Fig. 1.

Prefrontal neurons display selectivity indicative of a hierarchical cue to rule transformation during attentional switching.

(a) Schematic of task design. (b) Mice were trained to associate four cues with two rules. These cues were presented in two blocks, each containing two cues. An animal had to achieve at least 70 correct trials in a block before moving on to the next block. For details, see Methods. (c) Example peri-stimulus time histogram (PSTH) and raster plot (number of trials vs. time) for a regular-spiking (RS) PFC neuron that is selective to a LP noise. The black bar above the raster marks the cueing period, and the red arrowhead indicates the transient increase in spiking reliability. (d) Transient responses tile the duration of the delay period. Each color is a different cue-selective neuron.(e-f) Same as (c-d) but for PFC cue-invariant cells. (g-h) Same as (c-d) but for PFC fast spiking (FS) cells. Unlike RS cells, these neurons have persistent changes in firing rate over the delay period. Representative examples in d, f and h drawn from n = 5 mice (independent samples). (i) Classification accuracy over time relative to cue onset for a decoder trained to predict either rule (top) or cue context decoding (bottom) for PFC RS and FS neuronal populations. The asterisks denote the time point at which classification accuracy is significant (i.e. p < 0.05, permutation test from n = 5 biologically-independent mice) above chance (50% classification accuracy). (j) Classification accuracy (within delay period) scales with the number of neurons. Similar to (i), the asterisk indicates the number of neurons at which classification accuracy is significantly above chance levels (p < 0.05, permutation test from n = 5 biologically-independent mice). (k) Top: Schematic of Poisson generalized linear model (GLM). Bottom: Model prediction (grey) of the PSTH (black) for one example PFC neuron. EV, explained variance. (l) Left: Heatmap showing coupling probability between the four cue-selective cell PFC cell types. Right: Box-whisker plots comparing the coupling strengths of inputs to PFC cue-selective neurons from cue-selective neurons preferring the same or different cues (light gray, p = 1.23 x 10⁻⁴) or cue-invariant neurons (dark gray, p = 0.18 x 10⁻⁴). Bonferroni-corrected Kruskal-Wallis ANOVA with post-hoc rank-sum test relative to neurons preferring the same cues (n = 5 biologically-independent mice). (m) Same as (l) but characterizing the inputs to cue-invariant PFC neurons from cue-selective neurons preferring the same or different rules (p = 1.89 x 10⁻⁶) or cue-invariant neurons (p = 1.42 x 10⁻⁶). Bonferroni-corrected Kruskal-Wallis ANOVA with post-hoc rank-sum test relative to neurons preferring the same rules (n = 5 biologically-independent mice). (n) Cartoon schematic of how cue-invariant neurons gain their selectivity by pooling from cue-selective neurons across both cueing contexts. Data is shown as mean +/− 95% confidence interval (shaded error bars). Box plots: median (line), box edges, 95% confidence interval, whiskers, range.