A novel dynamic network imaging analysis method reveals aging-related fragmentation of cortical networks in mouse

Daniel A. Llano, Chihua Ma, Umberto Di Fabrizio, Aynaz Taheri, Kevin A. Stebbings, Georgiy Yudintsev, Gang Xiao, Robert V. Kenyon, Tanya Berger‐WolfbioRxiv 2019 · DOI

TL;DR

This paper introduces CommDy, a novel dynamic network analysis tool adapted from social network theory, to study brain activity. They demonstrate that aging in mice leads to a significant fragmentation and reduced connectivity in cortical networks, particularly in the auditory and motor cortices, suggesting a potential link to synaptic dysfunction.

What Problem Does This Paper Address?

Understanding how brain networks change with age is crucial, as aging is associated with cognitive decline. While structural changes in the aging brain are well-documented, functional changes, especially in network connectivity, are less understood. Existing methods for analyzing functional connectivity often average data over long periods, losing valuable dynamic information. This paper aims to address this gap by developing and applying a new method to capture the dynamic nature of brain networks and investigate how aging impacts them.

Methods

The core of this work is the adaptation of the Community Dynamic Inference Method (CommDy) for neuroimaging data. CommDy, originally from social network theory, identifies dynamic communities (functional clusters) by minimizing changes in community affiliations over time.

Here's a breakdown of their approach:

  • Data Pre-processing: Raw brain imaging data (flavoprotein autofluorescence) was transformed into a network model where pixels served as nodes. A weighted and thresholded network of pixel value correlations was generated.
  • Window Size and Threshold Determination: They systematically tested different time window sizes (25-200 frames) and correlation thresholds to find the optimal settings that captured meaningful network activity without spurious correlations. They settled on a 50-frame window and a 0.7 correlation threshold.
  • Static Community Detection: For each time step, the Louvain algorithm was used to identify static communities within the correlation networks. Louvain was chosen for its intuitive steps and scalability.
  • Dynamic Community Inference: CommDy was then applied to the time series of static communities to infer dynamic communities, revealing how network structures evolve over time.
  • Network Metrics: They utilized 10 CommDy network metrics to quantitatively describe network activity, allowing for comparisons between different conditions (e.g., young vs. aged brains).
  • Statistical Analysis: Univariate and multivariate analyses (including mixed-effects models and PCA) were used to compare network metrics between young and aged mice and to explore relationships with other factors like cortical thickness.

Datasets

The study utilized two primary datasets:

  1. Brain slice imaging data: From the auditory cortex of young (5.5 months) and aged (22 months) mice during paroxysmal depolarizations.
  2. In vivo imaging data: From the awake motor cortex of young (average 4.7 months) and aged (average 23.2 months) mice.

No new benchmarks were introduced; rather, existing imaging techniques were adapted for the novel CommDy analysis.

Key Insights and Ideas

  • CommDy as a powerful tool for dynamic brain network analysis: The paper demonstrates that CommDy, adapted from social network theory, can effectively capture the dynamic nature of brain activity, providing insights that static analyses miss.
  • Aging leads to network fragmentation: A central finding is that aging in mice is associated with a significant fragmentation of cortical networks, particularly in the auditory and motor cortices. This fragmentation is characterized by reduced connectivity.
  • Aging-related network changes mirror NMDA receptor blockade: The study found that network changes observed in aged mice closely resemble those induced by blocking NMDA receptors in young mice. This suggests a potential synaptic mechanism underlying age-related network fragmentation.
  • Cortical thickness is linked to network structure: Cortical thickness was found to be a significant predictor of network size and cohesion, consistent with the known age-related thinning of the cortex and its implications for connectivity.

Key Results

  • Reduced connectivity in aged brains: In both slice and in vivo preparations, aged mice showed significantly lower average node degree (connectivity) compared to young mice. For auditory cortex slices, the mean normalized node degree was 813.0 in young vs. 149.1 in aged animals (p=0.005).
  • Distinct network dynamics in young vs. aged: Dynamic analysis revealed that young animals exhibited coherent clusters of activity that spread and coalesced, while aged animals showed more heterogeneous and fragmented activity patterns.
  • NMDA blockade recapitulates aging effects: A random forest classifier trained on NMDA blockade data could distinguish between young and aged mouse brain activity with 74% accuracy. This suggests NMDA receptor hypofunction might contribute to age-related network changes.
  • Cortical thickness correlation: Cortical thickness was significantly correlated with normalized community size (R=0.687, p<0.0001) and normalized group size (R=0.687, p<0.0001).

Limitations and Future Work

  • Technical Considerations: The flavoprotein autofluorescence imaging used has a relatively slow time course (~0.5-1 sec), which might limit the detection of very rapid neuronal events. Additionally, using pixels as nodes, rather than defined regions or individual neurons, is a simplification, though the authors estimate each pixel represents ~50-100 neurons.
  • Generalizability: While CommDy was applied to auditory and motor cortices, its application to other brain regions and imaging modalities (like fMRI) is a promising avenue for future work.
  • Mechanism of Fragmentation: The specific substrate for the observed aging-related fragmentation remains unknown, though NMDA receptor hypofunction is a strong candidate. Further research could explore dendritic branching, intrinsic excitability, and synaptic properties.
  • Cohesion vs. Connectivity: The study notes that while connectivity drops in aging, the loss of cohesiveness might not be solely explained by reduced connectivity and requires further investigation.
  • Potential Utilities: The authors suggest CommDy could be valuable for studying neurological and psychiatric disorders, especially those characterized by dynamic fluctuations in symptomatology.

Summary updated

Abstract

Abstract Network analysis of large-scale neuroimaging data has proven to be a particularly challenging computational problem. In this study, we adapt a novel analytical tool, known as the community dynamic inference method (CommDy), which was inspired by social network theory, for the study of brain imaging data from an aging mouse model. CommDy has been successfully used in other domains in biology; this report represents its first use in neuroscience. We used CommDy to investigate aging-related changes in network parameters in the auditory and motor cortices using flavoprotein autofluorescence imaging in brain slices and in vivo . Analysis of spontaneous activations in the auditory cortex of slices taken from young and aged animals demonstrated that cortical networks in aged brains were highly fragmented compared to networks observed in young animals. Specifically, the degree of connectivity of each activated node in the aged brains was significantly lower than those seen in the young brain, and multivariate analyses of all derived network metrics showed distinct clusters of these metrics in young vs. aged brains. CommDy network metrics were then used to build a random-forests classifier based on NMDA-receptor blockade data, which successfully recapitulated the aging findings, suggesting that the excitatory synaptic substructure of the auditory cortex may be altered during aging. A similar aging-related decline in network connectivity was also observed in spontaneous activity obtained from the awake motor cortex, suggesting that the findings in the auditory cortex are reflections of general mechanisms that occur during aging. Therefore, CommDy therefore provides a new dynamic network analytical tool to study the brain and provides links between network-level and synaptic-level dysfunction in the aging brain.
BibTeX
@misc{llano2019novel,
  author = {Daniel A. Llano and Chihua Ma and Umberto Di Fabrizio and Aynaz Taheri and Kevin A. Stebbings and Georgiy Yudintsev and Gang Xiao and Robert V. Kenyon and Tanya Berger‐Wolf},
  title = {{A novel dynamic network imaging analysis method reveals aging-related fragmentation of cortical networks in mouse}},
  year = {2019},
  doi = {10.1101/836817},
  howpublished = {bioRxiv},
}
Figure