Temporal structure of task engagement organizes infra-slow BOLD dynamics

Task-based fMRI has traditionally characterized cognitive engagement through spatial patterns of BOLD activation, leaving less understood whether tasks also organize brain activity along the frequency dimension. Here, we tested whether temporally structured attention reshapes infra-slow BOLD dynamics by imposing frequency-specific organization on ongoing cortical activity. Across three independent fMRI datasets, task timing systematically shaped the BOLD power spectrum. In the main visual attention dataset, periodic stimulation produced sharp stimulus-locked peaks at 0.083 Hz and 0.125 Hz, with increased power across task-relevant frequencies and reduced power in task-irrelevant frequencies. Independent auditory and visual motion fMRI datasets showed analogous task-frequency alignment at their respective temporal scales, indicating that this effect generalizes across task modalities and stimulation frequencies. Importantly, stimulus-locked spectral peaks were spatially related to GLM-derived activation but persisted after removal of GLM-modeled responses, demonstrating that frequency-specific BOLD organization is not fully reducible to conventional task-evoked activation. These spectral features were functionally informative: task-specific frequencies distinguished cognitive states, predicted individual reaction time, and preserved subject-specific signatures across repeated runs more effectively than broader spectral ranges or GLM-derived features. Finally, Jansen-Rit neural mass simulations coupled to a Balloon-Windkessel hemodynamic model reproduced the emergence of stimulus-locked BOLD peaks showing that their expression depends on balanced excitatory, inhibitory, and filter gain regimes rather than simple amplification of external input. Together, these findings suggest that task-fMRI reveals not only regional localization, but also how temporally structured task engagement organizes infra-slow brain dynamics across task-specific and task-unspecific timescales.