Discrete circuits of the ventral hippocampus in threat-based learning and memory

Abstract

While the hippocampus has intrigued generations of neuroscientists for its contributions to cognitive and emotional processing, functional specialization along its longitudinal axis confers particular importance to the ventral hippocampus (vHPC) in affective regulation under normal and pathological conditions. In particular, vHPC is extensively linked to the encoding, expression and extinction of fear memories, which mediate behavioral adaptation to environmental threats. Despite decades of research, however, many questions remain about precisely what is encoded among specific populations of vHPC neurons and what brain systems cooperate in processing this information during fear regulation. Furthermore, as insights accumulate into the function of discrete output projections of vHPC, an important area of focus is how vHPC circuitry might be organized to support different output patterns through synaptic integration. Here we summarize current understanding of these issues based on contemporary circuit-based approaches and highlight potential clues to the anatomical and functional organization of synaptic networks that may help further understanding of vHPC as a system of interacting modules.

1. Introduction

Fear conditioning is a fundamental form of learning that animals use to detect and defend themselves from dangerous situations. Conditioned stimuli in this paradigm elicit an array of defensive reactions that are primarily constrained through extinction and figure prominently in psychiatric conditions (B. M. Graham & Milad, 2011; Jovanovic & Ressler, 2010; VanElzakker, Dahlgren, Davis, Dubois, & Shin, 2014). Among brain regions implicated in these processes, the hippocampus is highly conserved between humans and animal models, which in turn exhibit alterations in activity and structure linked to psychopathology (Fanselow & Dong, 2010). While much early work on this system in rodents has been focused on its dorsal subregion (dHPC) and its role in spatial navigation and episodic memory, the hippocampus exhibits substantial anatomic, molecular, and functional heterogeneity along its dorsal-ventral axis (or anteroposterior axis in humans) (Poppenk, Evensmoen, Moscovitch, & Nadel, 2013; Strange, Witter, Lein, & Moser, 2014). In particular, the ventral hippocampus (vHPC) possesses unique anatomical connections and stimulus-response properties, which are increasingly implicated in defensive behavioral control. Compared to dHPC, vHPC has also been more extensively involved in motivational and emotional aspects of learning, as well as their state-dependent regulation (Fanselow & Dong, 2010). This has invigorated efforts to unravel neural circuits in vHPC encoding threat versus safety under normal and pathological conditions. Of principal importance is how emotionally salient stimuli are represented by distinct neuronal populations and which vHPC outputs respond to discrete and contextual cues to engage relevant downstream networks.

In this review, we focus on the major output structure of vHPC, the ventral cornu ammonis 1 (vCA1) region, which is broadly connected with a distributed brain network underlying emotional regulation. The vCA1 receives information about multimodal sensory stimuli, interoceptive brain states, and spatiotemporal context through glutamatergic synapses of intra- and extrahippocampal pathways, which recruit subsets of principal excitatory neurons (PNs) at discrete phases of behavior. Early work conceptualized individual hippocampal PNs as having equal potential for mnemonic coding, distributed across the network, but recent work suggests the existence of more localized processing capabilities in which groups of projection-defined PNs differ not only in function but also in intrinsic wiring. Moreover, a low abundance of recurrent excitatory synapses allows for an outsized role of GABAergic interneurons in CA1 input-output integration. These features may be key to resolving the circuit logic underlying the multifaceted contributions of vHPC to experience-dependent control of defensive behavior.

2. Ventral HPC in Conditioned Fear Regulation

Learning about aversive experiences involves multisensory, spatiotemporal, and interoceptive information, which may engage different brain networks for encoding distinct memory components. Apart from such functional specialization, learning demands unique resources at different stages to support not only the acquisition and retrieval of learned associations but also their updating through extinction. Indeed, flexibility of conditioned fear behavior is a crucial evolutionary adaptation and a hallmark of successful recovery from trauma.

Stimulus associations in fear conditioning can be established from either a discrete cue or a context, which is a conjunctive representation of discrete elements (Maren, Phan, & Liberzon, 2013). While the hippocampus contains neurons that respond to both types of stimuli, it is thought to be more fundamentally adapted for cognitive maps, such as spatial trajectories. This is generally true for both dHPC and vHPC, but larger place fields exhibited by vHPC neurons suggest a relatively low capacity for this type of coding (Chockanathan & Padmanabhan, 2021; Jung, Wiener, & McNaughton, 1994; K. B. Kjelstrup et al., 2008; Komorowski et al., 2013; Royer, Sirota, Patel, & Buzsaki, 2010). At the same time, evidence indicates that vHPC is, to a much greater extent, engaged by experiences with emotional content (Bannerman et al., 2004; Fanselow & Dong, 2010). This lends mystery to precisely what types of representations are contained in vHPC, how they are used by the animal, and under what circumstances.

2a. Fear memory acquisition and expression

Fear conditioning can be readily acquired when a discrete conditioned stimulus (CS) is paired with an aversive unconditioned stimulus (US), either when these events directly coincide (delay conditioning) or when they are separated by a brief period (trace conditioning). Regardless of whether a discrete CS is available to signal the occurrence of a US, conditioning is also acquired to the context in which a US is experienced, but the underlying mechanisms may differ depending on whether the context is being used as a foreground or background cue. Lesions of vHPC before auditory fear conditioning, in which the context serves as a background cue, indicate that delay, trace and contextual fear are all impaired by vHPC lesions (Hunsaker & Kesner, 2008; Maren, 1999; Maren & Holt, 2004; Richmond et al., 1999; Rogers, Hunsaker, & Kesner, 2006; Trivedi & Coover, 2004; Yoon & Otto, 2007), but see (Trivedi & Coover, 2006). In foreground contextual conditioning, where no discrete CS is presented, vHPC lesions before learning generally do not affect conditioning (Bannerman et al., 2003; Biedenkapp & Rudy, 2009; K. G. Kjelstrup et al., 2002; Maren & Holt, 2004), although deficits have been reported when either foot shock or predator odor served as the aversive US (Pentkowski, Blanchard, Lever, Litvin, & Blanchard, 2006; M. E. Wang et al., 2013). Post-conditioning lesions of vHPC impair auditory-cued delay as well as trace conditioning (Maren & Holt, 2004; Yoon & Otto, 2007). In addition, in contrast to those performed before training, post-training lesions impair contextual memory acquired through foreground conditioning (Maren & Holt, 2004). This may indicate some potential for functional compensation when vHPC is absent during learning but not after it has participated in memory encoding.

While lesions are a blunt instrument with significant drawbacks, pharmacological approaches have yielded similar, albeit variable, effects. Inactivation of vHPC by the GABA_A agonist muscimol impairs both acquisition and retrieval of trace conditioning (Cox, Czerniawski, Ree, & Otto, 2013; Czerniawski, Yoon, & Otto, 2009; Esclassan, Coutureau, Di Scala, & Marchand, 2009; Gilmartin, Kwapis, & Helmstetter, 2012). Regarding delay conditioning, muscimol yields inconsistent outcomes. While some find pronounced deficits following pre-training infusions (V. M. Chen, Foilb, & Christianson, 2016; Esclassan et al., 2009; Maren & Holt, 2004), others report no effect (Bast, Zhang, & Feldon, 2001; Cox et al., 2013; W. N. Zhang, Bast, Xu, & Feldon, 2014). Likewise, following acquisition of delay conditioning, reports conflict on whether muscimol yields an increase or decrease in cue-elicited freezing (V. M. Chen et al., 2016; Maren & Holt, 2004; Sierra-Mercado, Padilla-Coreano, & Quirk, 2011; Sotres-Bayon, Sierra-Mercado, Pardilla-Delgado, & Quirk, 2012). Finally, muscimol prior to contextual fear conditioning attenuates memory expression according to some studies (Bast et al., 2001; Heroux, Horgan, Pinizzotto, Rosen, & Stanton, 2019; W. N. Zhang et al., 2014), but not others (Biedenkapp & Rudy, 2009; Maren & Holt, 2004).

Collectively, these studies demonstrate that the vHPC is involved in multiple forms of fear learning, but these contributions may be dispensable under certain behavioral conditions. Alternatively, manipulations targeted at the vHPC may sometimes preserve its critical functions due to the sheer size of this structure and the fact that discrete cell types and anatomical connections are not homogeneously distributed along its geometric axes. Indeed, outputs arise from major subregions of vHPC, including CA3, CA1, and subiculum, which are further organized into discrete domains. In this review, we concentrate on area CA1 due to its abundant projections to other fear-related structures, which represent plausible substrates for mnemonic control. It is therefore important that neurons in vCA1 display increased expression of activity-dependent immediate-early genes following cued and contextual fear conditioning (Brockway, Simon, & Drew, 2023; W. B. Kim & Cho, 2017; Lacagnina et al., 2024; Park & Chung, 2019; Shpokayte et al., 2022; Silva, Burns, & Graff, 2019), indicating that this region is active during memory encoding. In addition, fear conditioning correlates with plasticity in the stimulus-response properties of excitatory and inhibitory vCA1 neurons (Jimenez et al., 2020; Lacagnina et al., 2024; K. Li, Koukoutselos, Sakaguchi, & Ciocchi, 2024; Nguyen, Koukoutselos, Forro, & Ciocchi, 2023; Suthard et al., 2024), suggesting that dissection of specific vCA1 circuitry may help disambiguate the role of vHPC in memory encoding and potentially reconcile prior conflicting results.

2b. Fear memory extinction

Once a fear association has been acquired, it is subject to distinct forms of regulation affecting its strength, specificity, and expression. Extinction is one example of such behavioral updating that occurs when a conditioned stimulus is repeatedly encountered without aversive consequences, resulting in persistent fear attenuation. By restoring exploration of safe environments, extinction plays an important adaptive role in survival and facilitates recovery from trauma. However, expression of the original threat association can spontaneously resurface. Consequently, conditioning and extinction are considered complementary forms of learning that compete for control over behavior, a dynamic process in which studies paint a consistent picture of hippocampal involvement.

Some of the earliest evidence that vHPC regulates behavioral outcomes of extinction was obtained from muscimol infusions, which block the renewal of extinguished auditory fear (Hobin, Ji, & Maren, 2006). Subsequent work revealed that vHPC inactivation before extinction learning impairs retrieval of auditory fear memory as well as acquisition of extinction, as reflected by higher freezing levels during subsequent extinction retrieval (Sierra-Mercado et al., 2011). Consistent with its involvement in extinction learning, low-frequency stimulation of vHPC during contextual extinction training reduces fear during subsequent extinction retrieval (Cleren et al., 2013). This pattern defies a simple explanation in which vHPC mediates the storage and/or retrieval of either fear or extinction memories. Rather, vHPC activity contributes to both types of learning and can promote either a high or low fear state depending on behavioral history and the circumstances under which the stimulus is encountered. Among the unique possibilities is that vHPC might bias the selection of competing stimulus-outcome associations, which may aid refinement of defensive behavior to match environmental contingencies and restrict fear expression to increasingly circumscribed conditions.

A major factor vHPC relies on to discriminate these conditions is context. It has long been noted that extinction of auditory fear conditioning is readily expressed in the extinction arena, but fear returns when the same stimuli are encountered in other distinct environments, a phenomenon termed renewal (Maren et al., 2013). Thus, while conditioning to a discrete cue can be acquired and expressed in a context-independent manner, extinction relies on a contextual match to the training conditions for successful retrieval. In addition, relapse can be triggered by passage of time (i.e., spontaneous recovery), stress (i.e., reinstatement), and changes in interoceptive state, including those induced by pharmacological agents (Bouton, 2002; Clem & Schiller, 2016). Others have speculated that these triggers constitute features of the extero- and interoceptive context and, as such, they may share an underlying neurophysiological basis (Goode & Maren, 2014). If so, vHPC seems like an ideal system for representing this information, given its importance in contextual processing and its integration of autonomic, visceral, and stress-related signals. At least some of these factors bear relevance to both discrete and contextual stimuli, since perception of internal state and time can occur independent of physical context and lead to a mismatch between learning and retrieval, for example during reinstatement and spontaneous recovery. We are a long way from understanding how vHPC might integrate this information to influence distributed memory networks, and one major reason is that the computational logic of vHPC circuits remains very poorly understood.

3. Ventral CA1 long-range circuits in fear regulation

3a. Amygdala

Several interconnected amygdala regions are highly implicated in fear conditioning and extinction: the striatal-like central nucleus of the amygdala (CeA), and the cortex-like lateral (LA), basolateral (BLA), and basomedial (BMA) nuclei. vCA1 projections are diffusely targeted across these structures and form monosynaptic connections with resident cell populations. Imaging studies indicate that the vCA1 neurons projecting to BLA (vCA1-BLA), and potentially collateralizing to other nuclei, are preferentially activated by foot shocks, and following conditioning, their context-evoked activity becomes correlated with surrounding vCA1 populations (Jimenez et al., 2020). Activity of vCA1-BLA neurons is required for acquisition and retrieval of contextual fear conditioning (Brockway et al., 2023; Jimenez et al., 2020; W. B. Kim & Cho, 2017; C. Xu et al., 2016). However, this pathway can support either positive or negative contextual associations (Shpokayte et al., 2022), suggesting that vCA1-BLA neurons might comprise several distinct populations or otherwise acquire specific valence properties through experience-dependent plasticity. Synapses formed onto excitatory BLA neurons are potentiated following contextual association with a shock US (J. Graham et al., 2021; W. B. Kim & Cho, 2020). In addition, vCA1-BLA projections drive feedforward inhibitory transmission, which is crucial for theta-frequency entrainment of BLA circuits involved in memory encoding and retrieval (Bazelot et al., 2015; Hubner, Bosch, Gall, Luthi, & Ehrlich, 2014).

Ventral CA1 projections to BLA also contribute to distinct aspects of cued fear memory. Electrode recordings reveal that a subpopulation of BLA neurons develops CS responses during auditory fear conditioning, but during subsequent extinction, this firing is attenuated while other, previously silent, neurons become CS responsive (Herry et al., 2008). This indicates that extinction does not merely reverse conditioning-induced changes but is encoded as an orthogonal BLA representation. Indeed, subsequent retrieval and fear relapse elicit switching between these populations, signaling high and low fear states. Orthodromic stimulation of vCA1-BLA firing indicates that fear-related BLA neurons are more likely to receive synaptic connections from vCA1 (Herry et al., 2008). Context-dependent fear renewal induces c-Fos in vCA1-BLA neurons (Jin & Maren, 2015; Orsini, Kim, Knapska, & Maren, 2011), while disconnection lesions of vCA1-BLA pathways prevent renewal-induced freezing (Orsini et al., 2011). These data suggest that vCA1-BLA neurons rely on specific contextual information to support CS-related BLA activity, imposing a gate on fear expression. However, another report indicates that optogenetic silencing of vCA1-CeA but not vCA1-BLA terminals blocks context-dependent renewal of auditory fear (C. Xu et al., 2016). Both experimental limitations and species differences might play a role in this discrepancy, but one technical issue deserving more attention is the possibility that vCA1 neurons collateralize to multiple amygdala nuclei. Providing there are reciprocal connections between the amygdala and vHPC, deficits in fear renewal after anatomical disconnection might also result from disruption of BLA-vCA1 transmission.

Indeed, reciprocal BLA projections carry information in the opposite direction to vCA1, but not dCA1 (Tao et al., 2021), and can be activated by auditory CSs of either positive or negative valence (Beyeler et al., 2016). Of potential relevance, excitatory BLA PNs have been differentiated into functionally and genetically distinct subtypes with complementary anatomical distribution (J. Kim, Pignatelli, Xu, Itohara, & Tonegawa, 2016; X. Zhang, Kim, & Tonegawa, 2020). R-spondin 2+ (Rspo2+) cells, which are more concentrated in the anterior BLA (aBLA), promote contextual freezing, whereas protein phosphatase 1-regulatory inhibitor subunit 1B+ (Ppp1r1b+) neurons populate the posterior BLA (pBLA) and support context fear extinction in addition to appetitive conditioning. These results suggest that BLA-vCA1 neurons may have pre-specified valence properties and contribute to bidirectional fear regulation. However, because projections in this pathway arise predominantly from pBLA, it is possible that BLA-vCA1 transmission is primarily important for extinction. Indeed, photoinhibition of BLA-vCA1 projections impairs extinction of both cued and contextual fear as well as vCA1 coding of the extinction context (Nguyen et al., 2023).

3b. Medial prefrontal cortex

The rodent medial prefrontal cortex (mPFC) is extensively implicated in fear learning and comprises several cytoarchitectonic regions, including the prelimbic (PL), infralimbic (IL), dorsal peduncular (DP), and anterior cingulate cortices (ACC). Although their precise homology is debated, these areas form synaptic networks with BLA and vCA1 and, while they are relatively devoid of projections to the hippocampal fields, they maintain indirect connections via thalamic nuclei and receive abundant hippocampal input originating mainly from vHPC.

The dorsomedial mPFC, including the PL and cingulate area 1 (Cg1), plays critical roles in memory encoding and expression, involving plasticity and excitatory PNs as well as inhibitory interneurons (Herry & Johansen, 2014). Some of the earliest studies employing c-Fos labeling and anatomical disconnection suggested that vCA1-PL projections figure more prominently in fear expression after extinction, when they support context-dependent renewal of auditory fear conditioning (Jin & Maren, 2015; Orsini et al., 2011; Q. Wang, Jin, & Maren, 2016). However, selective manipulations of either vHPC or vCA1-PL transmission using chemo- and optogenetics have reached the opposite conclusion, namely that vCA1-PL activity contributes to inhibition of CS neuronal responses and associated freezing through putative interactions with PL GABAergic interneurons (Sotres-Bayon et al., 2012; Szadzinska et al., 2021; Vasquez et al., 2019). Likewise, experiments involving fear inhibition by an explicit safety cue indicate a requirement for vCA1-PL activity in mice and reveal that an analogous vHPC-dmPFC network is similarly modulated by safety in both mice and humans (Meyer et al., 2019). An important caveat is that the functional relationship of vCA1 and PL might be shaped by learning, since optogenetic stimulation of the vCA1-PL pathway attenuates freezing only after several extinction sessions, while having the opposite effect at an earlier time point (Szadzinska et al., 2021).

While c-Fos labeling suggests that neither contextual fear conditioning nor recall activates vCA1-PL neurons to a significant degree (Dixsaut & Graff, 2022; Santos, Kramer-Soares, Coelho, & Oliveira, 2023), recall increases c-Fos among vCA1 neurons with projections to both mPFC and BLA (W. B. Kim & Cho, 2017). This is consistent with evidence for preferential recruitment of vCA1 double-projecting neurons following context-dependent renewal (Jin & Maren, 2015), and thus relevant contextual information might be conveyed through a specialized subnetwork. Regardless of c-Fos expression patterns, selective manipulations of vCA1-PL projections indicate that they contribute to acquisition and expression of contextual fear responses during trace fear conditioning, without affecting CS encoding (Twining, Lepak, Kirry, & Gilmartin, 2020). In contrast, stimulation of vCA1-PL projections does not affect foreground contextual fear retrieval but reduces background contextual freezing during context renewal (Pompili et al., 2025). This warrants further examination of vCA1-PL projections in contextual fear, especially given evidence for its involvement of PL cell populations (DeNardo et al., 2019; Kitamura et al., 2017; Rozeske et al., 2018). The opposing effects of vCA1-PL on freezing after cued and contextual training suggest that either vCA1 supplies purely context-related information to PL or it can signal either positive or negative valence through discrete projections, a possibility that we discuss below.

In contrast to PL, most investigations of IL have emphasized its role in the extinction of contextual and auditory fear conditioning, suggesting a function complementary to that of PL. Indeed, extinction induces a coordinate shift in c-Fos labeling within mPFC, reducing the density of c-Fos+ neurons in PL and increasing their density in IL (Knapska & Maren, 2009). Silencing IL or its BLA projections interferes with extinction acquisition (Bukalo et al., 2015; Do-Monte, Manzano-Nieves, Quinones-Laracuente, Ramos-Medina, & Quirk, 2015). While studies disagree regarding the necessity of post-training IL activity for extinction retrieval (Adhikari et al., 2015; Bukalo et al., 2015; Do-Monte et al., 2015; H. S. Kim, Cho, Augustine, & Han, 2016; Luft, Popik, Goncalves, Cruz, & de Oliveira Alvares, 2024), chemogenetic activation of vCA1-IL neurons interferes with extinction recall, while silencing vCA1-IL neurons prevents context-dependent renewal after auditory fear extinction (Marek et al., 2018). Because vCA1-IL projections generate feedforward IPSCs through activation of parvalbumin interneurons (PV-INs) (Liu & Carter, 2018; Marek et al., 2018), a suggested mechanism for behavioral control entails suppression of extinction-related excitatory PNs, which may include those with projections to BLA and/or BMA (Adhikari et al., 2015; Bukalo et al., 2015; Do-Monte et al., 2015; H. S. Kim et al., 2016). However, excitatory actions of vCA1 projections could also contribute to extinction, particularly because vCA1-IL neurons form relatively strong synapses onto layer 5 PNs that support spiking activity in the presence of GABAergic inhibition (Liu & Carter, 2018).

In any case, while behavioral effects of vCA1-IL activity have been observed only after extinction, experience-dependent plasticity of vCA1 synapses is detected before this stage (Soler-Cedeno et al., 2019). Following auditory fear conditioning, vCA1 inputs exhibit decreased NMDA receptor-mediated transmission, an effect that is reversed upon extinction (Soler-Cedeno et al., 2019; Q. Wang et al., 2018). Extinction-induced reversal of NMDA receptor depression can be mimicked by ex vivo application of BDNF, which accumulates in vCA1 neurons during extinction learning and when infused into the IL, but not PL, generates a long-term reduction in CS-evoked freezing (Peters, Dieppa-Perea, Melendez, & Quirk, 2010; Rosas-Vidal, Do-Monte, Sotres-Bayon, & Quirk, 2014; Q. Wang et al., 2018). These results point to dynamic molecular changes within vCA1-IL circuits correlated with bidirectional fear regulation, but the impact of these changes on circuit activity remains unclear.

Several other mPFC regions also receive abundant projections from vCA1, but the potential role of these circuits in fear conditioning remains to be established. Both ACC and vHPC have been independently implicated in context fear generalization observed at remote time points after training (Bian et al., 2019; Cullen, Gilman, Winiecki, Riccio, & Jasnow, 2015; Frankland, Bontempi, Talton, Kaczmarek, & Silva, 2004; Ortiz et al., 2019). In both cases, projections to BLA contribute to generalization, but the potential role of vCA1-ACC transmission is unknown. DP is another target of vHPC that was recently identified as an important locus in auditory fear conditioning (Campos-Cardoso et al., 2024). Given the functional dichotomy of vCA1-PL and vCA1-IL projections, and the fact that DP possesses distinct cytoarchitecture from both areas, it would be interesting to test what influence vCA1-DP projections have on bidirectional fear regulation. Ultimately, our understanding of vCA1-mPFC functional interactions would benefit greatly from a more detailed mapping of vCA1 cell populations that send axon projections to these distinct areas and the degree to which they collateralize. This will help clarify which anatomical levels are involved in distinct functional outcomes of vCA1 activity.

3c. Hypothalamus and basal forebrain systems

While the corticoamygdala network has received more attention for its contributions to fear learning, vCA1 contains a far greater number of PNs that project to subcortical areas (Gergues et al., 2020). One such area with well-established roles in emotional regulation and motivation is the lateral hypothalamus (LH), which is targeted by a vCA1 population having negligible overlap with vCA1-BLA cells (Jimenez et al., 2018). Optogenetic stimulation of vCA1-LH neurons induces anxiety-like behavior but does not affect acquisition or expression of contextual fear memory. However, as revealed by real-time place preference, vCA1-LH activity is strongly aversive and may therefore modulate fear learning under yet untested conditions. A potential substrate for such effects would be orexin-containing LH neurons, which promote fear memory acquisition (Sears et al., 2013) and expression (Soya et al., 2017) through recruitment of locus coeruleus noradrenergic signaling. Following extinction, recruitment of LC neurons can induce fear reinstatement (Giustino, Fitzgerald, Ressler, & Maren, 2019) and might be sensitive to vCA1-LH orexinergic interactions. Another possibility is that vCA1-LH circuitry modulates conditioned fear under more complex scenarios, such as during conflicting motivational drives or after additional learning experience. For example, while LH GABAergic neurons are not required for cued fear conditioning in naïve animals, silencing these neurons interferes with fear conditioning in animals recently trained in a reward-based task (Sharpe, Batchelor, Mueller, Gardner, & Schoenbaum, 2021). Finally, although poorly characterized, reciprocal projections from LH have also been described (Noble et al., 2019). These studies suggest a more detailed examination of vCA1-LH interactions in fear learning may be warranted.

The most abundant target of vCA1 is the lateral septum (LS), which receives projections from roughly half of vCA1 PNs (Gergues et al., 2020). Although more extensively studied for its role in motivated, goal-directed behaviors (Besnard & Leroy, 2022), LS is increasingly identified as a highly stress-responsive system that participates in fear conditioning. Hippocampal projections to LS from dCA3/vCA3 and dCA1 are implicated in contextual (Besnard et al., 2019; Besnard, Miller, & Sahay, 2020) and auditory fear conditioning (M. Chen, Li, Shan, Yang, & Zuo, 2025; Opalka & Wang, 2020), respectively. While dCA3 projections promote contextual generalization, vCA3 projections inhibit freezing in conditioned contexts in part through recruitment of somatostatin (SST)-expressing neurons in the dorsal LS (Besnard et al., 2020). LS SST neurons are primarily active outside of contextual freezing bouts and may therefore interfere with conditioned fear expression (Besnard et al., 2019). While the function of vCA1-LS projections remains to be examined, they could form synapses with SST neurons and thereby mediate fear inhibition. However, another population that receives monosynaptic connections from vCA1-LS projections is enriched for corticotropin-releasing hormone receptor 2 (Crhr2) and promotes arousal, cortical activation, and defensive responding to unexpected stressors (Hashimoto et al., 2023), implying that vCA1 control over LS may not be so straightforward.

Located further rostral in the basal forebrain, the nucleus accumbens (NAc) is perhaps best associated with appetitive reinforcement, motivation, and reward, and is a major target of vCA1 PNs (Gergues et al., 2020). In signaled active avoidance, in which animals move to avoid aversive stimuli, both BLA and PL inputs to NAc play critical, albeit opposing roles (Diehl et al., 2020; Klune et al., 2025; Ramirez, Moscarello, LeDoux, & Sears, 2015). Although this type of learning depends on formation of a CS-US association, it culminates in an inhibitory rather than excitatory CS response within PL PNs, which presumably gates the activity of avoidance-related NAc populations (Diehl et al., 2020). While inactivation of vHPC disrupts non-signaled, context-dependent shuttling as well as extinction of 2-way active avoidance (Cavdaroglu et al., 2020; Oleksiak et al., 2024; Oleksiak et al., 2021), the potential role of vCA1-NAc projections remains to be examined.

Another major role for NAc supported by multiple recent studies is in processing a reward-related component of extinction learning. Confirming a long-standing theory about omission of expected aversive outcomes, extinction of auditory fear conditioning is associated with phasic activation of VTA dopaminergic neurons (Salinas-Hernandez et al., 2018), which in turn release dopamine in the anteromedial NAc (Salinas-Hernandez, Zafiri, Sigurdsson, & Duvarci, 2023). Following learning, dopamine terminals in the shell but not core subregion of NAc are also important for extinction consolidation (Luo et al., 2018), and once fully consolidated, evidence suggests that memory-related ensembles in the insular cortex (IC) mediate extinction retrieval via projections to NAc (Q. Wang et al., 2022). Given these extensive contributions, it seems likely that abundant projections from vCA1 might regulate extinction-related functions of NAc, and a relevant factor might be the overrepresentation of vCA1 neurons that project to both NAc and LS (Gergues et al., 2020).

The medial septum (MS) and HPC also share reciprocal connections. MS and the adjacent diagonal band of Broca (DB) provide ascending inputs to the HPC via the fimbria-fornix, forming the septo-hippocampal pathway (Amaral & Kurz, 1985; X. Li et al., 2018). Most septo-hippocampal projections are cholinergic; nonetheless, GABAergic and glutamatergic fibers contribute to a lesser extent (Khakpai, Nasehi, Haeri-Rohani, Eidi, & Zarrindast, 2013). In turn, the HPC provides feedback to MS through glutamatergic and GABAergic projections. Through the fimbria, septohipocampal projections supply acetylcholine (ACh) to the vHPC (Dutar, Bassant, Senut, & Lamour, 1995; Everitt & Robbins, 1997), where MS/DB cholinergic neurons innervate vCA1 and vCA3 (Wu et al., 2024). Cholinergic lesions specific to vHPC impair acquisition and expression of cued fear memory (Staib, Della Valle, & Knox, 2018), whereas nicotine infusion into the vHPC disrupts contextual and trace but not delay auditory fear conditioning (Kenney, Raybuck, & Gould, 2012; Raybuck & Gould, 2010). In addition, vHPC nicotine infusion impairs contextual fear extinction (Kutlu et al., 2018). While cellular substrates of cholinergic signaling within the vHPC remain largely unknown, recent evidence suggests that acetylcholine activates oriens lacunosum moleculare (OLM) interneurons to support type 2 theta oscillations (Mikulovic et al., 2018). Occurring in the 6-7.5 Hz range, type 2 theta is associated with stressful or anxiety-provoking situations and, intriguingly, augmenting this rhythm through optogenetic manipulation of OLM interneurons exerts an anxiolytic effect. Further dissection of cholinergic substrates may therefore provide unique insights into fear attenuation.

3d. Nucleus reuniens

Tucked within the midline thalamus, nucleus reuniens (RE) is attributed critical mnemonic functions that relate to its coordination of activity within cortico-hippocampal networks (for detailed review see: (Dolleman-van der Weel et al., 2019; Eichenbaum, 2017)). RE possesses bidirectional connections with both mPFC and CA1 neurons throughout the dorsoventral axis, providing an important disynaptic link between regions with otherwise limited connectivity. Interestingly, RE projections to CA1 are 10-fold more abundant in vHPC than in dHPC (Hoover & Vertes, 2012; Varela, Kumar, Yang, & Wilson, 2014; Vertes, Hoover, Do Valle, Sherman, & Rodriguez, 2006) which, together with the fact that most HPC projections to mPFC originate from vCA1, could hint at unique roles for RE circuits in the ventral hippocampal pole. Activity manipulations of RE neurons and their connections with these systems supports their involvement in the acquisition, generalization, and extinction of fear memory (Adhikari et al., 2015; Ramanathan, Jin, Giustino, Payne, & Maren, 2018; Ramanathan, Ressler, Jin, & Maren, 2018; Ratigan, Krishnan, Smith, & Sheffield, 2023; Totty et al., 2023; Troyner, Bicca, & Bertoglio, 2018; W. Xu & Sudhof, 2013). The precise circuit basis for these effects is incompletely understood and the subject of ongoing investigation. However, conceptual models posit that prefrontal-thalamo-hippocampal networks allow stimulus representations of the hippocampus to be accessed by the mPFC to support action selection. A postulated correlate of this functional coupling is the synchronous entrainment of network oscillations, which is observed during rule-learning (Benchenane et al., 2010), spatial working memory (de Mooij-van Malsen et al., 2023; O’Neill, Gordon, & Sigurdsson, 2013; Spellman et al., 2015), and sequential memory tasks (Jayachandran et al., 2023). Likewise, theta synchrony in the mPFC-RE-dCA1 network is critical for expression of fear extinction memory (Totty et al., 2023) and may function to suppress the reactivation of fear-related ensembles and/or recruit extinction-related cells (Hassell et al., 2025).

4. Potential organizing principles of vCA1 networks and emerging roles for GABAergic interneurons

While the above studies reveal vCA1 networks critical to fear conditioning and extinction (Table 1), an important unanswered question is how synaptic integration within vCA1 engages discrete outputs to support these interactions. A comprehensive model may not be within reach, but some recent findings suggest potential organizing principles for connectivity and functional interaction among discrete cell populations. These may provide a basis for relating activity dynamics to functional outputs underlying fear and other emotional states.

Table 1. Summary of specific vCA1 circuits implicated in bidirectional regulation of conditioned fear and reported findings.

Connections are grouped by afferent targets and efferent synaptic inputs of vCA1 neurons.

Connection Type	Area	Reference	Task	Findings
Afferent	BLA	Nguyen et al., 2023	Delay FC and extinction	• Opto-inhibition during extinction training does not impact extinction learning but reduces freezing during extinction test to both CS and context
	MS (cholinergic)	Staib et al., 2018	Delay FC and extinction	• vHPC cholinergic lesions reduce freezing during acquisition and retrieval test • No effect on extinction acquisition or retrieval
Efferent	BLA	Xu et al., 2016	Delay FC, extinction, and context renewal Contextual FC	• Opto-inhibition during context renewal reduces freezing • Opto-inhibition during contextual fear retrieval reduces freezing
		Jimenez et al., 2018	Contextual FC	• Opto-activation during training reduces freezing during retrieval • Opto-activation during retrieval also reduces freezing
		Kim & Cho, 2020	Contextual FC	• Chemo-inhibition during training reduces freezing at test
	CeA	Xu et al., 2016	Delay FC, extinction, and context renewal Contextual FC	• Opto-inhibition during CS fear retrieval does not impact CS-induced freezing • Opto-inhibition during context-dependent renewal reduces freezing • Opto-inhibition during contextual fear retrieval does not affect freezing
	PL	Vasquez et al., 2019	Delay FC, extinction, and context renewal	• Chemo-activation impairs context-dependent renewal • Chemo-activation does not impair fear retrieval
		Twining et al., 2020	Trace FC	• Opto-inhibition during training does not impact trace cued fear acquisition or retrieval • Opto-inhibition during training reduces contextual freezing
		Szadzinska et al., 2021	Delay FC and extinction	• Opto-activation after a single-extinction session increases freezing during extinction retrieval • Opto-activation following extensive extinction training prevents context-dependent renewal
	IL	Marek et al, 2018	Delay FC, extinction, and context renewal	• Chemo-inhibition attenuates context-dependent renewal • Chemo-activation impairs extinction retrieval
	LH	Jimenez et al., 2018	Contextual FC	• Opto-activation during test or retrieval have no effect on freezing

BLA: basolateral amygdala. MS: cholinergic medial septum. CeA: central amygdala. PL: prelimbic cortex. IL: infralimbic cortex. LH: lateral hypothalamus.

Along the radial axis of dorsal CA1, PNs form sublaminar subpopulations that can be distinguished based on morphology, electrophysiology, gene expression, and embryonic origin, as well as distinct response properties during spatial navigation, learning, and memory (Geiller, Royer, & Choi, 2017) (Figure 1A). Superficial PNs, which have a cell body located near the stratum radiatum, have more elaborate apical dendrites and express the marker protein calbindin, while deep PNs have larger somata and higher basal dendritic complexity (Bannister & Larkman, 1995; Lee et al., 2014; Masurkar et al., 2017). A potential implication is that these populations preferentially sample afferents targeting different CA1 domains. Indeed, superficial PNs possess a greater number of dendritic spines located in stratum lacunosum moleculare (s.l.m.) and exhibit stronger synaptic responses to perforant path stimulation (Masurkar et al., 2017), although this bias reverses along the transverse axis approaching the subiculum, complementing other evidence of anatomical and physiological variance along this spatial dimension (Beer et al., 2018; Henriksen et al., 2010; Igarashi, Ito, Moser, & Moser, 2014; Nakazawa, Pevzner, Tanaka, & Wiltgen, 2016). In addition to direct medial and lateral entorhinal input, s.l.m. receives projections from nucleus reuniens, which elicit excitatory responses in CA1 neurons (Dolleman-Van der Weel, Lopes da Silva, & Witter, 1997; Goswamee, Leggett, & McQuiston, 2021; Wouterlood, Saldana, & Witter, 1990).

Figure 1. Potential organization of vCA1 circuits of relevance to conditioned fear.

A, Similar to dCA1, the genetic and morphological properties, as well as synaptic connectivity of vCA1 PNs may depend on spatial location along the dorso-ventral and proximo-distal axis, which coincides with the antero-posterior axis in a coronal plane of section. In addition, PNs can occupy deep and superficial layers of vCA1, which independently correlate with distinct genetic, anatomic and morphological properties. Color and shade of excitatory PNs indicates functional variation along these geometric axes. B, Illustration of recent findings regarding parallel processing of affective information by anatomical subnetworks of vCA1 PNs. In assays of anxiety-like behavior, networks preferentially targeting and originating from the superficial and deep layers of vCA1 are implicated in approach versus avoidance, respectively. Relevant studies: (Pi et al., 2020; Sanchez-Bellot et al., 2022). C, Potential substrates for modulation of fear by extinction-related SST-INs. SST-INs recruited by extinction might preferentially inhibit fear-related PNs or, via suppression of firing in other interneuron (IN) populations, disinhibit extinction-related PNs. These outcomes might depend on the intrinsic wiring of fear- versus extinction-related PNs and the relative strength of transmission within local GABAergic microcircuits. Relevant study: (Lacagnina et al., 2024).

Whether laminar position confers similar functional differences in vCA1 is unknown, but recent work suggests that inputs and outputs with distinct functional relevance can be mapped onto deep and superficial vCA1 layers. For example, anterior and posterior BLA exert opposing control over exploratory behavior and target different laminar components of vCA1. aBLA-vCA1 projections innervate deep PNs to promote avoidance, while pBLA-vCA1 projections more strongly innervate superficial PNs to support approach (Pi et al., 2020) (Figure 1B). These outcomes are seemingly matched to the function of anatomically segregated BLA populations, wherein Rspo2+ neurons in the aBLA and Ppp1r1b+ neurons in the pBLA drive expression of fear and extinction memory, respectively (J. Kim et al., 2016; X. Zhang et al., 2020). Conversely, BLA-vCA1 inputs to sublaminar PNs could also evoke feedback to functionally discrete BLA PNs (AlSubaie et al., 2021). As with afferent connections, distinct outputs of vCA1 tend to arise from sublaminar clusters. Projections from superficial vCA1 preferentially target non-fast-spiking GABAergic interneurons in the mPFC and promote exploration, while those from deep vCA1 target both excitatory PNs and fast-spiking INs, and support avoidance (Sanchez-Bellot, AlSubaie, Mishchanchuk, Wee, & MacAskill, 2022). Thus, BLA, vCA1, and mPFC appear to be linked through parallel subnetworks. Surprisingly, however, BLA^vCA1 projections mostly avoid vCA1^mPFC PNs, forming few connections with them (AlSubaie et al., 2021). While the involvement of vCA1 PNs in approach-avoidance conflict is only suggestive of a role in fear memory, if functional subnetworks of vCA1 are accessible through such anatomical approaches, similar insights might be gained into behavioral switching between high and low fear, as well as other valence-related states (Shpokayte et al., 2022).

A potential advantage to clustering of functionally distinct PNs along the vCA1 radial axis, and perhaps other dimensions, is that they can be activated with greater selectivity by relevant input patterns, but this implies the existence of circuits capable of differentiating between PNs based on spatial organization. Preferential recruitment of PNs could emerge solely from morphological diversity and afferent connectivity of laminar subpopulations, as discussed above. However, this leaves unaddressed the question of how vCA1 inputs, conveyed by various afferent projections, might be further processed through local synaptic integration to support or otherwise modulate vCA1 output. Although synaptic connections between CA1 PNs exist, they are very sparse and largely restricted to the interlamellar plane perpendicular to the hippocampal trisynaptic pathway (Yang et al., 2014). By contrast, PNs are extensively innervated by GABAergic interneurons, which have an enormous capacity to sculpt local network activity. It is therefore interesting that lamina-specific PNs in dCA1 receive different levels of synaptic inhibition from GABAergic subtypes; in particular, deep-layer PNs exhibit stronger modulation by PV-INs but not CCK-INs (Lee et al., 2014), which may explain their unique response properties during sharp-wave ripples (Valero et al., 2015). Among deep-layer PNs, however, those with projections to amygdala or medial entorhinal cortex exhibit stronger input from PV-INs than those that project to mPFC, suggesting further specialization among sublaminar populations or the existence of additional spatial codes for PV-IN connectivity (Lee et al., 2014). In addition to exhibiting state-dependent modulation (Dudok et al., 2021), PV-INs derive most of their excitatory input from local glutamatergic cells and are therefore well-positioned to mediate competitive interactions between projection-specific PNs. Indeed, PNs receiving weaker PV-IN input provide stronger excitatory output to PV-INs, potentially reinforcing antagonism between anatomically segregated networks (Lee et al., 2014).

While it remains unclear whether anatomical connectivity of vCA1 interneurons exhibits similar bias with respect to laminar- and/or projection-specific PNs, emerging evidence suggests that vCA1 GABAergic subtypes play important roles in fear regulation. For example, trace fear conditioning is associated with cue-evoked suppression of SST-IN activity, whereas photoactivation of SST-INs during the cue period impairs memory acquisition (K. Li et al., 2024). A corresponding increase in cue-related VIP-IN activity develops in tandem with SST-IN suppression during trace conditioning and is required for learning, implying the existence of a disinhibitory interaction between VIP-INs and SST-INs. In contrast to trace fear acquisition, increased vCA1 SST-IN activity is observed following extinction of contextual fear conditioning and, furthermore, optogenetic manipulation of SST-INs but not PV-INs elicits bidirectional changes in freezing that are specific to the context in which extinction was acquired (Lacagnina et al., 2024). SST-IN activity increases at freezing offset, which may explain a shortening of discrete freezing bouts during fear extinction. A similar activation pattern has been observed in the dorsolateral septum as well as nucleus reuniens (Besnard et al., 2019; Silva et al., 2021), which contain cell populations that, like vCA1 SST-INs, signal transitions between immobility and movement following context fear extinction. This suggests that extinction recruits inhibitory mechanisms in vCA1 that limit the duration and/or stability of defensive behavior, perhaps as a microcircuit component of a functional subnetwork.

While vCA1 SST-INs appear to support a low-fear state, a critical unanswered question is how they modulate and potentially cooperate with surrounding PNs that convey the bulk of vCA1 output. A possible clue is that a large proportion of SST-INs preferentially target the apical dendritic compartment of PNs, where they can suppress the integration of entorhinal or other inputs targeting this domain and thereby disrupt the activity of fear-related PNs that rely on this transmission (Figure 1C). Whether SST-IN activity predominantly affects medial versus lateral entorhinal transmission might depend on whether PNs reside in superficial or deep layers of CA1 (Masurkar et al., 2017). In addition, like SST-INs in dorsal and intermediate CA1, vCA1 SST-INs could also synapse onto PV-INs and Schaffer-collateral-associated interneurons (Leao et al., 2012; Lovett-Barron et al., 2014), which would enable suppression of feedforward inhibition from CA3 afferents to facilitate recruitment of extinction-related cell assemblies or otherwise increase the gain of intrahippocampal processing (Figure 1C). SST-INs could therefore boost or attenuate activity of discrete PNs, depending on their dendritic morphology, connectivity, and/or stimulus response properties.

Interneurons defined by SST expression represent a useful test case for specialization of GABAergic transmission in the ventral hippocampus and its relationship to functional parcellation of PNs, but connectivity of this cell class is not uniform. A major subset of SST-INs are so-called oriens-lacunosum-moleculare (OLM) cells that arborize extensively within s.l.m. (Leao et al., 2012; Lovett-Barron et al., 2014), but these are comprised of discrete subtypes that are enriched for secondary markers, including the nicotinic α2 cholinergic receptor (Chrna2), neuron-derived neurotrophic factor (Ndnf), and thyroid transcription factor 1 (Nkx2-1), each exhibiting unique connectivity (Chamberland et al., 2024). For example, in dorsal CA1, Chrna2-INs exert equivalent inhibition of PV-INs and PNs, while Ndnf;;Nkx2-1-INs inhibit only PNs. In addition, other genetic subtypes of SST-INs, including bistratified and oriens-oriens interneurons, do not possess OLM morphology and therefore exhibit even more dissimilar wiring (Chamberland et al., 2024; Harris et al., 2018). Such heterogeneity is not exclusive to hippocampal SST-INs but is present among all conventionally defined GABAergic subtypes, including PV-INs and other CGE-derived IN classes (Booker & Vida, 2018; Pelkey et al., 2017). Relating GABAergic activity to the function of anatomically defined subnetworks will require dissection of these component populations. It is also worth noting that the taxonomy of hippocampal neurons was mostly developed in the dCA1, but differences in biophysical and morphological properties might contribute to computational specialization along the dorsoventral axis.

Another factor that may impact vCA1 circuit operations is biological sex, especially given that rodents exhibit sexual dimorphism in fear conditioning and extinction (Day & Stevenson, 2020; Romero, Acharya, Nabas, Marin, & Andero, 2025; Velasco, Florido, Milad, & Andero, 2019). Interestingly, vCA1 neurons projecting to BLA are twice as abundant in female compared to male rats and are enriched in perisomatic cannabinoid type-1 receptors (CB1Rs) (Huckleberry et al., 2023). Disruption of CB1R signaling increases cue-elicited freezing within a specific subgroup of females that expresses conditioned darting responses (Huckleberry et al., 2023). Brain-wide c-Fos labeling following fear-conditioning and extinction also suggests sex-specific alterations in vCA1 functional connectivity (du Plessis, Basu, Rumbell, & Lucas, 2022; K. Zhang et al., 2024). However, the extent to which sex affects the physiology of discrete vCA1 circuits remains largely unknown.

Ultimately, deeper insight into the intrinsic wiring and response properties of both glutamatergic and GABAergic populations will go a long way towards clarifying the role of vCA1 circuits in fear and extinction. Principles of synaptic connectivity would be a basis for relating the function of discrete neuronal subpopulations across different levels of analysis and enable testable predictions about how this logic generalizes to distinct learning conditions and pathological states. Such basic insights could open productive research avenues that reframe emotional regulation in terms of functional subnetworks with whose activity corresponds more directly to behavior, avoiding pitfalls of prior ambiguous work. Considering the importance of circuit-specific models of emotion and the high translational relevance of the hippocampal system, this would be a worthwhile endeavor.