Cortical Localization of the Sensory-Motor Transformation in a Whisker Detection Task in Mice

Examples of sensory, sensory-motor and motor single unit spiking activity. A, A sample sensory unit from S1. Top, Raster plots show spiking activity for all trials within a session, aligned to the stimulus onset (left) and the mouse’s RT (right). The trials in all raster plots are sorted according to the mouse’s RT. Middle, Average spiking rates across all trials. A transient peak immediately poststimulus is observable with stimulus alignment (left) but not with response alignment (right). Bottom, Trials were further grouped into slow, medium and fast RTs. The sensory peak overlaps in all groups when aligned to the stimulus onset (left) but varies when aligned to the RT (right). B, Same structure as panel A but for a sample sensory-motor unit in wMC. Middle, A transient sensory peak is observable with stimulus alignment (left), along with a sustained activity prominent in the response alignment (right). C, Same structure as panel A but for a sample motor unit in ALM. Middle, Response alignment shows prominent ramping activity immediately before the RT. Bottom, Unlike the sensory unit, the stimulus-aligned peak activity varies with RT (left), whereas the response-aligned peak activity overlaps for all RTs (right).

Quantification of target and distractor stimulus encoding. A, A sample S1 unit firing rate averaged across target (blue) and distractor (black) trials. This unit shows a prominent increase in spiking after target stimulus onset. Dashed lines reflect the poststimulus window used for quantification of sensory encoding. B, Illustration of the single trial prestimulus and poststimulus windows. C, Plot of prestimulus and poststimulus spike count distributions from target trials of the example unit shown in A. D, Plotting of the prestimulus and poststimulus cumulative distribution functions to create a ROC curve for the example unit shown in A. The AUC is transformed into a neurometric d-prime value. The large response in A is reflected in the large separation of prestimulus and poststimulus distributions in C and the highly convex ROC curve in D. E, Scatter plot of all single units in this recording session, plotting target stimulus d-prime versus distractor stimulus d-prime values (example unit indicated in red). Note that target d-prime values are more positively skewed than distractor d-prime values.

Sensory encoding and neurometric-psychometric comparisons across cortical regions. A, Distribution of single unit target and distractor d-prime values for all S1 (top, blue, n = 445 units), wMC (middle, green, n = 424 units), and ALM (bottom, yellow, n = 315 units) units. The average behavioral detection performance (behavioral d-prime) during these recording sessions is depicted by the red dashed lines (S1 = 2.5, wMC = 2.9 and ALM = 2.2). Note that S1 and wMC target d-prime values are highly positively skewed along the x-axis (target detection) but ALM units are not. B, Behavioral and neural d-prime measures across regions. Lines connecting columns within each set denote differences of statistical significance. Set 1, psychometric d-prime across all regions. Set 2, neurometric d-prime averaged across all single units within each region. Set 3, neurometric d-prime of summed spiking within each session averaged across all sessions. Set 4, neurometric d-prime of summed spiking of all units within each region. Combining units results in neurometric performance surpassing psychometric performance for S1 and wMC, but not ALM. C, Randomly selected units were added sequentially to determine the resulting d-prime values of pooled neuronal activity. Shown are the distributions from 300 iterations of each region. Increasing the number of combined units increased d-prime values, with the fastest rate of rise in wMC. D, Transformation of data in panel C, depicting the size of the neural pools achieving the corresponding d-prime values. Red arrows overlaying S1 and wMC data indicate the number of units needed to match behavioral performance. Fewer wMC units were required to match behavioral performance compared with S1 and ALM. The traces and shades in panels C, D are the mean ± SD.

Sensory and motor representations on hit trials across cortical regions. A, Schematic showing how the sensory and motor alignments were calculated; 100-ms windows, after stimulus onset (magenta arrow) and preceding RT (black arrow), were referenced as sensory (red) and motor (blue) epochs, respectively. Spike counts in these windows were compared with a prestimulus baseline (black). Right, Sensory-aligned versus motor-aligned values were plotted for each unit. Population measurements of each region included the sensory and motor alignment mean, variance (σ squared), and slope (sensory variance/motor variance). B, Sensory and motor alignment for all of the recorded units of S1 (left, n = 445), wMC (middle, n = 424), and ALM (right, n = 315). In each plot, the x-axis and the y-axis show motor and sensory alignment d-prime values, respectively. The dashed line indicates equal sensory and motor alignment. Note that S1 and wMC populations both show high variance along the unity line, whereas the ALM population shows high variance nearly exclusively along the motor-aligned axis. C, Each circle’s area is proportional to the mean value along the indicated axis. Statistically significant differences are indicated by bars (permutation statistics). Note the increase in both sensory and motor mean values from S1 to wMC and reduction in sensory mean value in ALM. D, Similar to C, with each circle’s area proportional to the variance of d-prime along the indicated axis.

Comparison of spike rates on hit versus miss trials. Colored plots denote hit trials, black plots denote miss trials. A, An example S1 session showing moderately higher hit-related spiking immediately poststimulus and during the response window. B, An example wMC session, showing robust increased and sustained hit-related spiking that emerges immediately poststimulus. C, An example ALM session, showing robust increased hit-related spiking that emerges late poststimulus. D–F, Average spike rates across all sessions for S1, wMC, and ALM recordings, respectively.

Choice probability within each cortical region. A, A schematic that shows the calculation of the choice probability. Choice probability was calculated by 50-ms sliding window, comparing spike counts on hit (red) versus miss (black) trials. B, Choice probability as a function of time for each region, with overlapping hit and miss distributions at 50% (horizontal dashed line). Data are averages of recording sessions (left, S1, n = 21 sessions; middle, wMC, n = 13 sessions; right, ALM, n = 9 sessions). Significant choice probability is indicated by bars above each plot, gray bars indicate significant positive choice probability (>50%) whereas purple bars indicate significant negative choice probability (<50%). Vertical dashed lines indicate latency to significant poststimulus choice probability. Red bars indicate ±1 SD of the sensory response latency for the same recording sessions. Left, S1 shows prestimulus negative choice probability and poststimulus positive choice probability at a latency of 165 ms. Middle, wMC shows poststimulus positive choice probability at a latency of 70 ms. Right, ALM shows poststimulus positive choice probability at 175 ms.