Amygdala
Referenced across the wiki long before it had a page. Five of this wiki’s sources characterize it incompatibly, which makes it a useful place to see locationist-vs-constructionist-brain-emotion play out on one structure.
Five readings
1. Fear centre (locationism). The canonical mapping fear → amygdala, inherited from the basic-emotions tradition and the fear-conditioning literature. This is the target Lindquist et al. (2012) set out to test — and it is not a position LeDoux endorses, despite his identification with amygdala fear research.
2. Salience detector (constructionism). Lindquist et al.’s meta-analysis found the amygdala functionally selective for fear perception (and disgust experience) but not functionally specific — it responds comparably to novel, uncertain, arousing, or salient stimuli of any valence. Reframed as a domain-general salience detector and folded into the core affect network. See also salience-network, where the amygdala is a node of the AIC/ACC-anchored network.
3. Nucleus-specific survival circuit component (LeDoux). LeDoux (2012) rejects “the amygdala” as an explanatory unit at all. The relevant objects are nuclei within specific circuits, and they differ by trigger modality and by learned vs. innate status. See survival-circuits, pavlovian-defense-conditioning.
4. Trigger structure for somatic states from primary inducers (Bechara & Damasio). Bechara & Damasio (2005) assign the amygdala one specific job in the somatic-marker-hypothesis: it couples the features of primary inducers — stimuli present in the environment — to their somatic states, via effector structures (hypothalamus, autonomic brainstem nuclei, ventral striatum, PAG). Fast, automatic, obligatory, short-lived, habituating quickly. The VM cortex handles the other half (thoughts and memories).
The load-bearing claim is developmental: because secondary induction works by re-activating patterns laid down by primary induction, amygdala damage starves it. Hence the iowa-gambling-task dissociation — amygdala patients generate no SCRs at all, not even to actual reward and punishment, while VM patients generate those but not anticipatory ones. When the amygdala is damaged, “the patient can no longer register how painful it feels when one loses money,” which in turn “misleads” the VM cortex about how painful a loss should feel.
This reading is worth keeping distinct from (1) despite the surface resemblance. It is not a claim that the amygdala is fear: the category of primary inducers includes lottery wins and the “aha” of solving a puzzle alongside snakes, so the proposed function is obligatory somatic triggering irrespective of valence. But note it is vulnerable to LeDoux’s objection in reading (3) — the three examples are grouped by obligatoriness alone, with no account of whether the same nuclei, routes, or learning histories are involved. Bechara & Damasio treat “the amygdala” as exactly the explanatory unit LeDoux denies exists. Nothing in this wiki tests which is right.
Reading (4) is not stable within its own programme
Added with the Damasio (1996) ingest, and it complicates the reading above rather than supporting it.
Nine years before the trigger account, the same author gives the amygdala the opposite position in the causal chain. In the 1996 network the amygdala is a central autonomic effector: ventromedial cortices activate somatic effectors in amygdala, hypothalamus, and brainstem nuclei. It is downstream — an output stage that executes a somatic state on the viscera, vascular bed, endocrine system, and non-specific neurotransmitter systems when VM calls for it.
| 1996 | 2005 | |
|---|---|---|
| position | downstream of VM | upstream / parallel to VM |
| function | executes a somatic state on the body | triggers a somatic state from a present stimulus |
| what depends on it | the effector arm of the body loop | the entire secondary dissociation |
Neither text acknowledges the other. The raw material for the 2005 role was already in hand in 1996 — the paper contains a fast subcortical route to autonomic centres direct from thalamus, bypassing cortex, for basic unconditioned stimuli like a startling noise or a flash of light, cited to Clugnet et al. (1988) and Farb, Ruggiero & LeDoux (1988), i.e. LeDoux’s own “low road.” But Damasio uses it for something else entirely: a predicted dissociation between responses to complex stimuli requiring cortical processing and basic stimuli not requiring it. The trigger architecture is not built on it.
So the wiki should treat “the amygdala’s role in the somatic-marker-hypothesis” as a moving target, not a fixed fourth reading. What is constant across both texts is only that the amygdala is where obligatory, body-directed responses get organized; whether it commands VM or answers to it changed, silently, between the two primary sources this wiki holds.
That instability strengthens LeDoux’s objection rather than weakening it. A structure whose position in a causal chain can invert without anyone noticing is a structure being used as a placeholder for “the fast automatic bit” — which is close to what reading (3) alleges.
A third timepoint, and the amygdala is not in it. The Bechara, Damasio & Damasio (2000) ingest supplies the midpoint text, which names the amygdala in its abstract as one of the “other cortical and subcortical components” of the decision system and then brackets it: “Here we focus only on the role of the orbitofrontal cortex.” So the structure is downstream in 1996, set aside in 2000, and upstream in 2005 — and the inversion happened in the last five years of that span, after the amygdala results the 2005 architecture rests on (Bechara et al. 1999) were already published and cited in the 2000 paper for something else. Recorded on primary-and-secondary-inducers.
The one amygdala result the 2000 paper does turn on
Not about triggering, and useful independently of which reading above survives. Twelve controls and six VM patients recalled neutral and emotionally charged pictures presented 1, 2, 4 or 8 times; both groups showed the normal emotional-recall advantage.
The authors’ reasoning: the amygdala is necessary for emotion to enhance memory (Cahill et al. 1995) and contributes to biasing and decision-making (Bechara et al. 1999a), so within the amygdala those two functions may be inseparable — but they dissociate downstream, at VM cortex, where a patient can use emotional content to remember while failing to use it to choose.
Two things worth keeping. It removes “cannot recall emotional events” as an explanation of the decision deficit. And it means “emotion modulates cognition” is at least two mechanisms sharing a structure upstream — a claim about functional architecture that the constructionist readings (2) and (5) would want to restate in their own terms and neither has. Hold it at the source’s discount: a null result at n = 6, reported in a review with no statistics and no power analysis.
(Housekeeping: the Farb of Farb, Ruggiero & LeDoux 1988 is Claudia Farb, LeDoux’s anatomist collaborator. No relation to norman-farb. Noted because this wiki’s raw/ tree makes the collision likely to recur.)
5. Uncertainty signal for a body budget (Barrett). Barrett (2017) removes the amygdala from emotion entirely:
I hypothesize that information from the amygdala to the cortex is not ‘emotional’ per se, but signals uncertainty (Whalen, 1998) about the predicted sensory input (via the basolateral complex) and helps to adjust allostasis (via the central nucleus) as a result.
Amygdala output is prediction error, sent to cortex to correct the internal model. The arousal that accompanies amygdala activity “can be considered a learning signal” (Li & McNally, 2014) rather than a component of an emotion. She extends the treatment to the neighbours: ventral striatal “reward prediction errors” convey that an input impacted allostasis more than expected and should be encoded; dopamine supports vigorous action and the learning needed to secure rewards, “rather than playing a necessary or sufficient role in rewards themselves” (opioids being more intrinsic to reward). And she reinterprets the monkey lesion data the fear reading depends on: amygdalectomized monkeys explore novel objects sooner, usually read as “lack of fear,” but an alternative is that the amygdala regulates exploration under uncertainty — “‘Fear’ is not necessary.”
Why reading (5) is not just reading (2) restated
Easy to file as constructionism-again, and it isn’t. The wiki’s reading (2) — Lindquist’s domain-general salience detector — is exactly what LeDoux attacks on resolution grounds in the section below: treating “the amygdala” as one thing with one broad job, when unit recordings show distinct appetitive and aversive populations inside it.
Barrett’s 2017 reading assigns different jobs to different nuclei — basolateral signals uncertainty, central nucleus adjusts the body budget. That is nucleus-specific in the way LeDoux demands, and it comes from the camp he says cannot deliver it. It does not answer his objection (she still infers function from imaging plus theory, and her claim that amygdala neurons are multipurpose rests on a personal communication — Cerf, 30 July 2015 — which the wiki cannot check). But it undercuts the specific complaint that constructionists treat the structure as homogeneous.
Note also that (5) is the only reading here that makes the amygdala’s job uncertainty about a prediction rather than a response to a stimulus. On readings (1)–(4) something happens and the amygdala reacts. On (5) the amygdala reports how much the model should be corrected. That difference is predictive-coding doing the work, not constructionism — and Barrett’s uncertainty proposal is in principle compatible with LeDoux’s circuits, since a circuit could compute uncertainty. Nothing in this wiki tests it.
The nucleus-level anatomy (LeDoux 2012, Figure 1)
| Pathway | Route |
|---|---|
| Unconditioned olfactory threat | vomeronasal → MEA → VMH → PMH → dorsal PAG |
| Unconditioned non-olfactory threat | sensory → LA → ABA → VMH-PM-PAG |
| Conditioned threat (any modality) | sensory → LA → (direct, and via BA/ABA/ITC) → CEA → medial CEA → ventrolateral PAG → freezing |
Different subnuclei of MEA, PMH, and PAGd handle conspecific vs. predatory threat. Damage to PAGvl disrupts freezing while PAGdl lesions enhance it, implying interaction between the regions. Homologous circuitry is reported in rabbits, nonhuman primates, reptiles, and birds; human imaging and lesion work supports an amygdala role in defense conditioning “at least to a first approximation,” though at resolution that “obscures circuit details.”
Appetitive functions, and the argument they support
The amygdala is not defense-only: LA/BA/ABA process learned food cues and relay to lateral hypothalamus, where sufficiently potent cues stimulate eating even in sated animals. CEA outputs suppress feeding to an aversive CS; MEA threat-odor areas suppress reproduction via VMH.
LeDoux uses this to make a methodological point aimed squarely at reading (2): that the amygdala serves both appetitive and aversive functions does not mean it processes them alike. Primate unit recordings show appetitive and aversive signals handled by distinct neuronal populations in lateral/basal amygdala (Paton et al. 2006; Belova et al. 2007, 2008; Morrison & Salzman 2010), and molecular imaging shows area-level activation similarity concealing microcircuit-level difference (Lin et al. 2011). His charge is that fMRI-based claims of shared mechanism — Lindquist’s and Barrett’s — outrun the resolution of the method.
This is the sharpest live methodological disagreement in the wiki’s emotion-anatomy material: selectivity-without-specificity at the voxel level (Lindquist) vs. specificity at the cellular level invisible to voxels (LeDoux). Both are consistent with the imaging data. They differ on what imaging data can license.
Route to arousal and the body
Central amygdala outputs drive the brain-wide neuromodulator systems and, peripherally, the sympathetic ANS (adrenal medulla) and the hpa-axis (cortisol) — the mechanism by which a detected threat becomes a whole-body global organismic state rather than a local response.
Selective for negative arousal, not valence — a lesion dissociation (Berntson et al. 2010)
Berntson, Norman, Bechara, Tranel & Cacioppo (2010) supply the causal, subtractive version of reading (2)‘s arousal claim, and it is unusually clean. Twelve amygdala-lesion patients (anterior temporal lobectomy) rated IAPS pictures on valence and arousal. Their valence ratings were indistinguishable from controls (d < 0.01) — they recognized and categorized the affective content — while their arousal ratings were selectively attenuated for unpleasant stimuli only (the normal arousal increment survived for pleasant pictures). Removing the amygdala costs the felt impact of aversive content without costing the judgment of what that content is.
Two things this pins down for the readings above. It is the wiki’s cleanest causal evidence that the amygdala’s evaluative job is registering arousal/emotional impact, especially for aversive stimuli (readings 2 and 4), dissociated from determining appetitive-vs-aversive valence — “the amygdala may not be necessary to determine whether and to what extent a stimulus is appetitive or aversive.” And it sits opposite the same study’s insula result (broad loss of both valence and arousal, both signs; see insular-cortex), making one paper a double dissociation of the two core-affect dimensions across two structures. The standard caveat is sharper than usual here: the amygdala group is confounded with the insula/control groups on aetiology (lobectomy vs. stroke) and age, so the authors treat it as ancillary and lean on its replication of Berntson et al. (2007).
An instance of reading (2): salience, trained up by compassion (Lutz et al. 2008)
Lutz et al. (2008) supply a clean instance of reading (2) — the amygdala as a domain-general detector of emotional salience — that also happens to be trainable. In their compassion-meditation study, bilateral amygdalae were among the regions where experts, more than novices, activated during meditation vs. rest, alongside the TPJ and IFG attentional-detection circuit; the authors gloss the amygdala’s role here as “appraisal of emotional stimuli,” priming experts to register the salience of another’s distress. Note this is the amygdala responding to salience in another person’s state — the empathy/salience-network use — and it is an fMRI area-level activation, so it is exactly the kind of result LeDoux’s reading (3) cautions against over-reading: voxel-scale amygdala activation to emotional sounds does not tell us which nuclei or populations are involved. Recorded as another data point in the selectivity-without-specificity dispute, from the contemplative-training literature.
A sixth role: memory consolidation (van der Kolk 1994)
Added with the van der Kolk (1994) ingest, which gives the amygdala a job the five readings above leave out: gatekeeper of memory encoding. In the 1990s dual-memory model van der Kolk builds on (his Figure 2, Table 2), the amygdala assigns emotional significance fast and the hippocampus does the slower declarative/contextual encoding — and the two interact non-linearly:
- Moderate amygdala activation enhances hippocampal long-term potentiation → hypermnesia for emotionally significant events.
- Excessive activation inhibits hippocampal function → the experience is not laid down as declarative memory at all but “stored in sensorimotor modalities: somatic sensations and visual images.”
The driver is norepinephrine input to the amygdala (McGaugh’s work), operating on an inverted-U: both too little and too much impair consolidation, which van der Kolk uses to explain PTSD’s coexisting hypermnesias and amnesias. See traumatic-memory.
This reading is compatible with several of the others rather than a rival to them — it is the amygdala-as-significance-tagger (readings 2-4) extended into what significance-tagging does to what gets remembered. But it is worth keeping distinct because it is the wiki’s only source in which the amygdala’s output is cast as modulating an encoding process rather than triggering a response or reporting uncertainty. Note the tension with reading (5): where Barrett’s central-nucleus output adjusts the body budget, van der Kolk’s amygdala output gates the hippocampus — the same structure, different downstream target, on frameworks 20 years apart. Table 2’s lesion double-dissociation (hippocampal damage: declarative memory lost, skill/immediate memory spared; amygdala damage: fear conditioning and affect-to-neutral-stimuli lost, declarative memory intact) is the 1994 evidence base, since complicated by later imaging.