Neural representation during associative threat learning in pulvinar divisions, MD, and LGN.

(a) Human fear conditioning paradigm during fMRI. (b-d) Boxplots with individual data points at the block level (gold diamonds indicate the means), and trial-wise activation shown as means ± SE and kernel density estimates (KDEs) for responses to CS+ and CS- in the pulvinar divisions (b), MD (c), and LGN (d). Block-level comparisons were assessed using paired t-tests, while trial-level effects were examined using a 2 × 2 repeated-measures ANOVA, followed by post hoc comparisons between CS+ and CS− across the four trials. Multiple comparisons were controlled using false discovery rate (FDR) correction. Conditioning sample size: n = 293. Detailed statistical parameters are provided in Supplementary Tables 1-2. Additionally, cross-validation analyses were conducted to assess the robustness of the CS+ vs. CS− signal in the anterior pulvinar and MD (Supplementary Figures 1-2 and Supplementary File 1). *p < 0.05, **p < 0.01, ***p < 0.001. SE, standard error; KDE, kernel density estimates; MD, mediodorsal thalamus; LGN, lateral geniculate nucleus; CS+, conditioned stimulus predicting shock; CS−, conditioned stimulus predicting no shock. Display items in panel (a) were created using BioRender.

Quantifying the relationships between the anterior pulvinar and MD during conditioning.

a. Hierarchical regression models of trial-wise relationships between the anterior pulvinar and MD activations while controlling for a potential effect of anatomical proximity. The effect of anatomical proximity was controlled by progressively adding other pulvinar divisions as controls. b. Results of a hierarchical model that included the LGN as an alternative control, distinct from the pulvinar. c. Comparison of activation levels in the anterior pulvinar and MD at both block-wise and trial-wise levels (means ± SE). Sample size in these analyses: n = 293. Supplementary analyses testing alternative hypotheses regarding the correlations between the MD and all pulvinar divisions are presented in Supplementary Figure 3. *p<0.05, **p<0.01, ***p<0.001 MD: Mediodorsal thalamus. LGN: Lateral geniculate nucleus. SE: Standard error.

A data-driven approach to characterizing the functional organization of pulvinar divisions during conditioning.

(a) Means ± SE of activation differences between pulvinar divisions. (b) Network analysis identified the medial pulvinar as a central node within the observed functional network, exhibiting the highest centrality measures. (c) Schematic visualization of activation patterns observed across pulvinar divisions. (d) Previous studies suggest that the inferior and lateral pulvinar are involved in processing basic visual information (Berman and Wurtz, 2008, 2010; Cortes et al., 2024), whereas the medial pulvinar has been associated with higher-order functions, including attention and working memory (Homman-Ludiye and Bourne, 2019). (e) Based on panels b-d, we propose a computational model in which medial pulvinar activation statistically accounts for the associations among activation in the other pulvinar divisions. (f) Statistical mediation analysis evaluating the proposed computational model. (g) Replication of the mediation model in an independent sample. This independent cohort (N = 114) was collected as part of a separate research project that included the same Day 1 fear-conditioning paradigm used in the present study. The cohort was used to assess whether the mediation findings could be replicated independently. Dashed paths in panels f and g represent statistically unstable paths, whereas solid paths indicate stable paths. Note: The functional relationships among pulvinar divisions during threat learning should be interpreted as computational dependencies derived from statistical associations rather than evidence of direct anatomical connectivity. These statistical associations may reflect indirect interactions involving broader corticothalamic and thalamocortical circuits (e.g., via visual cortical regions). Determining the anatomical mechanisms underlying these functional associations will require future studies. Supplementary analyses examining within-pulvinar functional associations while controlling for the effects of the neighboring thalamic nuclei included in this study (MD and LGN) are presented in Supplementary Figure 4. These analyses provide an additional assessment of the specificity of the observed functional associations but do not distinguish direct inter-nuclear interactions from indirect coupling through corticothalamic and thalamocortical pathways. *p<0.05, **p<0.01, ***p<0.001. SE, standard error; SC, superior colliculus. Display items in panels d and e were created using BioRender.

Neural representation during extinction learning in pulvinar divisions, MD, and LGN.

(a) Human extinction learning paradigm during fMRI. (b-d) Boxplots with individual data points at the block level (gold diamonds indicate the means), and trial-wise activation shown as means ± SE and kernel density estimates (KDEs) for responses to CS+E and CS- in the pulvinar divisions (b), MD (c), and LGN (d). Block-level comparisons were assessed using paired t-tests, while trial-level effects were examined using a 2 × 2 repeated-measures ANOVA, followed by post hoc comparisons between CS+ and CS− across the four trials. Multiple comparisons were controlled using false discovery rate (FDR) correction. Extinction sample size: n = 320. Detailed statistical parameters are provided in Supplementary Tables 4–5. *p < 0.05, **p < 0.01, ***p < 0.001. SE, standard error; KDE, kernel density estimates; MD, mediodorsal thalamus; LGN, lateral geniculate nucleus; CS+E, extinguished conditioned stimulus that no longer predicts shock; CS−, conditioned stimulus predicting no shock. Display items in panel (a) were created using BioRender.

Neural representation during extinction recall in pulvinar divisions, MD, and LGN.

(a) Human extinction recall paradigm in the fMRI conducted within safe contextual cues. (b-d) Boxplots with individual data points at the block level (gold diamonds indicate the means), and trial-wise activation shown as means ± SE and kernel density estimates (KDEs) for responses to CS+E and CS- in the pulvinar divisions (b), MD (c), and LGN (d). Block-level comparisons were assessed using paired t-tests, while trial-level effects were examined using a 2 × 2 repeated-measures ANOVA, followed by post hoc comparisons between CS+ and CS− across the four trials. Multiple comparisons were controlled using false discovery rate (FDR) correction. Extinction recall sample size: n = 412. Detailed statistical parameters are provided in Supplementary Tables 6–7. *p < 0.05, **p < 0.01, ***p < 0.001. SE, standard error; KDE, kernel density estimates; MD, mediodorsal thalamus; LGN, lateral geniculate nucleus; CS+E, extinguished conditioned stimulus that no longer predicts shock; CS−, conditioned stimulus predicting no shock. Display items in panel (a) were created using BioRender.

Neural representation during threat renewal in pulvinar divisions, MD, and LGN.

(a) Human threat renewal paradigm in the fMRI conducted within threat contextual cues in the original context where fear conditioning occurred. (b-d) Boxplots with individual data points at the block level (gold diamonds indicate the means), and trial-wise activation shown as means ± SE and kernel density estimates (KDEs) for responses to CS+E and CS- in the pulvinar divisions (b), MD (c), and LGN (d). Block-level comparisons were assessed using paired t-tests, while trial-level effects were examined using a 2 × 2 repeated-measures ANOVA, followed by post hoc comparisons between CS+ and CS− across the four trials. Multiple comparisons were controlled using false discovery rate (FDR) correction. Threat renewal sample size: n = 318. Detailed statistical parameters are provided in Supplementary Tables 8–9. *p < 0.05, **p < 0.01, ***p < 0.001. SE, standard error; KDE, kernel density estimates; MD, mediodorsal thalamus; LGN, lateral geniculate nucleus; CS+E, extinguished conditioned stimulus that no longer predicts shock; CS−, conditioned stimulus predicting no shock. Display items in panel (a) were created using BioRender.

Task-related thalamic connectivity with threat-related brain regions.

(a–d) Thalamic seed-to-ROI connectivity with the vmPFC, sgACC, dACC, amygdala, and hippocampus during conditioning (n = 293), extinction (n = 320), recall (n = 412), and renewal (n = 318). Red lines indicate significant positive connectivity for CS+ vs. CS− following FDR correction across regions (pFDR < 0.05); gray lines indicate non-significant effects. For each panel, boxplots and kernel density estimates show the distribution of connectivity values for CS+ and CS−. All thalamic regions of interest—including pulvinar divisions (anterior, inferior, lateral, medial), LGN, and MD—were analyzed using identical statistical procedures. To maintain figure clarity and focus on effects relevant to the main conclusions, we display only seeds showing at least one significant CS+ vs. CS− differences after FDR correction. Non-significant seeds did not show systematic condition-related differences. *p < 0.05, **p < 0.01. MD, mediodorsal thalamus; LGN, lateral geniculate nucleus; vmPFC, ventromedial prefrontal cortex; sgACC, subgenual anterior cingulate cortex; dACC, dorsal anterior cingulate cortex; CS+, conditioned stimulus predicting shock; CS−, conditioned stimulus predicting no shock. gPPI; generalized psychophysiological interaction; connectivity values are Fisher z–transformed correlation coefficients.

Proposed conceptual neurobehavioral models of thalamic involvement in associative threat learning and memory.

(a) Conceptual schematic illustration of thalamic circuitry during the acquisition of associative threat learning, highlighting interactions between pulvinar divisions, MD, and LGN with key brain regions implicated in threat-related processing. (b) Conceptual model illustrating a possible thalamic-dependent “toggle switch” underlying context-dependent safety and threat memory expression. The observed MD-dACC and anterior pulvinar-vmPFC connectivity patterns are consistent with possible preferential engagement of prefrontal circuits during extinction recall and threat renewal, respectively. This framework represents a theoretical integration of the present findings rather than evidence of causal neural mechanisms. It is based on the findings observed within each phase separately and should not be interpreted as a direct comparison between phases. Note (panel a): Pulvinar-cortical pathways, as well as sensory input pathways (e.g., visual inputs via the retina/LGN and nociceptive inputs via the spinothalamic tract), are not explicitly shown. The pulvinar-cortical pathways and major sensory input pathways are recognized anatomically but were not specified due to their heterogeneity and incomplete characterization at the level of pulvinar subnuclei. Their lack of specification should not be interpreted as a lack of anatomical or functional relevance. The bidirectional dashed gray arrows between the pulvinar and visual cortex schematically represent corticothalamic and thalamocortical pathways. Their precise organization and contribution to individual pulvinar subnuclei remain incompletely characterized and were not examined in the present study. MD: Mediodorsal thalamus. LGN: Lateral geniculate nucleus. vmPFC = Ventromedial prefrontal cortex. dACC = Dorsal anterior cingulate cortex. Hypo.: Hypothalamus. Hippo.: Hippocampus. Created using BioRender.