Participants
One hundred fourteen participants (79 TEs and 35 HCs) completed the study protocol (see Table 1 and Figure S4). Of the 79 TE participants, 41 had psychopathology at the level of a psychiatric diagnosis (TEPG), and 38 were trauma-exposed healthy controls with no diagnosis (TEHC; see Table 1 and Figure S4). Participants were recruited by local advertisements, websites, and word-of-mouth referrals and evaluated at the New York State Psychiatric Institute (NYSPI). The methods were performed in accordance with relevant guidelines and regulations and approved by NYSPI IRB. All participants provided written consent to take part in the study.
Behavioral ratings
This task consists of six types of stimuli including conditioned stimuli (CSs including CS+: danger cue; oCS-: o shape safe cue; vCS-: v shape safe cue) and three generalization stimuli (GSs including GS1, GS2, GS3). There were a string of colored crosshairs (blue, yellow, red, green, and purple) presented serially for a duration of 800 ms each, in a quasi-random order in the center of the viewing screen during 4 s presentation of each stimulus. The inter-trial interval (ITI) periods last 2.4 s (three crosshairs) or 4.8 s (four crosshairs). Participants were instructed to continuously monitor the stream of colored crosshairs and rate their perceived level of risk for shock as quickly as possible following each red cross using a three-button fiber optic response pad (Lumina LP-404 by Cedrus), where 0=‘no risk’, 1=‘moderate risk’ and 2=‘high risk’. Risk ratings were recorded with Presentation software (Neurobehavioral Systems). For half of CS/GS trials, one of five crosshairs was red, and the remaining trials included no red crosshairs. Additionally, on reinforced CS+ trials, the red crosshair never appeared in the fourth or fifth position to avoid interference from shock on behavioral responses. Finally, self-reported anxiety to CS+, oCS-, and vCS- were retrospectively assessed following the pre-acquisition, acquisition, and generalization phases using a 10-point scale. The behavioral risk rating analysis excluded non-learner participants, indicated by rating the vCS- higher than CS+ during task-related risk rating or the post-task questionnaire (28).
Design
The generalization paradigm included three phases (Figure S5): (i) pre-acquisition-consisting of 20 trials of each stimulus type (CS+, GS1, GS2, GS3, oCS- and vCS-), all presented in the absence of any shock US; (ii) acquisition-including 15 CS+, 15 oCS-, and 15 vCS-, with 12 of 15 CS+ co-terminating with shock (80% reinforcement schedule) and (iii) generalization-including an early generalization (EG) stage and late generalization (LG) stage, each comprised of 10 trials of each stimulus type (unreinforced CS+, GS1, GS2, GS3, oCS-, vCS-) and an additional 5 CS+ co-terminating with shock (33% reinforcement schedule) to prevent extinction of the conditioned response while leaving 10 unreinforced CS+ to index responses uninfluenced by the shock US. Trials of all three phases were arranged in quasi-random order such that no more than two stimulus-type of the same class occurred consecutively. An additional constraint for the generalization phase was the arrangement of trials into six blocks of 13 trials each (i.e., two unreinforced CS+, one reinforced CS+, two oCS-, two vCS-, two GS1, two GS2, two GS3) to ensure an even distribution of trial types throughout runs.
Image Acquisition and analysis
Seventy-seven participants were scanned using a 3T General Electric MR750, and 37 participants were scanned using a 3T General Electric PREMIER (GE Medical Systems, Waukesha, WI, USA) equipped with a 32-channel receive-only head coil. For each participant a high-resolution T1-weighted 3D BRAVO sequence was acquired using the following parameter: T1=450 mm, Flip angle =12°, field of view =25.6 cm, 256×256 matrix, slice thickness =1 mm. T2*-weighted echo-planar images (EPIs) depicting the blood-oxygen-level-dependent (BOLD) were acquired for each participant with TR=1.3 sec, TE=28 msec, FA =60°, FOV =19.2 cm, number of slices=27, slice thickness=4 mm. A head cushion was used to limit head motion.
fMRI Network Analysis
Group independent components analysis was performed using the GIFT toolbox (v3.0b, http://icatb.sourceforge.net), implemented as a MATLAB toolbox (Matlab 2020b, MathWorks Inc., Sherborn, MA, USA), to obtain functional networks that underlie fMRI data. The group spatial ICA is first performed on all participants at once, providing an independent component spatial map and a single associated ICA time course for every component, participant, and stage. Individual participants’ spatial maps and associated time courses corresponding to the group ICA maps were calculated, and significant between-group differences in the activity of the SN, ECN, and DMN networks during generalization were determined by a second-level analysis of the ICA results. To identify the components most involved in each trial type, a GLM analysis was performed. We examined the role of each ICN for each condition (CS-, GS, CS+ etc.) and how this differed according to trial type. These conditions were modeled using a GLM with the canonical hemodynamic response function (HRF) in SPM12 to examine the association between component time courses and different trial types. The resulting β-weights, a measure of each component’s trial-specific amplitude, were entered into statistical analysis to identify those components significantly more engaged in each condition.
Statistical Analysis
All statistical analysis was carried out in SPSS. Levels of conditioning were assessed with paired sample t-tests comparing behavioral risk ratings to CS+ vs. oCS-, and CS+ vs. vCS- in HC and TE separately during pre-acquisition, acquisition and generalization phases. The significance threshold for these behavioral analyses were set at p<.0167 to adjust for multiple comparisons using Bonferroni correction (corrected for 3 comparisons).
To examine the generalization phases over time (EG, LG), for both behavior and neural markers, we first assessed changes (delta) in behavioral and neural markers over the two stages (EG-LG) for trauma exposure by comparing TE and HC using a group (TE and HC) by stimulus-type (vCS-, oCS-, GS1, GS2, GS3 and unreinforced CS+) repeated measures ANOVAs, and then assessed the changes of behavior and neural markers over the two stages (EG-LG) of resilience by further comparing TEHC and TEPG with HC using a group (TEPG, TEHC, HC) by stimulus-type (vCS-, oCS-, GS1, GS2, GS3 and unreinforced CS+) repeated measures ANOVAs.
Next, we assessed the changes in steepness of the generalization gradients across early and late stages (i.e., changes/ delta of generalization magnitudes: EG-LG) measured by linear deviation scores (LDS) (26). LDS reflect the degree to which participant level gradients depart from linearity: LDS = ([CS+, CS-] ∕2) - [GS1, GS2, GS3] ∕3)), where [CS+, CS-] ∕2 reflects the theoretical, linear midpoint of the gradient, and [GS1, GS2, GS3] ∕3 the average response to GSs. This equation provides a single number index reflecting the steepness of generalization gradients, with larger values indicating stronger generalization. We then assessed the behavioral and neural markers of trauma exposure by measuring the delta of LDS across two stages using one-way ANOVA, and then assessed the behavioral and neural markers of resilience by further comparing TEPG and TEHC with HC using one-way ANOVA. Effects of covariate corresponding to different scanners, age and sex was used in all analysis as a covariate of no interest.
For assessing the neural markers of trauma exposure and resilience, the five intrinsic connectivity networks (ICN) meeting selection criteria were used (Table S1). These networks included the salience network (SN), left executive control network (LECN), right executive control network (RECN), anterior default mode network (a-DMN) and posterior default mode network (p-DMN). All neural imaging results were corrected for multiple comparison at p<0.01 (5 networks).