Retrosplenial cortex and its role in cue-specific learning and memory

Abstract

The retrosplenial cortex (RSC) contributes to spatial navigation, as well as contextual learning and memory. However, a growing body of research suggests that the RSC also contributes to learning and memory for discrete cues, such as auditory or visual stimuli. In this review, we summarize and assess the Pavlovian and instrumental conditioning experiments that have examined the role of the RSC in cue-specific learning and memory. We use the term cue-specific to refer to these putatively non-spatial conditioning paradigms that involve discrete cues. Although these paradigms emphasize behavior related to cue presentations, we note that cue-specific learning and memory always takes place against a background of contextual stimuli. We review multiple ways by which contexts can influence responding to discrete cues and suggest that RSC contributions to cue-specific learning and memory are intimately tied to contextual learning and memory. Indeed, although the RSC is involved in several forms of cue-specific learning and memory, we suggest that many of these can be linked to processing of contextual stimuli.

1. Introduction

The retrosplenial cortex (RSC) has become the focus of increased research interest over the past two decades (Figure 1), due in part to the discovery that structural and functional changes in RSC are associated with several common forms of mental illness. For example, the RSC is among the first cortical regions to exhibit signs of neurodegeneration and pathology associated with Alzheimer’s Disease (Buckner et al., 2005; Ma et al., 1994). Additionally, heightened excitability of RSC neurons has been linked to age-related cognitive decline (Bucci et al., 1998; Haberman et al., 2017) and alterations in RSC function and connectivity are associated with schizophrenia (Bluhm et al., 2009) and autism (Hogeveen et al., 2018). Accordingly, in 2009, Vann, Aggleton, and Maguire aptly asked “What does the retrosplenial cortex do?” These authors reviewed findings from neuroanatomical and animal studies, as well as human imaging studies, demonstrating a role for the RSC in a range of cognitive functions. In particular, however, they described a role for the RSC in spatial navigation as “incontrovertible” (pg. 799). Considering the reciprocal connections between the RSC and other regions involved in spatial navigation (e.g., hippocampal formation, anterior and lateral-dorsal thalamic nuclei), and that single-unit recording studies have identified spatial correlates within the RSC, it is no surprise that recent studies have continued to emphasize the RSC’s important role in spatial navigation (Alexander & Nitz, 2015; 2017; Czajkowski et al., 2014; Vedder et al., 2017). Nevertheless, a growing body of research, carried out primarily in laboratory animals, has also demonstrated a role for the RSC in a variety of non-spatial learning and memory paradigms. It is thus worth asking if the RSC contributes more to learning and memory than simply “space” and “place”.

Figure 1.

Number of articles published in five year intervals (through 2017) based on a PubMed search of “retrosplenial cortex.”

The purpose of this review is to assess the role of the RSC in cue-specific learning and memory. We use the term cue-specific learning and memory to refer to putatively non-spatial conditioning paradigms that involve discrete cues, typically visual or auditory stimuli. Thus, we will not explicitly review experiments that examine RSC contributions to spatial navigation, even if discrete cues are involved (e.g., Vedder et al., 2017). In addition, we recognize that in any learning and memory experiment, discrete cues are necessarily presented against a background of other cues, often referred to as the “context” (Bouton, 2010). We will only briefly review RSC contributions to contextual learning and memory, in and of itself (for a review, see Corcoran et al., 2018). We will, however, emphasize the role of the context when it may be especially important for influencing responding to discrete cues. Indeed, as will be described, we suggest that RSC contributions to cue learning are intimately tied to contextual processing.

In the first portion of this review we will briefly describe the connectivity of RSC in rats, with an emphasis on connections that may be particularly important for cue-specific learning. Next, with a specific focus on Pavlovian and instrumental conditioning experiments, we will review and assess the role of the RSC in associative learning and memory for discrete visual and auditory cues. These experiments fall into three general categories (each with its own sub-categories): 1) first-order conditioning, 2) discrimination learning, and 3) higher-order conditioning. We then review several ways in which contextual stimuli can influence learning and responding to discrete cues, and suggest a role for the RSC in cue learning that is intimately tied to contextual learning and memory.

2. Connectivity of the Retrosplenial Cortex

Here we summarize the connections of the RSC that make it well-suited to contribute to cue-specific learning and memory. We note that many of these connections are likely also important for spatial navigation and contextual learning and memory. For example, projections from the visual cortex to the RSC might be involved in cue learning, as well as aspects and contextual processing. Thus, we do not suggest that the anatomy described here is exclusively important for cue-specific learning and memory. Indeed, the RSC is ideally situated to process information relevant to contexts, cues, and reinforcing outcomes.

2.1. Connectivity with limbic and memory systems.

The RSC (Brodmann areas 29 and 30) in the rat brain is a large midline structure that extends for more than eight millimeters along the rostro-caudal access of the brain (Paxinos and Watson, 2009), and includes subregions identified by the presence (regions Rga and Rgb) or absence (region Rdg) of a distinct granule cell layer (Wyss & Sripanidkulchai, 1984; van Groen & Wyss, 1990a, 1990b). Neuroanatomical studies initially focused on characterizing the RSC’s connections with limbic structures and other regions involved in spatial learning and navigation (van Groen & Wyss, 1990a, 1992; 2003). For example, the RSC is strongly interconnected with the anterior and lateral-dorsal thalamic nuclei (Mathiasen et al., 2017; Sripanidkulchai & Wyss, 1986; van Groen & Wyss, 1992; van Groen & Wyss, 1990a), each of which contributes to spatial memory and navigation (Aggleton et al., 1995; Taube, 1995; Warburton & Aggleton, 1998). In addition, the RSC has connections with several components of the hippocampal-parahippocampal memory system, including the subiculum, postsubiculum, and area CA1 (van Groen & Wyss, 1990a, 1990b; 2003; Wyss & van Groen, 1992), as well as the postrhinal, perirhinal, and medial entorhinal cortices (Wyss & van Groen, 1992; Kerr et al., 2007; Burwell & Amaral, 1998). Each of these areas has also been shown to be important for both spatial learning as well as contextual learning and memory (Baldi, 2014; Bucci et al., 2000; Maren et al., 1997; Robinson & Bucci, 2012; Liu, 2001; 2002; Taube et al., 1992; Morris et al., 1982). A summary of the connections between the RSC and hippocampal-parahippocampal structures is available in a searchable RSC ‘connectome’ (Sugar et al., 2011). A simplified version of these connections is presented in Figure 2a.

Figure 2.

Simplified schematic of RSC connections. A) RSC connectivity with limbic and memory systems. B) RSC connectivity with sensory processing regions. C) RSC connectivity with regions involved in processing biologically-relevant stimuli.

2.2. Connectivity with sensory processing regions.

If the RSC contributes to cue-specific learning and memory, it might also be expected that it is interconnected with a variety of sensory regions (Fig. 2b). Consistent with this notion, the most well-described direct connections between the RSC and sensory regions involve its afferent and efferent connections with areas of visual cortex. Specifically, the RSC is strongly interconnected with visual areas 17 and 18a/b (van Groen & Wyss, 2003; Wyss & van Groen, 1992; Vogt & Miller, 1983). Although less work has considered whether the RSC receives other types of sensory input, Vogt and Miller (1983) reported that RSC also receives direct projections from auditory cortex. We recently replicated and expanded on those findings by revealing a substantial projection from secondary auditory cortex to all subregions of RSC (Todd et al., 2016b).

In addition to these direct connections between the RSC and sensory regions, the RSC also receives sensory information in an indirect fashion. For example, the RSC is interconnected with the claustrum (Figure 2b; Vogt & Miller, 1983; Todd et al., 2016b), which receives polymodal sensory input from areas including visual, auditory, and somatosensory cortices (Miller & Vogt, 1984; Carey & Neal, 1985; Sadowski et al., 1997; Beneyto & Prieto, 2001; White et al., 2017). In addition, the RSC is interconnected with the posterior parietal cortex (Olsen et al., 2017) an area that is essential for visuo-spatial processing, working memory, and a host of related cognitive functions (Bucci, 2009). These connections, as well as connections with the anterior thalamic nucleus, provide an additional, indirect means by which RSC is privy to polymodal sensory information.

2.3. Other connectivity relevant to learning and memory.

In many associative learning and memory paradigms, otherwise neutral sensory stimuli often become associated with reinforcing outcomes. Thus, if the RSC contributes to cue-specific learning, it might also be expected that the RSC receives information from brain regions and systems that process unconditioned stimuli (USs) such as food reward or shock. Consistent with this notion, the RSC contains dopaminergic terminals (Berger et al., 1985; Descarries et al., 1987; Fremeau et al., 1991) and receives direct, albeit sparse, input from areas including the ventral tegmental area and substantia nigra (Figure 2c; Loughlin & Fallon, 1984), which are critically involved in reward processing.

More generally, the RSC is also innervated by other diffuse neurotransmitter systems that are involved in learning and memory, including learning that involves discrete cues. For example, the RSC receives basal forebrain cholinergic projections from the medial septum (MS) and vertical limb of the diagonal band (VDM), which are known to modulate memory functions carried out by hippocampus and midline cortical regions (Baxter et al., 1997; Levey et al., 1984; Rye et al., 1984; van Groen & Wyss, 1990a; 1992; Tengelsen et al., 1992; Eckenstein et al., 1988). The RSC is also strongly interconnected with the anterior cingulate cortex (ACC) (van Groen & Wyss, 1990a; 1992; Wyss & van Groen, 1992; Shibata & Naito, 2008), as well as the anterior medial, anterior ventral, and anterior dorsal thalamic nuclei (Mathiasen et al., 2017), all of which have roles in associative learning and memory (e.g., Gabriel, 199).

Overall, the connectivity of RSC indicates that it receives both direct and indirect input from a variety of cortical and subcortical sensory regions, across multiple sensory modalities, and is modulated by subcortical dopaminergic and cholinergic systems. These connections, along with the connections to the medial-temporal lobe system, provide a solid anatomical foundation for the involvement of RSC in cue-specific learning and memory.

3. Retrosplenial Cortex Contributions to Cue-specific Learning and Memory

A growing body of evidence for the involvement of RSC in cue-specific learning and memory has emerged from studies that have tested the effects of various perturbations of RSC on behavior. In addition, Michael Gabriel and colleagues have examined the contribution of RSC to discriminative avoidance learning, using a combination of lesion and neural recording methods. In this review, we will focus on behavioral experiments that have manipulated the RSC in an effort to causally link RSC function to behavior (e.g., lesions, temporary inactivation), in part because studies of RSC activity during discriminative avoidance learning have been reviewed elsewhere (see Gabriel, 1993; Smith et al., 2018).

The most common approach has been to produce permanent lesions of the RSC (in rodents) prior to behavioral training. Studies of this sort have used a variety of lesion methods, such as passing tissue-damaging current through a fine electrode wire placed at one or more sites within the RSC (electrolytic lesion) or infusing a chemical through a small-gauge needle or pipette to destroy neurons (neurotoxic lesions). Other approaches involve more transient manipulations of RSC neurons by infusing a short-acting chemical inhibitor of neural activity (e.g., muscimol) or a compound that temporarily blocks protein synthesis (e.g., anisomycin). The advent of chemogenetics (Designer Receptors Exclusively Activated by Designer Drugs; DREADDs) and optogenetics has taken this approach to an even greater level of refinement by allowing for circuit-specific manipulations, and for perturbing neural function with sub-second levels of temporal precision. Each of these of the methods has been used to interrogate the functions of RSC in cue-specific learning and memory. Thus, we briefly consider the advantages and disadvantages to each method before turning to the behavioral studies in which they have been used to study the functions of RSC.

Permanent lesions provide a reliable and effective means of determining if a target region, such as the RSC, is necessary for a particular behavioral function. Because the damage is permanent, there is little chance that regrowth or recovery of neural activity can occur and produce false negatives (e.g., null effects). In addition, a relatively high percentage of the target area is typically affected. Electrolytic lesions have the drawback of damaging fibers of passage in addition to RSC neurons, especially if the lesion includes the cingulum bundle (a large fiber tract located just ventral to RSC). However, if a null effect is found following an electrolytic lesion, one can be reasonably certain that the target area is not essential for the behavior being tested. Neurotoxic lesions (e.g., infusions of ibotenic acid or NMDA), on the other hand, only affect neurons in the RSC, although fibers of passage in layer 1 often undergo mechanical damage by lowering the infusion needle or cannula. Thus, in considering the results of the lesion studies below, care must be taken to consider the lesion method when drawing conclusions about the involvement of the RSC in a particular behavior.

In some studies, lesions of RSC were carried out prior to behavioral training (“pre-training” lesions) while others have examined the effects of lesioning RSC after training (“post-training” lesions). In the case of pre-training lesions, an effect on behavior is often taken as strong evidence that the target region is necessary for that function. However, the opposite is not necessarily true. That is, a null effect of a pre-training lesion does not necessarily indicate that brain structure is uninvolved. Instead, it may be that some other brain system or learning strategy may have “compensated” for the loss of the target area. For example, it has been argued that the hippocampus normally has a primary role in contextual fear learning and memory. However, absent the hippocampus, alternate structures that are normally overshadowed by the hippocampus, can become engaged and can compensate for the loss of hippocampus, resulting in normal performance (see Fanselow, 2010; Rudy et al., 2004). As such, carrying out lesions after training can provide valuable information in the face of null results following pre-training lesions. For example, if RSC is normally involved and intact during learning, then subsequent damage to RSC would be expected to impact behavior, because the putative alternative regions or strategies were not deployed since RSC was available during learning.

Compared to permanent lesions, temporarily silencing neurons in the RSC allows for a more precise determination of when the RSC is needed for a particular behavior. For example, silencing the RSC during acquisition only can disambiguate effects on learning versus memory or performance, because RSC is back “on line” during subsequent memory tests. One drawback of this approach is that traditional inactivation methods, such as infusing a GABA receptor agonist into RSC, require the implantation of an indwelling cannula into the target areas. Considering the RSC is a very long structure (8mm along the rostro-caudal axis), it is challenging to implant enough cannula to inactivate the entire region without potentially causing significant mechanical damage, which is tantamount to a lesion. Also, since multiple cannulae are required in each hemisphere to bilaterally inactive the entire structure, null effects achieved through infusion into a single site in RSC are difficult to interpret. Additionally, the adjacency of RSC to hippocampus and subicular structures poses the risk of a lesion or infusion encroaching into these areas. These issues have recently been surmounted with the advent of chemogenetics (Armbruster et al., 2007; Urban & Roth, 2014). In some of the research we will consider below, a viral construct containing the DNA for a synthetic inhibitory (or excitatory) receptor can be surgically infused into the RSC, much like the case of neurotoxic lesion methods. The receptor is then expressed by RSC neurons and can be activated via systemic administration of a synthetic ligand, which results in silencing neurons for ~ 2 hours. In this way, activity throughout the entire RSC can be perturbed in a transient, reversible fashion during specific phases of behavioral training or testing (Robinson et al., 2014; Todd et al., 2016b). In addition, the fluorescent reporter provides a valuable indicator of the area that is affected, unlike infusions of traditional inactivating agents, the spread of which is difficult to assess.

In summary, the approaches that have been used to manipulate RSC each have advantages and disadvantages that should be considered when drawing conclusions regarding the causal involvement of RSC in cue-specific learning and memory. Perhaps the most suitable perspective is to consider the different approaches as complementary to one another in informing the functional role(s) of RSC. In addition, the results of causal manipulations should be considered in tandem with research that has examined multiunit neural activity in RSC during cue-specific learning, which we noted earlier (see Gabriel, 1993; Smith et al., 2018). Finally, only a few studies to date have sought to differentiate the function of several anatomically defined subregions of RSC (e.g., Rga, Rgb, and Rdg; Hindley et al, 2014; Pothuizen et al, 2009, 2010; Vann & Aggleton, 2005; van Groen et al., 2004). Indeed, the majority of the studies we consider here have treated the RSC as a whole or manipulated just a single site within RSC. To our knowledge, no studies have examined the role of RSC subregions in cue-specific learning and memory, and there is only preliminary evidence directly assessing RSC function along the rostro-caudal extent (Pullins et al., 2017). We now consider the role of the RSC in cue-specific learning and memory across a range of conditioning paradigms.

3.1. First-order Conditioning

3.1.1. Delay and trace conditioning.

In Pavlovian first-order conditioning, an initially neutral cue, such as a tone or a light, acquires the ability to elicit a conditioned response (CR) when it is repeatedly presented in such a way that it provides information about the occurrence of the unconditioned stimulus (US; Rescorla, 1988). It is typically assumed that this form of first-order conditioning results in an association between the internal representation of the cue, now the conditioned stimulus (CS), and the US. Indeed, this form of associative coding has been demonstrated in a variety of conditioning procedures (e.g., Holland & Rescorla, 1975; Rescorla, 1974). In perhaps the most frequently utilized version of this procedure, so called “delay” conditioning (e.g., Mackintosh, 1974), the CS is presented for a short period of time (e.g., 10 s) and ends with presentation of the US (see Figure 3a).

Figure 3.

Schematic of first-order conditioning procedures. Rectangles represent cues, with colors differentiating stimuli, and circles represent outcomes such as food or shock. A) Delay and trace conditioning. B) Recent vs. remote conditioning. C) Latent inhibition.

There is now a substantial amount of converging evidence that damage or inactivation of the RSC does not impair delay conditioning in experiments involving only a single cue. For example, electrolytic lesions (damaging cell bodies and fibers of passage) of the RSC made just prior to, or shortly after, tone-shock pairings do not affect expression of fear (e.g., freezing) to the tone CS at a later test (e.g., Keene & Bucci, 2008a). Likewise, Keene and Bucci (2008c), reported no impact of pre-training neurotoxic lesions (damaging cell bodies but sparing fibers of passage) on the expression of fear to a tone CS. More selective manipulations have produced results consistent with these findings. Blocking protein synthesis in the RSC before conditioning does not impair the acquisition of delay fear conditioning to an auditory cue (Kwapis et al., 2015). Further, blocking NMDA receptors selectively at the time of retrieval has no impact on the expression of fear to a previously conditioned tone (Corcoran et al., 2011; Kwapis et al., 2014, 2015). In a conditioned suppression procedure, Todd et al., (2017) found that fear conditioned to a light cue was not impacted by pre-training electrolytic lesions of the RSC. This lack of involvement of the RSC is not specific to aversive conditioning preparations; conditioned responding to a light cue paired with food reinforcement is also not impaired by pre-training electrolytic lesions of the RSC (Keene & Bucci, 2008b; Robinson et al., 2011). Thus, across a range of stimuli, reinforcers, and conditioning procedures, the RSC does not appear necessary for either the acquisition or retrieval of Pavlovian delay conditioning.

The involvement of the RSC in first-order Pavlovian conditioning, however, is sensitive to the temporal relationship between the CS and the US. When a short time interval (e.g., 20 s) separates the offset of the CS and the onset of the US, so called “trace” conditioning (see Figure 3a), the RSC is necessary for both conditioning and retrieval. For example, pre-training aspiration lesions of the RSC disrupt trace conditioning of the nictitating membrane response in rabbits (Solomon et al., 1986), and blocking protein synthesis in the RSC during conditioning impairs the acquisition of trace fear conditioning (Kwapis et al., 2015; Experiment 3). Further, blocking NMDA receptors during retrieval testing impairs the expression of fear to a previously conditioned trace CS (Kwapis et al., 2014, Experiment 3; Kwapis et al., 2015; Experiment 2). As noted, neither of these manipulations impact delay fear conditioning. Thus, in contrast to delay conditioning, the RSC is involved in both the acquisition and retrieval of trace fear conditioning.

3.1.2. Retrieval of remotely acquired conditioning.

One common feature of the experiments described in the preceding section is that they all involved testing soon after conditioning occurred. That is, there was only a short period of time between training and final testing. For example, in the experiments by Kwapis et al., (2015), testing occurred 1 day following initial conditioning. A recent series of experiments from our lab has demonstrated that, in contrast to recently-acquired delay fear conditioning, more “remotely” acquired delay conditioning (see Figure 3b) is RSC dependent. In our experiments, we assessed “remote” fear memory by testing after a relatively long consolidation interval (often 28-days). For example, Todd et al., (2016b) first trained rats with 3 pairings of a 10-s tone with a mild-footshock US (i.e., delay conditioning). Following a 28-day consolidation interval, different groups of rats received either electrolytic, neurotoxic, or sham lesions of the RSC. When fear expression (i.e., freezing) to the tone CS was then tested, rats with either electrolytic or neurotoxic lesions showed reduced fear to the tone relative to sham-lesioned animals. In a second study, using DREADDs to temporarily inactivate the RSC at the time of test, these authors also found that retrieval of remotely acquired trace fear conditioning was also RSC dependent (see Figure 4). Thus, retrieval of delay conditioning is RSC dependent at the remote time point, whereas retrieval of trace conditioning appears to be RSC dependent at both recent and remote time points. The necessity of the RSC for the retrieval of remotely-acquired delay fear conditioning is general across stimuli. For example, post-training electrolytic lesions of the RSC impair the expression of fear to a visual CS, when those lesions are made 28-days, but not 1-day, after training (Jiang et al., 2018).

Figure 4.

RSC inactivation at the time of test impairs retrieval of remotely acquired trace conditioning. Prior to training, one group of rats (Gi) received infusions of AAV-hSyn-HA-hM4Di-IRES-mCitrine into the RSC. This viral vector contained a gene for a synthetic inhibitory G-protein-coupled receptor (hM4Di) that suppresses neural activity when activated by systemic injection of clozapine-n-oxide (CNO). Control rats (GFP) received infusions into the RSC of a viral vector (AAV-hSyn-GFP) that did not contain the hM4Di receptor. All rats then received trace fear conditioning to an auditory cue. Fear to the tone was tested at a remote time point ( > 28 days later). During testing, all rats were received injections of CNO. B) Freezing during the training session (recorded in the post-shock period) and during the tone test session (recorded during tone presentations). Inactivation of the RSC (Group Gi) reduced freezing to the tone CS. * p < .05. Adapted from Todd et al., 2016.

Research from other laboratories has demonstrated an important role for secondary sensory cortices in the retrieval of remotely-acquired auditory fear conditioning. For example, Sacco & Sachetti (2010) demonstrated that post-training excitotoxic lesions of secondary auditory cortex impaired the retrieval of remote, but not recent, conditioning to an auditory cue. Since both secondary auditory cortex and the RSC are necessary for the retrieval of remotely acquired conditioning, it seems possible that these two regions are part of a functional circuit. We investigated this possibility by examining the effect of functionally disconnecting secondary auditory cortex (AuV) and the RSC. All rats first received Pavlovian delay fear conditioning involving three pairings of a tone CS and shock US. Following a 28-day retention interval, rats received unilateral lesions of both the AuV and RSC. One group of rats received ipsilateral (i.e., within the same hemisphere) lesions of the AuV and RSC. A second group of rats received contralateral (i.e., in opposite hemispheres) lesions of the AuV and RSC. Considering that projections from AuV to RSC are primarily ipsilateral (e.g., Todd et al., 2016b, Experiment 3), ipsilateral lesions were expected to disrupt communication between AuV and RSC in one hemisphere but leave communication intact in the non-lesioned hemisphere. In contrast, contralateral lesions would impair communication in both hemispheres. Thus, if communication between the AuV and RSC is important for the retrieval of remotely-acquired auditory fear conditioning, then contralateral lesions would be expected to have a larger impact on fear expression than ipsilateral lesions. This is precisely the pattern observed. During a nonreinforced test session, rats with ipsilateral lesions froze less to the tone than control rats that received sham lesions of AuV and RSC, and rats with contralateral lesions froze less than both the ipsilateral and sham lesioned rats. Based on this result, Todd et al., (2018) suggested that the retrieval of remotely-acquired auditory fear conditioning is dependent upon a functional circuit that includes the AuV and RSC. Future studies should examine this suggestion with more selective manipulations of this circuit, as well as connections between visual cortex and the RSC during retrieval of remotely-acquired fear to a visual cue.

There are several open questions regarding the nature of the RSC’s involvement in the retrieval of remotely-acquired conditioning. First, it is unknown if the RSC is active during initial conditioning, and if so, if this activity is necessary for later involvement in retrieval. Second, although the RSC is involved in retrieval of remote but not recent conditioning, it is also unknown if the involvement is temporally-graded. Studies to date have only compared retrieval at 1- vs. 28-days. Third, it is worth examining the nature of remotely-acquired and recently-acquired conditioning. For example, Todd et al., (2016b) suggested the involvement of RSC in the retrieval of remotely-acquired conditioning might coincide with a change in the content of memory, with remote memories being less detailed than recent memories. Although there is some evidence to support this notion (e.g., Pollack et al., 2018; Thomas & Riccio, 1979; for a review see Bouton et al., 1999), the involvement of the RSC has not been explicitly tested.

3.1.3. Extinction.

Conditioned responding established through first-order Pavlovian conditioning can be reduced through extinction (Pavlov, 1927). In extinction, repeated presentations of the CS in the absence of the US (Figure 3a) gradually weakens the strength of the CR. Similar to acquisition, the RSC’s role in extinction is dependent upon the temporal relationship between the CS and US during initial conditioning. That is, extinction of delay conditioning does not require the RSC, whereas extinction of trace conditioning does. For example, Kwapis et al. (2014, Experiment 4) first trained groups of rats with either delay or trace auditory fear conditioning. The next day, all rats received 40 extinction trials. During extinction, some rats received infusions of the NMDA receptor antagonist APV into RSC, while control rats received infusions of vehicle. In a drug free test the following day, trace fear conditioned rats that received APV infusions during extinction showed higher levels of responding to the CS relative to controls (i.e., less extinction). In contrast, delay conditioned rats that received infusions of APV were no different from controls. The higher levels of conditioned responding shown by trace conditioned rats suggests that blocking NMDA receptors in the RSC prevented the reduction of responding through extinction. One complication with this conclusion, however, is that the RSC is also necessary for the retrieval of trace fear conditioning (noted earlier). Thus, it is possible that extinction itself was not influenced by RSC NMDA receptor antagonism, but that a failure to retrieve the original conditioning precluded the ability of that conditioning to undergo extinction. Nevertheless, Kwapis et al., (2014, Experiment 3) also demonstrated increased ERK phosphorylation in the RSC during extinction, but not retrieval, of trace fear conditioning, consistent with the notion that the RSC undergoes plastic changes during extinction, over and above any contributions from retrieval.

Although conditioned responding decreases during extinction, the original CS-US association is not lost, but instead is offset by new inhibitory learning that is especially context specific (e.g., Bouton 2002, 2004). One line of support for this notion comes from studies of the renewal effect (Bouton & King, 1983). In renewal, conditioned responding will return when the extinguished CS is presented outside the context where extinction occurred. Although the findings of Kwapis et al., (2014) suggest that the RSC is not necessary for the extinction of delay fear conditioning, their experiment did not examine the RSC’s role in the contextual modulation of extinction learning. To do so, using a conditioned suppression method, Todd et al., (2017) examined the influence of pre-training electrolytic lesions of the RSC on renewal after extinction. In their first experiment, rats with either control or pre-training lesions of the RSC underwent delay fear conditioning with a visual CS in one distinct context (Context A). Next, the CS underwent extinction in a different context (Context B). Consistent with the findings of Kwapis et al., (2014), damage to the RSC had no impact on the extinction of delay fear conditioning. When extinction was complete, all rats were subsequently tested in both Context B and Context A. Conditioned responding continued to be low in Context B, but renewed in Context A, and there were no differences between groups in either context. Thus, damage to the RSC had no effect on either the rate of extinction learning, or the contextual modulation of extinction. A second experiment found consistent results when the lesions were made after extinction (Todd et al., 2017, Experiment 2).

Overall, the role of the RSC in extinction is similar to that of conditioning. The RSC is not necessary for the extinction of delay fear extinction to either auditory or visual cues. However, plasticity within the RSC is necessary for the extinction of trace fear conditioning to an auditory cue. Although experiments have examined both the acquisition of extinction and the contextual modulation of extinction for delay fear conditioning, no study to date has examined the contextual modulation of extinction after trace conditioning. Furthermore, there have been no studies examining extinction of remotely-acquired conditioning.

3.1.4. Latent inhibition.

In latent inhibition, pre-exposure to a cue alone slows down the acquisition of conditioned responding when the cue is subsequently paired with reinforcement (Figure 3c). The latent inhibition effect is sometimes described as a reduction in attention to the cue during pre-exposure that then impacts the rate of conditioning (e.g., Mackintosh, 1975; Pearce & Hall, 1980). Thus, this procedure is often used to investigate the neural mechanisms related to attention and conditioning. In a recent study, Nelson et al. (2018b) examined the impact of pre-training excitotoxic RSC lesions on latent inhibition. Both Sham and RSC lesioned rats first received multiple presentations of a single auditory cue, in the absence of a reinforcing outcome. Next, both groups of rats underwent conditioning, in which the pre-exposed cue and a novel cue were repeatedly paired with food reinforcement. Both lesioned and control rats demonstrated a latent inhibition effect; conditioning was slow for the pre-exposed cue relative to the non-pre-exposed cue, and there was no significant difference between groups. Indeed, as Nelson et al. (2018b) noted, if anything, the RSC lesioned rats showed a slightly more robust latent inhibition effect. These authors noted that the lack of effect of RSC damage on latent inhibition is consistent with other studies demonstrating RSC damage does not produce attentional deficits (Powell et al., 2017).

As noted, latent inhibition if often described as resulting from a decrease in attention to the pre-exposed cue. However, an alternative theoretical account is that during pre-exposure subjects encode either a stimulus - “nothing” association, or an inhibitory stimulus - “event” association (Westbrook & Bouton, 2010). During conditioning, the encoded pre-exposed memory is maintained, and competes with the newly acquired CS-US memory for control over performance of the conditioned response. Importantly, the association encoded during pre-exposure is modulated by the context, and thus the presence or absence of these contextual cues controls whether or not the memory encoded during pre-exposure will be retrieved. When viewed in this way, the learning that underlies latent inhibition is actually very similar to extinction (see Westbrook & Bouton, 2010). The finding that RSC lesions have no impact on latent inhibition is thus consistent with the fact that RSC lesions also have no impact on the extinction of delay conditioning. As Nelson et al. (2018b) suggest, future studies should further examine the role the RSC in latent inhibition, especially with respect to the influence of the context. This notion may be analogous to examining the RSC’s role in the contextual encoding of extinction (e.g., Todd et al., 2017).

3.2. Discrimination Learning

3.2.1. Discriminative conditioning.

As noted earlier, there is little involvement of the RSC in Pavlovian delay conditioning, in experiments in which a single CS is paired with the US. A slightly different procedure involves presenting two CSs (see Figure 5a), one of which predicts the occurrence of the US (CS+) and one that predicts its absence (CS−). Several studies have suggested a role for the RSC in such discriminative conditioning paradigms. For example, in a conditioned suppression paradigm, Todd et al. (2016a), trained rats to discriminate between two visual cues. One visual cue (flashing house light or steady panel light, counterbalanced) was paired with shock (CS+), whereas the other visual cue was presented alone (CS−). Rats with sham lesions of the RSC learned to discriminate between the CS+ and CS−; that is, they showed high levels of fear to the CS+ and low levels of fear to the CS−. In contrast, rats with pre-training electrolytic lesions of the RSC were severely delayed at learning the discrimination, initially showing high levels of fear to both the CS+ and CS−. This finding is consistent with earlier work by M. Gabriel and his colleagues. Gabriel, Sparenborg, and Stoler (1987) trained rabbits in a discriminative avoidance procedure with an auditory CS+ and CS−. When conditioned responding first reached criterion, responses to the CS+ did not reliably differ between control rabbits and rabbits with lesions of the RSC. However, lesions of the RSC did produce a reliable increase in CRs to the CS−. Thus, the experiments by Todd et al., (2016a) and Gabriel et al., (1987) both found that lesions of the RSC impaired the ability to inhibit responses during the CS−. Interestingly, Gabriel et al., (1987) found that pre-training electrolytic / aspirative RSC lesions produced an increase in discriminative avoidance CRs to the CS+ during the first conditioning session (see also Todd et al., 2016a), and a decrease in CRs to the CS+ during overtraining. Thus, at least in some cases, responding to the CS+ is also influenced by damage to the RSC. It is worth noting that the discriminate avoidance procedure used by Gabriel and colleagues includes a trace interval between CS offset and US onset.

Figure 5.

Schematic of discrimination learning procedures. Rectangles represent cues, with different colors differentiating stimuli, and circles represent outcomes such as food or shock. Stacked rectangles represent co-presentation of cues. A). Discriminative conditioning. B) Feature-negative discrimination training. C) Negative patterning.

At a general level, extinction of the CS+ following discriminative conditioning does not appear to be influenced by damage to the RSC, consistent with experiments using a single CS. For example, Todd et al., (2016a) reported no impact of RSC lesions on extinction of the CS+ (after discriminative training). Gabriel et al., (1987) also reported no statistically significant differences in the mean percentage of CRs in control rats and RSC lesioned rats during extinction of the CS+, though this conclusion is weakened by differences in performance in the overtraining sessions prior to extinction. However, in the experiment by Gabriel et al., (1987), damage to the RSC did alter the nature of the CR during extinction of the CS+; the mean duration of the CR to the CS+ was elevated during extinction for rats with lesions of the RSC vs. controls. As one exception, Berger et al., (1986) observed impaired extinction of the nictitating membrane response to a CS+ in rabbits with pre-training aspiration lesions of the RSC. However, this effect was observed during reversal learning, in which the original CS+ underwent extinction, and the original CS− was reinforced.

In summary, the results from discriminative conditioning experiments are generally consistent with first-order delay conditioning studies using only a single CS. The findings from discriminative conditioning experiments show either no effect (e.g., Todd et al., 2016a), or limited effects (e.g., Gabriel et al., 1987) of RSC damage on the acquisition and extinction of responding to a CS+. Further, both Gabriel et al., (1987) and Todd et al., (2016a) demonstrated that damage to the RSC influenced responding to the CS−, with RSC lesioned rats showing an increase in CRs to the CS−. However, both of these studies produced RSC damage via electrolytic lesions, raising the possibility that the behavioral effects observed may related to mechanical damage. Nevertheless, a meta-analysis of human fMRI data demonstrated RSC activity in response to safety cues (i.e., CS−) relative to fear cues (i.e., CS+; Fullana et al., 2015), which is consistent with the rodent studies.

Impaired discriminative conditioning could reflect a deficit in the acquisition / expression of associations between each cue and the outcome it predicts (e.g., CS-US or CS-no US), or a more general deficit in the ability to discriminate the auditory and visual cues themselves. However, the result of experiments examining the role of the RSC in other forms of discrimination learning suggest that damage to the RSC does not impair the ability to discriminate the sensory qualities of cues. For example, Nelson et al., (2014) trained rats on a complex appetitive discrimination. In one context, presses on the left lever, but not the right, were reinforced during the presentation of a steady panel light. Presses on the right lever, but not the left, were reinforced during a flashing panel light. All rats were concurrently trained on this discrimination, and a parallel discrimination in a second context with two auditory cues. Nelson et al. (2014) found that pre-training lesions of the RSC did not impair the original acquisition of these discriminations. Similar findings have been reported by Bussey et al. (1996, 1997) who found no impact of RSC lesions on the early phases of a conditional visual discrimination task in which correct operant responses were reinforced with food. Bussey et al. (1996, 1997) did report, however, that RSC lesions impaired performance at later stages of training, consistent with the findings of Gabriel et al. (1987). Nevertheless, the fact that RSC lesions have no impact on operant discriminations (Nelson et al., 2014) or only impact performance late in training (Bussey et al., 1996, 1997) suggests that RSC damage does not impair the processing of sensory qualities of cues.

3.2.2. Feature-negative discrimination learning.

The role of the RSC in discrimination learning has also been examined with procedures in which reinforcement or nonreinforcement of one cue depends upon the presence of a second cue (Figure 5b). For example, Keene & Bucci (2008b) trained rats on an appetitive feature-negative discrimination (e.g., Pavlov, 1927). On one set of trials, a tone (i.e., the “target”) was paired with reinforcement, but on other trials a tone and a light (i.e., the “feature”) were presented simultaneously and reinforcement did not occur. Control rats learned to discriminate the two types of trials, showing high levels of responding to the target alone, but low levels of responding to the feature-target compound. In contrast, rats with pre-training electrolytic lesions of the RSC were impaired on this discrimination, responding at similarly low levels on both types of trials. Using a slightly different procedure, Robinson et al. (2011) produced complementary results. In their experiment, rats were once again presented with tone-food trials. The difference was that on non-reinforced trials, the light preceded presentation of the tone. Specifically, the light was presented for 5 s, followed by a 5 s gap, then the tone was presented. Robinson et al. (2011) found that pre-training electrolytic lesions of the RSC impaired this discrimination. Once again, lesioned rats showed overall lower levels of responding, but importantly they did not discriminate between reinforced and non-reinforced trials. It is worth noting that in the experiments by Keene & Bucci (2008b) and Robinson et al. (2011), electrolytic lesions of the RSC produced impairments in responding to both the reinforced cue and the non-reinforced compound cue. That is, lesions produced an overall decrease in conditioned responding. Since the groups were at different points on the response scale, this makes comparison of the size of the discrimination across groups difficult.

The different procedures used by Keene & Bucci (2008b) and Robinson et al. (2011) are often thought to produce different forms of learning (for a review see Holland, 1992). When the two cues are presented simultaneously (Keene & Bucci, 2008b), the feature is typically thought to acquire a direct inhibitory association with the reinforcer, often described as “conditioned inhibition” (Rescorla, 1969; see also Rescorla, (1975) for a discussion of the nature of inhibitory associations). However, when the cues are presented serially (Robinson et al., 2011), the feature often does not acquire a direct inhibitory association with the reinforcer, but instead modulates the association between the target cue and the reinforcer, often described as “negative occasion-setting”. The fact that pre-training electrolytic lesions impair acquisition of both discriminations suggests a role for the RSC in both conditioned inhibition and negative occasion setting. However, neither the experiment by Keene & Bucci (2008b) nor Robinson et al. (2011) included independent tests to assess how the rats solved the discrimination. Therefore, it is not possible to confirm the status of the inhibitory learning in either study.

A more recent study by Nelson et al. (2018a) has examined the role of the RSC in an appetitive feature-negative discrimination and included direct assessment of the status of the putatively inhibitory cue. Rats with either Sham or pre-training neurotoxic lesions of the RSC received conditioning in which a tone was reinforced (A+), but on other trials the tone was presented in a simultaneous compound with a light stimulus (house-light off), and reinforcement did not occur (AX−). In contrast to the findings of Keene & Bucci (2008) and Robinson et al. (2011), Nelson et al. (2018a) found that neurotoxic lesions of the RSC had no impact on acquisition of the feature-negative discrimination. Furthermore, these authors included direct assessment of the inhibitory status of cue X (house-light off). Whether or not a cue has acquired inhibitory properties is typically assessed through tests of summation and retardation-of-acquisition (Rescorla, 1969; cf. Papini & Bitterman, 1993; Williams, Lolordo, & Overmeir, 1992). In summation, presentation of a conditioned inhibitor in combination with a conditioned excitor results in less overall conditioned responding than when the excitor is presented alone. In retardation tests, a conditioned inhibitor is slow to acquire conditioned excitatory properties relative to a control cue. To test for summation, Nelson et al. (2018a) first trained an additional excitor (panel light; cue B) and then tested cue B alone or in combination with cue X. Sham rats showed less responding to the compound of cue BX relative to cue B, whereas rats with lesions of the RSC showed similar levels of responding on both trials. Finally, to test for retardation, cue X and a relatively novel Y cue (magazine light) were paired with reinforcement. Both Sham and RSC lesioned rats acquired conditioned responding slower to cue X than to cue Y, and responding to each cue did not differ between groups. Based in part on these findings, Nelson et al. (2018a) suggested that the RSC is critically important when there is a mismatch between previously acquired learning and the current situation. In particular, they note that the RSC-lesion induced impairments are especially likely to emerge when animals are tested in the absence of reinforcement.

Overall, the role of the RSC in feature-negative discriminations is not clear. One possible reason for the mixed results is the lesion methods employed in each study. The studies by Keene & Bucci (2008b) and Robinson et al. (2011) both involved electrolytic lesions of the RSC, damaging cell bodies and fibers of passage. In contrast, neurotoxic lesions of the RSC, which presumably spare fibers of passage, do not impact feature negative discrimination (Nelson et al., 2018a). Thus, it is possible that damage to fibers of passage, and not cells within the RSC, are responsible to poor performance on feature-negative discriminations. It is additionally possible that the specific behavioral procedures used in each experiment may have influenced the involvement of the RSC. As just one example, Keene & Bucci (2008b) presented rats with 4 reinforced trials and 12 nonreinforced trials each session. On the other hand, Nelson et al. (2018a) presented rats with an equal number of reinforced and nonreinforced trials. These differences in the ratio of trial types may have subtly altered the learning mechanisms engaged during acquisition of the discrimination.

3.2.3. Negative patterning.

An additional discrimination procedure that involves multiple cues is negative patterning (Figure 5c). In this procedure, two different cues are reinforced when presented alone as elements but are non-reinforced when presented in compound. For example, Cue A and Cue B are reinforced when separately presented, but nonreinforced when presented as a single “AB” compound. Successful discrimination requires organisms to respond more to A and B when presented as elements, compared to when they are presented as a compound. To do so, the “AB” compound must necessarily be represented in some way as a unique stimulus (e.g., Pearce, 1987; 1994; Wagner & Rescorla, 1972). In the absence of such representation, the negative patterning discrimination would be unsolvable. Indeed, the strength of the association between each cue and the reinforcer might be expected to sum on “AB” trials (e.g., Rescorla & Wagner, 1972), and thus responding to “AB” would paradoxically be greater than to A or B alone (Bouton, 2016; Redhead & Pearce, 1995).

The fact that the negative patterning discrimination requires processing and integration of multiple cues would perhaps suggest a potential role for the RSC. To test this possibility, Nelson et al. (2018a) trained rats with pre-training excitotoxic lesions of the RSC on an operant negative patterning task. In this experiment, lever presses during the A or B stimulus were reinforced, but lever presses during the AB compound were not reinforced. Nelson et al. (2018a) reported that although RSC lesioned rats appeared to acquire the discrimination more slowly than control rats, there was no statistical evidence that RSC lesions in fact attenuated acquisition of the discrimination. Thus, the available evidence indicates that the RSC is not necessary for negative patterning discrimination learning. One caveat with pre-training lesions, especially when a behavioral deficit is not observed, is that other unspecified regions might compensate for the damaged region, making it difficult to detect a behavioral impairment. Thus, it may be important to examine the impact of post-training RSC lesions on the maintenance of the discrimination.

3.3. Higher-order Conditioning

3.3.1. Sensory preconditioning.

Discrete cues can acquire the ability to elicit conditioned responding even if they are not directly paired with reinforcement. In sensory preconditioning, two initially neutral cues, such as a tone and a light, are repeatedly presented together. One cue (e.g., the light, “S1”) is then repeatedly paired with reinforcement (Figure 6a). As a consequence of reinforcement of S1, the tone (“S2”) will elicit a conditioned response, despite never being directly paired with reinforcement. Sensory preconditioning has been taken as evidence for the formation of stimulus-stimulus (S-S) associations. That is, during the initial phase of the experiment, an association is formed between the two sensory cues (Rizley & Rescorla, 1972, Experiment 4).

Figure 6.

Schematic of higher-order conditioning procedures. Rectangles represent cues, with colors differentiating stimuli, and circles represent outcomes such as food or shock. A) Sensory preconditioning. B) Second-order conditioning.

To date, two studies have used sensory preconditioning to establish a role for the RSC in the acquisition of stimulus-stimulus associations. For example, Robinson et al. (2011) presented two groups of rats (RSC lesions vs. Shams) with two types of intermixed trials. On one set of trials, a 10 s auditory cue (Preconditioned cue; either tone or white noise) was immediately followed by a 5 s visual cue. On a second set of trials, an auditory cue (Unpaired cue; either tone or white noise) was presented alone. In the second phase of the experiment, the visual cue was then paired with food reinforcement. During the final test session, Sham lesioned rats responded more on Preconditioned cue trials than on Unpaired cue trials. In contrast, rats with electrolytic lesions of the RSC did not respond differentially on the two trial types. Robinson et al. (2011) concluded that pre-training lesions of the RSC impaired the ability to acquire S-S associations.

In the experiment by Robinson et al. (2011) permanent lesions of the RSC occurred prior to any behavioral training, thus impairing RSC function for all phases of the experiment. Interestingly, RSC lesions did not impact acquisition of conditioned responding to the visual cue paired with reinforcement, consistent with the idea that the RSC is not necessary for Pavlovian delay conditioning. Nevertheless, in a separate study, Robinson et al. (2014) demonstrated impaired sensory preconditioning when the RSC was selectively silenced with DREADDs, only during the preconditioning phase of the experiment. This finding is important because it indicates that RSC is necessary precisely for forming the association between S1 and S2. In addition, the results indicate that the deficit observed by Robinson et al. (2011) cannot be fully accounted for by the use of electrolytic lesions.

3.3.2. Second-order conditioning.

A second form of higher-order conditioning is second-order conditioning (SOC). SOC is conceptually similar to sensory preconditioning, with the exception that the phases of the experiment are “flipped”. In SOC, one cue (S1) is first paired with reinforcement. In the second phase of the experiment, S2 is then paired with S1 (Figure 6b). On the basis of this, S2 will elicit a conditioned response, despite the fact that it was never directly paired with the reinforcing outcome. Using a conditioned suppression procedure, Todd et al. (2016a) found that pre-training electrolytic lesions had no impact on SOC. Rats were first trained with one visual cue that predicted shock (V1+) and a control cue that predicted the absence of shock (V2−). In the second phase of the experiment, A1 was paired with V1 in a serial fashion (A1 then V1) and A2 was paired with V2 (A2 then V2). Conditioned suppression was acquired to A1, but not A2, indicating the development of SOC. However, SOC was acquired equally for control rats, and rats with pre-training lesions of the RSC.

At first glance, the results of Todd et al. (2016a) and Robinson et al. (2011; 2014) appear at odds. With similar procedures, the RSC is necessary for one form of higher-order conditioning (sensory preconditioning) but not another (second-order conditioning). Although sensory preconditioning and SOC are similar procedures, they may produce very different underlying associations. For example, while sensory preconditioning is often thought to result in S-S associations during phase 1 (e.g., Rescorla & Cunningham, 1978), SOC has been characterized as the formation of S-R associations during phase 2. Specifically, in SOC, instead of S2 being associated with S1, it appears that S2 becomes associated with the response elicited by S1 (Rescorla, 1973; 1977; cf. Winterbauer & Balleine, 2005). Additional experiments are required to directly test the role of appetitive and aversive reinforcement, and whether RSC involvement in higher-order conditioning is dependent upon S-S rather than S-R associations.

4. RSC Function: The Many Influences of Context

The preceding sections have provided evidence that the RSC contributes to learning and memory for auditory and visual cues in putatively non-spatial paradigms. Damage to the RSC produces deficits in a range of conditioning paradigms. However, when considering the most selective manipulations (e.g., neurotoxic lesions, temporary inactivation), the RSC has been found to consistently impair trace fear conditioning (acquisition, retrieval, and extinction), the retrieval of remotely acquired conditioning (both delay and trace), and sensory preconditioning. It has proven difficult, however, to conceptualize the role of the RSC in cue-specific learning. We suggest that RSC involvement in many aspects of cue-specific learning and memory can be understood in terms of a role for the RSC in contextual learning and memory, especially considering the multiple ways by which contexts can influence responding to discrete cues.

4.1. Multiple Roles of Context in Learning and Memory

Learning and memory for discrete cues inevitably occurs against an environmental backdrop composed of a variety of stimuli. This backdrop, or “context” is often operationally defined as the experimental apparatus, and contexts typically differ in their visual, tactile, and olfactory characteristics. With experience, these individual features of the context become integrated into a unified contextual representation (Fanselow, 2010). Once this context representation is acquired, it can then serve multiple functions (Holland & Bouton, 1999), including influencing responding to discrete cues. Below we describe several roles of the context in learning and memory, and highlight the role of the RSC.

4.1.1. RSC and contextual conditioning

Contexts can become directly associated with outcomes and subsequently elicit CRs. In contextual fear conditioning experiments, rats and mice will freeze when re-exposed to a context previously paired with mild-foot shock. There is considerable evidence that the RSC is involved in contextual fear conditioning (for a review, see Corcoran et al., 2018). For example, pre- and post-training lesions of the RSC attenuate contextual fear conditioning (Keene & Bucci, 2008a), RSC neurons are active during the retrieval of contextual fear conditioning (e.g., Tayler et al., 2013), and optogenetic activation of RSC neurons can drive expression of contextual fear conditioning (Cowansage et al., 2014). Furthermore, RSC involvement in the retrieval of contextual fear conditioning is long-lasting; retrieval of both recent and remotely acquired contextual fear memories is RSC dependent (Corcoran et al., 2011; Tayler et al., 2013; Todd et al., 2016b).

4.1.2. An indirect influence of the context

Apart from directly eliciting a conditioned response, contextual conditioning can also indirectly influence conditioning to a discrete auditory or visual cue. Anytime a cue is reinforced, all cues present, including the background context, can potentially acquire an association with the reinforcer. Furthermore, the amount of conditioning that occurs on a given trial will depend upon how well the reinforcer is already predicted by other co-present cues (e.g., Rescorla & Wagner, 1972). Thus, if the reinforcer is already strongly predicted by contextual cues, then little conditioning will accrue to a discrete auditory or visual cue when it is subsequently paired with reinforcement (Randich, 1981).

It is interesting to note that while Gabriel et al. (1987) reported potentiated responding to the CS+ during the first session of conditioning following RSC lesions, in that experiment all subjects had been extensively pre-exposed to both the CS and the US (Kang & Gabriel (1998) reported similar findings following combined subiculum and retrosplenial lesions). It is possible that prior exposure to the US produced strong contextual conditioning that then slowed conditioning to the CS. Considering the role of the RSC in contextual conditioning, it is likely that RSC lesions resulted in weak contextual conditioning following pre-exposure to the US, which would reduce the ability of the context to “block” learning about the CS+. Thus, the facilitated conditioning observed to the CS+ may have been indirectly influenced by a reduction in contextual conditioning resulting from RSC lesions.

4.1.3. Context-specific associations

In the preceding sections we have described how contextual conditioning (i.e., context-outcome associations) can directly elicit conditioned responding, and can also influence conditioning to a discrete auditory or visual cue. In addition, context-specific associations can develop between cues and the contexts in which they occur. For example, contexts and cues can be directly associated with each other, or they can be encoded together as a unique configuration. Contexts can also enter into hierarchical associations, by signaling the relationship between the cue and reinforcer (i.e., “occasion setting”). In all of these examples, cues and contexts interact to influence learned behavior (Holland & Bouton, 1999).

A variety of findings suggest a role for the RSC in context-specific associations. For example, Talk et al. (2005) found that pre-exposure to a CS (i.e., latent inhibition) reduced CS elicited neuronal activity in the RSC. As previously noted, pre-exposure to a CS is often thought to result in a context-specific association (Westbrook & Bouton, 2010). Consistent with this, latent inhibition of the avoidance response was abolished following context extinction (i.e., exposure to the context alone), which presumably weakened any prior association between the context and the CS. In addition, RSC activity also recovered following context extinction, leading Talk et al. (2005) to suggest that the RSC contributes to the contextual coding of latent inhibition. This suggestion is consistent with earlier recording studies demonstrating RSC neurons encode context-specific associations in complex context discriminations (Freeman et al., 1996). For example, in one study, rabbits were trained on an approach discrimination in which one CS was paired with water (CS+), and the other was nonreinforced (CS−). On alternating days, the same rabbits were trained on an avoidance discrimination, in which one CS was paired with shock (CS+) and the other was nonreinforced (CS−). The approach discrimination occurred within an operant chamber, whereas the avoidance discrimination occurred within a running wheel. Importantly, the approach CS+ served as the avoidance CS−, and the approach CS− served as the avoidance CS+. Rabbits were able to acquire and maintain both discriminations. More so, multi-unit activity in the RSC exhibited context-specific patterns of activity. That is, RSC units exhibited discriminative activity for both discriminations (greater activity to the CS+ than CS−) despite the fact that the CS+ for one discrimination served as the CS− on the other discrimination. Finally, electrolytic lesions of the RSC impair the ability to associate a discrete cue, such as a tone, with a unique context (Robinson et al., 2018). Thus, both electrophysiological and lesion analysis suggests a role for the RSC in context-specific associations.

4.2. Cues and Contexts: Putting it All Together

We have described multiple ways that the context in which cues are presented can influence learned behavior, and have suggested a role for the RSC in many of these context functions. We suggest that one function of the RSC is to integrate contextual information with information related to cues, and potentially reinforcing outcomes. Indeed, recent evidence from navigation studies suggests that RSC neurons can jointly encode reinforcing outcomes with specific locations (Vedder et al., 2017; see Smith et al., 2018). We further suggest that integration of cue and context information within the RSC is a slow process, consistent with Gabriel (1990), who described a gradual engagement of the RSC in discriminative avoidance conditioning (for a discussion of rapid vs. slow engagement of the RSC in learning and memory, see Smith et al., 2018).

4.2.1. Novel applications: remote memory retrieval and trace conditioning

As we have described, a number of studies have suggested a dissociable role for the RSC in the retrieval of recently vs. remotely acquired delay fear conditioning (Kwapis et al., 2014; Jiang et al., 2018; Todd et al., 2016). One important feature of these studies is that they typically involve only a few conditioning trials (e.g., 3 CS-US pairings). Cue information is often encoded along with the context early in training (i.e., after a few trials), but the role of the context is reduced with additional conditioning trials (see Rosas et al., 2014). Thus, in these experiments, it is possible that cue and context information was jointly encoded. If RSC engagement in delay conditioning is a gradual process, then with few conditioning trials the initial memory trace encoded within the RSC would be expected to be weak. Thus, there would be little contribution of the RSC to behavior until the memory trace gradually strengthens, in this case over the course of the prolonged retention interval (see Figure 7).

Figure 7.

Integration of cue, context, and outcome information within the RSC. Initially, connections within the RSC are weak (left panel). However, with a protracted consolidation period, the connections within the RSC strengthen.

The contribution of the RSC to trace conditioning, at both recent and remote time points, may also be related to contextual conditioning. For instance, trace conditioning produces greater contextual conditioning than delay conditioning (e.g., Marlin, 1981). Trace conditioning might also require the discrimination between two context alone periods (Bolles et al., 1978; see also Chowdury et al., 2005); the period between CS-offset and outcome delivery, and the period between the outcome and the subsequent CS. Further, Quinn et al., (2002) have noted that during conditioning the CS may become associated with the conditioning context itself. Thus, although we have proposed that RSC engagement in cue-specific learning is a gradual process, the RSC may be required for trace conditioning at a recent time point if trace conditioning relies more upon contextual processing than delay conditioning.

4.2.2. Predictions

We have suggested an overarching function of the RSC is the integration of cue and context information, and that engagement of the RSC occurs gradually. This perspective makes several novel predictions First, any procedure that reduces the influence of the context should also reduce the role of the RSC. For example, we have noted that studies demonstrating a role for the RSC in the retrieval of remote memory typically involve only a few conditioning trials and thus likely results in coding of the context along with the CS. If the influence of the context was sufficiently reduced, we would predict retrieval of remotely acquired conditioning to be RSC independent. Second, the idea that RSC involvement in remote memory retrieval is related to a gradual engagement of the RSC over the course of a prolonged retention interval leads to the prediction that blocking this engagement should ultimately reduce the role of the RSC in retrieval of remote memory. This could be accomplished using chronic administration of CNO to activate inhibitory DREADD receptors in the RSC during the consolidation period.

Third, if the RSC is involved in the integration of context and cue information, then manipulation of the RSC should attenuate the development of context-specific associations. For example, lesions or temporary inactivation of the RSC would be expected to reduce any influence of context otherwise observed. Finally, with respect to recently acquired conditioning, we predict that procedures that strengthen the memory trace, even when acquired at a recent time point, should promote RSC dependency. Indeed, Gabriel et al., (1987) reported pre-training lesions of the RSC produced deficits in responding to a CS+ following overtraining. The overtraining procedure may thus have strengthened the memory trace sufficiently to produce RSC involvement. It is especially important to note, however, that the discriminative avoidance procedure used by Gabriel produced context-dependency of conditioning even following many conditioning trials (Freeman et al., 1997). This is important because in many procedures additional conditioning trials can reduce the role of context, which according to our perspective should reduce the role of the RSC.

4.2.3. Unresolved Issues

Although we have described a role for the RSC in learning and memory for context-specific associations, the main exception to this is its role in extinction learning. Despite the fact that extinction results in context-dependent learning, several studies have demonstrated that lesions or temporary inactivation of the RSC have no impact on extinction of delay conditioning (Kwapis et al., 2014; Todd et al., 2016a, 2017). However, it would be premature to suggest that the RSC has no role in extinction learning. As noted earlier, Gabriel et al., (1987) reported a change in the form of the CR during extinction for rats with lesions of the RSC. In addition, Freeman et al., (1997) reported a suppression of RSC multi-unit activity in the RSC during extinction of discriminative avoidance learning, when extinction occurred in the original conditioning context. However, when extinction occurred in a novel context, and thus no mismatch between the context and the US, discriminative multi-unit activity persisted. That is, there continued to be more activity to the CS+ than the CS−. Finally, it may be the case that the RSC’s involvement in extinction is dependent upon its involvement in the initial conditioning. This was suggested by Kwapis et al., (2014), who found that the RSC was not involved in either the retrieval or extinction of delay conditioning, but was involved in both retrieval and extinction of trace conditioning.

4.3. Interim Summary

Learning and memory for discrete cues auditory or visual cues always occurs within a backdrop of contextual stimuli. Here we have summarized several ways by which contextual stimuli can influence learned behavior. In acknowledging the multiple roles for contextual stimuli in learning, memory, and behavior, we have suggested that RSC involvement in cue learning may in fact be related to its involvement in contextual learning and memory. Specifically, we have suggested that the RSC might gradually integrate information related to cues, context, and reinforcing outcomes. The integration of contextual information with cue learning may help explain the RSC’s role in the retrieval of remotely acquired conditioning, as well as trace conditioning. Finally, although this perspective generates novel predictions, there remains outstanding issues. For example, although we have described the RSC as integrating cue and context information, the precise nature of this integration is not clear. Indeed, we have described how cues and contexts can interact in several different ways, and it is possible that several of these roles are sub-served by the RSC.

In many ways this framework is an elaboration upon prior theoretical conceptualizations. In particular, the present framework draws heavily from the work of Gabriel and colleagues, whose analysis of RSC involvement in discriminative avoidance conditioning emphasized a role of the RSC in context-specific memory retrieval (Gabriel, 1993). Furthermore, the suggestion that cue and context information is encoded within the RSC and then gradually strengthens, either over time or with further conditioning, is perhaps consistent with a working model suggested by Miller et al., (2014). These authors suggested that early in training plasticity within the RSC encodes sensory information, which is then consolidated into a stable memory trace through hippocampal-RSC interaction. Finally, the current framework is also informed by the conceptualization of Bucci & Robinson (2014), that the RSC serves to integrate sensory information in the service of S-S associations. A role for the RSC in the formation of S-S associations is perhaps consistent with RSC involvement in contextual learning and memory, which involves integrating a variety of environmental stimuli into a cohesive context representation (see Bucci & Robinson, 2014). Indeed, a role for the RSC in sensory integration may help explain its contributions to some forms of discrimination learning (Bucci & Robinson, 2014; cf. Nelson et al., 2018a). However, the current conceptualization expands upon that of Bucci & Robinson (2014) by emphasizing the interaction between contexts and cues, and reinforcing outcomes. In addition, the current conceptualization is agnostic to the type of associations that are formed between contexts, cues, and outcomes. These associations could be S-S in nature, but they could also take other forms, such as configural or hierarchical associations.

It remains to be determined how our conceptualization of RSC contributions to cue-specific learning extends beyond Pavlovian and instrumental conditioning to spatial processing and navigation. However, there are some notable similarities. For instance, RSC neurons encode behavioral relevant auditory cues (Gabriel et al., 1993) as well as navigational cues (Vedder et al., 2017; for a discussion see Smith et al., 2018). Further, lesions of the RSC impair the detection of objects in novel spatial locations (“object in place” task; Van & Aggleton, 2002; Van et al., 2009), which may bear similarity to learning about a particular cue in a particular context. Nevertheless, future research is required to determine if RSC contributions to spatial and non-spatial processing is sub-served by a common process.

5. Is RSC Function Selective or Shared?

The majority of experiments examining RSC contributions to cue-specific learning and memory have focused the RSC itself, and have not examined connections between the RSC and other regions (cf. Todd et al., 2018). However, assessment of hippocampal and cortical contributions to RSC-dependent forms of cue-specific learning and memory may shed light on the larger networks supporting these functions.

5. 1. Hippocampal contributions to delay and trace conditioning

Initial experiments suggested little role for the hippocampus in learning about single cues that predicted aversive outcomes. For example, using a delay fear conditioning procedure, Kim and Fanselow (1992) reported no effect of post-training dorsal hippocampus (DH) lesions on conditioned freezing. More recently, however, it has become clear that the role of the DH in delay fear conditioning is parameter specific. For example, Quinn et al., (2008) reported that post-training lesions of the DH impaired freezing to a delay conditioned auditory cue when the training procedures produce a weak, but not strong, association between the CS and US. In contrast, the RSC typically has very little role in recently acquired delay fear conditioning.

The RSC and hippocampus might likewise have reciprocal roles in the retrieval of remotely acquired delay fear conditioning. Maren et al., (1997) observed that post-training lesions of the DH impaired retrieval of delay fear conditioning to an auditory cue. However, this deficit appeared to be at least somewhat temporally graded. Although a lesion deficit was following a 1- and 28-day retention interval, lesioned and sham rats did not differ following a 100-day retention interval. In contrast to delay conditioning, recently acquired trace conditioning appears to be dependent upon both the hippocampus and RSC (e.g., Kwapis et al., 2014, 2015; Beeman et al., 2013; Quinn et al., 2002). However, for remotely acquired trace conditioning, there once again is a dissociation between the RSC and hippocampus. While remotely acquired trace conditioning does not depend upon the DH (Beeman et al., 2013), it does depend upon the RSC (Todd et al., 2016).

The overall pattern of findings suggests a potential interplay between the RSC and hippocampus during delay and trace conditioning. Recently acquired delay conditioning is hippocampal dependent when the association is weak, but not strong, while the RSC has little role in recently acquired delay conditioning. The RSC, but not the hippocampus, has a critical role in the retrieval of remotely acquired delay conditioning. For trace conditioning, the hippocampus and RSC appear necessary for recent conditioning, however the RSC has a protracted role in the retrieval of trace conditioning, and the hippocampus does not. One possibility is that a cortico-hippocampal network encodes context-specific associations, and that the role of the RSC in this network increases as the memory strengthens over time.

5.2. Cortico-hippocampal contributions to sensory preconditioning

Perhaps related to its role in contextual learning and memory, there is some evidence that that damage to the RSC impairs subjects’ ability to link or associate environmental stimuli. For example, lesions or temporary inactivation of the RSC impairs sensory preconditioning in rats (Robinson et al., 2011, 2014). In addition, pre-training lesions of the perirhinal cortex impair sensory preconditioning in eyeblink conditioning in rats (Nicholson et al., 2000). The role of the hippocampus in sensory preconditioning, however, is not clear. For example, pre-training lesions of the hippocampus have been shown to impair sensory preconditioning of the rabbit’s nictitating membrane response (Port et al., 1987) and conditioned suppression in rats (Talk et al., 2002). In both of these studies, the sensory stimuli were either visual or auditory cues. In contrast, using a conditioned flavor aversion procedure, Ward-Robinson et al. (2001) reported no impact of pre-training hippocampal lesions on sensory preconditioning. Nevertheless, human imaging studies have suggested a role for the hippocampus in sensory preconditioning (Yu et al., 2014; cf. Wimmer & Shohamy, 2012), which has been incorporated into a recent theoretical perspective (Wikenheiser & Schoenbaum, 2016). On the whole, associations between sensory stimuli thus appear to be mediated by the hippocampus and cortical regions including the RSC.

7. Evidence from Human Imaging Studies

In their review, Vann et al., (2009) noted that human imaging studies have found RSC modulation in a “bewildering array of tasks or procedures” (pg. 797). Indeed, human imaging studies have suggested a role for the RSC in mental imagery, self-referential processing, imagining and planning for the future (for reviews see Chrastil, 2018; Vann et al., 2009). In particular, the RSC is consistently active during episodic memory retrieval (Chrastil, (2018; Vann et al., 2009). Ranganath and Ritchy (2014) have described the RSC as part of a “posterior-medial network” (PM system) that also includes the postrhinal cortex, mammilary bodies, anterior thalamic nuclei, pre- and para subiculum and the default mode network. They describe the PM system as contributing to a range of functions, including episodic memory, that share the common theme of the construction and utilization of a “situation model”. Ranganath and Ritchy describe situation models as similar to schemas, “that specify the gist of the spatial, temporal, and causal relationship that apply within a particular context” (pg. 721).

Episodic memories are often considered to contain “what”, “where”, and “when” information. Thus, information about a specific item, object, person or event (i.e., “what”) is encoded within a spatial (i.e., “where”) and temporal (i.e., “when”) context. Here we have emphasized that interactions between cues and contexts may be particularly important for engaging the RSC. Thus, the RSC’s role in episodic memory is consistent with our suggestion that the RSC’s role in cue-specific learning and memory is intimately tied to the context. However, future experiments may examine the role of RSC in the integration of “what”, “where”, and “when” information (e.g., Iordanova et al., 2009).

8. Where Does This Leave Us, and Where Do We Go From Here?

We have reviewed and assessed experiments examining the role of the RSC in Pavlovian and instrumental conditioning, and based on this review, we suggest the following summary points. First, a significant number of studies have consistently reported that the RSC is not necessary for the acquisition, retrieval, or extinction of Pavlovian delay conditioning when a single cue is paired with a single outcome and tested shortly after conditioning. Second, manipulations of the RSC consistently impair trace fear conditioning (acquisition, retrieval, and extinction), the retrieval of remotely acquired conditioning, and sensory preconditioning. Finally, manipulations of the RSC may impair some, but not all, forms of discrimination learning.

How should the role of the RSC in cue-specific learning and memory be best described? Several possibilities have been suggested, and it remains to be determined if RSC contributions to cue-specific learning can be distilled down to a single function, or if instead the RSC performs a range of functions. Prior conceptualization has emphasized RSC contributions to the formation of S-S associations (Bucci & Robinson, 2014), and cognitive control under conditions of response conflict (Nelson et la., 2014). It has also previously been suggested that RSC contributions to behavior may emerge when there is a mismatch between previous learning and the current contingencies (Nelson et al., 2018a). Here, we have suggested that the role of the RSC in cue-specific learning may be intimately tied to contextual learning and memory. According to this conceptualization, the RSC may be especially engaged in situations where cue and context information interact to influence learned behavior. We have further suggested that RSC engagement may occur slowly (e.g., Gabriel, 1990), perhaps over conditioning trials, or over the course of an extended consolidation interval.

Importantly, this conceptualization has provided testable predictions with respect to when the RSC should be involved in cue-specific learning and memory. In addition, we have noted empirical findings that are inconsistent with this approach. For example, future research is required to fully understand the role of the RSC in extinction, a form of context-specific learning. The possibility of shared function between the RSC and other regions, such as the hippocampus, may complicate the interpretation of lesion or inactivation studies, especially if damage to the RSC can be compensated for by other regions. Multi-unit activity recording within the RSC during discriminative avoidance learning has provided valuable information about the role of RSC in cue-specific learning and memory. However, additional in vivo physiology experiments (e.g., single unit recordings, calcium imaging) that examine RSC task coding may provide a more nuanced insight into the functions / computations performed within this region.

Box 1.

Common methods used to study retrosplenial contributions to cue-specific learning, memory, and behavior.