Optogenetic stimulation of BF-mPFC cholinergic projections during threats impairs discrimination of two probabilistic outcome contingencies.

(A) The maze used in the probabilistic spatial learning task (left). The probability of threat and reward assigned to each of two paths connecting two reward sites (right). Mice received air puffs 75% of the time on one of the paths (high-threat path) and 25% of the time on the other path (low-threat path). The color of the squares indicates the locations on the maze as specified in the illustration on the left. (B) The site of GRIN lens implantation and viral vector infusion. (C) Images showing ChrimsonR-tdTomato expressing cholinergic terminals (red) near the GCaMP6f-expressing neurons (green) in the mPFC (top). Images showing colocalization of ChAT and virally infected cells expressing tdTomato in the BF (bottom). The scale bar represents 100 µm. (D) The proportion of laps in which mice chose the low-threat path (% adaptive path choice, one thin line/mouse, mean ± s.e.m.). The table on top indicates the assigned threat probability for each path. Adaptive path choice in the pre-training (T5) and random stage (R1-5) was defined as choosing the path that was the low-threat path during the probabilistic stage (P1-5). (E) The movement speed of mice while they ran toward the threat sites in P5 (mean ± s.e.m.). (F) The approaching speed toward the threat sites on two paths (one thin line/mouse, mean ± s.e.m.). The speed was averaged over a 500-msec window before the threat site entry. (G) The distribution of the speed approaching the threat site in all laps in P5 in a representative mouse. Laps with the approaching speed below or above the median were categorized as “slow” and “fast” laps, respectively. Laps on the high-threat path (grey) were likely to be slow laps, while laps on the low-threat path (white) were likely to be fast laps, consistent with differential threat expectations between the paths. (H) The movement speed around the threat sites in fast and slow laps in P5 (mean ± s.e.m.). The speed in fast laps with threats and slow laps with omissions showed the mice’s reaction to surprising outcomes (red). Conversely, the speed in slow laps with threats and fast laps with omissions showed their reaction to expected outcomes (blue). (I) The averaged speed during a 500-msec window starting from the threat site entry (one thin line/mouse, mean ± s.e.m.). Laps were categorized into the combination of the approaching speed (fast or slow) and outcomes (threat or omission) and labeled as “expected” and “surprising” as defined in H.

Cholinergic terminal stimulation augments PL cell activity in response to threats.

(A) Field of views in a representative mouse from the last pre-training session (T5) and the fifth probabilistic session (P5; top). White contours represent registered cells of the corresponding session. Calcium traces of extracted cells (bottom). Scale bars represent 50 s × 5 z-score. (B) Pseudocolor plots showing the z-normalized cell activity aligned to the threat site entry during puff-omission (left), puff (middle), and puff and LED (right) laps. In all panels, cells were sorted by the response magnitude in puff and LED laps. (C) Z-normalized cell activity aligned to the threat site entry of threat-responding cells (mean ± s.e.m.). (inset) The proportion of cells with increased activity (blue) and those with decreased activity (green). (D) The distribution of the change in cell activity evoked by air puffs or air puffs and LED. Horizontal lines show the median. Z-normalized activity was averaged over a 500-msec window starting from the threat site entry in each lap type. The average activity in omission laps was subtracted from the average activity in puff laps or puff and LED laps.

Cholinergic terminal stimulation abolishes learning-induced strengthening of PL cell responses to rare threat omission.

(A) The changes in PL cell activity upon the threat site entry in tdTomato mice. a. Pseudocolor plots showing the z-normalized cell activity aligned to the threat site entry during the first (P1) and last session (P5) of the probabilistic stage. Laps were categorized into four types based on the path (high-threat path, ht-p; low-threat path, lt-p) and the threat state (threat delivery or omission). Cells were sorted by their activity in the threat site in each panel. b. Venn diagrams depicting the proportion of cells responding to threats and their omissions on the ht-p (blue) and the lt-p (green) as well as their overlap (yellow) in P5. The numbers show the percentage of cells in each category. c. The distribution of z-normalized activity in the threat sites in P1 (light grey) and P5 (dark grey) with horizontal bars depicting the median. (B) The same as A for PL cell activity in ChrimsonR-expressing mice.

PL cells differentiate outcome-related activity depending on outcome expectations.

(A) Pseudocolor plots showing the differentiation of outcome-related activity between expected and surprising outcomes during the first (P1) and last session (P5) of the probabilistic stage. The color depicts the difference in z-normalized cell activity between laps with expected outcomes and lap with surprising outcomes. Cells were sorted by the activity difference in the threat site in each panel. (B) Expectation-dependent differentiation of cell activity in response to threats (x axis) and their omission (y axis; one dot per cell) in P5. The z-normalized activity was averaged during a 500-msec window starting from the threat site entry. The averaged values in laps with expected outcomes were subtracted from those with surprising outcomes (differential activity). Blue squares highlight cells with the same response direction to surprising threats and omissions (unsigned prediction errors). Red squares highlight cells with an opposite response direction to surprising threats and omissions (signed prediction errors). (C) The averaged z-normalized activity across cells with different types of activity differentiation (mean ± s.e.m.). A cell was selected when its differential activity for threats or their omission was greater than 2 (Up) or smaller than -2 (Down). Cells with significant differential activity for both threat delivery and omission were then further categorized into four types depending on their response direction. The activity of cells with significant differentiation only for one outcome type was depicted in Fig S5. (D) The proportion of cells in various differential activity types. Cells selective for outcome expectations were categorized into four types as defined in C and Fig S5. Error bars show the upper and lower confidence limit (α = 0.05). PE, prediction errors. R5, the last session of the random stage.

Threat-evoked phasic cholinergic signals constrain learning from surprising outcomes in a stable but probabilistic environment.

(A) Sites of viral vector infusion and optic fiber implantation. (B) Representative images showing co-localization of ChAT and virally infected cells expressing tdTomato. The scale bar represents 100 µm. (C) The effect of cholinergic terminal stimulation on mice’s behavior during the probabilistic paradigm. a. Schematic representation of threat probability and light stimulation on each path. LED stimulation was applied to the PL to activate the BF cholinergic terminals during every threat delivery. b. The percentage of laps in which mice chose the low-threat path. c. Movement trajectories of a representative mouse from each group. The locations of threat and reward sites were specified with the same color scheme as in Fig 1A. (D) Same as C for the deterministic paradigm.

Enhanced threat-evoked phasic cholinergic signals impair the integration of outcome information over time.

(A) The probability of choosing the low-threat path (lt-p) after an air-puff delivery on the high-threat path (ht-p; left; one thin line/mouse, mean ± s.e.m.). Only mice with usable data in all five sessions were used for this analysis. (B) Correlation coefficients between the actual path choice and threat probability calculated with the different number of previous laps from representative mice. The greater positive values for the ht-p and negative values for the lt-p indicate stronger correlations with the estimated threat probability (see Fig S7C). (C) Numbers of mice showing significant correlations between the estimated threat probability and the choice of the ht-p (top).

Schematic representations of aversive prediction errors in the deterministic and probabilistic paradigms.

Aversive prediction errors become positive for unexpected threat delivery while negative for unexpected threat omission(97,98). Red crosses indicate prediction error signals with abnormally large amplitude. We argue that the role of outcome-evoked phasic cholinergic signals is to suppress learning from these occasional surprising outcomes.