Learning Fast and Slow: Increased cortical plasticity leads to memory interference and enhanced hippocampal-cortical interactions

Our brain is continuously challenged by daily experiences. Thus, how to avoid systematic erasing of previously encoded memories? While it has been proposed that a dual-learning system with “slow” learning in the cortex and “fast” learning in the hippocampus could protect previous knowledge from interference, this has never been observed in the living organism. Here, we report that increasing plasticity via the viral-induced overexpression of RGS14414 in the prelimbic cortex leads to better one-trial memory, but that this comes at the price of increased interference in semantic-like memory. Indeed, electrophysiological recordings showed that this manipulation also resulted in shorter NonREM-sleep bouts, smaller delta-waves and decreased neuronal firing rates. In contrast, hippocampal-cortical interactions in form of theta coherence during wake and REM-sleep as well as oscillatory coupling during NonREM-sleep were enhanced. Thus, we provide the first experimental evidence for the long-standing and unproven fundamental idea that high thresholds for plasticity in the cortex protects preexisting memories and modulating these thresholds affects both memory encoding and consolidation mechanisms.


Introduction 33
Since patient H.M. 1 we know that memories are supported in the brain by a dual- 34 learning system, but why this is the case remains unclear. Initially memories are stored 35 in the hippocampus via synaptic changes in this more plastic brain area, known as the 36 "fast learner" 2 . Later during sleep these hippocampal representations support 37 reactivations of recent memories in the neocortex, the "slow learner" in the brain. 38 Neocortical synapses are less plastic and therefore are thought to change only a little on 39 each reinstatement. Therefore, remote memory is based on over time accumulated 40 neocortical changes. Computational models testing why we have a dual-learning system 41 have proposed that the neocortex learns slowly to discover the structure in ensembles of 42 experiences 2-4 . Further, the hippocampus would then still permit rapid learning of new 43 items without disrupting this structure and therefore the dual system would protect our 44 memories from interference, when new memories would overwrite existing ones without 45 the dual system. Although these theories provides remarkable insights about learning and 46 knowledge extraction, they remains computational models with -until now -no direct 47 experimental support, due to the lack of a valid behavioral paradigms that enable 48 examining structured knowledge extraction in rodents as well interference effects. 49 To test if naturally restricted plasticity in the neocortex protects from memory 50 interference, we artificially increased plasticity in the prelimbic cortex via the 51 overexpression of an established plasticity-enhancer called regulator of G protein 52 signaling 14 of 414 amino acids (RGS14414) 5,6 . The overexpression of RGS14414 is 53 known to lead to increased BNDF and dendritic branching in the targeted area 5,6 and 54 thereby increase plasticity locally. This increased local plasticity makes memories, that 55 usually would not be retained, last longer and can rescue memory-deficits accompanying 56 aging or diseases 7-9 . However, until now, the prefrontal cortex had not been targeted and 57 it remained unknown how increasing plasticity would affect previously acquired 58 knowledge. We combined this plasticity manipulation in the prefrontal cortex with a novel 59 behavioral task -the Object Space task 10 -that allows the testing of semantic-like as 60 well as simple memories in rodents. 61 We show that increased cortical plasticity leads to better one-trial memory 62 performance. However, we observed that such enhanced fast learning is associated with 63 impaired semantic-like, cumulative memories. In alignment with these findings, 64 pharmacological experiments confirmed that these results were an outcome of local

76
Increasing cortical plasticity leads to more memory interference 77 Prelimbic plasticity was increased by the overexpression of RGS14414 5-9 (Fig. 1A, 78 B) and initially two behavioral experiments were performed using the Object Space task 79 10 . In this task the condition Overlapping tests for the extraction of an underlying structure 80 across five training trials, while the Stable condition tests for the simple memory of the 81 last experience (Fig. 1C). In both experiments, we examined the effect of an interference 82 trial 24h after initial training, with object configurations violating previously trained rules, 83 on a test trial 48h later (Fig. 1D). The design was such that in the Overlapping condition Overlapping condition emphasize that in controls cumulative memory expression was 92 protected from interference, but after increasing plasticity in the cortex interference effects 93 were observed. Further, the simple memory in the Stable condition did not last until test 94 in controls, however after increasing plasticity the memory lasted longer. 95 To show that these effects were a result of the interference, we conducted three 96 control experiments. Firstly, we performed a behavioral control, in which animals did not 97 experience the interference trial. Secondly, we performed a pharmacological control, in 98 which animals did experience the interference, but any subsequent plasticity-related 99 changes were inhibited in the cortex via the infusion of anisomycin, a protein synthesis 100 inhibitor. In these experiments RGS14-overexpressing animals showed discrimination 101 indices comparable to controls emphasizing the determinant role of the interference trial 102 in producing opposite outcomes. Thirdly, we added an additional control, in which animals 103 did not receive pretraining on the Object Space Task conditions, but instead were only 104 exposed to an object configuration for one-trial. When tested 48h later, increased cortical 105 plasticity led to an enhanced one-shot memory performance (Fig. 1E, p<0.05).

106
Together, the behavioral and pharmacological results show that increasing cortical 107 plasticity with RGS14-overexpression caused larger interference effects in a semantic-108 like memory task. Increased interference in this case was due to better memory for one-109 trial experiences and local changes in the prelimbic cortex. Thus, our experimental results 110 verify for the first time the hypothesis that lower cortical plasticity is critical to protect 111 previous knowledge from interference effects.

112
Memory interference is due to a higher learning rate 113 To characterize the build-up of a memory trace and its expression in the Object 114 Space Task, we previously developed 10 a computational model that progressively learns 115 place-object associations and makes decisions about which proportion of time to spend 116 exploring each object in order to minimize uncertainty about these place-object 117 associations. The model employs two parameters: a learning rate α, which determines 118 the balance between recent and remote memories, and a parameter β, which determines 119 the balance between neophilic (preference for more novel object location) and neophobic 120 (aversion for more novel object location) exploratory behaviors. Here, we fitted our model 121 on the behavioral data-set to find for each individual subject the values of the model 122 parameter set α and β that best fit the data. There was no difference in memory 123 expression (β). However, RGS14-overexpressing animals had systematically higher 124 learning rate (α) values (Fig. 1F, p<0.05). This indicates that exploration behavior in 125 RGS14-overexpressing animals was driven more by recent than remote memories in 126 contrast to controls.

127
Thus, the modelling results show that increasing cortical plasticity with RGS14-128 overexpression caused larger interference effects in a semantic-like memory task due to 129 a higher learning rate.   Sleep is supposedly the price the brain pays for plasticity 11,12 . The idea is that during 152 a waking episode, learning statistical regularities about the current environment requires 153 strengthening connections throughout the brain. This increases cellular needs for energy 154 and supplies, decreases signal-to-noise ratios, and saturates learning. Therefore, 155 subsequently during sleep, previous waking activity would lead locally to larger delta-156 waves (1-4Hz) and spontaneous activity during these oscillations in NonREM sleep 157 should renormalize net synaptic strength and restore cellular homeostasis 13 .

158
To test this, after viral-injection rats were implanted with hyperdrives containing 16 159 tetrodes targeting the hippocampus and prelimbic cortex (Fig. 2). We recorded neural 160 activity during training as well as sleep in the Object Space task (OS) and compared this 161 to a home cage control (HC). Surprisingly, RGS14-overexpressing animals showed less 162 NonREM sleep (p<0.05, Fig. 2C), which can be attributed to shorter bout lengths 163 (p<0.0001, Fig. 2D). These animals also presented with smaller amplitude delta-waves 164 (p<0.0001). In controls, we did observe that delta-waves occurred more after learning  In sum, in controls we could confirm the proposition that learning and therefore 170 plastic changes lead to a homeostatic response with increases in delta wave activity.

192
To summarize, RGS-overexpressing animals had more prelimbic neurons with 193 slower firing rates. These results provide the first causal evidence that increasing synaptic 194 plasticity shifts the neural firing towards the slow firing end of the neural firing spectrum. 195 Furthermore, because it is the faster-firing neurons that dominate upstate spiking activity 196 and therefore delta amplitude, the slowing of firing rates in the more plastic neurons is 197 most likely the cause of the smaller delta waves seen in these animals.  Increased hippocampal-cortical connectivity during wake and sleep 217 Interactions between the hippocampus and cortex are critical during encoding as 218 well as consolidation of memories 15 (Fig. 3A). During wake as well as REM sleep these 219 interactions take place in the theta domain and can be measured in theta coherence 16 .

220
In NonREM sleep, they can be captured in the coupling of cortical delta and spindle 221 oscillations with hippocampal ripples 17 . Different types of interactions between these 222 three oscillations have been reported; interactions between two oscillations such as delta 223 followed by spindle 18 , delta followed by ripple 19 , ripple followed by delta 20 , and spindles 224 with a ripple in their troughs 21 , but also three-oscillation interactions such as delta 225 followed by spindle with a ripple in the trough 22 , delta followed by ripple then spindle 17 , 226 ripple followed by delta and then spindle 20 . Interestingly, RGS14-overexpressing animals 227 presented with higher hippocampal-cortical theta coherence during both task and REM Increasing cortical plasticity leads to changes in hippocampal ripples 252 Next, we focussed on the hippocampal ripple oscillation in NonREM sleep, which is 253 linked to memory reactivation, and therefore is suggested to be a key player in the 254 memory consolidation process 14,23,24 . Increased cortical plasticity led to more, larger and 255 slower ripples (Fig.4A all p<0.0001). Further, learning in comparison to home cage led to 256 a decrease in ripple amplitude and increase in frequency in both animals' groups (each 257 p<0.0001). In RGS14 ripples were less phase-locked to the slow oscillation (p<0.0001) 258 and cortical delta power around ripples was decreased in comparison to controls (Fig.4B).  interactions, the response of cortical neurons to ripples remained lower than in controls.

316
The current prevalent view on hippocampal ripple oscillations is that they are essential 317 for memory consolidation. Furthermore, it is believed that they are generated mainly as  These and the other 32 behavioural animals had a 2-week surgery recovery and then went on to do the 342 habituation as well as training in the Object Space Task (all conditions counterbalanced within animal). The 343 8 electrophysiology animals (4 vehicle, 4 RGS) received a second surgery three weeks after the first one, 344 for hyperdrive implantation. During 2-3-week surgery recovery, tetrodes were slowly lowered to target area 345 before the animals also had habituation and training in the Object Space Task. 346 All data will be available on the Donders Repository. 347 348

Animals 349
Three-month-old male Lister Hooded rats weighing between 300-350 g at the experiment start 350 (Charles Rivers, Germany) were used in this study. Rats were pair-housed in conventional eurostandard during the light period (between 9:00-18:00). 362 All animal procedures were approved by the Central Commissie Dierproeven (CCD) and conducted 363 according to the Experiments on Animals Act (protocol codes, 2016-014-020 and 2016-014-022). 364 365 366 The RGS14414-and vehicle-lentivirus solutions (1.72 x 10 7 CFU/ml and 2.75 x 10 6 CFU/ml 367 respectively) were prepared and provided by Dr. Zafaruddin Khan at the University of Malaga (Malaga, 368

372
The animal's procedures related to the non-replicant lentiviral solution were approved and carried 373 out in compliance with institutional regulation.

375
Tetrode hyperdrive 376 A customized lightweight tetrode micro-drive was manufactured to implant 10 and 6 movable tetrodes 377 in the prelimbic cortex and hippocampus (HPC), respectively 30-32 . Two separate bundles of #33 polyimide 378 tubes (Professional Plastics,) were prepared: one of 2 columns x 5 rows for prelimbic cortex and 3 X 3 for 379 HPC. The bundles were fixed first to the customized 3D printed cannula and then into the customized 3D 380 printed body drive. The 3D printed cannula was designed according to the Rat Brain Atlas in Stereotaxic 381 Coordinates 33 for the correct placement of the bundles in the areas of interest. Inner tubes (#38 Polyimide 382 tubes; Professional Plastics) were placed inside the outer tubes and glued to the shuttle, which moves 383 through the body spokes thanks to an inox steel screw and a spring CBM011C 08E (Lee spring, Germany).

384
A total of 16 tetrodes were built, twisting four 10 cm polyimide-insulated 12 μm Nickel-Chrome wires (80 385 turns forward and 40 turns reverse) (Kanthal Precision, Florida) and fused by heat. Tetrodes were loaded 386 in the inner tubes, and their free ends were connected to a customized 64 channels, 24 mm round electrode 387 interface board (EIB) using gold pins (Neuralynx). Previously, 2 NPD dual row 32 contact connectors 388 (Omnetics) had been attached to the EIB. The tetrode tips were cut using fine sharp scissors (maximum 389 length 3.5 mm and 3 mm for prelimbic cortex and HPC, respectively) and fixed to the inner tubes in the 390 upper part. Tetrode tips were clean in distilled water and gold-plated (gold solution, Neuroalynx) using 391 NanoZ software to lower their impedance to 100-200 kΩ and improve the signal-to-noise ratio. The tetrode 392 tips were hidden at the same level as the bundle. The whole drive was covered with aluminum foil connected 393 to the ground to reduce the electrostatic interference during the recordings. The bottom of the micro-drive 394 was deepened in 70 % ethanol for 12 hours before brain implantation.

396
Stereotaxic surgeries 397 Lentivirus injection 398 Lentiviral solutions were infused in the prelimbic cortex using stereotaxic surgery under biosafety 399 level 2 conditions. The coordinates of the prelimbic cortex injection site were +3.2 mm AP, +/-0.8 mm ML 400 from Bregma, and -2.5 mm DV from dura mater, according to The Rat Brain Atlas from Paxinos and Watson 401 33 . The procedure was carried out under isoflurane inhaled anesthesia. Unconsciousness was induced at 5 402 % isoflurane + 1 l/min O2 and maintained at 1.5-2 % isoflurane +1 l/min O2. A 0.8 mm diameter craniotomy 403 was drilled above the target area in each hemisphere. The DV dura mater coordinate was measured before 404 performing the durotomy.

405
A 30 G dental carpule connected to a 10 μl Hamilton and an infusion pump (Micro-pump, WPI) was slowly 406 inserted into the brain target area (0.2 mm/min). A total volume of 2 μl of the lentiviral solution was infused 407 at 200 nl/min. After 5 minutes of diffusion, the needle was removed, and the incision was sutured. 408 For the anisomycin infusions, 32 animals were implanted with a 26 G bilateral guide cannula (3 mm 409 length and 1.6 mm inter-cannula distance; Plastic1 Technology, USA) above prelimbic cortex (AP +3.2 mm, 410 ML +/-0.8 mm from Bregma and DV -0.5 mm from dura mater 33 ). Three supporting screws 1 mm X 3 mm 411 were driven approx. 0.9-1 mm into the skull around the cannula. The bilateral cannula was slowly inserted 412 into the brain target area (0.2 mm/min) after dura matter was removed. The whole structure was attached 413 to the previously scratched skull by Metabond (Sun Medical, Japan) and simplex rapid dental cement 414 (Kemdent, UK). The lentivirus infusions took place similarly as explained above, but using a 30 G bilateral 415 internal cannula with a 2 mm projection length from the guide cannula (Plastic1 Technology, USA). The 416 final DV coordinate was -2.5 mm from the dura mater. After 5 minutes of diffusion, the internal cannula was 417 removed, the guide cannula was protected with a bilateral dummy cannula without projection and its cap. 418 Temperature, oxygen saturation, and blood pressure were monitored during the whole surgical 419 procedure. Some eye cream (Opthosam) was applied to protect the corneas during the intervention. At the 420 start and end of the surgery, 2 ml of 0.9% NaCl physiological serum was administered subcutaneously. As 421 analgesia, animals were administered 0.07 mg/ml carprofen in their water bottles two days before and three 422 days after surgery. Immediately before surgery, 5 mg/kg carprofen was sc injected. In addition, a mix of 4 423 mg/kg lidocaine and 1 mg/kg bupivacaine in a 0.9% NaCl physiological serum was administered sc locally 424 in the head. 425 After the viral injection, animals were housed individually in rat IVC cages for 14 days. Their weights 426 and status were monitored daily for the correct recovery of animals. Then, rats were pair housed with their 427 previous cagemate and moved to conventional housing.

429
Tetrode hyper-drive implantation 430 Twenty-one days after viral infusion, a second stereotaxic surgery took place for tetrode micro-drive 431 implantation in 8 animals. The procedure was similar as described above. In this intervention additionally, 432 a prophylactic 10 mg/kg sc injection of Baytril antibiotic was administered at the beginning of the surgery. 433 Two craniotomies (2x1 mm and 1x1 mm for prelimbic cortex and HPC, respectively) were drilled above the 434 target areas on the right hemisphere. The coordinates for the upper left corner of each craniotomy were: 435 AP +4.5 mm and ML -0.5 (prelimbic cortex) and AP -3.8 mm and ML -2 mm (HPC) from Bregma 33 . A 436 ground screw (M1x3) was placed on the left hemisphere in the cerebellum (AP -11 mm, ML +2 mm from 437 Bregma). In addition, six M1x3 mm supporting screws were driven and bound to the skull using Super-bond 438 C&B dental cement (Sun Medical, Japan). Carefully, the durotomies were performed, and the brain's 439 surface was exposed. Subsequently, the micro-drive was positioned on the brain's surface, and attached 440 to the skull and the screws by simplex rapid dental cement (Kemdent, UK). Then, tetrodes were slowly 441 screw-driven into the prelimbic area in prelimbic cortex (3 mm DV from brain surface) and the cortical layers 442 above the HPC (1.5 mm DV from brain surface). The dorsal hippocampal CA1 pyramidal layer was reached 443 progressively in the subsequent days. 444 445

Pharmacological infusion 446
Anisomycin (ANI) powder (Merck, Germany) was solved in 1 M HCl in 0.9% NaCl physiological 447 serum, and the pH was adjusted with 10 μl of 5 M NaOH. Aliquots were prepared and stored at -20ºC until 448 the moment of use. Immediately after the 24 h test in the object space task, animals were infused with 3 449 µl/hemisphere of a 25.6 µg/µl ANI solution or the solvent as control at 300 nl/min. An infusion pump and 450 two 10 μl Hamilton syringes connected through a PE10 tube (Plastic1;) to a customized 30 G bilateral 451 internal cannula with a 2 mm projection (Plastic1;) was used for the infusion. The internal cannula was 452 carefully removed after 3 minutes of diffusion time, and the dummy and cap were placed back. All the 453 animals from Experiment 2 received the infusion of both ANI and vehicle across different weeks for each 454 experimental paradigm.

456
Object Space Task 457 The Object Space Task (OST) is a newly developed behavioral paradigm to study simple and 458 semantic-like memories in rodents 34 . The task is based on the tendency of rodents to explore novel object-459 location in an open field across multiple trials. In these experiments, the OST took place as described 460 previously 34 at least 21 days after the viral infusion when the effect of the RGS14414 protein is observed 461 28,29 . Briefly, animals were handled for 5 consecutive days before and after surgery recovery. Then, rats 462 were The OST consists of two phases: a training phase of 5 training trials in which animals are exposed 467 to two identical objects (different across trials) for 5 minutes (45-55 min intertrial time); and a 468 interference/test phase consisting of a single 10 min' trial performed 24 h and 72 h after training. In the 469 stable condition, both object locations were fixed during the training trials, and one object location was 470 moved during the interference and tests sessions. In the overlapping condition, one object location was 471 fixed, and the other one moved across training trials. In the interference and test sessions, the same object-472 location pattern from the last training trial was repeated. 473 The open field was a wooden square 75 x 75 x 60 cm. For the task, but not for the habituation, we Rats involved in the electrophysiological recordings also performed two experimental control 493 conditions: homecage and random. The random condition was carried out as described previously 34 , so 494 there was a lack of repetitive object location patterns across different trials. In the homecage, the animal 495 was recorded for 7 h and 10 min in the sleep box (a whole training session recording), and the experimenter 496 kept the rat awake for the equivalent trial times.

498
In vivo electrophysiology recordings 499 In vivo freely moving extracellular recordings were executed during the OST and the resting periods. 500 One session per experimental condition (homecage, stable, overlapping, and random) was carried out per 501 animal. The local field potential (LFP) and single-unit activity detected by the 64 channels were amplified, 502 filtered, and digitized through two 32 channels chip amplifier headstages (InstanTechnology) connected 503 through the Intan cables and a commutator into the Open Ephys acquisition box. The signal was visualized 504 using the open-source Open Ephys GUI (sample rate 30 kHz). In addition, the headstage contains an 505 accelerometer to record the movement of the animals. 506 507 508 After all the recording sessions, the tetrode-implanted animals received brain electrolytic lesions 48 509 h before the transcardial perfusion to identify the electrode tips placement. Thus, a current of 8 µA for 10 s 510 was applied in two wires per tetrode using the stimulator with the animal under isoflurane inhaled 511 anesthesia. 512 513 Histology 514 Brain processing 515 After data collection, animals had overdoses with 150 mg/kg sodium pentobarbital ip. Rats were 516

Tetrode electrolytic lesions
transcardially perfused first with 80 ml of 0.1 M phosphate-buffered saline pH 7.4 (PBS) and then with 250 517 ml of 4% (w/v) paraformaldehyde in 0.1 M phosphate-buffered pH 7.4 (PFA). After brain extraction, it was 518 immersed in PFA overnight at 4ºC. Then, the brains were rinsed in PBS 3 times for 10 min and 519 cryoprotected by deepening in 20 ml of 30% (w/v) sucrose, 0.02% (w/v) NaN3 in PBS. Once brains sank 520 (after 2-3 days approx), they were frozen in dry ice and stored at -80ºC. Finally, 30 or 50 µm coronal 521 sections of target areas were obtained using the cryostat (SLEE medical, Germany), collected in 48-well 522 plates containing 0.02 % (w/v) NaN3 PBS and stored at 4ºC.

524
Immunohistochemistry 525 The overexpression of rgs14414 was checked by free-floating fluorescence immunohistochemistry. 526 First, the target sections were selected, rinsed in PBS, and incubated overnight at 4°C with the rabbit 527 polyclonal anti-RGS14 antibody (Novus biological, NBP1-31174; dilution 1:500). Then, the Alexa fluor® 528 488-conjugated goat anti-rabbit IgG (Life Technologies, A11008; dilution 1:1000) at room temperature for 529 2.5 h. Some drops of water-soluble mounting medium containing DAPI (Abcam, ab104139) were applied 530 for 5 min before placing the coverslip. Leica fluorescense microscope (Leica DM IRE2) and camera were 531 used to observe and photograph the samples.

533
Nissl staining 534 Coronal sections were AP sequentially mounted on gelatin-coated slides and incubated at 37 ºC 535 overnight. Slices were hydrated first in 0.1 M PBS pH 7.4 and then in Milli Q water for 20 min each. Next, 536 brain sections were stained in 0.7 %(w/v) acetate cresyl violet for 20 min and dehydrated in an increasing 537 ethanol gradient (water for 3 min, 70% ethanol for 20 s, 96% ethanol+acetic acid for 45 s, 100% ethanol for 538 5 min). Lastly, the tissue was immersed in xylene for 15 min, and the coverslip was placed using some 539 DePeX mounting medium drops. Cannula placement, infusion traces, or/and tetrode lesions were observed 540 and photographed under a light field microscope (Leica DM IRE2) and a camera. Behavioral data analysis 556 Object Space Task. 557 The total exploration time was calculated as the sum of the time spent exploring both object locations. 558 The discrimination index (DI) was computed by subtracting the familiar object location exploration time to 559 the novel object location and dividing it by the total exploration time. A DI > 0 means a preference for the 560 new object location and consequently memory from the previous episode. A DI = 0 shows no preference 561 for either the new object location or the fixed one. DI<0 means a preference for the stable object location. 562 563 The same computational model as in  was used (see article for more detailed 565 methods). In short, the model learns place-object associations and then translates this memory into an 566 exploratory behavior: the objects that were stably found at the same location have a very low uncertainty 567 and are thus either less attractive or more attractive (depending on the individuals) during exploration than 568 objects found at changing locations (high uncertainty in place-object association). The source code of the 569 computational model and model simulation/fitting procedures is available here: 570 https://github.com/MehdiKhamassi/ObjectSpaceExplorationModel 571 The model employs two different parameters: a learning rate α, which determines the speed of 572 memory accumulation; an inverse temperature β, which determines the strength and sign of memory 573 expression during exploratory behavior. 574 A low learning rate α (i.e., close to 0) means that the model will need numerous repetitions of the 575 same observation (i.e., in the Object Space Task, many trials observing the same place-object association) 576 to properly memorize it. In contrast, a high learning rate α (i.e., close to 1) means that the model quickly 577 memorizes new observations at the expense of old observations which are more quickly forgotten. As a 578 consequence, with a low learning rate the exploratory behavior generated by the model will mostly reflect 579 remote memories but not recent ones (semantic-like memory). Conversely, with a high learning rate, 580 exploratory behavior in the model will mostly reflect recent memories but not remote ones (episodic-like 581 memory). 582 Finally, an inverse temperature β close to zero means that the model does not strongly translate 583 memories into object preferences for exploration, thus showing little object preference. In contrast, a high 584 inverse temperature will mean that the model's exploratory behavior is strongly driven by differences in 585 relative uncertainty between place-object associations. A high positive inverse temperature (β > 0) will result 586 in neophilic behavior: the model spends more time exploring objects associated with high uncertainty (i.e., 587 novelty or constantly changing location); a high negative inverse temperature (β < 0) will result in neophobic 588 behavior: the model spends more time exploring objects with low uncertainty (stable/familiar objects). 589 The model was fitted to each mouse's trial-by-trial behavior using a maximum likelihood procedure 590 described in , and similar to state-of-the-art model fitting methods in cognitive 591 neuroscience (Collins & Wilson, 2019). In brief, this model fitting process found the best parameter values 592 for each subject that best explain the relative proportion of time spent exploring each object at each trial. 593 The main operations of the model are summarized in Figure 1F. Local field potential analysis 615 Signal preprocessing 616 For the following analyses, first a single channel was selected per brain area. For prelimbic cortex, the 617 channel with the largest slow oscillations was chosen. For hippocampus, the channel closest to the 618 pyramidal layer, which displayed noticeable ripples was selected. Both channels were originally acquired 619 at a sampling rate of 30 kHz and to avoid working which such a high rate, the channels were filtered with a 620 3 rd order Butterworth lowpass filter at 500 Hz to avoid signal aliasing and then downsampled to 1 kHz.

622
Theta coherence 623 Theta coherence was computed as the magnitude squared coherence using the mscohere function in 624 MATLAB and a custom-written script that collected the downsampled data from different animals and study 625 days. The magnitude squared coherence was calculated as follows: 626 627 where Cxy is the magnitude coherence, Pxy is the cross spectral density of the hippocampal and prelimbic 629 cortex signal, Pxx is the hippocampal signal, and Pyy is the prelimbic cortexsignal. The coherence analysis 630 focused on two periods of interest namely, WAKE and REM periods. Since REM sleeping periods might 631 occur several times over several sleeping cycles, REM periods from pre and posttrial sleep were first 632 extracted and concatenated together before running the analysis. For both Wake and REM periods, the 633 power and cross-power spectra were computed on overlapping time windows of 1 second with 80% overlap. 634 Then, the theta coherence was computed as the average value over the theta frequency range (5-12 Hz) 635 for the Object Space (Stable, Overlapping and Random) and home cage conditions. 636 637

Detection of spindles and delta waves 638
The downsampled prelimbic cortex channel (1kHz) was loaded and using a 3 rd order Butterworth 639 filter the signal was filtered to 9-20Hz for detecting spindles and to 1-6Hz for detecting delta waves. The 640 NonREM bouts were then extracted from the filtered signal and concatenated. The functions FindSpindles 641 and FindDeltaWaves from the Freely Moving Animal (FMA) toolbox http://fmatoolbox.sourceforge.net were 642 modified and used to detect the start, peak and end of spindles and delta waves respectively. The optimal 643 threshold was found for each animal by visually inspecting the detections and modifying the default 644 parameters of the functions when needed. The results were saved as timestamps with respect to the 645 concatenated NonREM signal in seconds. They were then used to find the timestamps with respect to the 646 recorded signal. This process was repeated for pre and post trial sleep periods in study days pertaining to 647 all animals in both treatment groups. 648 649

650
The downsampled channels (1kHz) of the hippocampal pyramidal layer were loaded and the 651 NonREM bouts were extracted. Using a 3 rd order Butterworth bandpass filter, the epochs of HPC signal 652 were filtered to a frequency range of 100-300Hz. A custom MATLAB function was used for detecting the 653 start, peak and end of the ripples by thresholding voltage peaks which lasted a minimum duration of 30 ms 654 above the threshold. The start and end of the ripple were determined as half the value of the selected 655 threshold. The standard deviations of concatenated NonREM bouts were computed individually for 656 presleep and all post trials in a study day. The average of these standard deviations was calculated to find 657 a single detection threshold per study day. An offset of 5 units was added to the threshold to reduce false 658 positives. This was repeated for all study days pertaining to all animals in both treatment groups.

660
Oscillations characteristics 661 The traces of each event detected (ripples, spindles, delta waves) were extracted using the start and 662 end timestamps obtained from the detectors. The traces of the events were filtered in their corresponding 663 detection frequency band. Characteristics such as the amplitude and mean frequency were calculated for 664 these filtered events using built-in and custom MATLAB functions. Namely, the amplitude of the events was 665 calculated by computing the envelope of the filtered trace using a Hilbert transform. The absolute value of 666 the result was taken and its maximum was found. The mean frequency of the filtered traces was computed 667 using the meanfreq function of MATLAB.

669
Detection of oscillation sequences 670 The sequences between ripples, spindles and delta waves were counted in various combinations to 671 study cortico-hippocampal coupling during NonREM sleep as done by Maingret et. al. 20 . The time 672 difference between the peaks of these events was compared to a fixed duration to establish if there was a 673 sequential relationship in the following combinations of oscillations: Delta-Spindle (D-S), Delta-Ripple (D-674 R), Ripple-Delta (R-D), Ripple-Delta-Spindle (R-D-S). For D-S a sequence was considered when the 675 duration between events was between 100-1300 ms, for D-R it was 50-400 ms and for R-D it was 50-250 676 ms. To find R-D-S sequences, the results of R-D and D-S were compared to find delta waves preceded by 677 a ripple and followed by a spindle. The results were saved as counts of each sequence for each post-trial. 678 679

Co-occurrence between ripples and spindles 680
The co-occurrence between ripples and spindles was computed by comparing the start and end 681 timestamps of both events. To consider co-occurrence between a ripple and a spindle, either one of the 682 following conditions had to be fulfilled. 1) A ripple had to start and end within the duration of the spindle. 2) 683 One of the events had to start or end within the duration of the other. Given that more than one ripple can 684 co-occur with the same spindle, we counted separately spindles co-occurring with spindles and spindles 685 co-occurring with ripples.

687
Slow oscillation phase 688 The downsampled prelimbic cortex signal was filtered in the 0.5 to 4 Hz range using a 3 rd order 689 Butterworth bandpass filter and its Hilbert transform was computed to find the phase angle of slow 690 oscillations in a range from 0° to 360°. The peaks of ripples and spindles were then used to find the 691 corresponding slow oscillation phase. This same signal was later used to find the phase during spikes 692 timestamps of cortical neurons. 693 694

Spectral analysis and Granger causality 695
A two-second-long window centered on each ripple peak was extracted from the hippocampus and the 696 prelimbic cortex channels respectively. All ripples across animals and conditions were combined per 697 treatment and their amplitude was computed by finding the maximum of their envelope computed with a 698 Hilbert transform. The median ripple amplitude was calculated and the corresponding two-second-long 699 windows of the 2000 ripples which amplitude was the closest to the median amplitude were included in the 700 following analysis. A notch filter at 50 Hz was applied to the ripple-centered windows using the 701 ft_preprocessing function from the Matlab-based Fieldtrip toolbox 35 . The Short-time Fourier transformation 702 was calculated to detect the changes of spectral power in hippocampus and prelimbic cortex with respect 703 to a time window of ± 1 second around each ripple. This was computed using the ft_freqanalysis function 704 from Fieldtrip with a 100 ms Hanning window and time steps of 10 ms, for a frequency range from 100 to 705 300 Hz with a 2 Hz step for hippocampus and from 0.5 to 20 Hz with a step of 0.5 Hz for prelimbic cortex. 706 The resulting spectrograms were averaged and displayed. To statistically compare spectrograms between 707 treatments, a nonparametric permutation test to correct for multiple comparisons with two-tailed pixel-based 708 statistics was computed using 500 permutations and a p-value of 0.05 36 . 709 To determine the predictive power between brain regions during ripples, the time-frequency Spectral 710 Granger Causality was computed for each directionality 37 . A window with length of 2.2 seconds centered 711 around each ripple peak was extracted for the simultaneous hippocampal and prelimbic cortex signals. The 712 length of this window was chosen to at least capture one cycle of 0.5 Hz activity. A two-second-long time-713 frequency non-parametric Spectral Granger causality was computed by implementing a Short-time Fourier 714 transform with a 500ms Hanning window with 10ms steps using the Fieldtrip functions ft_freqanalysis and 715 ft_connectivityanalysis respectively. To determine statistical differences between granger spectrograms we 716 created randomized trials by taking 400 random ripples per treatment and computing their time-frequency 717 granger causality as described above. The result was stored, and the procedure was repeated 30 times to 718 give a total of 30 randomized trials per treatment. We then used the trials of the rgs14 and control 719 treatments to determine significant statistical difference in each pixel of the time-frequency matrix by 720 of interneurons were 18 for RGS14 and 7 for Vehicle. In the following analyses, only pyramidal neurons 769 were used given the low number of interneurons detected.

771
Firing rate analysis 772 The spikes of each neuron were grouped by the sleep stage during which they fired. The total number 773 of spikes of a neuron during a specific sleep stage were determined by counting all the spikes occurring 774 during the sleep stage in question across pre-trial sleep, post-trial sleep and trials periods of a single day. 775 The cumulative amount of time spent in a specific sleep stage during the day was determined similarly. 776 Using these two values, the firing rate of a neuron during a sleep stage was computed by dividing the total 777 number of spikes during the sleep stage by the cumulative amount of time spent in seconds in the sleep 778 stage. The firing rate during each sleep and wake stage was calculated for all neurons. The firing rates of 779 neurons during the 'Wake' stage in the vehicle control condition were divided into 5 quantiles based on their 780 magnitudes (0-20%, 20-40%, 40-60%, 60-80%, 80-100%), as shown in Figure 2G. The upper-limit values 781 of firing rate differentiating the groups in Vehicle were then used as a threshold to divide the neurons in the 782 RGS treatments in 5 groups as well. The spike timestamps during NonREM sleep were collected across 783 the whole study day for each neuron and the corresponding slow oscillation phases during each spike 784 timestamps were extracted. The slow oscillation phase during spikes was calculated as described above 785 for ripples and spindles.

787
Cortical activity during ripples 788 Using the spike timestamps during NonREM sleep, the cortical pyramidal neuron response to 789 hippocampal ripples was computed in a 2 second window defined around the peak of each ripple. For each 790 ripple the activity of all pyramidal neurons detected during that day was extracted. The spikes timestamps 791 in the 2 second window were normalized to vary from -1 to 1 seconds, where 0 was the ripple peak. After 792 concatenating the normalized timestamps across all ripple-centered windows per neuron, they were binned 793 in 10 ms bins and the number of spikes was determined for each bin. Hence, for each neuron, a [1 x 200] 794 column vector was obtained. This vector was then z-normalized. The final z-scored vector was found by 795 averaging the z-scored vector over all neurons. The data was smoothed twice using the MATLAB smooth 796 function. To visualize the activity of prelimbic cortex pyramidal neurons around the ripple, the firing activity 797 of a neuron was determined from a 50 ms window around the peak of the ripple (-20 to +30 ms) by 798 quantifying the number of spikes. After compiling the firing activity around the ripple peak for each neuron 799 the values were sorted in an ascending order and was visualized using the MATLAB imagesc function.