Figure 1. Overview of the data analysis system
Daniel A. Wagenaar, Thomas B. DeMarse, Steve M. Potter
As part of our lab's effort to study learning in vitro by connecting living neural networks to a simulated body in a simulated environment (the Animat project), we have developed a set of tools for online and offline analysis of data collected from MEAs (multi electrode arrays).
We grow cultures of thousands of neurons dissociated from cortex of E18 rat embryo on dishes with a rectangular array of 60 microelectrodes spaced at 200 um. Although such cultures are enormously simplified model systems when compared to the intact brain, they still exhibit rich dynamics at a variety of timescales.
Recording from 60 electrodes at 25 kHz produces a data stream of 3 MB/s, so it is clear that significant online data reduction is a prerequisite for experiments of any duration. We developed a fully extendible modular toolset for online MEA data analysis, featuring spike detectors tailored to data streams with large stimulation artifacts, online data inspection, and hooks for connecting online feedback stimulation. These tools are implemented as a set of C++ programs with a common core library which we run under Linux on a four CPU SGI machine.
We used this system to make 20 minute recordings of spontaneous spiking activity in two dishes on 30 consecutive days starting on the second day after plating cells. Most of this activity occurred in dishwide bursts of varying magnitude and duration. There was large day-to-day variation in bursting behavior. On some days, the culture bursted quite regularly about 30 times per minute. On other days, the burst intervals were longer, or much less regular. Finally, on some days bursts mainly occurred in `superbursts' - periods lasting upto 30 seconds with burst rates as high as 60 bpm.
The spatiotemporal fine structure of bursts as well as inter-burst activity are currently under investigation.
