Active Licking Shapes Cortical Taste Coding

Abstract

Neurons in the gustatory cortex (GC) represent taste through time-varying changes in their spiking activity. The predominant view is that the neural firing rate represents the sole unit of taste information. It is currently not known whether the phase of spikes relative to lick timing is used by GC neurons for taste encoding. To address this question, we recorded spiking activity from >500 single GC neurons in male and female mice permitted to freely lick to receive four liquid gustatory stimuli and water. We developed a set of data analysis tools to determine the ability of GC neurons to discriminate gustatory information and then to quantify the degree to which this information exists in the spike rate versus the spike timing or phase relative to licks. These tools include machine learning algorithms for classification of spike trains and methods from geometric shape and functional data analysis. Our results show that while GC neurons primarily encode taste information using a rate code, the timing of spikes is also an important factor in taste discrimination. A further finding is that taste discrimination using spike timing is improved when the timing of licks is considered in the analysis. That is, the interlick phase of spiking provides more information than the absolute spike timing itself. Overall, our analysis demonstrates that the ability of GC neurons to distinguish among tastes is best when spike rate and timing is interpreted relative to the timing of licks.SIGNIFICANCE STATEMENT Neurons represent information from the outside world via changes in their number of action potentials (spikes) over time. This study examines how neurons in the mouse gustatory cortex (GC) encode taste information when gustatory stimuli are experienced through the active process of licking. We use electrophysiological recordings and data analysis tools to evaluate the ability of GC neurons to distinguish tastants and then to quantify the degree to which this information exists in the spike rate versus the spike timing relative to licks. We show that the neuron's ability to distinguish between tastes is higher when spike rate and timing are interpreted relative to the timing of licks, indicating that the lick cycle is a key factor for taste processing.

Keywords: coding; elastic shape analysis; gustatory cortex; licking; support vector machine; taste.

Figure 1.

A, Sketch showing a head-restrained mouse licking a spout to obtain gustatory stimuli or water. B, Left, Example of histologic section showing the tetrode tracks (magenta) in the GC. Right, Schematic reconstruction of the tetrode tracks of all the mice used in this study (red). In magenta the reconstruction of the track corresponding to the histologic section shown on the left. C, Top panel, Diagram of the taste delivery paradigm. Gustatory stimuli and water (T) are delivered after six consecutive dry licks (D) to the spout using a fixed ratio (FR) schedule. Each lick is denoted by a vertical line. Bottom panel, Raster plot of licking activity in a 2-s time interval centered at the taste delivery (time 0) from one experimental session. D, Top panel, Raster plot of spiking activity from a GC neuron centered at the taste delivery, where each tick mark represents a spike. Bottom panel, Corresponding raster plot of licking activity. Spike and lick tick marks are grouped together and color-coded with sucrose (S) in pink, NaCl (N) in green, citric acid (C) in orange, quinine (Q) in purple, and water (W) in brown.

Figure 2.

Binning and smoothing of neuronal spike trains. Top row, Raw spike train consisting of a {0, 1}-valued time series over 4000 ms with one spike count (S.C.) for each millisecond. Taste delivery occurs at time 0. Middle row, Binned versions of the same raw spike train with various bin widths. Responses are a spike count for the bin. Bottom row, Smoothed versions of the same raw spike train with various smoothing window sizes. Responses are in arbitrary units (A.U.).

Figure 3.

Illustration of the behavior of the Fisher–Rao metric. Each row shows a different example of aligning a pair of synthetic signals (i.e., solving the optimization problem described in the text), with responses measured in arbitrary units (A.U.). Panel A shows a pair of signals f (red, solid) and g (blue, dashed). Considering these signals as smoothed spike trains, it is clear that the spike trains have the same firing rate, but that they differ in phase. This is made precise using the ESA framework. Solving the ESA optimization problem produces an optimal alignment function γ, as shown in panel B. The time warped signal g ∘ γ is shown in panel C (blue, dashed), together with the signal f (red, solid), the traces are the same, indicating that f and g have the same rate code. The fact that the optimal alignment function γ is far from the identity tells us that the signals f and g differ significantly in phase. Panel D shows another pair of functions f (red, solid) and g (blue, dashed), and the optimal alignment function γ is shown in panel E. Observe that the graphs of f and g both enclose the same area; however, f has a single maximum, whereas g has two local maxima. Following alignment, there are still two peaks in g ∘ γ (panel F), versus one in f, so the difference in rate is preserved.

Figure 4.

SVM classification scores of all five tastants and full-length spike trains treated with varying smoothing and binning parameters. At each parameter level, the mean classification rate among the best-performing 1%, 5%, and 10% of neurons is plotted.

Figure 5.

Top row, Histogram of the SVM classification scores for all five tastes on the pre-taste and post-taste data (A) and a closer look at the post-taste scores (B). High post-taste scores are colored in blue, and the corresponding neurons are referred to as “coding neurons.” Bottom row, Quantile-quantile plots for the SVM scores using pre-taste data (C) and post-taste data (D). For reference, the red lines indicate the quantiles of a normal distribution.

Figure 6.

A–D, Support vector machine classification rates on selected taste pairs using post-taste test set spike trains. The distributions are heavily skewed toward good-performing coding neurons, colored in blue. E, A cumulative distribution histogram of the fraction of neurons that are labeled as coding neurons for one or more taste pairs.

Figure 7.

Rate-Phase code for the two tastants Water and NaCl. The orange line shows the best separation of the data based on mean rate only, the green line shows the best separation based on mean phase only, and the blue line is the best separating line based on both mean rate (R) and mean phase (P).

Figure 8.

Separation scores for selected taste pairs applied to the best 50 neurons for each stimulus pair (A, citric acid vs. sucrose. B, sucrose vs. salt. C, salt vs. water. D, water vs. quinine).

Figure 9.

SVM classification on four selected stimuli pairs (A, citric acid vs. sucrose. B, sucrose vs. salt. C, salt vs. water. D, water vs. quinine). In each case, the 50 best neurons based on SVM classification of original spike trains are used and the classification score is in blue. The classification scores using the aligned spike trains are in orange, and those using the alignment functions are in green. The ESA and SVM classification are performed on spike trains after administration of the stimuli and without consideration of lick timing.

Figure 10.

SVM classification on four selected stimuli pairs in which alignment is performed over interlick intervals (A, citric acid vs. sucrose. B, sucrose vs. salt. C, salt vs. water. D, water vs. quinine). In each case, the 50 best neurons based on SVM classification of original spike trains are used and the classification score is in blue. The classification scores using the interlick aligned spike trains are in orange, and those using the interlick alignment functions are in green.

Figure 11.

SVM classification scores for all five tastants. The classification uses either original spike trains (blue), aligned spike trains (orange), or alignment functions (green). A, Alignment is performed over the entire 2-s post-taste time interval. B, Alignment is performed on five individual post-taste interlick intervals.

Figure 12.

A, Box plot depicting the proportion of the “rate” (aligned; orange) and “temporal” (alignment functions, green) scores to the original SVM score, from two different types of ESA experiments (phase and randomized). Left boxes (phase), ESA is performed on spike trains in five interlick intervals. Right boxes (randomized), ESA is performed on spike trains in five random intervals. Table 1 below provides the information of each statistical comparison. B, Example raster plot of spiking during the first two interlick intervals in response to two tastants. At the bottom, the spike timings over all trials are represented as smoothed green or blue bumps. The overall SVM classification score for this neuron was 90%, and when classification was performed using aligned spike trains in interlick intervals, the score was 96.3%, and when using the alignment functions in the classification, the score was 89.3%.