Visual association learning induces global network reorganization

Highlights

• Visual association learning induces changes in global brain connectivity pattern.

• Learning induces similar connectivity pattern between trained and associated stimuli.

• Learning modulates distributed connectivity between high- and low-level networks.

Abstract

It has been proposed that visual learning is accomplished not only by neural plasticity in the visual cortex, but also by complex interactions between bottom-up and top-down processes that may induce global network reorganization. Here, we applied a multivariate analysis to functional connectivity (FC) patterns across the brain to investigate how visual association learning was achieved through large-scale network reorganization. Participants were trained to associate a set of artificial line-drawing objects with English letters. After three consecutive days of training, participants underwent a functional magnetic resonance imaging scan in which they were presented with the trained stimuli, untrained stimuli, and English words. By calculating pairwise FC between 189 nodes of 10 well-established networks across the brain, we found that the visual association learning induced changes in the global FC pattern when viewing the trained stimuli, rendering it more similar to the FC pattern when viewing English words. Critically, the learning-induced global FC pattern differences were mainly driven by the FC related to the high-level networks involved in attention and cognitive control, suggesting the modification of top-down processes during learning. In sum, our study provides one of the first evidence revealing global network reorganization induced by visual learning and sheds new light on the network mechanisms of top-down influences in learning.

Introduction

Extensive findings indicate that visual learning can improve behavior performance in the training tasks and modify the visual system of humans and animals (Li, 2016; Watanabe and Sasaki, 2015). For example, training to discriminate basic visual properties (e.g., grating orientation, contrast, motion direction) or more complex visual objects modifies neuronal responses in the early or higher-order visual cortical areas (Baker et al., 2007; de Beeck et al., 2006; Golby et al., 2001; Kobatake et al., 1998; Logothetis et al., 1995; A. Schoups, Vogels, Qian and Orban, 2001; A. A. Schoups, Vogels and Orban, 1995; Shiu and Pashler, 1992; Sigman et al., 2005). However, it is increasingly acknowledged that visual learning is a complex and constructive process and is not mediated by local changes in the visual cortex alone.

Accumulating evidence indicates that top-down influences play a pivotal role in visual learning. Behaviorally, psychophysics studies demonstrate that the learning effect of discriminating grating orientations or motion directions can transfer to untrained location or stimulus properties with double-training procedures, suggesting the involvement of high-level cognitive processes (Wang et al., 2016; Wang et al., 2012; Xiao et al., 2008). The role of top-down influences in visual learning is also supported by physiological and neuroimaging studies showing task-dependent modulation of responses in the visual areas when two different tasks are trained for an identical set of stimuli (Li et al., 2004; Song et al., 2010b). More direct evidence for the involvement of top-down processes in visual learning come from neuroimaging studies showing that training of visual tasks induced activation changes in high-level fronto-parietal areas responsible for attentional control and decision making, in addition to changes in the visual areas (Kahnt et al., 2011; Lewis et al., 2009; Mukai et al., 2007; Sigman et al., 2005). Visual learning also strengthens functional connectivity (FC) between the visual areas and fronto-parietal areas (Lewis et al., 2009; Mukai et al., 2007). Therefore, convergent evidence suggests that visual learning induces complex interactions between bottom-up sensory processing and top-down cognitive control, which may manifest as large-scale network reorganization in addition to local changes. Here we tested this hypothesis by exploring the plasticity of global FC patterns across the brain with a multivariate approach (Dresler et al., 2017; Tambini et al., 2017).

To do this, participants were trained to associate a set of artificial line-drawing objects with English letters for three consecutive days, and underwent an fMRI scan after learning. Then, we calculated pairwise FC between 189 nodes from 10 well-established networks across the whole brain (Cole et al., 2013; Power et al., 2011) when participants viewed English words, the trained and untrained stimuli respectively. Instead of examining univariate changes in the FC for the trained stimuli, we used an FC pattern similarity analysis (Dresler et al., 2017; Tambini et al., 2017) to compare the multivariate FC pattern for the trained stimuli to a template provided by the FC pattern for English words. We tested whether the global FC pattern for the trained objects became more similar to the learning template and, more importantly, whether the learning-induced changes in global FC pattern were driven by the connectivity related to the high-level control networks or low-level sensory networks. Although changes in local activation have usually been observed in low-level visual system, we predicted that the training would induce distributed differences in the FC pattern of the high-level control networks if top-down influences indeed play a critical role in visual learning.