Inhibiting LC-NE neurons or their terminals in the mPFC impair switching behavior.

(a) Schematic of DREADD inhibition in the LC and histological images showing DREADD(Gi) and TH (Tyrosine Hydroxylase) expression in the LC of a DBH-Cre mouse. (b) Task performance in the control (n = 4, WT) and test (n = 5) groups. Following systemic CNO injections, test group mice took more trials to complete extra-dimensional shift (EDS. Trial to reach the criterion: control vs. test, 14 ± 2 trials vs. 25 ± 2 trials, P = 0.028, t = -2.7). (c) Histology showing the expression of EYFP in LC-NE cell bodies (left) and their terminals (right) in the mPFC. AAV carrying Cre-dependent EYFP was injected in the LC of a DBH-Cre mouse. Scalebars: 100 μm. (d) Schematic of inhibiting LC terminals in mPFC and histology displaying cannula placement in the mPFC. (e) Task performance in the control (n = 7, WT) and test (n = 8) groups. Following localized CNO injection, test group mice took more trials to complete EDS (Trial to reach the criterion, control vs test: 16 ± 1 trials vs. 24 ± 2 trials, P = 2.0e-3, t = -3.9).

LC inhibition enhances mPFC engagement and broadens tuning.

(a) Illustration of miniscope recording in the mPFC with DREADD inhibition in the LC. (b) Top: Histology showing lens implant and GCaMP6f expression in the mPFC (prelimbic). Bottom: Snapshot of miniscope recording during behavior. (c) Example time series of fluorescence signals. Over 50 ROIs were acquired from this session. (d) Left to right: Example traces of individual mPFC neurons responding to choice (left), trial history (middle) and switch (right) based on activity prior to choice (gray bars). (e) Example behavioral progression. Each dot represents a trial. We define the initial mixed correct and incorrect trials (rule-learning) and the last set of consecutive correct trials (rule-following) as two different states in switching behavior. (f) Bar plots showing the percentage of mPFC neurons responding to task-related variables in the control (black) and test (red) groups. Control vs. test, choice responsive: 10% (59/593) vs. 17% (76/446), P = 7.7e-4; history responsive: 6% (34/593) vs. 13%, (57/446), P = 7.0e-5; switch responsive: 17% (102/593) vs. 25%, (111/446), P = 2.4e-3; overall fraction of responsive neurons: 27% (159/593) vs. 40% (178/446), P = 8.1e-6; the fraction of mixed tuning neurons among all responsive neurons: 20% (31/159) vs. 33% (59/178), P = 4.7e-3, Chi-squared test.

LC inhibition dampens mPFC population dynamics during switching.

(a) Population vectors of mPFC activity representing early (light color) and late (dark color) states in control (black, left) and test (red, right) groups. Each line represents a population vector from a subset of neurons. (b) Projection of population vectors in (a) onto the first two PCs. (c) Left: Euclidean distance (mean ± SEM) between state vectors aligned to choice for control (black) and test (red) groups. Arrows indicate maximal fluctuations prior to choice (peak). Right: Comparison of Euclidean distance quantified prior to choice for control (black) and test (red) groups. Control vs. test, 12.8 ± 0.05 vs. 8.9 ± 0.03, P = 6.8e-8, rank sum = 610, n = 20. Sample size represents number of bootstraps. (d) Comparison of peak Euclidean distance quantified prior to choice for control (black) and test (red) groups. Control vs. test: 4.1 ± 0.14 vs. 2.0 ± 0.07, P = 6.8e-8, rank sum = 610, n = 20. (e) Comparison of vector similarities between the early and late states for control and test groups. Correlation coefficient, control vs. test: 0.15 ± 0.03 vs. 0.95 ± 0.01, P = 6.8e-8, rank sum = 210, n = 20). Black and red dots indicate group mean in (c-e).

LC inhibition impairs mPFC encoding capacity of switching.

(a) Example behavioral state progression (solid curve: 0-early, 1-late) and hidden Markov model (HMM) predicted state progression (dashed curve) in a control session (black, left) and a test session (red, right). State prediction accuracy is 85% (control) and 71% (test). (b) Left: Cumulative distribution of the accuracy of predicting behavioral states in control (black) and test (red) groups. Sample size represents the total number of iterations that the model was tested (20 times per recording, 4 control mice and 5 test mice). Control vs. test: 0.89 ± 0.01 vs. 0.74 ± 0.02, P = 5.8e-7, rank sum = 9.0e3. Right: Cumulative distribution of the accuracy of predicting switch point in control (black) and test (pink) groups. Control vs. test: -4 ± 1 trials vs. -8 ± 1 trials, P = 4.2e-4, rank sum = 8.5e3. (c) Example sequences of animals’ choices (solid, top) and generalized liner model (GLM) predicted choices (dashed, bottom) in a control session (black, left) and a test session (red, right). Prediction accuracy is 82% (control) and 60% (test). (d) Cumulative distribution of the accuracy of predicting trial-by-trial choices in control (black) and test (red) groups. Control vs. test: 0.75 ± 0.01 vs. 0.68 ± 0.01, P = 6.0e-8, rank sum = 8.0e3.