Sleep targets highly connected global and local nodes to aid consolidation of learned graph networks

Abstract

Much of our long-term knowledge is organised in complex networks. Sleep is thought to be critical for abstracting knowledge and enhancing important item memory for long-term retention. Thus, sleep should aid the development of memory for networks and the abstraction of their structure for efficient storage. However, this remains unknown because past sleep studies have focused on discrete items. Here we explored the impact of sleep (night-sleep/day-wake within-subject paradigm with 25 male participants) on memory for graph-networks where some items were important due to dense local connections (degree centrality) or, independently, important due to greater global connections (closeness/betweenness centrality). A network of 27 planets (nodes) sparsely interconnected by 36 teleporters (edges) was learned via discrete associations without explicit indication of any network structure. Despite equivalent exposure to all connections in the network, we found that memory for the links between items with high local connectivity or high global connectivity were better retained after sleep. These results highlight that sleep has the capacity for strengthening both global and local structure from the world and abstracting over multiple experiences to efficiently form internal networks of knowledge.

Conflict of interest statement

The authors declare no competing interests.

© 2022. The Author(s).

Figures

Figure 1

Experimental procedure and task description. (A) In our within-subject design participants took part in two identical experimental sessions with parallel versions of the task and retention intervals containing either sleep or wakefulness. The learning phase started at 8:00 a.m. (or p.m.) and participants performed a learning task (see C) and a reward task (see F). After the 10-h retention interval, participants came back to the lab to complete the retrieval phase (see G). After 1 week, participants returned to perform the other experimental session with the remaining retention interval. (B) Representation of the undirected graph structure composed of 36 edges (black lines) linking 27 nodes (circles with examples of stimuli presented during the experience). Red nodes represent reinforced nodes, either positive (reward), negative (punishment) or neutral. This image was never shown to the participants, but determined the structure of the learned associations. (C) During learning, participants saw one planet at the bottom depicting their current position on the graph and three planets to choose from displayed at the top. After choosing, the choice was marked but only the correct planet (i.e., the one connected to the bottom planet according to the graph) moved down to replace the bottom planet. Then a new set of three planets appeared at the top prompting a new choice. Participants performed eight such choices (transitions) taking an eight-step route through the graph (an example route is indicated by arrows in (B). After each route, they received feedback on their performance and a new route started at a pseudo-random location on the graph. Participants performed 81 routes in total. (D) To construct the graph the graph-theoretical parameters of degree centrality (number of direct connections for a given node) and (E) closeness centrality (inversely proportional to the number of steps to every other node on the graph) were orthogonalized (i.e., allowing to independently asses their effect on retention) and a three-fold symmetry was pursued (to enable equal positions for the reinforced nodes). (F) During the reward task, participants were shown a planet representing one of the three reinforced nodes (reward, punishment and neutral). After 0.5–1 s a white square appeared on top of the picture. Dependent on the participants pressing the spacebar quickly enough, they were shown the outcomes at the bottom (for details see “Reward task” below). (G) During the retrieval task, participants were shown two planets taken pseudo-randomly from the graph network and had to answer whether they were directly connected or whether one, two or three and more planets were in between.

Figure 2

Retention performance. (A) Left, the retention measure was calculated by substracting the learning performance from retrieval performance. Details can be found in Supplementary Fig. 4. Right, mean overall retention performance for the sleep (blue) and the wake (red) condition. (B) Mean retention performance and (C) regression model for the different distances within the graph, (D) and (E) for the different levels of degree centrality of the nodes and (F) and (G) for the different levels of closeness centrality of the nodes. For the violin plots, the black dots represent the individual performance, the black bar represents the mean across participants, the black rectangle shows the 95% of a Bayesian highest density interval and the coloured shape displays the smoothed density. *p < 0.05, **p < 0.01.

Figure 3

Participants’ raw learning and retrieval data. (A) Learning curve across the 81 routes for the sleep and (B) the wake condition. Mean (thick line) and individual responses (thin lines) for correct response (green), the close distractor (yellow) and the distant distractor (red) smoothed by a five point moving average. The black arrow indicates when participants’ performance was significantly biased by the graph structure (i.e., accuracy of the close and distant distractor started to differ). (C) Retrieval task data for each distance (rows) in the sleep and (D) the wake condition. On the left the graph structure and an example (in green) is shown for each distance, which is defined by the number of edges between the pair of nodes tested. Within the circles, green lines represent correct connections and grey lines correspond to incorrect connections at that distance, whereas line thickness depicts how many participants gave the respective answer.

Figure 4

Reinforced nodes results. (A) Individual monetary balance curves during the reward task in the sleep and (B) the wake condition. Participants earned the amount reached at the end of the task. (C) Retention performance for the reinforced nodes. The black dots, bar and rectangle represent the individual performances, the mean and the 95% of a Bayesian highest density interval, respectively. The coloured shape shows the smoothed density. *p < 0.05.

Figure 5

Correlation of the results and the navigation score. (A) The Navigation Strategies Questionnaire (NSQ) score, the black dots, bar and rectangle represent the individual performances, the mean and the 95% of a Bayesian highest density interval, respectively. The coloured shape shows the smoothed density. (B) Relationship of the Navigation Strategies Questionnaire (NSQ) with learning performance, (C) retrieval performance and (D) retention performance for the sleep condition and with (E) learning performance, (F) retrieval performance and (G) retention performance for the wake condition. Regression lines (blue—sleep, red—wake) and black dots for the individual data points are shown.