We performed a series of analyses designed to identify prefrontal-subcortical pathways linked to reappraisal success. We defined reappraisal success as the decrease in reported emotion when applying a cognitive reappraisal strategy to aversive images (ReappNeg trials) vs. experiencing the natural emotional response to aversive images (LookNeg trials). Thus reappraisal success scores were the [LookNeg – ReappNeg] values for each participant. We chose emotion self-reports as an outcome because they are stable and reliable, they predict numerous mental and physical health outcomes (Gross and Munoz, 1995
; Moskowitz, 2003
; Tugade et al., 2004
), and no other measure provides a direct correlate of emotional experience. We applied MEPM and multivariate clustering methods to relate reappraisal success with reappraisal-related increases in brain activity—[ReappNeg – LookNeg] contrast values for each participant, which we refer to as reappraisal-related activation.
Analyses of emotion reports showed that versive images evoked substantially stronger negative emotion reports (mean = 3.60 points, SE = 0.10) than neutral images (mean = 1.33, SE = 0.03; paired t-test t(35) = 22.3, p < .0001). Reappraisal of negative images resulted in robust reductions in emotion reports (ReappNeg mean = 2.65, SE = 0.10), an average reduction of 0.95 points, paired t(35) = 8.83, p < .0001; see Supplementary Figure 2
for additional details). We refer to this reduction in reported emotion as reappraisal success
in subsequent analyses.
fMRI analysis proceeded in a sequential series of steps. (1) First, we identified regions showing significant reappraisal-related activation. (2) Then, we used standard statistical parametric mapping to locate voxels from Step 1 in which reappraisal-related activation predicted reappraisal success. From among the resulting regions, we focused specifically on the right vlPFC as a predictor in subsequent analyses. (3) The next step was to conduct mediation analysis in subcortical ROIs. We defined ROIs in the amygdala and NAC/VS and tested whether voxels in each region mediated the vlPFC-reappraisal success relationship. (4) To locate additional regions, we used MEPM to generate a whole-brain map of mediators (4a), and then used cluster analysis to group these mediators into coherent networks and tested their independence from one another (4b). (5) Finally, we considered whether other candidate PFC regions showed similar relationships with reappraisal success mediated by NAC and amygdala.
1. Reappraisal-related activations
Results from the [ReappNeg – LookNeg] contrast are shown in (orange/yellow) and Supplementary Table 1
, controlling the expected false discovery rate (FDR) in whole brain search at q
< .05 (Genovese et al., 2002
). Consistent with previous work, we observed increases in a number of PFC regions. Dorsal and ventral PFC were active bilaterally, including the inferior frontal junction (IFJ), inferior frontal gyrus (IFG), middle frontal gyrus (Brodmann’s Area [BA] 8/9), and anterior PFC (BA 10). Medial PFC regions included anterior cingulate (BA 24) and dorsomedial PFC (DMPFC) in and anterior to pre-SMA (BA 9). Posterior cortical regions included the inferior parietal lobule (IPL), angular gyrus, and middle and inferior temporal gyri extending into anterior temporal cortex (aTC). Activations appeared to be more reliable on the left, though we did not perform detailed analysis of laterality effects. Decreases during reappraisal (blue in ) were found in the parietal operculum, around SII, and in the parahippocampal cortex.
Results for the [ReappNeg – LookNeg] contrast
2. Identification of potential predictor regions: Correlates of reappraisal success
Voxel-wise regression analyses identified 8 regions in which reappraisal-related activation was significantly predicted by reappraisal analysis, shown in and reported in detail in (p < .005 and k = 3 contiguous voxels; see Methods
for details). Only voxels identified in Step 1 above were subjected to analysis. Reappraisal-success related regions included bilateral vlPFC (see also ) and temporal regions, pre-SMA and DMPFC, left inferior parietal lobule, and right caudate. The cluster in right vlPFC, identified by its position on the inferior frontal gyrus on the mean normalized anatomical image for our participants, was used as the predictor region of interest in subsequent mediation analyses. Analysis steps 3 and 4a use this region as a frontal predictor region in MEPM analyses. Analysis steps 4b and 5 explore the relationships between right vlPFC and other cortical regions correlated with reappraisal success.
Correlations between reappraisal activation [Rea - Neu] and reappraisal success [Neu - Rea] in reported experience
We chose vlPFC as focus region for three reasons. First, vlPFC – and especially right vlPFC – is well-known as a region critical for cognitive control in general, and response inhibition or the selection of information in particular (Aron and Poldrack, 2005
; Badre and Wagner, 2005
; Nee et al., 2007
; Phillips et al., 2003
). Second, studies of reappraisal and other cognitive means of regulating affective responses (Cunningham and Zelazo, 2007
; Lieberman et al., 2007
) have consistently activated right vlPFC. Third, activation measures in vlPFC have been shown to be abnormal in emotional disorders (Drevets et al., 1992
; Lawrence et al., 2004
). This choice was necessary to search for mediators, but we did not expect it to be the only important PFC region. As we describe below, subsequent analyses searched for additional PFC regions whose relationship with emotion was mediated by subcortical ROIs.
We also conducted a whole-brain exploratory search for correlations with reappraisal success to locate regions that may have been missed in our more constrained masked analysis (q < .05 FDR-corrected, p < .0005). The results largely confirmed those in our constrained search, including bilateral vlPFC and pre-SMA, but also identified additional regions shown in Supplementary Figure 3
, including the dorsal thalamus, caudate body, and caudate head extending into the NAC/VS, parahippocampal cortex, cingulate isthmus (which bridges the posterior cingulate and medial temporal cortices), and area around periaqueductal gray (PAG). Whole brain mediation and network analyses (Section 4a) identified some of these regions as significant mediators, and we explored their relationship to the key ROIs in our study in Section 4b below.
3. Testing the amygdala and NAC/VS as mediators using a priori ROIs
With the right vlPFC region identified as the predictor, and reappraisal success as the outcome (), the next step in the MEPM analysis was to search, voxel-by-voxel, for mediators of the rvlPFC-reappraisal success relationship within two subcortical ROIs that were of a priori interest: the amygdala and NAC/VS.
For a variable to be considered a significant mediator, we required that it reach statistical significance in each of three tests (at p < .005 and 3 contiguous voxels in each), which were conducted as part of the same path model. First, the predictor variable (rvlPFC) must be related to the mediator (voxel in amygdala or NAC/VS), which we refer to in the text and all Figures/Tables as path a
. Secondly, the mediator must be directly related to the outcome (reappraisal success), controlling for vlPFC (path b
). Finally, the mediation effect must be significant (effect a*b
), which amounts to a statistical test on the product of the a
path coefficients, or equivalently, a test that the predictor-outcome relationship (vlPFC-reappraisal success) is significantly reduced by including the mediator in the path model. The threshold of p < .005 and 3 contiguous voxels controlled the family-wise error rate at p < .05 corrected in each region (see Methods
According to standard conventions for mediation analysis, we refer to the overall predictor-outcome relationship as effect c, and the direct effect controlling for the mediator as c’. Thus, the a*b effect tests the significance of c – c’. In statistical reports and figures, a refers to the vlPFC-mediator relation, b refers to the direct mediator-reappraisal success relation (controlling for vlPFC), and a*b refers to the mediation effect.
For the amygdala analysis, we manually identified amygdala ROIs in gray matter on the average T1 image from our sample after warping to Montreal Neurologic Institute (MNI) space (left: [−24 0 −23], 5958 mm3
; right: xyz = [21 0 −22], 4354 mm3
(; see Methods
). MEPM analysis identified a portion of ventral left amygdala that negatively mediated the vlPFC-reappraisal success relation (xyz = [−24 0 −32], 319 mm3
). This portion of the left amygdala was positively associated with vlPFC, but predicted reduced
reappraisal success: aamy
= .57, Z = 4.02, bamy
= −.24, Z = −5.01, and abamy
= −0.14, Z = −3.30, all p < .001. This finding suggests that the vlPFC plays a role in generating
negative affective responses through the amygdala, perhaps due to PFC-related appraisal processes that support the generation of negative responses to emotional pictures. Supplementary Figure 4
shows path diagrams and scatterplots for path analyses in both amygdala and NAC/VS. In this and other Figures showing ROI analyses, the region contiguous with the significant cluster is shown at p < .005 (dark blue), p < .01 (light blue), and p < .05 (purple; all 2-tailed), including the extent beyond the ROI boundary.
Mediation analysis results within regions of interest (ROIs)
For the NAC/VS analysis, we manually identified ROIs in NAC/VS gray matter adjacent to the posterior ventral caudate head (left: xyz = [−10 14 −9], 2765 mm3
; right: xyz = [7 14 −9], 2818 mm3
(). MEPM analysis identified a left NAC/VS region that positively
mediated the vlPFC-reappraisal success relation (xyz = [−10 14 −14], 106 mm3
). The left NAC/VS was positively associated with both vlPFC and reappraisal success: aNAC/VS
= .42, Z = 2.86, bNAC/VS
=.29, Z = 3.20, and abNAC/VS
= 0.12, Z = 3.04, all p < .002. Supplementary Figure 4
shows the significant region and contiguous voxels at p < .005 (yellow), p < .01 (orange), and p < .05 (pink; all 2-tailed).
We then tested whether each region was a significant mediator of rvlPFC-reappraisal success correlations while controlling for effects attributable to the other region. We averaged over voxels in each region and included both amygdala and NAC/VS in the same path model. This analysis tests whether the results, shown in , indicated that each subcortical pathway makes an independent contribution to reappraisal success. rvlPFC was positively associated with both amygdala (aamy = .42, SE = .15, p < .01) and NAC/VS (aNAC/VS = .57, SE = .17, p < .001). Consistent with the individual mediation analyses, the two subcortical regions had opposite associations with reappraisal success: increasing amygdala activity was associated with reduced success (bamy = −.19, SE = .06, p < .001), whereas increasing NAC/VS activity was associated with increased success (bNAC/VS = .18, SE = .09, p < .05). Finally, both regions were significant mediators (abamy = −0.11, Z = −2.83, p = .01; abNAC/VS = 0.07, Z = 2.31, p = .002), indicating that each explains a part of the rvlPFC-reappraisal success covariance while controlling for effects attributable to the other mediator.
Because each pathway had an opposite effect on reappraisal success, this suggests that the rvlPFC-reappraisal success correlation is partially masked by their opposing effects. The correlation between vlPFC and reappraisal success was only moderately significant (r = .43, p = .018) without controlling for any additional regions, and was much stronger (r = 0.66, p < .0001) when controlling for amygdala activity.
These results demonstrate the advantage of the mediation approach over a simple bivariate (i.e., two-variable, no mediators or confounding variables) correlation approach. To more fully illustrate this advantage, we tested bivariate correlations with reappraisal success in the amygdala and NAC/VS. The left panel of shows the results in the amygdala ROIs (again including contiguous significant voxels extending outside the ROI), demonstrating significant negative correlations with reappraisal success (peak r = −.54, p < .005 one-tailed; ). However, a small subset of voxels show significant effects in the amygdala ROIs (1 voxel at p < .005, 3 at p < .01, and 5 at p < .05, one-tailed). shows direct amygdala-reappraisal success associations, controlling for right vlPFC. Much stronger effects are apparent (32 voxels in the amygdala at p < .005, 44 at p < .01, and 78 at p < .05, Two-tailed; ). The NAC/VS showed strong positive correlations with reappraisal success (, p < .0004, whole-brain FDR corrected q < .05), which were somewhat reduced but still strongly significant after controlling for vlPFC (). Overall, the results confirm our prior expectations that amygdala and NAC/VS activation would show negative and positive brain-behavior correlations, respectively, and they demonstrate the value of the path modeling approach.
Detail of effects in the amygdala (left) and nucleus accumbens (NAC)/ventral straitum (right)
4. Identifying mediating networks localized with MEPM analysis
One important issue with the preceding results is that activity in other emotion-related regions might also play a mediating role in the vlPFC-reappraisal success relationship. In order to examine how the amygdala and NAC/VS fit into distributed networks that might underlie reappraisal success, we performed a whole-brain search for mediators of the rvlPFC-reappraisal success relationship. Then, significant regions from this mediation effect map were subjected to component analysis and clustering into interconnected networks. Finally, the average contrast values within each network were entered into a path model as mediating variables, with one mediator per network. This analysis tested, whether multiple distributed networks make independent contributions to reappraisal success.
Results of the whole-brain search for mediators are shown in on medial and limbic surfaces (left and center panels, respectively) and on sagittal and coronal slices (right panel). These included several regions with positive indirect effects, including NAC/VS and neighboring subgenual cingulate and ventral striatum, pre-SMA, precuneus, superior frontal gyrus, dorsal caudate, and the cingulate isthmus. Regions showing negative indirect effects included bilateral amygdala, ventral anterior insula, rostral dorsal anterior cingulate (rdACC), and two regions in the subthalamic zone (near the ventral thalamus and hypothalamus). Stereotaxic coordinates and statistics for all mediators are shown in .
Mediation results in whole-brain search
Mediators of the right ventrolateral prefrontal cortex (vlPFC) relationship with reappraisal success in whole-brain search
Cluster analysis combined with nonparametric inference ((Etkin and Wager, 2007
); see Supplementary Materials
for details) revealed evidence of two clusters, one consisting of the positive mediators (yellow in ), including NAC/VS, and the other of the negative ones (blue in ), including amygdala. These networks are shown in the context of the path model in . Regions in each network are shown on brain surfaces, with solid black lines connecting pairs of regions that are significantly correlated (q < .05 FDR-corrected) and whose correlation is not mediated by any single other region. Thus, grouping the set of positive mediators and the set of negative ones seemed to provide a sensible first-order grouping of the results, and further sub-division was not warranted by the data.
To assess the overall organization of these regions and their relationship with vlPFC activity and reappraisal success, we next averaged over voxels from all the regions within in each network to obtain average [ReappNeg – LookNeg] contrast values for each network. We included the average scores from each network in the mediation model. This step was employed to avoid multi-colinearity problems that make many multivariable SEM models inestimable or unstable and difficult to interpret. Path coefficients and standard errors are shown in . As before, both the positive and negative mediating networks were positively correlated with rvlPFC, but their direct effects on reappraisal success had opposite signs: the amygdala/insula/rdACC network (blue) was associated with reduced success, and the NAC/VS/cingulate isthmus/pre-SMA/precuneus network was associated with increased success. Scatterplots and estimates of the probability distribution of the indirect effects (a*b,
) are shown for each network in .
5. Search for additional prefrontal predictor regions with MEPM analysis
In our previous analyses, we chose right vlPFC as a frontal predictor or ‘seed’ region based on prior anatomical and functional evidence. A natural question, however, is whether other frontal regions are similarly related to emotional experience through the cortical-subcortical pathways we describe. The MEPM analysis framework also provides a facility for searching for these mediated regions (i.e., regions that satisfy all requirements for mediated regions), given a chosen outcome and mediating pathway. We searched for frontal regions mediated by each of left NAC/VS and left amygdala (averaging over voxels in each region) controlling for the other (see Methods
for details). The results of each analysis are shown in , and coordinates and statistics for each significant region are listed in Supplementary Table 1
. The regions include the frontal regions that are part of the network shown in , including pre-SMA and SFS, as well as bilateral vlPFC and several other regions on the medial wall: medial frontal pole (MFP) and ventromedial PFC, including subgenual cingulate (SGACC) and medial orbitofrontal cortices (OFC). Notably, the rostral medial PFC was also strongly, positively correlated with reappraisal success in the bivariate correlation analysis (Supplementary Figure 2
), providing additional support for the involvement of the medial frontal cortex in successful downregulation of negative emotion. Thus, as expected, the right vlPFC is part of a set of frontal regions whose activation during reappraisal predicts drops in reported experience, to the degree that they co-activate with NAC/VS.
Frontal regions mediated by nucleus accumbens/ventral striatum (NAC)/VS and amygdala