The physiology of mental representations: from single neurons to circuits.

Mental representations enable the brain to transcend the immediacy of sensory input and operate on internal models of the world. Working memory (WM) provides the neural substrate that maintains, manipulates, and accesses these representations over behaviorally relevant timescales. Early studies identified persistent neuronal activity in the dorsolateral prefrontal cortex (dlPFC) as a key mechanism supporting WM, arising from specialized recurrent microcircuits in which excitatory and inhibitory interactions sustain information in the absence of external stimulation. Subsequent research has revealed a richer and more diverse set of mechanisms underlying WM representations. Depending on task demands, cognitive state, and behavioral context, information can be encoded through persistent activity, oscillations, activity-silent synaptic processes, and temporally structured neural sequences. These representational formats emerge from canonical cortical microcircuits composed of excitatory and inhibitory neurons whose physiological properties vary across brain regions. Interactions between WM, attention, and executive control ensure that internal representations are selectively maintained, updated, and flexibly deployed to guide reasoning, planning, and decision-making. At a broader systems level, WM representations distributed across prefrontal, association and sensory cortices may occupy low-dimensional neural manifolds whose geometry defines the content of conscious access, allowing memories, goals, and imagined scenarios to enter awareness even in the absence of sensory input. Together, current evidence suggests that WM representations are emergent properties of recurrent, hierarchical neural networks whose dynamics are continuously shaped by context, experience, and behavioral demands, providing a flexible neural architecture for adaptive cognition and conscious thought.