by Donald W Doherty
with Salvador Dura-Bernal and William W Lytton
Department of Physiology & Pharmacology
SUNY Downstate Medical Center, Brooklyn, NY 10962 USA
Avalanches display non-Poisson distributions of correlated neuronal activity that may play a significant role in signal processing. The phenomenon appears robust but mechanisms remain unknown due to an inability to gather large sample sizes, and difficulties in identifying neuron morphology, biophysics, and connections underlying avalanche dynamics. We set out to understand the relationship between power-law activity patterns, their values, and the neural responses observed from every neuron across different layers, cell populations, and the entire cortical column using a detailed mouse primary motor cortex model with 15 neuron types that simulated the full-depth of a 300μm diameter column with 10,073 neurons and ~18e6 connections. Self-organized and self-sustained activity from our simulations have power-law values which are in the range of -1.5 for avalanche size and -2.0 for duration as noted for both in vitro and in vivo neural avalanche preparations reported by Beggs and Plentz (2003).
We applied a 0.57 nA, 100 ms stimulus across 40 μm in diameter and full column depth at each of 49 gridded locations (40 μm) across the pia surface of our 400 μm diameter cylindrical cortical column. Stimuli applied to 4 locations (8.2%) produced no sustained responses. Self-sustained activity was seen in the other 45 locations, which always included activity in IT5B or IT5B and IT6. In 6 locations activity was restricted to IT5B or IT5B/IT6 alone (subcritical). Intermittent spread of activity from IT5B/IT6 across other neuron types and layers was seen in 24 locations (subcritical). In 15 locations, frequent spread of activity to other neuron types and layers was observed (avalanche size: ~-1.5). Cascades and avalanches were defined using binned spiking activity (1 ms bins). Each cascade or avalanche was composed of adjacent bins filled with one or more action potentials, preceded and followed by at least one empty bin. If the size distribution of spike clusters fit a power-law value of ~-1.5, then the clusters were termed avalanches. Otherwise, the spike clusters were termed cascades.
A prolonged 10 minute M1 simulation with different connectivity produced 15,579 avalanches during sustained activity after the initial 100 ms stimulation. The data’s size distribution fit the power-law value -1.51 and their durations fit -1.98. Again, IT5B/IT6 activity was constant and punctuated by more widespread activity. Four distinct patterns of activity spontaneously recurred and could be characterized by either 1) irregular spiking or 2) delta, 3) beta, or 4) gamma frequency-dominance. All large-scale avalanches were composed of 1 or a combination of these 4 recognizable patterns. Between the large-scale avalanches we saw smaller avalanches composed of irregular spiking activity (continuous IT5B/IT6 neuron activity) or fragments of frequency-dominated activity that sometimes transitioned to continuous IT5B and IT6 activity. Since cortical column activity with just IT5B and IT6 activity showed little correlation (very steep and narrow distributions of avalanche sizes and durations), we hypothesize that the addition of fragments of frequency-dominated activity result in more correlated activity and power-law values closer to -1.5 and -2.0 for size and duration respectively. In conclusion, an increase in frequency-dominated activity may turn neuronal cascades into neuronal avalanches.