NOTE-1061: Bo et al. — Deconstructing the brain bases of emotion regulation

Contribution

Earlier imaging work on reappraisal compared regulated with unregulated viewing and called what went up a “generator” and what went down a “target”. It could not show that a region was not engaged, so it could not say whether the regions active in reappraisal are specific to it, nor which emotion-related regions reappraisal leaves alone. This study uses Bayes factors, which can give evidence for a null, in two samples large enough for that (n = 182 and 176), and requires every effect to replicate in both. It finds four kinds of region, two of which (reappraisal-only and non-modifiable) needed the null to be established.

Key insight

A region can be classified by the pattern of what it does and does not do across two contrasts, emotion generation [Look Negative − Look Neutral] and reappraisal [Reappraise Negative − Look Negative], as an axiomatic definition: a reappraisal-specific region “must (1) respond more strongly when reappraising negative images than when merely looking at them, and (2) must not respond differentially to negative vs. neutral images in the absence of reappraisal demand” (p. 5). Strong evidence for the null in condition (2) is what turns active during reappraisal into selective for it.

Assumptions

• Two contrasts carry the processes. Emotion generation is Look Negative against Look Neutral; reappraisal is Reappraise Negative against Look Negative (p. 5). Look Negative is taken as the unregulated response.
• Three conditions stated by the authors (p. 23). “(1) brain activation from different conditions reflects different demand on appraisal processes, (2) Look negative vs look neutral can cover brain response of appraisal processes, (3) Appraisal processes do not introduce activity decreases.” They say each could fail, and what would follow.
• Thresholds. BF > 10 for an effect and BF < 1/10 for a null, with a JZS prior (Rouder et al. 2009), computed voxelwise from group t statistics (pp. 25–26). BF 10 corresponds to p < 0.0018 here (p. 8). A null at 10:1 needs at least 120 participants (p. 5; p. 26).
• Consensus maps. Each study’s binary map was smoothed (3 mm) and the two multiplied; a voxel enters the consensus map above a product of 0.01 (p. 26). The worst consensus voxel lies 4–4.4 mm from the nearest voxel in either study’s map (p. 27).
• The sample. A midlife community sample from the Pittsburgh area, mean age 42.7 and 40.1 (p. 21), from two existing projects (AHAB-2 and PIP). Each person saw 15 trials per condition of IAPS pictures, a 2 s cue (“Look” or “Decrease”), the picture for 7 s, and a 4 s rating of negative feeling on a 5-point scale (p. 24).

Key results

• Behaviour (p. 7). Negative pictures raised negative ratings over neutral ones (t = 63.8, d = 3.37). Reappraisal lowered them against looking (t = −17.5, d = 0.92). “81% of individual participants showed regulation effects in the same direction as the group findings.” The dominant strategies, by McRae et al.’s coding, were changing current circumstances, reality challenge and changing future consequences (Fig. 1D).
• Reappraisal only (p. 8; Fig. 2A). Sixteen regions replicated across both studies: anterior lateral prefrontal cortex (aDLPFC, aVLPFC), TPJ, IPL, posterior middle and anterior inferior temporal gyri, and superior cerebellum. By resting-state network, 47% default mode, 31% frontoparietal, 16% ventral attention. On the bar plots these regions sit at the same level for Look Neutral and Look Negative and rise only under reappraisal.
• Common appraisal (pp. 8–9; Fig. 2B). Fourteen regions, “the largest set of brain regions among system components”: DLPFC, VLPFC, SPL, pSTS, anterior insula, a dMPFC/aMCC/SMA cluster, precuneus and cerebellum. Reappraisal-only regions lie anterior and lateral to them, “suggesting a cortical gradient” (p. 9).
• Non-modifiable emotion generation (p. 10; Fig. 3A). “Most regions responsive to emotional stimuli were non-modifiable, including all subcortical emotion-related regions.” Thirty-two regions: amygdala, PAG, parabrachial nucleus, ventral putamen, hypothalamus, ventral striatum and pallidum, and in cortex lateral OFC, medial prefrontal regions, midcingulate and retrosplenial cortex, agranular area 6, parietal cortex, precuneus and ventral fusiform.
• Modifiable emotion generation (p. 10; Fig. 3B–C). Eight regions, “located exclusively in the cortex”: mainly higher retinotopic visual areas (V3, V4, V7, VO, LO, MT), intraparietal sulcus, postcentral gyrus and inferior frontal junction. “We did not detect any modifiable emotion regions in the subcortical cortex” (sic, p. 10).
• Amygdala and PAG (p. 11; Fig. 4). Every amygdala subregion and the PAG responded to negative pictures (all p < 0.0001), and none was reduced by reappraisal (all BF < 0.1). Basolateral and superficial amygdala were “numerically in the opposite direction as predicted by the downregulation hypothesis”. On the figure the reappraise bars equal or slightly exceed the look-negative bars.
• Reappraisal success (pp. 11–12; Fig. 5). Success, the drop in rated negative affect, correlated with reappraisal-related activity in reappraisal-only regions (r = 0.2, adjusted R² = 0.039) and common appraisal regions (r = 0.14), negatively with modifiable regions (r = −0.14), and not with non-modifiable ones (r = 0.015, p = 0.77). The best subset model held reappraisal-only and modifiable together (r = 0.26, adjusted R² = 0.063). A whole-brain robust regression (n = 358, FDR q < 0.05) added nucleus accumbens, pregenual ACC and posterior insula and somatosensory cortex (lower with success), which the four systems missed because they fell with negative emotion.
• Annotation (pp. 13–16; Figs 6–7). Against 50 Neurosynth topics, reappraisal-only maps were most selective for “response inhibition”, common appraisal for “cognitive conflict”, non-modifiable for “emotion face” and modifiable for “object recognition”. Against PET binding maps, reappraisal-only and common appraisal both went with CB1, 5-HT2A, mu-opioid, mGluR5 and GABA-A; non-modifiable regions went with 5HTT, NET, mu-opioid and D2. Both analyses are called hypothesis-generating.
• The authors’ readings (pp. 17–22).
  • Reappraisal-only regions are among the most expanded in humans and “positioned above unimodal sensory hierarchies and thus in a position to regulate them based on conceptual context” (p. 17).
  • Common appraisal regions “are presumably involved in the appraisal processes that give rise to negative emotion, and may be involved in spontaneous self-regulation in the absence of explicit regulatory instructions” (p. 18).
  • Summary: “emotion generation and reappraisal depend on many of the same cortical regions, in line with psychological theories that posit overlap between affective appraisal and reappraisal. However, reappraisal involves a distinct set of brain regions” (p. 20).
  • Subcortical regions “were neither influenced by reappraisal nor correlated with reappraisal success”, which “might indicate that subcortical affective processes are less modifiable by reappraisal than previously thought” (p. 20). Two reasons are offered for the difference from earlier amygdala findings: reinterpretation strategies dominate here, and the sample is older (pp. 20–21).
  • Modifiable visual regions are read as semantic and conceptual representation, not only shifted attention, since they are higher-order visual areas (p. 21).
  • The closing sentence: reappraisal “is not able to modulate subcortical-level affective experience as well as lateral prefrontal and orbitofrontal areas related to value construction” (p. 22).

Claims

id	claim	strength	support
C1	Reappraisal of aversive pictures lowers reported negative affect	strong	n = 358; d = 0.92; p. 7
C2	Some regions are engaged by reappraisal and show evidence of no response to negative pictures without it	strong for this task and these samples	BF > 10 and < 1/10, replicated in both samples; p. 8
C3	Most regions engaged by reappraisal are also engaged by negative emotion generation	strong as a count of regions in this task	p. 9; p. 18
C4	Amygdala, PAG, hypothalamus and other subcortical regions respond to negative pictures and show evidence of no change under reappraisal	strong for BOLD in this task	BF < 1/10 voxelwise and in ROIs; pp. 10–11
C5	The regions reappraisal reduces are cortical, mainly higher visual and parietal	moderate	p. 10; the modifiable set is the smallest (eight regions)
C6	Reappraisal-only and modifiable-region activity each predict reappraisal success; non-modifiable activity does not	moderate: r of 0.14–0.26	pp. 11–12
C7	Reappraisal-only regions sit at the top of the cortical hierarchy and serve executive control	weak: by annotation and citation	Neurosynth topics; Burt et al.; pp. 14, 17
C8	Reappraisal cannot modulate subcortical affective experience	weak as stated: experience was not measured subcortically	the authors’ reading of C4, p. 22
C9	Reappraisal systems relate to cannabinoid and serotonin binding	weak, exploratory	spatial correlation with PET atlases; pp. 15–16

Concepts

• Systems identification (axiomatic): defining a hypothetical process by the effects any signal encoding it must and must not show, after Caplin & Dean and Rutledge et al. (p. 5).
• The four components (pp. 5–6): “Reappraisal only” (reappraisal +, emotion null); “Common appraisal” (both +); “Non-modifiable emotion-generation” (emotion +, reappraisal null); “Modifiable emotion-generation” (emotion +, reappraisal −).
• Reappraisal: “reinterpreting the meaning of stimuli and events in service of a regulatory goal” (p. 3).
• Reappraisal success: a person’s drop in rated negative affect, Look Negative minus Reappraise Negative (p. 11).
• Bayes factor for the null: BF < 1/10, read as 10:1 evidence that the two conditions do not differ (p. 8).

Connections

The design and samples come from Gianaros et al. (2014, 2020, 2022). The axiomatic method is borrowed from Caplin & Dean (2008) and Rutledge et al. (2010). It sets itself against the meta-analyses that make the amygdala the main target of reappraisal (Buhle et al. 2014, a Wager co-authored paper; Frank et al. 2014), and agrees with the meta-analysis finding that reinterpretation does not reliably lower amygdala activity (Dörfel et al. 2014). Its framing question is Gross & Barrett (2011), “Emotion generation and emotion regulation: one or two depends on your point of view”, which THEORY-056’s promotion test names and the record has not filed. It also cites Mesquita & Frijda (2011) and Barrett 2017b (LIT-542) for the view that affect generation is widely distributed.

Bearing on the record

• The dispute between THEORY-056 and THEORY-680. This is the first data the record holds on the question that divides them, whether regulation is a process apart from generation. The paper gives each side a result.
  • For THEORY-056. Most of the reappraisal network is also engaged by looking at negative pictures without any instruction (C3). The authors read these as appraisal regions shared with generation, and allow that they “may be involved in spontaneous self-regulation”. That is what a regulation made of the same stuff as generation looks like, and the authors place it “in line with” theories of overlap, naming Gross & Barrett and Mesquita & Frijda.
  • Against THEORY-056. A separate set of regions is engaged by reappraisal and shows evidence of no engagement by emotion generation (C2). They are anterior prefrontal and temporo-parietal, annotated as response inhibition and executive control, and the best predictors of success (C6). The authors describe them as “positioned above unimodal sensory hierarchies and thus in a position to regulate them”. That is close to the distinct regulating system above the motive states that THEORY-056 denies. THEORY-056’s promote_when says in advance that “a difference in network activation between emotion and regulation tasks” cannot settle it, because one readiness may be built differently from another. By that clause the result does not refute it. But the clause then excludes the very evidence a separate regulator would leave in imaging, and the record should note that it makes THEORY-056 hard to test with imaging at all.
  • For THEORY-680, partly. C2 fits its claim that regulation includes processes distinct from generation. C4 and C5 fit its picture of reappraisal acting early, on meaning: what reappraisal changes is higher visual and parietal representation of the picture, not the subcortical response.
  • Against THEORY-680’s measurement. Its promote_when asks for each person’s unregulated response to be estimated independently. Here Look Negative is the unregulated baseline, and the authors themselves allow that common appraisal regions may carry spontaneous regulation during it (p. 18). That is the counterfactual problem THEORY-680 already names, met in practice. Only one strategy was tested, so nothing compares strategies by stage.
  • Neither promotion test is met. No lesion, interference or developmental dissociation (THEORY-056); no stage comparison (THEORY-680). Filed as THEORY-878.
• QUESTION-054 (is there a governing level?). The reappraisal-only regions are a candidate governing level for deliberate emotion regulation, in the anatomical sense: selective, high in the cortical hierarchy and predictive of success. It is a partial answer for one deliberate strategy only. It says nothing about effortless regulation, which is where THEORY-056 and THEORY-070 differ.
• CLAIM-412 (experience constituted by interpretive organization). The exchange cited this study as a limit on the constitutive reading: not every component of an emotion is open to reinterpretation. The reading supports a narrower statement. Reinterpretation lowered reported negative feeling while amygdala, PAG and hypothalamus responses to the pictures were unchanged, with evidence for the null. Whether that limits the claim depends on whether those responses are part of the experience.
  • If they are, reported experience changed while part of what constitutes it did not, which the claim must explain.
  • If they are not, the result shows what the claim allows: reported experience follows the reorganized cortical interpretation, not every process the stimulus drives.
  • The study cannot choose. It measures BOLD in regions and one rating per trial, and its closing phrase “subcortical-level affective experience” (p. 22) treats subcortical activity as experience without evidence. The reading that an unchanged subcortical component is the emotion, apart from a feeling that changed, is the split Barrett et al. reject (NOTE-1067). Filed as THEORY-871.
  • The exchange’s description should be corrected in one respect. The unmodulated system is not only subcortical. Most emotion-responsive regions were non-modifiable, cortical ones included (lateral OFC, medial prefrontal, midcingulate, parietal); the subcortical regions were the ones with none modifiable.
• CLAIM-411 (broad interpretation admits processing not shown to be experiential). Indirect support. The common appraisal regions are engaged without instruction and “may be involved in spontaneous self-regulation”, an appraisal that is not reported as such. Nothing here shows it is unconscious.
• CLAIM-414 (four facts that come apart). A loose parallel, mine. Reappraisal is an intervention on the interpretation, and the result separates where the interpretation is represented (reappraisal-only and common appraisal regions) from where it is effective (modifiable regions) and where it is not (non-modifiable ones). It is correlational imaging, not the construction the claim rests on.
• TERM-073 (interpretation). The paper’s “reappraisal” is interpretation in the term’s sense, a revised significance with consequences. Its “appraisal” regions are those engaged whether or not the revision is asked for.
• LIT-542, NOTE-434, C8 (Barrett 2017). Barrett’s hypothesis was that reappraisal works with the same predictions that construct the emotion, and that emotion and regulation are brain modes, not two systems. C3 supports the overlap. C2 is the part Barrett’s hypothesis did not predict. NOTE-434 reports Barrett’s Figure 4C overlap (anterior insula, vlPFC, MCC, SMA, posterior STS), and those are the common appraisal regions here.
• THEORY-057. Not tested: the study has one category, negative affect, and no variation across instances.
• THEORY-767. Not touched. The authors suggest that posterior insula and somatosensory regions lower with success “may be related to interoception” (p. 13), without a test.
• No instruction for machine-learning practice; nothing for the anthology.

Limitations

• Preprint, not the version of record. The published paper may differ; only the abstract was compared. The supplement was not read.
• One task, one strategy, one kind of stimulus. Instructed downregulation of negative IAPS pictures, 15 trials per condition. Nothing here says whether other strategies, upregulation or real-life emotion behave alike.
• “Unaffected” is a BOLD null at 2 mm voxels and region scale. Fast or small changes, or changes in pattern without a change in mean, would not show. The authors’ explanation, that subcortical processes are engaged before reappraisal can act, is offered, not tested.
• The component definitions rest on the authors’ three assumptions (p. 23), and they say the key limitation is that “emotion appraisal may not be sufficiently defined by current contrast”. The non-modifiable set is a null on one contrast, so it depends on Look Negative being an unregulated baseline.
• Experience is one rating per trial. Every claim about experience goes through it. C8 goes beyond it.
• Small individual-difference effects. The best model explains about 6% of the variance in success.
• Reporting slips. The discussion says “two large samples of over 200 participants each” (p. 17), against 182 and 176. “Subcortical cortex” (p. 10). The reappraisal condition is named “Reappraise”, “Regulate” and “Decrease” in different places. The methods give the Neurosynth topics as 50 and, a few sentences later, 52 (p. 27). None changes a result.

Open questions

• Do the reappraisal-only regions do the regulating, or do they only accompany it? Interference with them (stimulation, lesion) during reappraisal, with reported affect and the modifiable regions measured, would tell.
• Does the subcortical null hold for other strategies, for upregulation, and with designs sensitive to fast or patterned change?
• Is Look Negative free of regulation? A design that estimates each person’s unregulated response separately (THEORY-680’s test) would show whether common appraisal regions include spontaneous regulation.
• Does the subcortical response bear on what is felt? Studies that predict trial-by-trial experience from amygdala or PAG activity, with and without reappraisal, would show whether the unchanged response is part of the experience or apart from it.