(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



The thalamus encodes and updates context representations during hierarchical cognitive control [1]

['Xitong Chen', 'Department Of Psychological', 'Brain Sciences', 'The University Of Iowa', 'Iowa City', 'Iowa', 'United States Of America', 'Cognitive Control Collaborative', 'Iowa Neuroscience Institute', 'Stephanie C. Leach']

Date: 2024-12

Cognitive flexibility relies on hierarchically structured task representations that organize task contexts, relevant environmental features, and subordinate decisions. Despite ongoing interest in the human thalamus, its role in cognitive control has been understudied. This study explored thalamic representation and thalamocortical interactions that contribute to hierarchical cognitive control in humans. We found that several thalamic nuclei, including the anterior, mediodorsal, ventrolateral, and pulvinar nuclei, exhibited stronger evoked responses when subjects switch between task contexts. Decoding analysis revealed that thalamic activity encodes task contexts within the hierarchical task representations. To determine how thalamocortical interactions contribute to task representations, we developed a thalamocortical functional interaction model to predict task-related cortical representation. This data-driven model outperformed comparison models, particularly in predicting activity patterns in cortical regions that encode context representations. Collectively, our findings highlight the significant contribution of thalamic activity and thalamocortical interactions for contextually guided hierarchical cognitive control.

Funding: Research reported in this publication was supported by the Iowa Neuroscience Institute ( https://medicine.uiowa.edu/iowaneuroscience ) and the National Institute of Mental Health ( https://www.nimh.nih.gov ) under Award Number R01MH122613, awarded to K.H. This work was also conducted on an MRI instrument funded by the National Institutes of Health ( https://www.nih.gov ) under grant number 1S10OD025025-01. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability: The numerical data supporting this study are available in the supplementary file S1 Data. The neuroimaging data have been deposited in NeuroVault and can be accessed at https://identifiers.org/neurovault.collection:18728 . All raw data have been made publicly accessible on OpenNeuro ( https://openneuro.org/datasets/ds005600/ ). Additionally, the analysis code used in this study has been archived on GitHub ( https://github.com/HwangLabNeuroCogDynamics/Thalamocortical_Hierarchical_Control ), with a DOI generated via Zenodo (DOI: 10.5281/zenodo.14086485 ).

The current study focused on elucidating the role of the human thalamus in hierarchical cognitive control. We designed a paradigm with different levels of hierarchical task representations. Instead of using different tasks to establish hierarchy, we equated stimuli and rule complexity across different levels of hierarchical representations within one behavioral task design [ 5 ]. We had 3 main objectives. First, we aimed to characterize the detailed functional anatomy of thalamic activity that supports hierarchical cognitive control. Second, we aimed to determine the interaction between thalamic activity and cortical control representations. To accomplish this, we utilized a novel, data-driven, thalamocortical activity flow mapping analysis to test if thalamocortical interactions can predict cortical activity patterns related to cognitive control [ 31 , 33 – 35 ]. Third, we decoded the cognitive representations encoded by thalamic activity to determine the hierarchical task representation thalamocortical interactions most strongly contribute to.

Notably, the human thalamus is a critical but understudied component of this circuitry. Structurally, the basal ganglia does not directly project to the cortex, therefore its influence on cortical representations must be mediated by the thalamus [ 16 – 18 ]. Despite this anatomical relationship, many theoretical models omit it from the model. Furthermore, most existing models are primarily built on the known projections from the pallidum to the ventrolateral thalamus for allowing motoric representations to be influenced by higher-order control processes [ 12 , 19 – 22 ]. It remains unclear whether higher order thalamic nuclei, including the anterior, mediodorsal, and pulvinar nuclei, which have dense connectivity with frontal and parietal regions [ 23 – 25 ], are involved in hierarchical cognitive control in humans. This is because most past neuroimaging studies did not provide specific details on the thalamic functional anatomy. Furthermore, recent findings from animal models demonstrated that thalamocortical interactions can signal context changes [ 26 – 28 ]. In humans, the thalamus is ideally positioned to influence cortical representations via its extensive one-to-many and many-to-one thalamocortical connectivity motifs [ 29 ]. Specifically, the anterior, medial, and posterior thalamus exhibit strong converging connectivity with multiple frontal-parietal systems that have been implicated in control-related functions [ 30 ]. The behavioral significance of this thalamocortical connectivity profile is further affirmed by recent functional neuroimaging studies [ 31 , 32 ] and lesion evidence [ 32 , 33 ]. However, it remains unclear how thalamocortical connectivity supports hierarchical cognitive control.

To support flexibility, task representations are updated as context changes. One computation model suggests that the cortico-striatal-thalamic circuit is involved in updating cortical task representations [ 3 , 11 , 12 ]. According to this model, the basal ganglia receives control signals from the prefrontal cortex, and then uses these signals to disinhibit the thalamus to influence cortical representations via thalamocortical inputs [ 3 , 11 ]. Inspired by this model, several neuroimaging studies focused on characterizing the interaction between prefrontal cortex and the striatum [ 13 – 15 ].

One influential framework posits that cognitive control is supported by hierarchically organized task representations that are encoded by the functional organization of the prefrontal cortex [ 1 – 3 ]. In this framework, task representations are structured to encode contexts, goals, task-relevant features, and potential actions [ 4 ]. Furthermore, representations can be organized in a hierarchical manner, where more abstract contextual representations exert stronger influence over action representations to support cognitive flexibility [ 5 , 6 ]. Anatomically, it was proposed that this abstraction gradient is supported by a rostro to caudal organization in the lateral prefrontal cortex [ 2 , 7 – 10 ].

Cognitive control enables humans to adjust their behaviors flexibly in response to changing circumstances. For example, the sound of a ringing phone can elicit different reactions depending on the context. At home, one might answer, while driving might cause one to ignore it for safety. Everyday life is filled with scenarios that require us to modify our actions based on different contexts. How does our control system facilitate such contextually guided control?

Results

Hierarchical control task and behavioral results We designed a behavioral task to investigate hierarchical cognitive control (Fig 1A). In each trial, a task cue was presented for 0.5 second, followed by a picture probe for 2.5 seconds. Subjects had to respond to the probe by applying one of 2 specific response rules: the “scene rule” (determining whether the presented picture was a scene) or the “face rule” (determining whether the presented picture was a human face). The appropriate rule selection was based on different features of the cue, organized hierarchically. Specifically, the texture of the cue (solid versus hollow) referred to as the “context,” representing the highest level of attributes. This context information determined which mid-level “feature” subjects should select to determine the appropriate “response rule.” Each feature from the mid-level (shape or color) was then associated with either the face or scene rule. The face and scene rule then determines the visual stimuli category subjects should decide to attend to generate the correct motor response. This organization introduced three hierarchical levels of task switching: extra-dimension switch (EDS), inter-dimension switch (IDS), and Stay (Fig 1B). In EDS trials, subjects updated the highest-level task context. In IDS trials, subjects switched between task-relevant features within the same task context. In Stay trials, subjects performed the same task for 2 trials in a row. We hypothesized that as the hierarchical task switching demand increases from Stay to IDS to EDS, response time (RT) would also increase. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 1. Hierarchical control task. (A) The hierarchical control task, in which we introduced different levels of hierarchical task switching between trials. (B) The hierarchy of task representations. (C) Reaction time observed across the 3 hierarchical levels of task-switching. (D) Accuracy across the 3 hierarchical levels of task-switching. ** p < 0.01. n.s., nonsignificant. The error bar represents the 95% confidential interval. The black dot indicates the mean value, while the colored dot represents data from individual subjects. Lines connect the data points for each subject across different conditions. Data used for (C) and (D) can be found in S1 Data, specifically in the sheet labeled “Fig 1C and 1D”. https://doi.org/10.1371/journal.pbio.3002937.g001 Results from the one-way (Conditions: EDS, IDS, and Stay) repeated measures ANOVA (rmANOVA) showed that the hypothesized hierarchical structure of task representations significantly modulated subject’s RT (F(2,116) = 156.21, p < 0.0001, η2 = 0.11, Fig 1C). Subjects responded slowest for the EDS condition (mean ± SD: 1.16 s ± 0.22; EDS versus IDS: t(58) = 8.36, p < 0.001, Cohen’s d = 0.44; EDS versus Stay: t(58) = 16.70, p < 0.001, Cohen’s d = 0.83), followed by the IDS (mean ± SD: 1.07 s ± 0.20; IDS versus Stay: t(58) = 10.60, p < 0.001, Cohen’s d = 0.41) and Stay (mean ± SD: 0.99 s ± 0.19) conditions. Similar effects were observed for accuracy (F(2,116) = 13.70, p < 0.0001, η2 = 0.05; Fig 1D). Accuracy for the EDS condition (mean ± SD: 0.94 ± 0.05) was significantly lower than the Stay condition (mean ± SD: 0.97 ± 0.03; t(58) = −6.10, p < 0.001, Cohen’s d = −0.51), but not significantly different to the IDS condition (mean ± SD: 0.95 ± 0.04; t(58) = −1.88, p = 0.19, Cohen’s d = −0.20). The accuracy for the IDS condition was significantly lower than Stay (t(58) = −3.12, p = 0.008; Cohen’s d = −0.36). We next determined whether these effects were affected by repeating responses (choose “yes” or “no”) in the EDS and IDS conditions. Significant differences were found between EDS and IDS for both response switch (EDS versus IDS: t(58) = 6.61, p < 0.0001, Cohen’s d = 0.43; EDS versus Stay: t(58) = 15.01, p < 0.0001, Cohen’s d = 0.95; IDS versus Stay: t(58) = 11.93, p < 0.0001, Cohen’s d = 0.56) and response repeat (EDS versus IDS: t(58) = 8.14, p < 0.0001, Cohen’s d = 0.44; EDS versus Stay: t(58) = 13.38, p < 0.0001, Cohen’s d = 0.70; IDS versus Stay: t(58) = 6.08, p < 0.0001, Cohen’s d = 0.25; S1 Fig). We then examined whether task-switching effects were influenced by repeating cues in the Stay condition. Significant task-switching effects were observed when comparing both cue repeat (CR) and cue switch (CS) Stay trials for RT and accuracy. For RT, participants were significantly slower in the EDS condition compared to both Stay_CS (t(58) = 12.99, p < 0.0001, Cohen’s d = 0.62) and Stay_CR (t(58) = 16.80, p < 0.0001, Cohen’s d = 0.97; S1 Fig) trials. Similarly, participants were slower in the IDS condition compared to Stay_CS (t(58) = 4.31, p = 0.001, Cohen’s d = 0.19) and Stay_CR (t(58) = 12.32, p < 0.0001, Cohen’s d = 0.55). Additionally, participants responded more slowly in Stay_CS trials compared to Stay_CR trials (t(58) = 7.42, p < 0.0001, Cohen’s d = 0.36; S1 Fig). For accuracy, performance in the EDS condition was significantly lower than in Stay_CS (t(58) = −3.97, p = 0.002, Cohen’s d = 0.30) and Stay_CR (t(58) = −8.01, p < 0.0001, Cohen’s d = 0.70; S1 Fig) trials. In the IDS condition, accuracy was not significantly different from Stay_CS (t(58) = −0.125, p = 1.00, Cohen’s d = 0.01) but was significantly lower than in Stay_CR (t(58) = −4.94, p < 0.0001, Cohen’s d = 0.45). Furthermore, accuracy in Stay_CS trials was significantly lower than in Stay_CR trials (t(58) = −5.30, p < 0.0001, Cohen’s d = 0.46; S1 Fig). This observed pattern of reaction time and accuracy results suggest a hierarchical organization of task representations, consistent with findings from our previous published study that employed the same task with EEG [5]. Furthermore, these behavioral results pattern remained consistent when separated trials where the response and cue in the current trial was repeated from the previous trial.

Cortical evoked responses We found that the behavioral performance was influenced by the demand of hierarchical task switching. Subsequently, we explored how hierarchical task switching modulated cortical and subcortical activity. We first investigated cortical evoked responses elicited by different hierarchical task switching conditions. This analysis identified a set of cortical regions including the middle frontal gyrus, the inferior frontal sulcus, the precentral sulcus, the intraparietal sulcus, the postcentral sulcus, the insula, the medial frontal gyrus, the posterior cingulate gyrus, the lateral occipital gyrus, the fusiform gyrus, the posterior cingulate cortex, and the inferior temporal gyrus (EDS, IDS, and Stay conditions; Fig 2A; center of mass (CM) coordinates available in Table A, B, and C in S1 Text). We then contrasted the magnitudes of evoked responses between different task switching conditions (e.g., EDS-Stay, EDS-IDS, and IDS-Stay; Fig 2B). Specifically, we examined EDS versus Stay and EDS versus IDS to determine cortical regions involved in switching of task context, and IDS versus Stay for switching of task feature. We found stronger evoked responses in the frontoparietal and the temporal regions in response to changing task contexts (EDS-Stay and EDS-IDS, Fig 2B; CM coordinates available in Table D, E and F in S1 Text), included areas such as the rostral middle frontal gyrus, the inferior frontal sulcus, the insula, the precentral sulcus, the postcentral sulcus, the intraparietal sulcus, the medial frontal gyrus, the posterior cingulate gyrus, the precuneus, the cuneus, and the middle temporal lobe. We did not observe significant effects in these regions during switches in feature level task information (IDS-Stay). Instead, when contrasting task switching that involved feature level task information (IDS-Stay), we found significant clusters primarily distributed in the premotor cortex and the intraparietal sulcus. These cortical regions also showed significant modulations in response to updating context information (EDS-IDS). This pattern revealed an asymmetrical hierarchy consistent with previous studies [2], in which several regions in the caudal frontal cortex are more involved in contextual update, whereas premotor regions are responsive to both contextual and feature level updates of task representations. The overall cortical response pattern remained consistent when considering trials where the current trial’s response (choose “yes” or “no”) and cue were repeated from the previous trial (see S2 Fig). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 2. Cortical evoked response to hierarchical task-switching. (A) Cortical evoked responses for the three task-switching conditions. (B) Contrast between conditions. The results were first thresholded at a voxel-level threshold of p < 0.05, followed by cluster correction procedure with a cluster level threshold of p < 0.05, only showing clusters with a minimum cluster size (k) of 58 voxels. Clusters were defined as groups of voxels that are connected by sharing a face with their neighboring voxels. The group statistical maps presented in Fig 2 can be accessed at https://identifiers.org/neurovault.collection:18728. https://doi.org/10.1371/journal.pbio.3002937.g002

Thalamic activity profile responding to hierarchical task-switching We then investigated the detailed functional anatomy of thalamic activity in response to different levels of hierarchical task switching (EDS, IDS, and Stay conditions; unthresholded activity maps: Fig 3A, thresholded activity maps at cluster corrected p < 0.05: Fig 3B). Significant evoked responses were observed in the anterior, the ventral, the medial, and the posterior regions of the thalamus for all task switching conditions (Fig 3A and 3B). The negative response patterns observed in the medial and the posterior region of the thalamus may be attributed to a significant undershoot in the hemodynamic response functions (HRFs) of these nuclei (see S3 Fig). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 3. Thalamic voxel-wise evoked responses in response to hierarchical task-switching. (A) Unthresholded thalamic evoked response patterns (t-value) for the 3 hierarchical task-switching conditions. (B) Thresholded thalamic evoked response patterns. (C) Contrast between conditions. The results were first thresholded at a voxel-level threshold of p < 0.05, followed by cluster correction procedure with a cluster level threshold of p < 0.05, only showing clusters with a minimum cluster size (k) of 58 voxels. Clusters were defined as groups of voxels that are connected by sharing a face with their neighboring voxels. The color of contour in (B) and (C) refers to the nuclei described in (D). (D) Morel atlas of thalamic nuclei. Abbreviations for thalamic nuclei are as follows: AN, anterior nucleus; VM, ventral medial nucleus; VL, ventral lateral nucleus; MGN, medial geniculate nucleus; MD, mediodorsal nucleus; PuA, anterior pulvinar nucleus; LP, lateral posterior nucleus; IL, intralaminar nucleus; VA, ventral anterior nucleus; Po, posterior nucleus; LGN, lateral geniculate nucleus; PuM, medial pulvinar nucleus; PuL, lateral pulvinar nucleus; VP, ventral posterior nucleus. Similar results were also observed using an 8 mm smoothing kernel, voxel-level threshold of p < 0.001, and cluster threshold of p <0.05 (cluster size of 42 voxels), see S5 Fig. The group statistical maps presented in Fig 3 can be accessed at https://identifiers.org/neurovault.collection:18728. https://doi.org/10.1371/journal.pbio.3002937.g003 Our results revealed activity patterns within the thalamus for updating different hierarchical task representations. Thalamic voxels in the anterior, the ventroanterior, the ventromedial, the ventrolateral, the mediodorsal, and the intralaminar nuclei showed stronger evoked responses more selective for updating context representations (EDS-IDS; Fig 3C). In addition, we observed that voxels in the anterior, the ventroanterior, the ventrolateral, the mediodorsal, the intralaminar, the medial pulvinar, the lateral pulvinar, and the ventral posterior nuclei of the thalamus showed stronger evoked responses for the EDS-Stay contrast (Fig 3C). We did not observe any significant clusters for the IDS-Stay contrast, but only for the EDS-IDS and EDS-Stay contrasts. This suggests that the thalamus may be more selectively involved in hierarchical cognitive control that involves updating context representations. This overall thalamic response pattern remained consistent when considering trials where the current trial’s response and cue were repeated from the previous trial (see S4 Fig). In summary, the hierarchical cognitive control modulates the evoked response of thalamus. Additionally, thalamic nuclei in anterior, ventral, and medial regions displayed a preference of updating context-level task representation in hierarchical cognitive control task.

Thalamic voxels selectively encode contextual information Our results showed that thalamic activity is most strongly associated with updating contextual information during hierarchical cognitive control. We next seek to determine which level of hierarchical task representation is encoded in the thalamus. Utilizing multivoxel pattern analysis, we investigated whether we could decode context, color, shape, or response rule from multivoxel thalamic activity patterns. We observed above chance decoding of contextual information throughout the thalamus after applying a stringent cluster correction threshold of p < 0.001 (solid versus hollow; Fig 4). Highest decoding accuracy was found in the ventroanterior (0.52 ± 0.004), the ventrolateral (0.52 ± 0.004), the mediodorsal (0.52 ± 0.004), the intralaminar (0.52 ± 0.004), the medial pulvinar (0.52 ± 0.004), and the lateral posterior nuclei (0.52 ± 0.003). These thalamic nuclei also found to show strongest evoked response during the contextual update condition (EDS). Decoding performance for other information, color (blue versus red), shape (square versus circle), and task (face versus place tasks), was significantly weaker, with few significant voxels in anterior and medial regions (Fig 4). But these voxels did not survive after cluster-correction. These results suggest that thalamic activity is involved in encoding contextual information during hierarchical cognitive control. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 4. Thalamic activity encodes task contexts. Cluster-corrected decoding performance on Context at p < 0.001 (minimum cluster size = 343 voxels). Color, shape, and task were uncorrected. The group statistical maps presented in Fig 4 can be accessed at https://identifiers.org/neurovault.collection:18728. https://doi.org/10.1371/journal.pbio.3002937.g004

[END]
---
[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3002937

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.