Imaging Data Acquisition and Analysis

Neuroimaging data were acquired with a 3-T GE scanner (Waukesha, Wisc.) with an eight-channel head coil with 2.5 × 2.5 × 2.5 mm resolution and T₂* weighting (TR=2,300 ms, TE=25 ms, flip angle=50°, FOV=24 cm, matrix=96×96, 41 contiguous 3-mm interleaved axial slices). Coregistration and normalization used a highresolution three-dimensional magnetization-prepared rapid gradient echo scan (NEX=1, TE/TI=min/725 ms, FOV=22 cm, matrix=256×192, bandwidth=31.25 Hz per 256 voxels).

Processing in AFNI (Analysis of Functional Neuroimages) included slice timing correction, coregistration, and normalization and nonlinear registering of echoplanar data to anatomical scans. Data were smoothed (5 mm full width at half maximum) and scaled to 2.5-mm isotropic voxels. For motion correction, repetition time (TR) pairs with a Euclidean norm motion derivative >1 mm were censored prior to individual-level analyses. To be included in the analyses, no more than 20% of TRs across conditions could be censored.

Individual-level general linear models included regressors for correct trials across task conditions, incorrect trials, and for baseline drift and motion (i.e., rotational movement of roll, pitch, yaw, and motion displacement in the x, y, and z axes). Functional connectivity used generalized psychophysiolog- ical interaction (gPPI) to model connectivity between each anatomically defined amygdala in the AFNI Talairach Daemon atlas and other brain regions across each task condition. Separate individual-level general linear models were created for the right and left amygdala seeds. PPI terms for congruent, incongruent, and neutral conditions were the product of detrended and demeaned seed and trial condition regressors. Individual PPI general linear models used the same regressors for task-related changes in activation, in addition to the time series for the seed and the three PPI terms. With gPPI, individual differences in activation are controlled to better isolate task-specific differences in connectivity (23).

All analyses relied on an event-related design and focused on task-related amygdala-based connectivity. This focus reflected the consistency of previous findings (14–16) and the greater stability of amygdala-based connectivity than activation on the dot-probe task (24). Thus, the results presented in the main text emphasize omnibus statistical models testing for differences in amygdala-based gPPI functional connectivity across task condition. Other analyses appear in the Supplemental Results section of the online data supplement.

The results are presented in three sections examining how amygdala connectivity at baseline 1) differed between patients and healthy comparison subjects, 2) related to overall treatment response in patients, controlling for ABMT effects, and 3) related to ABMT-specific treatment effects. In the main text, connectivity findings are highlighted where consistent associations emerged across these three sets of analyses; this convergence occurred only for right amygdala connectivity. Other notable results appear in the data supplement, including between-group comparisons for amygdala activation, associations of age and sex with brain function, treatment-related changes in brain function, and differences in brain function related to clinical indices beyond either diagnosis or PARS treatment response.

Across all analyses, significant clusters were identified using both whole brain and region-of- interest approaches. With an initial threshold of p<0.005 followed by a gray matter-masked cluster correction, a whole brain cluster threshold of 1,063 mm³ was needed for a correction of p<0.05. This threshold was determined using 10,000 Monte Carlo simulations in AFNI’s 3dClustSim tool with the autocorrelation function correction. Based on findings from previous imaging studies with the dot-probe task, a region-of-interest approach was used to test for significant results specifically in the prefrontal cortex and the insula (14–16, 24). The cluster-wise threshold for the prefrontal cortex was based on a single prefrontal cortex mask, used in a previous study with the dot-probe task (24), that encompassed gray matter voxels anterior to a plane drawn at y=0 perpendicular to the anterior commissure-posterior commissure line. Also as in previous studies with the dot-probe task (15), the threshold for the right and left insulae was defined based on the insula Talairach Daemon atlas in AFNI. 3dClustSim produced a cluster-wise threshold size of 734 mm³ for the prefrontal cortex and 203 mm³ for each insula, for a correction of p<0.05. The group maps were also thresholded to include only data for which 90% of participants had valid data. All Talairach coordinates are presented in the left-posterior-inferior convention.

The first imaging analyses examined amygdala-based connectivity using individual- level connectivity values (PPI coefficients) for 105 participants (54 patients and 51 healthy comparison subjects). Connectivity values were subjected to a linear mixed-effects model using AFNI’s 3dMVM program (25) with baseline group (patients, healthy comparison subjects) as a between-subject variable and task condition (congruent, incongruent, neutral) as the within-subject variable.

The next imaging analyses examined relationships between connectivity and treatment response in 40 patients who had both usable pretreatment dot-probe fMRI data and a posttreatment clinical assessment. This set of analyses also used 3dMVM; posttreatment PARS rating was entered as a continuous variable, ABMT group (active, placebo) as a between-subject variable, and PPI coefficients for task condition (congruent, incongruent, neutral) as the within-subject variable. To control for baseline anxiety, pretreatment PARS rating was entered as a covariate.

Two interactions were tested within one model to yield two sets of results. First, the two-way condition-by-posttreatment PARS interaction was examined in patients as a group; this result maps connectivity related to overall CBT response, controlling for ABMT group and pretreatment PARS rating. Significant interactions were decomposed using partial correlation analyses between connectivity levels and posttreatment PARS rating. The second result considered connectivity related specifically to ABMT treatment response. This result pertained to the three-way condition-by-ABMT- by-posttreatment PARS rating interaction, mapping connectivity uniquely related to treatment differences in either the active or placebo ABMT group relative to the other group. Post hoc visualization relied on correlations between connectivity levels and posttreatment PARS rating for each of the two ABMT groups. The Fisher r-to-z transformation test was used to test for significant ABMT group differences between correlation coefficient magnitudes.