Spontaneous neuronal activity, in the form of "retinal waves", is present in the mammalian retina prior to light-driven activity, and is known to be required for synaptic refinement in the lateral geniculate nucleus (LGN) and the cortex. Because both the spatiotemporal patterning of retinal activity and the resulting synaptic refinement of retinal afferents in the LGN have been well characterized through numerous physiological and anatomical studies, activity-dependent retinogeniculate development presents an opportunity to connect how realistic system level activity drives local synaptic "learning rules" to give rise to the observed system-level refinement. However, making such a connection is difficult because synaptic plasticity is generally measured outside of the context of natural activity patterns, and it is not known how it applies in realistic circumstances. By studying the complex spatiotemporal patterning of retinal waves, we have designed experiments that focus on the informative aspects of these waves that can drive development. The resulting experiments reveal a novel learning rule at the retinogeniculate synapse that describes how realistic retinal activity results in long-term changes in the strength of retinogeniculate synapses (both potentiation and depression). Such a learning rule can explain how normal retinal wave activity drives multiple aspects of developmental refinement, as well as numerous experiments showing the effect of manipulations of retinal wave activity on retinogeniculate refinement. These studies highlight the importance of understanding synaptic plasticity in the system-level contexts: both of the "natural" activity patterns, and synaptic refinement that results.
Neuronal connections are often organized in a spatially precise manner to maintain the nearest neighbor relationships from an origin structure to its target, an arrangement called topographic mapping. A prominent model for studying how topographic maps are established during mammalian development is the projection of retinal ganglion cells to the superior colliculus. We have recently studied the roles of the molecular guidance cues ephrins and structured activity in the development of retinocollicular maps. By studying the overall structure of the retinotopic maps with intrinsic signal imaging and comparing the experimental data with a computational model, we have revealed the contributions of ephrin-As and activity-dependent mechanisms in retinocollicular map formation. These studies also provide a unique opportunity for studying the role of the retinotopic map, an orderly physical layout, in the physiological processing of sensory information.
The shapes of dendritic arbors are fascinating and important, yet the principles underlying their complexity and diversity remain unclear. Here, we analyzed basal dendritic arbors of 2171 pyramidal neurons sampled from mammalian brains and discovered three statistical properties: the dendritic arbor size scales with the total dendritic length, the spatial correlation of dendritic branches within an arbor has a universal functional form, and small parts of an arbor are self-similar. We proposed that these properties result from maximizing the repertoire of possible connectivity patterns between dendrites and surrounding axons while keeping the cost of dendrites low. We solved this optimization problem by drawing an analogy with maximization of the entropy for a given energy in statistical physics. The solution is consistent with the above observations and predicts scaling relations that can be tested experimentally. In addition, our theory explains why dendritic branches of pyramidal cells are distributed more sparsely than those of Purkinje cells. Our results represent the first step towards a unifying view of the relationship between neuronal morphology and function.
The onset of vision occurs when neural circuits in the visual cortex are immature, lacking the full complement of connections and the response selectivity that defines functional maturity. Direction selective responses are particularly vulnerable to the effects of early visual deprivation, but how stimulus driven neural activity guides the emergence of cortical direction selectivity remains unclear. To explore this issue we developed a novel motion training paradigm that allowed us to monitor the impact of experience on the development of direction selective responses in visually naive ferrets. Using intrinsic signal imaging techniques we found that training with a single axis of motion induced the rapid emergence of direction columns that were confined to cortical regions preferentially activated by the training stimulus. Using 2-photon calcium imaging techniques, we found that single neurons in visually na´ve animals exhibited weak directional biases and lacked the strong local coherence in the spatial organization of direction preference that was evident in mature animals. Training with a moving stimulus, but not with a flashed stimulus, strengthened the direction selective responses of individual neurons and preferentially reversed the direction biases of neurons that deviated from their neighbors. Both effects contributed to an increase in local coherence. We conclude that early experience with moving visual stimuli drives the rapid emergence of direction selective responses in visual cortex.
Patterned, spontaneous neural activity is found in many parts of the developing nervous system, and is generally thought to play an important role in its development and maturation. While there is increasing evidence that merely the presence of a certain level of neural activity alone is insufficient to support normal development, it is still unclear which specific properties of spontaneous activity are relevant. To answer this question, it is necessary to understand the defining features of spontaneous activity, and its interactions with developmental mechanisms. To address this issue, we used computational modelling and multielectrode array recordings to investigate in detail the properties of retinal waves, the propagating activity patterns observed in the developing retina of many vertebrate species before the onset of vision. Consistent with previous models and recent experimental findings, the analysis of the model shows that early-stage retinal waves are regulated by a refractory mechanism, which controls neural excitability in an activity-dependent manner. Extending these earlier findings, the model shows that a biophysically plausible model of the refractory mechanism, operating on multiple time scales, can gradually desynchronise the network activity while maintaining spatially correlated activity. This model predicts that the experimentally observed randomness of initiation sites, trajectories and sizes of retinal waves are best reproduced at points in configuration space where the synchronising effect of synaptic transmission is balanced by the desynchronising effect of the refractory mechanism. Then, wave sizes and lifetimes assume power-law distributions, and a mean field model shows that this network state corresponds to the critical point of a percolation phase transition, where the network undergoes a transition between purely local and global functional connectedness. An analysis of multielectrode array recordings confirms this prediction. These results indicate that early-stage retinal waves are regulated according to a very specific principle, which maximises randomness and variability in the resulting activity patterns. Moreover, the resulting activity contains events on all length scales, and is therefore unbiased with respect to scale or sequence of events, which may be an important prerequisite for the normal visual system development. Finally, the scale-free character of retinal waves might present the visual system with an early opportunity to adapt to input statistics later also encountered during natural vision. Joint work with Christopher Adams, David Willshaw and Evelyne Sernagor.
Retinal ganglion cells (RGCs) are the bottleneck of visual information leaving the eye; there are at least two dozen RGC subtypes, each responds to a unique aspect of the visual scene (motion, direction, color, etc.) and sends that information to the brain where it is processed into perceptions and behavior. To delineate the complete map of brain connections made by individual RGC subtypes, we carried out a genetic screen to identify mice with i) RGC mosaics selectively labeled with green fluorescent protein (GFP) and ii) no GFP+ cells in retinorecipient areas. We identified several transgenic mice that meet these criteria. The central projections or 'maps' of each RGC mosaic exhibit remarkable specificity; different RGC subtypes project their axons to different combinations of target nuclei and those axons occupy distinct depths, or layers across the full extent of their targets. Moreover, within each mosaic-specific layer, RGC axons form regularly tiled arrays. Developmental analysis revealed that both laminar-specificity and axon arbor tiling emerge through a process involving removal of inappropriate projections. Genetic manipulations indicate that whereas laminar specific mapping is hard wired, axonal tiling is critically dependent on patterned RGC activity (Huberman et al., Neuron, 2008). These data, combined with analysis of other RGC subtype specific GFP lines, are beginning to reveal a code for mapping axons from distinct RGC mosaics into specific targets, target layers and tiled arrays in the brain.
In higher mammals, preferred stimulus orientation is mapped smoothly across the visual cortex, except at "pinwheel" centers, where all orientation preferences coalesce. This functional organization of the orientation map emerges in ferrets around postnatal day 30 and, once formed, remains stable over the following weeks, as demonstrated by intrinsic signal imaging. Sparse electrical recordings from individual neurons, however, have revealed orientation selective single units about 10 days before the earliest orientation maps have been reported with intrinsic signal imaging, but whether these neurons are organized into an orientation map remains unknown. We use two-photon calcium imaging to study the development of orientation preference with cellular resolution in the primary visual cortex of ferrets. In the youngest ferrets exhibiting visual responses, almost all neurons responded strongly and nearly exclusively to horizontal stimuli. This unexpected regime of "all-horizontal" tuning lasted for about a week, P21-27. Subsequently, around the time of eye-opening, cells lost their all-horizontal tuning and responded largely unselectively to all orientations. Despite such broad tuning during this period, cells were already organized into a smooth map of orientation preference with occasional pinwheels. Later still, orientation selectivity improved further, but map structure remained largely similar. Thus, during the initial development of visual response properties, neurons in the visual cortex undergo dramatic and exquisitely orchestrated changes in orientation tuning as one regime of functional organization gives way to another. In particular, the transition from all-horizontal tuning to the familiar pinwheel arrangement implies considerable, previously unreported developmental changes in the neuronal circuits underlying the generation of orientation preference.
Neuronal circuits in the brain are shaped by experience during early postnatal development. In the visual system for example, closing one eye for a few days during a critical period of heightened plasticity causes loss of the cortical representation for the deprived eye and increases that for the non-deprived open eye. Since the pioneering work by Hubel and Wiesel, the changes induced in the visual cortex by such monocular deprivation (MD), termed ocular dominance plasticity (ODP), have been widely studied as a model for competitive, experience-dependent cortical plasticity.
Conventionally, studies on ODP have measured the relative responses to the two eyes in acute, invasive preparations, in which determining changes in absolute responsiveness is difficult. We have developed a procedure for minimally invasive, rapid imaging of intrinsic signal of the mouse cortex, enabling repeated longitudinal recording in individual animals. Such chronic imaging during the critical period showed a clear temporal separation of changes in visual cortical responses: first, a reduction in deprived-eye responses over 2-3 days with little change in open-eye responses; second, with continued deprivation, a dramatic increase in open-eye responses and a smaller increase in deprived eye responses; third, after ending deprivation by re-opening the closed eye in young mice, responses to each eye return to their original magnitude over a several days. This chronic recording method has allowed tracking changes in absolute response strength over time and more powerful comparisons for detecting altered plasticity.
The competitive nature of ODP has suggested that inputs from two eyes compete for a postsynaptic reward that mediates plasticity by differently affecting synapses serving the two eyes. Brain-derived neurotrophic factor (BDNF) has been proposed as such a retrograde signal, based on numerous observations. We tested this hypothesis using a recently developed chemical-genetic approach to inhibit TrkB kinase activity rapidly and specifically during the induction of cortical plasticity in vivo. Contrary to the model, TrkB inactivation during MD had no detectable effect on changes in cortical responses to the deprived eye or the open eye. The changes in cortical responses induced by MD can be reversed quickly by re-opening the eye to allow binocular vision for several days. This recovery was completely blocked by TrkB inhibition. These findings suggest a more conventional trophic role for TrkB signaling in the enhancement of responses or growth of new connections, rather than a role in competition. Indeed, we observed that deprived-eye responses were restored more rapidly when both eyes were open than when the occlusion was reversed, further suggesting binocular cooperation, rather than competition, for the process of recovery. The delay in the expression of second-phase changes has suggested a compensatory or homeostatic mechanism that would maintain the firing rates of cortical cells during continuing deprivation. To test this idea, we examined visual cortical plasticity in mice lacking tumour necrosis factor-alpha (TNF-α). TNF-α has been shown to mediate homeostatic scaling of excitatory postsynaptic potential amplitude following prolonged activity blockade in hippocampal neurons in vitro, while having no effects on Hebbian plasticity such as long-term potentiation and depression. We found that in TNF-α knockout mice the initial phase of plasticity, the loss of deprived-eye responses, was similar to that of wild type, whereas the second phase of plasticity in which open-eye responses increase was absent. Local pharmacological inhibition of endogenous TNF in the visual cortex of wild type mice phenocopied the knockout mice. These observations suggest that what had been thought of as experience-dependent competition in developing visual cortex is the outcome of two distinct processes, the second of which appears to depend on homeostatic synaptic scaling.
These results demonstrate mechanisms involved in 3 stages in the course of MD and recovery: an initial loss of deprived-eye responses, independent of TrkB and TNF-α signaling; a subsequent homeostatic increase of open-eye responses dependent on TNF-α and a recovery of deprived-eye responses that depends on TrkB kinase after restoration of visual inputs.
The young brain is structurally different from the adult brain and contains additional circuits that are formed by subplate neurons (SPNs). These neurons are among the earliest born cortical neurons; they reside in the white matter and disappear during development.
After the critical period ends - when SPNs are no longer present - only limited plasticity is present. Thus SPNs participate in types of synaptic plasticity that occur only during the critical period. To elucidate the developmental role of SPNs, we studied SPNs and their associated circuits in vitro in thalamocortical slices. We find that SPNs in mouse auditory cortex receive functional excitatory inputs from the medial geniculate nucleus (MGN) as early as P2 and that MGN inputs to SPN were capable of inducing action potentials in SPNs. We also find that SPNs provide functional input to neurons in developing layer 4. Thus, SPNs are tightly integrated into the developing thalamocortical circuit providing input to the eventual targets of thalamic projections. These results suggest that SPNs are a reliable relay of early spontaneous and sensory evoked activity and can thereby regulate cortical development and plasticity. Thus, premature loss of SPNs can lead to developmental abnormalities.
Using a combination of ablation experiments and computational models, we find that SPNs are required for the functional maturation of the cortical columnar organization, the development of intracortical inhibitory circuits, and the outcome of plasticity during the critical period.
Together these results show that SPNs act like a "teacher" helping thalamic neurons to make strong and precise connections to their cortical target neurons. By relaying thalamic input and controlling the balance of excitation and inhibition, SPNs can influence the correlations between thalamic and cortical activity and thereby synaptic plasticity. Together, this work provides a framework demonstrating that plasticity during the critical period is the product of a complex and dynamically changing circuit in which SPNs play a key role.
In developing brain, axons are capable to find and recognize their targets based on specific chemical cues, which are emitted into intercellular space or localized on cellular membranes. Further adjustment of connectivity is possible due to the presence of neural activity that is thought to modify synapses in an experience-dependent way. Formulating a model for development of connectivity matrix therefore poses a basic problem of combining various factors, such as Sperry chemoaffinity principle and Hebb rule, in the same approach. We will argue that the problem of network formation can be thought of as an optimization principle of an affinity potential that depends of the network graph and includes all these diverse factors. In a way combining Hebb and Sperry in the same approach is similar to formulating the Standard Model in particle physics that combines disparate forms of fundamental interactions.
The highly ordered wiring of retinal ganglion cell (RGC) neurons in the eye to their synaptic targets in the superior colliculus (SC) of the midbrain (the tectum in non-mammalian vertebrates) has long served as the dominant experimental system for the analysis of topographic neural maps. These maps are comprised of axonal connections in which the positional coordinates of a set of input neurons are mapped onto the corresponding coordinates of their targets. They are a feature of nearly all sensory modalities, including sight, touch, sound, taste and smell, and are seen throughout the nervous system. We describe a quantitative model for the development of one arm of the retinocollicular (retinotectal) map: the wiring of the nasal-temporal axis of the retina to the caudal-rostral axis of the SC. The model is based RGC-RGC competition that is governed by comparisons of the signaling intensity experienced by RGCs expressing differing levels of EphA receptor protein-tyrosine kinases, whose expression is exponentially graded across the nasal-temporal axis of the retina. These comparisons are made using ratios of, rather than absolute differences in, EphA signaling intensity. Molecular genetic experiments, exploiting a combinatorial series of EphA receptor knock-in and knock-out mice, confirm the salient predictions of this 'Relative Signaling' model, and demonstrate that it both describes and predicts topographic mapping.
The cells of the developing neocortex of the mouse are produced during a 6 day period that begins on embryonic day 10 and continues through E17. During this time, the cells of the ventricular zone that contains the major proliferative population, pass through 11 cell cycles as they sequentially produce the layers of the neocortex. With each pass through the cell cycle, the cell cycle itself becomes progressively longer, increasing from 8 to over 18 hours (almost exclusively from an increase in the length of G1), and a progressively larger proportion of the daughter cells exit the cell cycle, increasing from Q ("quitting fraction") = 0 to Q = 1. The initial process of neuron production is initiated in the ventrolateral portion of the neocortex at E10 and is propagated dorsomedially over the next ~40 hours. This sets up a gradient in cell cycle (and G1) lengths across the surface of the ventricular zone. This gradient by itself is insufficient to uniquely specify an identity for cells produced at any given time that is necessary to code of the arealization of the neocortex. There also exist transcription factor gradients across the surface of the ventricular zone. In the case of the gradient of Lhx2 quantitative measures shows that its gradient is also insufficient to uniquely specify a code for arealization of the neocortex. The two gradient systems taken together, however, can uniquely specify both a WHEN and WHERE for the ventricular zone so that cell class can be varied across the surface of the neocortex as the basis for a code for the arealization of the neocortex. The perspective proposed is by analogy to navigation on the high seas where both a clock (to provide the accurate time) and a sextant are necessary to determine position of a ship from the position of the stars. The cell cycle length is the analogue of the clock and the transcription factor concentration is the analogue of the sextant reading. The fact that cell cycle lengths vary across the surface of the ventricular zone in other developing brain regions, such as the retina and hippocampus, indicates that positional information from cell cycle timing is widely available in the developing nervous system and is a possible basis for areal development of cell class differences and connectional maps.
Supported by a grants from the NINDS, the NEI, and the New Jersey Commission on Spinal Cord Research.
Enriched sensory experience during development significantly affects maturation of brain connectivity and function. Wild type mice raised in an enriched environment from birth show accelerated visual system development. However, the extent to which connectivity can be modified by an enriched environment could not be assessed because visual system connectivity in these mice is essentially accurate from the outset.
Ephrin-A-/- mice have disordered visual system circuitry and an associated deficit in visuomotor behaviour. These mice have already been used to demonstrate the role of spontaneous retinal activity in refining topography during development. Here we show that an enriched visual environment in early postnatal life significantly improves the accuracy of retinocollicular topography in ephrin-A-/- mice. The ability to measure the specific behavioural consequences of these cellular changes provides a direct assessment of structure-function relationships in the brain.
Spontaneous activity of neurons serves as an important instructive signal in early development. Toward the peak of the critical period (CP) for induction of ocular dominance plasticity by monocular deprivation (MD), the visual cortex progressively becomes more sensitive to the imbalance of visual input from the two eyes. In the first part of the talk, we introduce a theory that explains two important developmental stages of visual plasticity in a unified manner: pre-CP and CP plasticity.
We test the hypothesis that the balance of spontaneously driven activity and visually driven activity, which is controlled by cortical inhibition, explains both pre-CP and CP plasticity. Experimental results (lab of T.K. Hensch), both intracellular and extracellular, showed modulations of spontaneous-to-visual ratio with the strength of inhibition. Our computer simulations with Hebbian synaptic plasticity with homeostatic scaling of synapses reproduced plasticity results in both periods under various visual-deprivation conditions. In particular, during the pre-CP, innate spontaneous input from the two eyes largely prevented the ocular dominance shift induced by MD, yet the change in visual correlations induced by MD slowed the rate of retinotopic refinement.
The reduction of spontaneous activity level caused by maturation of inhibition increased the sensitivity of the cortex to the visual stimulus and opened the CP. This scenario suggests a novel account of the results from G. Turrigiano's lab, showing that monocular deprivation in monocular cortex caused homeostatic plasticity in layers 2/3 of V1 during the CP but not before. This could result simply from inhibition-induced changes in cortical spontaneous activity, without any changes in the plasticity rules themselves.
In the second part of the talk, we model biological elements with different time constants that are involved in ocular dominance plasticity. We construct and simulate a simple model that includes three separable plasticity processes: (a) Rapid weakening of the deprived eye under MD -- not affected by blockade of TNF-alpha or TrkB. (b) Strengthening of the open eye after about 3 days of MD -- specifically prevented by blockade of TNF-alpha, which blocks a global form of homeostatic synaptic scaling, but not of TrkB. (c) Recovery from MD under binocular vision, specifically blocked by blockade of TrkB.
We modeled the total synaptic strength as the product of (1) A synapse-specific strength (e.g. the percent of potential AMPA receptor sites occupied), modulated by NMDA dependent LTD as well as by NMDA and BDNF dependent LTP. (2) A postsynaptic-cell-specific scaffolding factor (e.g. the number of potential AMPA receptor sites per spine), modulated by TNF-alpha mediated homeostatic plasticity.
The model captures the transient behaviors of ocular dominance plasticity, which many traditional models do not. First, the learning rule must depend on past as well as present synaptic strengths in order to reproduce the MD result in the monocular cortex. Second, the fast Hebbian component requires a built-in stabilization mechanism, e.g. maximal and minimal weight limits. Slow homeostatic plasticity cannot stabilize an unstable Hebbian component. Third, in order to robustly avoid an overshoot of synaptic strength under MD, homeostasis and LTP/LTD should control independent factors. In the present model, homeostasis and LTP/LTD control two independent factors that multiply to determine synaptic strength, allowing the homeostatic response to build up slowly without being overwritten by fast LTD.
Functional maps arise in developing visual cortex as response selectivities become organized into columnar patterns of population activity. This process entails increases in the magnitude of neuronal selectivity and the spatial coherence of neuronal response preference. Recent studies of developing orientation and direction maps indicate that both maps are sensitive to visual experience, but not to the same degree or duration. Direction maps have a greater dependence on early vision while orientation maps remain sensitive to experience over a longer period of cortical maturation. There is also a darker side to experience: abnormal vision through closed lids produces severe impairments in neuronal selectivity rendering these maps nearly undetectable. Thus, the rules that govern their formation and the construction of the underlying neural circuits are modulated - for better or worse - by early vision. Direction maps, and possibly maps of other properties that are dependent upon precise conjunctions of spatial and temporal signals, are most susceptible to the potential benefits and maladaptive consequences of early sensory experience.
Current views are that activity-based and molecular-based mechanisms act together to form the ordered retinotopic maps found in the vertebrate retinotectal or retinocollicular systems. One crucial question concerns the relative importance of each mechanism. I will describe the model given in Willshaw (Development, 2006) which was designed with recent experimental evidence in mind which involves genetic manipulation of the putative molecular mechanism, disturbing the retinocollicular projection in mouse. According to this model, retinal axons form such maps by effectively self-sorting according to the labels that they carry. This model leads to specific predictions about the state of the molecular labels on the colliculus.
In order to be able to evaluate the part played in neural map-making by any particular mechanism, it is essential to have a method for measuring the amount of order in a map. I will then describe my ongoing work to develop such a method. In this method, the internal order of the map is assessed separately from the global orientation of the map. I am using this method to compare the amount of local order within wildtype maps compared to those found in maps found in Beta2 knockouts.
Spatial alignment between sensory representations can be shaped and maintained through plasticity. We found that a Hebbian model can account for the synaptic plasticity resulting from displacement of the space representation of one input channel relative to that of another, even when the synapses from both channels are equally plastic. Surprisingly, although the synaptic weights for the two channels obeyed the same Hebbian learning rule, the amount of plasticity exhibited by the respective channels was highly asymmetric and depended on the relative strength and width of the receptive fields (RFs): the channel with the weaker or broader RFs always exhibited most or all of the plasticity. This asymmetry mimicked the vast asymmetry in plasticity across input channels often observed in vivo. In agreement with experimental observations, plasticity was enhanced by the accumulation of incremental adaptive adjustments to a sequence of small displacements. These same principles would apply not only to the maintenance of spatial registry across input channels, but also to the experience-dependent emergence of aligned representations in developing circuits.