Associative Learning and the Hippocampus
By Wendy A. Suzuki, PhD
Each day we learn and remember a myriad of new information from the names of new people we meet to the best dish to order at a particular restaurant to the location of a new bike path. This novel information is first acquired, and if strengthened through a process called consolidation, is eventually stored in long-term memory. The structures of the medial temporal lobe are essential for this ability to acquire new long-term memories for facts and events. This form of memory is called declarative memory in humans and relational memory in animals. Convergent findings from neuropsychological studies in humans, together with experimental lesion studies and neuroanatomical studies in animals have shown that the key medial temporal lobe structures important for declarative/relational memory include the hippocampus together with the surrounding entorhinal, perirhinal and parahippocampal cortices. While it is clear that the structures of the medial temporal lobe are essential for the acquisition of new declarative/relational memories, we still have only a rudimentary understanding of the normal patterns of neural activity that underlie this ability. To start to address this question, my laboratory has recorded the activity of individual neurons in the hippocampus as monkeys perform various memory demanding tasks. We have focused on one particular form of declarative/relational memory, called associative memory, which is defined as the ability to learn and remember the relationship between unrelated items such as the name of someone we have just met or the aroma of a particular perfume. Specifically, we examined the patterns of activity in hippocampal neurons as monkeys are in the process of forming new associations in memory. We hypothesized that if the hippocampus is important for the early formation of the new associations, we should see changes in neural activity that parallel behavioral learning. Our long-term goal is to understand the evolution of learning-related signals throughout the medial temporal lobe as memories are initially established, strengthened and eventually stored in long-term memory.
A Task Requiring New Associative Leaning
To examine the patterns of neural activity during associative memory formation, we trained two monkeys to perform a location-scene association task. In this task, animals were required to learn new associations between particular complex visual "scenes" and particular rewarded target locations. We know that the medial temporal lobe participates importantly in the normal performance of this task since damage to this region in monkeys produces significant impairment in the ability to learn new location-scene associations (Brasted et al., 2003; Brasted et al., 2002; Murray et al., 2000; Wise and Murray, 1999; Murray and Wise, 1996; Rupniak and Gaffan, 1987). A schematic representation of the task is shown in Figure 1. On each trial, monkeys are first shown 4 identical target stimuli superimposed on a complex visual scene (typically a picture of a real outdoor scene). Following a delay interval, during which the scene disappears but the targets remain on the screen, the animal is cued to make a single eye movement to one of the four peripheral targets on the screen. For each visual scene, only one of the 4 targets is associated with a juice reward. Each day, the animals learned 2-4 new location-scene associations by trial and error. The new location-scene associations were randomly intermixed with well-learned "reference" associations that the animals had seen for many months before the recording experiments began. Each of the 4 reference scenes was associated with a different rewarded target location (i.e., north, south, east or west). Responses to the reference scenes were used to control for possible motor-related activity in hippocampal cells.
Patterns of Hippocampal Activity During Learning of New Location-scene Associations
In our initial studies, we focused on the hippocampus, a medial temporal lobe structure long implicated in associative learning and memory (Eichenbaum and Cohen, 2001; Squire and Zola, 1996; Scoville and Milner, 1957). We first asked if hippocampal cells responded differentially to the different visual scenes used in the task. We found that 61% of the isolated hippocampal cells were engaged in the task in that they responded differentially to the different scenes (i.e., visually selective response). Moreover, consistent with our working hypothesis, we found that 28% of the selectively responding cells (18% of the entire population of hippocampal cells recorded) exhibited changes in neural activity across trials that were significantly correlated with the animal's behavioral learning curve for a particular scene. We called these cells "changing cells". Two categories of changing cells were observed. Sustained changing cells (54% of the population of changing cells) signaled learning with a change in neural activity that was maintained for as long as we were able to hold the cell (typically 30 min to 1 hour). Many of these cells exhibited dramatic increases in neural activity that paralleled the animal's behavioral learning curve for that association (Figure 2A). Importantly, these learning signals were highly selective in that a changing cell would typically only change its activity for one particular learned scene while the responses to other learned scenes did not change over time. One interpretation of the findings illustrated in Figure 2A is that the changing neural activity is related to learning. However, another possible interpretation is that this activity is related to learning a particular motor response (i.e., learning to respond to the north). According the this interpretation, the early correct trials may not have elicited much activity if the movements were made tentatively, but strong motor-related activity is observed once the animal starts responding consistently to the north. If this motor-based interpretation were correct, we would expect to see similar levels of activity from this cell in response to the reference scene with same north rewarded target location. This was never the case. In fact, changing cells typically responded with little or no activity to the reference scene with the same rewarded target location. These findings support the idea that the changing activity is related to learning of a new association between a scene and a target/eye movement and not learning of a particular motor response. Also consistent with this idea are findings from other control experiments in which the animals learned 2 consecutive sets of novel location-scene associations. We found that changing cells identified in the first set of learned location-scene associations never signaled learning of a second novel set of location-scene association even when the rewarded target location was the same (i.e., north target rewarded).
The remaining 45% of changing cells exhibited a different pattern of learning-related activity. These changing cells started out with a scene-selective response during either the scene or delay period of the task early in the session well before the animal learned the association. These cells signaled learning by returning to baseline activity and this return to baseline was typically anti-correlated with the animal's learning curve for that particular scene (Figure 2B). We called these cells baseline sustained changing cells. Importantly, the changes in neural activity seen in the baseline sustained cells were as selective for a particular learned scene as the sustained changing cells. Similar patterns of activity were never seen for the reference scene with the corresponding rewarded target location suggesting that these signals were not motor-driven.
Thus, both sustained changing cells and baseline sustained changing cells provide a highly selective signal for when a particular scene is learned. We hypothesize that these selective increases and decreases in neural activity that occur across the hippocampal population may constitute a hippocampal network learning signal. It will be important to examine the learning signals across larger numbers of hippocampal cells recorded simultaneously to better understand the interactions between cells during learning.
Timing of Hippocampal Activity During Learning of New Location-scene Associations
A critical issue for any study examining the neural correlates of behavior is defining the causal relationship between the patterns of neural activity observed and the behavioral output. Is the observed neural activity driving behavior or it is occurring downstream of the critical sites of origin? A variety of different approaches have been used to address this issue. For example, as mentioned above, lesions studies can implicate a particular brain area in the normal performance of a task, though lesion studies alone cannot determine the patterns of neural activity that underlie this function. Electrical stimulation studies have been used to test the effect of direct stimulation on animal's choice in sensory discrimination problems (Salzman et al., 1992; Salzman et al., 1990), but this approach has not been used in studies of hippocampal function. A third method that has been used to probe the relationship between neural activity and behavior is to examine the precise timing of the changes in neural activity relative to behavioral learning. We hypothesize that those selective neural changes that occur before behavioral learning is expressed may be involved in driving behavioral change while those that occur after behavioral learning may play a role in strengthening the newly formed association. To address this question, for all changing cells, we calculated trial number of learning and compared it to the calculated trial number of neural change (See Wirth et al., 2003 for detailed description of the behavioral and neural algorithms used). We found that hippocampal changing cells could both precede (14 examples) parallel (4 example) and lag (19 examples) behavioral learning (Figure 3). These finding suggest that the hippocampus participate in all stages of the learning process from several trials before behavioral learning is expressed, when the observed activity may be involved in driving the behavioral changes that underlie learning to several trials after learning, when the activity may be involved in strengthening the newly formed association.
Summary and Future Directions
We showed that cells in the hippocampus provide strong learning-related patterns of neural activity that participate in the initial formation of new associative memories. Because these changes can occur before, at the same time, or after learning, these findings suggest that there may be a gradual recruitment of a network of hippocampal neurons during the formation of new associative memories. Previous studies have shown that in addition to the hippocampus, cells in several other brain areas including the prefrontal cortex (Asaad et al., 1998), frontal motor-related areas (Brasted and Wise, 2004; Chen and Wise, 1995a; Chen and Wise, 1995b; Mitz et al., 1991) and striatum (Brasted and Wise, 2004) exhibit similar patterns of learning-related activity during similar associative learning tasks. An important long-term goal will be to understand how all these brain areas from the hippocampus to the motor related areas of the frontal lobe and striatum may work together to underlie the initial formation as well as the early strengthening and consolidation of new associative learning.
Asaad W. F., Rainer G., Miller EK (1998) Neural activity in the primate prefrontal cortex during associative learning. Neuron 21: 1399-1407.
Brasted P. J., Bussey TJ, Murray EA, Wise SP (2002) Fornix transection impairs conditional visuomotor learning in tasks involving nonspatially differentiated responses. J Neurophysiol 87: 631-633.
Brasted P. J., Bussey TJ, Murray EA, Wise SP (2003) Role of the hippocampal system in associative learning beyond the spatial domain. Brain 126: 1202-1223.
Brasted P. J., Wise S. P. (2004) Comparison of learning-related neuronal activity in the dorsal premotor cortex and striatum. Eur J Neurosci 19: 721-740.
Chen L. L., Wise S. P. (1995a) Neuronal activity in the supplementary eye field during acquisition of conditional oculomotor associations. J Neurophys 73: 1101-1121.
Chen L. L., Wise S. P. (1995b) Supplementary eye field contrasted with the frontal eye field during acquisition of conditional oculomotor associations. J Neurophys 73: 1122-1134.
Eichenbaum H., Cohen N. J. (2001) From Conditioning to Conscious Recollection. New York: Oxford University Press.
Mitz A. R., Godschalk M., Wise S. P. (1991) Learning-dependent neuronal activity in the premotor cortex: Activity during the acquisition of conditional motor associations. J Neurosci 11: 1855-1872.
Murray E. A., Bussey T. J., Wise S. P. (2000) Role of prefrontal cortex in a network for arbitrary visuomotor mapping. Exp Br Res 133: 114-129.
Murray E. A., Wise S. P. (1996) Role of the hippocampus plus subjacent cortex but not amygdala in visuomotor conditional learning in rhesus monkeys. Behav Neurosci 110: 1261-1270.
Rupniak N. M., Gaffan D. (1987) Monkey hippocampus and learning about spatially directed movements. J Neurosci 7: 2331-2337.
Salzman C. D., Britten K. H., Newsome W. T. (1990) Cortical microstimulation influences perceptual judgements of motion direction. Nature 346: 174-177.
Salzman C. D., Murasugi C. M., Britten K. H., Newsome W. T. (1992) Microstimulation in visual area MT: effects on direction discrimination performance. J Neurosci 12: 2331-2355.
Scoville W. B., Milner B. (1957) Loss of recent memory after bilateral hippocampal lesions. J Neurol Neurosurg Psych 20: 11-21.
Squire L. R., Zola S. M. (1996) Structure and function of declarative and nondeclarative memory systems. Proc Natl Acad Sci 93: 13515-13522.
Wirth S., Yanike M., Frank L. M., Smith A. C., Brown E. N., Suzuki W. A. (2003) Single neurons in the monkey hippocampus and learning of new associations. Science 300: 1578-1581.
Wise S. P., Murray E. A. (1999) Role of the hippocampal system in conditional motor learning: Mapping antecedents to action. Hippocampus 9: 101-117.
About the Author
Wendy A. Suzuki earned her PhD in Neuroscience from the University of California, San Diego in 1993 and did her post-doctoral work in the Laboratory of Neuropsychology at the National Institute of Mental Health. She is currently an Associate Professor of Neural Science and Psychology at New York University. Her research focuses on understanding the brain basis of memory. To address this question, she uses behavioral neurophysiology, recording the activity of individual neurons as monkeys perform various memory demanding tasks. Her recent work has focused on how cells in the hippocampus signal both the initial formation as well the long-term representation of associative memory. Suzuki is the recipient of the Lindsley Prize in Behavioral Neuroscience (1994) and the Troland award in Experimental Psychology (2004).