Luhe Li¹ (), Ilona M. Bloem^1,3, Stephanie Badde⁴, Wouter Schellekens⁵, Natalia Petridou⁵, Michael S. Landy^1,2, Jonathan Winawer^1,2; ¹Department of Psychology, New York University, ²Center for Neural Science, New York University, ³Netherlands Institute for Neuroscience, Amsterdam, the Netherlands, ⁴Department of Psychology, Tufts University, ⁵Radiology Department, Center for Image Sciences, UMC Utrecht, the Netherlands

Neural responses to sensory stimuli exhibit complex temporal dynamics. Recent studies demonstrated that temporal dynamics of visual neural responses, including sub-additive temporal summation and response reduction with repeated or sustained stimuli (adaptation), are well characterized by a delayed-normalization model. Do similar principles of temporal dynamics apply more generally to sensory coding? Here, we apply similar methods and modeling to the tactile domain. We used fMRI to measure responses to tactile stimuli within somatosensory cortex. Participants were presented with vibrotactile stimuli (110 Hz) simultaneously to all five finger pads of the non-dominant hand while they visually fixated a dot on the screen. We used an event-related design, in which on a single trial the vibrotactile stimuli either varied in duration (for a single stimulus) or in inter-stimulus interval (for pair stimuli), comparable to the visual studies. The single pulse durations and paired pulse intervals were 50, 100, 200, 400, 800, or 1200 ms. We estimated the underlying population neural response time courses from the fMRI BOLD response using deconvolution and computed the area under the response curve to estimate the total response. The results showed clear sub-additive temporal summation, comparable to responses in visual cortex. We modeled the neural time courses for all single pulse durations and paired pulse intervals with the delayed-normalization model and found that this model outperformed a linear prediction. Importantly, our results reveal similar temporal dynamics for visual and tactile neural responses; both are best explained by the delayed-normalization model. These findings suggest that delayed normalization constitutes a canonical neural computation across modalities.

Acknowledgements: EY08266 (MSL), R01MH111417(NP & JW)