Neural Coding of Perceived Odor Intensity

A, Latency of the first spike estimated using distributions of interspike intervals (Shusterman et al., 2011) for responses identified as sharp pooled across all odors and concentrations versus the latency of the peak PSTH for the same response. B, Difference in absolute PSTH latency between sharp responses to high and 3× lower concentrations versus the relative latency estimated using cross-correlation (see Materials and Methods).

MTC responses change with repeated sampling. A, Sniff-warped raster and PSTH plots of sharp excitatory (I), excitatory (II), and inhibitory (III) responses of single MTCs during the first, fourth, and seventh sniff cycles (shown as color shades). A schematic of the sniff waveform is shown above the plots. Gray shading and vertical dashed lines delineate inhalation period. Gray trace, Activity of the mitral/tufted cell during unodorized sniffs. B, Scatter plot comparing amplitude and latency of excitatory, sharp, and inhibitory responses on the seventh sniff following odor onset. Boxplots show marginal response distributions, as in Figure 1C . Color conventions as in Figure 1. C, Colored lines are normalized distributions of changes in latency, amplitude, and firing rate of sharp, excitatory, and inhibitory responses with adaptation (difference between first and seventh sniffs). Black solid and dashed lines are the same distributions for early and late responses. Notations are same as in Figure 1D .

Correlated changes in response features for concentration and adaptation. From left to right, plots show changes in the latency, amplitude, and mean firing rate. Points are individual response sets. Response types are indicated by color as in Figure 1. Box plots show distributions of response changes across cells for concentration and adaptation. Conventions are as in Figure 1. Reported r values are Spearman correlation coefficients computed independently for the three response types. Black arrows mark positions of the three example cells in Figures 1 and 2.

Spike count unlikely to code odor intensity. A, Average number of spikes observed on a single sniff for each unit as a function of odor concentration. B, Average number of spikes per sniff per cell observed on each sniff for the three tested concentrations and baseline (gray, baseline; light blue, 0.1; dark blue, 0.3; red, 1.0). Error bars indicate the SD across trials.

Principal component analysis of the population vector changes with concentration and adaptation. A, B, The full temporal population vectors plotted in the space of the first and second (A) and second and third (B) principal components. Large symbols, Average PC projection of all first sniffs (black) and seventh sniffs (gray); small symbols, projection of 10 independent subsets of the full dataset (shaded ovals, SD). Blank is cross symbol; concentration 0.1, 0.3, and 1.0, respectively, are circles, squares, and triangles. Black lines connect first sniffs of different concentrations. Gray lines connect first and seventh sniffs of the same concentration. Numbers denote the presented concentrations. C, D, Same as A and B, but for average firing rate population vector.

Adaptation increases concentration identification error. A, Results of classification analysis for concentration discrimination between four levels (0.0, 0.1, 0.3, 1.0): average probability of classification (empty circles) of temporal patterns of MTCs at the first sniff as a function of concentration mismatch between actual concentration and classified concentration (1 corresponds to correct classification, 3(10) is the classification mismatch for 1(2) step threefold concentration differences) for different numbers of cells (shading from lightest to darkest corresponds to 1, 2, 4, 8, 16, 32, and 67 cells). Solid lines are Gaussian fits of classification probability: , where p1 is a probability of correct classification, and σ is the concentration classification error in log10 units. Inset, Concentration classification error as a function of number of cells included in classification. Vertical dashed line: threefold concentration difference. B, Classification performance for all 67 cells for different sniffs following odor onset (black, sniff 1; gray, sniffs 2-7). Inset: concentration classification error for sniff 1 (black) vs later sniffs (gray). Dashed line: median for sniffs 2+.

Adaptation decreases the encoded odor concentration. Single-trial responses were classified based on their Euclidean distance to the average responses to the three concentrations presented on the first sniff and the average blank response. A, Schematics of the classification process for three concentrations (left, 0.1; middle, 0.3; right, 1.0). Responses on a given sniff and concentration (examples are shown in boxes) are classified against responses on the first sniff. The arrows from sniff 5 (shaded box) illustrate match probabilities between this sniff and responses on the first sniff. B, For each concentration (left to right), grayscale plots show the classifier match probability (see bar on right) for responses on a given sniff (x-axis) with the average concentration responses on the first sniff (y-axis). C, Equivalent concentration for each sniff calculated as the average match probability weighted by concentration (circles), and distributions of classification results: thin line is the 10-90% interval; and thick lines are the 25-75% interval.