Intrinsic dynamics shape responses to external stimulation in the human brain

When ongoing sensory stimulation reaches the brain, the resulting neural activity reverberates in its recurrent neural networks. How this intrinsic dynamic affects sensory responses is not well understood. To separate the immediate effect of the stimulus from the recurrent brain dynamic we used a new vector-autoregressive model with external input (VARX). Applying this analysis to intracranial recordings in humans, we find that the recurrent connectivity observed during rest is unaltered when humans are watching movies. The recurrent dynamic enhances and prolongs the responses of the brain to scene cuts, fixation onsets, and sound envelope. If one fails to account for these external inputs, then spurious connections appear in the “functional connectivity”. The model reproduces the prominent observation that an external stimulus can reduce intrinsic noise. The model also reveals that sensory areas have mostly outward connections, whereas higher-order brain areas have mostly incoming connections. By combining the concepts of “functional connectivity” and “encoding models” we introduce an analytical approach capable of revealing interactions between external stimulation and internal dynamics that are not apparent when analyzing these concepts in isolation.