24We used two-photon microscopy to study the role of ensembles of cerebellar molecular 25 layer interneurons (MLIs) in a go-no go task where mice obtain a sugar water reward if 26 they lick a spout in the presence of the rewarded odorant and avoid punishment when they 27 restrain from licking for the unrewarded odorant. When the animal was naï ve to the valence 28 of the stimulus (is the odorant rewarded?) responses of the MLIs did not differ between 29 odorants. However, as the animal became proficient, the rewarded odorant elicited a large 30 increase in Ca 2+ in MLIs, and the identity of the odorant could be decoded from the 31 differential response. Importantly, MLIs switched odorant responses when the valence of 32 the stimuli was reversed. Finally, licks diverged more slowly and mice did not become 33 proficient when MLIs were inhibited by chemogenetic intervention. Our findings support 34 a role for MLIs in learning valence in the cerebellum. 35 109response returned to basal levels shortly after the odorant was applied. Fig. 2f shows the 110 mean F/F (+CI) for the last second of odorant application for trials falling within different 111 percent correct performance ranges ( Supplementary Fig. 4). A generalized linear model 112 (GLM) analysis finds a statistically significant interaction between performance and the 113 odorant (S+ vs. S-) (p<0.001, 44 d.f., 4 mice). Therefore, as the animal learns the change 114 in Ca 2+ in MLIs diverges between S+ and S-odorants, suggesting a possible role for MLIs 115 in associative learning. 116 117Finally, we calculated the dimensionality of F/F responses for the MLI cell ensemble.
118The plots in Fig. 1c indicate that odorant responses are fairly uniform for the ROIs in the 119 FOV. In order to provide an estimate of the number of independent components comprising 120 the F/F responsiveness of the ensemble we calculated a quantitative measure of 121 dimensionality 27 (see Methods). Fig. 2g shows the dimensionality for four sessions with a 122 total number of ROIs ranging from 103 to 136. The dimensionality ranged from 2 to 6 and 123 was higher for the pre-odorant period compared to the odorant or reinforcement periods 124 (GLM p<0.01, 18 d.f.). This indicates that Ca 2+ ensemble activity becomes more uniform 125 after addition of the odorant. 126 127 The identity of the stimulus can be decoded from MLI activity after mice learn to 128 differentiate odorants. The finding that the MLIs developed divergent responses for S+ 129 vs. S-trials following learning raised the question whether ensemble neural activity 130 encodes for the identity of the stimulus when the animal becomes proficient differentiating 131 the odorants. The learning curve for a session where the mouse achieves >80% ~60 trials 132 is shown in Fig. 3a. Fig. 3b shows for this session plots of the first two principal 133 components for a principal component analysis (PCA) of F/F values for all MLIs for a 134 subset of trials with percent correct <65% (top, naï ve) or >80% (bottom, proficient) 135 calcul...