Unsupervised Phenotyping Reveals Disrupted Neural Firing Characteristics in the Anterior Thalamus and Surrounding Brain Regions Following Third-Trimester Equivalent Alcohol Exposure in Mice

Abstract

Acute binge-like third-trimester-equivalent alcohol exposure (TTAE) causes apoptosic neurodegeneration in brain regions necessary for spatial learning and memory, such as the anterior thalamus (AT), which encodes context-relevant information, including head direction. While we are beginning to understand the behavioral consequences of this exposure, the neural correlates of spatial cognition deficits are underexplored. Thus, we recorded a mixture of neurons from the AT and surrounding brain regions in mice with TTAE while they freely moved within a controlled environment. To model acute binge-like TTAE, C57BL/6J mice received 2 injections of 2.5 g/kg alcohol (or saline) on post-natal day (PND) 7. Subjects were then left undisturbed until the day of surgery as adults (>PND 60), when they were implanted with silicon or multi-wire arrays. Mice were placed in a circular 40 cm diameter arena under dim red light with 2 LED cues on the walls that rotated on a pseudo-random basis while electrophysiological data was recorded. Following data preprocessing, spike and waveform features were extracted from each putative single spiking unit. These features were reduced utilizing uniform manifold approximation and projection (UMAP). Following dimensionality reduction, we used agglomerative hierarchical clustering to find populations of neurons with similar features. Following this, we compared each feature based on treatment and found the features most important to disambiguate TTAE's impact on neural activity. TTAE was associated with decreases in mean firing rate, peak firing rate, rebound index, and tail decay constant, increases in alpha, peak to trough time, and repolarization time, and bidirectional differences as a function of neuronal subtype in burst index and rebound index. This suggests that TTAE produces long-lasting and fundamental differences in spiking features that can be observed in vivo and are amenable to intervention. Together, this dataset provides further clarifying criteria that can be utilized to diagnose and treat FASD.

Figure 1:

Schematic of methodology. A. Mice at post-natal day 7 were given either ethanol at 2.5 g/kg s.c. or saline; two hours were allowed to pass and a second dose of the same treatment was given. Mice were aged until at least post-natal day 60 prior to any experimentation. B. The anterior thalamus was targeted with recording electrodes. C. The mice were placed in a 40cm environment with rotating LED cues. D. Neurons were spikesorted and the resulting spike-and-wave features were extracted from each neuron within the dataset. Features extracted include the peak to through duration, refractory period, mean firing rate, mean ISI, C_v, burstiness, alpha, beta, maximum firing rate, and variance of the firing rate. E. All ten features were put into a single Nx10 matrix where N is the number of neurons. UMAP was performed to reduce dimensionality from 10D to 2D. Hierarchical agglomerative clustering was then performed to cluster the UMAP reduction.

Figure 2:

A. UMAP reduction with corresponding cluster identities is shown. B. A random forest classifier was trained to determine which features are most predictive of cluster identity; burstiness and peak to through time dominated over other features. C. A UMAP reduction overlaid with the treatment identity of each neuron is shown. D. A random forest classifier was trained to determine which features are most predictive of treatment. E. Neuron counts per cluster and treatment were tabulated.

Figure 3.

Features values were superimposed over the UMAP reduction for the (A) mean ISI, (B). CV, and burstiness (C). D. Treatment effects were assessed for the mean ISI, E. CV, and F. burstiness. A treatment effect was observed in burstiness for cluster 7.

Figure 4.

Features values were superimposed over the UMAP reduction for the (A) mean firing rate, (B). maximum firing rate, and variance of the firing rate (C). D. Treatment effects were assessed for the mean firing rate, E. maximum firing rate, and F. variance of the firing rate. A treatment effect was observed in cluster 2 for the mean firing rate. A treatment effect was observed in cluster 4 and cluster 7 for the maximum firing rate.

Figure 5.

Features values were superimposed over the UMAP reduction for the (A) alpha and B. beta values. C. Treatment effects were assessed for alpha and D. beta values. A significant effect was found in the alpha value for cluster 7.

Figure 6.

Features values were superimposed over the UMAP reduction for the (A) repolarization time (B) and peak to through time. C. The normalized waveforms for voltage were plotted for each cluster and treatment. D. Treatment effects were assessed for the repolarization time E. and peak to through time. Significant treatment effects were found in cluster 9, 10, and 11 for the repolarization time. Significant treatment effects were found in cluster 9 and 11 for the peak to through time.

Figure 7.

A. Representative 3D autocorrelograms for three clusters are shown. Features from the 3D autocorrelograms were extracted and treatment effects were assessed. B. The burst index was assessed; a significant treatment effect was observed in cluster 7 and cluster 9. C. The rebound index was assessed; a significant treatment effect was observed in cluster 7 and cluster 9. D. The tail decay constant was assessed; a significant treatment effect was observed in cluster 1, 5, 6, 8, 9, and 11.

Figure 8.

A. The FDR corrected p-values for all pairwise t-test comparisons were summarized. B. Cohen’s D was calculated. C. Cohen’s D was summed across all clusters to generate values for each feature that summarize which features discriminate treatments.