Clampfit

The electrophysiological and MEA data were analyzed using Clampfit version 11.1.0.23 (Molecular Devices) and a custom-made OriginLab program (OriginLab 2018b, Northampton, USA). For the patch-clamp and MEA recordings, comparisons between cell lines were performed with the non-parametric Kruskal–Wallis test followed by Dunn’s post-hoc multiple comparisons test (GraphPad Prism Version 5.02, San Diego, CA, USA). The Ca²⁺ peak parameters were analyzed using Clampfit (Molecular Devices). In these analyses, the instantaneous frequency was calculated by converting each interevent interval into a frequency and subsequently averaging these frequencies. In addition, Ca²⁺ traces were analyzed by identifying the signals as normal or abnormal and categorizing the abnormalities into subgroups. Data obtained from WT CMs (04511.WT and 04602.WT) were pooled as their Ca²⁺ peak parameter values were similar. Statistical analyses of Ca²⁺ and qRT-PCR data were conducted using the SPSS software version 24 (SPSS, Chicago, IL, USA). Comparisons within cell lines (before and after drug administration) were performed with non-parametric Wilcoxon and comparisons among cell lines (cell line-to-cell line comparison) with non-parametric Kruskal–Wallis with Bonferroni correction. Statistical significance levels are indicated as “ns” (not significant), * P < 0.05, ** P < 0.01, or *** P < 0.001.

Dose–response relationships
were described with standard Hill’s equation in the form: where [GABA] is the agonist
concentration, and n_h is Hill’s
coefficient. The time course of the macroscopic desensitization was
fitted with a single exponential function (Clampfit, Molecular Devices,
US) as:
The current deactivation
time course after application of saturating GABA pulse was fitted
with a biexponential function:
FR10 and FR500 parameters
were calculated as: where A_max refers to the current amplitude [pA], and A_x refers to the current value [pA] after x ms after its peak.
The deactivation time constant was calculated
as: where
The rise time (RT)
was calculated with the function build in Clampfit
(Molecular Devices, US) as 10–90% of the macroscopic current
onset.

To track limb movements during running, we used DeepLabCut (version 2.1.5.2, Mathis et al., 2018 (link), see Figures 3,4). We manually labeled the positions of the head and the 4 paws from 50 frames of each video. We then used 95% of the labeled frames to train the network using a ResNet-50-based neural network with default parameters for three training iterations. We validated with two shuffles and found that the test error for trot experiments was: 3.33 pixels and the train error: 2.37 pixels (image size: 1344 × 301). Similarly, we trained the network for gallop condition using a ResNet-50-based neural network with default parameters for one training iteration. We validated with two shuffles and found that the test error was: 3.43 pixels and the train error: 2.43 pixels (image size: 1936 × 230). These networks were then used to analyze videos from similar experimental settings. X and Y coordinates from the head and the four limbs were then extracted and interpolated to 10 kHz to match the EMG recordings. The latter were exported from LabScribe to Clampfit (Molecular Devices), and both sets of signals were merged in a single file, before being processed offline in Clampfit.