🧠 neuroscience

Causal network structure predicts memory organization and neural reinstatement across events

This study demonstrates that the causal structure of narratives guides memory organization and is neurally encoded by the default mode network, which reactivates causally related events across temporal gaps and modulates representational switching based on causal distance.

Original authors: Antony, J. W., Abbas, S., Babb, M. S., Reagh, Z., Ranganath, C.

Published 2026-07-24

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Antony, J. W., Abbas, S., Babb, M. S., Reagh, Z., Ranganath, C.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine your brain as a super-advanced movie theater. Every time you experience something new, your brain doesn't just record a flat video; it builds a complex, three-dimensional map of the story. Scientists have long known that our memories are organized like stories, but they've been arguing about how the brain decides what comes next. Is it just a simple timeline, like a calendar where Monday follows Sunday? Or is it more like a web of connections, where one idea pulls you toward another because they make sense together? This question sits at the intersection of psychology and neuroscience, two fields trying to figure out the secret code behind how we remember the past. While we know the brain has a "default mode" that kicks in when we're daydreaming or thinking about big ideas, it's been a mystery whether this specific brain network is the one building the causal bridges between our memories. Understanding this is crucial because it explains how we make sense of the world, turning a chaotic stream of moments into a coherent narrative we can actually use.

In this study, a team of researchers decided to test this by treating a classic TV show, Friends, like a laboratory experiment. They didn't just watch the show; they scanned the brains of 23 people while they watched a two-part episode featuring five different storylines happening at the same time. To figure out the "rules" of the story, another group of people watched the same clips and rated how much one event caused another, creating a giant map of cause-and-effect relationships. The researchers then asked the brain-scanned group to close their eyes and recall the episode as best they could.

The results were like finding a hidden instruction manual for the brain. When the participants tried to remember the show, they didn't jump from event to event based on who was in the scene or what time it was. Instead, their memories flowed almost perfectly along the lines of cause and effect. If Event A caused Event B, the brain was much more likely to jump straight from A to B in memory, even if other events happened in between. It turns out that the "top" cause-and-effect link was the strongest predictor of how people organized their memories, beating out time, characters, and even the general meaning of the words used.

But the real magic happened inside the brain scans. The researchers looked at a specific network of brain regions known as the Default Mode Network (DMN), which acts like the brain's internal storyteller. They found that the activity patterns in this network mirrored the causal map created by the raters. When a new event started, the brain didn't just start fresh; it "replayed" or reactivated the patterns of the previous event that caused the new one. It's as if, just as a new chapter begins, the brain quickly flips back to the previous page to remind itself of the plot before moving forward. This "stitching" happened specifically in regions like the medial prefrontal cortex and the angular gyrus.

Furthermore, the study showed that the brain knows when to switch gears. When two events were far apart in the cause-and-effect chain (meaning they were unrelated), the brain's activity patterns changed drastically at the boundary between them. But when events were tightly linked by cause and effect, the brain's pattern stayed more similar, keeping the story thread intact. This suggests that the brain uses these causal connections to decide whether to smoothly continue a story or to snap to a completely new one.

In short, the paper suggests that our brains are not just passive video recorders. Instead, they are active editors that build a complex web of cause-and-effect. The Default Mode Network appears to be the engine that stitches these moments together, reactivating the past to understand the present, ensuring that our memories make sense as a connected story rather than a random pile of snapshots.

Technical Summary: Causal Network Structure Predicts Memory Organization and Neural Reinstatement Across Events

Problem Statement
While causality is recognized as a fundamental component of narrative structure (e.g., "Because of that..."), its specific influence on the organization of episodic memory and the underlying neural mechanisms remain unclear. Previous research has established that temporal contiguity and semantic relationships organize memory, but it is unknown whether causal structure independently predicts how events are retrieved and how the brain represents transitions between them. Specifically, the study addresses whether causal networks shape memory recall transitions and whether the Default Mode Network (DMN) encodes these causal relationships, particularly at event boundaries where mental models are updated.

Methodology
The study employed a multi-experiment design involving fMRI and behavioral tasks using a two-part episode of the TV show Friends (Season 1, Episodes 16–17) as the stimulus. The stimulus featured five temporally interleaved storylines.

Causal Network Construction: A separate group of participants ( $N=6$ ) rated the cause-effect relationships between every pair of events in the stimulus (0–10 scale). These ratings were averaged to create a weighted causal network matrix, from which metrics like node degree (centrality) and weighted path length were derived.
fMRI Experiment: Participants ( $N=23$ ) watched the stimulus while undergoing fMRI scanning. Following encoding, they performed a free recall task, verbally recounting the episode in as much detail as possible.
Event Segmentation: Two additional groups ( $N=20$ and $N=25$ ) segmented the video into meaningful units or narrative shifts to validate behavioral event boundaries.
Neural Analysis:
- Regions of Interest (ROIs): Analyses focused on the Default Mode Network (DMN) components—medial prefrontal cortex (mPFC), angular gyrus (AG), and posterior medial cortex (PMC)—as well as V1, posterior hippocampus (postHC), and anterior hippocampus (antHC).
- Representational Similarity: Event patterns were averaged within ROIs, and representational similarity matrices were created. Multiple linear regression models tested whether causal structure (full matrix and "top" causal matrix) predicted neural similarity, controlling for time, storyline, semantic similarity, character, and place.
- Boundary Reactivation: Neural patterns at event boundaries (±2 TRs) were compared to patterns in the middle of past and future events to test for reactivation.
- Event Boundary Strength: Neural event boundaries were identified using Greedy State Boundary Search (GSBS) on Shared Response Model (SRM) reduced data. The strength of pattern changes at boundaries was correlated with the weighted path length between successive events in the causal network.

Key Results

Behavioral Memory Organization: In free recall, the transition from one recalled event to the next was best predicted by the "top" causal relationship (the strongest cause-effect link from the current event) compared to temporal order, character, place, or semantic similarity ( $p < 0.001$ for all comparisons). This suggests causality is the primary driver of memory organization in complex narratives.
Neural Representation of Causality: In the DMN (mPFC, AG, PMC), the representational similarity between events significantly correlated with the causal network structure, even after controlling for storyline, time, and semantic features. This effect was not observed in V1 or the hippocampus.
Reactivation at Event Boundaries: At event boundaries, neural patterns in the mPFC, AG, and PMC showed significant reactivation of prior, causally related events compared to future events or non-causally related past events. This reactivation was specific to causal links, suggesting boundaries serve as windows for integrating past and current experiences.
Causal Distance and Pattern Switching: The strength of neural pattern changes (event boundary strength) in the AG (and marginally in mPFC) was positively correlated with the weighted path length in the causal network. This indicates that transitions between events that are less causally connected (greater network distance) trigger stronger neural switching, while closely connected events result in weaker transitions.

Significance and Claims
The paper claims that causality plays a central role in narrative comprehension and memory, potentially superseding temporal and semantic factors in organizing recall. The findings suggest that the DMN performs the mechanisms required to construct complex causal networks:

Stitching: Reactivating causally related prior events at boundaries to link them into hierarchical threads.
Switching: Modulating the strength of pattern changes based on the causal distance between events, facilitating the separation of unrelated events.

The authors propose that humans utilize a "cognitive graph" of causal associations to navigate narratives, similar to how cognitive maps are used for spatial navigation. These results provide neural evidence that the brain actively constructs causal structures to support the comprehension and retrieval of real-world memories, positioning the DMN as a critical substrate for this process. The study concludes that causal structure should be a central component in conceptualizations of memory organization.

More like this