Treves et al. (2025) — Dynamic connectivity signatures of breath attention
The wiki’s first first-hand empirical study of the breath-counting task (BCT) with fMRI, and its most direct attempt to find the brain signature of paying attention to the breath in a population that meditation research most wants to treat — high-rumination adolescents. Its lasting contribution is a negative one, and a carefully controlled one: the brain-state markers that cleanly separate breath focus from rest turn out not to be the markers of moment-to-moment on-task attention, and none of them track how attentive a given adolescent is. It is the wiki’s cleanest cautionary tale against the search for a single, universal neural biomarker of focused attention.
Sample and design
72 English-speaking adolescents (54 female, 18 male; M age 15.7, range 13–18) from greater Boston, all with elevated rumination (Children’s Response Style Questionnaire ≥13) and enrolled in an ongoing RCT of an app-delivered mindfulness intervention — the baseline (pre-intervention) scans were analysed here. Diagnoses were common (generalized anxiety 43%, social anxiety 28%, major depressive episode 8%), consistent with a help-seeking, ruminative sample. In a 3T scanner they completed a 20-min breath-counting task (Levinson et al. 2014) and three resting-state runs. During the BCT, participants counted each breath with one button and pressed a second button to mark the end of every nine-breath cycle; a third “reset” button acknowledged a lost count. Breath-belt recordings let the authors (i) correct physiological noise with RETROICOR and (ii) define on-task periods (a cycle with 8 button presses and 8–10 breath peaks) versus off-task periods (everything else — miscounts and resets).
Static connectivity: breath counting vs rest
Compared to rest, breath counting showed elevated salience↔executive (SN–CEN) and within-SN connectivity, and, whole-brain, higher within-network connectivity with weaker cognitive-control↔subcortical and more negative cognitive-control↔DMN coupling. So the coarse static picture is the expected one — focused attention engages the control/salience machinery and pushes the DMN into anticorrelation. Static FC did not correlate with any self-report measure.
Four dynamic brain states
Sliding-window (42-TR ≈ 30 s) k-means clustering over the targeted salience/executive/DMN ROIs yielded four recurring brain states (Allen et al. 2014 dynamic-FC method):
| state | character | more prevalent in |
|---|---|---|
| 1 | hypoconnected (low connectivity overall) | rest |
| 2 | DMN-anticorrelated (DMN negatively coupled to the rest) | breath counting |
| 3 | SN-reduced (weak salience-network correlations) | — |
| 4 | hyperconnected (high connectivity overall) | breath counting |
States 1, 2 and 4 replicate those Mooneyham et al. (2017) found around a breath-meditation exercise; the DMN-anticorrelated state’s increase during breath counting matches Mooneyham’s finding that the same state rose after a mindfulness intervention. Dwell times were longer during breath counting for states 2 and 4, and there were proportionally more switches between states during the task than at rest — a robust dynamic difference between focused attention and rest.
The disconnect: task-vs-rest states ≠ on-task-vs-off-task states
The paper’s pivot, and the reason it earns a page rather than a citation. Using the breath-belt-derived on-task/off-task labels and mixed-effects logistic regression at the hemodynamic lag, the states that predicted within-task attention were not the states that separated task from rest:
- State 1 (hypoconnected) predicted on-task > off-task (OR=1.23) — even though it was the state more prevalent at rest. The authors flag this as unexpected and float a “flow” reading (high engagement, low error-monitoring).
- State 4 (hyperconnected) predicted off-task (OR=0.64) — the reverse of its task-vs-rest role.
- State 3 (SN-reduced) weakly predicted on-task (OR=1.04).
- State 2 (the DMN-anticorrelated state) — the single state that best distinguished breath counting from rest — showed no relationship to on-task vs off-task attention (OR=1.02, ns).
So the marker of “doing the breath task” and the marker of “currently attending to the breath” come apart. This directly qualifies the tidy story the default-mode-network and salience-network pages tell from Weng, Barrett and Farb: DMN anticorrelation rises with focused-attention practice, but here it does not index the moments of focus within that practice.
No individual-difference signal
Across participants, no brain-state measure related to behavioral accuracy, trait mindful attention (FFMQ “Acting with Awareness”), or ecological-momentary-assessment mindful attention. The authors read this together with the within-task result as a genuine caution: “there is a disconnect between (i) the states that we found to be more present during the breath counting task than rest and (ii) states that are related to within-task breath counting performance,” and it “remains unclear how these dynamic brain states relate to periods of attention and inattention.” This echoes Weng et al. (2020)‘s finding that attention-related activation patterns were highly distinct between individuals, and Kucyi et al. (2024)‘s that mind-wandering connectivity markers varied substantially by person. The paper’s constructive proposal is a turn to within-individual imaging — identifying each person’s own attention states and using them for individualized neurofeedback — rather than one biomarker for everyone.
A novel whole-brain state
In the whole-brain (53-network Neuromark) analysis, breath counting uniquely elevated a state with cognitive-control↔somatomotor correlations. The authors read this speculatively as inward attention to bodily sensations — attention to the breath coupling control and somatomotor cortex where thought and somatosensation usually decouple (the Schooler/Kucyi perceptual-decoupling account of mind-wandering). This is the study’s one positive, interoception-flavoured hint, and it is offered as a conjecture.
Why it matters to the wiki
This is the empirical, first-hand attention study the mind-wandering page noted was missing — though it enters as on-task/off-task attention during breath counting, not classical probe-caught mind-wandering. It supplies the wiki’s method pages on the breath-counting-task and dynamic-functional-connectivity, and it plants a firm caution beside the contemplative-neuroscience optimism elsewhere: the network changes that accompany breath-focused practice are real and replicable at the group level, yet on this evidence they neither mark the instant of attention nor distinguish the attentive adolescent from the distractible one. Held as a well-controlled null on universal biomarkers, from within the Gabrieli/Webb mindfulness-imaging programme rather than from its critics.