When the brain processes a stimulus or plans an action, neural activity does not simply switch on within a static, localized network. Instead, information moves through the brain like a physical wave, traveling across specialized cortical regions in a highly structured spatiotemporal sequence. This directional flow acts as a biological pipeline, transforming raw sensory inputs into high-level cognitive representations, and ultimately into motor execution.
To understand how the brain coordinates complex behavior, we must look past where things happen in the brain, and focus instead on when and in what order information shifts across the cortical landscape.
The Cortical Conveyor Belt: From Perception to Action
To coordinate complex behaviors, the brain routes information through a sequence of discrete functional states. For instance, during tasks like reading or picture naming, the brain must orchestrate an orderly transition from visual decoding to semantic retrieval, and finally to motor articulation.
By applying switching linear dynamical models to human intracranial recordings, Aram Giahi Saravani et al. demonstrated in Dynamic Brain Interactions during Picture Naming that speech production relies on a reliable sequence of non-overlapping "visual," "language," and "articulation" network states. Crucially, the variable duration of the language processing phase directly predicts the exact moment a person will start speaking on a trial-by-trial basis. This suggests that complex cognitive tasks are organized into strict, sequential temporal windows of communication.
This structured routing is also evident in reading and language comprehension. Leila Wehbe mapped the millisecond-by-millisecond pipeline of language processing in The Spatio-Temporal Representation of Natural Reading, showing a clear spatio-temporal sequence that progresses from raw visual recognition to semantic processing and high-level narrative integration.
When oral reading is required, Jinfeng Huang et al. showed in Revealing Spatiotemporal Brain Dynamics of Speech Production Based on EEG and Eye Movement that specific neural hubs facilitate this transition by pipelining information from the visual cortex to frontal motor regions over several hundred milliseconds. Similarly, generative modeling of neural dynamics by Yule Wang et al. in Exploring Behavior-Relevant and Disentangled Neural Dynamics with Generative Diffusion Models confirms a consistent, directional flow of information that starts in visual regions, passes through somatosensory feedback circuits, and terminates in motor execution areas.
Directional Trajectories Across Cortical Geography
This temporal sequencing manifests as physical, directional waves across the brain's anatomy. Two prominent trajectories dominate these dynamics: a posterior-to-anterior wave for sensory and cognitive processing, and a rostral-to-caudal shift for motor planning and execution.
In the temporal cortex, visual processing behaves like a physical conveyor belt. Using high-resolution intracranial recordings, Jessica Schrouff et al. revealed in Fast temporal dynamics and causal relevance of face processing in the human temporal cortex a fast, posterior-to-anterior temporal wave during face processing, characterized by measurable delays between successive anatomical stages.
A similar posterior-to-anterior spatial flow was observed by Luca Ambrogioni et al. in Dynamic decomposition of spatiotemporal neural signals during attentional preparation in somatosensory tasks, where frequency-specific power suppression sequentially sweeps across cortical regions. This directional wave is also mirrored in higher-order cognitive tasks; for example, Marcus R. Streit et al. tracked facial emotion recognition in Neurophysiological correlates of the recognition of facial expressions of emotion as revealed by magnetoencephalography, showing that it initiates in posterior temporal regions before cascading into the middle temporal cortex, amygdala, and prefrontal areas.
Over longer timescales, Rieko Osu et al. in Decoding emotion from EEG using machine learning and Zhongyang He et al. in Processing of Affective Pictures: A Study Based on Functional Connectivity Network in the Cerebral Cortex noted that emotional appraisal shifts dynamically between frontal and posterior/temporal networks as time progresses from initial stimulus exposure to sustained attention.
Once information reaches the frontal networks for action planning, the directional flow shifts along a rostral-to-caudal axis. Using high-density micro-electrocorticography ($\mu$ECoG), Leo Scholl et al. discovered in Mapping Eye, Arm, and Reward Information in Frontal Motor Cortices Using Electrocorticography in Nonhuman Primates that directional intent emerges first in rostral prefrontal areas during planning before physically migrating to caudal primary motor regions for execution.
This spatial progression is a hallmark of motor preparation: Baoguo Xu et al. found in Electroencephalogram Source Imaging and Brain Network Based Natural Grasps Decoding that the brain shifts from a highly communicative frontal-lobe network during the intention phase to a tightly localized parietal-lobe network during physical grasp execution.
[Sensory Input] ──> Posterior Temporal ──> Rostral Prefrontal ──> Caudal Motor ──> [Action]
(Perception) (Planning) (Execution)
Stage-Specific Regional Necessity
Why does the brain use these moving waves instead of keeping networks active continuously? The answer lies in efficiency and functional specialization: different regions are causally necessary only during precise, transient windows of a task.
Peter Zatka-Haas demonstrated this causal shifting in visual decision-making in Sensory coding and the causal impact of mouse cortex in a visual decision. While the visual cortex is strictly necessary early on for processing sensory evidence, its causal relevance fades as the frontal cortex takes over to compute the final decision.
In more complex behaviors requiring a delayed response, Vahid Esmaeili et al. showed in Rapid suppression and sustained activation of distinct cortical regions for a delayed sensory-triggered motor response that sensory processing areas are rapidly suppressed after the initial input phase, while frontal motor hubs are actively recruited to sustain the motor plan during the delay.
This time-varying requirement extends to subcortical and cerebellar structures. In Cerebellar disruption impairs working memory during evidence accumulation, Ben Deverett et al. showed that the cerebellum's functional contribution shifts dynamically across a task pipeline: it actively stabilizes working memory traces during evidence accumulation, then switches to influencing action preparation during the subsequent delay period.
Hierarchical Broadcast and Execution Cascades
When a plan transitions into an action, the spatiotemporal wave expands from a localized cortical signal into a brain-wide cascade. Susu Chen et al. tracked this phenomenon in Brain-wide neural activity underlying memory-guided movement by recording tens of thousands of neurons simultaneously, revealing that movement planning signals propagate through a structured broadcast mechanism. The signal begins in the motor cortex and midbrain, then sequentially reaches the thalamus, striatum, and medulla, vanishing the instant the action begins.
When this hierarchical flow breaks down, neurological symptoms can emerge. For example, Irene Neuner et al. demonstrated in Imaging the where and when of tic generation and resting state networks in adult Tourette patients that involuntary motor events, such as tics in Tourette Syndrome, follow an unconstrained cascade. Activity spreads prematurely from high-level cortical premotor structures to interoceptive regions, and finally down to subcortical circuits and the primary motor cortex.
By mapping these directional trajectories, modern neuroscience moves away from a static view of brain localization. Instead, it reveals a dynamic system where information is continuously handed off across space and time to turn perception into purposeful behavior.
Key Foundational Work
Decoding emotion from EEG using machine learning (2014) — contributed 1 concept elements to this direction.
Neurophysiological correlates of the recognition of facial expressions of emotion as revealed by magnetoencephalography (1999) — contributed 1 concept elements to this direction.
Imaging the where and when of tic generation and resting state networks in adult Tourette patients (2014) — contributed 1 concept elements to this direction.