Interoceptive sensors
The stage the wiki had been skipping. Berntson & Khalsa (2021) lay interoception out as sensors → pathways → networks → circuits → awareness, and this page is the first term. lamina-i-spinothalamocortical-pathway starts at the dorsal horn; insular-cortex and central-autonomic-network are networks; the method pages measure awareness. What converts a bodily state into a signal in the first place had no page.
Sherrington supposed the adequate stimuli for interoception were chemical. They are one class among many.
The inventory
Chemoreceptors. Widespread and heterogeneous. The classic is the carotid body’s sensitivity to pCO₂, pH and (less so) pO₂ for respiratory and cardiovascular reflexes, achieved partly through acid- and oxygen-sensitive TASK1/TASK3 potassium channels. Beyond it: central and peripheral glucoreceptors, thermoreceptors, gustatory receptors, and osmoreceptors — the last sensitive to local osmotic pressure (cellular dehydration) and the trigger for thirst.
Central sensing without a nerve. Some central neurons appear inherently chemoreceptive, and brain astrocytes — chemically but non-synaptically coupled to neurons — may mediate sensitivity to local pCO₂/pH, glucose and Na⁺, participating in the control of hunger and thirst. This is a route from bodily state to behaviour with no afferent fibre in it at all, and it is why “the interoceptive pathway” in the singular is a simplification.
Humoral receptors. Hormones acting paracrine (locally) and endocrine (at distance). Leptin from adipocytes and intestinal enterocytes suppresses hunger via the hypothalamic arcuate nucleus; ghrelin from gastrointestinal enteroendocrine cells stimulates it; cholecystokinin mediates satiety signalling. Adrenal hormones (epinephrine, cortisol, adrenomedullin) act at both central and peripheral sites. Consequentially, ANS and endocrine receptors sit on immune cells, which is the receptor-level basis for the vagal anti-inflammatory pathway and for sickness-behaviors.
Specialized mechanoreceptors. Baroreceptors are the prototype — stretch-sensitive neurons of the nodose and petrosal ganglia with processes in the aorta and carotid arteries, whose afferent volley triggers the compensatory fall in heart rate, contractility and vascular resistance. The transduction machinery is increasingly attributed to PIEZO1 and PIEZO2 ion channels, expressed in lung, bladder, gastrointestinal tract, sensory ganglia and skin. Berntson & Khalsa expect their role in interoception to be “broader than what has been documented so far.”
Free nerve endings. Aδ and C fibres projecting largely to lamina I, carrying pain, heat and cold. They have some inherent thermal and mechanical sensitivity but are strongly sensitized by biochemicals associated with tissue injury and immune reaction (prostaglandins) — so the same ending reports different things depending on the local inflammatory milieu. Newly added: specialized mechanosensitive Schwann cells forming a mesh around free nerve endings, expressing a PIEZO gene, whose optogenetic activation and inactivation respectively enhance and reduce mechanical pain (Abdo et al. 2019). Glia as part of the transducer, not its support.
The NIH’s version: three signal types, and an admission
Chen et al. (2021) cover the same stage in the same journal issue and reach for a signal-side taxonomy rather than a receptor-side one, sorting interoceptive signals into three kinds:
- Biochemical — from inorganics such as acidic ions through organic molecules to small peptides
- Mechanical — forces that alter structure, such as cellular shape, through stretch or tissue extension
- Thermal and electromagnetic — delivered at various frequencies across the spectrum
They coin interoceptors for the transducers themselves (chemoreceptors, humoral receptors, specialized mechanoreceptors, free nerve endings and nociceptors), and note that interoceptor location largely determines the route: interoceptors on neurons inside the brain receive their signals through non-neural circulatory or lymphatic delivery, while mechanical and thermal interoceptors at peripheral nerve termini feed the sensory ganglia directly. The classical neuroendocrine axes — HNS, HPA, HPG, HPT — are offered as the worked examples of interoceptive communication carried by non-neural systems.
The sentence worth keeping is the concession: “the biochemical identity of most interoceptors largely remains unknown, and only a few specific examples have been described to date.” The funder’s own review makes their systematic identification the first of its Outstanding Questions. That is a strong statement about how provisional this whole stage is — stronger than anything on this page from Berntson & Khalsa, who inventory what is known without foregrounding how little of the space it covers.
Three things this changes
1. The interoceptive/exteroceptive boundary is not drawn by receptor type. (Consolidated at interoception-exteroception-boundary.) Gustatory receptors are prototypic chemoreceptors, and Sherrington counted taste as interoceptive; the field mostly does not. Worse for a tidy line: functional bitter-taste receptors (TAS2Rs) and olfactory receptors are expressed on lung smooth muscle, vascular smooth muscle and cortical neurons, with ligands possibly bacterial (acyl-homoserine lactones, which open airways). A “broad chemoreceptor network responding to both endogenous and exogenous ligands” does not respect the body’s surface. Berntson & Khalsa’s tentative criterion is behavioural rather than anatomical — the swallow marks the transition, since ingestion is a decision to move a substance inside. Compare Petzschner et al.’s state-based criterion and the microbiome case, which strains both.
2. Much of this is recent. PIEZO baroreceptor work, the Schwann-cell pain result, neuropod transduction, gut TAS2Rs — all post-2015. The wiki’s theoretical pages (predictive coding, active inference, EPIC) are built on an afferent anatomy tacitly treated as classical and complete. It is neither, and the additions have not been small.
3. Signals with radically different physics share one label. A millisecond mechanoreceptor volley, a minutes-to-hours hormone gradient, an immune signal over days, and a microbial shift over months are all “interoceptive input.” Petzschner et al. make the timescale point computationally; this page is where it becomes physical. Any account that treats interoception as inference over a single signal type is underdescribing the periphery by a wide margin.
What it does not settle
Nothing here bears on the feedforward-vs-predictive-interoception question directly — a receptor inventory is compatible with either direction of processing. Its relevance is indirect and cautionary: the richer and more heterogeneous the transduction layer, the less plausible that “the interoceptive signal” is one quantity a prior can be weighed against, and the more the precision-weighting story owes an account of which channel is being weighted. See multisensory-integration, which Quigley et al. widen to cover interoceptive × interoceptive pairs for related reasons.