Figure 2. Category neurons.

(a-c) Characterization of category neurons in the hippocampus. (a) Example category neuron. The preferred category of this neuron was “animals”. Top: Peri-stimulus time histogram (PSTH, bin size = 200 ms, step size = 25 ms) during the first picture in each correct trial. Colored areas represent ± s.e.m. Bottom: raster plot with trials re-ordered into preferred and non-preferred categories. Stimulus and maintenance onset is at t = 0s (left and right, respectively). (b) Firing rates of category neurons from the hippocampus remained persistently active and were higher for preferred than unpreferred trials during the WM maintenance period. Top: Beta value extracted from the GLM for preferred/non-preferred regressor in units of “percent change to baseline” (−900 – −300 ms before first picture onset). Bottom: Distribution of FR differences between preferred (pref = 1) and non-preferred (pref = 0) trials across all hippocampal category neurons (each dot is a neuron, n = 104). FRs were baseline-normalized to represent percent change to baseline. (c) When the preferred category remained in WM, category neurons in the hippocampus had higher FRs in correct as compared to incorrect and in load 1 as compared to load 3 trials. (d-f) Characterization of category neurons in the amygdala. (d) Same as in (a) but for an example category neuron from the amygdala. (e) Firing rates of category neurons from the amygdala also remained persistently active and were higher for preferred than unpreferred trials during the WM maintenance period (n = 220). (f) Their FRs were higher in load 1 as compared to load 3 trials when their preferred category was maintained but there was no difference between correct and incorrect trials. In (b,c,e,f) we computed mixed-effects GLMs. Error bars represent standard errors of the coefficient. In (b,e) each dot is a neuron. ** p < 0.01, *** p < 0.001, ns = not significant.