Activation of serotonin neurons promotes active persistence in a probabilistic foraging task

Abstract

The neuromodulator serotonin (5-HT) has been implicated in a variety of functions that involve patience or impulse control. Many of these effects are consistent with a long-standing theory that 5-HT promotes behavioral inhibition, a motivational bias favoring passive over active behaviors. To further test this idea, we studied the impact of 5-HT in a probabilistic foraging task, in which mice must learn the statistics of the environment and infer when to leave a depleted foraging site for the next. Critically, mice were required to actively nose-poke in order to exploit a given site. We show that optogenetic activation of 5-HT neurons in the dorsal raphe nucleus increases the willingness of mice to actively attempt to exploit a reward site before giving up. These results indicate that behavioral inhibition is not an adequate description of 5-HT function and suggest that a unified account must be based on a higher-order function.

Fig. 1

The probabilistic foraging task. a Schematic drawing of the foraging task apparatus. Mice shuttle back and forth between two reward sites, located at the opposite ends of an elongated box, to obtain water rewards. b Example snapshot of foraging behavior. The one-dimensional location of an example mouse along the long axis of the box is plotted as a function of time. The ROIs around each water port are marked as dashed rectangles, and green and orange ticks above and below the trajectory mark nose-pokes into the right and left ports, respectively. The red asterisk marks an error trial. c Task events during a single trial. Each trial starts with an exit from one of the ROIs. Following shuttling to the other end, the mice would nose-poke multiple times and receive reward on some of the attempts on a probabilistic basis, before switching back. Green and orange rectangles mark rewards, gray rectangles mark omissions. d Example session, this time showing only the sequence of outcomes during nose-poking. Each column represents a single trial. Green/orange squares represent rewarded nose-pokes, gray squares, omissions

Fig. 2

Reward statistics and task performance. a In each trial, reward probabilities were drawn from one of the three exponentially decreasing functions, all sharing the same time constant but with a different scaling factor, labeled high, medium, and low. Dots mark hypothetical values; solid lines are averages derived from data. b Bar plot showing the average number of rewards in each of the trial types (n = 16 mice). Dashed lines mark maximal values (assuming mice stay in place and poke indefinitely). c Schematic drawing of an optimal-agent’s behavior during foraging. When plotting the average cumulative reward as a function of time from reward site exit, the average reward rate is the slope of the line that connects this curve at the time of leaving with the origin. Thus the slope of this line is maximal when it is tangent to the curve. Consequently, better or worse trials result in later or earlier leaving times, respectively (vertical lines and arrows). d Left: Scatter plot of reward rate at leaving vs. average reward rate. Each circle represents one mouse (n = 16). Dashed line is the unity diagonal and red line is a linear regression curve, with its correlation coefficient shown as well (p < 0.001). Right: Bar plot showing the average reward rate at leaving and the average reward rate. p > 0.05, Wilcoxon signed-rank test. e Cumulative distributions of the number of pokes per trial for the three trial types, averaged across mice (n = 16). f Bar plot showing the average number of pokes in each of the trial types. Bars on the left represent real data, and bars on the right represent shuffled data. Asterisk indicates significant effect (p < 0.05, ANOVA). g Cumulative distributions of the estimated reward probability after the last poke in a trial (i.e., at the time of switching) for the three trial types, averaged across mice (n = 16). h Bar plot showing the average reward probability after the last poke. Bars on the left represent real data, and bars on the right represent shuffled data. Asterisk indicates significant difference between trial types, p < 0.05, ANOVA

Fig. 3

The effect of DRN 5-HT photostimulation on switching behavior. a Scheme of the locations of ChR2-YFP expression and optic fiber placement (adapted with permission from ref. ). b Schematic diagram of task events during a single trail, also showing the period of photostimulation. Photostimulation was triggered by the first poke in 50% of correct trials and ended either when the mouse left the ROI or if 10 s had elapsed since the last poke. c Distributions of the number of pokes per trial for photostimulated trials (blue) and control trials (black) averaged across the population of SERT-Cre mice (n = 10). d Distributions of the number of pokes per trial for photostimulated (green) and control (black) trials, across the population of wild-type mice (n = 6). e Difference between average number of pokes in photostimulated and control trials for SERT-Cre (n = 10) and wild-type (n = 6) mice. Averages across mice are shown in filled circles. *p < 0.05, Wilcoxon rank-sum test. f Plot showing the average number of pokes in photostimulated and control trials for SERT-Cre mice (n = 10) as a function of consecutive numbers of photostimulated (blue) or control (black) trials. There was no effect of sequence length on behavior, p > 0.05, linear regression analysis

Fig. 4

Lack of effect of DRN 5-HT photostimulation on travel duration. a Schematic diagram of task events during a single trial showing the period of photostimulation. In the first protocol (photostimulation while poking), photostimulation was triggered by the first poke in 50% of correct trials, and in the second protocol (photostimulation while traveling), photostimulation was triggered by ROI exit in 50% of the correct trials and lasted for 2 s. We also defined two relevant time periods: (T1) time-to-leave—the interval between the last poke and ROI exit, and (T2) travel—the interval between ROI exit and subsequent ROI entry. b Left: Cumulative distributions of T1 durations for photostimulated (blue) and control (black) trials, averaged across the population of SERT-Cre mice (n = 10) in the photostimulation-during-poking protocol. Right: Bar plot showing the corresponding averages. *p < 0.05, Wilcoxon signed-rank test. c Left: Cumulative distributions of T2 durations for photostimulated (blue) and control (black) trials, averaged across the population of SERT-Cre mice (n = 10) in the photostimulation-during-poking protocol. Right: Bar plot showing the corresponding averages. p > 0.05, Wilcoxon signed-rank test. d Left: Cumulative distributions of T2 durations for photostimulated (blue) and control (black) trials, averaged across the population of SERT-Cre mice (n = 6) in the photostimulation-during-traveling protocol. Right: Bar plot showing the corresponding averages. p > 0.05, Wilcoxon signed-rank test

Fig. 5

Fitting behavioral data using the proportional hazards model. a Schematic drawing of proportional hazards model fitting pipe-line for two example consecutive trials. Nose-pokes (pooled across trials and sessions for each mouse independently) were used to fit a logistic regression model—the outcome of which was an estimated hazard rate that is reset at trial start and after each reward and is multiplicatively changed by the different coefficients’ values. This hazard rate can be viewed as an estimate of the probability of leaving after each nose-poke and can therefore be used to simulate mouse leaving decisions. Leaving decisions depicted as red circles. b Scatter plot of simulated vs. real average number of nose-pokes per trial. Each circle represents one mouse (n = 16). Dashed line is the unity diagonal and red line is a linear regression curve, with its correlation coefficient shown as well (p < 0.001). c Same as b for the effect of photostimulation on the number of nose-pokes per trial (p < 0.001). d The modeled hazard function for an example SERT-Cre mouse for phostostimulated and control nose-pokes. Note that decreased hazard means longer staying. e Cox regression photostimulation coefficient for SERT-Cre (n = 10) and wild-type (n = 6) mice. Averages across mice are shown in filled circles. *p < 0.05, Wilcoxon rank-sum test

Fig. 6

The effect of DRN 5-HT photostimulation on the microstructure of behavior. a Example trial represented in real time and starting from the last reward in that trial. Top: Each rectangle represents a single nose-poke (green marks the last reward and gray the subsequent omissions). Middle: estimated hazard for each nose-poke. Bottom: omission duty cycle (ODC). b Correlation between ODC and estimated hazard. The red line is a linear regression curve, with its correlation coefficient shown as well (p < 0.001). c Rewards reverse an overall (across trial) monotonic decrease in ODC (red) and an overall increase in hazard (black). The plot shows the average ODC and hazard as a function of poke number, aligned on the last reward and averaged across the population of mice (n = 16). Note that despite a decreasing (increasing) trend, the ODC (hazard) immediately after the last reward is higher (lower) than the one just before it. d Difference between average ODC and hazard immediately before the last reward compared to immediately after it across the population of mice (n = 16). *p < 0.05, Wilcoxon signed-rank test. e Left: The ODC for an example SERT-Cre mouse for phostostimulated and control nose-pokes aligned on last reward. The dashed lines are linear regression curves (p < 0.001 for the phostostimulation coefficient). Right: Regression photostimulation coefficients for SERT-Cre (n = 10) and wild-type (n = 6) mice. Averages across mice are shown in filled circles. *p < 0.05, Wilcoxon rank-sum test. f Left: The ODC for the same SERT-Cre mouse shown in d for phostostimulated and control nose-pokes as a function of estimated hazard. The dashed lines are linear regression curves, (p > 0.05 for the phostostimulation coefficient). Right: Regression photostimulation coefficients for SERT-Cre (n = 10) and wild type (n = 6) mice. Averages across mice are shown in filled circles. p > 0.05, Wilcoxon rank-sum test