Synaptic Mechanisms for Plasticity in Neocortex

Whisker Map Plasticity in S1

Tactile experience drives robust plasticity of the somatotopic map in S1 (Blake et al. 2002), with cellular mechanisms best analyzed in the whisker map in rodent S1. Whiskers are active tactile detectors, represented anatomically by cell clusters, called barrels, in cortical layer (L) 4. An orderly map of whisker receptive fields exists in S1, with neurons in each barrel-related column responding best to deflection of the corresponding whisker. Similar to other primary sensory areas, whisker input from thalamus primarily arrives in L4, which in turn projects to L2/3, and then to L5, which provide corticocortical and subcortical output, respectively (Lubke & Feldmeyer 2007).

Trimming or plucking a subset of whiskers causes S1 neurons to rapidly lose spiking responses to deprived whiskers and to more slowly increase responses to spared whiskers, thus weakening and shrinking the representation of deprived whiskers and strengthening and expanding the representation of spared whiskers within the map (Fox 1992, Diamond et al. 1993, Glazewski & Fox 1996). Plasticity is most robust in young animals but persists substantially into adulthood. Plasticity occurs at multiple sites and layers in S1, with L4 being a primary site of plasticity in neonates (<4–6 days of age), whereas in juveniles and adults, plasticity occurs most rapidly and extensively, and sometimes exclusively, in L2/3 (Diamond et al. 1994, Glazewski & Fox 1996, Stern et al. 2001, Drew & Feldman 2009). Depression of responses to deprived whiskers (termed response depression) and potentiation of responses to spared whiskers (response potentiation) have distinct dynamics, are separable developmentally and genetically, and can be differentially induced by different patterns of whisker deprivation. They therefore represent distinct functional components and mechanisms of plasticity (Glazewski et al. 2000, Fox 2002).

In young adults (<2 months), plasticity involves both response depression and response potentiation, whereas in older animals only response potentiation is present (Fox & Wong 2005). The net effect of this map plasticity is to dynamically reallocate cortical processing space from deprived inputs toward spared inputs, which may optimize sensory processing. For detailed review of S1 map plasticity, see Feldman & Brecht (2005) and Fox (2002).

Ocular Dominance Plasticity in V1

Ocular dominance plasticity occurs in many species (Hubel & Wiesel 1998), but cellular mechanisms have been best analyzed in rodents. Rodent V1 lacks discrete ocular dominance columns but has a small binocular region in which individual neurons exhibit visual responses to both eyes. Monocular deprivation during a critical period (19–32 days of age) causes a rapid loss of responses to the deprived eye, followed by a slower gain of responses to the open eye, leading to a physiological shift in ocular dominance. As in S1, ocular dominance plasticity occurs both in L4 and in L2/3, with separate sites and mechanisms for plasticity in these layers (Daw et al. 1992, Maffei et al. 2004, Liu et al. 2008, Maffei & Turrigiano 2008). Like S1, plasticity can occur in L2/3 before in L4, suggesting that L2/3 is a primary early locus for plasticity (Trachtenberg et al. 2000). Response potentiation and response depression are genetically separable, with similar pharmacological and genetic requirements to S1 (Sawtell et al. 2003, Kaneko et al. 2008b, Liu et al. 2008) and similar layer-specific dynamics and critical periods (Daw et al. 1992, Trachtenberg et al. 2000, Mrsic-Flogel et al. 2007). Ocular dominance plasticity persists in adult rodents but is slower and mediated mostly by response potentiation, like map plasticity in S1 (Sawtell et al. 2003, Hofer et al. 2006, Sato & Stryker 2008). In contrast, adult cats and primates show substantially less adult plasticity.

Physiological changes in ocular dominance in L4 and L2/3 are followed in L4 after several days (in cats) to weeks (in mice) by shrinking and expanding geniculocortical axon arbors representing the closed and open eye, respectively (Antonini & Stryker 1993, Antonini et al. 1999). Whether whisker experience similarly remodels single thalamocortical axons in S1 is not known. For detailed review of ocular dominace plasticity, see Fox & Wong (2005) and Hofer et al. (2006).

Use- and Correlation-Dependent Plasticity Without Deprivation

In another major form of plasticity, repeated activation of a specific sensory input (without deprivation) potentiates neural responses to that input. This is usually most robust in young animals. For example, exposing young rats to auditory stimuli enhances the representation of the presented frequencies and intensities in A1, thus altering auditory tuning curves and the tonotopic map (Keuroghlian & Knudsen 2007). Presentation of high-contrast oriented gratings to young mice similarly drives orientation-specific enhancement of visual responses in V1 (Frenkel et al. 2006). Similar potentiation occurs in adult S1 in response to temporal correlations between inputs, leading to neurons and map regions with strong joint representation of temporally correlated inputs (e.g., Diamond et al. 1993, Wang et al. 1995). In adult V1, temporally correlated, near-simultaneous stimuli drive systematic shifts in visual tuning related to stimulus order and timing (Dan & Poo 2006).

Learning-Related Plasticity in Adults

Whereas map plasticity in juveniles occurs rapidly in response to passive sensory experience, such plasticity is slower and more limited in adults, except when stimuli are actively attended and behaviorally relevant (e.g., during a perceptual learning task) or explicitly paired with positive or negative reinforcement or neuromodulation. Thus, classical conditioning using tone stimuli increases A1 responses to trained frequencies (Weinberger 2007), classical conditioning using whisker stimuli expands the representation of trained whiskers (Siucinska & Kossut 1996, 2004), and perceptual training on visual stimulus location or orientation discrimination tasks alters V1 tuning for the trained feature (Gilbert 1998, Dan & Poo 2006). Training can increase neural responses to reinforced stimuli, shift tuning curves toward (or away from) trained stimuli, or sharpen tuning curves to improve discrimination between stimuli. These changes in neural tuning are generally modest and do not cause large-scale changes in map topography, except with very extensive training (Blake et al. 2002, Karmarkar & Dan 2006).

Homeostatic Plasticity

In a distinct form of plasticity, substantial sensory overuse or deprivation drives compensatory, homeostatic changes that restore, at least partially, cortical activity to a set point level. For example, 24 hours of continuous whisker deflection weakens the S1 representation of the stimulated whisker, whereas visual deprivation increases visual responses in the deprived monocular zone of rodent V1 (Knott et al. 2002, Mrsic-Flogel et al. 2007). Turrigiano & Nelson (2004) propose that such homeostatic plasticity stabilizes mean cortical activity in the face of slowly changing input levels and in response to synaptogenesis and synapse elimination during development (Turrigiano & Nelson 2004). Homeostatic plasticity also occurs during more modest sensory manipulations, such as monocular closure in binocular visual cortex, and thus is likely to contribute to multiple types of cortical plasticity (Mrsic-Flogel et al. 2007, Maffei & Turrigiano 2008).

Five Common Components of Sensory Cortical Plasticity

The analysis above suggests five basic components of experience-dependent plasticity that can be identified across S1, V1, and A1. These are illustrated in Figure 1. The first component is rapid response depression to deprived inputs, and the second is slower response potentiation to spared inputs, when a subset of inputs are deprived. These components of plasticity are classically hypothesized to involve Hebbian weakening and strengthening of deprived and spared pathways, and to be driven by competition between active and inactive inputs, because less or no plasticity occurs when all inputs are deprived (Wiesel & Hubel 1965). Response potentiation must involve a competitive process because it is driven heterosynaptically by depriving neighboring inputs. (On the cellular level, this could be accomplished by classical heterosynaptic plasticity or by homeostatic plasticity or metaplasticity affecting all synapses on a neuron.) Whether response depression is a competitive process is less clear. In some cases, response depression requires neighboring, active inputs (Glazewski et al. 1998), which may heterosynaptically depress deprived inputs. However, response depression can also occur when all inputs are deprived (Wallace & Fox 1999, Kaneko et al. 2008b), which is more consistent with noncompetitive, homosynaptic plasticity driven by residual activity on deprived pathways (Rittenhouse et al. 1999, Frenkel & Bear 2004). The third component is potentiation of responses to active inputs during normal sensory use, and in response to temporal correlation between inputs. The fourth is potentiation of responses paired with reinforcement in adults. The third and fourth components are both consistent with Hebbian strengthening of active inputs but differ in dependence on attention or reward. These are driven homosynaptically or cooperatively by activity on active pathways and therefore appear functionally distinct from potentiation of spared inputs during deprivation-induced plasticity. The fifth component is homeostatic regulation of cortical activity in response to substantial increase or decrease in sensory input. The next sections review many of the known cellular plasticity mechanisms in cortex and suggest how they mediate these components of plasticity.

Long-Term Depression

LTD implements use-dependent, homosynaptic and heterosynaptic weakening and therefore may mediate response depression to deprived inputs. Multiple forms of LTD exist and may have different roles in plasticity (for review, see Malenka & Bear 2004, Massey & Bashir 2007). In NMDA receptor–dependent LTD (NMDA-LTD), calcium from postsynaptic NMDA receptors activates protein phosphatases including calcineurin, leading to dephosphorylation of specific sites on the AMPA receptor GluR1 subunit and internalization of synaptic AMPA receptors. In cortex, NMDA-LTD (defined by NMDA receptor involvement and AMPA receptor internalization) has been clearly observed at thalamocortical synapses in V1, and likely S1 (Feldman et al. 1998, Crozier et al. 2007), and at other synapses in sensory, anterior cingulate, entorhinal, and perirhinal cortex (Dodt et al. 1999, Toyoda et al. 2006, Deng & Lei 2007, Griffiths et al. 2008). A second major form is metabotropic glutamate receptor–dependent LTD (mGluR-LTD), of which several subforms exist (Egger et al. 1999, Renger et al. 2002, Barbara et al. 2003, Czarnecki et al. 2007). A third form of LTD involves cannabinoid type 1 (CB1) receptors (Chevaleyre et al. 2006). In CB1-LTD, postsynaptic calcium elevation and activation of group I mGluRs drive postsynaptic endocannabinoid synthesis, which signals retrogradely to presynaptic CB1 receptors, driving a long-lasting decrease in release probability (Chevaleyre et al. 2006). CB1-LTD occurs at many neocortical excitatory synapses (Sjostrom et al. 2003, Bender et al. 2006b, Nevian & Sakmann 2006, Crozier et al. 2007, Lafourcade et al. 2007). CB1-LTD is independent of post-synaptic NMDA receptors but may require presynaptic NMDA receptors, which exist at specific neocortical synapses (Sjostrom et al. 2003, Rodriguez-Moreno & Paulsen 2008).

Long-Term Potentiation

LTP has been proposed to underlie use-dependent and temporal correlation-dependent strengthening of sensory responses in juveniles, reinforcement-dependent strengthening of responses in adults, and strengthening of spared inputs during deprivation-induced plasticity. Many neocortical excitatory synapses exhibit LTP (Kirkwood & Bear 1994, Castro-Alamancos et al. 1995, Crair & Malenka 1995, Isaac et al. 1997, Markram et al. 1997, Buonomano 1999, Feldman 2000, Sjostrom et al. 2001, Froemke & Dan 2002, Hardingham & Fox 2006). Where characterized, cortical LTP is most often classical NMDA-LTP. In NMDA-LTP, calcium from postsynaptic NMDA receptors and other sources activates kinases including αCaMKII, which drive specific AMPA receptor phosphorylation, and insertion of GluR1-containing AMPA receptors into synapses (Malinow & Malenka 2002). Long-lasting LTP further involves CaMKII autophosphorylation, activation of CREB, and altered gene expression. NMDA-LTP in neocortex can be blocked by viral expression of a GluR1 C-terminal tail construct (GluR1-ct) that prevents activity-dependent GluR1 insertion (Takahashi et al. 2003, Hardingham & Fox 2006, Toyoda et al. 2007). NMDA-LTP causes appearance of AMPA receptor currents at immature synapses that express NMDA but not AMPA receptors (“silent synapses”), thus functionalizing these synapses (Isaac et al. 1997, Rumpel et al. 1998). A second form of neocortical LTP is expressed presynaptically by an increase in release probability, which alters short-term synaptic dynamics (Markram & Tsodyks 1996, Buonomano 1999, Eder et al. 2002). LTP at adult L4-L2/3 synapses in S1 appears to exhibit both a presynaptic component mediated by retrograde nitric oxide (NO) signaling and a postsynaptic component involving GluR1 insertion (Hardingham et al. 2003, Hardingham & Fox 2006). Additional forms of LTP may also occur (Daw et al. 2004).

Spike Timing–Dependent Plasticity and Learning Rules for Plasticity

In the past decade, a dizzying variety of LTP/LTD learning rules have been discovered that vary with cell type, synapse location on dendrites, background network activity, and neuromodulation (Sjostrom et al. 2008). Which learning rules are most relevant in vivo, and which spike train patterns or other aspects of neural activity trigger experience-dependent plasticity in vivo, remain largely unknown.

One learning rule that appears to mediate some types of experience-dependent plasticity in vivo is spike timing–dependent plasticity (STDP), in which the temporal sequence and interval between pre- and postsynaptic spikes drive plasticity. In classical STDP, pre-leading-post firing (0–20-ms interval) drives LTP, and post-leading-pre firing (0 to 20–50-ms interval) drives LTD, although STDP rules vary considerably across synapses and physiological states. STDP occurs at many neocortical synapses in vitro and can be induced experimentally in vivo by pairing sensory stimulation with precisely timed spikes (Meliza & Dan 2006, Jacob et al. 2007). STDP mechanisms are surprisingly diverse, involving NMDA-LTP and NMDA-LTD at some synapses (Froemke et al. 2005), NMDA-LTP and CB1-LTD at others (Sjostrom et al. 2003, Bender et al. 2006b, Nevian & Sakmann 2006) and mGluR-LTD at others (Egger et al. 1999). STDP has powerful, Hebbian-like computational properties that predict development and plasticity of sensory maps (Song & Abbott 2001). For detailed review of STDP, see Caporale & Dan (2008).

STDP is strongly implicated in one form of perceptual learning-related plasticity in V1. Sequentially flashing brief visual stimuli at two nearby retinotopic locations imposes specific spike timing on V1 neurons representing these locations. Repeated presentation of such stimuli to anesthetized adult cats at short, STDP-like intervals (~10 ms) alters the functional strength of synaptic connections between activated neurons and spatially shifts neuronal receptive fields in a manner consistent with STDP. In humans, the same conditioning procedure causes a shift in the perceived location of visual stimuli, again consistent with STDP (Fu et al. 2002). Similar conditioning with sequentially flashed oriented stimuli shifts V1 orientation tuning and alters perception of orientation (Yao & Dan 2001). Thus, STDP drives perceptual learning in V1 in response to appropriate timed visual stimuli. In A1, a similar conditioning procedure shifts frequency tuning of A1 neurons consistent with STDP (Dahmen et al. 2008). Other forms of perceptual learning that do not involve precise stimulus timing are less likely to involve STDP and may involve other mechanisms.

STDP may also drive LTD at L4-L2/3 synapses in response to sensory deprivation. L4-L2/3 synapses in S1 exhibit robust STDP in vitro (Feldman 2000, Nevian & Sakmann 2006) and undergo LTD in vivo in response to whisker deprivation (see above). During normal sensory responses in vivo, L4 neurons fire a few ms before cocolumnar L2/3 neurons, consistent with a serial relay of sensory information from L4 to L2/3. Simulated whisker deprivation acutely causes firing order to reverse, and firing to decorrelate, in deprived columns. These spike timing changes quantitatively predict STDP-LTD induction at L4-L2/3 synapses, suggesting that LTD is induced by STDP in vivo (Celikel et al. 2004). In V1, retinotopic plasticity following focal retinal lesions has also been reported to be more consistent with STDP than with standard correlation-dependent plasticity (Young et al. 2007).

Homeostatic Plasticity

Slower, non-Hebbian forms of plasticity exist that globally adjust synapse strength and neuronal excitability to maintain mean cellular activity at a set point level (Turrigiano & Nelson 2004). This homeostatic plasticity was discovered in cortical cultures in vitro, where experimentally increasing (or decreasing) network activity over hours to days causes a uniform, multiplicative decrease (or increase) in excitatory synapse strength, termed homeostatic synaptic scaling. Opposite plasticity occurs at some inhibitory synapses; here decreased activity causes a reduction in inhibitory synapse strength (Turrigiano & Nelson 2004). Intrinsic excitability, NMDAR content at synapses, and excitatory-inhibitory balance are also homeostatically regulated. Synaptic scaling is mediated by multiple cellular mechanisms that vary by cell type, time course, brain region, and developmental stage. In neocortex, scaling of excitatory synapses onto principal neurons is expressed primarily by regulating AMPA receptor insertion, similar to NMDA-LTP and LTD (Turrigiano & Nelson 2004). How cellular or network activity is read out to drive homeostatic plasticity is not clear. Recent work suggests that glial cells, which detect mean local network activity, trigger one form of synaptic scaling in hippocampus and visual cortex by secreting the cytokine tumor-necrosis factor-α (TNF-α) (Stellwagen & Malenka 2006, Kaneko et al. 2008b).

Homeostatic plasticity occurs in vivo and may explain homeostatic changes in sensory responses with substantial overuse or deprivation. In the monocular region of V1, contralateral eye closure dramatically reduces sensory activity and leads to increased network excitability and spontaneous firing (Desai et al. 2002, Maffei et al. 2004) and to increased visual responses to the closed eye (Mrsic-Flogel et al. 2007). Multiple homeostatic mechanisms are involved, including scaling up of excitatory synapse strength, cell-type specific changes in inhibitory circuits, and changes in intrinsic excitability, depending on the precise visual deprivation paradigm (Maffei & Turrigiano 2008). Synaptic scaling obeys layer-specific critical periods but persists in L2/3 through adulthood. Homeostatic synaptic and intrinsic plasticity occur in A1 with peripheral hearing loss (Kotak et al. 2005). In S1, synaptic scaling has not yet been observed, but sensory activation drives homeostatic changes in L4 inhibitory circuits (see below) and in glutamate transport by astrocytes (Genoud et al. 2006).

Homeostatic plasticity may also contribute to response potentiation to spared inputs during deprivation-induced map plasticity. In a homeostatic model, deprivation of a subset of inputs drives rapid Hebbian weakening on deprived pathways and more slowly drives a homeostatic increase in global synapse strength and/or intrinsic excitability, which increases responses to spared inputs (and partially offsets weakening of deprived inputs). Consistent with homeostasis, monocular deprivation slowly potentiates deprived-eye responses in neurons in binocular V1 that lack open-eye input, and binocular deprivation slowly potentiates responses to both eyes (Mrsic-Flogel et al. 2007). Moreover, knockout of TNF-α, which prevents homeostatic synaptic scaling in hippocampus, blocks potentiation of open-eye responses during ocular dominance plasticity (Kaneko et al. 2008b). Together, these findings strongly suggest that synaptic scaling is one mechanism driving potentiation of open-eye responses during ocular dominance plasticity.

Metaplasticity

A distinct class of plasticity is metaplasticity, that is, experience-dependent changes in synaptic plasticity rules themselves (Abraham & Bear 1996). In metaplasticity, experience-dependent alterations in inhibitory tone, dendritic excitability, NMDA receptor function, or neuromodulation alter the ability of future stimuli to drive LTP and LTD. In V1, visual experience regulates the capacity for LTP and LTD at L4-L2/3 synapses by regulating NMDA receptor subunit composition (Philpot et al. 2003, 2007). This form of metaplasticity is homeostatic: Visual deprivation biases LTP/LTD learning rules toward LTP so that subsequent activity tends to strengthen synapses and restore mean cortical activity. Such metaplasticity was hypothesized to counteract the inherently unstable, positive-feedback nature of Hebbian synaptic plasticity and may act during monocular deprivation to promote LTP by open-eye inputs, thereby driving response potentiation (Bienenstock et al. 1982, Bear et al. 1987). In S1, single whisker experience both drives NMDA-LTP at L4-L2/3 synapses and induces a form of metaplasticity in which a novel mGluR-LTP appears. This mGluR-LTP is required after initial potentiation to maintain synapse strength in vivo (Clem et al. 2008).

Plasticity of GABAergic Cells and Circuits

Although most research has focused on excitatory synapses and circuits as loci for cortical plasticity, recent findings demonstrate that GABAergic inhibitory neurons and circuits are highly plastic and play several important roles in sensory map plasticity.