Event boundaries in the brain: a review of Zheng et al. 2022

Author: Riley Yen
Mentor: Dr. Apoorva Bhandari
United Nations International School

A paper titled “Neurons detect cognitive boundaries to structure episodic memories in humans” (Zheng et al., 2022) examined the question of how different cognitive boundaries influence neural activity and memory processes. This influential study was one of the key steps to help scientists understand how different types of memory are processed and retained. Cognitive boundaries are moments that break up an experience into meaningful parts and are known to play a role in structuring memory. Little is known about how this process is realized in the human brain. The paper under discussion attempts to fill in this gap by identifying neurons that code for cognitive boundaries and show their relationship to subsequent memory. The aim of the current paper is to summarize and evaluate Zheng et al.’s study and discuss how this new knowledge can be used in the field. One thing to note about each participant was that the participants all had drug-resistant epilepsy on whom an invasive procedure was performed for medical reasons to implant depth electrodes in the brain to identify the source of the epilepsy. These depth electrodes were placed in the amygdala, hippocampus, and parahippocampal gyrus which are known to have a role in episodic memory, and could also be leveraged to sense the firing of neurons in those specific regions of the brain This allowed the researchers to collect a specific neuron’s firing pattern or rate (Zheng et al., 2022).

The study was comprised of 3 different tasks: an encoding task, a scene recognition task, and a time discrimination task to mimic the way memories are conceived and recounted in natural life. Participants were introduced to movie clips in the ‘encoding task’ with different boundaries. In the encoding stage, boundaries were placed in the movie clips and a no-boundary section (NB) was also presented. High boundaries (HBs) cut to an entirely different movie while short boundaries (SBs) were movies that cut to a different part in the movie. Each participant saw 90 different clips. After the encoding stage, they were assessed on scene recognition and time discrimination, followed by a confidence rating from 1 – 6. In the scene recognition task participants were shown a random image of either a film they saw or a new unseen one (foils). Participants answered the questions about the film as “old” or “new”. New films and foils were equated on luminosity, colours, similar objects, etc. The time discrimination task was another type of memory performance test, but instead of recognition, the aspect of memory tested was temporal order memory. Participants were shown 2 frames and had to decide which one was seen first in the original video (Zheng et al., 2022).

The signals measured from the depth sensors were evaluated for the changes in neural activity through principal component analysis (PCA) to examine dynamics associated with the different types of boundaries. Analyzing the neural activity with PCA can reveal how neurons interact with each other in relation to episodic memory and boundaries, how different boundaries affect the memory of individuals, and the role firing rate has in episodic memory depending on the cognitive boundaries (Zheng et al., 2022).

The results provide new insights into how memory is encoded and recalled. The study found that events followed by a cognitive boundary were recognized better in the recognition questions, but HBs decreased performance in time recognition (order of events). The recognition accuracy decreased when the boundaries were farther away from the tested frames, so the order of test frames that appeared immediately after a boundary in the original video was better discriminated. Impaired time discrimination in the HB was likely due to less memory strength since more strength is required for time discrimination. Thus, frames immediately after a boundary were better recognized and the strength of activity in boundary-reactive neurons was predictive of memory strength. In scene recognition, confidence and the time taken to answer stayed the same across boundaries. The recognition accuracy decreased when the boundaries were farther away from the frame. On the other hand, impaired time discrimination was likely due to less memory strength since more strength is required for time discrimination. It became evident that different types of responses to the stimuli were beneficial for different types of memory. In scene recognition, large neural shifts were beneficial for memory but not in time discrimination. The contrast in reactions and results depending on the tasks actively demonstrates how unique memory functions are processed, retained, and recalled differently (Zheng et al., 2022).

Over the course of the study, an unexpected discovery was made: there were 2 types of cells which had different patterns in activity and reactions. The experiment classified these different types of cells as ‘boundary’ and ‘event’ cells to study their properties in the enhancement or manufacturing of memory. It was found that the cells’ reactivity was predictive of how well the clips were remembered, Boundary cells were cells that responded to HBs and SBs while event cells were only responsive to SBs. The parahippocampal gyrus (26.47%) had the most boundary cells, amygdala (7.10%) and hippocampus (3.50%). Their responsiveness was all graphed into response time using the principal component analysis and electrodes which showed implications that event cells are reliant on boundary cells to function as there is a time lag for them to possibly communicate. When applied to time distinction and recognition, the two types of cells showed different results that alluded to different functions for different purposes. Event cells were more reactive after HBs compared to SBs as the brain started paying more attention after the narrative of the film was disrupted. SB neural shifts were only visible when boundary cells were included. HB neural shifts were only visible when event cells were included. In short, theta phase locking (neuron firing rhythms) did not play a role in recognition memory success but did increase the temporal order memory when studying the event cells. The firing rate in boundary cells predicted scene recognition and increased performance. Boundary and event cells were used to predict different things, demonstrating the functions they play in the different types of memory they benefit (Zheng et al., 2022).

The findings from this experiment have progressed the understanding of how memory is structured pertaining to the different types of memory and the different cells responsible. The understanding can be applied to other implications in the field. One application to learning about how memory works can help maximize a human’s ability to remember in school settings or professional settings to better retain information. Knowing the cognitive boundaries is beneficial for scene recognition, for example, it can be a guiding point for how lessons or meetings are structured to optimize neuro functions. Another application can be in the medical field where memory disorders such as dementia and Alzheimer’s have a better understanding to be studied as researchers continue to develop solutions based on the work in this study concerning boundary and event cells. Due to their work with drug-resistant epilepsy patients, the implanted electrodes can push forward research in identifying and diagnosing where the seizures stem from in the brain. Finally, optimizing how memory is encoded in a human’s brain can be applied to how it is used in AI technology. Gabriel Kreiman, an ophthalmology professor at Harvard University commented on the findings of the study and how in the field of IA technology, it can be shifted to “build better algorithms that will incorporate episodic-like memories in artificial intelligence” (Kreiman, 2023). This can not only be a very plausible application of the research, but it can also be a sign that there are many more uses for these findings in the future as AI technology continues to develop (Zheng et al., 2022).

The experiment has been supported by other researchers who have found similar results after the study was conducted. South Korean researchers, Yujin Rah, Jiyun Kim, and Sangah Lee discovered parallel ideas to how boundaries affect memory. They specifically studied how it impacts children’s memories and spatial mapping. They found that the participant’s memory was indeed impacted by the type of boundary (2D line, 3D wall, and aligned objects) and children’s memory was negatively affected by boundaries (Rah, Lee, and Kim, 2022). Another study used a more similar approach to the original experiment as they targeted the recognition task more in-depth. Alexis Campbell made use of fMRI instead of Zheng et al.’s original method of depth electrodes which can detect where the brain is most active instead of targeting specific neurons. This allowed Campbell to see a more general and wider picture of how episodic memory is encoded for recognition (Campbell, 2023).

These discoveries derived from Zheng’s experiment have opened unexplored opportunities for different research areas with new progress to be made. The discovery of event cells and boundary cells in the way they behave and the different functions they offer is one of the most remarkable things about this work as it is an example of how there are still realms of the brain to be discovered and how research can better suit the complexities of neuroactivity and episodic memory research. Zheng’s experiment has contributed to how time discrimination and scene recognition are both very distinct areas of episodic memory that operate uniquely. With their analysis using depth electrodes, Zheng and her team were able to hone in on a couple of neurons to graph and study the patterns in their firing rates which were predictive of the participant’s memory performance task results in accuracy. With all of these applications to this knowledge, it is undoubted that new information can be built off of the understanding of episodic memory and cognitive boundaries as there already has been a growing number of experiments that have been conducted to deepen this understanding (Zheng et al., 2022).


Campbell, Alexis. “Behavioral and Neural Effects of Event Boundaries on Implicit Associations in Episodic Memory.” Digital Collections, 2023, https://digitalcollections.wesleyan.edu/_flysystem/fedora/2023-07/1667.pdf.

Rah, Y. J., Kim, J., & Lee, S. A. (2022). Effects of spatial boundaries on episodic memory development. Child Development, 93, 1574–1583. https://doi.org/10.1111/cdev.13776

Zheng, J., & Kreiman, G. (2022, May 31). Neurons detect cognitive boundaries to structure episodic memories in humans | The Center for Brains, Minds & Machines. MIT CBMM. Retrieved 2023, from https://cbmm.mit.edu/video/neurons-detect-cognitive-boundaries-structure-episodic-mem ories-humans-video

Zheng, J., Schjetnan, A.G.P., Yebra, M. et al.. Neurons detect cognitive boundaries to structure episodic memories in humans. Nat Neurosci 25, 358-368 (2022). Retrieved 2023, from https://www.nature.com/articles/s41593-022-01020-w#citeas

About the author

Riley Yen

Riley is an 11th grader in Manhattan, New York. She is passionate about neuroscience, psychology, and molecular biology. She is particularly interested in topics such as human behavioural patterns, social decision-making, and forensic science which she wishes to further explore in her studies in the future. Her curiosity in this field is due to her fascination with the relationship between how society operates and how it is intertwined with neurological and biological processes in humans.

In her free time, Riley also enjoys writing and composing music, baking for her friends and family, and reading books about social sciences.