The brain basis of emotion: A meta-analytic review

Study at a glance

• Authors: Kristen A. Lindquist, Tor D. Wager, Hedy Kober, Eliza Bliss-Moreau, Lisa Feldman Barrett
• Year: 2012
• Journal: Behavioral and Brain Sciences
• DOI: 10.1017/S0140525X11000446
• Design: coordinate-based neuroimaging meta-analysis (density, chi-square, and logistic-regression analyses over ~90+ studies, 1990–2007) + BBS open peer commentary + authors’ response

Key findings

• Directly tests the locationist hypothesis (discrete emotion categories — fear, anger, disgust, sadness, happiness — map consistently and specifically onto distinct brain regions or inherited anatomical networks; e.g. fear=amygdala, disgust=insula, anger=OFC, sadness=ACC) against a psychological constructionist hypothesis (emotions emerge from domain-general psychological operations realized in networks not specific to emotion, let alone to any one category).
• Found no brain region showing both consistency AND specificity for any single discrete emotion category — every region activated for one category was also activated for at least one other. Functional selectivity (relative preference) was found; functional specificity (exclusivity) was not.
• The amygdala shows functional selectivity for fear perception and disgust experience but is equally active for other high-arousal, salient, or novel stimuli — consistent with a domain-general salience-detection role, not a dedicated ‘fear module.’
• The anterior insula shows functional selectivity for disgust perception/experience but is not specific to disgust; reframed as a hub for representing core affective feelings and interoceptive awareness (explicitly citing Craig 2002 and Craig 2009), active whenever people attend to and represent bodily states in awareness — not a disgust module.
• The ACC (pACC, sACC, aMCC) shows no consistent sadness-specificity; reframed as visceromotor regulation (pACC/sACC) and response-selection/executive-attention (aMCC) sites active across emotion categories generally.
• The OFC shows no anger-specificity; reframed as a heteromodal integrator of exteroceptive and interoceptive sensory information guiding context-sensitive behaviour.
• Identifies a domain-general ‘neural reference space for discrete emotion’ comprising (1) a core affect network (amygdala, insula, mOFC, lOFC, ACC, thalamus, hypothalamus, BNST, basal forebrain, PAG), (2) a conceptualization network (DMPFC, medial temporal lobe, retrosplenial/posterior cingulate — the ‘default network’), (3) a language network (anterior temporal lobe, VLPFC), and (4) an executive attention network (DLPFC, VLPFC) — plus visual cortex, consistently co-activated across all examined emotion categories.
• No inherited, anatomically fixed intrinsic (resting-state) networks corresponding to discrete emotion categories have been found in the broader network literature; the intrinsic networks that do exist (default, salience/personal-relevance, executive-control) resemble the constructionist psychological-operation networks, not locationist emotion-specific circuits.
• Concludes the century-old locationist intuition (traceable to the ‘basic emotions’ program) is not supported by the neuroimaging evidence, echoing James’s own 1890 view that emotion requires no dedicated brain centres beyond ordinary sensational/associational/motor elements.

Limitations

• Coordinate-based meta-analysis inherits reverse-inference risk and depends on the composition of the published literature; ~90%+ of fear-perception contrasts used startled faces and ~35% of disgust-experience contrasts used novel contamination/mutilation images — caricatured, high-arousal stimuli that may not represent naturalistic emotion instances.
• Functional selectivity is a weaker claim than specificity; the authors explicitly did not test whether emotion categories are represented by unique, distributed spatial-activation patterns (multivoxel pattern analysis) rather than univariate regional consistency — a possibility left fully open for a locationist rescue.
• As a BBS target article, drew substantial pushback in Open Peer Commentary: Scherer (component process/appraisal theory) argues the ‘locationist’ label is a ‘gross misrepresentation’ of most rival (especially componential appraisal) theories, which never assumed emotion-specific brain centres to begin with, and that the meta-analysis’s positive findings (e.g., ‘interoceptive representation in the insula’) are equally consistent with appraisal theory. Smaldino & Schank and others caution against fully abandoning ‘quasi-natural perceptual kinds’ for emotion expression/recognition even while accepting the anti-locationist neural conclusion.
• Stapleton’s commentary flags an unresolved ambiguity in the model itself: whether ‘core affect’ denotes (1) a psychological feeling state, (2) the neural representation of interoceptive information, or (3) the bodily state itself — a conflation risk the authors do not fully resolve in their response.

Connections

• Methods: Activation likelihood estimation (ALE), Meta-analytic reverse inference (Neurosynth decoding)
• Concepts: Basic emotions and natural kinds, Core affect, Cognitive appraisal (of emotion), Insular cortex, Anterior cingulate cortex (ACC), Salience network, Interoception, Autonomic specificity of emotion, Predictive coding
• Debates: Do discrete emotions map onto distinct brain regions or networks?, Do basic emotions have distinct autonomic signatures?
• Related studies: Craig (2002), Craig (2009), Seth & Friston (2016), Friedman (2010), MacCormack et al. (2021)

A Behavioral and Brain Sciences target article (with ~30 commentaries and an authors’ response) reporting a large coordinate-based meta-analysis of the human neuroimaging literature on discrete emotion, run to adjudicate between a locationist account (fear=amygdala, disgust=insula, anger=OFC, sadness=ACC — see Fig. 1 of the target article) and Barrett’s psychological constructionist (“conceptual act model”) account, in which emotion categories are not natural kinds but situated conceptualizations built from domain-general operations. First-authored by Kristen A. Lindquist (senior author Lisa Feldman Barrett); the same pairing later runs the account down the lifespan in MacCormack et al. (2021). See Basic emotions and natural kinds.

The constructionist model being tested

Four hypothesized “ingredients,” none specific to emotion: core affect (raw valenced/aroused bodily sensation), conceptualization (using stored prior experience to make sensation meaningful, “situated conceptualization”), language (emotion words anchor otherwise loosely-bounded categories), and executive attention (selects which representations combine). Emotions are the joint product of these operations interacting like “ingredients in a recipe,” not discrete triggered modules.

Method

Density analyses (which voxels show more consistent activation for one emotion category than others), chi-square analyses (functional selectivity vs. specificity), and stepwise logistic regressions (which categories/methods predict activity in a given region), applied across neuroimaging contrasts of the experience or perception of anger, disgust, fear, happiness, and sadness published 1990–2007.

Anatomy findings directly relevant to this wiki

• Amygdala (locationist: fear): functionally selective for fear perception and disgust experience, but not specific — equally responsive to novel, uncertain, arousing, or salient stimuli of any valence. Reframed as a salience detector, not a fear centre.
• Anterior insula (locationist: disgust): functionally selective for disgust, but not specific to it. The paper’s positive reframing lands squarely on this wiki’s existing anatomical model — it explicitly invokes Craig (2002) and Craig (2009): “the anterior insula plays a key role in representing core affective feelings in awareness… is thought to be involved in the awareness of bodily sensations (Craig 2002) and affective feelings (Craig 2009).” Increased activation also occurs during body-movement awareness, gastric distension, and orgasm — states with nothing to do with disgust — consistent with a general interoceptive-awareness role. See Insular cortex.
• ACC (pACC, sACC, aMCC; locationist: sadness): not sadness-specific. pACC/sACC reframed as visceral-regulation sites (consistent with Craig’s “limbic motor cortex” framing); aMCC reframed as a response-selection/executive-attention site receiving exteroceptive (thalamic) and interoceptive (insular) input to direct attention and motor response.
• OFC (locationist: anger): not anger-specific. Reframed as integrating exteroceptive and interoceptive sensory information for context-sensitive behaviour (consistent with Seth & Friston’s later “VMA” characterization of the OFC — see Visceromotor areas (VMAs)).
• PAG: consistently active across the whole “neural reference space” for emotion (even though scanner participants lie still), but with no category specificity — consistent with a general autonomic/arousal role rather than discrete Panksepp-style rage/fear/lust circuits.

Is this a contradiction of Craig, or a vindication?

Mostly the latter, with one real tension. Lindquist et al.’s meta-analytic rejection of “insula = disgust” and “ACC = sadness” is not a rejection of Craig’s own claims — Craig never proposed disgust-specificity; his claim was always that the insula (AIC) supports general interoceptive/affective awareness, and Lindquist’s data support exactly that reading, citing Craig directly as source. The genuine tension is with Craig’s forebrain emotional asymmetry hypothesis (right AIC ↔ sympathetic/survival/withdrawal emotions like pain, anger, anxiety; left AIC ↔ parasympathetic/affiliative/approach emotions — see Insular cortex), which is a category-cluster-to-hemisphere locationist claim in Lindquist et al.’s sense, and would be predicted to fail the same functional-specificity tests this meta-analysis applies to single regions. Flagged here as an unresolved cross-paper tension, not merged or resolved.

New debate raised

This paper’s own target/commentary/response structure makes explicit a debate this wiki did not yet have a page for — not “do emotions have distinct autonomic signatures” (Do basic emotions have distinct autonomic signatures?, about peripheral physiology) but whether discrete emotions map to distinct brain structures or networks at all. See the new Do discrete emotions map onto distinct brain regions or networks? debate, which also records Scherer’s commentary that the locationist/constructionist dichotomy mischaracterizes componential appraisal theory, and that “any evidence for brain mechanisms subserving categorization/conceptualization… does not privilege any theory on why or when” — a methodological caution about what a meta-analysis of this kind can and cannot decide between rival theories.

Bridge to interoceptive inference

Mog Stapleton’s commentary (see limitations) explicitly connects Lindquist et al.’s core-affect construct to predictive coding (Bar, Friston) — proposing that unexpected interoceptive sensations are intrinsically motivating because the nervous system’s function is error-reduction, and that precision/gain control (Feldman & Friston 2010) implements attention over prediction-error units. Written four years before Seth & Friston (2016) formalized exactly this move for interoception, it is an early, independent bridge between the constructionist and predictive-coding literatures represented in this wiki.