Astrocytes induce desynchronization and reduce predictability in neuron-astrocyte networks cultured on microelectrode arrays

Abstract

In this article, we aim to study how astrocytes control electrophysiological activity during neuronal network formation. We used a combination of spike/burst analysis, classification of spike waveforms based on various signal properties and tools from information theory to demonstrate how astrocytes modulate the electrical activity of neurons using microelectrode array (MEA) signals. We cultured rat primary cortical neurons and astrocytes on 60-electrode MEAs with different neuron/astrocyte ratios ranging from 'pure' neuronal cultures to co-cultures containing 50% neurons and 50% astrocytes. Our results show that astrocytes desynchronize the network and reduce predictability in the signals and affect the repolarization phase of the action potentials. Our work highlights that it is crucial to go beyond standard MEA analysis to assess how astrocytes control neuronal cultures and investigate any dysfunction that could potentially result in neuronal hyperactivity.

Keywords: astrocytes; ionic buffering; microelectrode arrays; neuronal networks; signal analysis.

Figure 1.

Waveform selection and features. (a) Waveform features used for the waveform analysis (modified from [41]) and (b) the decision tree of the algorithm used for the waveform detection. (c) Examples of the five possible wave shapes: RS, FS, TS, CS and PS.

Figure 2.

Immunostaining of neurons and astrocytes on coverslips. (a) In the first row MAP2 is represented in cyan for the NS, 80/20 and 50/50 cultures. In the second row, GFAP is shown in red and in the third row the merge of MAP2, GFAP and DAPI images of the same ROI for each culture is presented. The scale bar is 200 µm. (b) The relative cell counts of all ROIs and all coverslips are presented. The bars represent the percentage of neurons or astrocytes over the total number of DAPI-labelled nuclei. One-way ANOVA test results for the comparisons between neuron and astrocyte amounts in the culture are shown in black; the differences between the different cultures are shown in cyan for the neurons and red for the astrocytes. ^∗$<$0.05; ^∗∗$<$0.01; ^∗∗∗∗$<$0.0001. (c) Representative raster plots of the measurements from the same MEAs at 14, 19 and 28 DIV, respectively. The figure represents the spiking activity of a snippet of 20 channels (rows) in the 100 s time window; the spike times are represented by black dots.

Figure 3.

Activity development analysis. (a) SR (Hz), (b) BR (Hz), (c) BD (ms), (d) % spikes in bursts, (e) ISIs (ms) and (f) IBIs (ms) for NS (in blue), 80/20 (in magenta) and 50/50 (in yellow) cultures for 14, 19 and 28 DIV. The black bars in the violin plots represent the median. For the 50/50 co-culture in (a) and (b), the median and some of the points are hidden in the logarithmic scale towards the 0 value. One-way ANOVA test: ns, non-significant;.^∗$<$0.05; ^∗∗$<$0.01; ^∗∗∗$<$0.001; ^∗∗∗∗$<$0.0001.

Figure 4.

C-E planes and iAAFT surrogates analysis. (a–c) C-E values of the single channels in the cultures at DIV14, DIV19 and DIV28, respectively. NS cultures are shown in blue, 80/20 in magenta and 50/50 in yellow. The values for 50/50 were all clustered in the lower right corner, at high entropy and low complexity values, for all the DIVs tested. (a1–c1) Empirical CDF of the complexity values of the cultures on each measurement DIV, compared with the empirical CDF of the complexity levels of the relative generated surrogates (subscript s in the labels). The NS_s are shown in light blue, the 80/20_s in pink and the 50/50_s in light yellow. (a2–c2) Empirical CDF of the entropy measures of the cultures on each DIV, compared with the empirical CDF of the entropies of the relative generated iAAFT surrogates. The original complexity and entropy values from the C-E planes, and the entropy and complexity empirical CDFs were compared at each DIV with a two-sample Kolmogorov–Smirnov test.

Figure 5.

Waveform analysis. (a–c) Prevalence of the spike waveforms in the NS, 80/20 and 50/50 cultures (blue, magenta and yellow) at 14, 19 and 28 DIV, respectively. RS, regular spiking; FS, fast spiking; CS, compound spiking; TS, triphasic spiking; and PS, positive spiking. (d,e) Spike amplitude in µV at 14, 19 and 28 DIV for the negative and positive spikes, respectively. (f) Spike duration in ms, counted as the time from the spike peak to the 2nd peak (or to the end of the time window in the absence of the 2nd peak). (g,h) Positive and negative slopes, respectively, to reach the spike peak for the NS, 80/20 and 50/50 cultures at 14, 19 and 28 DIV. The white bars in the violin plots represent the median. One-way ANOVA test: ns, non-significant; ^∗$<$0.05; ^∗∗$<$0.01; ^∗∗∗$<$0.001; ^∗∗∗∗$<$0.0001.

Figure 6.

Random forest prediction. (a–c) Signals with the training set in black, the test set in blue and the random forest prediction in green, for NS, 80/20 and 50/50 cultures, respectively. (d) Pearson’s correlation analysis between the real values and the predicted values for all channels from all MEAs for NS (in blue), 80/20 (in magenta) and 50/50 cultures (in yellow). The black lines represent the medians and the whiskers show the minimum and maximum points of the datasets. One-way ANOVA test: ^∗$<$0.05; .^∗∗∗∗$<$0.0001.