Volume 92, Issue 4, Pages 902-915 (November 2016) Distinct Roles of Parvalbumin- and Somatostatin-Expressing Interneurons in Working Memory  Dohoung Kim, Huijeong Jeong, Juhyeong Lee, Jeong-Wook Ghim, Eun Sil Her, Seung-Hee Lee, Min Whan Jung  Neuron  Volume 92, Issue 4, Pages 902-915 (November 2016) DOI: 10.1016/j.neuron.2016.09.023 Copyright © 2016 Elsevier Inc. Terms and Conditions

Figure 1 Working Memory Task (A) Figure 8-shaped maze. The maze had five doors (gray solid lines) and six photobeam sensors (green dotted lines). The water reward (8 μL) was provided at two reward sites (blue circles) in correct trials. (B) Delayed non-match-to-place task. Shown is an example sequence of the task. Sample phase, the animals were forced to visit a randomly determined target site (no reward provided); delay period, the animals waited for 3 s on the central track; and choice phase, the animals were rewarded by choosing the target different from the one they visited during the sample phase. (C) Learning curve. The animals’ performance (% correct choice) during the training, before surgery, is shown (mean ± SEM across animals; n = 17). The linear regression indicated significantly improving performance (slope = 1.9; t test and p = 4 × 10−17) over sessions. Neuron 2016 92, 902-915DOI: (10.1016/j.neuron.2016.09.023) Copyright © 2016 Elsevier Inc. Terms and Conditions

Figure 2 Optical Tagging and Unit Classification (A) Histological verification of ChR2 expression. Shown are coronal sections of a PV-Cre (left) and a SOM-Cre (right) mouse brain expressing ChR2-eYFP (green) stained with anti-PV (left) or anti-SOM (right) antibody (red) along with DAPI (blue). A high-magnified image of a PV- (left) or SOM-expressing (right) neuron is shown in the inset box. Secondary motor cortex, M2; anterior cingulate cortex, Cg; prelimbic cortex, PL. (B) The diagrams show the proportions of neurons expressing ChR2 and/or PV/SOM in the prelimbic cortex. (C) The extent of ChR2 expression in each animal is indicated in green, with dark color indicating overlapping areas across animals. The vertical bars denote locations of optical probes, and each red circle indicates the location of each tetrode in the last recording session. The diagrams are coronal section views of mouse brains (1.98, 1.94, and 1.78 mm anterior to bregma). Modified with permission from Elsevier (Franklin and Paxinos, 2007). (D) The recorded neurons were classified into type I (filled circles) and II (filled triangles) based on the mean firing rate, half-valley width, and peak-valley ratio. The open squares denote unclassified neurons. The purple symbols indicate light-activated neurons in PV-Cre mice, and the yellow and green symbols denote light-activated neurons in SOM-Cre mice. (E) Example neuronal responses to light stimulation. Top, raster plot. Each row is one trial and each dot is one spike. The blue bar denotes a period of light stimulation. Bottom, peri-stimulus time histogram (PSTH). The time 0 denotes the time of light stimulus (5-ms duration) onset. The averaged spike waveforms of spontaneous (black) and optically driven (blue) spikes recorded through four tetrode channels are shown (inset). The scale represents 250 μs and 50 μV. See also Figures S1 and S2. Neuron 2016 92, 902-915DOI: (10.1016/j.neuron.2016.09.023) Copyright © 2016 Elsevier Inc. Terms and Conditions

Figure 3 Delay-Period Activity of PV and SOM Neurons (A) Raster plots and spike density functions (Gaussian kernel with σ = 100 ms was applied to each spike) during four different behavioral stages (trial onset, sample target, delay, and reward; the animal’s approximate position at the onset of each stage is indicated by gray shading on top) are shown for example PV, ns-SOM, and ws-SOM neurons. The trials were divided into correct left choice (deep blue), incorrect left choice (light blue), correct right choice (deep red), and incorrect right choice (light red) trials and plotted separately. The time 0 denotes the onset of each behavioral stage (the time of breaking a photobeam sensor). (B–D) Target-dependent firing of PV and SOM interneurons during delay period. (B) Distributions of the index of target preference for the entire delay-period activity (3 s) of PV and SOM (ns-SOM and ws-SOM combined) neurons. (C) Distributions of the magnitude of target preference (absolute value of the index of target preference) for the entire delay-period activity of PV and SOM neurons. ∗significant difference between PV and SOM neurons (t test and p < 0.05). (D) Time courses of the magnitude of target preference are shown for PV and SOM neurons with a smaller analysis time window (0.5 s window advanced in 0.1 s time steps). The red squares indicate significant differences between PV and SOM neurons (t test and p < 0.05). (E–G) Target-dependent firing was separately examined for ns-SOM and ws-SOM interneurons. The same format as in (B)–(D). The bar graphs represent mean and SD (B and E), or mean and SEM (C and F), and shading represents SEM (A, D, and G). See also Figures S3–S6. Neuron 2016 92, 902-915DOI: (10.1016/j.neuron.2016.09.023) Copyright © 2016 Elsevier Inc. Terms and Conditions

Figure 4 Results of Decoding Analysis The identity of sample target was decoded based on delay-period activity. (A) The relationship between ensemble size and correct decoding of sample target based on neuronal ensemble activity during the entire delay period. The neural decoding was performed for different trial types (different combinations of the left/right sample and choice targets) and then averaged according to correct and error trials. Those neurons with mean delay-period activity <0.5 Hz were excluded from the analysis. The black color shows decoding results for correct trials, and the gray color shows decoding results for error trials. (B) The identity of sample target was decoded based on ensemble activity of untagged type II neurons (n = 334) in a 0.5 s time window that was advanced in 0.1 s steps. The vertical lines denote the onset and offset of the delay period. (C) Neural decoding of sample-target based on type II neuronal ensemble activity during different phases of the delay and post-delay periods. ∗significantly different from chance (50%) level (t test and p < 0.05). The bar graphs represent mean and SEM (C), and shading represents SEM (A and B). Neuron 2016 92, 902-915DOI: (10.1016/j.neuron.2016.09.023) Copyright © 2016 Elsevier Inc. Terms and Conditions

Figure 5 Responses of PV and SOM Neurons to Reward (A and B) Mean normalized firing rates in rewarded (correct, cyan) and unrewarded (error, gray) trials are shown for PV and SOM (ns-SOM and ws-SOM combined) neurons (A) and separately for ns-SOM and ws-SOM neurons (B). (C) Distributions of the index of reward-dependent firing for PV, ns-SOM, and ws-SOM neurons. (D) Experimental setting for the probabilistic Pavlovian conditioning task (top) and temporal structure of the task (bottom). The odor cues were paired with probabilistic reward delivery. (E–G) Reward-related responses of PV and SOM neurons during the Pavlovian conditioning task. The same format as in (A)–(C). The reward onset was aligned to the first licking response after delay offset. (H) Mean GLM coefficients for trial outcomes (colors, reward; gray, no reward). The bar graphs represent mean and SEM (C and G), and shading represents SEM (A, B, E, F, and H). ∗significantly different from zero (t test and p < 0.05). See also Figure S7. Neuron 2016 92, 902-915DOI: (10.1016/j.neuron.2016.09.023) Copyright © 2016 Elsevier Inc. Terms and Conditions

Figure 6 Other Task-Related Responses of PV and SOM Neurons (A) Mean normalized firing rates of PV, ns-SOM, and ws-SOM neurons during four different behavioral stages (trial onset, sample target, delay, and reward). The bottom plot shows mean running speed. The trials were divided into correct (blue) and incorrect (gray) trials. The time 0 denotes the onset of each behavioral stage. The bar graphs show mean firing rates before (gray) and after (black) each stage onset (1 s each) in correct trials. (B) Scatterplots showing mean firing rates before (abscissa) and during (ordinate) the delay period (3 s each). The numbers indicate slopes of the regression lines. (C) Frequency histograms showing correlations between movement speed and neuronal activity. The empty and filled bars denote neurons that were significantly and insignificantly correlated with movement speed, respectively. (D) The magnitude of target preference during trial onset, sample, and reward stages (first 2 s following each stage onset). ∗significant difference between PV and SOM neurons (t test and p < 0.05). (E) Examples of licking and neuronal responses following reward onset (the time of the first licking response since delay offset) in the Pavlovian conditioning task. The lick/spike raster and lick/spike density functions are shown for each example. Blue and gray represent rewarded and unrewarded trials, respectively. (F) Frequency histograms showing correlations between lick rate and neuronal activity. The empty and filled bars denote neurons with significant and insignificant correlations with licking response, respectively. (G) Mean GLM coefficients for lick onset and offset. The bar graphs represent mean and SEM (A and D), and the horizontal error bars (C and F) and shading (A, E, and G) represent SEM. Neuron 2016 92, 902-915DOI: (10.1016/j.neuron.2016.09.023) Copyright © 2016 Elsevier Inc. Terms and Conditions

Figure 7 Light Stimulation Effects of PV and SOM Neurons on Delay-Period Activity of Type I and II Neurons (A) Examples of simultaneously recorded untagged type I and type II neuronal activity with (deep colors) and without (light colors) light stimulation (correct trials only). The red and blue indicate delay-period activity associated with ipsilateral and contralateral target choices, respectively (top, raster plots and bottom, spike density functions [σ = 100 ms]). (B) Mean % suppression of delay-period activity by light stimulation for type I and type II neurons in PV-Cre and SOM-Cre mice. ∗significant difference between PV- and SOM-Cre mice (t test and p < 0.05). (C) Examples showing light stimulation-induced activation of optically tagged PV and SOM neuronal activity during the delay period. (D) Time courses of light stimulation effect on type II neurons. (E) Effect of light stimulation on delay-period activity in ipsilateral versus contralateral choice trials. Shown are values for the index of target-dependent response suppression (positive and negative values indicate preferential suppression of ipsilateral and contralateral choice-associated delay-period activity, respectively). The delay period was divided into two halves (early, 0–1.5 s and late, 1.5–3 s) and the effect of light stimulation was separately examined for each half. The bar graphs represent mean and SEM (B and E) and shading represents SEM (A, C, and D). See also Figure S8. Neuron 2016 92, 902-915DOI: (10.1016/j.neuron.2016.09.023) Copyright © 2016 Elsevier Inc. Terms and Conditions

Figure 8 Light Stimulation Effects of PV and SOM Neurons on Behavioral Performance (A) Performance during learning. The animals (nine PV-Cre and eight SOM-Cre mice) were trained in the working memory task (delay duration, 3 s) for 9 days. (B) Effects of light stimulation on behavioral performance of PV- and SOM-Cre mice. The behavioral performances with (+) and without (−) optical stimulation in the same block are shown together (thin gray lines) for each animal. The behavioral data were grouped together according to delay duration/stimulation intensity conditions (three sessions each). The squares and error bars represent mean and SEM, respectively. ∗significantly different (two-way ANOVA followed by Bonferroni’s post hoc test and p < 0.05). Neuron 2016 92, 902-915DOI: (10.1016/j.neuron.2016.09.023) Copyright © 2016 Elsevier Inc. Terms and Conditions