Rats remember items in context using episodic memory

Summary

Vivid episodic memories in people have been characterized as the replay of unique events in sequential order [1–3]. Animal models of episodic memory have successfully documented episodic memory of a single event (e.g., [4–8]). However, a fundamental feature of episodic memory in people is that it involves multiple events, and notably, episodic-memory impairments in human diseases are not limited to a single event. Critically, it is not known whether animals remember many unique events using episodic memory. Here we show that rats remember many unique events and the contexts in which the events occurred using episodic memory. We used an olfactory memory assessment in which new (but not old) odors were rewarded using 32 items. Rats were presented with 16 odors in one context and the same odors in a second context. To attain high accuracy, the rats needed to remember item in context because each odor was rewarded as a new item in each context. The demands on item-in-context memory were varied by assessing memory with 2, 3, 5, or 15 unpredictable transitions between contexts, and item-in-context memory survived a 45-min retention-interval challenge. When the memory of item in context was put in conflict with non-episodic familiarity cues, rats relied on item in context using episodic memory. Our findings suggest that rats remember multiple unique events and the contexts in which these events occurred using episodic memory and support the view that rats may be used to model fundamental aspects of human cognition.

Graphical Abstract

Results and Discussion

A fundamental attribute in human cognition is the ability to remember multiple unique events using episodic memory [1, 9, 10]. In an everyday example, imagine you are watching an episode of your favorite television show when you see actors that look vaguely familiar make cameo appearances. You know that you have seen the actors before in another show or movie, but you cannot remember where. This common scenario illustrates two types of memory situations; one involves episodic based remembering of items in context, in this case, the cameo actors and the movies or series you originally saw them in, whereas the other involves a vague sense of familiarity of the actor but without memory of the episode or context in which they occurred. Episodic memory involves remembering an event and the contextual details of the episode, whereas familiarity is the somewhat vague judgment that an item is known [3, 11–14], as highlighted in the example above.

Importantly, disorders of episodic memory (such as Alzheimer’s disease) are often not limited to the loss of a single event, but rather, present with the widespread impairment of multiple episodic memories which makes the disorder debilitating [15–18]. Increasingly, animal models of human memory are developed to gain insight and understanding of basic biological mechanisms of memory and to validate therapeutic approaches to treat memory diseases that are observed in humans [19, 20]. Prior work on animal models of episodic memory has been successful at documenting that animals are capable of remembering a single event [4–8]. Although it is possible for single events to contain multiple features, it is currently unknown whether animals can remember many episodic memories. Animal models of episodic memory have used a variety of animals, such as scrub jays [4], rats [5–8], and non-human primates [21]. How to define episodic memory in animal models has been a subject of debate [22, 23], but a major view about episodic memory is that it involves memory of an event and the contextual details of the episodic (i.e., item in context information) [13]. Our approach focuses on the objective content of episodic memory [19, 24–27] rather than subjective experiences that are thought to accompany episodic memory in people, while carefully eliminating non-episodic memory alternatives. In the current study, we asked rats to identify specific items and the contexts in which the items were encountered for multiple unique events (i.e., item-in-context memory).

We exploited the well-established proficiency of rats with olfactory information [28, 29]. Rats were individually tested in two distinctive circular arenas with “food holes” covered by scented opaque lids (see Supplemental Experimental Procedures). In our approach, rats were trained to pick the “new” odor and avoid the “old” odor when presented with pairs of odors across multiple contexts. Specifically, one odor was always “new” (S+) to the current context while the other was “old” (S−; i.e., it had already been presented within that context earlier in the day). A new odor item was presented to individual rats using a scented lid placed on top of a small cup, together with an old odor placed on a different cup (at randomly selected locations in the arenas). We defined a choice as displacement of a lid. Selection of a new item was rewarded with food, whereas selection of the old item was not. In preliminary training, rats were presented with multiple odors across two contexts in rapid succession (see Figure S1 and Supplemental Experimental Procedures); to make the contexts distinctive, we used two circular arenas that differed in diameter, black-and-white patterns on the floor, number of holes on the floor, and access to visual cues outside the arena. Initially, we presented the rats with a set of 16 odors in the first context (Context A). Next, in the second context (Context B), the same 16 odors were presented as “new”, despite the earlier presentation of these odors in the other context. Training accuracy was high (Table S1). However, because the odors are presented in each context sequentially, high accuracy could be attained by selecting the least familiar item (i.e., by using a semantic rule [30], namely “avoid familiar items”), without using episodic memory of items in context.

To investigate memory of item in context, we rapidly interleaved presentations of new odors across contexts by presenting an additional transition (Context A→Context B→Context A; Figure 1; Experiment 1). Initially, rats were presented with half of the items (i.e., 8 odors) in the first context, followed by all 16 items in a second context (i.e., Context B presentation included items that had and had not been presented in Context A). Finally, the rats returned to the first context where they received the remaining half of the odors (i.e., the 8 items that had not yet appeared in the initial context). Each day, 16 odors were randomly selected from a pool of 40 odors, and odors were presented in random order. Figure 2 illustrates how small changes in the order of items provide insights into the type of memory that may support selection of the “new” item. In Figure 2A, the animal could find the “new” item by selecting the least familiar odor, without using episodic memory. To illustrate how familiarity cues can promote selection of the “new” item, consider a snapshot of four items presented across two contexts (Figure 2A). For a particular pair of odors (e.g., banana and basil, depicted as yellow and green in Figure 2A), one item (banana) but not the other (basil) occurs in the first context. Next, both items occur in the second context (notably basil followed by banana). Finally, upon return to the first context, a choice between banana and basil is offered; basil is rewarded because it is “new” to Context A (i.e., the item-in-context correct choice). Because banana was rewarded after basil in the second context in Figure 2A, when the choice occurs, banana would be more familiar than basil. Thus, an animal that used a semantic rule (“avoid familiar items”) would successfully choose the basil in the final context (i.e., the correct item-in-context choice in Figure 2A) based on judgments of relative familiarity. Figure 2A illustrates that in some odor sequences, a correct choice of the “new” item could be based on familiarity or remembering item-in-context. Although accuracy in selecting the new item is high (Table S1) when both familiarity and item-in-context cues are available, these data do not isolate item-in-context memory while eliminating the use of familiarity cues. Importantly, a small change in the sequence of odors unconfounds these two alternatives, as described next.

Figure 1. Schematic of odor sequence presented across multiple contexts.

Schematic illustrating the sequence of new odors (depicted as colors) in Experiment 1 with 2 context transitions. The “new” item (i.e., odor) in each trial is rewarded (S+, denoted by “√”), whereas the “old” item is not rewarded (S−). In the first segment, the rats were presented with half of the items in Context A. Next, the rats were presented with all of the odors in Context B, including those already presented in Context A. Finally, in Context A, the rats were presented with the remaining half of the items that had not yet been presented in Context A. Six of 16 S+ items in each context are shown. See also Figure S1 and Table S1.

Figure 2. Dissociating episodic item-in-context memory from familiarity cues.

(A) Familiarity and item-in-context memories are confounded in (A). Banana and basil odors are depicted here as yellow and green, respectively. Initially, we presented banana in Context A, and we presented both basil and banana in Context B. Notably, basil was not presented in Context A, and, importantly, basil occurred before banana in Context B. Finally, the memory test occurred in Context A. In the memory test, the rats were presented with a choice between banana and basil in Context A. Critically, both familiarity and item-in-context memories would lead an animal to choose basil, the correct choice in the memory test. A small change in the order of items in Context B unconfounds these two alternatives, as shown in (B).

(B) Familiarity and item-in-context memories are dissociated in (B). Strawberry and blueberry odors are depicted here as red and blue, respectively. Initially, we presented strawberry in Context A, and we presented both strawberry and blueberry in Context B. Notably, blueberry was not presented in Context A, and strawberry occurred before blueberry in Context B. Finally, the memory assessment occurred in Context A. In the memory assessment, the rats were presented with a choice between strawberry and blueberry. The correct choice, based on item in context, is blueberry because it has not yet been presented in Context A. Blueberry is rewarded when chosen in this test and our measure of accuracy is the proportion of choices of the rewarded item. Importantly, prior to the memory assessment, blueberry was presented more recently than strawberry. Consequently, in the memory assessment, strawberry would be less familiar relative to blueberry. Thus, an animal that relied on judgments of relative familiarity would choose the strawberry in the memory assessment. By our measure of accuracy, this choice results in accuracy below chance. By contrast, an animal that relied on item-in-context memory would choose blueberry in the memory assessment, which results in above chance accuracy. Notably, this memory assessment dissociates item-in-context memory (above chance) from judgments of relative familiarity (below chance).

(A and B) The presence of additional odors (not shown) is identified by “…” in the schematic. The schematic focuses on S+ items (denoted by “√”) by omitting comparison S− items prior to the memory assessment. Trials depicted in (A) and (B) were randomly intermingled throughout daily testing.

To dissociate episodic memory from familiarity judgments, in the last segment, we identified sequences of odors that put familiarity cues and item-in-context memory in conflict. For a particular pair of odors (e.g., strawberry and blueberry, depicted as red and blue in Figure 2B), we presented one item (strawberry) but not the other (blueberry) in the first context. Next, both items were presented in the second context (notably strawberry followed by blueberry). Finally, the memory assessment occurred upon return to the first context. In the memory assessment, the rats were confronted with a choice between strawberry and blueberry. The correct choice, based on item in context, is blueberry because it has not yet been presented in the first context; indeed, blueberry is rewarded when chosen in this test and our measure of accuracy is the proportion of choices of the rewarded item. Importantly, prior to the memory assessment, blueberry was presented more recently than strawberry (see Figure 2B). Consequently, in the memory assessment, strawberry would be less familiar relative to blueberry. Thus, an animal that relied on judgments of relative familiarity would choose the strawberry in the memory assessment (i.e., following the semantic rule “avoid familiar items”). By our measure of accuracy, such a choice would result in accuracy below chance in the memory assessment shown in Figure 2B. On the other hand, an animal that relied on item-in-context memory would choose blueberry in the memory assessment, which would in turn result in above chance accuracy. Notably, this memory assessment dissociates item-in-context memory (above chance) from judgments of relative familiarity (below chance). Because the arrangement shown in Figure 2B dissociates item-in-context memory and judgments of relative familiarity, we used the memory assessment shown in Figure 2B for each experiment (data shown in Figure 3), and excluded the arrangement shown in Figure 2A.

Figure 3. Rats remember item-in-context using episodic memory.

(A) Item-in-context memory is shown by above chance accuracy following 2, 3, 5, and 15 context transitions. Initial performance comes from the first two memory assessments in each experiment; subsequent memory-assessment performance is labeled terminal. See also Table S1.

(B) Item-in-context memory survives a long retention-interval challenge.

Data are shown as mean +1 SEM.

To test whether the rats were relying on item-in-context episodic memory or non-episodic judgments of familiarity, we examined the rats’ accuracy (Figure 3A) in the memory assessment (Figure 2B). Notably, the most compelling evidence for our dissociation of memory lies within the rats’ initial memory-assessment performance when these conditions are novel (i.e., before receiving feedback from rewards in the novel condition). Consequently, the critical data for evaluating item-in-context memory comes from the initial trials (labeled Initial in Figure 3A). If rats were relying on item-in-context episodic memory, performance in the initial memory assessment will be above chance, whereas using familiarity cues would produce below chance performance. When the identity of items in context was put in conflict with familiarity cues, initial performance was above chance (t(11)=7.71, p<0.001; Figure 3A; Experiment 1). Similar to the initial performance, high accuracy was observed in subsequent memory assessments (labeled Terminal in Figure 3A). High accuracy provides compelling evidence that rats relied on episodic item-in-context memory rather than judgments of familiarity.

Our data suggest that, after performing in the second context, the rats remembered the items presented in the first context. To establish that rats can also remember items from the second context, we added an additional context transition. We divided the day into four segments by using three context transitions (Context A→Context B→Context A→Context B). In sequence, we presented the rat with half of the items in each context during the first two segments. For the third and fourth segments, the rat returned to the first and second context, respectively, and was presented with the remaining half of the items in each. Therefore, in order to attain high accuracy in the fourth-segment memory assessment, the rat must be able to identify which items had and had not previously appeared in each context. Further, this approach allowed us to recreate novel conditions, which again, allowed us to examine performance, prior to the opportunity for new learning. When the identity of item in context was put in conflict with familiarity cues in the novel fourth segment, initial performance in the fourth segment was above chance (t(11)=10.65, p<0.001; Figure 3A; Experiment 2). Similarly, high accuracy was observed in subsequent memory assessments (terminal performance). These data suggest that the rats remember items in both contexts.

We recreated novel conditions again by adding two new transitions. Days were divided into six segments (Context A→Context B→Context A→Context B→Context A→Context B), which produced novel transitions in the last two segments. When we recreated novel conditions that included not one, but two new context transitions, initial performance was above chance (t(10)=8.86, p<0.001; Figure 3A; Experiment 3) and high accuracy was also observed in subsequent memory assessments (terminal performance).

Next, we challenged the rats’ memory further by randomly arranging memory assessments that were maximally unpredictable from trial to trial (i.e., by randomly determining whether a context transition would occur after each trial). The novelty of our conditions was enhanced because context transitions were unpredictable item to item and day to day. Under these conditions, it is not possible to use previous transitions to predict the next transition. On average, 15 unpredictable context transitions occurred each day. When item in context was put in conflict with familiarity cues using unpredictable transitions, initial memory-assessment performance was above chance (t(8)=10.0, p<0.001; Figure 3A; Experiment 4) and high accuracy was also observed in subsequent memory assessments. These data suggest that the rats relied on item-in-context memory when the change in contexts was unpredictable.

Episodic memory is posited to be a part of long-term memory [31]. To test the hypothesis that rats were relying on long-term episodic memory in our approach, we asked if the rats’ performance could survive a 45-min retention-interval challenge inserted between memory encoding and memory assessment. We used three context transitions (as in Experiment 2) and inserted the 45-min delay between the third and fourth segments; the data for the 0-min delay come from Experiment 2. In memory assessments after 0- and 45-min retention intervals, performance was above chance (t(11)=9.4, p<0.001 and t(8)=3.6, p<0.01, respectively; Figure 3B; Experiment 5) with no significant decline in performance as a function of delay (t(8)=−1.3, p=0.24). These data suggest that the rats were able to remember item in context following a long-retention interval challenge, and are consistent with the hypothesis that rats relied on long-term episodic memory.

It is noteworthy that in our memory assessment (Figure 2B), the familiarity choice was rewarded twice (red in Figure 2B) and the item-in-context choice was rewarded once (blue in Figure 2B). This observation raises an important concern, namely that the rats may have chosen the correct item in our memory assessment (blue in Figure 2B) because the incorrect item had been rewarded twice whereas the correct item was only rewarded once. Alternatively, the number of rewarded presentations may not impact accuracy. To address this issue, we compared performance on other occasions, when the item was rewarded in one or two contexts (see Figure 4 and Supplemental Experimental Procedures). In this analysis, choice between two items (depicted as orange and purple in Figure 4) was compared after a single rewarded presentation of the incorrect item (orange, Figure 4A) vs. two rewarded presentations of the incorrect item (Figure 4B). The addition of a rewarded presentation did not significantly impact the accuracy in selecting the correct item (t(12)=1.45, p=0.17). Because the absence of evidence (from the traditional null hypothesis significance test) is not necessarily evidence for the absence of a reward-frequency effect, we used a Bayesian statistical approach. Bayesian statistics can be used to “prove the null hypothesis” [32, 33], which in this case corresponds to the hypothesis that equivalent performance occurs when the number of rewarded presentations is varied. The JZS Bayes factor is 4.0; that is, the null hypothesis is 4 times more likely than the hypothesis that reward frequency impacts performance. A Bayes factor of this size is described as substantial evidence that the null hypothesis is correct [32]. Thus, these data provide substantial evidence that our memory-assessment performance (Figure 3) is not influenced by the number of rewarded presentations.

Figure 4. Schematic of odor sequences to compare the impact of rewarding an item once or twice.

An item (orange) is rewarded once (A) or twice (B) prior to a common memory test (choice between orange and purple in Context A). The other item (purple) is rewarded once in Context B. The second presentation of orange occurs after the memory test in (A). The trials depicted in (A) and (B) come from Experiment 2 (Contexts A→B→A→B).

Four lines of evidence suggest that rats remember multiple items in context using episodic memory. First, rats remember at least 32 items in context. Second, episodic-memory performance can sustain at least 15 transitions between contexts. Third, item-in-context memory survives a long retention-interval challenge. And, fourth, we conducted four transfers to novel, unpredictable context transitions, which document that our data cannot be explained by learning rules that govern predictable changes in context (Figure 3A; initial versus terminal (F(1,8)=2.21, p=0.18); number of context transitions (F(3,24)=1.77, p=0.18); interaction (F(3,24)=.28, p=0.84). Moreover, all of our data come from memory assessments which dissociate item-in-context memory from non-episodic judgments of familiarity. Thus, the high level of accuracy (84% across Experiments 1–4) provides dramatic evidence for episodic memory that rules out non-episodic contributions to performance. Moreover, our data cannot be explained by the ability to detect the presence of pellets under S+ lids because we conducted unbaited probes and observed high levels of accuracy (see Supplemental Experimental Procedures).

Our data suggest that rats remember the context in which odors were presented. When the memory of items in context was put in conflict with familiarity cues, rats relied on item-in-context memory rather than familiarity. We conclude that rats remember multiple unique events and the contexts in which these events occurred using episodic memory. Our findings enhance the translational potential for utilizing animal models of episodic memory to both explore the biological mechanisms of memory and memory disorders and to validate therapeutic approaches for disorders of episodic memory. Moreover, our findings suggest that the ability to represent numerous episodic memories is quite old in the evolutionary timescale. More broadly, our work supports the view that rats may be used to model fundamental aspects of human memory.