Behavioral tasks

At the start of each daily behavior session, mice were transiently anesthetized with isoflurane and head-fixed under the 2-photon microscope. 9 whiskers (rows C-E, arcs 1-3) were inserted into a 3 ×3 array of calibrated piezoelectric actuators, centered on the D2 whisker. Whiskers were not trimmed, and were threaded into tubes on the piezos, held by soft glue. Deflections were applied 5 mm from the face. A drink port with capacitive lick sensor recorded licks. Paw guards prevented paw contact with whiskers, piezos, or drink port. After whisker insertion, mice recovered from anesthesia and began the behavioral task. Mice had to suppress licking to initiate a behavioral trial. Training was performed in total visual darkness (using 850 nm IR illumination for behavioral monitoring). Uniform white noise (77.4 ± 0.5 dB) was continuously applied to mask sounds from piezo actuators and drink port opening (drink port: 58.9 ± 0.7 dB; all-whisker deflection: 69.2 ± 0.7 dB; single whisker deflection: 59.7 ± 0.5 dB; background noise floor without white noise: 58.7 ± 0.5 dB; all sound levels were measured at the location of animal’s ears). Tasks were controlled by an Arduino Mega 2560 and custom routines in Igor Pro (WaveMetrics).

Each trial contained a 0.5 s baseline period, 0.5 s stimulus period, and 1.5 s response window. One randomly chosen stimulus was applied per trial, either 1 of 9 single whisker deflections, all-whisker deflection, one of 2 tones, or a blank (no stimulus). Whisker stimuli were ramp-return rostrocaudal deflections (300 µm, 5 ms rise/fall time, 10 ms duration). A train of 5 deflections (100 ms interval) was used to reliably evoke GCaMP signals. The all-whisker stimulus was simultaneous deflection across all 9 whiskers. Tone stimuli were a single tone pip (2 or 8 kHz, 200 ms duration, 82.4 ± 0.2 dB) delivered from a nearby speaker. Each trial was followed by a 3 ± 1 s interval before the mouse could initiate the next trial. Thus, consecutive stimuli were separated by > 5.5 ± 1 s.

On S+ trials (all-whisker deflection), water reward (2–4 µl) was automatically dispensed 300 ms into the response window. Licking was not required to dispense reward. Water was not dispensed on S− trials. Licking above a threshold rate during the response window was defined as a lick response, and scored as a hit on an S+ trial and a false alarm (FA) on an S− trial. FAs and misses were not punished. Mice learned that S+ stimuli predicted reward, evidenced by licking in S+ trials prior to reward delivery (Supplementary Fig. ^1c).

The structure of each trial was the same as one-versus-all whisker discrimination task, except the S+ stimulus was one of the two tones.

Each trial contained a 0.5 s baseline period, 0.5 s stimulus period, 1 s delay period, and 1.5 s response window. One randomly chosen single whisker stimulus (1 of the 9 whiskers) or a blank was presented per trial. The mouse was rewarded if the D1, D2, or D3 whisker was presented, and not rewarded for any other whisker or blank. Mice had to withhold licking during the post-stimulus delay period, and then lick during the response window. Water reward was only dispensed in response to licks during the response window. In order to train the response window, the reward volume was linearly increased as a function of first lick time in the period 0-500 ms into the response window. For licks after this time, a constant maximal reward was delivered (Fig. 8a). No additional cues were given to indicate the delay period. This delay-dependent reward increment encouraged animals to withhold their licking following stimulus delivery, and to receive a larger water reward if they licked later. No punishment was given if animals licked during the delay period, but that trial was aborted (i.e., no reward was delivered) and excluded from analysis.