NOTE-761: Theriault et al. — Allostasis at the core of brain function

Contribution

The paper gathers the Barrett group’s anatomical, imaging and physiological work into one argument for an “allostasis-first” view: the organ’s job is to coordinate and regulate the body’s systems, and mental activity is “a means to the allostatic end”. It adds an applied example: a hypothesis that falling glucose uptake before cognitive decline in Alzheimer’s disease protects waste clearance and pH at the cost of cognition. It offers a definition of stress as anticipated metabolic outlay, and a position on whether allostasis can be measured (Box 3).

Key insight

Read the brain’s anatomy before its psychology. The regions that issue visceromotor commands and receive interoceptive signals form one connected system, sit at the top of the cortex’s laminar gradient, and include the hubs that imaging studies have labelled with psychological functions. On that reading, the default-mode and salience networks are the allostatic system under other names, and thought, feeling and perception are ways of meeting the body’s predicted needs.

Assumptions

• Efficient metabolic regulation is “a fundamental selection pressure”, so an allostatic architecture should be “evident in the core of the brain”. The argument is from expected design to found anatomy.
• Predictive processing as the reading of cortical signal flow. Centrifugal signals are predictions, centripetal ones prediction errors. The laminar anatomy is cited; the computational reading is hypothesis.
• “Resting-state” BOLD largely reflects interoceptive signalling, with exteroceptive input reduced, not psychological activity such as self-referential thought. This is asserted from cited work.
• Allostasis is not a measurable quantity. “It is not reducible to any single measure or small set of measures”; it is “a conceptual tool to frame more focused, tractable questions” (Box 3).
• The scope is stated as modest. “Our goal in this paper was not to claim our ‘allostasis-first’ hypothesis is correct, but more modestly to justify asking scientific questions about the links between brain, body, and mind in a different way” (Summary and Conclusion).

Key results

All results are reviewed, mostly from the group’s earlier papers.

• An intrinsic allostatic system (Figs. 1–2). 21 seeds: 14 allostatically relevant subcortical regions and the 7 cortical regions receiving their monosynaptic projections in non-human tract-tracing (aMCC, pACC, sgACC, ventral anterior insula, dorsal mid and posterior insula, hippocampus). Resting-state connectivity at 7 T (N = 90), replicating 3 T (N = 550), “replicated 100% of the monosynaptic cortico-cortical connections, 94% of the monosynaptic subcortico-cortical connections, and 96% of the monosynaptic subcortico-subcortical connections”. The system extends to regions not seeded, including primary motor cortex and cerebellum.
• Overlap with named networks (Fig. 2). The system overlaps the central autonomic network, the skeletomotor network, the salience/cingulo-opercular network and a heat-pain signature. k-means on the 21 seed maps gives two motifs, “default-mode-like” and “salience/cingulo-opercular-like”, which overlap in the connectome’s “rich club” hubs. Hence: “Scientists have been studying the allostatic system all along but under different names”, and efforts to assign psychological functions to these networks “are destined to fail”.
• The limbic-to-sensory gradient (Fig. 3). From sensory to limbic cortex, layer 2/3 pyramidal neurons become larger, sparser and more connected, and layer 4 thins to agranular. Read as dimensionality reduction: sensory signals are “increasingly compressed into low dimensional, multimodal summaries” towards the limbic edge, and limbic signals decompress outwards. Predictions flow from deep layers of less laminated to superficial layers of more laminated cortex, and errors the reverse. NEFH, PV and SST expression follow the gradient. Interoception, olfaction and taste arrive already compressed, which places them “higher on the predictive hierarchy than the distance senses”.
• Two-way centrifugal flow (Fig. 4). Limbic control signals go down to the vagus and spinal cord as visceromotor commands and out to sensory cortex as their “efferent copies”, which act as sensory predictions. The ACC projects directly to V1.
• Interoception shapes brain-wide signalling. Respiration couples oscillations across many regions and modulates hippocampal sharp-wave ripples; vagal stimulation enhances memory consolidation and mood; people saccade more in systole and fixate more in diastole, and visual search is slower and less accurate in systole. Visceral and heartbeat-evoked potentials track detection and attention (Box 2).
• Stress redefined (Box 1). “Stress is a brain-body state associated with an anticipated allostatic perturbation involving a significant increase in energy utilization”, value-neutral. It is felt as stressful when predicted and actual outlay mismatch, or the system is depleted. The box restates that “affective experience, or mood, functions as a low-dimensional allostatic barometer”, and that an illness disrupting the brain’s model of the body “will involve affective symptoms … [and] some dysfunction in energy regulation”.
• Alzheimer’s as allostatic trade-off. Anaerobic glucose metabolism is ~15× less glucose-efficient and acidifies tissue; as vascular waste clearance declines with age and amyloid, lowering glucose uptake may protect pH. Cited support: GLUT1 and GLUT3 density and glucose uptake “decline following amyloid accumulation, but before the appearance of cognitive decline”. Hence glucose-restoring therapies might rescue cognition while “worsening the more fundamental disease process”. The brain’s “core function is not to think, or to feel, or even to sense”.
• Concepts and mental causation (Box 3). Concepts are “emergent, lower dimensional features that describe and causally constrain” the allostatic network; all animals have concepts, and “concepts do not necessarily need to be conscious”. Psychological interventions act on the body through them (CBT and cardiovascular risk, conditioned immune responses, emotion granularity and cardiorespiratory differentiation).

Claims

id	claim	strength	support
C1	Visceromotor and interoceptive regions form one intrinsic system whose connectivity reproduces tract-traced anatomy	moderate	the group’s 3 T and 7 T studies, reported; not read here
C2	That system contains the hubs of the default-mode and salience networks, so findings labelled by psychological function are also findings about bodily regulation	moderate for the overlap, weak for the reinterpretation	Fig. 2 overlap maps; the inference is the paper’s
C3	The limbic-to-sensory laminar gradient carries predictions outward and compressed errors inward, with allostasis at the top	weak	cited anatomy; the computational reading is hypothesis
C4	Interoceptive rhythms modulate brain-wide signalling and sensory sampling	moderate	cited reviews and studies (respiration, cardiac phase)
C5	Affect or mood is a low-dimensional summary of allostatic state	weak	restated from the group’s earlier papers, Box 1
C6	Cognitive decline in Alzheimer’s disease may be an allostatic trade-off protecting waste clearance and pH	weak	a hypothesis, supported by the timing of GLUT changes
C7	The brain’s primary function is allostasis, and psychological phenomena are means to it	weak	the synthesis; the authors disclaim claiming it is correct

Concepts

• Allostasis. Predictive regulation that balances the demands of many bodily systems, “stability through change”; distinguished from homeostasis, reactive return to a set point.
• Interoception. The brain’s predictive modelling of the body’s sensory conditions, part of allostasis because it supports visceromotor control as somatosensation supports skeletomotor control.
• Allostatic system. The 21-seed network and the regions intrinsically connected with it.
• Centrifugal / centripetal signals. Limbic-to-sensory (prediction, decompression) / sensory-to-limbic (prediction error, compression).
• Stress. Anticipated large energetic outlay, value-neutral.
• Allostatic load. Rejected as “an unhelpful abstraction”.

Connections

It extends Barrett’s theory of constructed emotion (LIT-542, read in NOTE-434), whose agranular-to-granular prediction scheme (C3 there) is here generalised from emotion to all brain function, and Sterling’s allostasis, quoted at the outset. It sits with Seth’s interoceptive “beast machine” (LIT-135) and invites dialogue with enactivism and autopoiesis (Box 3). Kleckner et al. (2017), the 3 T source of C1, is cited in NOTE-434 as well.

Bearing on the record

• QUESTION-059 (Deferred). The LIT was filed as “the bodily-mechanism option: a feature a functional organization might lack”. The reading does not support that use.
  • The paper is about what the brain is for, not about what makes a state felt. Its only claims about feeling are restatements: affect as “a low-dimensional allostatic barometer” (Box 1) and stress becoming “psychologically ‘stressful’” on mismatch.
  • It does not say an organization lacks the mechanism, or that feeling needs a body. Its terms are functional: prediction, regulation, compression. A functionalist could take every claim in it as a description of a role.
  • Its one remark on consciousness is that matrix thalamic inputs to upper cortical layers are “considered to be causally important to levels of consciousness”, cited and not developed. It is relevant only as the newest statement of the allostatic frame in which THEORY-045’s Barrett half is set.
• THEORY-045. Consistent, and adds nothing to the felt-affect claim. It strengthens the “prediction” side of THEORY-045’s open “prediction or map” question only by restating it. It also puts allostatic generation in limbic cortex and brainstem nuclei together (Fig. 1), which neither settles nor sharpens the disagreement THEORY-045 records with Damasio’s brainstem-only generation.
• THEORY-057. Box 3’s account of concepts as low-dimensional constraints on allostasis, and the claim that networks were named for psychological functions they do not specifically serve, are the constructionist view extended from emotion to all mental categories.
• THEORY-093, THEORY-090. Its subcortical list (PAG, PBN, NTS, hypothalamus, monoamine nuclei) overlaps the substrates those accounts name, but the paper makes no claim about where feeling arises.
• THEORY-101. Veit’s “coordinated regulatory state” and this paper’s allostasis are close in content; the paper does not identify feeling with regulation, so it is not a second source for THEORY-101.
• No instruction for machine-learning practice; nothing for the anthology. The Alzheimer’s section is a hypothesis for clinical research, not a practice recommendation for ML.

Limitations

• A review, mostly of the authors’ own work. The decisive evidence for C1 and C2 is in cited papers not read here.
• The central thesis is disclaimed. The authors do not claim “allostasis-first” is correct, and give no result that would refute it. Allostasis is said to be unmeasurable by any single variable.
• Overlap is not function. That the allostatic system overlaps the default-mode and salience networks shows shared topography; that the networks’ psychological labels are “destined to fail” goes beyond it.
• The Alzheimer’s account is speculative. It rests on the order of GLUT changes and cognitive decline, and on mouse work, and predicts harm from glucose-restoring therapies without data.
• No claim about experience. Nothing in it bears on whether, or why, allostatic states are felt.

Open questions

• What finding would count against allostasis-first? A brain region or network whose activity is reliably dissociated from any bodily regulatory demand while carrying a psychological function would.
• Does the default-mode network’s activity in tasks track interoceptive or metabolic variables once task demands are matched?
• Does lowering glucose uptake in early Alzheimer’s disease protect pH and clearance, and do glucose-restoring therapies worsen long-run outcomes as predicted?