Source-linked AI summary
The specificity and robustness of long-distance connections in weighted, interareal connectomes
Richard F. Betzel, Danielle S. Bassett
TL;DR
The paper asks whether costly long-distance brain connections are generic topological shortcuts or instead serve a more specific functional role. Using five weighted interareal connectomes and linearized Wilson-Cowan simulations, it finds that these connections contribute minimally to shortest-path distance while diversifying connectivity and supporting functional diversity. The authors conclude that long-distance connections are not merely shortcuts but help support diverse and complex brain dynamics.
Problem
The study addresses whether long-distance connections primarily reduce topological distance, despite evidence that weighted shortest paths favor strong short-range connections and that long-distance architecture is specific.
Method
The authors analyze five weighted interareal network datasets and use linearized Wilson-Cowan dynamics to compute neural-activity covariance and functional diversity.
Results
Across species and scales, long-distance connections contribute minimally to shortest-path communication, have markedly different neighbor connectivity profiles, exhibit similar connectivity patterns, and their removal decreases mean participation coefficient.
Takeaways & Limitations
Long-distance connections are not merely topological shortcuts; they introduce diverse neighbor configurations and support diverse and complex brain dynamics.
Takeaways & Limitations
The study assumes that structural networks alone can support meaningful claims about the functional properties of nervous systems and brain areas.
Abstract
from arXiv · showhide
Brain areas' functional repertoires are shaped by their incoming and outgoing structural connections. In empirically measured networks, most connections are short, reflecting spatial and energetic constraints. Nonetheless, a small number of connections span long distances, consistent with the notion that the functionality of these connections must outweigh their cost. While the precise function of these long-distance connections is not known, the leading hypothesis is that they act to reduce the topological distance between brain areas and facilitate efficient interareal communication. However, this hypothesis implies a non-specificity of long-distance connections that we contend is unlikely. Instead, we propose that long-distance connections serve to diversify brain areas' inputs and outputs, thereby promoting complex dynamics. Through analysis of five interareal network datasets, we show that long-distance connections play only minor roles in reducing average interareal topological distance. In contrast, areas' long-distance and short-range neighbors exhibit marked differences in their connectivity profiles, suggesting that long-distance connections enhance dissimilarity between regional inputs and outputs. Next, we show that -- in isolation -- areas' long-distance connectivity profiles exhibit non-random levels of similarity, suggesting that the communication pathways formed by long connections exhibit redundancies that may serve to promote robustness. Finally, we use a linearization of Wilson-Cowan dynamics to simulate the covariance structure of neural activity and show that in the absence of long-distance connections, a common measure of functional diversity decreases. Collectively, our findings suggest that long-distance connections are necessary for supporting diverse and complex brain dynamics.
INTRODUCTION
Brain networks balance predominantly short, low-cost wiring against a small set of costly long-distance connections whose function remains debated. Across five weighted connectomes, the paper argues that long-distance connections diversify regional inputs and outputs rather than primarily serving as generic topological shortcuts.
- INTRODUCTION: Most brain connections are short because spatial embedding keeps total wiring cost low, although a small proportion of long connections may provide additional functionality.Low wiring cost limits the material and metabolic expense of forming, using, and maintaining connections.
- INTRODUCTION: The prevailing account treats long-distance connections as bridges that reduce topological distance and facilitate rapid, efficient interareal communication.The paper questions this account because weighted shortest paths are dominated by strong short-range connections and because generic shortcuts imply non-specificity.
- INTRODUCTION: Empirical evidence that long-distance architecture is conserved across and replicable within individuals instead suggests that these connections are highly specific.This specificity challenges the idea that any long-distance connection is functionally interchangeable with any other.
- INTRODUCTION: The study analyzes five weighted interareal network datasets spanning mouse, Drosophila, macaque, and human connectomes to characterize long-distance connectivity.It first examines spatio-structural architecture and reports consistency across species.
- INTRODUCTION: Long-distance connections are proposed to diversify specific brain areas’ inputs and outputs, thereby promoting complex network dynamics.The paper tests this proposal by comparing long-distance and short-range neighbors and by simulating neural dynamics.
RESULTS
Across five weighted interareal connectome datasets, distance consistently shapes network architecture, while long-distance connections contribute little to shortest weighted paths. Instead, they distinguish regional connectivity profiles, show non-random redundancy, and support functional diversity in dynamical simulations.
- Distance shapes weighted network architecture: Connection weights and connectivity-profile similarity were negatively correlated with Euclidean distance across all five datasets.Both relationships were statistically significant with maximum p < 10−15, FDR-corrected.
- Similarity of connection length distributions shapes areal function: Connection-length distributions varied across cortical areas, with dorsal attention and limbic systems showing the greatest diversity in both human datasets.Diversity was quantified using the interquartile range of each area’s connection-length distribution.
- Similarity of connection length distributions shapes areal function: Clusters based on connection-length distributions recapitulated aspects of functional-system organization, linking spatial connection statistics with regional function.The analysis compared k-means clusters of connection-length distributions with functional-system assignments.
- Long connections contribute little to shortest, weighted paths: Long-distance connections played relatively minor roles in shortest-path structure, while removing strong short connections had consistently greater effects than removing long connections.This pattern held across the weighted interareal datasets and across definitions using the shortest or longest 5%, 10%, 20%, and 25% of connections.
- Distance shapes weighted network architecture: Long-distance and short-range neighbors had more dissimilar connectivity profiles than expected by chance, suggesting that long connections provide unique inputs and novel output targets.Standardized similarity-score distributions were consistently negative, with the cumulative distribution reaching 95% before a positive value in most datasets and thresholds.
- Redundancy of long-distance connectivity: Mean similarity among areas’ long-distance connectivity profiles consistently exceeded randomized-network values, indicating non-random redundancy in long-distance architecture.The comparison preserved degree sequence and edge-weight distribution exactly while approximately preserving connection-length structure and its relation to weight.
- Long-distance connections lead to diverse patterns of functional coupling: Removing long-distance connections decreased average participation coefficient, whereas removing the same number of short-range connections increased it across all network datasets.The participation coefficient was computed from covariance matrices generated by a linearized Wilson-Cowan model.
DISCUSSION
Across five weighted interareal connectomes, the paper argues that long-distance connections are not primarily topological shortcuts but instead diversify connectivity and support complex dynamics. The discussion also qualifies these conclusions by noting uncertainties in connection-weight interpretation, composite network representativeness, edge weighting, and tractography.
- Linearized dynamics and participation coefficient: Removing long-distance connections decreases mean participation coefficient, whereas removing the same number of short-range connections increases it across all network datasets.Participation coefficient is used as a measure of functional diversity in covariance structures simulated from linearized Wilson-Cowan dynamics.
- Interpreting the functional roles of long-distance connections: Shortest paths are dominated by strong short-range connections, so long-distance connections contribute minimally to reducing average interareal topological distance.This challenges the traditional view of long-distance connections as the brain’s main integrative shortcuts.
- Interpreting the functional roles of long-distance connections: Long-distance neighbors have markedly different connectivity profiles from short-range neighbors, allowing regions to interact through novel input-output configurations across species and scales.The consistency of this pattern is suggestive of an evolutionarily conserved communication mechanism.
- The specificity of long-distance connections: Long-distance connectivity profiles show non-random similarity beyond degree, distance, weight, and weight-distance distributions, indicating redundant pathways and an underlying latent structure.Possible contributors include spatial location, cytoarchitecture, transcriptional similarity, higher-order topology, and developmental timing.
- Interpreting connection weights: The conclusions depend on interpreting structural connection weights functionally, although weights may not directly reflect interareal communication efficacy.Communication also depends on axonal diameter and myelination, and the appropriate edge-weighting scheme remains unknown.
- Limitations: The study assumes that structural networks alone support meaningful claims about functional properties and treats shortest-path structure as functionally important despite ongoing challenges to that view.Disruptions to shortest paths have been associated with disease, motivating their study despite the role remaining unclear.
- Limitations: The networks are composites of many single-subject observations, while diffusion imaging and tractography have biases that complicate detection of long-distance corticocortical tracts.The paper also notes that results may depend on the unknown correct edge-weighting scheme.
CONCLUSION
Long-distance connections are presented as more than topological shortcuts: they diversify neighbors, support functional diversity, and exhibit similarity patterns that may promote robustness.
- Long-distance connections introduce diversity among brain areas’ neighbors and can be related to brain function in human data.
- Removing long-distance connections decreases functional diversity in simulations of brain dynamics.
- Long-distance connections exhibit degeneracies, with many areas showing similar long-distance connectivity patterns.
- The authors speculate that this degeneracy may confer robustness to the system.
MATERIALS AND METHODS
The study analyzes weighted interareal networks across multiple species and imaging or reconstruction strategies to assess the generality of its findings.
- The analysis covers mouse, Drosophila, macaque, and human weighted interareal network datasets.
- The datasets differ in imaging modality, reconstruction technique, and connection weighting scheme.
- This processing variability was used to examine whether findings were universal and robust to acquisition and processing schemes.
Network datasets
The study combines structural network datasets from mouse, Drosophila, macaque, and human brains, using species-specific reconstruction procedures and weighted interareal representations.
- Mouse: The mouse network contained N = 112 areas linked by weighted, directed interareal axonal projections.Weights represented normalized connection densities.
- Drosophila: The Drosophila network was reconstructed from images of 12,995 projection neurons in the female brain.
- Macaque: The macaque dataset was based on retrograde tract-tracing experiments and analyzed a [29 × 29] weighted, directed connectivity matrix.
- Human: Human structural networks were reconstructed from diffusion-weighted MRI using deterministic tractography and group-representative composites of 30 subjects.
- Human: Human parcellations included low- and high-resolution atlases with Nlow = 82 and Nhigh = 1000 areas.
Network analysis
Network analysis represents interareal connectivity through weighted profiles, spatial distances, shortest paths, modular structure, and measures of neighbor diversity and long-distance redundancy.
- Connectivity representation: Weighted connectivity matrices represent connection strengths as wij, while directed-area profiles include both incoming and outgoing connections.
- Shortest weighted paths: Shortest weighted paths are computed after transforming connection weights into lengths with lij(α) = w−α.
- Shortest weighted paths: Mean weighted path length measures the average cost of using shortest paths for communication.
- Connectivity-profile similarity: Cosine similarity compares connectivity profiles to assess functional relatedness and the uniqueness of short- versus long-distance neighbors.
- Long-distance redundancy: Long-distance structural degeneracy is quantified by average pairwise cosine similarity among connectivity profiles in the long-distance-only network, where larger values indicate greater degeneracy.
- Modularity maximization: Modularity maximization partitions nodes into modules by maximizing within-module excess connectivity relative to a null model.The objective uses bij = wij−γ·pij, with γ controlling the relative importance of expected weights and detected module resolution.