2.5 Data analysis
Data were preprocessed and analyzed using the BrainVoyager QX 2.0 software package (Brain Innovation, Maastricht, The Netherlands). Preprocessing of the functional data included slice time correction (using sinc interpolation), 3-dimensional rigid-body motion correction (using trilinear-sinc interpolation), spatial smoothing with a FWHM 4-mm Gaussian kernel, linear-trend removal, and temporal high-pass filtering (fast-Fourier transform based with a cutoff of 3 cycles/time course). The functional data sets were coregistered to high-resolution, within-session, T1-weighted anatomical images which were in turn normalized to Talairach space (Talairach & Tournoux, 1988
), to create 4-dimensional data sets. While it is possible that this normalization process could differentially affect younger participants whose brains may differ more from the Talairach template than the brains of older participants, it was nevertheless important to ensure that participants’ data could be effectively combined and statistically assessed. Additionally, normalizing to Talairach space allows comparison of the present findings to prior adult studies. Kang et al. (2002)
provided an empirical validation of normalization for analysis of fMRI data from children. They found very small differences (relative to the resolution of fMRI data) in the spatial correspondence among several brain loci between young children and adults after a standard, nonlinear transformation that warped child and adult fMRI data into a common adult Talairach space. These and other similar findings (Burgund et al., 2002
) support the use of a common, adult stereotactic space in this study. An in-house script was used to identify (and exclude) participants for whom, after removing volume acquisitions where movement between two volumes or integrated movement over 4 volumes exceeded 1mm, more than 25% of the data was removed from the entire experiment or one experimental condition.
To confirm that participants understood and performed the task, behavioral ratings grouped by experimental condition (look-gross, look-neutral, decrease-gross, and increase-gross) were averaged in each participant. These average ratings were then compared in group-wise paired-samples t-tests. The first t-test compared ratings for look-gross to ratings for look-neutral in order to confirm that participants responded to the emotional nature of the stimuli. Additional t-tests compared ratings for look-gross to ratings for decrease-gross and increase-gross, respectively. These two t-tests were performed in order to confirm that participants experienced a change in their emotional reactions to the stimuli when instructed to modulate their reaction to the gross pictures.
To investigate brain regions modulated during the experimental paradigm, a random-effects multi-participant general linear model (GLM)-based analysis was performed. Regressors were defined as boxcar functions peaking during each of the four experimental conditions (predictors of interest), as well as three additional boxcar functions peaking during, instruction, affect rating, and “relax” periods (predictors of no interest). These boxcar functions were convolved with a double-gamma hemodynamic response function (HRF) time-locked to the onset of the 4-second image display for the experimental conditions, and to the 6-second instruction period, the 6-second affect rating, and the 2-second “relax” period, respectively. To additionally account for motion during each scan, functions of all of the 3 directions and 3 translations of movement from each participant were included in each single-participant GLM-based analysis as additional predictors of no interest. In all whole-brain analyses, a mask was used to restrict analyses to only voxels located within the brain, determined by the extent of the MNI brain normalized to Talairach space.
To identify brain regions modulated by the emotional nature of the stimuli, brain activation in the contrast of look-gross > look-neutral was assessed at a statistical threshold of p
< .05, corrected for multiple comparisons with a cluster threshold of 34 contiguous functional voxels (Forman et al., 1995
; Xiong et al., 1995
). This cluster threshold was calculated by the BrainVoyager cluster-threshold estimator plugin performing 1000 iterations of a monte-carlo simulation to correspond to α < .05.
To identify brain regions modulated by efforts to emotionally regulate (increase and decrease), a random-effects analysis was performed on the conjunction of both regulation contrasts (decrease-gross > look-gross and increase-gross > look gross). This conjunction analysis was assessed at a statistical threshold of p < .05, corrected to α < .05 with a cluster threshold of 34 contiguous functional voxels.
To identify regions modulated by each emotion regulation strategy individually, brain activation in the contrasts of decrease-gross > look-gross and increase-gross > look-gross were assessed separately, each at a statistical threshold of p < .05, corrected to α < .05 with a cluster threshold of 41 contiguous functional voxels. To explore the effects of each emotion regulation task specifically on regions responsive to gross pictures, a region of interest (ROI) mask was created using regions identified as more active to gross (versus neutral) pictures (identified in this same participant group, in the contrast of look-gross > look-neutral in the multi-participant random-effects GLM analysis at a statistical threshold of p < .05, k = 34). Restricting the analyses to only voxels in the ROI mask, the same contrasts of decrease-gross > look-gross and increase-gross > look gross were assessed at the same statistical threshold as the whole-brain analyses.
More specific ROI analyses were performed in the bilateral insula and amygdala, regions of a priori
interest because of their implicated roles in processing negative (and particularly gross) stimuli. Specifically, we chose to investigate the insula given its role in the processing of disgust and our focus on disgust-inducing images (Calder et al., 2000
; Ibañez et al., 2010
; Lane et al., 1997
; Phillips et al., 1997
; Schafer et al 2005
; Wicker et al., 2003
). We selected the amygdala because prior studies have consistently demonstrated that activity in the amygdala is modified by cognitive reappraisal efforts (e.g., Eippert et al., 2007
; Harenski & Hamann, 2006
; Kober et al., 2010; Koenigsberg et al., 2010
; McRae et al., 2010
; Ochsner et al., 2002
; Ohira et al., 2006
; Schaefer et al., 2002
). These ROIs were functionally defined from the multi-participant random-effects GLM analysis in the contrast of look-gross > look-neutral at a more stringent threshold of p
< .01 which allowed us to discriminate these specific ROIs. Average difference beta values in the contrasts decrease-gross > look-gross and increase-gross > look gross were calculated for each of the four ROIs and statistically tested in their variance from zero (representing no modulation by emotion regulation) using one-sample t
In the two regions we found to be significantly decreased by down-regulation (right insula, left amygdala), we used difference values (decrease-gross – look-gross) calculated for each participant as an index of successful down-regulation. We used these values as a covariate in the whole-brain analysis of decrease-gross > look-gross to identify regions where activation significantly correlated with the degree of successful regulation in each participant. Specifically, we looked for regions showing an inverse correlation with the covariate, indicating that increased activation in these regions predicted a larger decrease in insula or amygdala activation during regulation. This covariate analysis was assessed at a statistical threshold of p < .05, with a cluster threshold of 10 contiguous functional voxels. We used a more liberal cluster threshold as the low number of active voxels in this analysis precluded the use of BrainVoyager’s cluster-threshold estimator plugin.
The age range in the current study’s participant sample allowed for the examination of brain regions in which activation correlated with age while viewing gross pictures, as well as during emotion regulation. To this end, whole-brain voxel-wise analyses were performed with chronological age as a covariate in each of the three contrasts of interest (look-gross > look-neutral, decrease-gross > look-gross, increase-gross > look-gross). These covariate analyses were assessed at a statistical threshold of p <.05, with a cluster threshold of 34 contiguous functional voxels.
To further elucidate the nature of the age correlations identified in whole brain covariate analyses, we performed similar correlations in anatomically defined regions which overlapped with areas we found to significantly correlate with age and for which we had a priori
hypotheses about their importance in the current study, specifically the amygdala and insula. We visualized age correlations with activation in the contrast of decrease-gross versus look-gross in the left amygdala, and increase-gross versus look-gross in the left insula in two separate scatter plots. The left amygdala ROI was defined by the Talairach database (Lancaster et al., 1997
), while the left insula ROI was defined by manually drawing insular gray matter on the Montreal Neurological Institute (MNI) 152 standard brain and then converted to Talairach space by normalizing the MNI brain, as previously described (Deen et al., 2010
). Difference beta values for increase-gross – look-gross and decrease-gross – look-gross were calculated for the left insula and left amygdala, respectively, and plotted against age to visually inspect the correlation patterns for outliers or binary grouping patterns.