Default mode network

The wiki’s third intrinsic network, and the last to get a page — it had appeared only as the thing the right anterior insula switches away from in Craig (2009). Barrett (2017) gives it a central role.

Barrett’s hypothesis: the DMN is the internal model

I hypothesize, as others do (Mesulam, 2002; Hassabis and Maguire, 2009; Buckner, 2012), that the default mode network is necessary for the brain’s internal model.

Whatever else gets mapped to DMN activity — mind-wandering, memory, prospection, self-reference, semantics — the simulations initiated there “cascade to create concepts that eventually categorize sensory inputs and guide movements in the service of allostasis.”

Barrett is careful about what this does and does not say, and the care is the interesting part. Two disclaimers:

• Concepts are not stored in the DMN. “I’m saying that the default mode network represents efficient, multimodal summaries, from which a cascade of predictions issues through the entire cortical sheet, terminating in primary sensory and motor regions. The whole cascade is an instance of a concept.” The DMN holds part of the pattern, not the pattern.
• This does not make the DMN an emotion network. Domain-general networks “can be mapped to many psychological categories at the same time.” That is the point of them.

The hypothesis is “partially consistent” with the semantic-network reading of the DMN (Binder et al., 2009): Barrett’s addition is that simulations are “more than just multimodal sensorimotor summaries; they are fully embodied brain states,” which they become as the summaries cascade out and particularize.

The species-general claim, and the footnote defending it

Worth recording because it shows the shape of the theory. Rats have a DMN and, as far as anyone can tell, cannot engage in mental time travel (Hsu et al., 2016). If the DMN were for mental time travel, that would be awkward. Barrett’s footnote says it “in no way disconfirms the hypothesis that the network is running an internal model of the animal’s world in the service of allostasis.”

That is a reasonable reply and also a demonstration of the falsifiability worry recorded on theory-of-constructed-emotion: the network’s function is specified at a level general enough that a species lacking the headline capacity is not a counterexample.

Relation to the wiki’s other networks

Barrett’s division of labour across the three intrinsic networks:

network	job
default mode	hosts the internal model; initiates the prediction cascade
salience	precision — decides which prediction errors are worth encoding
frontoparietal control	sustains simulations over longer timescales; suppresses low-prior ones

She reports (Table 2) that DMN–salience interconnectivity tracks the intensity of emotional experience, as distinct from arousal (Raz et al., 2016) — one of the few entries in her supporting table that is a direct positive prediction rather than a reinterpreted null.

Figure 6 of the source presents the two networks as a single large-scale system for allostasis and interoception, interconnected by rich-club hubs (ventral anterior insula, midcingulate, parahippocampal gyrus, postcentral gyrus, cuneus) and reaching subcortically to PAG, PBN, NTS, ventral striatum, hypothalamus and thalamus. Mapped in 280 participants and replicated in a further 270 — though the anatomy is credited to Kleckner et al., which was “in press” at the time.

The DMN as narrative self-reference, and a way to switch off it (Farb et al. 2007)

The clinical readings below are foreshadowed by Farb et al. (2007), the earliest Farb source on this page and the one that first ties the DMN’s midline machinery to narrative self-reference. Reading trait adjectives in a narrative-focus mode (“what does this say about me”) recruited the mPFC/PCC midline exactly as the resting/mind-wandering DMN does — the paper cites Gusnard, Raichle and Mason to make the point that narrative self-reference is the default mode. Its contribution is showing this default is not obligatory: an experiential focus, especially after mindfulness training, deactivates the mPFC and shifts toward a right viscerosomatic network, decoupling the insula from the vmPFC self-model. So the DMN’s self-referential mode has, on this account, a trainable alternative — the seed of the “disengage the DMN” clinical aim that Weng et al. later make an explicit treatment target below.

The DMN as a clinical predictor: self-reference that will not disengage

Farb et al. (2011) give the DMN’s self-referential role a prospective clinical stake. In remitted depressed patients watching a sad film, reactivity in medial prefrontal cortex (BA 24/32 — a core DMN hub) predicted relapse over 18 months (r = .68) and tracked trait rumination, while sensory (visual) reactivity protected against relapse and tracked acceptance. The signature of risk was specifically a failure to disengage self-referential processing: dorsal and ventral mPFC, normally anticorrelated, became positively coupled in the patients who relapsed. So the DMN here is not merely “the internal model” in the abstract (Barrett) — it is the substrate of the ruminative, self-evaluative mode whose over-engagement forecasts a return of depression (see cognitive-reactivity). The wiki thus meets the DMN twice: as Barrett’s body-budgeting internal model, and as the clinical site of pathogenic self-focus — the same anatomy, read once as generative machinery and once as a mode that can get stuck on the self.

The earlier Farb et al. (2010) paper shows the other half of this: that the midline/self-referential recruitment is not fixed but trainable down. Untrained controls responding to a sad film activated the midline network (vmPFC, dmPFC, posterior cingulate/precuneus) plus left-language areas — the elaborative mode — while MBSR completers, at equal felt sadness, activated it less and kept the interoceptive insula online instead. Read together, the two Farb papers frame the DMN’s sadness-evoked self-reference as both prospectively pathogenic (2011: mPFC reactivity → relapse) and plastic (2010: mindfulness lowers it), which is the clinical bet MBCT is built on.

The mentalizing overlap, and a state that activates it to stimuli (Lutz et al. 2008)

Lutz et al. (2008) add a third clinical-to-contemplative angle, and it cuts across the usual story that the DMN deactivates during tasks. The “mentation” network they found recruited by compassion meditation — TPJ, pSTS, mPFC, precuneus/PCC, the machinery for reading others’ mental states — “overlap[s] with the proposed ‘default mode’ or ‘resting state’ networks,” as they note explicitly. What is unusual is the direction: experts, in the compassionate state, showed these regions activating to emotional sounds where at rest they deactivated to them — a state-dependent reversal the paper flags as the adepts’ ability to selectively produce DMN deactivation (rest) and activation (meditation) in precuneus and TPJ.

So the wiki now meets the DMN three ways: Barrett’s body-budgeting internal model, Farb’s pathogenic/plastic self-focus, and here as the substrate of mentalizing that a cultivated positive state can drive up rather than suppress. The through-line is that these midline/parietal regions are not simply “task-off” machinery — they host self- and other-modelling, and a contemplative state directed at others turns that modelling toward the emotional sounds instead of away from them.

Frontal DMN self-reference: associated with relapse, but not the predictor, and its connectivity is inert (Wu et al. 2026)

Wu et al. (2026) test the DMN’s self-referential role as a preregistered relapse hypothesis, and the result is a partial miss worth recording precisely because the hypothesis is the standard one. Deriving a frontal DMN ROI (dmPFC) from the dorsal-nexus account (Sheline et al. 2010), they measured dysphoric self-referential activity (Self-Negative minus Self-Positive) during a self-referential encoding task in remitted patients:

• Frontal DMN dysphoric activity was associated with relapse status (b=0.32 [0.09, 0.55]), relapsers showed higher baseline dysphoric DMN activity, and a Relapse×Time interaction reached significance for the frontal DMN only (not the salience ROI).
• But frontal DMN dysphoric activity did not independently predict relapse timing in Cox survival models, and DMN–salience connectivity — plus within-DMN connectivity — was unrelated to relapse throughout.

This is a direct, negative test of the model on which this whole page’s clinical readings rest. The dorsal-nexus story (Sheline; Bertocci et al. 2023) holds that depression vulnerability lives in heightened DMN–salience coupling; Wu et al. find the coupling inert and conclude “elevated prefrontal activation during self-referential processing may be more critical for relapse vulnerability than heightened inter-network coupling per se.” The DMN’s self-referential activation marks a relapser (consistent with Farb 2011’s mPFC-reactivity finding), but on this cohort it is not the robust prospective predictor — the whole-brain somatosensory and subgenual markers are (see two-factor-model-of-relapse-vulnerability). A caution for the “DMN hyperconnectivity = depression risk” generalization, from within the Farb corpus that most relies on the DMN-as-pathogenic reading.

A DMN hub that is also body-mapped (Zeharia et al. 2019)

One of the midline/parietal regions the DMN keeps recruiting — the precuneus (with PCC) — turns out to carry a somatotopic body map. Zeharia, Hofstetter et al. (2019), mapping 20 moved body parts with phase-encoded fMRI, find a whole-body toes-to-tongue gradient across the precuneus, confirming Penfield & Jasper’s neglected 1954 prediction. This is not a DMN result per se — the somatotopy is sensorimotor, and the anterior precuneus is the most sensorimotor-connected third (Margulies et al. 2009). But it complicates any picture of the precuneus as purely an abstract self-model: part of the “self-processing hub” is a literal map of the body surface. The authors close by proposing the precuneus homunculus and the same group’s insula homunculus (Zeharia et al. 2012) as a putative node “linking body and mind,” coupling self-referential (DMN) and body-related representation — the Farb / Lutz mind-body thread reached from the somatomotor side. Recorded as the authors’ conjecture, not a demonstrated interoceptive function.

Disengaging it as a stated clinical aim (Weng et al. 2021)

The three readings above are descriptive. Weng et al. (2021) make DMN disengagement an explicit treatment target: breath-focused meditation decreases activation in anterior mPFC, posterior cingulate and posterior inferior parietal lobule, and the authors state that “disengaging from self-related thought is particularly important from a clinical perspective due to the cognitive and emotional flexibility it cultivates.” Their proposed mechanism for meditation is a trade — engage cognitive-control and interoception networks, disengage the DMN — with the salience-network switching between them.

This lands neatly on the Farb material above and should be read against it. Farb (2011) makes DMN reactivity prospectively pathogenic; Farb (2010) makes it trainable down; Weng et al. now make training it down the point of the intervention. The wiki’s caution stays attached: the 2022 RCT found the prefrontal factor moved equally under a therapy that trains body awareness and one that does not, so “meditation lowers DMN engagement” is a mechanism claim that has not yet been shown to be specific to the interoceptive route.

Worth also holding beside the Lutz result one section up, which is the awkward case for any “meditation deactivates the DMN” generalization: compassion meditation drove precuneus/TPJ mentalizing regions up to emotional sounds. Different practice, opposite direction — a reason to specify which contemplative practice is meant before making network claims about “meditation.”

The kind of self-generated thought that is not the DMN (Banellis et al. 2026)

Every section above inherits the field’s assumption that self-generated thought is DMN business — Barrett’s internal model, Farb’s ruminative self-focus, Lutz’s mentalizing, Weng’s disengagement target. Banellis et al. (2026) look for the intrinsic connectivity supporting body-oriented spontaneous thought in 489 participants and do not find the DMN.

What they find instead, via cross-validated CCA over 216 parcels (out-of-sample r = 0.354): thalamus and striatum coupling to primary somatosensory and motor cortex, right thalamus the dominant hub (53 high-loading edges), with insula, cingulate and paracingulate — the interoceptive-allostatic regions — recruited alongside. The mode’s topography aligns with Margulies’ Gradient 1, the unimodal-to-transmodal axis, with poles at somatomotor cortex and at transmodal frontal/cingulate cortex.

Two things follow for this page.

Mind-wandering is not one network’s output. The paper’s own framing is that canonical accounts emphasize DMN and frontoparietal control, and that body-wandering “differs from” that. Their mental-time-travel factor — the classical past/future mind-wandering dimension — showed negligible connectivity overlap with the body-wandering substrate (r = 0.061 against r = 0.431 for the body factor). Two dimensions of the same resting stream, two substrates.

The “decoupling” story does not cover it. The DMN literature treats mind-wandering as perceptual decoupling — attention withdrawn from sensory input (Smallwood et al. 2011, pupillometry). Body-wandering runs the other way: it engages sensorimotor and subcortical circuitry during internal experience. Whatever the DMN is doing when the mind drifts, it is not the whole of what drifting is.

A caution against overreading this. Banellis et al. did not test the DMN and find it absent; they ran an unconstrained whole-brain CCA on body-item-loaded thought and the DMN did not surface at the top 1% of loadings. That is weaker than a null result about the DMN, and it is one multivariate mode in one cohort. What it licenses is the narrower claim that this wiki’s habit of reading “self-generated thought” as “DMN” is at least incomplete.

The DMN unregulated: stereotypy in frontotemporal dementia (Farb et al. 2012)

The clinical readings above are about the DMN being over-engaged by self-focus (Farb 2011) or a target for disengagement (Weng). Farb et al. (2012) show a third clinical face: the DMN running unregulated because its brake has degenerated. In behavioural-variant frontotemporal dementia, as the salience network dissolves and stops switching attention in and out of the default mode, the paper found elevated DMN connectivity in the right angular gyrus that tracked stereotypy (and apathy) — rigid, repetitive behaviour read as the DMN’s habit/self-narrative machinery freed from the salience network’s contextual veto: “abnormally heightened DMN connectivity may in turn promote maladaptive behaviors driven by self-generated narratives and habit rather than what is emotionally salient and appropriate to environmental context.”

This is the pathological endpoint of the salience↔DMN antagonism the wiki records under Barrett and Weng. Where those sources describe the salience network deciding when the DMN should be engaged, Farb (2012) shows the behavioural cost when that decision can no longer be made — the angular-gyrus DMN node hyperconnected, and stereotyped behaviour as the output. It also seats the DMN in the same “prefrontal isolation” story as the anterior insula: both the salience apex and the DMN are dysregulated in FTD, in opposite directions (insula disconnected, DMN over-connected), consistent with the loss of the switching relationship between them.

DMN anticorrelation rises with breath focus — but does not mark the moment of attention (Treves et al. 2025)

Every clinical and contemplative reading above leans on a tidy inference: focused-attention practice pushes the DMN into anticorrelation, and that anticorrelation is the good, on-task signature. Treves et al. (2025) test the second half of that inference with dynamic connectivity during a breath-counting task in 72 high-rumination adolescents, and it does not hold.

The first half survives: a DMN-anticorrelated brain state was more prevalent, dwelt in longer, and switched to more often during breath counting than at rest — replicating Mooneyham et al. (2017), whose same state rose after a mindfulness intervention. But when the authors used breath-belt physiology to label on-task versus off-task periods within the task, the DMN-anticorrelated state showed no relationship to whether the adolescent was currently attending (OR=1.02, ns). The states that did predict within-task attention were a hypoconnected state (on-task) and a hyperconnected state (off-task) — neither of which was the state that separated task from rest. And no brain state related to any individual difference in attention (accuracy, trait FFMQ, EMA).

So DMN anticorrelation, on this evidence, is a marker of being in the practice, not of the instant of focus and not of who is more attentive. This is a boundary condition on the whole “disengage the DMN” program the Weng and Farb sections build: the target moves at the group/condition level as advertised, yet it does not index moment-to-moment or between-person attention here. The authors’ constructive reading — that attention’s neural signature may be too person-specific for a single biomarker, echoing Weng et al. (2020)‘s highly individual patterns — is recorded on the study page.

Why this page matters to the wiki’s convergence story

Note what has happened to the region set. The wiki records a three-way convergence on nearly the same regions from independent starting points — the salience-network (functional connectivity, Seeley), the core affect network (meta-analytic, Lindquist), and the VMAs (cytoarchitectural, Seth & Friston).

Barrett’s Figure 6 adds the DMN to that set and reframes the whole thing: these are not three descriptions of an emotion system that happen to coincide. They are one system, and it is not for emotion — it is for budgeting the body. The convergence the wiki had been treating as a striking coincidence between literatures is, on this account, what you would expect, because all three were measuring the allostasis/interoception system and calling it emotion.