Prediction of Choice from Competing Mechanosensory and Choice-Memory Cues during Active Tactile Decision Making

Abstract

Perceptual decision making is an active process where animals move their sense organs to extract task-relevant information. To investigate how the brain translates sensory input into decisions during active sensation, we developed a mouse active touch task where the mechanosensory input can be precisely measured and that challenges animals to use multiple mechanosensory cues. Male mice were trained to localize a pole using a single whisker and to report their decision by selecting one of three choices. Using high-speed imaging and machine vision, we estimated whisker-object mechanical forces at millisecond resolution. Mice solved the task by a sensory-motor strategy where both the strength and direction of whisker bending were informative cues to pole location. We found competing influences of immediate sensory input and choice memory on mouse choice. On correct trials, choice could be predicted from the direction and strength of whisker bending, but not from previous choice. In contrast, on error trials, choice could be predicted from previous choice but not from whisker bending. This study shows that animal choices during active tactile decision making can be predicted from mechanosensory and choice-memory signals, and provides a new task well suited for the future study of the neural basis of active perceptual decisions.SIGNIFICANCE STATEMENT Due to the difficulty of measuring the sensory input to moving sense organs, active perceptual decision making remains poorly understood. The whisker system provides a way forward since it is now possible to measure the mechanical forces due to whisker-object contact during behavior. Here we train mice in a novel behavioral task that challenges them to use rich mechanosensory cues but can be performed using one whisker and enables task-relevant mechanical forces to be precisely estimated. This approach enables rigorous study of how sensory cues translate into action during active, perceptual decision making. Our findings provide new insight into active touch and how sensory/internal signals interact to determine behavioral choices.

Keywords: behavior; computational modeling; decision making; mouse; sensory-motor integration; whisker system.

Figure 1.

The three-choice object localization task. A, Left, Schematic of the experimental preparation, showing the three pole locations (circles) and the two lick ports. Both lick ports and pole location are color coded consistently with B. Whisker movements and whisker-pole interactions were filmed with a high-speed camera (1000 frame/s). Right, Schematic of a correct go trial to illustrate the trial structure (colored bars, defined in Materials and Methods). Whisker angle, whisker curvature, and whisker–pole touches were extracted from the high-speed video. Mouse choice was monitored by measuring the time of first lick. B, Trial-choice outcomes and how they were rewarded/punished. C, Mouse behavior during an example experimental session. Whisker angle (left) and whisker curvature (right) for each whisker-tracked trial (see Material and Methods). In the top panels, trials are sorted according to chronological order during the session. In the bottom panels, trials are sorted first by pole location, and, within each pole location, by mouse choice.

Figure 2.

The three-choice object localization task is whisker-dependent. A, Top, Task performance of mouse 33 during the course of training. The mouse was initially trained with all its whiskers intact. The whiskers were progressively trimmed to one whisker and, finally, as a control, to none. Colored lines indicate the protocol the mouse was trained on each day: lick (cyan), go–no go (red), lick left–lick right (green), lick left–lick right–no lick (gold; protocols are detailed in Materials and Methods). When cyan and red lines overlap, it indicates that the protocol was switched to the go–no go protocol during the same behavioral session. Bottom, Total number of trials performed each day. B, Stable performance for each mouse during AB trials of the full task with a single whisker. Stable sessions were selected as detailed in Materials and Methods. Purple dots show performance in each session, and large black dots and black error bars show mean and SD across selected sessions, respectively. Gray dots and gray error bars show chance performance and 95% confidence interval on chance, respectively. C, Task performance during AB trials in the five sessions before (purple) and two to three sessions after (black) whisker trimming, for each of five whisker-trimming tests. D, Grand mean task performance on sessions before (dark purple) and after (black) whisking trimming. Error bars indicate SD. *p = 0.0013 (t test). Dotted lines, average chance range (see Materials and Methods).

Figure 3.

Whisking kinematics and bending during the task. A, Example trajectories of whisker angle and whisker curvature for posterior (left), middle (middle), and anterior (right) pole locations in two mice (top and bottom). B, Whisking amplitude in 200 whisker-tracked trials (see Materials and Methods) for an example mouse, relative to the onset of pole movement (vertical broken line). C, Mean (thick line) + SD (thin line) whisking amplitude across whisker-tracked trials of each mouse. Whisking amplitude significantly increased after pole onset (200 ms interval before and after pole onset; t test, p = 4 · 10⁻⁴).

Figure 4.

Presence/absence of touch cannot account for mouse choice. A, Probability of correct choice as a function of pole location, for each mouse (gray lines). Red dots indicate that performance of a given mouse was outside 95% confidence interval on chance (5000 shufflings). B, Probability of touch as a function of pole location. Gray lines indicate individual mice. C, Task performance of mouse and PAT classifier along with choice consistency. Red dots indicate that the classifier/mouse performance or choice consistency was significantly higher than chance. Black circles report mouse behavioural data. Orange circles report PAT classifiers outcome. Empty circles and error bars show means and SDs across mice, respectively. *p = 1.8 · 10⁻⁴ (t test).

Figure 5.

Whisker bending magnitude and direction account for mouse choice. A, Probability of each touch type as a function of pole location (mean and SD across mice). B, Performance of touch-based classifiers compared to mice. Red dots indicate that corresponding classifier/mouse performance or choice consistency was significantly greater than chance for the given mouse. Orange circles show mean classifier performance and choice consistency of the PAT classifier. Light blue circles show mean classifier performance and choice consistency of the touch type classifie Error bars indicate SD. C, Mean Δκ₉₅ of each touch type as a function of pole location for all mice. Error bars indicate SEM. D, Performance of touch type- and bending-based classifiers compared to mice. Red dots indicate that corresponding classifier /mouse performance or choice consistency was significantly greater than chance for that mouse. Dark blue circles show mean classifier performance and choice consistency of the Δκ₉₅–touch type classifier. Light blue circles are same data as in B). Error bars indicate SD. E, Single mouse values of touch type only and touch type-Δκ₉₅ choice consistency. *p ≤ 0.05 (t test).

Figure 6.

Previous choice predicts error trials. A, Sequence of 31 consecutive trials performed by an example mouse. Red, blue, and gray rectangles respectively indicate trials in which the pole location was anterior, middle, and posterior (the top row), and in which the mouse made posterior, middle, and anterior choices (bottom row). Triangles indicate error trials, and dark red triangles indicate error trials in which the choices in the previous and current trials were identical (i.e., the mouse perseverated). B, Probability of perseveration for each mouse (gray lines) under different conditions: considering all trials (left), correct trials only (middle), and error trials only (right). The green line indicates the example mouse in A. Black circles indicate the means; black error bars indicate SD across mice. *p ≤ 0.0167 (t test; Bonferroni correction, n = 3). Gray bars indicate the chance interval (10,000 shuffling, 95% confidence interval). C, Performance of classifiers predicting mouse choice on correct trials only (left) and error trials only (right). Blue and yellow circles indicate mean values for the Δκ₉₅–touch type classifier and previous choice classifier respectively. Small dots are single mouse values. Red indicates that the classifier performance value for the mouse was above chance. Error bars indicate SD across mice. *p ≤ 0.05 (t test). D, Probability of perseveration during error trials depending on whether the previous trial was a correct trial or an error trial. *p = 3.4 · 10⁻⁴ (t test). Black error bars indicate chance intervals of each mouse (10,000 shufflings, 95% confidence interval).