Interoception in anxiety and depression

Study at a glance

• Authors: Martin P. Paulus, Murray B. Stein
• Year: 2010
• Journal: Brain Structure and Function, 214, 451-463
• DOI: 10.1007/s00429-010-0258-9
• Design: Narrative review and theoretical model (no new data). Extends the authors’ 2006 insular model of anxiety with belief-based self-processing, alliesthesia and afferent signal-to-noise; evidence is a selective review of neuroimaging studies.

Key findings

• Proposes that depression and anxiety are ‘not simply interoceptive disorders but are altered interoceptive states as a consequence of noisily amplified self-referential interoceptive predictive belief states’.
• Three extensions of the 2006 model: (1) beliefs about the self (mPFC, temporo-parietal junction) shape how anticipated bodily signals are evaluated; (2) positive Alliesthesia amplifies the valence of initially sub-threshold interoceptive signals and ties them to predicted aversive outcomes; (3) people at risk have a reduced signal-to-noise ratio in interoceptive afferents, so they cannot tell predictive body signals from ongoing fluctuation.
• The insula supplies a ‘body prediction error’ (current minus anticipated body state); anterior cingulate, dorsolateral prefrontal and orbitofrontal cortex then work constantly to separate predictive from non-predictive signals, and this top-down overactivity is experienced as worry, ‘aimed at providing increased prediction accuracy’. The prediction error motivates withdrawal in depression and avoidance in anxiety.
• Makes testable predictions: anxious and depressed people should show a reduced ability to report interoceptive afferents accurately, exaggerated responses to aversive afferents even when they predict nothing, and poorer performance on body-guided tasks such as the Iowa Gambling Task.
• States that the causal order is unknown: feeling states could be altered by exaggerated afferent valence or by exaggerated beliefs, and patients might split into predominantly bottom-up and predominantly top-down subtypes.

Limitations

• No new data. The supporting evidence is insula (and ACC, amygdala) activation during anticipation of aversive events and self-relevant negative processing, mostly with non-interoceptive stimuli (aversive pictures, faces, monetary loss, pain expectation). It shows the insula in aversive anticipation more than it shows altered interoception.
• The central behavioural prediction (reduced interoceptive accuracy in anxiety) was stated without a behavioural test, and the wiki’s later evidence is mixed at best: heartbeat counting accuracy is higher in panic samples, unrelated to anxiety at meta-analytic scale, and respiratory sensitivity is lower with anxiety (see Placement).
• Key terms (belief, noise, prediction error, alliesthesia) are defined verbally, not formally; the formal version came later (Paulus, Feinstein & Khalsa 2019).

Connections

• Concepts: Interoceptive psychopathology, computational psychosomatics, Alliesthesia, Insular cortex, Interoceptive inference, Interoceptive trait prediction error (ITPE), Negative self-referential bias (NSB)
• Debates: Is more interoceptive awareness better?, Does somatic feedback guide decisions?, inferential?
• Related studies: Khalsa et al. (2018), Paulus & Khalsa (2021), Harrison et al. (2021), Desmedt et al. (2022), Wilzok et al. (2023)

The paper behind the sentence Interoceptive psychopathology is built around, cited about 30 times in the wiki before this ingest, mostly through the roadmap. Reading it first-hand confirms the anticipation story and adds two things the second-hand version dropped.

The model

Something happens in the body, say a heartbeat or a breath. In people prone to anxiety or depression, three things go wrong at once:

• The signal is noisy. Reduced signal-to-noise in the afferents makes it hard to tell a meaningful bodily change from ordinary fluctuation.
• It gets amplified. Through positive Alliesthesia, a weak signal takes on strong negative valence (“something is wrong with my heart”).
• Beliefs about the self decide what it means. The same fast heartbeat fits an “excitement” schema or an “I am in danger” schema, and the medial prefrontal self-system picks.

The insula compares the anticipated body state with the current one, producing a body prediction error. Because the signals are noisy, control regions (ACC, dlPFC, OFC) work constantly to separate what predicts something from what does not. That effort is felt as worry. Depression and anxiety are then “altered interoceptive states” produced by noisy, amplified, self-referential predictions.

What the second-hand version dropped

It is not only a prediction account. The wiki’s Martin P. Paulus page summarized the thesis as locating dysfunction in interoceptive prediction rather than sensing. The source puts a sensing problem, low afferent signal-to-noise, at the centre alongside prediction, and says outright that it cannot tell which comes first; it floats bottom-up and top-down subtypes of patients. The later LIBR version (hyperprecise priors, context rigidity) is the top-down half of this.

It predicts reduced accuracy. The first testable prediction is that anxious and depressed people should be worse at reporting their interoceptive signals, while reacting more to aversive ones. It is the same accuracy-down, reactivity-up shape as the Interoceptive trait prediction error (ITPE). The wiki’s evidence on it is split by channel and instrument: heartbeat counting accuracy is higher in panic samples (Is the heartbeat-counting task a valid measure of cardiac perception?), flat against anxiety across 133 studies (Desmedt et al. 2022), while respiratory sensitivity is lower with anxiety (Harrison et al. 2021). So the prediction holds for the one channel with a clean psychophysical measure and fails for the channel with the contested one.

How much evidence it carries

As a framework, a lot of influence; as evidence, little. Its neuroimaging review shows the insula active when people anticipate aversive events or process negative self-relevant material, but most of those stimuli are pictures, faces, money and pain cues, not bodily signals. It is the source of a research programme (the anticipation paradigms that end in Wilzok et al. 2023) rather than a result.

Placement

Held first-hand on Interoceptive psychopathology, Martin P. Paulus (with the sensing-plus-prediction correction), computational psychosomatics and the new concept page Alliesthesia. Its prediction of poorer Iowa Gambling Task performance links it to Does somatic feedback guide decisions?. No hard contradiction; one correction of emphasis on the Paulus page.