Dynamic functional connectivity patterns associated with dementia risk

Sophie Dautricourt, Julie Gonneaud, Brigitte Landeau, Vince D. Calhoun, Robin de Florès, Géraldine Poisnel, Salma Bougacha, Valentin Ourry, Edelweiss Touron, Elizabeth Kuhn, Harriet Demintz-King, Natalie L. Marchant, Denis Vivien, Vincent de La Sayette, Antoine Lutz, Gaël Chételat, Eider M. Arenaza‐Urquijo, Florence Allais, Claire André, Julien Asselineau, Alexandre Bejanin, Pierre Champetier, Anne Chocat, Marion Delarue, Stéphanie Egret, Francesca Felisatti, Eglantine Ferrand Devouge, Éric Frison, Marc Heidmann, Thien Huong Tran, Gwendoline Le Du, Valérie Lefranc, Florence Mézenge, Inès Moulinet, Cassandre Palix, Léo Paly, Anne Quillard, Géraldine Rauchs, Stéphane Rehel, Florence Requier, Caitlin Ware, Sebastian Baez Lugo, Olga Klimecki, Patrik Vuilleumier, Thorsten Barnhofer, Fabienne Collette, Éric Salmon, Pascal Delamillieure, Martine Batchelor, Axel Beaugonin, Francis Gheysen, Harriet Demnitz‐King, Tim Whitfield, Corinne Schimmer, Miranka WirthAlzheimer 2022 · DOI

TL;DR

This study moves beyond "static" snapshots of brain activity to show that the way our brain networks fluctuate over time—known as dynamic functional network connectivity (DFNC)—is a sensitive indicator of dementia risk. By analyzing 127 cognitively healthy older adults, researchers found that the time spent in specific "connectivity states" correlates strongly with modifiable risk factors like LDL cholesterol and blood pressure, as well as markers of cognitive reserve.

What Problem Does This Paper Address?

While we know that modifiable risk factors (like cardiovascular health and lifestyle choices) account for nearly 40% of dementia cases, the exact neural mechanisms remain fuzzy. Most neuroimaging research relies on static functional connectivity, which averages brain activity over several minutes. This approach assumes the brain’s "wiring diagram" is constant during a scan.

However, the brain is inherently dynamic. The authors argue that by ignoring these moment-to-moment fluctuations, we miss crucial information about how the brain maintains "cognitive reserve"—the ability to withstand age-related changes or disease. This paper investigates whether the dynamics of three key networks—the Default Mode (DMN), Salience (SN), and Executive Control (ECN)—can serve as a better barometer for dementia risk in people who don't yet show symptoms.

Methods

The researchers employed a multi-step pipeline to capture brain dynamics:

  1. Resting-state fMRI: Participants were scanned while lying still.
  2. Independent Component Analysis (ICA): This was used to decompose the brain data into 30 components, from which 7 were selected to represent the DMN, SN, and ECN.
  3. Sliding Window Analysis: Instead of one average, they used a 43-second "window" that slid across the 10-minute scan in 2.4-second increments, creating a series of connectivity matrices over time.
  4. K-means Clustering: They grouped these thousands of matrices into four distinct "states" or recurring patterns of connectivity.
  5. Stepwise Regression: Finally, they modeled how time spent in these states related to 15 dementia risk/protective factors (e.g., BMI, LDL, cognitive activity) and neuroimaging biomarkers (e.g., amyloid load, brain perfusion).

Datasets

The study utilized baseline data from the Age-Well randomized controlled trial, part of the Medit-Ageing European project.

  • Participants: 127 cognitively unimpaired older adults (mean age ~69).
  • Validation: To ensure the results weren't a fluke of their specific brain map, they replicated the analysis using the Neuromark atlas, a standardized template for brain networks.

Key Insights and Ideas

  • Dynamics Trump Statics: One of the most important findings was that DFNC showed significant associations with risk factors where static connectivity showed none. This suggests that the flexibility and timing of network interactions are more sensitive to early pathology than the average strength of those connections.
  • The "Weakly Connected" Trap: State 1 (the most frequent state, occurring 51% of the time) was characterized by very low connectivity across all networks. Spending more time here was linked to lower lifetime cognitive activity and higher LDL cholesterol. This might represent a "low-efficiency" mode where the brain struggles to coordinate complex tasks.
  • Connectivity as Reserve: State 3 ("Strongly Connected") was the "healthy" counterpart. Time spent in this state correlated with higher scores on the Preclinical Alzheimer’s Cognitive Composite (PACC5) and higher early-life cognitive activity. This suggests that a brain capable of entering and maintaining high-connectivity states has better cognitive reserve.
  • The Perfusion Link: A lower number of transitions between states was strongly associated with lower brain perfusion (blood flow). This provides a direct physiological link between vascular health and the brain's ability to switch between functional configurations.

Key Results

The study identified four recurring states:

  • State 1 (51%): Weakly connected. Linked to increased risk (high LDL, low cognitive activity).
  • State 2 (18%): SN-negatively connected. Linked to increased risk (high blood pressure, depression, low BMI).
  • State 3 (16%): Strongly connected. Linked to reduced risk (high early-life cognitive activity, better PACC5 scores).
  • State 4 (15%): DMN-negatively connected. Linked to reduced risk (high mid-life stimulating activities, lower LDL).

Quantitatively, the "Strongly Connected" state (State 3) was a significant predictor of cognitive performance ($\beta = 1.80, p = 0.02$). Furthermore, the "flexibility" of the brain—measured by the number of state transitions—was a robust predictor of brain perfusion ($\beta = 25.4, p = 0.0001$).

Limitations and Future Work

The authors acknowledge several caveats:

  • Directionality: Because the study is cross-sectional, they cannot prove that high LDL causes the brain to spend more time in a weakly connected state; they only know they happen together.
  • Scope: The analysis focused on only three networks. While these are the "big players" in Alzheimer's, other networks (like the visual or motor networks) might also show relevant dynamic changes.
  • Missing Factors: Some known risk factors, such as air pollution and hearing loss, were not included in the models.

Future research should look at longitudinal data to see if these dynamic patterns can actually predict who will decline cognitively over several years, potentially serving as a "early warning system" for clinical trials.

Summary updated

Abstract

DFNC states are differentially associated with dementia risk and could underlie reserve.
BibTeX
@inproceedings{dautricourt2022dynamic,
  author = {Sophie Dautricourt and Julie Gonneaud and Brigitte Landeau and Vince D. Calhoun and Robin de Florès and Géraldine Poisnel and Salma Bougacha and Valentin Ourry and Edelweiss Touron and Elizabeth Kuhn and Harriet Demintz-King and Natalie L. Marchant and Denis Vivien and Vincent de La Sayette and Antoine Lutz and Gaël Chételat and Eider M. Arenaza‐Urquijo and Florence Allais and Claire André and Julien Asselineau and Alexandre Bejanin and Pierre Champetier and Gaël Chételat and Anne Chocat and Sophie Dautricourt and Robin de Florès and Marion Delarue and Stéphanie Egret and Francesca Felisatti and Eglantine Ferrand Devouge and Éric Frison and Julie Gonneaud and Marc Heidmann and Thien Huong Tran and Elizabeth Kuhn and Gwendoline Le Du and Brigitte Landeau and Valérie Lefranc and Antoine Lutz and Florence Mézenge and Inès Moulinet and Valentin Ourry and Cassandre Palix and Léo Paly and Géraldine Poisnel and Anne Quillard and Géraldine Rauchs and Stéphane Rehel and Florence Requier and Edelweiss Touron and Denis Vivien and Caitlin Ware and Sebastian Baez Lugo and Olga Klimecki and Patrik Vuilleumier and Thorsten Barnhofer and Fabienne Collette and Éric Salmon and Vincent de La Sayette and Pascal Delamillieure and Martine Batchelor and Axel Beaugonin and Francis Gheysen and Harriet Demnitz‐King and Natalie L. Marchant and Tim Whitfield and Corinne Schimmer and Miranka Wirth},
  title = {{Dynamic functional connectivity patterns associated with dementia risk}},
  booktitle = {Alzheimer},
  year = {2022},
  doi = {10.1186/s13195-022-01006-7},
}
Figure