Visceromotor areas (VMAs)
Introduced in Seth & Friston (2016) as the anatomical apex of the interoceptive-inference hierarchy: AIC, ACC, subgenual cortex (SGC), and OFC. VMAs are proposed to embody a generative model of interoceptive responses, issuing descending predictions that, once “unpacked” at the lowest hierarchical level, function as homeostatic set-points for autonomic reflexes.
The structural argument
The paper’s distinctive move is anatomical rather than purely functional: VMAs are agranular or dysgranular — they lack a well-formed granular layer IV, the canonical cytoarchitectural target of feedforward, prediction-error-carrying afferents in laminar models of predictive coding (Barbas-style). Lacking that input layer, VMAs are structurally suited to be a source of descending predictions rather than a recipient of ascending error — giving Seth’s predictive reading of interoception a structural foothold beyond the circumstantial functional evidence marshalled in Seth (2013). See feedforward-vs-predictive-interoception.
Connectivity
VMAs receive ascending viscerosensory projections from posterior and mid-insula, and send descending connections to subcortical, brainstem, and spinal targets involved in visceromotor control — notably the periaqueductal grey (PAG) and the parabrachial nucleus (PBN). Visceromotor efferents also directly innervate viscerosensory areas, potentially providing efference copy/corollary discharge that helps construct the ascending interoceptive prediction errors those same predictions are meant to explain away.
Convergence with the constructionist “core affect” network
The VMA set (AIC, ACC, SGC/subgenual, OFC) overlaps substantially with the core affect network that Lindquist et al.’s (2012) neuroimaging meta-analysis identifies empirically (amygdala, insula, mOFC, lOFC, ACC, thalamus, hypothalamus, BNST, basal forebrain, PAG) as the domain-general substrate of emotional experience across categories. Two independent research programs — a predictive-coding anatomical argument (Seth & Friston) and a psychological-constructionist meta-analytic one (Lindquist et al.) — converge on nearly the same set of regions from different starting assumptions and different evidence (cytoarchitecture vs. cross-study activation consistency), though they do not cite one another.
Barrett makes the convergence deliberate — and extends the gradient
The section above records the convergence as a coincidence between literatures. Barrett (2017) closes the loop from the constructionist side: she runs the same cytoarchitectural argument, from the same source (Barbas’s structural model of corticocortical connections), and arrives at the same regions — calling them “limbic” or “visceromotor regions” rather than VMAs.
What the two papers agree on, almost exactly: agranular cortices (ACC, ventral anterior insula) lack a developed layer IV, are “cytoarchitecturally arranged to send but not receive prediction signals within the cerebral cortex,” and relay descending predictions to the internal milieu via central amygdala, ventral/dorsal striatum, and central pattern generators across hypothalamus, PBN, PAG and the solitary nucleus. Dysgranular cortices (midcingulate, mPFC, vlPFC, premotor) also issue visceromotor predictions.
Barrett’s extension, and it is the more surprising claim: the laminar gradient is continuous, so it does not stop at the limbic/sensory boundary. Because primary motor cortex is less granular than primary sensory cortices, it sends them predictions (following Adams, Shipp & Friston, 2013). And because primary interoceptive cortex (mid-to-posterior dorsal insula) is less granular than primary visual, auditory and somatosensory cortex, Barrett hypothesizes that interoceptive cortex forwards sensory predictions to the exteroceptive senses.
Taken seriously that is a directional anatomical claim with real content: your model of your viscera predicts what you will see. Not affect “colouring” perception as a metaphor — a specific proposal about which cortex constrains which, propagating across one or multiple synapses. Hence: “all action and perception are created with concepts. All concepts contribute to allostasis and represent changes in affect, not just those that construct the events that feel affectively intense.”
The wiki should record that this is deduced from a gradient, not measured. Barrett offers it as a hypothesis licensed by three decades of tract tracing plus an engineering principle (compute locally, relay only what is needed to assemble a larger pattern — Sterling & Laughlin, 2015). It is the kind of claim that could be wrong in an interesting way, which distinguishes it from most of the theoretical material here.
What this does to the interoceptive hierarchy
A difference worth not smoothing over. Seth & Friston place the VMAs at the apex of an interoceptive hierarchy — the top of a system for inferring and regulating visceral states. Barrett’s gradient has no interoception-specific hierarchy at all: there is one cortical gradient from agranular to granular, and interoceptive cortex sits partway up it, receiving predictions from limbic cortex and sending them to exteroceptive cortex. Interoception is not a hierarchy with a top; it is a region of a single sheet.
Both are Barbas-derived and neither is obviously wrong. But “the apex of the interoceptive hierarchy” and “one continuous gradient with the interoceptive cortex partway along it” are different architectures, and they make different predictions about whether interoceptive and exteroceptive processing can be dissociated. Nothing in this wiki tests it. See feedforward-vs-predictive-interoception.
The field adopts the visceromotor picture (Khalsa et al. 2018)
The Khalsa et al. (2018) roadmap takes the visceromotor architecture as settled consensus, listing the efferent processing regions as “anterior insula, anterior cingulate, subgenual cingulate, orbitofrontal, ventromedial prefrontal, supplementary motor, and premotor areas” — the same VMA set, slightly widened. Its psychopathology figure reproduces Barrett & Simmons’s EPIC model, in which the agranular visceromotor cortex issues allostatic predictions while the granular mid/posterior insula returns prediction error (see insular-cortex). What the roadmap adds is a clinical stake for the architecture: if the agranular regions are structurally arranged to send-but-not-receive prediction, then the pathological case is that one-way insensitivity failing — interoceptive priors decoupling from posteriors, producing the “noisy afferent interoceptive inputs” the roadmap names as a candidate mechanism of mental illness. The cytoarchitecture the wiki had been tracking as a theoretical argument becomes, here, the substrate a disorder is proposed to break. See computational-psychiatry, interoceptive-psychopathology.
Quadt, Critchley & Garfinkel (2018) restate the EPIC loop in the same terms and add the useful detail that the agranular regions’ relative insensitivity to ascending error is what permits abstract, future-directed (allostatic) prediction rather than reactive homeostasis — the send-but-not-receive architecture is a feature, not a bug, until it breaks. Their depression application is the clearest worked case of the break: visceromotor cortical dysfunction overpredicts metabolic demand, and the downweighting of the resulting error is what locks the faulty predictions in place (see computational-psychiatry).