Neural Representation of Associative Threat Learning in Pulvinar Divisions, Lateral Geniculate Nucleus, and Mediodorsal Thalamus in Humans

29 Understanding the neural mechanisms underlying associative threat learning is essential 30 for advancing behavioral models of threat and adaptation. We investigated distinct 31 activation patterns across thalamic pulvinar divisions, lateral geniculate nucleus (LGN), 32 and mediodorsal thalamus (MD) during the acquisition of associative threat learning in 33 the MRI. We revealed parallel thalamic learning systems within the anterior pulvinar and 34 MD, supporting distinct mechanisms of automatic survival vs. more deliberate learning. 35 Additionally, o ur findings support a novel hierarchical pulvinar model during fear 36 conditioning: the medial pulvinar mediate s basic threat information from the inferior and 37 lateral divisions to the anterior pulvinar for integrative learning. Pulvinar divisions and MD 38 support extinction learning . These regions also process salience and modulat e 39 safe/threat memory expression during extinction recall and threat renewal . The LGN 40 sustains feedforward processing of anticipated visual input throughout all threat phases. 41 This study extends dominant brain models of threat learning and memory, reframing our 42 understanding of distinct thalamic roles in these psychological processes. 43

Thalamus; Pulvinar; Lateral geniculate nucleus; Mediodorsal thalamus; 44 Threat learning; Conditioning; Extinction; Recall extinction; Threat renewal; Emotional 45 memory. 46 47 48 49 50 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 3

51 52 53 Over a century ago, Ivan Pavlov provided foundational behavioral evidence for 54 associative learning, demonstrating that pairing a neutral stimulus with an unconditioned 55 stimulus could elicit a conditioned response(Pavlov, 1904 ). This discovery laid the 56 groundwork for modern learning theories and has been particularly influential in 57 understanding how humans learn to associate neutral stimuli with threat s, a process 58 known as associative threat learning . This survival-oriented learning plays a key role in 59 shaping a broad range of behavioral and emotional response s, such as avoidance, 60 decision-making under threats, and fear regulation(Badarnee et al., 2025; Kolling et al., 61 2014; Korn and Bach, 2019, 2018; Milad et al., 2013) . Research on t he neural 62 mechanisms underlying associative threat learning has primarily focused on brain regions 63 involved in emotional regulation, such as the amygdala, hippocampus, and prefrontal 64 cortex (PFC)(Fullana et al., 2015; Krasne et al., 2021; Milad and Quirk, 2012; Vuilleumier 65 et al., 2003) . More r ecently, the thalamus has received growing attention in threat 66 learning(Lithari et al., 2015; Penzo et al., 2015; Ramanathan et al., 2018; Ramanathan 67 and Maren, 2019; Ratigan et al., 2023; Totty et al., 2023) , since its role has been 68 reconsidered beyond the traditional view of a mere relay station. Yet, the distinct 69 contribution of this complex structure to associative threat learning remains poorly 70 understood. 71 72 The thalamus is a hub structure in the mammalian brain that plays multiple critical roles, 73 ranging from basic sensory processing to higher-order functions and threat 74 learning(Halassa and Sherman, 2019; Hummos et al., 2022; Hwang et al., 2017; 75 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 4 Saalmann and Kastner, 2011; Sherman, 2016, 2007) . It has long been proposed that 76 threat detection is mediated by the pulvinar and the lateral geniculate nucleus (LGN)(Carr, 77 2015; Pessoa and Adolphs, 2010; Silverstein and Ingvar, 2015) , two major nuclei of the 78 visual thalamus (Arcaro et al., 2015; Casanova and Chalupa, 2023; Takakuwa et al., 79 2021). The pulvinar projects directly to the amygdala and is believed to subconsciously 80 facilitate rapid survival reactions to threats via a subcortical path (‘low road’)(Carr, 2015; 81 Pessoa and Adolphs, 2010; Rafal et al., 2015; Wei et al., 2015) . The LGN, on the other 82 hand, is part of a slower but more accurate neural pathway , connecting peripheral 83 information to cortical regions for comprehensive processing (‘high road’). Additionally, 84 the involvement of the mediodorsal thalamus (MD) in threat learning has also been 85 proposed(Lee and Shin, 2016; Lee et al., 2011) . Although this nucleus is not typically 86 classified as part of the visual thalamus , it serves as a high-order region reciprocally 87 connected to the amygdala and PFC while playing a critical role in executive 88 functions(Hwang et al., 2020; Li et al., 2022; Mukherjee et al., 2021; Wolff and Halassa, 89 2024). 90 91 The general implication of these nuclei in threat processing has been demonstrated in 92 primates and rodent models. It has been shown that exposing primates to snake images 93 elicited increased neuronal firing within the pulvinar(Le et al., 2013). Excitation of neurons 94 within the LGN enhanced the acquisition of eyeblink conditioning in rodents (Halverson 95 and Freeman, 2010) . Inhibiting this nucleus was associated with impaired conditioned 96 responses(Shi and Davis, 2001; Steinmetz et al., 2013), and activating GABA neurons in 97 the LGN reduced freezing response to an overhead dark shadow that mimics a real-world 98 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 5 predator in rats(Salay and Huberman, 2021). Reduced freezing was also observed after 99 MD lesions(Li et al., 2004) and when the connections with the anterior cingulate cortex 100 were ablated(Zheng et al., 2020). Lesions within the MD are also associated with impaired 101 fear extinction(Lee et al., 2011) , and injecting gabazine , a modulator of extra-synaptic 102 GABA receptors, into the MD facilitated extinction(Paydar et al., 2014). 103 104 Evidence regarding the involvement of the thalamus in the context of threat learning in 105 humans is relatively sparse. Much of our knowledge comes from studies focusing on 106 attention, perception, and decision -making. For example, the LGN has been mainly 107 viewed as a transmitter of peripheral information to other brain regions and found to be 108 associated with selective attention and anticipation of visual stimuli (Mahoney and 109 Schmidt, 2024; O’Connor et al., 2002; Saalmann and Kastner, 2009). The MD contribution 110 to perception, spatial attention, and decision -making has also been reported (Griffiths et 111 al., 2022; Wurtz et al., 2011) . Beyond cognitive research, s ome human studies have 112 specifically investigated the pulvinar ’s engagement in threat detection. Individuals with 113 increased fiber density in the pulvinar-amygdala pathway showed an enhanced ability to 114 recognize fearful faces(McFadyen et al., 2019). A patient with a complete lesion in the left 115 pulvinar showed slower responses to threatening images when stimuli were presented 116 on the ipsilesional, but not contralesional, field(Ward et al., 2005). In agreement with this, 117 unseen stimuli (fearful vs. happy faces) presented on the blind side of patients with 118 hemianopia moderated their performance when the pulvinar was spared but not when it 119 was lesioned(Bertini et al., 2018). 120 121 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 6 Specific thalamic contribution to threat processing is still largely unknown, and 122 translational research on this topic is limited. We aim to address these critical gaps in our 123 understanding by investigating the distinct neural representations of associative threat 124 learning in the human pulvinar divisions, LGN, and MD. We analyzed the neuroimaging 125 data of 293 -412 controls. All participants underwent a two -day threat learning 126 paradigm(Milad et al., 2009, 2007; Wen et al., 2024, 2022) while in the fMRI scanner. The 127 conditioned stimulus ( CS+) (e.g., red light) was associated with an electric shock 128 (unconditioned stimulus US, 62.5% reinforcement rate), while a control light (e.g., blue) 129 was never paired with the US (CS -). This conditioning phase occurred in a computer-130 displayed visual context A (e.g., an office). The extinction learning included presenting 131 the CS+ and CS - with no US reinforcement in a distinct context B (e.g., bookcase). In 132 extinction recall and threat renewal, the extinguished stimuli were presented with no US 133 reinforcement within contexts B (safe contextual cues) and A ( threat contextual cues), 134 respectively. Schematic illustrations of each phase of the paradigm are presented in Figs. 135 1a, 4a-6a. 136 137 We focus on neural activation patterns within pulvinar divisions, LGN, and MD while 138 acquiring the CS-US association. As associative learning is a rapid psychological 139 process(Konrad et al., 2024), we analyzed the first four trials of all threat learning phases. 140 We compared brain activation (BOLD) to the CS+ vs. CS- at the block level by averaging 141 the activation across all four trials and at a trial-by-trial level to provide finer activation 142 temporal resolution. This dual approach allows us to capture the general neural 143 responses associated with associative threat learning and provides new insights into trial-144 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 7 level dynamics (Wen et al., 2022) . The current dominant neurocognitive thalamic 145 models(Sherman, 2007; Sherman and Guillery, 2006) highlight the thalamus’s role in 146 mediating cortical-cortical communications and facilitating high-order functions. Based on 147 this view, we anticipate distinct pulvinar, MD, and LGN roles. The LGN, as a first -order 148 nucleus, is expected to serve relay function s, transmitting information with minimal 149 integrations. The pulvinar and MD, on the other hand, as higher -order nuclei, are likely 150 involved in more complex processing, potentially integrating threat-related information. 151 152

153 Parallel functional representation of associative threat learning in the anterior 154 pulvinar and MD during conditioning 155 The anterior pulvinar and MD contribute to early associative threat learning, as evidenced 156 by increased functional activation in response to CS+ compared to CS- at the block level 157 (Fig. 1b-c). The trial-wise analysis revealed a distinct activation pattern. In the first trial, 158 we observed similar activation levels for both CS+ and CS- (no significant difference), but 159 in the subsequent trial, the BOLD signal in both regions was heightened specifically for 160 CS+ (Fig. 1b-c, Extended Tables 1-2). In our paradigm, the electric shock was paired with 161 CS+ at the end of the CS presentation. The similarity in BOLD responses to CS+ and CS- 162 during the first trial likely reflects an initial equivalence in the emotional valence of the 163 stimuli. The gradual increase in activation for CS+ by the second trial suggests a shift in 164 the emotional valence, indicating rapid associative learning. 165 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 8 The similarity in trial-level activation patterns in the anterior pulvinar and MD raises the 166 question of whether this apparent similarity truly reflects a parallel functional contribution 167 to associative threat learning . To test this possibility, w e first quantified the trial-wise 168 relationships using Pearson correlation coefficients between the two regions . The 169 observed correlations support consistent co -activation across all corresponding trials (r 170 values ≥ 0.66, p < 0.001, Fig. 2a, left). However, this approach could not exclude the 171 possibility that the shared variance is a consequence of shared anatomical proximity , 172 particularly because both regions are part of the same brain structure, i.e., the thalamus. 173 To control for this confound, we conducted a hierarchical regression model in four steps 174 for each trial . We modeled MD activation as the target and progressively ad ded the 175 pulvinar division s as covariates . Despite small variations in regression weights (beta 176 values) across control models, the anterior pulvinar consistently exhibited the highest 177 beta estimat es (Fig. 2a, middle ). Importantly, these values remain consistent across 178 10,000 bootstrapping iterations with replacements, suggesting that the anterior pulvinar 179 effect is robust and stable across resampled datasets (Fig. 2a, right). Additionally, the 180 anterior pulvinar explained the largest proportion of variance in MD activation (r2 = 43-181 60.7%) relative to other pulvinar divisions, which negligibly contributed to this parameter 182 (pie charts, Fig. 2a). Finally, applying the same analytical approach while controlling for 183 the LGN activation as an alternative anatomical control beyond the pulvinar itself, yielded 184 similar results (Fig. 2b). Together, these analyses support preferential trial-wise functional 185 co-activations between the anterior pulvinar and MD , independent of shared anatomical 186 proximity. 187 188 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 9 Although co-activation does not necessarily imply similar activation magnitude, we further 189 tested the differences in activation levels in the anterior pulvinar and MD. We used a t-190 test to capture the differences in the overall activation at the block level and repeated 191 measures analysis of variance ( RM-ANOVA) to capture the activation differences at the 192 trial level. The results showed that the overall MD activation in response to CS+ was 193 higher than the anterior pulvinar response (p-value < 0.001, Fig. 2c). The trial -wise 194 analysis revealed heightened activation in the MD compared to the anterior pulvinar , 195 particularly during trials 3 and 4 (pFDR-values < 0.05, Fig. 2c). The increased activation 196 within the MD potentially reflects its role in deeper processing of threat relatively to the 197 anterior pulvinar. 198 199 A data -driven approach reveals hierarchical functional processing of threat in 200 pulvinar divisions 201 The anatomical and functional interconnections between pulvinar divisions remain 202 insufficiently characterized. To address this gap and understand the pulvinar’s role in 203 early associative threat learning, we integrated findings from complementary analyses 204 into a novel model designed to investigate the functional relationships between pulvinar 205 divisions. 206 207 We first defined the activation within each pulvinar division as the differences in BOLD 208 signal at the block level in response to CS+ compared to CS -. We then performed RM -209 ANOVA to test whether the pulvinar divisions were engaged with different activation 210 levels. We found no activation differences, suggesting similar processing levels of CS+ 211 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 10 information across all pulvinar divisions (Fig. 3a). Using network analysis, w e explored 212 the underlying functional dynamics between the divisions. We employed the EBICglasso 213 method(Foygel and Drton, 2010) to estimate a sparse Gaussian graphical model. To 214 assess the stability of the network and provide robust estimates of edges and centrality 215 measures, we conducted 10,000 bootstrap resamples. This approach allowed us to 216 compute 95% confidence intervals, ensuring the reliability of our findings. The combined 217 use of EBICglasso and bootstrapping is essential for accurately capturing the dynamics 218 of pulvinar division interactions. The resulting network included four nodes and five edges. 219 The network plot pointed to stable direct edges connecting the lateral pulvinar to the 220 anterior pulvinar and indirect edges connecting both the lateral and inferior pulvinar to the 221 anterior through the medial pulvinar (Fig. 3b). Additionally, the medial pulvinar exhibited 222 the highest centrality values, a measure that captures the node importance within a 223 network, indicating a potential hub role of this division in associative threat learning (Fig. 224 3b). 225 These underlying dynamics and the activation timing in the trial-wise analyses (overview 226 in Fig. 3c and details in Fig. 1b) led us to propose a functional model of the relationships 227 between pulvinar divisions. Specifically, the increased activation induced by CS+ during 228 trial 1 in the medial, inferior, and lateral pulvinar suggests that these divisions process 229 CS+ information before the anterior pulvinar, which exhibited a delayed BOLD response 230 starting in trial 2. The medial pulvinar mediates connections within the pulvinar network 231 and is associated with elevated centrality values. Together, these findings highlight the 232 hub role of the medial pulvinar , integrating CS+ information at a higher level , compared 233 to sensory-driven processing in the inferior and lateral pulvinar. This possible hierarchical 234 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 11 organization is supported by previous evidence showing that the inferior and lateral 235 pulvinar are associated with processing basic sensory information(Berman and Wurtz, 236 2010, 2008; Cortes et al., 2024) , while the medial pulvinar is implicated in higher -order 237 processing, including attention and working memory(Homman-Ludiye and Bourne, 2019) 238 (Fig. 3d). 239 This data-driven approach provides a new perspective on the pulvinar divisions’ functional 240 specialization during associative threat learning. We hypothesize that, during this 241 process, the activation in the medial pulvinar mediates the functional relationships 242 between the inferior and lateral divisions with the anterior pulvinar (Fig. 3e). To test this, 243 we conducted a mediation model analysis and evaluated the model robustness using the 244 k-fold cross-validation method. The sample ( N = 293) was randomly divided into three 245 groups of 91, 96, and 106 subjects. Each sub -sample was used to test the model while 246 the remaining groups served as a training phase. This resulted in six iterations of the 247 mediation model. The 95% Confidence Intervals for each path coefficient were estimated 248 using 10,000 bootstrapping replications. This choice was made to achieve greater 249 precision and stability in the confidence interval estimates. Finally, applying the mediation 250 analysis to another independent cohort (N = 114) confirmed the model’s applicability. The 251

in Figs. 3f-3g support the mediating role of the medial pulvinar, demonstrating 252 significant indirect paths from the lateral and inferior divisions to the anterior pulvinar. 253 Detailed parameters are presented in Extended Table 3. 254 Pulvinar divisions and MD support extinction learning while preserving threat 255 memory across contextual cues 256 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 12 After our comprehensive examination of the thalamic contribution to threat conditioning, 257 we moved forward to examine the neural contributions of the same regions to extinction 258 learning. During this phase, w e found higher activation in response to CS+ than CS- 259 across all pulvinar divisions and MD at the block level. The trial -wise analysis 260 demonstrated that these differences are mainly driven by the first two trials (Figs. 4b and 261 4c). Together, the pulvinar divisions and MD showed a similar activation pattern where 262 the BOLD signal induced by the CS+ is significantly diminished after trial 2 , highlighting 263 these nuclei’s engagement in rapid extinction learning. Detailed parameters are 264 presented in Extended Tables 4-5. 265 266 During extinction recall, the anterior pulvinar and MD exhibited similar functional patterns 267 at both block and trial levels, with increased activation to CS+ compared to CS - during 268 the first two trials. In contrast, the medial and inferior pulvinar maintained the BOLD signal 269 to CS+ in the second trial while showing reduced responses to CS -. We observed no 270 activation differences in the lateral pulvinar (Fig. 5b-c, Extended Tables 6-7). This finding 271 highlights the involvement of most pulvinar divisions and MD in sustaining threat memory 272 under safe contextual cues and suggests retrieval suppression of the extinguished threat. 273 274 Changing the contextual cues to a threat background where the original associative 275 learning occurred elicited increased activation in response to CS+ in the anterior, inferior, 276 and lateral pulvinar divisions along with the MD (Fig. 6b-c, Extended Tables 8-9). These 277 regions are, therefore, involved in the reactivation of threat memory, a hallmark process 278 in the threat renewal phase. 279 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 13 280 Consistent LGN activation patterns across all threat learning phases align with 281 principles of feedforward processing. 282 At the block level, we observed elevated activation in response to CS+ compared to CS- 283 across conditioning, extinction, recall, and renewal. These differences were driven by the 284 first trial, during which an increased BOLD signal in response to CS+ was followed by a 285 decline in subsequent trials ( Figs. 1d, 4d-6d, Extended Tables 2, 5, 7, and 9 ). These 286 findings suggest that the LGN plays a consistent role across different phases of threat 287 learning, potentially reflecting early attentional processing or anticipation of upcoming 288 visual stimuli. 289 290 Thalamic connectivity underlies threat learning and memory 291 We tested the relationships between each thalamic nucleus and core brain regions within 292 the ‘fear circuit’(Herry et al., 2010; Maren and Quirk, 2004; Milad et al., 2014; Milad and 293 Quirk, 2012; Tovote et al., 2015) by seeding the nuclei to target the amygdala, 294 hippocampus, ventromedial prefrontal cortex ( vmPFC), subgenual anterior cingulate 295 cortex (sgACC), and dorsal anterior cingulate cortex ( dACC). During conditioning, the 296 anterior pulvinar exhibited positive connectivity with the amygdala, vmPFC, and 297 hippocampus (all pFDR < 0.05 and all 95% bias -corrected and accelerated (BCa) 298 confidence intervals (CI) of the mean differences from 10,000 bootstrap resamples 299 excluded zero; Cohen’s ds = 0.14, 0.14, and 0.15, with corresponding 95% parametric 300 CIs also excluding zero ) (Fig. 1e). The connectivity with the amygdala likely supports 301 encoding the emotional valence of the newly learned CS-US associations , w hile the 302 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 14 engagement of the hippocampus suggests prioritizing the contextual CS+ information. 303 The connectivity with the vmPFC may underlie a process of top-down control release to 304 enhance fear expression or encode threat information for tracking and supporting 305 decision-making in future encounters. 306 During extinction learning, we found increased LGN-sgACC connectivity (pFDR < 0.05 and 307 95% BCa CIs of the mean differences from 10,000 bootstrap resamples excluded zero ; 308 Cohen’s d = 0.15, 95% parametric CI [0.035, 0.256] ), suggesting sgACC involvement in 309 emotional regulation ( Fig. 4e). Interestingly, i n recall and renewal, we found that the 310 learners, i.e., the anterior pulvinar and MD contribute to support ing either safe or 311 threatening memory , depending on contextual cues . Specifically, the MD -dACC 312 connectivity supported safe memory in recall (pFDR < 0.05 and 95% BCa CI of the mean 313 differences from 10,000 bootstrap resamples excluded zero ; Cohen’s d = 0.13, 95% 314 parametric CI [0.031, 0.225]; Fig. 5e), whereas anterior pulvinar -vmPFC supported 315 threatening memory in renewal (p FDR < 0.01 and 95% BCa CI of the mean differences 316 from 10,000 bootstrap resamples excluded zero ; Cohen’s d = 0. 20, 95% parametric CI 317 [0.084, 0.306]; Fig. 6e). 318

319 We examined the neural representation of associative threat learning within the pulvinar 320 divisions, LGN, and MD, providing new insights into thalamic involvement in this adaptive 321 behavior in humans. We identified distinct roles among these thalamic nuclei with respect 322 to their activation profiles during threat learning and memory. The anterior pulvinar and 323 MD exhibited parallel activation patterns consistent with associative learning, reflecting 324 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 15 their roles in automatic survival responses and deliberate threat processing. We propose 325 a novel hierarchical model for processing threat information in the pulvinar divisions. The 326 medial pulvinar mediates basic sensory information from the inferior and lateral divisions 327 to the anterior pulvinar for higher-order integrative learning. Both pulvinar and MD were 328 involved in extinction and showed activation consistent with the salience processing of 329 threat-related memories during extinction recall and threat renewal. The LGN primarily 330 represented feedforward processing, anticipating upcoming visual stimuli throughout all 331 phases of threat learning. We integrated these insights into schematic models underlying 332 the emotional and behavioral expression of threat learning and memory , providing a 333 neural framework for studying related human behaviors (Fig. 7a-b). 334 The anterior pulvinar and MD co-activation patterns demonstrate a parallel contribution 335 to threat learning. The similar activation in response to CS+ and CS - at the first trial, 336 followed by a heightened activation specific to CS+ in the next trial, is consistent with the 337 acquisition of the CS-US association. The responses’ similarity to both types of CS during 338 the first trial suggests an equivalent initial emotional valence. The gradual increase in the 339 activation, specifically to CS+ by the end of trial 1, suggests a shift in the emotional 340 valence of CS+, indicating rapid associative learning. This parallel specialized role 341 highlights the anterior pulvinar and MD as central thalamic hubs for integrating CS -US 342 information, specifying the two-system model proposed by LeDoux and Pine(LeDoux and 343 Pine, 2016) . Briefly, fear processing, a ccording to this model, involves two distinct 344 pathways: a subcortical route for rapid threat detection and species -specific defense 345 reactions and a cortical route for deliberate threat evaluation and conscious fear 346 experience. Learning within the anterior pulvinar appears to support rapid automatic 347 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 16 processes of the subcortical pathway. This aligns with previous reports that demonstrated 348 the role of the pulvinar-amygdala pathway in encoding negative emotions in 349 humans(Bridge et al., 2015; Koller et al., 2019; Kragel et al., 2021; McFadyen et al., 2019; 350 Rafal et al., 2015) . Indeed, the increased connectivity that we observed with the 351 amygdala, hippocampus, and vmPFC during threat conditioning aligns with this role. 352 Specifically, in this context, the amygdala likely contributes to tagging a negative 353 emotional valence to CS+ and triggering fight, flight, or freeze reactions (Adolphs et al., 354 1994; Costa et al., 2022; Fanselow and LeDoux, 1999; Phelps and LeDoux, 2005; Wen 355 et al., 2024, 2022) . The connectivity with the hippocampus likely facilitates contextual 356 encoding of the environmental characteristics of the aversive event(Maren, 2001; Maren 357 et al., 2013). In turn, the connectivity with vmPFC appears to support salience encoding 358 for tracking future encounters with the learned threat(Battaglia et al., 2020) and facilitating 359 decision-making and emotion regulation(Milad and Quirk, 2012, 2002; Nejati et al., 2021). 360 On the other hand, MD activation aligns more closely with the cortical pathway, facilitating 361 conscious and deliberate threat encoding. This is supported by our findings, which 362 showed increased MD activation in response to CS+ , compared to the anterior pulvinar 363 during threat learning , i ndicating a broader or deeper processing of the CS -US 364 association. Additionally, this interpretation is supported by substantial evidence pointing 365 to well -established anatomical pathways connecting the MD with the PFC and 366 demonstrating its role in decision-making and learning (Behrens et al., 2003; Hwang et 367 al., 2020; Li et al., 2022; Wolff and Halassa, 2024) . This is a long with reports that 368 underscored the contribution of the MD-cortical loops to supporting a conscious 369 experience in humans(Griffiths et al., 2022; Whyte et al., 2024). Together, these findings 370 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 17 advance our understanding of the two-system model of threat processing, highlighting the 371 anterior pulvinar and MD role in the acquisition of the CS-US association and suggesting 372 that these regions underlie a parallel conscious vs. unconscious learning proposed by 373 LeDoux and colleagues(LeDoux and Pine, 2016), Fig. 7a. 374 Our findings demonstrate that different pulvinar divisions hierarchically contribute to the 375 integration of CS-US association in the anterior pulvinar through bottom -up processing. 376 Although the anatomical projections within pulvinar divisions are not well characterized, 377 our functional data-driven approach revealed a critical role for the medial pulvinar as a 378 hub region facilitating communication among pulvinar divisions. Specifically, our trial-wise 379 analysis showed that the medial, inferior, and lateral pulvinar process threat -related 380 information earlier than the anterior pulvinar. The anterior pulvinar exhibited activation in 381 response to CS+ starting at trial 2, whereas activation in the other pulvinar divisions 382 occurred as early as trial 1. Network modeling further identified the medial pulvinar as a 383 central hub, facilitating communication with other divisions. Using a robust mediation 384 model, we demonstrated that the medial pulvinar appears to mediate the relay of sensory 385 CS+ information from the inferior and lateral pulvinar to the anterior division for higher-386 order integra tive learning . This hierarchical model aligns with previous evidence 387 suggesting that the inferior and lateral pulvinar are associated with processing basic 388 sensory information(Berman and Wurtz, 2010, 2008; Bridge et al., 2015; Cortes et al., 389 2024). W hile the medial pulvinar , which receives projections from deep layers of the 390 superior colliculus, supports more advanced functions such as attention and working 391 memory(Bridge et al., 2015; Homman-Ludiye and Bourne, 2019). These findings provide 392 novel insights into the functional specialization of pulvinar divisions during threat learning, 393 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 18 suggesting feedforward functions in the inferior and lateral pulvinar and high -order 394 integration in the anterior pulvinar. 395 The pulvinar divisions and MD are actively engaged in extinction learning, as indicated 396 by increased activation in response to CS+ during the first two trials, followed by a 397 diminished response. This decline in activation likely reflects a shift in the emotional 398 valence of the CS+ to match that of the CS-, suggesting a rapid extinction process in the 399 human thalamus. Despite the involvement in extinction, these nuclei remained sensitive 400 to the extinguished threat during safe contextual cues in extinction recall and exhibited 401 increased activation to the extinguished CS+ when the stimuli were presented with in 402 threat contextual cues during renewal. This pattern suggests engagement in retrieval 403 suppression during recall , along with threat salience processing during both recall and 404 renewal. The MD and pulvinar continue to monitor and evaluate the extinguished threat 405 across contexts rather than relying on static safety memories . These nuclei appear to 406 engage in dynamic evaluation, learning, and decision -making during future encounters 407 with the extinguished CS+. 408 The classical Pavlovian model proposes that extinction learning forms a new safety 409 memory, competing with the original threat memory acquired during conditioning(Bouton, 410 2002; Milad and Quirk, 2002) . Contextual changes often favor either the safe or threat 411 memory. Maren and colleagues (Maren et al., 2013; Maren and Quirk, 2004) described 412 the hippocampal-prefrontal-amygdala model for contextual memory. The hippocampus 413 projects to the basolateral amygdala, vmPFC, and dACC. Although the vmPFC supports 414 safe memory recall by projecting to the intercalated cells (which inhibit the central nucleus 415 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 19 of the amygdala ), the dACC enhances threat memory renewal by projecting to the 416 basolateral amygdala (which activates the central nucleus of this structure). Our findings 417 suggest that the thalamic connectivity may influence the balance between recalling safe 418 vs. threat memory by modulating the competitive interaction between the dACC and 419 vmPFC. The increased MD-dACC connectivity during recall may reflect a thalamic-driven 420 reconfiguration of prefrontal circuits, enabling the vmPFC to become more functionally 421 dominant and promote safe memory expression. On the other hand, the anterior pulvinar-422 vmPFC connectivity during renewal may destabilize the influence of the vmPFC, allowing 423 the dACC to regain dominance and facilitate threat memor y retrieval . The thalamic 424 connectivity, thus, appears to orchestrate a context-dependent functional balance 425 between the dACC and vmPFC. This balance may serve as a neural "toggle switch" 426 between safe vs. threat memory expression. We integrated this suggested flexible 427 responding mechanism to changing environmental contexts in Maren’s circuit model of 428 emotional memory, Fig. 7b. 429 The LGN contribution is consistent across all phases of threat learning, as evidenced by 430 our trial -wise results, which pointed to increased activation in response to CS+ that 431 diminished immediately after the first trial. This distinct pattern indicates readiness for 432 imminent visual stimuli, regardless of the actual emotional valence, as conditioning, 433 extinction, recall, and renewal elicited similar BOLD patterns. This aligns with previous 434 studies that highlighted the LGN’s engagement in selective attention and anticipation of 435 visual stimuli (Mahoney and Schmidt, 2024; O’Connor et al., 2002; Saalmann and 436 Kastner, 2009). The LGN, as a first-order nucleus(Cortes et al., 2024), primarily supports 437 a feedforward function, while the pulvinar, as part of the visual thalamus, is more closely 438 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 20 associated with a broader functional processing(Cortes et al., 2024) including emotional 439 valences of stimuli. This distinction underscores their joint but specialized contributions 440 to adaptive threat responses. 441 Although distinct thalamic roles in threat learning have been proposed, fMRI data do not 442 fully capture the complexity of this structure. Pulvinar divisions, MD, and LGN each 443 contain different neuron subtypes and f iner anatomical subdivisions, which may serve 444 diverse functions. Future advancements, such as higher-resolution human brain atlases, 445 could improve our ability to study these nuclei with better anatomical precision. 446 Additionally, since different sensory modalities preferentially engage distinct thalamic 447 nuclei, the specific thalamic roles we described may not be consistent across different 448 experimental designs, particularly in studies using auditory threat learning. The pulvinar 449 divisions’ relationships during conditioning are purely functional and might be supported 450 by direct or indirect anatomical projections. Finally, given the indirect nature of fMRI data 451 and the absence of direct brain signal manipulations, our findings should not be 452 interpreted as evidence of causality. Further research is needed to examine causal 453 mechanisms underlying dynamic neural representations of threat learning within the 454 thalamus. 455 This study’s insights raise critical future questions at the intersection between 456 neuroscience, psychology, and mental health , particularly regarding the neural 457 mechanisms underlying associative threat learning . First, the crucial role of the anterior 458 pulvinar and MD during the acquisition of the CS-US association sets the stage for 459 developing more precise brain interventions, focusing on refining maladaptive fear 460 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 21 reactions. Future studies could explore the optimal parameters for applying non-invasive 461 techniques to inhibit these thalamic regions during or immediately after fear conditioning. 462 This approach is particularly promising for clinical populations and individuals exposed to 463 high-risk environments, such as emergency medical staff, firefighters, and paramedics. 464 Second, the distinct contributions of the MD -dACC and anterior pulvinar-vmPFC circuits 465 to safe vs. threatening memory suggest potential intervention targets for prioritizing one 466 memory over the other . Activating the MD -dACC may facilitate recall ing safe memories 467 while engaging the anterior pulvinar-vmPFC circuits might reinforce fear memory 468 pathways. Experimental designs in controlled laboratory settings could target these 469 circuits to identify specific parameters for enhancing safe memor ies. This avenue holds 470 therapeutic potential , particularly in conditions such as aviophobia ; activating a safe 471 memory circuit before a flight might suppress fear relapse during the actual flight. 472 Third, the anterior pulvinar-MD relationships raise questions related to the wide range of 473 human behaviors, focusing on understanding neural gateways between conscious and 474 unconscious learning. Studying the dynamics of information flow across this promising 475 pathway could deepen our understanding of how sensory and cognitive information 476 transits between conscious and unconscious states and bring s about behaviors . This 477 research line may pave the way for innovative learning methods and reveal 478 neurocognitive mechanisms underlying the acquisition of new information in humans. 479 Together, our findings and emerging research directions underscore the thalamic nuclei’s 480 vital role as a hub for fear acquisition and memory processing, offering promising avenues 481 for theoretical advancements and clinical applications. As the primary neural gateway to 482 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 22 the human brain, the thalamus is the first station for all sensory information except 483 olfactory input s. Modulating its functions is potentially an impactful strategy to induce 484 widespread functional changes across the brain and influence different mental and 485 behavioral expressions. 486

487 Participants 488 We analyzed the fMRI data of 293-412 human subjects of both sexes, aged 18-70 years 489 old (M = 32.17 ± 13.1). All participants were proficient in English, right -handed, and had 490 normal or corrected-to-normal vision. The exclusion criteria included a history of seizures 491 or significant head trauma, current substance abuse or dependence, metal implants, 492 pregnancy, breastfeeding, or positive urine toxicology screen for drugs of abuse. We 493 followed the latest version of Helsinki's declaration, and all procedures were approved by 494 the Partners HealthCare Institute Review Board of the Massachusetts General Hospital, 495 Harvard Medical School. All subjects provided written informed consent before taking part 496 in the study. Results from this dataset have been published elsewhere with a different 497 focus(Marin et al., 2020; Wen et al., 2024, 2022). The current results are novel and have 498 not been previously published. 499 500 Experimental procedure 501 502 Participants underwent two -day sessions of a validated threat learning paradigm in the 503 MRI scanner (Figs. 1a, 4a-6a); the experimental contexts consisted of visual scenes on 504 a computer display. On day 1, participants underwent a Pavlovian conditioning phase in 505 which a neutral stimulus in context A (e.g., red light in an office) was paired with a 500ms 506 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 23 electric shock (US) with a partial reinforcement rate of 62.5%. Another neutral stimulus in 507 context A (e.g., blue light in the same office) was also presented but never paired with the 508 shock (CS-). Participants were guided to select a highly annoying but nonpainful shock 509 level during a pre -experiment calibration stage, and we used that personalized level 510 during conditioning. On the same day, we also conducted extinction learning in context B 511 (e.g., red and blue lights in a casebook background). The stimuli were repeatedly 512 presented with the removal of the expected reinforcement (i.e., no shock). 513 514 On day 2, the participants underwent two phases of a memory test. The first is extinction 515 recall, including presenting the extinguished CS+ and CS - within safe contextual cues 516 (context B used during extinction learning). The second is threat renewal, in which the 517 stimuli were presented within threat contextual cues (context A used in conditioning). 518 Both extinction recall and threat renewal included no US reinforcement. Across phases 519 of threat learning , the duration of each trial was 6s, and the inter -trial intervals with a 520 fixation screen ranged between 12 to 18 s (15 s on average). Finally, to control for 521 potential confounds related to the experimental design, we pseudorandomized and 522 counterbalanced the order of CS+ and CS− across phases and between subjects. 523 524 MRI data acquisition and preprocessing 525 Two MRI settings were used to acquire the neuroimaging data. The first is a Trio 3T whole-526 body MRI scanner (Siemens Medical Systems, Iselin, New Jersey) using an 8 -channel 527 head coil. The functional data in this setting were acquired using a T2* weighted echo -528 planar pulse sequence with these parameters: TR = 3.0s, TE = 30 ms, slice number = 45, 529 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 24 voxel size = 3 × 3 × 3 mm3. The second setting was also in the same scanner using a 32-530 channel head coil. The functional images were obtained using a T2* weighted echo-planar 531 pulse sequence using TR = 2.56s TE = 30 ms, slice number = 48, voxel 532 size = 3 × 3 × 3 mm3. The anatomical brain images were collected using a T1 -weighted 533 MP-RAGE pulse sequence , parcellated into 1 × 1 × 1 mm3 voxels. Elastic bands were 534 affixed to the head coil device to reduce head motions. 535 536 Using the default pipeline in fMRIPrep, version 20.0.2 (Esteban et al., 2020, 2019) , we 537 preprocessed the data and applied correction of slice timing, realignment of the functional 538 images, and coregistration. In addition, the data was normalized into the Montreal 539 Neurological Institute (MNI) space and smoothed with a 6 -mm full-width half-maximum 540 Gaussian kernel. 541 542 Activation analyses 543 We applied the least-squares-based generalized linear model (GLM) for each participant 544 to estimate the BOLD response to CS+ and CS - using SPM 12. We estimated the beta 545 values for each voxel during each learning phase in the paradigm. Overall, the model 546 included 32 regressors for the CS+ and CS-, a regressor for the context, and a regressor 547 for shock in conditioning but not in other phases . The GLM also included six head 548 movement parameters (x, y, z directions, and rotations). This first-level analysis resulted 549 in contrast maps that we used to estimate the variability of these maps across all subjects 550 at the group-level analysis. We then used the contrast maps from the group-level analysis 551 to extract the averaged values across the voxels within predefined masks of the pulvinar 552 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 25 divisions, MD, and LGN. These outputs were used to compare the BOLD response during 553 the first four CS+ and four CS- trials. We averaged the activation for each CS type across 554 trials to obtain the block-level activation. 555 556 The statistical analyses included t-tests, RM -ANOVA, network analyses, mediation 557 models, and hierarchical regression models. All were conducted using JASP versions 558 0.18.3 and 0.19.3(JASP Team, 2024). False discovery rate (FDR) correction was applied 559 across analyses. RM-ANOVA was used to analyze trial-by-trial BOLD responses as the 560 same participants were measured across the trials. Assumptions of sphericity were 561 checked, and Greenhouse-Geisser corrections were applied where necessary. 562 563 Connectivity analyses 564 We computed the connectivity values using the CONN functional connectivity toolbox , 565 version 22. a, for the MathWorks MATLAB program (Nieto-Castanon and Whitfield -566 Gabrieli, 2022; Whitfield -Gabrieli and Nieto -Castanon, 2012) . We first segmented the 567 brain images into different tissues of gray matter, white matter, and CSF. Then, we applied 568 the standard denoising pipeline in the CONN toolbox (Nieto-Castanon, 2020) to the 569 functional images to control the effect of potential confounding parameters, using a 570 component-based noise correction method . Finally, bandpass frequency filtering of the 571 BOLD time series between 0.008 Hz and 0.09 Hz was also applied. 572 We evaluated differences in connectivity between CS+ and CS− conditions using 573 generalized psychophysiological interaction (gPPI) analyses. We defined the first eight 574 trials of each condition as a block. The pulvinar divisions, MD, and LGN were defined as 575 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 26 individual seed regions, and connectivity was assessed with target regions, including the 576 dACC, sgACC, vmPFC, amygdala, and hippocampus. The model included seed BOLD 577 signals as physiological factors, boxcar signals characterizing the task conditions 578 convolved with an SPM canonical hemodynamic response function as psychological 579 factors, and the product of the two as psychophysiological interaction terms. Functional 580 connectivity changes were quantified using Fisher-transformed correlation coefficients of 581 the psychophysiological interaction terms. At the group level, we used a GLM to assess 582 task-related connectivity changes across participants. Differences in connectivity 583 between conditions were evaluated using paired t -tests, applying false discovery rate 584 (FDR) correction at p < 0.05 . 10,000 bootstrap resamples were used to obtain BCa CI , 585 providing more robust estimates of the connectivity mean differences. 586 Masks 587 We defined the pulvinar divisions, MD, and LGN nuclei using predefined masks based on 588 the neuroanatomical guidelines of the Automated Anatomical Labelling Atlas(Rolls et al., 589 2020). We also applied the same atlas guidelines to determine the masks for the two 590 other anatomical regions; the amygdala and hippocampus. The masks for the functional 591 regions were created using Neurosynth (Yarkoni et al., 2011) and the keyword 592 ‘conditioning.’ For each region, we created 8mm spheres around the following identified 593 peak coordinates: vmPFC (MNIxyz = −2, 46, −10), sgACC (MNIxyz = 0, 26, −12), and 594 dACC (MNIxyz = 0, 14, 28). 595 596 597 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 27 Acknowledgments 598 This work was supported by the National Institute of Mental Health grants 599 R01MH123736, R01MH125198, R33MH111907, R01MH097880, and R01MH097964 to 600 M.R.M. 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 28

639 Adolphs R, Tranel D, Damasio H, Damasio A. 1994. Impaired recognition of emotion in 640 facial expressions following bilateral damage to the human amygdala. Nature 1994 641 372:6507 372:669–672. doi:10.1038/372669a0 642 Arcaro MJ, Pinsk MA, Kastner S. 2015. The Anatomical and Functional Organization of 643 the Human Visual Pulvinar. Journal of Neuroscience 35:9848–9871. 644 doi:10.1523/JNEUROSCI.1575-14.2015 645 Badarnee M, Wen Z, Hammoud MZ, Glimcher P, Cain CK, Milad MR. 2025. Intersect 646 between brain mechanisms of conditioned threat, active avoidance, and reward. 647 Communications Psychology 3:32. doi:10.1038/S44271-025-00197-7 648 Battaglia S, Garofalo S, di Pellegrino G, Starita F. 2020. Revaluing the Role of vmPFC 649 in the Acquisition of Pavlovian Threat Conditioning in Humans. Journal of 650 Neuroscience 40:8491–8500. doi:10.1523/JNEUROSCI.0304-20.2020 651 Behrens TEJ, Johansen-Berg H, Woolrich MW, Smith SM, Wheeler-Kingshott CAM, 652 Boulby PA, Barker GJ, Sillery EL, Sheehan K, Ciccarelli O, Thompson AJ, Brady 653 JM, Matthews PM. 2003. Non-invasive mapping of connections between human 654 thalamus and cortex using diffusion imaging. Nat Neurosci 6:750–757. 655 doi:10.1038/NN1075;KWRD=BIOMEDICINE 656 Berman RA, Wurtz RH. 2010. Functional Identification of a Pulvinar Path from Superior 657 Colliculus to Cortical Area MT. Journal of Neuroscience 30:6342–6354. 658 doi:10.1523/JNEUROSCI.6176-09.2010 659 Berman RA, Wurtz RH. 2008. EXPLORING THE PULVINAR PATH TO VISUAL 660 CORTEX. Prog Brain Res 171:467. doi:10.1016/S0079-6123(08)00668-7 661 Bertini C, Pietrelli M, Braghittoni D, Làdavas E. 2018. Pulvinar lesions disrupt fear-662 related implicit visual processing in hemianopic patients. Front Psychol 9:412050. 663 doi:10.3389/FPSYG.2018.02329/BIBTEX 664 Bouton ME. 2002. Context, ambiguity, and unlearning: Sources of relapse after 665 behavioral extinction. Biol Psychiatry 52:976–986. doi:10.1016/S0006-666 3223(02)01546-9 667 Bridge H, Leopold DA, Bourne JA. 2015. Adaptive pulvinar circuitry supports visual 668 cognition. Trends Cogn Sci 20:146. doi:10.1016/J.TICS.2015.10.003 669 Carr JA. 2015. I’ll take the low road: The evolutionary underpinnings of visually triggered 670 fear. Front Neurosci 9:165437. doi:10.3389/FNINS.2015.00414/BIBTEX 671 Casanova C, Chalupa LM. 2023. The dorsal lateral geniculate nucleus and the pulvinar 672 as essential partners for visual cortical functions. Front Neurosci 17:1258393. 673 doi:10.3389/FNINS.2023.1258393/BIBTEX 674 Cortes N, Ladret HJ, Abbas-Farishta R, Casanova C. 2024. The pulvinar as a hub of 675 visual processing and cortical integration. Trends Neurosci 47:120–134. 676 doi:10.1016/J.TINS.2023.11.008 677 Costa M, Lozano-Soldevilla D, Gil-Nagel A, Toledano R, Oehrn CR, Kunz L, Yebra M, 678 Mendez-Bertolo C, Stieglitz L, Sarnthein J, Axmacher N, Moratti S, Strange BA. 679 2022. Aversive memory formation in humans involves an amygdala-hippocampus 680 phase code. Nature Communications 2022 13:1 13:1–16. doi:10.1038/s41467-022-681 33828-2 682 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 29 Esteban O, Ciric R, Finc K, Blair RW, Markiewicz CJ, Moodie CA, Kent JD, Goncalves 683 M, DuPre E, Gomez DEP, Ye Z, Salo T, Valabregue R, Amlien IK, Liem F, Jacoby N, 684 Stojić H, Cieslak M, Urchs S, Halchenko YO, Ghosh SS, De La Vega A, Yarkoni T, 685 Wright J, Thompson WH, Poldrack RA, Gorgolewski KJ. 2020. Analysis of task-686 based functional MRI data preprocessed with fMRIPrep. Nat Protoc 15:2186–2202. 687 doi:10.1038/S41596-020-0327-3 688 Esteban O, Markiewicz CJ, Blair RW, Moodie CA, Isik AI, Erramuzpe A, Kent JD, 689 Goncalves M, DuPre E, Snyder M, Oya H, Ghosh SS, Wright J, Durnez J, Poldrack 690 RA, Gorgolewski KJ. 2019. fMRIPrep: a robust preprocessing pipeline for functional 691 MRI. Nat Methods 16:111–116. doi:10.1038/S41592-018-0235-4 692 Fanselow MS, LeDoux JE. 1999. Why we think plasticity underlying pavlovian fear 693 conditioning occurs in the basolateral amygdala. Neuron 23:229–232. 694 doi:10.1016/S0896-6273(00)80775-8 695 Foygel R, Drton M. 2010. Extended Bayesian Information Criteria for Gaussian 696 Graphical Models. Adv Neural Inf Process Syst 23. 697 Fullana MA, Harrison BJ, Soriano-Mas C, Vervliet B, Cardoner N, Àvila-Parcet A, Radua 698 J. 2015. Neural signatures of human fear conditioning: an updated and extended 699 meta-analysis of fMRI studies. Molecular Psychiatry 2016 21:4 21:500–508. 700 doi:10.1038/mp.2015.88 701 Griffiths BJ, Zaehle T, Repplinger S, Schmitt FC, Voges J, Hanslmayr S, Staudigl T. 702 2022. Rhythmic interactions between the mediodorsal thalamus and prefrontal 703 cortex precede human visual perception. Nature Communications 2022 13:1 13:1–704 11. doi:10.1038/s41467-022-31407-z 705 Halassa MM, Sherman SM. 2019. Thalamo-cortical circuit motifs: a general framework. 706 Neuron 103:762. doi:10.1016/J.NEURON.2019.06.005 707 Halverson HE, Freeman JH. 2010. Ventral lateral geniculate input to the medial pons is 708 necessary for visual eyeblink conditioning in rats. Learn Mem 17:80–85. 709 doi:10.1101/LM.1572710 710 Herry C, Ferraguti F, Singewald N, Letzkus JJ, Ehrlich I, Lüthi A. 2010. Neuronal circuits 711 of fear extinction. European Journal of Neuroscience 31:599–612. 712 doi:10.1111/J.1460-9568.2010.07101.X 713 Homman-Ludiye J, Bourne JA. 2019. The medial pulvinar: function, origin and 714 association with neurodevelopmental disorders. J Anat 235:507–520. 715 doi:10.1111/JOA.12932 716 Hummos A, Wang BA, Drammis S, Halassa MM, Pleger B. 2022. Thalamic regulation of 717 frontal interactions in human cognitive flexibility. PLoS Comput Biol 18:e1010500. 718 doi:10.1371/JOURNAL.PCBI.1010500 719 Hwang K, Bertolero MA, Liu WB, D’Esposito M. 2017. The Human Thalamus Is an 720 Integrative Hub for Functional Brain Networks. Journal of Neuroscience 37:5594–721 5607. doi:10.1523/JNEUROSCI.0067-17.2017 722 Hwang K, Bruss J, Tranel D, Boes AD. 2020. Network Localization of Executive 723 Function Deficits in Patients with Focal Thalamic Lesions. J Cogn Neurosci 724 32:2303. doi:10.1162/JOCN_A_01628 725 JASP Team. 2024. JASP (Versions 0.18.3 and 0.19.3). 726 Koller K, Rafal RD, Platt A, Mitchell ND. 2019. Orienting toward threat: Contributions of 727 a subcortical pathway transmitting retinal afferents to the amygdala via the superior 728 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 30 colliculus and pulvinar. Neuropsychologia 128:78–86. 729 doi:10.1016/J.NEUROPSYCHOLOGIA.2018.01.027 730 Kolling N, Wittmann M, Rushworth MFS. 2014. Multiple neural mechanisms of decision 731 making and their competition under changing risk pressure. Neuron 81:1190–1202. 732 doi:10.1016/J.NEURON.2014.01.033/ATTACHMENT/A9984A25-8961-453B-8E1A-733 778EAD3A0ADE/MMC1.PDF 734 Konrad C, Neuhoff L, Adolph D, Goerigk S, Herbert JS, Jagusch-Poirier J, Weigelt S, 735 Seehagen S, Schneider S. 2024. Associative learning via eyeblink conditioning 736 differs by age from infancy to adulthood. Communications Psychology 2024 2:1 737 2:1–13. doi:10.1038/s44271-024-00176-4 738 Korn CW, Bach DR. 2019. Minimizing threat via heuristic and optimal policies recruits 739 hippocampus and medial prefrontal cortex. Nature Human Behaviour 2019 3:7 740 3:733–745. doi:10.1038/s41562-019-0603-9 741 Korn CW, Bach DR. 2018. Heuristic and optimal policy computations in the human brain 742 during sequential decision-making. Nature Communications 2018 9:1 9:1–15. 743 doi:10.1038/s41467-017-02750-3 744 Kragel PA, Čeko M, Theriault J, Chen D, Satpute AB, Wald LW, Lindquist MA, Feldman 745 Barrett L, Wager TD. 2021. A human colliculus-pulvinar-amygdala pathway 746 encodes negative emotion. Neuron 109:2404-2412.e5. 747 doi:10.1016/J.NEURON.2021.06.001/ATTACHMENT/2E7063C2-7363-4DE3-9A2B-748 F75C55D9DD96/MMC3.PDF 749 Krasne FB, Zinn R, Vissel B, Fanselow MS, Franklin Krasne CB, Howland-Rose V, 750 Battersby D, King D. 2021. Extinction and discrimination in a Bayesian model of 751 context fear conditioning (BaconX). Hippocampus 31:790–814. 752 doi:10.1002/HIPO.23298 753 Le Q Van, Isbell LA, Matsumoto J, Nguyen M, Hori E, Maior RS, Tomaz C, Tran AH, 754 Ono T, Nishijo H. 2013. Pulvinar neurons reveal neurobiological evidence of past 755 selection for rapid detection of snakes. Proc Natl Acad Sci U S A 110:19000–756 19005. doi:10.1073/PNAS.1312648110/SUPPL_FILE/PNAS.201312648SI.PDF 757 LeDoux JE, Pine DS. 2016. Using Neuroscience to Help Understand Fear and Anxiety: 758 A Two-System Framework. https://doi.org/101176/appi.ajp201616030353 759 173:1083–1093. doi:10.1176/APPI.AJP.2016.16030353 760 Lee S, Shin HS. 2016. The role of mediodorsal thalamic nucleus in fear extinction. J 761 Anal Sci Technol 7:1–5. doi:10.1186/S40543-016-0093-6/FIGURES/1 762 Lee Sukchan, Ahmed T, Lee Soojung, Kim H, Choi S, Kim DS, Kim SJ, Cho J, Shin HS. 763 2011. Bidirectional modulation of fear extinction by mediodorsal thalamic firing in 764 mice. Nature Neuroscience 2011 15:2 15:308–314. doi:10.1038/nn.2999 765 Li K, Fan L, Cui Y, Wei X, He Y, Yang J, Lu Y, Li W, Shi W, Cao L, Cheng L, Li A, You B, 766 Jiang T. 2022. The human mediodorsal thalamus: Organization, connectivity, and 767 function. Neuroimage 249:118876. doi:10.1016/J.NEUROIMAGE.2022.118876 768 Li XB, Inoue T, Nakagawa S, Koyama T. 2004. Effect of mediodorsal thalamic nucleus 769 lesion on contextual fear conditioning in rats. Brain Res 1008:261–272. 770 doi:10.1016/J.BRAINRES.2004.02.038 771 Lithari C, Moratti S, Weisz N. 2015. Thalamocortical interactions underlying visual fear 772 conditioning in humans. Hum Brain Mapp 36:4592. doi:10.1002/HBM.22940 773 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 31 Mahoney HL, Schmidt TM. 2024. The cognitive impact of light: illuminating ipRGC circuit 774 mechanisms. Nature Reviews Neuroscience 2024 25:3 25:159–175. 775 doi:10.1038/s41583-023-00788-5 776 Maren S. 2001. Nuerobiology of Pavlovian fear conditioning. Annu Rev Neurosci 777 24:897–931. doi:10.1146/ANNUREV.NEURO.24.1.897/CITE/REFWORKS 778 Maren S, Phan KL, Liberzon I. 2013. The contextual brain: implications for fear 779 conditioning, extinction and psychopathology. Nature Reviews Neuroscience 2013 780 14:6 14:417–428. doi:10.1038/nrn3492 781 Maren S, Quirk GJ. 2004. Neuronal signalling of fear memory. Nature Reviews 782 Neuroscience 2004 5:11 5:844–852. doi:10.1038/nrn1535 783 Marin MF, Hammoud MZ, Klumpp H, Simon NM, Milad MR. 2020. Multimodal 784 Categorical and Dimensional Approaches to Understanding Threat Conditioning 785 and Its Extinction in Individuals With Anxiety Disorders. JAMA Psychiatry 77:618–786 627. doi:10.1001/JAMAPSYCHIATRY.2019.4833 787 McFadyen J, Mattingley JB, Garrido MI. 2019. An afferent white matter pathway from 788 the pulvinar to the amygdala facilitates fear recognition. Elife 8. 789 doi:10.7554/ELIFE.40766 790 Milad MR, Furtak SC, Greenberg JL, Keshaviah A, Im JJ, Falkenstein MJ, Jenike M, 791 Rauch SL, Wilhelm S. 2013. Deficits in Conditioned Fear Extinction in Obsessive-792 Compulsive Disorder and Neurobiological Changes in the Fear Circuit. JAMA 793 Psychiatry 70:608–618. doi:10.1001/JAMAPSYCHIATRY.2013.914 794 Milad MR, Pitman RK, Ellis CB, Gold AL, Shin LM, Lasko NB, Zeidan MA, Handwerger 795 K, Orr SP, Rauch SL. 2009. Neurobiological Basis of Failure to Recall Extinction 796 Memory in Posttraumatic Stress Disorder. Biol Psychiatry 66:1075–1082. 797 doi:10.1016/J.BIOPSYCH.2009.06.026 798 Milad MR, Quirk GJ. 2012. Fear Extinction as a Model for Translational Neuroscience: 799 Ten Years of Progress. Annu Rev Psychol 63:129. 800 doi:10.1146/ANNUREV.PSYCH.121208.131631 801 Milad MR, Quirk GJ. 2002. Neurons in medial prefrontal cortex signal memory for fear 802 extinction. Nature 2002 420:6911 420:70–74. doi:10.1038/nature01138 803 Milad MR, Rosenbaum BL, Simon NM. 2014. Neuroscience of fear extinction: 804 Implications for assessment and treatment of fear-based and anxiety related 805 disorders. Behaviour Research and Therapy 62:17–23. 806 doi:10.1016/J.BRAT.2014.08.006 807 Milad MR, Wright CI, Orr SP, Pitman RK, Quirk GJ, Rauch SL. 2007. Recall of fear 808 extinction in humans activates the ventromedial prefrontal cortex and hippocampus 809 in concert. Biol Psychiatry 62:446–454. doi:10.1016/J.BIOPSYCH.2006.10.011 810 Mukherjee A, Lam NH, Wimmer RD, Halassa MM. 2021. Thalamic circuits for 811 independent control of prefrontal signal and noise. Nature 2021 600:7887 600:100–812 104. doi:10.1038/s41586-021-04056-3 813 Nejati V, Majdi R, Salehinejad MA, Nitsche MA. 2021. The role of dorsolateral and 814 ventromedial prefrontal cortex in the processing of emotional dimensions. Scientific 815 Reports 2021 11:1 11:1–12. doi:10.1038/s41598-021-81454-7 816 Nieto-Castanon A. 2020. FMRI denoising pipeline. Handbook of functional connectivity 817 Magnetic Resonance Imaging methods in CONN 17–25. 818 doi:10.56441/HILBERTPRESS.2207.6600 819 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 32 Nieto-Castanon A, Whitfield-Gabrieli S. 2022. CONN functional connectivity toolbox: 820 RRID SCR_009550, release 22. CONN functional connectivity toolbox: RRID 821 SCR_009550, release 22. doi:10.56441/HILBERTPRESS.2246.5840 822 O’Connor DH, Fukui MM, Pinsk MA, Kastner S. 2002. Attention modulates responses in 823 the human lateral geniculate nucleus. Nature Neuroscience 2002 5:11 5:1203–824 1209. doi:10.1038/nn957 825 Pavlov I. 1904. Nobel Lecture. NobelPrize.org. Available at: 826 https://www.nobelprize.org/prizes/medicine/1904/pavlov/lecture/ 827 Paydar A, Lee B, Gangadharan G, Lee S, Hwang EM, Shin HS. 2014. Extrasynaptic 828 GABAA receptors in mediodorsal thalamic nucleus modulate fear extinction 829 learning. Mol Brain 7:39. doi:10.1186/1756-6606-7-39 830 Penzo MA, Robert V, Tucciarone J, De Bundel D, Wang M, Van Aelst L, Darvas M, 831 Parada LF, Palmiter RD, He M, Huang ZJ, Li B. 2015. The paraventricular thalamus 832 controls a central amygdala fear circuit. Nature 519:455–459. 833 doi:10.1038/NATURE13978 834 Pessoa L, Adolphs R. 2010. Emotion processing and the amygdala: from a “low road” to 835 “many roads” of evaluating biological significance. Nature Reviews Neuroscience 836 2010 11:11 11:773–782. doi:10.1038/nrn2920 837 Phelps EA, LeDoux JE. 2005. Contributions of the amygdala to emotion processing: 838 From animal models to human behavior. Neuron 48:175–187. 839 doi:10.1016/J.NEURON.2005.09.025/ASSET/31E14E48-ACAC-4132-ACD2-840 31705A61F19C/MAIN.ASSETS/GR4.JPG 841 Rafal RD, Koller K, Bultitude JH, Mullins P, Ward R, Mitchell AS, Bell AH. 2015. 842 Connectivity between the superior colliculus and the amygdala in humans and 843 macaque monkeys: virtual dissection with probabilistic DTI tractography. J 844 Neurophysiol 114:1947–1962. doi:10.1152/JN.01016.2014 845 Ramanathan KR, Jin J, Giustino TF, Payne MR, Maren S. 2018. Prefrontal projections 846 to the thalamic nucleus reuniens mediate fear extinction. Nature Communications 847 2018 9:1 9:1–12. doi:10.1038/s41467-018-06970-z 848 Ramanathan KR, Maren S. 2019. Nucleus reuniens mediates the extinction of 849 contextual fear conditioning. Behavioural Brain Research 374:112114. 850 doi:10.1016/J.BBR.2019.112114 851 Ratigan HC, Krishnan S, Smith S, Sheffield MEJ. 2023. A thalamic-hippocampal CA1 852 signal for contextual fear memory suppression, extinction, and discrimination. 853 Nature Communications 2023 14:1 14:1–17. doi:10.1038/s41467-023-42429-6 854 Rolls ET, Huang CC, Lin CP, Feng J, Joliot M. 2020. Automated anatomical labelling 855 atlas 3. Neuroimage 206. doi:10.1016/J.NEUROIMAGE.2019.116189 856 Saalmann YB, Kastner S. 2011. Cognitive and Perceptual Functions of the Visual 857 Thalamus. Neuron 71:209. doi:10.1016/J.NEURON.2011.06.027 858 Saalmann YB, Kastner S. 2009. Gain control in the visual thalamus during perception 859 and cognition. Curr Opin Neurobiol 19:408. doi:10.1016/J.CONB.2009.05.007 860 Salay LD, Huberman AD. 2021. Divergent outputs of the ventral lateral geniculate 861 nucleus mediate visually evoked defensive behaviors. Cell Rep 37. 862 doi:10.1016/j.celrep.2021.109792 863 Sherman SM. 2016. Thalamus plays a central role in ongoing cortical functioning. 864 Nature Neuroscience 2016 19:4 19:533–541. doi:10.1038/nn.4269 865 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 33 Sherman SM. 2007. The thalamus is more than just a relay. Curr Opin Neurobiol 866 17:417–422. doi:10.1016/J.CONB.2007.07.003 867 Sherman SMurray, Guillery RW. 2006. Exploring the thalamus and its role in cortical 868 function 484. 869 Shi C, Davis M. 2001. Visual pathways involved in fear conditioning measured with fear-870 potentiated startle: behavioral and anatomic studies. J Neurosci 21:9844–9855. 871 doi:10.1523/JNEUROSCI.21-24-09844.2001 872 Silverstein DN, Ingvar M. 2015. A multi-pathway hypothesis for human visual fear 873 signaling. Front Syst Neurosci 9:137726. doi:10.3389/FNSYS.2015.00101/PDF 874 Steinmetz AB, Buss EW, Freeman JH. 2013. Inactivation of the ventral lateral geniculate 875 and nucleus of the optic tract impairs retention of visual eyeblink conditioning. 876 Behavioral Neuroscience 127:690–693. doi:10.1037/A0033729 877 Takakuwa N, Isa K, Onoe H, Takahashi J, Isa T. 2021. Contribution of the Pulvinar and 878 Lateral Geniculate Nucleus to the Control of Visually Guided Saccades in Blindsight 879 Monkeys. Journal of Neuroscience 41:1755–1768. doi:10.1523/JNEUROSCI.2293-880 20.2020 881 Totty MS, Tuna T, Ramanathan KR, Jin J, Peters SE, Maren S. 2023. Thalamic nucleus 882 reuniens coordinates prefrontal-hippocampal synchrony to suppress extinguished 883 fear. Nature Communications 2023 14:1 14:1–12. doi:10.1038/s41467-023-42315-1 884 Tovote P, Fadok JP, Lüthi A. 2015. Neuronal circuits for fear and anxiety. Nature 885 Reviews Neuroscience 2015 16:6 16:317–331. doi:10.1038/nrn3945 886 Vuilleumier P, Armony JL, Driver J, Dolan RJ. 2003. Distinct spatial frequency 887 sensitivities for processing faces and emotional expressions. Nature Neuroscience 888 2003 6:6 6:624–631. doi:10.1038/nn1057 889 Ward R, Danziger S, Bamford S. 2005. Response to visual threat following damage to 890 the pulvinar. Curr Biol 15:571–573. doi:10.1016/J.CUB.2005.01.056 891 Wei P, Liu N, Zhang Z, Liu X, Tang Y, He X, Wu B, Zhou Z, Liu Y, Li J, Zhang Y, Zhou X, 892 Xu L, Chen L, Bi G, Hu X, Xu F, Wang L. 2015. Processing of visually evoked 893 innate fear by a non-canonical thalamic pathway. Nature Communications 2015 6:1 894 6:1–13. doi:10.1038/ncomms7756 895 Wen Z, Pace-Schott EF, Lazar SW, Rosén J, Åhs F, Phelps EA, LeDoux JE, Milad MR. 896 2024. Distributed neural representations of conditioned threat in the human brain. 897 Nature Communications 2024 15:1 15:1–14. doi:10.1038/s41467-024-46508-0 898 Wen Z, Raio CM, Pace-Schott EF, Lazar SW, LeDoux JE, Phelps EA, Milad MR. 2022. 899 Temporally and anatomically specific contributions of the human amygdala to threat 900 and safety learning. Proc Natl Acad Sci U S A 119:e2204066119. 901 doi:10.1073/PNAS.2204066119/SUPPL_FILE/PNAS.2204066119.SAPP.PDF 902 Whitfield-Gabrieli S, Nieto-Castanon A. 2012. Conn: a functional connectivity toolbox for 903 correlated and anticorrelated brain networks. Brain Connect 2:125–141. 904 doi:10.1089/BRAIN.2012.0073 905 Whyte CJ, Redinbaugh MJ, Shine JM, Saalmann YB. 2024. Thalamic contributions to 906 the state and contents of consciousness. Neuron 112:1611–1625. 907 doi:10.1016/J.NEURON.2024.04.019/ASSET/AA96B159-0E29-4D7D-B3C8-908 E302EA442A73/MAIN.ASSETS/GR2.JPG 909 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 34 Wolff M, Halassa MM. 2024. The mediodorsal thalamus in executive control. Neuron 910 112:893–908. doi:10.1016/J.NEURON.2024.01.002/ASSET/8EBBDE5D-737B-911 479F-966E-07AE8F092DB6/MAIN.ASSETS/GR3.JPG 912 Wurtz RH, McAlonan K, Cavanaugh J, Berman RA. 2011. Thalamic pathways for active 913 vision. Trends Cogn Sci 15:177–184. doi:10.1016/J.TICS.2011.02.004 914 Yarkoni T, Poldrack RA, Nichols TE, Van Essen DC, Wager TD. 2011. Large-scale 915 automated synthesis of human functional neuroimaging data. Nature Methods 2011 916 8:8 8:665–670. doi:10.1038/nmeth.1635 917 Zheng C, Huang Y , Bo B, Wei L, Liang Z, Wang Z. 2020. Projection from the Anterior 918 Cingulate Cortex to the Lateral Part of Mediodorsal Thalamus Modulates Vicarious 919 Freezing Behavior. Neurosci Bull 36:217–229. doi:10.1007/S12264-019-00427-920 Z/METRICS 921 922 923 924 925 926 927 928 929 930 931 932 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 35 Figure captions 933 Fig 1. Neural representation of associative threat learning in pulvinar divisions, MD 934 and LGN. 935 a. Human f ear conditioning paradigm in the fMRI. b-d. Means ± SE of activation in 936 response to CS+ vs. CS - at both block-wise and trial -wise levels within the pulvinar 937 divisions (b), MD (c), and LGN (d). e. ROI-to-ROI connectivity . Right: the red line 938 represents significant positive connectivity to CS+ vs. CS - (PFDR < 0.05), while the gray 939 lines indicate no n-significant connectivity. Thalamic nuclei that showed no significant 940 connectivity with other regions (i.e., amygdala, hippocampus, etc.) were omitted from the 941 visualization. Left: boxplots and kernel density estimates illustrate the distribution of 942 connectivity values in response to CS+ and CS-. 943 SE: Standard error. 944 MD: Mediodorsal thalamus. 945 LGN: Lateral geniculate nucleus. 946 CS+ vs. CS-: Conditioned stimulus predicting shock vs. no shock. 947 ROI: Region of interest. 948 *p<0.05, **p<0.01, ***p<0.001 949 Display items in panel a were created using BioRender (BioRender.com). 950 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 36 Fig 2. Quantifying the relationships between the anterior pulvinar and MD during 951 conditioning. 952 a. Hierarchical regression models of trial-wise relationships between the anterior pulvinar 953 and MD activations while controlling for a potential effect of anatomical proximity. The 954 effect of anatomical proximity was controlled by progressively adding other pulvinar 955 divisions as controls. b. Results of a hierarchical model that included the LGN as an 956 alternative control, distinct from the pulvinar. c. Comparison of activation levels in the 957 anterior pulvinar and MD at both block-wise and trial-wise levels (means ± SE). 958 *p<0.05, **p<0.01, ***p<0.001 959 MD: Mediodorsal thalamus. 960 LGN: Lateral geniculate nucleus. 961 SE: Standard error. 962 963 964 965 966 967 968 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 37 Fig 3. A d ata-driven approach to understanding the functional relationships 969 between pulvinar divisions during conditioning 970 a. Means ± SE of a ctivation differences between pulvinar divisions. b. Network analysis 971 reveals that the medial pulvinar serves as a central hub, mediating interactions among 972 pulvinar divisions and exhibiting increased centrality measures. c. Schematic 973 visualization of activation patterns we observed in pulvinar divisions. d. Previous studies 974 suggest that the inferior and lateral pulvinar are involved in processing basic visual 975 information, while the medial pulvinar is associated with higher -level functions, including 976 working memory. e. Based on b-d, we hypothesize that the medial pulvinar mediates the 977 relationships with other divisions. f. Mediation analysis supports our hypothesis (panel e). 978 g. Validation of the mediation model on an additional independent sample. Dashed paths 979 in panels f and g represent statistically unstable paths, while the continuous paths 980 indicate stable paths. 981 SE: Standard error. 982 SC: Superior colliculus. 983 *p<0.05, **p<0.01, ***p<0.001 984 Display items in panels d and e were created using BioRender (BioRender.com). 985 986 987 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 38 988 Fig 4. Neural representation of extinction learning in pulvinar divisions, MD and 989 LGN. 990 a. Human extinction learning paradigm in the fMRI. b-d. Means ± SE of activation in 991 response to extinguished CS+ vs. CS- at both block-wise and trial-wise levels within the 992 pulvinar divisions (b), MD (c), and LGN (d). e. ROI-to-ROI connectivity. Right: the red line 993 represents significant positive connectivity to extinguished CS+ vs. CS - (PFDR < 0.05), 994 while the gray lines indicate no significant connectivity. Thalamic nuclei that showed non-995 significant connectivity with other regions (i.e., amygdala, hippocampus , etc. ) were 996 omitted from the visualization. Left: boxplots and kernel density estimates illustrate the 997 distribution of connectivity values in response to extinguished CS+ and CS-. 998 SE: Standard error. 999 MD: Mediodorsal thalamus. 1000 LGN: Lateral geniculate nucleus. 1001 Extinguished CS+ vs. CS-: A conditioned stimulus that no longer predicts shock vs. a 1002 stimulus that was never paired with shock. 1003 ROI: Region of interest. 1004 *p<0.05, **p<0.01, ***p<0.001 1005 Display items in panel a were created using BioRender (BioRender.com). 1006 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 39 1007 Fig 5. Neural representation of extinction recall in pulvinar divisions, MD and LGN. 1008 a. Human extinction recall paradigm in the fMRI conducted within safe contextual cues. 1009 b-d. Means ± SE of a ctivation in response to extinguished CS+ vs. CS - at both block -1010 wise and trial-wise levels within the pulvinar divisions ( b), MD (c), and LGN (d). e. ROI-1011 to-ROI connectivity. Right: t he red line represents significant positive connectivity to 1012 extinguished CS+ vs. CS - (PFDR < 0.05), while the gray lines indicate no n-significant 1013 connectivity. Thalamic nuclei that showed no significant connectivity with other regions 1014 (i.e., amygdala, hippocampus, etc.) were omitted from the visualization. Left: boxplots and 1015 kernel density estimates illustrate the distribution of connectivity values in response to 1016 extinguished CS+ and CS-. 1017 SE: Standard error. 1018 MD: Mediodorsal thalamus. 1019 LGN: Lateral geniculate nucleus. 1020 Extinguished CS+ vs. CS -: A conditioned stimulus that no longer predicts shock vs. a 1021 stimulus that was never paired with shock. 1022 ROI: Region of interest. 1023 *p<0.05, **p<0.01, ***p<0.001 1024 Display items in panel a were created using BioRender (BioRender.com). 1025 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 40 1026 Fig 6. Neural representation of threat renewal in pulvinar divisions, MD and LGN. 1027 a. Human threat renewal paradigm in the fMRI conducted within threat contextual cues in 1028 the original context where fear conditioning occurred . b-d. Means ± SE of activation in 1029 response to extinguished CS+ vs. CS- at both block-wise and trial-wise levels within the 1030 pulvinar divisions (b), MD (c), and LGN (d). e. ROI-to-ROI connectivity. Right: the red line 1031 represents significant positive connectivity to extinguished CS+ vs. CS - (PFDR < 0.05), 1032 while the gray lines indicate non-significant connectivity. Thalamic nuclei that showed no 1033 significant connectivity with other regions (i.e., amygdala, hippocampus , etc. ) were 1034 omitted from the visualization. Left: boxplots and kernel density estimates illustrate the 1035 distribution of connectivity values in response to extinguished CS+ and CS-. 1036 SE: Standard error. 1037 MD: Mediodorsal thalamus. 1038 LGN: Lateral geniculate nucleus. 1039 Extinguished CS+ vs. CS -: A conditioned stimulus that no longer predicts shock vs. a 1040 stimulus that was never paired with shock. 1041 ROI: Region of interest. 1042 *p<0.05, **p<0.01, ***p<0.001 1043 Display items in panel a were created using BioRender (BioRender.com). 1044 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 41 1045 Fig 7. Neurobehavioral models of thalamic involvement in associative threat 1046 learning and memory. 1047 a. Schematic illustration of thalamic circuitry during the acquisition of associative threat 1048 learning, highlighting interactions between pulvinar divisions, MD, and LGN with key brain 1049 regions involved in fear expression. 1050 b. A thalamic-dependent “toggle switch” regulates the retrieval of safety vs. threat-related 1051 memory. The MD -dACC connectivity modulates the interaction between dACC and 1052 vmPFC, promoting vmPFC dominance during extinction recall. In contrast, the anterior 1053 pulvinar-vmPFC connectivity promotes dACC dominance, enhancing the expression of 1054 threat memory during threat renewal. 1055 MD: Mediodorsal thalamus. 1056 LGN: Lateral geniculate nucleus. 1057 vmPFC = Ventromedial prefrontal cortex. 1058 dACC = Dorsal anterior cingulate cortex. 1059 V1, V2, and V4: Primary, secondary, and fourth visual areas. 1060 TEO, TE: Temporal cortex regions. 1061 Hypo.: Hypothalamus. 1062 Hippo.: Hippocampus. 1063 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint 42 1064 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials Blocks comparison Trial-by-trial comparison ** *** *** * *** ** * * ***** ** *** *** *** *** CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** *** CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal Mediodorsal thalamus 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** *** *** Anterior Medial Inferior Lateral Pulvinar divisions Lateral geniculate nucleus Blocks comparison Trial-by-trial comparison Blocks comparison Trial-by-trial comparison b c d a Anterior pulvinar Amygdala dACC vmPFC sgACC Hippocampus e Connectivity Anterior pulvinar vmPFC Amygdala Hippocampus * * * .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint b 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials ** * 0.0 0.1 0.2 0.3BOLD signal Anterior Pulvinar MD *** Block comparison Trial-by-trial comparisons c Anterior pulvinar MD r = 0.66*** Trial 2 Anterior pulvinar MD r = 0.66*** Raw correlation Hierarchical regression R2 change Standard 𝛃 estimations -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1x104 bootstraps 𝛃 estimations -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Trial 1 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 2 3 4 models 1 2 3 4 models 44.1% 1 2 3 4 models 1 2 3 4 models 43% Are the activation levels significantly different between anterior pulvinar and MD? *** *** *** *** *** ** *** *** *** *** a R2 change = 0.006 R2 change = 0.01 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1x104 bootstraps 𝛃 estimations Conservative control: Other pulvinar divisions Alternative control: LGN Inferior pulvinar Anterior pulvinar Lateral pulvinar Medial pulvinar LGN *** *** *** *** -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Trial 3 Anterior pulvinar MD r = 0.77*** Anterior pulvinar MD Trial 4 r = 0.78*** -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 2 3 4 models 1 2 3 4 models 1 2 3 4 models 1 2 3 4 models 59.4% 60.7% *** R2 change = 0.03 R2 change = 0.001 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 *** *** *** *** *** *** *** *** *** * * *** *** *** *** *** *** *** *** *** * ** *** *** *** * * ** *** *** *** *** Is the observed similarity in activation between the anterior pulvinar and MD statistically significant, or could it be attributed to anatomical proximity? 1 2 3 4 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials Anterior pulvinar MD .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint A = anterior pulvinar B = medial pulvinar C = inferior pulvinar D = lateral pulvinar BOLD signal = CS+ minus CS- e B-C B-A D-A B-D C-D C-A Sample Bootstrap mean Edges A B C D C B A B C D 1x104 Bootstraps Centrality differences based on 1x104 Bootstraps B C Strength Closeness B C D A B C Betweenness Not stable Stable ba Activation Centrality measures Z-scores Trial 1 Trial 2 Trial 1 Trial 2 CS+ CS- Anterior pulvinar Other sub-pulvinars Schematic representation Our findings of the activation pattern in conditioning Network plots Anterior pulvinar may integrate information after initial processing in other sub-pulvinars. Medial pulvinar may act as a relay or convergence point between the sensory-oriented inferior and lateral pulvinars and the integrative anterior pulvinar. Understudied paths Lateral pulvinar Medial pulvinar Inferior pulvinar Anterior pulvinar … dc Previous studies showed: Network stability A B C D 0.0 0.1 0.2 0.3BOLD signal -2 -1 0 1 2 A B C D -0.4 -0.2 0 0.2 0.4 0.6 0.8 | | | | | | | test Hypothesis Mediation model using K-fold cross-validation techniques f Lateral pulvinar Medial pulvinar Inferior pulvinar Anterior pulvinar 0.49 0.25 0.48 -0.12 0.73 Training Test Training Test Training Test 3 folds Random allocation into 3-folds Sample N = 293 Sub-samples each n ~ 91- 106 Path coefficient Overall sample Aggregated paths coefficients -0.4 -0.2 0.0 0.2 0.4 0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 R2 Sample Lateral pulvinar Medial pulvinar Inferior pulvinar Anterior pulvinar 0.42 0.25 0.49 -0.05 0.71 Independent cohortg N = 114 R2 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint Blocks comparison Trial-by-trial comparison Mediodorsal thalamus Anterior Medial Inferior Lateral Pulvinar divisions Lateral geniculate nucleus Blocks comparison Blocks comparison Trial-by-trial comparison b c CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials ** ** CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials ** * * CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials * CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal ** Trial-by-trial comparison CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** ** Extinguished a d LGN Hippocampus Amygdala dACC vmPFC sgACC e Connectivity LGN sgACC * .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint Mediodorsal thalamus Anterior Medial Inferior Lateral Pulvinar divisions Lateral geniculate nucleus Blocks comparison Blocks comparison Trial-by-trial comparison b c Extinguished CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials ** * * Trial-by-trial comparisonBlocks comparison CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials ** CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** ** CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal *** Trial-by-trial comparison CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials * ** d a vmPFC Mediodorsal thalamus Amygdala dACC sgACC Hippocampus e Connectivity Mediodorsal thalamus dACC * .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint Mediodorsal thalamus Anterior Medial Inferior Lateral Pulvinar divisions Lateral geniculate nucleus Blocks comparison Blocks comparison b c Extinguished Blocks comparison CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** ** Trial-by-trial comparison CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials * CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials ** 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** ** * CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal *** Trial-by-trial comparison 1 2 3 4 -0.2 0.0 0.2 0.4 0.6 0.8BOLD signal Trials *** CS+ CS- 0.0 0.1 0.2 0.3 0.4BOLD signal * Trial-by-trial comparison a d Amygdala dACC vmPFC sgACC Anterior pulvinar Hippocampus e Connectivity Anterior pulvinar vmPFC ** .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint Lateral pulvinar Medial pulvinar Inferior pulvinar Anterior pulvinar Superior colliculus Hippo. Amygdala vmPFC LGN V1 TEOV2 V4 TE MD Potential neural gate for conscious and subconscious threat learning Co-activation Integrative simplified functional thalamic circuitry during the acquisition of associative threat learning Motor dACC Behavioral output CS Sensory inputs Associative threat learning Emotional output Fight Flight Freeze Fear Hypo. + Hormonal response Extinction recall - safe cues Hippocampus MD Anterior pulvinar MD Anterior pulvinar Amygdala ExcitationInhibition Threat memory expression Safe memory expression Threat renewal - threat cues Dynamic model of safe vs. threat memory retrieval using contextual cues: A thalamic-dependent “toggle switch” a b Hippocampus .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 14, 2025. ; https://doi.org/10.1101/2025.07.09.663823doi: bioRxiv preprint