Your brain learned to fear without asking your conscious mind for permission. A sound, a smell, a particular face can trigger a cascade of physiological responses before you even recognize what you’re reacting to. This is fear conditioning, the brain’s ancient system for associating neutral stimuli with danger.
For decades, neuroscience textbooks placed the amygdala at the center of this process. New research complicates that picture considerably. A landmark mega-analysis of 2199 individuals across nine countries found that the amygdala did not show significant activation during fear conditioning in fMRI scans[s]. Meanwhile, star-shaped support cells called astrocytes, long dismissed as mere scaffolding, turn out to be active participants in encoding and extinguishing fear memories.
How Fear Conditioning Works
Fear conditioning follows a simple pattern: pair a neutral stimulus (a tone, a light, a room) with something unpleasant (a shock, a loud noise), and the brain learns to associate the two. After conditioning, the formerly neutral stimulus alone triggers fear responses: increased heart rate, sweating, heightened alertness.
This process happens fast. A single pairing can create a lasting association. The brain prioritizes threats, and what gets encoded as dangerous tends to stick.
The ENIGMA consortium’s mega-analysis revealed that fear conditioning consistently engages brain regions within the “central autonomic-interoceptive” or “salience” network[s]. This network includes the dorsal anterior cingulate cortex (dACC), a strip of tissue along the brain’s midline involved in conflict monitoring and autonomic control, and the anterior insular cortex, which processes bodily sensations and emotional awareness.
The amygdala’s absence from this list challenges decades of assumptions. Neither the left nor the right amygdala showed significant activation during fear conditioning in the mega-analysis[s]. This does not mean the amygdala is uninvolved, but it suggests the story is more complex than the textbook version. Rodent studies consistently implicate the amygdala, but fMRI in humans tells a different tale. The salience network may be doing more of the heavy lifting than previously recognized.
The Astrocyte Revolution
Neurons get most of the attention, but astrocytes are a major class of glial cell in the brain. These cells wrap around synapses, regulate neurotransmitter levels, and maintain the blood-brain barrier. Until recently, their role in learning and memory was underappreciated.
A 2026 Nature paper demonstrated that basolateral amygdala astrocytes dynamically track fear state and support fear memory retrieval and extinction[s]. Using calcium imaging in mice, researchers watched astrocyte activity rise during fear memory retrieval and fall during extinction. Disrupting astrocyte calcium signaling impaired the brain’s ability to encode and retrieve fear memories.
The findings reveal a key role for astrocytes in the generation and adaptation of fear-state-related neural representations, revising neurocentric models of critical amygdala-mediated adaptive functions[s]. This is not a minor addendum: astrocytes are active participants in fear learning, not passive bystanders.
Unlearning Fear: How Extinction Works
Fear conditioning creates problems when the learned associations no longer match reality. A car crash survivor who flinches at every sudden noise, a veteran who cannot tolerate fireworks: these are adaptive responses gone maladaptive. The brain has a built-in mechanism for updating these associations, called extinction.
Extinction is not erasure. The original fear memory remains; what changes is the brain’s response to it. A new “safety” memory forms and competes with the fear memory for expression. This is why extinguished fears can return: place someone back in the original context, and the fear memory can reassert itself, a phenomenon called renewal.
Intracranial recordings in epilepsy patients revealed that amygdala theta oscillations during extinction learning signal safety rather than threat[s]. In the observed effect, theta-band power (3-10 Hz) increased during exposure to safe stimuli and decreased during threatening ones. The mutual competition of fear and extinction memory traces provides a mechanistic basis for clinically important phenomena such as fear renewal and extinction retrieval[s].
The prefrontal cortex plays a crucial role in extinction. Studies using transcranial magnetic stimulation found that both memory reactivation and intact prefrontal cortex function were necessary for the short-term fear amnesia after a retrieval-extinction protocol[s]. Disrupting dorsolateral prefrontal cortex activity blocked the immediate suppression of fear, revealing a two-stage process: rapid cognitive suppression followed by slower reconsolidation-based updating.
Fear Generalization: When the Brain Overapplies Lessons
Fear conditioning would be useless if it only worked for exact matches. A mouse that learned to fear one specific cat would not survive long. The brain generalizes: fear spreads to similar stimuli and contexts.
In mice, this generalization runs through an ascending circuit from the basolateral amygdala to the anterior cingulate cortex. Synaptic plasticity within the ACC and signaling from basolateral amygdala inputs during fear learning are necessary for generalized fear responses to novel contexts[s]. The ACC extracts general features of a threatening experience rather than specific contextual information[s].
One striking finding from that mouse work: schematic learning can occur in the prefrontal cortex after single-trial learning[s]. Standard models often describe schematic learning as requiring repeated exposure over time. A single intense experience can apparently shortcut this process, encoding a general “this type of situation is dangerous” rule immediately.
The Extended Fear Network
The amygdala forms extensive connections with the bed nucleus of the stria terminalis (BNST), ventral hippocampus, and medial prefrontal cortex, establishing essential neural circuits that underlie the expression of anxiety-like behaviors and emotional memory[s].
These regions divide labor in important ways. The BNST specializes in processing unpredictable threat contexts, contrasting with amygdala nuclei handling discrete threats[s]. A specific cue predicting danger activates the amygdala; vague, sustained anxiety about uncertain threats engages the BNST. This distinction matters clinically: generalized anxiety disorder may involve BNST dysfunction more than amygdala dysfunction.
Activation of the basolateral amygdala to ventral hippocampus pathway increases anxiety-like behaviors, whereas its inhibition exerts anxiolytic effects[s]. Similar to how the brain chemistry of addiction involves dysregulated reward circuits, fear and anxiety involve dysregulated threat-detection circuits.
When Fear Conditioning Goes Wrong: PTSD
Post-traumatic stress disorder can involve conditioned fear responses that resist extinction. Traumatic memories intrude unbidden; cues that remotely resemble the original trauma trigger full-blown fear responses. Current medications target serotonin pathways but help only a subset of patients[s].
Recent research identified a new mechanism: excessive GABA produced by astrocytes impairs the brain’s ability to extinguish fear memories[s]. PTSD patients had unusually high levels of GABA and reduced cerebral blood flow in the medial prefrontal cortex. These findings emerged from brain imaging studies of more than 380 participants[s].
The source of this excess GABA? Astrocytes, producing it via the enzyme monoamine oxidase B (MAOB). A drug called KDS2010, which selectively blocks MAOB, reversed PTSD-like symptoms in mice[s]. The drug has passed Phase 1 safety trials in humans[s]; as of May 2026, ClinicalTrials.gov lists Phase 2 KDS2010 studies for obesity and Alzheimer’s disease, while PTSD evidence remains preclinical[s][s].
Brain activation patterns during fear conditioning differ between healthy individuals and those with anxiety-related or depressive disorders, with distinct profiles characterizing specific disorders such as post-traumatic stress disorder[s]. The same basic mechanism, fear conditioning, manifests differently depending on clinical status.
What This Means
Fear conditioning is one of the brain’s most fundamental learning mechanisms. Understanding it illuminates not only anxiety disorders but also how the brain perceives threats generally. Similar mechanisms likely operate in the neuroscience of grief and other emotional learning.
The shift from a neuron-centric view to one that includes astrocytes opens new therapeutic avenues. If glial cells contribute to fear learning and extinction, drugs targeting glial function might help patients who do not respond to conventional treatments.
The ENIGMA consortium’s finding that task design and clinical status drive most variability in fear conditioning suggests that methodology matters enormously in this field. What looks like inconsistent results across studies may reflect different experimental setups capturing different aspects of the same underlying process.
Fear conditioning happens below the level of conscious awareness. Decisions precede conscious awareness in many domains of brain function, and threat learning is no exception. The brain decides what is dangerous before “you” have any say in the matter. Understanding this process is the first step toward changing it.
Neural Correlates of Fear Conditioning
Fear conditioning pairs a conditioned stimulus (CS+) with an unconditioned stimulus (US, typically electric shock) to create an associative memory. Post-conditioning, the CS+ alone elicits conditioned responses (CRs): skin conductance increases, startle potentiation, autonomic activation. Control conditions use an unpaired CS- to establish specificity.
The ENIGMA consortium’s pre-registered mega-analysis pooled harmonized fMRI data from 43 samples across 21 laboratories (n=2199, including 1888 healthy controls and 311 with anxiety-related or depressive disorders)[s]. Linear mixed-effect models revealed that fear conditioning consistently engages brain regions within the central autonomic-interoceptive or salience network[s].
Significant CS+ > CS- activation clusters included: bilateral anterior and mid insular cortices, secondary somatosensory cortex (SII), dorsolateral prefrontal cortex (dlPFC), lateral premotor cortices, dorsal anterior cingulate cortex (dACC) extending to pre-supplementary motor area (SMA), ventral posterior cingulate cortex (PCC), dorsal precuneus, dorsal striatum (caudate nucleus), globus pallidus, ventral tegmental area, mediodorsal thalamus, and midbrain tegmentum.
Region-of-interest analysis found that neither the left amygdala (t = 1.93, p = 0.054, Cohen’s d = 0.129) nor the right amygdala (t = 1.57, p = 0.116, Cohen’s d = 0.117) showed significant activation during fear conditioning[s]. Significant CS+ < CS- deactivations occurred in default mode network regions: vmPFC, mPFC, subgenual cingulate, angular gyri, parahippocampi, and hippocampi.
Astrocyte Contribution to Fear Memory Encoding
Basolateral amygdala astrocytes dynamically track fear state and support fear memory retrieval and extinction[s]. In vivo fiber photometry in GfaABC1D-cyto-GCaMP6f-expressing mice revealed robust astrocyte Ca2+ responses to US (not CS) presentation during conditioning, then to CS presentation during early extinction (fear retrieval). CS-related astrocyte Ca2+ activity (AUC, transient frequency) decreased with extinction and recovered on fear renewal in the conditioning context.
Chemogenetic manipulation using hM3Dq DREADDs in BLA astrocytes demonstrated bidirectional effects. hM3Dq actuation (CNO injection before extinction) markedly increased Ca2+ activity within 10 minutes but thereafter decreased and remained low for at least 2 hours. This constrained astrocyte dynamics reduced CS-related freezing during early extinction, consistent with impaired memory retrieval. Astrocyte Ca2+ signalling enables neuronal encoding of fear memory retrieval and extinction, and readout through a BLA-prefrontal circuit[s].
In vivo one-photon cellular-resolution BLA neuronal (GCaMP7f) Ca2+ imaging combined with astrocyte hM3Dq actuation showed that disrupting astrocyte Ca2+ activity reduced the number of conditionally CS-excited neurons. PCA decomposition indicated that CS-related BLA neuronal activity resided in a lower-dimensional state-space and was less dynamic when astrocyte Ca2+ signaling was disrupted.
Extinction Mechanisms: Theta Oscillations and Competing Memory Traces
Intracranial EEG recordings in 49 epilepsy patients undergoing a novel fear and extinction paradigm revealed that amygdala theta oscillations during extinction learning signal safety rather than threat[s]. Theta (3-10 Hz) power was significantly higher for CS- than CS+ trials in the amygdala during extinction (pcorr = 0.005), with effects driven by both theta power increases above baseline for CS+- items (previously threatening, now safe) and theta power reductions below baseline for CS++ items (still threatening).
Representational similarity analysis showed that extinction memory traces are characterized by stable and context-specific neural representations that are coordinated across the extinction network[s]. Context specificity during extinction learning predicted the reoccurrence of fear memory traces during subsequent testing, while reoccurrence of extinction memory traces predicted safety responses. The mutual competition of fear and extinction memory traces provides a mechanistic basis for clinically important phenomena such as fear renewal and extinction retrieval[s].
Dual Mechanisms of Fear Amnesia
A retrieval-extinction protocol produces fear suppression through two temporally and mechanistically distinct pathways[s]. The memory retrieval-extinction protocol prevents the return of fear expression shortly after extinction training (30 min), and this short-term effect is memory reactivation dependent[s].
Continuous theta-burst stimulation (cTBS) to dlPFC demonstrated that both memory reactivation and intact prefrontal cortex function were necessary for the short-term fear amnesia[s]. TMS to dlPFC selectively blocked the short-term effect without affecting long-term (24 h) reconsolidation-based updating.
The differences in temporal scale, cue specificity, and cognitive control ability dependence between the short- and long-term amnesia suggest that memory retrieval and extinction training trigger distinct underlying memory update mechanisms[s]. Short-term amnesia generalizes across cues and depends on cognitive control ability; long-term amnesia is cue-specific and depends on protein synthesis during reconsolidation.
Fear Generalization Circuit: BLA → ACC
Synaptic plasticity within the ACC and signaling from basolateral amygdala inputs during fear learning are necessary for generalized fear responses to novel contexts[s]. Chemogenetic inactivation of BLA terminals in ACC during conditioning blocked generalization to novel contexts without affecting fear expression in the training context.
The ACC did not encode specific fear to the training context, suggesting this region extracts general features of a threatening experience rather than specific contextual information[s]. Arc expression in ACC was elevated following contextual fear conditioning but not immediate shock (which produces no context representation), confirming activity-dependent plasticity.
A striking finding: schematic learning can occur in the PFC after single-trial learning[s]. Standard models attribute schematic learning (extraction of statistical regularities across episodes) to gradual hippocampal-to-cortical transfer over weeks. Single-trial encoding of generalizable threat schemas in ACC challenges this timeline.
Extended Amygdala: BNST vs CeA Dissociation
The amygdala forms extensive connections with BNST, ventral hippocampus (vHPC), and medial prefrontal cortex (mPFC), establishing essential neural circuits that underlie the expression of anxiety-like behaviors and emotional memory[s].
Within the extended amygdala paradigm, the BNST specializes in processing unpredictable threat contexts, contrasting with amygdala nuclei handling discrete threats[s]. The BNST functions as a neural switchboard integrating stress-related information, with its interaction with the HPA axis facilitating stress hormone release.
Optogenetic studies revealed that activation of the BLA-vHPC pathway increases anxiety-like behaviors, whereas its inhibition exerts anxiolytic effects[s]. Similar to how brain chemistry of addiction involves dysregulated mesolimbic dopamine, anxiety involves dysregulated amygdala-hippocampal-prefrontal communication.
PTSD Pathophysiology: Astrocytic GABA and MAOB
Excessive GABA produced by astrocytes impairs the brain’s ability to extinguish fear memories[s]. MRS imaging in over 380 participants revealed that PTSD patients had unusually high levels of GABA and reduced cerebral blood flow in the medial prefrontal cortex[s]. GABA levels decreased in patients who showed clinical improvement.
Postmortem human brain tissue and PTSD-model mice revealed that astrocytes, not neurons, produced abnormal GABA amounts via monoamine oxidase B (MAOB). KDS2010, a selective reversible MAOB inhibitor, reversed PTSD-like symptoms in mice, reduced GABA levels, restored mPFC blood flow, and re-enabled extinction mechanisms[s]. KDS2010 has cleared Phase 1 safety testing, but its PTSD evidence remains preclinical; as of May 2026, listed Phase 2 trials are for obesity and Alzheimer’s disease[s][s][s].
Brain activation patterns during fear conditioning differ between healthy individuals and those with anxiety-related or depressive disorders, with distinct profiles characterizing specific disorders such as post-traumatic stress disorder and obsessive-compulsive disorder[s]. Normative modeling revealed that variance in fear conditioning neural signatures arises primarily from task design and clinical status rather than individual differences like age or trait anxiety.
Implications
Fear conditioning engages the salience network more consistently than the amygdala in human fMRI, challenging textbook models derived from rodent lesion studies. Astrocytes actively contribute to fear memory encoding and extinction, opening therapeutic targets beyond neurons. The prefrontal cortex mediates both rapid cognitive suppression and slower reconsolidation updating of fear memories. In mice, the ACC-BLA circuit enables single-trial schematic learning of generalized threat representations.
These findings inform exposure therapy design: context-dependent extinction memory formation suggests that extinction training in multiple contexts may reduce renewal. The competition between fear and extinction traces, mediated by theta oscillations, provides a mechanistic target. Astrocyte-targeted interventions like MAOB inhibitors offer a novel approach for treatment-resistant PTSD.
Fear conditioning operates below conscious awareness. Understanding how decisions precede conscious awareness in threat learning explains why cognitive insight alone often fails to extinguish conditioned fears: the fear circuit learns and expresses independently of explicit knowledge. The neuroscience of grief, social cognition, and other emotional domains likely share overlapping mechanisms with fear conditioning, suggesting that these findings have broad implications for understanding how the brain perceives threats and encodes emotionally significant experiences.
