Neural Correlates of Top-down and Bottom-up Communication in Sensory Networks

Abstract

What we perceive is not an exact reflection of our surroundings, but rather an interpretation of the noisy and ambiguous information provided by the senses. In the visual system, bottom-up communication occurs in a hierarchical manner, with early areas processing basic features like form and color, and higher areas processing more complex features. On the other hand, top-down communication travels upstream from higher to lower areas, transmitting prior knowledge and influencing our perception of the world. The impact of top-down prior knowledge on perception can be categorized in at least two distinct processes: prediction and attention. Prediction can be understood as prior knowledge about the likelihood of certain stimuli occurring in a particular context or location. On the other hand, attention can be understood as prior knowledge about the relevance of specific stimuli or locations to current behavioral goals. In the present work, we developed two new EEG experiments that investigate top-down and bottom-up communication in sensory networks, controlling either sensory predictions or spatial attention, respectively. Neuroimaging studies have shown that visual predictions can modulate early visual cortex in a retinotopic specific way. However, the time course of prediction related visual cortex activation is not known yet. In the first study (Chapter II), we implemented a novel event-related potential (ERP) paradigm. A trial started with one of two sounds, which each was associated with one visual stimulus location: either the top left visual field, or the bottom right visual field (‘Standards’). Visual stimuli comprised of gratings, which were presented 750 ms after sound onset. In a small number of trials, the visual stimulus occurred at the unexpected location (‘Deviants’), or was omitted (‘Omissions’). Standards and Deviants elicited a C1 effect, that is, a polarity reversal for lower vs. upper visual field presentation in the latency range between 50-100 ms post stimulus onset. Spatially specific Deviant and Omission effects started with a latency of 150 ms and 230 ms, respectively. The first spatially selective modulation of visual processing was observed as early as 70 ms, as reflected by the Visual predictive signal. Spatially specific Negative and Positive error signals emerged with a latency of 150 and 320 ms, respectively. These results suggest that visual predictions control visual cortex activity in a spatially specific manner, but only after the first sweep of visual processing. However, visual predictions do not elicit neural responses that mimic stimulus-driven activity, but rather, seem to affect early visual cortex via distinct neural mechanisms. Past research on the relationship between oscillatory activity and attention have consistently found that alpha-power decreases on the contralateral side to the attended stimulus, while showing increases on the ipsilateral side. In contrast to the direction of alpha-lateralization, gamma power shows the complementary pattern. Interestingly, there appears to be a close relationship between alpha and gamma modulation, with alpha power controlling gamma power. In the second study (Chapter III), we used a spatial attention paradigm to investigate whether alpha and gamma lateralization might reflect top-down and bottom-up communication, respectively, as mapped onto endogenous and exogenous spatial attention. We presented letter cues that served as directional (L/R) endogenous cues, which predicted the correct side of the next upcoming target with an 80% probability. Before target presentation, we flashed bright frames around bilateral presented dynamic grating stimuli. These frames served as exogenous cues, which had no predictive value. The presentation of bilateral dynamic grating stimuli resulted in a strong, sustained, band-limited response in the gamma range. During the anticipation of visual stimulation, endogenous spatial attention was found to result in lateralized alpha- and gamma band responses. However, these effects were not observed after target presentation, which may seem surprising considering the consistent behavioral effects seen with valid and invalid cues. We speculate that the attentional effects were narrow, limited to the task-relevant spatial location, that is the gray circle that was subtended by the dynamic grating stimuli. Therefore, the inhibitory effects of alpha desynchronization observed during the visual stimulation might have been local and did not spread to other regions. Furthermore, since the spatial location of the dynamic grating stimuli were not task relevant the visually induced gamma response was potentially not affected by attention. Overall, both the behavioral and electrophysiological results suggest that endogenous attention (reflecting top-down communication) and exogenous attention (reflecting bottom-up communication) may operate independently from each other. In conclusion, the findings of this dissertation provide evidence that the initial cortical response in visual processing is independent of top-down control and bottom-up error signals only emerge after the effects of top-down communication. It could be postulated that modulatory top-down effects during anticipation are more widespread, involving a broader assembly of neurons, while modulatory effects during stimulation are limited to more local areas that are behaviorally relevant.