Enhanced perceptual task performance without deprivation in mice using medial forebrain bundle stimulation

Abstract

Perceptual decision-making tasks are essential to many fields of neuroscience. Current protocols generally reward deprived animals with water. However, balancing animals' deprivation level with their well-being is challenging, and trial number is limited by satiation. Here, we present electrical stimulation of the medial forebrain bundle (MFB) as an alternative that avoids deprivation while yielding stable motivation for thousands of trials. Using licking or lever press as a report, MFB animals learnt auditory discrimination tasks at similar speed to water-deprived mice. Moreover, they more reliably reached higher accuracy in harder tasks, performing up to 4,500 trials per session without loss of motivation. MFB stimulation did not impact the underlying sensory behavior since psychometric parameters and response times are preserved. MFB mice lacked signs of metabolic or behavioral stress compared with water-deprived mice. Overall, MFB stimulation is a highly promising tool for task learning because it enhances task performance while avoiding deprivation.

Keywords: Go/No-go; audition; behavior; lever pressing; licking; medial forebrain bundle; motivation; mouse; perception; perceptual decision-making; reinforcement learning; satiation.

Graphical abstract

Figure 1

Protocol for perceptual discrimination learning with MFB reward (A) Medial forebrain bundle (MFB) implantation site of bipolar electrode fixed with dental cement to skull. (B) Photo of head-fixed mouse showing head post and MFB electrode (left) and mouse with 5-mm cranial window over the auditory cortex, head-post, and MFB electrode (right), illustrating that the small MFB electrode can be combined with recording methods. (C) Structure of the Go/No-go step of the discrimination task. (D) Scheme of head-fixed mouse during task: pressing the lever with the forepaw after Go sound elicits stimulation via the implanted electrode. (E) Calibration curve for an example mouse. Measured probability of pressing the lever during 1.5-s response window after Go sound for MFB stimulation of different voltages (black dots) and sigmoid curve fit to data (gray line). Arrow indicates voltage used for learning. (F) Traces of lever press for 30 Go and 30 No-go trials from an example mouse after training (top) and the mean lever press trace for all trials (bottom). For each Go trial, the time of reward (50 ms after lever press) is shown by a green dot, and for each No-go trial, trials punished by a time out are indicated by a red dot. The distribution of reward timings is shown in green. (G–I) As in (D)–(F) for licking. See also Videos S1 and S2.

Figure 2

MFB-rewarded mice learn tasks at equal rates to water-deprived mice and reach higher peak accuracy on hard tasks (A–C) Learning curves for water-deprived (blue) and MFB-rewarded (red and yellow) mice performing the pure tone (PTvsPT) discrimination task ([A] n = 7, 8, 8), the pure tone psychometric task (PTvsPT-PS) requiring classification of high and low pure tones ([B], n = 7, 8, 8), and the pure tone vs. frequency modulated (PTvsFM) discrimination task ([C], n = 7, 4). Shaded areas indicate SEM. See Figure S4 for independent evolution of Go vs. No-go accuracy. (D–F) Number of days necessary to reach criterion accuracy for water-deprived and MFB-rewarded mice for the three tasks as in (A)–(C) (Wilcoxon rank-sum test, water vs. MFB lever p values: 0.74, 0.019, 0.29, rank-sum statistic: 52.5, 79, 48; water vs. MFB lick p values: 0.37, 0.004, rank-sum statistic: 48,75; n = 7, 8, 8; 7, 8, 9; 7, 4). The vertical bar in (F) indicates that variability was higher for PTvsFM task for water-deprived mice (Levene test, p value: 0.0156, Levene statistic: 8.8). For water-deprived mice that never reached criterion in the PTvsFM task, the maximum day of training is shown as a cross. (G) Average psychometric function for pure tone frequency classification, drawn from the best 800 consecutive trials of each mouse for the water-deprived and MFB-rewarded mice (n = 7, 8, 8). Shaded areas indicate SEM. (H) Accuracy of the best session for water-deprived and MFB-rewarded mice for the PTvsFM discrimination task. (Wilcoxon rank-sum test, p value: 0.02, rank-sum statistic: 30, n = 7, 4). See also Video S3.

Figure 3

MFB-rewarded mice perform thousands of trials per session (A) Example sessions accuracy on Go (green) and No-go (red) trials for a water-deprived and an MFB lever-pressing mouse in the PTvsPT task. Go accuracy drops for the water-deprived mouse after 240 trials, whereas the MFB-rewarded mouse is still performing above the criterion after 2,000 trials. (B and C) Average of Go (green) and No-go (red) accuracies for water-deprived (B) and MFB lever-pressing (C) mice in the PTvsPT task. For MFB mice, in order to cumulate sessions of varying duration (600–2,000 trials), time was normalized between 0 and 1 (n = 7, 8). Shaded areas indicate SEM. (D) Trial count of several sessions in the PTvsPT-PS task (>80% accuracy) with varying trial duration. The black line corresponds to the maximum possible number of trials mice could reach given the session duration (supposing no punishment intervals or delayed trials by licking outside of trials). Shaded areas provide the range of trial numbers found with other strategies in the literature: optogenetic stimulation of dopaminergic fibers (DA), water deprivation, and water deprivation by adding citric acid to drinking water. (E) Accuracy over 200 trials for different trial speeds for MFB-rewarded mice performing the PTvsPT task by lever pressing or licking. Note that 500-ms inter-trial intervals provide the same accuracy as 5000-ms-spaced trials (n = 4, 6). (F) Trial count for 4-h session with 2-s ITI for MFB-rewarded mice performing the PTvsPT task by lever pressing or licking. Each point is one session (lever: n = 8 sessions from n = 6 mice, lick: n = 9 sessions from n = 6 mice).

Figure 4

MFB-stimulated mice show no signs of increased agitation or impulsivity (A) Movement level for hit and false alarm (FA) trials during the psychometrics task. The difference between the two curves is indicative of the movement induced by the reward since hit trials contain movement induced by the response (licking/lever press) and reward (water delivery or MFB stimulation), whereas FA trials only contain movement induced by the response. The movement level for each mouse is normalized by the peak of FA movement, and therefore the absolute levels of movement between MFB and water cannot be compared (see STAR Methods for details, n = 4, 8, 8, error bars are SEM). (B) Average accuracy on Go and No-go trials in psychometrics task during the three best sessions for water-deprived (left) and MFB-rewarded lever-pressing and licking mice. Note that all groups are biased toward Go-responding, although the effect is weaker in lever-pressing MFB mice (Wilcoxon signed rank test, p values = 0.04, 0.19, 0.02; signed rank statistic: 26, 28, 34; n = 7, 8, 8). (C) Ratio of Go to No-Go accuracy showing that all groups have similar levels of impulsivity based on this measure. (Wilcoxon rank-sum test, p values: 0.12, 0.77, 0.19; rank-sum statistic: 70, 59, 55; n = 7, 8, 8). (D) Average wait period between trials for water-deprived and MFB-rewarded mice that captures the tendency of mice to lick/lever press during the wait period. MFB-rewarded mice are significantly more restrained during the wait period (Wilcoxon rank-sum test, p values: 6.2 E-4, 1.5 E-4, 0.003; rank-sum statistic: 98, 100, 41; n = 7, 8, 8).

Figure 5

Improved mouse well-being using MFB stimulation relative to water deprivation protocol (A) Weight evolution starting from the week prior to any behavioral experiments (isolated point) and across the first 5 weeks during which mice were trained five days a week. (n = 7, 8 mice, error bars are SEM). In this panel and following, only data from MFB lever-pressing mice were collected. (B) Weight evolution averaged across days of the week (n = 7, 8 mice, error bars are SEM). (C) Top: temporal distribution of periods of activity during the lights-off period for each session for a cage containing two animals (black bar indicates a period during which wheel running was detected). Bottom: mean activity throughout the lights-off period for MFB and water-deprived mice (blue and red) with the mean activity of control mice (gray) with no interventions overlayed. (D) Percent of time spent active throughout the whole night for MFB and water-deprived animals as well as control animals with no interventions (Wilcoxon rank-sum test, p values: 0.007, 0.02; rank-sum statistic: 73, 58; n = 7, 7, 6 sessions). (E) Superposed trajectories for mice during the open field session. (F) Ratio of total distance traveled (left) and number of entries into the center of the field (right) before and after training during 5-min exploration of the open field (Wilcoxon rank-sum test, p values: 0.039, 0.028; rank-sum statistic: 83, 89; n = 7, 8 mice).

Figure 6

Similar task dependance of discrimination times for MFB and water-reward protocols (A) Average lever-press probability for PTvsPT and PTvsFM tasks across all Go and No-go trials for MFB-rewarded mice. Arrow shows the estimated discrimination time defined as the earliest time at which p value of a t test for the difference between the Go and No-go traces (bottom panel) is below 0.01. Shaded areas are SEM. (B) Discrimination time as defined in (A) for each animal. (n = 8, 4 Wilcoxon rank-sum test, p value = 0.03; rank-sum statistic: 23). (C) Average lick probability for PTvsPT and PTvsFM tasks across all Go and No-go trials for water-deprived mice. (D) Discrimination time, defined as earliest time where Go and No-go lick signals are significantly different, for the PTvsPT and PTvsFM tasks (n = 7, 4; Wilcoxon rank-sum test, p value = 0.001; rank-sum statistic: 23; only animals that reached criterion accuracy for PTvsFM task are included).