Occipital, parietal, and frontal cortices selectively maintain task-relevant features of multi-feature objects in visual working memory

Highlights

• Non-spatial features can be maintained in the parietal and frontal cortices.

• Feature-specific representations in visual working memory are highly selective.

• Only task-relevant features are maintained during visual working memory.

Abstract

Previous studies have shown that information held in visual working memory is represented in the occipital, parietal, and frontal cortices. However, less is known about whether the mnemonic information of multi-feature objects is modulated by task demand in the parietal and frontal regions. To address this question, we asked participants to remember either color or orientation of one of the two colored gratings for a delay. Using fMRI and an inverted encoding model, we reconstructed population-level, feature-selective responses in the occipital, parietal and frontal cortices during memory maintenance. We found that not only orientation but also color information can be maintained in higher-order parietal and frontal cortices as well as the early visual cortex when it was cued to be remembered. Conversely, neither the task-irrelevant feature of the cued object, nor any feature of the uncued object was maintained in the occipital, parietal, or frontal cortices. These results suggest a highly selective mechanism of visual working memory that maintains task-relevant features only.

Introduction

Visual working memory (VWM) refers to the ability to maintain and manipulate visual information in mind for a brief period of time when it is no longer in view (Curtis and D’Esposito, 2003, Khan and Muly, 2011, Luck and Vogel, 2013). As humans have highly limited cognitive resources, VWM performance is largely influenced by the ability to selectively attend to, memorize, and manipulate information according to task demand. The parietal and frontal cortices have long been considered to play a role in these processes, such as selecting task-relevant information, and ignoring or even suppressing task-irrelevant information, in order to prioritize the current task goals and optimize the deployment of limited cognitive resources (D’Esposito et al., 2000; Miller and Cohen, 2001).

In the neurophysiology literature on non-human primates, there has been ample evidence showing that parietal and frontal neurons can exhibit feature selectivity when the feature is task-relevant (Bichot et al., 1996; Buschman et al., 2012; Toth and Assad, 2002). More recent studies have also shown that such feature-selective tuning can be fine-scaled and modulated by task demand (Ibos and Freedman, 2014). In addition, human neuroimaging work using multi-voxel pattern analysis (MVPA) has shown that the parietal (Albers et al., 2013, Bettencourt and Xu, 2016, Christophel et al., 2012) and frontal (Albers et al., 2013, Lee et al., 2013) cortices, as well as the early visual cortex (Harrison and Tong, 2009, Serences et al., 2009), can represent the feature-specific information held in VWM. Furthermore, recent work using an encoding technique showed that feature-specific representations can be reconstructed at the population-level from activities in some of the parietal and frontal regions during VWM maintenance (Ester et al., 2015). This work has provided a novel tool of examining feature-selective responses during VWM maintenance in higher-level cortical areas.

However, it remains to be examined if these feature-specific representations can reflect the selective maintenance of task-relevant information, especially when multiple features are presented simultaneously. Although it has been previously shown that manipulation of attention to visual features can modulate feature-selective signals in the early visual cortex (Jehee et al., 2011) as well as in the parietal and frontal cortices (Ester et al., 2016, Liu et al., 2011) during visual attention tasks, few studies have examined how the feature-selective mnemonic representations are modulated by task demand in the parietal and frontal cortices (for early visual cortex see Serences et al., 2009).

In the current study, we aimed to systematically investigate the mnemonic representations of both task-relevant and task-irrelevant information in the occipital, parietal, and frontal cortices when multi-feature objects were in view. We measured participants’ BOLD activity using fMRI while they performed a delay-estimation task on one of the visual features (color or orientation) of multi-feature objects. We examined mnemonic representations of color in particular because unlike orientation, color does not contain any spatial component; therefore, the memory-related signals in the parietal and frontal cortices are less influenced by spatial attention or motor-related signals which have also been implicated in frontoparietal regions (Kastner et al., 2007, Schluppeck et al., 2005, Silver et al., 2005). Using an inverted encoding model (Brouwer and Heeger, 2009, Brouwer and Heeger, 2011, Ester et al., 2015, Sprague et al., 2015), we reconstructed population-level feature representations during the delay phase. Our results revealed that VWM representations in the occipital, parietal, and frontal cortices are strongly modulated by task, and that only task-relevant mnemonic representations can be maintained in the early visual and high-level parietal and frontal cortices, suggesting the highly selective nature of VWM representations.