The ability of neurons to generate persistent activity in the absence of stimulation is of fundamental importance to complex behaviors. Deprived of this faculty, our brain would be completely at the mercy of sensory inputs of the moment; there would be no internal state of the mind, nor association and integration across time. Higher cognitive processes, such as working memory, decision-making or goal-directed behavior, would be impossible.
Since Niki in Japan and Fuster in the USA published their seminal papers in 1971, neurophysiologists have documented memory-related persistent activity in many brain regions. Only recently, however, has the study of cellular and circuit mechanisms of persistent activity come to the fore. The goal for this special issue of Cerebral Cortex is to provide an overview of this emerging field.
The persistence time (up to 10 s) of sustained firing activity is orders of magnitude longer than the biophysical time constants (tens of milliseconds) of fast electrical signals in neurons and at synapses. For this reason, persistent activity is believed to be generated by feedback dynamics, or reverberation, in a neural circuit. This idea is made precise in theoretical work, most explicitly by Amit, where persistent activity is described in terms of ‘dynamical attractors’. The mathematical term ‘attractor’ simply means any self-sustained and stable state of a dynamical system, such as a neural network. For example, according to this picture, in a working memory system, the spontaneous state and stimulus-selective memory states are assumed to represent multiple attractors, such that a memory state can be switched on and off by transient inputs. This formulation is plausible, inasmuch as stimulus-selective persistent firing patterns are dynamically stable and approximately tonic in time (e.g. across a delay). To assess whether attractor networks can be realized in the brain, biologically constrained models of persistent activity were needed, which became possible only recently thanks to the advances in quantitative neurophysiology. A biophysically realistic model designed for a particular system, say the prefrontal cortex, can reproduce a large set of neural data from behaving animals only under restricted conditions. This gives rise to insights into, and testable predictions about, the structure and operation of the neural circuit under study.
Recent anatomical and physiological studies have begun to reveal feedback mechanisms at play in brain regions that are presumably capable of persistent activity. Specifc scenarios have been formulated and tested, like the role of NMDA receptors at local synaptic connections, bistable or multi-stable dynamics of single neurons, dopamine modulation of synaptic transmission and plasticity. The development of active slice preparations with ongoing firing patterns has allowed us to study in detail the biophysics of self-sustained, albeit nonspecific, neural activity (it remains a challenge to produce switchable and selective patterns of persistent activity in a slice preparation.) Another topic that has spurred considerable interest is neural systems that, in the form of a continuous family of persistent activity states, encode an analog quantity such as head direction or eye position, spatial location of a visual cue, or the frequency of a vibrational stimulus.
In some sense, converting a pulse-like transient input into a tonic persistent activity can be viewed as an integral calculation. However, time integral requires more than simple switches. To integrate over a long-lasting input or continuously varying inputs, neurons must ramp slowly in time. Such integration can be achieved by slow reverberatory dynamics, instead of fast switches. Time integration is increasingly recognized as of general relevance to diverse neural systems: oculomotor neurons that integrate transient saccades into sustained internal representation and memory of eye position; head-direction cells that compute a time integral of head angular velocity signals to encode and maintain the animal’s directional heading during navigation; cortical circuits that slowly accumulate sensory data in a decision-making process. The implications are twofold. First, persistent activity is a widespread phenomenon, therefore should be studied in many different systems. More and simpler preparations offer new opportunities for elucidating the biophysical mechanisms of persistent activity. Secondly, time integration represents a unifying concept for understanding why persistent activity may confer neural networks with the capability of not only memory storage, but also interesting ‘higher-level’ computations that cannot be accomplished otherwise.
These and other topics are covered in the contributions to this special issue. It is our hope that many key questions discussed in these pages will stimulate further research in the future. What are the critical experimental tests for the attractor model? Are there alternative theoretical frameworks for describing persistent activity? What are the defining properties of a neural microcircuit that enable it to generate stimulus-selective persistent activity, or/and compute time integration? What are the relative contributions of recurrent synaptic circuitry versus single cell dynamics to the generation of persistent activity? Are NMDA receptors critical to the generation of persistent activity? How is the inhibitory circuitry organized in persistent activity networks? What are the regulatory mechanisms for the robustness of neural networks with a continuum of persistent activity states? How are persistent activity patterns shaped and modified by experience and reward-based learning?
This special issue was co-edited with Dr Patricia Goldman-Rakic, who passed away suddenly on July 31, 2003. Pat was a pioneer in the studies of persistent activity, working memory and prefrontal function. Her prescient and massive contributions will have a lasting impact to our field. Throughout her career, Pat’s work was uniquely marked by its explicit emphasis on the cellular, molecular and circuit basis of higher brain functions. In her view, computational principles and biophysical mechanisms of persistent activity, discussed in this issue, might hold the key to understanding the neural basis of cognition as well as certain psychiatric disorders.