GABAergic mechanisms regulated by miR-33 encode state-dependent fear

Some stress-related memories are state-dependent: they cannot be retrieved unless the brain is in the same state as during initial encoding. The authors show that hippocampal extrasynaptic GABAA receptors, regulated by miR-33, support state-dependent contextual fear conditioning by altering the processing of context memories within the extended hippocampal circuit. Fear-inducing memories can be state dependent, meaning that they can best be retrieved if the brain states at encoding and retrieval are similar. Restricted access to such memories can present a risk for psychiatric disorders and hamper their treatment. To better understand the mechanisms underlying state-dependent fear, we used a mouse model of contextual fear conditioning. We found that heightened activity of hippocampal extrasynaptic GABAA receptors, believed to impair fear and memory, actually enabled their state-dependent encoding and retrieval. This effect required protein kinase C-βII and was influenced by miR-33, a microRNA that regulates several GABA-related proteins. In the extended hippocampal circuit, extrasynaptic GABAA receptors promoted subcortical, but impaired cortical, activation during memory encoding of context fear. Moreover, suppression of retrosplenial cortical activity, which normally impairs retrieval, had an enhancing effect on the retrieval of state-dependent fear. These mechanisms can serve as treatment targets for managing access to state-dependent memories of stressful experiences.


Gaboxadol induces state-dependent fear
To test this hypothesis, we used the specific agonist gaboxadol to increase the activity of extrasynaptic GABA A receptors 11 . Gaboxadol injected intrahippocampally (i.h.) either before training ( Fig. 1a; n = 6 mice per group for the 0.5 µg per hippocampus dose and n = 7 mice per group for all other doses; F 5,35 = 26.798, P < 0.001) or before memory testing ( Fig. 1b; n = 7 mice per group for the 0 and 0.125 µg per hippocampus groups and 8 mice per group for 0.25 and 0.5 µg per hippocampus groups; F 3,26 = 20.594, P < 0.001) dose-dependently impaired contextual freezing, an index of learned fear 12 . These freezing impairments could be interpreted as impaired learning, memory retrieval or fear expression. However, when mice were injected with gaboxadol both before training and testing (G-G group), freezing was indistinguishable from that of vehicle controls (V-V group) and was significantly higher than that of the groups receiving gaboxadol only before training (G-V group) or before the test (V-G group; n = 7 mice per group for V-V, G-V and V-G and 8 mice per group for G-G; F 3,25 = 4.481, P < 0.05; Fig. 1c). This effect was replicated in a within-subject study with mice trained on vehicle or gaboxadol and then tested on or off drug on alternate tests ( Fig. 1d; n = 7 mice per group; within-subject effects were F 1 = 9.584, P < 0.01 for vehicle and F 1 = 9.581, P < 0.01 for gaboxadol). Thus, gaboxadol did not impair memory processes, but instead induced state-dependent contextual fear conditioning. At the lowest dose used to trigger state-dependent fear, gaboxadol did not affect locomotor activity or tone-dependent fear conditioning (Supplementary Fig. 1a,b), consistent with the preferential role of the hippocampus in contextual fear versus cue-dependent learning 13,14 . Muscarinic cholinergic receptors have also been implicated in state-dependent learning 6 , but antagonism of these receptors by scopolamine impaired memory without generating state-dependent effects (Supplementary Fig. 2). These findings suggest that state-dependent contextual fear is particularly sensitive to manipulations of GABAergic mechanisms. a r t I C l e S Gaboxadol mediates state-dependent fear via PKC bIII GABA A receptor function is closely linked to the activity and phosphorylation of protein kinase C (PKC) 15 . Of several PKC isoforms, gaboxadol infused before fear conditioning only enhanced the phosphorylation of PKCβII at S660 (Fig. 2a,b; F 2,11 = 3.813, P < 0.05). Similarly, PKCβII at S660 was upregulated when gaboxadol was injected both before conditioning and before the memory test or only before the memory test (Fig. 2c, n = 5 hippocampi per group; F 12 = 7.29, P < 0.01; full-length blots are presented in Supplementary Fig. 3a,b), suggesting that this isoform is involved in both encoding and retrieval of state-dependent context fear. Accordingly, inhibition of PKCβII before the memory test did not affect freezing in the vehicle control ( Fig. 2d; n = 8 mice per group for V-V and V-G, n = 9 mice per group for PKC βII-inhibitor; F 2,22 = 0.964, P = 0.399), but it did block retrieval in the gaboxadol-treated group, as shown by reduced freezing when inhibition of PKCβII preceded injection of gaboxadol ( Fig. 2d; n = 8 mice per group for V-V and V-G, n = 9 mice per group for PKC δII-inhibitor; F 2,22 = 7.375, P < 0.01). Inhibition of PKCδ had no effect on either group, indicating that PKCβII is specifically involved in gaboxadolmediated retrieval. These findings reveal an important role for PKCβII in retrieval of state-dependent fear and suggest that manipulations of this kinase could modify access to fear-inducing memories.
miR-33 regulates the effects of gaboxadol on state-dependent fear GABA A receptors regulate the expression of specific microRNAs 16,17 , which, in turn, may regulate GABA A receptor function 18,19 . We therefore investigated whether the actions of gaboxadol on tate-dependent fear involve microRNA-mediated mechanisms. Using microarrays, we first compared the microRNA profiles in the hippocampi of naive mice and mice exposed to fear conditioning, and identified 19 differentially expressed microRNAs that were associated    (c) Phosphorylation of PKCβII (S660) was also increased when gaboxadol was injected before both conditioning and the retrieval test (G-G group) or only before the retrieval test (V-G group) when compared with the V-V control group. Bottom, representative immunoblots.
(d) Effect of pre-test infusion of PKCβII (0.25 µg per hippocampus) and PKCδ (0.5 µg per hippocampus) inhibitors on freezing behavior in mice injected with vehicle or gaboxadol before conditioning and PKC inhibitors preceding vehicle or gaboxadol before the memory test. *P < 0.5, ***P < 0.001 when compared with corresponding control groups (one-way ANOVA or t test for c).
Error bars represent s.e.m. in all panels. npg a r t I C l e S with fear conditioning (Supplementary Fig. 4a). Five of these micro-RNAs are predicted to target four or more mRNAs encoding GABA A receptors ( Supplementary Fig. 4b), of which miR-33 was the only one whose level changed in response to gaboxadol (Fig. 3a). Contrary to the increase of miR-33 after fear conditioning off gaboxadol, the level of this miRNA significantly decreased 1 and 24 h after fear conditioning on gaboxadol (n = 4 hippocampi per group, F 2,9 = 6.189, P < 0.05). To determine whether changes of miR-33 levels are involved in state-dependent fear, we produced lentiviral vectors carrying miR-33 (LV-miR-33), which served to increase the level of miR-33 ( Supplementary Fig. 5a-c). To downregulate miR-33, we used miR-33 locked nucleic acid (LNA) inhibitor (miR-33-LNA; Supplementary  Fig. 5d-f) and compared its effects to those of scrambled miR LNA (miR-S-LNA) (Supplementary Fig. 6). In control mice injected with a lentivirus carrying scrambled miRNA (LV-SCR), gaboxadol (0.5 µg per hippocampus) induced state-dependent fear (Fig. 3b), as revealed by high freezing levels in the V-V and G-G groups, but not in the V-G and G-V groups (n = 8 mice per group; F 3,28 = 18.360, P < 0.001).
miR-33 regulates the levels of GABA-related proteins Overexpression of miR-33 significantly reduced mRNA expression of several predicted targets related to GABA A function: Gabra4, which encodes the α4 subunit of extrasynaptic GABA A receptors, Gabrb2, which encodes the β2 subunit, and Slc12a5, which encodes the chloride symporter KCC2 ( Fig. 4a; n = 4 hippocampi per group; GABRA4: F 3,12 = 5.334, P < 0.05; KCC2: F 3,12 = 4.782, P < 0.05; GABRB2: F 3,12 = 3.662, P < 0.05). miR-33 inhibition reduced the effects of LV-miR-33 on mRNA levels, but did not increase mRNA levels by itself. At the translational level, both LV-miR-33 (Fig. 4b) and miR-33-LNA (Fig. 4c) Fig. 7a-c). The levels of GABA-unrelated proteins relevant for fear conditioning, such as NMDAR subunits or protein kinases, were not affected by LV-miR-33 (Supplementary Fig. 7d,e; NMDAR: F 2,24 = 2.116, P = 0.115; kinases: F 2,24 = 0.31, P = 0.861). Using a pulldown assay with biotinylated miR-33 or control miRNAs (scrambled miRNA and miRNA with a mutated seed sequence), we validated Gabrb2 (F 2,6 = 11.78, P << 0.01) and Kcc2 (F 2,6 = 66.9, P < 0.001) mRNAs as targets for miR-33 using an immortalized hippocampal cell line (n = 3 samples per group; Fig. 4d). Because this cell line does not express Gabra4 mRNA, we performed quantitative PCR using Gabra4 as a negative control. Together, these findings suggest that endogenous hippocampal miR-33 mainly affects the translation of these mRNAs into proteins and only disrupts mRNA stability when    npg a r t I C l e S miR-33 levels are elevated above baseline. This is consistent with previously reported data with other miR-33 targets 20 . In addition to its direct targets, miR-33 inhibition also affects the levels of other proteins, as revealed by proteomic analysis of pooled hippocampal lysates. Of the nine identified candidate proteins ( Supplementary  Fig. 8), synapsin-2 (Syn2) is of particular interest because it is a key regulator of asynchronous GABA release, which is thought to generate tonic currents mediated by extrasynaptic GABA A receptors 21 .
We confirmed that Syn2a is not a miR-33 target (Fig. 4d) (Fig. 4e). Although significant, the changes of all examined GABA-related proteins were not marked, suggesting that the multitude of affected GABA A -related targets, rather than the magnitude of their changes, contributes the most to the behavioral sensitivity to gaboxadol.
Gaboxadol changes stimulus processing in the external hippocampal circuit Contextual fear conditioning depends on neuronal plasticity in both the hippocampus and areas receiving direct hippocampal projections 22 . To establish whether gaboxadol  (e) Effects of LV-miR-33 and miR-33-LNA on the levels of Synapsin-2 proteins. *P < 0.05, **P < 0.01, ***P < 0.001 versus corresponding control groups; # P < 0.05 versus LV-miR-33 + miR-33-LNA groups (one-way ANOVA). Gabra4 a mRNA was not found in the input and was used as a negative control; Syn2a b mRNA was readily detectable in the input, but was not captured by biotinylated miR-33. Error bars represent s.e.m. in all panels. (c) Quantification of cFos-positive neurons in adjacent brain sections. Significant changes were found in the lateral septum, dentate gyrus, RSC cortex and EC cortex. *P < 0.05 versus vehicle, # P < 0.05 and ## P < 0.01 versus V-FC (one-way ANOVA). Error bars represent s.e.m. in all panels. npg a r t I C l e S alters plasticity in the extended hippocampal circuit, we compared the levels of early growth factor 1 (EGR-1) and cFos during encoding of context fear in the presence or absence of gaboxadol. These immediate early genes were selected on the basis of their well-documented roles in plasticity relevant for encoding of contextual fear 23,24 . We first established the main projections from the rostral-dorsal hippocampus by injecting SynaptoTag. This viral vector carries mCherry, which was expressed in the entire neuron, and Syn2-EGFP, which was only expressed at axon terminals (Supplementary Fig. 9a,b). We infused gaboxadol or vehicle through the same cannula 1 month later, performed fear conditioning and then collected brains for immunohistochemical analyses of early growth factor 1 (EGR-1) and cFos responses. Consistent with previous findings 25 , the most prominent sites of rostral-dorsal hippocampal projections were the lateral septum, retrosplenial cortex (RSC) and entorhinal cortex (EC) (Fig. 5a). Gaboxadol significantly increased the number of EGR-1-positive neurons in showing average EGR-1 responses in mice of the V, V-FC and G-FC groups. The significant differences were confirmed with an independent replicate with 3 mice per group. We noted some variability in the CA1 response, which showed a trend toward reduced EGR-1 levels in the first experiment, and significantly reduced levels (P < 0.05) in the second experiment.  Figure 7 Inactivation of RSC at memory retrieval enhances state-dependent fear. (a) Mice from the V-V and G-G groups showed similar freezing levels when RSC activity was intact. Inactivation of RSC with i.p. injection of CNO before the memory test significantly impaired freezing in the V-V group, but enhanced freezing in the G-G group. *P < 0.05 versus V-V (t test), # P < 0.05 and ## P < 0.01 versus corresponding group on a previous test (paired t test). Error bars represent s.e.m. (b) Top, a typical right-field image of a RSC-containing brain slice, from a wild-type mouse that had previously been stereotaxically injected in the RSC with AAV carrying the DREADD and the red fluorescent protein mCherry. All slices were first analyzed for virus spread and only slices of mice with a similar expression pattern of mCherry (n = 4 mice) were used for recordings. Bottom, epifluorescence image showing the expression pattern of infected RSC neurons. (c) Application of CNO to DREADD-expressing RSC neurons reduced their excitability by hyperpolarizing them and increasing the threshold current for firing. Example traces from a whole-cell recording of a DREADD-expressing layer 5 pyramidal neuron in RSC, sampled in current-clamp mode. Under control conditions (gray traces), with the cell at its resting membrane potential, current was injected via the patch pipette in 1-s-long steps (amplitudes of 50, 100 or 150 pA, from bottom to top) to assess the neuron's suprathreshold excitability. Responses to the same family of current steps were acquired again after bath application of 100 nM CNO (red traces) and showed both a hyperpolarization of the resting membrane potential and a large reduction in the number of action potentials evoked by the step stimuli. npg a r t I C l e S the dentate gyrus (F 2,8 = 4.802, P < 0.05) and in the lateral septum (F 2,8 = 6.879, P < 0.05), the main subcortical projection target of the dorsal hippocampus (Fig. 5b). On the contrary, the number of these neurons was significantly reduced in the cortical targets, the RSC (F 2,8 = 11.293, P < 0.01) and EC (F 2,8 = 7.046, P < 0.05). The trend of cFos responses to gaboxadol was similar, although a significant difference was only observed for the lateral septum (P < 0.05; Figs. 5c and 6a,b,  and Supplementary Fig. 10).
RSC suppresses gaboxadol-induced state-dependent fear These changes in immediate early gene responses in the extended hippocampal circuit suggest a shift of information flow from the hippocampus to its subcortical rather than cortical targets during encoding of state-dependent context fear. This is consistent with the view that state-dependent memories are subcortical in nature and are not dependent or even suppressed by cortical activity 26 . We tested this hypothesis by inactivating RSC, which has a major role in retrieval of context-dependent fear-provoking memory 27 , during retrieval tests in the presence or absence of gaboxadol. We first infused an adenoviral vector carrying the inhibitory receptor of the designer receptors exclusively activated by designer drugs (DREADD) family, AAV8. hSyn.hM4D(Gi).mCherry, into the RSC and 6 weeks later exposed the mice to fear conditioning on vehicle or gaboxadol (Fig. 7a). The vehicle mice were then tested on vehicle (V-V) and gaboxadol mice on gaboxadol (G-G) when RSC was intact (intraperitoneal (i.p.) injection of vehicle) or inactivated (i.p. injection of the DREADD ligand clozapine-n-oxide, CNO). When RSC was intact, both groups froze similarly, as found in the previous experiments ( Fig. 7a; n = 8 mice per group; t 14 = 0.17, P = 0.72). However, when RSC was inactivated by CNO, the V-V group froze significantly less, whereas the G-G group froze significantly more, than on the previous test, revealing significant effects of RSC inactivation (F 1,14 = 17.46. P < 0.01) and interaction between RSC inactivation and gaboxadol treatment (F 1,14 = 19.19, P < 0.01) (Fig. 7b-f). Together, these findings demonstrate that statedependent memories acquired under gaboxadol are best retrieved when RSC is inactivated. Thus, cortical mechanisms required for the retrieval of normally acquired memories are not required, and even impair retrieval of gaboxadol-induced state-dependent memories.

DISCUSSION
Since its discovery 26 , state-dependent learning has been demonstrated in studies with humans 28,29 and animals 5 ; however, the data have not always been consistent 30 . Drugs that induce state-dependent operant conditioning or passive avoidance, such as benzodiazepines, NMDAR antagonists or scopolamine 31,32 , have proved ineffective in fear conditioning [33][34][35] . This may be a result of the very narrow dose-range that allows investigation in the absence of side effects, which commonly include changes of locomotor activity and thus confound analyses of freezing behavior. Alternatively, state-dependent regulation of context fear may be restricted to fewer neurobiological mechanisms, such as the hippocampal GABAergic pathway that we identified.
Under control (normal) conditions, fear conditioning critically depends on hippocampal glutamate receptors and cAMP-dependent protein kinase (PKA) signaling [36][37][38] . From this perspective, our experiments can be viewed as a comparison of fear responses encoded in two different states: a glutamate receptor/PKA-mediated state in control mice versus an extrasynaptic GABA A receptor/PKCβII-mediated state in gaboxadol-injected mice. These states seem to be separated by an amnestic barrier because, in both groups, retrieval of fear-provoking memory was confined to the state in which fear conditioning occurred. Notably, contextual fear acquired with or without gaboxadol showed many phenomenological similarities, such as contextual specificity, lack of generalization and comparable freezing levels. Nevertheless, the molecular mechanisms underlying fear conditioning in the presence or absence of gaboxadol were different, as revealed by the finding that PKCβII signaling and miR-33 had significant roles in GABAergic mechanisms of context fear, but showed no involvement in controls.
Fear conditioning in the presence and absence of gaboxadol induced opposite changes of the levels miR-33 and its GABA-related targets, which lasted at least up to 24 h post-training. Gabra4, Kcc2 and Gabrb2 mRNA or protein levels changed inversely with the level of miR-33, and these changes were consistent with the direction of behavioral susceptibility to gaboxadol. This is consistent with observations that microRNAs simultaneously target several functionally related mRNAs and proteins 39 . An unexpected observation was that levels of Syn2a, whose mRNA is not predicted to interact with miR-33, followed this pattern, suggesting that miR-33 indirectly regulates Syn2a through one of its primary targets.
Although some microRNAs are necessary for conditioning and extinction of fear 40,41 , we found that miR-33 did not affect fear conditioning or anxiety-or depression-like behavior (data not shown) under normal conditions. Instead, it regulated the threshold for GABA A receptor-mediated state-dependent fear. These findings are particularly important in view of the observations that these receptors control brain states ranging from sleep to heightened affect and psychosis 42 , and that brains of patients suffering from major depression and psychosis show consistent alterations of miR- 33 (refs. 43,44) along with a disruption of the glutamatergic and GABAergic balance 45 . These findings suggest that some brain microRNAs can be predisposing factors for specific brain states and corresponding mental states or disorders.
Extrasynaptic GABA A receptors are especially abundant in interneurons of the dentate gyrus 46 ; thus, an increase of EGR-1 responses in this hippocampal subfield by gaboxadol is most likely a result of disinhibition of the EGR-1 response to fear conditioning. To our surprise, such enhanced activity did not result in an overall enhancement of immediate early gene responses in the hippocampal projections. Instead, EGR-1, and in part cFos responses, were enhanced in subcortical targets of hippocampal inputs, but suppressed in cortical targets. These results experimentally support the conclusions of Girden and Cullar 26 , who, in their seminal work on state-dependent learning under curare, suggested that conditioning under curare is subcortical in nature and does not require, or is even suppressed by, cortical activity. Accordingly, we found that RSC suppresses the retrieval of state-dependent memories, even though it is required for retrieval of normally acquired memories.
Taken together, our results identify a molecular pathway involving extrasynaptic GABA A receptors, PKCβII signaling and miR-33 as a mediator of state-dependent encoding and retrieval of contextual fear. This pathway enhanced EGR-1 responses in the dentate gyrus of the hippocampus and promoted subcortical while inhibiting cortical processing of context memories. Our evidence for multiple mechanisms of fear has important implications for the treatment of patients experiencing fear and anxiety comorbid with other mental disorders 47 , as, despite similar manifestations, these symptoms may require disorder-specific therapies.

METhODS
Methods and any associated references are available in the online version of the paper.