Extended Data Figure 7. Satiation by consumption of Ensure can be labile; InsCtx neuronal responses to the three visual cues during Sated-2 + AgRP and Sated-2; saline injections do not mimic hunger in InsCtx; comparison of InsCtx responses in hungry mice across two daily sessions; time course of decoding accuracy and the necessity of neuronal identity for decoding accuracy.

A, Example licking raster plots from a single session, for food cue trials presented during hunger and satiety. Intermingled aversive and neutral cue presentations are omitted for clarity. Note cue-evoked licking during hunger but not during satiety. B, Examples licking raster plots from two mice, for a session in which each mouse was not fully sated and thus partly re-engaged in the task for a substantial number of trials after initial satiation. Note that in such cases we stopped the experiment, and re-sated the mice with delivery of additional Ensure until they were fully sated (operationally defined as lack of voluntary licking for Ensure, see Methods). C, Heatmap of the average response to the three visual cues during Sated-2 + AgRP and Sated-2 for each neuron (one row per neuron) that was significantly responsive to at least one cue during Sated-2 + AgRP. Vertical dashed lines: visual cue onset. Horizontal dashed lines: grouping of neurons by the cue that evoked the strongest response. D, Saline injections did not mimic hunger in InsCtx. Average response magnitude of InsCtx neurons across hunger, satiety, and saline injections (n = 109 neurons from two mice, different mice from those in c). *P = 0.03 (Hungry); NS, not significant (P ≥ 0.5, Sated-1, Sated-2, Sated-2 + saline injections); Kruskal–Wallis test. Pairwise comparisons across states (Hungry-1 versus Sated-1): food cue, P = 2.4 × 10⁻⁵; aversive cue, P = 0.02; neutral cue, P = 1.7 × 10⁻⁵. Pairwise comparisons across states (Sated-2 versus Sated-2 + saline injections): P ≥ 0.15 for all three cues; **P < 0.02; NS, not significant; Mann–Whitney U-test. Mean ± s.e.m. Note the population response bias to the food cue in Hungry-1, but not in Sated-2 + saline injections. Furthermore, while there was a food cue bias in the fraction of responsive neurons per cue during hunger (Hungry-1; food cue, 0.16; aversive cue, 0.10; neutral cue, 0.12), saline injections in sated mice did not induce a similar food cue bias (‘Sated-2 + saline inj.’; food cue, 0.05; aversive cue, 0.07; neutral cue, 0.05; n = 109 neurons from two mice). E, Heatmap of all food- cue-responsive neurons during two consecutive imaging sessions, both performed while the mice (n = 4) were hungry. Note that some neurons were similarly responsive on both days, while others were not. F, Average decoding accuracy. The decoder was trained on data from entire populations of neurons recorded during the first ‘Hungry’ session, and tested on other ‘Hungry’ data from the same neurons on the same day, or from the same neurons recorded on the following day (‘Hungry-next-day’ data; n = 4 mice). Note that decoding accuracy was similar. G, Average time course of accuracy of population decoding of whether a food cue was presented versus other cues (see text for details). *All time points that were significantly different from chance (33%) for ‘Sated-2 + AgRP stim.’ (P ≤ 0.03, paired t-test, Holm–Bonferroni correction for multiple comparisons, n = 4 mice). Decoding was performed on single-trial responses of simultaneously imaged ensembles (90–98 neurons per mouse, n = 4 mice). H, a, Overall food cue decoding accuracy after shuffling of neuronal identity. Data are averages of 100 shuffles for each mouse separately (n = 4 mice). Note that shuffling neuronal identity decreased overall decoding accuracy to chance levels. All values are mean ± s.e.m. across mice. H, b, Average time course of accuracy of decoding the food cue versus other cues, after shuffling of neuronal identity (see text for details). Data are averages of 100 shuffles, performed separately for each mouse (n = 4 mice). Note that shuffling neuronal identity decreased decoding accuracy to chance levels, and that decoding accuracy did not increase with time. Thus, decoding accuracy stemmed from the specific pattern of responses across the population and not from global biases in response strength or response type.