Brewer GJ, Boehler MD, Ide AN, Wheeler BC (2009) Chronic electrical stimulation of cultured hippocampal networks increases spontaneous spike rates. Journal of Neuroscience Methods.
Neural cultures developing in vitro lack the incoming stimulation/information of their natural (in vivo) environment. In this paper Brewer and collegues applied chronic electrical stimulation to cultures of E18 rat hippocampal neurons developing on 60 channel multi electrode arrays.
30 uA paired pulses (50 ms ISI; biphasic, 100 us/phase duration, positive first) with a 5s wait between pulse pairs were delivered to 30 of the 60 electrodes, one-by-one, in a semi-random sequence for 0 (control), 1 or 3 hrs per day, at 7 11 12 14 18 19 and 21 days in vitro. There were five cultures in each condition. Three minutes of activity were recorded and analysed on day 21. My only problem with these methods is that the cultures recieved stimulation just before recordings were made. This confounds the 'chronic' impact of stimulation during network development - the observed effect could be acute. Recordings further from the time of the last stimulation sequence are required to rule out acute effects.
Interestingly, 1 h of stimulation had a greater impact on almost all spike and burst measures compared to 3 h stimulation, which often was not significantly different from control. Spikes per burst, burst duration, burst rate (bursts/minute) and total spike rate were all up to 2-fold higher in the 1 h stimulation group. However, intra-burst spike frequency was significantly higher in the 3 h group, although bursts in this group were shorter than in the 1 h group. In other words, bursts in the 3 h stimulation group were sharper, more distinct. This was perhaps the most interesting finding in the paper as far as I am concerned. In all groups, 90% of spikes occurred in bursts, with no difference between conditions. Other reserachers have reported similarly high levels of bursting in neuron cultures. "can we reasonably conclude that information coding occurs in bursts and less so in individual or smaller groups of action potentials?" ask the authors. If so, chronic electrical stimulation during development certainly has considerable impact on information coding.
The authors also make some interesting observations on the impact of distance-from-stimulating-electrode on spike rate. Weirdly, they find that spike-rate decreases with increasing proximity to stimulating electrodes in the 1 h stimulation group, but increases (albeit modestly) with increasing proximity to stimulating electrodes in the 3 h stimulation group. Another piece of the puzzle.
In conclusion, the presence of incoming stimulation/information clearly changes the behaviour of developing neural networks. Izhikevich and Edelman similarly found that input was required for activity to emerge and maintain itself in their vast thalamocortical network models.
It's hard to know what to make of these cultured or modeled, essentially random neural networks. They need input and can discriminate complex output in ways your average CPU cannot. They also produce complex, often chaotic output that changes over time with some regularity (Wagenaar's report on superbursts is relevant here). But the question of how to link input to output, output to input, in ways that produce self-organizing, useful, meaningful or even intelligent results has not yet been addressed (as far as I'm aware).