Introduction

The ability to remember aversive things and avoid them is crucial for the survival of humans and animals. Memory, including aversive memory, is inherently malleable. However, when aversive memory is related to addictive drug withdrawal, it can generate a motivational state leading to compulsive drug taking [1]. And also, a fear memory induced by strong trauma may result in a long period of emotional disorder [2]. Extinction training was developed for use as a therapy to interfere with these types of pathological memories [3,4,5]. The recall-extinction procedure is very efficient for the change of initial memory trace [6,7,8,9,10,11]. This suggests that memory recall is a hub between acquisition, recall, and extinction, and recall induces a memory into a relatively labile condition under which the memory is easy to be modified [12,13,14]. However, the effect of the recall-extinction procedure does not always seem to be replicable in terms of memory modulation [15, 16]. Whether the memory is successfully recalled may be one of the important factors that influence the extinction efficacy [17, 18]. In this study, we were looking for a specific circuit that is involved in the recall-extinction procedure and that can be potentially manipulated to enhance this learning. Successful revelation to this circuit would contribute to understanding the mechanisms underlying memory updating and provide a potential strategy for removing pathological memory.

The dentate gyrus (DG) engages in memory encoding and recall, as well as extinction [19, 20]. The DG seems to be a noteworthy brain region for investigating the role of memory recall in bridging encoding and extinction processes. Although there are inconsistent opinions showing the exclusive role of the DG in recall [21, 22], a recent report further isolated the function of DG in memory expression, that is most directly revealed when the similar contexts need to be distinguished [23]. In other words, the DG plays a role in the accuracy of memory recall. The different cell ensembles in the DG that are encoding different context may together engage accurate memory recall, which is also called pattern separation. The idea is supported by the studies that activating cell ensembles in DG genetically tagged during acquisition is sufficient to drive memory recall, whereas silencing these ensembles prevents it [24,25,26,27]. A study on fear memory further revealed that the degree of fear reduction is revealed to be positively correlated with the reactivation of recall-induced engram cells in DG during the extinction process [28], suggesting the role of DG cells for linking the processes of recall and extinction of memory. Thus, we hypothesized that the cell ensemble in the DG encoding the learned information might be potential neural targets for memory modulation. The approaches that induce this cell ensemble into active or plastic state may contribute to the cellular and synaptic remodeling in the DG and facilitate the extinction learning [29].

Noradrenergic (NA) signaling is a critical mechanism for reconsolidation of emotional memories in rodent and human models [30]. The function of NA mainly goes through NA receptors in hippocampal neurons [31, 32]. Rather than producing direct excitatory or inhibitory effects on postsynaptic neurons, NA transmitters regulate such effects produced by other neurotransmitters, such as glutamate and gamma amino butyric acid [33]. NA cells in the locus coeruleus (LC) are the main source of NA in the brain and send dense projections to the DG [34,35,36]. A recent study isolated a key function of NA signaling in the LC-DG circuit in contextual aversive generalization through selective manipulation and pharmacological intervention to LC-DG circuit during memory acquisition [37], highlighting the crucial role of LC-DG in memory formation. However, the modulatory role of the LC-DG circuit in acquired memory was still unknown. Since the short-term increase of engram cell excitability in DG enhances the subsequent recall of specific memory content in response to cues [38], we postulated that preheating the DG cells by activating LC-DG circuit would be a potential approach that facilitates the activation number or excitability of the DG cells during recall process and further improves the extinction learning, depending on a recall-extinction procedure.

To test the hypothesis, we established a naloxone-precipitated conditioned place aversion (CPA) model in morphine-dependent mice, which is a robust associative model for pathological aversion memory [4, 39, 40]. In this behavioral model, repeated opioid exposure induces an opioid dependent state; and then, the animals are trained to acquire an associative memory between negative withdrawal emotion and the conditioning context [4, 5]. Naloxone is used to trigger the withdrawal response and the associated emotion. In the present study, we firstly clarified that the LC and the DG subarea of hippocampal formation, but not the cornu ammonis (CA1, CA2, or CA3), were activated during CPA memory recall. And then, using chemogenetic tools, we found an impairment in naloxone-precipitated CPA expression when we inhibited LC-DG circuit, which suggests the necessity of the LC-DG activation in the memory recall. We next revealed that optogenetic activation of the LC-DG circuit followed by a recall-extinction procedure promoted the process of extinction. In further investigations with the Fos-targeted recombination in activated populations (FosTRAP) strategy, we showed that activating the LC-DG-facilitated reactivation of DG engram cells in the CPA memory contributed to the extinction learning, and the engram cell population involved in above memory modulation was associated with the memory recall rather than the acquisition.

Materials and methods

Subjects

Male wild-type and FosTRAP2 mice (8–10 weeks old) were used in this study. All mice were maintained on a circadian 12 h light–12 h dark cycle (light on at 7:00 am). After stereotactic surgery, each mouse was kept in one cage for at least 2 weeks before subsequent experiments. Experiments were conducted during the light phase. All procedures were approved by the Animal Care and Use Committee in Institute of Psychology, Chinese Academy of Sciences.

Stereotaxic surgery

Mice were anesthetized and placed in a stereotaxic frame. A glass micropipette was attached to a 10 μl microsyringe using a silicone tube. The injection speed was maintained at 0.1 μl/min. After finishing the injection, the glass micropipette stayed in brain for 20 min before it was withdrawn slowly. The virus injection volumes for the DG and LC were 0.4 μl/side and 0.3 μl/side, respectively. The virus titer in all experiments was around 1.0 × 1013 vg/m. Bilateral DG regions were targeted at −2.0 mm AP, ±1.4 mm ML, and −2.0 mm DV (virus injection) or −1.5 mm DV (fiber implanted). Bilateral LC regions were targeted at −5.6 mm AP, ±0.9 mm ML, and −3.6 mm DV.

Apparatus for behavioral assessments

Acrylic dumbbell-like boxes with three chambers were selected for naloxone-precipitated CPA experiments. A white chamber (16.5 × 15 × 15 cm) with hole floors and a black chamber (16.5 × 15 × 15 cm) with grid floors were located at the ends. The central gray chamber (9.5 × 5 × 15 cm) was equipped with a slippy floor. The floor of all chambers was white. Between the two adjacent chambers, there was a door that could be closed as needed. In optogenetic stimulation-induced real-time CPA experiments, the same boxes as those in the naloxone-precipitated CPA model were used. The box was divided from the middle into two chambers. The walls and floors of the chambers were decorated according to the experimental requests.

Naloxone-precipitated CPA in morphine-dependent mice

Morphine hydrochloride solution (dissolved in 0.9% saline) was continuously administered for 7 days. From day 1 to day 4, morphine hydrochloride solution was injected into the mice intraperitoneally (i.p.) twice a day (12 h intervals). From day 5 to day 7, morphine hydrochloride solution was injected into the mice once a day. The dose of morphine hydrochloride was increased from day 1 to day 5 in the following amounts: 12 mg/kg, 24 mg/kg, 36 mg/kg, 48 mg/kg, and 60 mg/kg [41, 42]. The last dose (60 mg/kg) was maintained on days 6 and 7. On day 3, the pre-test was performed 5 h after the first morphine injection. Mice were placed in three-chamber behavioral boxes and allowed to move freely for 15 min. In the present setup, mice generally preferred staying in the black chamber; this was used for naloxone-paired training and referred to as the naloxone-paired chamber. By contrast, the white chamber was referred to as the saline-paired chamber. In particular, saline-paired training was performed 5 h after the first morphine injection on days 4 and 6. Mice were limited to the saline-paired chamber for 30 min following injection of 0.2 ml saline. Naloxone-paired training was performed 5 h after the first morphine injection on days 5 and 7. Mice were limited to the naloxone-paired chamber for 30 min following being injected with naloxone solution (0.3 mg/kg). A post-test was performed on day 8, and the mice were allowed to move freely in the three-chamber behavioral box for 15 min.

The naloxone-paired results on both the pre-test and post-test, CPA score, and ΔCPA score were obtained using the following formula:

$$\begin{array}{l}Naloxone - paired({{{{{{{\mathrm{\% }}}}}}}}) = \frac{{Duration_{Naloxone - paired}(s)}}{{Duration_{Naloxone - paired}\left( s \right) \,+\, Duration_{Saline - paired}(s)}} \times 100\\ CPA\;score = Naloxone - paired({{{{{{{\mathrm{\% }}}}}}}})_{post - test} - Naloxone - paired({{{{{{{\mathrm{\% }}}}}}}})_{pre - test}\\ \Delta CPA\;score = Naloxone - paired({{{{{{{\mathrm{\% }}}}}}}})_{extinction} - Naloxone - paired({{{{{{{\mathrm{\% }}}}}}}})_{post - test}\end{array}$$

Real-time CPA

Two-chamber boxes were used in the real-time CPA experiment. Via a 20 min pre-test, the innate chamber bias was identified for each mouse and selected as its laser-paired chamber. On the second day, real-time CPA training was performed, and the mice were placed in the box and allowed to move freely for 20 min. Laser stimulus was automatically applied once the mice entered the laser-paired chamber and lasted until the mice left.

Laser-paired (%) of both pre-test and training was calculated as:

$$Laser - paired({{{{{{{\mathrm{\% }}}}}}}}) = \frac{{Duration_{Laser - paired}(s)}}{{Duration_{Laser - paired}\left( s \right) + Duration_{Non - laser - paired}(s)}} \times 100$$

Optogenetic activation and suppression

Mice were handled for 3–4 days prior to experiments and habituated to fiber-optic cables. Laser stimulation for neuronal activation and inhibition was administered with the following parameters: for activation (473 nm, 20 Hz, 5 ms pulse-width, 1 s light duration and 4 s intervals in each 5 s segment); for inhibition (589 nm and 2900 ms of constant stimulation with 100 ms intervals). The light power from the fiber tip was around 20 mW. The stimulation lasted for 10 min in each session.

Chemogenetic suppression

To selectively manipulate the LC-DG circuit, AAV(2/9)-Syn-Cre-EGFP and AAV(2/R)-Syn-DIO-hM4D(Gi)-mCherry were respectively injected into the LC and DG bilaterally. The naloxone-precipitated CPA model was established as described above, and on day 8, clozapine-N-oxide (CNO; 5 mg/kg [43,44,45]) or vehicle (0.2 ml) was injected into each mouse 40 min prior to the CPA post-test. The information for virus was shown in Supplementary Table S1, and similarly hereinafter.

Calcium imaging by fiber photometry

AAV(2/9)-Syn-DIO-GCaMP6s was bilaterally injected into the DG of FosTRAP2 mice, followed by fiber (230 μm core, NA 0.5) implantation in the bilateral DG region. A commercialized fiber photometry system (ThinkerTech Inc., Nanjing, China) was used to record the calcium signal from the event-tagged cells. Calcium-dependent fluorescence signals, depending on a calcium influx, were excited by 480 nm LED light (40 μW). The baseline fluorescence was acquired prior to each signal recording (5 min). The sample rate was 50 Hz. Data analysis was performed with MATLAB software (R2021b). Signals were normalized to the baseline to calculate as dF/F = (Fsignal − Fbaseline)/Fbaseline.

Immunohistochemistry

Mice were perfused with cold 4% paraformaldehyde (PFA) 90 min after behavioral tests, and the brains were post-fixed in 4% PFA for 6 h. And then the gradient of dehydration was performed to transfer the brains from 20 to 30% sucrose solution in phosphate-buffered saline (PBS) at 4 °C. Coronal sections (40 μm) were prepared. Floating slices were rinsed three times (3 × 15 min) in PBS and blocked in PBS with 2% Triton X-100 and 10% normal goat serum (NGS) for 2 h at room temperature (RT). After the samples were rinsed three times, incubation with primary antibody in 5% NGS and PBS solution was performed at 4 °C overnight (rabbit anti-cFos, 1:1000). After another three times rinsed, slices were incubated with a secondary antibody (anti-rabbit, 1:500) for 2 h at RT. Slices were washed again and dyed with 4′,6-diamidino-2-phenylindole (DAPI) for 20 min. After washing, slices were transferred onto pathological-grade slides and imaged with fluorescence microscopy. Cell counting analysis was performed using ImageJ software (1.51j8). The DG granule cells were outlined as a region of interest according to the DAPI signal in each slice. The number of EGFP+ or cFos+ cells was calculated by thresholding EGFP or cFos immunoreactivity above background levels. The overlap chance level was calculated as chance level = (EGFP+/DAPI) × (cFos+/DAPI) [28, 46].

Statistical analysis

Statistical analyses were performed by using Prism7.0 software. t-tests, one-way and two-way analysis variance (ANOVA), and simple liner regression were engaged in data analysis. Data were considered statistically significant when p < 0.05. The corresponding statistical analysis information for each experiment is shown in Supplementary Table S2.

Results

Activation of LC-DG circuit is necessary for aversive memory recall

To clarify whether the LC engages in aversive memory recall and which subregions of dorsal hippocampus are involved, we used an immunohistochemical method to compare the cFos expression in the LC and the dorsal CA1, CA2, CA3, and DG subregions between the mice with naloxone-precipitated CPA expression and the controls. Figure 1A shows behavioral procedure. To develop a morphine-dependent condition, mice were subjected to 7 days of morphine intraperitoneal injections, and from the third day of the injection regimen, a procedure to establish naloxone-precipitated CPA was started. Mice were randomly divided into four groups after pre-test: Mor-Nal (morphine injection and naloxone-paired conditioning), Mor-Sal (morphine injection and saline-paired conditioning), Sal-Nal (saline injection and naloxone-paired conditioning), and Sal-Sal (saline injection and saline-paired conditioning). Behavioral results showed that the CPA score of the Mor-Nal group significantly decreased compared to the other groups (Fig. 1B; see Supplementary Table S2 for more details on effect statistics, similarly hereinafter). Coronal slices of dorsal hippocampus around the coordinates shown in Fig. 1C were selected for cFos staining. Imaging results showed that the density of cFos positive cells had significantly increased in LC and DG (but not in CA1, CA2, or CA3) in the Mor-Nal group compared to the controls (Fig. 1E). Figure 1D shows representative images for DG and LC. These observations indicate that the LC and DG, rather than other dorsal hippocampal subregions, are involved in aversive memory recall. To test whether the CPA recall process is necessary for the change of cell activity in the DG and LC, we compared cFos levels between the groups both with and without a post-CPA test. Both the DG and LC regions were significantly activated with memory recall, when compared to no-recall (kept in their home cages) (Supplementary Fig. S1A–C).

Fig. 1: Enhanced activity in DG and LC during aversive memory recall.
figure 1

A Procedure for naloxone-precipitated CPA training in morphine-dependent mice. For the first 2 days, morphine solution was injected while the mice were in their home cage. From day 3 to day 7, morphine administration was accompanied with the establishment of a CPA model. Twenty-four hours after the pre-test, CPA training began with two sessions of naloxone-paired training and two sessions of saline-paired training, which were conducted alternately for one session per day. Each training was started 5 h after the first morphine injection. On day 8, the mice were sacrificed for cFos immunohistochemistry (IHC) 1.5 h after post-test. B Naloxone-paired training in morphine-dependent mice induced CPA expression; the groups were Mor-Nal (morphine administration and naloxone-paired on day 5 and 7), Mor-Sal (morphine administration and saline-paired on day 5 and 7), Sal-Nal (saline administration and naloxone-paired on day 5 and 7), and Sal-Sal (saline administration and saline-paired on day 5 and 7) (one-way ANOVA followed by Tukey’s multiple comparisons test; F3,28 = 16.53, p < 0.0001; n = 9 in Sal-Sal and Sal-Nal, and n = 7 in Mor-Sal and Mor-Nal). C Delineation of dorsal hippocampal subfields of interest [95]. Blue, CA1; yellow, CA2; green, CA3; red, DG. D Representative images showing cFos expression in the DG and LC. E The densities of cFos+ cells significantly increased in the LC and DG, but not in CA1, CA2, or CA3 in Mor-Nal mice (one-way ANOVA followed by Tukey’s multiple comparisons test; DG: F3,20 = 7.307, p = 0.0017; LC: F3,20 = 22.09, p < 0.0001). Data are expressed as means ± s.e.m. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

Full size image

To probe whether LC-modulated neuronal activation in the DG is required for withdrawal memory recall, we bilaterally injected an AAV expressing hM4Di [AAV(2/R)-Syn-DIO-hM4D(Gi)-mCherry] (see Supplementary Table S1 for more details on virus information, similarly hereinafter) into the DG and AAV(2/9)-Syn-Cre-EGFP into the LC of wild-type mice (Fig. 2A). CNO was administered (i.p.) at a dose of 5 mg/kg, 40 min before the CPA test. In the control groups, the virus for injecting into the DG was replaced by AAV(2/R)-Syn-DIO-mCherry (in mCherry-Sal and mCherry-CNO groups), or CNO (i.p.) was replaced by saline (i.p.) before the behavioral test (in hM4Di-Sal groups). The protein of chemogenetic suppression was successfully expressed in the LC (Fig. 2B); in morphine-dependent animals without hM4Di exposure or CNO injection, naloxone-paired conditioning produced CPA establishment (Fig. 2C, D), whereas the mice with hM4Di and CNO injections no longer showed aversive behavior to naloxone-paired chambers (Fig. 2C, D). Besides behavioral observations, imaging results also showed a reduction in cFos expression in the DG of hM4Di-CNO naloxone-trained morphine-dependent mice, relative to hM4Di- or CNO-lacking conditioned mice (Fig. 2E, F). In further detailed analyses for behaviors during the CPA test, we did not find any difference in locomotion between different groups (Supplementary Fig. S2A); naloxone-trained mice with morphine dependence—but not saline exposure—showed a significant increase in rearing behavior, while LC-DG suppression had no further effects on this behavioral change (Supplementary Fig. S2B). The LC-DG circuit may play no part in naloxone-induced anxiety-like responses in morphine dependence [47]. This implication is consistent with previous findings that lesioning of NA neurons in the LC does not influence acute morphine withdrawal syndrome or naloxone-precipitated aversion [48]. We also conducted a chemogenetic experiment using a fear memory procedure and found that LC-DG suppression inhibited the expression of contextual fear memory (Supplementary Fig. S2C). Thus, the circuit of LC-DG may be involved in modulating negative memories such as naloxone-precipitated withdrawal-associated place aversion and fear memory recall.

Fig. 2: LC-modulated DG activation was essential for aversive memory recall.
figure 2

A Diagram of the virus injection strategy for selectively chemicogenetic manipulation in the LC-DG circuit. B Representative images showing the range of virus infection in the LC and the overlap of hM4D(Gi)-mCherry and Cre-EGFP expression in the LC. C The suppression of the LC-DG pathway inhibited naloxone-precipitated CPA expression in morphine-dependent mice (mCherry-Sal, n = 7; the rest groups, n = 8) compared to Sal-Nal treatment (n = 8). The hM4Di-CNO group also showed decreased CPA scores compared to the other groups in the morphine-naloxone treatment condition (two-way ANOVA followed by Tukey’s multiple comparisons test; F3,55 = 6.127, p = 0.0011). D There were no significant changes in the percentage of duration in the naloxone-paired chamber between the pre-test and post-test in the Sal-Nal treatment groups (upper panel: two-way ANOVA followed by Tukey’s multiple comparisons test; F3,56 =1.1025, p = 0.3885); this percentage of duration is significantly lower in the post-test in the Mor-Nal treatment groups, except the hM4Di-CNO group (lower panel: two-way ANOVA followed by Tukey’s multiple comparisons test; F3,54 = 5.400, p = 0.0025). E The representative images show cFos expression in the DG. F Chemogenetic suppression of the LC-DG pathway blocks DG activation, which is associated with CPA memory recall (two-way ANOVA followed by Tukey’s multiple comparisons test; F3,55 = 6.661, p = 0.0006). Data are expressed as mean ± ss.e.m. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

Full size image

Activation of LC-DG circuit in recall-extinction procedure facilitates aversion extinction

Utilizing whole-cell recordings (see Supplementary Methods) combined with optogenetics, we demonstrated that activating LC terminals in the DG increased the excitability of DG cells in CPA-modeled mice (Supplementary Fig. S3). A FosTRAP strategy (see Supplementary Methods) and cFos immunohistochemical methods were used [49] (Supplementary Fig. S4A), and a suitable duration between the TRAPing event and animal sacrifice was identified (Supplementary Fig. S4B–D); 10–14 days after TRAPing was therefore chosen for full expression of the event-tagged fluorescent protein.

According to reconsolidation theory [12, 13] and the role of LC-DG circuit in aversive memory recall, we hypothesized that artificially activating the circuit before memory recall could promote the subsequent extinction process, and reactivation of the cells tagged during acquisition might possibly contribute to extinction learning. To test this idea, firstly, wild-type mice were bilaterally injected with AAV(2/9)-Syn-ChR2-mCherry or AAV(2/9)-Syn-mCherry in LC and implanted with light fibers in DG (Supplementary Fig. S5A, B). In recall-extinction procedure, mice were allowed to move freely in a CPA box for 10 min to recall the memory and for 40 min to establish the extinction, with an hour interval between the two steps [31, 50, 51]. All mice were subjected to a 10 min laser stimulation in the DG terminals of LC inputs before the extinction or recall-extinction procedure (Supplementary Fig. S5B). Behavioral data for the last 10 min of the extinction session were used for ΔCPA score calculation. The mice in ChR-Rec-Ext group exhibited higher ΔCPA scores (the difference in CPA scores between extinction and post-test) than controls (Supplementary Fig. S5C, D), suggesting that LC-DG activation improves extinction learning through modulating recall process, but without any direct influence on extinction in this protocol. In addition, behavioral analysis showed no significant effect of LC-DG activation on the recall itself (Supplementary Fig. S5E).

Subsequently, FosTRAP2 mice were used to reveal the mechanisms underlying LC-DG-facilitated extinction learning at the engram cell level. The mice were bilaterally injected with AAV(2/9)-Syn-ChR2-mCherry or AAV(2/9)-Syn-mCherry in LC and AAV(2/9)-DIO-EGFP in DG; light fibers were bilaterally implanted in DG. The groups of mCh-Ext, mCh-Rec-Ext and ChR-Rec-Ext were included. To tag the DG cells active during acquisition of the naloxone-precipitated morphine-withdrawal memory, we injected tamoxifen 6 h before the second naloxone-paired session. Ten days after the CPA modeling, the mice underwent an extinction or recall-extinction session following a 10-min laser stimulation, depending on the groups (Supplementary Fig. S6A). After behavioral test, the densities of EGFP+ and cFos+ cells in the DG were examined (Supplementary Fig. S6B). The LC-DG activation followed by recall caused an increase in DG activity during the extinction (Supplementary Fig. S6C), while there was no correlation between DG neuronal activity and behavioral extinction (Supple. Fig. S6E). The circuit activation failed to improve the reactivation of acquisition-tagged cells during the extinction (Supplementary Fig. S6D, F). These imply that promoted extinction process by the LC-DG activation is neither a result of enhanced DG activation during the extinction, nor related to the reactivation of acquisition-induced cell ensemble in the DG.

We then considered whether more recall-induced DG cells were activated during the extinction following the LC-DG activation. Protocol is shown in Fig. 3A. FosTRAP2 mice were injected bilaterally with AAV(2/9)-Syn-ChR2-mCherry or AAV(2/9)-Syn-mCherry in LC and AAV(2/9)-DIO-EGFP in DG; light fibers were bilaterally implanted in DG. We tagged recall-activated cells and stained extinction-activated cells in the DG with cFos antibodies. Tamoxifen was injected 6 h before the CPA recall. After 10 days, behavioral assessment was performed based on the recall-extinction procedure, and a 10-min laser stimulation was administered just before the CPA recall. Imaging representatives are respectively illustrated in Fig. 3B. Activation of the LC-projecting terminals in the DG did not change the density of the active neurons (Fig. 3C) during the CPA recall, while increased the cFos+ cell density during the extinction (Fig. 3D). We also revealed significant increases in the proportion of merged cells with the EGFP+ cells in the ChR-Rec-Ext group, relative to the mCh-Rec-Ext group (Fig. 3E). By further analyses, we showed statistically significant positive correlation between the ΔCPA score and the proportion of merged cells among the EGFP+ cells (Fig. 3G), while no notable correlation between the ΔCPA score and the density of cFos+ cells (Fig. 3F). These results indicate that the LC-DG activation-promoted extinction learning is mainly attributable to the enhanced population size of the reactivated recall-tagged DG engram cells during extinction.

Fig. 3: Changes in population size of reactivated recall-tagged DG cells in extinction.
figure 3

A Procedure for testing the colocalization of cells tagged by recall (EGFP+) and extinction (cFos+) in FosTRAP2 mice. B Representative images showing colocalization between EGFP+ and cFos+ cells in the DG. C No differences in the density of EGFP+ cells in DG between mCh-Rec-Ext group (n = 6) and ChR-Rec-Ext group (n = 9) (unpaired t-test; t14 = 1.261, p = 0.2279). D Compared to the mCh-Rec-Ext group, the ChR-Rec-Ext group showed a higher density of cFos+ cells in DG (unpaired t-test; t14 = 3.697, p = 0.0024). E The ChR-Rec-Ext group showed a higher proportion of cells merged with the EGFP+ cells in DG (unpaired t-test; t14 = 5.597, p < 0.0001). F The data from the two groups were pooled for correlation analysis; no correlation between the density of the cFos+ cells and the ΔCPA score was found (n = 15) (simple linear regression; F 1, 13 = 1.426, p = 0.2538; Pearson correlation coefficient; r = 0.314). G There was a positive significant correlation between the proportion of merged/EGFP+ cells and the ΔCPA score (n = 15) (simple linear regression; F 1, 13 = 14.55; p = 0.0021; Pearson correlation coefficient; r = 0.727). H Procedure to test the effect of selectively inhibiting recall-tagged cells on LC-DG-facilitated extinction. Ten days after FosTRAP strategy enablement, recall-tagged cells specifically expressed EGFP (ChR-EGFP-Ext group; n = 7) or NpHR-EGFP (ChR-NpHR-Ext group; n = 8). A laser with 589 nm wave was performed to the mice during the recall process. These mice had ever been subjected to a 473 nm stimulation before the recall-extinction procedure. I Specific inhibition to recall-tagged cells in the DG suppressed the extinction learning induced by the LC-DG activation before the recall-extinction procedure (unpaired t-test, t13 = 3.924, p = 0.0017). J The duration percentage in naloxone-paired chambers was calculated in the post-test and extinction processes (two-way ANOVA followed by Tukey’s multiple comparisons test; F1, 26 = 5.676, p = 0.0248). Data are expressed as means ± s.e.m. **, p < 0.01; ***, p < 0.001.

Full size image

To further identify the importance of recall-induced DG cells for improved extinction learning by the LC-DG circuit activation, we tested the effects of selectively inhibiting this cell ensemble on behavioral extinction. FosTRAP2 mice were used. Experimental protocol is shown in Fig. 3H. AAV(2/9)-Syn-ChR2-mCherry was bilaterally injected into the LC of all mice and AAV(2/9)-DIO-NpHR-EGFP (ChR-NpHR group) or AAV(2/9)-DIO-EGFP (ChR-EGFP group) was bilaterally injected into DG, and the light fibers were bilaterally implanted in DG. Tamoxifen was injected 6 h before CPA recall to initiate the expression of cre recombinase in recall-induced cells. NpHR-EGFP or EGFP were turned on in these cells, depending on groups. After 10 days, a recall-extinction procedure was performed following the 473 nm laser stimulation to the LC terminals in the DG. During the recall period, 589 nm laser stimulation was subjected to all mice. Result showed that the suppression of recall-induced DG cells reversed the facilitating efficiency of the LC-DG activation in extinction learning (Fig. 3I, J), indicating a crucial role of recall-induced DG cells for the memory modulation in the recall-extinction procedure.

Manipulations for LC-DG activation increase calcium signals in recall-induced DG cells during extinction

In order to reinforce the finding that LC-DG activation contributes to the reactivation of recall-tagged DG cells in the extinction, we conducted two more experiments, in which a FosTRAP strategy and a fiber photometry imaging method were used. At first, we injected AAV(2/9)-Syn-ChR2-mCherry or AAV(2/9)-Syn-mCherry in LC and AAV(2/9)-DIO-GCaMP6s in DG and implanted light fibers in the DG of FosTRAP2 mice. After the model establishment of naloxone-precipitated CPA, the mice underwent a CPA recall procedure for tagging recall-activated DG cells with GCaMP proteins. After 10 days, the recall-extinction procedure was performed, and calcium images were recorded (Fig. 4A). The LC-DG circle activation before the recall-extinction procedure promoted the extinction process (Fig. 4B, C). We recorded the calcium signals of recall-induced DG cells for the last 10 min of extinction process (Fig. 4D). Results showed an increased area under the curve (AUC) and event frequency in ChR-Rec-Ext group, compared to mCh-Rec-Ext group (Fig. 4E, G), indicating that the population of recall-tagged cells in the DG exhibited greater activity after the LC-DG activation [52, 53]. We further revealed a significant correlation between the degree of CPA extinction and the intensity of calcium signals in recall-induced DG cells (Fig. 4F, H). These findings suggest that promoted reactivation of recall-tagged DG cells forms the mechanism that underlies LC-DG-modulated aversive memory attenuation.

Fig. 4: Changes in calcium signals of reactivated recall-tagged DG cells in extinction.
figure 4

A Procedure for testing the change in calcium signals of the recall-tagged cells during the extinction process with optogenetic LC-DG stimulation (473 nm) before the recall-extinction procedure. B There was a significant difference in the ΔCPA scores between the mCh-Rec-Ext group (n = 9) and ChR-Rec-Ext group (n = 9) (unpaired t-test; t10 = 3.410, p = 0.0003). C The duration percentage in naloxone-paired chambers was calculated in the post-test and extinction processes (two-way ANOVA followed by Tukey’s multiple comparisons test; F1,32 = 17.00, p = 0.0002). D The average dF/F of calcium signals during the last 10 min of the extinction process. E The area under curve (AUC) in the ChR-Rec-Ext group significantly increased, compared with the mCh-Rec-Ext group (unpaired t-test; t16 = 4.338, p =  0.0005). F The data from the two groups were pooled for correlation analysis; a significant correlation between the AUC and the ΔCPA score was found (n = 18) (simple linear regression; F 1, 16 = 9.205, p = 0.0079; Pearson correlation coefficient; r = 0.604). G The event frequency increased in the ChR-Rec-Ext group, compared with the mCh-Rec-Ext group (unpaired t-test; t16 = 4.756, p < 0.0001). H These was a significant positive correlation between the event frequency and the ΔCPA score (n = 18) (simple linear regression; F 1, 16 = 5.573, p = 0.0313; Pearson correlation coefficient; r = 0.508). I Procedure for testing the change in the calcium signals of the recall-tagged cells during the extinction process with exposure to a novel environment before the recall-extinction procedure. J The ΔCPA score in the mCh-Nov-Rec-Ext group (n = 6) was higher than those in the mCh-Rec-Ext (n = 8) and NpRH-Nov-Rec-Ext groups (n = 6) (one-way ANOVA followed by Tukey’s multiple comparisons test; F2,17 = 13.03, p = 0.0004). K The duration percentage in naloxone-paired chambers in the post-test and extinction processes was calculated and shown (one-way ANOVA followed by Tukey’s multiple comparisons test; F2,34 = 10.45, p = 0.0003). L The average dF/F of the calcium signals during the last 10 min of the extinction process. M The AUC in the mCh-Nov-Rec-Ext group significantly increased, compared with the other two groups (one-way ANOVA followed by Tukey’s multiple comparisons test; F2,17 = 8.423, p = 0.0029). N There was a significant positive correlation between the AUC and the ΔCPA score (n = 20) (simple linear regression; F 1, 18 = 10.11, p = 0.0052; Pearson correlation coefficient; r = 0.600). O The event frequency in the mCh-Nov-Rec-Ext group significantly increased, compared with the other two groups (one-way ANOVA followed by Tukey’s multiple comparisons test; F2,17 = 11.57, p = 0.0007). P There was a significant positive correlation between the event frequency and the ΔCPA score (n = 20) (simple linear regression; F 1, 18 = 11.69, p = 0.0031; Pearson correlation coefficient; r = 0.628). Data are expressed as means ± s.e.m. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

Full size image

The aim of the following experiment was to extensively explore behavioral interventions to memory extinction, potentially with the LC-DG activation mechanism. Experimental protocol was similar to the previous experiment, but with a replacement of 473 nm laser administration by 10 min of exposure to a novel environment (different to the home cage or CPA training chamber) (Fig. 4I); previous study revealed the function of novel environment exposure in the LC excitation [54]. FosTRAP2 mice were injected with AAV(2/9)-Syn-NpHR-mCherry into LC and AAV(2/9)-DIO-GCaMP6s into DG of the mice in NpHR-novel-recall group, while the virus carrying NpHR was replaced by AAV(2/9)-Syn-mCherry in recall and novel-recall groups. All mice were implanted with light fibers in DG. The 589 nm laser stimulations were respectively performed to the mice during novel environment exposure (mCh-Nov-Rec-Ext and NpHR-Nov-Rec-Ext groups) and in their home cages (mCh-Rec-Ext group). Results showed that novel exposure before recall-extinction improved the extinction learning and that this change was reversed by optogenetic LC-DG suppression (Fig. 4J, K). In the analysis of the calcium imaging data, we revealed that increased calcium signals in recall-tagged DG cells with novel exposure were reduced to the control level by LC-DG suppression (Fig. 4L, M, O). Furthermore, the degree of CPA extinction and the intensity of the calcium signals in the DG engram cells showed a positive correlation (Fig. 4N, P). These findings indicate the novel exposure is a noninvasive approach to modulate the memory extinction with a LC-DG activation potential.

In addition, both optogenetic LC-DG activation and novel environment exposure induced significant increases in calcium signals in recall-tagged DG cells (Supplementary Fig. S7), suggesting that the activity level of recall-induced cells in the DG is important for memory updating during a recall-extinction procedure. Nevertheless, to more clearly map the dynamic process of memory engrams regulated by the LC-DG circuit, more observations and analyses are needed at the cell and synapse levels using refined neuronal recording techniques [55].

Recall-induced DG cells carries memorized information

A sparse colocalization of acquisition- and recall-tagged cell ensembles (Supplementary Fig. S8) implies that a new ensemble of DG cells, different from that engages in the memory acquisition, has been recruited to recall process. According to above findings, the active and plastic state of this cell ensemble contributes to the following memory updating. Nevertheless, it was unknown whether this cell ensemble induced by recall mainly represented memorized (aversive) information or new learning (non-aversive) messages.

To answer this, we bilaterally injected AAV(2/9)-Syn-DIO-ChR2-EGFP and implanted light fibers into the DG of FosTRAP2 mice. Protocols are shown in Fig. 5A (acquisition-induced ChR expression in DG cells) and Fig. 5D (recall-induced ChR expression in DG cells). Twenty-four hours after the last morphine administration, naloxone-paired training in context A was performed. Tamoxifen was injected respectively before the naloxone-paired conditioning (acquisition-induced ChR expression in DG cells) and before the conditioned-context exposure (recall-induced ChR expression in DG cells). After 10 days, real-time CPA was conducted in context B, which was distinct from context A. Results showed that the acquisition-tagged (Fig. 5B) and recall-tagged (Fig. 5E) mice displayed lower bias toward a laser-paired chamber (Supplementary Videos S1, S2), indicating that activating DG cells induced by either the acquisition or the recall contributes to establishing CPA behavior. We further excluded the effects of the blue light flash in the chamber on the behavior with a light beam blockade (Fig. 5C, F; Supplementary Videos S3, S4) and with the substitution of the control virus AAV(2/9)-Syn-DIO-EGFP (Supplementary Fig. S9). These findings support that the DG cell ensemble induced by the recall process dominantly represents memorized information but not new learning message.

Fig. 5: Both of acquisition- and recall-tagged cell ensembles carried aversive memory information.
figure 5

A, D. Procedure for the real-time CPA process by optogenetically stimulating acquisition-tagged cells (A) or recall-tagged cells (D) in FosTRAP2 mice. Morphine administration and naloxone-precipitated aversion acquisition and recall were conducted in context A. Laser-paired real-time CPA training and laser (blocked)-paired real-time CPA training were performed in contexts (B, C), respectively, which were two-chamber boxes that contained differently textured floors and different visual cues on the walls. B, E. The duration in laser-paired chamber significantly decreased with training for both of acquisition-tagged (paired t-test; t5 = 2.841, p = 0.0362) (B) and recall-tagged (paired t-test; t5 = 4.816, p = 0.0057) (E) ChR2 expression (n = 6, right panel). Representative motion traces are shown for the pre-test and laser-paired training (left panel). C, F No effect was found of the duration in laser-paired chamber with laser (blocked)-paired training in either of acquisition-tagged (paired t-test; t5 = 0.4330, p = 0.6830) (C) or recall-tagged (paired t-test, t5 = 0.0890, p = 0.9326) (F) ChR2 expression (n = 6, right panel). Representative motion traces are shown for the pre-test and laser-paired training (left panel). Data are expressed as means ± s.e.m. *, p < 0.05; **, p < 0.01.

Full size image

Discussion

As a potential technique for targeting the reconsolidation of maladaptive memories with a behavioral intervention, extinction training within the reconsolidation window has attracted considerable research interest [12, 13, 18]. However, it is still unclear why the recall process plays a necessary role in the recall-extinction procedure and whether the neural circuit that engages in memory recall functions as an interventional target for promoting memory modulation. Via a series of double labeling and calcium imaging experiments combined with selective neural manipulation, we revealed the importance of LC inputs in the DG for memory recall; we also identified that the population of memory recall- but not acquisition-associated DG engram cells functionally responded to extinction efficacy, which was improved by the LC-DG activation.

Dynamic profiles of memory engram in DG

The DG is commonly thought to be critical during hippocampal memory acquisition, but its contribution to memory recall has been widely debated. Earlier publications showed that DG suppression failed to block the expression of conditioned context fear [21, 22], whereas a recent study implied that the DG had certain contributions to the recall process of fear memory [23]. In this study, DG suppression impaired the ability of mice to distinguish between environments similar to the shock-linked context [23]. The finding supports the present results that mice with reduced DG activity induced by chemogenetic suppression of the LC-DG circuit would not discriminate between saline- and naloxone (withdrawal)-paired contexts (Fig. 2). Together with the previous studies [38, 56], our findings (Fig. 1) suggest that the function of the DG rather than CA1 region in memory recall could be revealed in an animal model with discrimination needs, such as place-paired conditioning preference/aversion or pattern separation.

Beyond the consideration for behavioral models, inconsistent findings regarding the role of the DG in memory recall may be attributed to distinct functional ensembles of cells in the DG. The cell ensembles which are associated with the memory process may certainly represent the memory engram [57]. The DG has been demonstrated as one of the main areas in which the ensemble of engram cells is located [58]. By the activity-dependent neural tagging approach, studies marked the activated cells during memory acquisition or expression, which were identified as the engram cells [38], and clarified their functions in memory generation and storage [59]. However, an interesting phenomenon is currently attracting attention—specifically, the proportion of the overlap between acquisition- and recall-tagged cells in the DG is in the range of 2–10% [59,60,61]. Although the values fluctuate as there may be different intervals between the tagged event and the animal sacrifice, it generally seems that distinct ensembles of DG cells are in charge of memory acquisition and recall. We demonstrated and extended the indication with the data that CPA expression could be driven by optogenetic stimulation of DG neurons, which had been activated during the aversive memory acquisition or recall process (Fig. 5). These findings suggest that a small proportion of learning-tagged DG neurons may stably encode the features of the environment, whereas another subset of cells undergoes memory-dependent reorganization [62]. In addition, DG activity during the learning or early stages of consolidation is necessary to establish the ensemble of the DG neurons that supports memory recall [27, 63]. Thus, identification of the cell types and interactions of different types of cells in the DG during different memory stages in future research may contribute to revealing neuronal mechanisms underlying the construction and updating of memorized information.

Memory extinction is one of the important patterns of memory update [64]. There are two different viewpoints that explain how extinction occurs. One idea is that extinction is a process of new memory acquisition (extinction memory), where the new memory competes with the original memory [65]; another supports that the original memory is recoded during the extinction process, where engram cells related to the acquisition or recall process reactivate in the extinction stage [28, 66]. The coexistence of these two mechanisms has been accepted by more and more studies [67, 68]. We obtained some interesting findings that the reactivation of neurons tagged during the later memory recall rather than the acquisition stage has a greater influence on memory extinction (Fig. 3), indicating that the extinction process requires the recall of hippocampally stored information [28] and supporting the previous implication that initially acquired memory and extinction memory belong to two types of distinct engram, at least in the DG [69]. Nevertheless, a close relationship between the acquisition-activated cell population and memory extinction has also been implicated in the amygdala area [66, 70]. These neural region-dependent findings related to memory information flow indicate the complexity in the memory network.

Modulatory role of LC-DG circuit for memory updating

The efficacy of the recall-extinction procedure in memory modulation is good phenomenological evidence for a close neuronal relationship between recall and extinction [10, 71, 72]. During the time window after recall, in which the memory is labile, it is easy to change or remove [73]. The definition of memory reconsolidation was proposed two decades ago [13], and since then, many preclinical and clinical studies have demonstrated the value of the reconsolidation window for memory modulation; moreover, several behavioral intervention strategies have emerged, including the recall-extinction procedure [9, 10, 74]. Unfortunately, the effect of this behavioral procedure does not always seem replicable [15, 16]. Thus, a better understanding of the mechanisms underlying the recall-extinction procedure may contribute to stabilizing its effectiveness. Based on a recent study that shows the importance of the population size of reactivated engram DG cells in the extinction process for fear memory reduction [28], we extensively reveal a neural circuit that enables facilitating the reactivation of DG engram cells during the recall-extinction procedure (Fig. 3). The recall process is easily influenced by environmental factors such as cue availability and cue competition [75], our data will contribute to promoting the recall-extinction procedure and help enhance the likelihood of its success in memory modulation. The DG is a promising target for neural manipulations for this aim.

Previous studies reveal that direct activation or inhibition of the DG region induced little effect on memory recall [23], changing the microenvironment around the DG engram cells through endogenous or exogenous approaches may be a new perspective [31, 38, 76, 77]. Following the “NA hot-pot” theory raised by Mather et al. [78], which posits the role of the LC as an NA brain source [34,35,36] with an anatomic and functional relationship with the hippocampus [79,80,81,82], we supposed that the LC may be a potential upstream neural region to the hippocampus involved in memory modulation. Via a series of observation, gain-of-function and loss-of-function approaches, we have extensively revealed that LC-DG activation may provide a preheating effect on the DG engram cells and improve the update of initial memory (Figs. 3, 4). In the present study, LC-DG does not appear to play a role in anxiety or somatic signs during opioid withdrawal as measured by rearing behavior. This seems consistent with the previous report that the lesion of NA neurons of the LC did not alter naloxone-precipitated morphine withdrawal [48]. Nevertheless, the proposed involvement of the LC-DG pathway in aversive memories such as withdrawal or fear conditioning is likely related to the well-known role of the LC in attention and arousal [83, 84], which may be associated with stress or novel exposure [54, 83,84,85]. Thus, the LG-DG pathway may be a neural base for influencing, promoting or permitting cognitive activity through modulating behavioral states [85]. For example, activation of the LC is associated with generating novel hippocampal sequences (e.g., remapping) [86] and promoting new learning such as extinction [87]. Collectively, our findings highlight a point that inhibiting the LC-DG circuit can reduce the retrieval of aversive memory but to modify at the engram level, this requires the opposite—activation of the circuit.

This study still has some limitations. Besides aversive withdrawal memory, the DG neurons receiving neural signaling from the LC may engage in other forms of extinction learning, such as fear memory [88] and reward memory [89]. These suggest that the LC-DG circuit is functionally basic for memory modulation. Moreover, both of NA and dopamine neurotransmitters originating from the LC neurons may play important role in this cognitive process [88,89,90]. Further investigations at the level of the transmitter and receptors regarding the function of the LC-DG circuit in the recall-extinction procedure would provide insight for drug development. In addition, Fos positivity is a good marker for time-averaged neuronal activity [91,92,93], but it also means that poorer temporal resolution cannot address whether the timing of the activation of the cells within the ensemble may play an important role. Thus, a well-designed behavioral program and in vivo imaging methods with high temporal and time resolution need to be utilized in further studies [94].

In summary, our study reveals that LC inputs in the DG play the role of modulation for memory recall and that manipulation of the LC-DG circuit promotes the recall-extinction with the underlying enhancement in the population size of reactivated recall-induced DG cells. These findings could be instrumental in understanding the neural mechanisms underlying the behavioral efficacy of the recall-extinction procedure and provide a potential intervention route for refining some pathological memories.