In searching for cellular mechanisms by which nervous systems store learned information, it is helpful to differentiate properties which are genetically programmed from those which are conferred by the interaction of an organism with its environment. With mechanisms of learning, as with those underlying developmental transformations, a chronology must be reconstructed - a sequence of processes which arise as a consequence of preceding conditions. To construct such a chronology, the cellular site at which learned information is stored must be accessible to analysis so that we may watch it during sequential transformations, and so that we can measure with appropriate morphological, biophysical and biochemical techniques the nature and magnitude of the critical changes.
Accessibility to sites of associative learning has been elusive for vertebrates, in which extracellular potentials have been recorded as correlates of acquisition and retention of conditioned behaviour (Woody, Vassilevsky & Engel, 1970; Woody & Black-Cleworth, 1973; Woody etal. 1974; Berger, Alger & Thompson, 1976; Cohen & MacDonald, 1976; Berger & Thompson, 1977, 1978; Brons & Woody, 1980; Thompson et al. 1982; Woody, 1982). It has also been largely true for invertebrates such as the bee, locust and the gastropod molluscs Pleurobranchaea, Umax and Aplysia. Hoyle and his colleagues recorded changes of impulses in certain locust motor neurones during learning (Hoyle, 1982). However, these changes have only been measured unequivocally for minutes, and their site of origin, and thus their accessibility, have not been determined. Furthermore, the synaptic organization c® the neural elements which mediate the training response remains obscure.
Workers using Aplysia and Pleurobranchaea have had more success in identifying neurones within relevant sensory pathways (Castellucci, Pinsker, Kupferman & Kandel, 1970; Gillette & Davis, 1977; Gillette, Gillette & Davis, 1980). However, a close and specific correlation of neuronal changes with learned avoidance of food in Pleurobranchaea (as opposed to its being in a food-satiated state), has not been made, as it has for vertebrate conditioning. Nor has it been possible to measure correlated changes in identified neurones or neuronal aggregates of Limax or Aplysia during or after associative training. For dissected and reduced preparations, short-term correlations (Carew, Hawkins & Kandel, 1983) have been made, but their relevance to the neurophysiology of living animals must be established (see for example, Kanz, Eberley, Cobbs & Pinsker, 1979).
Furthermore, the type of learning involved with some of these gastropods has not been entirely resolved. Some assert that non-associative behavioural changes such as habituation and sensitization can underlie associative learning, such as classical conditioning (Kandel & Schwartz, 1982), although any relevance of non-associative to associative mechanisms remains to be demonstrated. Even those behavioural changes called ‘associative’ require further study before their ‘associative’ nature can be considered certain.
Mpitsos & Collins (1975) trained Pleurobranchaea to avoid food substances by pairing the food with shocks to the head. Unpaired presentations did not produce an effect. These experiments could be simplified by controlling the different states of satiation which might result from the training. Long-term sensitization of a neophobic reaction to the food substances by shock should also be examined. The recent experiments of Gelperin and his colleagues withLimax (Gelperin, Wieland & Barry, 1984; Sahley, Rudy & Gelperin, 1984) require these same controls, since all measures of avoidance behaviour following pairing of a food substance with quinine were made by comparison with responses to a ‘safe’ food. A neophobic reaction to a ‘safe’ food (on which the animal had been raised) might be entirely habituated, while quinine could sensitize the neophobic reaction to a food substance with which it was paired during training. An apparently associative behavioural change (i.e. one dependent on the pairing of a food substance with quinine or an electric shock) might be simulated by particular combinations of two non-associative behavioural effects such as habituation and sensitization.
In recent experiments on Aplysia, tactile stimulation of the gill withdrawal response, which habituates, has been paired with electrical stimulation of the tail, which causes sensitization (Carew, Walters & Kandel, 1981). This causes the animal to withdraw its gill more rapidly and reliably. However, the possibility exists that the shock-induced sensitization prevents the touch-induced habituation. This is suggested by the observation that repeated presentation of the shock alone produces, to a significant degree, the same effect as repeated, paired, presentations. Other characteristics of associative learning, such as the stimulus specificity shown for Aplysia (Carew et al. 1981), and the blocking shown forLimax (Sahley et al. 1984), have yet to be clearly distinguished from combinations of effects induced by single stimuli.
With these difficulties in mind, we have worked for some years with the nudibranch mollusc Hermissenda crassicomis to localize sites of neuronal change which could play a causal role in associative learning, and which are accessible to quantitative analysis. We first found sites of convergence between three sensory pathways: the visual, vestibular and chemosensory (Alkon & Fuortes, 1972; Alkon, 1973, 1974a, 1980a, 1983; Alkon, Akaike & Harrigan, 1978). With simultaneous intracellular recordings from identified pre-and postsynaptic elements in thousands of adult nervous systems we constructed a working blueprint of the circumoesophageal nervous system (Fig. 1). Within this blueprint, the flow of sensory information could be traced from the input stages through the integrative centres to the output motor cells of relevant muscle groups. The effects of sensory stimuli such as light (transduced by five photoreceptors in each eye) and rotation (a gravitational stimulus transduced by thirteen hair cells in each statocyst, a primitive vestibular organ) could be studied at every level of integration: sensory cells, interneurones and motor neurones (Fig. 2A). Based on the known convergences between the visual and statocyst pathways, and the specific responses of the network to light and rotational stimuli with specific temporal relationships, we hypothesized that Hermissenda could be conditioned by repetition of paired light and rotational stimuli.
In fact, it has been possible to demonstrate most of those features which characterize vertebrate classical conditioning. Rotation, which serves as an unconditioned stimulus, elicits a consistent and reliable ‘clinging’ response (Alkon, 19746) manifest by measurable contraction of the ‘foot’ (particularly the caudal half of the ventral surface). In addition, the animal moves towards a light source (Alkon, 1974b). This is accompanied by lengthening of its foot, again particularly the caudal half. After light and rotation are repeatedly paired (Alkon, 1974b) (i.e. maximal light precedes maximal rotation by TOs), light elicits an entirely new response: contraction of the foot (Fig. 3), the response reflexly elicited by rotation. As a result of the conditioning procedure, light has taken on the meaning of rotation as measured by a new learned response (Lederhendler, Gart & Alkon, 1983). This learning behaviour is associative in that it is not produced by explicitly unpaired light and rotation, or light and rotation occurring with a randomly-varying interval during training (Crow & Alkon, 1978). It is stimulus specific, increases as a function of practice (i.e. it shows acquisition), can be retained for many weeks, extinguishes (Richards, Farley & Alkon, 1983) and shows savings. It also has a requirement for contingency (as well as contiguity) of the associated stimuli (Farley & Kern, 1983). This was shown by the degradation of the learning by interpolation of unpaired light or rotation stimuli. Finally, there is no long-term habituation or sensitization following training.
Intracellular recordings at each level of integration within the visual system of conditioned and control animals uncovered many learning-correlated changes of neuronal properties and response characteristics (Crow & Alkon, 1980; Farley & Alkon, 1982; Lederhendler, Goh & Alkon, 1982; West, Barnes & Alkon, 1982). Such recordings were made from progressively reduced preparations including living animals, isolated nervous systems and individual neurones isolated from all others, of conditioned and control animals. By this means it was possible to reconstruct a sequence of responses and changes during training.
Paired light and rotation (but not unpaired stimuli) enhance and prolong the depolarization of Type B photoreceptors during and following a light step. Repeated pairings cause progressively greater light-elicited depolarizations, an accumulation of membrane depolarization (Alkon, 1980b) and an increased input resistance (Fig. 2B). After a training session (e.g. of 50 trials) the Type B cell remains depolarized for many minutes (Alkon, 1980b; Crow & Alkon, 1980). On days following training (the retention period) the Type B cell from conditioned animals is no longer depolarized but remains more excitable. In response to injection of positive current or to light stimuli (Fig. 4), the conditioned Type B cell shows a greater depolarization (West et al. 1982). The increased impulse activity accompanying light-evoked depolarization in turn causes more inhibition of the medial Type A photoreceptor and, thereby, decreases excitation of interneurones which receive monosynaptic EPSPs from it (Farley & Alkon, 1982; Lederhendler et al. 1982; Goh & Alkon, 1984). Motor neurones which are monosynaptically excited by these interneurones are inhibited in this way by the Type B photoreceptor. In conditioned animals, increased light-evoked impulses in Type B cells decreases the excitation of motor neurones which in living animals cause turning toward a light source.
A number of other experiments taken together provide strong evidence that changes of Type B cells have a causal role in the production of associative learning. For example, the changes in Type B cells predict motor neurone changes (Lederhendler et al. 1982). The changes are present in Type B cells of conditioned but not control animals, physically isolated from all impulses and synaptic interactions with other neurones - i.e. these changes are intrinsic to the soma membrane (West et al. 1982). When these same changes are produced in Type B cells of living animals by pairing current injections (simulating the synaptic effects of stimulus pairing) with light, the learned behaviour measured on subsequent days is produced (Farley et al. 1983). This behavioural change does not follow unpaired light and current, or sham procedures.
Given that the Type B cell is a site for storage of learned information, is it accessible to biophysical and biochemical analysis? Fortunately, substantial changes induced by the conditioning were found to be intrinsic to the soma membrane of the Type B cell. These changes were sufficiently large and the spike initiating zone and synaptic endings were electrically close enough, to account for many of the observed changes of impulse activity and thus for the modified response of the visual pathway to light.
Since the roughly spherical soma can be isolated from the small diameter (~l-2μm and short axon, it is well suited for voltage-clamp studies (Alkon, 1979). Pharmacological blockers and substitution of ions permitted the separation of six distinct ionic currents across the soma membrane of a Type B cell (Fig. 5).
An early rapidly-activating and inactivating outward K+ current (Shoukimas & Alkon, 1980; Alkon, Lederhendler & Shoukimas, 1982zz), blocked by 1-3 mmol 1-1 4-aminopyridine. It resembles IA currents previously identified (Connor & Stevens, 1971).
A slowly-activating outward K+ current with much less inactivation, blocked by lOOmmoll-1 tetraethylammonium ion. It is similar to delayed rectifying currents described for many neurones (Shoukimas & Alkon, 1980).
A slowly-activating outward K+ current with substantial and prolonged inactivation (Fig. 6), which depends on elevation of intracellular Ca2+ and is therefore blocked by injection of Ca2+-chelators such as ethylene glycocol tetracetic acid. It can thus be termed a Caz+-dependent K+ current or Ic (Alkon, Farley, Hay & Shoukimas, 1983b; Alkon, Farley, Sakakibara & Hay, 1984).
A rapidly-activating voltage-dependent inward Ca2+ current which shows no inactivation and is blocked by 2-4mmol I-1 Cd2+ (Fig. 7; Alkon et al. 1983b, 1984).
A light-induced inward Na+ current which rapidly inactivates (Fig. 5; Alkon, 1979).
A light-induced outward K+ current which substantially inactivates and results from a light-induced release of intracellular Ca2+ (Fig. 8). This current differs from the third one in that it arises from a light-induced rather than a voltage-dependent rise of intracellular Ca2+.
‘Blind’ experiments on Type B somata indicate that increase of excitability induced by conditioning arises from a persistent decrease of voltage-dependent K+ current (Fig. 9; Alkon et al. 1982a; Forman et al. 1984). These include the IA and the Ic*1+ - K+ outward K+ currents. A number of other studies were undertaken to determine some of the biophysical steps leading to this long-term reduction of K+ currents. It was already known that progressively enhanced light-evoked depolarization and cumulative membrane depolarization of the Type B cell occurs during acquisition of associative learning (Alkon, 1980b). During prolonged depolarization, prolonged elevation of intracellular Ca2+ was demonstrated by differential absorption spectrophotometry (Connor & Alkon, 1982, 1984). Elevation of intracellular Ca2+ by injection of Ca2+ causes prolonged reduction of both IA (Fig. 10; Alkon, Shoukimas & Heldman, 1982b) and currents (Alkon et al. 1984; Alkon & Sakakibara, 1984).
The biophysical sequence (Fig. 11) underlying the transformation of the Type B membrane during learning is most probably as follows: repeated pairings of light and rotation elicit cumulative membrane depolarization. This arises from (1) the genetically specified synaptic organization and (2) the response of this network during the interaction of the animal with its environment during the training stimuli. Cumulative depolarization is accompanied by prolonged elevation of the intracellular calcium within the soma, and perhaps in the axon and terminal branches. Elevated calcium causes inactivation of voltage-dependent K+ currents (IA and most probably ) which outlasts by many days the cumulative depolarization and elevation of intracellular Ca2+. Reduction of voltage-dependent K+ currents increases the excitability thus permitting an enhanced depolarization to a light stimulus. The greater depolarization and the greater number of impulses that will be generated cause (1) greater inhibition of the medial Type A photoreceptor, (2) increased inhibition of ipsilateral interneurones, (3) decreased depolarization of ipsilateral motor neurones and finally (4) decreased turning toward a light source.
The change from transient (lasting minutes to hours) cumulative depolarization and elevated intracellular Ca2+ to a longer lasting reduction of open voltagedependent K+ channels can most reasonably be thought to be mediated by biochemical steps. Ca2+-calmodulin phosphorylation of proteins which regulate or are a part of the K+ channels themselves is implicated. Electrophoretic analysis revealed a difference in phosphorylation of low molecular weight proteins from the eyes of conditioned but not control Hermissenda (Neary, Crow&Alkon, 1981). Phosphorylation of these proteins was in addition shown to be Ca2+-dependent (Alkon, 1983). Phosphorylation of one of these proteins (25 000 Mr) also changed under conditions which inactivated the same voltage-dependent K+ current (IA) that was reduced during days on which the associative learning was retained (Neary & Alkon, 1983). Finally, iontophoresis of a Ca2+-calmodulin-dependent protein kinase (phosphorylase kinase) caused a long-lasting reduction of the IA current (Fig. 12) which required some initial elevation of intracellular Ca2+ (Acosta-Urquidi, Alkon & Neary, 1982; Acosta-Urquidi, Neary & Alkon, 1984).
The essential link between the stimulus pattern (here paired light and rotation) and the change of ionic channels with learning is the training-induced voltage change of the Type B cell. This voltage change is an integrated effect of the entire visualstatocyst network as it responds to pattern repetition. This voltage change, and the calcium-mediated reduction of K+ channels it causes, contrasts with other neurohumorally-mediated phenomena. Such generalized physiological states as arousal are accompanied in vertebrates, for example, by increased levels of circulating epinephrine that exerts diffuse effects via neural receptors. In recent years neurohumoral agents such as octopamine, dopamine and serotonin have been suggested to have a similar role in invertebrates. Serotonin and octopamine, for example, have profound effects on the development and the ability to sustain muscular tension fcrayfish (Jacobs & Atwood, 1981), and on neuromuscular junctions of the lobster (Kravitz et al. 1980). Serotonin also produces effects on Aplysia similar to those observed after delivering a noxious stimulus such as shock or a ‘pinch’ (Camarado, Siegelbaum & Kandel, 1984).
These effects of pharmacological agents differ in many respects from the learning of an association between two specific stimuli as analysed for Hermissenda. First, a learned stimulus association is quite specific, while ‘state’ changes elicited by neurohumoral agents are generalized to many targets throughout the body. Second, the associative learning in Hermissenda persists for many weeks, while the pharmacological effects do not outlast the presence of the neurohumoral agent. Thus, there is no evidence that the neurohumorally-induced cellular changes involve storage of information, which persists beyond the initial experience of, for example, an arousal state. Nor mechanistically would the persistence of arousal effects be useful for the animal unless the arousing stimuli had been associated with other temporally-related events. Moreover, the effects of serotonin and octopamine are mediated by cyclic nucleotide phosphorylation (Batelle & Kravitz, 1978) rather than Ca2+-calmodulin-dependent phosphorylations involved in the learned association in Hermissenda (Acosta-Urquidi et al. 1982, 1984; Alkon et al. 1983a). The serotonin-induced effects on Aplysia neurones are also independent of calcium (Siegelbaum, Camardo & Kandel, 1982). It may, however, be premature to compare directly calcium-dependent modification of Hermissenda membrane channels during associative learning with mechanisms of arousal and sensitization. Thus, although serotonin has been thought to be responsible for long-term reduction of potassium channels during sensitization, their prolonged reduction has never been correlated with long-lasting sensitization of intact animals. Furthermore, it has been recently shown that serotonin is not released or synthesized from presynaptic interneurones which have long been thought to convey the sensitizing effects on the synaptic endings of the sensory cell (Hawkins et al. 1981; Kistler et al. 1983).
The integrated response of a population of neurones during conditioning results in long-term (days) changes of specific membrane currents within identified neurones. Prolonged elevation of intracellular calcium during conditioning causes a persistent increase of excitability by reducing K+ currents (IA and probably I(v+ - K+) in the membranes of identified somata. This Ca2+-mediated reduction of K+ currents, which encodes a learned stimulus association is thought to involve changes of Ca2+ -calmodulin - dependent phosphorylation of distinct membrane proteins. These changes are contrasted with the short-term regulation of currents by neurohormones during altered behavioural states such as arousal.
Ultimately some intelligent and testable extrapolation from relatively simple to more complex nervous systems may become possible. Most promising to date are the properties of ionic channels within neural membranes. The voltage-dependent calcium channels of the Hermissenda Type B photoreceptors have been found in Paramecium (Naitoh & Eckert, 1974), spinal cord motor neurones (Barrett & Barrett, 1976), dendrites of inferior olive neurones (Llinas & Yarom, 1981a,b) and pyramidal cells of the hippocampus (Clark & Wong, 1983). Voltage-dependent K+ currents such as IA and Ica2+-K+, are equally widespread. Prolonged membrane depolarization is known to occur in vertebrate nervous systems and could modify membrane excitability in exactly the same way as in Hermissenda. The necessary physiological prerequisites exist within vertebrate neural systems to use an invertebrate mechanism for associative learning. It will be extremely interesting to induce prolonged depolarization in cells with these prerequisites in brain slices to see if prolonged excitability changes, analogous to those of Hermissenda, would result. Encouragement has already been provided by the finding of prolonged elevation of intracellular calcium in somata of hippocampal pyramidal cells in response to a depolarizing stimulus (Krnjevic, Morris, Reifenstein & Ropert, 1982). It will be far more difficult, of course, to correlate such biophysical changes with associative learning of a cat or dog and then to prove that these changes are intrinsic to particular neurones. It will more difficult still to reach any conclusion as to what portion of a discrete association learned by a cat or dog resides within a given set of neurones and is critical for its recall.