PERSPECTIVES OPINION
The cortex as a central pattern generator Rafael Yuste, Jason N. MacLean, Jeffrey Smith and Anders Lansner Abstract | Vertebrate spinal cord and brainstem central pattern generator (CPG) circuits share profound similarities with neocortical circuits. CPGs can produce meaningful functional output in the absence of sensory inputs. Neocortical circuits could be considered analogous to CPGs as they have rich spontaneous dynamics that, similar to CPGs, are powerfully modulated or engaged by sensory inputs, but can also generate output in their absence. We find compelling evidence for this argument at the anatomical, biophysical, developmental, dynamic and pathological levels of analysis. Although it is possible that cortical circuits are particularly plastic types of CPG (‘learning CPGs’), we argue that present knowledge about CPGs is likely to foretell the basic principles of the organization and dynamic function of cortical circuits.
Neuroscience has traditionally been dominated by an empiricist approach that partly stems from the influential work of Sherrington. In his research, Sherrington succeeded in identifying basic spinal reflexes that govern some of the functions of the nervous system and emphasized the role of sensory inputs in these reflexes1. In an extreme empiricist view, the nervous system would be seen to remain inactive until it is engaged by sensory inputs, at which time it would process the sensory information and generate an appropriate output. This simple input/output, reflexive view of the brain
proved to be very useful and generated the concept of the receptive field 2 , whereby the function of a neuron is defined by the sensory inputs that drive it to a maximum level of excitation and action potential generation. The idea of receptive fields and the resultant branch of scientific inquiry for which this model is used has proved to be extremely fertile and has greatly increased our understanding of neural circuits, not only in sensory regions, but also more generally with regard to brain function. Concurrently, a different tradition in neuroscience has emphasized the endogenous function of neuronal circuits 3 . Brown, a disciple of Sherrington, described how the isolated spinal cord in decerebrate and deafferented cats can generate patterns of activity that resemble those generated during motor behaviour, such as walking, in the absence of any sensory inputs4. This discovery led Brown to suggest the existence of a self-contained functional circuit, or central pattern generator (CPG), in which sensory inputs primarily have a modulatory role in the basic intrinsic function of that circuit. In his words, “the fundamental unit of activity in the nervous system is not that which we term the spinal reflex ... the functional unit is the mutually conditioned activity of the linked antagonistic efferent neurons” 4 . This view, incidentally, was partly endorsed by Sherrington towards the end of his life when he argued in his Nobel lecture that the complexity of “central circumstances” could influence the outcome of spinal reflexes5.
NATURE REVIEWS | NEUROSCIENCE
The concept and discovery of CPG circuits has shaped and driven research on the neuronal substrates of invertebrate and vertebrate motor systems6–9. However, because CPGs are generally thought of as specialized circuits that are specific to the generation of oscillatory or alternating motor patterns, the concept of selfcontained circuits and the accompanying theories and experimental approaches have largely been restricted to the motor pattern generation field. In this article, we expand these ideas to the neocortex and review compelling evidence for similarities between cortical circuit properties and brainstem and spinal cord CPGs at the anatomical, biophysical, functional and even pathological levels. We restrict ourselves to comparisons between these two types of circuit, although similar arguments could be made with respect to other parts of the CNS. We consider two classic CPGs — spinal circuits that underlie locomotion in lower vertebrates and brainstem networks that generate breathing movements in the mammalian CNS — in which network organization is beginning to be understood in terms of the integration of cellular and circuit properties. We argue that the experimental information that is derived from the analysis of CPGs is likely to be prescient for neuroscientists who are exploring the basic principles of the organization and dynamic function of cortical circuits. Our argument further elaborates the proposal by Llinás that CNS evolution represents the progressive “encephalization” of motor rhythms3. Anatomical similarities
Operationally, for our purposes, CPGs can be defined as functional circuit modules that generate intrinsic patterns of rhythmic activity independently of sensory inputs9. They typically transform excitatory tonic driving input into detailed spatiotemporal patterns of oscillatory activity. The basic
VOLUME 6 | JUNE 2005 | 477
© 2005 Nature Publishing Group
PERSPECTIVES
a Spinal central pattern generator module
Excitatory Inhibitory
b Cortical microcircuit
Excitatory Inhibitory
Figure 1 | Similarities between vertebrate central pattern generators and neocortical circuits. Central pattern generator (CPG) modules in both the brainstem and the spinal cord (a; schematically illustrated here with simplified architecture) contain excitatory interneurons that are functionally embedded within inhibitory networks. This is also true for the cortex (b), although a piece of cortex would contain many intermingled CPGs. In these recurrent circuits, a combination of intrinsic and circuit properties mediates neuronal bistability, which can lead to rich intrinsic dynamics, including oscillations (as indicated) and other forms of attractor dynamics.
circuit modules are organized into neuronal populations that function to collectively coordinate the activity of numerous cells in time and space and produce behaviourally important output7. Although not required for the generation of circuit activity, phasic and other temporal signals that arise from sensory afferent inputs modulate the intrinsic activity of CPGs; therefore, CPG circuits also function as substrates for the integration of sensory input. An underlying assumption
478 | JUNE 2005
is that there is precise connectivity between specific circuit components — indeed, the idea of specific components being wired together in selective ways has provided an essential theoretical foundation for the examination of CPGs10. Specifically, in terms of the design of the circuit modules, a basic building block of vertebrate spinal locomotor 11,12 and brainstem respiratory13 CPGs is an ‘excitatory core’ or ‘kernel’ of mutually excitatory
| VOLUME 6
interneurons, with recurrent excitation being provided by AMPA (α-amino-3hydroxy-5-methyl-4-isoxazole propionic acid) receptors (AMPARs) and NMDA (Nmethyl-d-aspartate) receptors (NMDARs) in glutamatergic synapses. Each hemisegment of the spinal cord has such a core, and reciprocal (glycinergic) inhibition between contralateral hemisegments produces left–right alternation of motor output (FIG. 1a). In the brainstem respiratory pattern generator, an excitatory core that is distributed bilaterally in the pre-Bötzinger complex14,15 also uses fast recurrent excitatory transmission for burst generation and synchronization 16 , with inhibitory connections that terminate activity and generate alternating phases of inspiratory and expiratory motor output17. Similarly, in the cortex, precise connections and recurrent excitatory circuits seem to be present. Since the work of Cajal and Lorente de Nó, researchers have emphasized that there are many distinct classes of neocortical neuron 18,19 . Furthermore, although it is still unclear to what extent cortical circuits are wired in a specific way20,21, there is compelling evidence that at least some excitatory and inhibitory connections are extremely selective in choosing postsynaptic targets, or at least targets located in specific cortical layers22–25. Recurrent excitation is also prominent in the cortex26–28 (FIG. 1b), where it occurs both locally (for example, within a minicolumn) and distally through cortico-cortical connections and intracortical horizontal fibre systems20. In layer 4 of the primary visual cortex, it has been estimated that 94% of excitatory synapses originate in the cortex itself, the vast majority of which probably arise locally29. Analogous to the spinal and brainstem excitatory interneuron networks, neocortical excitation is glutamatergic and involves both AMPAR- and NMDAR-gated channels. Lateral inhibition is mediated through different types of local inhibitory GABA (γ-aminobutyric acid)-containing inter neuron, which provide feedforward, as well as feedback, inhibition. This is notably different from the lower vertebrate spinal cord, in which inhibition crucially involves glycinergic synaptic interactions, although the GABA system is active during spinal CPG activity and is an integral part of the locomotor network11. Both GABAand glycine-mediated mechanisms also control inhibitory synaptic interactions in the higher vertebrate brainstem respiratory CPG17.
www.nature.com/reviews/neuro
© 2005 Nature Publishing Group
PERSPECTIVES Although this description of cortical and brainstem/spinal cord circuits is general, allowing for analogy, the idea of a similar fundamental circuit organization in these different networks with recurrent excitatory, lateral and inhibitory connections still stands. As suggested by Lorente de Nó30, an excitatory network core together with inhibitory networks, in which the excitatory network is functionally embedded, represents a basic functional module for generating endogenous spatiotemporal patterns with rich dynamics, including several activity states. The dynamic tuning and control of this module by interactions with sensory inputs at the level of excitatory and/or inhibitory circuit components further expands the repertoire of patterned activity. Biophysical similarities
There are important biophysical similarities between brainstem/spinal and neocortical circuits in terms of both principal excitatory cells and inhibitory neurons. Among all neurons, there is substantial overlap in the types of ion channel and the membrane potential behaviours that are expressed. As described above, the primary excitatory neurotransmitter in all of these circuits is glutamate. In spinal CPGs, glutamate activates NMDARs, which are widely expressed in all of the excitatory networks. The unique function of the NMDAR as a coincidence detector, which requires both a sufficiently depolarized membrane potential and the binding of glutamate to become active, imparts nonlinear properties to the current–voltage relationship of the neuron. Although most investigators are aware of the role of NMDARs in Hebbian synaptic plasticity, the fact that this nonlinearity also produces bistability that can give rise to oscillatory behaviour is normally overlooked. Indeed, oscillations in CPGs are generated by a combination of network interactions and intrinsic cellular properties (NMDAR-mediated bistability, persistent sodium conductances31,32 and, possibly, other inward currents, such as calcium-activated nonspecific cation currents 33,34 ). Network interactions, which are fundamentally inhibitory, dynamically regulate the oscillations. Furthermore, the subthreshold-activated voltage-dependent cellular conductances that promote bistability and oscillations also promote synchronization by the nonlinear amplification of synaptic input, which is important for both the generation and transmission of oscillatory activity.
These same cellular properties/conductances are also present in neocortical neurons, and probably underlie the observed oscillatory behaviour and synchronization that is found in the cortex. In fact, neocortical circuits can be described as a complicated system of oscillation generators that function through both intrinsic and circuit mechanisms3. Oscillations in cortical circuits are prevalent in vivo, although the dynamic range of thalamo-cortical oscillations under normal conditions might be skewed towards higher frequencies than those observed in CPGs (the frequency range for motor oscillations is ~0.1–20 Hz and the range for cortical oscillations is ~1–80 Hz)35. Cortical oscillations are state dependent (for example, sleep/wake states and sensory input-dependent states)3, and have been linked to perceptual awareness, attention, consciousness and disease states36–38. However, except for the link between lowfrequency cortical oscillations and motor patterns3, and their obvious role in increasing synchronized activity39, the behavioural relevance of cortical oscillations is still not as well documented as it is in the brainstem and spinal cord. Interestingly, during the oscillatory states of network activity in motor systems, synchronization in the spike discharge of interacting neuronal populations on faster timescales occurs at frequencies of >20 Hz40, which are characteristic of cortical networks. Therefore, the higher frequency synchrony in CPGs and cortical networks might reflect similar dynamics of excitatory and/or inhibitory interactions.
“There are important biophysical similarities between brainstem/spinal and neocortical circuits in terms of both principal excitatory cells and inhibitory neurons.” The mechanisms responsible for the generation of oscillations could also reflect a similar basic design. The brainstem and spinal cord CPGs are driven by tonic brainstem input, which converts tonic drive to a phasic output. Surprisingly, this tonic excitatory drive can be converted into a rhythmic output in the absence of inhibition. This was implied by computational modelling studies and subsequently verified experimentally in
NATURE REVIEWS | NEUROSCIENCE
the lamprey: the isolated spinal cord can produce rhythmic activity in the physiological frequency range, independent of inhibition, which primarily results from repolarizing conductances41,42. The presence of an excitatory ‘kernel’ network with intrinsic oscillatory properties that can be tuned by tonic excitation was shown early on for the respiratory network14. Simulation of this excitatory network indicates that the dynamic interactions of neuronal burstgenerating currents and synaptic currents can produce synchronized oscillations that can be regulated over a wide dynamic range31 by tonic excitation. In both systems, inhibition acts with the excitatory network to dynamically regulate the rhythm and coordinate the activity of spatially distributed populations that are required for motor pattern formation. In the neocortex, research on the mechanisms of oscillations has focused on the role of GABA-containing interneurons, as oscillatory rhythms in cortical circuits seem to be modulated by inhibitory interneurons, whereby different types of interneuron sculpt the rhythms of different temporal frequencies43,44. Nevertheless, theoretical and computational studies45–47 indicate that it is also possible that slow cortical oscillations crucially depend on interactions between mutual excitation and cellular adaptation due to slow repolarizing currents, as in the CPGs, which are possibly augmented by synaptic depression. In the cortex, inhibition is still necessary to avoid epileptic mass activation and, in general, has an important role in the dynamic regulation of circuit activity. It should be noted that although rhythmic activity is possible in both systems following the blockade of inhibition, this activity is more analogous to pathological than functional states. A classic model for producing epileptic-type activity in slices of the neocortex can be produced by the application of GABA antagonists48, and, in the mammalian spinal cord, co-activation of flexors and extensors on both the ipsi- and contralateral spinal cord occurs following the blockade of glycinergic transmission49. Both results indicate that rhythm generation can operate independently of inhibition; however, spatiotemporal pattern formation in both types of circuit crucially depends on the presence of inhibition. Neurons in the brainstem and spinal cord show bistable membrane behaviours, which are commonly referred to as plateau potentials 11 . These have been proposed to amplify synaptic inputs, and might act to emphasize and de-emphasize inputs
VOLUME 6 | JUNE 2005 | 479
© 2005 Nature Publishing Group
PERSPECTIVES depending on their timing. Recently, a correlate of bistable membrane behaviour, in this case termed ‘up’ and ‘down’ states, has also been described in the striatum and neocortex both in vivo and in vitro50–54. It is still unclear whether this bistability arises from intrinsic55 or circuit56 mechanisms or a combination of the two. The up and down states, which act as membrane potential ‘set points’, probably influence the response of neocortical neurons to stimuli in vitro and in vivo and also control their internal dynamics54,57. Although the terms used to describe bistability differ, it is likely that this phenomenon in both the spinal cord and the neocortex is generated by the same conductances and serves the same basic computational function. NMDARs, which are widely expressed in both circuits, are ideally suited to mediate bistability in these cases. In addition, in the brainstem pre-Bötzinger complex, persistent sodium currents (NaP) occur ubiquitously58 and can mechanistically underlie neuronal burst generation and termination as well as the sustained tonic spiking of respiratory neurons32. Therefore, conductances such as NaP that endow cells with bistable properties give rise to numerous activity states, including silence, oscillatory bursting and tonic firing. NaP is also expressed in cortical neurons and contributes to the generation of persistent activity59. The relative contributions of these and other currents in generating bistability in cortical neurons remain to be determined. It should also be noted that in all of these systems, periods of sustained depolarization can be terminated by the same repolarizing currents, such as the Ca2+-dependent K+ channels, independent of inhibitory input. A common component of CPG circuits is the presence of pacemaker neurons — that is, neurons that can repetitively fire action potentials or generate oscillatory bursts of action potentials independently of the network, in the absence of any synaptic input60. Pacemaker properties, which reflect bistability that produces cellular oscillations or sustained tonic spiking, are prevalent in every CPG that has been studied so far. Pacemaker cells in all systems generate spontaneous rhythmic bursting activity through various combinations of conductances7. In the lower vertebrate and mammalian spinal cord, the role of pacemaker neurons is still poorly understood, although pacemaker properties are present even in motor neurons, and probably contribute to rhythm and pattern generation11. NaP seems to be essential for pacemaking in many systems61, including
480 | JUNE 2005
the respiratory pre-Bötzinger complex16,58. In the neocortex, spontaneous pacemaking activity can still be detected in the presence of synaptic blockers and relies on NaP59. Heterogeneous assemblies of excitatory neurons with currents such as NaP that promote pacemaker behaviour at the cellular level have rich dynamics31 with many oscillatory states, including pacemaker-driven oscillations, emergent network rhythms and quasi-periodic states. Dynamic similarities: attractors
An attractor network is a feedback neural network whose activity (either spontaneous or stimulus driven) has intrinsic dynamics that settle on stable points (attractors)62. A ‘Hopfield attractor’ would, therefore, represent a stable pattern of activity in the network. Networks might have many attractors that could be visualized as basins in an energy landscape that represents the network activity. These stable states are emergent properties of the circuit that are, therefore, largely invisible to the single neuron. Attractors could represent solutions to a specific computation63, such as orientation selectivity 64, eye position stability65, working memory 66 or decision making in Bayesian networks 67. The recurrent excitatory network that would correspond to the excitatory core of the spinal cord and brainstem CPGs is highly distributed in the neocortex. Therefore, instead of a pair of excitatory cores that are distributed bilaterally, as found in a spinal segment or in the respiratory pre-Bötzinger complex, there could be many superimposed structures, each of which would be analogous to a Hebbian cell assembly 68 or Hopfield attractors62, distributed over a large cortical area. In fact, when considered as a whole, the entire spinal circuit that is responsible for coordinated activity and motor output also represents a highly distributed interconnected system of excitatory–inhibitory circuit modules with short- and long-range connections. The neocortical excitatory core could be kept together by the intracortical and corticocortical fibre systems that originate in layers 2/3 and 5. A key feature is that at least part of this connectivity is presumably shaped during development by experience-dependent Hebbian synaptic plasticity, rather than being genetically encoded. The repolarizing neuronal currents discussed above, together with inhibitory interactions, could terminate an active attractor state. Control could be imposed by sensory afferent input or facilitation from other sources that bias activation towards more stimulated assemblies.
| VOLUME 6
This could lend a continuous dynamic to the attractor network and implement associative memory 62,69. In the neocortex, simulations of a network model with a minicolumnar–macrocolumnar arrangement have shown dynamics that are reminiscent of cortical up and down states47. In addition, in the neocortex, evidence for microstates has been observed in multichannel electroencephalography recordings that show stimulus-locked periods of stability that are separated by transient shifts of short duration; it is possible that these states also have a behavioural correlate70. Although not generally analysed in this theoretical framework, spinal cord and brainstem CPG activity has obvious behavioural correlates that can also be classified as having periods of stability and transition. An oscillatory cycle can be understood as a dynamic attractor. As indicated, the functional significance of oscillations, although prevalent in the neocortex, remains open to debate. Nevertheless, besides oscillations, other forms of intrinsic dynamics have been described in neocortical circuits both in vitro and in vivo71,72. Some of these spontaneous dynamics seem to follow precise temporal sequences, although the existence and potential function of these temporal patterns of activity remain unclear. Indeed, precise temporal sequences of action potentials, both within a single neuron and among neurons, are commonly found in CPGs and are widely accepted as necessary to implement fine control of motor patterns. In any case, both the intrinsic and evoked dynamics of activity in CPGs and the neocortex settle at stable points, and can be interpreted as evidence for attractor-governed neural networks. In the spinal cord, for example, these stable points punctuate trajectories that can be segmented into states, some of which correspond to different phases of the step cycle during walking. Modulatory similarities
Monoaminergic and sensory modulation is also a common characteristic that is shared between CPGs and cortical circuits. Examples in CPGs include the modulation of oscillatory frequency, of temporal coordination among different populations of active cells, of the amplitude of network activity and of the gating of CPG input/ output11. For instance, serotonin has been shown to modulate action potential afterhyperpolarization, plateau potentials and NMDA-mediated bistability in neurons in the mammalian spinal cord73. All of these cellular effects will partly determine the
www.nature.com/reviews/neuro
© 2005 Nature Publishing Group
PERSPECTIVES output of the CPG. Serotonergic inputs to CPGs increase the excitability of the networks, augment oscillation frequency and can induce plasticity in network operation74. Switching between stable states of network operation (for example, switching coordinated motor patterns for different modes of locomotion) is under sensory afferent and neurochemical modulatory control, which makes CPGs multifunctional and dynamically plastic11. Switching between cortical activity states is also under modulatory control, as shown, for example, by the role of dopamine on working memory in monkeys75. Therefore, in CPGs and probably in the neocortex, despite the presence of hard wiring and dedicated combinations of cellular properties that support stable ‘default’ patterns of network activity, modulation reconfigures network dynamics and transforms activity states to modify behaviour9,11. Pathological similarities
Finally, the likelihood of common organizing principles between CPGs and cortical circuits is further strengthened by the similar basic pathophysiology of hyperexcitability. This might be a consequence of the strong excitatory recurrent connectivity that is found in both types of circuit, as well as the prevalence of NMDARs and other burstgenerating conductances, such as persistent sodium, which, if unbalanced (by inhibitory mechanisms), can produce massive synchronized discharges in the excitatory cells. Whereas epilepsy is commonly found in the neocortex, tetanic syndromes in spinal cord circuits (although they use a different clinical nomenclature) essentially reflect a nucleus of excitatory neurons that have epileptiform-like discharges. Indeed, intracellular recordings from tetanic spinal cord neurons and epileptic neocortical cells are very similar 48,76–78. Differences: plasticity
It could be argued that there are also fundamental differences between spinal and brainstem CPGs and cortical circuits. Developmental hard-wiring between circuit components is an accepted idea in the spinal cord field, as predicted by the concept of the CPG. However, this has still not been clearly shown in the neocortex, although there is mounting evidence for such hard wiring between identifiable circuit components in the neocortex24,25. At the same time, it is widely accepted that the connectivity in the cortex can be shaped and determined by activity-dependent Hebbian plasticity79. In fact, it could be argued that the cortex
is a special type of CPG that is based on Hebbian assemblies and is specialized for learning and storing or retrieving memories (a ‘learning CPG’ or ‘memory CPG’). Rather than implement locomotor phases, cortical attractors would encode memories. Indeed, dendritic spines, which are thought to implement calcium-dependent learning rules that are restricted to individual inputs, are extremely prevalent in cortical circuits and mediate most excitatory connections80. However, they are relatively rare in spinal cord circuits and other CPGs (notice the paucity of spines in the spinal cord18, although see REFS 81,82).
“An important goal for combined experimental, theoretical and computational cortex research should be to discriminate between theories of cortical function to either falsify or delineate their applicability.” Furthermore, in the neocortex, the widespread occurrence of synaptic plasticity, together with the great diversity of neuronal types and properties, could generate circuits with a constant state of flux without any stable states83. Such ‘liquid-state’ dynamics could allow the implementation of real-time computational strategies that are particularly adept at solving nonlinear computations84. This view of the cortex seems to be inconsistent with the more rigid CPG framework. Nevertheless, we argue that Hebbian plasticity does not preclude the presence of predefined neuron classes and connections with predefined dynamics. It is possible that the skeleton of the circuit — that is, the stronger connections that probably transmit the essential information for fast processing to ensure the survival of the animal — has a predetermined structure and has relatively little plasticity. By contrast, finer connections, which are designed to fine-tune the dynamics and match them to the outside world, are more plastic and could even be constantly changing. By analogy, it is possible to imagine that the main interconnections resemble predetermined highways that connect cities, whereas the
NATURE REVIEWS | NEUROSCIENCE
smaller country roads could be the principal source of plasticity. This argument is consistent with the current view of the role of neuronal activity in developmental plasticity as one of refining a basic pattern of connections, determined through a genetic/ molecular process85. Alternatively, it is also possible that we currently underestimate the role of synaptic or structural plasticity in spinal and brainstem CPGs. In fact, in CPGs, Hebbian and other forms of synaptic plasticity have been clearly shown. So far, the best examples of plasticity can be found in motor reflex pathways, which might also involve the activation of CPGs; this is best typified by the Aplysia californica gill-withdrawal reflex86,87. However, in cases of continuously active synapses, such as those found in classic invertebrate CPGs, forms of plasticity similar to classical conditioning might not occur. Yet, even if this is the case, it does not preclude Hebbian plasticity in circuits that are not continuously active, such as the mammalian locomotor CPG. In short, it remains unclear whether the modification of circuit properties by learning is essential for motor CPG function74,88, as these circuits seem to be designed primarily to generate innate stereotypical patterns of neural activity, and functional plasticity can be achieved through neuromodulatory control 9,11 . Therefore, although CPGs are clearly functionally plastic, the role of learning in refining innate function remains to be established. Cross-fertilization and predictions
In summary, we argue that both brainstem and spinal cord CPGs and cortical circuits have recurrent excitatory kernels, are built with bistable neurons, are embedded within inhibitory circuits and are endowed with attractor dynamics. Although it is possible that cortical circuits are inherently more plastic than traditional CPGs, and some of the properties discussed here have not yet been shown in all regions of the neocortex, we suggest that these many similarities are deep and not accidental. Rather, they are likely to reflect basic organizing principles that are common to circuits that function to generate robust and spatially patterned rhythmic activity on different timescales. An important goal for combined experimental, theoretical and computational cortex research should be to discriminate between theories of cortical function to either falsify or delineate their applicability. Moreover, crossfertilization between these different fields of study could lead to considerable advances in both. For example, if neocortical circuits do
VOLUME 6 | JUNE 2005 | 481
© 2005 Nature Publishing Group
PERSPECTIVES resemble brainstem and spinal cord circuits in their design and basic properties, they would also be expected to resemble them developmentally and to result from similar genetic mechanisms. In the spinal cord, there is strong evidence for different classes of neuron, and it is becoming clear that the expression of a small set of homeobox transcription factors determines cell type identity and controls not only the dendritic and axonal morphology of the neuron, but also its functional properties and connectivity patterns10,89. Therefore, if this lesson is applied to the neocortex, it would be expected that neocortical neurons would also be grouped into distinct functional, anatomical, morphological and molecular classes, which would be determined by the expression of specific patterns of transcription factors. Already, neuroanatomists have highlighted the idea that there are different classes of neocortical neuron and that the morphology of the axons are the best predictors to classify them19. These theories are being confirmed by quantitative studies that include wellcorrelated differences in both morphological and physiological characteristics90. Returning to spinal cord differentiation, we argue that the axonal morphology and connection patterns actually reflect the particular choice of transcription factor that is expressed by the neuron. Therefore, anatomists such as Lorente de Nó and Cajal might have indirectly been scoring cellular differences in transcription factor expression when sorting neurons on the basis of their axonal morphology. It might even be predicted that similar transcription factors, membrane proteins or extracellular matrix components mediate the differentiation and synaptic targeting of both neocortical and spinal cord neurons. Another area of cross-fertilization could be the application of experimental approaches and analytical tools that are used in the CPG field to studies of neocortical function. For example, phase-resetting experiments91 — in which an endogenous rhythm is reset or altered by the introduction of a sensory input that occurs at a particular phase in the cycle — have been used in CPG studies92. However, they have not been systematically used in studies of neocortical circuits93 that have focused on analysing the receptive field of the neuron rather than on the effect of sensory stimulation on ongoing activity and the phase dependency of this effect. Furthermore, dynamic system analysis94, which is often used in CPG studies and for the analysis of other oscillatory systems, could be applied more extensively to studies
482 | JUNE 2005
of the neocortex to determine the aspects of neocortical function that involve bifurcations that lead to changes in circuit dynamics95,96. A similar argument could be made in reverse, whereby approaches or lessons that are learned from studying the neocortex could be applied to the study of CPG circuits. For instance, because of the obvious complexity of cortical circuits, researchers who study the cortex have explored many theories about computation, even in the absence of identified components and circuit structure. Examples include the strategies that have been developed to quantify and infer informational content97,98 or to estimate the certainty of a future event in a Bayesian manner99. Such approaches are not commonly used in the study of CPG circuits, and could help us to better understand the dynamic organization of the circuit and to predict possible cellular components. In addition, the application of simultaneous recordings from synaptically connected neurons in the neocortex is revealing rules of connectivity and systematic correlations between neuronal morphologies and molecular markers90. On the basis of the analogy with the spinal cord, we have also argued that one important aspect of cortical dynamics is represented well by the attractor network model. Attractor states can become quasi-stable, lasting only a few hundred milliseconds, owing to the action of intrinsic repolarizing currents and synaptic depression. This indicates several interesting avenues of research that connect the cellular and network level dynamics with important experimental cognitive phenomena. Indeed, the timescale of fast motor rhythms and cortical attractor dynamics are similar (hundreds of milliseconds70). We predict that the neural mechanisms that underlie a phenomenon such as the attentional blink100, for example, can be better understood in terms of the attractor dynamics of the cortical circuits that are involved. In summary, it seems that the comparison of both types of circuit is not only of academic interest, but might also lead to important changes in perspective. Our argument is one of analogy and lacks direct experimental verification at present, so we do not consider this issue to be settled. Instead, we would like to open the debate and cross-fertilize these two fields. It is possible that the similarities that we have highlighted also apply more generally to other regions of the nervous system. It seems that the time is ripe for the re-examination of the old empiricist/nativist debate in neocortical circuits in light of the similarities between vertebrate CPGs and cortical circuits discussed here.
| VOLUME 6
Rafael Yuste and Jason N. MacLean are at the Department of Biological Sciences, Columbia University, 1212 Amsterdam Avenue, Box 2435, New York 10027, USA. Jeffrey Smith is at the Cellular and Systems Neurobiology Section, Laboratory of Neural Control, Porter Neuroscience Research Center, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, Maryland 20892, USA. Anders Lansner is at the Department of Computer Science and Communication, Stockholm University and Royal Institute of Technology (KTH), SE-100 44 Stockholm, Sweden. Correspondence to R.Y. e-mail:
[email protected] doi:10.1038/nrn1686 1. 2.
3. 4.
5.
6.
7.
8.
9. 10.
11.
12.
13.
14.
15.
16. 17.
18. 19. 20.
21.
22.
Sherrington, C. S. The Integrative Action of the Nervous System (Yale Univ. Press, New York, 1948). Hartline, H. K. The response of single optic nerve fibres of the vertebrate eye to illumination of the retina. Am. J. Physiol. 121, 400–415 (1938). Llinás, R. I of the Vortex: From Neurons to Self (MIT Press, Cambridge, Massachusetts, 2002). Brown, G. On the nature of the fundamental activity of the nervous centres; together with an analysis of the conditioning of rhythmic activity in progression, and a theory of the evolution of function in the nervous system. J. Physiol. (Lond.) 49, 18–46 (1914). Sherrington, C. S. Inhibition as a Coordinative Factor (Nobel Lecture) [online] (12 December 1932). Grillner, S. & Zangger, P. How detailed is the central pattern generation for locomotion? Brain Res. 88, 367–371 (1975). Selverston, A. General principles of rhythmic motor pattern generation derived from invertebrate CPGs. Prog. Brain Res. 123, 247–257 (1999). Harris-Warrick, R. M. & Marder, E. Modulation of neural networks for behavior. Annu. Rev. Neurosci. 14, 39–57 (1991). Marder, E. & Calabrese, R. L. Principles of rhythmic motor pattern generation. Physiol. Rev. 76, 687–717 (1996). Kiehn, O. & Butt, S. J. Physiological, anatomical and genetic identification of CPG neurons in the developing mammalian spinal cord. Prog. Neurobiol. 70, 347–361 (2003). Grillner, S. The motor infrastructure: from ion channels to neuronal networks. Nature Rev. Neurosci. 4, 573–586 (2003). Lansner, A., Ekeberg, Ö. & Grillner, S. in Neurons, Networks, and Motor Behavior (eds Stein, P. S. G., Grillner, S., Selverston, A. I. & Stuart, D. G.) 165–171 (MIT Press, Cambridge, Massachusetts, 1997). Smith, J. C. in Neurons, Networks, and Motor Behavior (eds Stein, P. S. G., Grillner, S., Selverston, A. I. & Stuart, D. G.) 97–104 (MIT Press, Cambridge, Massachusetts, 1997). Smith, J. C., Ellenberger, H. H., Ballanyi, K., Richter, D. W. & Feldman, J. L. Pre-Bötzinger complex: a brainstem region that may generate respiratory rhythm in mammals. Science 254, 726–729 (1991). Reckling, J. C. & Feldman, J. L. PreBötzinger complex: hypothetical site and kernel for respiratory rhythm generation. Ann. Rev. Physiol. 60, 385–405 (1998). Koshiya, N. & Smith, J. C. Neuronal pacemaker for breathing visualized in vitro. Nature 400, 360–363 (1999). Richter, D. W. & Spyer, M. Studying rhythmogenesis of breathing: comparison of in vivo and in vitro models. Trends Neurosci. 24, 464–472 (2001). Ramón y Cajal, S. La Textura del Sistema Nerviosa del Hombre y los Vertebrados 1st edn (Moya, Madrid, 1899). Lorente de Nó, R. La corteza cerebral del ratón. Trab. Lab. Invest. Bio. (Madrid) 20, 41–78 (1922). Douglas, R. J., Martin, K. A. C. & Markram, H. in The Synaptic Organization of the Brain (ed. Shepherd, G. M.) 499–558 (Oxford Univ. Press, Oxford, UK, 2004). Kalisman, N., Silberberg, G. & Markram, H. The neocortical microcircuit as a tabula rasa. Proc. Natl Acad. Sci. USA 102, 880–885 (2005). Gilbert, C. & Wiesel, T. N. Morphology and intracortical projections of functionally characterised neurons in the cat visual cortex. Nature 280, 120–125 (1979).
www.nature.com/reviews/neuro
© 2005 Nature Publishing Group
PERSPECTIVES
23. Somogyi, P., Tamas, G., Lujan, R. & Buhl, E. Salient features of synaptic organisation in the cerebral cortex. Brain Res. Brain Res. Rev. 26, 113–135 (1998). 24. Kozloski, J., Hamzei-Sichani, F. & Yuste, R. Stereotyped position of local synaptic targets in neocortex. Science 293, 868–872 (2001). 25. Silberberg, G., Gupta, A. & Markram, H. Stereotypy in neocortical microcircuits. Trends Neurosci. 25, 227–230 (2002). 26. Lorente de Nó, R. Studies on the structure of the cerebral cortex. J. Psychol. Neurol. 45, 381–438 (1932). 27. Lorente de Nó, R. in Physiology of the Nervous System (ed. Fulton, J. F.) 228–330 (Oxford Univ. Press, New York, 1949). 28. Douglas, R. J. & Martin, K. A. C. Neuronal circuits in the neocortex. Annu. Rev. Neurosci. 27, 419–451 (2004). 29. Ahmed, B., Anderson, J. C., Douglas, R. J., Martin, K. A. C. & Nelson, J. C. Polyneuronal innervation of spiny stellate neurons in cat visual cortex. J. Comp. Neurol. 341, 39–49 (1994). 30. Lorente de Nó, R. Analysis of the activity of the chains of internuncial neurons. J. Neurophysiol. 1, 207–244 (1938). 31. Butera, R. J., Rinzel, J. & Smith, J. C. Models of respiratory rhythm generation in the pre-Bötzinger complex. II. Populations of coupled pacemaker neurons. J. Neurophysiol. 82, 398–415 (1999). 32. Butera, R. J., Rinzel, J. & Smith, J. C. Models of respiratory rhythm generation in the pre-Bötzinger complex. I. Bursting pacemaker neurons. J. Neurophysiol. 82, 382–397 (1999). 33. Shah, M. M. & Haylett, D. G. K+ currents generated by NMDA receptor activation in rat hippocampal pyramidal neurons. J. Neurophysiol. 87, 2983–2989 (2002). 34. Pena, F., Parkis, M. A. & Ramirez, J. M. Differential contribution of pacemaker properties to the generation of respiratory rhythms during normoxia and hypoxia. Neuron 43, 105–117 (2004). 35. Steriade, M., Jones, E. G. & Llinás, R. R. Thalamic Oscillations and Signaling 431 (John Wiley & Sons, Somerset, Chichester, 1990). 36. Gray, C. M., Konig, P., Engel, A. K. & Singer, W. Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties. Nature 338, 334–337 (1989). 37. Crick, F. & Koch, C. Some reflections on visual awareness. Cold Spring Harbor Symp. Quant. Biol. 55, 953–962 (1990). 38. Jeanmonod, D. et al. Neuropsychiatric thalamocortical dysrhythmia: surgical implications. Neurosurg. Clin. N. Am. 14, 251–265 (2003). 39. Engel, A. K., Fries, P. & Singer, W. Dynamic predictions: oscillations and synchrony in top-down processing. Nature Rev. Neurosci. 2, 704–716 (2001). 40. Parkis, M. A., Feldman, J. L., Robinson, D. M. & Funk, G. D. Oscillations in endogeneous input to neurons affect excitability and signal processing. J. Neurosci. 23, 8152–8158 (2003). 41. Hellgren-Kotaleski, J., Grillner, S. & Lansner, A. Neural mechanisms potentially contributing to the intersegmental phase lag in lamprey. I. Segmental oscillations dependent on reciprocal inhibition. Biol. Cybern. 81, 317–330 (1999). 42. Cangiano, L. & Grillner, S. Fast and slow locomotor burst generation in the hemispinal cord of the lamprey. J. Neurophysiol. 89, 2931–2942 (2003). 43. Buzsaki, G. & Chrobak, J. J. Temporal structure in spatially organized neuronal ensembles: a role for interneuron networks. Curr. Opin. Neurobiol. 5, 504–510 (1995). 44. Klausberger, T. et al. Brain-state- and cell-type-specific firing of hippocampal interneurons in vivo. Nature 421, 844–848 (2003). 45. Fuhrmann, G., Segev, I., Markram, H. & Tsodyks, M. Coding of temporal information by activity dependent synapses. J. Neurophysiol. 87, 140–148 (2002). 46. van Vreeswijk, C. & Hansel, D. Patterns of synchrony in neural networks with spike adaptation. Neural Comput. 13, 959–992 (2001). 47. Fransén, E. & Lansner, A. Low spiking rates in a population of mutually exciting pyramidal cells. Network Comput. Neural Sys. 6, 271–288 (1995). 48. Gutnick, M. J., Connors, B. W. & Prince, D. A. Mechanisms of neocortical epileptogenesis in vitro. J. Neurophysiol. 48, 1321–1335 (1982). 49. Cowley, A. & Schmidt, B. J. Effects of inhibitory amino acid antagonists on reciprocal inhibitory interactions during rhythmic motor activity in the in vitro neonatal rat spinal cord. J. Neurophysiol. 74, 1109–1117 (1995). 50. Wilson, C. J. & Groves, P. M. Spontaneous firing patterns of identified spiny neurons in the rat neostriatum. Brain Res. 220, 67–80 (1981). 51. Steriade, M., Nunez, A. & Amzica, F. A novel slow (~1Hz) oscillation of neocortical neurons in vivo: depolarizing and
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63. 64.
65.
66.
67.
68. 69.
70.
71.
72.
73.
74.
75. 76. 77.
78.
hyperpolarizing components. J. Neurosci. 13, 3252–3265 (1993). Cowan, R. L. & Wilson, C. J. Spontaneous firing patterns and axonal projections of single corticostriatal neurons in the rat medial agranular cortex. J. Neurophysiol. 71, 17–32 (1994). Sanchez-Vives, M. & McCormick, D. Cellular and network mechanisms of rhythmic recurrent activity in neocortex. Nature Neurosci. 3, 1027–1034 (2000). Cossart, R., Aronov, D. & Yuste, R. Attractor dynamics of network UP states in neocortex. Nature 423, 283–289 (2003). Egorov, A. V., Hamam, B. N., Fransen, E., Hasselmo, M. E. & Alonso, A. A. Graded persistent activity in entorhinal cortex neurons. Nature 420, 173–178 (2002). Shu, Y., Hasenstaub, A. & McCormick, D. A. Turning on and off recurrent balanced cortical activity. Nature 423, 288–293 (2003). Anderson, J., Lampl, I., Reichova, I., Carandini, M. & Ferster, D. Stimulus dependence of two-state fluctuations of membrane potential in cat visual cortex. Nature Neurosci. 3, 617–621 (2000). Del Negro, C. N., Koshiya, N., Butera, R. J. & Smith, J. C. Persistent sodium current, membrane properties, and bursting behavior of pre-Bötzinger complex inspiratory neurons in vitro. J. Neurophysiol. 88, 2242–2250 (2002). Mao, B. Q., Hamzei-Sichani, F., Aronov, D., Froemke, R. C. & Yuste, R. Dynamics of spontaneous activity in neocortical slices. Neuron 32, 883–898 (2001). Llinás, R. R. The intrinsic electrophysiological properties of mammalian neurons: insights into central nervous system function. Science 242, 1654–1664 (1988). Raman, I. & Bean, B. P. Ionic currents underlying spontaneous action potentials in isolated cerebellar Purkinje neurons. J. Neurosci. 19, 1663–1674 (1999). Hopfield, J. J. Neural networks and physical systems with emergent collective computational abilities. Proc. Natl Acad. Sci. USA 79, 2554–2558 (1982). Hopfield, J. J. & Tank, D. W. Computing with neural circuits: a model. Science 233, 625–633 (1986). Ben-Yishai, R., Lev Bar-Or, R. & Sompolinsky, H. Orientation tuning by recurrent neural networks in visual cortex. Proc. Natl Acad. Sci. USA. 92, 3844–3848 (1995). Seung, H. S., Lee, D. D., Reis, B. Y. & Tank, D. W. Stability of the memory of eye position in a recurrent network of conductance-based model neurons. Neuron 26, 259–271 (2000). Durstewitz, D., Seamans, J. K. & Sejnowski, T. J. Neurocomputational models of working memory. Nature Neurosci. 3 (Suppl.), 1184–1191 (2000). Sandberg, A., Lansner, A., Petersson, K. & Ekeberg, O. A Bayesian attractor network with incremental learning. Network 13, 179–194 (2002). Hebb, D. O. The Organization of Behaviour (Wiley, New York, 1949). Lansner, A., Fransén, E. & Sandberg, A. Cell assembly dynamics in detailed and abstract attractor models of cortical associative memory. Theor. Biosci. 122, 19–36 (2003). Lehmann, D., Strik, W. K., Henggeler, B., Koenig, T. & Koukkou, M. Brain electric microstates and momentary conscious mind states as building blocks of spontaneous thinking: 1. Visual imagery and abstract thoughts. Int. J. Psychophysiol. 29, 1–11 (1998). Abeles, M., Bergman, H., Margalit, E. & Vaadia, E. Spatiotemporal firing patterns in the frontal cortex of behaving monkeys. J. Neurophysiol. 70, 1629–1638 (1993). Ikegaya, Y. et al. Synfire chains and cortical songs: temporal modules of cortical activity. Science 304, 559–564 (2004). Schmidt, B. J. & Jordan, L. M. The role of serotonin in reflex modulation and locomotor rhythm production in the mammalian spinal cord. Brain Res. Bull. 53, 689–710 (2000). Feldman, J. L., Mitchell, G. S. & Nattie, E. E. Breathing: rhythmicity, plasticity, chemosensitivity. Annu. Rev. Neurosci. 26, 239–266 (2003). Goldman-Rakic, P. S. Cellular basis of working memory. Neuron 14, 477–485 (1995). McNamara, J. Emerging insights into the genesis of epilepsy. Nature 399, A15–A22 (1999). Ruscheweyh, R. & Sandkuhler, J. Epileptiform activity in rat spinal dorsal horn in vitro has common features with neuropathic pain. Pain 105, 327–338 (2003). Van Den Pol, A. N., Obrietan, K. & Belousov, A. Glutamate hyperexcitability and seizure-like activity throughout the brain and spinal cord upon relief from chronic glutamate receptor blockade in culture. Neuroscience 74, 653–674 (1996).
NATURE REVIEWS | NEUROSCIENCE
79. Wiesel, T. N. Postnatal development of the visual cortex and the influence of the environment. Nature 299, 583–592 (1982). 80. Yuste, R. & Majewska, A. On the function of dendritic spines. Neuroscientist 7, 387–395 (2001). 81. Cameron, W. E., Averill, D. B. & Berger, A. J. Morphology of cat phrenic motoneurons as revealed by intracellular injection of horseradish peroxidase. J. Comp. Neurol. 219, 70–80 (1983). 82. Semba, K. et al. Ultrastructure of pacinian corpuscle primary afferent terminals in the cat spinal cord. Brain Res. 302, 135–150 (1984). 83. Maass, W., Natschlager, T. & Markram, H. Real-time computing without stable states: a new framework for neural computation based on perturbations. Neural Comput. 14, 2531–2560 (2003). 84. Jaeger, H. & Haas, H. Harnessing nonlinearity: predicting chaotic systems and saving energy in wireless communication. Science 304, 78–80 (2004). 85. Katz, L. C. & Shatz, C. J. Synaptic activity and the construction of cortical circuits. Science 274, 1133–1138 (1996). 86. Kandel, E. R., Schwartz, J. H. & Jessell, T. M. Principles of Neural Science (Elsevier, New York, 1991). 87. Carew, T. J., Hawkins, R. D., Abrams, T. W. & Kandel, E. R. A test of Hebb’s postulate at identified synapses which mediate classical conditioning in Aplysia. J. Neurosci. 4, 1217–1224 (1984). 88. Ji, R., Kohno, T., Moore, K. & Woolf, C. J. Central sensitization and LTP: do pain and memory share similar mechanisms? Trends Neurosci. 26, 696–705 (2003). 89. Tsuchida, T. et al. Topographic organization of embryonic motor neurons defined by expression of LIM homeobox genes. Cell 79, 957–970 (1994). 90. Toledo-Rodriguez, M. et al. Correlation maps allow neuronal electrical properties to be predicted from singlecell gene expression profiles in rat neocortex. Cereb. Cortex 14, 1310–1327 (2004). 91. Winfree, A. T. The Geometry of Biological Time (Springer, New York, 2001). 92. Glass, L. & Mackey, M. C. From Clocks to Chaos. The Rhythms of Life (Princeton Univ. Press, New Jersey, 1988). 93. Netoff, T. I. et al. Synchronization in hybrid neuronal networks of the hippocampal formation. J. Neurophysiol. 93, 1197–1208 (2005). 94. Fiedler, B. (ed.) Handbook of Dynamical Systems (Elsevier, New York, 2002). 95. Pinsky, P. F. & Rinzel, J. Intrinsic and network rhythmogenesis in a reduced Traub model for CA3 neurons. J. Comput. Neurosci. 1, 39–60 (1994). 96. Pinto, D. J., Jones, S. R., Kaper, T. J. & Kopell, N. Analysis of state-dependent transitions in frequency and longdistance coordination in a model oscillatory cortical circuit. J. Comput. Neurosci. 15, 283–298 (2003). 97. Rieke, F., Warland, D., de Ruyter van Steveninck, R. & Bialek, W. Spikes, Exploring the Neural Code (MIT Press, Cambridge, Massachusetts, 1997). 98. Gold, J. I. & Shadlen, M. N. Banburismus and the brain: decoding the relationship between sensory stimuli, decisions, and reward. Neuron 36, 299–308 (2003). 99. Barlow, H. Redundancy reduction revisited. Network 12, 241–253 (2001). 100. Raymond, J. E., Shapiro, K. L. & Arnell, K. M. Temporary suppression of visual processing in an RSVP task: an attentional blink? J. Exp. Psychol. Hum. Percept. Perform. 18, 849–860 (1992).
Acknowledgements This review resulted from the 2004 Dahlem Workshop entitled ‘Microcircuits: the interface between neurons and global brain function’. We thank the organizers and co-participants in this workshop for their input, and the anonymous reviewers for their constructive criticisms. Work in our laboratories is supported by the Kavli Institute for Brain Science (R.Y.), the National Institutes of Health (R.Y., J.N.M. and J.S.) and the Swedish Science Council (A.L.).
Competing interests statement The authors declare no competing financial interests.
Online links FURTHER INFORMATION Lansner’s homepage: http://www.nada.kth.se/~ala Smith’s homepage: http://intra.ninds.nih.gov/lab.asp?org_ id=34 Yuste’s homepage: http://www.columbia.edu/cu/biology/ faculty/yuste/index.html Access to this interactive links box is free online.
VOLUME 6 | JUNE 2005 | 483
© 2005 Nature Publishing Group
