Exact distinction of excitatory and inhibitory neurons in neural networks: a study with GFP-GAD67 neurons optically and electrophysiologically recognized on multielectrode arrays

Abstract

Distinguishing excitatory from inhibitory neurons with multielectrode array (MEA) recordings is a serious experimental challenge. The current methods, developed in vitro, mostly rely on spike waveform analysis. These however often display poor resolution and may produce errors caused by the variability of spike amplitudes and neuron shapes. Recent recordings in human brain suggest that the spike waveform features correlate with time-domain statistics such as spiking rate, autocorrelation, and coefficient of variation. However, no precise criteria are available to exactly assign identified units to specific neuronal types, either in vivo or in vitro. To solve this problem, we combined MEA recording with fluorescence imaging of neocortical cultures from mice expressing green fluorescent protein (GFP) in GABAergic cells. In this way, we could sort out “authentic excitatory neurons” (AENs) and “authentic inhibitory neurons” (AINs). We thus characterized 1275 units (from 405 electrodes, n = 10 experiments), based on autocorrelation, burst length, spike number (SN), spiking rate, squared coefficient of variation, and Fano factor (FF) (the ratio between spike-count variance and mean). These metrics differed by about one order of magnitude between AINs and AENs. In particular, the FF turned out to provide a firing code which exactly (no overlap) recognizes excitatory and inhibitory units. The difference in FF between all of the identified AEN and AIN groups was highly significant (p < 10−8, ANOVA post-hoc Tukey test). Our results indicate a statistical metric-based approach to distinguish excitatory from inhibitory neurons independently from the spike width.

Mesoporous silica nanoparticles trigger mitophagy in endothelial cells and perturb neuronal network activity in a size- and time-dependent manner

“…During this process, we applied the following procedures: 1) removal of spike using the Mahalanobis threshold (range 1.8-1.4), with P-value of multivariate analysis of variance (ANOVA) sorting quality statistics ,0.01 among the identified units; and 2) when the previous procedure led to excessive spike invalidation, we manually removed the spikes invading the adjacent-unit ellipsoids (the latter method was very effective in decreasing P-values, with only a limited number of erased spikes). On the basis of an unsupervised learning approach consisting of data-reducing principal-component analysis based on the Fano factor (FF) as a feature and followed by k-means clustering, 25,26 we statistically identified two neuronal clusters: inhibitory (i, black) and excitatory (e, red) neurons. Cluster processing was enriched by means of an outlier-removal procedure that discarded the units whose Mahalanobis distances from the centroid of the cluster were greater than the fixed threshold at 1.4.…”

“…For each parameter, data were computed in half-hour time segments and identified as time points in the experiment's plots corresponding to the control and treatment time course. 22,24,25,28 analysis of MsNP protein coronas by sDs-Page MSNPs (0.25 mg/mL) of each size were incubated under constant movement with 1 mL of completed cell-culture medium with 5% or 10% of serum at 37°C for 24 hours. After incubation, samples were centrifuged at 1,000× g for 20 minutes at room temperature to remove unbound proteins from NP surfaces.…”

“…Such in vitro activity can be recorded by MEA technology in a noninvasive way, with good temporal and spatial resolution, as described previously. 22,24,25 Briefly, raw analogue signals were sampled at 40 kHz using MEA-1060-INV preamplifiers (bandwidth 0.1-8 kHz; MultiChannel Systems) and an MEA Workstation (bandwidth 0.1-8 kHz; Plexon, Dallas, TX, USA). Offline Sorter software (Plexon) was used to clean data appropriately from artifacts and assign spikes into multiunit (1-4) electrodes by using principal-component analysis-based software.…”

The different effects of high-frequency stimulation of the nucleus accumbens shell and core on food consumption are possibly associated with different neural responses in the lateral hypothalamic area

“…5). The FF value was used to distinguish excitatory and inhibitory neurons in neural networks in previous papers (Becchetti et al, 2012;Lintas, 2014). Inhibitory neurons typically had a higher FF value and a thin waveform; thus, we could regard BFR neurons as inhibitory neurons in the neural networks.…”

“…Digitized neural activity was isolated into single units based on the difference in the waveform size and shape (Li et al, 2013). The Fano factor value (FF, variance of 200-ms-unit spike-counts in a 4-s window divided by their mean) of each neuron was computed to distinguish between excitatory and inhibitory neurons in neural networks in the LHA (Teich, 1989;Becchetti et al, 2012). The reliability of the spike cluster separation was quantitatively determined based on the refractory period in the autocorrelograms (Gage et al, 2010); for each neuron, stimulation artifact was removed using Hu's method (Hu et al, 2011).…”