Comparative strategies in the investigation of neural networks

Comparative Strategies in the Investigation of Neural Networks

DONALD KENNEDY Department o f Biological Sciences, Stanford Universi ty , Stanford, Cal i fornia 94305

ABSTRACT Comparative studies on nervous systems, though infrequently undertaken for the purpose of comparison, have yielded some important generalities about the formats of nervous networks, and about the cell biology of certain neural types. In the first category, it is clear that convergent evolutionary processes arrived at very similar networks to accomplish reciprocal and lateral inhibition, and load-compensation in “resistance reflexes.” A newer general network format is described, command-derived inhibit ion, in which the central nervous elements controlling a rapid movement deliver presynaptic inhibition to the terminals of sensory neurons that carry reafferent excitation from the movement. It is argued that such circuits occur in several groups of animals, and that they include as a special class the efferent inhibitory neurons innervating acoustico-lateralis receptors in vertebrates.

The properties of circuit elements that now seem to constitute useful generali- zations include size principle (the inverse relationship between size and excitability in a variety of neurons), and the late differentiation of sensory neurons, failure to decussate, and their inability to mediate inhibition. Many other generalities have emerged, only to fall; one conclusion from such searches is that many supposedly “basic” properties of cell types or neural circuits are in fact not phylogenetically conservative, however much the physiologist may expect them to be.

I hope the other participants in this symposium forgive me for revealing that some of us “comparative” physiologists pa- rade under false colors. Our proper role is held to be the analysis of adaptation - that marvelous process that suits system to environmental demand. We argue that special evolutionary contexts will reveal function in its starkest form: the super- cooled fish, the kidney up against it in the desert, the moth fleeing the bat.

But many of us are closet generalists instead. While pretending to seek out the unique products of adaptive radiation, we really hope to find ourselves! It is our se- cret, Walter Mitty wish that in the oddest animal (preferably from the most exotic, far-off place) we can describe a physiological mechanism that exists in everything from abalones to zebras, including especially man, but which mysteriously has eluded everyone else. This great truth ex- plains why so many reputable comparative physiologists are actually one-animal chau-

J. EXP. ZOOL., 194: 35-50.

vinists. We get away with this masquerade only because medical physiologists are so anxious to have a category into which to put us.

This revelation applies especially to the neurophysiologists, whom we find engaged in an international treasure hunt for the best preparation. The criteria are simple enough : the nervous system should con- sist of not more than a hundred large cells, each richly pigmented; its owner should have a generation time of less than an afternoon. When the Mystery Animal is discovered, all of us will leave locusts, crickets, crayfish, Aplysia, Tritonia, Pleu- robranchaea, Navanax, Helisoma, and the leech, gleefully abandoning our pretense. The purpose of this paper is to examine what we will have left behind - it is a

1 Work from the author’s laboratory was supported by grants from the National Institutes of Health, U. S. Public Health Service. The collaboration of Ronald Calabrese, Jeffrey J. Wine and Robert S. Zucker on some of the recent experiments reported here is gratefully acknowledged.

35

36 DONALD KENNEDY

sifting through the rummage of explora- tion.

Nervous systems can be examined com- paratively at three levels. The first, which will not concern us here, is that of gross structure: tracts, connectives and brain parts. This highest level of nervous system organization obviously offers much; but so far the comparative information we have is mainly in terms of anatomy. I will con- centrate on two deeper levels that seem more functionally meaningful.

The first is that of nerve networks. In investigating an ensemble that controls a particular behavior, we focus on circuit properties: that is, on the arrangement of connections among elements that allow the operation to be performed in a certain way. What we try to deduce is the network format. This format is really independent of the basic units; that is, a circuit could be made of very different starting materials (sensory neurons, motoneurons, etc.) and yet accomplish exactly the same result.

Second, there are the elements themselves. Here we inquire whether specific properties can be identified with a certain kind of element. Do epithelially derived sensory cells have universal qualities? Do motoneurons? Is the ability to produce chemically mediated synaptic inhibition limited to cell types having a particular origin? If we could answer such questions, we might be able to predict much more quickly how nervous systems evolve solutions to problems in network design.

NETWORK FORMAT

Obviously, only a fraction of the classes of neural networks found in higher animals have been investigated in the depth necessary to say whether their formats have anything in common. Nevertheless, several connection patterns appear so ubiq- uitous that they must be regarded as preferred solutions to a given kind of op- erational demand. I will discuss three ex- amples.

( 1 ) Lateral inhibition networks Reciprocal inhibition between popula-

tions of neurons that normally discharge out of phase with one another is perhaps the oldest proposal to account for recipro-

cal motor output. Originally invoked to explain descending respiratory drive in mammals, it was also suggested as a mechanism for the alternating output to ele- vator and depressor muscles in insect flight (Wilson and Waldron, ’68). While it seems likely that such connections do function in the ventilatory control centers, a t least of lower vertebrates, reciprocally- coupled “half-centers” have not been found in locomotor control systems.

A more general case of inhibitory in- terconnection, lateral inhibition was first identified in the Limulus eye by Hartline et al. (’56) and has since been shown to be responsible for the enhancement of sta- tionary brightness contrast in the vertebrate visual system (for review see Dowl- ing and Werblin, ’71) . It probably serves analogous functions in pitch discrimina- tion in vertebrate audition. The inhibition between contralateral auditory interneurons in insects is similar, though only pairs of cells are involved (Katsuki and Suga, ’60). Although the contrast-enhanc- ing properties of such arrangements for sensory systems have received the most emphasis, similar networks occur in circuits for generating movement; motor units innervating synergistic indirect flight muscles in flies are linked by mutual inhibitory connections that maintain phase relations between the units (Wyman, ’69).

( 2 ) Resistance reflexes and active movements

In arthropods and vertebrates, many skeletal muscles are equipped with receptors that are associated with specialized muscle strands, the latter arranged in parallel with the main muscle. The sensory neurons discharge when the muscle is stretched passively, but are silenced during active contraction because the receptor muscle is unloaded by shortening of the parallel “working” muscle fibers. In the muscle spindles of vertebrates, the specialized receptor muscle strands receive private motor innervation from gamma (= fusimotor) efferent neurons; activity in these elements has the effect of increasing sensitivity of the spindle to imposed stretch. In the spinal cord, the effect of spindle excitation is to excite motoneurons returning to the muscle of origin, and to

COMPARATIVE STRATEGIES IN THE INVESTIGATION OF NEURAL NETWORKS 37

inhibit antagonist motoneurons. Sherring- ton called such circuits myostat ic reflexes, but the term resistance reflex is more de- scriptive because the effect is to maintain muscle length against applied stretch.

At first it was supposed that the resistance reflexes were mainly postural; that is, that their main function was to con- serve primary orientation against passive influences. More recently, a role for resistance reflexes in active movements has been demonstrated. In the mammalian respiratory system, muscle spindles associated with inspiratory intercostal muscles discharge on inspiration, not expiration; this is a surprising result, because activity in the muscle containing the spindle should unload the latter, and reduce the frequency of sensory discharge. The ob- servation makes sense only if fusimotor efferent neurons are activated along with motoneurons innervating the working muscle fibers. Such coactivation has been confirmed in respiratory muscle spindles (Critchlow and von Euler, ’63) and also in those of the human hand (Vallbo, ’71). Under these conditions, loading of the system (achieved in the case of respiration by obstructing the airways) should produce differences between the velocity of contraction of the spindle intrafusal fibers and that of the extrafusal fibers. The lag in the latter should be translated into spindle tension, which is thus proportional to load; in turn, excitation will be fed to the muscle reflexly, to overcome that load. Convincing evidence for this kind of compensation in the respiratory system has been provided by Corda et al. (’65).

Two important features of such circuits in mammals are: (1) parallel stretch receptors excite motoneurons returning to the muscle, but do no t excite the fusimotor supply to the receptor itself; (2 ) central commands for voluntary movements acti- vate the motoneurons to the muscle and the efferent nerve supply to the receptor together.

Muscle receptor organs in the decapod crustacea ( Alexandrowicz, ’51 ; Eyzaguirre and Kuffler, ’55) resemble vertebrate muscle spindles in their parallel arrangement with the dorsal abdominal extensor muscles, in their excitation by stretch, and in the possession of an efferent innervation

(Kuffler, ’54). Some years ago we analyzed their central connections : these are arranged in exactly the same way as those of the muscle spindle, with the central pro- jections of the afferent neuron supplying excitation to an identified motoneuron that innervates the parallel extensor muscles (Fields, ’66). The efferent innervation of the receptor muscle is shared with ordinary extensor muscle fibers, so coactivation during “voluntary” extension is auto- matic if it involves motoneurons that innervate both muscles. In later experiments (Fields et al., ’67) we showed that central command pathways selectively ac- tivate these motor elements. As would be predicted, the muscle receptor afferents discharge during active extension, just as the inspiratory intercostal spindles fire during inspiration (Fields, ’66). A final parallel is that the central connections from the receptor afferents are restricted to a motoneuron that innervates only the working muscles, and not the receptor muscle. As in the mammalian respiratory system, there is evidence that central commands are actually load-compensated (So- kolove, ’73). A circuit diagram is shown in figure 1.

Down to the finest detail of organization, the motor circuit parallels that in which the vertebrate muscle spindles are involved. Yet no elements in the systems bear any phylogenetic relationship to one another. From separate starting materials, arthropod and vertebrate have assembled load-compensating networks that can be represented by identical circuit diagrams, yet share few if any elementary properties.

( 3 ) Command-derived inhibition All organisms that engage in sudden,

violent movements run an attendant risk. Such behavior inevitably excites receptors associated with the moving parts, especially the muscles and joints. Put in this way, the notion is hardly surprising: indeed, one of the functions of muscle and joint receptors, as we have seen, is to de- tect the stresses or position changes gen- erated by programmed movements, and to initiate error- or load-compensating reflexes. A different sort of problem arises when the activated receptors are external ones that mediate fixed action patterns

38 DONALD KENNEDY

Reccptor muscle / r

I MpoI I I Shared motoneu r-on

Command Ciber Fig. 1 Diagram of the load-compensating circuit for abdominal extension in the crayfish.

Central command interneurons selectively excite motoneurons that innervate both the functional extensor muscles and the muscle associated with the muscle receptor organ (MRO). Tension in the latter is converted into excitation of a motoneuron ( # 2 ) which innervates the functional muscles selectively. For discussion,

critical to the organism's survival (e.g., escape movements). Movements that produce intense, reafferent excitation might evoke inappropriate responses that would interfere with the movement already in progress, or prevent the detection of sig- nals impinging during its execution. Be- cause the afferent synapses often lose efficacy during intense bouts of activity, periods of insensitivity might follow each movement - especially in aquatic animals, where cutaneous tactile receptors, neuromast cells in the lateral line organs, and mechanoreceptors associated with hairs are exquisitely sensitive to deforma- tions of the medium. In two groups of aquatic animals, central pathways linked to the system of motor command appear to exert a control over recurrent excitation. In Xenopus and the dogfish, active movements are preceded by efferent discharge to the lateral line receptors; these prepara- tory discharges have the effect of inhibit- ing the spontaneous activity in lateral line afferents (Russell, '71; Roberts and Rus- sell, '72). The firm association of this event with subsequent movement argues for its

see text (From Kennedy, '69).

derivation from the command pathways themselves, although direct evidence is lacking.

A central mechanism for achieving the same outcome has been worked out for decapod crustaceans. Roberts ('68) first showed that the lateral and medial giant fibers of crayfish, which are the command elements for the tail-flip escape responses, produced recurrent inhibition directed against themselves and also against other central neurons. Krasne and Bryan ('73), in an ingenious and important experiment, demonstrated that this pathway also inhibited the first-order synaptic junctions between afferents from hair mechanoreceptors and the central interneurons. Post- synaptic changes in membrane conduc- tance are a component of this inhibition, but Krasne and Bryan showed that the inhibitory pathway also protects the afferent synapses against the decrement in efficacy they normally show during repetitive activity. This result seems consistent only with a direct effect upon the presynaptic terminals themselves that reduces the


amount of transmitter they release in response to each impulse.

In our laboratory we have recorded in- tracellularly from the afferent axons themselves in order to show directly that the inhibition is presynaptic and to unravel its mechanism. Before describing these experiments, it is necessary to emphasize some differences between the organization of sensory endings in arthropod neuropile and that in the spinal cord; these differences will be returned to later in the discussion. Electron micrographs of crustacean neuropile show frequent synaptic specializations along the course of appar- ently unbranched axons. This arrangement is true not only for large identified interneurons (Krasne and Stirling, '72); it also applies to many of the smaller processes from which one obtains fortuitous longitudinal sections. Furthermore, it applies equally to different types of presynaptic specializations, as judged by their vesicle complement. We therefore conclude that sensory axons, as well as central elements, must have this configuration. It is consistent with the appearance of sensory fibers filled with cobalt or other intercellular dyes: they appear to run directly into the center of ganglionic neuropile and to end suddenly, without much branching (San- deman and Okajima, '73; Calabrese, '75). And in at least one case, large identified sensory neurons - the afferents from the muscle receptor organs - make synapses in ganglia through which they pass without branching at all.

In all of these cases, junctions occur where post-synaptic neurons (known to branch wherever they receive input) send processes to contact the afferent axons. The latter possess varicosities, seen in electron micrographs as pouches of vesicles that expand beyond the characteristic axonal ring of peripheral mitochondria (fig. 2). A given axon may synapse with post-synaptic elements at a series of closely-spaced points along its length. A microelectrode penetrating an afferent axon in the region of neuropile where synapses occur with interneurons is therefore likely to be quite close to one or more sites of transmitter release. In the largest axons, which have diameters of 30 pM in the peripheral nerve just before they enter the

ganglion, all such sites would be within a space constant of the recording locus.

The results of our experiments (Ken- nedy et al., '74) are consistent with that conclusion. As a source of presynaptic inhibition, we used the lateral giant (LG) fibers; these axons are activated by mechanoreceptors on the abdomen, and mediate a special kind of tail-flip response that somersaults the animal and moves it away from the source of the stimulus (Wine and Krasne, '72). Thus they provide the commands for an escape response which is known to habituate (Krasne and Wood- small, '69) ; the habituation itself occurs at the very junctions that link the primary afferents to first-order interneurons (Zucker, '72b).

Responses recorded within a primary afferent neuron in the sixth abdominal ganglion are shown in figure 3. Stimula- tion of LG produces a large depolarization, which is augmented upon hyperpolariza- tion and decreased by depolarization of the membrane by injected current. The primary afferent depolarization mediated by LG activity has a time-course similar to that of the inhibition of excitatory post- synaptic potentials produced in interneurons by fourth-root stimulation (Kennedy et al., '74). Furthermore, the depolarization in primary afferent neurons produces substantial shunting of impulses if prop erly timed, as shown in figure 4. The inhibition is thus clearly presynaptic, as in the vertebrate spinal cord; moreover, it may operate by reducing the absolute amplitude of the presynaptic impulse, and thus limiting its capacity to release transmitter.

Other sensory pathways also generate presynaptic inhibition, and they appear to do so by activating a limited population of interneurons - the same ones that mediate the recurrent effect produced by LG stimulation. Depolarizations in primary afferent fibers are evoked by shocking peripheral nerve roots, but occlusion is shown when these responses are added to those evoked by LG activity. It appears that only half a dozen interneurons are direct sources of such primary afferent depolarization, and that both sensory pathways and the motor command pathways con- verge upon them. Indeed, preliminary evi-

40 DONALD KENNEDY

Fig. 2 Longitudinal ( a ) and transverse ( b ) sections through presynaptic axons in the neuropile of the third abdominal ganglion of the crayfish. Synaptic vesicles are located pe- ripherally, in expansions of the unbranched axons; these varicosities are outside the ring of mitochondria that normally encircle the periphery of the axon, and lack microtubules. Mag- nification: ( a ) x 10,800; (b) x 27,200.


f - --

Fig. 3 Primary afferent depolarization recorded within an afferent fiber in the neuropile of a crayfish abdominal ganglion, in response to stimulation of the lateral giant neuron. In the top record, a depolarizing current pulse is passed through the recording microelectrode using a bridge circuit; in the bottom record the pulse is hyperpolarizing. The middle record shows the response at resting membrane potential. cm, current monitoring trace; ic, intracellular record. Vertical calibration, 10 mv, 1 na; horizontal calibration, 10 msec. From Kennedy, Calabrese and Wine, '74.

A

dence of two kinds suggests that these inhibitory interneurons distribute their effects both to the afferent and post-synaptic sides of first-order synapses : the drug pic- rotoxin blocks both pre- and post-synaptic effects, and similar temporal irregularities are observed in both when they are simul- taneously recorded. If these observations stand up, then the central inhibitory interneurons resemble some crustacean peripheral inhibitors in that they act upon terminals as well as upon the postsynaptic membrane.

Recurrent inhibition from motor command pathways appears to be of widespread occurrence. When crickets walk, input from the caudal cerci to central interneurons is inhibited; it seems likely, though unproven, that the influence is partly presynaptic (Murphey and Palka, '74). Inhibitory efferent impulses to semi- circular canals in larval amphibians (Klinke and Schmidt, '70) are associated with attempted or executed movements. In frogs, the electrical signs of presynaptic inhibition may be evoked by antidromic stimulation of motoneurons; the effect, significantly, is much larger from hindlimb extensors (the pathways involved in the jump) than from flexors (Carpenter and Rudomin, '73). In each of these instances, the recurrent inhibition is associated with movements that would evoke violent reafferent activity.

L

Fig. 4 Shunting of the amplitude of an impulse in a primary afferent fiber by repetitive stimulation of the lateral giant. The lower trace monitors the activity in the connective adja- cent to the ganglion i n which the microelectrode recording is made. In A, a root shock evokes activity in the afferent neuron (upper trace); in B, the impulse so produced rides on the compound primary afferent depolarization evoked by repetitive stimulation of the lateral giant neuron (large responses in the lower trace). The absolute amplitude of the impulse from baseline to crest is smaller in B; more important, the absolute value of membrane potential reached at the peak is also less. Calibration: 20 mv (upper trace), 10 msec. From Kennedy, Calabrese and Wine, '74.

42 DONALD

Taken together, these findings from a diverse group of vertebrates and invertebrates suggest a rationale for presynaptic inhibition that has not been widely recog- nized. An interneuronal system to link motor commands with the suppression of reafference would provide a widespread network for distributing inhibition to sensory endings in the central nervous system; in a later stage, such a network could ~ e r v e to mediate inhibitory interactions among input pathways.

PROPERTIES OF CIRCUIT ELEMENTS

At the level of single neurons and junctions, the value of the comparative approach seems less dea r than for networks. The basic inventory of neuronal abilities appears, on first glance, to be roughly similar at all levels. But if certain neuronal types - identified by place of origin, morphology, or phylogenetic homology - ex- hibit constraints in their interactions, then we can predict corresponding limitations in the properties of the nerve circuits in which they participate. In what follows, I shall try to pull together some emerging generalities about the properties of such neuronal elements, and survey the impli- cations for nervous organization. This is a dangerous activity, of course, because for every premature generalizer the world contains at least one exception-ferret.

Cell size There is a general correlation among

neurons of all types between size, threshold, and tendency to accommodate. This tendency was first noted by Bullock ( ’ 5 3 ) ; it was documented in detail for mammalian motoneurons by Henneman and his co-workers (Henneman et al., ’65) and for crustacean motoneurons by Davis (’71). The size principle attributes the higher thresholds and more rapid adaptation (phasic nature) of large cells to their lower effective input resistance.

In a number of invertebrates and lower vertebrates, neurons of exceptionally large size - “giant” cells - have evolved. In nearly every instance they serve equivalent functions, as command elements for fast motor responses. For these purposes the physiological correlates of large size are

KENNEDY

strikingly appropriate: one or a few impulses trigger the responses, and these are evoked only at high levels of input. Earlier speculation about the selective advantage of size in such neurons has focussed on the conduction speed offered by large diameter. But there are compelling arguments against this explanation. In the most com- pletely known escape circuit, that involving the lateral giant cells in crayfish (Zucker et al., ’71), the minimum pathway from receptors to muscles includes four synapses, as well as sensory and motor conduction time from the periphery along non-giant axons. The reflex time saved by converting the command element from non-giant to giant size (that is, increasing its diameter from 40 to 200 pM) is, at the most conservative estimate, about one msec. - in a reflex with a latency of well over 15 msec!

It seems more likely that size confers a different kind of advantage. We do know that the giant fibers, like other central neurons in crustacea, make their output connections without branching, Many of these connections, including all of those with the fast flexor motoneurons, are electrical (Zucker, ’72c), and may be presumed to re- quire substantial areas of membrane contact. Possibly, given their style of connection, the only way in which these command interneurons can accept enough post-synaptic processes is by increasing their diameter.

Whatever the explanation for their size, it has important secondary outcomes for input processing, as indicated by the reconstruction in figure 5. A Iarge axon, associated with a dendritic tree of ordinary dimensions (Remler et al., ’68) presents a very low cable resistance to excitatory post-synaptic potentials initiated on the dendritic branches. This requires that very Iarge synaptic potentials be produced there. Zucker (’72a) showed that EPSP’s of over 50 mV in amplitude recorded in the LG dendrites were still subthreshold for dis- charging the axon. At this leveI of depolarization, the summation of chemicaI EPSP’s would be highly non-linear, because the equilibrium potential for the EPSP is closely approached. This may explain why the input to these cells is en-

COMPARATIVE STRATEGIES IN THE INVESTIGATION OF NEURAL NET WORKS 43

Fig. 5 Reconstruction of the lateral giant neuron. The cell was injected with Procion Yellow, and serial sections were made through the third abdominal ganglion; fluorescent pro- files were sketched on plastic sheets from each section, and the sheets were stacked to scale and photographed. A motoneuron was injected in the same preparation; its soma appears in the foreground of the reconstruction. The lateral giant neuron soma is dimly seen in the background, which is caudal; all the dendritic structures shown are part of LG, as is the axon emerging at the front. Reconstruction made by Allen Selverston.

tirely electrical. Because convergence is high, a large proportion of the dendritic membrane may be involved in low-resistance connections; if these were non-recti- fying it would further lower the effective input resistance of the neuron and raise its threshold. Recently, however, we have shown that most of the junctions are rec- tifying: afferent terminals that make excitatory electrical connections with LG do not respond to antidromic spikes in LG (Calabrese, '74).

A large neuron having relatively low effective input resistance, therefore a short time-constant, and excited only by brief electrical EPSP's will respond only to intense and well-synchronized input. These properties make the LG neuron an effective high-pass filter. A single impulse in

LG is sufficient to evoke a full program of escape behavior. By contrast, command interneurons for postural movements are smaller cells that must discharge continu- ously in order to produce their motor effects (Evoy and Kennedy, '67).

MORPHOLOGY AND INTEGRATIVE CAPACITY

Apart from size, the distinctive charac- teristics of a given neuronal type are morphology and embryonic origin; both have an important impact on function. This relationship - the one linking cellular ge- ometry with synaptic properties, integrative capacity, and other features - has only recently been explored intensively in a wide range of nervous systems. The first intracellular recordings from large somata

44 DONALD KENNEDY

in invertebrates, and the first microelectrode investigations of neuropile, are less than two decades old; and the total information available on this subject has surely more than doubled in the time since the new intracellular staining techniques (Pro- cion yellow, cobalt, and horseradish perox- idase) were introduced (see, e.g., Kater and Nicholson, ' 7 3 ) . Some generalities are now possible, though subject to the same cautions stated earlier.

Sensory neurons It was once easy to distinguish between

the somatic sensory neurons of invertebrates and vertebrates. In the former - at least in the arthropods and a scattering of less well-studied cases - sensory cell bodies were peripheral, located under the sensory structures they innervate, and derived from epithelial cells. In the latter they were para-central and derived from the neural crest, with a few exceptions like the neuromasts of the acoustico-lateralis system. This distinction has been blurred by the discovery in molluscs (Byrne et al., '74), arthropods (see below), and annelids (Nicholls and Baylor, '68) that some ap- parently first-order sensory neurons - in each case a minority of the total -have cell bodies located in the central nervous system. These cells differ in various other ways from typical populations of afferent elements in the same organisms. The central cells lack the basic bipolar form of invertebrate sensory neurons; they appear to differentiate early, and to show fixed numbers and highly specific connections. As an illustration, the large receptor cells in the crustacean Emerita shown in figure 6A (and related joint receptors in the basal segments of other crustacean limbs) have multiple processes in the neuropile of the ganglion in which they are found; large dendrites project to the periphery and conduct only graded potentials into the ganglion (Paul, '72). By contrast, peripheral sensory neurons in crustacea are bipolar, even when their somata are dis- placed toward the central nervous system in the ganglionic roots (fig. 6B), and their dendrites are electrically excitable (Mellon and Kennedy, '64; Pabst and Kennedy, '67). Such typical sensory neurons in ar-

thropods are added throughout life (Gymer and Edwards, '67; Kennedy, '73).

Obviously the position of the soma does not by itself yield a reliable diagnosis of which kind of cell one is dealing with. In an example just cited, the somata of bipolar sensory neurons can occur almost centrally. Conversely, at least one peripheral sensory neuron appears to resemble the central type in all respects save its location : the sensory cell belonging to the muscle receptor organ of Alexandrowicz ('51 ) has multiple processes, including dendrites that produce graded generator potentials, and it is differentiated early in development.

Most of the general statements made below apply to the "classical" somatic sensory neurons in invertebrates, and to the vertebrate afferents that have somata in the dorsal root ganglia.

1. Sensory neurons do not decussate. In vertebrates, all known interactions between afferent fibers and central elements occur on the side of entry; the midline of the spinal cord is not crossed by sensory cell processes. The same thing can now be said for several major classes of afferents in crustacea (e.g., Calabrese, '74) : cobalt staining of sensory roots fails to reveal crossing processes (cf. Sandeman and Oka- jima, '73), and all interneurons that receive bilateral input in a segment do so by sending processes of their own to the other side of the ganglion. So far, similar restrictions apply to sensory neurons in the other bilaterally symmetrical phyla.

2. Sensory neurons cannot inhibit. Al- though other kinds of neurons show a broad range of synaptic competence, sensory neurons produce only synaptic excitation : in both vertebrates and invertebrates, inhibition is always accomplished by in- serting an interneuron. First established for the mammalian spinal cord, by the demonstration that antagonist inhibition upon motoneurons from Ia spindle afferents was disynaptic (Eccles et al., '56), this generality has been strengthened by work on a variety of systems in which pure sensory pathways produce, in central neurons, only EPSPs at a latency sufficiently short to be considered monosynaptic. In the crayfish, we have been able to search fairly intensively, in the course of other

COMPARATIVE STRATEGIES 1N THE INVESTIGATION OF NEURAI, NETWORKS 45

Fig. 6 A. Sensory cell from a proprioceptor in the last abdominal ganglion of Emerita; reconstruction from a Procion Yellow injection (From Paul, '72). B. Sensory cell from a cu- ticular receptor in the crayfish, stained with methylene blue; the soma is located along a ganglionic root (From Pabst and Kennedy, '67).

experiments, for unitary inhibitory post- synaptic potentials in post-synaptic central neurons. They occur often, but are always associated with activity in interneurons rather than in afferent fibers. The same appears to be true in annelids (Nicholls and Purves, '70). An interesting ability that sensory neurons do have, at least in crustacea, is to make excitatory synapses that are either chemical or electrical: we have shown that the same axon can pro-

duce the two types of actions on different post-synaptic cells.

Interneurons Interneurons - defined broadly as cells

having somata and all processes confined to the central nervous system -- may be too diverse to generalize about. Indeed, they are sometimes not even easy to iden- tify. In molluscs, nearly all neurons with central somata have axons that project

46 DONALD KENNEDY

centrifugally in peripheral nerves, and it is often unclear whether or not they innervate muscles or other effectors directly. At first the challenge to the molluscan neurophysiologists was to prove that some particular cell was a motoneuron; having met that successfully (e.g., Carew et al., ’74), they are now being asked to show that some cells are not!

1. Do interneurons specialize in excitation or inhibition? One question that can be approached is whether particular interneurons tend to produce a consistent synaptic action upon all the elements to which they are connected. In the form first offered by Dale, for the vertebrates, this principle held that all processes of a single neuron released the same transmitter. It was later supplemented by Eccles to insist that the sign of all the synaptic actions would also be the same. The addendum has not survived comparative scrutiny. In molluscs, Kandel and his associates (for review, see Kandel and Gardner, ’72) have shown that many putative interneurons produce diverse synaptic actions upon different (indeed, even upon the same) fol- lower cells. Double-action interneurons, however, remain a molluscan specialty de- spite fairly serious prospecting in arthropods; other phyla are poorly known in this regard.

2. Electrical connections are more common than we thought. Bennett, whose own work has done so much to widen our view of electrical transmission in the nervous system, has reviewed the occurrence of such junctions (Bennett, ’72). He points out that the main advantages of electrical transmission are speed (hence, presum- ably, the utility of electrotonic synapses in “escape” systems like the one described earlier in this paper), reciprocity, and temporal stability.

In our own work on crayfish we have been surprised by the emerging ubiquity of electrical junctions involving interneurons, even where the advantages listed above seem of doubtful relevance. All excitatory connections involving interneurons above first-order level in the abdominal tactile system are electrical: the circuit has the form of a cascade in which only the first junctions (at which sensory fibers are the presynaptic elements) operate

chemically. It is true of arthropod interneurons SO far (and let me emphasize that the number of cases is meager) that they make excitatory connections only electrically, and reserve chemistry exclusively for inhibition.

Motoneurons Motoneurons are more thoroughly known

electrophysiologically than any other cell type. In vertebrates, they excite muscle or other effectors by excitatory chemical transmission; where they connect with other cells in the central nervous system using chemical synapses, these are excitatory and employ the same transmitter (Ec- cles, ’64). This rule applies everywhere else, although the number of thoroughly analyzed cases is small. Interaction between motoneurons also occurs by direct electrical excitatory connections, in vertebrates (Grinnell, ’66), annelids (Ort et al., ’74), and arthropods (Evoy et al., ’67; Kendig, ’68; Mulloney and Selverston, ’74; Tatton and Sokolove, ’75). In a few cases these interactions are strong, notably where the motor output is a relatively ste- reotyped pattern and the function of a given motoneuron therefore bears a fixed relation to others. In general, however, motoneuron connections are weak, and they have not been found in molluscs at all.

Peripheral inhibitory neurons have a wide distribution in invertebrates (for review see Kennedy and Davis, ’ 75) , but they have never been found in a vertebrate. Like conventional motoneurons, they achieve their peripheral effects chemically. Electrical connections centrally between such cells are common, and may be quite strong (Evoy et al., ’67); where they exist, the same neuron technically has a double action, electrical excitation and chemical inhibition - a combination that probably occurs among interneurons as well, although it hasn’t been found yet. Finally, peripheral inhibitory neurons may have two targets, the muscle fiber membrane and the terminals of the motoneurons; in other words, pre- and post-synaptic inhibition is accomplished by processes of the same cell. We believe this may also be true of central inhibitory interneurons in crustacea, although the well-known pharmacological differences between pre-


and post-synaptic inhibition in the vertebrate spinal cord make it an unlikely pos- sibility there (for review see Schmidt, ’71).

Diversity A final note concerns the frailty of gen-

eralities - including all of the above! Of all of the areas of neurobiology, none has proven more tempting to generalizers than synaptic chemistry - beginning with the super-generality that “transmission at the synapse is chemical.” The kinds of problems that have arisen from more contem- porary generalities are well illustrated by the situation in arthropods, where a long list of experiments on crustaceans and insects have suggested that glutamate is the excitatory transmitter a t neuromuscular junctions. Acetylcholine, which plays this role in vertebrates, is instead now suggested as the sensory transmitter (Barker et al., ’72). But in the past two years, two crustacean systems have turned up in which the case for acetylcholine as a neuromuscular transmitter appears quite good. One is the junction between the largest motoneuron and the slow abdominal flexor muscles in the crayfish, which responds to iontophoresed ACh, but not glutamate, and can be blocked by curare (Futamachi, ’72). The other is a new junction between a motoneuron of the stomatogastric ganglion in the lobster and the skeletal muscle it innervates, at which biochemical as well as physiological evidence implicates acetylcholine (Marder, ’74).

Such results are disappointments only for those who take unity too seriously. The take-home lesson of comparative physiology is that to talk about t he neuromuscular junction - even t he crustacean neuromuscular junction - is a mistake if we mean by it that we expect them all to conform to exactly the same rules. On the other hand, the phylogeny of cell types and of neuronal ensembles clearly does impose some limits on the direction of their future evolution. One way in which we can come to know these constraints is by venturing generalities and then having the exceptions noted. This will help us understand other systems, because we can approach a network problem more efficiently if we can predict that “this kind of cell is likely to be able to do this, but not that.” To this process of refining experimental strategies

and making them more economical, both the premature generality-maker and the exception-ferret make important contribu- tions.

LITERATURE CITED Alexandrowicz, J. W. 1951 Muscle receptor or-

gans in the abdomen of Homarus vulgaris and Palinurus vulgaris. Quart. J. Microscop. Sci.,

Barker, D. L., E. Herbert, J. G. Hildebrand and E. A. Kravitz 1972 Acetylcholine and lobster sensory neurons. J. Physiol., 226: 205-229.

Bennett, M. V. L. 1972 A comparison of electrically and chemically mediated transmission. In: Structure and Function of Synapses, G. D. Pamas and D. P. Purpura. eds. Raven. New

92: 163-199.

- ,

YO;^, pp. 221-256. Bullock, T. H. 1953 Comparative aspects of

some biological transducers. Fed. Proc., 12: 666-672.

1974 Receptive fields and response properties of mechanoreceptor neurons innervating siphon, skin and mantle shelf in Aplysia. J. Neurophys- iol., 37: 1041-1064.

Calabrese, R. L. 1975 Crayfish mechanorecep- tive interneurons: I. The nature of ipsilateral excitatory inputs. J. Comp. Physiol., in press.

Carew, T. J., H. Pinsker, K. Rubinson and E. R. Kandel 1974 Physiological and biochemical properties of neuromuscular transmission between identified motoneurons and gill muscle in Aplysia. J. Neurophysiol., 37: 1020-1040.

Carpenter, D. O., and P. Rudomin 1973 The organization of primary afferent depolarization in the isolated spinal cord of the frog. J. Phys- iol., 229: 471493.

Corda, M., G. Eklund and C. von Euler 1965 External intercostal and phrenic a-motor responses to changes in respiratory load. Acta Physiol. Scand., 63: 391-400.

Critchlow, V., and C. von Euler 1963 Intercos- tal muscle spindle activity and its gamma motor control. J. Physiol., 168: 820-847.

Davis, W. J. 1971 Functional significance of motoneuron size and soma position in swim- meret system of the lobster. J. Neurophysiol., 34: 274-288.

Dowling, J. E., and F. S. Werblin 1971 Synap- tic organization of the vertebrate retina. Vision Res., Suppl. 3: 1-15.

Eccles, J. C. 1964 The Physiology of Synapses. Academic Press, New York, p. 316.

Eccles, J. C., P. Fatt and S. Landgren 1956 The central pathway for the direct inhibitory action of impulses in the largest afferent nerve fibers to muscle. J. Neurophysiol., 19: 75-98.

Evoy, W. H., and D. Kennedy 1967 The central nervous organization underlying control of antagonistic muscles in the crayfish. I. Types of command fibers. J. Exp. Zool., 165: 223-238.

1967 Discharge patterns of neurones supplying tonic abdominal flexor muscles in the crayfish. J. Exp. Biol., 46: 393-411.

Eyzaguirre, C., and S. W. Kuffler 1955 Proc- esses of excitation in the dendrites and in the

Byrne, J., V. Castellucci and E. R. Kandel

Evoy, W. H., D. Kennedy and D. M. Wilson

48 DONALD KENNEDY

soma of single isolated sensory nerve cells of the lobster and crayfish. J. Gen. Physiol., 39:

Fields, H. L. 1966 Proprioceptive control of posture in the crayfish abdomen. J. Exp. Biol., 44: 455-468.

1967 Refiex role played by efferent control of an invertebrate stretch receptor. J. Neurophysiol., 30: 859-874.

Futamachi, K. J. 1972 Acetylcholine: possible neuromuscular transmitter in crustacea. Sci- ence, 175: 1373.

Grinnell, A. D. 1966 A study of the interaction between motoneurons in the frog spinal cord. J. Physiol., 182: 612-648.

Gymer, A,, and J. S . Edwards 1967 The development of the insect nervous system. I. A n analysis of postembryonic growth in the termi- nal ganglion of Acheta domesticus. J. Morphol., 123: 191-197.

1956 Inhibition in the eye of Limztlus. J. Gen. Phys- iol., 39: 651-673.

Henneman, E., G. Somjen and D. 0. Carpenter 1965 Functional significance of cell size in spinal motoneurons. J. Neurophysiol., 28: 560- 580.

Kandel, E. R., and D. Gardner 1972 The synaptic actions mediated by the different branches of a single neuron. In: Neurotransmitters, Res. Publ. A.R.N.M.D., V. 50, Ch. 4, pp. 91-146.

Kater, S., and C. Nicholson 1973 Intracellular Staining in Neurobiology. Springer-Verlag, New York.

Katsuki, Y., and N. Suaa 1960 Neural merha-

87-119.

Fields, €3. L., W. H. Evoy and D. Kennedy

Hartline, H. K., H. C. Wagner and F. Ratliff

I . .. nism of hearing in insects. J. Exp. Biol., 37: 279-290.

Kendig, Joan J. 1968 Motor neurone coupling in locust flight. J. Exp. Biol., 48: 389404.

Kennedy, D. 1969 The control of output by central neurons. In: The Interneuron. M. A. Brazier, ed. Univ. Calif. Press, pp. 21-36.

__ 1973 Connections among neurons of different types in crustacean nervous systems. In: Neurosciences: Third Study Program. F. 0. Schmitt and F. Worden, eds. M.I.T. Press, Cam- bridge, Mass.

1974 Presynaptic inhibition: primary afferent depolarization in crayfish neurons. Science, 186: 451-454.

Kennedy, D., and W. J. Davis 1975 Motor control in invertebrates. In: Handbook of Physiol- ogy, 2nd Ed., V. I, Neurophysiology. E. R. Kan- del, ed. In press.

Klinke, R., and C. L. Schmidt 1970 Efferent influence on the vestibular organ during active movements of the body. Pflugers Arch. ges. Physiol., 318: 325-332.

Krasne, F. B., and J. S . Bryan 1973 Habitu- ation: regulation via presynaptic inhibition. Science, 182: 582-584.

Krasne, F. B., and Ch. A. Stirling 1972 Synap- ses of crayfish abdominal ganglia with special attention to afferent and efferent connections of the lateral giant fibers. 2. Zellforsch., 127:

Kennedy, D., R. L. Calabrese and J . J. Wine

526-544.

Krasne, F. B., and K. S. Woodsmall 1969 Wan- ing of the crayfish escape response as a result of repeated stimulation. Anim. Behav., 17: 4 16-424.

Kuffler, S . W. 1954 Mechanisms of activation and motor control of stretch receptors in lobster and crayfish. J. Neurophysiol., 17: 558- 574.

Marder, E. 1974 Acetylcholine: a transmitter in the lobster stomatogastric system. Abstr. 4th Ann. Mtg., SOC. for Neuroscience, p. 428.

Mellon, DeF., and D. Kennedy 1964 Impulse origin and propagation in a bipolar sensory neuron. J. Gen. Physiol., 47: 487499.

Mulloney, B., and A. I. Selverston 1974 Orga- nization of the stomatogastric ganglion of the spiny lobster. I. Neurons driving the lateral teeth. J. Comp. Physiol., 91: 1-32.

Murphey, R. K., and J. Palka 1974 Efferent control of cricket giant fibers. Nature, 248:

Nicholls, J. G., and I). A. Baylor 1968 Specific modalities and receptive fields of sensory neurons in the CNS of the leech. J. Neurophysiol.,

Nicholls, J. G., and D. Purves 1970 Monosyn- aptic chemical and electrical connections between sensory and motor cells in the central nervous system of the leech. J . Physiol., 209: 647-667.

1974 Neuronal control of swimming in the medicinal leech. 11. Identification and connections of motor neurons. J. Comp. Physiol., 94: 121-154.

Pabst, H., and D. Kennedy 1967 Cutaneous mechanoreceptors influencing motor output in the crayfish abdomen. Z. vergl. Physiol., 57: 190-208.

Paul, D. H. 1972 Decremental conduction over “giant” afferent processes in an arthropod. Sci- ence, 176: 680-682.

Remler, M. P., A. I. Selverston and D. Kennedy 1968 Lateral giant fibers of crayfish: location of somata by dye injection. Science, 162: 281- 283.

Roberts, A. 1968 Recurrent inhibition in the giant-fibre system of the crayfish and its effect on the excitability of the escape response. J. Exp. Biol., 48: 545-567.

Roberts, B. L., and I. J. Russell 1972 The activity of lateral line efferent neurons in sta- tionary and swimming dogfish. J. Exp. Biol.,

Russell, I. J. 1971 The role of the lateral-line efferent system in Xenopus Zaevis. J . Exp. Biol.,

Sandeman, D. C., and A. Okajima 1973 Stato- cyst-induced eye movements in the crab ScyZZa serrata. J. Exp. Biol., 59: 17-38.

Schmidt, R. F. 1971 Presynaptic inhibition in the vertebrate central nervous system. Ergeb. Physiol. Biol. Chern. Exp. Pharmakol., 63: 20- 101.

Sokolove, P. G. 1973 Crayfish stretch receptor and motor unit behavior during abdominal extension. J. Comp. Physiol., 84: 251-266.

Tatton, W. L., and P. G. Sokolove 1975 Analy- sis of postural motoneuron activity in cray-

249-25 1.

31: 740-756.

Ort, C. A., W. B. Kristan, Jr. and G. S. Stent

57: 435-448.

54: 621-641.


fish abdomen. 11. Coordination by excitatory and inhibitory connections between motoneurons. J . Neurophysiol., 38: 332-346.

Vallbo, A. B. 1971 Muscle spindle response at the onset of isometric voluntary contractions in man. Time differences between fusimotor and skeletomotor effects. J. Physiol., 218: 4055431.

Wilson, D. M., and I. Waldron 1968 Models for the generation of the motor output pattern in flying locusts. Proc. Inst. Elec. Electron. Eng., 56: 1058-1064.

Wine, J. T., and F. B. Krasne 1972 The organization of escape behavior in the crayfish. J. Exp. Biol., 56: 1-18.

Wyman, R. J. 1969 Lateral inhibition in a motor output system. I. Reciprocal inhibition in dip-

teran flight motor system. J. Neurophysiol., 32: 297-306.

Zucker, R. S. 1972a Crayfish escape behavior and central synapses. I. Neural circuit exciting lateral giant fiber. J. Neurophysiol., 35: 599- 620.

19721, Crayfish escape behavior and central synapses. 11. Physiological mechanisms underlying behavioral habituation. J. Neuro- physiol., 35: 621-637.

1972c Crayfish escape behavior and central synapses. 111. Electrical junctions and dendrite spikes in fast flexor motoneurons. J. Neurophysiol., 35: 638-651.

Zucker, R. S., D. Kennedy and A. I. Selverston 1971 Neuronal circuit mediating escape responses in crayfish. Science, 173: 645-650.

Documents

Comparative strategies in the investigation of neural networks