Chemogenetic stimulation of hippocampal adult-born neurons improves long-term memory accuracy


 Hippocampal adult neurogenesis is involved in many memory processes from learning, to remembering and forgetting. However, whether or not the stimulation of adult neurogenesis can improve memory performance remains unclear. Here, using a chemogenetic approach that combines selective tagging and specific activation of distinct adult-born neuron populations, we demonstrate that this activation can improve remote memory accuracy and strength. These results open up new avenues for remedying memory problems that may arise over time.

reconsolidation 8 and forgetting 9 . These studies suggest that memory deficits observed in memory pathologies could be linked to a disturbance of adult neurogenesis.
Nevertheless, relatively few studies have shown any beneficial effect on memory resulting from the stimulation of adult neurogenesis other than those resulting from the genetic enhancement of the neurogenic pool 9,10 . Because it is known that adult neurogenesis is homeostatically regulated 3 , the addition of new neurons can lead to compensatory mechanisms 11 . Thus, knowledge about the effect of specifically activating a given population of new neurons, already present at the time of learning, on subsequent memory processes is lacking. Here we tested whether chemogenetic stimulation of neuronal activity of new neurons during retrieval could enhance longterm memory retention in two different hippocampal-dependent tasks.

Results
We first wanted to analyze long-term spatial memory in the water maze (WM). Towards this end, 2 months-old adult rats were trained to locate a hidden platform in a fixed location in the water maze (WM) for 6 days. Then two groups of animals were tested respectively 48 hours or 4 weeks later (Fig 1a). The results showed that, at both tests, latency to cross the position of where the platform was during training was not different between groups and not different from the latency on the last training day (Fig 1b). This shows that the spatial information was retained at both delays. However, when we considered the number of annulus crossings or the percentage of time spent in the target zone, results showed that the performances were impaired when rats were tested 4 weeks after training. Both annulus crossings and time spent in zone were higher in the target zone compared to those in the other zones at 48h, whereas no difference was observed when memory was tested at 4 weeks (Fig 1 c,d). This demonstrates that the passage of time disrupted spatial memory accuracy.
To further determine whether stimulation of adult-born neurons could improve memory accuracy, we developed a GFP retrovirus (RV) into which we inserted a DREADD-Gs construct (Gs-GFP-RV)(see Suppl Methods). This tool acts initially by infecting new granular cells at their birth. Several weeks later, when they are fully integrated into the network, these same cells can then be activated upon binding of the synthetic ligand Clozapine-N-Oxide (CNO). To ensure that Gs-GFP-RV injections into the dentate gyrus (DG) had no impact on new neuron survival, rats were injected in the DG with either the Gs-GFP-RV or with a control retrovirus (GFP-RV). At the same time, both groups of animals were injected with bromodeoxyuridine (BrdU) to study cell genesis.
Six weeks later, the animals were sacrificed in order to quantify the number of surviving BrdU-positive cells. No differences were observed between the groups, demonstrating that injection of the Gs-GFP-RV has no impact on BrdU-labeled cell survival ( Fig. 2a-b).
To further confirm that CNO injections specifically activate Gs-GFP-RV cells, rats were injected with Gs-GFP-RV in the left DG and with a control GFP-RV in the right DG. Six weeks later, all animals received an IP injection of CNO (1mg/Kg) and were sacrificed two hours later (Fig. 2c). We determined the activation of the infected cells by analyzing the expression of the immediate early gene Zif268 in the GFP labeled cells. We found that 98.8% of the cells infected with the Gs-GFP-RV were activated by CNO compared to 1.72% of the cells injected with a Ctrl-retrovirus (Fig. 2d). Finally, we performed whole cell recordings of Gs-GFP-RV or control GFP-RV-infected cells (Fig. 2e) and assessed changes in cell excitability after CNO application (Fig. 2f). Local perfusion of CNO quickly and reversibly enhanced Gs-GFP-RV-infected cell activity, which was seen by an increase in both resting potential and action potential firing rate (Fig. 2g). No such effect was seen in GFP-RV-infected cell activity (Fig. 2h).
We then sought to determine whether stimulating adult-born neurons during retrieval could promote memory retention. We have recently shown that adult-born neurons that were immature (1 week-old) or mature (6 weeks-old) at the time of training were both activated by remote memory retrieval. However, only the immature population was necessary for remote memory reconsolidation 8 . We therefore first focused on the population of neurons born one week before training (Fig 3a). We have previously shown that the survival and dendritic development of this latter population is increased by learning 12 and it has been hypothesized that these immature neurons could be primed by experience 14,15 . Recent data showed that this immature population is required for maintaining the memory trace after retrieval 8 . However, whether the stimulation of neurons that were immature at the time of learning is capable of improving memory retrieval is not known. Two month-old rats were injected with either the Gs-GFP-RV or the GFP-RV. One week later, these rats were trained in the water maze (Fig 3a). All rats learned to find the platform (Fig 3b). Four weeks after training, we performed a remote retention test, one hour prior to which all rats received an IP injection of CNO. We found that stimulation of this population enhanced memory retention. Compared to GFP rats, the number of annulus crossings and the time spent in the target zone were higher in the Gs-GFP rats (Fig 3c,d,e). Gs-GFP, but not GFP, rats showed also a clear preference for the target zone compared to the others zones with performances significantly above the chance level. These results demonstrate that stimulation of adult-born neurons that were immature at the time of learning could enhance memory accuracy. To confirm that dentate granular neurons were correctly transduced, the number of GFP positive cells was estimated in the dentate gyrus. This analysis showed that 3156 ± 295 cells were GFP positive and distributed along the DG septo-temporal axis (Fig3f).
We then performed the same experiment, but this time, we targeted the population of adult-born neurons that were mature (6 weeks-old) at the time of training (Fig 4a).
This specific population is known to be activated by spatial learning and retrieval and its pharmacological ablation disrupted spatial memory learning 13,14 . We have also recently shown that chemogenetic silencing of this population impairs recent spatial memory retrieval 8 . Hence, two month-old rats were injected with either the Gs-GFP-RV or the GFP-RV. Six weeks later, they were trained in the water maze. All groups learned to find the platform (Fig 4b). Four weeks after training, we performed a remote retention test, one hour prior to which all rats received an IP injection of CNO (Fig 4a). Stimulation of mature adult-born neurons infected with the Gs-GFP-RV enhanced retention, since the Gs-GFP-RV, but not the GFP-RV, rats made more crossings and spent more time in the target zone compared to those in the other zones (Fig 4c,d,e). The neurons transduced by the Gs-GFP-RV (1303 ± 150 cells) were distributed along the DG septo-temporal axis (Fig 4f).
One could argue that the stimulation of the dentate gyrus, no matter the population targeted, could enhance memory retention. To determine whether the effect observed on memory was specific to the stimulation of adult-born neurons, we targeted dentate granular neurons generated postnatally. To do so, rats were injected with either the Gs-GFP-RV or the GFP-RV, 3 days after birth (P3). Then rats were trained in the WM at 2 months of age. Four weeks later, CNO was injected one hour before a probe test (Fig 5a).
In this case, the stimulation of postnatally-generated granular neurons had no effect on memory retention (Fig 5b-e). We estimated the number of neurons transduced by the Gs-GFP-RV and we found that 7588 ± 1255 cells were GFP positive and distributed along the DG septo-temporal axis (Fig 5f) . These results indicate that the beneficial memory effects of chemogenetic stimulation does not depend on the number of DG cells transduced.
Altogether these data reveal that chemogenetic stimulation of adult-born neurons generated before learning, and not developmentally-born granular neurons, enhances remote memory accuracy which can be lost with the passage of time.
We then investigated whether chemogenetic stimulation of adult-born neurons could enhance long-term memory in another hippocampal-dependent task, i.e, contextual fear conditioning (CFC). Towards this end, two month-old rats were injected with either the Gs-GFP-RV or the GFP-RV. One week later, these rats were trained in contextual fear conditioning (Fig 6a). All rats received two mild foot-shocks over a period of 4 min (Fig   6b). During conditioning, both groups had the same level of freezing (Fig 6c). Five weeks later, when the targeted cells reached the age of 6 weeks-old, we performed a remote retention test, one hour prior to which all rats received an IP injection of CNO. Freezing was measured over 3 min (Fig 6b). The results showed the level of freezing was higher in the Gs-GFP rats compared to that of GFP control animals (Fig 6d). Finally, we estimated the number of neurons transduced by the Gs-GFP-RV and we found that 1799 ± 311 cells were GFP positive and distributed along the DG septo-temporal axis (Fig   6e,f). These data demonstrate that stimulating adult-born neurons that were present at the time of learning increases the strength of remote memory.

Discussion
Here, we show that stimulating adult-born neurons generated before learning can rescue memory accuracy in a navigational task and promote fear response. Our data underline a new role for a population of immature neurons at the time of learning. While this specific population is not activated by learning and is not necessary for recent memory retrieval 8 , its silencing results in disruption of memory maintenance after reactivation. Altogether, these results add to previous findings demonstrating that decreasing adult neurogenesis before training impairs remote but not recent spatial memory 15 and suggest that adult-born neurons are important for consolidation, retrieval and reconsolidation.
Although activation of neurons which are immature at the time of learning leads to the reengagement of the network and eventually to memory retrieval, they do not meet the definition of "engram cells" 16 , because they are not activated by learning 8 . They are nonetheless undergoing structural changes in response to learning 12 . As a result these immature cells become part of the engram network but further experiments targeting the mature cells that did encode the information are needed to understand how immature neurons could, together with the engram cells, be engaged in the engram network.
The stimulation of the population of neurons that were mature at the time learning also induces an increase of spatial memory accuracy. These neurons are activated by spatial learning and memory retrieval 13 and their silencing disrupts recent memory recall 8 . We can speculate that this population was part of the initial engram that encoded the spatial information. However, remote memory stabilization after reactivation does not depend on this population 8 .
The positive effect observed on remote memory seems to be specific to adult-born neurons stimulation. In fact, stimulation of postnatally-generated cells does not improve memory accuracy. It should be noted that more neurons were transduced with the retroviruses in P3 pups than in adult rats, ruling out the hypothesis that the lack of effect observed could be due to a low number of neurons transduced. The results are consistent with previous finding showing that postnatally-generated neurons are not critical for spatial learning 17 , nor for recent 18 or remote retrieval 8 . Together, these data confirm that postnatally-generated neurons are not involved in spatial learning in adult rats.
Finally, using another hippocampal-dependent task, i.e contextual fear conditioning, we found that stimulating the population of new neurons that was immature at the time of learning could enhance contextual fear response. This demonstrates that the beneficial effect of adult-born neurons stimulation is not restricted to spatial memory but may be generalized to hippocampal-dependent memory.
Altogether these findings promote the idea that adult neurogenesis is essential for memory accuracy 19 and could be a valuable tool to compensate for the loss of precision and detail induced by the passage of time. Since the dentate gyrus is known to govern the reactivation of remote memory trace 20 and to diminish generalization of remote fear memories 21 , our results confirm that adult-born dentate neurons are key players in the role of the dentate network.
Finally, promoting memory is a relevant outcome when it comes to aging. During aging, not only the rate but also the responsiveness of adult hippocampal neurogenesis is altered 22,23 . Here, we propose that increasing the activity of adult-born neurons is a promising strategy for prevention or treatment of memory loss.

Plasmids and retroviruses
The Gs DREADD was cloned by PCR using pcDNA5/FRT-HA-rM3D(Gs) (Addgene #45549; 24 ) as a template (See table S2 for PCR primers) and then inserted into the BamHI site of a CAG-IRES-GFP retroviral backbone 25 . The resulting construct CAG-Gs-IRES-GFP was sequencing using specific primers (table S2)

Retroviral injections
Adult rats were anaesthetized with 3% isoflurane and placed in the stereotaxic frame, where they were maintained on 2% isoflurane for the duration of the surgery. Analgesia was provided by a subcutaneous injection of Metacam (1mg/Kg). Retroviruses were stereotaxically injected (2µL per injection site at 0.3µL/min) into the dentate gyrus of adult rats with a microcapillary pipette connected to a micro-syringe pump

Water maze procedures
The apparatus consisted of a circular plastic swimming pool (180 cm diameter, 60 cm height) that was filled with water (20±1°C), rendered opaque by the addition of a white cosmetic adjuvant. Two days before training, the animals were habituated to the pool for one minute. During training, animals were required to locate a submerged platform (16 cm diameter) hidden 1.5 cm under the surface of the water in a fixed location, using the spatial cues available within the room. All rats were trained for four trials per day (90 s with an inter-trial interval of 30 s, with release from one of three starting points selected in a pseudorandom sequence each day) during 6 days. If an animal failed to locate the platform itself, it was placed on the platform by the experimenter at the end of the trial.
The time to reach the platform was recorded with a video camera that was fixed to the ceiling of the room and connected to a computerized tracking system (Videotrack, Viewpoint, Lyon, France) located in an adjacent room.
Four weeks after learning, rats were submitted to a retention test in the water maze.
During the test, rats were put in the water maze for 60 s in the absence of the platform.
Performances were assessed using several parameters: the amount of time spent in and number of entries (annulus crossing) into each zone. Zones were defined as an ideal circle (30 cm diameter) located at the original platform location (Target zone; T) and the three equivalent areas in each of the other quadrants (other zones; O). The grouped heat-maps were created with Ethovision XT 10, Noldus (Nantes, France). The color of a pixel represents the average proportion of a track that is found at that location.

Contextual fear conditioning
The fear conditioning chamber consisted of a squared conditioning chamber (30 cm X 30 cm X 40 cm) of a brightness of 10 lux, containing a stainless steel grid floor (Imetronic, Marcheprime, France). Rats were habituated to handling for 3 days before conditioning. On conditioning day, rats were placed in the chamber for 2 min for habituation. Then they received 2 foot shocks (0.8 mA each, 1sec) 60 sec apart (wait 1) and they were left undisturbed for another 60 sec period (wait 2). Chambers were cleaned with 70% ethanol between each rat. On testing day, rats were placed in the chamber for 3 min. Behavioral data were automatically collected using infrared beams spaced 1 cm apart in the x and y planes, located at the floor of the chambers. Freezing behavior was recorded following the cessation of movement for at least 2 sec.

CNO delivery
The DREADD ligand CNO (Clozapin-n-Oxyde, Enzo Life Sciences, Lyon, France #BML-NS105) was dissolved in a saline solution and delivered via one ip injection of 1mg/kg in rats 60 min before the test. Mosaic pictures were taken using a SP8 confocal system with a 20X multi-immersion lens (digital zoom of 1.2).

Electrophysiological recordings
Infected animals were deeply anesthetized (167mg/Kg ketamine and 16,7mg/Kg xylazine) and sacrificed. Dissected brain was immediately immerged in ice-cold Signals were analyzed offline (Clampfit software, pClamp 10, Molecular devices, UK). For statistical analysis, "Vehicle" data were collected during the last 60 seconds before CNO perfusion, then "CNO" data were collected after 45 seconds of CNO treatment.

Statistical analysis
The data (mean±SEM) were analyzed using the Student t-test (two-tailed) and two ways ANOVA which was followed by the Tukey'comparison test when necessary. All analyses were carried out using the software GraphPad Prisms 6 and 8. Competing Interests: The authors declare no competing interests.

Data availability
All data supporting the findings of this study are provided within the paper and its supplementary information. A source data file is provided with this paper. The CAG-Gs-IRES-GFP retroviral construct is available upon request to the authors after MTA approval. All additional information will be made available upon reasonable request to the authors.  to that of the other zones. Rats that were tested 48h after training spent significantly more time in the target zone compared to the other zones and the amount of time was higher than that of chance level. No difference was observed when animals were tested 4 weeks after training (zone X group interaction F(1,15)=7.145; p=0.017; zone effect F(1,15)=12.60; p=0.0029; group effect F(1,15)=2.762; p=0.1173) (**p<0.01; # p<0.05 compared to chance level). e-Density plot for grouped data: The color level represents the lowest to the highest location frequency in pixels. All data shown are mean ± s.e.m.