Habenula contributions to negative self-cognitions

Abstract

Self-related cognitions are integral to personal identity and psychological wellbeing. Persistent engagement with negative self-cognitions can precipitate mental ill health; whereas the ability to restructure them is protective. Here, we leverage ultra-high field 7T fMRI and dynamic causal modelling to characterise a negative self-cognition network centred on the habenula - a small midbrain region linked to the encoding of punishment and negative outcomes. We model habenula effective connectivity in a discovery sample of healthy young adults (n = 45) and in a replication cohort (n = 56) using a cognitive restructuring task during which participants repeated or restructured negative self-cognitions. The restructuring of negative self-cognitions elicits an excitatory effect from the habenula to the posterior orbitofrontal cortex that is reliably observed across both samples. Furthermore, we identify an excitatory effect of the habenula on the posterior cingulate cortex during both the repeating and restructuring of self-cognitions. Our study provides evidence demonstrating the habenula's contribution to processing self-cognitions. These findings yield unique insights into habenula's function beyond processing external reward/punishment to include abstract internal experiences.

Fig. 1. Cognitive restructuring paradigm.

a In each of the task blocks, participants were first shown a common negative self-cognition statement on the screen (e.g., “I am incompetent in the things I do”, “My value depends on my body shape”) for 4 s. Each block featured a unique statement. b Next, participants were given 9 s to decide and select either to restructure or repeat the negative statement that was on the screen. Participants indicated their choice via an MRI-compatible button box, which moved the black cursor to the elected choice. A counter was presented under each option, indicating the remaining number of blocks they could select the respective strategy, ensuring equal numbers of statements being restructured or repeated throughout the task. c Once the decision period lapsed, participants were shown the same statement accompanied by an instruction to engage in their chosen strategy for 12 s (restructure = challenge condition; repeat = repeat condition). d A jittered fixation cross was then presented for an average of 6 s (rest condition) before the next block commenced. The replication sample group underwent an abbreviated version of this paradigm, containing 16 blocks that did not include negative self-cognition statements about food and body image.

Fig. 2. Task-based neural activation and construction of the DCM model space.

a, b The heatmaps display the general linear model (GLM) results of the cognitive restructuring fMRI paradigm (P_FDR < 0.05, K_E ≥ 10), with the colour bars representing the t-statistics of the single-sample t test (one-tailed). The warm colour map shows brain regions with increased activity during the restructuring of negative self-cognitions compared to repeating (Challenge > Repeat). The cool colour map highlights structures showing increased activity during the repetition of negative cognitions versus restructuring (Repeat > Challenge). These results confirmed the engagement of the habenula in negative self-cognition processing and were used to inform DCM model node selection. c Consecutive coronal views of the neural activation results of the Repeat > Challenge contrast are presented on the MNI152 template to highlight the habenula cluster showing increased activity during the repeating of negative self-cognitions compared to restructuring. d The line graph plots the group-level blood-oxygen level dependent (BOLD) response (GLM-estimated) of the habenula region-of-interest across the key conditions and when averaged across task epochs (rest, repeat, rest, challenge). The habenula showed sustained activity during the repeating of negative self-cognitions (repeat condition) and evoked response during cognitive restructuring (challenge condition). e Based on the GLM results and past literature on habenula connectivity, the bilateral habenula, right pOFC, PCC, and the hippocampus were selected as model nodes for DCM analysis. f A DCM model centred on the habenula was constructed and estimated for each individual. The model assumed (1) bidirectional endogenous connections between the habenula and the other network regions (grey arrows), in addition to their self-inhibitory connectivity (not shown here); (2) driving input of the experimental stimuli into all network nodes (yellow arrows); and (3) modulatory effects of the restructuring and repetition of negative self-cognitions on the bidirectional connections between the habenula and the PCC, hippocampus, as well as the pOFC (dashed blue arrows). 3-D brain rendering were constructed in BrainNet Viewer with the MNI152 template brain. DCM dynamic causal model, dlPFC dorsolateral prefrontal cortex, dmPFC dorsomedial prefrontal cortex, HC hippocampus, L left, PCC posterior cingulate cortex, pOFC posterior orbitofrontal cortex, pSMA pre-supplementary motor area, R right, vlPFC ventrolateral prefrontal cortex.

Fig. 3. Habenula effective connectivity in the discovery sample.

a Intrinsic connectivity and task-induced modulation of connections between the habenula and the network nodes that demonstrated strong evidence (posterior probability > .95) for a non-zero group effect are illustrated on the MNI152 template brain with BrainNet Viewer. Intrinsic connectivity is represented with solid arrows while dashed arrows depict modulatory effects by each of the task conditions-of-interest – the restructuring (CHAL condition) or repeating (REP condition) of negative self-cognitions. Red arrows represent excitatory intrinsic effective connectivity or positive modulatory effects. Blue arrows show inhibitory intrinsic connectivity or negative task-induced modulation. b Task-induced changes in habenula effective connectivity are plotted for connectivity associated with significant modulatory effects. The first column shows the average effective connectivity throughout the task for each pathway (intrinsic connectivity; A-matrix). This represents the context-independent influence from the habenula to the PCC (blue) and the pOFC (green). The second and third columns represent the net effective connectivity of each pathway under the repeat (REP) and challenge (CHAL) conditions, respectively. This is calculated by adding or subtracting the modulatory effects from the corresponding intrinsic connectivity parameter (A-matrix + B-matrix) as the experimental input is mean-centred in our model. All connectivity estimates are in units of Hz denoting the rate of change in neural activity in the input regions (e.g., PCC, pOFC) due to neural response of the output region (i.e., habenula). Source data are provided as a Source Data file. CHAL, challenge condition, Hb habenula, HC hippocampus, Hz hertz, L left, PCC posterior cingulate cortex, pOFC posterior orbitofrontal cortex, R right, REP repeat condition.

Fig. 4. Habenula effective connectivity in the replication sample.

a Intrinsic and task-induced modulation of habenula connectivity with strong evidence (posterior probability > .95) for a non-zero group effect are displayed on the MNI152 template brain using BrainNet Viewer. Solid arrows represent intrinsic connectivity, while dashed arrows illustrate modulatory effects by the task conditions – the restructuring (CHAL condition) or repeating (REP condition) of negative self-cognitions. Red arrows show excitatory intrinsic connectivity or positive modulatory effects. Blue arrows indicate inhibitory intrinsic connectivity or negative task-induced modulation. b Changes in habenula effective connectivity are plotted for connectivity showing significant task-induced modulation. The first column quantifies the average effective connectivity throughout the task for each pathway (intrinsic connectivity; A-matrix), representing the context-independent influence from the habenula to the PCC (blue) and the pOFC (green). The second and third columns show the net effective connectivity of each pathway under the repeat (REP) and challenge (CHAL) conditions. As the experimental input is mean-centred in our model, this is calculated as the intrinsic connectivity parameter plus the modulatory effect of the corresponding task condition (A-matrix + B-matrix). All connectivity estimates are in units of Hz denoting the rate of change in neural activity in the input regions (e.g., PCC, pOFC) due to neural response of the output region (i.e., habenula). c The bar graph depicts modulatory connectivity of the discovery model (n = 45; grey) superimposed with the posterior expectation of the replication model (n = 56; orange). The bars represent the Bayesian model-averaged (BMA) connectivity strength estimates of the corresponding network connection in the models, and the whiskers show the 95% confidence interval (CI) centring the discovery model parameter estimates derived from the posterior covariance matrix (spm_plot_ci.m). An asterisk is placed above the bars for connectivity replicated across the models, which are identified based on the replication model connectivity with posterior expectations that are within the 95% CI of the discovery model estimate and surpass the posterior probability threshold (> .95). Source data are provided as a Source Data file. CHAL challenge condition, Hb habenula, HC hippocampus, Hz hertz, L left, PCC posterior cingulate cortex, pOFC posterior orbitofrontal cortex, R right, REP repeat condition.

Fig. 5. Generation and evaluation of individual habenula ROIs.

a The MAGeTbrain algorithm produced high-resolution individualised habenula masks (red outline) using the 7-Tesla anatomical images (0.75 mm isotropic). Three example habenula masks are presented here and displayed on their corresponding T1-weighted whole-brain image in the native space. b The individualised habenula masks were normalised to the MNI space using the DARTEL flow fields produced during anatomical image pre-processing. These high-resolution normalised masks were used as reference label images with which we evaluated the functional resolution habenula masks. c To create the habenula ROIs for functional analysis, each individual’s anatomical-resolution habenula mask were resampled to the functional resolution (2 mm isotropic) and transformed to the standard space. An iterative volume optimisation procedure was developed to minimise the impact of down-sampling on the habenula mask. In brief, adjustment was made to the re-binarising threshold in each iteration to ensure that the resultant habenula ROI had a volume that is within 10% of the reference label image (i.e., normalised habenula mask in anatomical resolution) after the removal of CSF voxels. d The subject-specific habenula ROIs in functional resolution were further validated via an rsFC analysis. Results of this analysis replicated the habenula functional connectivity pattern reported in past studies. ACC anterior cingulate cortex, CSF cerebrospinal fluid, DARTEL Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra, FSL FMRIB Software Library, L left, MAGeT Multiple Automatically Generated Templates brain segmentation, PCC posterior cingulate cortex, pSMA pre-supplementary motor area, R right, rsFC resting-state functional connectivity, SPM Statistical Parametric Mapping, VOI volume-of-interest, VTA ventral tegmental area.