A Biologically Inspired Neural Network Model to Gain Insight Into the Mechanisms of Post-Traumatic Stress Disorder and Eye Movement Desensitization and Reprocessing Therapy

Abstract

Eye movement desensitization and reprocessing (EMDR) therapy is a well-established therapeutic method to treat post-traumatic stress disorder (PTSD). However, how EMDR exerts its therapeutic action has been studied in many types of research but still needs to be completely understood. This is in part due to limited knowledge of the neurobiological mechanisms underlying EMDR, and in part to our incomplete understanding of PTSD. In order to model PTSD, we used a biologically inspired computational model based on firing rate units, encompassing the cortex, hippocampus, and amygdala. Through the modulation of its parameters, we fitted real data from patients treated with EMDR or classical exposure therapy. This allowed us to gain insights into PTSD mechanisms and to investigate how EMDR achieves trauma remission.

Keywords: amygdala; computational modeling; eye movement desensitization and reprocessing therapy; post-traumatic stress disorder; prolonged exposure.

Figure 1

Model architecture: units and connections of the model. EMDR acts on the strength of the inhibition from the PFC (vmPFC in PE and dlPFC in EMDR) to the amygdala through the parameter ϕ, and on the learning rate of the connections between the sensory units and the PFC through the parameter ψ.

Figure 2

Effect of trauma establishment in hippocampus and amygdala, compared to the control. In the first trial, the baseline activity of the hippocampus (A) and amygdala (B) were measured during the stimulation of the V1 or the V2 sensory units. In trial 2, pattern 1 (A1-S1-V1) was coactivated with the trauma unit; as a control, pattern 2 (A2-S2-V2) was activated with the trauma unit turned off. During trial 3 and succeeding trials, V1 and V2 were repeatedly activated to investigate the dynamics of the hippocampus and amygdala response. After memory establishment in trial 2, V2 stimulation induces the activation of the hippocampus (A, white dots, trials 3–30), but not of the amygdala (B, white dots, trials 3-35). On the other hand, the trauma-associated stimulus V1 induces both hippocampus (A, black dots, trials 3–35) and amygdala (B, black dots, trials 3–35). Over time, the non-traumatic memory trace is lost (A, white dots, trial 35), while traumatic associated stimulus V1 is persistently capable to activate the hippocampus (A, black dots, trial 35) and amygdala (B, black dots, trial 35).

Figure 3

Model memory formation in PTSD and control conditions. (A) The graph corresponds to trial 3 of Figure 2 in the condition with the trauma. V1 activates the hippocampal unit H3 and the cortical units S1 and A1 encoding the other elements of pattern 1. (B) The graph corresponds to trial 35 of Figure 2 in the trauma condition. The memory trace continues to be activated by V1 even after many trials. (C) The graph corresponds to trial 3 of Figure 2 in the control condition with no trauma. V2 activates H2, a different unit of the hippocampus with respect to V1, and also the other cortical elements of the pattern, S2 and A2, although to a lesser extent with respect to PTSD [note the different scales in (A,C)]. (D) The graph corresponds to trial 35 of Figure 2 in the control condition. After 35 trials, the memory trace is lost and V1 does not elicit activations (the memory has been transferred to the cortex).

Figure 4

Effect of the high excitability of vmPFC on the triggering of post-traumatic stress disorder (PTSD). Hippocampus (A) and amygdala (B) activity before (trial 1: V1 input), during (trial 2: V1-A1-S1-Trauma input) and after (trial 3–35: V1 input) trauma, in a model where the vmPFC firing threshold is halved (white dots) compared to the model shown in Figure 2 acquiring the PTSD (black dots).

Figure 5

Effect a mild emotional engagement on memory retention. Hippocampus (A) and amygdala (B) activity with an activation protocol as in Figure 2, except that the vmPFC firing threshold is halved (model robust with respect to PTSD) and that during the second trial the emotional (Trauma) unit is weakly activated (1/10 of activation) to represent a mild (non-traumatic) emotional engagement of the amygdala.

Figure 6

Effect of PE on hippocampus (A) and amygdala (B). The first 10 trials were as in Figure 1: trial 1 was the baseline V1 activation of hippocampus and amygdala; trial 2 marked the trauma experience (A1-S1-V1-Trauma); trials 3–10 involved the re-experience of V1 to observe the effect of the trauma. After trial 10, however, we administered PE for 20 sessions (20 trials in the model). PE consisted in the activation of the hippocampal memory trace (unit H3 from the Figures 3A and B), coupled with the activation of vmPFC. Each time point in trials 11–30 corresponds to a V1-activation test after a PE session (trial). After the 20 trials of therapy, hippocampal activation is reduced to a plateau (A) and amygdala activation becomes zero (B): the memory has been freed from the negative emotional load.

Figure 7

Patients data fitting. (A) Root square MSE for the fitting of the PE data from Nijdam et al. (2012). (B) Root square MSE for the fitting of the EMDR data from Nijdam et al. (2012).

Figure 8

Eye movement desensitization and reprocessing therapy is compared to PE therapy. Each dot in (A,B) represents a simulation to fit the data of PE (A) and EMDR (B) in Nijdam et al. (2012). The size and the color (from blue to yellow) of the dots are proportional to the root square MSE: the best fitting model is marked by a red dot. (C) Comparison between the symptoms remission curves obtained with the model, using the parameter combinations indicated by the red dots in graphs (A) PE and (B) EMDR, and the actual experimental data from Nijdam et al. (2012). For the real data, the PTSD index is represented by the normalized IES-R score (Nijdam et al., ; refer to Section 3.4); in the model, the PTSD index represents the normalized amygdala unit activation following the presentation of the reminder cue. (D) PFC activation in correspondence to the simulation trials in (C).

Figure 9

Model weight modification due to trauma and therapy. (A) Connection weights linking the sensory cortices to the PFC. (B) Connection weights linking the hippocampus to the amygdala.