Prefrontal control of the amygdala during real-time fMRI neurofeedback training of emotion regulation

Abstract

We observed in a previous study (PLoS ONE 6:e24522) that the self-regulation of amygdala activity via real-time fMRI neurofeedback (rtfMRI-nf) with positive emotion induction was associated, in healthy participants, with an enhancement in the functional connectivity between the left amygdala (LA) and six regions of the prefrontal cortex. These regions included the left rostral anterior cingulate cortex (rACC), bilateral dorsomedial prefrontal cortex (DMPFC), bilateral superior frontal gyrus (SFG), and right medial frontopolar cortex (MFPC). Together with the LA, these six prefrontal regions thus formed the functional neuroanatomical network engaged during the rtfMRI-nf procedure. Here we perform a structural vector autoregression (SVAR) analysis of the effective connectivity for this network. The SVAR analysis demonstrates that the left rACC plays an important role during the rtfMRI-nf training, modulating the LA and the other network regions. According to the analysis, the rtfMRI-nf training leads to a significant enhancement in the time-lagged effect of the left rACC on the LA, potentially consistent with the ipsilateral distribution of the monosynaptic projections between these regions. The training is also accompanied by significant increases in the instantaneous (contemporaneous) effects of the left rACC on four other regions - the bilateral DMPFC, the right MFPC, and the left SFG. The instantaneous effects of the LA on the bilateral DMPFC are also significantly enhanced. Our results are consistent with a broad literature supporting the role of the rACC in emotion processing and regulation. Our exploratory analysis provides, for the first time, insights into the causal relationships within the network of regions engaged during the rtfMRI-nf procedure targeting the amygdala. It suggests that the rACC may constitute a promising target for rtfMRI-nf training along with the amygdala in patients with affective disorders, particularly posttraumatic stress disorder (PTSD).

Conflict of interest statement

Competing Interests: Wayne Drevets, M.D., is an employee of Johnson & Johnson, Inc., and has consulted for Myriad/Rules Based Medicine and for Eisai, Inc. The other authors have declared that no competing interests exist. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials.

Figures

Figure 1. Regions of interest for the effective connectivity analysis.

Six brain regions exhibited a significant enhancement in functional connectivity with the left amygdala during the rtfMRI neurofeedback training with positive emotion induction . They included: the left rostral anterior cingulate cortex (rACC, BA 24), bilateral dorsomedial prefrontal cortex (DMPFC, BA 9), bilateral superior frontal gyrus (SFG, BA 6,8), and right medial frontopolar cortex (MFPC, BA 10). The 10 mm diameter regions of interest (ROIs) in those areas are projected onto the standard anatomical template (TT_N27) in the stereotaxic array of Talairach and Tournoux .

Figure 2. Learned enhancement of control over BOLD activity and emotion induction.

(A) Mean BOLD signal activity of the left amygdala during the rtfMRI neurofeedback (rtfMRI-nf) training for the experimental group (EG). The EG subjects received rtfMRI-nf based on the BOLD activity in the left amygdala ROI. Each bar represents a group average (mean±sem) of percent BOLD signal changes for the Happy Memories condition vs Rest condition for each of the six experimental runs: Rest (RE), Practice (PR), Run 1 (R1), Run 2 (R2), Run 3 (R3), and Transfer (TR). The enhancement in the left amygdala activity (red) was accompanied by increased activities of the left rACC (magenta), the right DMPFC (orange), as well as the other ROIs depicted in Fig. 1. (B) Lack of learned control over BOLD activity of the left amygdala and other regions for the control (sham) group (CG). The CG subjects received sham rtfMRI-nf based on BOLD activity in the left horizontal segment of the intraparietal sulcus (HIPS), presumably not involved in emotion regulation.

Figure 3. Interactions within the network suggested by the multivariate VAR analysis.

Results of the multivariate first-order vector autoregression (VAR) analysis for the network of seven ROIs depicted in Fig. 1. The four subplots show meta-analytic group statistics for path coefficients for the following groups and contrasts. (A) Experimental group (EG), neurofeedback Run 3. (B) Control group (CG), Run 3. (C) Difference between Run 3 and Rest for EG. (D) Difference between Run 3 for EG and Run 3 for CG. Red arrows denote augmentation effects (path coefficient α>0), and blue arrows – inhibition effects (path coefficient αq<0.05, and dotted arrows – to effects with 0.05≤q<0.1. In (C) and (D), solid arrows correspond to results with uncorrected P<0.05, and dotted arrows – to results with 0.05≤P<0.1.

Figure 4. Effects of the rACC on the other six network regions suggested by the multivariate VAR analysis.

Average path coefficient values (mean±sem) describing the effects of the left rACC on the other six network regions based on the analysis illustrated in Fig. 3. The results for each of the six experimental runs are shown in red for the experimental group (EG) and in blue for the control group (CG).

Figure 5. Schematics of structural models used in the multivariate SVAR analyses.

(A) An example of a star model for instantaneous effects. A model of this kind was defined for each of the seven ROIs and examined in the multivariate structural vector autoregression (SVAR) analysis. (B,C) Two models for instantaneous effects, Model I and Model II, that provided the best χ2 fits to the experimental group data in the SVAR analyses for the system of three ROIs. A total of 24 structural models were optimized and compared for the system consisting of the left amygdala, the left rACC, and the right DMPFC (see text for details).

Figure 6. Interactions suggested by the multivariate SVAR analyses for seven ROIs.

(A) Results of the multivariate first-order structural vector autoregression (SVAR) analysis for the network of seven ROIs using a star model for instantaneous effects of the left rACC (Fig. 5A). (B) Results of a similar SVAR analysis using a star model for instantaneous effects of the left amygdala. For the experimental group (EG), average path coefficients (mean±sem) for the instantaneous effects are depicted in magenta and denoted EG0, and those for the lagged effects are depicted in red and denoted EG1. For the control group (CG), average path coefficients for the instantaneous effects are shown in cyan and denoted CG0, and those for the lagged effects are shown in blue and denoted CG1.

Figure 7. Interactions suggested by the multivariate SVAR analyses for three ROIs.

Results of the multivariate SVAR analyses for the system of three ROIs – the left rACC, the left amygdala, and the right DMPFC – with the models for instantaneous effects depicted in Fig. 5 B,C. (A) Effects that are common to both Model I (Fig. 5B) and Model II (Fig. 5C). (B) Interactions between the left rACC and the left amygdala in the SVAR analysis with Model I (Fig. 5B). (C) Interactions between the left rACC and the left amygdala in the SVAR analysis with Model II (Fig. 5C). Notations are the same as in Fig. 6.