The cholinergic system influences the local network by diffuse transmission rather than by synaptic transmission (Descarries et al., 1997; Yamasaki et al., 2010). This property is related to the fact that ACh is released from the varicosities that are distributed along the cholinergic axons and that these varicosities show only rare synaptic organizations at the ultrastructural level (Umbriaco et al., 1994; Vaucher and Hamel, 1995; Mechawar et al., 2000). However, the modulation of the cortex by ACh is not widespread and is primarily selective and adapted to the local microfunction due to the differential distribution of varicosities along the cholinergic axons (Zhang et al., 2011) and the differential distribution of the cholinergic receptor subtypes on different neuronal targets. Moreover, ACh release might be triggered by local neuronal activity to induce locally restricted rather than generalized action of the cholinergic system (Laplante et al., 2005). The variety of the cholinergic receptors and their distributions convey subtype-specific functions (Thiele, 2013; Groleau et al., 2014). In V1, AChRs exhibit differential subtype densities across the cortical layers (I-VI) on both excitatory (Gulledge et al., 2009; Thiele, 2013) and inhibitory neurons (Hashimoto et al., 1994). The distinct actions of cholinergic receptors can be related to differences in the conductances of the ionotropic receptor nAChRs for Na⁺, K⁺ (α₄β₂) and Ca²⁺ (α₇) (Rang, 2003) and in the intracellular pathways of the different subtypes of the G-protein coupled mAChRs. Amongst the five mAChR subtypes identified, the M1, M3 and M5 mAChRs are coupled with Gq/11 proteins, which activate phospholipase C and lead to increases in intracellular Ca²⁺ and the M2 and M4 mAChRs are bound with Gi protein that inhibits adenylyl cyclase, which leads to a decrease in cAMP, the inhibition of voltage-gated Ca²⁺ channels and an increased K⁺ efflux (Caulfield and Birdsall, 1998; Wess, 2003). In addition, M1 promotes the opening of NMDARs and induces LTP in the hippocampus (Buchanan et al., 2010; Giessel and Sabatini, 2010).

Stimulation of the cholinergic system in V1 can be achieved via the administration of ACh analogs (e.g., carbachol), cholinergic receptor agonists (e.g., nicotine and selective mAChR drugs) or cholinesterase inhibitors or through electrical or optogenetic stimulation of the cholinergic neurons that project to V1. The cholinergic neurons that project to V1 are located in the basal forebrain (BF), particularly the ventral pallidum, substantia innominata and the horizontal limb of the diagonal band of Broca (HDB; Gaykema et al., 1990; Laplante et al., 2005). Although the nucleus basalis magnocellularis is the main cholinergic nucleus of the BF which innervates the cortical mantle, it projects only weakly to V1 (Luiten et al., 1987; Vaucher and Hamel, 1995); nevertheless, some studies report that the stimulation of this nucleus might induce functional changes in the visual cortex (Goard and Dan, 2009; Pinto et al., 2013). Moreover, although there are GABAergic neurons in the BF, many studies have confirmed that the effects of BF stimulation are identical to those of intracerebral injections of ACh agonists and are primarily mediated by the cholinergic fibers (Dauphin et al., 1991; Ma and Suga, 2005; Dringenberg et al., 2007; Kocharyan et al., 2008; Kang and Vaucher, 2009). There are also intrinsic cholinergic neurons that represent only 10–15% of the total cortical innervation (Eckenstein et al., 1988; Chédotal et al., 1994), and the involvement of these neurons in cortical processing remains unclear.

Cortical responses to sensory stimuli transmitted by the LGN are amplified during learning and experience-dependent plasticity to emphasize relevant information (Sarter et al., 2005; Wang et al., 2013). These thalamic afferents are of prime relevance because they define the receptive fields and other properties of V1 neurons. Complex information is extracted according to its properties (e.g., orientation) via projections to different columns (in primates) or specific cells (in rodents). Cholinergic activation in this layer induces a general increase in responsiveness regardless of the features of the visual stimuli (e.g., orientation; Disney et al., 2012), which allows the cortex to respond reliably to weak stimulation (Disney et al., 2007). ACh increases the thalamocortical input through presynaptic nAChRs on the thalamocortical fibers (Gil et al., 1997; Disney et al., 2007; Figures 2A, 4). The M1 mAChR also amplifies the spiny stellate cell/pyramidal cell response through a postsynaptic intracellular pathway (Gu, 2003), but inhibition through the M4 mAChR has also been observed on spiny neurons in the somatosensory cortex (Eggermann and Feldmeyer, 2009). Interestingly, the cholinergic facilitation of thalamocortical inputs in sensory cortex slices is ACh-concentration dependent. High doses of ACh enhance the thalamocortical afferents both in vitro and in computational models (Hasselmo, 2006; Deco and Thiele, 2011). Together, these results indicate that, under conditions of high levels of ACh release, the enhancement of the thalamocortical inputs in layer IV facilitates the transmission of sensory information and induces experience-dependent plasticity (e.g., learning).

In addition to the enhancement of thalamocortical inputs, ACh might modulate intracortical connectivity either by suppressing lateral inhibition (Kimura and Baughman, 1997; Metherate et al., 2005; Metherate, 2011) or suppressing the spread of the excitation of thalamic inputs (Kimura et al., 1999; Silver et al., 2008). The presynaptic mAChRs that are located on the glutamatergic fibers induce a suppression of the intracortical neurons (Gil et al., 1997), although the inhibition of GABAergic terminals induces a disinhibition of the pyramidal cells (Ji and Dani, 2000; Christophe et al., 2002; Seeger et al., 2004; Salgado et al., 2007). Intracortical connectivity modulates the response intensity and the output of V1 neurons (Figure 3). The lateral connections also synchronize the firing of similar neuronal populations (Gilbert and Wiesel, 1989; Lien and Scanziani, 2013), which allows for lateral correlation between neurons with similar orientation preferences during typical perceptual learning tasks (e.g., the Vernier acuity test) (Ramalingam et al., 2013). The differential action of ACh on lateral connections might simultaneously enhance specific modules of the same orientation (lateral correlation) while depressing adjacent irrelevant modules (McGuire et al., 1991; Stettler et al., 2002). A recent study using optogenetics showed that inhibition of the intracortical excitatory neurons leads to a receptive field reduction (Li et al., 2013), and this finding is consistent with the effect of ACh release in V1 (Roberts et al., 2005; Zinke et al., 2006) and the increases in the population receptive fields of M1/M3 mAChR knock-out mice (Groleau et al., 2014). Furthermore, an ACh esterase inhibitor reduces surround suppression in a perceptual study in humans (Kosovicheva et al., 2012), which could be indicative of a weakening of lateral connections. Hasselmo (2006) proposed that high ACh levels suppress the magnitude of feedback excitation, whereas low ACh levels result in weaker afferent input to the cortex. Similarly, Deco and Thiele (2011) also proposed that high ACh levels decrease the intracortical interactions and that low ACh increase these interactions. The hypothesis of these authors was confirmed in an in vitro study that showed that the enhancement of the recurrent cortical activity in low-dose ACh conditions was independent of the thalamocortical input (Wester and Contreras, 2013). Together, these results suggest that during intense ACh release, the intracortical connections are inhibited, which relieves the sensory cortices from recurrent connections. However, in low concentration of ACh situations, the lateral connections might amplify the thalamocortical activity amongst similarly tuned neurons.

These effects have primarily been recorded within layer II/III; however, in layers I, V and VI, which are primarily involved in feedback mechanisms, ACh might also influence feedforward processing by interacting with neurons in layers IV and II/III (De Pasquale and Sherman, 2012). Layer I neurons are densely innervated by the cholinergic projections (Vaucher and Hamel, 1995; Mechawar et al., 2000). It has been shown that inhibitory actions mediated by AChRs can suppress layer II/III (Zinke et al., 2006; Alitto and Dan, 2012; Soma et al., 2013b) and layer V pyramidal neuron activity (Lucas-Meunier et al., 2009; Amar et al., 2010) and can also inhibit the cortical GABAergic network and thus result in the disinhibition of the majority of the cortical layers (Christophe et al., 2002). It has been observed that local ACh application primarily suppresses the activity of layer VI neurons (Disney et al., 2012), which can alter the activation of all of the layers of V1 in a linear manner via the intracortical pathway (Olsen et al., 2012) and alter the activation of the thalamocortical fibers (Cudeiro and Sillito, 2006; Sillito et al., 2006). Cholinergic action might thus disinhibit the activities of other layers by suppressing layer VI. Topical injections of ACh into layer V produce the predominant effect of facilitation of the regular and fast-spiking cells (Soma et al., 2013b), although local ACh activation seems to decrease excitatory drive through presynaptic M1 mAChRs (Kimura and Baughman, 1997) and to increase inhibitory drive through M3 mAChRs (Amar et al., 2010). Similarly, an increase in the activation of GABAergic neurons activation in layer V has been observed following repetitive BF/visual pairing (Kang et al., 2014). Layer V pyramidal neurons send dense projections to the superior colliculus and diverse thalamic nuclei that are involved in focused attention.

Finally, ACh can promote the co-activation of different cortical areas and layers which might be an efficient method for the selection of visual information via a summation of the temporally coincident presynaptic spikes (Fries et al., 2007). It has been shown that visually driven gamma power is differentially distributed across the layers of V1 (Xing et al., 2012) and that gamma oscillations can be induced by cholinergic stimulation (Rodriguez et al., 2004; Bhattacharyya et al., 2013).

In conclusion, BF stimulation that facilitates the release of ACh in multiple layers of V1 might act in diverse manners and results in the enhancement of visual stimulus-driven responses. The pre-amplified responses of layer IV are filtered out by GABAergic neurons of layer II/III to transfer task-relevant information to higher visual cortical areas. The activated synaptic connections can be modulated by layers V and VI or by the feedback mechanism of layer I. Differential responses across layers might be integrated by the synchronization of their activities in the gamma-band to facilitate visual processes.

ACh has been suggested to control the balance between bottom-up and top-down processing through attentional mechanisms (Yu and Dayan, 2002, 2005; Sarter et al., 2005). This influence is mediated by pre-synaptic thalamocortical nAChRs (Gil et al., 1997; Disney et al., 2007). Attention that is prompted by the properties of a stimulus, i.e., the saliency of the stimulus relative to the background, is said to be bottom-up attention, whereas attention that is prompted by the voluntary direction of focus toward a specific stimulus is defined as top-down attention. Although it can be difficult to separate bottom-up and top-down attentional control (Ansorge et al., 2010; Egeth et al., 2010; Eimer and Kiss, 2010; Theeuwes, 2010), some studies have shown that cholinergic activity influences bottom-up attention. The effect of ACh on bottom-up attention might occur not only in V1 but also in early processing areas such as the thalamus. For example, the direct injection of 192-IgG saporin into the BF causes a complete loss of cholinergic projections to the neocortex but causes restricted fiber lesions when injected into V1. The injection of 192-IgG saporin into the BF but not V1 affects performance in the sustained attention task (McGaughy and Sarter, 1998, 1999). In addition, compared to controls and ex-smokers, human smokers have been shown to exhibit increased subcortical activity during an attentional task (Nestor et al., 2011). These data indicate that attentional dysfunction following cholinergic lesions might be due to the disruption of detection processes that are independent of V1. However, there is no direct evidence of cholinergic enhancement effect in bottom-up attention in human studies (Rokem et al., 2010). In contrast, there is a growing body of evidence showing that ACh is involved in top-down attention. Direct effects of ACh on attention in the visual cortex have been measured (Herrero et al., 2008; Bauer et al., 2012). Specifically, Herrero et al. provided direct evidence that ACh in V1 enhances the cortical response to an attentional demand (Herrero et al., 2008). It has also been shown that lesions to the cholinergic system impair attention performance and increase neuronal activity in the PFC upon the presentation of distractors (which trigger top-down attention) (Gill et al., 2000). Taken together, these results indicate that ACh can facilitate task-relevant learning in V1 by promoting attentional states in both top-down and bottom-up manners.

Response gain modulation by ACh has frequently been observed (Disney et al., 2007; Aggelopoulos et al., 2011; Bhattacharyya et al., 2013; Soma et al., 2013a) and follows the gain control model at least in terms of the contrast-response function. Increasing thalamocortical pathway input in a context-independent manner while context-dependent intracortical suppression occurs might facilitate the transmission of information related to novel stimuli. In V1, context-dependent (i.e., increases in the maximal response) and independent (i.e., increases in the baseline response) gain control due to cholinergic effects have both been observed (80% and 20%, respectively) without any laminar bias (Soma et al., 2013b). These findings could be related to the optimization of the gain of supragranular pyramidal cells controlled by ACh which could result in the detection of novel stimuli and hence perceptual learning (Moran et al., 2013). Interestingly, gain modulation was proposed as function that underlies of attentional control (Keitel et al., 2013) and network connectivity (Haider and McCormick, 2009). The high gain that results from the amplification of the responses of excited neurons is similar to attention processes (Servan-Schreiber et al., 1990; Eldar et al., 2013) and hence facilitates learning. Taken together, these results suggest that ACh might assist in visual perceptual learning via modulation of cortical responses through gain control in both stimulus-dependent and -independent manners.

At the neuronal level, ACh has been shown to contribute to cortical plasticity through both the acute and long-term modulation of synaptic responses (Sato et al., 1987; Soma et al., 2012). The impairment of learning by cholinergic antagonists is similar to the effect of blocking cortical plasticity mechanisms and LTP with NMDA receptor (NMDAR) antagonists (Morris et al., 1986; Artola and Singer, 1987; Cooke and Bear, 2010). In most situations, LTP in the visual cortex induced by high theta-burst stimulation (100 Hz) (Heynen and Bear, 2001; Dringenberg et al., 2007) has been found to be NMDAR-dependent. Interestingly, cholinergic system-induced cortical plasticity has also been found to be NMDAR-dependent (Verdier and Dykes, 2001; Dringenberg et al., 2007; Kang and Vaucher, 2009) but independent of theta-burst stimulation (Kirkwood et al., 1999; Figure 1B). Previous studies in hippocampal slices have shown that NMDAR opening during LTP induction is facilitated by mAChR activation (Buchanan et al., 2010) and administration of ACh to pyramidal neurons (Shinoe et al., 2005). Additionally, NMDAR-dependent long-term facilitation of synaptic responses is associated with ACh release in V1, and LTP is impaired in the visual cortices of mAChR knock-out mice (Origlia et al., 2006).

Another contribution of the cholinergic system to cortical plasticity mechanisms in V1 is the alteration of the excitatory and inhibitory (E-I) balance (Figure 1C). The excitatory and inhibitory synaptic inputs tend to equilibrate during maturation to optimally tune the neurons according to sensory experiences (Hensch et al., 1998; Sun et al., 2010) during the critical period; i.e., the post-natal time window during which mammals visual cortices are highly plastic that terminates with the maturation of the neurons. It has been proposed that disrupting the E-I balance can re-open the critical period after maturation (Hensch, 2004). Neuromodulation can also disrupt the E-I balance and contribute to cortical plasticity. Recent studies have also demonstrated numerous examples of cortical plasticity that are modified by the inhibitory system (Hensch, 2005). The onset of the critical period is accelerated by GABAA inhibitory receptor activation (Fagiolini and Hensch, 2000; Iwai et al., 2003). Conversely, it is also possible to re-induce plasticity after the critical period by reducing the inhibitory drive via the injection of GABAA receptor antagonists (Harauzov et al., 2010). As the inhibitory system is strongly modulated by the cholinergic system through the protein Lynx1 (Takesian and Hensch, 2013), which acts as a brake on nAChR-dependent plasticity (Morishita et al., 2010), by nAChRs (Christophe et al., 2002; Arroyo et al., 2012), or by mAChRs (Salgado et al., 2007), cholinergic activation might modulate the E-I balance and facilitate cortical plasticity in adults that would promote perceptual learning. An interaction between the cholinergic and GABAergic systems has been shown to occur following BF stimulation that increases the activation of Parvalbumin-positive (PV+) neurons through mAChRs (Dotigny et al., 2008; Alitto and Dan, 2012). Interestingly, Alitto and Dan used an optogenetic method to show that the nAChRs on vasoactive intestinal peptide-positive (VIP+) neurons and layer I neurons can inhibit excitatory and PV+ neurons (Christophe et al., 2002).

The cholinergic modulation of V1 thus promotes cortical plasticity through LTP-like long-term enhancement of synaptic responses to subsequent presentations of a visual stimulus and through control of the excitatory-inhibitory balance that regulate the strength of cortical output and internal connectivity. The cortical plasticity induced by cholinergic stimulation could transfer the acute cholinergic effect into long-term scale to produce visual precision.

As mentioned above, ACh can induce NMDAR-dependent long-term modifications of postsynaptic glutamatergic neurons which are related to memory formation. The opening of the NMDAR launches a second messenger cascade and guides the expression of synaptic glutamate receptors (Regehr and Tank, 1990; Zhong et al., 2006) but also activates autoregulated kinases that confer a persistent improved response of the neuron to the stimulus. Immunohistochemistry for the c-Fos, which is an immediate early gene and also a transcription factor for synaptogenesis genes, has revealed that c-Fos is increased in layer II/III pyramidal neurons following a repetitive BF/visual stimulation (Kang et al., 2014), which may be indicative of the formation of new synapses and LTP mechanisms. Repetitive pairing of the cholinergic and visual stimulation also induces morphological reorganization, i.e., increase in the numbers of cholinergic varicosities in the proximity of the neurons that are sensitive to the orientation of the stimulus (Zhang et al., 2011). This increased number of cholinergic inputs, along with postsynaptic mechanisms, would increase and consolidate the response of the activated neurons to ameliorate its long-term efficiency. Thus repetitive cholinergic stimulation might enhance the encoding of the memory and morphological modifications.

We suggest that selection of decisive inputs is controlled by the cholinergic system and contributes to the specific enhancement of a particular stimulus in perceptual learning. Modulation of the orientation selectivity of the neurons provides a great example of the possible improvement of perceptual sensitivity. Training of the rat to a preferred or a non-preferred orientation might increase the cortical response for this orientation (Cooke and Bear, 2010; Figure 4). These mechanisms are facilitated by repetitive cholinergic activation, which improve orientation discrimination of human or rats (Rokem and Silver, 2010; Kang et al., 2014). Repetitive cholinergic stimulation coupled with a certain orientation stimulus might favor the discrimination of this stimulus by two different cellular mechanisms (Figure 4). First, ACh can harmonize the activation of the whole dendritic tree of layer II/III neurons to preserve their orientation selectivity and confer responsiveness to new orientation—the dendrites of the layer II/III neurons receive inputs randomly over all of their branches, some of which are selective for the neurons’ un-preferred orientations (Jia et al., 2010). Second, the cholinergic system can enhance orientation discrimination through its interaction with the GABAergic system which assists in the sharpening (Isaacson and Scanziani, 2011) of the convergent input in the layer II/III neurons (Nassi and Callaway, 2009) but also filters out task-relevant information during perceptual learning (Roberts and Thiele, 2008). PV+ and somatostatin-positive (SOM+) GABAergic neurons are particularly involved in orientation tuning in V1 (Atallah et al., 2012; Wilson et al., 2012). It has been shown that the specific activation of PV+ neurons in V1 improves orientation discrimination abilities in awake rats during perceptual learning (Lee et al., 2012) and repetitive coupling of ACh to visual stimulation activates the V1 GABAergic neurons (Dotigny et al., 2008; Kang et al., 2014).

Thus repetitive cholinergic pairing to sensory training enhances the cortical response to trained feature of the sensory stimulus that increases the influence of the feedforward afferent.

Repetitive cholinergic stimulation first promotes attentional mechanisms that are necessary to perceptual learning (Ahissar and Hochstein, 1993; Schoups et al., 2001; Li et al., 2004; Mukai et al., 2007). These attentional processes might be also related to synchronization in the gamma band (30–90 Hz) (Fries et al., 2008) induced by repetitive cholinergic stimulation which has been proposed to facilitate the transfer of the visual information to higher visual areas. ACh can also promote task-irrelevant perceptual learning that occurs in the absence of conscious effort (Skrandies and Fahle, 1994; Watanabe et al., 2002; Gutnisky et al., 2009). Compared to task-relevant learning, which utilizes focused attention as reinforcement, studies of task-irrelevant learning have suggested that reward serves as the reinforcement signal (Seitz et al., 2009; Chubykin et al., 2013). During task-irrelevant learning, the response to a feature on which attention was not directed can also be enhanced (Watanabe et al., 2001; Giordano et al., 2009; Gutnisky et al., 2009). Interestingly, rewards can affect the visual response in V1 (Shuler and Bear, 2006), and the cholinergic system can influence reward timing expectancy (Chubykin et al., 2013). To reconcile studies showing a role of attention in perceptual learning or not, Roelfsema proposed that the attentional feedback signal related to the cholinergic system that enhances the plasticity of task-relevant features in the visual cortex also causes the inhibition of task-irrelevant features so that their plasticity is switched off (Roelfsema et al., 2010).

To a cognitive point of view, by modulating synaptic transmission in V1 and modifying the cortical dynamics, ACh can also participates in the perceptual inference to increase the strength of the representation of trained stimuli and reduce the sensory noise (Yu and Dayan, 2002) and induce sensory precision (Moran et al., 2013). It might suppress the top-down sources in the balance between top-down and bottom-up information integration in V1 (Yu and Dayan, 2005). This is in agreement with a recent study demonstrating that the cholinergic enhancement reduces the connectivity strength between cortical regions involved in attention and V1 (Ricciardi et al., 2013) and reduce the activity in frontoparietal regions (Furey et al., 2008). This suggests an increased neural efficiency in the processing of the trained stimulus that leads to an improved perceptual task performance (Ricciardi et al., 2013) linked to an automation of the cortical processing and a reduction of the attentional load required to process the trained stimulus (Furey, 2011).

Together, the findings from recent work using different techniques suggests that cholinergic pairing induces perceptual learning via different mechanisms that include the following: (1) the use of the layer II/III GABAergic system to filter the pre-amplified response from layer IV; (2) NMDAR-dependent modification at the postsynaptic level to induce long-term augmentations of individual neurons, and an increase in the numbers of cholinergic varicosities to facilitate ACh release; and (3) changes in the efficiency of the connectivity between cortical areas and bottom-up and top-down control.