Thalamocortical excitability modulation guides human perception under uncertainty

The study asks how the brain adapts sensory processing when the relevance of specific stimulus features is uncertain. The authors propose that uncertainty triggers a shift from phasic, alpha-rhythm-based selective gain control (favoring selective enhancement and distractor suppression) toward elevated tonic excitability that increases sensitivity to multiple features simultaneously. They further hypothesize that the thalamus acts as a subcortical hub coordinating these cortical state switches, potentially via neuromodulatory influences (e.g., noradrenergic arousal). The work aims to identify how uncertainty during encoding impacts subsequent evidence accumulation, cortical excitability dynamics (rhythmic and aperiodic), arousal, and thalamic activity in humans.

Prior work links selective attention to gain modulation in sensory cortex and phasic inhibitory control via alpha rhythms, which can gate high-frequency (gamma) activity and sensory gain. Elevated sensitivity to high-dimensional inputs may alternatively be supported by transient increases in the cortical excitation/inhibition (E/I) ratio, reflected in flatter 1/f spectral slopes and increased signal irregularity (sample entropy). Neuromodulatory systems, particularly noradrenergic arousal, are known to alter cortical states and sensitivity to stimuli. The thalamus has long been implicated in attentional control and neuromodulatory integration, with known roles in modulating system excitability via rhythmic and aperiodic dynamics and in supporting cognitive flexibility. However, non-invasive human evidence for thalamic contributions to rapid cortical state adjustments under uncertainty has been lacking. The study builds on this literature to test whether uncertainty shifts cortical dynamics from alpha-rhythmic to aperiodic excitability increases, and whether thalamic activity indexes and potentially coordinates these changes.

Design: A multi-modal within-subject EEG and fMRI experiment in healthy young adults performing the same task in two sessions (EEG n=47; fMRI n=42). Pupil diameter was recorded during EEG as a proxy for arousal. Task (MAAT): Dynamic visual display of small moving squares with four attributes (color red/green, motion direction left/right, size large/small, saturation high/low). On each trial, valid pre-stimulus cues (1–4 features) indicated the set of potentially probed features, manipulating probe uncertainty while keeping visual input constant. Stimulus duration 3 s; subsequent 2-AFC probe on a single cued feature; block structure (8 trial blocks) with fixed cue set per block. Behavior/Modeling: Hierarchical drift-diffusion modeling (HDDM) estimated drift rate, non-decision time (NDT), and boundary separation. The final EEG-informed model allowed load-related changes in drift and NDT with fixed threshold across conditions. Model selection used DIC; reliability examined across sessions. EEG: 64-channel active scalp EEG (1 kHz acquisition; preprocessing with filtering, ICA artifact removal, channel/epoch cleaning, re-referencing). Decision-related signatures: centro-parietal positivity (CPP) and contralateral mu-beta suppression. Spectral analyses: wavelet-based low-frequency (2–15 Hz) and multitaper high-frequency (45–190 Hz) power; multivariate partial least squares (PLS) relating time–frequency–space matrices to task conditions and behavior, yielding a spectral power modulation component (SPMC). Rhythm-specific metrics via eBOSC to separate narrowband rhythmic events from the aperiodic background. Alpha–gamma phase-amplitude coupling (PAC) estimated during detected alpha episodes. Aperiodic dynamics: time-resolved sample entropy (alpha-notched) and 1/f slope (2–64 Hz; excluding 7–13 Hz and SSVEP range). SSVEP at 30 Hz assessed as control. Cluster-based permutation tests used for spatiotemporal inferences. Pupillometry: EyeLink 1000 (1 kHz); phasic arousal quantified via first temporal derivative of pupil diameter; cluster-based permutation tests for load effects; median derivative 0–1.5 s post-stimulus extracted for post hoc analyses. fMRI: 3T Siemens TrioTim, multiband EPI (TR=645 ms; voxel size 3 mm isotropic); extensive preprocessing (motion correction, smoothing, high-pass filter, ICA-based denoising, regression of motion/WM/CSF, DVARS censoring). First-level GLM with separate regressors for load conditions; second-level multivariate PLS analyses: task PLS for load-related BOLD modulation; behavioral PLS relating voxel-wise linear load slopes to individual EEG/behavioral/pupil indices. Thalamic time courses visualized within Morel atlas mask; nuclei-level mapping via Morel parcellation and projection zones. Statistics: Within-subject centering for visualization; paired t-tests with Benjamini–Hochberg correction; repeated-measures partial correlations controlling for main effect of load; permutation testing and bootstrap ratios for PLS solutions.

• Behavior: Increasing number of relevant targets slowed RTs and reduced accuracy. Drift-diffusion modeling showed robust linear decreases in drift rate with probe uncertainty (EEG session: linear b = -0.57, 95% CI [-0.65, -0.49], t(46) = -13.94, p ≈ 2.9e-16). NDT increased with load (b = 0.07, 95% CI [0.06, 0.08], t(46) = 23.27). CPP slopes decreased with uncertainty (linear b ≈ -2e-6, 95% CI [-2.8e-6, -1.5e-6], t(46) = -7.05, p = 7.2e-8), mirroring drift decreases.
• Individual differences: Single-target drift correlated with CPP slope (r = 0.52, p = 3.59e-4). Drift decreases covaried with shallowing CPP slopes (r = 0.34, p = 4.87e-5). Greater drift decreases were observed in those with higher single-target evidence (EEG r = -0.93; MRI r = -0.88; both p < 1e-13). Higher absolute drift across loads associated with faster RTs and higher accuracy.
• Cortical excitability (EEG): Multivariate PLS revealed stimulus-evoked increases in theta and gamma power and alpha desynchronization; the same pattern scaled up with probe uncertainty (SPMC; permuted p < 0.001). Stronger SPMC expression related to higher single-target drift (r ≈ 0.53) and larger drift decreases with load (partial r = -0.40, p = 1.12e-6). Alpha–gamma PAC was significant during alpha episodes; alpha episode duration decreased with uncertainty (linear b = -1.17, 95% CI [-1.81, -0.53], t(46) = -3.67, p = 3.6e-5), indicating reduced phasic alpha engagement as more features became relevant.
• Aperiodic dynamics: Sample entropy increased with uncertainty over posterior–occipital sites (linear b = 0.002, 95% CI [0.001, 0.003], t(44) = 4.22). 1/f spectral slopes shallowed (occipital cluster; linear b = 0.008, 95% CI [0.003, 0.012], t(43) = 3.06). SampEn correlated strongly with PSD slope across conditions (r = 0.78, p = 7e-11) and with load-related slope changes (r = 0.44, p ≈ 4.9e-8).
• Arousal (pupil): Phasic pupil derivative increased with uncertainty (linear b = 0.14, 95% CI [0.09, 0.19], t(46) = 5.58). Pupil increases tracked SPMC increases (partial r = 0.22, p = 0.01) and related to higher baseline drift (r = 0.31), larger drift decreases (partial r = -0.17, p = 0.05), and smaller NDT increases (partial r = -0.21, p = 0.01).
• Thalamic BOLD: Task PLS (LV1) showed load-related BOLD increases in fronto-parietal and salience networks and thalamus; individual brain scores correlated with larger drift decreases (r = -0.36, p = 5.11e-5). Behavioral PLS identified a single LV predominantly loading on anterior/midline thalamic nuclei with fronto-parietal projections; greater thalamic BOLD modulation associated with higher baseline drift (r = 0.75, bsCI [0.72, 0.86]), larger drift reductions (r = -0.60, bsCI [-0.78, -0.54]), lower baseline NDT (r = -0.37), larger SPMC (r = 0.31), greater pupil increases (r = 0.67), and higher entropy modulation (r = 0.22). Relations with 1/f slope were not stable at the individual level.

Findings show that contextual uncertainty during encoding reduces the precision of information available for subsequent single-feature decisions (lower drift), while concurrently shifting cortical dynamics from selective, phasic alpha-rhythmic control toward heightened tonic excitability (alpha desynchronization, increased gamma, higher entropy, shallower 1/f slopes). These cortical state changes co-occur with increased phasic arousal and are strongly indexed by thalamic BOLD upregulation, especially in anterior/midline nuclei with fronto-parietal projections (e.g., mediodorsal thalamus). The results support a model in which the thalamus integrates neuromodulatory signals to regulate cortical excitability states, enabling flexible multi-feature sampling under uncertainty at the expense of per-feature evidence strength. The dorsal fronto-parietal and midcingulo-insular networks increased activity with load, consistent with salience and cognitive control engagement. Alpha–gamma coupling indicates preserved phasic control when possible, but reduced alpha engagement as more features become relevant. Overall, thalamocortical interactions appear central to rapid regime switches between rhythmic selective gating and aperiodic excitability that optimize processing for current contextual demands.

This study provides convergent behavioral, EEG, pupillometric, and fMRI evidence that thalamocortical neuromodulatory processes guide adaptive shifts in cortical excitability under uncertainty. As probe uncertainty increases, cortex moves from alpha-rhythmic selective gain control toward elevated tonic excitability, facilitating concurrent sampling of multiple features but reducing per-feature evidence for later decisions. Thalamic activity, particularly in anterior/midline nuclei with fronto-parietal projections, robustly tracks individual differences in these excitability shifts, arousal, and decision parameters. Future work should dissociate roles of specific thalamic nuclei (e.g., mediodorsal vs pulvinar) under manipulations of sensory evidence, directly assay neuromodulatory contributions (noradrenergic vs cholinergic), and examine clinical relevance of excitability modulation and E/I control in disorders characterized by uncertainty processing deficits.

• The MAAT manipulation targeted probe uncertainty without directly manipulating sensory evidence (e.g., motion coherence), limiting dissociation of thalamic subnuclei roles (e.g., mediodorsal vs pulvinar) in sensory vs contextual uncertainty.
• Spatial localization within thalamic nuclei is challenging with BOLD fMRI; precise attribution to specific nuclei remains tentative.
• Individual-level relations between 1/f slope changes and BOLD modulation were unstable, possibly due to noisier estimates of aperiodic slopes.
• Neuromodulation was inferred indirectly via pupil diameter; no direct pharmacological or receptor-specific measures were obtained to separate noradrenergic and cholinergic influences.
• Sample comprised healthy young adults; generalizability to other age groups or clinical populations is not established.