Self-organized and self-sustained ensemble activity patterns in simulation of mouse primary motor cortex

Abstract

The idea of self-organized signal processing in the cerebral cortex has become a focus of research since Beggs and Plentz¹ reported avalanches in local field potential recordings from organotypic cultures and acute slices of rat somatosensory cortex. How the cortex intrinsically organizes signals remains unknown. A current hypothesis was proposed by the condensed matter physicists Bak, Tang, and Wiesenfeld² when they conjectured that if neuronal avalanche activity followed inverse power law distributions, then brain activity may be set around phase transitions within self-organized signals. We asked if we would observe self-organized signals in an isolated slice of our data driven detailed simulation of the mouse primary motor cortex? If we did, would we observe avalanches with power-law distributions in size and duration and what would they look like? Our results demonstrate that a brief unstructured stimulus (100ms, 57μA current) to a small subset of neurons (about 181 of more than 10,000) in a simulated mouse primary motor cortex slice results in self-organized and self-sustained avalanches with power-law size and duration distributions and values similar to those reported from in vivo and in vitro experiments. We observed 4 cross-layer and cross-neuron population patterns, 3 of which displayed a dominant rhythmic component. Avalanches were each composed of one or more of the 4 population patterns.

Figure 1.

Raster plot of excitatory neuron firing in the first 1 minute of a 10 min M1 simulation. (A) Continuous (wrapped) 60 s raster plot (24 rows of 2.5 s each; red: L2–4; black: L5–6) Sustained activity was initiated by a brief stimulus (100 ms; 0.57 nA/cell, 1.8% of cells; top left). Asterisks denote the start of selected delta events. Row 1 line is expanded in B; row 3 line above beta event; row 8 line above gamma event. (B) Detail of first 1.64s (row 1 blue line in A).

Fig 2.

Recurring population events. A-C. Raster plots for delta, beta, gamma patterns respectively; red: excitatory; blue: inhibitory cells). D-F. Average spike histograms showing 4 periods of delta, 3 of beta, 3 of gamma (note different time scales) G-I. Characteristic initiation for each pattern. (in D-I -- IT2/3:blue; IT4:yellow; IT5A:green; IT5B:purple; PT5B:red; IT6:brown)

Fig 3.

Self-sustained activity showed oscillatory ~1 Hz activity. (A) Instantaneous spike rates over 10 min, all neurons; mean firing rate is 4.7 Hz (horizontal line); CV 1.53. (B) Detail of first 100 s above red line in A. (C) Spectrograms (top 0–50 Hz; middle 0–5 Hz) and raster (10 s, red line in B; population labels at right).

Fig 4.

Oscillatory activity patterns make up individual avalanches. Continuous sequences of raster plots divided into individual avalanches. Top row is one continuous sequence. Bottom two rows compose another continuous sequence. A, B. Spike rate spectrograms of segments marked above show strong gamma activity, beta activity, respectively.

Figure 5.

Power-law fits for all neurons in our 10 minute simulation of an M1 cortical column. A. Avalanche size probability density distribution with the number of avalanches normalized (y-axis) and the size of the avalanche (number of spikes; x-axis). Power-law fit equals −1.52 (red dashed line; sigma = 0.035, D = 0.028). B. Avalanche duration probability density distribution showing the number of avalanches normalized (y-axis) and their durations (milliseconds; x-axis). Power-law fit equals −1.61 (red dashed line; sigma = 0.088, D = 0.067). C. Same as A except only Irregular and Fragment avalanches. Power-law fit equals −1.53 (sigma = 0.036, D = 0.033). D. Same as A except only Irregular and Fragment avalanches. Power-law fit equals −1.39 (sigma = 0.045, D = 0.063). Analyzed using the Python powerlaw package.

Figure 6.

Log-log probability density distributions of avalanche size (first column) and duration (second column) for all avalanches (gray) and each of four avalanche types.