Brain, Behavior and Evolution 2000; 55:241-247. Issue on Identifiable Neurons in Invertebrates edited by J.L. Leonard
Revisiting the Concept of Identifiable Neurons
Theodore Holmes Bullock
Department of Neurosciences and Neurobiology Unit, Scripps Institution of Oceanography
University of California, San Diego, La Jolla, Calif., USA
Identifiable . Equivalent cells . Addressable neurons . Redundancy . Giant neurons . Eutely . Command cells . Revolutions . Evolution of nervous systems . Circuits
Although eutely in nematodes was known, giant neurons in several taxa and unique motor neurons to leg muscles in decapod crustaceans, the idea that many animals have many identifiable neurons with relatively consistent dynamical properties and connections was only slowly established in the late 1960s and early 1970s. This has to be one of the important quiet revolutions in neurobiology. It stimulated a vast acquisition of specific information and led to some euphoria in the degree and pace of understanding activity of nervous systems and consequent behavior in terms of neuronal connections and properties. Some implications, problems and opportunities for new discovery are developed. The distribution of identifiable neurons among taxa and parts of the nervous system is not yet satisfactorily known. Their evolution may have been a case of several independent inventions. The degree of consistency has been quantified only in a few examples and the plasticity is little known. Identified neurons imply identifiable circuits but whether this extends to discrete systems, functionally definable, seems likely to have several answers in different animals or sites. Very limited attempts have been made to extend the concept to cases of two or ten or a hundred fully equivalent neurons, on all kinds of criteria. These attempts suggest a much smaller redundancy and vaster number of types of neurons than hitherto believed. Theory as well as empirical information has not yet interpreted the range of systems from those with small sets of relatively reliable neurons to those with large numbers of parallel, partially redundant units. The now classical notion of local circuits has to be extended to take account and find roles for the plethora of integrative variables, of evidence for neural processing independent of spikes and classical synapses, of spatial configurations of terminal arbors and dendritic geometry, of modulators and transmitters, degrees of rhythmicity (regularity varying several orders of magnitude), and of synchrony. Adequate language and models need to go beyond "circuits" in any engineering sense. Identifiable neurons can contribute to a broad spectrum of issues in neurobiology.
The notion that neurons can be individually named, and distinguished from any others in the same animal, and can be identified with individual neurons in other specimens of the same species, was well established, but not recognized as a widespread principle, long before the summary of invertebrate nervous systems by Bullock and Horridge (1965). Three classes of identifiable neurons had been known for decades and each class was regarded as quite exceptional or atypical among animal groups.
First were giant neurons in a few taxa, such as the cells of Mauthner in the medulla of many teleosts, the median and lateral giant nerve fibers of earthworms, crayfish and shrimp and the dozen or so giant fibers, one per stellar nerve in the mantle of squid - and a few other giant axons in minor phyla. Second was the whole nervous system of the eutelic nematode, Ascaris, shown by Goldschmidt in a series of monographs before 1910 to have consistently 162 neurons in the central nervous system in fixed positions and relations. The third class was the pattern of efferent axons to leg muscles in decapod crustaceans, shown by Wiersma in a series of papers in the 30s and 40s to be only two to six per major muscle and individually distinguished by their physiological properties and effects on the muscle. This case, like the others, appeared to be an exceptional specialization, in a category of odd occurrences, like the rare instances of syncytial neurons. It seemed unlikely to be common, having such unfamiliar features as multiple innervation of each muscle fiber, multiterminal innervation of each axon, parallel branching of all the axons to a given muscle, inhibitory axons to skeletal muscles and slow, medium and fast excitatory axons to the same muscle fibers.
The idea that such identifiable neurons were special cases of a widespread general class had hardly been considered when the Bullock and Horridge treatise went to press in 1963. Apart from two giant cells and some categories of cells in the visceral ganglion of the gastropod, Aplysia, defined by their responses to transmitter drugs or their tendency to spontaneous activity, it was not until the late 60s and early 70s that a few nongiant cells were recognized and named as consistent for the species. The names of Arvanitaki and Tauc are especially important in this advance, soon followed by Kerkut and others who identified a number of cells in another gastropod, Helix. Arthropods were close behind, led by crayfish and locusts where the previously known motor neurons, giant neurons and possibly unique command cells were soon joined by unique and identifiable nongiant interneurons. The 30-celled stomatogastric ganglion of lobsters forged ahead of the longer known 9-celled cardiac ganglion in identifiable cell properties. The early and middle 70s was a yeasty time of finding and characterizing cell after cell in other insects, crustaceans, polychaetes, leeches, nudibranchs and fresh species of tectibranchs and pulmonates. The important contributors can be found in the literature lists in this volume, as well as among the chapter authors. This personal essay does not attempt to be historical in selecting key references from a gradually accumulating data base or assigning credit carefully. It suffices for my purposes to say: it gradually became clear that some major development was emerging, at least for certain taxonomic groups.
By the mid- and late 1970s it was widely realized that several major taxa have a modest set of neurons that are identifiable by a combination of their connections and dynamic properties even more than by their position, shape or dendritic branches. Intracellularly transported markers, such as dyes and molecules that permit imaging played a crucial role while patient and skillful listing of physiological and pharmacological characteristics by penetration of recognized cells did the rest. The evidence has increased and spread - not so much to new classes of animals as to further classes of neurons in the same animals.
The gradual recognition of the widespread reality of identifiable neurons and their consistent connectivity and dynamic properties - in the mid- and late 1970s - must be reckoned as one of the major revolutions in neuroscience. It led to the tentative forays, then the hesitant speculation and finally the exuberant conviction that we have a good chance actually to unravel the whole neural circuitry for one and then another piece of normal behavior, including simple reflexes, and spontaneous decision-making and pattern-formation. The torrent of new information largely accounts for this and another collection of papers (Leonard, 2000b).
As that trickle became a stream, the concept and implications of it came under discussion, even more gradually. These implications and the problems they raise are chiefly unsettled today and are my main concern here. At this point, I turn from superficial history to a selection of issues and assessments.
First is the topic just alluded to - distribution. We still do not know how many of all the neurons in pulmonates, opisthobranchs, leeches, insects, crustaceans and lampreys are identifiable. The small ones and those otherwise hard to record singly are more difficult to identify consistently from individual to individual. Some experts have opined that most of the neurons in some of these groups are probably potentially identifiable, apart from certain clusters such as the bag cells in Aplysia. The present state of knowledge about the distribution of identifiable neurons among the taxa is curious in that sister groups of those just listed appear to lack this kind of cell. We do not know of identifiable neurons in platyhelminths, echinoderms, prosobranchs, pelecypods, cephalopods, arachnids, oligochaetes, vertebrates and many minor groups, except for giant and near-giant neurons in some species. The frontier is not yet fully explored and our information is inadequate on the existence of identifiable neurons in these groups, besides the giants here and there. I am sure surprises are in store in respect to distribution, both in the animal groups and in the anatomical regions involved. If improved information confirms the present impression that the phenomenon of unique neurons is common in many taxa and not in many others, it will sharpen the evolutionary issue of what functional consequences distinguish these two sets of taxa and what adaptive significance they may represent. "Successful" taxa are in each of the two groups; the advantages of identifiable cells or of their absence are not obvious
Another class of issues is subsumed under the question: how far does the consistent specification of a given, named neuron extend, comparing for example the branching of dendrites or the functional connectivity to other identified cells? This has been carefully studied in several favorable cases and is consequently fairly well known: it seems that within the defining feature-set for a given, named neuron there is a considerable range of variation between exemplars, e.g. in finer dendritic anatomy and in dynamic properties.
Less clear, however, are the questions whether the demonstration of identified neurons implies discrete, identifiable circuits of them and whether this means discrete systems for discrete behaviors. Much evidence has pointed, on the contrary, to systems of neurons with fuzzy membership, participating in a given behavior to variable degrees and each cell participating in several behaviors. Are these two, apparently contrary concepts, the discrete and the distributed forms of participation, both realized in different cases? My belief - or bet - is "yes", that nature presents a spectrum of forms of organization. A long debate over the reality of command cells and the proper limits of that concept may fit into this framework (Kupferman and Weiss 1978; Eaton and Didomenico 1985; Didomenico and Eaton 1988; Larimer 1988; Frost and Katz 1996).
Another view of the membership issue has been to define a category of addressable neurons, which include not only the single, identifiable cells but also those that belong to a group living close together, sharing functions to some degree (Bullock 1978). This term has not been widely taken up, although it is occasionally useful.
An effort to be more precise, at least in principle, was to widen the issue of enumerating distinctive cell types by recognizing the intermediate cases, where instead of single, unique identifiable cells, there are sets of equivalent cells - two or three or ten or a hundred fully equal, and distinguishable from all other sets of equivalent cells (Bullock 1980, 1993, chapter 7) . The idea led to estimating how many such sets there are, which is tantamount to the number of kinds of neurons there are. It was a major point of this exercise that, whereas authors usually speak of dozens, scores or hundreds of kinds of neurons, based on their anatomy, or their modulators or their input pattern preferences, it would be more appropriate to specify each "kind" on all of the possible criteria, including receptive and output fields.
Neurons of similar cytology, chemistry and dynamic properties but whose sensory fields are distinct, even though overlapping, on fingers, face and elsewhere, as well as those whose efferent fields are distinct, though overlapping even within a single finger muscle are functionally different kinds of neurons, in terms of genetic and epigenetic and plastic specification and of sensations and motor abilities. The rough estimates, offered as targets for correction, were 3,000 in Aplysia, 20,000 in a lobster, two million in a rat and many times more in humans (see page 131 of Bullock 1993). This concept may be more relevant in evolution than size of brain or number of cells and is worth new attention, although the bases for the estimates are necessarily rough. Neurons of the same equivalence set are fully redundant and the educated guess is that sets of more than a few thousand fully redundant neurons are probably rare. Relatively common is partial redundancy which means incomplete overlap and hence non-equivalent sets. Cephalopods and many vertebrates seem to have more of the large redundant sets and the partially overlapping sets and the addressable rather than unique, identifiable neurons than do the pulmonates, insects and leeches. We can only hypothesize that this achieves better signal detection in noise and hence more subtle discriminations and finely graded repertoires of response, since we lack good measures of these basic ethological entities.
The 1970s, 80s and 90s witnessed an enormous contribution in laboriously unraveled "circuits," justifying the bold, early assumption that circuits for chosen parts of natural behavior can now be worked out, because of the consistent properties of identifiable neurons that allow cumulative study of many preparations. What started as euphoria, based on a bizarre feature of some hairless invertebrates, became orthodox canon. Sometimes one had the impression that circuit chasing was the be-all and end-all of integrative neurobiology. The unfortunate workers on vertebrate animals, sadly, could not be expected to truly understand their behavior, since it lacked the advantage of identifiable neurons. Some enthusiasts defined the science of neuroethology as the explanation of natural behavior in terms of circuits of known cells (Hoyle 1984), perhaps influenced by the coincidence in timing of the rise of neuroethology as a recognized field and the spurt in knowledge of identifiable neurons.
Only the cognescenti in this field can fully appreciate the richness of the information that has accumulated. Suffice it to point to the 300 or so citations in Comer and Robertson's bibliography in this volume, just on insects. Databases already exist and a federation of them is under discussion. This will be a great step for comparative and evolutionary work as well as for computational neurobiology - an active area among these workers.
Besides unraveling circuits, other new opportunities and researchable questions became available with the realization that identifiable neurons are not outliers in rare and exceptional animals. Insights into development, plasticity, modulation and the relation of neurons to behavior are among the domains where identifiable neurons have contributed to general neurobiology. Many examples are in this book - like the findings that a neuron's sufficiency to mediate a behavior is not evidence of its necessity and that a neuron's role in behavior depends on its state (Larimer, 2000; Leonard, 2000a).
New levels of problems also began to arise as knowledge of simple circuits increased. Shepherd (1972) and Rakic (1975) introduced the idea of "local circuits," based on the emerging evidence of subthreshold and spikeless interactions in axonal terminals, dendrites and somata, mostly quite nonlinear. This term and the reality of such local, graded processes have become standard parts of the prevailing paradigm, for both identifiable cells and others. However, the issue is seldom articulated - probably because there is no near hope of answering it: how much of neural processing is independent of spikes and classical synaptic potentials. Our ability to recognize significant signals in graded, nonspiking, not-all-or-none, often slow changes, electrical, chemical or mechanical is so asymmetrical with the ease of detecting and interpreting as a neural signal even one spike, that the prevailing paradigm of neural operations, and of theory and modeling is heavily weighted on the latter side. We have not taken the retina seriously enough! Identifiable neurons are a prime opportunity for such basic neurobiological questions - if we can find them in systems like the retina, where several successive orders of neurons process information with few or no all-or-none impulses. The need is to record both fast and slow events simultaneously from closely spaced extracellular fields and single cells, intracellularly, to try to bridge this gap.
Is there no limit to the number of variable parameters of neurons whose integration determines their responses? The answer seems to be "no", as every year new interactions are reported among the countless dynamic properties of membranes, channels, transmitters, modulators, terminals, somata, dendrites, glomeruli, baskets, slower and faster events, their localization, synchrony, rhythmicity - each in a range of degrees, and commonly nonlinear. These are not peculiar to identifiable neurons but complicate the interpretation of even simple circuits. Discovering new dependencies is a major and active front of advances, but the already long list represents also a major obstacle to achieving explanation or prediction of the activity even in well known circuits of identifiable neurons. Selverston (1980) pointed out how frustrating it was to attempt to explain the simple behavior of the stomatogastric ganglion of a lobster, although the knowledge of the connectivity and properties of each of its 30 neurons was seemingly virtually complete.
Coming to terms with a level of complexity beyond any other known (except systems of brains), is a sine qua non of modern brain research! Strategy and tactics, perforce, have become more and more a personal choice. Simplification is necessary in order to model, explain or predict and to deal with a manageable system - and that means accepting the high probability that one chooses incorrectly which parameters are important. Identifiable neurons can help substantially in this choice since it may be safe to assume, as an approximation, that parameters in which they vary between specimens are likely to be less important - but this strategy has not been much exploited explicitly.
A permutation of the problems raised by sheer complexity is that even simple neural systems seem able to achieve the same behavior in various ways. Identifiable neurons have been particularly important in revealing evidence of this. Hoyle (1977d) became convinced that locusts can perform what seemed to be the same kind of walking in any of several ways; that is, the ientified neurons participate in a variety of phases and degrees. Some current authors dispute this.
The basic issue whether there are rules, for example in the use of certain transmitters or modulators or synaptic properties for certain types of behaviors or of classes of subsystems, has made very little progress. Identifiable neurons would seem to offer advantages but so far information is too meager to make correlations with significant numbers of examples. Perhaps the compilations alluded to above will help.
This situation applies as well to questions such as which systems have useful degrees of plasticity and which systems depend on numbers of parallel units to average out significant stochastic variation as opposed to systems employing small sets of relatively reliable neurons (Bullock, 1993; Buchanan, 2000; Eatin et al., 2000; Larimer, 2000). On present information, each type of system is found to be adequately working in some animals and subsystems.
An impression exists in some quarters that stochastic
variation has become more evident with the increase in knowledge of behavior and its substrata and hence that behavior is chaotic and indeterminate. My view, without denigrating the significance of the issue, is that most evidence for such conclusions is in a class with the conclusion that people are stochastic because x% of them vote Republican or choose red automobiles and y% vote Democratic or choose white cars. In a certain sense, these may be called stochastic but what is more heuristic to me is evidence that some systems are highly reliable, whether because of averaging large numbers of unreliable components or otherwise. Very few people pay $0.51 or $0.49 when the sign says "Fifty cents" and very few cats fail to land on their feet when dropped. Unexplained variation in behavior is weak evidence for noisy indeterminacy but strong evidence for unknown modulating factors.
Identifiable neurons and simple circuits call for a new terminology and imagery to go beyond that of circuitry, since "circuits" in the electronic engineering sense do not depend on the great number of kinds of elements (neurons, transmitters and modulators, functional personalities and proclivities), or the "local circuit" features and the diversity of geometric arrays of axonal terminal arbors and dendritic ramifications so characteristic of neural organization (Bullock 1993, 1997). Although we still lack compelling evidence of their relative importance, I would tentatively emphasize the extent of extrasynaptic and non-classical synaptic, graded communication between neurons, the evidence of field effects, both chemical and electrical, out to several microns or tens of microns, as in the neural control of smooth muscle and the Mauthner's cell axon cap, in axonal varicosities and axonal free endings in neuropile. Active as well as passive roles of neuroglial cells are now documented in a wide array of preparations (Laming 1998). The imagery and simplifications of circuits of axonal spikes and classical synapses are useful in first approximations but cannot be considered realistic in the general case, any longer. Mpitsos (2000) presents other arguments for a similar conclusion.
The reality of identifiable neurons, already well established, is abundantly manifested in this volume - and the concept is alive and well. Its many implications, associated problems and opportunities, some of them mentioned here, are variously developing or awaiting exploitation. The only certainty is that these developments or exploits will impact general neurobiology, including vertebrate and mammalian understanding and, more than likely, lead to overturns of long favored orthodoxies on how brains work.
Buchanan, J.T. (2000) Contributions of identifiable neurons and neuron classes to lamprey vertebrate neurobiology. Prog. Neurobiol., in press.
Bullock, T.H. (1978) Identifiable and addressed neurons in the vertebrates. In: Neurobiology of the Mauthner Cell (ed. by D.S. Faber, and H. Korn), Raven Press, New York, pp. 1-12.
Bullock, T.H. (1980) Reassessment of neural connectivity and its specification. In Information Processing in the Nervous System, (ed. by H.M. Pinsker and W.D. Willis, Jr.), Raven Press, New York, pp. 199-220.
Bullock, T.H. (1993) How Do Brains Work? Papers of a Neurophysiologist. Birkhäuser, Boston.
Bullock, T.H. (1997) Signals and signs in the nervous system: the dynamic anatomy of electrical activity is probably information-rich. Proc. Natl. Acad. Sci., USA, 94:1-6.
Bullock, T. H. and G. A. Horridge. (1965) Structure and function in the nervous systems of invertebrates. W.H. Freeman, San Francisco.
Comer, C.M. and R.M. Robertson (2000) Identified nerve cells and insect behavior. Prog. Neurobiol., in press.
Didomenico, R. and R.C. Eaton. (1988) Seven principles for command and the neural causation of behavior. Brain Behav. Evol., 31:125-140.
Eaton, R. C. and R. Didomenico. (1985) Command and the neural causation of behavior: a theoretical analysis of the necessity and sufficiency paradigm. Brain Behav. Evol., 27:132-164.
Eaton, R.C. R.K.K.Lee, and M.B.Foreman (2000) The Mauthner cell and other identified neurons of the brainstem escape network. Prog. Neurobiol., in press.
Frost, W. N. and P. S. Katz (1996) Single neuron control over a complex motor program. Proc. Natl. Acad. Sci. USA, 93:422-426.
Hoyle, G. (1977) Identified Neurons and Behavior in Arthropods. Plenum Press, New York.
Hoyle, G. (1984) The scope of neuroethology. Behav. Brain Sci., 7:367-412.
Kupferman, I. and K. R. Weiss (1978) The command neuron concept. Behav. Brain Sci., 1:3-39.
Laming, P. (1998) Neuroglia and Animal Behavior. Cambridge Univ Press.
Larimer, J.L. (1988) The command hypothesis: a new view using an old example. Trends Neurosci., 11:506-510.
Larimer, J.L. (2000) The interneurons of the abdominal positioning system of the crayfish. How these neurons were established and their use as identified cells and command elements. Brain Behav. Evol., 55:241-247.
Leonard, J.L. (2000a) Network architectures and circuit function: testing alternative hypotheses in multifunctional networks. Brain Behav. Evol., 55:248-255.
Leonard, J.L. (2000b) 'Simpler' Systems in Comparative and Integrative Neurobiology. Prog.Neurobiol., in press.
Mpitsos, G.J. (2000) Attractors: architects of network organization? Brain Behav. Evol., 55:256-277.
Rakic, P. (1975) Local circuit neurons. Neurosci. Res. Prog. Bull., 13:290-446.
Selverston, A. I. (1980) Are central pattern generators understandable? Behav. Brain Sci., 3:535-571.
Shepherd, G.M. (1972) The neuron doctrine: a revision of functional concepts. Yale J. Biol. Med., 45:584-599.