling with anti-syntaxin 1 constitute preliminary evidence that there are numerous synapses in the inner nerve ring, including some involving rootlet neurones. Although there may be many synapses on each neurite in the zone of overlap, and the zone of overlap may be more extensive than revealed by our carboxyfluorescein fills, the rootlet PSPs recorded from the bases of motor giant axons in elevated [Mg²⁺] and [Mn²⁺] exhibit a clean waveform without inflections or ripples and do not appear to be spatially summed composite events. A possible answer to this conundrum is that the fast rise of the rootlet PSP may reflect a regenerative contribution from the same kind of T-type Ca²⁺ channels that are present in the motor giant axon membrane (Meech and Mackie, 1993a, 1995). This would explain the unexpected sensitivity of the rootlet PSP to divalent ions compared with the insensitivity of chemical transmission at the neuromuscular junction (see Kerfoot et al., 1985).

The discovery of the rootlet pathway means that we can now distinguish three types of input into the motor giant axons. Slow or P-PSPs, which generate Ca²⁺ spikes, clearly represent input from the pacemaker system because they invariably follow the characteristic P potentials recorded extracellularly. Similarly, input from the ring giant axon can be identified by the fact that it follows ring giant spikes, which are readily recorded both intra- and extracellularly. The rootlet PSP, although resembling the ring giant PSP, can be distinguished from the latter by the fact that it is not preceded by ring giant spikes. In fact, we were not able to record an extracellular correlate of the presynaptic rootlet potentials. This is perhaps to be expected given the histological picture of the rootlet system as a small group of fine neurites. The resemblance between rootlet PSPs and ring giant PSPs is not surprising, given that both serve precisely the same purpose, that of raising the motor giant to the threshold for Na⁺ spikes.

In the course of investigating the rootlet pathway, it became clear that excitation of the motor giant axons triggers events in the pacemaker system, so that P potentials propagate around the margin in a close time relationship with those conducted in the rootlet pathway. As the P system in its turn synapses with the motor giants, producing PSPs in them, these events appear postsynaptically and complicate the waveform of spikes produced by the nearly simultaneous rootlet input. Humps in the spike after-potential have been seen previously (see Fig. 1A in Kerfoot et al., 1985) but their significance was puzzling, because recordings of the spike in its ‘pure’ form showed a smoothly declining after-potential (see Roberts and Mackie, 1980). It is now clear that the humps are P-PSPs.

The mechanism whereby the tentacles contract during escape responses evoked by subumbrellar stimulation has not been worked out in detail, but we know from previous work (Mackie and Meech, 1995a) that triggering of P potentials singly or in bursts will occur when motor giants fire, and this will result in activation of the relay system and the pathways to the tentacles; the tentacular contractions produced will be graded events of the type seen during slow swimming rather than the violent contractions associated with escape behaviour in which the ring giant spreads the response.

Because the rootlet system and P pathways conduct at similar velocities and the two PSPs appear almost synchronously, the spike may appear to rise from a P-PSP, but analysis of such cases always shows that a rootlet PSP was present and actually generated the spike. Under experimental conditions (elevated [Mg²⁺]), the two events can sum to produce a spike, although neither would come close to the spike threshold on its own. Thus, in a preparation containing elevated [Mg²⁺] (Fig. 6A,B), a 14 mV rootlet PSP surmounting a 15 mV P-PSP almost reached spike threshold. This might suggest that summing of the two sorts of input contributes to spike production under normal conditions, but this seems unlikely, or at least unnecessary, because the rootlet PSP can easily achieve spike threshold on its own (Fig. 5D). Where a P-PSP surmounted an abortive rootlet PSP, spikes were never generated even though the summed event surpassed spike threshold. It appears that Na⁺ spike production calls for a rapidly rising PSP, not just one that exceeds the threshold.

When a motor giant axon immediately adjacent to the one serving for the recordings is stimulated, the rootlet PSP is recorded ahead of the slow PSP, but when the stimulated axon is two or more axons away, the slow PSP is seen coincidently with or leading the rootlet PSP. The rootlet pathway from the immediately adjacent axon involves only one synapse, while that going through the pacemaker system must involve two (Fig. 7). The additional synapse would delay the arrival of the P event and its PSP. How does the P event catch up with and overtake the rootlet event when more distant axons are stimulated? We envisage the P neurones as long units that may extend without interruption through the territories of several motor giant rootlet systems. Once impulses are initiated in the P system, they will travel for long distances along the margin at the high conduction velocity characteristic of this system (40–60 cm s⁻¹), whereas the rootlet pathway will be interrupted by synapses that introduce a delay at each overlap zone. It must be said, however, that we do not know how long the P neurones are, how many there are and how they are kept in synchrony so that they appear to propagate as a single unit. The fact that synchrony sometimes breaks down in elevated [Mg²⁺] (Fig. 4D) suggests that they are laterally interconnected by synapses and ‘piggyback’ each other when the synapses are operating normally.

An interesting aspect of this nervous system is the feedback loop operating between the motor giants and the P system. Synaptic transmission occurs between the two systems in both directions. This means that whenever a motor giant fires, whether because it is stimulated itself, because other motor giants are stimulated or because the ring giant axon is stimulated, the P system will also be excited after a delay of approximately 1 ms, and at least one P impulse will be seen. If the P system is about to produce a spontaneous burst, a burst will be triggered, and a series of P-PSPs will be recorded in the motor giant (as in Fig. 4B). While a single P-PSP occurring within a few milliseconds of a spike in the motor giant is unlikely to have any behavioural consequences, the later events in a triggered P burst would certainly evoke Ca²⁺ spikes and slow swimming in an animal in its natural habitat. This may explain the observation that escape swims are sometimes followed by slow swims.

Bidirectional transmission between the rootlet neurones and the P system could be brought about by separate synapses polarized in opposite directions or by symmetrical synapses (i.e. single synapses with vesicles on each side), as described in Cyanea capillata (Anderson and Grünert, 1988). Symmetrical synapses have not been reported between rootlet neurones and pacemaker neurones, but occur elsewhere in the nervous system of A. digitale (Mackie, 1989) and might be present here. Our recordings showing feedback from the P system to the motor giants are very reminiscent of the recordings of Anderson (1985) obtained by inserting electrodes into neurones on either side of symmetrical synapses in C. capillata in which, if synaptic transmission from neurone A to neurone B produced a spike in B, a PSP appeared in the form of a hump on the declining slope of the presynaptic (A) spike. Anderson and Spencer (1989) review the general occurrence and possible significance of symmetrical synapses in cnidarians and other animals.

In Fig. 7, we present an updated version of our previous wiring diagram (see Fig. 8 in Mackie and Meech, 1995a) modified to include the rootlet pathway. In addition to the rootlet pathway, there are four physiologically recognizable, through-conducting systems of neurones running around the margin and directly concerned with swimming or with the contractions of the tentacles that accompany swimming. The pacemaker system initiates slow, rhythmic swimming; the ring giant axon coordinates escape swimming; and the relay and carrier systems transfer information from the pacemakers to the ring giant and tentacles during slow swimming. Tentacle contractions themselves are of two sorts, twitch responses mediated by the tentacle giant axons (which also provide sensory input to the ring giant) and slow contractions mediated by a diffuse, slowly conducting tentacle system. Two other pathways concerned with locomotion have been described since the original wiring diagram appeared and are included in Fig. 7, the touch-sensitive exumbrellar epithelium, excitation in which inhibits the locomotory pacemakers (Mackie and Singla, 1997), and a plexus of nitric-oxide-synthase-reactive neurones (NO system) located in the tentacles and outer nerve ring. Nitric oxide, presumably secreted by the NO neurites in the nerve ring, excites the swimming pacemakers (Moroz et al., 1997). Including these two systems along with the motor giants and the rootlet neurones brings the total number of known conduction pathways involved in swimming and tentacle contraction to ten but, in the interests of simplicity, we have omitted several components from Fig. 7: the lateral and basal plexus neurones that spread excitation across the subumbrellar muscle sheet (Kerfoot et al., 1985), the vibration-sensitive hair cell mechanoreceptors that excite the ring giant (Arkett et al., 1988) and the statocyst nerves, whose input regulates the directionality of locomotion (Mackie, 1980). Including these components brings the total to thirteen.

To this list of physiologically distinct (and in some cases histologically identified) systems involved in locomotion and tentacle control may be added at least one more well-defined through-conduction pathway, the manubrial system, composed of FMRFamide-immunoreactive nerves, which mediates feeding responses between the margin and the manubrium (Mackie et al., 1985; G. O. Mackie, unpublished observations), bringing the total of known conduction pathways to fourteen. While many details of the circuitry remain to be worked out, it seems doubtful that any major conduction system still remains undetected because we can assign all the electrical signals recorded from the nerve rings to one or other of the known systems, and most aspects of the animal’s behaviour can be explained on the basis of interactions among them. The work on A. digitale fully supports the view expressed by other workers (e.g. Satterlie and Spencer, 1987; McFarlane et al., 1989) that cnidarian nervous systems, far from being ‘simple’, can be as complex as those of arthropods.

G.O.M. gratefully acknowledges the support of the Natural Sciences and Engineering Research Council of Canada.

R.W.M. thanks the Wellcome Trust for travel funds. We thank the Director and staff of Friday Harbor Marine Station, University of Washington, where this work was carried out, for providing excellent facilities.