The amygdala mediates the facilitating influence of emotions on memory through multiple interacting mechanisms

Abstract

Emotionally arousing experiences are better remembered than neutral ones, highlighting that memory consolidation differentially promotes retention of experiences depending on their survival value. This paper reviews evidence indicating that the basolateral amygdala (BLA) mediates the facilitating influence of emotions on memory through multiple mechanisms. Emotionally arousing events, in part by triggering the release of stress hormones, cause a long-lasting enhancement in the firing rate and synchrony of BLA neurons. BLA oscillations, particularly gamma, play an important role in synchronizing the activity of BLA neurons. In addition, BLA synapses are endowed with a unique property, an elevated post-synaptic expression of NMDA receptors. As a result, the synchronized gamma-related recruitment of BLA neurons facilitates synaptic plasticity at other inputs converging on the same target neurons. Given that emotional experiences are spontaneously remembered during wake and sleep, and that REM sleep is favorable to the consolidation of emotional memories, we propose a synthesis for the various lines of evidence mentioned above: gamma-related synchronized firing of BLA cells potentiates synapses between cortical neurons that were recruited during an emotional experience, either by tagging these cells for subsequent reactivation or by enhancing the effects of reactivation itself.

1. Introduction

As we go about our lives, we experience a variety of events, most of which we forget. By contrast, we tend to form long lasting and vivid memories for emotionally arousing or stressful events. This review will examine the basis for the differing fate of neutral vs. emotionally laden memories. Although many reviews on this topic have been published, none focused on the neurophysiological mechanisms underlying the facilitation of memory by emotions. We will draw from various disciplines, reviewing established mechanisms as well as emerging concepts. Because literally thousands of studies have been published on some of these topics, this review cannot be exhaustive. Instead, we will describe these areas in broad strokes, often relying on excellent reviews that appeared previously.

We will first consider the concept of memory consolidation (Section 2), the notion that memories are not fully formed right after an experience but change over time. Supporting this concept is evidence that the brain is endowed with systems determining to what extent memories are consolidated, even though they are not their storage site. This will lead us to focus on the role of the amygdala and its recruitment by stress hormones (Section 3), which index the arousing consequences of our experiences and act retrogradely to affect memory of those experiences. Next, asking how the amygdala affects memory consolidation (Section 4) will steer us toward less established territory. We will consider various possibilities, including evidence that the synapses formed by amygdala neurons facilitate synaptic plasticity where memories are stored, and that amygdala activity promotes interactions between the neocortex and hippocampus. Next, we will relate the spontaneous recollections of emotional experiences during wake and sleep (Section 5) to the notion of replay and speculate about their potential role in the consolidation of emotionally laden memories. This will lead us to consider the role of paradoxical sleep and theta oscillations in the consolidation of emotional memory (Section 6), including evidence that spontaneous network events seen in different sleep stages support off-line replay of salient waking experiences. We will conclude with recommendations for future research (Section 7), drawing attention to questions and phenomena that have been overlooked.

2. Memory consolidation

The concept of memory consolidation originated from an unexpected observation made by Müller and Pilzecker (1900). Subjects tasked with learning lists of nonsensical syllable pairs reported experiencing involuntary and intrusive mental rehearsal of the paired associates in between training sessions. This led Müller and Pilzecker (1900) to speculate that such post-learning mental rehearsal might serve to reinforce the associations between syllables. Supporting this idea, they found that recall of a recently learned list of syllables was impaired when subjects learned other material shortly after the first learning. While such retroactive interference lends itself to another interpretation, namely that learning new information disrupts old memory traces, subsequent work bolstered the ideas of Muller and Pilzecker.

Indeed, many other types of post-learning interventions were later shown to interfere with recent memories, including electroconvulsive shocks (Duncan, 1949), drug injections (McGaugh, 1966), and electrical stimulation of particular brain regions (McGaugh and Gold, 1976). Critically, some of these interventions actually enhanced memories, which is incompatible with the interference account. Consistent with the consolidation model, the effect of these post-learning treatments diminished with time from the original learning. Also, because these interventions occurred after training but long before testing retention, their impact could not be ascribed to performance deficits at encoding or during the recall tests.

Among the post-learning treatments found to facilitate long-term retention were the systemic administration of stress hormones or their pharmacological analogs (reviewed in McGaugh and Roozendaal, 2002). These findings offered an explanation for the commonplace observation that emotional experiences are typically remembered better than neutral events (Christianson, 1992). They suggested that when released during an emotionally arousing experience, stress hormones could act retrogradely to affect memory of that event. The extent to which a memory would be consolidated would depend on the arousing consequences of an experience, as indexed by the release of stress hormones (Gold and McGaugh, 1975). Thus, memory consolidation would be adaptive in that it allows for differential retention of experiences depending on their survival value.

As detailed below, emotional arousal facilitates different types of memories. This is significant given that different types of memories depend on distinct neuronal systems, all of which receive dense projections from the basolateral complex of the amygdala (BLA; reviewed in Pitkanen, 2000). Memories are typically divided into two major categories: procedural and declarative (Sherry and Schacter, 1987). Declarative memories are further subdivided into episodic memories for events individuals experience in a specific context, and semantic memories for facts that are not linked to specific episodes. By contrast, procedural memory comprises a variety of phenomena, including habits, perceptual and motor skills, unconscious changes in response propensities like priming, and classically conditioned responses (Packard and Knowlton, 2002). A key feature of procedural memories is that they require extensive training, such that the learned behaviors become habitual, inflexible, and stimulus-specific.

Whereas many types of procedural memories depend on the striatum, the storage of novel episodic memories was shown to depend on the hippocampus (Scoville and Milner, 1957). Consistent with this, lesion of the rodent hippocampus interferes with the formation of contextual fear memories (Kim and Fanselow, 1992). Importantly, the hippocampus’ role in memory is time-limited, a feature thought to reflect a system-level consolidation process, whereby memories are progressively transferred from the hippocampus to the cortex for long-term storage (Alvarez and Squire, 1994). By contrast, disrupting neocortical activity at specific sites can interfere with the retrieval, consolidation, and encoding of memories, provided the targeted region processes the relevant information (Sacco and Sacchetti, 2010; Letzkus et al., 2011; Teixeira et al., 2006).

3. Role of the amygdala and its recruitment by stress hormones

3.1. Evidence implicating the amygdala in the facilitation of memory by emotions

How could a peripheral signal like the release of adrenal stress hormones affect memory consolidation in the brain? A breakthrough in this field occurred when it was realized that pre-training lesions of the BLA, but not of its central nucleus (CeA), prevents the facilitation of memory produced by the peripheral administration of glucocorticoids (Roozendaal and McGaugh, 1996; Roozendaal et al., 1996). Consistent with this, intra-BLA post-training infusions of drugs that increase or decrease the activity of BLA neurons, such as GABA receptor antagonists and agonists, respectively facilitated or impaired long-term memory formation (Castellano et al., 1989; Giachero et al., 2015; Huff et al., 2006; Salinas and McGaugh, 1995). As for stress hormones, the impact of manipulating BLA activity was maximal when it occurred right after training and then decayed with longer delays.

An influential study illustrates how the BLA contributes to memory consolidation. Here, Packard et al. (1994) compared the effects of post-learning amphetamine infusions in the BLA, hippocampus, or striatum on two versions of the water maze task, with a hidden or visible platform. Infusions of amphetamine into the caudate or hippocampus, respectively enhanced retention of the cued or spatial version of the task but had no effect on the other version. By contrast, immediate post-learning infusions of amphetamine in the BLA improved long-term memory in both versions of the task. Yet, intra-BLA lidocaine infusions performed shortly before the recall test did not reverse the effect of amphetamine on either task. Together, these findings indicate that the BLA facilitates the consolidation of different types of memory and that these memories are not stored in the BLA but in other structures, which receive inputs from the BLA.

Of note, BLA activity also regulates the consolidation of emotional memories in humans. For instance, humans with bilateral amygdala lesions do not show a facilitation of memory by emotions (Adolphs et al., 1997; Cahill et al., 1995). Moreover, the amount of amygdala activation at encoding correlates with long-term recall of emotionally arousing or neutral material (Cahill et al., 1996; Canli et al., 2000; Hamann et al., 1999).

Overall, the findings reviewed above suggest that in emotionally arousing conditions, stress hormones are released, causing an increase in the activity of BLA neurons, which in turn facilitates memory formation in target structures of the amygdala.

3.2. Recruitment of the amygdala by emotional arousal

In normal (non-experimental) circumstances, adrenal stress hormones are released under central control, as part of the organism's response to emotional arousal and stress. This response is complex, involving the release of multiple neuromodulators that act throughout the nervous system via varying complements of G-protein coupled receptors and associated signaling pathways, ultimately altering intrinsic neuronal excitability and synaptic transmission. Since reviewing this wide-ranging material is beyond the scope of this review, we will focus on the direct actions of adrenal stress hormones on the BLA, specifically on glucocorticoids given that adrenaline does not cross the blood-brain barrier.

Glucocorticoids readily permeate the blood–brain barrier and bind to two kinds of intracellular receptors: the high-affinity mineralocorticoid receptors (MRs), occupied at basal glucocorticoid levels, and the low-affinity GRs, which become fully activated at stress levels of glucocorticoids (de Kloet et al., 1990; Joëls, 2018). Moreover, through low-affinity GRs, glucocorticoids exert delayed and long-lasting effects by altering gene expression (Reichardt and Schutz, 1998). In addition to the classical delayed effects of glucocorticoids, rapid actions have also been reported.

Because of technical limitations associated with patch-clamp recordings, the delayed actions of glucocorticoids must be studied using between-cell comparisons; only the rapid effects can be studied using within-cell comparisons. Using the latter approach, it was found that corticosterone elicits a rapid and long-lasting increase in the frequency of miniature excitatory post-synaptic currents (Karst et al., 2010; Karst and Joëls, 2016) and the opposite in spontaneous GABAergic currents (Di et al., 2016).

By contrast, between-cell comparisons revealed that corticosterone produces a dramatic increase in the excitability of principal BLA neurons through multiple parallel mechanisms. These include a marked decrease in spike frequency accommodation (Fig. 1A–C), a slight depolarization of the resting potential, a rise in input resistance, and a reduction in the amplitude of GABA-A inhibitory postsynaptic potentials due to a positive shift of the chloride equilibrium potential (Duvarci and Paré, 2007). In addition, an increase in the amplitude of high-voltage-activated Ca²⁺ currents was reported (Karst et al., 2002).

Fig. 1.

Corticosterone (CORT) enhances the intrinsic excitability of principal BLA neurons. (A) On the basis of spike-frequency adaptation, BLA cells were classified into three groups: cells with rapid adaptation -RA, with intermediate adaptation - IA, and slow adaptation – SA. (B), Proportion of BLA cells in the RA, IA, and SA groups among CORT- (red, 100 nM) versus vehicle-treated (black) cells. (C) Graph plotting the number of evoked spikes (y-axis) as a function of injected current (x-axis) in CORT-treated (red) and vehicle-treated (black) cells matched for input resistance. CORT-treated cells fired significantly more spikes in response to 160–200 pA current injections. Group means ± SEM. *p < 0.05; **p < 0.001. (D) Firing rate (y-axis) of two different BLA neurons (D1,2) as a function of time (x-axis, 10-sec bins). Vertical lines indicate when a tone plus unexpected footshock were presented. Black and red circles, data obtained in W and SWS, respectively. Insets: Average firing rates ± SEM in W (black) and SWS (red), before (left, Pre) and after (right, post) the footshock. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

3.3. Changes in BLA activity induced by emotional arousal

Given the effects of corticosterone in vitro, one would expect emotionally arousing or stressful experiences to cause long-lasting changes in the firing rate and ensemble activity of BLA neurons. To test this prediction, one study used extracellular recordings of BLA neurons in a feline analog of the inhibitory avoidance task (Pelletier et al., 2005). After a single inescapable footshock, the firing rate of ∼50% of BLA neurons increased for ∼2 h and became more synchronized (Fig. 1D). Separately comparing pre-vs. post-shock firing rates during wakefulness (W) and slow-wave sleep (SWS) revealed that firing rates increased in both states, albeit much more in W (∼450%) than SWS (∼70%; Fig. 1D, insets). While most of these cells reached their peak firing rate after a long delay (30–50 min; Fig. 1D1), in ∼25% of them they peaked within seconds of the footshock (Fig. 1D2; Pelletier et al., 2005).

4. How amygdala activity affects memory consolidation?

4.1. The BLA can facilitate memory indirectly, by recruiting neuromodulatory systems

There is evidence that the BLA affects memory consolidation through multiple mechanisms. We consider them in turn below. First, the BLA can recruit neuromodulatory systems, allowing it to exert a widespread indirect influence over structures which, in rodents, are devoid of direct BLA afferents like the visual cortex or dentate gyrus. In support of this idea, the BLA projects to the CeA, which in turn targets most neuromodulatory systems of the brainstem (Hopkins and Holstege, 1978).

Consistent with this, many neuromodulators are released during emotional arousal, including dopamine (Hori et al., 1993), noradrenaline (NA; Tanaka et al., 1991), serotonin (Kawahara et al., 1993), and acetylcholine (ACh; McIntyre et al., 2003). In keeping with these findings, β-adrenergic (Ferry and McGaugh, 2000) and muscarinic receptor (Power et al., 2000) antagonists were reported to block the potentiating effects of emotional arousal on memory.

Both, NA and ACh exert complex presynaptic and postsynaptic effects on BLA neurons. Via α2 receptors, NA presynaptically inhibits glutamate release in the BLA. By contrast, via β-adrenoreceptors, NA decreases spike frequency accommodation and presynaptically enhances excitatory synaptic transmission (Huang et al., 1996; Ferry et al., 1997).

Similarly, ACh causes multiple parallel effects in the BLA. Via muscarinic receptors, it presynaptically inhibits glutamate release (Yajeya et al., 2000), it reduces several K⁺ conductances (Washburn and Moises, 1992), it activates a Ca²⁺-independent mixed cationic conductance, and it enhances a hyperpolarization-activated inward rectifier K⁺ current (Yajeya et al., 1999). Via nicotinic receptors, ACh also elicits a depolarization of GABAergic BLA interneurons (Washburn and Moises, 1992) and a presynaptic enhancement of glutamate and GABA release (Girod et al., 2000; Barazangi and Role, 2001). Further complicating this picture, an optogenetic study reported that the effects of ACh on principal BLA neurons depend on how active they are (Unal et al., 2015). When they fire at low rates, light-induced activation of basal forebrain cholinergic axons elicits muscarinic IPSPs. By contrast, when they fire tonically at ≥6 Hz, the same stimuli elicit prolonged current-evoked afterdepolarizations that overcome the muscarinic IPSPs.

In fact, much evidence suggests that the amygdala can facilitate synaptic plasticity in cortical networks via the recruitment of basal forebrain cholinergic neurons (reviewed in Weinberger, 2004). For instance, blockade of muscarinic receptors interferes with the facilitating effects of BLA stimulation on long-term potentiation (LTP) of thalamocortical synapses (Dringenberg et al., 2004) and perforant path inputs to the dentate gyrus (Akirav and Richter-Levin, 1999; Frey et al., 2001; Ikegaya et al., 1995). Similarly, selective (immunotoxic) lesions of basal forebrain cholinergic neurons prevent the memory facilitation produced by NA infusions in the BLA (Power et al., 2002).

While the above studies dealt with the immediate impact of NA and ACh on BLA neurons, there is also evidence that the memory enhancing effects of these modulators depend on mechanisms that persist long after their concentration has returned to control levels. Indeed, many neuromodulators activate intracellular signaling cascades and related kinases that recruit transcription factors, like the cAMP response element binding protein. In turn, these transcription factors stimulate or suppress genes coding for various proteins involved in the regulation of neuronal excitability and synaptic plasticity, like the activity-regulated cytoskeleton-associated protein (Benito and Barco, 2010; Ehrlich and Josselyn, 2016). However, discussing this extensive literature is beyond the scope of the present review.

4.2. BLA inputs promote heterosynaptic plasticity

There is also evidence that the synapses formed by BLA axons with memory supporting neurons are endowed with unusual properties that promote heterosynaptic plasticity (Fig. 2). Popescu et al. (2007) reported that the NMDA-to-AMPA ratio is much higher at BLA than cortical synapses onto principal striatal neurons (Fig. 2D). As a result, activation of BLA synapses greatly facilitated LTP induction at corticostriatal synapses (Fig. 2B,C,E). Remarkably, this facilitation developed even when BLA and cortical inputs were separated by 0.5 s during LTP induction. In a follow up study, Popescu et al. (2010) found that the signal bridging the 0.5 s gap between BLA and cortical inputs was Ca²⁺-induced Ca²⁺ release (CICR). Thus, it appears that due to the high density of NMDA receptors postsynaptic to BLA axon terminals and associated waves of CICR, BLA inputs can facilitate the induction of corticostriatal plasticity without requiring precise coincidence of the two inputs. Whether these properties characterize other BLA projections remains to be tested.

Fig. 2.

NMDA-dependent facilitation of corticostriatal plasticity by BLA inputs. (A1) Experimental set-up on a coronal section of the guinea pig brain: stimulating electrodes (dots) in cortex and BLA elicited responses recorded in medium spiny striatal neurons (pipette). (A2) Occlusion test: cortical stimuli 1 and 2 were delivered independently (left) or at the same time (right). EPSCs elicited by the paired stimuli (right) had the same amplitude as the sum of the independently evoked EPSCs (left, dashed lines). (B, top) Pairing BLA and cortical stimuli produces a marked potentiation of responses evoked by the paired cortical stimulus (black lines) but not by an unpaired cortical stimulation site (red lines). (B, bottom) Slope of cortically-evoked EPSPs (y axis) over time (x axis; average of 11 cells). (C) Pairing of two cortical stimulation sites produces LTP but of markedly lower magnitude than with BLA stimuli (average of 16 cells). (D) NMDA-to-AMPA ratio is higher at BLA than cortical inputs to medium spiny striatal neurons. (D1) To estimate the NMDA-to-AMPA ratio, in a Mg²⁺-free aCSF at −90 mV, picrotoxin, CNQX, and AP5 were added sequentially to the perfusate while stimulating cortical (left) or BLA (right) inputs converging on the same medium spiny neuron. The ratio of the pharmacologically isolated components was then computed. Red lines in D2 and red bars in D3 show the NMDA-to-AMPA ratio in experiments where MK-801 (5 μM) was used to selectively block NMDA receptors at BLA synapses. (E) From left to right, corticostriatal LTP with paired BLA–cortex stimuli (n = 11), at an unpaired cortical site (n = 5), with pairing of two cortical sites (n = 16), with paired BLA–cortex stimuli in the presence of AP5 (n = 5), and with paired BLA–cortex stimuli but with MK-801 in the pipette solution (n = 5). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

In the above in vitro studies, BLA and cortical inputs were activated by electrical stimulation and paired with post-synaptic depolarization. To determine how the BLA and striatum interact in vivo, another study analyzed local field potential (LFP) and unit activity in the BLA and striatum (Popescu et al., 2009). Coherence of BLA and striatal LFPs was highest in the gamma range and coupling of unit activity in the BLA and striatum rose when gamma power increased. Intra-BLA infusions of muscimol caused a drop in striatal gamma power. Moreover, as learning progressed on a stimulus-reward task, BLA-striatal gamma coupling increased selectively in relation to the rewarded stimulus (Popescu et al., 2009).

4.3. BLA activity promotes interactions between the neocortex and hippocampus

The perirhinal cortex receives sensory inputs from a variety of associative neocortical areas (Witter and Groenewegen, 1986) and, together with the entorhinal cortex, serves as the main route for impulse traffic into and out of the hippocampus (Fig. 3A; Witter et al., 2000), a mediator of memory consolidation (Nadel and Hardt, 2011). Importantly, perirhinal transmission of neocortical and entorhinal inputs normally occurs with a low probability (de Curtis and Paré, 2004) because of major feed-forward inhibitory pressures in the connections between the perirhinal and entorhinal cortices (Pinto et al., 2006). However, the BLA contributes dense glutamatergic projections to the rhinal cortices (Krettek and Price, 1977; Pitkänen et al., 2000), raising the possibility that it can regulate the formation of declarative memories by modifying impulse traffic into and out of the hippocampus at the level of the rhinal cortices.

Fig. 3.

BLA activity facilitates perirhinal-entorhinal interactions. (A) Schematic of the connections between the neocortex and hippocampus. Note stepwise progression from the neocortex to the hippocampus through perirhinal areas 35 and 36 and the entorhinal cortex. (B-D) Activity of an entorhinal neuron around perirhinal (B), BLA (C) or perirhinal and BLA (D) firing as shown in cross-correlograms (top) and rasters (bottom). Hip, hippocampal formation.

Paz et al. (2006) tested this possibility by performing simultaneous unit recordings in the BLA, perirhinal, and entorhinal cortices. Cross-correlating perirhinal and entorhinal spiking generally yielded little evidence of correlated activity (Fig. 3B and C). However, restricting the analyses to rhinal spikes that occurred in close temporal proximity to BLA firing revealed that hidden in the cross-correlations were periods of enhanced rhinal interactions that prevalently occurred when BLA cells were active (Fig. 3D). These results indicated that the strong inhibition normally limiting perirhinal-entorhinal interactions can be counteracted by BLA inputs. Importantly, Paz et al. (2006) observed that this effect was strongest when unexpected rewards were presented, when the activity of BLA neurons was most synchronized.

Together, the findings of Paz et al. (2006) imply that during emotionally salient events, inputs to the medial temporal lobe cause a synchronization of BLA activity, which in turn produces a depolarization of rhinal neurons. As a result, impulse transmission from the perirhinal to the entorhinal cortex is facilitated, thereby enhancing the processing of sensory cues and supporting the formation of declarative memories.

4.4. Gamma oscillations facilitate memory by promoting firing synchrony among BLA neurons

A recurring observation in the studies reviewed above is that emotionally arousing learning episodes not only increase the discharge rates of principal BLA neurons but also their firing synchrony. In this context, it should be noted that BLA cells have unusually low spontaneous firing rates in most conditions (on average ∼0.2 Hz; Amir et al., 2018). Thus, an increase in firing synchrony may be required to compensate for their low levels of activity. In fact, as reviewed below, there is strong evidence that gamma oscillations enhance the firing synchrony of BLA neurons and that this effect is key to the facilitating influence of the BLA on memory.

Gamma oscillations in the BLA are not continuous but appear in LFP recordings as short spindle-shaped bursts of 4–12 cycles (Fig. 4A). Critically, the entrainment of BLA neurons by LFP oscillations of various frequencies is maximal in the gamma range (Fig. 4B and C; Amir et al., 2018). As in the cerebral cortex (Whittington et al., 2000), gamma in the BLA arises from reciprocal interactions between principal cells and fast-spiking interneurons (FSIs; Amir et al., 2018; Headley et al., 2021). A hallmark of this mechanism is that principal cells preferentially fire at the negative peak of gamma, when depolarization is maximal, followed shortly by FSIs (Fig. 4D). As the inhibition generated by FSIs wanes, principal cells regain the ability to fire, and start another gamma cycle. These interactions could be reproduced by a biophysically and connectionally accurate BLA model where disrupting the synapses between principal cells and FSIs eliminated gamma (Feng et al., 2019). Moreover, current source density analyses of in vivo recordings revealed that local current sources underlie BLA gamma (Headley et al., 2021), arguing against volume conduction from adjacent regions, as is the case for the nearby nucleus accumbens (Carmichael et al., 2017).

Fig. 4.

Gamma oscillations synchronize the activity of BLA neurons. (A) Example of spontaneously occurring gamma bursts in the BLA. (B) Peri-event histograms of neuronal discharges for a representative principal cell (B1) and fast-spiking interneuron (B2) around large amplitude (≥2 SD) high gamma oscillatory peaks. (C) Entrainment of unit activity (y-axis) as a function of frequency (x-axis) for principal cells (median of entire sample). (D) Distribution of preferred high-gamma firing phase for principal cells (D1) and FSIs (D2). (D3) Average firing phase in relation to high-gamma for principal cells (blue) and interneurons (red). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Across a range of behaviors and states, BLA gamma is enhanced during periods of vigilance and ruminative processing (Amir et al., 2018; Headley et al., 2021). Both processes enhance memory, suggesting a potential link with memory consolidation. Indeed, inhibitory avoidance training (Fig. 5A–C), a type of contextual fear conditioning, enhanced BLA gamma for ∼30 min post-training (Fig. 5D; Kanta et al., 2019). Moreover, the strength of gamma predicted subsequent memory (Fig. 5D–F), with stronger gamma correlating with greater avoidance.1

Fig. 5.

Elevated BLA gamma power during consolidation of emotional memories. (A) Inhibitory avoidance training stages. (B) Latency to enter dark compartment during training (empty bars) and retention (solid bars) for good (orange, n = 7 rats) and poor learners (purple; n = 6 rats). Gray lines: individual rats. (C) Distribution of IA performance. (D) BLA spectrograms before and after training in good (top) and poor (bottom) learners. (E) Average power change (30 min post-minus pre-training) for good (orange) and poor learners (purple). (F) Correlation between change in mid-gamma power and performance (n = 13 rats). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Huff et al. (2013) were the first to test the role of gamma frequency activity in the BLA on memory consolidation. Immediately following training on an inhibitory avoidance task, excitatory optogenetic stimulation of principal BLA cells at gamma (40 Hz), but not beta (20 Hz) frequencies, significantly enhanced avoidance of the shock sector during the retention test. Conversely, optogenetic inhibition of the BLA right after training impaired subsequent avoidance.

Unfortunately, these manipulations also altered BLA firing rates, so they did not specifically test the role of gamma. To address this, Kanta et al. (2019) used a closed-loop optogenetic approach that allowed manipulations of gamma amplitude with no off-target effects (Fig. 6). An excitatory opsin was expressed preferentially in FSIs of the BLA, which were then stimulated either in- or out-of-phase with the endogenous gamma rhythm. In-phase stimulation enhanced gamma, while out-of-phase stimulation suppressed gamma (Fig. 6F1,2). Critically, artificially boosting BLA gamma following training on a spatial memory task enhanced memory consolidation, while depressing gamma impaired it (Fig. 6G; Kanta et al., 2019).

Fig. 6.

Closed-loop control of BLA gamma oscillations demonstrates their role in the consolidation of spatial memories. (A) Behavioral apparatus for the hole-board task, a dry version of the water maze task. Each day and in a single trial, rats must remember the location of a food pellet buried in one of 36 holes using only distal cues. On retention trials performed 60 min post-training, the lower the latency to reach the correct well, the better the memory. (B) Training stages. (C) Normalized latency to food reward across trials (n = 13 rats). Shaded orange region reflects probe trial latency. (D) Normalized 2nd trial latency for different Training-Retention intervals. (E) BLA spectrogram showing increased gamma power right after training on the hole-board task. (F) Spontaneous gamma bursts were either suppressed (F1) or amplified (F2) depending on the timing of blue light stimuli (peak vs. trough). (G) Normalized latency in the second trial depending on timing of the light stimuli (x-axis). Boxes show lower quartile, median and upper quartile; whiskers show the lowest and highest non-outlier observations. Gray lines show individual subject performance across treatment types. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

BLA is thought to modulate memory through its downstream targets. For instance, the intermediate/ventral hippocampus receives a prominent projection from the BLA (Pikkarainen et al., 1999), and this pathway has been implicated in memory consolidation (Wang et al., 2017). When rats learn that a specific path on a linear track is associated with an aversive stimulus, coordinated ensembles of BLA and dorsal hippocampal neurons representing that path are re-engaged during post-training SWS (Girardeau et al., 2017). This reactivation occurred during sharp waves ripples, a high-frequency oscillatory population burst in CA1 that has been causally linked to memory consolidation (Girardeau et al., 2009).

While the link between BLA gamma and hippocampal sharp wave ripples has not been explored, indirect evidence supports their co-occurrence. Sharp wave ripples are associated with the ‘replay’ of previously experienced trajectories by CA1 place cells. Replay events are elicited when rats pause while approaching the shock sector in an inhibitory avoidance task (Wu et al., 2017). Compellingly, BLA gamma oscillations are at their strongest during the same behavioral state (Headley et al., 2021).

More generally, there is substantial evidence that gamma oscillations support communication between the BLA and hippocampus, along with memory consolidation. First, BLA afferents target both principal cells and FSIs in CA1 (Felix-Ortiz et al., 2013), a microcircuit motif that promotes gamma synchrony (Womelsdorf et al., 2014; Zemankovics et al., 2013). Second, the hippocampus reciprocates this connection, with similar engagement of feedforward inhibition (Hübner et al., 2014), a type of reciprocal connectivity that promotes coordinated gamma bursts (Palmigiano et al., 2017). Third, electrical and optogenetic stimulation of BLA at gamma frequencies elicits gamma rhythmic responses in CA1 in rats (Ahlgrim and Manns, 2019; Bass and Manns, 2015) and similar results have been obtained in humans with intracranial stimulation (Sendi et al., 2021). Finally, optogenetic stimulation of BLA afferents in the ventral hippocampus immediately after training on inhibitory avoidance enhanced subsequent memory (Huff et al., 2016), an effect that was specific to gamma frequency stimuli. Notably, theta frequency stimulation of the BLA does not elicit comparable responses in CA1 (Ahlgrim and Manns, 2019), nor does it promote the encoding of contextual information (Wahlstrom et al., 2018).

5. Spontaneous recollections of emotional experiences during wake and sleep

As reviewed above, emotional arousal induces persistent increases in the firing rates, synchrony, and gamma entrainment of BLA neurons (Bauer et al., 2007; Paz et al., 2006; Pelletier et al., 2005; Popescu et al., 2009). Moreover, these changes in BLA activity were shown to alter neuronal interactions in subcortical and cortical targets of the BLA (Paz et al., 2006; Popescu et al., 2009). Hence, emotional arousal should be associated with BLA driven alterations in cognitive activity, which may contribute to the facilitation of memory by emotions.

Although there is no definitive data linking BLA activity patterns to specific cognitive events (however see Kreiman et al., 2002; Mormann et al., 2011, 2019), it is clear that emotional arousal impacts cognition. For instance, a well-established consequence of emotional arousal is rumination, whereby subjects report repeatedly experiencing intrusive memories in the aftermath of traumatic or emotional experiences (Brown and Kulik, 1977; Nolen-Hoeksema and Morrow, 1991; Wilkinson, 1983). While spontaneous intrusive recollections are particularly prevalent among subjects afflicted with post-traumatic stress or major depressive disorder, “normal” subjects experience them too (Reynolds and Brewin, 1998). In fact, you will recall that Müller and Pilzecker (1900) were led to the concept of memory consolidation because some of their subjects reported experiencing involuntary mental rehearsal of material to be learned.

Since they constitute additional opportunities for rehearsal, could these intrusive recollections facilitate memory consolidation? Ferree and Cahill (2009) examined this question by showing participants emotional or neutral films, testing recall of these films days later, and relating it to the number of intrusive recollections participants had experienced for each film. They found that the frequency of intrusive recollections was higher for emotional than neutral films and that it was positively correlated with memory. Together, these results suggest that because emotional arousal raises the frequency of intrusive recollections, rumination boosts memory for emotional compared to neutral stimuli.

Consistent with the observation that emotional arousal not only alters the firing pattern of BLA neurons during wakefulness but also during sleep (Pelletier et al., 2005), there is evidence that rumination also occurs during sleep. For instance, Eichenlaub et al. (2018) analyzed the dream content of subjects awoken during rapid-eye movement (REM) sleep and found that dreams preferentially incorporate those recent waking experiences that were emotionally intense (Eichenlaub et al., 2018), consistent with many prior studies (Cartwright et al., 2006; Malinowski and Horton, 2014; Propper et al., 2007).

Overall, the above indicates that emotional experiences are particularly prone to being re-experienced during wakefulness and incorporated into dreams, promoting consolidation of the corresponding memories. However, the underlying mechanisms are unknown. Here, we propose a speculative mechanism based on the high expression of NMDA receptors at BLA synapses and their ability to facilitate plasticity at other inputs converging on the same neurons (Popescu et al., 2007). According to this model, the synchronized gamma-related firing of principal BLA cells would potentiate the synapses between cortical neurons that were recruited during an emotional experience, effectively tagging these cells for subsequent coordinated reactivation during wake and sleep. By biasing the cortical network to reactivate prior activity patterns, this mechanism could drive plasticity in the hippocampus and indirectly assist hippocampal replay.

6. Role of REM sleep and theta oscillations in the consolidation of emotional memories

6.1. Influence of sleep on memory consolidation

Evidence that sleep affects memory consolidation was first obtained in early forgetting studies where it was noted that subjects remembered more items when the retention interval included a period of sleep compared to wakefulness (reviewed in van Ormer, 1932). While these findings were ascribed to reduced interference with memory traces in sleep than waking (Jenkins and Dallenbach, 1924), several observations challenged this interpretation. For instance, many studies showed that restricting early vs. late sleep stages (respectively enriched in SWS or REM sleep) resulted in different degrees of forgetting (Plihal and Born, 1997, 1999), which the interference account cannot explain unless one assumed that sleep stages differ in the amount of interference they cause. Also, it was found that depending on the learning task, learning resulted in differing proportions of time spent in SWS and REM sleep, with the magnitude of these changes often correlating with long-term recall (Hanlon et al., 2009; Hellman and Abel, 2007; Rasch and Born, 2013; Smith, 1996).

While it is now well established that sleep facilitates memory consolidation, whether different sleep stages preferentially affect long-term recall of different types of memories remains highly controversial (Genzel et al., 2015; Rasch and Born, 2013; Vertes and Eastman, 2000). This question was often studied by comparing recall between retention intervals that included the first or second half of sleep and, as a control for differences in the circadian rhythm, with waking retention intervals at corresponding times. Using this approach, REM sleep was found to be beneficial to the consolidation of procedural memories (priming, mirror tracing skill, etc.) whereas SWS appeared favorable to the consolidation of declarative memories (for instance see Plihal and Born, 1997, 1999). However, some procedural tasks, often involving the acquisition of sequences of movements, were not found to benefit from REM sleep (Genzel et al., 2015). Moreover, patients with major depressive disorder taking REM suppressing medications did not show enhanced memory suppression (Vertes and Eastman, 2000).

By contrast, the facilitating influence of REM sleep on the consolidation of emotional memories is well established (Genzel et al., 2015; Rasch and Born, 2013). Critically, this appears to be true for emotional memories that belong to the declarative (e.g. emotional stories) and procedural (e.g. classical conditioned responses) categories, whether negatively or positively valenced (Hu et al., 2006; Nishida et al., 2009; Perogamvros and Schwartz, 2012; Wagner et al., 2001, 2002). For instance, REM sleep was found to be beneficial to the recall of emotional scripts (Wagner et al., 2001) and images (Hu et al., 2006; Nishida et al., 2009; Wagner et al., 2002), and to the recall (Kumar and Jha, 2012) and extinction of conditioned fear responses (Fu et al., 2007; Pace-Schott et al., 2009).

6.2. Theta rhythms, REM, and the BLA

REM sleep features sustained theta oscillations that have been linked to memory consolidation. For instance, memory for emotional items following a nap is modulated by the prevalence of REM theta over right frontal EEG sites (Nishida et al., 2009; Sopp et al., 2017). Moreover, directly eliminating theta without disturbing REM sleep, by optogenetically silencing GABAergic neurons in the medial septum, impairs consolidation of contextual fear memory (Boyce et al., 2016).

Given their shared importance to the consolidation of emotional memories, there should be a link between REM theta and the BLA. Consistent with this, the amygdala is metabolically active during REM sleep (Maquet et al., 1996) and the firing rates of its neurons increase (Girardeau et al., 2017). In the same vein, during theta the BLA interacts with other structures implicated in memory formation. Specifically, the BLA exhibits coherent theta with the hippocampus (Hegde et al., 2008; Karashima et al., 2010; Seidenbecher et al., 2003), and its neurons phase-lock to entorhinal theta (Paré and Gaudreau 1996). Moreover, hippocampal theta preferentially entrains dendrite targeting calbindin-positive BLA cells, who regulate the responsiveness of principal BLA neurons to extrinsic inputs (Bienvenu et al., 2012).

Since theta is shared across regions during REM sleep, theta-band coordination between the BLA and its synaptic partners should influence consolidation. Supporting this prediction, theta coherence between the BLA, medial prefrontal cortex, and hippocampus is enhanced during REM sleep following cued fear conditioning (Popa et al., 2010). Indeed, the strength of theta coherence predicted subsequent fear memory. In subjects with the strongest fear memories, the direction of these interactions was from the hippocampus to the BLA, and then from the BLA to the medial prefrontal cortex, as revealed by Granger causality analysis (Popa et al., 2010). Moreover, these interactions can be experimentally induced by manipulations that mimic stressful events. For instance, infusing corticosterone, a stress hormone, into the dorsal raphe promotes the coupling of theta oscillations between the BLA and medial prefrontal cortex (Merino et al., 2021).

Unfortunately, few studies have tested whether inter-regional theta coordination affects memory consolidation. A notable exception is Wahlstrom et al. (2018), who examined how post-training optogenetic activation of BLA afferents to the medial entorhinal cortex affects consolidation of spatial and contextual memories. Only theta frequency stimulation had an effect, enhancing memories for escape location in a Barnes maze, and the details of a context. These effects were specific for spatial/contextual memories, with no enhancement for the association between a known context and a shock, nor for a cue that signaled an escape location. This specificity for memory content agrees with the theory that the BLA facilitates memory consolidation by enhancing local processing of relevant information in downstream regions, in this case the medial entorhinal cortex, which is crucial for spatial processing.

7. Recommendations for future research

Our understanding of the BLA's role in the emotional enhancement of memory is unbalanced. Numerous studies investigated how stress hormones influence the BLA to enhance consolidation via interactions with its downstream targets (McGaugh and Roozendaal 2002). In contrast, scant attention was paid to the underlying synaptic and circuit dynamics. Several lines of research would go a long way toward filling this deficit.

Does the BLA modulate memory-related plasticity in its downstream targets, and if so, how? One possibility is that persistently elevated BLA activity during sleep promotes synaptic plasticity when experience-related ensembles in its targets are spontaneously reactivated. For this to work, BLA activity does not need to be precisely coordinated with ensemble reactivation. The BLA's facilitatory effect could operate on a relatively slow time scale. This may be the case in the striatum, where amygdalostriatal synapses are enriched in NMDA receptors, lengthening the time window for heterosynaptic potentiation (Popescu et al., 2007). Whether this applies to other BLA targets, such as medial prefrontal or entorhinal cortex, remains to be determined.

An alternative possibility is that the BLA precisely coordinates ensemble activity in regions involved in memory storage. There is circumstantial evidence for this. During rest, numerous cortical areas exhibit coordinated replay with the hippocampus, which is thought to support memory consolidation (Ji and Wilson 2007, 2007lafsdóttir et al., 2016), along with the BLA (Girardeau et al., 2017). Moreover, the BLA can modulate interactions between cortical areas on a fast timescale (Paz et al., 2006), which may enrich cortical replay events. Thus, it is plausible that BLA amplifies cortico-cortical interactions during hippocampal replay of emotional events.

An outstanding issue is whether the role of the BLA in memory consolidation differs between REM and SWS. Emotional memories are multifaceted, with declarative and arousing elements benefiting from slow wave and REM phases, respectively (Rasch and Born 2013). The rhythms that define these sleep states, delta and theta, have gamma oscillations nested within them. Given that emotional memory consolidation depends upon BLA gamma (Kanta et al., 2019), it may have the same interaction with its targets in both sleep states. In that case, differences in consolidation of emotional memories between sleep phases may stem mainly from the different constellations of areas recruited downstream of BLA (Headley and Paré 2017). If this is the case, then the BLA's role in consolidation during sleep, and perhaps emotional behavior in general, would reflect the same underlying mechanism acting upon a rotating cast of regions.

Author contributions

DP and DBH contributed equally to writing this manuscript and preparing the figures.

Declaration of competing interest

The authors declare that they have no conflict of interest, financial or otherwise.

Handling Editor: R Victoria Risbrough

Data availability

This is not applicable to this manuscript since it is a review paper