Psychologically traumatic events can have complex long-term effects on mental and physical health as well as impair personal and occupational functioning (Disner et al., Reference Disner, Kramer, Nelson, Lipinski, Christensen, Polusny and Sponheim2017; Hoge et al., Reference Hoge, Terhakopian, Castro, Messer and Engel2007; Pizarro et al., Reference Pizarro, Silver and Prause2006). Military populations, in particular, experience a variety of life-threatening events that put them at an elevated risk for stress-related psychopathology. Returning US military personnel from Operations Enduring Freedom, Iraqi Freedom, and New Dawn endorsed high rates of exposure to explosive blasts in addition to posttraumatic stress disorder symptomatology (PTSS; Hoge et al., Reference Hoge, Castro, Messer, McGurk, Cotting and Koffman2004, Reference Hoge, McGurk, Thomas, Cox, Engel and Castro2008; Schell & Marshall, Reference Schell and Marshall2008). Consequently, mild traumatic brain injury (mTBI) and posttraumatic stress disorder (PTSD) have been described as signature injuries of these conflicts (Hoge et al., Reference Hoge, McGurk, Thomas, Cox, Engel and Castro2008; Sayer, Reference Sayer2012; Stein et al., Reference Stein, Kessler, Heeringa, Jain, Campbell-Sills, Colpe and Ursano2015; Warden, Reference Warden2006). A diagnosis of PTSD is often confounded by mTBI, which complicates attempts to isolate clinical complaints to either PTSS or the after-effects of physical injuries, particularly those sustained during traumatic explosive blast events (Harvey & Bryant, Reference Harvey and Bryant1998). In line with the theme of this special issue, we present findings identified through the lens of dimensional personality assessment to better understand the separable effects of PTSS and blast-induced mTBI (bmTBI) on neurophysiological responses among military veterans. We aimed to investigate candidate brain mechanisms that may explain co-occurring cognitive and emotional dysregulation specific to PTSS, taking into account the effects of bmTBI. Understanding the neurobiological underpinnings of maladaptive behavior after trauma could advance interventions that promote restoration of personal and occupational functioning and facilitate reintegration into civilian society.

Examining dimensions of symptomatology and personality in traumatized populations allows for a differentiation of how various aspects of self-reported experience map onto abnormal brain responses. Emerging research focusing on “subthreshold” PTSD (i.e., symptoms not meeting a full-threshold clinical diagnosis) has provided important insights into affective dysregulation, behavioral dysfunction, and suicide risk observed in traumatized populations (Cukor et al., Reference Cukor, Wyka, Jayasinghe and Difede2010; Jakupcak et al., Reference Jakupcak, Conybeare, Phelps, Hunt, Holmes, Felker, Klevens and McFall2007, Reference Jakupcak, Hoerster, Varra, Vannoy, Felker and Hunt2011; Marshall et al., Reference Marshall, Olfson, Hellman, Blanco, Guardino and Struening2001; Zlotnick et al., Reference Zlotnick, Franklin and Zimmerman2002). It is therefore justified to conceptualize the psychological after-effects of trauma as existing on a spectrum of severity without clear cut-offs between sick and well (Forbes et al., Reference Forbes, Haslam, Williams and Creamer2005; Ruscio et al., Reference Ruscio, Ruscio and Keane2002). Statistical modeling of a dimension of PTSS may better reflect the range of possible maladaptive responses to trauma, and enhance statistical power to detect associations with other related aspects of psychopathology (Grove, Reference Grove1991). Furthermore, dimensional models of PTSS appear capable of uncovering the associations with neurobiological systems sometimes not observable when using categorical diagnoses alone (Disner, Marquardt, Mueller, Burton, & Sponheim, Reference Disner, Marquardt, Mueller, Burton and Sponheim2018; Lieberman, Gorka, Funkhouser, Shankman, & Phan, Reference Lieberman, Gorka, Funkhouser, Shankman and Phan2017; Marquardt et al., Reference Marquardt, Goldman, Cuthbert, Lissek and Sponheim2018; Moran, Reference Moran2016).

In this investigation, we aimed to more precisely characterize neural abnormalities associated with PTSS and blast exposure by using scales from the Minnesota Multiphasic Personality Inventory-2–Restructured Form (MMPI-2-RF; Tellegen & Ben-Porath, Reference Tellegen and Ben-Porath2011). Two sets of indices composed of similar items – the Personality Psychopathology Five–Restructured Form (PSY-5-RF; Harkness et al., Reference Harkness, McNulty, Finn, Reynolds, Shields and Arbisi2014) and the Higher Order scales (H-O; Sellbom, Ben-Porath, & Bagby, Reference Sellbom, Ben-Porath and Bagby2008; Tellegen & Ben-Porath, Reference Tellegen and Ben-Porath2011) invoke personality frameworks from different traditions (to aid readability, see Table 1 for a list of acronyms). PSY-5-RF uses a structure similar to normative personality instruments developed using exploratory dimensionality reduction techniques (Harkness, McNulty, & Ben-Porath, Reference Harkness, McNulty and Ben-Porath1995). By organizing items into five independent groupings, the scales resemble a clinical version of the commonly applied “Big Five” factor model of personality (e.g., Costa & McCrae, Reference Costa and McCrae1985). These include internalizing traits such as negative emotionality/neuroticism (NEGE-r) and introversion/low positive emotionality (INTR-r); externalizing traits such as aggressiveness (AGGR-r) and disconstraint (DISC-r); and disorganized thought processes reflective of psychoticism (PSYC-r). In contrast, the H-O scales were derived through an examination of the higher-order structure of the constituent Restructured Clinical (RC) scales within MMPI-2-RF (Ben-Porath, Reference Ben-Porath2012; Sellbom et al., Reference Sellbom, Ben-Porath and Bagby2008). Self-report on these items is organized into three core groupings: emotional/internalizing dysfunction (EID), behavioral dysfunction (BXD), and thought dysfunction (THD). A three-factor H-O structure resembles the higher-order three-factor structure of psychiatric diagnoses (Kotov et al., Reference Kotov, Ruggero, Krueger, Watson, Yuan and Zimmerman2011), and presents a more broad characterization of current functioning compared to PSY-5-RF scales. At the same time, the H-O scales can be further expanded into constituent RC scales for more detailed characterizations.

Table 1.

Acronyms and their definitions

The PSY-5-RF and H-O scales provide complementary frameworks for understanding PTSS-related expressions of internalizing and externalizing dysfunction. Across various studies of personality, PTSD has been most strongly linked to high neuroticism, followed by low conscientiousness and low extraversion (Kotov, Gamez, Schmidt, & Watson, Reference Kotov, Gamez, Schmidt and Watson2010). Variability in these or similar dimensions of personality may help explain differences across people in terms of their PTSS and psychiatric comorbidities (Miller, Reference Miller2003; Miller, Greif, & Smith, Reference Miller, Greif and Smith2003; Miller et al. Reference Miller, Wolf, Reardon, Greene, Ofrat and McInerney2012). Many individuals with posttraumatic stress commonly endorse elevated negative emotionality (Miller, Kaloupek, Dillon, & Keane, Reference Miller, Kaloupek, Dillon and Keane2004). However, those with greater internalizing symptoms (e.g., mood disorders) also report lower positive emotionality/extraversion, while those with externalizing symptoms (e.g., disruptive substance use) report reduced constraint/conscientiousness. For example, NEGE-r was the primary PSY-5-RF index separating military veterans with PTSD from comparison controls in one post-deployment sample (Arbisi, Polusny, Erbes, Thuras, & Reddy, Reference Arbisi, Polusny, Erbes, Thuras and Reddy2011). Furthermore, EID and all of its component RC subscales (RCd – demoralization; RC2 – low positive emotions; and RC7 – dysfunctional negative emotions), along with RC4 (antisocial behavior), distinguished PTSD from controls (Arbisi et al., Reference Arbisi, Polusny, Erbes, Thuras and Reddy2011). In particular, RC7 (dysfunctional negative emotions), which includes items about intrusive thoughts, rumination, and nightmares, has been replicated as an important distinguishing index (Wolf et al., Reference Wolf, Miller, Orazem, Weierich, Castillo, Milford, Kaloupek and Keane2008). It may be that distress and generalized impairment commonly observed with bmTBI may be accounted for by emotional disruptions associated with comorbid elevations in PTSS. To our knowledge, no study to date has used the PSY-5-RF or H-O scales to explain PTSS-related cognitive and neurophysiological dysfunction using functional magnetic resonance imaging (fMRI).

Findings regarding the physiology of emotion regulation provide a neuroanatomical framework for understanding anxiety and traumatic stress-related cognitive impairments. Such perspectives (e.g., Etkin, Büchel, & Gross, Reference Etkin, Büchel and Gross2015; Gross, Reference Gross2015) posit that brain regions central to the generation of negative affect are modulated through pathways of model-based mechanisms of explicit regulation or model-free mechanisms of implicit regulation (Figure 1). The dorsolateral prefrontal cortex (dlPFC) may take an explicit role in regulating anxiety by changing a person’s model (i.e., their understanding) of threat (e.g., from “bad for me” to “good for me”; Corbetta & Shulman, Reference Corbetta and Shulman2002). Furthermore, a high cognitive load in various attentional control paradigms has been strongly associated with dlPFC activation (Curtis & D’Esposito, Reference Curtis and D’Esposito2003; Tsuchida & Fellows, Reference Tsuchida and Fellows2008). DlPFC may assist with compensatory processing during states of anxiety by actively inhibiting responses to distractors, shifting of attention, and updating working memory. Eysenck’s attentional control theory suggests that threat processing may become disruptive in certain circumstances when efforts to disengage using explicit attentional control depletes the same cognitive resources required for other relevant tasks (Eysenck, Derakshan, Santos, & Calvo, Reference Eysenck, Derakshan, Santos and Calvo2007). In line with this, individuals with PTSS commonly report hyperactive threat processing in daily life. Consequentially, they must use goal-directed attentional systems to manage the psychological impact of trauma-related cues from their external (e.g., unpleasant reminder images) and internal (e.g., personal worries, negative evaluations, memories of trauma) environments. These individuals may feel particularly compromised when other tasks in their lives require use of those same cognitive inhibition and shifting abilities (Berggren & Derakshan, Reference Berggren and Derakshan2013; Miyake et al., Reference Miyake, Friedman, Emerson, Witzki, Howerter and Wager2000). In other words, initial maladaptive processing of task-irrelevant threat and overburdened compensatory responses may explain some aspects of cognitive dysfunction among individuals with elevated PTSS.

Figure 1.

Neuroanatomical framework of emotional regulation with study regions of interest.

Note: On the leftmost column, left dlPFC depicted at x = −36, y = 44, z = 22 (top left); left amygdala depicted at x = −24, y = −3, z = −18; vmPFC-sgACC depicted at x = 0, y = 22, z = −14. A systems neuroscience model of explicit and implicit emotion regulation is displayed for the study regions of interest adapted from (Etkin et al., 2015). Red outlines and arrows suggest possible influences of anxiety in accordance with attentional control theory (Eysenck et al., 2007). vmPFC = ventromedial prefrontal cortex, sgACC = subgenual anterior cingulate cortex, dlPFC = dorsolateral prefrontal cortex.

For tasks that require little cognitive effort, explicit regulatory compensation may not be employed. However, anxiety-related deficits may still be observable via other neural mechanisms. For example, Fales and colleagues (Reference Fales, Barch, Burgess, Schaefer, Mennin, Gray and Braver2008) observed that greater anxious symptomatology was associated with broadly reduced neural activity at rest. Implicit regulatory regions such as vmPFC-sgACC have also been posited to regulate emotional responding in a model-free (i.e., experience-based) manner. These areas also play a role in generating and updating inhibitory responses through new perceptions of stimuli (e.g., Eysenck et al., Reference Eysenck, Derakshan, Santos and Calvo2007; Öhman & Mineka, Reference Öhman and Mineka2001), and are implicated as updating abilities appear to be consistently disrupted in PTSD. These results also suggest a greater cognitive resource utilization with increased anxiety, making it conceivable that dlPFC-mediated compensatory activity may influence vmPFC-sgACC activity. The amygdala has a well-validated role in attentional capture from affective information (LeDoux, Reference LeDoux2012; Öhman, Flykt, & Esteves, Reference Öhman, Flykt and Esteves2001; Pessoa & Ungerleider, Reference Pessoa and Ungerleider2004; Vuilleumier, Reference Vuilleumier2005; Vuilleumier & Huang, Reference Vuilleumier and Huang2009), and contributes to the expression of anxiety. It would be expected that higher levels of anxiety and presentation of threat cues would be associated with increased amygdala activation.

In the current study, we examined the associations for dimensional measures of PTSS and bmTBI with brain responses within an emotion regulation neural system composed of dlPFC, vmPFC-sgACC, and amygdala during an N-back task involving the manipulation of cognitive and affective load. We used a multiply-mediated moderation framework to account for brain activity with respect to personality dysfunction using the PSY-5-RF and H-O scales. The affective manipulation within the N-back task was anticipated to partially deplete goal-directed attentional resources necessary for disengagement from threatening images. In this compromised cognitive state, individuals with elevations in PTSS were expected to be inefficient at inhibiting their automatic processing of threatening images.

2. Combat N-back task protocol

The N-back task stimuli consisted of single letters centered on a screen superimposed over task-irrelevant neutral or combat background images (Figure 2). Participants were tasked with identifying target letters during counter-balanced manipulations of cognitive load (0-back vs. 2-back) and affective content (neutral vs. combat images). During 0-back trials, participants pressed a response button when designated target letters appeared (e.g., Target = “A,” FAHRALPKAQ). For the 2-back condition, participants indicated when a sequentially presented letter was identical to the letter presented two screens before (e.g., GLPLFGNRNR). Low-arousal and intermediate-valence (i.e., neutral pleasantness) background images were selected using the International Affective Picture System (IAPS; Lang, Bradley, & Cuthbert, Reference Lang, Bradley and Cuthbert2005) based on published ratings (IAPS identifiers 2383, 2393, 2880, 7050, 7080, 7175, 7205, 7224, 7550, 7705). Ten images of aversive Operation Iraqi Freedom–related combat scenes were also selected from a larger set of stimuli used in a previous study of post-deployment functioning (Marquardt et al., Reference Marquardt, Goldman, Cuthbert, Lissek and Sponheim2018). These images depicted scenes with threatening enemy combatants, civilian injuries, and roadside bombings. Participants previously rated these combat scenes as highly arousing and unpleasant.

Figure 2.

Affective N-back task design.

Note: The emotional N-back task design of the study. Example of trials from the 2-back condition (left). Examples of the combat and neutral affective background image conditions are displayed using substitute images from freely available online sources (right).

N-back letters and background images were presented simultaneously for 1000 ms followed by inter-trial intervals of 1100 ms with centrally located crosshairs. Participants were allowed to respond until the onset of the next stimulus, but were asked to make their selections as quickly as possible while ignoring the background images. Task trials advanced regardless of the responses provided by participants to prevent a possible negative reinforcement for quicker button presses. Task trials were administered in blocks of 10 with 2–4 target trials per block and 32 total blocks administered across two separate runs. Overall, participants viewed eight total blocks from each combination of experimental manipulations, and experienced 25 0-back neutral, 26 2-back neutral, 25 0-back combat, and 24 2-back combat target trials interspersed with non-target trials in those same blocks. Non-target trials did not necessitate responses. Stimulus delivery was controlled by E-Prime software (Psychology Software Tools, 2012, Sharpsburg, PA).

2.1. Functional MRI

Image acquisition. All fMRI scans were conducted at the Center for Magnetic Resonance Research at the University of Minnesota on a Siemens Trio 3 T scanner (Siemens, Erlangen, Germany) with a 32-channel receive-only phased array radio frequency head coil. Anatomical scans with 1 mm isotropic resolution (224 coronal slices) were obtained using a T1-weighted MPRAGE sequence (TR = 2530 ms, TE = 3.65 ms, T1 = 1100 ms, flip angle = 7 degrees, FOV = 176 × 256 mm). Functional scans were gradient echo EPI images consisting of 348 volumes with 2 mm isotropic resolution, each with 64 slices (TR = 1320 ms, TE = 30 ms, flip angle = 90 degrees, multiband factor = 4, matrix = 106 × 106, FOV = 212 mm, 2 mm thick).

Preprocessing. Image analysis was performed using Analysis of Functional Neuroimages (AFNI) software (Cox, Reference Cox1996). Each subject’s data were motion-corrected such that all subsequent volumes of both N-back runs were registered to the first volume of first N-back run. Data were smoothed with a Gaussian blur of 4.5 mm full-width-at-half-maximum using AFNI’s 3dBlurToFWHM. Distortion reduction was achieved using the top-up command from the FMRIB Software Library (FSL) and an EPI scan with identical parameters to those of the task, but with the opposite phase-encoding direction and thus the opposite pattern of distortion (Andersson et al., Reference Andersson, Skare and Ashburner2003; Smith et al., Reference Smith, Jenkinson, Woolrich, Beckmann, Behrens, Johansen-Berg and Matthews2004). Within the framework of a general linear model, blood oxygen level-dependent (BOLD) responses were analyzed at the individual subject level to produce separate beta weights for each of the 0-back neutral, 0-back combat, 2-back neutral, and 2-back combat block conditions. These beta weights were treated as dependent variables in the subsequent multi-subject analyses. Six motion parameters, five degrees of Legendre polynomials to account for baseline drift, and the instructions subjects read during the scan were modeled in GLM as regressors of no interest. Motion parameters were used to compute the Euclidean norm (Enorm) of change in head position from one volume to the next. Volumes with Enorm values >.5 were censored from GLM. Clusters of activation data were warped to the MNI space prior to a region-of-interest (ROI) analysis (Evans et al., Reference Evans, Collins and Milner1992).

Regions of interest. Primary data for the present analyses included brain responses from five distinct brain areas: right and left amygdala, right and left dlPFC, and vmPFC-sgACC. These ROIs were selected to characterize activity within neural structures involved with affective responding and cognitive control processes. Masks of these ROIs were then used to extract parameter estimates from the combat N-back fMRI task. Bilateral amygdala and vmPFC-sgACC ROIs were defined using the Harvard–Oxford MNI probabilistic atlas. ROIs for left dlPFC (−36, 44, 22) and right dlPFC (34, 44, 32) were defined as 5-mm spheres centered upon coordinates from a meta-analysis of fear conditioning (Fullana et al., Reference Fullana, Harrison, Soriano-Mas, Vervliet, Cardoner, Àvila-Parcet and Radua2016).

2.2. Data analysis

Analyses were conducted using a series of evolving mixed-effects multilevel path models in Mplus 6 (Muthén & Muthén, Reference Muthén and Muthén2019). Analyses started from a basic task effect model to a task effect model moderated by CAPS and MN-BEST scores, before ending with a task effect model with tests of mediated moderation from CAPS and MN-BEST scores through personality dimensions. Participants were analyzed at level two with their neural activity for each of the four task conditions nested at level one. Predictor variables were z-scored before being included in the models to produce standardized betas as estimates, which can be interpreted as a measure of the size of the effect (Lorah, Reference Lorah2018).

Model 1: Task effect model. To examine the task effects on neural activity irrespective of individual differences measures, we estimated a mixed-effect multilevel path model predicting left/right amygdala, left/right dlPFC, and vmPFC-sgACC activity for each of the 0-back neutral, 2-back neutral, 0-back combat, 2-back combat blocks. To create a 2-by-2 factorial design, dummy variables were created for a cognitive load factor (0 = 0-back, 1 = 2-back) and an affect factor (0 = neutral, 1 = combat). Consequently, the 0-back neutral condition was included as the model intercept (β0) with additional fixed effects of cognitive load (β1), affect (β2), and cognitive load-by-affect interaction (β3). Individual variance components were also included by estimating random effects (intercept and slopes) for each of the predictors in the model.

Model 2: Moderation model. We tested for moderation as a function of individual differences in bmTBI severity and PTSS effects across task manipulations. Fixed effects from model 1 were included as well as MN-BEST blast severity, CAPS total severity scores, and their interaction as independent predictors and moderators of those fixed effects.

Models 3a and 3b: Mediated moderation models. Using multiply-mediated moderation, we modeled the degree to which bmTBI and PTSS moderation effects on neural responding could be explained by MMPI-2-RF H-O scores. Mediating variables included EID, THD, and BXD scale scores. When a significant direct effect was observed between MN-BEST or CAPS severity variables and an H-O scale, we planned to remove that particular H-O scale and substitute in its component RC scales in a separate, follow-up model (model 3b). Indirect effects for models 3a and 3b were estimated by testing the product of A and B paths. Partial-versus-full mediation was determined by examining whether or not the C’ path (i.e., the direct pathway of original moderator after co-varying for candidate mediators) remained significant following the identification of a significant mediator. Mediation models were also examined for non-traditional mediation effects such as complementary mediation, competitive mediation, indirect-only mediation, and direct-only mediation (Zhao et al., Reference Zhao, Lynch and Chen2010). We applied family-wise Bonferroni correction to model 3b to control the type I error rate. Families were defined as groups of closely related questions, a list of which can be found in Table S16 of Supplemental Materials.

Parallel analyses using PSY-5-RF in place of H-O indices demonstrated convergent results. Given that the H-O scales were expandable by their RC scales, allowing for potentially greater descriptive resolution, we describe the results for H-O scales here, while PSY-5-RF analyses are detailed in Supplemental Materials (see Model S3).

3. Results

3.1. Demographics and clinical characteristics

Of the 93 participants (M_age = 34.35, SD_age = 8.36) analyzed within the fMRI models, 90 (96.77%) were male and 85 (91.40%) were non-Hispanic/white. Clinically, nine (9.68%) met DSM-IV-TR criteria for PTSD and not bmTBI; 25 (26.88%) met the criteria for bmTBI only and not PTSD; seven (7.53%) met the criteria for both PTSD and bmTBI; and 52 (55.91%) did not meet the criteria for either PTSD or bmTBI. However, of the latter group of 52, 17 had at least once met the criteria for PTSD in their lifetime, and nine met the criteria for subthreshold PTSD (i.e., meeting the criteria for some but not all PTSD symptom domains), indicating that these categories masked discernible impairments better detected with the CAPS dimensional symptom measure. The 52 who did not meet the criteria for either diagnosis exhibited average MN-BEST impact severity ratings of 2.17 (SD = 2.41) and current CAPS severity ratings of 21.83 (SD = 13.49), indicative of modest blast exposure and mild-moderate subthreshold PTSD symptomatology, on average. Descriptive statistics of CAPS and MN-BEST for the full sample are shown in Table 2.

Table 2.

Sample statistics of CAPS and MN-BEST scores (N = 93)

Note: ^asubset of individuals exposed to an explosive blast (n = 49); CAPS: Clinician Administered PTSD Scale, MN-BEST: Minnesota Blast Exposure Screening Tool

3.2. Task performance

A performance path model (Table S1) was first run to examine the effects of cognitive load and the effects of our affect manipulation using combat image content on performance indices (i.e., d-prime and reaction time). In sum, cognitive load significantly decreased d-prime (β = −0.504, p < .001) and increased reaction times (β = 0.471, p < .001), whereas combat images significantly interacted with the levels of cognitive load to predict d-prime (β = −0.471, p = .001) and reaction times (β = −0.23, p = .001). Re-estimating models at each level of cognitive load, simple effects analysis revealed that combat images decreased d-prime only under cognitive load (β = −0.298, SE = 0.094, p = .002, 95% CI = [−0.482, −0.114]), and increased reaction times only when cognitive load was absent (β = 0.357, SE = 0.073, p < .001, 95% CI = [0.213, 0.501]). We followed this with a performance moderation path model (Table S2) to examine the moderation of these effects by PTSS and bmTBI severity. In sum, PTSS increased the reaction times to combat images (β = 0.133, p = .018) and predicted significant changes in d-prime to cognitive load-by-affect interaction (β = −0.329, p = .008). Simple effects revealed that PTSS predicted decreased d-prime in the 0-back neutral condition (β = −0.459, SE = 0.193, p = .017, 95% CI = [−0.838, −0.081]), but not in the 2-back neutral condition (β = 0.052, SE = 0.216, p = .809, 95% CI = [−0.372, 0.477]). Performance indices failed to show moderation by bmTBI. Fuller statistical details of these models are provided in Table S3 of Supplemental Materials.

3.3. Effects of cognitive load and affective picture background (model 1)

Model 1 fixed and random effect estimates are displayed in Tables 3, S4, and S5 as well as Figures 3A and 4A. Bilateral amygdala (left: β = 0.307, p = .001; right: β = 0.339, p <.001) and vmPFC-sgACC (β = 0.438, p <.001) exhibited activations in the 0-back neutral condition. These activations became significant deactivations for bilateral amygdala (left: β = −0.717, p < .001; right: β = −0.761, p < .001) and vmPFC-sgACC (β = −0.879, p < .001) under cognitive load. Bilateral dlPFC exhibited deactivations in the 0-back neutral condition (left: β = −0.257, p < .001; right: β = −0.329, p < .001) that became significantly activated under cognitive load (left: β = 0.538, p < .001; right: β = 0.699, p < .001). From the 0-back neutral to 0-back combat condition, bilateral amygdala (left: β = 0.272, p = .005; right: β = 0.199, p = .020) and vmPFC-sgACC (β = 0.134, p = .049) exhibited increased activations, while bilateral dlPFC activation was not significantly changed (p > .263). Finally, cognitive load-by-affect interactions were found for left amygdala (β = −0.337, p = .002) and vmPFC-sgACC (β = −0.264, p = .006). Simple effect estimates for affect manipulation were derived by estimating models for 0-back and 2-back conditions separately. In the 0-back condition, combat images were associated with significant increases in left amygdala activity (β = 0.252, SE = 0.096, p = .009, 95% CI = [0.064, 0.439]) and marginal increases in vmPFC-sgACC activity (β = 0.122, SE = 0.068, p = .074, 95% CI = [−0.012, 0.256]). In the 2-back condition, combat images were no longer significantly associated with left amygdala activity (β = −0.047, p = .580) and was associated with marginally decreased vmPFC-sgACC activity (β = −0.120, SE = 0.072, p = .099, 95% CI = [−0.262, 0.022]).

Table 3.

Model 1 Estimates of task effects by region of interest

Note: Standardized beta values are displayed. 372 observations. ^†p < .10. *p < .05. **p < .01. ***p < .001. vmPFC = ventromedial prefrontal cortex, sgACC = subgenual anterior cingulate cortex, dlPFC = dorsolateral prefrontal cortex.

Figure 3.

Model 2: PTSS and bilateral amygdala activity effects.

Note: Means and standard errors are displayed. (Panel A) Functional magnetic resonance imaging (fMRI) activations within regions of interest during cognitive load (0-back, 2-back) and affective (neutral, combat) manipulations. (Panel B) Moderation effects are displayed for individuals +1 and −1 standard deviations on posttraumatic stress disorder symptomatology (PTSS). PTSS moderated associations of left and right amygdala, and vmPFC-sgACC via blunted activation during the 0-back neutral condition. During the 2-back neutral condition, these same individuals with high PTSS also produced attenuated deactivations within the left and right amygdala. Additionally, high PTSS was associated with reduced right dlPFC activation during the 2-back condition. vmPFC = ventromedial prefrontal cortex, sgACC = subgenual anterior cingulate cortex, dlPFC = dorsolateral prefrontal cortex, BOLD = blood-oxygen-level-dependent.

Figure 4.

Statistical model effects.

Note: Simplified path diagram for the evolving mixed-effects multilevel path statistical models. (4A) Significant within-subject effects for the task manipulations are depicted. (4B) Only significant moderating effects of posttraumatic stress symptom and mild brain injury severity are depicted. (4C) Only significant paths to study effects via RC scales are shown. RC2 partially mediates the CAPS Severity moderation of bilateral amygdala under cognitive load. RCd indirectly and competitively mediates right amygdala activity under cognitive load. Cog. = 2-back neutral image condition predictor, vmPFC = ventromedial prefrontal cortex, sgACC = subgenual anterior cingulate cortex, dlPFC = dorsolateral prefrontal cortex, PTSD = posttraumatic stress disorder, RC2 = Low Positive Emotions.

3.4. Moderation of brain activation by PTSD symptoms and history of bmTBI severity (model 2)

There were no significant interaction effects between PTSS and bmTBI severity, or between PTSS or bmTBI severity and the cognitive load-by-affect predictor for any ROI. There was also no significant association between PTSS and bmTBI severity. Therefore, all of these paths were removed from model 2 and analyses hereafter.

Estimates of model 2 fixed effects of PTSS and bmTBI severity to ROIs are displayed in Tables S6 and S7 and Figures 3B and 4B. Model 2 revealed that PTSS moderated bilateral amygdala, vmPFC-sgACC, and right dlPFC activity. Bilateral amygdala activity was also moderated by bmTBI. Specifically, greater PTSS was associated with less activity in bilateral amygdala (left: β = −0.259, p = .001; right: β = −0.283, p = .001) and vmPFC-sgACC (β = −0.182, p = .011) in the 0-back neutral condition, and was significantly associated with reduced bilateral amygdala deactivation (left: β = 0.295, p < .001; right: β = 0.304, p < .001) and right dlPFC activation (β = −0.150, p = .023) in the 2-back neutral condition. Greater bmTBI severity (while controlling for PTSS levels) was associated with reduced bilateral amygdala activity to combat images in the 0-back combat relative to the 0-back neutral condition (left: β = −0.201, p = .002; right: β = −0.205, p = .001).

3.5. Effect of personality on the relationship between PTSS and brain activations: Multiply-mediated moderation (models 3a and 3b)

Model 3a A-path fixed-effect estimates to H-O scales from PTSS and bmTBI are displayed in Tables S8 and S9, respectively. PTSS was associated with increased EID (β = 0.416, p < .001), while bmTBI severity was associated with decreased EID (β = −0.193, p = .022).

Estimates for B-path fixed effects from H-O scales to ROIs are displayed in Tables S10 and S11. These analyses revealed that greater EID was concurrently associated with significantly increased activity in bilateral amygdala under cognitive load (left: β = 0.220, p = .012; right: β = 0.185, p = .036) and increased activity in right dlPFC activity to combat images (β = 0.081, p = .020) over and above PTSS and bmTBI. Moreover, greater THD was associated with reduced vmPFC-sgACC activity in the 0-back neutral condition (β = −0.146, p = .028) and reduced left dlPFC activity to combat images (β = −0.053, p = .049) over and above PTSS and bmTBI.

Indirect effect estimates (shown from PTSS in Table S12 and from bmTBI in Table S13) showed that EID was found to significantly mediate PTSS-related cognitive load effects on left amygdala (β = 0.092, p = .041), and evidence that EID mediated the PTSS-related cognitive load effect on right amygdala was marginally significant (β = 0.077, p = .073).

Model 3a was re-estimated as model 3b after breaking EID into its component RC scales: RCd, RC2, and RC7. Analysis of A-paths (shown in Tables S8 and S9) revealed that increased PTSS was independently associated with increased RCd (β = 0.364, p < .001), RC2 (β = 0.387, p < .001), and RC7 (β = 0.393, p < .001) scores, while increased bmTBI severity was independently associated with decreased RCd (β = −0.199, p = .021) and RC2 (β = −0.272, p < .001) scores.

Analyses of B-paths (shown in Tables S10 and S11) revealed that RC2 was associated with increases in left (β = 0.372, p < .001) and right (β = 0.506, p < .001) amygdala activity under cognitive load over and above PTSS and bmTBI. B-paths were also initially found associating increased RCd with right amygdala deactivation under cognitive load, increased RC2 with vmPFC-sgACC activation under cognitive load, increased RC7 with vmPFC-sgACC activation under cognitive load, and increased RC7 with left amygdalar deactivation to combat images, but these paths did not survive Bonferroni correction (p’s > .085).

Indirect effect estimates revealed that RC2 was found to fully mediate PTSS-associated increases in left (β = 0.144, p = .004) and right (β = 0.196, p = .001) amygdala activity under cognitive load (Figure 4C and Table S12), as their corresponding C’ paths were not significant after Bonferroni correction (p’s > .205). RC2 was also found to be an indirect-only mediator from bmTBI severity to decreases in left (β = −0.101, p = .005) and right (β = −0.138, p = .002) amygdala activity under cognitive load (Figure 4C; Table S13). Indirect-only mediation was also initially found in PTSS to right amygdala deactivation under cognitive load through RCd, as well as PTSS to increased vmPFC-sgACC activity under cognitive load through both RC2 and RC7, but these paths did not survive Bonferroni correction (p’s > .080).