Abstract
Based on recent molecular genetics, as well as functional and quantitative anatomical studies, the basal forebrain (BF) cholinergic projections, once viewed as a diffuse system, are emerging as being remarkably specific in connectivity. Acetylcholine (ACh) can rapidly and selectively modulate activity of specific circuits and ACh release can be coordinated in multiple areas that are related to particular aspects of cognitive processing. This review discusses how a combination of multiple new approaches with more established techniques are being used to finally reveal how cholinergic neurons, together with other BF neurons, provide temporal structure for behavior, contribute to local cortical state regulation, and coordinate activity between different functionally related cortical circuits. ACh selectively modulates dynamics for encoding and attention within individual cortical circuits, allows for important transitions during sleep, and shapes the fidelity of sensory processing by changing the correlation structure of neural firing. The importance of this system for integrated and fluid behavioral function is underscored by its disease-modifying role; the demise of BF cholinergic neurons has long been established in Alzheimer's disease and recent studies have revealed the involvement of the cholinergic system in modulation of anxiety-related circuits. Therefore, the BF cholinergic system plays a pivotal role in modulating the dynamics of the brain during sleep and behavior, as foretold by the intricacies of its anatomical map.
- basal forebrain population dynamics
- cortico-cortical coherence
- cholinergic engram of fear
- dynamics of encoding and retrieval
- signal and noise correlation
Introduction
The basal forebrain (BF) is composed of structures including the medial septum, ventral pallidum, vertical and horizontal diagonal band nuclei (VDB, HDB), substantia innominata/extended amygdala (SI/EA), and peripallidal regions; these structures contain a heterogeneous mixture of neuron types that differ in transmitter content, morphology, and projection pattern. A prominent feature of the mammalian BF is the presence of a collection of aggregated and nonaggregated, large neurons, many of which contain choline acetyl transferase (ChAT), the critical enzyme in the synthesis of acetylcholine (ACh); these neurons project to the cerebral cortex, the hippocampal complex, and the amygdala. BF areas rich in cholinergic neurons also contain GABAergic, glutamatergic, and peptidergic interneurons and projection neurons in rodents and primates (Gritti et al., 2003; Záborszky et al., 2015b,c). The collection of large, darkly stained (in Nissl sections) neurons, many of which are cholinergic in primates including humans, are referred to as the nucleus basalis of Meynert.
The highly complex BF system has been implicated in cortical activation, affect, attention, sensory coding, motivation, and memory, and in disorders such as Alzheimer's disease (AD), Parkinson's disease, schizophrenia, autism, attention deficit disorder, and drug abuse (Picciotto et al., 2012; Ballinger et al., 2016; Zhang et al., 2016). Although the original description of the cholinergic system in the early 1980s acknowledged that the BF corticopetal system shows some topographic organization (Price and Stern, 1983; Saper, 1984), anatomical data contributed to the BF cholinergic projections being lumped together as part of the “diffuse cortical projection systems” (Saper, 1987). Recent anatomical studies revealed that the cholinergic projection to the neocortex is not diffuse, but instead is organized into cortical target-specific groups of cholinergic neurons that receive a specific combination of inputs (Záborszky et al., 2015a; Gielow and Záborszky, 2017). Advances in refined pharmacological techniques have defined a phasic ACh release in a spatially selective fashion in attention and sensory processing and tonic ACh release over broad cortical areas in a brain-state-dependent manner (Parikh et al., 2007; for review, see Ballinger et al., 2016). The observation of spatially specific rapid ACh release (cholinergic “transients”) in addition to slower and spatially broader ACh release may reflect the existence of two functionally distinct types of cholinergic neurons (Unal et al., 2012) and could support two different aspects of information processing (Sarter and Kim, 2015).
The timing of ACh release is important for actions throughout the brain, including cortex, hippocampus, and amygdala. Selective cholinergic activation in the prefrontal cortex (PFC) on the scale of subseconds to seconds is associated with cue detection and cue-triggered changes in goal-oriented behavior (goal-driven attention) (Parikh et al., 2007; Gritton et al., 2016; Howe et al., 2017), whereas changes on the scale of minutes may occur over the entire cortex to support more general arousal (Parikh et al., 2007). In the sensory cortex, ACh increases the signal-to-noise ratio of evoked responses and contributes to a change in the correlation structure of intracortical noise (Pinto et al., 2013; Minces et al., 2017), thereby enhancing neuronal response reliability (stimulus-driven attention). The acute enhancement of signal-to-noise ratio could arise from cholinergic depolarization of pyramidal cells and interneurons coupled with presynaptic inhibition of glutamatergic and GABAergic transmission (Patil and Hasselmo, 1999). Over a longer timescale, cholinergic signaling in the hippocampus enhances synaptic plasticity, including LTP (Blitzer et al., 1990; Burgard and Sarvey, 1990) short-term depression, and long-term depotentiation (Huerta and Lisman, 1996), all of which are considered to be cellular substrates of memory and are affected by the precise timing of ACh release in the target area (Blitzer et al., 1990; Burgard and Sarvey, 1990; Gu and Yakel, 2011). Finally, cholinergic signaling in the basolateral amygdala has a state-dependent and largely inhibitory effect on pyramidal cell firing (Unal et al., 2015). This may be important in state-dependent optimization of emotionally salient memories (Jiang et al., 2016). One recent study describes a biologically detailed tissue model of neocortical microcircuitry predicting how ACh affects different types of neurons, synapses, and global network states (Ramaswamy and Markram, 2018) and earlier models addressed the modulation of hippocampal function (Hasselmo et al., 1995; Hasselmo, 2006).
Anatomical organization of the BF cholinergic system
Using retrograde tracers deposited into disparate cortical areas to map labeled cells in the BF, it was demonstrated that the BF has a complex topographic organization consisting of segregated and overlapping pools of projection neurons. Furthermore, the extent of overlap of BF-projecting populations seems to depend on the degree of connectivity between their cortical targets (Záborszky et al., 2015a).
Preliminary studies (P Varsanyi and L Záborszky, unpublished data) are revealing that BF neurons that project to different cortical targets are organized in BF clusters according to both topographical and functional principles (Fig. 1). For example, cell populations projecting to cortical areas representing different body parts, such as S1 whisker and S1 hindlimb, have very low BF spatial overlap. In contrast, in some BF areas neurons projecting to S1 are clustered with neurons that project to the corresponding M1 regions, whereas other areas of the BF contain only M1 projecting neurons. Similarly, V1 or V2 projecting cells are mostly segregated (see also Huppé-Gourgues et al., 2018), but in specific locations are intermixed with cells projecting to various association areas, including retrosplenial, medial prefrontal, and orbitofrontal cortex. Also, auditory projecting cells may be intermixed with specific somatosensory, perirhinal, prorhinal, and insular cortex projecting cells.
Results of a clustering model using 20 cortical ontology categories and 29 BF clusters. Each cluster is indicated by different color. White arrow points to a large purple cluster containing 1292 cells of which 15% project to retrosplenial, mPFC and OFC and 12% to V2. Medial part is the septum, lateral part toward the globus pallidus.
It is envisaged that, in specific clusters, neighboring cholinergic cells may innervate single cortical targets or collateralize to double or triple targets (Li et al., 2018). We propose that the clusters in the BF may serve as “modules” that redistribute information from specific locations in the BF to subsets of associated cortical areas, allowing spatially selective modulation of individual or joint cortical areas.
BF cholinergic neurons are modulated by diverse inputs
Using electron microscopy, BF cholinergic neurons have been shown to receive synaptic inputs from the ventral and dorsal striatum, hypothalamus, amygdala, and brainstem tegmentum. Additionally, adrenaline, noradrenaline, dopamine, GABA, glycine, Vglut1, Vglut2, orexin, somatostatin (SOM), neuropeptide Y (NPY), substance P, and enkephalin synapses were identified on BF cholinergic neurons (Záborszky and Gombkoto, 2018). Cortical inputs to cholinergic cells were also recently suggested based on monosynaptic virus tracing studies (Do et al., 2016; Hu et al., 2016). For example, it seems that target-identified cholinergic cells receive a specific combination of cortical inputs: motor cortex projecting cholinergic neurons receive a substantial S1–S2 input, whereas cholinergic cells innervating the medial PFC (mPFC) do not receive such input. Conversely, the allocortex contributes ∼11% of the total input to the cholinergic neurons projecting to mPFC and only ∼1% to the cholinergic neurons projecting to the motor cortex (M1/M2) (Gielow and Záborszky, 2017).
The input dynamics to the BF cholinergic neurons remain open for investigation because no study to date has investigated the convergence of two or more types of afferents and there is a lack of electrophysiology data about how the various inputs sculpt neuronal firing properties. Cholinergic neurons possess extensive local collaterals (Záborszky et al., 2002); cholinergic–cholinergic synapses were described in the septum (Bialowas and Frotscher, 1987) and optogenetic stimulation of cholinergic neurons in slice inhibits spiking of other cholinergic neurons (CT Unal and L Záborszky, unpublished data). The continuously shifting input patterns to cholinergic neurons may reflect environmental salience and behavioral demands that ultimately shift the oscillatory dynamics of the BF for maximal coordination with cortical targets (Quinn et al., 2010; Tingley et al., 2015).
Local neurons in the BF and their interconnections
As indicated by the complex temporal dynamics of the aggregate of BF neurons, cholinergic neurons exist in the company of a diversity of cell types of the BF (Záborszky and Gombkoto, 2018). For example, neighboring GABAergic neurons are a diverse cell population in the BF and their axons frequently surround other GABAergic, cholinergic, and glutamatergic neurons in rodents (Záborszky et al., 1986; Henderson et al., 2010). Axons of parvalbumin (PV)-containing GABAergic BF neurons in rats possess few collaterals and some of these synapse with cholinergic dendrites (Záborszky and Duque, 2000), although in optogenetic experiments, stimulation of PV cells did not induce responses in neighboring cholinergic cells (Xu et al., 2015). Cortically projecting BF PV/GABAergic cells are important in regulating cortical gamma band oscillations (Kim et al., 2015), which increases the fidelity of sensory and perceptual coding (Harris and Thiele, 2011; Beaman et al., 2017).
In the rat, many cholinergic neurons projecting to the basolateral nucleus of the amygdala are located in the ventral pallidum and express Vglut3, suggesting that they may release both glutamate and ACh (Nickerson Poulin et al., 2006). Some of the glutamatergic input to cholinergic neurons (Hur et al., 2009) originates in locally arborizing Vglut2 neurons because optogenetic stimulation of Vglut2 neurons excites cholinergic cells (Xu et al., 2015).
A reversed cortical EEG phase relationship exists between BF local NPY and cholinergic cell firing, as demonstrated in anesthetized rats in vivo (Duque et al., 2000). NPY injection into the BF induces changes in cortical EEG in both anesthetized and freely moving rats, suggesting that cholinergic output is regulated by local NPY neurons (Tóth et al., 2005, 2007). NPY local neurons synapse on cholinergic neurons both in the SI/HDB area (Záborszky et al., 2009) and in the caudal globus pallidus/SI area (Nelson and Mooney, 2016). NPY, via NPY Y1 receptors, inhibits the majority of cholinergic neurons. (Záborszky et al., 2009).
SOM has been identified in synapses on cholinergic projection neurons (Záborszky, 1989). A portion of these SOM-containing terminals may originate from local neurons (Záborszky and Duque, 2000). Using in vitro patch-clamp techniques, studies suggest that SOM presynaptically inhibits both GABA and glutamate release onto BF cholinergic neurons (Momiyama and Záborszky, 2006). SOM neurons inhibit Vglut2, cholinergic, and PV neurons and receive excitatory input from cholinergic and Vglut2 cells (Xu et al., 2015).
Cholinergic network firing in behaving rodents
Verified cholinergic neurons in behaving mice were first recorded in 2015 using optogenetic tagging during an auditory sustained attention task (Hangya et al., 2015). Cholinergic neurons responded to primary reinforcers, including innate reward and punishment, with remarkable speed and precision, similarly to responses of putative cholinergic neurons described in the primate (Monosov et al., 2015). Activity in verified cholinergic neurons was also recorded via microendoscopic calcium imaging in mice during spontaneous innate and learned behaviors (Harrison et al., 2016). In contrast to GABAergic or Vglut2 neurons, cholinergic neurons became active at the onset of running and licking and also in response to overt punishment regardless of behavioral context (Harrison et al., 2016). The robust activation of BF cholinergic neurons by movement (Harrison et al., 2016) might mediate the running-induced gain increases evident in sensory cortex (Fu et al., 2014; McGinley et al., 2015).
High-resolution electrophysiological recording of basalo-cortical networks in awake, behaving rats with optogenetically tagged cholinergic neurons revealed functional connections compatible with meso-scale and large-scale anatomical networks (Záborszky and Gombkoto, 2018). Extracellular spikes were recorded simultaneously in the BF, in the orbitofrontal cortex (OFC), and in the visual association cortex (V2). Several putative functional connections within the BF and between BF and specific cortical areas can be recognized (see Fig. 3 in Záborszky and Gombkoto, 2018) using short-latency temporal interactions (Fujisawa et al., 2008). The cholinergic influence in the cortex (Gombkoto et al., 2016) supports the hypothesis that cholinergic modulation of cortical microcircuits is layer specific, corresponding to a layer-specific receptor pattern (Muñoz and Rudy, 2014; Verhoog et al., 2016; Obermayer et al., 2017). Optogenetic stimulation in the BF modulates gamma coherence at spatially specific locations in V2–OFC cortical areas (P Gombkoto and L Záborszky, 2016; P Gombkoto and L Zaborszky, unpublished data), suggesting that the cholinergic system is capable of behavior-dependent modulation of corticocortical functional connectivity, enabling information exchange between interconnected cortical regions. This could involve the cholinergic modulation of feedback synaptic connections weakening local influence and enhancing the influence of longer-range feedforward connections between cortical regions (Hasselmo and Cekic, 1996). These electrophysiological findings, together with the clustered organization of the BF projection system, suggest that the BF could coordinate activity in remote but associated cortical areas, which is consistent with experiments showing coordinated ACh release in PFC and hippocampus (Teles-Grilo Ruivo et al., 2017). This evidence coincides with recent studies indicating that the firing rate dynamics of individual populations of BF neurons align with the local oscillations in the BF in a nested fashion (“multiplexing”), demonstrating a process whereby coordination of local events in the BF might maximize transfer of information to cortical regions (Tingley et al., 2018).
Relation of cholinergic firing to brain states and networks
Three basic cortical states are associated with differential BF activity: wake, slow-wave sleep (SWS), and rapid eye movement (REM) sleep (McCormick et al., 2015). Cholinergic neurons show greater activity during waking and REM sleep compared with SWS (Lee et al., 2005; Xu et al., 2015). According to recent studies, finer distinctions can be made in terms of internal cortical dynamics, pupil diameter, and responsiveness to external stimuli (Harris and Thiele, 2011; Vyazovskiy et al., 2011; McGinley et al., 2015). In global SWS, when BF cholinergic firing is low, LFPs and unit firing are characterized by simultaneous slow waves and reduced firing rates in multiple cortical areas and in the hippocampus. Cholinergic inhibitors induce a similar LFP and cell firing state in the hippocampus and cortex of waking animals. Cholinergic agonists will switch the pattern from synchronized slow waves to asynchronous patterns characterizing wakefulness (Jones, 2005). Indeed, unihemispheric sleep in sea mammals, which is characterized by slow waves in only one hemisphere, shows differences in ACh levels between hemispheres, with the synchronous “sleeping” side showing low ACh levels and the asynchronous “awake” side showing high ACh levels (Lapierre et al., 2007). Even within a hemisphere during waking, specific areas may individually display signs of local sleep (Vyazovkiy et al., 2011). Evidence exists that hippocampal theta activity during REM sleep is characterized by especially high ACh levels (Marrosu et al., 1995). The hippocampus and neocortex can display simultaneously starkly different cholinergically controlled EEG/LFP activity patterns (Emrick et al., 2016; Duran et al., 2018), suggesting that separate functional populations of BF neurons project to these areas that are temporally coordinated (see also Teles-Grilo Ruivo et al., 2017). During waking, increased gamma power, enhanced sensory responsiveness, and decreased low-frequency oscillations in the cortex are often paralleled with movement or whisking (Niell and Stryker, 2010; Eggermann and Feldmeyer, 2009; Lee et al., 2014; Harrison et al., 2016; Nelson and Mooney, 2016).
Analysis of population single-unit spiking dynamics in V2 and OFC (mentioned above), allowed segregation of states into three groups: local irregular and global UP and DOWN states (Gombkoto et al., 2018). Cholinergic and noncholinergic neurons of the BF showed either increased modulation or decreased modulation according to cortical state or BF activity ignored the cortical state change. Figure 2 depicts a cholinergic neuron (BF-ID52) that ignores UP/DOWN states in V2 but is synchronized to UP/DOWN states to OFC neurons (OFC-ID64). These functional connections may be based on anatomical projections between specific cholinergic neurons and distinct cortical regions.
A, B, Multisite recording of field and unit activity in the awake rat from BF (green), OFC (red), and V2 (blue) displaying high-amplitude slow fluctuations: UP and DOWN states. Global DOWN states were localized from all structures (yellow areas) and separated from the local DOWN state of V2 (A, blue area) and from OFC (B, red area). Stars above spikes denote cholinergic firing from BF. C, Peristimulus time histogram (PSTH) of neurons from V2 (first column), OFC (second column), and BF (third column) triggered (red dashed lines) by the onset of the V2 DOWN state. Subplots at upper right corner of each of PSTH shows the spike waveform from silicon electrode arrays. D, PSTH of the same neurons from V1, OFC, and BF triggered by the onset of the DOWN state from OFC locally.
The mechanism linking cortical arousal, movement-related activity and pupil microdilations remains unexplained but may in part be due to the coordination of the cholinergic system with the noradrenergic system (Reimer et al., 2016; Larsen and Waters, 2018) and the presence of specific brainstem inputs to cholinergic neurons. For example, PFC projecting cholinergic neurons (Gielow and Záborszky, 2017) seem to receive information about pupil diameter and reflex gaze coordination and can broadcast this information to frontal cortex, potentially to modulate attention. Input to cholinergic neurons in the pedunculopontine tegmental, cuneiform, and parabrachial nuclei, largely corresponding to the mesencephalic locomotor region, are good candidates to convey fast movement-related information that accompany cortical membrane desynchronization and arousal (Kaur et al., 2013; Bennett et al., 2014; Lee et al., 2014; Nelson and Mooney, 2016).
In contrast to movement related states, reflective states in humans are said to rely on the default mode network (DMN), supporting functions, including memory, consciousness, and self-reflection (Gusnard et al., 2001; Buckner et al., 2008; Christoff et al., 2009), with similar circuits also described in monkeys and rodents (Lu et al., 2012). Procholinergic drugs (in which systemic administration limits the interpretation) suppress activity in regions that overlap with the DMN (Bentley et al., 2011). Cholinergic stimulation increases task-related activity in dorsolateral frontal and posterior parietal regions, suggesting that the BF shifts processing states from the DMN to those regions that support processing of external events. The association of the approximate human homolog of the medial septum-diagonal band to activation of the DMN (Yuan et al., 2018) might reflect the anatomical connection between cholinergic neurons in this compartment and the hippocampus, cingulate cortex, and precuneus. Recently, Nair et al. (2018) observed increased gamma activity in the BF of rats during quiet wakefulness/grooming and suppression of this activity during active exploration of an unfamiliar environment. The investigators suggest that the changing gamma band activity during these behaviors reflects a role for the BF projections to the anterior cingulate cortex (ACC) in switching between internal (DMN) and external events.
Mapping the cholinergic engram of fear and anxiety
ACh plays a crucial part in the formation of fear memories and might contribute to anxiety-like behaviors. The central nucleus of the amygdala (CeA) registers unexpected events in the environment via its projections to cholinergic neurons in the BF that are ultimately essential to an animal's ability to increment attention to unexpected events (Chiba et al., 1995; Gallagher and Chiba, 1996; Avery et al., 2012). Through this relay, emotionally salient events in the environment can quickly affect cortical learning (Baxter and Chiba, 1999) via the cholinergic enhancement of LTP and LTD cited above. Also timed with such events are changes in the ensemble dynamics of the BF (Quinn et al., 2010; Tingley et al., 2015) that may affect the basolateral amygdala (BLA), eliciting changes in intrinsic dynamics and excitability (for review, see Knox, 2016).
Neurons within the BLA are activated during fear behaviors and threat processing (Reijmers et al., 2007; LeDoux, 2012; Amir et al., 2015). BLA neuronal ensembles activated by recall of a fear memory become part of a fear engram (Nonaka et al., 2014); that is, a subset of previously active neurons engaged during learning are reengaged during fear recall. Moreover, direct stimulation of threat-induced, engram-enrolled BLA neurons can elicit fear behavior (Liu et al., 2012; Redondo et al., 2014). The mechanisms that underlie the increased activity of these neuronal ensembles include both changes in intrinsic excitability of neurons and synaptic plasticity (Byrne et al., 1991; Nonaka et al., 2014; Sehgal et al., 2014; Yiu et al., 2014).
BF cholinergic input to the BLA (Carlsen et al., 1985) modulates plasticity of cortico-amygdala synapses and is critical for mediating proper acquisition, recall, and extinction of threat memories (Jiang et al., 2013, 2016). These findings raise the question of whether a coordinated cholinergic input to the BLA constitutes a cholinergic engram that is essential to the acquisition of conditioned fear (M Ananth, P Rajebhosale, L Jiang, G Lopez-Hernandez, S Wang, N Desai, A Jone, L Role & D Talmage, unpublished data). This question was addressed using a conditional and inducible gene expression system, allowing examination of the activity of an early immediate gene, cFos, which is associated with long-term adaptive changes within cholinergic neurons, by infecting the BF of cFos-tTA, cFos-shGFP, and ChAT-IRES-Cre triple transgenic mice with a Cre-dependent virus: AAV-TRE-DIO-mCherry-2a-mut.tTA (Reijmers et al., 2007; Liu et al., 2012). This inducible system allows for permanent marking of active cholinergic neurons during specific time periods (Fig. 3) through viral expression. Infected cholinergic neurons that were active during the first behavior (e.g., fear acquisition) expressed mCherry and those that were reactivated by a previously conditioned stimulus (tone-associated recall) were colabeled with GFP. Electrophysiological examination of these cFos-positive cholinergic neurons revealed increased excitability following threat recall compared with cholinergic neurons that were not activated by threat recall. In other experiments, cholinergic-specific, activity-dependent tags allowed quantification of specific populations of BF cholinergic neurons activated by threat. These data support the existence of a modulatory engram for conditioned fear learning within the cholinergic BF.
BF cholinergic neurons are activated during fear conditioning. A, Transmitted light image showing regions profiled in B and C at bregma at 1.34 mm. EP, Entopeduncular nucleus; ic, internal capsule; GP, globus pallidus. B, AAV-TRE-mCherry-2a-tTA* injected in the nucleus basalis area of a ChAT-tauGFP × cFos-tTA mouse. Doxycycline transiently removed from the diet opens a labeling window during fear conditioning. White dotted inset delineates region magnified in C. C, Activated cholinergic neurons coexpress mCherry (red) and GFP (green) (yellow arrows). Cholinergic neurons are labeled by GFP. All neurons activated during fear conditioning are labeled by mCherry.
Cue-dependent (associative) fear recall is only one aspect of threat-related behaviors. The other aspect involving sustained fear responses is thought to be nonassociative akin to anxiety in humans. It is clearly of great interest to discern circuit mechanisms that might differentiate these associative and nonassociative modules of threat response profiles (Liu et al., 2012; Felix-Ortiz et al., 2013; Kim et al., 2013; Adhikari et al., 2015; Jimenez et al., 2018). Brain regions involved in mediating anxiety-like behaviors in rodents and humans, such as the BLA, CeA, bed nucleus of the stria terminalis, and the ventral hippocampus, are intimately connected to the BF cholinergic system. Predator odors activate these regions within the rodent brain and lead to an anxiety-like behavioral response (Staples, 2010). Lesion and pharmacological studies have shown the influence of cholinergic signaling in the BLA on predator odor-induced freezing behavior (Power and McGaugh, 2002). Mapping the cholinergic activity during exposure to predator odor, identified specific and distinct subsets of activated BF cholinergic neurons, supporting their involvement in modulation of anxiety-related circuits (P Rajebhosale, D Talmage & W Role, unpublished data). Additionally, the number of cholinergic neurons activated during conditioned fear correlated with an animal's behavioral performance (M Ananth, P Rajebhosale, L Jiang, G Lopez-Hernandez, S Wang, N Desai, A Jone, L Role & D Talmage, unpublished data).
In sum, a combination of viral labeling strategies, rapid acquisition microscopy with high-resolution data extraction (Boor Boor et al., 2018), and electrophysiology provides evidence that specific populations of cholinergic neurons are active during threat-related behaviors and recall (M Ananth, P Rajebhosale, Jiang L, G Lopez-Hernandez, S Wang, N Desai, A Jone, L Role & D Talmage, unpublished data); that recall-activated cholinergic neurons are more electrically excitable than cholinergic neurons not engaged in recall; that inhibition of cholinergic signaling in the BLA -via either optogenetic or chemogenetic approaches during fear acquisition disrupts recall-induced activation of the BLA (Jiang et al., 2016; M Ananth, P Rajebhosale, L Jiang, G Lopez-Hernandez, S Wang, N Desai, A Jone, L Role & D Talmage, unpublished data); and that distinct populations of cholinergic neurons may be critically involved in anxiety-related circuits (P Rajebhosale, D Talmage & L Role, unpublished data).
Cholinergic neuromodulation affects cellular, synaptic, network, and cognitive functions in the cortex
ACh released from BF axons to the cortex during waking causes cellular effects that appear to enhance memory encoding and attention (Hasselmo, 2006) by enhancing the influence of sensory afferent inputs on cortical activity while reducing the influence of internal corticocortical connections, thereby reducing the internal cortical dynamics associated with memory retrieval (Hasselmo and Bower, 1992; Hasselmo et al., 1995; Hasselmo and Cekic, 1996; Hasselmo, 1999, 2006; Eggermann and Feldmeyer, 2009). Cholinergic enhancement of afferent input is mediated by nicotinic ACh receptors at thalamic axonal inputs to cortex (Vidal and Changeux, 1993; Gil et al., 1997; Metherate and Hsieh, 2003; Disney et al., 2007; Kruglikov and Rudy, 2008; Picciotto et al., 2012). The neuronal spiking response is further enhanced by muscarinic depolarization of cortical neurons (Krnjević et al., 1971; Cole and Nicoll, 1984) and reduction of spike frequency accommodation (Madison et al., 1987; Schwindt et al., 1988) that could enhance memory encoding and attention to sensory afferent input. ACh also causes a strong presynaptic inhibition of excitatory recurrent connections within cortical structures including hippocampus (Valentino and Dingledine, 1981; Herreras et al., 1988; Hasselmo et al., 1995; Vogt and Regehr, 2001), piriform cortex (Hasselmo and Bower, 1992), and neocortex (Bröcher et al., 1992; Hasselmo and Cekic, 1996; Hsieh et al., 2000; Eggermann and Feldmeyer, 2009), which could reduce interference from previously encoded memories. Other modulators such as norepinephrine have also been shown to depolarize neurons and regulate synaptic transmission and plasticity, which could underlie the evidence for noradrenergic enhancement of attention and memory function (Hasselmo et al., 1997). Other neurotransmitter systems are active under different sets of conditions and states and do not cause the same neurophysiological activity pattern, so the BF cholinergic system, with its complex anatomy, is a unique system without which attention, memory, sleep patterns, and a host of other functions would be compromised.
Network effects can also be explained by cholinergic modulation of a canonical microcircuit consisting of PV, SOM, and VIP interneurons for cortical inhibition and disinhibition of principal cell activity (Fu et al., 2014; Bell et al., 2015a; Schmitz and Duncan, 2018; Fig. 4). Nicotinic input depolarizes interneurons expressing VIP, which could disinhibit cortical circuits by inhibiting interneurons that express PV and SOM (Fu et al., 2014; Bell et al., 2015b). In contrast, SOM-positive interneurons are depolarized by muscarinic ACh receptors and SOM- and PV-positive interneurons directly inhibit excitatory cells (Kuchibhotla et al., 2017; Young et al., 2017; Desikan et al., 2018). As noted above, cholinergic modulation also affects synaptic plasticity by: enhancing induction of long-term potentiation (Blitzer et al., 1990; Bröcher et al., 1992; Patil et al., 1998), inducing LTD (Williams and Johnston, 1990; Kirkwood et al., 1999; Jo et al., 2010), and biasing spike-timing-dependent plasticity toward LTD (Seol et al., 2007; Brzosko et al., 2017).
Schematic drawing of cholinergic modulation of a cortical microcircuit. ChAT-positive cholinergic neurons in the BF modulate excitatory glutamatergic PYR cells and three types of inhibitory GABAergic interneurons, VIP-positive, SOM-positive, and PV-positive interneurons. Whereas PV and SOM neurons directly inhibit pyramidal neurons, VIP neurons inhibit SOM and PV neurons, resulting in disinhibition of pyramidal cell activity. Cholinergic modulation acts via nicotinic receptor (N, red) to enhance thalamic input and depolarize VIP neurons and via muscarinic (M, green) ACh receptors to depolarize pyramidal cells and PV and SOM interneurons and cause presynaptic inhibition at excitatory and inhibitory feedback synapses.
Modeling demonstrates how ACh sets appropriate dynamics for encoding new memories without interference from retrieval of previous memories (Hasselmo, 1995, 2006), enhances the signal-to-noise ratio in cortical circuits (Patil and Hasselmo, 1999), and enhances the representation of information by desynchronization and decorrelation of neuronal activity (Pinto et al., 2013; Chen et al., 2015; Minces et al., 2017). For example, modeling shows how cholinergic depolarization of interneurons could enhance GABAergic inhibition of spontaneous background activity, whereas the depolarization of pyramidal cells, suppression of spike frequency accommodation, and presynaptic inhibition of feedback GABAergic transmission can enhance the response to afferent sensory signals (Patil and Hasselmo, 1999). Selective optogenetic stimulation of BF cholinergic neurons increases the signal-to-noise ratio in cortex (Minces et al., 2017). Further computational analyses revealed that neurons achieve this by slightly increasing the signal and generally decreasing the dependency commonly observed between signal and noise correlations. Evidently, this change in the correlation structure of visual cortical neurons increases the encoding capacity of the network (Minces et al., 2017).
Cholinergic muscarinic presynaptic inhibition of excitatory recurrent synapses in the neocortex (Hasselmo and Cekic, 1996; Gil et al., 1997) could account for reduced correlation of neuronal responses during cholinergic input (Goard and Dan, 2009; Pinto et al., 2013) and reduced extent of spatial integration in visual cortex (Roberts et al., 2005; Silver et al., 2008). The muscarinic enhancement of spiking response by depolarization and reduction of spike frequency accommodation could underlie the enhanced response to attended stimuli (Herrero et al., 2008) and enhanced perceptual performance (Pinto et al., 2013).
During active waking and REM sleep, the hippocampus exhibits prominent theta-rhythm oscillations (4–10 Hz) associated with higher measured levels of ACh (Marrosu et al., 1995; Monmaur et al., 1997; Zhang et al., 2010; Gu et al., 2017). The cholinergic enhancement of theta rhythm (Dannenberg et al., 2015) may contribute to encoding, given data showing correlation of encoding with larger theta amplitude. Phase-amplitude coupling of oscillations with theta rhythm may also be important for encoding of new memories (Hasselmo et al., 2002, Colgin et al., 2009) and muscarinic antagonists impair this theta-gamma coupling (Newman et al., 2014). BF cholinergic neurons excite BF GABAergic projection neurons (Yang et al., 2014; Dannenberg et al., 2015) that promote wakefulness (Zant et al., 2016) and theta oscillations. BF oscillatory activity in the theta band can also serve to organize spiking activity and other oscillatory frequencies during behavior, conveying precisely organized information to the cortex (Tingley et al., 2018).
During slow-wave sleep, lower levels of ACh (Marrosu et al., 1995) result in a change in network dynamics relative to waking. The dominant influence of afferent input during high levels of ACh is replaced by lower presynaptic inhibition of excitatory feedback and stronger excitatory feedback potentials (Winson and Abzug, 1977), perhaps underlying the increase of sharp-wave ripples in hippocampus (Buzsáki et al., 1983; Vandecasteele et al., 2014). This enhancement of feedback excitation during low levels of ACh in SWS have been proposed to allow dynamics for consolidation (Buzsáki, 1989; Wilson and McNaughton, 1994; Hasselmo, 1999) and increased ACh during sleep has been shown to impair consolidation (Gais and Born, 2004).
Blockade of ACh receptors has been shown to break down of the spatial specificity of place cells (Brazhnik et al., 2003) and grid cells (Newman et al., 2014), possibly linked to increased excitatory feedback or loss of intrinsic modulation (Heys et al., 2010). This could contribute to the loss of grid cell spatial coding during inactivation of medial septum (Brandon et al., 2011; Koenig et al., 2011). Furthermore, ACh contributes to path integration by facilitating a consistent representation of directionality in head direction cells (Yoder et al., 2017). In summary, the cellular effects of ACh via presynaptic inhibition of excitatory feedback and enhancement of the spiking response to afferent input could contribute to the enhancement of both memory encoding and attention (Hasselmo, 2006).
Consistent with the network effects of cholinergic modulation described here, systemic injection of muscarinic cholinergic antagonists in humans impair sustained attention performance (Wesnes and Warburton, 1984) and impair the encoding of words into memory while sparing retrieval (Ghoneim and Mewaldt, 1977; Peterson, 1977). Learning of new motor actions and motor maps is substantially impaired without BF cholinergic input to cortex (Conner et al., 2003, 2005). Selective manipulation of cholinergic neurons in animal studies causes impairments of sustained attention (McGaughy and Sarter, 1998; Sarter et al., 2005) and ACh release increases with attentional demand (Himmelheber et al., 2000). Selective lesions of cholinergic neurons cause stronger attentional impairments (Chiba et al., 1995, 1999; Fox et al., 2003) than memory impairments (Baxter et al., 1996), but hippocampal memory may involve parallel systems because combined lesions of cholinergic and GABAergic inputs cause stronger effects on memory (Pang et al., 2001) and lesions of cholinergic neurons impair association of objects with spatial location (Cai et al., 2012; Easton and Eacott, 2013).
Cholinergic system in aging and dementia
Large-scale recordings of the BF indicate that specific cell assemblies of the BF track the ongoing behavioral activities of an animal (Tingley et al., 2014, 2015), organizing cell firing activity according to LFPs that correspond differentially to different stimuli and feedback from the environment (Quinn et al., 2010; Tingley et al., 2015, 2018). This underscores the importance of BF ensemble activity in maintaining fluid activity in a changing environment and adapting behavior accordingly.
In AD, cholinergic neuron number is reduced and, not surprisingly, patients with mild cognitive impairment (MCI) to early AD can be distinguished from typical older adults based on their visual scanning and selective attention (Sahakian et al., 1989). Interestingly, both reduced activation of DMN (Sorg et al., 2007; Buckner et al., 2008; Palop and Mucke, 2010) and loss of neurons in basal nucleus of Meynert (Iraizoz et al., 1991; Lehéricy et al., 1993) were reported in AD and the rostral (septal) volume of the cholinergic space shows a reduction in MCI patients that is positively correlated with hippocampal atrophy (Cantero et al., 2017). This also aligns with early changes in the inhibitory networks of the hippocampus that emerge during MCI (Bakker et al., 2012). Interestingly, the cholinergic space shows reduction even during normal aging in humans (Záborszky et al., 2008, 2015c).
Therefore, just as efficient operation of the BF supports functionality, as the network degrades, problems in those areas emerge in the form of an impaired ability to keep track of ongoing behavioral activities, the inability to increment behavior to respond to unexpected events, an inability to effectively decrement attention to irrelevant events, a failure of efficient information transfer to cortex in support of effective sensory coding and new learning, problems with sleep and memory consolidation, and a lack of fluid switching between encoding and retrieval, leading to disorientation, confusion, and lapses of memory and awareness (for review, see Albers et al., 2015; Bondi et al., 2017).
Future directions
Future studies should investigate whether the putative monosynaptic or oligosynaptic BF ensembles change dynamically during specific behavioral epochs and how firing patterns of selected neurons in the BF affect local cortical circuit dynamics in particular layers. The ability to systematically manipulate the constituent inputs to the BF might also lead to a fuller understanding of the circuit dynamics that allow for the ensemble behavior of the BF. For example, these manipulations could elucidate the potential role of cholinergic neurons in regulating the coding of specific dimensions relevant to behavior and the specificity and timing of the regulation of attention, encoding, and retrieval and consolidation. Future studies can clarify the individual and interactive role of the collective function of the cholinergic neurons in providing a final common pathway to cortical activation and function.
Footnotes
This work was funded in part by the National Institutes of Health (Grant R01 MH060013 to M.H.; Grant R01 MH061492 to M.H.; Grant R01 MH060670 to G.P.; National Institute of Neurological Disorders and Stroke Grant (NS023945) to L.Z.); the Office of Naval Research (Grant MURI N00014-16-1-2832 and Grant DURIP N00014-17-1-2304 to M.H.); the Deutsche Forschungsgemeinschaft (Grant 322014644 to H.D.); the National Science Foundation (Grants SMA 1041755 and IIS 0910485 to A.A.C.); R01 NS022061 to L.W.R; and U01, MH109014 to L.W.R., D.A.T. Helping with the revision by Dr. Candice Chavez is greatly appreciated.
The authors declare no competing financial interests.
- Correspondence should be addressed to Laszlo Záborszky, Center for Molecular and Behavioral Neuroscience, Rutgers University, 197 University Avenue, Newark, NJ 07102. laszloz{at}newark.rutgers.edu
