Everyday human cognition and behavior is accomplished via the coordinated efforts of numerous complex processes. From retaining information in memory to maintaining long-term personal goals, human behavior is multifaceted. In light of this complex nature, there exists massive variability from human to human in their ability to perform and implement these behaviors. In this dissertation, I present three experiments that elucidate the neural processes underlying this variability. Each experiment involved the collection and analysis of functional magnetic resonance imaging (fMRI) data. In one of these experiments, subjects performed a cognitive task during fMRI, while the other two assessed the relationship between baseline neural activity and complex cognition.

The first chapter of this dissertation uses behavioral modeling and fMRI to assess individual differences in the ability to learn and subsequently implement a hierarchically-structured rule set. I show that humans are capable of learning and implementing the complex hierarchical rule and that neural activity across multiple networks supports this ability. First, I find that brain regions across frontal and parietal cortices support the initial discovery of the hierarchical rule. Next, activity across a cingulo-opercular network of brain regions supports the generalization of this knowledge to novel settings.

I present evidence in the second chapter that individual differences in cognition and behavior are not only predicted by the patterns of coordinated neural activity across the entire brain, but that novel temporally-varying analysis approaches provide additional predictive power unobtainable with previous approaches. Here, I analyzed fMRI data collected while subjects performed no explicit task, a procedure referred to as “resting-state” fMRI. Subjects also completed a set of cognitive computer tasks that measured complex cognitive abilities such as working memory and cognitive control. By applying both traditional time-invariant and novel time-varying graph theoretical analyses to the resting-state fMRI data, I was able to predict individual differences in the cognitive abilities measured outside the MRI scanner. Moreover, the novel time-varying analysis revealed relationships to behavior that better captured task-specific behavioral variability.

The final chapter examines the ability of resting-state graph theoretical approaches to predict cognitive abilities related to attention and inhibitory control. Moreover, neural measures were interrogated alongside measures of human physiological functioning. Here, I find that all attentional and inhibitory abilities are accurately predicted across all neural and physiological measures. Specifically, neural measures indicative of brain-wide activity patterns predicted attentional accuracy and global inhibition ability, while neural measures reflecting activity patterns of particular neural networks additionally predicted attentional reactivity. Further, measures of physiological functioning were able to predict individual aspects of inhibitory control.

Together, the three experiments presented here contribute to our knowledge of how neural activity patterns ultimately beget complex cognition and behavior. Using multiple complimentary analytical approaches, I find evidence for the role of multiple neural networks in explaining the differences in cognitive abilities across the healthy human population.

Theories of working memory and attention postulate that the prefrontal cortex (PFC) provides top-down signals to other brain regions in order to keep behaviorally relevant sensory information activated and to suppress competing task-irrelevant information. Although there are numerous studies that suggest the PFC is a source of top-down modulation of posterior brain regions, the vast majority of these studies offer only indirect evidence in support of this claim. Additionally, while this executive control is usually thought of as being beneficial to the task at hand, there is reason to believe that there are certain circumstances in which this control is deleterious to performance. Here we provide direct evidence using transcranial magnetic stimulation and neuroimaging that 1) the PFC is a source of top-down control of early visual regions and 2) interfering with PFC function can sometimes lead to paradoxical improvements in task performance.

Verbal working memory refers to the limited-capacity store responsible for maintaining and manipulating task-relevant information in a verbal form over short time-periods. The cortical regions that underlie this type of memory can be dissociated into control and storage regions. Control processes may involve operations such as the identification and selection of relevant information and the assembly and initial execution of articulatory rehearsal programs. In contrast, storage regions are where the actual task-relevant information is rehearsed. Presumably, control and storage regions transfer information among each other within a functionally connected network.

To study the cortical network formed during the delay period of a verbal working memory task, the first study was specifically designed to investigate functional connectivity. While the functional connectivity results counterintuitively led to a null result, it was demonstrated that middle frontal gyrus (MFG) and superior parietal lobule (SPL) were active during encoding and the start of the delay period, as well as at the end of the delay period, suggesting they were involved in control processes. In contrast, it was demonstrated that inferior frontal gyrus (IFG), premotor (PM) and Sylvian-parietal-temporal junction (area Spt) were active throughout encoding and the entire delay period, suggesting they were involved in storage/rehearsal processes.

In a second follow-up study, results demonstrated that control regions MFG and SPL were not sensitive to manipulations in rehearsal rate, but exhibited non-linear memory load effects. Conversely, storage/rehearsal regions IFG, PM and area Spt demonstrated approximately linear rehearsal rate and memory load effects. Additionally, memory load effects were found to diminish through time in all areas, while rehearsal rate effects persisted through the entire delay period in storage/rehearsal areas. It was also revealed that rehearsal rate and memory load were confounded behaviorally as subjects strategically increased their rehearsal rate to manage an increased memory load, until the load became so great that their rehearsal rate began to decrease. It was also discovered that high capacity individuals have less activity in MFG, IFG and SPL, but more activity in PM.

In a third study, substitute tasks were made for the encoding period (listen word) and delay period (subvocal rehearsal) of standard verbal working memory tasks and the timing and network properties of the active cortical regions were investigated. Results demonstrated that during the listen word condition activity in primary auditory cortex preceded area Spt, and that primary auditory cortex and area Spt formed a functional network. In contrast, during the subvocal rehearsal condition activity occurred first in PM, then in area Spt, followed by IFG. These three regions also formed a functional network in the subvocal rehearsal condition. Together these experiments provide novel insights into the cortical mechanisms underlying verbal working memory.

An abundance of research has demonstrated that the brain is both organized into discrete, highly specialized, regions and that distributed regions work together in the form of large-scale networks. It remains unclear, however, how to reconcile these two views of brain function. In this thesis, I present several projects that investigated the specialized properties of individual regions, how these change depending on cognitive demands and external disruption, and how these changes interact with modulations across large-scale networks. Across these projects, I use a combination of methods including psychophysics, univariate and multivariate modeling of functional Magnetic Resonance Imaging (fMRI) data, Transcranial Magnetic Stimulation (TMS), and the study of patients with focal lesions.

The first section of this thesis focuses on visual attention to high-level face stimuli. Here, I demonstrate that highly specialized representations exist for individual faces, but that engaging in sustained top-down attention can cause selective modulations in these representations. This suggests that large-scale networks can interact with, and modify, the properties of local regions in an extremely selective manner.

The second section of the thesis examines how disrupting local regions, through focal brain lesions or TMS, impacts the structure of large-scale networks. The first two chapters in this section examine how acute and chronic disruptions modulate the magnitude of interactions throughout cognitive control networks. An additional project explores methodological considerations involving the variability in TMS effects and how these might be measured most effectively. The final project in this section examines how focal brain lesions also affect the organization of distributed networks. This section illustrates how the properties of local regions can modulate both the magnitude and organization of distributed networks.

In sum, these projects demonstrate that large-scale networks can both drive and be driven by the highly specialized properties of local regions. This research will hopefully provide some insight into how brain regions may concurrently demonstrate highly specialized brain activity and participate in distributed brain networks, and flexibly modify these interactions, in order to carry out complex behavioral functions.

A key challenge in neuroscience is to synthesize our understanding of neural structure, dynamics, and information processing in both health and disease. While substantial progress has been made toward such a synthesis at the microscopic scale of single neurons and their connections, signicant work remains to be done at the macroscopic scale of interacting brain regions. While the eld has successfully mapped the macroscale connections bridging cortical columns and regions, and has also systematically described basic features of macroscale cortical electrodynamics across perceptual, cognitive, and brain states, neuroscientists still currently lack a mathematically-specic understanding of how these macroscale networks and electrodynamics underpin large-scale neural computation and communication. Toward the end of advancing our mathematical understanding of this relationship between large-scale brain networks, dynamics, and information ow, we here present: a tool for quantifying, in bits, how much information is integrated across large-scale brain networks; a tool for tracking the presence and degree of chaos in neural electrodynamics; and evidence that macroscale cortical circuits optimize their information-carrying capacity during conscious states by operating near edge-of-chaos criticality.

Visual cognition applies temporal smoothing to its inputs, which creates a serial dependence between successive representations at the focus of attention. This is thought to promote perceptual stability. While the benefits of serial dependence have been assumed, evidence that perception itself is altered has been limited. In the first chapter of this dissertation, I vary the delay between stimulus and response in a spatial delayed response task to investigate whether serial dependence occurs at the time of perception or later in working memory. I find that behavioral responses made immediately after viewing a stimulus are on average veridical. Only as memory demands increase is a blending of past and present information apparent in behavior, reaching its maximum with a memory delay of six seconds.

In the second chapter, I explore potential neural-circuit mechanisms of serial dependence. I consider two possible substrates of the effect: stable persistent activity during the memory delay and dynamic “activity-silent” synaptic plasticity. I find that networks endowed with both strong reverberation to support persistent activity and dynamic synapses can closely reproduce behavioral serial dependence. Specifically, elevated activity drives synaptic augmentation, which biases activity on the upcoming trial, giving rise to a spatiotemporally tuned shift in the population response. My hybrid neural model is a theoretical advance beyond abstract mathematical characterizations of working memory and demonstrates the power of biological insights to provide a quantitative explanation of human behavior.

The model developed in Chapter 2 proposes that serial dependence is due in part to synaptic augmentation, which is especially prominent in prefrontal cortex. In the third chapter, I investigate whether the bias in behavior depends on activity in three separate nodes of prefrontal cortex (PFC) – the frontal eye fields, the dorsolateral PFC, and the anterior PFC near the frontal pole. I find that transcranial magnetic stimulation (TMS) to these nodes causes reductions in serial dependence consistent with the model’s predictions. In contrast, TMS to posterior sites – either primary somatosensory cortex or posterior parietal cortex – fails to alter the magnitude of the behavioral effect. This general result holds across TMS protocols (online vs. offline) and tasks with different stimulus and response types.

Visual working memory, a complex cognitive process that is essential for goal-directed behaviors, allows the internal maintenance and manipulation of detailed visual information that is no longer available in the environment. The neural processes and representations that support this essential ability remain the focus of much research and debate. In this dissertation, I present three experiments that tested key predictions of a sensory recruitment model of visual working memory, which proposes that the same regions responsible for primary sensory processing are recruited to maintain precise sensory details over short delays. All of these experiments involved the collection of functional magnetic resonance imaging (fMRI) data while participants performed cognitive tasks requiring visual working memory. The primary fMRI analyses described here utilize two multivariate approaches that model the information content of distributed patterns of brain activity: inverted encoding models and multivoxel pattern analysis.

The first chapter of this dissertation uses an inverted encoding model to reconstruct simple orientation information held in working memory and examines the effect of subsequent visual input on these memory representations. Here, I show that visual working memory representations are flexible and can dynamically adjust to meet task demands. First, I find that the early visual areas maintain precise orientation information over a delay, but that these representations are susceptible to bias from visual interference. Further, I find that the intraparietal sulcus redundantly represents orientation information in anticipation of possible distraction and continues to do so if visual interference renders early visual cortical representations unreliable.

In the second chapter, I present strong evidence in favor of a sensory recruitment model of visual working memory for complex images like human faces. Here, I use an encoding model to characterize distributed response patterns in face-selective regions of inferior temporal cortex during face perception. I show that by inverting this perceptual encoding model and applying it to data from a memory delay, I can successfully reconstruct face information held in working memory. Importantly, this perception-to-memory generalization implies a common representational structure between perceptual and mnemonic codes.

The final chapter examines the role of lateral prefrontal control processes in shaping successful visual working memory performance, by examining fMRI data collected after disruption of inferior frontal gyrus activity with transcranial magnetic stimulation. Here, I provide causal evidence that the lateral prefrontal cortex modulates the selectivity of working memory representations in extrastriate visual cortex. In addition, I find that the inferior frontal gyrus is involved in determining the task-relevance of visual input and communicating that information to a network of regions involved in further processing during visual working memory.

Together, these experiments contribute to our understanding of how the human brain represents both basic visual features and complex images like human faces and visual scenes. I find evidence for the recruitment of sensory regions in the maintenance of precise visual information in memory and for top-down prefrontal and parietal control processes that shape the selectivity and distractor-resistance of these mnemonic representations.