“Integrative learning” promotes learning but not memory in older rats

Introduction

Traditional learning focuses on a gradual, from the part to the whole, style. However, this type of learning may not fit well with the new era of information age, in which there may be too much to be learned and the access to any part of them become easy and available. In order to explore a novel way of learning that may fit better with the needs of the new era, we proposed the concept of “Integrative Learning”, which refers to “the process of actively integrating learning materials to achieve an efficient and in-depth understanding and mastery of knowledge under the effect of meta-cognition—it is the psychological process of learning that highly integrates the meta-cognitive and cognitive processes” (Yin, Wu & Lian, 2020). In other words, under the guidance of this learning concept, individuals learn by first encountering and grasping the overall landscape of an area of knowledge, then diving into the details of each part of the area while consolidating the basics and connecting the details to the appropriate position of the overall landscape, and finally forming an integrative map of the area of knowledge, thus exhibiting a “whole-part-whole” pattern of learning (Fig. 1). In contrast, non-integrative learning such as progressive learning learn by directly encountering and grasping the basics and details of an area of knowledge, then advancing to a higher level of the area and so forth, and finally forming a finished map of the area of knowledge, thus exhibiting a “part-part-whole” pattern of learning (Fig. 1).

The concept of “integrative learning” was proved to be able to provide more efficient knowledge comprehension and long-term migration in an animal behavioral model using a modified 14-unit composite T-maze (Yin, Wu & Lian, 2020), originally designed by Edward Chase Tolman, the famous psychologist who proposed the concept of “cognitive map” (Tolman, 1948). Briefly, forty 1-month-old (short as 1 Mo hereafter) Sprague-Dawley (SD) rats, half male and half female, were randomly divided into the “Integrative Learning” (IL) group and the “Progressive Learning” (PL) group. For the IL group, the whole path was clear from the beginning, providing the subjects with opportunities to explore the whole maze without blockage; whereas for the PL group, the whole path was divided into three segments (and the 12-session learning stage was divided into three sub-stages corresponding with the opening of each new segment), providing the subjects with opportunities to explore each segment step by step (Fig. 2). The study found that the IL group initially made a lot of errors due to the difficulty of the task but the errors made soon dropped sharply as learning went on, and maintained at a meaningfully lower level compared with the PL group. Although the study did not find significant differences among groups during the memory retention test 1 week after the learning stage, the IL group performed markedly better during the following test stages designed to evaluate long-term knowledge transfer and adaptation (Nokes, 2009). In other words, when the correct route was reversed or modified, the IL group could still find the target with fewer errors and less confusion, suggesting that “Integrative Learning” can better assist knowledge transfer and goal-directed behavior (Frese & Sabini, 2021). The study also found that the males benefited more from “Integrative Learning” than the females, possibly due to sex differences in risk taking and/or associative learning in rats (Jolles, Boogert & van den Bos, 2015).

Although “Integrative Learning” was found to be more effective in young rats, it is unclear whether the phenomenon still holds in older rats. Aging brought changes to cognitive abilities such as verbal learning, memory, and problem solving (Arenberg, 1973; Kubanis & Zornetzer, 1981), which might be a result of changes in brain plasticity (Mahncke, Bronstone & Merzenich, 2006; Bloss et al., 2011), although some argues that aging brains may have remarkable reorganization and plastic adaptation in its own right (Holman & de Villers-Sidani, 2014). In animal models, research has shown environmentally-induced changes in the brains of older rats which were different from those in young rats (Cummins et al., 1973). Therefore, it is reasonable to speculate that effects of “Integrative Learning” might be different between young and older rats.

Here, we adapted our original design of “Integrative Learning” (Yin, Wu & Lian, 2020) and tested its effects on older rats. A 14-unit composite T-maze was used as the experimental setup and a series of consecutive experiments and analysis were conducted for the purpose of comparing the effects of different learning modes and exploring the intrinsic mechanisms of the learning effect differences on the behavioral level.

Materials and Methods

Study design

The study adopted a two (Learning mode: Integrative Learning/Progressive Learning, or IL/PL) by two (Age: 1/12 Mo) inter-group design, with the number of sessions and the number of total errors made (sum of detection errors and entry errors) as quantitative variables, animal characteristic behavior and tracking routes as qualitative variables. In order to explore the effects of different learning modes, three stages of tasks were designed: the learning stage, the 1-week-later memory retention test stage and the Gestalt (reverse) transfer learning stage (Fig. 3). Detailed task design instructions can be found in the Appendix S1.3 of Yin, Wu & Lian (2020).

Results

The graphic results of the learning stage, memory retention test stage and Gestalt transfer learning stage are shown in Figs. 4A–4D, respectively. As shown in Fig. 4A and Table 1, during the learning stage, the results of repeated measurement ANOVA with age and learning mode as intergroup variables and session as an intragroup variable show that the main effect of age was significant in the number of total errors made, and the number of total errors made of the 12 Mo groups is significant more than the 1 Mo groups. At the same time, the interaction between session and learning mode was significant (further simple-effect analysis can be found in Table S1). As shown in Fig. 4B and Table 2, the results of complete random analysis of variance with age and learning mode as intergroup variables showed that the main effects of learning mode and age were both significant in the number of days to learning success. The number of days to learning success of the IL-1 Mo group was significantly shorter than that of the PL-1 Mo group, while the number of days to learning success of the 12 Mo groups was generally more than that of the 1 Mo groups (further simple-effect analysis can be found in Table S2). However, unlike the 1 Mo groups which showed significant differences between the IL mode and the PL mode all across three learning sub-stages (Table 3), there was no significant differences between the IL mode and the PL mode in the 12 Mo groups except for Sub-stage Three, during which the IL mode showed an advantage by making significantly fewer total errors than the PL mode (Table 4) and exerting much less effort exploring the maze (Fig. S2). Details for interaction and subsequent simple-effect analysis can be found in Tables S3 and S4.

In order to elucidate why in the 12 Mo groups, the differences between the two learning modes were not as vast as in the 1 Mo groups, we further conducted a heatmap analysis for the Sub-stage Two (Session 4–7). This is because during this sub-stage the subjects had been accustomed to the apparatus but had not fully mastered the correct route, providing us the opportunity to see what exactly were different between different age groups. As can be shown in Fig. 5, the IL-1 Mo group spent much less time at the learned routes and mostly at the critical turning points of the unlearned routes, whereas the PL-1 Mo group almost evenly spent their time across the opened segments. In contrast, both the IL-12 Mo group and the PL-12 Mo group distributed their time across the whole maze, possibly because they were able to climb the maze wall, but the IL-12 Mo group had a higher density of activities in the lower-part area of the maze, which was in closer proximity to the start and the end points of the maze, whereas the PL-12 Mo group was more similar to the PL-1 Mo group in having their most activities in the opened segments of the maze, although the PL-12 Mo group was able to occasionally “break the barrier” to the unopened segment because they could climb the wall. These results demonstrate that the 12 Mo groups are different from their young counterparts in their learning progress and regularity, albeit the differences were mainly manifested in the IL groups.

During the memory retention test stage, the results of repeated-measurement ANOVA with age and learning mode as intergroup variables and session as an intragroup variable show that the main effect of age was significant in the number of total errors made (Table 5). The young groups performed overall better than the 12 Mo groups, but the different learning modes had no effects on this stage (Fig. 4C).

In the Gestalt transfer learning task, the results of repeated-measurement ANOVA with age and learning mode as intergroup variables and session as an intragroup variable showed that the main effects of age and session were significant in the number of total errors made (Table 6). The 1 Mo groups performed overall better than the 12 Mo groups, though the differences between the two learning modes were not significant, mainly because unlike the 1 Mo groups, the IL-12 Mo group did not perform better than the PL-12 Mo group (Fig. 4D).

Results from an aggregated heatmap analysis for the 12 Mo groups throughout the experiments can be seen in Fig. 6. Few differences can be seen between the two learning modes during the memory retention test stage and the Gestalt transfer learning stage, albeit the obvious different distribution of activities between the two learning modes during the learning stage may reflect a slight advantage for the IL group.

In order to further elucidate the patterns, the number of total errors made during Sub-stage Three (Session 8–12) of the learning stage, memory retention test stage and Gestalt transfer learning stage was averaged for each rat (Fig. S1). Then non-repeated measure ANOVAs were conducted using the phases of the experiment as a within-subject factor, age and learning mode as between-subject factors (Table S5). The results show that the main effect of the phase of the experiment was significant (F_(2,62) = 24.29, p < 0.001, η2 = 0.439), and post-hoc multiple comparison tests reveal that the number of total errors made in the Gestalt transfer learning stage was significantly more than the other two phases of the experiment (Table S6). The main effect of age was also significant (F_(1,31) = 45.139, p < 0.001, η2 = 0.593), and post-hoc multiple comparison tests reveal that the number of total errors made by the 12 Mo groups was significantly more than that of the 1 Mo groups (Table S7). Further simple-effect analysis (Table S8) revealed that during Sub-stage Three (Session 8–12) of the learning stage, the PL-12 Mo group made significantly more total errors than the IL-12 Mo group (F_{(1, 31)} = 11.01, p < 0.01), whereas the IL-1 Mo group and the IL-12 Mo group did not have significant differences (F_{(1, 31)} = 1.48, p = 0.233). During the memory retention test stage, the two learning modes did not have significant differences between each other. During the Gestalt transfer learning stage, the two learning modes did not have significant differences between each other either.

Discussion

Maze learning in rats is a classic problem in the history of modern psychology (Tolman & Honzik, 1930; Tsai, 1930; Buel, 1935; Honzik, 1936; Tolman, 1938, 1948; Morris, 1984; Andrade et al., 2001; Dudchenko, 2001). Tsai (1930) used to address the effects of two different instructional methods in maze learning experiments with rats, proving that repeated instructions could inhibit individual learning, which fundamentally challenged the relationship between teaching and learning. The result was cited by Tolman (1938) to overthrow the law of exercise—proposed by the well-respected educator and psychologist, Thorndike, who considered that any stimulus-response combination will gradually be strengthened during trial-and-error learning through regular practice and application. The impact of that experiment is still meaningful today (Wang et al., 2021). The difference between Tsai’s experiment and ours is that Tsai found that adding more instructions was less effective while we found that adding more segmentation/limitation was less effective—we both found that whatever the experimenter as an instructor do to “help” the subject (the learner) to learn the task in an easier manner, it turns out to be a less effective one. Learning seems to ultimately depend on the integrative efforts made by the learner under a self-adapting mode, but not necessarily on the instructor¹ . From the viewpoint of the neurobiology of learning, the results indicate the brain may not just be a learning machine that learns by simply strengthening corresponding synapses via repetition of the same information every day, nor does it learn by adding synaptic connections step by step—it may actually learn better by forming “schema” of the “whole” spectrum of information and then adjust the weights and connections of the synaptic network as it encounters stimuli (new and old) every day (Tse et al., 2007; Farzanfar et al., 2022).

While integrative learning was clearly advantageous in the 1 Mo groups (Yin, Wu & Lian, 2020), data from the current study suggest that such advantages may only present in the last segment of the learning stage in older rats but not in the Gestalt transfer learning stage. Was it due to differences in the learning process or was it because older rats had deteriorating memory systems to support higher-order functioning required by the IL mode? The heatmap analysis suggests that the differences do lie in the learning process especially for the IL-12 Mo group—indeed both old groups learned to climb the wall during the learning stage. Kostić & Tošković (2022) designed an experiment to determine which factor among path, time and effort was the key determinant of rodent behavior in exploring a maze to reach food, and they found that the subject always chose paths requiring less effort. Their findings provided a clue for why the older rats had attempted to climb the wall to reach the food-box—this may be less arduous for them compared with young rats. However, while climbing the wall might help reaching the food-box faster in some cases, it clearly slowed down the overall learning progress as compared to the 1 Mo groups. Nevertheless, the IL-12 Mo group may still benefit from the IL scheme from the beginning, resulting in a clear advantage over the PL-12 Mo group during the last segment of the learning stage, albeit this advantage did not extend further into the Gestalt transfer learning stage as in the 1 Mo groups, which suggest that the critical function supporting the linkage between learning and memory may be deteriorating in older rats.

It also worths noting that the IL groups did not simply have more repetitions of the whole route than the PL groups. First of all, the IL group and the PL group both had 12 sessions of learning, and for each session they could conduct free exploration within 15 min before reaching the foodbox and consuming food, upon which time the individual trial immediately ended and the subject was fetched out of the maze. Therefore, even though the PL group only had 5 days to experience the whole route, they had apprehended two thirds of the maze for 7 days and their total learning efforts (as reflected by the travel distance and the travel time) during Sub-Stage Three (Session 7–12) was much more than the IL group (Fig. S2) as the IL group quickly reached the foodbox with much fewer errors. Secondly, the IL group did not simply repeat during the 12 learning sessions. In fact, after they grasped the idea of the whole maze through initial exploration, they spontaneously segmented the whole route, focusing on learning the first “artificial”² segment to facilitate transferring of the memory (Fig. 5 of Yin, Wu & Lian, 2020), after which they started repetition to consolidate what they had learned through memory transfer to the less familiar parts of the routes (Fig. 5 of this article). Indeed, the segment fixation test of Yin, Wu & Lian (2020) further proved the objectivity of this memory transfer hypothesis, during which both the IL group and the PL group adopted the original route of the first segment and the new (direct) route of the second and the third segment, which suggest that the autonomous consolidation of the first segment for easier transfer later lie in the nature of learning. At the same time, the PL group was facing a new segment during the initial 1 or 2 sessions of each sub-stage, during which they focused on exploration but not exploitation—however, during the rest of each sub-stage the PL group also entered a repetitive mode, consolidating what they have learned, as reflected by the number of total errors made in Fig. 4A and session by session heatmap analysis (see Supplemental Data of this manuscript and Yin, Wu & Lian, 2020).

In sum, on one hand, the repeated learning of the PL group occurred 1 or 2 sessions right after a new segment was opened, while the repeated learning of the IL group occurred after consolidation of the “artificial” first segment in the three-segment maze—considering the actual travel distance and travel time they had in the maze for each session, the two groups only had substantial differences in when repeated learning occurred but did not have substantial differences in the total amount of repeated learning of the whole route. On the other hand, even if the whole route was opened to the IL group from the very beginning, the rats did not simply repeat the correct route and remember it, but instead they transferred the memory of a typical segment of the route to help learn the other segments to optimize the utilization of memory, which have been proved to be no different between IL and PL—the critical difference between the two learning modes, as we argue, lie in the fact that the IL group got to learn the whole route from the beginning and thus did the segmentation by themselves while the PL group’s learning was constrained by pre-determined segmentation by the experimenter who might have thought that this would help the subjects learn better—this difference results in a different trajectory in how their memory is construed in their brain, as shown in Fig. 1.

This study is not without limitations—we could have designed a new maze with higher walls to avoid the problem of climbing in the 12 Mo groups—however, the high wall represents another technical challenge in tracking the subject that at the time we had no means to overcome. Therefore, as an informative study, we retain the original design of the apparatus to compare the effects of the different learning modes between the 1 Mo and the 12 Mo groups. Future studies could repeat our experiments in a more adapted apparatus that could accommodate subjects of different ages and sizes for confirmatory purposes. On the other hand, in this study we only had the 12 Mo male group to be compared with the 1 Mo male group due to the consideration that the high long-term maintenance cost of animals in our facility did not outweigh the potential benefits in answering research questions given that our previous study (Yin, Wu & Lian, 2020) found that the IL mode was more effective in the male group than in the female group. Future studies shall compare the effects of IL on the older female group with the older male group because older female rats may have different learning patterns from older males (Stouffer & Barry, 2014).

Conclusions

In this study, we adapted the original design of Yin, Wu & Lian (2020) to investigate whether “Integrative Learning” was still advantageous in older rats. We found that the IL learning mode still promotes learning but not memory in older rats, possibly due to the fact that those higher-order cognitive abilities that support meta-cognition, long-term retention and knowledge migration might be deteriorating in older rats. Further studies on neural correlates of such differences between the two learning modes in both young and older groups should provide us with further insights on the mechanisms and neural dynamics of such changes across different learning and memory stages. Furthermore, deep learning algorithms that feed on the real-time tracking points of the subjects under different learning modes could help formulate the theory of “Integrative Learning” in a way that it could provide more precise predictions for the relationship between learning mode and long-term outcome.

Supplemental Information