Mesencephalic representations of recent experience influence decision making

Abstract

Decisions are influenced by recent experience, but the neural basis for this phenomenon is not well understood. Here, we address this question in the context of action selection. We focused on activity in the pedunculopontine tegmental nucleus (PPTg), a mesencephalic region that provides input to several nuclei in the action selection network, in well-trained mice selecting actions based on sensory cues and recent trial history. We found that, at the time of action selection, the activity of many PPTg neurons reflected the action on the previous trial and its outcome, and the strength of this activity predicted the upcoming choice. Further, inactivating the PPTg predictably decreased the influence of recent experience on action selection. These findings suggest that PPTg input to downstream motor regions, where it can be integrated with other relevant information, provides a simple mechanism for incorporating recent experience into the computations underlying action selection.

Keywords: decision-making; mouse; neuroscience; pedunculopontine tegmental nucleus; sensorimotor.

Figure 1.. Behavior is influenced by previous trials.

(A) Timing of behavioral events for two consecutive trials (‘previous’ and ‘current’) of the spatial choice task. Gray box shows pre-stimulus epoch, the primary focus of our neurophysiological analyses. (B and C) Behavioral performance conditional on whether the choice on the previous trial was left (B) or right (C), and rewarded or unrewarded, for 1 example mouse (12 sessions). Gray circles and lines include all trials and are identical in B and C, shown for comparison with the conditional data (black). Lines show best-fit logistic functions using the Simple model. (D) Influence of previous choice and outcome on choice behavior for 7 mice (M1-M7), estimated with the Extended model, 1 trial back. Error bars, 95% confidence intervals. (E and F) Improvement in predicting choices realized by including in the behavioral model choices and outcomes 1 trial back (E) and 2 trials back (F) for all seven mice (corresponding to M1-M7 in D). Each symbol represents 1 session. For each session, simulated choices on a test set of trials (50% of trials in the session) were separately generated by three models (Simple model; Extended model, 1 trial back; and Extended model, 2 trial back) in which all regression coefficients were estimated from the remaining trials (in all sessions) for that mouse. The accuracy of each model’s prediction of choices in the test set was calculated as the percentage of trials in which the predicted choice matched the actual choice. This process was repeated, with a new test set, 50 times/session. Each symbol shows the average, across repeats for each session, of the improvement in accuracy of the Extended model, 1 trial back over the Simple model (E), and of the Extended model, 2 trials back over the Extended model, 1 trial back (F). *p<0.0001, one-tailed paired Student’s t test across sessions per mouse. DOI: http://dx.doi.org/10.7554/eLife.16572.003

Figure 2.. PPTg activity during the pre-stimulus epoch is influenced by choices and outcomes on previous trials.

(A and B) Rasters (top) and PSTHs (bottom) for one example neuron grouped by choice (A) and outcome (B) on the current trial. Fifty pseudorandomly selected ipsilateral and contralateral trials (A) and reward and no-reward trials (B) are shown in the rasters; all trials are included in the PSTHs. Activity is aligned to the time of odor valve open (blue lines) and sorted within each group by the time since odor port entry (orange ticks). PSTHs show average firing rate across all trials in each group, smoothed with a Gaussian filter (σ = 15 ms). Shading, ± s.e.m. Gray bar shows mean pre-stimulus epoch. (C and D) Preference during the pre-stimulus epoch for choice (C) and outcome (D) on the current trial across the population of neurons. Arrowheads indicate preferences for example neuron shown in A and B. (E, F, G and H) As in A, B, C and D with respect to 1 trial back instead of current trial. Data from same example neuron are shown. (I, J, K and L) As in A, B, C and D with respect to two trials back instead of current trial. Data from same example neuron are shown. DOI: http://dx.doi.org/10.7554/eLife.16572.004

Figure 3.. Dynamics of the influence of the choice and outcome on the previous and current trial on PPTg activity throughout the trial.

(A) For the example neuron shown in Figure 3, 95% confidence intervals are shown for each regression coefficient calculated in 100 ms bins (shifted by 10 ms), aligned to three trial events. (B) Average regression coefficients across all neurons that exhibited a significant preference for choice on the previous trial. Ribbons reflect mean ± SEM ${\beta }_{ Choice}^1$ (black) ${\beta }_{ Choice}^0$ (red) coefficient values as a function of time. (C) Fraction of neurons with activity in each 100 ms bin significantly influenced by choice on the previous or current trial, aligned as in A. (D) As in B, with respect to outcome instead of choice. (E) As in C, with respect to outcome instead of choice. DOI: http://dx.doi.org/10.7554/eLife.16572.005

Figure 4.. PPTg representation of previous choice affects behavior.

(A) Probability that the mouse chose the reward port corresponding to the previous-trial choice preference of an example neuron, plotted as a function of the firing rate of that neuron during the pre-stimulus epoch. Firing rates were normalized to the maximum for the neuron, separately within each previous choice condition. Only current trials with an ambiguous sensory cue are shown. Dashed line, best linear fit to raw data (antipreferred choice, y = 0; preferred choice, y = 1). Circles represent mean ± s.e.m. fraction preferred choices for binned normalized firing rates and are shown for display only (not used to fit the line). (B) Slope of best-fit line calculated as in A for all neurons exhibiting a significant preference for choice on the previous trial during the pre-stimulus epoch. Slopes were determined to be significantly different from 0 (p<0.05) with a Monte Carlo permutation test with 1000 repeats. Significance of slopes was identical when the data were fit to a logistic function rather than a line as shown in A. Arrowhead indicates slope for example neuron shown in A. (C) Fraction of significantly positive (white) and negative (black) slopes, and non-significant (gray) slopes (calculated as in B), separately for each mouse, numbered as in Figure 2. DOI: http://dx.doi.org/10.7554/eLife.16572.006

Figure 5.. Inactivating the PPTg causally affects behavior.

(A) Behavioral performance during example sessions after which either muscimol (black circles) or 0.9% saline (gray circles) was infused into the left PPTg. Saline was infused during the session before (gray circles, solid gray line) and after (gray triangles, dashed gray line) the muscimol session. One session was performed per day. Lines show best-fit logistic functions using the Simple model. Error bars, ± s.d. Behavioral data shown here and in all subsequent figures were collected from well-trained mice (Materials and methods). (B) ${\beta }_{Muscimol}$ calculated for all 3 mice and 30 sets of sessions. Positive values represent leftward (ipsilateral) influence of muscimol. Arrowhead indicates ${\beta }_{Muscimol}$ for example sessions shown in A. (C) Behavioral influence of the previous choice calculated separately for muscimol and saline sessions (same sessions shown in B). *p=0.021, one-tailed Mann-Whitney U test. DOI: http://dx.doi.org/10.7554/eLife.16572.007