The generation of cfos-transactivator (tTA) × tetO-tau lacZ double transgenic mice has been described previously (Reijmers et al., 2007). The double transgenic mice were backcrossed to C57BL/6J mice and maintained as heterozygotes. All mice were bred in social groups (2–5 mice per cage), provided with food and water ad libitum, and fed with a doxycycline (Dox) diet (50 mg/kg chow) from pregnancy. After behavioral treatment (fear conditioning or home cage (HC)) during the off-Dox time window, they were switched to a higher concentration of Dox diet (1 g/kg chow) to quickly suppress further expression of tau-lacZ induced by unrelated behavioral stimuli. Mice that were 11–24 weeks old at the onset of experiments were used. All procedures were approved and conducted in accordance with guidelines of Kyoto University and Osaka University on the care and use of laboratory animals.

All behavioral experiments were conducted during the light period of the light/dark cycle. At the start of the experiments, the mice were individually housed and subjected to handling sessions for 3 days. For contextual fear conditioning, the mice were placed in a novel rectangular chamber (context A: 25 cm × 33 cm × 28 cm) with white plastic side walls, a transparent plastic top, and front and rear walls with a stainless steel grid floor under a 100 lux light in a sound-proof room (O’ Hara and Co., Ltd). Three foot-shocks (2 s, 0.75 mA) were administered at time points of 208, 298 and 388 s after the animals were placed in the chamber. They were returned to their HC 30 s after the final shock. For the memory retrieval test, the mice were placed in the conditioned context (context A) or in a novel context (context B) for 180 s. For context B, the mice were placed in a chamber consisting of a checkered cylindrical wall and a transparent plastic top with a flat plastic floor under a 50 lux light. For context C, mice were placed in a novel triangular chamber made of opaque white Plexiglas (33 cm × 29 cm × 40 cm) with a flat plastic floor with a lemon oil scent under a 20 lux red light in a different sound-proof room. Freezing was scored and analyzed automatically using a CCD camera-based system (TimeFZ4, O’ Hara and Co., Ltd). Images were recorded from the top of each chamber using a camera at two frames per second. For the analysis of images, the gap area (pixel) between the contour of the mouse in one frame and that in the next frame was identified. Freezing behavior was considered to occur if the gap area was under 20 pixels for two continuous seconds. Freezing scores were expressed as the ratio of the freezing period to the experimental period. The discrimination index was calculated using the following formula: (freezing % in context A)/(average of freezing % in context B).

Brains were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) at 4°C overnight and sectioned at a thickness of 40 μm using a vibratome (Leica). Free-floating slices were permeabilized with 0.15% Triton X-100 in 5% bovine serum albumin (BSA)/PBS at room temperature for 30 min and then rinsed with PBS. Permeabilized slices were incubated with primary antibodies (rabbit anti-ZIF antibody, Santa Cruz Biotechnology; mouse anti-β-Galactosidase antibody, Promega) at 4°C overnight. Next, the slices were rinsed with PBS three times for 10 min and incubated with secondary antibodies (goat anti-rabbit Alexa Fluor 594, Molecular Probes; goat anti-mouse Alexa Fluor 488, Molecular Probes) at 4°C overnight. The slices were then rinsed with PBS for 10 min and subsequently incubated with 4′,6-diamidino-2-phenylindole (DAPI, Molecular Probes) at room temperature for 5 min. Next, the slices were rinsed with PBS three times for 10 min and mounted on a glass slide with Prolong Gold antifade reagent (Molecular Probes). For the quantification of immunoreactive cells, fluorescent images were acquired using a FV1000 confocal laser scanning microscope (Olympus), using a 40× objective lens at a speed of 12.5 μs per pixel by sequential illumination with UV (405 nm), Argon (488 nm) and HeNe (559 nm) lasers.

Three to four images were analyzed using the ImageJ software (NIH) for each animal in each region. Structures of ROI were anatomically defined according to The Mouse Brain in Stereotaxic Coordinates (Franklin and Paxinos, 2008; see Figure 3E). Along the anteroposterior axis, sections between level of −1.70 mm and −2.06 mm from the bregma were selected for the analysis of the dorsal hippocampus and the somatosensory cortex (SS). Along the mediolateral axis of the dorsal hippocampal CA1, three fixed positions comprising proximal and distal part were equally selected for imaging. As for the dorsal dentate gyrus (DG), four fixed positions comprising upper and lower blade were equally selected for imaging. As for the SS, three fixed positions comprising all layers were equally selected for imaging. Acquired 16-bit images were converted to 8-bit images, then threshold level was set to a range of 40–150 for the tau-lacZ immunoreactive images. As for ZIF immunoreactive images, threshold was set to a range of 70–255 (DG), 80–150 (CA1) and 70–255 (SS). The number of cells expressing DAPI, tau-lacZ or ZIF was quantified manually using the thresholded images. All quantification was performed blind to experimental conditions. We calculated double and chance levels as follows: double = (total number of cells co-labeled with tau-lacZ and ZIF)/(total number of DAPI positive cells); chance level = (total number of tau-lacZ positive cells/total number of DAPI positive cells) × (total number of ZIF positive cells/total number of DAPI positive cells) × 100. The double-to-chance ratio was calculated as double/chance level.

All data were statistically analyzed using Prism 5 (GraphPad software) and are presented as the mean ± SEM.

During contextual fear conditioning, animals learn to associate a neutral context with an aversive event, such as foot shocks. Contextual fear memory is often generalized over the passage of time (Houston et al., 1999; Biedenkapp and Rudy, 2007; Wiltgen and Silva, 2007). To corroborate this phenomenon in our experimental condition, mice were subjected to electrical foot shocks in a training chamber (context A: Figures 1A,B) to elicit a long-term contextual fear memory. Then, they were placed in a conditioned chamber (context A) or a novel chamber (context B: Figures 1A,B) at 1 or 9 days after conditioning, and their freezing scores were measured as an index of contextual memory retrieval (Figure 1C). Naive cohorts of animals were used for each test to avoid any experience-related influences. The freezing scores of the retrieval test were significantly higher in context A compared to context B at 1 day after training (two-way ANOVA: main effect of context, F_(1,46) = 17.450, p < 0.001; Tukey’s post hoc test: p = 0.002; Figure 1C). However, at 9 days, exposure to both context A and context B evoked robust freezing, and there was no significant difference between the contexts (two-way ANOVA followed by Tukey’s post hoc test: p = 0.204; Figure 1C). In addition, the analysis of discrimination index showed significant differences among the time points (unpaired t-test: p < 0.001; Figure 1D). These results suggest that the contextual fear memories in our experimental condition were initially specific to the context, but became less specific within 9 days.

The observed behavioral changes during memory retrieval tests over 9 days raised a possibility that different populations of cells might be activated during retrieval 1 day and 9 days after conditioning. To investigate this possibility, we visualized the specific neuronal ensembles activated in response to fear learning and retrieval in a single mouse brain. We used a tetracycline-regulated tTA transgenic system, in which neuronal activity induces activation of the c-fos promoter (Reijmers et al., 2007; Matsuo et al., 2008). In the absence of Dox, the tTA drives tetO promoter-linked tau-lacZ expression in those neurons selectively activated by behaviorally relevant events (Figure 2A). It also elicits the expression of the bi-directional tetO promoter-linked mutant tTA (tTA^H100Y), which can trigger the tetO promoter even in the presence of Dox (Reijmers et al., 2007). Therefore, once the particular neurons are activated during a given time window in the absence of Dox, they can be labeled with tau-lacZ persistently due to the positive feedback loop system created by tTA^H100Y (Figures 2A,B; Reijmers et al., 2007).

To validate the induced expression of tau-lacZ in response to fear-conditioned (FC) learning, transgenic mice were trained with fear conditioning in context A after removing Dox for 4 days (Figure 2B). Soon after training, mice were fed with Dox diet to suppress further induction of tau-lacZ in neurons that were unrelated to the learning (Reijmers et al., 2007). A control group received the same Dox treatment, but the mice were kept in their HC. Immunohistochemical analysis revealed a sparse expression of tau-lacZ in the brain (Figures 2C,D), and the proportion of cells labeled with tau-lacZ was significantly larger in the FC mice 1 day after learning compared with the control HC group in the DG, the CA1 and CA3 regions of the dorsal hippocampus, the SS, and the prelimbic cortex (PrL; DG: p = 0.003; CA1: p < 0.001; CA3: p = 0.014; SS: p < 0.001; PrL: p < 0.001, unpaired t-test; Figure 2E).

When the trained animals are exposed to a conditioned chamber (context A) or a novel chamber (context B) after training, the neurons activated by memory retrieval should express endogenous immediate early genes (IEGs) such as zif268 (ZIF; Saffen et al., 1988; Cole et al., 1989) 1–2 h later (Figures 2B,C,F). Thus, the previously activated neurons at the time of learning can be identified by long-lasting tau-lacZ expression, while neurons activated by retrieval would be labeled with ZIF within the same brain slice (Figures 2B,C). ZIF was used as an alternative neuronal activity marker due to the difficulty of discriminating endogenous c-Fos and transgenic c-Fos-EGFP (Reijmers et al., 2007; Matsuo et al., 2008) when we performed c-Fos immunohistochemistry. In this system, double-labeled cells should represent cells reactivated during the memory retrieval test.

We analyzed the cells reactivated during the retrieval test 1 day after contextual fear conditioning, when animals did not show generalized fear memory. The proportion of tau-lacZ positive and ZIF positive cells was not significantly different between context A and context B in all regions analyzed (tau-lacZ; DG: p = 0.543, CA1: p = 0.343, SS: p = 0.917. ZIF; DG: p = 0.873, CA1: p = 0.602, SS: p = 0.960, unpaired t-test; Figures 3B,C). Then, the proportion of double-labeled cells was quantified to analyze the ratio of doubles to chance as an index of reactivation. In the dorsal hippocampal DG, the proportion of reactivated granule cells was at the chance level when mice were re-exposed to conditioned context A 1 day after learning (p = 1.000, Wilcoxon signed rank test; Figure 3D, Table 1). Remarkably, we found that the proportion of reactivated DG cells was significantly below the chance level in the mice that were placed in a novel context (context B) for the retrieval test 1 day after conditioning (p = 0.001, Wilcoxon signed rank test; Figures 3A,D). These findings suggest that different neuronal populations in the DG were positively recruited for contextual representations when mice were exposed to a distinct context that they had not previously encountered.

In contrast to the DG cells, the proportion of reactivated dorsal CA1 pyramidal cells was not significantly different from the chance level when they were exposed to a distinct context B (p = 0.677, Wilcoxon signed rank test; Figure 3D). However, the proportion of reactivated CA1 pyramidal cells was significantly higher than the chance level when animals were returned to the same context 1 day after training (p = 0.041, Wilcoxon signed rank test; Figures 3A,D) though the reactivation indices were not significantly different between context A and context B (p = 0.443, Mann Whitney test). Several reports demonstrated that the same population of CA1 pyramidal cells activated during conditioning tended to be reactivated when the memory was retrieved in the trained context (Deng et al., 2013; Tayler et al., 2013).

Multimodal information, including sensory inputs, shapes contexts. We then examined the reactivation of neuronal ensembles in one of the primary sensory cortices, the SS. The proportion of reactivated cells in the cortex was significantly higher than the chance level in the mice re-exposed to context A at 1 day after learning (p = 0.001, Wilcoxon signed rank test; Figures 3A,D), whereas the percentage of reactivated cells during the retrieval test in context B was at the chance level (p = 0.910, Wilcoxon signed rank test; Figure 3D). Collectively, these results suggest that the hippocampal DG as well as the SS, were likely to distinguish the two different contexts at the ensemble activity level at the same time that the animals were able to discriminate the contextual differences at the behavioral level.

Next, we examined the proportion of reactivated neuronal ensembles during the retrieval test 9 days after conditioning, when mice exhibited generalized fear memory. A substantial number of tau-lacZ labeled cells was detectable 9 days after training in the dorsal part of the hippocampal DG, the CA1 region, and the SS (Figures 4A,B), and it was comparable between context A and context B (DG: p = 0.758, CA1: p = 0.620, SS: p = 0.599, unpaired t-test; Figure 4B). Similarly, the proportion of ZIF positive cells was not significantly different between context A and context B (DG: p = 0.531, CA1: p = 0.495, SS: p = 0.103, unpaired t-test; Figure 4C). In the DG, the proportion of reactivated cells was not different from the chance level for both retrieval contexts (context A: p = 0.967; context B: p = 0.487, Wilcoxon signed rank test; Figures 4A,D, Table 1). Similarly, preferential reactivation of hippocampal CA1 neuronal ensembles was not observed in either the conditioned context A or the unconditioned context B (context A: p = 0.077; context B: p = 0.301, Wilcoxon signed rank test; Figures 4A,D). Moreover, the reactivation indices in the DG and CA1 were not significantly different between context A and context B (DG: p = 0.977; CA1: p = 0.298, Mann Whitney test; Figure 4D). These findings suggest that neuronal ensembles in the hippocampus were not specifically reactivated to the context during the retrieval test at this time point.

Unlike the hippocampus, the proportion of reactivated cells in the SS was significantly higher than the chance level in the mice re-exposed to context A at 9 days after learning (p = 0.009, Wilcoxon signed rank test; Figures 4A,D), whereas the percentage of reactivated cells during the retrieval test in context B was at the chance level (p = 0.198, Wilcoxon signed rank test; Figure 4D). These results indicate that a similar set of ensembles in the SS was preferentially reactivated when the mice were returned to the same environment, even though the animals appeared to be unable to distinguish the contexts at the level of behavioral output.

Although we found a loss of unique reactivation pattern of DG neurons during retrieval 9 days after conditioning, the reduced number of tau-lacZ labeled cells at this time point raised a possibility that the result was an artifact of biased reduction of the labeling. Thus, we further analyzed the reactivation pattern of neuronal ensembles in a different novel context (context C: Figure 5A) where animals showed no significant freezing even 9 days after fear conditioning (one-way ANOVA: F_(2,32) = 12.67, p < 0.001; Tukey’s post hoc test, A vs. C: p < 0.001, B vs. C: p = 0.008; Figure 5B). In this condition, the proportion of reactivated DG cells was significantly below the chance level in the mice that were placed in context C (p = 0.031, Wilcoxon signed rank test), while the proportions in the hippocampal CA1 area and the SS were both at the chance level (CA1: p = 0.625, SS: p = 0.117, Wilcoxon signed rank test; Figures 5C,D). These results are not consistent with the reactivation pattern observed in context B at the same time point (9 days after conditioning), suggesting that the loss of DG ensemble reactivation pattern was not due to an artifact of reduced tau-lacZ labeling over days but rather an alteration of DG ensemble activities over time.