Neural Mechanisms of Preparation in Expectation and Selective Attention

Abstract

The world is complex and constantly changing. We are bombarded by a large number of stimuli that generate ambiguous patterns in our sensory receptors, to which we usually need to give quick and effective responses. To cope with this complexity and ambiguity, among other things, humans use the ability to anticipate what is going to happen before it happens. Indeed, it is widely believed that preparing to perceive impending information improves performance (Barton et al., 2006). However, some fundamental questions arise from these seemingly self-evident claims. How does preparation improve performance? Does it involve the representation of specific information? And, most importantly, is proactive preparation a unified process or does it employ dissociable mechanisms in different cognitive contexts? This thesis frames preparation as a brain process of internal or endogenous origin, top-down, referring to the origin of information in higher regions of the cortex, which would descend perceptual areas. This definition contrasts with another branch of bottom-up mechanisms, in which the trigger of the activity is in the stimuli of the environment. There are at least two possible sources of information that can lead to this endogenous anticipation: our goals and objectives, and the knowledge we have of the regularities of the environment. The activity associated with goals is related to the internal control of specific resources that allow us to effectively perform different tasks. This proactive cognitive control (Braver, 2012) is associated with brain regions such as the frontoparietal network or the multiple demand network (Dosenbach et al., 2007). Among its main capabilities we find the orientation of attention towards relevant information. This selective attention acts as a filter that allows us to better process necessary information while ignoring distracting information. Crucially, it has been shown on numerous occasions that while we select relevant information, this information is represented in the visual cortex, through patterns of activity associated with the stimuli we are attending to (Peelen & Kastner, 2011; Stokes et al., 2009). Moreover, when presented with stimuli that match what is selected, both increases in activity in perceptual regions (Kastner et al., 1999) and greater accuracy in processing those stimuli are often found (Woolgar et al., 2015). On the other hand, models that explain the effects of prior knowledge on the statistical regularities of the environment, or expectations, have gained much relevance in recent years. These predictive processing models (Feldman & Friston, 2010; Friston, 2005) highlight how at all times the brain acts as a machine that anticipates what is most likely. These top-down predictions would be contrasted with bottom-up information from sensory receptors. From the contrast of the two, prediction errors emerge, which are then used to update the following predictions and thus give the brain a flexible mechanism capable of adapting in different circumstances. These models also describe how probable information can be represented in the visual cortex before bottom-up stimulation (de Lange et al., 2018; Kok et al., 2017). However, in contrast to selective attention, here when a probable stimulus is perceived, smaller activations are usually found (Feuerriegel et al., 2021). With respect to the precision of these representations, the evidence is still conflicting, and data have been found pointing to both sharpened accuracy (Kok et al., 2012) and a dampening of these representations (Richter et al., 2018). Although these two phenomena, selective attention and expectation, are associated with different effects during stimulus processing, both involve anticipatory representations in visual cortex. However, it has not been investigated whether these representations share the same processing mechanisms, or whether, on the contrary, preparation is a complex and heterogeneous phenomenon that depends on the cognitive context in which it occurs. Therefore, the main goal of this thesis is to examine the nature of preparation, and to evaluate how it changes across different contexts. To this end, we designed a task in which different cues provided information about target stimuli. These cues could indicate the relevance (attention) or probability (expectation) of the stimuli. This task was implemented in two experiments in which data were collected from different neuroimaging tools. The analyses focused primarily on multivariate techniques (Hebart & Baker, 2018), which are particularly sensitive and effective in studying patterns of brain activity that inform how information is represented. In the first two studies, we investigated proactive preparation, exploring how anticipated information is represented in the brain, and whether and how it changes across relevance and probability contexts. The first study focused on the temporal dynamics of preparatory activity. For this purpose, brain activity was recorded with Electroencephalography (EEG). The results showed that several events take place during the preparatory window. First, the perceptual characteristics of the signal are processed. Then, during most of the trial, global differences associated with the mechanisms deployed during attention and expectation were found. Finally, the anticipatory category was represented throughout the trial, especially towards the end. Crucially, although both attention and expectation showed similar decoding accuracies and magnification effects, the results of one did not generalize to those of the other. This suggests that although anticipatory information is represented in the two conditions, the nature of such representations differs. Finally, we sought to explore the level of similarity between preparatory representations and stimulus perception. The two conditions differed in the similarity of patterns between the cue and the target; specifically, the preparation during attention was more similar to the target activity. This first experiment left an important question open: what are the neural substrates underlying the representations encountered during anticipation? We addressed this question in the second experiment, in which we adapted the behavioral paradigm to functional magnetic resonance imaging (fMRI). Again, we found evidence of anticipatory encoding that was different for attention and expectation. The results revealed increased activations for attention in early visual regions, whereas the posterior cingulate cortex was more active for expectation in this window. In addition, anticipatory information showed large differences in the involvement of different visual and frontoparietal regions during anticipation, suggesting that different networks exhibit distinguishable mechanisms in attention and expectation. In addition, we found data indicative of category representation in the ventral visual cortex, partially replicating the results of the EEG experiment. We found that the attended and probable categories were represented in nearby, though mostly different, regions. Again, cross-classification was not possible between attended and expectation, implying that the representational nature of preparation in the two conditions was different. When comparing the similarity between anticipatory representations and stimulus perception, we found that these similarities occurred in different regions depending on the condition. In addition, a voxel selectivity analysis revealed that only expectation showed some evidence of cue-to-target generalization, and it did so in the ventral visual cortex. Finally, the representation of anticipated stimuli did not correlate with behavior in either condition. Finally, after studying the nature of top-down preparation, we investigated how the accuracy of prior induced representations changed through attention and expectation. Specifically, we set out to study whether cues sharpen or dampen sensory representations. To do so, we re-analyzed the data from the two experiments to draw joint conclusions. As a preliminary step, we set out to find evidence of separate activation differences for attention and expectation. These analyses showed converging evidence of increased activations for both likely and relevant stimuli. Regarding sharpening and dampening effects, the results mostly supported a sharpening effect for attention, and a marked dampening in expectation. In addition, we examined whether the dampened representations would generalize better to the preparatory interval, as they should be poorer in both cases. Our hypothesis held true, as expectation trials generalized to the signal, whereas relevant targets did not. This again suggests sharpening for attended stimuli and damping for probable stimuli. In the fMRI experiment we studied the involvement of frontoparietal regions in the sharpening and damping effects. Although we were able to successfully decode target categories in some frontoparietal areas, these regions showed no evidence of sharpening or damping for either condition. Finally, we concluded that the results were not consistent with equivalent effects for attention and expectation, suggesting that not only during anticipation but also during target processing relevance and probability appear to act through different mechanisms. Taken together, our results show that preparation is a complex event, in which the representation of different stimuli is highly dependent on the context in which they occur. This preparation, moreover, affects the fidelity with which perceived stimuli are represented, which is also context-dependent. Finally, this thesis highlights the differences between selective attention and expectation, two processes that, although sharing similarities, involve different mechanisms and manners of representing information.