- S, Goldman-Rakic, Section of Neurobiology, Yale University School of Medicine, New Haven, Connecticut 06510
In the presence of normal sensory and motor capacity, intelligent behavior is widely acknowledged to develop from the interaction of short- and long-term memory. While the behavioral, cellular, and molecular underpinnings of the long-term memory process have long been associated with the hippocampal formation, and this structure has become a major model system for the study of memory (Bliss and Lomo, 1973; McNaughton and Nadel, 1990; Squire and Zola-Morgan, 1991), the neural substrates of specific short-term memory functions have more and more become identified with prefrontal cortical areas (Goldman-Rakic, 1987; Fuster, 1989). The special nature of working memory was first identified in studies of human cognition (e.g., Norman, 1970; Baddeley, 1986), and modern neurobiological methods have identified a specific population of neurons, patterns of their intrinsic and extrinsic circuitry, and signaling molecules that are engaged in this process in animals. In this article, I will first define key features of working memory and then descdbe its biological basis in primates.
Distinctive Features of a Working Memory System
Working memory is the term applied to the type of memory that is active and relevant only for a short period of time, usually on the scale of seconds. A common example of working memory is keeping in mind a newly read phone number until it is dialed and then immediately forgotten. This process has been captu red by the analogy to a mental sketch pad (Baddeley, 1986) an~l is clearly different from the permanent inscription on neuronal circuitry due to learning. The criterion–useful or relevant only transiently-distinguishes working memory from the processes that have been variously termed semantic (Tulving, 1972) or procedural (Squire and Cohen, 1984) memory, processes that can be considered associative in the traditional sense, i.e., information acquired by the repeated contiguity between stimuli and responses and/or consequences. If semantic and procedural memory are the processes by which stimuli and events acquire archival permanence, working memory is the process for the retrieval and proper utilization of this acquired knowledge. In this context, the contents of working memory are as much on the output side of long-term storage sites as they are an important source of input to those sites. Considerable evidence is now at hand to demonstrate that the brain obeys the distinction between working and other forms of memory, and that the prefrontal cortex has a preeminent role mainly in the former (Goldman.Rakic, 1987). However, memory-guided behavior obviously reflects the operation of a widely distributed system of brain structures and psychological functions, and understanding the prefrontal component is but one part of the grand design. Working memory in its most elementary form, the ability to keep events «in mind» for short periods of time, has been studied in nonhuman primates by delayed-response paradigms. Whereas in humans, facts and events accessed from long-term memory stores can be instigated by verbal instructions, in experiments with animals, the information to be processed has to be provided by the experimenter. In the case of the classical delayed response task, the subject is shown the location of a food morsel that is then hidden from view by an opaque screen. Following a delay period of several seconds, the subject chooses the correct location out of two or more choices. Thus, the subject must remember where the bait had been placed a few seconds earlier, and the correct response is guided by a representation of the prior stimulus rather than the stimulus itself. Furthermore, as the location of the bait changes randomly from trial to trial, another critical feature of the delayed-response task is that the correct response on any given trial cannot be predicted from the preceding trial, and consequently, information must be updated on a trial-to-trial basis. The underlying principle of delayed response operates in many cognitive paradigms, including the match-to-sample or nonmatch-to-sample tasks commonly used to test hippocampal function in monkeys (Mishkin, 1982; Squire and Zola-Morgan, 1991). In these tasks, as in spatial delayed-response tasks, the animal must defer its response, update it on the basis of constantly changing stimulus items, and execute the correct response based on the memory of the most recent one. A similar working memory process may be the basis of a rat’s performance in the Morris water maze (Morris, 1981) or radial arm maze (Olton, 1984), particularly when visual and/or olfactory cues are not available to guide the animal’s responses.
Cellular Correlate of Working Memory: Neurons with Memory Fields
A major advance in our understanding of prefrontal cortex came in the early seventies, when electrophysiological studies were performed for the first time in awake behaving monkeys trained on delayed-response tasks (Fuster and Alexander, 1971; Kubota and Niki, 1971). These studies revealed that neurons in the prefrontal cortex become activated during the delay period of a delayed-response trial, and suggested that the prefrontal neurons examined were the cellular correlate of a mnemonic event. The evidence for prefrontal neurons in mnemonic processing has been accumulating steadily over the past 25 years. Most recently, an oculomotor version of the classical delayed response paradigm has allowed more exacting analysis of prefrontal neurons under controlled conditions.