Learning Enhances Sensory Processing in Mouse V1 before Improving Behavior

Abstract

A fundamental property of visual cortex is to enhance the representation of those stimuli that are relevant for behavior, but it remains poorly understood how such enhanced representations arise during learning. Using classical conditioning in adult mice of either sex, we show that orientation discrimination is learned in a sequence of distinct behavioral stages, in which animals first rely on stimulus appearance before exploiting its orientation to guide behavior. After confirming that orientation discrimination under classical conditioning requires primary visual cortex (V1), we measured, during learning, response properties of V1 neurons. Learning improved neural discriminability, sharpened orientation tuning, and led to higher contrast sensitivity. Remarkably, these learning-related improvements in the V1 representation were fully expressed before successful orientation discrimination was evident in the animals' behavior. We propose that V1 plays a key role early in discrimination learning to enhance behaviorally relevant sensory information.SIGNIFICANCE STATEMENT Decades of research have documented that responses of neurons in visual cortex can reflect the behavioral relevance of visual information. The behavioral relevance of any stimulus needs to be learned, though, and little is known how visual sensory processing changes, as the significance of a stimulus becomes clear. Here, we trained mice to discriminate two visual stimuli, precisely quantified when learning happened, and measured, during learning, the neural representation of these stimuli in V1. We observed learning-related improvements in V1 processing, which were fully expressed before discrimination was evident in the animals' behavior. These findings indicate that sensory and behavioral improvements can follow different time courses and point toward a key role of V1 at early stages in discrimination learning.

Keywords: behavior; discrimination learning; mouse vision; primary visual cortex.

Figure 1.

Classical conditioning of orientation discrimination. A, Experimental setup. Head-fixed mouse on a spherical treadmill in front of a monitor showing oriented gratings. Reward was delivered through a drinking spout equipped with a lick sensor. B, Discrimination learning paradigm. At irregular intervals, one of two oriented gratings was presented at a randomly selected contrast level. White arrow indicates the drift direction of the grating. The presentation of one grating (315 degrees) was immediately followed by a fluid reward; the other grating (45 degrees) had no consequences. C–E, Licking behavior of one animal in three example sessions at different learning stages (Mouse M22). Top, Licks to the rewarded stimulus. Bottom, Licks to unrewarded stimulus. Vertical bars represent stimulus onset and offset. E, Horizontal bars represent the time windows used to quantify the strength of anticipatory licking.

Figure 2.

Orientation discrimination learning is characterized by a sequence of distinct stages. A–C, Cumulative LIs across trials to the rewarded (blue) and the unrewarded (red) stimulus for 3 example mice (M22, M28, M42). Circles indicate significant change points. C, The second to last slope change for the unrewarded stimulus is concealed by the large number of trials but becomes evident in the difference (H). D, Distribution of the number of change points for the rewarded stimulus across mice (n = 15). E, Ideal-observer analysis decoding stimulus presence from lick rates in naive versus later (intermediate and trained) stages. F–H, Same as A–C, difference in cumulative LIs for rewarded and unrewarded orientations. I, Same as D, for the difference in cumulative LIs across mice. n/a, Number of animals that failed to learn. J, Ideal-observer analysis decoding stimulus orientation from LIs in early (naive and intermediate) versus trained stages. Data points indicate individual mice. Error bars indicate 95% CIs.

Figure 3.

Orientation discrimination performance at different levels of stimulus contrast. A, B, Cumulative LIs in response to the rewarded stimulus at different contrast levels for 2 example mice (M110, M42). Black dots indicate significant change points. C, Area under the ROC curve as a function of stimulus contrast. At each contrast level, stimulus presence is decoded from lick rates in the intermediate and trained stage. Circles represent individual animals. Black line indicates the mean across animals. D, Same analysis as in C, but showing contrast levels of 1% versus 2% only. Error bars indicate 95% CIs. E, Rank of the first change point in the cumulative LIs as a function of stimulus contrast. Black line indicates the mean rank across animals. F, G, Same as A, B, but for difference in cumulative LIs to rewarded and unrewarded stimuli. H, Same as C, decoding stimulus orientation from LIs in the trained stage. I, Same as D, for data in H. J, Same as E, for change points extracted from the difference in LIs. N = 7 mice.

Figure 4.

Transient optogenetic suppression of V1 neurons impairs behavioral performance. A, Left, LFP (black traces) superimposed on the CSD profile. Right, CSD traces. Bold line indicates the base of layer 4. B, V1 activity was suppressed by unilateral photo-simulation of PV⁺ inhibitory interneurons expressing ChR2. C–E, Raster plots of three example neurons from different V1 layers (units M120-33-8, -28, -47). Cyan represents V1 suppression by PV⁺ photo-activation. Black represents control condition. Stimulus conditions are separated by gray horizontal lines. Symbols and vertical lines on the left indicate stimulus orientation (blue drop represents rewarded; red cross represents unrewarded) and contrast (black represents 40%; gray represents 6%). F, Cumulative LIs for one mouse (M117). Left, Behavioral performance for rewarded (dark blue) versus unrewarded (red) orientation during V1 suppression (dashed lines) and in control condition (solid lines). Right, Rewarded orientation only, at 40% (black) and 6% (gray) contrast. G, Same as F, for Mouse M199.

Figure 5.

V1 neurons show improved discriminability and sharper orientation tuning already in the intermediate learning stage. A, B, Spike rasters and density functions in response to the rewarded (blue) and unrewarded (red) stimulus for two example neurons (units M22-14-16, M81-3-31). Insets, Preferred (black), rewarded (blue), and unrewarded (red) orientations. C, Neural discriminability (d′) during task performance in the naive (gray), intermediate (orange), and trained stage (green). Bins represent preferred orientation relative to the rewarded orientation. D, d′ values computed from responses to the same stimuli during orientation tuning measurements. E, Mean pairwise differences in d′ between task and tuning measurements across orientation bins. F, Cumulative distribution of relative orientation preferences across learning stages. G, Learning-related changes in tuning width as a function of orientation preference relative to the rewarded orientation, for all neurons recorded during the naive, intermediate, and trained stage. H, Tuning width (σ) and laminar location for neurons in G. Trends (vertical lines) were computed with locally weighted, robust regression (lowess). I, Absolute d′ values during task performance versus measurements of orientation tuning, separately for each learning stage. Included are only neurons from bins centered on 0 degrees or −90 degrees in C. J, Same for differences in firing rates between rewarded and unrewarded stimuli. K, Same for pooled SD; n = 11 mice; numbers of neurons per learning stage are given in C. Error bars indicate ±1 SEM.

Figure 6.

Contrast sensitivity in the population of V1 neurons. A–C, Contrast responses of three example V1 neurons (units M110-30-5, M22-14-46, M154-6-15) for the rewarded (blue) and unrewarded (red) stimulus. Because the two stimuli are not identical, these contrast responses are strongly determined by the neuron's preferred orientation (inset, black line). Data are mean ±1 SEM. A, Dotted lines indicate contrast sensitivity, defined as the contrast at half the maximum response. D, Contrast sensitivity for the rewarded stimulus across the depth of cortex, separately for each learning stage. E, Summary statistics of contrast sensitivity for the rewarded stimulus. Center marks represent medians. Box edges represent the 25th and 75th percentiles. Error bars indicate the range. F, Same for unrewarded stimulus. N = 11 mice. Conventions as in Figure 5.

Figure 7.

Running behavior during orientation discrimination learning. A, Two single-trial speed traces (Mouse M36). Dashed line indicates the speed threshold (1 cm/s) used to determine whether the animal was running or not. B, Percentage of trials during which animals were running around the onset of the rewarded (blue) or unrewarded (red) stimulus in the naive stage. Shaded areas represent ±1 SEM across recording sessions. Dashed line indicates stimulus onset. Black bar represents the time window used for the analysis in E. C, Same for intermediate stage. D, Same for trained stage. E, Bottom, Average percentage of run trials within 0.1–1.5 s after stimulus onset across learning stages (N, Naive; I, intermediate; T, trained). Top, Mean pairwise differences between rewarded and unrewarded conditions across recording sessions, together with their SEMs. F–H, Same as B–D, during measurements of orientation tuning but under identical sensory stimulation. I, Same as E, during measurements of orientation tuning. J, Maximum firing rate of V1 neurons during locomotion and stationary trials. K, Contrast sensitivity of V1 neurons during locomotion and stationary trials. L, Schematic summary of the effects: locomotion, on average, increases the maximum response without affecting contrast sensitivity.

Figure 8.

Learning-related changes in V1 processing are not an artifact of eye movements. A, Eye position was measured by tracking the pupil under infrared illumination. Top, Example image acquired by the eye-tracking camera. White spot indicates the cornea reflection of the infrared LED. Green cross represents the estimate of the pupil center. Bottom, Example traces for vertical and horizontal eye position. Dashed lines indicate saccades. B, Saccade frequency within 0.1–1.5 s after stimulus onset across learning stages. Data points indicate recording sessions. Error bars indicate ± 1 SEM. C, Distributions of vertical eye position. D, Distributions of horizontal eye position.

Figure 9.

Learning-related changes in V1 processing are not an artifact of licking behavior. A, B, Examples of perievent spike histograms triggered on licks. Firing rates of neuron 1 are weakly reduced with licking; firing rates of neuron 2 are not affected (M53-1-8, M117-22-21). C, Cumulative distributions of lick-modulation indices across the population of neurons during the naive (gray), intermediate (orange), and trained stage (green).

Figure 10.

VEPs during measurements of orientation tuning. A, VEPs from one recording session in response to gratings of six different orientations (M110–36). VEPs for the rewarded and unrewarded stimulus are shown in blue and red. B, Amplitudes of VEPs in A. C, Normalized VEP amplitude across stimulus orientations for all sessions recorded in the trained stage. Gray lines indicate individual recording sessions, normalized to the mean amplitude across orientations. Black line indicates the mean across sessions.