Summary: A new model gives a richer, testable understanding of how activity is coordinated between neurons in neural circuits and promises improved prediction of brain dynamics.
Source: University of Pittsburgh
New model promises better prediction of brain dynamics by extending balanced network theory
For decades neuroscientists have worked to explain the brain’s apparently noisy and variable activity. Over the past 20 years, balanced network theory—based on a tight balance of excitation and inhibition in recurrently connected neural networks—has become a leading framework for explaining that variability. Researchers at the University of Pittsburgh have now extended this theory to incorporate realistic spatial wiring, producing a model that offers deeper, experimentally testable predictions about correlated neuronal variability and brain dynamics.
The expanded model, reported online in Nature Neuroscience on October 31, accurately accounts for a range of experimental observations about the highly variable responses of neurons in awake animals. By including distance-dependent connectivity and the spatial structure of synaptic projections, the model explains how neuronal correlation patterns vary across space and time and how those patterns arise from local circuit architecture.
“Brain activity often looks highly random, which seems like an odd way to perform computations,” said Brent Doiron, associate professor of mathematics at the University of Pittsburgh and senior author on the paper. “To understand how neural computation emerges, we need to know how network dynamics depend on network architecture. This work brings us significantly closer to that goal.”
Previous balanced network models captured how the timing and strength of excitatory and inhibitory inputs shape variability, but they relied on simplifying assumptions that ignored the fact that real brains have distance-dependent wiring. Neurons close to one another are far more likely to connect than neurons separated by larger distances, and those spatial statistics affect how activity propagates through a circuit.
“Early models either produced activity that was effectively random, unlike the brain, or activity that was overly synchronized, similar to a seizure,” Doiron explained. “They couldn’t generate the nuanced patterns we see in cortex.” Matthew Smith, assistant professor of ophthalmology and a co-author, added: “In a balanced regime neurons sit in a delicate state — like balancing on tiptoe — so small perturbations cause large fluctuations in firing. Capturing that sensitivity requires realistic network structure.”
The updated model explicitly incorporates both temporal dynamics and spatial connectivity, and it predicts the correlations between neuronal firing as a function of distance. When the authors compared model predictions to electrophysiological recordings from the superficial layers of visual cortex, they found close agreement. Nearby neuron pairs showed strong positive correlations; pairs at intermediate separations were negatively correlated (when one fired more the other fired less); and pairs far apart were largely independent. This distinctive, non-monotonic spatial profile of correlated neuronal variability emerges naturally from networks with balanced excitation and inhibition and distance-dependent lateral projections.
The correspondence between model and data suggests the approach can be used to identify neural signatures that predict changes in brain state, learning, attention, or disease. “Any serious theory of brain computation must account for noise in the neural code,” Doiron said. “Shifts in neuronal variability accompany attention and learning and are also signatures of disorders such as Parkinson’s disease and epilepsy. A model that links connectivity to variability helps interpret those signals.”
While the study focused on visual cortex recordings, the investigators anticipate that the model generalizes across cortical areas and mammalian species. The same principles should apply to areas involved in auditory, olfactory, or somatosensory processing, wherever lateral projections and local wiring shape correlated variability. Independent recordings from mouse visual cortex have revealed similar spatial correlation patterns, supporting the broader applicability of the theory.
Authors on the paper include Brent Doiron and Matthew Smith, with contributions from Jonathan Rubin (mathematics), Robert Rosenbaum (now at the University of Notre Dame), and Adam Kohn (Albert Einstein College of Medicine). The research was supported by National Science Foundation grants under the BRAIN Initiative, as well as funding from the National Eye Institute, Research to Prevent Blindness, the Eye and Ear Foundation of Pittsburgh, and the Simons Foundation.
Source: John Fedele, University of Pittsburgh
The spatial structure of correlated neuronal variability
Shared variability is common across cortical populations, and it is often attributed to overlapping synaptic inputs. However, the precise link between correlated variability and local circuit architecture has remained unclear. Combining computational models with in vivo recordings, the authors extend balanced network theory to include distance-dependent connectivity. They show that localized lateral projections produce weak correlations, while broader lateral projections yield a characteristic spatial correlation pattern: nearby neuron pairs are positively correlated; pairs at intermediate distances are negatively correlated; distant pairs are weakly correlated. This non-monotonic relationship between correlation and distance is detected in recordings from macaque primary visual cortex. Incorporating spatially structured connectivity therefore significantly improves the ability of balanced network theory to explain correlated neuronal variability.
Reference: Robert Rosenbaum, Matthew A. Smith, Adam Kohn, Jonathan E. Rubin, and Brent Doiron. “The spatial structure of correlated neuronal variability.” Nature Neuroscience. Published online October 31, 2016. doi:10.1038/nn.4433
Image credit: NeuroscienceNews.com image used for illustrative purposes. Please cite the original research and the University of Pittsburgh press release when referencing this summary.