RESEARCH DOSSIER
Neural Circuits of Threat, Fear, and Defensive Behavior
Abstract. Threat-related neural circuits connect detection and learned prediction of danger to context-sensitive selection of freezing, flight, avoidance, and physiological mobilization. This fundamental research monograph critically synthesizes conditioning experiments, amygdala microcircuit manipulations, periaqueductal-gray pathways, extended-amygdala dynamics, hippocampal context processing, prefrontal extinction regulation, human lesion evidence, and neuroimaging meta-analyses. It distinguishes threat probability from temporal certainty, defensive motor expression from subjective fear, and neural correlates from causal mechanisms. Particular attention is paid to threat imminence, safety discrimination, response competition, species translation, clinical interpretation, computational models, and replication. The method is a source-attributed critical narrative synthesis of original experiments and published reviews; it introduces no newly validated circuit theory and reports no original experiments.
Keywords: threat processing; fear conditioning; defensive behavior; amygdala; central amygdala; basolateral amygdala; bed nucleus of the stria terminalis; periaqueductal gray; freezing; flight; defensive action selection; threat imminence; fear extinction; safety learning; fear generalization; contextual fear memory; human neuroimaging; optogenetics; neural circuits; fear and anxiety; Artificial Era
Introduction: Defensive Behavior as a Circuit-Level Problem
Threat processing is a family of operations through which an organism detects possible harm, estimates its proximity and probability, and organizes actions that may reduce exposure. The relevant neural circuitry coordinates sensory discrimination, associative memory, autonomic change, attention, motor preparation, and behavioral selection. A looming object, a learned shock-predictive tone, and an uncertain future danger recruit overlapping operations under different informational conditions. Consequently, the explanatory object of this monograph is the causal organization of defensive responses across connected brain systems rather than the assignment of a single emotional label to a single anatomical structure. The foundational circuit literature maps amygdalar, cortical, hypothalamic, and brainstem contributions while emphasizing their reciprocal interactions Tovote et al. (2015) Tseng et al. (2023).
A successful neural explanation must connect four levels. At the environmental level an event creates an exposure or contingency; at the computational level a system estimates threat, uncertainty, controllability, and action value; at the circuit level identified populations change firing, synaptic efficacy, and connectivity; at the behavioral level freezing, avoidance, escape, monitoring, or physiological mobilization can be observed. These levels are linked experimentally through manipulations that distinguish detecting an aversive cue from recalling it, expressing immobility from being unable to move, and choosing escape from merely increasing arousal. The evidence is strongest when behavior, neural recording, and causal intervention converge on the same proposed mechanism Fanselow (1994) Tovote et al. (2016) Fadok et al. (2017).
This monograph is a critical narrative synthesis of original experiments, mechanistic reviews, human imaging studies, and quantitative evidence syntheses. It follows the development of threat learning from contingency studies to cell-type-specific interventions, then examines response-selection pathways and the translation of animal mechanisms to human experience. Its central questions concern which neural computations are shared across defensive states, which mechanisms depend on imminence and context, how safety is learned, and what current methods can establish causally. Scientific interpretation remains proportional to the design of each study: evoked signals, successful decoders, circuit perturbations, and subjective reports offer different evidence about defensive organization LeDoux & Pine (2016) Fullana et al. (2016) Cornwell et al. (2025).
Defining Threat, Fear, Anxiety, and Defensive Response
Threat is an organism-relative condition in which environmental information indicates a possibility of harm, loss, or biological disturbance. In experiments its operationalization depends on a defined aversive outcome, an explicit threat probability, a temporal schedule, and an opportunity to act. Fear commonly denotes a relatively acute response to a detected or anticipated threat; anxiety often denotes apprehension under more sustained or uncertain conditions. Defensive behavior refers to measurable activities that manage exposure, including orienting, freezing, flight, fight, risk assessment, concealment, and active avoidance. These terms are useful only when the investigator reports how each was measured, since equivalent freezing durations can arise from different computations Davis et al. (2010) Shackman & Fox (2016) Tseng et al. (2023).
Conditioned defensive responding results from experience with a predictive relationship, whereas innate responses may be elicited by biologically significant sensory features without the same experimentally demonstrated association. Both depend on past evolutionary and individual history, internal state, species-specific affordances, and task constraints. A stimulus does not possess an invariant neural category called threat: the same auditory sound can signify safety in one context and danger in another. Defensive intensity is therefore jointly conditioned on the organism's representation of contingency, contextual memory, available routes of escape, and recent stress. Claims about an innate detector should specify the modality, developmental history, and population investigated Rescorla (1968) Maren et al. (2013) Gu et al. (2025).
Subjective fear has an additional measurement problem. A participant may report fear without freezing, and an animal may freeze in a protocol whose subjective qualities cannot be directly assessed. LeDoux and Pine proposed distinguishing the neural systems coordinating defensive survival responses from the systems supporting conscious reports of fear, a distinction that organizes experiments without settling every theory of feeling. Human amygdala lesion findings further show that threat recognition, fear induction, and experiential reporting can dissociate according to task and exposure. Accordingly, the present study uses 'defensive circuit' for operational mechanisms and retains 'fear' for contexts in which affective experience or established disciplinary terminology is itself at issue LeDoux & Pine (2016) Feinstein et al. (2011) Adolphs et al. (2005).
Historical Development: Conditioning, Emotion, and Brain Mechanisms
Experimental analysis of learned threat gained a decisive methodological advance with the distinction between pairing and contingency. Rescorla's 1968 analysis of the probability of shock during and outside a conditioned stimulus showed why counting paired events alone cannot define the information a cue conveys. This principle made threat learning testable through controlled changes in event probability and background rate. Subsequent neurophysiological work asked how such predictive relationships alter activity in defined pathways, especially auditory inputs reaching the lateral amygdala. The history therefore joins an experimentally specified learning relation to recordings of neural change under that relation Rescorla (1968) Quirk et al. (1995).
Research on defense also developed through ethological observation of whole behavioral sequences. Rather than reducing every response to a conditioned reflex, investigators considered the transitions from cautious monitoring to freezing, escape, attack, or post-threat recovery. Fanselow organized these responses as a defensive behavior system whose output changes with ecological circumstances and the imminence of danger. This perspective brought behavioral topography, movement opportunities, and the spatial relationship to a threat into mechanistic inquiry. A rat freezing in a chamber, for example, is displaying one available action within a constrained environment, not the entire repertoire of natural defense Fanselow (1994) Mobbs et al. (2007).
By the turn of the twenty-first century, converging lesion, recording, and functional imaging studies supported a multilevel account of emotional learning involving the amygdala, prefrontal cortex, and connected autonomic systems. Cell-type targeting and optogenetics subsequently allowed experimental control of defined pathways, refining older regional accounts of fear expression. The field now studies how inhibitory microcircuits switch between responses, how projection targets organize specific output programs, and how neural activity changes over the seconds preceding different kinds of threat. Historical models remain valuable for the questions they generated, while newer evidence supplies finer anatomical and computational distinctions LeDoux (2000) Haubensak et al. (2010) Tovote et al. (2016) Cornwell et al. (2025).
From Threat Probability to Action: A Multilevel Research Framework
Neural threat studies need experimental designs that separate the presence of an aversive event from its predictability. A conditioned stimulus paired with shock is informative only relative to a comparison distribution of shocks; a safety cue is meaningful when it predicts reduced danger under specified circumstances. Temporal certainty adds another dimension: knowing exactly when an aversive event will arrive differs from knowing only that it may arrive. Threat imminence, spatial accessibility, intensity, and controllability change the possible action policies. These variables should be crossed when feasible instead of treated as interchangeable indicators of fear Rescorla (1968) Davis et al. (2010) Cornwell et al. (2025).
The neural system can be described as a sequence of constrained transformations: sensory signals are filtered by attention and state, associative circuits retrieve or update predictions, contextual networks restrict the circumstances in which a prediction applies, and premotor and autonomic pathways express a selected defense. The sequence is analytically useful but real circuits are recurrent. Amygdalar inputs can influence cortical monitoring, prefrontal projections can alter amygdalar expression, and bodily changes can return information that affects subsequent perception. Explanations therefore require timing as well as connection diagrams; simultaneous activation alone establishes neither signal direction nor functional necessity Tovote et al. (2015) Maren et al. (2013) Tseng et al. (2023).
Threat-related outcomes are also multidimensional. Freezing time indexes a defensive posture; skin conductance indexes a sympathetic response; startle potentiation indexes a reflex modulation; avoidance choices index action selection; reported unpleasantness indexes an experiential judgment. Each measure may vary independently with instructions, contextual memories, motor demands, or learning stage. A circuit claim gains specificity when a manipulation selectively changes a prespecified outcome and leaves appropriate control behaviors intact. The strongest designs incorporate baseline mobility, sensory acuity, reward motivation, and shock sensitivity controls, helping distinguish mechanisms of threat computation from general impairments in performance Davis et al. (2010) Fullana et al. (2016) Tovote et al. (2016).
The Distributed Anatomy of Defensive Control
The amygdala is a group of anatomically and functionally distinct nuclei, including lateral and basal subdivisions implicated in associative representations and central subdivisions with major downstream defensive influence. Its pathways connect with hippocampal context systems, medial prefrontal cortex, the bed nucleus of the stria terminalis, hypothalamic structures, and periaqueductal gray. The periaqueductal gray itself contains functionally differentiated zones with connections to medullary and spinal control pathways. This organization offers multiple intervention points between recognizing a cue and executing an action, so a manipulation in one node can alter an entire behavioral chain LeDoux (2000) Tovote et al. (2015) Tovote et al. (2016).
The central extended amygdala includes the central amygdala and neighboring bed nucleus of the stria terminalis, which integrate information about anticipated danger and shape autonomic and behavioral output. Ventromedial and dorsal medial frontal networks participate in appraisal, learning, and regulation under task-dependent conditions, while insular and midcingulate cortices are frequently engaged during anticipation and interoceptive monitoring. Hippocampal systems contribute representations of place, time, and context that determine which threat or safety association is relevant. These regions participate in overlapping networks rather than fixed one-to-one assignments of emotions to anatomical locations Shackman & Fox (2016) Maren et al. (2013) Cornwell et al. (2025).
An anatomical label is not yet a mechanism. Projection-specific evidence identifies the origin and target of a pathway; recording indicates the signals its cells express; perturbation tests whether activity changes behavior under defined conditions. Strong models integrate these methods with verified cellular identity, time-locked events, and controls for activity outside the intended circuit. Cross-species comparisons must account for differences in cortical elaboration and available defensive behaviors. A distributed anatomical map provides a hypothesis space for causal testing, which becomes more informative as investigators specify what information enters each circuit, what transformations occur, and which actions its outputs can recruit Haubensak et al. (2010) Fadok et al. (2017) Tseng et al. (2023).
Basolateral Amygdala Plasticity and the Encoding of Aversive Associations
Basolateral amygdala populations integrate sensory inputs with information about the occurrence and significance of aversive outcomes. Their experimentally observed changes in firing and synaptic efficacy during conditioning support a role in developing cue-outcome relations. These populations are heterogeneous: neurons with distinct projection targets and response profiles can encode different learned valences and behavioral possibilities. The key empirical proposition is that a change within a neural ensemble influences how a predictive cue later controls defensive behavior, which is a stronger statement than a regional activation contrast Namburi et al. (2015) Quirk et al. (1995) Tovote et al. (2015).
Namburi and colleagues reported that circuit-defined amygdala projections participate differentially in positive and negative associations. Their experiments illustrate why a manipulation labeled simply 'amygdala activation' can combine distinct mechanisms with contrasting behavioral effects. Neural ensembles are embedded in recurrent networks containing inhibitory cells, afferents that carry contextual and motivational information, and projections capable of altering multiple downstream effectors. Thus a conditioning-related change in basolateral activity should be assessed with projection targeting, temporally specific intervention, and careful controls for the animal's capacity to express the learned response Namburi et al. (2015) Burgos-Robles et al. (2017).
An essential limitation concerns the inferred content of a representation. A neuron whose response increases to a shock-predicting cue may encode estimated threat probability, salience, uncertainty, arousal, action preparation, or combinations of these variables. Separating them requires a task in which probability changes without changing stimulus intensity, or action demands change while predicted outcome remains constant. Cross-validated decoding can show that information is recoverable, but cannot alone establish that downstream systems read it in the decoded manner. Mechanistic inference improves when an explicitly specified computational variable predicts learning and targeted perturbations alter that variable's behavioral consequences Rescorla (1968) Tovote et al. (2015) Tseng et al. (2023).
Central Amygdala Inhibitory Microcircuits and Defensive Gating
The central amygdala links learned and unlearned danger signals to networks controlling defensive expression, but its internal organization cannot be represented as a uniform output switch. Work in mice by Haubensak and colleagues identified genetically accessible inhibitory microcircuit elements that gate conditioned fear. Ciocchi and colleagues likewise showed functionally differentiated populations whose activity relates to conditioned freezing and its control. These causal and observational results changed the unit of explanation from the entire central nucleus to interacting cell types that regulate one another and project to distinct effectors Haubensak et al. (2010) Ciocchi et al. (2010).
Disinhibition is particularly important. Because many central amygdala neurons are inhibitory, activation of one population can suppress another inhibitory population and thereby increase the activity of a downstream output. This arrangement creates switch-like changes in behavioral expression without requiring a unitary fear center. The function of a connection depends on the identity of its presynaptic and postsynaptic cells, the state of competing pathways, and the timing of cue information. Cell counts, molecular markers, and broad anatomical location alone are insufficient substitutes for functional intervention during an appropriately controlled behavioral task Haubensak et al. (2010) Ciocchi et al. (2010) Tovote et al. (2016).
Microcircuit findings also impose interpretive constraints. Artificial stimulation may recruit synchronous firing patterns unlike the naturally evoked signals that occur during conditioning, while inhibition may suppress both information processing and motor execution. Behavioral freezing can decrease because an animal changes its threat estimate, selects escape, or loses the motor pathway necessary for immobility. Robust causal work therefore combines targeted activation and suppression, independent recording, sensory and motor controls, and replication across task configurations. The claim of circuit necessity should specify the particular species, response, context, and manipulation rather than being generalized to the whole human experience of fear Tovote et al. (2015) Tovote et al. (2016) Fadok et al. (2017).
The Central Extended Amygdala and the Temporal Dynamics of Uncertain Threat
Fear and anxiety have often been modeled through the distinction between brief responses to predictable danger and prolonged responses to uncertain threat. The central amygdala and bed nucleus of the stria terminalis are commonly assigned different temporal roles in this account, supported by animal paradigms involving predictable versus sustained aversive cues and human psychophysiological measures. Davis and colleagues developed this phasic-versus-sustained framework as a productive explanatory hypothesis, while subsequent accounts emphasized substantial anatomical, physiological, and functional overlap between extended-amygdala subdivisions Davis et al. (2010) Shackman & Fox (2016).
Recent human data require particular care with any strict anatomical partition. Cornwell and colleagues examined 220 adults using an anticipatory paradigm that separately manipulated temporal certainty and safety. A shared network involving central extended amygdala, periaqueductal gray, anterior insula, and midcingulate cortex showed sustained activity during uncertain anticipation and greater late activity before certain aversive encounters. Within their measurements the bed nucleus and central amygdala displayed statistically similar threat dynamics rather than exclusive sustained and phasic profiles. This result revises overly simple regional mappings while retaining temporal predictability as a meaningful dimension of defensive organization Cornwell et al. (2025).
The methodological lesson is that a region's average response across an entire trial can conceal time-dependent dynamics. Functional imaging estimates depend on the hemodynamic response model and on the epochs selected for comparison; apparent sustained versus transient activation must therefore be evaluated using temporal models capable of distinguishing them. Anatomically constrained regions and independent comparisons help, but small human nuclei remain challenging to resolve and their activation does not establish specific cell-type signaling. The strongest synthesis holds together historical animal causal evidence, human anticipatory dynamics, and explicit uncertainty about whether equivalent computations are implemented by homologous cellular populations Davis et al. (2010) Shackman & Fox (2016) Cornwell et al. (2025).
Innate Threat Cues, Defensive Salience, and Evolutionary Continuity
Some threats elicit organized defensive responses without explicit laboratory conditioning. Rapidly approaching visual stimuli, predator-associated odors, and intense auditory events can engage defensive networks in species- and context-dependent ways. The scientific problem is to distinguish sensitivity to biologically significant input features from a claim that the animal possesses a predetermined conscious interpretation of that input. Sensory pathways vary, and the same defensive output can be triggered through different upstream mechanisms. Reviews of modality-triggered defense emphasize the importance of tracing feature-selective routes and confirming the actual experience available to each species Gu et al. (2025) Tseng et al. (2023).
Comparative approaches reveal recurrent organizational principles: assessment of distance and direction, transition between immobility and movement, and coordination of autonomic support with defensive action. Yet ecological affordances differ greatly among a rodent in an enclosed chamber, a freely moving primate, and a human confronting a symbolic or social danger. The absence of a particular response in one experimental arrangement may reflect inaccessible escape routes rather than absent defensive capability. Translational experiments should therefore evaluate whether the species can deploy the relevant action, whether sensory information is comparable, and whether threat intensity has been calibrated to the organism Fanselow (1994) Mobbs et al. (2007) Tseng et al. (2023).
Evolutionary continuity is best treated as a testable proposition about circuitry and function. Conserved cellular or anatomical components may support different patterns of behavior when embedded in different ecological networks, while functionally similar responses can arise from partially distinct neural routes. Comparative interpretation benefits from homologous tracing, transcriptomic cell characterization, matched perturbations, and matched behavioral variables. The argument that a human circuit serves defense gains force from convergence across these levels, but animal freezing remains a behavioral measurement with its own validity limits. Such distinctions prevent a narrative of universal fear wiring from outrunning comparative evidence Tovote et al. (2015) Gu et al. (2025) Tseng et al. (2023).
Periaqueductal Gray and Brainstem Execution of Defensive Action
The periaqueductal gray occupies an important position between forebrain threat circuits and brainstem motor and autonomic systems. Its longitudinal and dorsoventral organization supports differentiated defensive effects, and its projections reach premotor regions capable of implementing changes in posture, respiration, and coordinated movement. Research on freezing has identified a central-amygdala-to-ventrolateral-periaqueductal-gray pathway whose inhibitory organization can disinhibit excitatory descending outputs. The evidence therefore gives a cellular account of how a learned cue can recruit a motor program without requiring the entire forebrain network to specify every muscle command Tovote et al. (2016) Tovote et al. (2015).
Tovote and colleagues combined pathway tracing, recordings, and optogenetic interventions to study freezing-related midbrain circuits. Their findings linked central amygdala inhibition of particular periaqueductal populations to the recruitment of downstream excitatory outputs and demonstrated interactions with pathways relevant to flight. The result is a mechanistically specific example of causality: manipulating a defined connection altered a measurable defensive pattern, and complementary physiological data supported the proposed circuit logic. Such evidence is more informative than a regional lesion alone, although the manipulated cells may participate in additional tasks and the observed behavior is conditional on the experimental setting Tovote et al. (2016).
The brainstem is also a site of integration rather than a final passive output relay. Local inhibitory interactions, sensory feedback, and descending signals can change how defensive programs are selected or sustained. Freezing often includes coordinated physiological adjustments, and escape requires synchronized orientation and locomotion. An explanation that stops at identifying a forebrain 'fear neuron' leaves these motor and feedback mechanisms unspecified. Future studies must track activity across multiple connected populations during naturalistic transitions and distinguish effects on action initiation, maintenance, termination, and subsequent recovery Tovote et al. (2016) Tseng et al. (2023) Fadok et al. (2017).
Selection Between Freezing, Flight, and Confrontation
An animal facing danger can remain immobile, move away, seek shelter, or confront the source, depending on threat distance, available routes, learned contingencies, and current action costs. These behaviors cannot be arranged as a single measure of greater or lesser fear: an increase in flight may coincide with a decrease in freezing even when the estimated hazard remains high. Fadok and colleagues identified a competitive inhibitory circuit implicated in selecting active versus passive defensive responses. Their findings complement periaqueductal-gray studies by placing response competition within a broader network that evaluates and implements alternative motor policies Fadok et al. (2017) Tovote et al. (2016).
Behavioral selection can be studied by allowing more than one valid response in the same paradigm, measuring transition probabilities and their timing relative to sensory events. A freezing-only chamber can reveal much about immobility but cannot independently estimate the preference for escape when escape is unavailable. Richer environments allow investigators to test whether an intervention specifically biases the choice of action, shifts the threshold at which flight begins, or broadly affects locomotion. Measurements of approach trajectories, head orientation, shelter use, and response latency provide information that a single percentage of time spent freezing may conceal Fanselow (1994) Fadok et al. (2017) Mobbs et al. (2007).
Competitive mechanisms are likely to be distributed. Defensive action depends on forebrain estimates, basal-ganglia and prefrontal selection processes, hypothalamic and brainstem state regulation, and embodied opportunities. A causal account must therefore determine which node chooses among actions, which implements a selected movement, and which updates the expected outcome of that movement. Disabling a motor channel can reveal a necessary executor without identifying the locus of valuation. Conversely, changing cue discrimination may alter behavior without directly changing any motor implementation pathway. This distinction between policy selection and motor expression organizes future cross-region perturbation experiments Tovote et al. (2015) Tseng et al. (2023) Fadok et al. (2017).
Threat Imminence, Defensive Distance, and Human Action
The imminence of danger changes the relevant decision problem. At a distance an organism may gather information, monitor the environment, or prepare an avoidance route; near contact it may be more adaptive to execute rapid escape or another immediate defense. Fanselow's defensive-system framework emphasized that distinct behaviors are elicited at different stages of a threat encounter. In humans, Mobbs and colleagues investigated threat proximity with a virtual-predator task and reported a change in the balance of prefrontal and periaqueductal-gray engagement as danger approached Fanselow (1994) Mobbs et al. (2007).
The human proximity result connects a laboratory manipulation of threat distance with activity in networks expected from comparative neurobiology. However, the task's explicit rules, artificial environment, and blood-oxygen-level-dependent signals restrict inference. A change in functional imaging signal does not reveal which neurons initiated the behavior, which projection carried the signal, or whether the same transition occurs during uncontrolled real-life emergencies. Model-based analysis can ask whether proximity, expected cost, and control opportunity explain behavior better than elapsed time, visual intensity, or motor demand Mobbs et al. (2007) Fullana et al. (2016).
Threat imminence also interacts with subjective appraisal and personal experience. A participant anticipating an unavoidable aversive event may show prolonged monitoring, whereas the availability of a controllable response can change both action policy and reported distress. In this sense, distance is one component of an anticipated transition from a relatively reversible situation to an urgent encounter. A complete neural theory must specify how information about escape options reaches forebrain and midbrain control networks. Direct comparisons across different degrees of control, sensory certainty, and temporal predictability would clarify which aspects of the proximity effect generalize across tasks and populations Cornwell et al. (2025) Mobbs et al. (2007) Tseng et al. (2023).
Avoidance Learning, Escape, and the Computation of Control
Avoidance is a behavior that prevents or reduces exposure to an expected adverse event, whereas escape terminates or reduces a harmful event that has already begun. Both require learned relations among cues, actions, and outcomes. A person may successfully avoid an event and therefore receive no direct evidence about whether the event would have happened, creating a persistent inferential problem. In experimental models, response-contingent safety and changes in threat probability must be distinguished because action performance can itself influence the observations available for later learning Rescorla (1968) Maren et al. (2013) Tseng et al. (2023).
Neural investigations of avoidance require tasks that offer meaningful action alternatives and explicitly record both successful prevention and opportunities to update threat beliefs. Passive freezing is insufficient to infer how an animal values a safety-producing action. Investigators must distinguish learned knowledge of an action-outcome contingency from rapid suppression of movement, increased locomotor vigor, or an aversive motivational drive. The same circuit manipulation might affect escape latency, threat estimation, or the ability to execute a response. Causal specificity demands separate control tasks for each candidate interpretation Fanselow (1994) Fadok et al. (2017) Tovote et al. (2016).
Controllability is also dynamic. A defensive response that works in one setting can become ineffective when contingencies change, and repeated unsuccessful action may alter subsequent exploration. The nervous system must update the relation between expected harm and possible intervention while preserving enough learning to respond efficiently when circumstances recur. This requirement links threat circuitry to broader learning, executive control, and action-selection systems, while leaving their detailed computational integration an active research question. Longitudinal protocols that manipulate actual control independently of perceived control are especially informative because subjective belief and experimental contingency may diverge Maren et al. (2013) Tseng et al. (2023) LeDoux & Pine (2016).
Autonomic, Endocrine, and Interoceptive Components of Defense
Defensive responses commonly include cardiovascular, respiratory, pupillary, endocrine, and gastrointestinal changes alongside action. Amygdalar, hypothalamic, and brainstem systems influence these adjustments through autonomic and neuroendocrine pathways. Their function can be understood as coordinating bodily resources with anticipated or actual demands: vigilance may alter heart rate, escape may increase metabolic mobilization, and immobility may involve a characteristic but context-sensitive physiological profile. Measurements of physiology are therefore informative when collected with behavior and timed to distinct stages of threat, although no single bodily variable provides an exhaustive measure of fear Davis et al. (2010) Tseng et al. (2023).
Interoceptive feedback may in turn influence perception of threat and judgments of bodily state. Human imaging often identifies anterior insular and cingulate responses during anticipation, but these activations can reflect attention, salience, uncertainty, nociceptive expectancy, action preparation, and several other partially overlapping processes. The field benefits from designs that manipulate bodily signals independently, record peripheral physiology continuously, and ask whether neural changes mediate a subsequent behavioral or subjective effect. Regional blood-flow responses alone cannot determine the full causal direction between bodily change and appraisal Fullana et al. (2016) Cornwell et al. (2025) Lindquist et al. (2012).
Stress physiology also has different timescales from immediate motor defense. Rapid sympathetic changes, slower endocrine activity, and plasticity following repeated exposure can each modify later behavior. Treating them as a single 'fight-or-flight reaction' obscures the timing and regulatory specificity essential to mechanistic studies. A comprehensive experiment can combine high-temporal-resolution measures with delayed recovery sampling and examine how responses habituate or sensitize. This monograph treats these processes as inputs and outputs of defensive networks while reserving detailed hypothalamic-pituitary-adrenal biology for the distinct Research Dossier dedicated to stress neurobiology Tseng et al. (2023) Davis et al. (2010).
Hippocampal Context, Occasion Setting, and Threat Memory
A defensive association learned in one environment may be inappropriate in another. Hippocampal systems support representations of spatial, temporal, and contextual relations that help determine which learned meaning is currently relevant. Context dependence becomes especially important when an aversive cue is later presented without the expected outcome, since successful behavior requires recognizing the conditions under which safety learning applies. Maren and colleagues integrated evidence that hippocampal-prefrontal-amygdala interactions contribute to the acquisition, expression, and extinction of conditioned fear. The same cue can therefore evoke different responses without its sensory identity changing Maren et al. (2013) Bouton (2004).
Context may function as background information, a retrieval cue, or an occasion setter that selects among competing predictive relations. Investigators distinguish these possibilities by varying whether a context itself predicts shock, indicates which cue-outcome relation applies, or merely accompanies training. Return of defensive behavior after a context change is an important experimental result because it demonstrates that extinction performance depends on retrieval circumstances. However, a contextual behavioral effect does not by itself localize the memory to hippocampus; network interventions and task-specific hippocampal measurements are needed to evaluate the proposed mechanism Bouton (2004) Maren et al. (2013).
The neurobiological problem concerns how contextual information influences downstream cue interpretation over time. Hippocampal outputs interact with prefrontal and amygdalar regions, and the composition of active neuronal ensembles may change as an animal encounters familiar or novel situations. Studies of hippocampal engrams provide one approach to investigating sparse or distributed populations associated with a contextual memory. The inference that a particular ensemble contains a complete memory requires further tests because stimulation can initiate retrieval or behavior without establishing that all relevant information is exclusively stored in that ensemble Vasudevan et al. (2024) Maren et al. (2013).
Extinction Learning and the Medial Prefrontal Regulation of Defense
Extinction occurs when a previously predictive cue is repeatedly encountered without the anticipated aversive outcome and the conditioned response consequently decreases. A central finding is that extinction often creates new learning about the circumstances of nonreinforcement while leaving elements of the original association available for later retrieval. Renewal, spontaneous recovery, and reinstatement are different behavioral demonstrations of this persistence. Bouton's analysis of contextual control explains why an immediate decline in freezing is insufficient evidence that the initial learning has been erased Bouton (2004) Maren et al. (2013).
Medial prefrontal cortex participates in the retrieval and expression of extinction memories under defined conditions. Milad and Quirk reported neurons whose firing signaled memory for fear extinction, supporting a role for prefrontal contributions to the regulation of conditioned responding. Subsequent human work linked ventromedial prefrontal and amygdala activity to extinction learning. The causal interpretation depends on experimental stage and task: cortical signals during expression of safety can differ from those during acquisition of a new prediction, and prefrontal contributions may vary with context and response demands Milad & Quirk (2002) Phelps et al. (2004).
Extinction is therefore a network process involving the updating and retrieval of cue meaning, the selection of an appropriate behavioral policy, and the modulation of output circuits. A reduction in amygdala activation is neither a necessary universal marker nor a sufficient proof of successful safety learning. Effective tests separate cue discrimination, context memory, expectancy, and motor response, then examine whether circuit manipulations change later retention or only immediate expression. Human meta-analyses reveal distributed extinction-related activity and support cross-study regularities while retaining heterogeneity in timing, outcomes, and analysis choices Fullana et al. (2018) Maren et al. (2013) Bouton (2004).
Safety Signals, Inhibition, and the Problem of Learned Nonthreat
Safety has its own informational structure. A cue can signal that an aversive event is less likely than expected, that it is delayed, or that a specific action can prevent it. A stimulus that is merely absent during shock trials need not have acquired the same inhibitory meaning as a well-tested safety signal. Experimental safety learning is best established through controlled contingency tests, including whether the putative safety cue reduces responding when combined with a known threat cue. These criteria reduce the risk of equating familiarity or weak arousal with active neural inhibition of a learned threat expectation Rescorla (1968) Bouton (2004).
Prefrontal, hippocampal, and amygdalar networks can contribute to discrimination between safety and danger, but their roles depend on the procedure used to generate the contrast. Human neuroimaging work has identified distributed activation and deactivation patterns in conditioning paradigms, and some of these changes may reflect processing of cues associated with the absence of danger. Yet an imaging contrast between danger and safety conditions cannot establish whether neurons encode inhibition, relative expected value, attention, or expectancy violation. Causal inference requires separate experimental variables and precise behavioral tests of conditioned inhibition Fullana et al. (2016) Phelps et al. (2004).
Safety signals pose a particular translational challenge because individuals may report feeling safe while continuing autonomic vigilance, or show low physiological response while consciously expecting danger. Experimental measures should be triangulated rather than collapsed into one construct. A productive design evaluates whether safety generalizes to a novel context, survives distraction, and remains effective after a change in threat probability. Such tests can distinguish a durable inhibitory association from temporary response suppression. The mechanistic goal is to explain when safety information reaches defensive output circuits and why it sometimes fails to control behavior Bouton (2004) Maren et al. (2013) Cornwell et al. (2025).
Fear Generalization, Discrimination, and Boundary Conditions
Generalization occurs when defensive learning about one event influences responding to another event with related sensory, conceptual, or contextual features. It can be adaptive if the new event carries similar danger, yet excessive breadth can create costly false alarms. Discrimination is the complementary ability to reserve a stronger response for signals with greater predictive relevance. Generalization gradients, obtained by systematically varying stimulus similarity, quantify these effects. A plausible neural mechanism involves the interaction of perceptual representations, memory retrieval, attention, and defensive output rather than a separate generalized-fear organ Rescorla (1968) Maren et al. (2013) Webler et al. (2021).
Human imaging studies of fear generalization have identified distributed networks, but pooled evidence is constrained by the relatively small number of comparable experiments and differences in stimulus spaces, instructions, and statistical thresholds. Webler and colleagues synthesized a limited human neuroimaging literature and proposed a working model of the circuits involved. These results are informative for hypothesis generation; they cannot determine that a region alone causes generalized fear or that every clinical group expresses the same pattern. Studies should report whether the gradient reflects a true change in threat expectancy or merely perceptual uncertainty Webler et al. (2021).
Generalization research is strengthened by explicitly modeling the structure of the stimulus space. Physical similarity, category membership, social meaning, and context familiarity may produce different gradients despite similar response magnitudes. Longitudinal designs can examine whether generalization narrows with discrimination training and whether a change reflects relearning of outcome likelihood, increased perceptual precision, or improved safety retrieval. Human-rodent translation requires matching the relevant information rather than assuming that two visually similar cues carry equivalent significance. These constraints delimit where circuit-level analogies are justified and where broader cognitive mechanisms become decisive Webler et al. (2021) Fullana et al. (2016) Maren et al. (2013).
Threat Memory Updating, Reconsolidation, and Prediction Error
Retrieving a threat association may provide an occasion for updating it when new information violates the expected outcome. Prediction error denotes a discrepancy between what the organism expected and what occurred; its magnitude and meaning depend on the animal's current model of the situation. Some experiments suggest that specific retrieval conditions allow subsequent pharmacological or behavioral interventions to alter expression of an established learned fear response. Sevenster and colleagues examined the dependence of a pharmacologically induced fear-memory effect on prediction error, highlighting why retrieval by itself is not an adequate operational definition of destabilization Sevenster et al. (2013).
Interpretation of reconsolidation experiments must distinguish temporary performance changes from lasting alteration of the original association. A reduction in conditioned responding after intervention can also reflect extinction, contextual learning, changes in arousal, or impaired expression. Researchers therefore test retention after delays, renewal across contexts, spontaneous recovery, and the dependence of an effect on the timing and content of retrieval. Different training strengths and retrieval protocols may produce different boundary conditions, so a single positive finding cannot establish a general procedure for erasing unwanted fear memories Sevenster et al. (2013) Bouton (2004) Maren et al. (2013).
At the circuit level memory updating creates a demanding causal question: which pathways signal the violation, which synapses change, and how does the resulting representation influence later selection of defensive actions? Dopaminergic and other neuromodulatory systems can contribute to threat and safety learning, but such observations must be integrated with local amygdala plasticity, context retrieval, and prefrontal regulation. No one measure of surprise establishes a unitary teaching mechanism. Future work combining prediction-error manipulations, cell-type-resolved recordings, and longitudinal response tests could determine when retrieval modifies storage, accessibility, or downstream expression Duvarci (2024) Sevenster et al. (2013) Tovote et al. (2015).
Experimental Methods: Behavioral, Cellular, and Circuit Causal Inference
Scientific claims about defensive circuitry are grounded in a hierarchy of complementary methods. Pavlovian conditioning manipulates the predictive relations between a cue and an adverse outcome, and instrumental paradigms manipulate the relation between an available action and subsequent safety. Video tracking, electromyography, startle, autonomic recording, and subjective expectancy can then quantify distinct response components. Electrophysiology provides millisecond-scale activity, fiber photometry estimates signals from genetically targeted populations, and functional imaging resolves activity across a wider anatomical field. None of these measurements independently proves that a recovered neural pattern constitutes the mechanism producing a behavior Tovote et al. (2015) Fullana et al. (2016) Tseng et al. (2023).
Cell-type-specific perturbation provides a more direct test of causal contribution, especially when a projection is stimulated or suppressed during a defined stage of learning or response execution. Optogenetic findings on central-amygdala and periaqueductal-gray circuits show the value of joining pathway anatomy, physiological response, and behavior. But stimulation parameters, viral targeting, expression levels, and off-target effects affect interpretation. Investigators should distinguish sufficiency under artificial activation from necessity during natural behavior and should independently confirm expression specificity and baseline motor performance Haubensak et al. (2010) Tovote et al. (2016) Fadok et al. (2017).
Rigorous study design requires randomized trial order where appropriate, prespecified primary outcomes, masked scoring, sufficient animal or participant sampling, and independent replication. Inference is strengthened by testing opposite causal manipulations, checking whether an effect survives changes in context, and reporting all measured response dimensions. A successful replication need not reproduce identical behavior when a task changes important informational or ecological constraints; those differences should be incorporated into the mechanistic explanation. Statistical significance alone cannot resolve whether the proposed neural variable was selectively manipulated or whether an observed behavioral change has the intended psychological meaning Tovote et al. (2015) Fullana et al. (2018) Tseng et al. (2023).
Human Neuroimaging and the Limits of Reverse Inference
Human functional imaging enables investigation of distributed responses to instructed and learned threats while preserving access to verbal reports. Fullana and colleagues' meta-analysis pooled 677 participants from 27 independent fear-conditioning studies and found a broad, reproducible pattern of activity across regions associated with central autonomic and interoceptive processes. This synthesis is particularly important because it evaluates convergence beyond one laboratory and included original statistical maps from some contributing studies. Nonetheless, the pooled studies varied in tasks, sample composition, modeling choices, and response measures, placing limits on how narrowly one can localize a specific mechanism Fullana et al. (2016).
A common analytical error is reverse inference: observing activity in a region associated with fear and concluding that fear must therefore have occurred. The same amygdala, insula, or cingulate region may participate in attention, valuation, salience, interoception, or other operations. Meta-analytic evidence for common activation patterns supports an association with the experimental manipulation, not a uniquely diagnostic mental process. Additional uncertainty arises from spatial averaging, individual anatomy, indirect hemodynamic measurements, and dependence on analysis specifications. The solution is stronger experimental separation of competing processes and convergent measurement rather than an assumption of one region per emotion Lindquist et al. (2012) Fullana et al. (2016).
Human lesion studies provide a different kind of evidence. Adolphs and colleagues showed that deficits in facial fear recognition following amygdala damage were associated with atypical attention to diagnostic eye information. Feinstein and colleagues investigated the induction and subjective experience of fear in a patient with bilateral amygdala damage, providing evidence about the dependence of specific fear responses on amygdala function. Such rare cases are informative natural experiments, but lesion timing, plasticity, preserved pathways, task exposure, and the small number of individuals restrict generalization. Combining lesions with physiology and imaging offers a more discriminating interpretation of the circuit's role Adolphs et al. (2005) Feinstein et al. (2011) LeDoux & Pine (2016).
Computational Models of Defensive Prediction and Action
Computational accounts attempt to formalize how incoming evidence changes a threat estimate and how that estimate guides action. Associative models begin with differences between expected and actual aversive outcomes, while Bayesian and state-inference approaches represent uncertainty about current context or hidden environmental causes. Decision models evaluate prospective responses in relation to danger probability, action cost, and the anticipated consequences of avoidance or escape. These formulations are useful because they generate measurable trialwise predictions, allowing investigators to ask which variable most closely explains response timing and neural activity Rescorla (1968) Maren et al. (2013) Brochard et al. (2025).
Two models can fit the same freezing curve while implying very different neural mechanisms. One may posit stronger threat expectancy, another weakened safety retrieval, and a third lower cost for immobility. Parameter identifiability therefore matters: recovery of a fitted quantity from simulated behavior is an essential preliminary check before interpreting it as a neural computation. Out-of-sample prediction, perturbation experiments, and factorial dissociation of expectancy, uncertainty, and motor opportunity provide more informative tests than matching one aggregate response measure. The explanatory target is a mechanistically constrained account that predicts transitions among available defensive actions Brochard et al. (2025) Fadok et al. (2017) Cornwell et al. (2025).
A useful cross-level computational experiment couples neural recording to manipulated information. For example, threat probability may be altered while stimulus identity remains constant, or time-to-threat may be manipulated while actual outcome intensity is held fixed. Researchers then examine whether the candidate neural signal tracks the predicted variable, whether circuit intervention shifts its influence on choice, and whether the same relationship remains after learning has stabilized. The interpretation remains sensitive to unobserved states, nonlinear dynamics, and feedback. Artificial models of defense can clarify computational possibilities, but their algorithmic success is separate from empirical evidence that a biological circuit implements the same calculation Brochard et al. (2025) Cornwell et al. (2025) Tseng et al. (2023).
Developmental, Individual, and Ecological Variability
Defensive systems develop through the interaction of neural maturation, experience, and the ecology of the organism. Conditioning, contextual discrimination, and regulation may show different trajectories because sensory capacities, memory systems, motor opportunities, and social supports change across development. A protocol established in adult laboratory animals cannot be presumed to characterize a child or adolescent human population. Comparative research should document the developmental stage of the organism, the exact training history, and what actions the environment makes available. The neural correlates of a response must be interpreted in relation to these conditions Maren et al. (2013) Tovote et al. (2015) Tseng et al. (2023).
Individual differences also arise through variation in threat sensitivity, prior exposures, generalization gradients, and regulatory strategies. A person reporting high subjective anxiety may show physiological responses that differ from another person making a similar report, and such discrepancies can carry scientific information. Group-mean imaging maps conceal anatomical and functional variation, while small studies can generate unstable estimates of between-person associations. Larger preregistered designs, test-retest measures, and participant-level models are necessary for establishing whether a proposed neural marker predicts future behavior with clinically meaningful accuracy Fullana et al. (2016) Webler et al. (2021) Cornwell et al. (2025).
Ecological validity matters because defensive actions occur in physical and social environments. A laboratory task with unavoidable shock removes opportunities for real-world avoidance, negotiation, and safety seeking, while a free-moving task may introduce locomotor and attentional confounds. Realistic paradigms should carefully calibrate risk, remain ethically appropriate, and measure what participants can actually do. The theoretical objective is to identify mechanisms that generalize across changing opportunities without erasing informative context dependence. Evidence that a circuit is activated during one imposed laboratory choice does not establish an invariant role in all contexts of danger Fanselow (1994) Mobbs et al. (2007) Brochard et al. (2025).
Artificial Era: Functional Comparison and Computational Boundaries
The Artificial Era brings a comparative question into focus: how might a nonbiological agent detect risk, predict harmful outcomes, allocate resources, and select protective actions? In artificial systems, these functions can be implemented through hazard classifiers, uncertainty estimation, memory, planning, and intervention policies. Their computational organization can be described and tested without assuming biological homology or subjective fear. The comparative value lies in identifying which formal problems, such as uncertain prediction, risk-sensitive action, and avoidance-induced information loss, occur in both biological and artificial decision architectures Brochard et al. (2025) Rescorla (1968).
Biological defensive circuits have distinctive constraints: embodied sensory systems, evolved homeostatic needs, neuromodulatory physiology, developmental plasticity, and descending motor coordination. Artificial systems may reproduce particular information-processing functions with different substrates and different consequences. A functional resemblance between a risk-averse software policy and animal avoidance establishes an analogy at a specified computational level, while claims about artificial experience require independent operational and philosophical arguments. Research should therefore report the level of comparison—algorithm, behavior, architecture, embodiment, or phenomenology—and avoid silently transferring results between them LeDoux & Pine (2016) Tseng et al. (2023) Brochard et al. (2025).
Cross-domain experiments could be informative where they expose systematic failures. For instance, an artificial decision system might maintain a risk estimate after repeated safe experiences, overgeneralize a hazard category, or avoid gathering data needed to revise its own predictions. These are researchable patterns of model behavior, not automatically equivalents of clinical anxiety. Carefully specified computational simulations can help design biological hypotheses about prediction, exploration, and controllability. The neuroscientific standard remains independent evidence that identified biological pathways implement the proposed mechanisms; simulation is a source of testable alternatives, not proof of biological identity Brochard et al. (2025) Maren et al. (2013).
Conclusion
Neural circuits of threat, fear, and defensive behavior form a distributed, dynamic system that learns predictive relations, integrates context and uncertainty, selects among actions, and coordinates physiological implementation. The amygdala participates in cue-outcome learning and defensive gating; central-amygdala inhibitory circuitry helps control expression; periaqueductal-gray networks translate selected signals into motor programs; hippocampal and prefrontal systems shape context-dependent retrieval and safety regulation. These conclusions rest on converging but nonidentical methods, including conditioning, recording, pathway perturbation, lesions, imaging, and quantitative synthesis Quirk et al. (1995) Haubensak et al. (2010) Tovote et al. (2016) Maren et al. (2013).
The strongest contemporary evidence resists simple regional assignments. The neural distinction between certain and uncertain threat depends on temporal dynamics as well as anatomy, and cell-type-resolved experiments show that individual nuclei contain diverse and competing functional pathways. Human subjective fear, autonomic change, learned threat expectancy, freezing, and escape represent related dimensions that may dissociate. A mature explanation therefore specifies the measurable response, the stage of learning, the information available to the organism, and the causal role of an identified circuit. These requirements guide credible translation from animal models to human studies Cornwell et al. (2025) LeDoux & Pine (2016) Fullana et al. (2016) Fadok et al. (2017).
The central open problems concern the integration of distributed neural activity into adaptive action, the generalizability of causal circuit results across contexts and species, the relation between defensive regulation and conscious emotion, and the conditions under which learned danger gives way to durable safety. Computational models sharpen these questions only when tested against identifiable variables and experimentally discriminating interventions. Research progress will depend on richer behavioral options, temporally precise multiregion measurements, robust human replication, and transparent evidence appraisal. This synthesis documents the present mechanistic foundation of defensive neuroscience and the work required to make its explanations more complete Tseng et al. (2023) Brochard et al. (2025) Fullana et al. (2018).
References
- Adolphs, R., Gosselin, F., Buchanan, T. W., Tranel, D., Schyns, P., & Damasio, A. R. (2005). A mechanism for impaired fear recognition after amygdala damage. Nature, 433, 68–72. https://doi.org/10.1038/nature03086
- Bouton, M. E. (2004). Context and behavioral processes in extinction. Learning & Memory, 11(5), 485–494. https://doi.org/10.1101/lm.78804
- Brochard, J., Dayan, P., & Bach, D. R. (2025). Critical intelligence: Computing defensive behaviour. Neuroscience & Biobehavioral Reviews, 174, Article 106213. https://doi.org/10.1016/j.neubiorev.2025.106213
- Burgos-Robles, A., Kimchi, E. Y., Izadmehr, E. M., Porzenheim, M. J., Ramos-Guasp, W. A., Nieh, E. H., Felix-Ortiz, A. C., Namburi, P., Leppla, C. A., Presbrey, K. N., Anandalingam, K. K., Pagan-Rivera, P. A., Anahtar, M., Beyeler, A., & Tye, K. M. (2017). Amygdala inputs to prefrontal cortex guide behavior amid conflicting cues of reward and punishment. Nature Neuroscience, 20, 824–835. https://doi.org/10.1038/nn.4553
- Ciocchi, S., Herry, C., Grenier, F., Wolff, S. B. E., Letzkus, J. J., Vlachos, I., Ehrlich, I., Sprengel, R., Deisseroth, K., Stadler, M. B., Müller, C., & Lüthi, A. (2010). Encoding of conditioned fear in central amygdala inhibitory circuits. Nature, 468, 277–282. https://doi.org/10.1038/nature09559
- Cornwell, B. R., Didier, P. R., Grogans, S. E., Anderson, A. S., Islam, S., Kim, H. C., Kuhn, M., Tillman, R. M., Hur, J., Scott, Z. S., Fox, A. S., DeYoung, K. A., Smith, J. F., & Shackman, A. J. (2025). A shared threat-anticipation circuit is dynamically engaged at different moments by certain and uncertain threat. Journal of Neuroscience, 45(16), e2113242025. https://doi.org/10.1523/JNEUROSCI.2113-24.2025
- Davis, M., Walker, D. L., Miles, L., & Grillon, C. (2010). Phasic vs sustained fear in rats and humans: Role of the extended amygdala in fear vs anxiety. Neuropsychopharmacology, 35, 105–135. https://doi.org/10.1038/npp.2009.109
- Duvarci, S. (2024). Dopaminergic circuits controlling threat and safety learning. Trends in Neurosciences, 47(12), 1014–1027. https://doi.org/10.1016/j.tins.2024.10.001
- Fadok, J. P., Krabbe, S., Markovic, M., Courtin, J., Xu, C., Massi, L., Botta, P., Bylund, K., Müller, C., Kovacevic, A., Tovote, P., & Lüthi, A. (2017). A competitive inhibitory circuit for selection of active and passive fear responses. Nature, 542, 96–100. https://doi.org/10.1038/nature21047
- Fanselow, M. S. (1994). Neural organization of the defensive behavior system responsible for fear. Psychonomic Bulletin & Review, 1(4), 429–438. https://doi.org/10.3758/BF03210947
- Feinstein, J. S., Adolphs, R., Damasio, A., & Tranel, D. (2011). The human amygdala and the induction and experience of fear. Current Biology, 21(1), 34–38. https://doi.org/10.1016/j.cub.2010.11.042
- Fullana, M. A., Harrison, B. J., Soriano-Mas, C., Vervliet, B., Cardoner, N., Àvila-Parcet, A., & Radua, J. (2016). Neural signatures of human fear conditioning: An updated and extended meta-analysis of fMRI studies. Molecular Psychiatry, 21, 500–508. https://doi.org/10.1038/mp.2015.88
- Fullana, M. A., Albajes-Eizagirre, A., Soriano-Mas, C., Vervliet, B., Cardoner, N., Benet, O., Radua, J., & Harrison, B. J. (2018). Fear extinction in the human brain: A meta-analysis of fMRI studies in healthy participants. Neuroscience & Biobehavioral Reviews, 88, 16–25. https://doi.org/10.1016/j.neubiorev.2018.03.002
- Gu, H., Zhao, F., Liu, Z., & Cao, P. (2025). Defense or death? A review of the neural mechanisms underlying sensory modality-triggered innate defensive behaviors. Current Opinion in Neurobiology, 92, Article 102977. https://doi.org/10.1016/j.conb.2025.102977
- Haubensak, W., Kunwar, P. S., Cai, H., Ciocchi, S., Wall, N. R., Ponnusamy, R., Biag, J., Dong, H.-W., Deisseroth, K., Callaway, E. M., Fanselow, M. S., Lüthi, A., & Anderson, D. J. (2010). Genetic dissection of an amygdala microcircuit that gates conditioned fear. Nature, 468, 270–276. https://doi.org/10.1038/nature09553
- Kozlowska, K., Walker, P., McLean, L., & Carrive, P. (2015). Fear and the defense cascade: Clinical implications and management. Harvard Review of Psychiatry, 23(4), 263–287. https://doi.org/10.1097/HRP.0000000000000065
- LeDoux, J. E. (2000). Emotion circuits in the brain. Annual Review of Neuroscience, 23, 155–184. https://doi.org/10.1146/annurev.neuro.23.1.155
- LeDoux, J. E., & Pine, D. S. (2016). Using neuroscience to help understand fear and anxiety: A two-system framework. American Journal of Psychiatry, 173(11), 1083–1093. https://doi.org/10.1176/appi.ajp.2016.16030353
- Lindquist, K. A., Wager, T. D., Kober, H., Bliss-Moreau, E., & Barrett, L. F. (2012). The brain basis of emotion: A meta-analytic review. Behavioral and Brain Sciences, 35(3), 121–143. https://doi.org/10.1017/S0140525X11000446
- Maren, S., Phan, K. L., & Liberzon, I. (2013). The contextual brain: Implications for fear conditioning, extinction and psychopathology. Nature Reviews Neuroscience, 14, 417–428. https://doi.org/10.1038/nrn3492
- Milad, M. R., & Quirk, G. J. (2002). Neurons in medial prefrontal cortex signal memory for fear extinction. Nature, 420, 70–74. https://doi.org/10.1038/nature01138
- Mobbs, D., Petrovic, P., Marchant, J. L., Hassabis, D., Weiskopf, N., Seymour, B., Dolan, R. J., & Frith, C. D. (2007). When fear is near: Threat imminence elicits prefrontal-periaqueductal gray shifts in humans. Science, 317(5841), 1079–1083. https://doi.org/10.1126/science.1144298
- Namburi, P., Beyeler, A., Yorozu, S., Calhoon, G. G., Halbert, S. A., Wichmann, R., Holden, S. S., Mertens, K. L., Anahtar, M., Felix-Ortiz, A. C., Wickersham, I. R., Gray, J. M., & Tye, K. M. (2015). A circuit mechanism for differentiating positive and negative associations. Nature, 520, 675–678. https://doi.org/10.1038/nature14366
- Phelps, E. A., Delgado, M. R., Nearing, K. I., & LeDoux, J. E. (2004). Extinction learning in humans: Role of the amygdala and vmPFC. Neuron, 43(6), 897–905. https://doi.org/10.1016/j.neuron.2004.08.042
- Quirk, G. J., Repa, C., & LeDoux, J. E. (1995). Fear conditioning enhances short-latency auditory responses of lateral amygdala neurons: Parallel recordings in the freely behaving rat. Neuron, 15(5), 1029–1039. https://doi.org/10.1016/0896-6273(95)90092-6
- Rescorla, R. A. (1968). Probability of shock in the presence and absence of CS in fear conditioning. Journal of Comparative and Physiological Psychology, 66(1), 1–5. https://doi.org/10.1037/h0025984
- Sevenster, D., Beckers, T., & Kindt, M. (2013). Prediction error governs pharmacologically induced amnesia for learned fear. Science, 339(6121), 830–833. https://doi.org/10.1126/science.1231357
- Shackman, A. J., & Fox, A. S. (2016). Contributions of the central extended amygdala to fear and anxiety. Journal of Neuroscience, 36(31), 8050–8063. https://doi.org/10.1523/JNEUROSCI.0982-16.2016
- Tovote, P., Fadok, J. P., & Lüthi, A. (2015). Neuronal circuits for fear and anxiety. Nature Reviews Neuroscience, 16, 317–331. https://doi.org/10.1038/nrn3945
- Tovote, P., Esposito, M. S., Botta, P., Chaudun, F., Fadok, J. P., Markovic, M., Wolff, S. B. E., Ramakrishnan, C., Fenno, L., Deisseroth, K., Herry, C., Arber, S., & Lüthi, A. (2016). Midbrain circuits for defensive behaviour. Nature, 534, 206–212. https://doi.org/10.1038/nature17996
- Tseng, Y.-T., Schaefke, B., Wei, P., & Wang, L. (2023). Defensive responses: Behaviour, the brain and the body. Nature Reviews Neuroscience, 24, 655–671. https://doi.org/10.1038/s41583-023-00736-3
- Vasudevan, K., Hassell, J. E., Jr., & Maren, S. (2024). Hippocampal engrams and contextual memory. Advances in Neurobiology, 38, 45–66. https://doi.org/10.1007/978-3-031-62983-9_4
- Webler, R. D., Berg, H., Fhong, K., Tuominen, L., Holt, D. J., Morey, R. A., Lange, I., Burton, P. C., Fullana, M. A., Radua, J., & Lissek, S. (2021). The neurobiology of human fear generalization: Meta-analysis and working neural model. Neuroscience & Biobehavioral Reviews, 128, 421–436. https://doi.org/10.1016/j.neubiorev.2021.06.035