Software analysis for multi-site recordings

The dynamics of sensory-evoked and spontaneous up-states have been recently compared, in multi-site recordings in vivo and in vitro, by using various types of statistical analysis such as the spike rate (SR) or the Fano factor (FF). We here describe and validate a novel computational method to classify into statistically different states the spontaneous reverberating activity recorded from long-term (12-18 days-in-vitro) cultured cortical neurons (from 60-site multi-electrode arrays, MEA). State classification was performed by spike number time histograms (SNTH, or other burst features) of excitatory and inhibitory neuron clusters and revealed that also the number of engaged neurons and the FF histograms are linked to the identified firing modes. To improve the characterization of each state we also computed the firing spike histograms (FSH) which revealed a new facet of the firing probability of clusters. Surprisingly, we found that the FF is an enhanced feature as compared to the autocorrelation function (ACF) to clusterize units. In exemplary functional experiments we show that: i) up to 4 different states can be identified in 30’ timesegments, ii) in adjacent time-segments novel states can be recognized, iii) up to 6-7 states can be safely categorized during several hours of recordings, iv) they disappear after a short pharmacological stimulation being replaced with novel states active and living up to 6-8 h. In conclusion, we believe that this novel procedure better characterizes the number of functional states of a network and opens up the possibility of predicting the elementary “vocabulary” used by small networks of neurons.