Although much progress has been made since neurologist Richard Restack called the brain one of science’s last frontiers, the functions of some brain areas remain mysterious. Foremost among these is prefrontal cortex (PFC), a region that is much reduced in size in most other primates, is among the last areas to develop in human children, and yet is active in almost every cognitive task.
In general, prefrontal cortex is associated with higher-order cognition, such as those processes involved in planning, strategizing, self-monitoring, self-regulation, and more generally, the coordination of behavior to achieve goals. Although Brodmann identified several structurally distinct subregions of prefrontal cortex in the early 1900’s, these same prefrontal regions have been particularly resistant to functional characterization for several reasons. Among these are the fact that prefrontal activity is seen in almost every task, that many prefrontal regions are completely interconnected with most other prefrontal regions, and that activation nearly always spans multiple Brodmann areas. In addition to these difficulties, some theories suggest that prefrontal cortex may forever elude functional characterization because its computations are intrinsically amorphous: the same regions may perform a variety of computations depending on the current task demands (known as the “adaptive coding” hypothesis).
With these obstacles, it is no wonder that the cognitive neuroscience literature is so equivocal on the topic of prefrontal computation. But because prefrontal cortex is thought to be the seat of higher cognition, the prospect of computationally or cognitively specifying its processes is particularly tantalizing.
To that end, a 2003 Nature Reviews Neuroscience paper by Wood & Grafman reviews the basic neurology of prefrontal cortex, along with 8 distinct theoretical perspectives on what cognitive processes or representations prefrontal cortex is actually responsible for.
First, the neurology on which these theories are based:
Wood & Grafman review how ventromedial PFC (vmPFC) is richly and reciprocally connected with “brain regions that are associated with emotional processing (amygdala), memory (hippocampus) and higher-order sensory processing (temporal visual association areas), as well as with dorsolateral PFC,” whereas dorsolateral PFC (dlPFC) is reciprocally connected “with brain regions that are associated with motor control (basal ganglia, premotor cortex, supplementary motor area), performance monitoring (cingulate cortex) and higher-order sensory processing (association areas, parietal cortex).” Thus the authors characterize vmPFC as more involved with the integration of emotional information, whereas dlPFC is more involved in the control or regulation of behavior.
Wood & Grafman also note that PFC neurons are more “spiny” than others, allowing them to receive more excitatory input, which may contribute to the characteristically abstract nature of the stimuli that elicit maximal prefrontal activity. In addition, PFC activity shows a different temporal profile than more sensory-specific and posterior regions of cortex, such that PFC neurons can remain active over extended periods of time even in the complete absence of sensory input.
What follows is a summary of Wood & Grafman’s review of PFC computational and representational theories:
Adaptive Coding Hypothesis
As mentioned above, one hypothesis about PFC function is that it’s a kind of chalkboard on which neural activity self-organizes to represent features of the current task. In this view, PFC representations are temporary and mutable, as are functional divisions between various prefrontal regions.
Attentional Control Model
Several decades ago, Norman and Shallice proposed an extremely influential model of executive control that has subsequently come to be viewed as a model of prefrontal cortex. According to this model, a “contention scheduling” system merely assigns priorities to different “threads” or “schemas” for execution, while a “supervisory attentional” system can adjust those priorities if the current task is novel or otherwise inapprioriate for standard priority assignments.
Wood & Grafman review a connectionist model of PFC in which PFC regions integrate sensory and motor processing, thus mediating stimulus-response mappings with input from reinforcement learning and current motivational state. Note that there are several other connectionist models of prefrontal function, this being just one!
Structured Event Complex (SEC) framework
Wood & Grafman review their own theory of PFC function, in which PFC stores “event complexes” consisting of schemas or episodes that include aspects of “thematic knowledge, morals, abstractions, concepts, social rules, event features, event boundaries and grammars.” Each of these various aspects is stored in a distinct region of PFC, specified by the SEC theory, and consistent with regional connectivity with more stimulus-specific posterior regions.
Guided Activation Theory
Wood & Grafman also review Miller & Cohen’s theory of PFC function, in which the PFC stores “task-specific rules, attentional templates and goals” to bias the activity in more posterior regions in order to achieve current goals. As presented here, this theory seems very similar to the Supervisory Attentional Control system proposed by Norman & Shallice.
Somatic Marker Hypothesis
According to this view, vmPFC mitigates the convergence of information from hippocampus, amygdala and posterior cortex for the purposes of decision making. The representations supported by PFC are “somatic markers” which link previous behavioral states with goal-related outcomes.
Temporal Organization Model
Yet another view of PFC function is that it is responsible for the temporal sequencing of behavior through mechanisms of behavior monitoring, short-term memory, inhibition, attention, and motor schemas. vmPFC involves these same mechanisms, but with an additional emotional component. Each of these functions is tied to distinct regions of PFC; for example, inhibition may be localized to medial PFC.
Working Memory model
Perhaps the most influential model of prefrontal processing is that proposed by Goldman-Rakic, in which PFC and a network of other regions (parietal cortex, anterior cingulate, and limbic regions) perform “active maintenance” of representations in the absence of sensory input. Wood & Grafman characterize this view as mostly limited to the dlPFC, and thus as an incomplete model of PFC in general.
Wood & Grafman do not review several other models of prefrontal function, perhaps because many of them are more limited in scope than those presented above. For example, in this article, Koechlin, Ody, and Kouneiher proposes a compelling “cascade model” for lateral PFC function; this paper by Ramnani & Owen suggests a role for anterior PFC in evaluating the outcome of two “cognitive operations”; and this paper by Hazy, Frank & O’Reilly proposes a computational model of dopaminergic interactions between prefrontal cortex and basal ganglia.
Clearly much progress has been made since the publication of Wood & Grafman’s review paper – more recent developments in our understanding of prefrontal function, exemplified by these and other papers, will be discussed in my next few posts. But I hope that this post has served as a good introduction to the variety of perspectives brought to bear on understanding one of the most elusive and tantalizing regions of the brain: the prefrontal cortex.