Figures and data
Mice display flexible discrimination learning in a Go/No-go task in an experimental setup allowing for task-based fMRI.
(A) Illustration of behavioral setup and task design. Odors were presented pseudo-randomly for 1 second with a response window lasting 2 seconds from odor onset. (B) Learning curve showing the averaged ratio of correct responses (Hit and Correct Rejection events) during task acquisition (n = 12). (C) Averaged reaction time of lick responses to Hit trials as a function of session. (D) Learning curve showing averaged ratio of correct responses (n = 6). Blue line indicates the last acquisition session prior to reversal. Purple line indicates the reversal phase. (E) Averaged reaction time of lick responses to Hit trials as a function of sessions during the reversal phase. (F) Pairwise comparison of the averaged reaction time for the first session in acquisition vs. reversal. (G) Averaged probability of choosing to lick in response to ‘go’ trials (n = 6). The gray dashed line indicates the rule reversal onset (95% CIacquisition = [0.580, 0.966], 95% CIreversal = [0.157, 0.942]). (H) Decision variable (Qlick – Qno-lick) parameter of a representative animal as computed by the Q-learning model throughout the experiment. The yellow line represents odor A and the orange line represents odor B. The gray dashed line indicates the time of rule reversal. Data are shown as mean ± SEM (B, C, D and E). *p < 0.05.
Neural substrates of Q-learning signals during go/no-go task acquisition.
Group-level parametric one-sample t statistic maps showing BOLD response correlates of the decision variable values from the Q-learning model (n = 12 mice). The red color spectrum indicates areas with positive correlations to decision variable values, while blue colors indicate negative correlations. Maps are presented on the averaged raw fMRI data (spin-echo echo planar imaging) and annotated based on the Allen Mouse Brain Atlas; p < 0.05, corrected for multiple comparisons using family-wise error correction, voxel extension of 5. ACB, nucleus accumbens; AI, agranular insular area; AON, anterior olfactory nuclei; BST, bed nucleus of the stria terminalis; GU, gustatory cortex; DLS, dorsolateral striatum; DMS, dorsomedial striatum; GPe, globus pallidus externus; TT, tenia tecta; d, dorsal; v, ventral.
Neural substrates of Q-learning signals during reversal learning.
Group-level non-parametric maps showing BOLD response correlates of the decision variable values from the Q-learning model (n = 6 mice). The red color spectrum indicates areas with positive correlations to the decision variable values, while blue colors indicate negative correlations. Maps are presented on an average raw fMRI data (spin-echo echo planar imaging) and annotated based on the Allen Mouse Brain Atlas; p < 0.05, corrected for multiple comparisons using family-wise error correction, voxel extension of 3. ACB, nucleus accumbens; AI, agranular insular area; AON, anterior olfactory nuclei; DMS, dorsomedial striatum; GPe, globus pallidus externus; GU, gustatory cortex; HIPP, hippocampus; MS, medial septum; ORB, orbitofrontal cortex; PAG, periaqueductal gray.
Opposing roles of the periaqueductal gray (PAG) and nucleus accumbens (ACB) in mediating adaptive behavior in proficient mice.
Region of interest (ROI) analysis showing the time course fMRI BOLD response in PAG (top) and ACB (bottom) for the fifth session in the reversal stage. (A) ROI masks are presented on an average raw fMRI image (spin-echo echo planar imaging). (B) BOLD fMRI Responses to Hit (lick response to go trials; green) and (C) Correct Rejection (no lick to no-go trials; blue) are shown. The thick lines represent the group averaged response and the thin lines show individual animals. The gray boxes at time zero depict the odor stimulus timing (1 s). Group mean area under the curve (AUC) of the fMRI response (filled circles) and individual animals (open circles) demonstrate consistent responses at the group and individual animal levels.
PAG responses do not contribute to adaptive behavior in proficient mice during Acquisition.
(A) Pairwise comparison of behavioral performance during Acquisition vs. Reversal for sessions 4/5. (B) FIR responses for Hit condition during Acquisition (left). Pairwise comparison of AUC values showing individual animals for Acquisition vs. Reversal (right). (C) FIR responses for Correct Rejection condition during Acquisition (left). Pairwise comparison of AUC values showing individual animals for Acquisition vs. Reversal (right). For FIR responses, the thick lines represent group averages and the thin lines show individual animals. The gray boxes at time zero depict the odor stimulus timing (1 s). *p < 0.05, #p < 0.1, n.s. no significance. Insets show PAG region of interest mask presented on an average raw fMRI image (spin-echo echo planar imaging).
Group parameters computed by the Q-learning algorithm.
Fitting parameters are shown for the two modeling groups: acquisition only (blue) and acquisition following reversal (purple; n = 6 per group). (A) Learning rate parameter alpha. (B) Beta estimates for intercept (bias to lick), odor 1 (go stimulus), odor 2 (no-go stimulus) and control for changes in airflow during final valve opening (nitrogen).
Group performance for the different discrimination conditions during Reversal.
Behavioral performance rates across five consecutive sessions during the Reversal phase divided by the lick response of the animal to each cue type: Hit (lick to go odor; green), Miss (no lick to go odor; yellow), False Alarm (FA, lick to no-go odor; red) and Correct Rejection (CR, no lick to no-go odor; light blue). Filled circles indicate group mean and open circles indicate individual subjects (n = 6).
Spatial localization of the periaqueductal gray across fMRI and atlas space.
(A) Periaqueductal gray mask (yellow) defined based on regions showing a significant BOLD response in the whole-brain analysis and further used in the ROI analysis. The ROI mask is presented on an average raw fMRI data (spin-echo echo planar imaging), shown as sequential coronal slices with a slice thickness of 300 µm. (B) Coronal images (left, atlas; right, two-photon tomography) taken from the Allen mouse brain connectivity atlas that correspond to the spatial location of the fMRI data. The region highlighted in purple denotes the Periaqueductal gray as defined by the Allen Institute. Image identification numbers are 87 (bottom) and 90 (top) in the reference atlas.
