Short Excitatory Connections Turn Wiring Constraints Into Bursting, Modular Brain-Like Dynamics

Short excitatory projections in the brain have long been viewed as a compromise forced by physics: wiring that stays local because long fibers cost space, energy, and material. A new computational study argues the opposite, showing that these brief connections are not just tolerated by neural networks but actively reshape how they behave. In simulations of balanced excitation and inhibition, restricting a fraction of excitatory neurons to short, distance-dependent projections amplified responsiveness, drove burst activity, and produced modular functional organization without any pre-built modular structure.

The research, published in Frontiers in Computational Neuroscience, comes from a team led by Claus C. Hilgetag of University Medical Center Hamburg-Eppendorf and Boston University, together with colleagues at ShanghaiTech University and Nantong University. The authors built conductance-based integrate-and-fire networks of 5,000 neurons, 4,000 excitatory and 1,000 inhibitory, embedded in a two-dimensional sheet with periodic boundaries. Neurons were randomly positioned, and the global connection probability was held at 0.1. The key manipulation was simple: a chosen proportion of excitatory neurons, called SEP neurons, had their outgoing connections rewired so that connection probability decayed with physical distance, while the remaining neurons kept distance-independent random connectivity.

Critically, the rewiring preserved each neuron’s out-degree. The number of outgoing projections per neuron stayed identical to the baseline random network; only where those projections landed changed. This control matters because it isolates the effect of spatial targeting from any change in how much a neuron connects. SEP neurons were scattered uniformly across the entire sheet rather than clustered into a sub-region, and the network contained no predefined structural modules. The excitatory-inhibitory composition of 80 percent and 20 percent follows experimentally reported cortical values, and the model parameters were adapted from the widely used Vogels and Abbott framework for balanced networks.

The biological motivation is well established. Connection probability in rat layer 2/3 neocortex drops systematically from 0.09 within 25 micrometers to 0.01 beyond 100 micrometers, and synapses from excitatory neurons typically sit a mean of only 67 micrometers from the postsynaptic soma. Many excitatory cell types, including layer 4 spiny stellate cells, have characteristically short axonal arbors and operate primarily within local columns. These neurons differ from long-projecting counterparts in gene expression, connectivity, firing properties, and developmental trajectories, which the authors interpret as functional specialization rather than developmental truncation.

When the researchers stimulated a single randomly selected neuron and measured how the perturbation spread, networks with 1,000 SEP neurons responded far more strongly than networks with 500. Response magnitude and propagation probability differed significantly across the majority of distance bins, and fitting an exponential decay to the propagation curves yielded a substantially larger decay length for the denser SEP condition. In other words, signals traveled farther and more reliably when more excitatory neurons kept their connections local. Neurons in the high-SEP network also showed both higher baseline activity and larger perturbation-induced changes, with a broader dynamic range visible in the survival distribution of firing rate changes.

The mechanism the authors propose centers on shared inputs. By concentrating outgoing connections among nearby neurons, SEP neurons increase the proportion of common presynaptic inputs that postsynaptic targets receive. This raises response correlations and creates a positive feedback loop that amplifies network activity. The finding aligns with theoretical predictions that distance-dependent recurrent connectivity in balanced excitatory-inhibitory networks generates spatially structured correlations. The authors frame SEPs as correlation amplifiers: structural elements that reshape network communication rather than redundant wiring.

That amplification showed up most vividly in burst dynamics. Networks with 1,000 SEP neurons generated significantly more frequent and longer bursts than those with 500, a difference robust across a wide range of burst-detection thresholds. More striking was the spatial pattern. Although SEP neurons were uniformly distributed and the architecture contained no clusters, burst events formed distinct spatial clusters that emerged spontaneously from the dynamics. The locus of burst activity was not fixed; it drifted across the two-dimensional neural field over the course of the simulation. Tracking cluster identity across consecutive 500-millisecond windows showed that clusters in high-SEP networks persisted over significantly more windows, making them both larger and more temporally stable even as their location migrated.

To rule out hidden structural modularity, the team computed Louvain modularity on both structural and functional connectivity across SEP densities of 0, 500, and 1,000, plus a benchmark network with 10 predefined modules, using 100 realizations per condition. Structural modularity stayed uniformly low at every density, including SEP = 1,000, with a value of 0.066 plus or minus 0.001, well below the conventional threshold of 0.3. Functional modularity was substantially higher at 0.158 plus or minus 0.155 across all realizations, and 0.27 plus or minus 0.10 excluding low-activity trials, and was uncorrelated with structural modularity within each condition. The correlation coefficient was 0.01 at SEP = 1,000, and the lack of correspondence held at the level of individual realizations across all three conditions. The clustering, in short, reflected SEP-driven recurrent dynamics rather than latent structure.

The effect depended specifically on locality. When the researchers fixed the number of SEP neurons at 1,000 and instead extended the spatial decay parameter of their outgoing connections, both burst count and functional modularity declined progressively, approaching the levels seen in networks lacking local SEP structure altogether. An activity-matched control, in which a 500-SEP network was driven to the same population rate as the 1,000-SEP network, showed only a marginal increase in functional modularity, confirming that SEP structure itself, not elevated firing rate, drives the emergence of column-like organization. The authors also note a boundary condition: when recurrent excitation grows strong enough to push the network into a hypersynchronous, seizure-like state, bursting spreads network-wide, correlation contrast between modules collapses, and functional modularity falls back toward lower-SEP levels. The modular regime they describe sits intermediate between quiescent weakly correlated activity and global hypersynchrony.

Overall excitability scaled non-linearly with SEP density. The maximal firing rate per network rose from below 5 Hz to approximately 30 Hz across the tested range, and trial-to-trial variability in firing rates increased with higher SEP density, suggesting greater sensitivity to initial conditions and local fluctuations. The authors are careful about scope. The model is abstract: it does not capture dendritic nonlinearities, active conductances, backpropagating action potentials, or the effects of short inhibitory projections, and the correspondence with cortical columns is offered as a structural hypothesis rather than a direct finding, since real columns also involve laminar architecture and sensory selectivity. Still, the convergence is suggestive. Recent synapse-resolution analysis of Drosophila connectomes shows that empirical power-law distance dependence maximizes information communication and sustains a balance between integration and segregation, consistent with the view that short-range wiring may be a broader design principle. Within the model class tested, the central results also reproduced in networks twice as large as those used in the main analyses. The study points toward a parsimonious account of how the brain might convert a simple physical constraint, short wires, into the machinery of column-like organization, burst propagation, and distributed computation, all from a single distance-dependent rule.

Subject of Research: Computational modeling of how short excitatory projections shape burst activity and modular functional organization in balanced neural networks

Article Title: Short excitatory projections promote burst activity and modular functional organization in neural networks

Article References: Hussainyar, M. S., Yun, D., Li, D., Guan, J.-S., & Hilgetag, C. C. (2026). Short excitatory projections promote burst activity and modular functional organization in neural networks. Frontiers in Computational Neuroscience, 20, Article 1965686. https://doi.org/10.3389/fncom.2026.1965686

Image Credits: AI Generated

DOI: 10.3389/fncom.2026.1965686

Keywords: short excitatory projections, neural networks, burst activity, functional modularity, cortical columns, shared presynaptic inputs, signal propagation, balanced excitation and inhibition, computational neuroscience, integrate-and-fire model, distance-dependent connectivity, recurrent amplification