Dissecting cell-type specific pathways in medial entorhinal cortical- hippocampal network for episodic memory

Abstract

Episodic memory, which refers to our ability to encode and recall past events, is essential to our daily lives. Previous research has established that both the entorhinal cortex (EC) and hippocampus (HPC) play a crucial role in the formation and retrieval of episodic memories. However, to understand neural circuit mechanisms behind these processes, it has become necessary to monitor and manipulate the neural activity in a cell-type-specific manner with high temporal precision during memory formation, consolidation, and retrieval in the EC-HPC networks. Recent studies using cell-type-specific labeling, monitoring, and manipulation have demonstrated that medial EC (MEC) contains multiple excitatory neurons that have differential molecular markers, physiological properties, and anatomical features. In this review, we will comprehensively examine the complementary roles of superficial layers of neurons (II and III) and the roles of deeper layers (V and VI) in episodic memory formation and recall based on these recent findings.

Graphical Abstract

The entorhinal cortex (EC) and hippocampus (HPC) play a crucial role in the formation and retrieval of episodic memories. Recent studies using cell-type-specific labeling, monitoring, and manipulation have demonstrated that medial EC (MEC) contains multiple excitatory neurons that have differential molecular markers, physiological properties, and anatomical features. In this review, we will comprehensively examine the complementary roles of superficial layers of neurons (II and III) and the roles of deeper layers (V and VI) in episodic memory formation and recall based on these recent findings.

Introduction

The medial entorhinal cortex (EC, MEC) is the structure which receives input from and sends output to the hippocampus (HPC). Accumulating evidence have shown that the EC-HPC network is crucial for formation of episodic memory, which is the ability to remember and store information about temporally organized events and their temporal-spatial relationship (Sugar & Moser, 2019; Tulving, 1972). The critical contribution of hippocampal structures to episodic memory was initially revealed in the human study with the patient H.M., who had bilateral resection of medial temporal lobes due to epilepsy (Scoville & Milner, 1957). Importantly, the histological validation of surgical performance on the patient H.M.’s brain revealed a lesion of a large portion of the EC while a significant amount of residual hippocampus remained (Annese et al., 2014). Likewise, lesion or neural inactivation of the HPC and MEC in animals shows impairment of memory of “when”, representing the timing and sequence of events and also the association of temporally segregated events (Ainge et al., 2007; Dias et al., 2021; Fortin et al., 2002; Fortin et al., 2004; Heys et al., 2020; Jacobs et al., 2013; Kitamura et al., 2014; McEchron et al., 1998; Robinson et al., 2017; Suh et al., 2011; Vo et al., 2021), and “where”, representing spatial orientation, navigation and context (Hales et al., 2021; Hales et al., 2014; Hales et al., 2018; Honey & Good, 1993; Kitamura et al., 2012; Morris et al., 1982; Nakazawa et al., 2004; Phillips & LeDoux, 1992; Steffenach et al., 2005). Correspondingly, findings of place cells (Ekstrom et al., 2003; O’Keefe & Dostrovsky, 1971) and time cells (Kraus et al., 2013; MacDonald et al., 2011; Marks et al., 2019; Pastalkova et al., 2008; Robinson et al., 2017) in the HPC as well as grid cells (Fyhn et al., 2004; Moser et al., 2008) and recently found time-coding cells (Heys & Dombeck, 2018; Kraus et al., 2015) in the MEC imply that spatial and temporal information is processed in the MEC-HPC circuit for memory formation.

In the MEC-HPC circuit, there are multiple cell types defined anatomically and genetically, comprising distinct neural microcircuits. Cells in the superficial layers of MEC (II and III) send input to the HPC and cells in the deep layers (V and VI) receive output from the HPC. To further understand how the storage and association of episodic memory elements are implemented by the MEC-HPC circuit, it is essential to dissect the role of cell-type specific microcircuits on memory through cell-type-specific tagging and manipulation (Marks et al., 2021; Marks et al., 2022; Yamamoto et al., 2021). In this review, we focus on anatomical connections of different cell types in MEC-HPC circuit and discuss roles of the cell-type specific pathways in the spatial and temporal aspects of episodic memory in rodents, although other cortical/subcortical areas also interact with HPC and EC, such as direct projections with olfactory area, retrosprenial cortex, amygdala, thalamus, and many other areas, or indirect communication with other sensory areas via perirhinal and postrhinal cortices (Aggleton et al., 2004; Cappaert et al., 2015; Kerr et al., 2007; Moscovitch et al., 2016; Nilssen et al., 2019; Ranganath & Ritchey, 2012).

Neural circuits in EC-HPC networks

From a classical perspective, the primary neurons in MEC send projections to HPC via two pathways: one is referred to as the tri-synaptic pathway (MEC layer II (MECII) → dentate gyrus (DG) → CA3 → CA1), and another is the monosynaptic pathway (MEC layer III (MECIII) → CA1) (Amaral & Witter, 1989; Steward & Scoville, 1976; Witter et al., 2000). The inputs from MEC are integrated in CA1 and return to MEC layer V/VI (MECV/ MECVI) directly or indirectly via subiculum (Naber et al., 2001; Witter et al., 2000), or to other cortical areas (Cenquizca & Swanson, 2007; Fanselow & Dong, 2010; J. I. Terranova et al., 2022). However, since the previous studies found the connections with anterograde/retrograde tracers, brain-area lesions, or electrical stimulation in an acute brain slice, the specificity of genetically defined cell-type could not be investigated.

The recent findings of cell-type specific circuits in MEC-HPC are summarized in Figure. In MECII, there are two types of primary cells. One is the stellate cell expressing Reelin sending their projections to DG, CA3, and CA2 (Kohara et al., 2014; Varga et al., 2010). Another primary cells in MECII are the pyramidal cells expressing Calbindin (Cal) and Wolfram syndrome 1 (Wfs1), organizing a unique hexagonal pattern of patchy clusters, or “islands”, in MECII (Kitamura et al., 2014; Ray et al., 2014; Varga et al., 2010). Kitamura et al., (2014) found Cal/Wfs1⁺ cells send their axons into striatum lacunosum (SL) in hippocampal CA1, placed in a border between striatum radiatum (SR) and striatum molecular (SM) (Delpech et al., 2021; Kitamura et al., 2014; Ohara et al., 2019; Shu et al., 2016; Surmeli et al., 2015; Zutshi et al., 2018). While SL was identified in early anatomical studies (Hamlyn, 1963), the projection pattern and morphology of CA1 pyramidal neurons are similar to the neighboring SM layer, so they are taken as one in-dividable layer as striatum lacunosum-molecular (SLM) by the most of researches for decades (Hjorth-Simonsen & Jeune, 1972). Recently, it was shown that the synapse protein lifetime is shorter in SL than the SR and SM which indicates high turn-over rate relating to learning-induced plasticity (Bulovaite et al., 2022), suggesting unique function of this layer. The axons of Cal/Wfs1⁺ cells were found to project to a population of interneurons in SL (SL-INs), providing feed-forward inhibition to regulate excitation of CA1 pyramidal cells induced by MECIII inputs (Kitamura et al., 2014). The minor projections are reported directly to CA1 pyramidal cells (Delpech et al., 2021; Kitamura et al., 2014), while these electrophysiological properties and roles in cognitive functions are not known yet. The projection pattern of MECIII to CA1 was investigated using transgenic mice in which the expression level of the Cre-loxP recombination was robust in MECIII (Kitamura et al., 2014; Kohara et al., 2014; Suh et al., 2011). While the previous study with electric stimulation suggested monosynaptic input to CA2 from ECIII (Chevaleyre & Siegelbaum, 2010), the observation of the MECIII cell-specific pathway showed direct projection to CA1 pyramidal cells in the SM layer (Kitamura et al., 2014; Kohara et al., 2014) but MECIII cell axons terminated sharply at the end of the CA1 region without projecting to CA2 (Kohara et al., 2014). The above MECII Cal/Wfs1⁺ and MECIII to HPC circuits mainly correspond to the temporo-ammonic pathways. While MECII/III cells also project to CA1 via alvear pathway (Bell et al., 2021; Deller et al., 1996), its detailed cell-type specific connectivity and role in memory are not known yet.

Figure. Summary of MEC-HPC circuit connectivity.

Hippocampal CA1 (CA1) pyramidal cells receive inputs from medial entorhinal cortex (MEC) via tri-synaptic pathway and direct pathway. Tri-synaptic pathway starts from MEC layer II (MECII) Reelin⁺ cells projecting to dentate gyrus (DG), hippocampal CA3, and CA2, then they project to CA1. MECIII cells send monosynaptic excitatory projections to CA1. MECII Cal⁺/Wfs1⁺ cells project to interneurons in stratum lacunosum (SL-INs) of CA1, and SL-INs send inhibitory projections to CA1 pyramidal cells. CA1 returns projections to MEC layer V (MECV) directly or via subiculum. In MECV, MEC layer Vb (MECVb) cells receive output from CA1 and send projections to MEC layer Va (MECVa) as well as MECIII and MECII. MECVa cells send outputs to telencephalic structures including neocortex and amygdala, while also sending a copy of their outputs back to the CA1. MECII Reelin⁺ cells → DG/CA3/CA2 contributes to contextual learning and CA2 contributes to social memory. While MECIII cells → CA1 contributes to temporal association learning, MECII Calbindin⁺ (Cal⁺)/Wolfram syndrome 1⁺ (Wfs1⁺) cells → CA1 regulate temporal association learning. CA1 → MECV pathways contribute to formation and recall of recent memory, and MECV to telencephalic structures contribute to consolidation of remote memory. Black lines indicate local projections in MEC or projection of MECVa to CA1, which cell-type specific roles in memory functions are not known.

Connectivity with EC deeper layers is still being intensively investigated. The hippocampus innervates the deep layer neurons in MEC layer V (MECV) and to some extent in layers II and III. Also, MECV functions as a hippocampal output relay center that transmits the processed hippocampal information to the telencephalic structures like neocortex for long-term storage (Cappaert et al., 2015; Kosel et al., 1982; Rosene & Van Hoesen, 1977; Witter et al., 2017). MECV was suggested to have two sublayers from its morphological properties (Boccara et al., 2015; Canto & Witter, 2012; Lorente de Nó, 1933), and indeed can be divided into layer Va (MECVa) and Vb (MECVb) using molecular markers. MECVb neurons preferentially receive output from CA1 and subiculum (Ramsden et al., 2015; Surmeli et al., 2015), and a recent study revealed the dorsal HPC preferentially targets MECVb while the ventral HPC tends to target MECVa (Ohara et al., 2023). In addition, while layer Vb neurons send their projections to layer Va, III, II of EC (Nilssen et al., 2019; Ohara et al., 2018; Surmeli et al., 2015; Witter et al., 2017) and to thalamus (Surmeli et al., 2015) (but not observed in another study (Ohara et al., 2018), MECVa neurons have telencephalic projections to neocortex and amygdala (Kitamura et al., 2017; Surmeli et al., 2015) and also direct projection to stratum pyramidale deep layer and SL of CA1 (Tsoi et al., 2022). The neural circuits involving MEC layer VI (MECVI) are unclear. A recent study by Ben-Simon et al., (2022) demonstrated that excitatory neurons in MECVI are labeled with the marker of cortical layer VIb, termed MEC layer VIb. MEC layer VIb cells receive input from hippocampal CA1, thalamus, and claustrum, and they send their projections to hippocampal CA3/CA1/DG (Ben-Simon et al., 2022).

Role of Reelin⁺ cells in MECII

Encoding and recalling information of spatial context is an important aspect of organizing episodic memory. The hippocampal tri-synaptic pathway (ECII Reelin⁺ cells → DG → CA3 → CA1) has been thought to process contextual information (Gilbert et al., 2001; Leutgeb et al., 2004; McHugh et al., 2007; Nakashiba et al., 2009; Nakashiba et al., 2008; Nakazawa et al., 2003; Oreilly & Mcclelland, 1994). In the tri-synaptic pathway, accumulating evidence indicates that DG operates pattern separation, a process that minimizes overlapping or similar experiences into distinct neural representations (Gilbert et al., 2001; McHugh et al., 2007). On the other hand, the recurrent network of CA3 is thought to operate pattern completion, a process that reconstructs memory representation from partial or degraded input cues (Nakashiba et al., 2008).

MECII is known to have grid, head-direction, spatial border modulated, and speed cells (Brandon et al., 2013; Fyhn et al., 2004; Iwase et al., 2020; Moser et al., 2008; Rowland et al., 2018; Sun et al., 2015; Tukker et al., 2022) as well as environmental-feature cells (Diehl et al., 2017), suggesting contextual information is already processed before starting tri-synaptic pathway. However, cell-type specific role in the ability of contextual discrimination had not been well understood. Kitamura et al. (2015) investigated the distinct roles of Reelin⁺ and Cal/Wfs1⁺ cells in MECII using in vivo Ca²⁺ imaging while the mice explored in different contexts. During freely moving, the Reelin⁺ cells, but not Cal/Wfs1⁺ cells, represented context-specific activities. Furthermore, using contextual fear conditioning, in which an animal receives an aversive stimulus in a specific context, optogenetic inhibition of Reelin⁺ cells impaired context-specific fear memory and neural representation in CA3 (Kitamura, Sun, et al., 2015). Similarly, Tannant et al. (2018) used cell-type-specific inhibition of MECII Reelin⁺ cells with tetanus toxin light chain and showed deficits in learning of cue- and path integration-based estimation of reward location. They also showed deficits in object location recognition whereas novel object recognition was intact (Tennant et al., 2018). Artificial depolarization of Reelin⁺ cells, but not hyperpolarization, induced remapping in CA1 place cells and deficits in the water maze task (Kanter et al., 2017). Thus, MECII Reelin⁺ cells process contextual information signals and send them to the tri-synaptic pathway for further processing for the formation and retrieval of contextual memory. In the tri-synaptic pathway, CA2 was found to be crucial for social memory (Hitti & Siegelbaum, 2014; Middleton & McHugh, 2020; Oliva et al., 2020); however, recent findings imply that a lateral entorhinal cortex (LEC) to CA2 direct pathway, rather than MEC to CA2, is involved in social memory (Lopez-Rojas et al., 2022).

Role of MECIII excitatory neurons

Association of temporally segregated events, called temporal association learning (TAL), is another crucial element for organizing episodic memory (Kitamura, 2017; Wallenstein et al., 1998). Impairments in TAL have been demonstrated in EC lesions (Esclassan et al., 2009; Ryou et al., 2001) and muscimol infusion into LEC (Morrissey et al., 2012), as well as lesions in hippocampal CA1 (Kesner et al., 2005; McEchron et al., 1998), although these experiments cannot differentiate the role of the distinct cell types. Suh et al., (2011) performed loss-of-function experiments using transgenic mice expressing tetanus toxin light chain (TeTX) specifically in MECIII cells and demonstrated the contribution of MECIII cells to TAL, as assessed by the paradigm of trace-fear-conditioning (TFC) (Suh et al., 2011). The TFC paradigm consists of conditioning tone stimulus followed by a temporal gap of 20–30-s before aversive foot-shock (Chowdhury et al., 2005; McEchron et al., 1998; Solomon et al., 1986). The mice in which MECIII cells were inhibited showed deficits in TAL during TFC test, despite having no deficit in normal contextual fear conditioning (Suh et al., 2011). Inhibition of CA3-CA1 synaptic transmission using mice expressing TeTX in CA3 (Nakashiba et al., 2008), on the other hand, did not affect deficits in TAL (Suh et al., 2011). Consistently, Kitamura et al. found that optogenetic inhibition of MECIII cells decrease TFC test performance and 4-Hz stimulation of these cells increase the test performance (Kitamura et al., 2014), while optogenetic inhibition of Reelin⁺ cells did not affect TFC test performance (Kitamura, Sun, et al., 2015). Although several studies indicate CA3 and DG may still have a limited role in temporal information processing (Keinath et al., 2020; Kesner et al., 2000; Salz et al., 2016), these findings suggest that the direct input from MECIII to CA1, but not the tri-synaptic pathway, is crucial for forming TAL.

While the neural activity mechanism to support TAL is not yet understood, TAL is considered to require such neural activity to bridge the temporal gap during TAL. One possible candidate is persistent firing activity (Fuster, 1973; Lin et al., 2020; Zylberberg & Strowbridge, 2017), which is defined as the ability to continuously spike even after the termination of an external stimulus. Persistent activity is observed in MECIII in in vitro study (Yoshida et al., 2008). This persistent activity depends on a metabotropic glutamatergic receptor 1 (mGluR1) and muscarinic 5-HT receptor (Egorov et al., 2002; Yoshida et al., 2008), and injection of muscarinic receptor antagonist into EC deficits TFC (Esclassan et al., 2009). In addition, the cocktail of antagonists for muscarinic receptor and for mGluR1 inhibits TFC in control mice but did not affect in MECIII-TeTX mice (Suh et al., 2011). Also, while high gamma oscillations generated by MECIII→CA1 pathway are crucial for successful execution of spatial memory, they are reduced by chemical blockade of persistent activity, indicating the relationship of the persistent activity and memory recall using MECIII→CA1 pathway (Yamamoto et al., 2014). Yamamoto and Tonegawa further found ripple burst activity in MEC and CA1 are reduced in MECIII-TeTX mice, suggesting role of MECIII cells in longer episodes such as that required in TAL (Yamamoto & Tonegawa, 2017). Moreover, plateau potential in CA1 neurons, a sustained membrane potential depolarization after the triggering stimulus, can be another candidate to serve for bridging temporal gap in TAL. The plateau potential is suggested to be generated by correlated input to CA1 from ECIII and CA3 (Bittner et al., 2015; Takahashi & Magee, 2009), and it induces plasticity in CA1 neurons in behavioral timescale, suggesting that ECIII activity induces CA1 plateau potentials to allow behavioral timescale plasticity (Bittner et al., 2017; Grienberger & Magee, 2022). Recently, Ahmed et al. (2020) found neural representation of CA1 during temporal gap in TAL is different between after the conditioned and control stimulus, suggesting neural encoding of trace period after the cue in CA1 (Ahmed et al., 2020). Similarly, Souza et al. also found sequential firing in CA1 accompanied by persistent firing in prefrontal cortex (PFC) in TFC; however, a short (1-sec) temporal gap was adopted in the study (Souza et al., 2022) thus the learning may be different from hippocampal-dependent TAL with a longer temporal gap (Chowdhury et al., 2005). Together, MECIII cells cause long-sustained neural activity changes in themselves and in CA1 cells, which can support TAL by bridging the temporal gap during events.

Role of Cal/Wfs1⁺ excitatory neurons in MECII

The second principal cells in MECII are Cal/Wfs1⁺ cells. The existence of different cell types in MECII (Alonso and Klink, 1993) have been thought to contribute to two distinct information processing within the MEC and signal transmission to the hippocampus (Alessi et al., 2016; Alonso & Klink, 1993; Sasaki et al., 2015). The Cal/Wfs1⁺ cells form unique periodic cell-clusters surrounded by Reelin⁺ cells (Kitamura et al., 2014; Ray et al., 2014; Varga et al., 2010). These patchy cell-cluster structures are also stained by cytochrome oxidase (Kitamura et al., 2014), a metabolic marker for neuronal activity (Wong-Riley, 1989), and found in many species including rodents, bats, monkeys, and humans with a manner of relatively preserved size and periodicity (Naumann et al., 2016). Correspondingly, a nonrandom microvasculature pattern surrounding the patch structures was observed on the surface of the human entorhinal cortex (Solodkin & Van Hoesen, 1996). This periodic structure is known to be lost during Alzheimer’s disease (AD) (Kordower et al., 2001; Solodkin & Van Hoesen, 1996). A recent study showed an accumulation of phosphorylated tau-protein in Cal/Wfs1⁺ cells in early AD followed by propagation to CA1 (Delpech et al., 2021). The relationship of patchy structure and AD suggest a critical role of Cal/Wfs1⁺ cells in organizing episodic memory. Since similar patchy structures can be observed not only in the EC but also in other cortical areas (Ichinohe et al., 2003; Ichinohe & Rockland, 2004; Marcondes et al., 2019; Rockland, 2021; Wyss et al., 1990), the periodic patchy structure of Cal/Wfs1⁺ cells may have some advantages for processing spatial or temporal information. Interestingly, staining with synaptic zinc visualizes another patchy structure complementary with Cal/Wfs1⁺ cells clusters (Ray et al., 2017), indicating the existence of two different module structures in MEC (Naumann et al., 2018; Witter & Moser, 2006). Furthermore, Cal/Wfs1⁺ cells and Reelin⁺ cells are connected from different types of interneurons (Nilssen et al., 2019; Szocs et al., 2022; Varga et al., 2010). Also, within the local circuit of MECII, Cal/Wfs1⁺ cells send inputs to Reelin⁺ cells and to local inhibitory neurons, but do not receive inputs from Reelin⁺ cells (Fuchs et al., 2016; Tukker et al., 2022; Winterer et al., 2017). Optogenetic stimulation of MECII Cal/Wfs1⁺ cells provide strong inhibition to neurons in MECII and MECIII following brief excitation to neurons in MECII (Zutshi et al., 2018). These suggest that MECII Cal/Wfs1⁺ cells process spatial or temporal information differently from other cell types.

In TAL, it is crucial to have a system to tightly regulate associative learning for adaptation to beneficial or threatening events in external environment (Kitamura, 2017). Too weak association of temporally distinct events will result in failure of memory formation, whereas too strong association will result in inappropriate linkage of events and interference of other useful associations. Kitamura et al., (2014) found that regulation of TAL was performed by Cal/Wfs1+ cells (Kitamura et al., 2014). Their distinct role from MECIII cells was shown by optogenetics cell-type-specific activation and inhibition of Cal/Wfs1⁺ cells and MECIII cells. While MECIII inhibition decreased the performance of TFC test with a 20-sec temporal gap, inhibition of Cal/Wfs1⁺ cells during TFC learning caused enhanced performance. On the other hand, activation of Cal/Wfs1⁺ cells decreased the performance of TFC test. However, both activation and inhibition of Cal/Wfs1⁺ cells did not affect normal contextual fear conditioning (Kitamura et al., 2014). Inhibition of Cal/Wfs1⁺ cells further demonstrated an enhanced association of events between a longer temporal gap of 60-sec, suggesting Cal/Wfs1⁺ cells inactivation is important for formation of larger episodic memory (Yokose et al., 2021). Therefore, these results indicate Cal/Wfs1⁺ cells regulate TAL which is performed by controlling CA1 pyramidal neuron excitation induced by MECIII inputs.

Role of Va excitatory neurons

MECV could be considered as a micro-hub which receives hippocampal afferents and is implicated in hippocampus-based memory (Rozov et al., 2020; van Haeften et al., 2003; Witter et al., 2017). Anatomically, the neurons of this layer have a wide range of connections ranging from local intrinsic connections towards other layers in the MEC to long-range projections to the telencephalic structures (Kitamura et al., 2017; Ohara et al., 2018; Swanson & Kohler, 1986). Based on genetic markers and connections, MECV could be divided into two sublayers: LVa projections to telencephalic structures and LVb as a local projection to superficial layers in EC (Surmeli et al., 2015; Witter et al., 2017). The layers Va and Vb differ in cell size and cell density (Boccara et al., 2015; Canto & Witter, 2012; Lorente de Nó, 1933), and have specific transcription factors such as Etv1 and Ctip2 respectively (Blankvoort et al., 2022; Ramsden et al., 2015; Surmeli et al., 2015).

Memories are thought to be temporary stored in the EC-HPC circuit followed by transforming into the neocortex as a long-lasting state, which is referred to as systems consolidation of remote memory (Frankland & Bontempi, 2005; Kitamura & Inokuchi, 2014; J. Terranova et al., 2022; Terranova et al., 2019; J. I. Terranova et al., 2022; Tonegawa et al., 2018). Using cell-type specific labeling and optogenetics experiments, Kitamura et al. (2017) found the projection of MECVa neurons to medial PFC (mPFC) is critical for remote memory formation (Kitamura et al., 2017). In this study, axon terminal inhibition of MECVa to mPFC during day 1 of the contextual fear conditioning paradigm resulted in the impairment of remote memory formation but not recent memory. The MECVa neurons also transmit hippocampal information to the basolateral amygdala (BLA), a critical brain region involved in contextual fear conditioning. However, layer Va projection to BLA is critical in the retrieval of recent memories but not remote ones. These experiments show the multifunctional role of layer Va neurons in memory consolidation in the neocortex. Layer Va neurons also play an active role in the mPFC engram maturation (Kitamura et al., 2017). Even though the engrams for context memory formation are established in mPFC and HPC in the initial memory formation stage, a reactivation in mPFC engrams happens at a much later stage, after the functional maturation of mPFC engrams. It is postulated that the transfer of hippocampal information to mPFC, resulting in their engram maturation, is mediated via the HPC-MECVa pathway. The inactivation of the MEC pathway to mPFC during memory acquisition affect the development of engram cells in mPFC, pointing the necessity of the activity of MECVa neurons for long-term storage of memory in mPFC (Kitamura et al., 2017). A recent study has also shown that the MEC to PFC activity is mediated by neurotrophic factors (Hong et al., 2022). These results show that the tropomyosin receptor kinase B (TrkB) deletion of MEC-PFC pathway impaired the remote memory recall, also shown by significantly reduced cFos⁺ cells in PFC. Another study found similar gene expression patterns of MEC and PFC, suggesting functional coherence between MECV/VI and PFC (Ramsden et al., 2015). In addition, recent studies have shown that layer Va sends a copy of their telencephalic outputs back to the hippocampal CA1 indicating its specific role in simultaneously influencing both the hippocampus and the neocortex (Tsoi et al., 2022). Thus, layer Va act as a microhub in both providing a neocortical output as well as sending a memory-copy to the CA1, much similar to the efference copy. MEC is also postulated to play a major role in transitive inference, as optogenetic inhibition of MECVa terminals in anterior cingulate cortex during REM sleep induced inference in mice which were not trained adequately to find the relationship between items (Abdou et al., 2021). Thus, MECVa also plays a role in reorganizing knowledge, which is important in flexible behaviors.

Role of Vb excitatory neurons

In addition to that MECVb neurons project to MECVa, MECVb also projects to contact pyramidal neurons in the MECII and MECIII (Nilssen et al., 2019; Ohara et al., 2018; Surmeli et al., 2015; Witter et al., 2017). Thus, the intrinsic connectivity of MECVb to the superficial layers might function as a re-entry circuit to the hippocampus. This circuit of reverberation is thought to be critical in the temporal storage of episodic information. It is interesting to note that the layer Vb neurons do not innervate telencephalic structures (Ohara et al., 2018; Surmeli et al., 2015). Instead, layer Vb neurons preferentially receive projection from the CA1 and subicular regions of HPC than Va, and receive local connections from MECII Reelin⁺ cells (Surmeli et al., 2015). Additionally, EC neurons from layer Vb projects to layers Va, although such projection is more prominent in LEC and ventral MEC compared to dorsal MEC (Ohara et al., 2021; Ohara et al., 2018). Also, the dorsal CA1/subiculum innervates the dorsal MECVa/Vb (Rozov et al., 2020), whereas the ventral CA1/subiculum innervates both the dorsal and ventral MECVa/Vb (Ohara et al., 2023). This peculiar connectivity of MECV might be helpful in separate processing of information arising from dorsal/ventral HPC and integration of them in MECV. Roy et al. found direct input to MECV from CA1 is crucial for memory formation and indirect input to MECV from CA1 via subiculum is important for memory retrieval (Roy et al., 2017); however, the role of MECVa and Vb for memory formation and retrieval cannot be dissected yet since Va and Vb both receive inputs from CA1 and subiculum even there are preferences.

Role of VI excitatory neurons

Very few studies have conducted on the functional role of MEC layer VI (MECVI) in learning and memory. In neocortex, although the cytoarchitecture of layer VI is complex, studies have categorized layer VI into two separate sub layers called layer VIa and VIb based on their different morphological and genetic properties (Molyneaux et al., 2007; Tasic et al., 2018). The neocortical VIb subplates neurons also express biomarkers like Ctgf, Nxph4, and Cplx3 (Hoerder-Suabedissen & Molnar, 2013; Hoerder-Suabedissen et al., 2013; Hoerder-Suabedissen et al., 2009). Ben-Simon et al. found that MECVI neurons are labeled by the markers of neocortical VIb neurons, while the markers of neocortical VIa are not expressed in ECVI (Ben-Simon et al., 2022), suggesting similarity of cell-type populations to neocortical layer VIb. A rabies-based transsynaptic retrograde and Cre-dependent AAV labeling experiments injecting into DG, CA3, and CA1 suggests that the MECVI neurons monosynaptically innervate all the subfields of the hippocampus including DG, CA3, and CA1 (Ben-Simon et al., 2022). However, it is better to be noted that further experiments of anterograde tracing from MECVI are needed to investigate the projection pattern and density of MECVI cell axons within HPC subfields. The ex vivo electrophysiological experiment in Ben-Simon et al. 2022 also showed that MECVI neurons are glutamatergic and monosynaptically connect with CA3 pyramidal cells. The optogenetic activation of MECVI axons exhibits a slow decay along with absolute synaptic depression in CA3 pyramidal neurons, thus giving rise to plateauing of EPSPs in the hippocampal CA3 pyramidal neurons (Ben-Simon et al., 2022). The plateau potentials observed in the MECVI-CA3 are comparable with that of CA1 plateau potentials which is critical for place field formation and associative memories, possibly suggesting the crucial role of the pathway in spatial representation and memory (Bittner et al., 2015; Bittner et al., 2017; Grienberger & Magee, 2022; Zhao et al., 2022). They have further established that the such pathway plays a key role in the spatial information coding where optogenetic inhibition of the MECVI-CA1 pathway leads to a decrease in the spatial information score, but the firing rate and behavioral correlates remain unchanged. To examine the role of MECVI neurons on spatial learning, Ctgf⁺ population in the whole brain was ablated. The mutant mice fail to learn new reward locations and had trouble in forgetting the reward location they learned before the ablation. However, further region-specific experiments are required, since the ablation is not only targeted to MECVI population but also other cortical Ctgf⁺ population. MECVI neurons receive the projections from multiple brain regions, including the central cortical hub, claustrum, and provide output to all subregions of the HPC (Ben-Simon et al., 2022; Wang et al., 2017), showing the richness and diversity of information transfer through this pathway. These new findings are in contrast with the existing hypothesis on deep layers of EC which were thought to channelize the hippocampal output by selectively processing the information from hippocampus and subiculum and relaying it to the telencephalic structures.

Recent studies from deep layers of EC (both V and VI) providing projections to/from the hippocampus suggest that deep layers could function to affect both the neocortex and the hippocampus simultaneously, proposing the need for identifying the functional role of the deep layers in memory. More circuit manipulation experiments would establish their specific role in memory.

Role of inhibitory neurons

While we have focused on the roles of excitatory connections, MEC-HPC circuit also contains long-range inhibitory connections (Melzer et al., 2012; Ye et al., 2018). Subpopulation of inhibitory interneurons in stratum oriens of CA1 and DG send their axons preferentially to MECI/II and III rather than deeper layers. Also, subpopulation of inhibitory interneurons in MEC project to SLM and SR of CA1 and DG. The both long-range inhibitory projections have an ability to modulate the activity of target brain areas (Melzer et al., 2012). Additionally, 83% of HPC-projecting fast-spiking cells, putative parvalbumin-positive neurons, in MECII and III were found to be modulated with animal’s speed (Ye et al., 2018), suggesting to have a role in representation of self-location (Kropff et al., 2015). Such long-range inhibitory projections were found in LEC to CA1, which provides temporally-precise disinhibition in CA1 and contribute to both context and object recognition memory (Basu et al., 2016). However, such inhibitory projections are minor in MEC than LEC (Basu et al., 2016), and whether the long-range inhibitory connections in the MEC-HPC modulate episodic memory is not yet known.

There are variety of local inhibitory networks in HPC (Bezaire & Soltesz, 2013; Klausberger & Somogyi, 2008; Pelkey et al., 2017) and MEC (Tukker et al., 2022), and their role in memory processing has been discussed (Caroni, 2015; Hattori et al., 2017; Jeong & Singer, 2022; Topolnik & Tamboli, 2022). In CA1, somatostatin-positive (SOM⁺) interneurons are found to be required for contextual fear conditioning learning through dendritic inhibition of pyramidal cells at SLM while inactivation of parvalbumin-positive (PV⁺) interneurons has no effect (Lovett-Barron et al., 2014). Activation of a subpopulation of oriens lacunosum-moleculare interneurons impairs object learning, whereas their inhibition enhances the learning (Siwani et al., 2018). Also, since SL-INs receive inputs from Cal/Wfs1⁺ cells (Kitamura et al., 2014) and are connected by gap junctions (Price et al., 2005), SL-INs are supposed to be responsible for regulating TAL by broadly suppressing MECIII input to CA1 at SLM (Kitamura, Macdonald, et al., 2015; Kitamura et al., 2014). In MEC, silencing of PV⁺ cells reduced spatial selectivity of grid cells and speed modulation in speed cells in contrast silencing of SOM⁺ cells impairs the spatial selectivity of cells with discrete aperiodic firing fields, suggesting different roles in spatial information processing (Miao et al., 2017). However, more detailed functional characterization of cell-type specific inhibitory networks remains to be investigated.

Cell-type specific oscillatory communication in MEC-HPC

Neural oscillations provide temporal windows of synchronizing or desynchronizing neural firing that help communication between brain areas (Bonnefond et al., 2017; Buzsaki & Draguhn, 2004; Canolty & Knight, 2010). Because different cell-type pathway in MEC-HPC differently contributes to memories, understanding oscillatory activity of cell-type specific pathways will help understanding how multiple pathways in MEC-HPC communicate with target regions processing and integrating episodic memory information components.

The firing pattern of individual cell-type has been investigated in vitro, or with in vivo intracellular or juxtacellular recording combined with anatomical cell-type identification (Canto & Witter, 2012; Canto et al., 2008). Anatomically identified stellate cells and pyramidal cells in MECII correspond to MECII Reelin⁺ and Cal⁺/Wfs1⁺ cells (Fuchs et al., 2016). Although MECII stellate cells have intrinsic membrane properties corresponding to theta oscillation (4–12 Hz) (Alonso & Klink, 1993; Nolan et al., 2007), such properties are weak or absent in MECII pyramidal cells (Alonso & Klink, 1993), MECIII and MECV/VI cells (Dickson et al., 1997; Hamam et al., 2002; Quilichini et al., 2010). Nevertheless, regardless of intrinsic membrane properties, spontaneous firing of MECII pyramidal cells is highly theta-locked while MECII stellate and MECIII cells are only weakly phase-locked to theta oscillation (Burgalossi et al., 2011; Ray et al., 2014; Tang et al., 2014; Tang et al., 2015). There is a variety of phase preferences in MECV/VI cells to theta or gamma oscillations, and they are generally silent (Burgalossi et al., 2011; Gerlei et al., 2021; Quilichini et al., 2010). Since MECII pyramidal cells receive cholinergic input (Ray et al., 2014), a candidate for their strong theta phase-locking is the cholinergic input from medial septum. Importantly, phase preferences of MECII pyramidal and stellate/MECIII cells are 180 degrees opposite (Quilichini et al., 2010; Tang et al., 2015; Valero & de la Prida, 2018), suggesting their different role in information processing. Because the medial septum also sends GABAergic and glutamatergic projections to MEC (Fuchs et al., 2016; Gonzalez-Sulser et al., 2014), the difference of the phase preference of MECII pyramidal and stellate cells can originate from difference of direct/indirect projection from medial septum.

It has been suggested that HPC and EC communicate using theta and gamma oscillations (Colgin, 2015; Colgin et al., 2009; Igarashi et al., 2014; Yamamoto et al., 2014). Recently, Fernandez-Ruiz et al. demonstrated coupling of theta and gamma oscillations of DG and input pathways of MEC → DG and LEC → DG. Applying signal source separation technique to local field potential data, they showed putative MEC to DG pathway coupled at fast gamma (100 to 150 Hz) range with hippocampal theta oscillation, while putative MEC to DG pathway coupled at slow gamma (30 to 50 Hz) range with different theta phases. In addition, DG cells firing showed phase-synchrony with fast gamma during spatial learning task in contrast to slow gamma during object learning task (Fernandez-Ruiz et al., 2021). This implies that MECII Reelin⁺ cells send contextual information signals to DG through fast-gamma activity coupled with specific theta phase.

In CA1 of HPC, lesion or inhibition of EC result in impairment of theta wave at SLM of CA1, indicating CA1 receive theta input from EC (Buzsaki, 2002; Zutshi et al., 2022). Additionally, theta oscillatory activities are shown in the putative pathway-specific activity of ECII/III to CA1, CA2/3 to CA1 as well as ECII to DG (Lopez-Madrona et al., 2020); however, while MECII Cal⁺/Wfs1⁺ cells show strong theta modulation in contrast to MECIII, it is not known which cell-type pathway, MECII Cal⁺/Wfs1⁺ cells or MECIII cells, directly contributes to CA1 theta oscillation at SLM. Also, 60–100 Hz gamma oscillation of putative EC to CA1 pathway is segregated by CA1 theta phase from 30–60 Hz gamma oscillation of putative CA3 to CA1 pathway (Fernandez-Ruiz et al., 2017; Schomburg et al., 2014). Yamamoto et al. found 60–120 Hz gamma synchrony between CA1 and MEC at successful execution of spatial working memory, together with inhibition of MECIII-CA1 pathway reduced the gamma oscillation and task performance (Yamamoto et al., 2014). These studies suggest that MECIII also produces theta-coupled gamma signals to CA1 for episodic memory, although its participation in temporal memory and interaction with MECII Cal⁺/Wfs1⁺ cells to CA1 pathway are not clear.

Hippocampal sharp-wave ripple (SWR) oscillation is critical for memory consolidation and recall (Buzsaki, 2015; Ego-Stengel & Wilson, 2010; Girardeau et al., 2009; Jadhav et al., 2012). While neurons in deep layers of MEC are generally silent, hippocampal SWR increases firing of deep layer cells in bursts, in contrast to the minor effect on superficial layers (Chrobak & Buzsaki, 1994, 1996; Gerlei et al., 2021; Roth et al., 2016). Considering that increase of HPC-cortical ripple-related coupling is critical for memory consolidation (Maingret et al., 2016) and that the MECVa has role in memory consolidation, the pathway CA1 → MECVa/Vb → telencephalic structures may have a function to relay SWR activities from HPC to neocortex during memory consolidation. Also, current-source density analysis of local field potential of EC after subiculum stimulation showed that the current sink initially appears in MECV/VI followed by the sink in MECIII, while the sink in MECIII is not observed in the case of weak stimulation (Kloosterman et al., 2003). Similarly, HPC stimulation cause the current sink in MECV/VI, MECIII followed by SLM of HPC CA1 (Kloosterman et al., 2004). Correspondingly, Yamamoto and Tonegawa found ripple bursts co-occur in CA1 and MEC, and blockade of MECIII → CA1 reduces ripple bursts and extended replays (Yamamoto & Tonegawa, 2017). These imply MEC deep layers have a function to return intense SWR activity of CA1 via CA1 → MECVb → MECIII → CA1 pathway. Simultaneously, MECIII cells induce local up-state generating MECVb cells activated (Beed et al., 2020) that may facilitate the relay of SWR in this pathway.

These oscillatory activities are thought to be generated by inhibitory networks, especially PV⁺ cells are considered to play key role and SOM⁺ cells are important for gamma and ripple oscillations (Amilhon et al., 2015; Bartos et al., 2007; Bezaire et al., 2016; Buzsaki, 2015; Buzsaki & Wang, 2012; Korotkova et al., 2010; Royer et al., 2012; Schlingloff et al., 2014; Stark et al., 2014). Recently, Zhou et al. (2022) found there is little evidence of cross-regional gamma-phase synchrony between MEC and HPC, suggesting local gamma oscillations are generated by interregional theta-inputs and local interneurons, rather than directly receiving interregional gamma-inputs (Zhou et al., 2022). Although different types of interneurons firing of at distinct times in local oscillations (Klausberger & Somogyi, 2008; Valero & de la Prida, 2018), the contribution of other interneuron types for oscillatory generation is less known. Also, it is not known whether HPC and MEC share the same mechanism of oscillatory rhythm generation. The activity of neurogliaform cells in CA1, which may be corresponding to SL-INs, decreases coupling between pyramidal cell firing and gamma oscillations without affecting the overall level of the pyramidal cell activity (Sakalar et al., 2022), suggesting neurogliaform cells selectively decreased the influence of inputs from EC. Asymmetric local inhibitory projection of MECV fast-spiking interneurons to MECVa and MECVb cells, favoring MECVb cells, indicates that hippocampal SWR regulates MECVb cell excitation rather than MECVa (Rozov et al., 2020). The contributions of local inhibitory networks to HPC-MEC communication warrant further investigation.

Brief conclusion and future direction

In this review, we have discussed the roles of different cell-type specific circuits in the MEC-HPC network. Contextual information is largely sent from Reelin⁺ cells in MECII to HPC through the tri-synaptic pathway that contributes to pattern separation and completion. On the other hand, a fine association of temporally distinct events is provided via the direct input from MECIII cells and Cal/Wfs1⁺ cells in MECII. Specifically, MECIII cells have a critical role in temporal association, whereas Cal/Wfs1⁺ cells regulate the temporal association implemented by MECIII using feed-forward inhibition to CA1. Contextual and temporal information are integrated in CA1 and then return to MECVb. Parts of the input into MECVb are then sent to telencephalic neocortex or BLA via MECVa, which contributes to memory consolidation.

The detailed cell-type specific anatomical connections and their function of MEC still need to be investigated. Particularly, studies of memory functions of MECVb and MECVI are yet limited. Also, commissural input in MEC superficial layers was recently found to be essential for memory retrieval of object displacement (Caputi et al., 2022); however it is not addressed which cell-type is involved.

While we have focused on MEC-HPC network, LEC is also critical for formation of episodic memory. In addition to that the LEC processes “what” information, or representation of objects (Connor & Knierim, 2017), social information (Lopez-Rojas et al., 2022), and egocentric information (Wang et al., 2018), it is also suggested to process temporal information (Lin et al., 2020; Montchal et al., 2019; Morrissey et al., 2012; Sawatani et al., 2023; Tsao et al., 2018). Together with the finding that time cells in HPC do not differ with MEC lesion (Sabariego et al., 2019) and LEC cells represent time information differently from MEC (Sawatani et al., 2023; Tsao et al., 2018), LEC may contribute to temporal memory in a different manner than MEC. Interestingly, Cal/Wfs1⁺ cells patchy structures observed in MEC are not found in LEC, but instead, Reelin⁺ cells form cell clusters in LEC layer II (Naumann et al., 2018; Naumann et al., 2016). While long-range projection of Cal/Wfs1⁺ cells in MEC is mainly to hippocampal CA1, Cal/Wfs1⁺ cells in LEC also send their projection to more wide-spreading forebrain areas (Leitner et al., 2016; Ohara et al., 2019). Furthermore, anatomical studies suggest that MEC and LEC may not be completely segregated because they are connected to each other (Kohler, 1988). Particularly, recent studies found that Reelin⁺ excitatory neurons in LEC layer II (Witter et al., 2017) project to inhibitory neurons in MEC layer I (Vandrey et al., 2022), both Cal⁺ cells in MEC and LEC respectively send projections to LEC and MEC (Ohara et al., 2019), and MEC somatostatin⁺ interneurons selectively inhibit LEC layer IIa cells (Reelin⁺ cells) (Nilssen et al., 2022). Thus, investigation of the role of cell-specific circuits in LEC in temporal memory will further help understand the neural mechanisms of episodic memory.

While classical methods for investigating neural circuit connectivity and functions, such as lesions, electrical stimulation and micro-injections of neuronal tracers lack the spatial and temporal resolution, the development of genetic targeting and manipulation techniques has enabled the study of connectivity and functions of genetically defined cell-type pathways (Luo et al., 2008; Yamamoto et al., 2021). Recently developed neuroscience techniques will further contribute to understand the EC-HPC network. While cell-type-specific molecular markers have been intensively identified (Kitamura et al., 2014; Kohara et al., 2014; Luo et al., 2008; Ramsden et al., 2015; Suh et al., 2011), high-throughput cell-type classification and targeting cells of which markers are currently not identified, such as SL-IN in CA1 or subpopulations of entorhinal pyramidal cells, will be aided by recent advances in transcriptome methods like multiplexed fluorescence in-situ hybridization (Chen et al., 2015; Lein et al., 2017; Luo et al., 2018; Moses & Pachter, 2022) or the coupling with lineage tracing (Matho et al., 2021; Musall et al., 2023; Zeng, 2022). Although retrograde trans-synaptic tracing has been conducted with rabies viruses (Callaway & Luo, 2015; Wickersham et al., 2007), development of anterograde trans-synaptic tracer (Tsai et al., 2022) will efficiently elucidate targeting cells from specific cell-type in EC. Moreover, while conventional tracing experiments require patch-clamp experiments to investigate actual monosynaptic connections (Kitamura et al., 2014; Kohara et al., 2014), the technique of green fluorescent protein reconstitution across synaptic partners (GRASP) allows the identification of the location of synaptic connectivity in neural circuits (Kim et al., 2011). The combination of dual-color GRASP and engram-tagging technique (Choi et al., 2018; Lee et al., 2023) or neuronal inhibition marker (Dong et al., 2023) during different learning tasks (e.g., space or object vs. time) will visualize the difference of synaptic connectivity relating to different types of memory. The oscillatory activity of cell-type specific pathway can be further elucidated by extracting pathway-specific activity from local-field-potential data using blind-source separation technique (Fernandez-Ruiz et al., 2012; Fernandez-Ruiz et al., 2017; Fernandez-Ruiz et al., 2021; Herreras et al., 2015; Makarova et al., 2011; Schomburg et al., 2014) or by utilizing genetically encoded voltage indicators (Cho et al., 2020; Marshall et al., 2016). These technologies promote understanding of detailed cell-type specific connections in MEC-HPC circuit and their contributions to episodic memory.

Abbreviations

AD

Alzheimer’s disease

BLA

basolateral amygdala

CA1

hippocampal area Cornu Ammonis 1

Cal

calbindin

CA2

hippocampal area Cornu Ammonis 2

CA3

hippocampal area Cornu Ammonis 3

Ctip2

COUP-TF interacting protein 2

Cplx3

complexin 3

Ctgf

connective tissue growth factor

DG

dentate gyrus

EC

entorhinal cortex

ECIII

entorhinal cortex layer III

Etv1

ETS variant 1

GABA

gamma-aminobutyric acid

GRASP

green fluorescent protein reconstitution across synaptic partners

HPC

hippocampus

LEC

lateral entorhinal cortex

MEC

medial entorhinal cortex

MECII

medial entorhinal cortex layer II

MECIII

medial entorhinal cortex layer III

MECV

medial entorhinal cortex layer V

MECVa

medial entorhinal cortex layer Va

MECVb

medial entorhinal cortex layer Vb

MECVI

medial entorhinal cortex layer VI

mGluR1

metabotropic glutamatergic receptor 1

mPFC

medial prefrontal cortex

Nxph4

neurexophilin 4

PFC

prefrontal cortex

PV

parvalbumin

REM

rapid eye movement

SL

striatum lacunosum

SL-IN

interneurons in stratum lacunosum

SLM

striatum lacunosum-molecular

SM

striatum molecular

SOM

somatostatin

SR

striatum radiatum

SWR

sharp-wave ripple

TAL

temporal association learning

TeTX

tetanus toxin light chain

TrkB

tropomyosin receptor kinase B

TFC

trace-fear-conditioning

Wfs1

Wolfram syndrome 1