Altered predictive control during memory suppression in PTSD

Individuals with post-traumatic stress disorder (PTSD) often exhibit maladaptive avoidance behaviors, stemming from an aberrant anticipation of future threats. This study investigates how this predictive processing impairment affects the control of memory, a key function of the dorsolateral prefrontal cortex (DLPFC). Previous research suggested a generalized dysfunction of the inhibitory control system in PTSD, impacting the regulation of unwanted memories. However, the role of prediction in this disruption remained unclear. The researchers hypothesized that the inhibitory control of memories relies on predictive inferences, and that the interplay between predictive and reactive control processes is crucial for understanding PTSD pathogenesis. They posited that in PTSD, the probability of intrusive re-experiencing (beliefs) might be overemphasized, leading to excessive predictive avoidance that overrides online memory signals. Alternatively, the reduced inhibitory control could be limited to reactive processes that address the online emergence of intrusive memories, due to heightened sensitivity to prediction errors (PE). The study aimed to track these hidden computations during a think/no-think (TNT) memory suppression task using meta-Bayesian modeling, analyzing their influence on memory control connectivity markers. The same participants from a previous study—exposed (with and without PTSD) and non-exposed individuals to the 2015 Paris terrorist attacks—were used. The goal was to determine whether predictive and reactive control mechanisms are differentially impacted in PTSD, and whether these computational processes modulate the connectivity between the inhibitory control system and memory regions.

Existing literature highlights the importance of aberrant predictive processing in PTSD, specifically concerning threat anticipation and avoidance behaviors. Bayesian models provide a framework for understanding these predictive failures, suggesting that maladaptive associations between environmental cues and threatening outcomes can compromise accurate prediction of aversive events. Studies have also indicated a potential generalized dysfunction in the inhibitory control system of PTSD patients. This dysfunction, manifested in the inability to suppress unwanted memories, might be rooted in the disruption of mechanisms normally involved in active forgetting. A previous study by the same research group demonstrated compromised downregulation of intrusive memories in PTSD individuals during a memory suppression task, with reduced functional coupling between control and memory brain networks. However, the computational underpinnings of this connectivity disruption, particularly the role of predictive processing, remained unexplored. The current study bridges this gap by integrating computational modeling with brain connectivity analyses to examine the interplay between predictive and reactive control in memory suppression among PTSD patients.

The study employed a think/no-think (TNT) memory suppression task within an fMRI setting. Participants first learned neutral word-object pairs. In the TNT task, they were presented with cue words and instructed to either freely recall (think trials) or suppress (no-think trials) the associated images. Brain activity was recorded during the suppression phase. Participants rated the presence or absence of intrusive memories after each trial. Three computational models of increasing complexity were used to track beliefs about upcoming intrusions and the resulting prediction errors (PE): the Rescorla-Wagner (RW) model, the Kalman filter (KF), and the two-level hierarchical Gaussian filter (HGF). These models assumed different sources of beliefs (state, item, or combined) based on trial history or item-specific memories. Model validation involved falsification, belief trajectory recovery, model recovery, and parameter recovery analyses to ensure reliability and validity. The best-performing model (HGF) was used to estimate trial-by-trial beliefs and PEs. Dynamic causal modeling (DCM) was then used to investigate the influence of these estimates on effective connectivity between the inhibitory control system (anterior and posterior middle frontal gyrus, MFG) and memory regions (rostral and caudal hippocampus, rHIP and cHIP, and precuneus, PC). Forty-two DCM models were built, encompassing different hypotheses about top-down versus bottom-up modulation of control by beliefs and PEs. Bayesian model selection (BMS) and averaging (BMA) were used to compare model fits and extract group-specific connectivity parameters. Finally, correlations were computed between the imbalance in predictive and reactive control and PTSD symptom severity, controlling for total symptom severity and accounting for transdiagnostic anxiety and affect-related symptoms.

Model validation revealed that only the HGF model accurately simulated intrusion beliefs and reliably distinguished beliefs formed from trial or item history, or a combination of both. Bayesian model selection indicated that, across all three groups (PTSD+, PTSD−, and non-exposed), beliefs were best explained by a combined state-item model. DCM analysis showed overwhelming evidence that beliefs and positive prediction errors (PE+) modulated top-down control from the MFG to memory regions. Crucially, individuals with PTSD showed a disproportionately negative coupling between the MFG and the hippocampus during predictive control compared to reactive control, while non-exposed individuals and individuals with PTSD but no avoidance symptoms exhibited a balanced response. This imbalance in PTSD was specifically associated with avoidance and intrusion symptoms, but not with general anxiety or negative affect. The imbalance in predictive and reactive control wasn't a simple matter of one being disrupted while the other remained unaffected. A geometric interpretation, treating predictive and reactive control as orthogonal forces, showed a significant imbalance favoring predictive control in the hippocampus for the PTSD+ group, more pronounced than in the other groups.

The findings suggest that altered predictive processing is a central mechanism in PTSD, linking impairments in memory function and inhibitory control. In PTSD, the brain prioritizes predictions (beliefs) about future intrusions over online error correction (PE+), leading to excessive predictive control and reduced reactive control. This imbalance specifically correlates with avoidance and re-experiencing symptoms of PTSD, not general symptoms of anxiety or negative affect, which further highlights the specific role of predictive control in the pathological mechanisms of PTSD. The lack of correlation between predictive and reactive control suggests that these are independent processes, rather than a single continuum. The observed imbalance, even when considered as orthogonal forces, suggests two distinct yet coordinated mechanisms within the same neurobiological system, leading to a disruption in the ability to optimally balance memory suppression.

This study provides compelling evidence for a novel mechanism underlying PTSD, highlighting the crucial role of disrupted predictive processing in the pathophysiology of the disorder. The imbalance in predictive and reactive control of memory processing, particularly in the hippocampus, specifically contributes to the core symptoms of avoidance and intrusive re-experiencing. Future research should explore the developmental trajectory of this predictive control imbalance and its contribution to traumatic memory formation and maintenance. Furthermore, interventions targeting the restoration of this balance, possibly through therapies focused on modifying beliefs and enhancing reactive control, may offer promising avenues for PTSD treatment.

The study was conducted with a specific sample population exposed to a single traumatic event. The generalizability of these findings to other trauma types and populations requires further investigation. The reliance on self-reported intrusion ratings introduces subjective bias. Although efforts were made to address potential confounds such as the influence of general anxiety and negative affect on the observed imbalance, future studies could employ alternative methodologies such as physiological measures to further validate the findings. Furthermore, the study investigated the mechanisms underpinning intrusive memories and their avoidance, but did not explicitly address the formation and persistence of traumatic memory traces. Future research would need to more directly investigate this relationship.