ISBN 3-938345-05-5
6th Int. Meeting on Substrate-Integrated Microelectrodes, 2008 1
Changes in bursting caused by learning
Jan Stegenga
*, Joost le Feber, Enrico Marani, Wim L C Rutten
Biomedical Signal and Systems, Institute for Biomedical Technology, Enschede, the Netherlands * Corresponding author. E-mail address: j.stegenga@utwente.nl
Abstract
The localization of learning in dissociated cultures to the stimulation-evaluation electrode pair was studied. The cultures were trained using the Conditional Repetitive Stimulation (CRS) algorithm, in which repetitive focal stimulation is ended when a preset ratio of desired responses is achieved. We found that CRS can be used to strengthen an initially weak stimulus-response relationship. We used estimations of the instantaneous firing frequencies per electrode during spontaneous network bursts (NBs), called phase profiles, to determine the spatial extent of the changes required to establish a new stimulus-response relationship in the network. We found significant changes in the profiles, both on the stimulated and observed electrode pair, but also at numerous other sites. The results indicate that most of the changes are uncontrolled and that the whole network is involved during learning.
1 Introduction
Multi electrode arrays (MEAs) are an important tool in studying the processing of information in net-works of neurons. Central to this research are the neu-ral code and the mechanisms underlying learning and memory. The neural code allows the network to sense its environment and act on it. Learning represents modifications to the networks’ connectivity as a con-sequence of experience. Various stimulation para-digms aiming to change network connectivity, meas-ured by action potential firings in reaction to test stimuli, have been introduced [1-3]. Most of these were open-loop, in that stimulation did not depend on network activity. These showed that, under certain conditions, changes lasting for ≥30 min could be in-duced. A certain amount of control over the changes that were induced came with the introduction of the conditional repetitive stimulation (CRS) algorithm by Shahaf et al [4]. This was the first successful closed-loop algorithm, as stimulation was stopped when the network response fulfilled a predefined goal. The network could thus be trained to incorporate a new input-output relationship defined between a stimula-tion electrode and an evaluastimula-tion electrode. However, the role of the rest of the network has not been inves-tigated. We investigated the influence of learning on spontaneous bursting, with emphasis on the per elec-trode contributions. To this end, we made profiles of the firing rate during bursts. A profile of summed ac-tivity was called a burst profile (BP), and profiles of single electrode activity were called phase profiles (PPs). The size and shape of the profiles was highly dependent on culture and age, and changed on a time-scale of several hours [5]. The relative stability of the profiles during spontaneous development allowed the
comparison between profiles before and after CRS with normal development.
2 Methods
We used cultures of newborn rat neocortical cells, as described elsewhere [5]. The training procedure was based on that developed by Shahaf et al [4]. Shortly, low frequency stimulation (ISI: 1 to 5 s) was applied to a single electrode as long as a predefined desired response at an evaluation electrode was ob-served in less than 2 out of the 10 last stimuli. A re-sponse was defined as the presence of one or more spikes in a certain time window after stimulation (e.g. 50 to 80 ms) at a non-stimulated electrode. Only elec-trodes with a small initial response ratio (<0.1) within the response window were chosen for evaluation.
Spontaneous NBs before and after CRS experi-ment were detected by analyzing the Array-Wide Spiking Rate (AWSR, the sum of activity over all electrodes). Next, we estimated the instantaneous AWSR during a burst by convolving spike-occurrences with a Gaussian function (standard devia-tion 5-10 ms). We followed the same procedure for each electrode site as well. When bursts were less in-tense, we used time-averages of several bursts by aligning them to peak AWSR, in order to get a more stable estimation of the per site firing frequency.
Similarity between two profiles was calculated by the mean squared value of their difference. We used t-tests with α=0.05 for determining significant changes between sets of profiles (i.e. before-set and after-set). In 4 experiments, the spontaneous activity before a CRS experiment was longer than the CRS experiment. These, and the associated CRS experiments were used to determine any changes that occurred spontaneously (control) and during CRS, respectively.
ISBN 3-938345-05-5
2 6th Int. Meeting on Substrate-Integrated Microelectrodes, 2008
3 Results
3.1 Training
Typical learning curves showed a fast initial de-crease in the number of stimuli required to elicit the desired response with a responsiveness >0.2 (N, see figure 1). Often, a period with higher N was observed, before settling on a low value. In many cases we found a decrease in array wide response to stimuli during the learning experiment.
Fig. 1. Learning curves. Top) Learning curve of one experiment. Bottom) Average learning curve (12 experiments; 9 cultures).
3.2 Profile changes
All BPs changed significantly during CRS, while 2 out of 4 changed significantly also during control. Figure 2 shows how the BPs in one experiment changed shape.
Fig. 2. : Change in burst profiles immediately before and after CRS (top two traces, black equals before, gray after). The lower two traces are immediately before and after an equal period of spontane-ous development. All curves are 45 minute averages. The single asterisk in the upper right corner indicates that changes were sig-nificant only during CRS.
The percentage of PPs that changed significantly increased from 46% during control to 57% during CRS. On average there were 13 active electrodes. Fig-ure 3 shows the changes in PPs in the same experi-ment as in figure 2.
Fig. 3. Phase profiles of the most active part of the MEA shown in the same layout as the electrodes. The shaded plot is at the location of the stimulus electrode (i.e. 74), evaluation was at electrode 73. Double asterisks denote profiles that changed both during CRS and control; single stars only changed significantly during CRS.
3 Discussion
Application of CRS learning resulted in an in-creased number of profile changes. The CRS protocol, although it evaluated only one (effective) connection, thus resulted or required the change of PPs on many electrodes. This should be taken into consideration if one were to train multiple connections using the CRS algorithm.
Acknowledgement
We thank Remy Wiertz for his work on culture preparation and maintenance.
References
1. Jimbo, Y., H.P. Robinson, and A. Kawana, Strengthening of synchronized activity by tetanic stimulation in cortical cultures: application of planar electrode arrays. IEEE Trans Biomed Eng, 1998. 45(11): p. 1297-304. 2. Madhavan, R., Z.C. Chao, and S.M. Potter, Plasticity of
recur-ring spatiotemporal activity patterns in cortical net-works. Phys Biol, 2007. 4(3): p. 181-93.
3. Ruaro, M.E., P. Bonifazi, and V. Torre, Toward the neuro-computer: image processing and pattern recognition with neuronal cultures. IEEE Trans Biomed Eng, 2005. 52(3): p. 371-83.
4. Shahaf, G. and S. Marom, Learning in networks of cortical neurons. J Neurosci, 2001. 21(22): p. 8782-8. 5. Stegenga, J., et al., Analysis of cultured neuronal networks
using intraburst firing characteristics. IEEE Trans Bio-med Eng, 2008. 55(4): p. 1382-90. 0 20 40 60 80 100 Numbe r of stimul i 10 15 20 25 30 35 40 45 50 55 Iterations Average Example 5 0 20 40 60 100 0. 1 /s 100 ms ** ** ** * ** ** ** ** ** * ** ** ** ** ** ** ** ** ** ** ** ** ** **
