1. Volume 97, Issue 4, Pages 898-910.e6 (February 2018)
Prefrontal-Periaqueductal Gray-Projecting Neurons Mediate Context Fear Discrimination 
Robert R. Rozeske, Daniel Jercog, Nikolaos Karalis, Fabrice Chaudun, Suzana Khoder, Delphine Girard, Nânci Winke, Cyril Herry 
Neuron 
Volume 97, Issue 4, Pages e6 (February 2018)
DOI: /j.neuron
Copyright © 2018 Elsevier Inc. Terms and Conditions

2. Figure 1
Modifying Sensory Elements Controls Contextual Fear Expression
(A) Experimental protocol.
(B) Schematic depiction of the paradigm developed to study contextual fear discrimination. Auditory, visual, and olfactory sensory elements were manipulated to produce contexts that more, or less, resembled the conditioned context. The conditioned context A contained tonic white noise, house lights, and vaporized lime odor. Control mice were tested for 12 min in context A.
(C) Configuration ABCA’ consisted of 3 min sequential exposures to four contexts while the mouse remained in the same testing chamber. Context B was identical to context A except the lime odor was aspirated from the testing chamber. Context C was the most distinct from context A, specifically the audible white noise was removed, the house lights were dimmed, and the lime odor was aspirated. Context A’ was identical to context A.
(D) Configuration CAC’A’ was used to assess timing effects of context C presentation. Context C’ was identical to C and all other contexts were as described previously.
(E and F) Left: dynamics of freezing behavior 24 hr (E) and 48 hr (F) after conditioning for mice tested in configuration ABCA’ and control mice tested in configuration A (bin size = 30 s; day 3: ABCA’ group, n = 22; A group, n = 14; day 4: ABCA’ group, n = 13; A group, n = 9). Right: corresponding average freezing values during context exposure 24 hr (E) and 48 hr (F) after conditioning reveal robust freezing behavior in contexts A, B, and A’ and a significant reduction in freezing behavior in mice exposed to context C compared to controls (day 3: two-way repeated-measures ANOVA [RM ANOVA], F(3, 102) = 13.09, p < ; day 4: two-way RM ANOVA, F(3, 60) = 11.96, p < , Bonferroni multiple comparison post hoc tests). Testing in configuration ABCA’ on day 4 was to assess novelty effects of context C.
(G) Left: dynamics of freezing behavior 5 days after conditioning for mice tested in configuration CAC’A’ and control mice tested in configuration A (bin size = 30 s; day 7: CAC’A’ group, n = 13; A group, n = 9). Right: corresponding average freezing values during context exposure are shown. Context C produced fear discrimination independently of timing, and presentation of context A’ produced a robust renewal of freezing (two-way RM ANOVA, F(3, 60) = 29.58, p < , Bonferroni multiple comparison post hoc tests). Data are expressed as mean ± SEM (∗p < 0.05 and ∗∗∗p < 0.001).
Neuron  , e6DOI: ( /j.neuron )
Copyright © 2018 Elsevier Inc. Terms and Conditions

3. Figure 2
Involvement of the dmPFC during Contextual Fear Discrimination and Context Transitions
(A) A schematic representing single-unit recordings in the dmPFC during context fear behavior.
(B) A heatmap of neuronal activity during exposure to configuration ABCA’ (n = 285 units; bin size = 5 s).
(C) A schematic of the method used for population analysis. Left: the heterogeneous activity of n recorded units during behavior is normalized by Z score, and the instantaneous population vector (iPV; STAR Methods) is calculated for units 1 through n from time 0 to t. Right: each unit represents one dimension in an n-dimensional population space with the color-coded iPV.
(D) Principal component analysis (PCA)-based two-dimensional (2D) projection of the iPV for dmPFC units recorded during exposure to configuration ABCA’ before conditioning (n = 60 units).
(E) PCA-based 2D projection of the iPV for dmPFC units recorded during contexts ABCA’ after conditioning (n = 285 units).
(F) Quantification of Mahalanobis distances in the full dimensional space among A/B, B/C, and C/A’ contexts for the pre- and post-conditioning groups using the resampling procedure (STAR Methods). Exposure to contexts ABCA’ following conditioning differentially engaged the dmPFC population as compared to pre-conditioning exposure.
(G) Schematic of the method used to analyze context transitions. Top: the Euclidean distance (d) between the iPV at a given time (t) and the centroids of two neuronal population clusters, contexts C1 and C2, is calculated. Bottom left: distances between iPV and C1 or C2 centroids are then plotted as a function of time. Bottom right: the difference of the iPV distance between C1 and C2 centroids is finally plotted as a function of time.
(H) Raster plot of individual freezing epochs for all mice during post-conditioning ABCA’. Superimposed is the probability of freezing (n = 22 mice).
(I) Top: transition between contexts A and B was not associated with a significant change in the probability of freezing. Bottom: transition between contexts A and B was not associated with a significant change in dmPFC network activity. Dotted black horizontal lines in top and bottom panels indicate significance thresholds (mean ± 5 SD).
(J) Transition between contexts B and C was associated with a significant reduction in freezing probability (top) and a significant alteration in dmPFC population activity (bottom). Dotted red vertical line indicates time of significant change in freezing probability, and dotted blue vertical line indicates time of significant change in dmPFC population activity.
(K) Transition between contexts C and A’ was associated with a significant increase in freezing probability (top) and a significant change in dmPFC population activity (bottom). Data are expressed as mean ± 95% confidence interval, so an absence of overlap indicates significance (#p < 0.05 and ##p < 0.01 based on the confidence interval).
Neuron  , e6DOI: ( /j.neuron )
Copyright © 2018 Elsevier Inc. Terms and Conditions

4. Figure 3
A Subpopulation of dmPFC PNs Is Significantly Activated during Contextual Fear Discrimination
(A) Left: firing rate dynamics of dmPFC PNs (bin size = 5 s) during contexts ABCA’ (n = 212 units) and control group context A for 12 min (n = 116 units), normalized to 0–3 min. Right: corresponding average dmPFC PN firing rate 24 hr after conditioning reveals a significant increase in firing activity in context C compared to the control group (Wilcoxon rank-sum tests, p < 0.001, Bonferroni multiple comparison post hoc test).
(B) Bootstrap-resampling method used to identify neurons significantly active in a context. Among the neurons selected, a larger fraction than expected by chance was highly active in context C (two-tailed binomial test, p < , Bonferroni multiple comparison post hoc test), whereas a smaller fraction than expected by chance was highly active in context A (two-tailed binomial test, p < 0.05, Bonferroni multiple comparison post hoc test).
(C) The firing rates of PNs selected by the bootstrap method as significantly active in context C were elevated in C compared to contexts B and A’ (Friedman’s rank test, p < , Dunn’s multiple comparison post hoc test).
(D) PNs selected by the bootstrap method as significantly active in context C (n = 112) with activity restricted to non-freezing periods in the 2D PCA space. Inset: analysis in the full dimensional space revealed a 2-fold increase in Mahalanobis distances for context C in comparison to contexts A/B.
(E) Firing rate of PNs restricted to non-freezing periods during ABCA’. PNs selected by the bootstrap method as active during context C (C active) displayed a selective and significantly higher firing rate in context C as compared to contexts A, B, and A’ (Friedman’s rank test, p < , Dunn’s multiple comparison post hoc test). Firing rate for C active PNs was also significantly elevated compared to units not selected by the bootstrap method as active in context C (non-active) for all contexts (Wilcoxon rank-sum tests, p < , Bonferroni multiple comparison post hoc test).
(F) Top: representative firing rate of a PN selected by the bootstrap method as highly active in context C (bin size = 5 s). Gray bars represent freezing epochs and at the top is a raster plot of firing rate. Middle: heatmap of the normalized firing activity (Z score) of all PNs selected by the bootstrap method as highly active in context C is shown (bin size = 5 s). Bottom: mean firing activity of all PNs that were selected by the bootstrap method as significantly active during context C is shown (bin size = 5 s). Data are expressed as mean ± SEM (∗p < 0.05 and ∗∗∗p < 0.001).
Neuron  , e6DOI: ( /j.neuron )
Copyright © 2018 Elsevier Inc. Terms and Conditions

5. Figure 4
Identification of the dmPFC-l/vlPAG Pathway Activated during Contextual Fear Discrimination
(A) A schematic of electric antidromic stimulation method used to identify dmPFC-l/vlPAG-projecting neurons.
(B) Top and middle: antidromic spikes recorded in the dmPFC following stimulation in the l/vlPAG (10 superimposed traces) demonstrating fixed latency and high fidelity. S, stimulation artifact; A, antidromic spike. The red trace illustrates a collision between a spontaneously occurring and an antidromic spike. Bottom: antidromic spikes were recorded in the dmPFC following high-frequency stimulation (250 Hz) in the l/vlPAG (10 superimposed traces).
(C) Distribution of spike latencies for antidromically identified dmPFC-l/vlPAG units. Inset: pie chart illustrates the proportion of antidromically identified dmPFC-l/vlPAG units that were context C active (C active) and not active in context C (non-active).
(D) Left: schematic of electrolytic lesion sites of dmPFC recordings in mice (n = 5) with antidromically identified neurons. Right: lesion sites of antidromically stimulated sites in the l/vlPAG are shown. Corresponding recording sites in the dmPFC and lesion sites in the l/vlPAG are color coded. Numbered labels indicate distance (mm) relative to bregma.
(E) A schematic of the intersectional infection strategy for photo-identification.
(F) Expression of GFP in dmPFC-l/vlPAG-projecting neurons using the intersectional infection strategy.
(G) Top: representative peristimulus time histogram (PSTH) of light-evoked inhibition of a dmPFC-l/vlPAG unit. Bottom: averaged PSTH (Z score) of dmPFC-l/vlPAG units inhibited during 200 ms of yellow light stimulation is shown (n = 7 neurons, bin size = 50 ms).
(H) Termination sites of optic fiber tips for photo-identification of dmPFC neuron experiments. All optrodes (n = 5) were implanted in the left hemisphere. Numbered labels indicate distance (mm) relative to bregma.
(I) Average freezing behavior during configuration ABCA’ in mice used for electric and photo-identification. Mice showed contextual fear discrimination in context C compared to contexts A, B, and A’ (n = 10 mice; one-way RM ANOVA, F(3, 27) = 8.820, p < 0.001, Bonferroni multiple comparison post hoc test).
(J) Representative dmPFC photo-identified PN that was active during context C (bin size = 5 s). At the top is a raster of the firing rate.
(K) Among the photo- and antidromically identified dmPFC-l/vlPAG-projecting neurons, the majority (n = 17/23) displayed significantly elevated firing during context C (Wilcoxon rank-sum tests, p < 0.001, Bonferroni multiple comparison post hoc test).
(L) A schematic of electric antidromic stimulation method used to identify dmPFC-BLA-projecting neurons.
(M) Left: location of recording sites in the dmPFC for BLA antidromic stimulation (n = 5 mice). Right: lesions of antidromically stimulated sites in the BLA are shown. Numbered labels indicate distance from bregma (mm). Recording and corresponding antidromic stimulation sites are color coded.
(N) Freezing behavior during configuration ABCA’ of mice submitted to the antidromic identification of dmPFC-BLA-projecting neurons revealed fear discrimination during context C (one-way ANOVA, F(3, 12) = 5.886, p < 0.05, Bonferroni multiple comparison post hoc test difference between context C versus A and A’).
(O) The firing rates of identified dmPFC-BLA-projecting neurons (n = 9 cells) were similar in contexts A, B, C, and A’ (one-way ANOVA, F(3, 24) = 0.375, p > 0.05). Data are expressed as mean ± SEM (∗p < 0.05 and ∗∗p < 0.01).
Neuron  , e6DOI: ( /j.neuron )
Copyright © 2018 Elsevier Inc. Terms and Conditions

6. Figure 5
Photo-Stimulation of dmPFC Axon Terminals in l/vlPAG Modulates Unit Activity
(A) Left: placement of fiber optic tips in mice expressing ChR2 or GFP controls in the dmPFC-l/vlPAG pathway. Right: placement of fiber optic tips in mice expressing ArchT or GFP controls in the dmPFC-l/vlPAG pathway is shown.
(B) A subset of mice expressing ArchT or ChR2 in the dmPFC-l/vlPAG pathway was implanted with optrodes, and l/vlPAG single units were recorded. Units recorded were plotted based on the following extracellular electrophysiological properties: firing frequency, trough-peak latency, and the area under the peak of the spike waveform.
(C) No significant differences were observed in the firing frequency of l/vlPAG units recorded in ArchT- or ChR2-expressing mice (Wilcoxon rank-sum test, p > 0.05).
(D) Comparison of area under the peak revealed no difference between ChR2- and ArchT-expressing mice (unpaired t test, t(26) = 1.17, p > 0.05).
(E) Trough-peak latency was compared between ChR2- and ArchT-expressing mice and no differences were found (unpaired t test, t(26) = , p > 0.05).
(F) Left: over half (8/14, 57%) of l/vlPAG units were excited during pre-synaptic 473-nm photo-stimulation (bin size = 5 s, baseline to context A). Right: blue light photo-stimulation significantly increased l/vlPAG unit activity in a subset of light-activated cells compared to non-light-activated cells (Wilcoxon rank-sum test, p < 0.01, Bonferroni multiple comparison post hoc test).
(G) Left: mice implanted with optrodes in the l/vlPAG and expressing ArchT in the dmPFC-l/vlPAG pathway received 593 nm photo-stimulation. Over half of l/vlPAG units (9/14, 64%) were inhibited following yellow light (bin size = 5 s, baseline to context A). Right: yellow light stimulation to pre-synaptic terminals significantly reduced l/vlPAG unit activity in a subset of light-inhibited cells compared to non-light-inhibited cells (Wilcoxon rank-sum test, p < 0.01, Bonferroni multiple comparison post hoc test). Box-and-whisker plots represent the median, interquartile range, and extreme values; other data are expressed as mean ± SEM (∗∗p < 0.01).
Neuron  , e6DOI: ( /j.neuron )
Copyright © 2018 Elsevier Inc. Terms and Conditions

7. Figure 6
Optogenetic Manipulation of dmPFC-l/vlPAG Neurons Controls Contextual Fear Expression
(A) Top: schematic describing infection and optogenetic strategy. Middle and bottom: images of GFP labeling in the dmPFC of a CaMKIIα-Cre mouse expressing ArchT (middle) or GFP (bottom).
(B) Experimental protocol.
(C) Freezing behavior during yellow light activation before fear conditioning in control (GFP, n = 5) and ArchT-expressing (n = 9) mice (two-way RM ANOVA, F(2, 24) = 3.00, p > 0.05).
(D) Freezing behavior of the same mice following yellow light activation during context C 24 hr following conditioning (two-way RM ANOVA for context, F(3, 36) = 25.61, p < , Bonferroni multiple comparison post hoc test).
(E) Freezing behavior during blue light activation before fear conditioning in GFP control (n = 7) and ChR2-expressing (n = 7) mice (Wilcoxon rank-sum tests, p > 0.05).
(F) Freezing behavior of the same mice following blue light activation during context B 24 hr following conditioning (two-way RM ANOVA, F(3, 36) = 19.47, p < , Bonferroni multiple comparison post hoc test). Data are expressed as mean + SEM (∗p < 0.05 and ∗∗∗p < 0.001).
Neuron  , e6DOI: ( /j.neuron )
Copyright © 2018 Elsevier Inc. Terms and Conditions