Natural phasic inhibition of dopamine neurons signals cognitive rigidity

When animals unexpectedly fail, their dopamine neurons undergo phasic inhibition that canonically drives extinction learning—a cognitive-flexibility mechanism for discarding outdated strategies. However, the existing evidence equates natural and artificial phasic inhibition, despite their spatiotemporal differences. Addressing this gap, we targeted a GABAA-receptor antagonist precisely to dopamine neurons, yielding three unexpected findings. First, this intervention blocked natural phasic inhibition selectively, leaving tonic activity unaffected. Second, blocking natural phasic inhibition accelerated extinction learning—opposite to canonical mechanisms. Third, our approach selectively benefitted perseverative mice, restoring rapid extinction without affecting new reward learning. Our findings reveal that extinction learning is rapid by default and slowed by natural phasic inhibition—challenging foundational learning theories, while delineating a synaptic mechanism and therapeutic target for cognitive rigidity.


J-K:
Pauses vs other features.Correlation between all other metrics and %PSI; format as above.Note that this figure shows raw (non-normalized) anticipatory licking, whereas the main-text figures display normalized anticipatory licking, with day-10 anticipatory as a constant of normalization for each animal.B: Behavioral inclusion criteria.We required robust anticipatory licking to cue A (raw anticipatory licking greater than 0.2) and low responsiveness to cue B probes (less than 30% of cue A anticipatory licking).A total of 9 mice (3 dd HTP and 6 + HTP) were excluded for lack of cue A responsiveness (blue).Of the remaining 27 mice, only 3 mice (2 dd HTP and 1 + HTP) were excluded for probe generalization (green).Thus 89% (24 of 27) successfully discriminated cue B. C: Locomotion.Ratio of the average treadmill RPM post-gabazine DART (day 11-12) divided by pre-gabazineDART (day 8-10).Individual mice (circles), group means (thin horizontal lines), mean-difference bootstrap (pink distribution), and 95% CI of the two-sided permutation test (vertical black bar).As previously reported ( 43), disinhibition of VTADA neurons enhances locomotion.D: Locomotion vs histology.Correlation between RPM ratio and Alexa647 DART capture in the dorsal VTA of + HTP mice (n=12).Mice (circles), regression ±95% CI (line and shading).Pearson's r 2 = 0.08, P = 0.4 indicates no significant correlation.E: Locomotion vs reward learning.Correlation between RPM ratio and our two measures of reward learning: conditioning AUC (left) and extinction AUC (right).Mice (circles; n=12 dd HTP; n=12 + HTP) and regression ±95% CI (line and shading).Pearson's tests show no significant correlation (r 2 and P values as indicated).F: Extinction learning vs histology.Correlation between extinction (AUC) and Alexa647 DART capture in the dorsal VTA of + HTP mice (n=12).Mice (circles), regression ±95% CI (line and shading).Pearson's r 2 = 0.36, P = 0.04 indicates a significant correlation, with higher levels of target engagement corresponding to faster rates of extinction.

Fig. S2 :
Fig. S2: Supporting data for in vivo electrophysiology A: Control dd HTP metrics.Analysis of tonic, burst, and pause stability in dd HTP mice.Steady-state Δnorm (1-hr post-gabazine DART ) with individual cells (circles), group means (thin horizontal lines), mean-difference bootstrap (grey distribution), and 95% CI of the two-sided permutation test (vertical black bar), comparing dd HTP cells to zero.

Fig. S3 :
Fig.S3: Supporting data for Pavlovian extinction and conditioning assay A: Training sessions.Lines and shading show anticipatory licking (fraction of time that beam is broken during the cue), mean ± SEM over mice (n = 12 dd HTP; n = 12 + HTP).Both + HTP and dd HTP mice develop robust anticipatory licking to cue A across training, while exhibiting little to no background licking during silent trials.Note that this figure shows raw (non-normalized) anticipatory licking, whereas the main-text figures display normalized anticipatory licking, with day-10 anticipatory as a constant of normalization for each animal.B: Behavioral inclusion criteria.We required robust anticipatory licking to cue A (raw anticipatory licking greater than 0.2) and low responsiveness to cue B probes (less than 30% of cue A anticipatory licking).A total of 9 mice (3 dd HTP and 6 + HTP) were excluded for lack of cue A responsiveness (blue).Of the remaining 27 mice, only 3 mice (2 dd HTP and 1 + HTP) were excluded for probe generalization (green).Thus 89% (24 of 27) successfully discriminated cue B. C: Locomotion.Ratio of the average treadmill RPM post-gabazine DART (day 11-12) divided by pre-gabazineDART (day 8-10).Individual mice (circles), group means (thin horizontal lines), mean-difference bootstrap (pink distribution), and 95% CI of the two-sided permutation test (vertical black bar).As previously reported (43), disinhibition of VTADA neurons enhances locomotion.D: Locomotion vs histology.Correlation between RPM ratio and Alexa647 DART capture in the dorsal VTA of + HTP mice (n=12).Mice (circles), regression ±95% CI (line and shading).Pearson's r 2 = 0.08, P = 0.4 indicates no significant correlation.E: Locomotion vs reward learning.Correlation between RPM ratio and our two measures of reward learning: conditioning AUC (left) and extinction AUC (right).Mice (circles; n=12 dd HTP; n=12 + HTP) and regression ±95% CI (line and shading).Pearson's tests show no significant correlation (r 2 and P values as indicated).F: Extinction learning vs histology.Correlation between extinction (AUC) and Alexa647 DART capture in the dorsal VTA of + HTP mice (n=12).Mice (circles), regression ±95% CI (line and shading).Pearson's r 2 = 0.36, P = 0.04 indicates a significant correlation, with higher levels of target engagement corresponding to faster rates of extinction.

Fig. S5 :
Fig. S5: Supporting data for within-mouse behavioral correlations A: Visual inspection of random order of interleaved trials.Each row is one mouse, sorted by extinction from the slowest to fastest (top to bottom).Columns indicate the first 25 trials after the rule-change, with trial type indicated by color (yellow = cue A extinction, blue = cue B conditioning).No visually discernable pattern is apparent in either + HTP mice (n=12) or dd HTP mice (n=12).B. Initial training vs later reward learning.Correlation between initial learning rates (AUC over training days 1-2) and our two measures of reward learning: extinction AUC (top) and conditioning AUC (bottom).Mice (circles; n=12 dd HTP; n=12 + HTP) and regression ±95% CI (line and shading).Pearson's tests show no significant correlation (r 2 and P values as indicated).C: Phenotypic spectrum across pooled controls.Conditioning-AUC vs extinction-AUC measured withinmouse.Individual mice (circles), regression fit (line), are shown for mice pooled from ongoing control experiments.All data are from dd HTP mice infused with various ligands, including DART.2 , blank.1 DART.2 ,

Table S2 . Detailed Author Contributions
HTP or dd HTP virus injected VTA DA neurons.Performed spontaneous and evoked action potentials recording by current-clamp, evaluated which gabazine.7 DART.2 effects on membrane excitability and endogenous channels from + HTP virus injected VTA DA neurons.Performed AMPAR-and NMDARmediated eEPSCs recording by voltage clamp and evaluated gabazine.7DART.2effectsonexcitatorysynapticfunction in + HTP positive VTA DA neurons.Performed evoked action potentials recording in + HTP positive VTA DA neurons, tested the effects of blank.1 DART.2 + Alexa647.1DART.2.Wrote original draft of methods for electrophysiology in brain slices, and prepared original version of Fig.S1.Reviewed and provided feedback on manuscript.Designed and cloned different variants of the HTP plasmid constructs, including the optimized variant in the experiment.Assisted with validating HTP expression and HTL capture in dopamine neurons.Separately validated locomotor effects from tethering gabazine.7DART.2onVTA dopamine neurons.Reviewed and provided feedback on manuscript. +