Title: Spontaneous Correlated Activity in Developing Neural Circuits
Abstract: Development of precise neural circuitry is traditionally thought to involve two separate phases. In the early activity-independent phase, neurons develop distinct phenotypes, migrate to characteristic positions within a developing structure, and establish an initial set of connections. In the later, activity-dependent phase, neural activity is thought to drive refinement of the initially crude connectivity (8Katz L.C. Shatz C.J. Science. 1996; 274: 1133-1138Crossref PubMed Scopus (2269) Google Scholar). Activity-dependent mechanisms have been demonstrated in many brain areas during early postnatal life when immature circuits are first driven by sensory experience. During this period, experimental manipulations that perturb normal firing of sensory afferents result in marked changes in neural circuits. Recent findings indicate, however, that neural activity may influence some of the developmental processes at stages earlier than previously thought. There is growing evidence that electrical and chemical activity can profoundly influence a variety of intracellular processes, including establishment of cell phenotype, activation of enzymes, and gene expression (reviewed by 1Berridge M.J. Neuron. 1998; 21: 13-26Abstract Full Text Full Text PDF PubMed Scopus (1673) Google Scholar). In the developing visual system, even before there is any sensory experience, spontaneously generated action potentials in the retina are required for the formation of retinogeniculate connections (14Penn A.A. Riquelme P.A. Feller M.B. Shatz C.J. Science. 1998; 279: 2108-2112Crossref PubMed Scopus (385) Google Scholar), evidence that immature circuits generate activity patterns that drive developmental processes (19Yuste R. Semin. Cell Dev. Biol. 1997; 8: 1-4Crossref PubMed Google Scholar). This minireview highlights several interesting functional properties common to three very different developing neural circuits: retina, spinal cord, and hippocampus. All three generate spontaneous action potentials during development. Although the structure of spontaneous activity differs in detail, they share a global or "macroscopic" activity pattern: the spontaneous activity consists of rhythmic bursts of action potentials that are correlated across tens to hundreds of cells and occur with a periodicity on the order of minutes. I propose that these similar properties arise from two general organizing principles: (1) highly interconnected excitatory synapses may be responsible for generating the rhythmic activity, and (2) homeostatic mechanisms may act to regulate the overall level of network activity. Last, I suggest a role for macroscopic aspects of these patterns in activity-dependent development. In the developing retina, periodic spontaneous bursts of action potentials are generated in retinal ganglion cells (see Figure 1A) and propagate from one cell to the next in a wavelike manner (reviewed by 17Wong R.O.L. Annu. Rev. Neurosci. 1999; 22: 29-47Crossref PubMed Scopus (390) Google Scholar). Retinal waves are an extremely robust phenomenon, observed in a large variety of vertebrate species, including chick, turtle, mouse, rabbit, rat, ferret, and cat. Although the precise circuitry underlying the propagation differs substantially across species (reviewed by 17Wong R.O.L. Annu. Rev. Neurosci. 1999; 22: 29-47Crossref PubMed Scopus (390) Google Scholar), more striking are the similarities in the characteristic activity and developmental programs. Wave periodicity and velocity in all species are comparable. In early postnatal ferret (<P10), cholinergic synaptic inputs mediate the depolarization of ganglion cells during waves, and blocking cholinergic transmission prevents all wave activity (4Feller M.B. Wellis D.P. Stellwagen D. Werblin F.S. Shatz C.J. Science. 1996; 272: 1182-1187Crossref PubMed Scopus (410) Google Scholar, 14Penn A.A. Riquelme P.A. Feller M.B. Shatz C.J. Science. 1998; 279: 2108-2112Crossref PubMed Scopus (385) Google Scholar). The sole source of acetylcholine (ACh) in the retina is a subclass of interneurons called starburst amacrine cells, which, like ganglion cells, are excited via synaptic input during waves (20Zhou Z.J. J. Neurosci. 1998; 18: 4155-4165PubMed Google Scholar). Though the source of synaptic input to starburst amacrine cells during waves is unknown, these results indicate that a complex network of amacrine and ganglion cells connected via excitatory chemical synapses mediates retinal waves. The synaptic circuitry that drives retinal waves changes during postnatal development. In ferrets older than P18, waves are insensitive to cholinergic antagonists and instead are blocked by glutamate receptor antagonists, which probably act at early synaptic connections between bipolar cells and amacrine or ganglion cells (17Wong R.O.L. Annu. Rev. Neurosci. 1999; 22: 29-47Crossref PubMed Scopus (390) Google Scholar). Despite this developmental change in synaptic circuitry, spatiotemporal properties of waves remain remarkably similar. However, there is some change in the frequency of events in different classes of ganglion cells, dictated in part by a changing modulatory role of GABA (17Wong R.O.L. Annu. Rev. Neurosci. 1999; 22: 29-47Crossref PubMed Scopus (390) Google Scholar). Retinal waves in developing chick retina are quite similar to the waves seen in ferrets. First, at all the ages studied in chick (E8–E112Catsicas M. Bonness V. Becker D. Mobbs P. Curr. Biol. 1998; 8: 283-286Abstract Full Text Full Text PDF PubMed Google Scholar; E13–E1818Wong W.T. Sanes J.R. Wong R.O.L. J. Neurosci. 1998; 18: 8839-8852PubMed Google Scholar), retinal waves exhibit spatiotemporal properties similar to those in ferrets—they are periodic, occurring roughly every 2 min, and they propagate over regions of the retina containing hundreds to thousands of cells. Second, chick retinal waves are mediated in part by excitatory synaptic transmission, though gap junctions may also be involved. (Whether gap junctions play a role in wave propagation in the mammalian retina is unknown.) Third, the synaptic mechanisms driving chick retinal waves change with age. At the earliest ages studied (E8–E11), cholinergic antagonists substantially reduce the frequency of rhythmic activity (2Catsicas M. Bonness V. Becker D. Mobbs P. Curr. Biol. 1998; 8: 283-286Abstract Full Text Full Text PDF PubMed Google Scholar); however, at older ages (past E11), waves are blocked completely by a combination of glutamate antagonists but are unaffected by ACh antagonists (18Wong W.T. Sanes J.R. Wong R.O.L. J. Neurosci. 1998; 18: 8839-8852PubMed Google Scholar). Together, these results demonstrate that a variety of strategies used by the circuits that drive retinal waves at different ages and in different species can lead to the generation of very similar macroscopic patterns of activity. Spontaneous activity has also been observed in motor neurons of the developing chick spinal cord using an isolated spinal cord–hindlimb preparation (13O'Donovan M. Chub N. Semin. Cell Dev. Biol. 1997; 8: 21-28Crossref PubMed Scopus (47) Google Scholar, 12Milner L.D. Landmesser L.T. J. Neurosci. 1999; in pressGoogle Scholar). This rhythmic activity occurs during the stages of development when motor neuron axons are pathfinding in the plexus region and innervating the skeletal muscle (E4–E11). The activity is characterized by periodic events or "episodes," which in turn are comprised of multiple bursts of action potentials (see Figure 1B). Calcium imaging experiments at ages >E10 reveal that individual events initiate in the ventrolateral part of the cord, propagate dorsomedially, and involve tens to hundreds of neurons. Many properties of the spinal cord circuit that drives this rhythmic activity are similar to the retinal circuit that mediates waves. First, the rhythmic activity recorded in motor neurons is driven by excitatory synaptic input from local circuit interneurons. Second, the primary excitatory transmitter driving the activity switches from acetylcholine at E5 (12Milner L.D. Landmesser L.T. J. Neurosci. 1999; in pressGoogle Scholar) to glutamate at E10 (3Chub N. O'Donovan M.J. J. Neurosci. 1998; 18: 294-306PubMed Google Scholar). Third, at the earliest ages studied, gap junction antagonists reversibly block the rhythmic activity, implying that gap junctional coupling between interneurons and motor neurons might play a role in regulating network excitability as seen in the developing chick retina. However, there is no direct evidence for gap junctional coupling past E10. Fourth, developmental changes in synaptic circuitry alter some aspects of the spinal cord rhythm in a cell class–specific way. At E4–E5, rhythmic activity occurs every 2 min with the flexor and extensor fibers firing synchronously. By E10, the activity is characterized by episodes that occur every 6–8 min with the flexor and extensor muscles firing out of phase. Despite these changes in the temporal pattern, the basic global pattern of activity—periodic action potentials correlated over a large number of cells—remains the same. Remarkably, even at a single development age, spinal cord circuits can produce appropriate rhythmic activity via more than one mechanism. Blockade of cholinergic transmission in the early stages of spinal cord development (E5) suppresses rhythmic activity, but after some delay it is restored (12Milner L.D. Landmesser L.T. J. Neurosci. 1999; in pressGoogle Scholar). The restored activity is mediated by GABAergic transmission. Similar effects are seen in E11 spinal cord; after blocking glutamatergic transmission, spontaneous rhythmic activity is reestablished through GABA and glycine transmission (3Chub N. O'Donovan M.J. J. Neurosci. 1998; 18: 294-306PubMed Google Scholar). These observations have given rise to the hypothesis (3Chub N. O'Donovan M.J. J. Neurosci. 1998; 18: 294-306PubMed Google Scholar) that the firing properties of the network may be regulated by homeostatic mechanisms—the network compensates for the reduction in excitatory activity by increasing the efficacy of functional transmitter systems. This hypothesis is explored further below. Spontaneous bursts of action potentials are produced by neurons in the CA1 region of the developing hippocampus, occurring between the ages of P1 and P4 (5Garaschuk O. Hanse E. Konnerth A. J. Physiol. (Lond.). 1998; 507: 219-236Crossref Scopus (265) Google Scholar). These periodic events, or "early network oscillations" (ENOs), are defined by a synchronous increase in intracellular calcium in pyramidal cells and interneurons throughout the region, in contrast to the propagating activity seen in the developing retina and spinal cord. Similar to the rhythmic activity in the spinal cord, individual ENOs, which occur approximately every 60 s, consist of multiple bursts of action potentials with an intraburst interval of 2–4 s (see Figure 1C). Similar periodic depolarizing events have been observed in the CA3 region of the hippocampus during development (9Leinekugel X. Medina I. Khalilov I. Ben-Ari Y. Khazipov R. Neuron. 1997; 18: 243-255Abstract Full Text Full Text PDF PubMed Scopus (330) Google Scholar). The circuitry that drives ENOs is completely contained within the CA1 region since spontaneous calcium increases are observed even when the CA1 and CA3 regions are surgically isolated from each other and from the rhythm-generating hilar interneurons that drive correlated epileptiform bursting in the adult. Whole-cell voltage clamp recordings conducted simultaneously with fluorescence imaging experiments show that calcium transients are correlated with a barrage of synaptic inputs, indicating that ENOs are driven by synaptic transmission, similar to retinal waves and spinal cord rhythms. Although a combination of the transmitters GABA and glutamate provides the excitatory synaptic input that depolarizes CA1 neurons during ENOs, GABAA antagonists completely block all spontaneous transients recorded using calcium imaging. This suggests that spontaneous activation of a circuit containing interneurons is required for generation of ENOs. These oscillations disappear at the stage of development when GABA becomes inhibitory rather than excitatory, indicating that GABA is the primary source of excitation driving ENOs, unlike in the retina and spinal cord where GABA plays only a modulatory role. As discussed above, the developing retina, hippocampus, and spinal cord exhibit spontaneous, rhythmic, action potential–based activity that is correlated over a large number of cells via excitatory synaptic connections. Importantly, well-defined pacemaker circuits do not appear to drive the rhythmic activity. This stands in contrast to many other periodically active networks, such as circuits that coordinate motor activity, where rhythmic behavior is generated by a small set of cells with reciprocal excitatory and inhibitory connections that function as a pacemaker unit (reviewed by 7Harris-Warrick R.M. Marder E. Selverston A.I. Moulins M. Dynamic Biological Networks. MIT Press, Cambridge, MA1992Google Scholar) and circuits such as the thalamus and inferior olive, where the frequency of events is controlled by a slow depolarizing current called Ih in individual cells (10Luthi A. McCormick D.A. Neuron. 1998; 21: 9-12Abstract Full Text Full Text PDF PubMed Scopus (291) Google Scholar). Without pacemaker circuits, how can a network generate rhythmic activity? One likely possibility is that the oscillatory properties in these developing circuits arise because each network contains a critical number of excitatory connections (13O'Donovan M. Chub N. Semin. Cell Dev. Biol. 1997; 8: 21-28Crossref PubMed Scopus (47) Google Scholar). This is demonstrated in cultures of synaptically connected neocortical pyramidal cells, where spontaneous events are mediated by excitatory synaptic transmission and spontaneous episodic behavior is evident (15Robinson H.P.C. Kawahara M. Jimbo Y. Torimitsu K. Kuroda Y. Kawana A. J. Neurophysiol. 1993; 70: 1606-1616PubMed Google Scholar). This highly simplified network provides a powerful demonstration that, in the absence of a pacemaker, excitatory connections can generate periodic activity (Figure 1D). In networks with highly interconnected or "recurrent" excitatory architecture, what dictates the periodicity? 13O'Donovan M. Chub N. Semin. Cell Dev. Biol. 1997; 8: 21-28Crossref PubMed Scopus (47) Google Scholar propose a model in which a network of spontaneously active cells connected recurrently through purely excitatory synapses will have a steady increase of activity until a threshold is reached. An episode of correlated firing then will occur, leaving the network depressed so the activity of the network is dramatically reduced. The cycle then begins again as the network slowly recovers from this "network depression." In this scenario, the periodicity of the activity is set by the time constant of the network's recovery from depression. A possible source of this network depression could be depletion of the readily releasable pool of transmitter, though no experiments have tested this hypothesis directly. A second intriguing aspect of these developing circuits is that mechanistic details of the excitation may not be critical for determining the spatiotemporal patterns of spontaneous activity. For instance, retinal waves exist at different ages and in different species, though the details of the circuits that generate the activity vary widely. In addition, as described above, spontaneous rhythmic activity in the spinal cord is reestablished in a period of several hours following blockade of the transmitter receptors that ordinarily provide the excitatory drive that mediates the activity. Though mechanisms specific to each circuit are probably responsible for the details of the activity pattern, such as frequency of firing within a burst and burst duration, these cellular properties do not influence the slow rhythms that characterize the output pattern of the network. In this view, no exact, highly precise circuitry needs to exist to produce the activity pattern generated by the network as long as the circuit has sufficient recurrent excitatory connections to reach the requisite level of excitability. What are the cellular mechanisms that regulate this feedback control of network excitability? Recently16Rutherford L. Nelson S. Turrigiano G. Neuron. 1998; 21: 521-530Abstract Full Text Full Text PDF PubMed Scopus (372) Google Scholar showed that the strength of synaptic connections between cultured neocortical cells is regulated by the level of spontaneous activity. Chronic blockade of all synaptic transmission in these cultures led to an increase in excitatory synaptic strength, while elevation of intrinsic firing rates by removal of inhibition led to a decrease in excitatory synaptic strength. These long-lasting changes in synaptic strength indicate that a cell can modulate the amount of excitation it receives on the time scale of hours or days through some type of feedback, perhaps involving neurotrophins (16Rutherford L. Nelson S. Turrigiano G. Neuron. 1998; 21: 521-530Abstract Full Text Full Text PDF PubMed Scopus (372) Google Scholar). In the circuits described in this minireview, it is possible that a similar homeostatic mechanism could explain the consistent generation of correlated activity patterns in circuits whose components change with development and vary between species (retina) or experimental manipulations (spinal cord). For example, increasing synaptic strength in a recurrently connected network exhibiting rhythmic behavior would increase the frequency of spontaneous events by increasing the network's rate of recovery from depression (as proposed by 13O'Donovan M. Chub N. Semin. Cell Dev. Biol. 1997; 8: 21-28Crossref PubMed Scopus (47) Google Scholar). Most current models of activity-dependent development assume that synaptic refinement is driven by neural activity that is correlated tightly in space (i.e., among neighboring cells) and in time (on the order of milliseconds) (8Katz L.C. Shatz C.J. Science. 1996; 274: 1133-1138Crossref PubMed Scopus (2269) Google Scholar). However, the spontaneous activity patterns described here have a periodicity on the order of minutes and spatial correlations among tens to hundreds of cells. The similarity in activity among these circuits raises the possibility that these broad temporal and spatial correlations may be important for guiding some aspects of circuit development. What sort of information might be encoded in the slow periodicity of spontaneous activity? In each of the three circuits, periodic bursts of action potentials lead to substantial increases in intracellular calcium concentration in the participating neurons. There is growing evidence that periodic changes in intracellular calcium occurring on the order of minutes can influence profoundly a variety of intracellular processes, including gene expression and the establishment of cell phenotype (1Berridge M.J. Neuron. 1998; 21: 13-26Abstract Full Text Full Text PDF PubMed Scopus (1673) Google Scholar). Indeed, in the developing spinal cord, the frequency of spontaneous calcium transients in growth cones regulates the rate of axon extension (6Gomez T.M. Spitzer N.C. Nature. 1999; 397: 350-355Crossref PubMed Scopus (397) Google Scholar). Hence, the periodicity of circuit activation may be tuned to the periodicity of intracellular signaling required to ensure normal maturation of neurons in either the rhythmic circuit itself or in its target tissue. This is not to argue that development of the retina, spinal cord, and hippocampus requires the same temporal information. In fact, the details of the spontaneous activity that are specific to each circuit may be relevant to the maturation of that circuit. For instance, in the case of the developing retina, at ages greater than P21, ON ganglion cells participate in only one-third of the waves that propagate through the OFF ganglion cells (17Wong R.O.L. Annu. Rev. Neurosci. 1999; 22: 29-47Crossref PubMed Scopus (390) Google Scholar). This difference in the periodicity might be critical in the establishment of the unique anatomical and functional specificity of these two classes of cells within the retina or the segregation of their inputs in the LGN. The spatial pattern of the activity can similarly encode important information. Synchronous activation of cells (e.g., hippocampus) contains no distinct spatial information regarding the relative positions of cells involved in each event, whereas propagating activity (e.g., retina and spinal cord) synchronizes the activity of subsets of cells, thereby encoding their relative positions. For instance, a single retinal wave will synchronize firing of cells along a wavefront with a particular orientation on the retina, generating an activity pattern that might be used to establish orientation selectivity in visual cortical neurons (reviewed by 11Miller K.D. Prog. Brain Res. 1994; 102: 303-318Crossref PubMed Scopus (33) Google Scholar). Activity patterns averaged over a large number of waves would lose orientation information but would maintain highly correlated firing among neighboring neurons, information that might be used in the establishment of topographic projections. Whether mechanisms that process activity correlated across large numbers of neurons exist in the developing brain and whether these correlations are critical for the normal formation of adult circuits remains to be determined. The search is on for pharmacological and genetic manipulations that change the frequency at which rhythmic events occur, significantly alter the distance over which cells are correlated, or transform the activity into uncorrelated firing of neurons. If found, these manipulations could be used in vivo to determine directly which aspects of patterned spontaneous activity are relevant to neural development.