Reward Modulates Visual Responses in the Superficial Superior Colliculus of Mice

Abstract

The processing of sensory input is constantly adapting to behavioral demands and internal states. The drive to obtain reward, e.g., searching for water when thirsty, is a strong behavioral demand and associating the reward with its source, a certain environment or action, is paramount for survival. Here, we show that water reward increases subsequent visual activity in the superficial layers of the superior colliculus (SC), which receive direct input from the retina and belong to the earliest stages of visual processing. We trained mice of either sex to perform a visual decision task and recorded the activity of neurons in the SC using two-photon calcium imaging and high-density electrophysiological recordings. Responses to visual stimuli in around 20% of visually responsive neurons in the superficial SC were affected by reward delivered in the previous trial. Reward mostly increased visual responses independent from modulations due to pupil size changes. The modulation of visual responses by reward could not be explained by movements like licking. It was specific to responses to the following visual stimulus, independent of slow fluctuations in neural activity and independent of how often the stimulus was previously rewarded. Electrophysiological recordings confirmed these results and revealed that reward affected the early phase of the visual response around 80 ms after stimulus onset. Modulation of visual responses by reward, but not pupil size, significantly improved the performance of a population decoder to detect visual stimuli, indicating the relevance of reward modulation for the visual performance of the animal.SIGNIFICANCE STATEMENT To learn which actions lead to food, water, or safety, it is necessary to integrate the receiving of reward with sensory stimuli related to the reward. Cortical stages of sensory processing have been shown to represent stimulus-reward associations. Here, we show, however, that reward influences neurons at a much earlier stage of sensory processing, the superior colliculus (SC), receiving direct input from the retina. Visual responses were increased shortly after the animal received the water reward, which led to an improved stimulus signal in the population of these visual neurons. Reward modulation of early visual responses may thus improve perception of visual environments predictive of reward.

Keywords: decision task; mouse; reward; superior colliculus; vision.

Figure 1.

Superficial superior colliculus neurons were imaged during a visual decision task. A, Experimental setup (top left), correct choices for various stimulus conditions (top right), and time course of each trial (bottom). B, Psychometric curve. Percentage of left (pink), right (blue), or NoGo (yellow) choices (22 sessions in six mice) depending on stimulus contrast (or difference in contrast between stimuli). Trials in which the animal was disengaged, i.e., more than or equal to three consecutive NoGo trials, were discarded. Average performance, i.e., percentage of correct choices: 64.9 ± 1.4%. C, Reaction time (mean ± SEM across sessions, median per session) measured as time from go cue to time when stimulus reached its target position. Only Go trials were considered. D, Average frame showing one imaging plane of one two-photon imaging session (same session as data in E and F). E, Example dataset showing task events (stimulus contrast of 50% or 100%, feedback), calcium traces (z-scored) of 270 simultaneously recorded neurons (sorted by responsiveness to task events, as in Fig. 7A), and animal behavior (licks, pupil size, wheel turns). F, ON and OFF receptive fields (top, continuous outlines show Gaussian fits at 1.5 standard deviations (STDs)) of three SC neurons and task stimulus position (dashed outlines at 1 STD of Gaussian stimulus mask). Calcium traces (baseline subtracted) of same neurons to three contrasts of contralateral task stimulus. G, Receptive field (RF) positions and sizes (1.5 STDs, mean of short and long STD of elliptic Gaussian fit) of all recorded neurons (for which RF could be mapped) relative to stimulus position (331 neurons, 12 sessions in 5 mice). H, I, RF-stimulus overlap of all neurons with mapped RF (H) and all neurons with significant response to task stimulus (I).

Figure 2.

Visual responses in the superficial superior colliculus are modulated by previous feedback and pupil size. A, Calcium traces of sSC neuron in response to visual stimuli of different contrasts after animal received negative (left) or positive (right) feedback. Shaded region: 0.0–0.5 s after stimulus onset, window used to determine response amplitudes (in B, D). Horizontal line: at 0 ΔF/F. Scale bar: 0.2 ΔF/F (recorded traces were scaled to range from 0 to 1). B, Visual responses (same neuron as in A) after negative (black) or positive (orange) feedback, fitted with a hyperbolic ratio function. Responses were baseline subtracted (mean of 0–0.5 s before stimulus onset). C, D, Similar as A, B for a different neuron (same scalebar). Responses were split according to pupil size. E, Mean calcium traces in response to 100% contrast stimuli after negative (left) or positive (right) feedback. Only neurons with significant modulation by previous feedback are shown (80 neurons). Traces were scaled so that mean responses to negative feedback equal 1. F, Same as in E but for modulation by pupil size (79 neurons). G–J, Gain of visually responsive neurons (407) for previous feedback (G), pupil size (H), action (I), and outcome (J). Black dots: significantly different gains (p < 0.05, permutation test). Neurons with significantly increased/decreased responses: 63/17 following reward (G), 32/47 during large pupil (H), 17/17 during Go trials (I), 8/5 before correct choices (J). K–N, Cumulative distributions of response modulation (black line). Gray shade: 2.5th to 97.5th interval of null distribution.

Figure 3.

Decoding from visual responses improves after reward. A, Decoding scores of logistic regression models to detect presence of visual stimulus contralateral to the recorded SC neurons, tested on trials after positive versus negative feedback. Significant difference in predictive power per dataset (black dots) and across datasets (star) determined with permutation test (p < 0.05, 16 sessions in 4 mice). B–D, Same as in A, but trials were split by different task variables. E, Decoding scores for neural responses after positive versus negative feedback, only considering trials with fixed pupil size (left), fixed action (middle), or fixed outcome (right). Significant score differences across sessions marked by star (p < 0.05, permutation test, 16 sessions in 4 mice). F, Decoding scores of logistic regression models trained to predict trial outcome (correct or incorrect) based on visual stimulus in the current trial (0-0, L-0, or 0-R) and outcome of the previous three trials (T-1, T-2, T-3). All trained models performed better than chance (p < 0.05, permutation test, 16 sessions in 4 mice). G, Coefficients of logistic regression models to predict trial outcome. Black: coefficients are significantly different from 0 (p < 0.05, Wald test). Gray lines show medians of coefficients across sessions. In 10 of 16 sessions, reward in the previous trial increased the likelihood of correct outcome (see “T-1”).

Figure 4.

Modulation by previous feedback cannot be explained by changes in pupil size, eye position or licking. A, Example trace of pupil size and trial outcomes. B, Pupil size (mean across trials) following positive versus negative feedback (t₍₂₁₎ = 1.523, p = 0.143, paired t test). C, Response modulation (RM) by pupil size versus previous feedback (same data as shown in Fig. 2K,L). Black dots: pupil size and previous feedback significantly contributed to gain modulation. Tests: all data: Pearson's r = 0.23, p < 0.05, 407 neurons; black dots only: Pearson's r = 0.20, p = 0.30, 27 neurons. D, Eye positions at time of stimulus onset after animal received positive (green) or negative (purple) feedback during one task session (241 and 213 trials). E, Eye movements in vertical (left) and horizontal (right) direction at time of stimulus onset after animal has received positive or negative feedback. F, Lick rate locked to feedback onset (left) and stimulus onset (right) following positive or negative feedback (241 and 213 trials, 1 example session). G, Lick rate (18 sessions) before (−0.5–0 s) and after (0–0.5 s) visual stimulus onset following positive or negative feedback. Mice licked more following rewarded than nonrewarded trials (F_(1,64) = 6.25, p < 0.05), but lick rate was not significantly different between prestimulus and poststimulus periods (F_(1,64) = 0.002, p = 0.95). Test: two-way ANOVA [factors: time (prestimulus/poststimulus) × feedback]. H, I, Response modulation (RM) by previous feedback (H) and pupil size (I) for all trials (“control”) versus for no-lick trials (no licks in response window). RMs were not significantly different (feedback: z = 0.59, p = 0.56, 80 neurons; pupil size: z = 1.76, p = 0.08, 79 neurons; LMM). Only neurons significantly modulated by previous feedback or pupil size were considered. J, K, Same as in F, G but for whisking. Whisking was not significantly different between prestimulus and poststimulus periods (F_(1,64) = 0.27, p = 0.60, 17 sessions in 4 mice). Test: two-way ANOVA [factors: time (prestimulus/poststimulus) × feedback].

Figure 5.

Modulation by previous feedback is independent of slow response fluctuations and reward history. A, Quantification of visual response fluctuation (Materials and Methods, see Response fluctuation analysis). B, Auto-correlogram of fluctuation trace (mean across visually responsive neurons in one session). Gray shade: 2.5th to 97.5th percentile interval of null distribution. Inset, same data on smaller x-axis. C, Histogram of largest absolute lags with significant correlation strengths. D, E, Response modulation (RM) by previous feedback (D) and pupil size (E) as determined previously (“control”) and when accounting for response fluctuations. RMs were not significantly different (feedback: z = −0.238, p = 0.81, 99 neurons; pupil size: z = −0.193, p = 0.85, 89 neurons; LMM). Only neurons significantly modulated by previous feedback or pupil size were considered. F, Quantification of cumulative reward history. G, Cumulative reward history for trials with previously positive and negative feedback for two example sessions (left and right). Median reward histories (horizontal lines) are different in the right example (p < 0.0001, t test). H, Median reward history across trials with previously positive or negative feedback. Trials with previously positive reward had slightly more positive reward histories (z = 4.803, p = 0.001, LMM). I, Calcium traces of two example neurons aligned to stimulus onset for trials following reward. Trials with reward histories larger than the median (dark green) or smaller than the median (light green) were pooled. J, Response modulation (RM) by feedback as determined previously (“control”) and when accounting for reward history. RMs were not significantly different (z = −0.293, p = 0.77, 407 neurons, LMM). Black dots: neurons with significant feedback modulation. K, Response modulation (RM) by feedback in the previous trial (T-1) and by feedback experienced two trials earlier (T-2). Black dots: neurons with significant feedback (in T-1) modulation (80 neurons). RMs for T-1 were significantly different from 0 (t = 3.21, p < 0.01, t test), whereas RMs for T-2 were not (t = 0.60, p = 0.54, t test).

Figure 6.

Modulation by previous feedback extends to deep SC and affects early phase of visual responses. A, Coronal brain slice (DAPI staining) with tracks of electrophysiological probes (red) and overlay of brain area borders (left). Border between sSC and dSC was estimated at 310 µm below SC surface. Section of Allen CCF mouse atlas aligned to brain slice (right). B, Spike waveforms recorded on nearby channels of the Neuropixels probe for two sSC neurons. C, D, Spike times (top) and firing rate (bottom) aligned to stimulus onset (from same neurons as in B) after animal received negative (left) or positive (right) feedback. E, Firing rates in response to 100% contrast stimuli after negative (left) or positive (right) feedback. Only neurons with significant modulation by previous feedback are shown (32 neurons). F, Same as in E but for modulation by pupil size (16 neurons). G, H, Gain of visually responsive neurons (158, 15 sessions in 4 mice) for previous feedback (G) and pupil size (H). Black dots: significantly different gains (p < 0.05, permutation test). Neurons with significantly increased/decreased responses: 24/8 following reward (G), 12/4 during large pupil (H). I, J, Cumulative distributions of response modulation (black line). Gray shade: 2.5th to 97.5th interval of null distribution. K, Depth of neurons within SC modulated by different task variables. Number of significantly modulated sSC/dSC neurons: 16/16 by previous feedback, 6/10 by pupil size, 7/8 by action, and 6/9 by outcome. Dashed line: border between superficial and deep superior colliculus. L, Firing rates of two neurons aligned to stimulus onset after positive or negative feedback. M, Latency of peak visual response for neurons significantly modulated by previous feedback (black) and other neurons (gray). Inset, Earliest significant difference in firing rates after positive versus negative feedback, relative to response peak (p < 0.05, t test with Bonferroni–Holm correction). Only neurons with significantly larger responses after positive (green) or after negative (purple) feedback are considered.

Figure 7.

SC neurons respond to various task events. A, Mean calcium traces (traces were z-scored across recording) of 1751 task responsive neurons locked to onsets (dashed lines) of the visual stimulus (at 100% contrast), auditory go cue, wheel move, feedback, and licking. Order of neurons is the same across all plots. Neurons were first grouped by the first event they responded to (order of tested events as shown in plot), and then sorted by response amplitude. Responses to licking (last column) shown for 269 neurons (18 sessions, 5 mice); lick times not recorded for the rest of the sessions (indicated by light gray). B, C, Similar to A, but for mean firing rates of 184 task responsive sSC neurons (B) and 422 task responsive dSC neurons (C) recorded using electrophysiology. In A–C, traces were first z-scored across each recording. D, Number of sSC neurons (recorded with two-photon imaging) with significant responses to a specific task event (diagonal, numbers are percentage of task responsive neurons) or pairs of task events (below diagonal). Note that lick data were recorded for a subset of 2902 neurons. E, F, Similar to D, but for sSC (E) and dSC (F) neurons recorded using electrophysiology.