In silico Hierarchical Clustering of Neuronal Populations in the Rat Ventral Tegmental Area Based on Extracellular Electrophysiological Properties

Abstract

The ventral tegmental area (VTA) is a heterogeneous brain region, containing different neuronal populations. During in vivo recordings, electrophysiological characteristics are classically used to distinguish the different populations. However, the VTA is also considered as a region harboring neurons with heterogeneous properties. In the present study, we aimed to classify VTA neurons using in silico approaches, in an attempt to determine if homogeneous populations could be extracted. Thus, we recorded 291 VTA neurons during in vivo extracellular recordings in anesthetized rats. Initially, 22 neurons with high firing rates (>10 Hz) and short-lasting action potentials (AP) were considered as a separate subpopulation, in light of previous studies. To segregate the remaining 269 neurons, presumably dopaminergic (DA), we performed in silico analyses, using a combination of different electrophysiological parameters. These parameters included: (1) firing rate; (2) firing rate coefficient of variation (CV); (3) percentage of spikes in a burst; (4) AP duration; (5) Δt₁ duration (i.e., time from initiation of depolarization until end of repolarization); and (6) presence of a notched AP waveform. Unsupervised hierarchical clustering revealed two neuronal populations that differed in their bursting activities. The largest population presented low bursting activities (<17.5% of total spikes in burst), while the remaining neurons presented higher bursting activities (>17.5%). Within non-high-firing neurons, a large heterogeneity was noted concerning AP characteristics. In conclusion, this analysis based on conventional electrophysiological criteria clustered two subpopulations of putative DA VTA neurons that are distinguishable by their firing patterns (firing rates and bursting activities) but not their AP properties.

Keywords: VTA; dopaminergic neurons; electrophysiology; hierarchical clustering; neurophysiology.

Figure 1

Dataset acquired using in vivo extracellular recordings of ventral tegmental area (VTA) neurons in anesthetized rats. (A) VTA location on a rat coronal brain section, adapted from Paxinos and Watson (2007). Insets: examples of coordinate locations of some neurons, reconstructed from stereotaxic measurements. All units are in mm. (B) Number of spontaneously-active neurons encountered during electrode descents. (C) Representative example of burst activities during in vivo extracellular electrophysiology. Red action potential (AP) waveforms are included within burst events while black AP waveforms are outside of burst events. Firing-rate distributions using linear (D1) or logarithmic (D2) scales. Firing-rate logarithmic values (E). Red curves represent fitted Gaussian curves. Note that firing rates are skewed (asymmetric Gaussian, in D1,D2) but also lognormally-distributed (symmetric Gaussian, in E). Across the 291 neurons, AP (F) and Δt₁ (G) durations were normally distributed. The direct linear relationship between AP and Δt₁ durations (H). Representative inter-spike interval distributions and AP waveforms of regularly (I) and irregularly (J) firing neurons. The horizontal bars represent 1 ms. Note the differences in coefficient of variation (CV; firing rate coefficient of variation). (K) Typical recording example of VTA dopaminergic (DA) neurons, displaying a stable firing rate over time. The arrow indicates the location of the AP waveform represented on the right.

Figure 2

Unsupervised clustering of the 269 non-high-firing neurons. (A) Initial segregation of neurons with firing rates above 10 Hz. (B) Principal component analysis (PCA) using all six electrophysiological variables revealed that the first two principal components accounted for 62% of the total variance within the dataset. Arrows indicate vectors of the six variables. For clarity, these arrows and their respective variables are drawn again at the top left. (C) Silhouette analysis determined that the optimal number of clusters within our dataset is 2. (D) Unsupervised agglomerative hierarchical clustering with two clusters, using Euclidean distances and Ward’s method for optimal branching. (E) Methodology flow chart used in the present study.

Figure 3

Segregation of VTA neuronal populations according to their electrophysiological properties. (A) Firing rates are significantly higher in cluster 1 than clusters 2 and 3. Neurons from cluster 2 displayed significantly lower firing rates than neurons from cluster 3. (B) Neurons from cluster 3 displayed significantly greater bursting activities than neurons from either cluster 2 or 3. (C) Firing rates and bursting activities in neurons from each cluster. High-firing neurons belonging to cluster 1 presented very minimal bursting activities. Generally, neurons belonging to cluster 3 (n = 64) presented burst activities above 17.5% of all spikes in bursts, while neurons belonging to cluster 2 (n = 205) displayed burst activities below 17.5%. Here, these three clusters are segregated. (D) Total action potential (AP) durations are significantly lower in the high-firing neurons (cluster 1) than in neurons from cluster 2 or 3. (E) Similarly, Δt₁ durations from cluster 1 were significantly lower than cluster 2 or 3. The dashed line represents the well-established cut-off criterion to distinguish between DA and non-DA neurons at 1.1 ms (Ungless and Grace, 2012). (F) Presence/absence of notched AP waveforms in neurons belonging to clusters 1, 2, or 3. Statistical results were performed using Chi-square tests. (G) Neurons responsive to DA pharmacology (apomorphine 10–70 μg/kg, quinpirole 20–60 μg/kg or methylphenidate 2–4 mg/kg) were found in both cluster 2 and 3. Note that some neurons were not tested. (H) Within all neurons included in the present study (n = 291), great heterogeneity of AP waveforms was observed. One-way Kruskal–Wallis analysis of variance (ANOVA) results are given as following: ***p < 0.001 vs. cluster 1 and ^$$$p < 0.001 vs. cluster 2. (I) Typical recording example of VTA neurons where the dopamine D₂ receptor agonist quinpirole (50 μg/kg, iv) induces autoreceptor-mediated negative feedback, which is reversed by the dopamine D₂ receptor antagonist eticlopride (0.2 mg/kg, iv). Note that quinpirole greatly increased the AP amplitude. The arrow indicates the location of the AP waveform represented on the right. All of the histograms represented here are box-plots (box and whiskers). The “+” sign represents the mean. The horizontal bar represents the median value. The interquartile range (Q1–Q3) is displayed with a vertical rectangle (“box”), while the thin lines (“whiskers”) represent the intervals between the lowest/highest values and the Q1/Q3 quartiles, respectively. Outlying values are represented individually with open circles but were not excluded from all analyses. Asterisks (***) are used to indicate significantly different values vs. cluster 1, while dollars ($$$) are used to indicate significantly different values vs. cluster 2. ns: non-significant.