The neural dynamics of deficient memory control in heavily traumatized refugees

Victims of war, torture and natural catastrophes are prone to develop posttraumatic stress disorder (PTSD). These individuals experience the recurrent, involuntary intrusion of traumatic memories. What neurocognitive mechanisms are driving this memory disorder? Here we show that PTSD symptoms in heavily traumatized refugees are related to deficits in the effective control of memory retrieval. In a think/no-think task, PTSD patients were unable to forget memories that they had previously tried to suppress when compared to control participants with the same trauma history but without PTSD. Deficits in voluntary forgetting were clinically relevant since they correlated with memory intrusions in everyday life. Magnetoencephalography (MEG) recorded during suppression attempts revealed that PTSD patients were unable to downregulate signatures of sensory long-term memory traces in the gamma frequency band (70–120 Hz). Thus, our data suggest that the inability to suppress unwanted memories through modulation of gamma activity is related to PTSD symptom severity.


Supplementary behavioral data
In case a stimulus was endorsed as old in the item recognition task, a blank screen followed for 500 ms after which subjects were asked for their source rating. To this end, the letters L (left) and R (right) appeared on the respective positions on the screen for up to 3000 ms and participants had to specify on which side of the centrally presented door the target was shown during study by pressing a corresponding button within display time.
The source memory test proved unsuccessful since the obtained source memory judgements to recognition hits did not differ from 50 % chance level in all three conditions This was true also if we tested for this separately in both groups (Controls: ts(12) ≤ 0.739, ps ≥ 0.474, ds ≤ 0.151; PTSD: ts(10) ≤ 1.567, ps ≥ 0.148, ds ≤ 0.320). Furthermore, there was no group difference between Controls and PTSD patients in neither of the conditions (ts(22) ≤ 1.032, ps ≥ 0.313, ds ≤ 0.421). Thus, further analyses of behavioral and MEG data involving source memory performance were considered unreliable and therefore omitted. This observation is possibly due to symptoms of overgeneral episodic memory in individuals affected by trauma and depression 1 , making it difficult for our participants to retrieve specific details of previous experiences.  In order to test whether these results replicate our previous findings in healthy students 2,3 , we followed up on these effects. ERFs and theta power for the NT condition were

Gamma power in the T condition
We also tested for a modulation of gamma power with repeated attempts to retrieve the desired memory representation in the T condition. We obtained a non-significant positive cluster at left-hemispheric sensors spanning from 950 to 1200 ms in a two-way interaction analysis with the factors Run x Group within the T condition (T sum = 131.672, P corr = 0.059; see Supplementary Data and Fig. S3).
In order to get an idea about how to explain lower memory performance in the T condition in PTSD patients we also followed up on the non-significant two-way interaction effect in the T condition across runs (see Fig. S3a). Gamma power in the T condition The most pronounced difference in source activity was observed in the left precentral cortex (MNI coordinates: -54, 0, 20). In addition, we observed widespread significant clusters in the left hemisphere spanning from the frontal gyri to the post-central gyrus, the insula, the middle cingulum and occipital, parietal, and inferior, middle and superior temporal areas that are involved in memory retrieval and visual processing (see Fig. S3c).

Effects in the alpha frequency band
To even more specifically follow up on previous studies, in particular with respect to the neural markers of inhibitory control, we calculated Group x Condition interaction effects for the alpha (10-14 Hz) frequency band 3,8 . No significant effects emerged in this analysis (T sum < 28.2978, P corr > 0.7322) and the same was true for a three-way interaction between Run x Group x Condition (T sum < 201.503, P corr > 0.1873)