Wu et al. (2026) — Functional brain biomarkers of self-referential bias in depression relapse
The fourth paper in the wiki’s Farb depression-relapse arc, and the one that pushes the two-factor model past its prefrontal-centric framing. Where Farb et al. (2022) scanned relapse vulnerability during a sad-vs-neutral film (dysphoric mood induction), this paper scans the same trial cohort (NCT01178424, OSF s4n8j) during a Self-Referential Encoding Task — asking not how a recovered depressive reacts to sadness, but how they process self-relevant information, and which part of that processing forecasts relapse. First-authored by Liliana C. Wu with Segal and Farb; ingested from raw/papers/Farb/.
Design
Remitted, recurrent-MDD outpatients (parent trial n=155; neuroimaging subsample n=99) were randomized to MBCT or Well-Being Cognitive Therapy (WB-CT; Fava) and scanned before and after eight weeks of prophylactic psychotherapy, then followed for two years (primary outcome: time to DSM-IV MDE by SCID + HRSD>=16). A final analytic sample of 81 completed both scans with adequate task variance; 27 relapsed, 54 did not. In the scanner they completed the SRET (Derry & Kuiper 1981) — judging whether positive and negative trait adjectives described themselves (self), another person (other), or were upper/lowercase (case, a non-evaluative perceptual baseline) — with a validated dysphoric mood induction interleaved, the same induction used in 2010/2011/2022. Negative self-referential bias (NSB) was the behavioural index: the ratio of negative to positive endorsement in the self condition. Preregistered on OSF (s4n8j); analyses controlled for antidepressant medication, past episodes, and residual symptoms.
The design reuses the 2022 static/dynamic logic: reactivity common to both scans estimates a static (trait) marker; reactivity change across scans estimates a dynamic (treatment-responsive) marker. The novelty is the probe — self-referential encoding rather than sad-film reactivity — and a set of preregistered frontal DMN / salience-network ROIs derived from the dorsal-nexus account of depression (Sheline et al. 2010).
What the behavioural bias does — and does not — predict
The first result is a dissociation the wiki should hold onto. NSB predicted concurrent symptoms but not relapse. More negatively biased self-endorsement went with higher residual depressive symptoms (b=0.10 [0.06, 0.14], p<.001) and, longitudinally, with less decentering — but it had no direct association with relapse status. Behavioural self-referential bias, the paper argues, “may primarily reflect a conscious, reportable tendency that tracks concurrent symptom burden without necessarily driving episode recurrence.” The thing that forecasts relapse is neural, not the endorsement pattern the patient can report.
The frontal DMN/SN ROIs: associated with relapse, but not predictive of it
The preregistered hypothesis was dorsal-nexus: that relapse would be predicted by greater dysphoric self-referential activity in frontal DMN (dmPFC) and salience network (ACC), and by stronger DMN–SN connectivity. The result is a partial, instructive miss:
- Dysphoric NSB activity in both frontal DMN (b=0.32 [0.09, 0.55]) and frontal SN (b=0.36 [0.09, 0.63]) was associated with relapse status, and relapsers showed higher baseline dysphoric activity in both; a Relapse×Time interaction reached significance for frontal DMN only.
- But neither network’s dysphoric activity independently predicted relapse timing in Cox survival models, and DMN–SN connectivity was unrelated to relapse — as were within-DMN and within-SN connectivity. Contrary to prominent models proposing that depression vulnerability lives in heightened inter-network coupling (Bertocci et al. 2023; Sheline et al. 2010), “elevated prefrontal activation during self-referential processing may be more critical for relapse vulnerability than heightened inter-network coupling per se.”
So the frontal machinery marks a relapser but does not robustly forecast the episode on its own — which is exactly the gap the whole-brain analysis fills from the sensory side.
The whole-brain markers: prefrontal risk, somatosensory protection
Beyond the a priori ROIs, whole-brain static and dynamic analyses (Self>Case) produced the paper’s headline three regions:
| marker | region | direction in relapsers | single-marker HR |
|---|---|---|---|
| static | left subgenual cingulate | greater activation | HR 1.81 [1.20, 2.72], p<.004 |
| static | right lateral occipital cortex | greater activation | HR 2.07 [1.12, 3.84], p<.02 |
| static | right postcentral gyrus / S1 | greater deactivation (activation protective) | HR 0.26 [0.12, 0.58], p<.001 |
| dynamic | left supramarginal gyrus | attenuated deactivation over treatment in non-relapsers | (protective; see combined model) |
The subgenual cingulate is the region “consistently implicated in persistent depressive risk and targeted in therapeutic interventions” (Drevets, Mayberg); right lateral occipital cortex is a more recent MDD site of perceptual disruption, and its baseline reactivity also predicted worsening residual symptoms over treatment (b=0.28, p=.02). The somatosensory marker is the one that ties this paper to 2022: greater postcentral deactivation during self-referential processing marks relapse, and greater postcentral activation protects.
The combined model: somatosensory markers win
The paper’s central claim rests on the combined Cox model incorporating both static and dynamic ROIs and adjusting for ADM, past episodes, and baseline + post-treatment residual symptoms (Fig. 4). Of every neural and clinical predictor entered, only two survived as indicators of reduced relapse risk:
- right postcentral gyrus (static, somatosensory): HR 0.16 [0.049, 0.52], p=.002
- left supramarginal gyrus (dynamic, somatosensory-associative): HR 0.12 [0.029, 0.51], p=.004
The subgenual/prefrontal marker, predictive on its own, dropped out once the somatosensory markers were included — meaning “somatosensory markers may capture unique protective variance in this sample that is not fully accounted for by prefrontal indices alone.” Higher right postcentral activation and greater treatment-related increases in left supramarginal activation predicted lower relapse likelihood even after clinical covariates. Concordance C = 0.83.
The valence split: elaboration vs embodiment
A conjunction analysis separated the two poles of self-referential valence, and they land on the wiki’s familiar axis. Self-criticism (activation to negative endorsement + deactivation to positive) recruited a broad prefrontal/posterior elaborative set — orbitofrontal/insula, cerebellum, left DLPFC (a CEN node, giving exploratory support to the dorsal-nexus story), left PCC, left lateral occipital. Self-affirmation recruited posterior somatosensory regions — right postcentral, right inferior temporal, right lingual, right posterior supramarginal, left precuneus, left superior parietal lobule. This is the narrative-vs-experiential dyad reappearing inside self-referential valence: negative self-evaluation as prefrontal elaboration, positive self-relation as somatosensory embodiment.
How it extends the two-factor model
Read against Farb (2022), this is continuity with a shift of emphasis — not a contradiction (see two-factor-model-of-relapse-vulnerability):
- The static somatosensory-deactivation vulnerability replicates. Greater somatosensory deactivation marking relapse is “a replicated finding in this cohort during dysphoric mood induction (Farb et al., 2022)” — now shown during self-referential processing too, in the same patients.
- The treatment-responsive (dynamic) factor is here on the sensory side. In 2022 the dynamic, treatment-moved factor was prefrontal reactivity (left-DLPFC reactivity that fell in non-relapsers). Here the protective dynamic marker is somatosensory — left supramarginal activation increasing over treatment in non-relapsers. The treatment-responsive pole has moved from turning prefrontal elaboration down to bringing somatosensory embodiment up.
- The combined model inverts the 2022 hierarchy. In 2022, static somatosensory + dynamic prefrontal markers both survived jointly. Here, when both are entered, the somatosensory (static + dynamic) markers dominate and the prefrontal marker drops out. The paper’s conclusion: relapse vulnerability reflects “not only heightened prefrontal negative self-processing but also insufficient recruitment of somatosensory systems supporting embodied self-experience” — a more comprehensive model than the prefrontal-centric account, and one that argues (as the 2022 page’s open problem anticipated) for interventions targeting embodied sensory representation directly rather than only prefrontal reappraisal.
The interoception boundary, held again
As with every Farb relapse paper, the protective “sensory” pole here is dominated by somatosensory cortex (postcentral, supramarginal) — exteroceptive touch and multisensory integration, not the insula proper. The orbitofrontal/insula territory appears in the self-criticism map, not the protective somatosensory one. So the interoceptive reading remains a generalization: the paper establishes that bodily/sensory self-representation is protective and its deactivation forecasts relapse, and frames this as “embodied self-experience” grounded in “ongoing bodily signals,” but the strongest single predictors are somatosensory rather than interoceptive. The theoretical bridge the paper leans on (Farb 2007, 2015; Paulus & Stein 2010; Seth 2013) is that accessible interoceptive/somatosensory representation supports flexible present-moment self-processing, while reduced access contributes to dysregulated, narrative-rigid self-processing — the simulation-map framework’s clinical face.
Not a duplicate of Farb (2022)
Same trial, same cohort, same 2-year follow-up, overlapping authorship (Segal, Farb) — but a distinct study: different journal issue (NeuroImage: Clinical 51, 2026, 104025 vs 34, 2022, 102969), different first author (Wu vs Farb), a different scanning paradigm (SRET vs sad-film mood induction), and a different question (self-referential bias and its neural markers vs mood-reactivity biomarkers). It cites Farb (2022) repeatedly and treats the somatosensory-deactivation/relapse link as a finding it replicates. Recorded here as the fourth relapse paper, complementary to 2022, not a re-report of it.
Provenance
Norman Farb is this wiki’s author (see norman-farb); recorded as provenance, not weighted as authority — the convention used for farb-2022-relapse-biomarkers and farb-2011-relapse-prediction. Senior author Segal declares book royalties (Guilford), workshop fees, and Mindful Noggin revenue related to his MBCT co-founder role; Wu, Livingston, and Farb declare none. Funded by CIHR (Grant 243812). Deidentified behavioural and clinical data are public on OSF (s4n8j); fMRI-derived data on request. First author Liliana C. Wu and second author Jordan L. Livingston are given no researcher pages — co-authors, not independently load-bearing on the wiki yet.