Working memory enables
Working memory enables the storage of information in a readily available state. This system is an integral component of general cognitive ability, as evidenced by robust correlations between individual differences in working memory performance and scores on tests of general intelligence (e.g., Stanford-Binet IQ; Cowan 2001) and scholastic achievement (Daneman & Carpenter, 1980). Moreover, disruptions in working memory performance are common in several psychiatric and neurological disorders, including Attention-Deficit/Hyperactivity Disorder, Parkinson’s Disease, and Schizophrenia (e.g., Fallon et al., 2015; Fried et al., 2015; Van Snellenberg et al., 2016). Thus, the basic architecture of working memory is of interest to scientists, educators, and clinicians alike.
As a general-purpose workspace, working memory provides a critical bridge between incoming sensory signals and anticipated future behaviors. However, despite strong conceptual links between working memory and motor control, extant studies of working memory have focused on understanding mechanisms of storage independent of motor outputs (e.g., Ester et al., 2013; Ester et al., 2015), or vice versa (e.g., Cisek & Kalaska, 2005; Li et al., 2016). Consequently, we lack a basic understanding of how the contents of working memory are used to implement appropriate memory-guided behaviors. This is a critical knowledge gap, as the ability to create and sustain links between mnemonic representations and task-relevant motor plans may be a key limiting factor in working memory performance and general cognitive ability. A greater understanding of how working memory representations are used to guide overt behaviors can inform (a) research into the basic structure of working memory, (b) research into limitations on complex cognition, and (c) research into how internal representations are mapped onto motor outputs (e.g., for the development of brain-computer interfaces). Motivated by these goals, the focus of this proposal is to understand how working memory representations are utilized to produce efficient memory-guided behaviors.
Leading models of working memory propose a “distributed and redundant” coding structure, where redundant representations of to-be-remembered information are encoded by a heterogeneous network of cortical areas, including motor circuitry (e.g., Christophel et al., 2017). However, this model is silent on the question of how working memory representations are used to generate behavioral outputs. We propose a “response-specific” coding structure, where memory representations are encoded jointly by sensory areas and cortical areas responsible for generating appropriate motor outputs (e.g., an eye movement or manual response; Figure 1). This architecture would enable rapid and adaptive memory-guided behavior through direct interactions between the contents of working memory and neural circuits responsible for generating an overt response.
The response-specific coding model makes several testable predictions. First, the model predicts that cortical regions responsible for generating task-relevant behaviors (e.g., a manual response) encode stronger stimulus-specific working memory representations than cortical regions responsible for generating other behaviors (e.g., a verbal response). The experiments described in Aim 1 will test this prediction by manipulating response demands in a working memory task while holding other all other experimental factors constant. Second, the model predicts strong relationship between stimulus-specific working memory representations and circuits for motor control. An extreme version of the model predicts that, when possible, memory representations and motor plans are stored as cointegrated units, such that such that activating or prioritizing a memory representation leads to an automatic prioritization of an associated motor plan (or vice versa). This arrangement would be advantageous in controlled settings where memory-motor mappings remain constant, but potentially disadvantageous in more complex settings where memory representations and motor plans must be constantly updated to match changing environmental circumstances. Thus, although stimulus-specific memory representations may be preferentially encoded by effector-specific cortical regions, it is unclear whether these representations are stored independently of motor plans. Aim 2 will leverage fMRI and EEG to examine the degree of overlap between memory representations and motor plans.