Multiple memory systems

From Scholarpedia
Norman M. White (2007), Scholarpedia, 2(7):2663. doi:10.4249/scholarpedia.2663 revision #62791 [link to/cite this article]
Jump to: navigation, search
Curator and Contributors

1.00 - Norman M. White

Contents

Multiple Parallel Memory Systems in the Brain

The multiple memory systems theory is based on evidence that different kinds of information are processed and stored in different parts of the brain. One version of this idea is illustrated in Figure 1.
New 1.png
Neural activity originating in external and internal receptors (Input) flows through several parallel brain systems (shown in yellow), each of which is specialized to extract a different kind of information from the ongoing activity. Each system has a central structure (shown in red) that performs its information processing functions independently of the others. Interactions among the systems occurs at the level of their inputs which come from many of the same sources, and their outputs which converge to produce thought and behavior.

One system, with the caudate nucleus as its central structure, represents constant stimulus-response (S-R) relationships that lead to successful outcomes (i.e., reinforcement such as food or escape from an aversive event). Information processed and stored in this system (called Procedural memory) tends to produce the response whenever the stimulus is encountered (often referred to as "habit learning"). A second system, with the amygdala as its central structure, represents relationships between neutral stimuli and rewarding or aversive emotional states (S-Af, or stimulus-affect associations). This form of memory is also known as Pavlovian, or Classical conditioning. The third system, with the hippocampus as its central structure, represents relationships among stimuli (S-S) and events, or pure cognitive information. This kind of stored declarative information can be used flexibly to generate different behaviors depending on immediate circumstances (or context). In humans, this system may support conscious recollection of previous events. In some situations the converging outputs of the systems may promote similar behaviors or complementary parts of complex behaviors. This is co-operation among the systems. In other situations the systems can promote different behaviors that compete with each other.

The idea that different kinds of information are stored in different parts of the brain originated with the observation that brain damage in humans often has specific effects on the kinds of memories that can be formed and recalled. In perhaps the best known case of this kind, patient HM underwent bilateral excision of a major portion of the temporal lobes of his brain (affecting the hippocampus and related structures) as a treatment for epilepsy. After his surgery, HM was unable to recall current experiences after a few minutes, and could remember only a few events of any kind that had occurred during his lifetime. He was unable to learn to find his way around in a new neighborhood and could not describe the layout of his apartment (Milner & Penfield, 1955; Milner, 1959). At the same time he learned to solve a simple maze (procedural S-R memory)(Milner et al., 1968) and was able to acquire a conditioned eye blink response (Woodruff-Pak, 1993)(a Pavlovian conditioned response); other patients with medial temporal lobe impairments have acquired normal conditioned fear responses (Bechara et al., 1995). These observations suggest that declarative memory depends on the medial temporal lobes, but that S-R and S-Af memories depend on other part(s) of the brain.

Dissociations in Rats

These three kinds of memory were explicitly dissociated by Packard, Hirsh & White (1989) and by McDonald & White (1993) using rats as subjects. The rats were tested using an 8-arm radial maze (consisting of an octagonal central platform with arms radiating from each side in a sunburst pattern, as illustrated in Figure 2).
Triple dissociation of memory systems in rats
The rats were hungry and were allowed to forage for food on the maze in three different conditions (shown in Figure 2). Each condition provided the rat with a different kind of information about the location of food on the maze. Normal rats learned to locate the food easily in all three conditions. However, the ability of rats to locate the food in each condition was severely impaired by specific lesions to a different one of three brain systems (the hippocampus, caudate nucleus (aka dorsal striatum in the rat) and amygdala systems). This finding led to the conclusion that each brain system was critical for processing one of the three kinds of information, but was not required for processing the other two kinds of information.

The Win-Shift task provides the rats with S-S information about the location of the food. In this task a single food pellet is placed at the end of each arm (red dots in Figure 2). A rat is placed on the center platform of the maze and allowed to move around freely. The rat must retrieve each pellet without re-entering arms it has already visited; this requires the rats to remember which arms it has visited as the trial progresses (sometimes called working memory). To do this, the rat must be able to identify the arms. The task is designed to eliminate local cues in the arms and force the rats to rely on visual cues in the maze environment to identify the arms. They do this by acquiring S-S information about the relationships of the arms to the environmental cues, a form of information called a spatial map (O'Keefe & Nadel, 1978). Each arm entry is determined by a flexible combination of the spatial map and working memory of arms already entered (note that this information changes after each arm entry). Performance on the win-shift task was impaired by lesions to the hippocampus or to the fimbria-fornix (a part of the hippocampal system), but not by lesions to the dorsal striatum or amygdala.

In the Win-Stay task lights at the entrance to each arm inform the rats about the location of the food. Each time a rat is placed on the maze a different set of four randomly selected arms is lit and only those arms contain food. When a rat obtains the pellet at the end of a lit arm a second pellet is placed there. When the second pellet in any arm is eaten the light on that arm is extinguished and no more pellets are placed there. Therefore, the rats obtain 8 pellets by entering each of 4 arms twice. No arm-identification, spatial learning or working memory are required to locate the food. The rat must simply acquire an S-R association in which the arm lights are the stimuli and entering the arm is the response. Every time this sequence is repeated it is followed by reinforcement (eating the food in the lit arm). Performance on the Win-Stay task was impaired by lesions of the dorsal striatum, but not by lesions of the fimbria-fornix or amygdala.

In the Conditioned Cue Preference (CCP) task a rat is confined at the end of an arm with a supply of food. Confining the rat to a small space and preventing it from moving around on the maze has two consequences. First, it prevents the rat from acquiring hippocampally mediated spatial information about the arm location. Second, the rat never learns to run into the arm from the center platform to obtain the food (which is S-R learning mediated by the dorsal striatum). However, the rat can acquire an association between the environmental stimuli visible from the arm and the rewarding consequences of eating the food. On alternate days the rats are confined to another arm on the opposite side of the maze with no food from which they view a different set of stimuli. When given a choice between the food-paired and unpaired arms with no food in either one normal rats chose to spend more time in the food-paired arm, demonstrating a preference for approaching and remaining in the vicinity of food-paired stimuli over responding to stimuli that have not been paired with food. This arm discrimination was impaired by lesions of the amygdala, but not by lesions of the fimbria-fornix or dorsal striatum.

In summary, acquisition of S-S information was impaired by lesions of the hippocampus-system (or by lesions of the hippocampus itself), acquisition of S-R information was impaired by lesions of the dorsal striatum, and acquisition of S-Af information was impaired by lesions of the amygdala. Each of the lesions affected the acquisition of only one kind of information.

The hippocampus, dorsal striatum and amygdala do not process or store information in isolation; each of these structures performs these functions together with numerous other afferent and efferent brain areas that constitute systems (this idea is illustrated in Figure 3). The triple dissociation suggests that the hippocampus, dorsal striatum and amygdala are the central structures in three information processing memory storage systems which may operate simultaneously and in parallel with at least some degree of independence.

Dissociations in Humans

Research in humans using a variety of perceptual, motor, and cognitive learning tasks has provided evidence of dissociations between different forms of memory that are parallel to the findings with rats. In a study of perceptual skill learning, people were trained to read words presented in mirror-reversed text and their reading speed was measured (Cohen & Squire, 1980). Normal participants improved their mirror reading skill with practice, and when tested afterwards were able to recognize correctly which words they had seen during training. People who were amnesic (due to Korsakoff's syndrome, thalamic damage, or electroconvulsive therapy) showed normal learning of the skill; however, they were impaired at recognizing the words that they had seen during training. The opposite pattern was observed in patients with Huntington's disease, which damages the striatum (or caudate nucleus); these patients were impaired at learning the mirror-reading skill but had no problem remembering the words (Martone et al., 1984). Similar dissociations have been observed in motor skill learning (Heindel et al., 1989). These findings dissociate declarative and procedural memory in humans, suggesting they have the similar anatomical substrates as in rats.

Other studies have examined classification learning using a “weather prediction" task. Participants were trained to classify stimuli (sets of shapes) into one of two categories by making different responses for “rain” or “sunshine” based on trial-by-trial feedback or reinforcement, consisting of information about whether or not their response was “correct”). To make cognitive learning of the correct predictions difficult the feedback was inconsistent (or probabilistic) while still providing sufficient repeated reinforcement of the correct response to produce S-R learning. Initial studies of amnesic patients with damage to the medial temporal lobe (including the hippocampus) showed they were able to learn the classification task as well as normal controls (Knowlton et al., 1996), though more recent studies with younger amnesiacs have qualified this conclusion (Hopkins et al., 2004). Patients with Parkinson’s disease (which disrupts the dopamine system and affects the striatum) were unable to learn the classification task, but recalled details of the task normally (Knowlton et al., 1996). These findings constitute another double dissociation of declarative and procedural memory and their anatomical substrates, the hippocampus and caudate nucleus, in humans.

When normal participants learned the weather prediction task while undergoing fMRI imaging of brain activity, activation of the caudate nucleus increased while activation of the hippocampus decreased (Poldrack et al., 2001). When the feedback learning version of the task was compared to a version in which subjects learned by observation of the correct pairings with no feedback, the hippocampus was more active and the caudate nucleus was less active. These results suggested that the caudate may play a particular role in the processing of reinforced S-R (feedback) learning, and subsequent studies have confirmed this by showing that patients with Parkinson's disease are impaired at learning by feedback but normal at learning the same information by observation (Shohamy et al., 2004).

In another study, a patient with Urbach-Weithe disease, which primarily affects the amygdala, and a patient with Alzheimer’s disease were compared on a Pavlovian conditioning task. The patients were shown a series of colored slides; one color was followed by an electric shock, which produces an unconditioned change in skin conductance. Subsequent presentation of the colored slide that appeared just before the shock to a normal participant produced a conditioned increased skin conductance in the absence of the shock (Pavlovian conditioning, or S-Af learning). This conditioned response did not occur in the Urbach-Weithe patient; however he was able to recall details of the experimental situation. The Alzheimer’s patient acquired a normal conditioned response to the colored slide, but was unable to recall details of the experimental situation (Bechara et al., 1995). These observations dissociate hippocampus-based declarative memory from amygdala-based Pavlovian conditioning, corresponding to the dissociation of these two kinds of information processing in rats.

Other similar dissociations of processing/memory for different kinds of information in both humans and rats have been reported.

Co-Operation and Competition Among Memory Systems

As shown in Figure 2, performance of the Win-Stay task was impaired by lesions of the caudate nucleus but was significantly improved compared to the performance of normal rats by lesions of the fimbria fornix. This improvement in performance produced by the fimbria-fornix lesions is attributed to competition between the outputs of the dorsal striatum and hippocampal memory systems. As suggested by the impaired win-stay performance of the rats with caudate lesions, the caudate system is critical for learning to respond to the lights. At the same time, the hippocampus was independently acquiring information about the spatial location of the food, as suggested by the effect of lesions to this system on performance of the win-shift task. However, most or all of the spatial information about the location of the food acquired on any given trial was wrong on the next trial because the lights and food were moved to different randomly selected arms on every trial. Therefore, in normal rats information acquired by the hippocampus system on one trial would tend to produce errors on the next trial. This tendency would compete with and interfere with the tendency to enter lit arms promoted by output from the caudate system. Eliminating hippocampus-based learning with lesions of the fimbria-fornix eliminated this source of errors and improved performance on the task. It has also been shown that putting a curtain around the maze during win-stay training improves performance because it attenuates the spatial cues, reducing their influence on the rats’ behavior.

A number of other examples of competition among memory systems have been reported, both in rodents and in humans (Poldrack & Packard, 2003). Figure 3 is a diagram of how these competitive interactions may occur.
Memory system inputs, outputs and interactions
According to this scheme information about the external world is transduced into neural activity by exteroceptors and transmitted to the directly and via the thalamus to sensory areas of the cerebral cortex. These cortical areas transmit information to the central structures of all three memory systems. The systems also receive information from interoceptors about deprivation states, rewarding and aversive states, and internal states produced by hormones and drugs. The internal microstructure of each system is specialized to represent certain kinds of relationships among these inputs. These informational relationships are instantiated by S-S, S-R and S-Af associations.

The relational information in any situation is likely to correspond most closely to the information processing specialization of one of the systems, so that system is likely to provide the most coherent representation of the situation. In turn the system with the most coherent representation of a situation is likely to store that representation. When the situation is encountered on a future occasion, the stored representation is activated and conditions the information arriving at the system, producing a learned behavior (i.e., one that is influenced by the organism’s previous experience in the situation).

As shown in Figure 3, the outputs of the systems converge on high level motor output structures. The system with the most coherent representation produces the strongest output and wins the competition for control of behavior. However, as discussed in the case of competition between spatial and S-R learning in the win-stay task, another system with a less coherent representation of the situation can also produce output that is sufficiently strong to influence behavior, often to the detriment of optimal responding in the situation (White & McDonald, 2002).

The model operates on the assumption that there is no direct transfer of information among the systems. However, as shown in Figure 3, all of the systems have outputs that reach the cortex, via either the pre-frontal cortex or the anterior thalamus. This output could affect the cortex, conditioning the way it represents incoming information on future occasions. Since all of the systems receive information from the same parts of the cortex, it is possible for output from one system to condition information that later reaches another system. In this way the systems could influence each other via the cortex without exchanging information directly.

References

Bechara, A., Tranel, D., Damasio, H., Adolphs, R., Rockland, C., & Damasio, A. R. (1995). Double dissociation of conditioning and declarative knowledge relative to the amygdala and hippocampus in humans. Science, 269, 1115-1118.

Cohen, N. J. & Squire, L. R. (1980). Preserved learning and retention of pattern-analyzing skill in amnesia: dissociation of knowing how and knowing what. Science, 210, 207-209.

Heindel, W. C., Salmon, D. P., Shults, C. W., Walicke, P. A., & Butters, N. (1989). Neuropsychological evidence for multiple implicit memory systems: A comparison of Alzheimer's, Huntington's, and Parkinson's disease patients. Journal of Neuroscience, 9, 582-587.

Hopkins, R. O., Myers, C. E., Shohamy, D., Grossman, S., & Gluck, M. (2004). Impaired probabilistic category learning in hypoxic subjects with hippocampal damage. Neuropsychologia, 42, 524-535.

Knowlton, B. J., Mangels, J. A., & Squire, L. R. (1996). A neostriatal habit learning system in humans. Science, 273, 1399-1402.

Martone, M., Butters, N., Payne, M., Becker, J., & Sax, D. S. (1984). Dissociations between skill learning and verbal recognition in amnesia and dementia. Archives of Neurology, 41, 965-970.

McDonald, R. J. & White, N. M. (1993). A triple dissociation of memory systems: hippocampus, amygdala and dorsal striatum. Behavioral Neuroscience, 107, 3-22.

Milner, B. (1959). The memory defect in bilateral hippocampal lesions. Psychiatr.Res.Rep.Am.Psychiatr.Assoc., 11, 43-58.

Milner, B., Corkin, S., & Teuber, H.-L. (1968). Further analysis of the hippocampal amnesic syndrome: 14-year follow-up study of H.M. Neuropsychologia, 6, 215-234.

Milner, B. & Penfield, W. (1955). The effect of hippocampal lesions on recent memory. Trans.Am.Neurol.Assoc., 42-48.

O'Keefe, J. & Nadel, L. (1978). The Hippocampus as a Cognitive Map. Oxford: Oxford University Press.

Packard, M. G., Hirsh, R., & White, N. M. (1989). Differential effects of fornix and caudate nucleus lesions on two radial maze tasks: evidence for multiple memory systems. Journal of Neuroscience, 9, 1465-1472.

Poldrack, R. A., Clark, J., Pare-Blagoev, E. J., Shohamy, D., Creso, M. J., Myers, C. et al. (2001). Interactive memory systems in the human brain. Nature, 414, 546-550.

Poldrack, R. A. & Packard, M. G. (2003). Competition among multiple memory systems: converging evidence from animal and human brain studies. Neuropsychologia, 41, 245-251.

Shohamy, D., Myers, C. E., Onlaor, S., & Gluck, M. A. (2004). Role of the basal ganglia in category learning: how do patients with Parkinson's disease learn? Behavioral Neuroscience, 118, 676-686.

White, N. M. & McDonald, R. J. (2002). Multiple parallel memory systems in the brain of the rat. Neurobiology of Learning & Memory, 77, 125-184.

Woodruff-Pak, D. S. (1993). Eyeblink classical conditioning in H.M.: Delay and trace paradigms. Behavioral Neuroscience, 107, 911-925.

Internal references

See Also

Brain, Classical Conditioning, Cortical Memory, Instrumental Conditioning,Learning, Memory, Synaptic Plasticity, Working Memory

Personal tools
Namespaces

Variants
Actions
Navigation
Focal areas
Activity
Tools