We have all experienced the recall of a specific memory when we are exposed to a similar situation; the feeling of a warm breeze on a beach might remind you of a summer day from your childhood when you enjoyed swimming with your family and a barbecue. It occurs because the target (the act of swimming with your family in this case), was memorized alongside multiple types of information that occurred at that moment (a warm breeze on the beach in this case). While we may focus on a specific event or person in the moment, many other pieces of information occurring simultaneously around the target become enmeshed in memory formation. Those other information streams can serve as a hint to facilitate recall of the target memory. Together, these are referred to as the “context” (Godden and Baddeley, 1975; Smith, 1979; Smith et al., 2004; Manns and Eichenbaum, 2009; McKenzie et al., 2014; Robin et al., 2018; Libby et al., 2019; de Voogd et al., 2020). Context can shape our decisions and our recall processes (Godden and Baddeley, 1975; Smith, 1979; Smith et al., 2004; Robin et al., 2018; de Voogd et al., 2020), and has been shown to be an important first step in processing and rebuilding of episodic memories, helping to streamline object representations (Manns and Eichenbaum, 2009; McKenzie et al., 2014; Libby et al., 2019), and has a role in the determination of motivation and valuation of actions and items (Zeithamova et al., 2018). Context includes external information and internal states (Spear, 1973; Bouton, 1993; Rudy et al., 2004; Maren et al., 2013). In laboratory observations of contextual learning, the external components collected by sensory systems such as visual information (e.g., a color of paint on the wall of a room, dark or brightness of light), odor, sound, and touch (e.g., texture of a floor) contribute to the formation of spatial-contextual information for animals. Thus, the external components prove easier to manipulate, and are often used as the primary means of modifying contexts. We can describe the context as the specifics of a place that are not the place itself. For example, it can be understood that the context changes if the color of paint on the walls in the room is changed, but the space of the room itself does not. The Internal elements can include emotions (e.g., happiness, fear, sadness, and anger etc.) and hormonal states such as hunger or stress experienced within the situation. These internal sorts of contextual stimuli are important; just as ones experience of a situation may vary with the context of an environment (dark vs. light) for example, ones experience of an event may differ depending on whether one is stressed or calm, or whether one is angry. The effects of these internal states can predispose a circuit to be more or less responsive to a given cue (Jezek et al., 2010; Maren et al., 2013; Tye, 2018). In animal behavioral models, there are various types of behavioral experiments in which an animal uses these “contextual” data to perform a task. Broadly, multiple tasks in which animals use place memory can be included as part of a context dependent memory paradigm such as contextual fear conditioning (CFC), passive avoidance, mazes, open field experiments etc. In this review, we focus primarily on CFC paradigms which are one of the most common behavioral tasks for assessing contextual memory, and highlight the brain regions that are involved in contextual memory processing. Among the many brain regions involved in contextual memory, the EC-hippocampal network has been widely studied across a number of behavioral paradigms, and is known to have a strong role in episodic memory formation. As a result, a great deal is known about the physiological processes that support learning and memory within this region (Scoville and Milner, 1957; Mahut et al., 1982; Kim and Fanselow, 1992; Phillips and LeDoux, 1992; Frankland et al., 1998; Eichenbaum, 2000; Tulving, 2002; Sutherland et al., 2008; Kitamura et al., 2012; Pilkiw et al., 2017). To that end, we begin with a main focus on the EC-HPC network as the most well studied region involved in contextual memory before shifting our focus to the large array of associated regions. We then discuss the contributions of engram assemblies both within and outside of the EC-HPC network to context-dependent memory.

Many different brain regions are involved in contextual fear memory acquisition, consolidation, and retrieval (Figure 2). In particular, there have been many reports on the necessity of EC-HPC formation in contextual memory processes. The HPC has been shown to be involved in the differentiation of contexts (Frankland et al., 1998), and integration of contextual information with learned behaviors and responses (Good and Honey, 1991; Honey and Good, 1993; Freeman et al., 1997; Smith et al., 2004; Kim et al., 2012). Lesion of the HPC disrupts CFC (McEchron et al., 1999) in a manner that is time sensitive, with recently acquired memories being destroyed by lesion, and longer term contextual memories being undamaged (Kim and Fanselow, 1992; Anagnostaras et al., 1999; Lee et al., 2017). Pharmacological inactivation of the dorsal HPC impairs context-dependent memory recall (Corcoran and Maren, 2001; Barrientos et al., 2002). On the other hand, there are also reports that contextual fear memories can be acquired even in the absence of the HPC. These findings raise the question of how critical the HPC is in the learning and recall of contextual fear memory. Several groups have performed contextual fear conditioning in HPC lesioned rodents to demonstrate that animals could indeed acquire contextual memory even in the absence of the HPC, however, this hippocampal independent conditioning required more shock events or more learning sessions to account for decreased efficiency of acquisition (Wiltgen et al., 2006; Lehmann et al., 2009). Depending on conditions of the experiment, these hippocampal independent memories can be less robust; disappearing more quickly that those generated with the hippocampus intact (Zelikowsky et al., 2012), while other studies demonstrate that memories generated without the hippocampus can be robust, lasting as long as thirty days (Lehmann et al., 2009; Gidyk et al., 2021). These results suggest that the acquisition of contextual fear may have redundant or compensatory systems, potentially explaining why hippocampus independent learning requires more shock events or more learning sessions (Wiltgen et al., 2006; Lehmann et al., 2009; Chaaya et al., 2018). The brain has a robust ability to compensate for damage, and these reports are presumed to be the result of the ability to acquire contextual fear memory by supplementing the absence of the HPC with other brain regions to counterbalance the disruption of normal operations. Entorhinal Cortex (EC), a major contributor to the functionality of the hippocampal circuit, is also known to play a key role in contextual memory processes, and it is reported that the lesion of EC results in decreased contextual learning, but not avoidance learning (Freeman et al., 1997). Chemical inactivation of the MEC also leads to alterations in contextual reconsolidation and extinction, supporting the regions’ role in processing contextual information (Baldi and Bucherelli, 2014). These two regions (EC-HPC) are part of a tightly interconnected network, which is collectively the most studied region involved in processing of contextual and episodic memories.

Taken together, these reports highlight that the EC-HPC formation plays a core role in contextual memory processes with a broader network supporting the expression of contextual memories following their initial formation. In the following sections, we mainly focus on EC-HPC formation as one of best studied contributing brain regions in the processes of contextual memory, and discuss their neuronal circuit connectivity and physiological mechanisms in relation to contextual memory. We describe other relevant brain regions, and how the individual components are involved in CFC as well.

In addition to the EC-HPC network, a number of other brain regions were reported to contribute to contextual memory formation (Figure 2), either by contributing to the acquisition and retrieval of context memory directly, or regulating EC-HPC networks.

The basolateral amygdala (BLA) has been demonstrated to play a role in contextual fear conditioning, primarily by generating the emotional valence associated with the context-shock pairing. HPC interacts with the BLA via MECVa to encode contextual information during encoding of contextual fear memory. Not only the dorsal HPC (dHPC), but also the ventral CA1 (vHPC) projections to the Basal amygdala, paired with aversive stimuli, contribute to encoding conditioned fear memory (Kim and Cho, 2020). Hippocampal projections to the amygdala are involved in recent memory recall and association of contextual stimuli with pain. BLA receives foot-shock US information via the ascending pain pathways including the periaqueductal gray (PAG) and the parabrachial nucleus (PBN) and auditory CS information via thalamus (Kim et al., 2013; Lee, 2015; Pellman and Kim, 2016) and projects not only back to the HPC, but to the PFC (Kitamura et al., 2017).

Kitamura et al. (2017) demonstrated that in addition to receiving amygdalar input, PFC also receives contextual fear information from HPC via MECVa cells and that PFC engram cells are generated during the CFC encoding period. With optogenetic manipulation of both pathways, HPC-MECVa to PFC and BLA to PFC, it was revealed that both pathways are crucial for the generation and maturation of PFC engram cells following CFC, and are key for formation and recall of remote memory. Moreover, contextual activity in the PFC has been demonstrated to directly influence amygdalar activity (Orsini et al., 2011, 2013; Maren et al., 2013; Zhang et al., 2018).

Beyond the role of the Medial Geniculate Nucleus (MGN) in routing tonal information to the amygdala (Lee, 2015), the thalamic nuclei are also known to be involved in processing contextual information. Fear memories stored in PFC are either heightened or dampened in the nucleus reuniens (NR) of the thalamus prior to routing to HPC, resulting in a control of generalization of fear memory to differential contexts (Richter-Levin and Akirav, 2000; Xu and Sudhof, 2013; Rozeske et al., 2015). Additionally, a sparse inhibitory projection from CA3 to the anterodorsal thalamic nucleus is necessary for contextual fear memory retrieval at remote, but not recent time points (Vetere et al., 2021).

The Retrosplenial cortex (RSC) contributes to spatial navigation, head direction, as well as learning and memory for discrete cues such as auditory or visual stimuli. Lesion studies demonstrate that RSC processing contributes to configuration of contextual representation building (combining of elements). This is particularly interesting considering the previously demonstrated roles of RSC in hippocampal place field remapping (Wyss and Van Groen, 1992; Cooper and Mizumori, 2001; Mao et al., 2017; Todd et al., 2017; Alexander et al., 2020). Optogenetic inhibition of dHPC terminals in the RSC during memory acquisition causes impairment of subsequent memory performance (Opalka and Wang, 2020).

The Lateral septum (LS) is downstream from the HPC and has been implicated in various functions such as mood, motivation, anxiety, and spatial behavior, as reviewed in Wirtshafter and Wilson (2021). HPC terminal inhibition in LS during memory acquisition or retrieval tells us the dHPC-LS pathway plays a critical role in both contextual memory acquisition and retrieval (Opalka and Wang, 2020; Opalka et al., 2020). It thought that LS is important for matching context with behavior by taking contextually relevant information from the hippocampus and routing that information to the amygdala (Penick and Solomon, 1991; Staiger and Nurnberger, 1991; Deng et al., 2019; Wirtshafter and Wilson, 2021). Additionally, LS and full basal forebrain lesions can result in inappropriate pairing of context and movement response (Vouimba et al., 1998; Knox and Keller, 2016).

Recently the involvement of neuromodulatory inputs in contextual memory has begun to attract attention. It was reported that the Locus coeruleus (LC) tyrosine-hydroxylase-expressing (TH⁺) neurons project strongly to the HPC, and release dopamine. Optogenetic activation of LC TH⁺ neurons increased dopamine release into dHPC, promoting novelty associated spatial learning and memory (Kempadoo et al., 2016; Takeuchi et al., 2016). We expect further elucidation of relevant brain regions and understanding of the mechanism of contextual memory from the perspective of the entire brain.

Neural representations of context, dependent on the repeated activation of neural ensembles, remain stable over a period of time to facilitate recall and integration. It is understood that repetitive exposures to the same context activates the same set of cells as assessed using the expression of immediate early genes (IEGs) such as c-Fos, Arc (activity-regulated cytoskeleton-associated protein) or Zif268 (Wilson and McNaughton, 1993; Radulovic et al., 1998; Guzowski et al., 1999; Guzowski, 2002). German evolutionary zoologist Richard Semon first posited the term “engram” to describe a hypothetical physical substrate for memory storage (Semon, 1921; Schacter et al., 1978). While Semon’s idea was impossible to prove in the early twentieth century, recent technologies have made it possible to revisit his idea and the definition was updated to fit our modern understanding of memory. The following three criteria must be met for an event to be considered as engram (Josselyn, 2010; Tonegawa et al., 2015, 2018; Denny et al., 2017; Josselyn and Tonegawa, 2020). An engram (i) is an enduring and significant physical or chemical alteration to a neural network (ii) due to activity in a subset of neurons caused by episodic stimuli (iii) that can be reactivated following presentation of all or part of the original stimulus set, leading to memory recall. Technological advances have allowed for the targeting of individual cell subtypes by gene expression, which can include constitutively active proteins or more transient expression caused by neuronal activity (Guzowski et al., 1999; Guzowski, 2002; Han et al., 2007, 2009; Reijmers et al., 2007; Guenthner et al., 2013; Roy et al., 2017a).

By utilizing IEG promoters to allow activity-dependent labeling of cells, and pairing this technique with optogenetically driven expression of opsins in previously activated cells, Liu et al., in 2012 made first observations of an “engram cell” under the modern definition (Reijmers et al., 2007; Liu et al., 2012). DG cells which had been active during CFC were targeted with ChR2, and stimulated with blue light at 20Hz were reactivated, and this reactivation artificially induced contextual memory recall without re-exposure to the original context information (Liu et al., 2012; Figure 4). Engram cells have since been found to have synapses with other engram cells resulting from the same memory event in the HPC (Choi et al., 2018). In this way, engrams can be further described as networks of individual engram cells which act together to store the components of memories, and reactivation of that network is both necessary and sufficient for recalling that encoded memory (Ryan et al., 2015; Tonegawa et al., 2015; Holtmaat and Caroni, 2016; Denny et al., 2017; Figure 4). Thus, we can understand an engram as the physical record of a representation, with the activated ensemble becoming reactivatable within a given set of stimuli (context) resulting in recall (Kelemen and Fenton, 2010; Jezek et al., 2011; Tayler et al., 2013). Importantly, engrams are distinctly encoded independently of other memory modalities (e.g., A spatial engram or a contextual engram) (Tanaka et al., 2018; Ghandour et al., 2019; Marks et al., 2021). Moreover, engram activation can even be sufficient to induce a false memory in the form of generalization of a fear response in one context onto a neutral context (Ramirez et al., 2013; Ohkawa et al., 2015; Oishi et al., 2019), or even to construct a representation of a context that has never been physically experienced (Garner et al., 2012). Kitamura et al. (2015) demonstrated how hippocampal engram cells are activated during memory recall. They found that the neural input from Reelin⁺ cells, but not CalB⁺ cells, in MEC drives contextual information into the HPC, is necessary for the formation and recall of CFC memory. They further showed that the neural input from Reelin⁺ cells in MEC is necessary for the reactivation of memory engram cells encoding CFC memory in the hippocampal CA3. These results indicate that the hippocampal trisynaptic pathway driven by Reelin⁺ cells in MEC layer II is crucial for CFC memory.

Recently, the question of when engram cells are formed during the process of memory acquisition and long-term storage has come into focus. Kitamura et al. (2017) found that engrams are actually formed simultaneously in the early stage of context memory acquisition in both the HPC and the medial prefrontal cortex (mPFC) where long-term memory is stored (Kitamura et al., 2017). This result went against the Standard Consolidation Model, which states that newly acquired episodic memories are initially stored in only HPC and then transferred to cortex as long-term memories, accompanied by “deletion” of HPC engrams with time. The mPFC engram cells which were generated in the early stage are not reactivated during the recent memory recall state but exist as “silent engrams” like those previously found by Ryan et al. (2015). These silent engrams become functionally mature with time, simultaneous with transition of the memory into the basolateral amygdala (BLA) which is necessary for fear memory, whereas HPC engram cells gradually became silent with time (Kitamura et al., 2017). This is an important finding, as it begins to explain how the long term storage of contextual memory remains stable, despite the observed shift in the active cell population makeup of HPC-stored representations over time (Mankin et al., 2012; Ziv et al., 2013); The hippocampal representations may need to be supported by representations constructed from engrams elsewhere in the brain, and that the maturation of the mPFC engrams over time may compensate for the decorrelation of hippocampal representations. Matos et al. (2019) demonstrated that activity of a small subset of mPFC neurons is sufficient and necessary for remote memory expression, and selective disruption of cAMP response element-binding protein (CREB) function in mPFC engram cells after CFC impairs remote memory formation (Matos et al., 2019), while another recent study suggests that the neural population in mPFC activated during memory recall changes over time (DeNardo et al., 2019). While further studies will be necessary for the understanding of mPFC engram cells for remote memory formation, these studies suggest that contextual information activates a set of neural ensembles and generates engrams in multiple brain regions during learning that are functionally connected (Kitamura et al., 2017; Roy et al., 2017b; Choi et al., 2018; DeNardo et al., 2019; Matos et al., 2019; Pignatelli et al., 2019). Recent advances in technology have the potential to generate an even deeper understanding of engram ensembles. Roy et al. (2019) applied a combination of genetic access to activated neurons labeled by the c-Fos promotor with a tamoxifen dependent Cre recombinase CreER^T2 through the whole brain in a Fos-TRAP (targeted recombination in active populations) mouse line (Guenthner et al., 2013). All neurons activated during the CFC protocol within a narrow (< 12 h) time window were labeled with tdTomato fluorescence, then those brains were made transparent using tissue clearing techniques and imaged intact at the whole-brain scale. The mapping and subsequent optogenetic manipulations revealed new contextual engram populations in the anteromedial thalamus and the nucleus reuniens. They found that simultaneous chemogenetic reactivation of these multiple engram assemblies conferred a greater level of memory recall than reactivation of a single engram ensemble (Roy et al., 2019). Engram cells encoding CFC memory have also been identified in RSC (Cowansage et al., 2014), and optogenetic activation of those engram cells facilitates the systems consolidation of CFC memory (de Sousa et al., 2019). These results show that CFC engrams exist in brain regions other than the classically highlighted areas. Understanding the mechanisms of contextual memory at the systems level, such as ensemble activity or synchrony between brain regions will be of great interest for future studies.

In this review, we have discussed the behavioral neuroscience of contextual fear conditioning in the EC-HPC network. Contextual memory has been found to be a unique part of episodic memory, encoding information about the conditions within a space in such a way as to serve as a hierarchical basis for encoding and indexing memories. Contextual information is fed to the HPC largely by Reelin⁺ cells in MEC layer II via the trisynaptic pathway. This information is then used to form unique ensembles of activity that have the ability to nest within representations of locations. Questions still remain as to the nature of the local processing of contextual data within the HPC, but recent advances in technologies for cell type specific genetic manipulations and for in vivo recording of cellular activity can provide powerful opportunities for investigation of these local circuit mechanisms. For example, the in vivo visualization of calcium transients during behavior, or the stable recording of cellular arrays across multiple brain regions simultaneously that have recently become possible have the potential to yield critical new insights into contextual learning processes as context dependent memory is composed of multiple elements, and simultaneous measurement of physiological neuronal activity could tell us which sub-component is driving the changes in neuronal activities. At the same time, utilizing cell type specific techniques to visually identify neuronal activity in specific subsets of hippocampal neurons could be of great use in further developing our understanding of the local processes that shape contextual encoding.

WM, TK, and SO conceived this project. WM, JY, TK, and SO wrote the manuscript and approved the final manuscript. WM and SO designed and created the figures. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.