Cell type-specific optogenetic techniques reveal neural circuits crucial for episodic memories

Abstract

The formation and maintenance of episodic memories is important for our daily life. Accumulating evidence from extensive studies with pharmacological, electrophysiological and molecular biological approaches has shown that both entorhinal cortex (EC) and hippocampus (HPC) are crucial for the formation and recall of episodic memory. However, to further understand the neural mechanisms of episodic memory processes in the EC-HPC network, cell type-specific manipulation of neural activity with high temporal resolution during memory process has become necessary. Recently, the technological innovation of optogenetics combined with pharmacological, molecular biological and electrophysiological approaches have significantly advanced our understanding of the circuit mechanisms for learning and memory. Optogenetic techniques with transgenic mice and/or viral vectors enable us to manipulate the neural activity of specific cell populations as well as specific neural projections with millisecond-scale temporal control during animal behavior. Integrating optogenetics with drug-regulatable activity-dependent gene expression systems has identified memory engram cells, which are a sub-population of cells that encode a specific episode. Finally, millisecond pulse stimulation of neural activity by optogenetics has further achieved i) identification of synaptic connectivity between targeted pairs of neural populations, ii) cell-type specific single unit electrophysiological recordings, and iii) artificial induction and modification of synaptic plasticity in targeted synapses. In this chapter, we summarize technological and conceptual advancements in the field of neurobiology of learning and memory as revealed by optogenetic approaches in the rodent EC-HPC network for episodic memories.

1, Introduction

In the mammalian brain, massive amounts of multisensory stimulation are combined and stored as a series of experienced events for acquiring and remembering episodes. Learning and memory process affect a wide variety of adaptive behaviors and contribute to higher cognitive function. In humans and rodents, the entorhinal cortical (EC) -hippocampal (HPC) circuit is required for the formation and recall of episodic memory, which extends into the spatial/contextual, object/individual and temporal domains (Eichenbaum 2000; Tulving 2002). Episodic memory can be formed as an integration between multiple elements individually remembered and linked to form a singular memory unit (Fortin et al. 2002; Naya and Suzuki 2011; MacDonald et al. 2011; MacDonald et al. 2013; Kitamura et al. 2014; Sakon et al. 2014; Allen et al. 2016; Aronov et al. 2017). These individual components include the “what” in the form of objects, the “where” representing spatial orientation, navigation, and context (Moser et al. 2008; Smith and Bulkin 2014), the “when” indicated by timing sequencing of events and linking of events across time (Kitamura et al. 2014), and the “who”, or social memory involving recall of individual conspecifics and relationships (Allsop et al. 2014). Furthermore, while memory acquisition initially requires the EC-HPC network, the acquired memory is gradually consolidated in neocortical networks for permanent storage (Frankland and Bontempi 2005; Tonegawa et al. 2018). To further understand the neural mechanisms by which the brain stores each component (what, where, who, and when) of episodic memories in the EC-HPC network and how each component is integrated to complete an episodic memory, we need to understand neural circuits and their activity during the formation, recall and consolidation (Box1) of episodic memories.

Box1: Memory consolidation.

The process of transforming a newly acquired memory into a more long-lasting stable state is referred as memory consolidation (Squire 1986; Eichenbaum 2000). Several theories have attempted to explain the neurobiological mechanisms that underlie this process. The most commonly accepted theory is the Standard Consolidation Theory (SCT). Beginning with Scoville and Milner’s study of patients with hippocampal lesions (Scoville and Milner 1957), many studies of memory-impaired patients and animals have provided strong evidence that our memories are initially stored within HPC and, with the passage of time, slowly consolidated in extra-hippocampal (neocortical) structures (Squire 1986; Zola-Morgan and Squire 1990; Kim and Fanselow 1992). For example, rats receiving electrolytic lesions to HPC one day after a contextual fear memory task do not retain the memory, whereas rats receiving a lesion 4 weeks later retain the memory (Kim and Fanselow 1992). However, there are conflicting reports about the role of HPC on the remote memories (Squire 1992; Nadel and Moscovitch 1997; Bayley et al. 2005). For example, HPC damage not only disrupts recently acquired memory but also impairs recall of long-term memory in human studies (Nadel and Moscovitch 1997). These contrary pieces of evidence have led to a new theoretical formulation; the Multiple Trace Theory (MTT). According to MTT, reactivation of a memory creates additional memory traces in HPC, and these multiple traces of a particular memory give greater robustness against partial HPC disruption (Nadel and Moscovitch 1997; Winocur and Moscovitch 2011). These conflicts had not been resolved until the emergence of new technologies enabled examination of the functionality of memory traces of a particular memory in a specific area (Kitamura et al. 2017).

In 2005 a novel technology termed optogenetics was invented. The light responsive cation channel, channelrhdopsin-2 (ChR2), chloride and proton pumps such as halorhodopsin (NpHR) and archaerhodopsin (Arch) have revolutionized the way that neuroscientists can interrogate the roles of neural circuits on learning and memory process by temporally and reversibly manipulating neural activity during animal behavior (Fig. 1b) (Boyden et al. 2005; Zhang et al. 2006; Chow et al. 2010; Yizhar et al. 2011). These opsin proteins can be expressed in selective brain regions by injection of viral vectors under the control of cell type-specific promoters (Sohal et al. 2009; Stuber et al. 2011). Furthermore, transgenic animals expressing Cre-recombinase in a specific cell population with the virus-mediated delivery of vectors encoding a Cre-inducible opsin allows us to manipulate neural activity in specific cell populations in the brain (Fig. 1a) (Cardin et al. 2010; Cohen et al. 2012; Kitamura et al. 2014). Retrograde transport of replication-deficient rabies virus (Wickersham et al. 2007; Chatterjee et al. 2018) and Adeno-associated viruses (AAVs) (Tervo et al. 2016) can be utilized for selective labeling of a defined neural population projecting to a virus-injected area (Fig. 1a) (Zhang et al. 2013; Kitamura et al. 2015b). Selective light delivery to axon terminals can be also used as an projection-specific neural activity control (Fig. 1a) (Tye et al. 2011). Activity-dependent cell labeling with optogenetic stimulation has identified memory engram cells, which are a sub-population of cells that encode a specific episode (Box2; Fig. 1c[iv]) (Reijmers et al. 2007; Liu et al. 2012). Finally, millisecond-timescale control of neural stimulation by optogenetics enables us to examine i) identification of synaptic connectivity between targeted pairs of neurons, ii) cell type-specific single unit electrophysiological activity and iii) artificial induction and modification of synaptic plasticity in targeted-synapses (Fig. 1c). In this chapter, we focus on recent advances in our understanding of neural circuits and the neural processes underlying the formation, recall and consolidation of episodic memory with emphasis on cell type-specific optogenetic approaches.

Figure 1. Advantages of optogenetic stimulation for learning and memory studies.

(a) Left panel; Population-specific opsin expression enables population-specific manipulation of neural activity by local light delivery. Each area surrounded by a dotted line represents a brain region. Red area represents a virus-injected brain area. Middle panel; Projection-specific light illumination enables selective manipulation of the neural activity at axon terminals (manipulating neural output). Right panel; Retrograde transport of virus enables labeling of specific neural population projecting to the virus-injected brain area. (b) Left panel; Specific neural activities in a defined period of behavioral paradigms can be optically manipulated. Figure is an example of the event-triggered optogenetic stimulation in a fear conditioning paradigm. Right panel; Neural-activity-triggered stimulation by light enables us to examine the functional roles of neural activities during the photo-stimulating period. Figure is an example of the burst-firing-triggered optogenetic stimulation and its control optogenetic stimulation. (c) Advantages of millisecond pulse stimulation of neural activity by optogenetics. i) Patch clamp recording with millisecond pulse stimulation by optogenetics enables examination of the synaptic connectivity between targeted pairs of neural populations (Petreanu et al. 2007). ii) in vivo electrophysiology with millisecond pulse stimulation to specific neural populations expressing opsin enables examination of cell type-specific single unit electrophysiological recording (Cohen et al. 2012). iii) Artificial induction and modification of synaptic plasticity in targeted-synapses by high-frequency light-pulse optogenetic stimulation (Nabavi et al. 2014). iv) Activity-dependent cell labeling with millisecond pulse stimulation by optogenetics enables induction of artificial memory recall by optogenetic activation of memory engram cells (Liu et al. 2012). (a-c) Red or gray neurons represent opsin-expressing or non-expressing cells. Blue tetragons or squares represent light delivered periods/areas.

Box2: Memory Engram.

In 1904, German scientist Richard Semon coined the term “engram” to conceptualize the physical substrate of memory, which he defined as “the enduring though primarily latent modification in the irritable substance produced by a stimulus“ (Semon 1904; Semon 1921). The term engram is equivalent to another commonly used term, “memory trace”. In order to be considered as such, an engram must meet three criteria: it is an enduring physical and/or chemical change that occurs in a neural network in the brain (criterion 1) as a result of activation of neuronal subpopulations by episodic stimuli (criterion 2) and can be subsequently reactivated by stimuli that were part of the original set of encoded stimuli, resulting in the recall of the original memory (criterion 3). The current conception is that individual neurons act as the substrate of engram formation, becoming “engram cells” (Josselyn 2010; Tonegawa et al. 2015). Over a century later, rapid development of molecular genetics and biological technologies allows us to examine Semon’s theory by labeling specific neuronal subpopulations activated during defined periods and manipulating or recording their activities (Reijmers et al. 2007; Han et al. 2009; Liu et al. 2012). Reijmers et al found that a subpopulation of neurons activated during the acquisition of a fear memory are preferentially reactivated during the recall of that memory in the amygdala of the mouse brain by examining the expression of the immediate early gene (IEG), c-Fos (Reijmers et al. 2007). Han et al found that a subpopulation of neurons artificially overexpressing the cAMP response element-binding (CREB) protein were preferentially activated during a subsequent contextual fear conditioning training, and became a part of the putative array of engram cells (Han et al. 2007). They also demonstrated that selective ablation of these putative engram cells impairs the recall of the associated fear memory in mice, but ablation of a random population of neurons in the same region does not (Han et al. 2009). Finally, Liu et al showed that optogenetic reactivation of a subpopulation of dentate gyrus granule cells which were previously activated during the acquisition of a contextual fear memory is sufficient to induce memory recall (Liu et al. 2012). This phenomenon, identified in the dentate gyrus granule cells, which stores a contextual fear memory meets all three of the criteria for a true engram (Josselyn 2010; Tonegawa et al. 2015).

2, Identification of novel hippocampal neural circuit by optogenetics

EC-HPC neuronal networks contain two major feed-forward excitatory pathways. One is called the hippocampal tri-synaptic pathway; it runs from EC layer II (EC2) to granule cells in the dentate gyrus (DG) to CA3 and to CA1 pyramidal cells, and the other pathway is called the hippocampal direct pathway from EC layer III (EC3) to CA1 pyramidal cells (Fig. 2) (Amaral and Witter 1989; Witter et al. 2000; Nakazawa et al. 2003). In both pathways, glutamatergic inputs from EC eventually converge onto the CA1 region. The CA1 directly projects to other neocortical areas or indirectly connects through subiculum and EC layer V (EC5) to various brain structures (Steward and Scoville 1976; Amaral and Witter 1989). Patch-clamp recordings with electrical stimulation in an acute brain slice have identified many synaptic connections in the EC-HPC networks (Andersen et al. 1966; Winson and Abzug 1977; McNaughton and Barnes 1977; Empson and Heinemann 1995; Scharfman 1995; Chevaleyre and Siegelbaum 2010a; Masurkar et al. 2017; Hashimotodani et al. 2017). However, electrical stimulation of afferent fibers in acute brain slices may activate any afferent fibers passing through the area; there is no specificity of stimulation to distinguish individual components present within the tract. Expressing ChR2 in a specific neural population allows us to optically stimulate targeted-afferent fibers expressing ChR2 in the brain (Fig. 1a). Using patch clamp recording paired with optogenetic light-pulse stimulation of specific neural fibers, it has become possible to assess the existence and/or strength of synaptic connections between targeted pairs of neurons (Fig. 1c[i]) (Petreanu et al. 2007). For example, using cell type-specific Cre transgenic mouse lines, Kohara et al. discovered a new mono-synaptic pathway from DG cells to CA2 pyramidal cells through abundant longitudinal projections. They expressed ChR2 in DG granule cells and conducted patch clamp recordings in CA2 neurons with optogenetic pulse stimulation of DG granule cells. This resulted in the observation of mono-synaptic responses from CA2 neurons after a brief pulse of blue light (460nm) was delivered to the brain slice (Kohara et al. 2014), indicating that DG granule cells directly project to CA2 pyramidal neurons (Fig. 2). In addition, contrary to previous observations made with electrical stimulation (Chevaleyre and Siegelbaum 2010b; Jones and McHugh 2011), Kohara et al. observed there is no direct input from EC3 to CA2 (Kohara et al. 2014), indicating that CA2 mainly receives input coming from the tri-synaptic pathway (Fig. 2). In 2014, Kitamura et al. discovered a novel direct projection from Wolfram syndrome 1 (Wfs1) positive pyramidal cells (Wfs1⁺ cells) in EC2 to GABAergic interneurons in stratum lacunosum (SL-INs) of hippocampal CA1 (Fig. 2) (Kitamura et al. 2014). Optogenetic terminal stimulation of ChR2-expressing Wfs1⁺ cells in EC2 with in vitro patch-clamp recordings revealed a monosynaptic glutamatergic input onto SL-INs. Furthermore, simultaneous recordings of CA1 pyramidal cells and SL-INs paired with optogenetic axonal stimulation from EC3 cells demonstrated the existence of a novel gating circuit in CA1 controlled by the Wfs1⁺ cells of EC2 (Kitamura et al. 2014). Similarly, Basu et al. found long-range projecting GABAergic interneurons in EC, which mediate gating mechanisms in CA1 (Basu et al. 2016). Utilizing cell type-specific optogenetic manipulation, Leão et al. discovered that hippocampal oriens lacunosum-moleculare cells differentially modulate inputs coming from CA3 and EC3/EC2 (Leão et al. 2012). These results demonstrate that patch-clamp electrophysiology combined with the millisecond pulse stimulation of neural activity with optogenetics enables the discovery of novel neural circuits, even in the well-known EC-HPC network.

Figure 2. EC-HPC neural circuits for episodic memory.

EC2 contains two populations of excitatory neurons, Reelin⁺ stellate cells and Wfs1⁺ pyramidal cells. Reelin+ cells project to hippocampal DG, CA3, and CA2, while Wfs1⁺ cells project to GABAergic interneurons in SL (SL-INs) of hippocampal CA1 region. Pyramidal cells in EC3 directly project to CA1 pyramidal cells. Hippocampal DG granule cells project to CA3 and CA2. Hippocampal CA3 project to CA2 and CA1. CA2 projects to CA1. CA1 projects to subiculum as well as EC5. Reelin⁺ stellate cells in EC2 drive contextual information to hippocampal DG and CA3 circuits to form contextual memory (blue). In the tri-synaptic circuits, DG contributes to the pattern separation process, while CA3 contributes to the pattern completion. Temporal association signals from EC3 cells to CA1 contribute to temporal association learning (green). On the other hand, Wfs1⁺ pyramidal cells in EC2 project to GABAergic interneurons in CA1, which gate inputs from EC3 to CA1 (red), and inhibit temporal association learning While the pathway from CA1 to EC5 via dorsal subiculum is involved in the memory recall as well as update, the direct pathway from CA1 to EC5 is crucial for memory encoding (yellow).

3, Neural circuit mechanisms for episodic memory by optogenetics

3.1, Neural circuits for contextual memory

3.1.1, Loss-of-function analysis by optogenetics

Contextual memory is comprised of multimodal information including spatial cues, sound, odor, texture, and more (Spear 1973; Maren et al. 2013). A common behavioral test in animal studies is contextual fear conditioning (CFC), in which an animal is given aversive footshocks paired with a specific context (Pavlov 1927; Phillips and LeDoux 1992; Kim and Fanselow 1992; Kim et al. 1993; Frankland et al. 1998; Sutherland et al. 2008; Goshen et al. 2011; Maren et al. 2013; Kitamura et al. 2017). During the conditioning and recall of CFC, contextual information has been thought to proceed through the hippocampal tri-synaptic circuit from EC2 to DG to CA3 and then to CA1 (Marr 1971; O’Reilly and McClelland 1994; Treves and Rolls 1994; Dudchenko et al. 2000; Gilbert et al. 2001; Nakazawa et al. 2003; Leutgeb et al. 2004; Leutgeb et al. 2007; McHugh et al. 2007; Nakashiba et al. 2008; Nakashiba et al. 2009). Accumulating evidence from theoretical (Marr 1971; McNaughton and Morris 1987), anatomical (Marr 1971; O’Reilly and McClelland 1994; Treves and Rolls 1994), physiological (Leutgeb et al. 2004; Leutgeb et al. 2007; McHugh et al. 2007), and behavioral studies using loss-of-function animals (Dudchenko et al. 2000; Gilbert et al. 2001; Nakazawa et al. 2003; McHugh et al. 2007; Nakashiba et al. 2008; Nakashiba et al. 2009) indicates that the functional role of the EC2-DG-CA3 feed-forward excitatory pathway is pattern separation; a process allowing discrimination of overlapping inputs, and that the functional role of the recurrent connections in CA3 is pattern completion; a process for recall of complete memories on the basis of incomplete sets of cues (Box3). Furthermore, Kitamura et al. found that the Reelin⁺ stellate cells (Reelin⁺ cells) in EC2 show context-specific neural activity during exploration of distinct contexts (Kitamura et al. 2015b) by using in vivo calcium imaging, indicating that the entorhinal cortex drives contextual information to the hippocampal circuit. To examine the role of EC2 input into the DG (and subsequently CA3) on contextual memory, they expressed ArchT (Han et al. 2011) in the Reelin⁺ cells of EC2 by retrograde AAV infection from the axon terminals in DG, and inhibited the neural activity of Reelin⁺ cells in EC2 with green light illumination (561 nm) in EC during either conditioning or testing (Fig. 1b). They found that optogenetic inhibition of the Reelin⁺ cells during conditioning or testing in CFC impaired contextual fear memory and reduced context-specific neural activation in CA3 pyramidal cells (Kitamura et al. 2015b). These results indicate that contextual information is already encoded by the Reelin⁺ cells in EC2 and neural input from Reelin⁺ cells in EC2 to the DG and CA3 is crucial for the formation and recall of contextual fear memory.

Box3: Pattern separation and pattern completion.

As a memory system, HPC must be able to generate distinct patterns of activation of neural networks, even from highly similar cues, at the time of storage, and HPC must be able to retrieve the full stored patterns of activation of neural networks from noisy and partial cues at the time of recall. The process of transforming overlapping or similar inputs into more distinct outputs is called pattern separation. The process of reconstructing incomplete or partial sets of inputs into complete stored representations is called pattern completion. Earlier theoretical studies provide an insightful theoretical framework for associating functional properties of memory with the mechanisms of pattern separation and pattern completion (McNaughton and Morris 1987; O’Reilly and McClelland 1994). Based on the anatomical organization of the hippocampal tri-synaptic pathway, in which a large number of granule cells in DG project to a smaller number of recurrently connected CA3 cells and form powerful synaptic connection with large synapses located very close to the soma of CA3 cells, these studies predicted that the EC2-DG-CA3 feed-forward excitatory pathway works as the pattern separator, and the recurrent connections in CA3 works as the pattern completer (Marr 1971; McNaughton and Morris 1987; O’Reilly and McClelland 1994; Treves and Rolls 1994). These theories were supported by physiological observations a within each hippocampal subregion in awake, behaving animals (Leutgeb et al. 2004; Leutgeb et al. 2007; McHugh et al. 2007), and by the behavioral performance of mice with genetic/pharmacological inactivation or lesion of discrete components of the EC-HPC network (Dudchenko et al. 2000; Gilbert et al. 2001; Nakazawa et al. 2003; McHugh et al. 2007; Nakashiba et al. 2008; Nakashiba et al. 2009). For example, McHugh et al. examined the role of the granule cells in the DG using a transgenic mouse line in which synaptic plasticity in EC2-DG synapses was attenuated by the DG-specific conditional knockout of N-methyl-D-aspartate (NMDA) receptor NR1 They found that the mutant mice showed impairment in the discrimination of similar contexts and reduced the context-specific activities of CA3 pyramidal cells (McHugh et al. 2007).

While the EC2-DG-CA3 circuit processes contextual memory and sends it to hippocampal CA1 pyramidal cells, how does contextual information further proceed from hippocampal CA1 pyramidal cells to basolateral amygdala for the formation and recall of contextual fear memory? CA1 pyramidal cells in dorsal HPC directly project to EC5 as well as dorsal subiculum (dSub) (Amaral and Witter 1989). Excitatory neurons in dorsal subiculum project to EC5 and other cortical and subcortical brain regions including amygdala (Ding 2013) (Fig. 2). EC5 neurons have also extensive projections to the neocortex as well as basolateral amygdala (Sürmeli et al. 2015; Kitamura et al. 2017). Studies with functional magnetic resonance imaging of human subjects suggested that the DG and CA subfields are activated during episodic memory formation, whereas dSub is more active during memory recall (Gabrieli et al. 1997; Eldridge et al. 2005). On the other hand, in rodents, chemical lesions of the CA1 or dSub caused impairment in the acquisition of place navigation (Morris et al. 1990). However, given the close anatomical proximity between CA1 and dSub in rodents, lesions might be difficult to target to a specific hippocampal subregion. With transgenic mice expressing Cre recombinase in excitatory neurons of dSub or CA1 combined with the injection of Cre-dependent AAV expressing Arch, Roy et al. examined the role of dSub and CA1 neurons and their circuits in the formation and recall of CFC (Roy et al. 2017). First, optogenetic cell body inhibition of dSub neurons impaired the recall of CFC but not learning. Axonal terminal inhibition of dSub neurons at EC5 targeted by the implantation of optic fibers also impaired the recall of CFC but not learning. Similarly, optogenetic terminal inhibition of CA1 neurons at dSub inhibited the recall of CFC but not learning. In contrast, optogenetic terminal inhibition of CA1 neurons at EC5 during testing of CFC did not impair the recall of CFC memory, while optogenetic terminal inhibition of CA1 neurons at EC5 during CFC acquisition resulted in inability to recall CFC memory. In addition, selective optogenetic terminal inhibition of dSub terminals at EC5 impaired the updating of CFC memory (Roy et al. 2017). These results showed that the di-synaptic pathway from CA1 to EC5 via dSub is selectively required for memory recall and updating, whereas the direct pathway from CA1 to EC5 is essential for memory formation. Furthermore, by using optogenetic terminal inhibition of EC5 neurons at basolateral amygdala, Kitamura et al., showed that the EC5 to basolateral amygdala pathway is crucial for the formation and recall of CFC (Kitamura et al. 2017). In summary, cell type-specific and/or projection-specific optogenetic manipulation of neural activity revealed that Reelin⁺ cells in EC2 drive contextual information to HPC, and contextual memory is formed in EC2-DG-CA3 circuits (Fig. 2). However, after EC2-DG-CA3 circuits processed contextual memory to hippocampal CA1 pyramidal cells, two distinct neural pathways (CA1-dSub-EC5 pathway or CA1-EC5 pathway) to basolateral amygdala are differentially involved in the formation or recall of contextual fear memory.

3.1.2, Gain-of-function analysis by optogenetics

The previous section discussed the identification of specific neural circuits crucial for formation and recall of contextual fear memory. However, these studies did not provide direct evidence as to whether these circuits store memory traces for CFC. In the early 19^th century, Richard Semon proposed that our brains retain traces for a specific memory, which are generated by and endure through modification produced by a stimulus. These long-term memory traces are referred to as an engram (Box2). Until recently biological basis of a true engram as defined by Semon’s theory, a stably maintained neural subpopulation which is activated during acquisition of specific memory and reactivated during memory recall, had not been identified. In 2012, optogenetic technology combined with activity-dependent cell labeling using immediate early gene promoters to drive opsin expression in cells activated during an episode made it possible to identify engram cells for a specific episode (Reijmers et al. 2007; Liu et al. 2012). As a first demonstration of engram cells, Liu et al labeled hippocampal DG cells activated during CFC with ChR2 in doxycycline-dependent manner. They found that optogenetic reactivation of these cells in DG by 20 Hz pulse stimulation of blue light via optic fibers implanted into the brain is sufficient to induce memory recall without natural recall cues but optogenetic reactivation of cells labeled in a context not associated with fear did not (Fig. 1c[iv]) (Liu et al. 2012). These results demonstrate that artificial optogenetic reactivation of a selective neural subset activated during learning is sufficient for the recall of specific episodic memory, indicating a direct evidence for the existence of memory engram for CFC in the hippocampal DG cells. Subsequent studies also showed that memory engram cells can be formed in various brain regions for different types of memories, however, the optimal frequency of light pulse stimulation for artificial memory recall is different in each brain region, indicating that the temporal pattern of neural activity in engram cells is also important for memory recall (Fig. 1c[iv]) (Ramirez et al. 2013; Redondo et al. 2014; Cowansage et al. 2014; Ohkawa et al. 2015; Ryan et al. 2015; Okuyama et al. 2016; Kitamura et al. 2017; Abdou et al. 2018).

3.1.3, Optogenetic manipulation of synaptic plasticity

Since we now are able to optogenetically activate targeted-neural projections with high frequency light pulse stimulation (Fig. 1a) (Lin et al. 2009), it has become possible to artificially induce and modulate synaptic plasticity to directly address causality between long-term potentiation (LTP)- or long-term depression (LTD)-mediated synaptic plasticity (Box4) and memory expression (Fig. 1c[iii]). Nabavi et al. demonstrated that high-frequency (100 Hz) or low-frequency (1 Hz) light pulse stimulation of axon terminals expressing mammalian codon optimized ChiEF (oChIEF) (Lin et al. 2009; Lin et al. 2013), a ChR2 variant that can respond to high-frequency optical stimuli, induced LTP or LTD in the brain of behaving animals. By using the optical LTP/LTD protocol, they found that induction of optical LTD at synapses of auditory cortex neurons onto lateral amygdala after auditory fear conditioning disrupted the recall of auditory fear memory. Furthermore, induction of optical LTP restored the conditioned fear response which had been disrupted by optical LTD (Nabavi et al. 2014). By using optical LTD protocols combined with activity-dependent cell labeling in auditory cortex and amygdala, Abdou et al. also found that induction of optical LTP by stimulating axon terminals from auditory cortical neurons activated during tone fear conditioning disrupted the recall of the auditory fear memory, but did not affect other auditory fear memory associated with a different tone frequency, indicating the recall of two distinct auditory fear memories required synapse-specific activation (Abdou et al. 2018). These studies demonstrated that LTP/LTD-mediated synaptic plasticity directly links with memory formation and memory expression.

Box4: Long-term potentiation and long-term depression.

In 1949, Canadian psychologist Donald Hebb proposed an elegant model explaining how the function of neurons contributed to learning and memory processes (Hebb 1949). In this model, he postulated a synaptic modification for learning and memory that occurs as a consequence of coincidence between pre- and postsynaptic activity. However, experimental evidence showing plasticity of synapses in the mammalian brain predicted from the model was lacking. In 1966, Lomo found that brief, high-frequency electric stimulation of the perforant path fibers to the hippocampal DG caused a rapid and long-lasting increase in the strength of these synapses (Lomo 1966). This long-lasting strengthening of synapses and the response of a postsynaptic nerve cell to stimulation is called long-term potentiation (LTP) (Bliss and Lomo 1973). By contrast, prolonged, low-frequency electric stimulation of the Schaffer collateral to the hippocampal CA1 caused long-lasting decrease in the strength of these synapses (Dudek and Bear 1992). This long-lasting weakening of synapses and the response of a postsynaptic nerve cell to stimulation called long-term depression (LTD). This LTP/LTD-mediated synaptic plasticity is a prevailing cellular model for learning and memory.

3.1.4, Optogenetic tagging for in vivo electrophysiology

In vivo electrophysiology in behaving animals is useful to investigate the behavioral correlates of the activity patterns of individual neurons. The combination of optogenetics and in vivo electrophysiology with single unit recording has emerged as a versatile method for identifying specific neuronal populations within blind extracellular recordings in behaving animals by expressing ChR2 in a specific neuronal population and inferring the light-responsive cell population from its reaction to the stimulus, referred to as “optogenetic tagging” (Fig. 1c[ii]) (Cohen et al. 2012). By combining transgenic mice expressing Cre recombinase under the control of cell-specific promoter and in vivo electrophysiology, Cohen et al. labeled dopaminergic and GABAergic neurons in ventral tegmental area with ChR2, and then identified their cell-types based on their responses to optical stimulation during in vivo single unit recording (Cohen et al. 2012). Ciocchi et al. combined optogenetic tagging with optogenetic terminal stimulation of light-responsive cells, and found that the ventral CA1 routes anxiety-related information preferentially to the prefrontal cortex and goal-related information preferentially to the nucleus accumbens (Ciocchi et al. 2015). Tanaka et al. applied optogenetic tagging combined with activity-dependent cell labeling to examine the electrophysiological profile of memory engram cells in hippocampal CA1 (Tanaka et al. 2018). These studies demonstrated cell type-specific single unit electrophysiological recording by optogenetic in behaving animals.

3.2, Neural circuits for temporal association memory

Remembering the timing of distinct events and associating temporally discontinuous events are crucial processes for the formation of episodic memories. We refer to this aspect of memory encoding as temporal association learning to engage a diverse set of temporally segregated information as an episode. Great progress in our understanding of temporal association learning has been made by using classical Pavlovian conditioning (Solomon et al. 1986; Maren 2001). In animal studies, a trace fear conditioning paradigm has been widely used to assess the associative learning of temporally discontinuous events (Solomon et al. 1986; Moyer et al. 1990; McEchron et al. 1998; Suh et al. 2011). The training session of classical trace fear conditioning (TFC) consists of a 20 sec tone as conditioned stimulus (CS) followed by a 20 sec delay, after which a foot shock as unconditioned stimulus (US) is delivered to the animal subject. In test sessions, the extent of association of two experiences, CS and a shock, were assessed by duration of freezing after delivery of the tone without shock. A number of studies with lesions as well pharmacological and genetic manipulations showed that the EC-HPC network is necessary for establishing CS–US associations across the temporal gap (Solomon et al. 1986; Moyer et al. 1990; McEchron et al. 1998; Hasselmo and Stern 2006; Kitamura et al. 2015a; Kitamura 2017). Suh et al examined roles of the EC3-CA1 direct pathway for bridging two events across the temporal gap by using mutant mice in which glutamatergic synaptic transmission from EC3 cells to CA1 pyramidal cells is blocked by tetanus-toxin (TeTX) (Suh et al. 2011). They found that the mutant mice showed deficits in association of temporally discontinuous events in TFC but they associated concurrently experienced events. On the other hand, inhibition of the CA3-CA1 pathway with TeTX did not affect TFC. In 2014, Kitamura et al. optogenetically manipulated the direct pathway from EC3 to CA1 during TFC. They found that optogenetic inhibition of the pathway during conditioning (tone+trace+shock) caused deficits in TFC, while optogenetic activation of the direct pathway from EC3 to CA1 by ChR2 stimulation enhanced TFC, indicating that the EC3 to CA1 direct pathway during TFC is crucial for driving temporal association learning (Kitamura et al. 2014).

Like most cognitive and motor phenomena, temporal association should be regulated for optimal adaptive benefit in animals. In 2014, Kitamura et al. discovered a novel excitatory input from Wfs1⁺ cells in EC2 to GABAergic neurons in SL which gates excitatory EC3 input into CA1 pyramidal cells (Kitamura et al. 2014) as mentioned above (Section 2). They examined the role of the excitatory input from Wfs1⁺ cells in EC2 into CA1 on TFC, and found that optogenetic terminal activation of EC2 input in CA1 during the entire training period and the delay plus foot shock period inhibited TFC, whereas activation of the same pathway during only CS period did not affect (Kitamura et al. 2014). On the other hand, optogenetic terminal inhibition of EC2 input in CA1 during TFC enhanced TFC. Optogenetic light activation/inactivation with temporal control identified the functional role of two direct glutamatergic inputs from EC to CA1 for specific moment during temporal association learning.

3.3, Neural circuits and neural process for memory consolidation

Memory consolidation is a process by which a newly acquired memory is transformed into a more long-lasting stable state (Box1) (Lechner et al. 1999). Memories are thought to be initially stored within the EC-HPC network (recent memory) and, over time, slowly consolidated within the neocortex for permanent storage (remote memory) (Marr 1971; Squire 1986; Kim et al. 1993). This process is known as systems consolidation of memory. Several theories have attempted to explain the neurobiological mechanisms that underlie consolidation of episodic memories. The Standard Consolidation Theory (SCT) argues that newly acquired episodic memories are initially stored in the HPC and are then transferred to the cortex for long-term storage (Squire 1986). While there is much evidence to support SCT, comparably strong evidence refutes SCT (Squire 1992; Nadel and Moscovitch 1997; Bayley et al. 2005). These inconsistencies have led to the Multiple Trace Theory (MTT). Unlike SCT, MTT argues that the HPC is always required for the recall of episodic memory that initially required the HPC for the formation. Furthermore, memory stored in the neocortex might be important for the recall of generalized semantic memory, whereas memory stored in the HPC might be important for the recall of detailed memory (Nadel and Moscovitch 1997). A possible mechanism that can explain these results is that there are memory traces in both the HPC and neocortex, and that either trace can be used for memory recall depending on an individual’s situation and condition (Kitamura and Inokuchi 2014; Tonegawa et al. 2018; Terranova et al. 2019). Goshen et al. showed that the acute optogenetic inhibition of the hippocampal CA1 neurons timed to the onset of recall impaired the recall of CFC in both recent and remote memory tests, whereas pharmacological hippocampal inhibition (tested 30 mins after drug infusion) or prolonged optogenetic inhibition before and during memory recall impaired only recent memory recall (Goshen et al. 2011). These results could be explained if memory traces coexist in both the hippocampus and neocortex and that either memory trace can be used for memory recall depending on the animal’s situation and condition during recall. By using activity-dependent cell labeling with ChR2 to identify memory engram cells as mentioned above (Section 3.1.2), Kitamura et al. examined the existence of memory engrams in the HPC and medial prefrontal cortex (mPFC) at recent and remote time points after CFC (Kitamura et al. 2017). They identified that mPFC engram cells are rapidly generated during the acquisition phase of contextual fear conditioning, however, mPFC engram cells are not reactivated during recent memory recall and are not necessary for the recall of recent memory. The memory engram cells in this state are referred to as silent engrams (Tonegawa et al. 2018; Terranova et al. 2019). After contextual fear conditioning, the mPFC silent engram cells mature over time with support from HPC engram cells. On the other hand, HPC engram cells rapidly generate as active state, which are reactivated during memory recall and necessary for memory recall. However, HPC engram cells gradually become “silent,” such that they can be only reactivated by artificial optogenetic stimuli but not by using natural recall cues. These data provide new evidence for a unified theory of memory consolidation between SCT and MTT. In summary, the systems consolidation process consists of two major steps: generation of silent memory engram in the PFC during learning and following functional maturation of the memory engram network by inputs from engram cells in EC-HPC network lasting a couple of weeks (Kitamura et al. 2017; Tonegawa et al. 2018; Terranova et al. 2019). However, even at remote time points, HPC engram cells still exist but in a silent state. These recent studies with optogenetic neural activation in engram cells revealed that responsible neural circuits and their engrams for a specific memory are functionally organized during the systems consolidation process.

Accumulating physiological evidence has suggested that high-frequency field oscillations, referred to as sharp-waves ripples (SWR), which are spontaneously observed in the hippocampal CA1 during slow-wave sleep and immobile quiet awake state, may contribute to the memory consolidation process (O’Keefe 1976; Buzsáki et al. 1983; Buzsáki et al. 1992). Discovery of the sequential reactivation of the firing sequences that represents recently acquired memory during sleep SWR events in hippocampal CA1 further supports the idea (Wilson and McNaughton 1994; Skaggs and McNaughton 1996; Karlsson and Frank 2009). Several groups examined the role of SWR on memory consolidation, and found that the disruption of SWR with electric stimulation showed the causal behavioral relevance of hippocampal SWR on the stabilization of recently acquired memory (Girardeau et al. 2009; Ego-Stengel and Wilson 2010; Jadhav et al. 2012). Nakashiba et al. also tested the effect of the inhibition of CA3 to CA1 output with TeTX (Nakashiba et al. 2009). Although the chronic blockade of CA3 output with TeTX did not alter the number of SWR events, the blockade reduced the intrinsic frequency of SWR in CA1 and impaired the remote memory formation of CFC (Nakashiba et al. 2009). However, two recent studies applied optogenetics to acutely inactivate the CA3 to CA1 pathway to re-examine the role of the tri-synaptic pathway on SWRs. An acute optogenetic silencing of CA3 terminals in CA1 drastically suppresses SWR incidence in both awake and sleep state in mice and rats (Yamamoto and Tonegawa 2017; Davoudi and Foster 2019), suggesting that the compensatory changes by prolonged suppression with TeTX underscored the importance of CA3 output on the generation of SWRs. The real-time optogenetic manipulation triggered by SWR occurrence in behaving animals and acute brain slices further showed that hippocampal SWRs in quiet awake and slow-wave states stabilize newly learned memory by down-regulation of unrelated synaptic weight (Fig. 1b) (Norimoto et al. 2018). These studies indicate that hippocampal SWRs are crucial for memory consolidation, and may contribute to the generation and maturation of PFC engram cells as well as silencing hippocampal engram cells for the formation of neocortical remote memory.

4, Conclusion

In this chapter, we have reviewed recent updates about neural circuit mechanisms of episodic memory revealed by cell type-specific optogenetic approach. Reelin⁺ cells in EC2 drive contextual information to HPC, and contextual memory is formed in EC2-DG-CA3 circuits (Fig. 2). During CFC, sub-populations of DG granule cells are activated as memory engrams and the activated cells during CFC are necessary and sufficient for recall of a CFC memory. However, after EC2-DG-CA3 circuits deliver processed contextual memory information to hippocampal CA1 pyramidal cells, two distinct neural pathways (CA1-dSub-EC5 pathway or CA1-EC5 pathway) to basolateral amygdala are differentially involved in the formation or recall of contextual fear memory. While memory engram cells are rapidly generated after CFC, memory engram cells are also formed in mPFC, but they are not responsible for memory recall by natural recall cues. Rather, mPFC engram cells gradually mature with time to form remote CFC memories. On the other hand, the tri-synaptic pathway is dispensable for temporal association learning. The direct pathway from EC3 to CA1 is necessary for TFC, and other direct pathway from Wfs1⁺ cells in EC2 into CA1 inhibits TFC by the suppression of the EC3 input into the CA1 pyramidal cells through feed-forward inhibition. These studies suggest that each sub-region and projection in the EC-HPC networks may have distinct functional roles in the formation, recall and consolidation of episodic memory. As a technical development, millisecond-timescale control of neural stimulation by optogenetics has facilitated i) identification of synaptic connectivity between targeted paired of neurons, ii) cell-type specific single unit electrophysiological recording and iii) identification of memory engram cells, and iv) artificial induction and modification of synaptic plasticity in targeted synapses.

While optogenetics has significantly advanced the field of neurobiology of learning and memory, there are several issues which need to be improved. For example, excitation of ChR2 in high-expressing neurons triggers firing with relatively high reliability, whereas low-expressing neurons need longer exposure of blue light to acquire reliable neuronal firing, and that longer exposure can result in phototoxicity (Wade et al. 1988; Lin 2011; Kravitz and Kreitzer 2011; Yizhar et al. 2011; Owen et al. 2019). However, high levels of opsin expression have been also linked to neuronal defects and toxicity (Zimmermann et al. 2008; Zhao et al. 2008; Lin 2011; Yizhar et al. 2011; Miyashita et al. 2013). Another issue is a misidentification error due to the variety of spike latency between directly excited cells and indirectly excited cells (Kravitz et al. 2013). It might be difficult to separate direct or indirect light-responsive cells by latency parameter criteria in combination with use of in vivo electrophysiology (Kravitz et al. 2013; Buzsáki et al. 2015). The silencing of neural activities also suffers from the same technical issue due to the reduction of firing rate of their connected partners (Senzai and Buzsáki 2017). Another issue is the production of heat with light illumination. Because optogenetic inactivation needs to continuously illuminate green light to inhibit neuronal activity, the light that is emitted from the optical fibers can cause heating. In fact, Owen et al. showed that neural firing rates were reduced by continuous illumination in correlation with the extent of temperature increase (Owen et al. 2019). Importantly, they showed that continuous illumination without expression of opsin affected animal behavior as well as neural firing rates. Thus, it is crucial to consider light-delivery parameters without the expression of opsins.

Nevertheless, optogenetic approaches have been significantly advancing day by day. For example, an activity-dependent cell labeling technique with doxycycline control, which was mentioned above in Section 3.1.2., has low temporal resolution (a couple of days) to labels activated cells due to the control by doxycycline diet. However, newly developed technologies enable the labeling neurons with increasing intracellular calcium concentration by the delivery of blue light (Lee et al. 2017; Moeyaert et al. 2018), which would produce higher temporal resolution for labeling cells activated during a specific moment. Although their sensitivity and stability should be tested in vivo, the development of these light- and activity-dependent cell labeling techniques will enable us to examine neural mechanisms of the formation, recall and consolidation of episodic memory with higher spatial-temporal resolution.

List of Abbreviations

AAV

Adeno-associated virus

Arch

Archaerhodopsin

CFC

Contextual fear conditioning

ChR2

Channelrhdopsin-2

CS

Conditioned stimulus

DG

Dentate gyrus

dSub

Dorsal subiculum

EC

Entorhinal cortex

EC2

Layer II of entorhinal cortex

EC3

layer III of entorhinal cortex

EC5

layer V of entorhinal cortex

HPC

Hippocampus

IEG

Immediate early gene

LTD

Long-term depression

LTP

Long-term potentiation

mPFC

Medial prefrontal cortex

MTT

Multiple Trace Theory

NMDA

N-methyl-D-aspartate

NpHR

Halorhodopsin

NR1

N-methyl-D-aspartate receptor

Reelin+ cells

Reelin positive stellate cells

SCT

Standard consolidation theory

SL

Stratum lacunosum

SL-IN

Stratum lacunosum interneuron (GABAergic)

SWR

Sharp-wave ripples

TeTX

Tetanus-toxin (light chain)

TFC

Trace fear conditioning

US

Unconditioned stimulus

Wfs1

Wolfram syndrome 1

Wfs1+ cell

Wolfram syndrome 1 positive (pyramidal) cell