Neural mechanisms of fear memory precision and generalization: from auditory cortex to amygdala

Abstract

Identifying neural networks that infer safety versus danger is of great interest across multiple disciplines. Recent research has started to elucidate the brain networks mediating fear discrimination and generalization. Here, we focus on the contribution of the auditory cortices to auditory fear memory and discrimination processes. We first review studies showing that the primary auditory cortex (Te1) is involved in fear memory for tones with complex perceptual features only. Conversely, Te1 is essential for fear discrimination across all types of acoustic stimuli. These studies support the idea that Te1 plays a key role in the precision of auditory fear memory and in discriminating between aversive sounds and other, non-aversive sounds. We also discuss potential neuronal mechanisms within Te1 that may support these functions. Next, we review studies showing that higher-order auditory cortices, particularly those located in the posterior auditory field, constitute a critical hub for the encoding of auditory fear memories, regardless of the physical characteristics of the tones. We present evidence suggesting that these cortices may be involved in encoding the emotional and motivational significance that sounds acquire through experience. We also summarize studies suggesting that these areas might contribute to fear discrimination, although more compelling evidence is needed. We then discuss how descending pathways from auditory cortices may influence neuronal processes within the basolateral amygdala that participate in fear discrimination. We conclude by proposing that a better understanding of these sensory cortical mechanisms may inform novel strategies to address pathological fear generalization in fear- and anxiety-disorders.

Introduction: fear memory precision and generalization processes

When encountering danger, animals, including humans, encode the sensory information present in the scene to avoid future exposure to similar threats. This learning capacity is conserved across all mammals. In laboratory settings, this process is typically studied using a sensory stimulus, such as a sound or light, as a conditioned stimulus (CS), which is paired with a painful unconditioned stimulus (US), commonly a footshock. Subsequent re-exposure to the CS alone is sufficient to elicit defensive behaviors, such as freezing or fight-or-flight responses [1]. This form of learning is known as fear or threat conditioning.

Emotional memories not only influence future behavior upon re-encountering the same sensory stimuli but can also affect behavior toward entirely novel stimuli whose emotional significance is unknown. Encountering a new stimulus is one of the greatest challenges for organisms: which sensory stimuli should be approached, and which should be avoided? Since it is not feasible to evaluate every possible outcome of a novel situation, the nervous system must rely on predictive mechanisms to assess potential threats in real time. To this end, organisms compare sensory information from the current experience with stored memories of past experiences. When novel stimuli resemble previously encountered threats, defensive behaviors, such as avoidance or immobility, are triggered; that is, fear is generalized to the new stimulus. In contrast, when stimuli are perceived as sufficiently dissimilar, alternative behaviors such as exploration and curiosity prevail.

Perceptual similarity is not the sole factor governing fear generalization. The intensity of the initial traumatic experience also modulates generalization, with more intense trauma leading to broader generalization [2–4]. Moreover, generalization is influenced by the environmental context in which novel stimuli are encountered, as well as the time elapsed since the traumatic event [5–9]. Thus, fear generalization is not merely a passive function of stimulus properties, but rather an active process integrating multiple dimensions to generate flexible and adaptive defensive responses [7, 10, 11].

Fear generalization can be studied experimentally by first conditioning subjects to associate a CS with a US, and then presenting a series of stimuli that vary in perceptual similarity to the original CS. The strongest fear response typically occurs in reaction to the CS itself and gradually diminishes as stimuli become less similar. The slope of this decline, often referred to as the generalization gradient, serves as an index of generalization, with flatter gradients indicating greater generalization [10–13]. This paradigm, sometimes termed non-discriminative conditioning, models real-life situations in which threats are encountered under conditions lacking fine sensory discrimination [14–19]. An alternative approach is discriminative fear learning, wherein one stimulus (CS + ) is paired with the US, while another stimulus (CS − ) is not associated with any aversive outcome. Upon subsequent exposure, the CS+ elicits defensive responses, whereas the CS− does not; that is, animals will distinguish between threatening stimuli and non-threatening stimuli that have been encoded as safe. Compared to non-discriminative conditioning, the discriminative paradigm requires more training trials and results in higher learning specificity and lower generalization to other frequencies [10, 15].

These paradigms are currently employed in both animal models (like rodents) and humans and have led to similar observations. However, in humans, implicit and explicit fear memories are stored and mediated by distinct neural circuits [20, 21], which can be functionally dissociated [22]. Hence, fear memory and generalization can be investigated within a multi-level perspective, including the implicit level, i.e., by measuring autonomic reactions such as the electrodermal activity [23], and the explicit level, i.e., by assessing episodic memory [24–26]. Interestingly, when exposed to a new stimulus after threat learning, implicit and explicit recognition patterns may diverge; that is, one individual may show generalization of cognitive responses despite expressing specificity of electrodermal reactions. This evidence suggests that, in humans, fear generalization does not stem from a unified process generating an integrated implicit–explicit outcome, but originates from the parallel engagement of multiple systems, each eliciting distinct and potentially dissociable responses to incoming stimuli, ultimately giving rise to divergent predictions regarding potential threats [27].

The neural mechanisms underlying fear generalization are highly complex and not yet fully elucidated. They involve a distributed network of cortical and subcortical structures, including the prefrontal cortex and the amygdala. Sensory cortices also play a key role in this process. This review aims to advance the understanding of how sensory cortices contribute to fear discrimination and interact functionally with the amygdala. In particular, we will focus on the auditory cortices and their role in generalization processes following auditory fear conditioning.

The Te1 cortex and fear learning specificity and discrimination processes

In mammals, the auditory cortex is subdivided into a central core and a surrounding belt region. The central core, corresponding to area Te1 of Zilles [28], is widely recognized as the primary auditory cortex, while the surrounding regions are classified as non-primary or higher-order auditory cortices [28–30] (Fig. 1A and Table 1). Te1 exhibits a clear tonotopic organization and primarily receives inputs from the tonotopically organized division of the medial geniculate nucleus (MGN), namely the MGv [31–33]. A convergent series of studies, initially using irreversible lesion techniques [31, 34–37] and later employing reversible inactivation methods [15, 38–41] has demonstrated that Te1 is not required for the formation of auditory fear memories when simple pure tones are used as conditioned stimuli (CS), as animals show freezing responses comparable to controls. In contrast, Te1 is necessary for the retention of CS–US associations when the CS consists of complex auditory stimuli, such as frequency-modulated sweeps or tone pips [41–45]. These findings are consistent with the established role of Te1 in analyzing the physical properties of sounds [29, 30] and support the view that Te1 is critical for fear learning when complex acoustic features must be processed, thereby contributing to the precision of auditory fear memory.

Fig. 1. Organization of primary and higher-order auditory cortices in rodents.

A Lateral view of the rodent cortex illustrating primary and higher-order auditory areas based on the Zilles atlas [28, 29]. The primary auditory cortex (Te1, pink) is surrounded by the secondary (Te2, light blue) and tertiary (Te3, green) auditory cortices. Approximate rostrocaudal extent of the auditory cortical fields is shown relative to bregma coordinates. Orientation: anterior, posterior, dorsal, and ventral. B In the Zilles atlas, Te2 is located posterior to Te1 [28, 29]. C Subsequent nomenclature defined higher-order auditory regions as the dorsal secondary auditory cortex (AuD), the ventral secondary auditory cortex (AuV, blue), and the temporal association cortex (TeA. cyan) [30]. Orientation in coronal sections: medial, lateral, dorsal, and ventral. Oc2L secondary lateral occipital cortex: V2L: secondary lateral visual cortex; PRh: perirhinal cortex; Ect: ectorhinal cortex.

Table 1.

Anatomical and functional homology and key cross-species differences in auditory fear memory precision, discrimination, and generalization.

Topic	Animal studies (rodents)	Human studies	Homology / key differences (telegraphic)
PRIMARY AUDITORY CORTEX (Te1)
Primary auditory cortex	Te1; tonotopy; primary auditory core	Heschl’s gyrus (A1/BA41, part of A2/BA42)	Conserved primary auditory core; human borders less distinct
Fear acquisition vs precision	Not required for simple-tone acquisition; required for complex CS and discrimination	A1/A2 patterns discriminate CS+ vs CS−	Functional convergence
Experience-dependent cortical plasticity	CS-specific tuning shifts; altered frequency representation	Pattern-based CS + /CS− discrimination (fMRI)	Shared representational plasticity
Inhibitory control (E/I balance)	PV/SOM/VIP interneurons regulate specificity; PV loss → generalization	No cell-type resolution	Not known
HIGHER-ORDER AUDITORY CORTICES (Te2/Te3; AuD/AuV/TeA)
Higher-order auditory cortices	Te2 and Te3; AuD/AuV/TeA; weak tonotopy; limbic/frontal connectivity	STG/STS association cortex; TVA	Functional convergence?
Role in fear memory	Required for fear memory retention (simple and complex CS), encode emotional meaning	Encode emotional auditory meaning; persistence during extinction	Convergent function
Role in fear discrimination	To be defined	To be defined	To be defined
NETWORK INTERACTIONS AND OSCILLATORY MECHANISMS
Auditory–mPFC interactions	Te1–mPFC gamma synchrony supports discrimination	PFC–auditory coupling	Shared network; directionality animal-only
Auditory cortex–BLA pathways	Ascending connections between BLA and auditory cortex, required for encoding; Descending inputs mainly from higher-order cortex; required for retrieval	Structural connections between both primary and higher-order cortex with amygdala	Functional convergence?
Oscillatory signatures	Theta (fear/generalization), gamma (discrimination)	Mostly fMRI based evidence	Oscillations established only in animals

Importantly, several of these studies have also shown that Te1 inactivation impairs the ability to discriminate between the CS and non-conditioned sounds, and this occurs when both pure and more complex tones are used. Under these conditions, animals tend to display generalized defensive responses to stimuli that differ from the CS and have not been paired with the US [35, 37, 39–41]. Only two studies reported different results, showing that lesions or reversible inactivation of the auditory cortex did not affect the generalization gradient [12, 15]. One possible explanation is that the contribution of auditory cortex to fear discrimination may depend on the specific behavioral paradigm employed and on whether perceptual acuity or conditioned fear responses are assessed. Nevertheless, the large majority of studies support a critical role of Te1 in maintaining fear memory precision and fear discrimination across a broad range of acoustic stimuli, both simple and complex.

The role of auditory cortex cellular organization in fear learning precision and discrimination

An increasing number of studies have begun to unravel the neural mechanisms by which Te1 processes fear-associated auditory stimuli and contributes to fear discrimination. Pioneering work by N.M. Weinberger and colleagues showed that fear learning induces stimulus-specific changes in cortical responsiveness to the CS frequency, which may persist for days to weeks and constitute a form of “physiological memory” [46] (Fig. 2). This plasticity “selectively facilitates responses to tonal CSs at the expense of other frequencies, producing tuning shifts toward and to the CS and expanded representation of CS frequencies in the tonotopic map of A1“ [46]. Similar results have been observed in associative incentive learning in rodents [47] and bats [48].

Fig. 2. Plasticity-related remapping of the auditory cortical map after emotional learning.

Neuronal responses in primary auditory cortex (Te1) are shown before conditioning, immediately after training, and 1 h later following differential conditioning with an aversive CS+ and a non-reinforced CS − . Conditioning selectively enhances responses to the CS+ frequency while suppressing responses to the pre-training best frequency and the CS − , with effects strengthening after consolidation, revealing stimulus-specific cortical reorganization. Adapted from ref. [46].

Subsequent studies confirmed that threat learning is associated with increased neuronal activity in response to the CS, but not to tones which have not been paired with the US [14, 49, 50]. This concept has been further expanded in other tasks where neurons in Te1 exhibit frequency selectivity in their responses to tones [51–57], and modify their tuning properties with auditory learning [58–61]. These findings further support the involvement of Te1 in tone frequency discrimination, revealing that the underlying mechanism may rely on a selective change in the representation of the frequency paired with the CS, at the expense of frequencies of sounds not paired with any emotional consequence. In addition to this mechanism, Te1 may contribute to discrimination through distinct neuronal clusters that respond separately to CS+ and CS– [62]. On the other hand, it has also been shown that fear learning induces a remapping of neuronal responses to tones not paired with the US but perceptually similar to the CS onto the same response mode as the CS. This co-mapping can predict the degree of behavioral generalization to a non-conditioned stimulus similar to the CS [49].

These activity-dependent changes in Te1 can occur at the cellular level through numerous mechanisms, such as nucleus-wide chromatin reorganization and specific changes in gene expression [63, 64], the expression of specific membrane receptors such as IGF2 [65], and synaptic remodeling [65–67]. However, the present review focuses specifically on circuit- and systems-level mechanisms; detailed descriptions of the underlying molecular and cellular processes can be found in other reviews [68].

All the studies mentioned above have focused on excitatory neurons. However, recent evidence indicates that inhibitory interneurons are also crucial for both CS encoding and fear discrimination. As in other cortical regions, Te1 contains several types of inhibitory interneurons, including parvalbumin-positive (PV), somatostatin-positive (SOM), and vasoactive intestinal polypeptide-expressing (VIP) cells (Fig. 3A). Letzkus and colleagues [43] showed that acetylcholine activates layer 1 interneurons, which inhibit layer 2/3 PV cells, resulting in disinhibition of excitatory neurons and enabling learning of complex tones. Similarly, in mice engaged in a sound-guided behavioral task, pyramidal cell excitatory responses are enhanced, while the activity of inhibitory SOM-cells is diminished [69] (Fig. 3B).

Fig. 3. Schematic representation of the principal inhibitory microcircuits in the primary auditory cortex (Te1) and their role in fear memory and fear discrimination.

A Schematic illustration of the main classes of GABAergic interneurons in Te1 and their characteristic connectivity. Parvalbumin-positive (PV) interneurons provide fast perisomatic inhibition to pyramidal neurons. Somatostatin-positive (SOM) interneurons primarily target distal dendrites of pyramidal neurons. Vasoactive intestinal polypeptide–expressing (VIP) interneurons preferentially inhibit other interneurons, resulting in disinhibition of pyramidal neurons. B Schematic representation of the main effects induced by increased or decreased activity of SOM and PV cells on auditory fear memory and discrimination.

Geffen and colleagues further demonstrated that an increase in the activity of PV-interneurons shapes frequency discrimination acuity and the frequency specificity of auditory fear conditioning [14, 15, 50]. Suppressing PV activity during conditioning increased fear generalization, whereas activating PVs preserved learning specificity [14]. Furthermore, the activity of VIP-positive interneurons can differentiate between tones associated with aversive versus rewarding outcomes [70]. Thus, GABAergic interneurons within the Te1 participate in both the encoding of CS properties and fear discrimination process (Fig. 3B). Interestingly, a recent study identified within the auditory cortex, encompassing both primary and higher-order areas, a previously unknown direct inhibitory circuit originating from SOM-positive neurons and projecting directly onto BLA neurons [71]. These neurons can control spike timing in BLA pyramidal neurons, although their precise functional role remains to be elucidated.

It is well established that acetylcholine (ACh) released from the basal forebrain, particularly from neurons of the nucleus basalis of Meynert (NBM) [72], shapes the activity of both excitatory and inhibitory neurons across primary and higher-order auditory cortices [46, 72]. A seminal study by Froemke et al. [73] demonstrated that pairing sensory stimulation with NBM activation rapidly reduces synaptic inhibition within seconds, followed by enhanced excitation and long-lasting reorganization of cortical tuning curves, consistent with the formation of a memory trace. Accordingly, regulation of inhibition emerges as a core mechanism through which ACh remodels cortical receptive fields [74]. This cholinergic control enables the auditory cortex to associate tones with aversive outcomes [43], even when the stimuli are separated by several seconds [75]. Moreover, Ach is also crucial for fine auditory discrimination [76]. Consistently, blockade of NBM cholinergic projections to the auditory cortex, or of nicotinic receptors within the cortex, impairs fear discrimination between aversive and neutral tones [62].

The role of oscillatory communication in fear learning precision and discrimination

At the population level, the coordinated activity of large groups of excitatory and inhibitory neurons gives rise to brain rhythms [77]. Specific oscillatory patterns have been linked to fear learning and discrimination across brain areas. Low-theta rhythms emerge in the amygdala, hippocampus, and prefrontal cortex during CS-evoked freezing [78–81], whereas transitions between fear and safety states correspond to shifts in higher theta band frequencies (4–12 Hz) [81–85]. In Te1, fear discrimination is associated with a local increase of neuronal activity in the slow gamma range (40–70 Hz), while this enhanced gamma activity is absent when animals generalize fear to novel tones [16] (Fig. 4). Interestingly, gamma oscillations are known to emerge from interactions between excitatory neurons and inhibitory interneurons, particularly PV-positive cells [86–90], which, as previously highlighted, play a critical role in fear discrimination within Te1.

Fig. 4. Proposed functional interactions underlying fear memory retention and discrimination across the auditory cortices, basolateral amygdala (BLA), and medial prefrontal cortex (mPFC).

During fear memory retention to a conditioned auditory stimulus (CS + ) (red lines), higher-order auditory cortices, but not primary auditory cortex (Te1), preferentially convey threat-related information to the BLA, exhibiting predominant cortical-to-BLA theta coherence. During fear discrimination (blue lines), direct Te1–mPFC coupling emerges with slow-gamma (40–70 Hz) coherence, while mPFC–BLA interactions are characterized by theta synchrony. Contributions of higher-order auditory cortices during discrimination cannot be excluded. Known neurotransmitters are indicated. The dashed boundary between Te1 and higher-order auditory cortices is schematic.

Beyond local processing, oscillations facilitate communication between distant brain regions [77, 81]. Gamma synchrony supports cortico-cortical interactions, which are essential for attention and cognition [86–92]. Consistently, animals discriminating threatening from novel tones exhibit gamma-band synchronization between Te1 and the prelimbic medial prefrontal cortex (mPFC), with the strength of gamma synchrony predicting behavioral outcomes (freezing versus no freezing) [16]. Silencing Te1 inputs to the mPFC shifts behavioral responses toward fear generalization [16]. Supporting the broader view of prefrontal and auditory cortex cooperation to solve behaviorally relevant tasks, training ferrets to respond to specific tones leads to dynamic coupling between prefrontal regions and Te1, presumably to enhance the detection and behavioral relevance of task-associated sounds [93] (Fig. 4). Similarly, neurons in both the prefrontal cortex and Te1 contribute to selecting a target sound among competing auditory stimuli [94]. Most recently, it has been found that inputs from the mPFC are necessary within Te1 for generating predictive coding and prediction error signals regarding incoming tones [95]. Together, these findings support the idea that neurons in Te1 participate in recognizing and discriminating tones that have acquired behavioral significance. Moreover, to perform this function, a functional and reciprocal interplay between the Te1 and mPFC is essential.

All the aforementioned studies were performed in experimental animal models. However, despite being few, human studies seem to confirm these findings. Using functional MRI, Staib and Bach [96] found that Heschl’s gyrus, which encompasses the primary auditory cortex (A1, Brodmann Area 41) and parts of the secondary auditory cortex (A2, Brodmann Area 42), supports the encoding of threat predictions evoked by both simple (pure tones) and complex (frequency sweep) auditory CSs. In a subsequent study [97], the authors also employed multiple frequency sounds (musical chords composed of triads of pure tones) and precisely mapped anatomical A1 locations and boundaries for each participant. They found that BOLD patterns localized in A1 discriminated between CS+ and CS-, both when they consisted of simple sounds and musical chords, relative to similar neutral cues which were not paired with the US. Hence, these findings demonstrated that human A1 encodes and actively forms CS-US predictive associations.

Recent data also suggest that human A1 is primarily engaged in sensory-driven encoding of affective features of stimuli. Intracranial EEG recordings [98] have shown that Heschl’s gyrus neural responses are predominantly activated by the emotion (among them, fear) expressed through the speaker’s voice, reflecting the acoustic properties of the cues rather than the emotion subjectively perceived by the listener.

In summary, converging evidence indicates that Te1 is not essential for the acquisition of fear memories when simple auditory stimuli are used, but plays a critical role in maintaining the precision of auditory fear memories and enabling discrimination between threatening and non-threatening sounds. These functions rely on experience-dependent cortical plasticity, tightly regulated inhibitory circuits, and neuromodulatory control. Together, these findings position Te1 as a key cortical substrate for auditory fear memory specificity rather than for fear learning per se.

Involvement of higher-order auditory cortices in auditory fear memory and discrimination processes

The cortical regions surrounding Te1 are generally considered non-primary or higher-order auditory cortices. The term “higher-order” serves as both an anatomical and functional designation, reflecting a predominance of higher-order thalamic inputs, enhanced afferents from multisensory-processing brain areas, and robust connectivity with non-auditory structures, including frontal cortices, subcortical nuclei such as the amygdala, and neuromodulatory centers. In accordance with this characterization, higher-order auditory cortices display a less sharply defined tonotopic organization, receive projections from the non-tonotopically organized divisions of the medial geniculate nucleus (MGN) [33, 99, 100], and establish extensive connections with both other neocortical areas and subcortical nuclei [29–31].

Several higher-order auditory areas have been delineated in the mammalian brain (Fig. 1A). In rats, Zilles identified the primary auditory cortex (Te1), a tertiary auditory cortex (Te3) located ventrally to Te1, and a secondary auditory cortex (Te2) positioned more posteriorly [28, 29] (Fig. 1A and B and Table 1). Subsequent anatomical nomenclature designated the higher-order auditory cortices as the dorsal secondary auditory cortex (AuD), situated dorsally to Te1, the ventral secondary auditory cortex (AuV), located ventrally to Te1, and the temporal association cortex (TeA), which occupies the most posterior and ventral aspects of the auditory cortex adjacent to AuV [30] (Fig. 1C). A comparable topographic organization has been described in mice [101]. Furthermore, some investigations propose an even more detailed functional parcellation of the auditory cortex into five subfields [61, 102, 103].

As outlined in the previous section, Te1 appears dispensable for auditory fear learning when pure tones are paired with an unconditioned stimulus (US). In contrast, converging lines of evidence demonstrate that higher-order auditory cortices are necessary for the formation of auditory fear memories, independent of the acoustic complexity of the stimulus. Early lesion studies revealed that permanent ablation of the secondary auditory cortex (Te2), encompassing the posterior AuV and TeA, impairs retention of fear memories elicited by pure tones in rats [36, 104]. These results have been corroborated by studies employing reversible inactivation [38, 84, 105] optogenetic [84, 106], chemogenetic [93, 107], and disconnection techniques [108].

Further research in mice focusing on the TeA confirmed its involvement in the formation of auditory fear memories [42, 102, 109, 110]. It is important to note, however, that the precise anatomical and physiological boundaries separating TeA from the adjacent AuV remain difficult to delineate. Moreover, a recent study demonstrated considerable variability in the size and shape of auditory cortices across individual mice [111]. Thus, although these findings predominantly implicate TeA, potential contributions from adjacent AuV cannot be excluded [109]. Indeed, when the roles of TeA and AuV were directly compared, both structures were shown to be necessary for fear memory formation [42].

These cortices are required 24 h after training during non-differential conditioning with complex tones and differential conditioning involving both CS+ and CS − [42, 102, 109]. In contrast, in rats subjected to non-differential conditioning with a pure tone as the CS, the involvement of these cortices emerges only after several days [36, 84, 105], suggesting that simpler forms of auditory learning may be initially supported by alternative neural sites.

Compelling evidence indicates that higher-order auditory cortices contribute to the assignment of affective valence to auditory stimuli. Compared to Te1, these regions display enhanced responsiveness to sounds paired with emotionally salient events [47, 83, 112–114]. Notably, in Te2, distinct neuronal populations selectively respond to tones associated with either appetitive or aversive stimuli. Pharmacogenetic blockade of neurons activated by aversive stimuli impaired subsequent recognition of aversive tones while sparing the recognition of pleasant tones, and vice versa [115]. Moreover, local infusion of cholinergic agonists such as nicotine into Te2 differentially modulated aversive versus appetitive learning [116, 117]. Emotional processing in Te2 also appears to be mediated by specialized neural circuits, which vary depending on the valence of auditory stimuli [83, 118, 119].

The ability to distinguish between emotionally distinct auditory cues suggests a potential role for higher-order auditory cortices in fear discrimination. In line with this hypothesis, neuronal population dynamics in these cortices during differential fear conditioning show distinct trajectories for CS+ and CS − [109]. Moreover, auditory discriminative learning induces robust activity-dependent plasticity within these regions [61]. This role is not limited to fear learning, as recent studies have highlighted the involvement of these areas in socially and emotionally salient auditory processing. For example, Tasaka and colleagues [120] demonstrated that TeA is essential for the encoding and recognition of pup ultrasonic vocalizations (USVs) in mother mice. USVs convey distress signals and prompt maternal retrieval behaviors. Compared to other cortical regions (including Te1), TeA exhibited a higher tagging rate by USVs. Chemogenetic silencing of USV-activated neurons in TeA reduced maternal responsiveness, indicating a causal role in maternal behavior [120]. Similarly, in zebra finches, the secondary auditory cortex is required for the formation of sound preferences. Dopaminergic modulation during song playback in this area significantly altered auditory preferences [121, 122]. These findings underscore a conserved role for higher-order auditory cortices in processing emotionally relevant auditory information across vertebrates. Furthermore, the acquisition of emotional meaning by sounds may enable the secondary auditory cortex to shape context-appropriate behavioral responses [123]. Despite this evidence, however, the precise neuronal mechanisms through which higher-order auditory cortices contribute to fear discrimination remain poorly understood, warranting further investigation.

Similar results have been reported in humans, where vocal expressions of emotions consistently activate several subregions in the higher-order auditory cortices [124–128]. The activity of some of these subregions is driven specifically by the emotional valence expressed in speech [128]. For example, a study [129] probed the role of the amygdala and the temporal voice area, a region located along the superior temporal sulcus and gyrus within the secondary auditory cortex, in detecting and responding to the acoustic characteristics of emotional vocalizations. The authors found that the bilateral temporal voice area discriminated anger from fear through the decoding of emotional and physical features conveyed by human voice. Multivariate decoding approaches [130] have also shown that distributed activity patterns within bilateral auditory cortices reliably discriminate fear-related from other music-evoked emotional states. However, the analysis did not explicitly differentiate between primary and higher-order auditory fields, as fear-related patterns extended across Heschl’s gyrus, planum temporale, planum polare, and superior temporal gyrus, thereby limiting anatomical specificity.

An fMRI study using fear-related stimuli and combining eigenvector centrality mapping and psychophysiological interaction analyses [131] revealed emotion-characteristic functional connectivity of anterior and posterior auditory association cortex with the insula, cingulate cortex, and striatum, but also with somatosensory, visual, motor, and attentional structures. To the contrary, primary auditory fields display stronger intra-auditory coupling. This pattern supports a preferential role of high-order auditory regions (belt, parabelt and superior temporal association areas) as influential nodes within affective processing networks. In parallel, a recent structural and diffusion tensor tractography (DTI) investigation [132] further demonstrates that the nonprimary auditory cortex maintains extensive cortico-cortical connections with the ipsilateral orbitofrontal, entorhinal, perirhinal, parietal, insular, primary motor, primary somatosensory, primary visual, and posterior cingulate cortices, as well as with the bilateral anterior cingulate cortex. Moreover, subcortical connections were observed with the ipsilateral amygdala and claustrum, the contralateral septal nucleus, and the pedunculopontine nucleus. Although these connecting pathways may reflect both afferent and efferent fibers from and to nonprimary auditory cortex, these data position this region within complex cortico-cortical and cortico-subcortical networks underpinning multisensory integration, cognitive and emotional processing, memory, and attentional control.

Complementary evidence from Apergis-Schoute and colleagues [133] demonstrated that, during fear extinction, the BOLD signal progressively decreased in the amygdala along with the autonomic conditioned reactions. In contrast, the representation of fear memory persisted in higher-order auditory cortices during extinction learning and extinction recall, potentially maintaining the original emotional content of the threat and discriminating it from the encoded safe stimulus. Moreover, the observation of a patient showing a focal lesion of the left auditory association cortex (Brodmann Area 22) demonstrated this region’s causal role in the conceptual decoding of everyday items with highly-relevant acoustic features, but not in processing the auditory knowledge of musical instruments [134]. This evidence further supports the idea that, in humans, this region is essential for representing the conceptual meaning associated with auditory stimuli.

In summary, higher-order auditory cortices surrounding Te1 are required for the formation and retention of auditory fear memories, regardless of the acoustic complexity of the CS. Overall, these cortices emerge as critical hubs for assigning emotional meaning to sounds and shaping context-appropriate behavioral responses. Their precise contribution to fear discrimination, however, remains to be fully elucidated.

A schematic comparison of the functional roles of Te1 and higher-order auditory cortices in fear memory precision, discrimination, and generalization is provided in Tables 1 and 2.

Table 2.

Functional comparison between primary (Te1) and higher-order (HO) auditory cortices in fear memory precision, discrimination, and generalization.

Domain	Te1 (Primary auditory cortex)	Higher-order auditory cortices (HO)
Cortical level	Primary sensory cortex	Secondary / associative auditory cortex
Anatomical position	Core auditory field	Belt and parabelt regions
Thalamic inputs	Tonotopic MGv	Non-tonotopic MGN divisions
Tonotopic organization	Well defined	Weak or absent
Acoustic processing	Precise encoding of physical sound features	Integration of complex auditory information beyond basic acoustic features
Fear learning (simple tones)	Not required	Required for long-term retention
Fear learning (complex sounds)	Required	Required
Fear memory precision	Maintains stimulus-specific representations	Supports fear memory through associative and emotional value encoding
Fear discrimination	Required across stimulus complexity	Not fully established
Fear generalization	Limits overgeneralization	May influence generalization via affective coding
Neuronal plasticity	Frequency-specific tuning shifts; map reorganization	Experience-dependent plasticity linked to valence
Interneuron involvement	PV + , SOM + , VIP interneurons regulate specificity	Poorly characterized
Oscillatory signatures	Gamma activity associated with discrimination	Theta synchronization with amygdala
mPFC interaction	Strong functional coupling	Less characterized
Projections to BLA	Sparse; not essential	Major cortical input to BLA
Emotional valence encoding	Limited	Prominent
Evidence in humans	A1 discriminates CS+ vs CS−	Secondary auditory cortex encodes emotional meaning

Basolateral amygdala–auditory cortex interactions in fear memory precision and discrimination processes

The amygdala, and in particular the BLA nuclei, is a key hub in regulating responses to aversive stimuli across all mammals. On the one hand, it regulates the expression of behavioral defensive responses toward dangerous stimuli. On the other hand, it is also necessary to form memories of traumatic events.

Numerous studies have revealed that the BLA is also involved in regulating fear discrimination processes and encoding safety signals [135–139]. When animals are trained to discriminate between an aversive CS+ and a CS–, some neurons in the BLA begin to respond to the CS+ but not to the CS– [138, 140–144]. The synapses of these neurons show long-term potentiation (LTP) associated with the formation of aversive memories to the CS + [142], and blocking this pathway causes amnesia for the CS + [142].

In addition to neurons responding to the CS + , however, neurons that increase their activity in response to the CS- [138, 142, 144], or to novel sounds different from the CS + [93] have also been identified in BLA. Critically, inhibition of these cells leads subjects to generalize fear to safety signals [93, 144].

Importantly, non-threatening and safety-predicting cues engage neuromodulatory systems classically associated with reward, novelty, and positive valence, including dopaminergic signaling within the amygdala. Beyond fear learning, dopaminergic signaling in the amygdala is broadly involved in processing motivational salience, including responses to rewarding and novel stimuli, which often share functional properties with non-threatening or safety-predicting cues. In these cells, learning safety signals induces the expression of the early gene Neuronal PAS Domain Protein 4 (NPAS4), which promotes dopamine D4 receptor expression, thereby engaging dopaminergic signaling mechanisms that support the consolidation of non-threatening memories [144]. Other studies have provided further support for the role of dopamine in fear discrimination within the BLA [141, 145] and other regions of the amygdala [146]. Consistent with this view, a large proportion of BLA neurons that respond to the CS– also fire in response to reward-predicting cues [138], suggesting that safety signals, novelty, and reward partially rely on overlapping neural substrates within the BLA. This overlap provides a mechanistic framework for the involvement of dopaminergic signaling in fear discrimination, as dopamine is critically implicated in both reward processing and the attribution of motivational salience.

The aforementioned neurons are in part excitatory cells [93, 144]. Additional mechanisms involved in fear discrimination within the BLA rely on inhibitory GABAergic cells. Increased fear generalization has been observed in mice deficient in glutamic acid decarboxylase 65 (Gad65 − /−) [147–149], a condition associated with persistent high theta synchronization between the amygdala and hippocampus [149] and an altered excitation/inhibition (E/I) balance [148].

These neuronal mechanisms undergo profound changes when subjects shift from discrimination to fear generalization. Neurons that previously encoded safety become silent [93, 144]. Neurons that initially responded to the CS+ but not to the CS– start responding to both stimuli [140], losing their specificity and increasing the overall number of neurons responding to danger [140, 143, 150]. Finally, during discrimination processes, BLA activity is characterized by an increase in theta frequency during CS+ presentations but not during CS– presentations, whereas the shift to generalization is associated with the emergence of theta rhythms during both CS+ and CS– presentations [81, 136] (Fig. 4).

Human studies partially support these findings. For example, Pannese and colleagues [129] argued that the amygdala also discriminates emotional vocally-expressed information, both through the detection of the affective valence in signals pre-processed by the auditory cortex and through the direct processing of acoustic characteristics of the voice. Concerning generalization, in a recent study [151], the authors recorded single amygdalar neurons in patients with surgically-implanted intracranial electrodes to characterize their role in overgeneralization towards aversive stimuli. Single-neuron patterns reflected and predicted the extent of subsequent behavioral generalization as well as the broadening of generalization gradients in a recognition choice task (i.e., familiar vs new), for aversive but not appetitive auditory cues. However, previous studies observed a negative generalization profile (i.e., a decreasing gradient of activation along with an increasing perceptual similarity to the CS + ) in the amygdala [152, 153], while other studies [11] failed to track any generalization pattern in this structure. Further human studies are therefore needed to disentangle this inconsistency among findings.

Taken together, all the findings presented so far raise the question of whether and how the auditory cortices and the BLA interact during fear discrimination responses. Some studies have begun to investigate the potential interplay between Te1 and the BLA during fear discrimination. Optogenetic blockade of the auditory cortex centered on Te1 prevents the activation of BLA neurons responding to the CS+ during memory retrieval [62]. Conversely, another study reported that direct excitatory projections from Te1 to the BLA were not necessary for successful fear discrimination [154]. Thus, whether direct projections from Te1 to the BLA are necessary for fear discrimination is still a matter of debate.

On the other hand, as previously mentioned, the functional interaction between the mPFC and Te1 is essential for fear discrimination [154], as are the functional interactions between the mPFC and BLA [81, 136]. Moreover, optogenetic blockade of prelimbic terminals projecting to the BLA impairs fear discrimination [155]. Altogether, these findings suggest that Te1 and mPFC may directly interact during fear discrimination, with direct projections from the mPFC, but not from Te1, driving BLA activity to support discrimination processes. Future studies are needed to further investigate this possibility.

An alternative, although not mutually exclusive, hypothesis stems from anatomical and functional evidence showing that most auditory cortical inputs to the BLA originate from higher-order auditory cortices rather than from Te [42, 106, 109]. During the recall of auditory threat memories, the posterior regions of the auditory cortex and the BLA exhibit high levels of synchronization in the theta frequency range (4–12 Hz) [38, 105]. This communication is characterized by a predominant cortical-to-BLA directionality [105, 109], and the proportion of time that higher-order auditory cortex theta activity leads BLA activity correlates with faster freezing responses to the CS [38, 105]. Critically, optogenetic or chemogenetic disruption of these descending cortical projections to the BLA abolishes retrieval-evoked BLA neuronal activity and prevents fear memory retrieval [38, 42, 84, 105, 106, 109, 156] (Fig. 4).

Thus, these descending pathways may serve as a crucial route for transmitting information about auditory threat stimuli to the BLA. Given their involvement in processing aversive stimuli, it is conceivable that this connectivity also supports the discrimination between threatening and safe stimuli. However, to the best of our knowledge, no study has specifically addressed whether and how descending connections between higher-order auditory cortices and the BLA are involved in fear discrimination processes. Clarifying this point represents an important avenue for future research.

In summary, BLA plays a central role in encoding both threat and safety signals and dynamically shifts its neuronal activity patterns during transitions between fear discrimination and fear generalization. While direct projections from Te1 to the BLA do not appear to be strictly necessary for fear discrimination, functional interactions between Te1, the medial prefrontal cortex, and the BLA are essential for maintaining fear discrimination. In contrast, descending projections from higher-order auditory cortices to the BLA are critical for auditory fear memory retrieval. Whether these descending pathways also convey safety-related information or actively support fear discrimination remains an open question for future research.

The major auditory cortical and amygdalar circuits implicated in fear memory precision, discrimination, and generalization are summarized in Fig. 4 and Table 3.

Table 3.

Summary of major auditory cortical and cortical–amygdala circuits involved in fear memory precision, discrimination, and generalization.

Circuit / pathway	Main structures and cells	Key mechanisms	Primary function
Local Te1 microcircuits	Excitatory neurons; PV + , SOM + , VIP interneurons	E/I balance; interneuron-mediated inhibition; activity-dependent plasticity; Acetylcholine; Dopamine	Fear discrimination and memory specificity
Local Te2 (HO auditory cortex) microcircuits	Te2 / posterior AuV–TeA; Excitatory neurons; PV + , SOM + , VIP interneurons	Valence-specific ensemble recruitment; activity-dependent plasticity; interneuron-mediated gating of ensemble selectivity; Acetylcholine; Dopamine	Encoding of affective/motivational value of sounds; support of auditory fear memory storage
Corticocortical top-down pathway	Te1 ↔ prelimbic mPFC	Gamma-band synchrony; predictive coding	Modulation of sensory processing during learning and retrieval; fear discrimination
Descending auditory–amygdala pathway	HO auditory cortices → BLA	Theta synchronization; cortical-to-BLA drive	Transmission of auditory fear memory signals
Prefrontal–amygdala pathway	mPFC → BLA	Control of BLA neuronal ensembles	Regulation of fear discrimination and safety signaling
Intra-BLA fear memory circuits	BLA excitatory and inhibitory neurons	Hebbian synaptic plasticity (LTP); recruitment of CS + -responsive BLA ensembles; Noradrenaline	Fear memory formation and expression
Intra-BLA safety circuits	BLA excitatory and inhibitory neurons	NPAS4–D4R signaling; E/I balance; Dopamine	Encoding of safety signals

Conclusions and future directions

In this review, we summarized past and recent evidence supporting the hypothesis that auditory cortices play a critical role in maintaining the precision of auditory fear memories and enabling auditory fear discrimination. We discussed studies suggesting that Te1 is not required for simpler forms of auditory fear learning but is essential for preserving the accuracy of fear memories and distinguishing threatening tones from neutral or safe auditory stimuli. The underlying mechanisms may involve activity-dependent plasticity of excitatory neurons, alterations in the E/I balance, and changes in rhythmic oscillatory activity.

However, several critical questions remain unresolved. It is still unclear whether, and through which mechanisms, Te1 actively participates in the encoding of safety signals, rather than merely sharpening representations of threat-predicting stimuli. Classical studies by Weinberger and colleagues demonstrated that discriminative fear learning induces a reorganization of Te1 receptive fields, characterized by enhanced responses to the CS+ frequency and a concomitant suppression of responses to the CS − . More recent work by Geffen and others has shown that parvalbumin-positive interneurons are essential for maintaining stimulus-specific representations and preventing fear generalization. Despite these advances, it remains unknown whether plasticity within Te1 directly supports safety learning per se, or whether reduced responses to the CS− simply emerge as a by-product of threat-focused cortical tuning. A key open question is therefore whether distinct neuronal populations or circuit motifs within Te1 selectively encode safety signals, and how such representations contribute causally to fear discrimination. Moreover, it is crucial to determine whether stress and traumatic experiences may disrupt these mechanisms, thereby contributing to the emergence of psychopathologies. Future studies will be key to addressing these critical issues.

Beyond Te1, several higher-order auditory cortices have also been implicated in fear learning. Based on recent findings, we propose that these regions may encode the affective-motivational value that tones acquire through experience. However, the degree to which higher-order auditory cortices contribute specifically to fear discrimination remains unclear, and further investigation is required to elucidate this point.

Finally, we examined whether and how Te1 and higher-order auditory cortices interact with the BLA during fear memory and discrimination processes. Current evidence suggests that descending projections from higher-order auditory cortices, but not from Te1, are necessary for the retention of fear memories and possibly for fear discrimination. Yet, several questions remain open. In particular: i) what functional circuits, if any, connect Te1 to the BLA during fear discrimination, given that direct anatomical projections do not appear essential for this function? ii) Are descending pathways from higher-order auditory cortices to the BLA specifically involved in discrimination, and if so, through what mechanisms? For instance, do distinct descending pathways mediate threat versus safety signals, or are these signals conveyed through different activity patterns within the same pathways?

Understanding the mechanisms by which auditory cortices contribute to fear discrimination has significant translational implications. Deficits in fear discrimination, leading to excessive and inappropriate generalization of fear to neutral or safe stimuli, may underlie the development of PTSD and other anxiety-related conditions [7, 10, 13, 18, 81, 157–160].

The role of generalization in pathological anxiety has been recognized since the early 20^th century, notably through the seminal “Little Albert” experiment by Watson and Rayner, which demonstrated the generalization of conditioned fear responses from a white rat to a broad range of furry objects. Individuals suffering from PTSD often have a history of traumatic events that disrupt normal sensory filtering and emotional regulation, leading to exaggerated fear responses even to neutral or safety-associated cues [7, 10, 11, 13, 158, 159, 161, 162].

For example, an auditory event-related potential (ERP) study [163] with PTSD patients and trauma-exposed control participants found a potentiated mismatch negativity in the former group, potentially highlighting a more pronounced tendency to redirect automatic attention processes towards unattended incoming auditory cues in PTSD. Moreover, a study that recruited youth with partial PTSD observed that a functional network centered on the superior temporal gyrus, which contains high-order auditory areas, predicted individual outcomes following trauma-focused therapy [164].

Thus, identifying the neuronal mechanisms within sensory cortices that are altered by trauma, and that may contribute to aberrant sensory processing and excessive fear generalization, holds a substantial promise for developing novel and more effective therapeutic interventions for these debilitating disorders.

Author contributions

E.M., F.C., and B.S. wrote the manuscript. L.M., A.R., and C.G. contributed to manuscript revision and final editing. All authors discussed the manuscript and provided critical feedback.

Funding

B.S. discloses support for this work from the Fondazione CariVerona (grant number 2018.0780), Banca d’Italia (grant number D63C23000630005), Italian Ministry of University and Research (MIUR) (grant “Progetti di ricerca di Rilevante Interesse Nazionale (PRIN)” 2022), and “Piano Nazionale di Ripresa e Resilienza” (PNRR) “Mnesys” Spoke 4. This study is part of the MIUR Project “Dipartimenti di Eccellenza” of the Rita Levi-Montalcini Department of Neuroscience (University of Turin). All the other authors declare no relevant funding.

Competing interests

The authors declare no competing interests.