Periodic Reporting for period 3 - SolutionSleep (Understanding creativity and problem solving through sleep-engineering)
Période du rapport: 2020-02-01 au 2021-07-31
Innovative problem solving is critical for all spheres of organised endeavour, including science and industry, and thus forms the cornerstone of a successful society. Such creative thinking often re-quires suppression of preconceptions and restructuring of existing knowledge. Pioneering work has shown that sleep facilitates problem solving, but exactly how, and which sleep characteristics are important, remain to be determined. We know that recent experiences are replayed in sleep, and that in Slow Wave Sleep (SWS) this replay integrates new knowledge with old. The role of such replay in Rapid Eye Movement (REM) sleep, a stage which is strongly linked to creativity, is unknown. I have proposed a model which combines physiology, behavioural studies, and computational modelling to make testable predictions about the complimentary contributions of memory replay in REM and SWS to problem solving. SolutionSleep tests this model through explicit manipulation of memory replay in sleep. I will use a very recently developed technique to explicitly trigger memory replay, a pioneering method for quantifying triggered replay, and cutting-edge approaches for manipulation of neural oscillations through neurostimulation. I expect two key results: first, I will uncover the principles of how memory replay in REM and SWS, combines with specific neural oscillations to promote both long-term memory and creative problem solving. This will involve development of a computational model which will enable optimised experimental design, paving the way for efficient future investigation of ways to enhance innovation through sleep engineering. Second, I will develop methods for boosting these sleep processes in a selective, targeted manner. Immediate consequences will include a translational project to facilitate everyday problem solving. My findings will revolutionise under-standing of sleep and how it impacts upon some of our most important cognitive abilities—memory and problem solving.
The project is composed of 3 work packages. Broadly speaking, WP2.1 relates to studies using a technique in which sounds are used to trigger memory reactivation in sleep called Targeted Memory Reactivation or TMR. In the first part of this work package we develop machine learning method to detect TMR cued reactivation. In the second part, a series of studies used TMR to test the impacts of different types of memory reactivation in different sleep stages on creative processes. WP2.3 relates to the manipulation of sleep oscillation through carefully targeted manipulations (external stimuli or other manipulations). WP2.2 relates to computational modelling of memory consolidation in sleep, and has yet to start.
WP 2.1.1 Here, we use EEG classifiers to examine the properties of neural reactivation during sleep. Our aim was to improve our existing classifiers. To date, we have show that neural reactivation classifies more successfully in slow wave sleep than stage 2 sleep and that targeted memory reactivation loses impact when repeated many times. We have developed a method for classification using phase locking between different electrodes values. We have simplified our pipeline, replicated our initial result, and found that reactivation occurs both immediately (~100ms) after the TMR cue and one second later, and published at theoretical model explaining this. Furthermore, we have found that immediate reactivation is significantly more likely after TMR cues which start just after the slow oscillation trough, suggesting a strong phase dependency of TMR.
WP 2.1.2: Here, we aim to use TMR to test the differential impact of memory reactivation in SWS and REM upon abstraction and the formation of unexpected novel connections. We have supported our theoretical model by showing that TMR in REM but not slow wave sleep facilitates formation of indirect associations, and that this facilitation is enhanced after two weeks. We followed this with two studies of abstraction and analogical reasoning. Our initial behaviour-only study using this task showed performance gains across a night of sleep compared to day wake. We have also completed a full TMR study, and shows that reactivation of the task in REM, but not slow wave sleep (SWS), leads to performance gains.
WP 2.3: Here, we explicitly manipulate the neural oscillations of sleep and examine the impact on abstraction/integration and analogical reasoning tasks. We have conducted a careful analysis of how closed loop auditory stimulation in SWS impacts on oscillations. Our work demonstrates the optimal window for such stimulation, and how this differs across participants. Additionally, we have developed a method to predict online, in real time, which SOs will respond to stimulation in terms of increasing subsequent spindles and SO power with ~80% accuracy. We are also conducting a study of REM manipulation through auditory stimulation.
Finally, we have conducted a study in which we attempt to increase REM by simply asking participants to wear eye masks. With 75 datasets we see a clear advantage in reaction time on the day after wearing an eyemask compared to no mask.
WP 2.1.1 Here, we use EEG classifiers to examine the properties of neural reactivation during sleep. Our aim was to improve our existing classifiers. To date, we have show that neural reactivation classifies more successfully in slow wave sleep than stage 2 sleep and that targeted memory reactivation loses impact when repeated many times. We have developed a method for classification using phase locking between different electrodes values. We have simplified our pipeline, replicated our initial result, and found that reactivation occurs both immediately (~100ms) after the TMR cue and one second later, and published at theoretical model explaining this. Furthermore, we have found that immediate reactivation is significantly more likely after TMR cues which start just after the slow oscillation trough, suggesting a strong phase dependency of TMR.
WP 2.1.2: Here, we aim to use TMR to test the differential impact of memory reactivation in SWS and REM upon abstraction and the formation of unexpected novel connections. We have supported our theoretical model by showing that TMR in REM but not slow wave sleep facilitates formation of indirect associations, and that this facilitation is enhanced after two weeks. We followed this with two studies of abstraction and analogical reasoning. Our initial behaviour-only study using this task showed performance gains across a night of sleep compared to day wake. We have also completed a full TMR study, and shows that reactivation of the task in REM, but not slow wave sleep (SWS), leads to performance gains.
WP 2.3: Here, we explicitly manipulate the neural oscillations of sleep and examine the impact on abstraction/integration and analogical reasoning tasks. We have conducted a careful analysis of how closed loop auditory stimulation in SWS impacts on oscillations. Our work demonstrates the optimal window for such stimulation, and how this differs across participants. Additionally, we have developed a method to predict online, in real time, which SOs will respond to stimulation in terms of increasing subsequent spindles and SO power with ~80% accuracy. We are also conducting a study of REM manipulation through auditory stimulation.
Finally, we have conducted a study in which we attempt to increase REM by simply asking participants to wear eye masks. With 75 datasets we see a clear advantage in reaction time on the day after wearing an eyemask compared to no mask.
The main expected progress points are listed below:
1) Determine whether triggered memory reactivation in REM or SWS can facilitate gist abstraction, the formation of conceptual hierarchies, analogical reasoning, or problem solving through the integration of different types of information.
2) Determine whether inhibiting or facilitating various sleep oscillations [e.g. slow oscillations, sleep spindles, and theta oscillations] can facilitate or impair the above processes.
3) Develop EEG classifiers which can accurately detect cued memory reactivation during non-REM and REM sleep.
4) Relate the memory reactivation detected by the above classifiers to performance improvements/deficits in the above listed abilities.
5) Develop a computational model which captures the concept of memory replay in sleep. This should include representations of the hippocampus and neocortex and how replay is coordinated between these two structures in non-REM sleep, but decoupled in REM sleep.
6) Ensure the model captures main properties of the system, such as the fact that new information is learned in wake, then this is followed by offline replay and consolidation across a night of sleep with REM and non-REM iteratively interleaved, and predicts some of our observed behavioural results.
7) Use the computational model and its outputs to design further experiments using triggered memory reactivation and manipulation of oscillations.
1) Determine whether triggered memory reactivation in REM or SWS can facilitate gist abstraction, the formation of conceptual hierarchies, analogical reasoning, or problem solving through the integration of different types of information.
2) Determine whether inhibiting or facilitating various sleep oscillations [e.g. slow oscillations, sleep spindles, and theta oscillations] can facilitate or impair the above processes.
3) Develop EEG classifiers which can accurately detect cued memory reactivation during non-REM and REM sleep.
4) Relate the memory reactivation detected by the above classifiers to performance improvements/deficits in the above listed abilities.
5) Develop a computational model which captures the concept of memory replay in sleep. This should include representations of the hippocampus and neocortex and how replay is coordinated between these two structures in non-REM sleep, but decoupled in REM sleep.
6) Ensure the model captures main properties of the system, such as the fact that new information is learned in wake, then this is followed by offline replay and consolidation across a night of sleep with REM and non-REM iteratively interleaved, and predicts some of our observed behavioural results.
7) Use the computational model and its outputs to design further experiments using triggered memory reactivation and manipulation of oscillations.