Part 2:Epigenetic Regulation Of The Circadian Gene Per1 Contributes To Age-related Changes in Hippocampal Memory

Contact: Audrey Hu audrey.hu@wecistanche.com

Knockdown of Per1 impairs long-term memory in young mice. Next, to test whether upregulation of Per1 in the dorsal hippocampus is necessary for long-term memory formation, we infused siRNA targeting Per1 into the dorsal hippocampus of young mice 48 h before 10-min OLM training. Infusion of Per1 siRNA produced a signiﬁcant reduction of PER1 protein in the dorsal hippocampus, as measured 2 h after the ﬁnal test session (Fig. 4d, Supplementary Fig. 5). This relatively modest knockdown of PER1 protein (~30%) produced severely impaired memory formation for OLM; mice infused with Per1 siRNA showed no signiﬁcant increase in DI between training and testing and dis- played signiﬁcantly less preference for the moving object during testing than control mice (Fig. 4e) with no effect on total object exploration at test (Fig. 4f) and no difference in movement during the pre-training habituation sessions (Supplementary Fig. 8c). This demonstrates, for the ﬁrst time, that local disruption of a core circadian clock gene selectively within the hippocampus can impair memory formation.

Per1 overexpression improves memory in aging mice. Finally, to determine whether overexpression of Per1 in the dorsal hippocampus is sufﬁcient to ameliorate age-related memory impairments, we locally upregulated Per1 using two complementary methods. First, we used a lentivirus expressing wild-type Per1 with a v5 epitope tag (pLVX-v5Per1) (Fig. 5a, Supplementary Fig. 6a). To conﬁrm that this plasmid over-expresses Per1, we transfected HT22 cells with pLVX-v5Per1 or pLVX-EV and measured Per1 mRNA expression. At both 24 and 48 h after transfection, Per1 mRNA was signiﬁcantly increased in cells transfected with pLVX-v5Per1 relative to cells transfected with the control plasmid (Fig. 5b). Transfection of pLVX-v5Per1 also led to a signiﬁcant increase in mRNA for Per2 (a Period clock family member that interacts with Per1) at 24 h (Supplementary Fig. 6b), although this increase was far smaller than the observed increase in Per1 (at 24 h, Per1: 466-fold increase over EV, Per2: 1.6-fold increase over EV). Per1 overexpression did not affect the transcription of the nearest downstream gene Hes7, however (Supplementary Fig. 6c).

To determine whether Per1 overexpression improves memory performance in aging mice, pLVX-v5Per1 was infused into the dorsal hippocampus of 18-m.o. mice 2 weeks before the behavior. pLVX-v5Per1 mice showed signiﬁcantly improved memory for OLM at test relative to pLVX-EV (empty vector) controls (Fig. 5c). Only pLVX-v5Per1 mice showed a signiﬁcant increase in preference for the moving object at rest relative to training with no observed group differences in total exploration at test (Fig. 5d) or in movement during habituation (Supplementary Fig. 8d).

Following behavior, hippocampal tissue from a subset of animals was processed for RNA sequencing to conﬁrm the presence of the pLVX-EV and pLVX-v5Per1 transcripts in the appropriate groups in vivo (Supplementary Fig. 7a, b, f, g).

To complement this approach, we also used the CRISPR/dCas9 Synergistic Activation Mediator (SAM) system33 to drive transcriptional activation of Per1 in the dorsal hippocampus. This system consists of three lentiviral components: a catalytically inactive Cas9 (dCas9) fused to a VP64 transcriptional activation domain with a GFP tag (dCas9-VP64-GFP), a modiﬁed single- guide RNA (sgRNA) targeting Per1 with two MS2 RNA aptamers that can recruit the third component, an MS2- double transcriptional activator (MS2-p65-HSF1) fusion protein (Fig. 5e). Control animals received a control sgRNA lacking the 20 nucleotide sequence required to target the SAM system to Per1 (referred to as ctrl sgRNA). Assembly of these components at the Per1 promoter allows the three effector domains (VP64, p65, and HSF1) to drive Per1 transcription. To conﬁrm, the effectiveness of Fig. 5 Overexpression of Per1 in the dorsal hippocampus ameliorates age-related impairments in object location memory. a Schematic of lentivirus construct used to overexpress v5-tagged Per1 (pLVX-v5Per1) compared to the empty vector control (pLVX-EV). b Per1 mRNA was signiﬁcantly increased 24 and 48 h after transfection of pLVX-v5Per1 in HT22 cells compared to cells transfected with EV (Two-way ANOVA: Group x Timepoint interaction (F(1,8) = 52.8, p < 0.001), Sidak’s post hoc tests, ***p < 0.001, n = 3, 3, 3, 3). c 18-m.o. mice given hippocampal infusions of pLVX-v5Per1 showed signiﬁcantly better memory for OLM than mice given pLVX-EV control virus (Two-way ANOVA: virus x session interaction, (F(1,29) = 7.15, p < 0.05), Sidak’s post hoc tests, **p < 0.01, ***p < 0.001, n = 16, 15, all males). d Total exploration was similar for both groups at test (t(29) = 0.57, p = 0.57). e Schematic of CRISPR/dCas9 Synergistic Activation Mediator (SAM) system used to drive Per1 transcription. Top: Individual components of SAM. Bottom: Components assembled at the Per1 promoter, driving Per1 transcription. f Per1 mRNA was signiﬁcantly increased 48 h after transfection of CRISPR-SAM components in HT22 cells compared to cells transfected with the control sgRNA (Two-way ANOVA: Group x Timepoint interaction (F(1,8) = 14.77, p < 0.01), Sidak’s post hoc tests, **p < 0.001, n = 3,3,3,3). g 18-m.o. mice given hippocampal infusions of the CRISPR-SAM system with sgRNA targeting Per1 showed signiﬁcantly better memory for OLM compared to EV control mice with control sgRNA (Two-way ANOVA: Virus x Session interaction (F(1,30) = 5.83, p < 0.05), Sidak’s post hoc tests, ***p < 0.001, n = 17, 15, all males). h Total exploration was similar for both groups at test (t(30) = 0.41, p = 0.68). Data are presented as mean ± SEM in the CRISPR-SAM system, we transfected HT22 cells with all three plasmids, harvested the cells 24 or 48 h later, and measured Per1 mRNA expression. By 48 h after transfection, Per1 mRNA was signiﬁcantly increased in the group given Per1 sgRNA compared to controls (Fig. 5f), conﬁrming that the CRISPR-SAM system effectively drives Per1 mRNA expression. We observed no change in expression for either Per2 mRNA (Supplementary Fig. 6e) or Hes7 mRNA (Supplementary Fig. 6f), indicating that the CRISPR-SAM system drives selective overexpression of the target gene, Per1.

Next, 18-m.o. mice were given intra-hippocampal infusions of the CRISPR-SAM lentiviruses (Supplementary Fig. 6d) and trained in OLM 2 weeks later. As observed with our pLVX- v5Per1 overexpression, CRISPR-SAM-mediated Per1 overexpression signiﬁcantly improved memory performance in Per1 sgRNA-infused mice relative to control animals (Fig. 5g) without affecting total exploration (Fig. 5h) or movement during habituation (Supplementary Fig. 8e). Only mice given the Per1 sgRNA showed a signiﬁcant increase in preference for the moving object at rest relative to training, indicating intact memory for the object locations (Fig. 5g). Following behavior, hippocampal tissue from a subset of animals was processed for RNA sequencing to conﬁrm the presence of the CRISPR transcripts in the appropriate groups in vivo (Supplementary Fig. 7c–g). Together, these results demonstrate that overexpression of Per1 in the dorsal hippocampus is sufﬁcient to ameliorate age-related impairments in long-term object location memory. Per1 is, therefore, a key gene that is critical for long-term memory formation, is regulated by HDAC3, and is impaired in the aging brain.

Discussion

Our results show that deletion or disruption of the repressive histone deacetylase HDAC3 can ameliorate age-related impairments in both long-term memory and synaptic plasticity. Further, deletion of HDAC3 restores experience-induced expression of the circadian gene Per1 in the dorsal hippocampus. As hippocampal PER1 expression is critical for long-term memory formation (Fig. 4) and overexpression of Per1 in the hippocampus ameliorates age-related memory impairments (Fig. 5), PER1 is a potential mechanism through which deletion of HDAC3 improves memory and synaptic plasticity in aging mice. More broadly, age-related disruption of Per1 might connect age-related impairments in both long-term memory and circadian rhythmicity, depending on the structure.

One key ﬁnding from the current study was that age-related impairments in hippocampal LTP could be ameliorated with HDAC3 deletion or disruption. This is consistent with recent work from the Sajikumar lab demonstrating that pharmacological blockade of HDAC3 can also ameliorate age-related impairments in associative hippocampal LTP34. Interestingly, we found that slices from old HDAC3ﬂox/ﬂox and HDAC3(Y298H) animals failed to reach the same level of potentiation as slices from young HDAC3ﬂox/ﬂox or HDAC3(Y298H) animals (Fig. 2c, f), suggest- ing that aging brains may have a lower plasticity ceiling than young brains. One possible explanation for this difference in potentiation is that age-related loss of synaptic contacts in CA122 could lower the plasticity ceiling in the aging hippocampus, as the fewer available synapses would become saturated more quickly than the relatively abundant synapses in the young DH. If this is the case, strengthening the stimulation protocol or providing spaced stimulation bouts35 should not further enhance LTP in the aging hippocampus, as no additional synapses are available. Further work will be necessary to determine the mechanism underlying this difference.

Our RNA sequencing results demonstrated that only a small subset of genes ﬁt the criteria of being restored in the aging brain by HDAC3 deletion (Fig. 3e). Thus, rather than recapitulating the gene expression proﬁle of the young brain, deleting HDAC3 in the aging brain restored experience-induced expression of a few key genes that are critically important for long-term memory formation, including Per1. Per1 appears to be directly regulated by HDAC3, as focal deletion of HDAC3 restored experience-induced Per1 expression and HDAC3 is physically removed from the Per1 promoter in response to OLM training. Further, Per1 expression is necessary for memory, as siRNA-mediated knock-down of PER1 protein in the DH impaired long-term memory formation in young mice, and overexpression improved memory for OLM in aging mice. Notably, both lentiviral systems used to overexpress Per1 only transfected a small number of cells in the CA1b area of the dorsal hippocampus (Supplementary Fig. 6a, d), making it necessary to conﬁrm the presence of these constructs with RNA-sequencing (see methods). As previous work has shown that this precise region of the dorsal hippocampus is critical for OLM36, this demonstrates that even a small focal alteration of Per1 within a memory-relevant structure can impact long-term memory formation. This extends previous research showing that nonspeciﬁc deletion of Per1 throughout the brain can impair memory formation in young mice6,28,29. Abnormal HDAC3-mediated repression of Per1 in the aging brain is therefore a key event that could lead to age-related impairments in both long-term memory formation and circadian rhythmicity. Although this study clearly demonstrates that Per1 is an important gene through which HDAC3 regulates long-term memory in the aging brain, other genes most likely contribute to the memory-enhancing effects of HDAC3 deletion. Other genes proposed to mediate the effects of HDAC3 on synaptic plasticity and memory include NF-κB34, FMRP37, and Nr4a226. In the current study, we identiﬁed an additional three genes (Fig. 3f:Nr4a1, Egr1, Tsc22d3) that, like Per1, show impaired expression in the aging brain that is improved by HDAC3 deletion. How these different genes uniquely contribute to the memory-enhancing effects of HDAC3 deletion is currently unclear. Notably, all of these genes interface with the CBP/CREB pathway, which is critical for long-term memory formation38,39 and is both up-and downstream of PER16,28. As PER1 is upstream of CREB phosphorylation6, it is possible that HDAC3-mediated repression of Per1 reduces CREB phosphorylation, ultimately impairing the transcription of CREB-mediated genes like Fmrp40,41, and the Nr4a gene family members Nr4a1 and Nr4a226. Understanding the complex dynamics between HDAC3, PER1, and these other molecular players will be an important goal of future research.

In the current study, we used two complementary methods to overexpress Per1 in the aging brain: overexpression of a full-length Per1 cDNA construct using pLVX-v5Per1 and transcriptional activation of endogenous Per1 using the CRISPR-SAM system. Although pLVX-mediated overexpression drove a large increase in Per1 mRNA (Fig. 5b), this was accompanied by an increase in Per2, another Period family member (Supplementary Fig. 6b). The expression of the downstream gene Hes7, however, was unchanged by pLVX-v5Per1 (Supplementary Fig. 6c). The CRISPR-SAM system, on the other hand, produced a more subtle increase in Per1 mRNA that avoided off-target enhancements of either Per2 or Hes7 expression (Supplementary Fig. 6e–f). As both approaches similarly improved long-term memory in aging mice, it is unlikely that the off-target increase in Per2 observed with pLVX-v5Per1 was the cause of the observed improvement in long-term memory.

Circadian effects on long-term memory are traditionally believed to stem from dysregulation within the SCN, which then drives alterations in peripheral structures involved in memory formation, like the dorsal hippocampus. Little is known about the role of individual circadian clock genes in the DH, despite a clear connection between circadian rhythmicity and long-term memory formation. Memory is closely linked to time-of-day, as plasticity-related gene cascades show circadian oscillations3,6 and memory can be acquired more easily at certain periods of the circadian cycle. For example, context fear conditioning is acquired more strongly during the daytime, when MAPK phosphorylation levels peak3. Further, it is well-documented that aging is accompanied by a breakdown of circadian rhythms, presumably due to changes in the central circadian clock, the suprachiasmatic nucleus (SCN) (for review1). How this disruption in the circadian clock relates to the age-related impairments in memory is an open question. Our results suggest that HDAC3 limits experience-induced Per1 in the aging hippocampus, possibly contributing to the observed impairments in long-term memory. Epigenetic repression of Per1 may therefore represent an important interface between age-related impairments in both circadian rhythmicity and long-term memory formation.

Of the core canonical circadian clock genes, Per1 is uniquely poised to dramatically affect hippocampal long-term memory. Per1 appears to be predominantly involved in SCN output pathways and plays a key role in peripheral clocks downstream of the SCN, such as the hippocampus42. Further, recent work suggests that Per1 may “gate” spatial memory formation throughout the day/night cycle by controlling CREB phosphorylation6,28,29. Indeed, hippocampal Per1 mRNA upregulation has been observed after context or spatial learning in at least three other RNA-seq studies, including work in rats, indicating that Per1 is typically upregulated after learning across species20,43,44. Along with the results of the current study, this work indicates that PER1 expression is critically important for hippocampal long-term memory formation; reductions in PER1 that occur at night29 or with aging could impair hippocampal memory. To date,

Cistanche-improve memory13

research implicating Per1 in memory formation has exclusively relied on global knockouts that disrupt Per1 expression in the core circadian clock and other regions in addition to memory-relevant structures like the hippocampus6,28,29,45,46, making it impossible to determine whether hippocampal PER1 is speciﬁcally required for memory formation. Indeed, global Per1 deletion does affect circadian rhythmicity in some reports42. Here, we show for the ﬁrst time that reducing PER1 expression directly in the dorsal hippocampus can impair memory in young mice whereas local overexpression of Per1 in the dorsal hippocampus can improve memory in aging mice. As selective deletion HDAC3 (which regulates Per1) in the dorsal hippocampus had no effect on circadian activity patterns in the current study (Supplementary Fig. 4), and even electrolytic lesions of the dorsal hippocampus are insufﬁcient to affect circadian rhythmicity47, it seems unlikely that manipulating Per1 within the dorsal hippocampus affects the function of the central circadian clock. Nonetheless, it is possible that experience-induced increases in Per1 or virus-mediated overexpression of Per1 can affect the circadian oscilla- tion of other molecular players, such as other clock genes, within hippocampal neurons, even without affecting circadian activity patterns. Understanding how Per1 functions to alter memory formation, including identifying these potential interacting partners, will be the target of future work.

Together, these results demonstrate that the core circadian clock gene Per1 plays a key role within local memory structures to alter memory formation, a role that is independent of its function in the SCN. More generally, this challenges the traditional hypothesis that circadian changes in memory formation are driven by alterations in the core circadian clock and instead supports the hypothesis that circadian clock genes play a more autonomous role in hippocampal cells, possibly gating memory formation based on the time of day6.

Methods

Mice. Young adult mice were between 2 and 4 months old at the time of testing and aging mice were between 18 and 20 months old. All mice were C57BL/6J or maintained on a C57BL/6 background (HDAC3+/+ and HDAC3ﬂox/ﬂox mice). Mice had free access to food and water and lights were maintained on a 12 h light/ dark cycle. All behavioral testing was performed during the light cycle. All experiments were conducted according to US National Institutes of Health guidelines for animal care and use and were approved by the Institutional Animal Care and Use Committee of the University of California, Irvine.

Surgery. Mice were anesthetized with isoﬂurane (induced, 4%; maintained1.5–2.0%) and placed in the stereotaxic. Injection needles were lowered to the dorsal hippocampus at a rate of 0.2 mm/15 s (AP, −2.0 mm; ML, ± 1.5 mm, DV, − 1.5 mm relative to Bregma). Two minutes after reaching the target depth, 1.0 μl of virus or siRNA was infused bilaterally into the DH at a rate of 6 μl/h. For the CRISPR-SAM lentiviral infusions, a cocktail of the three viruses was infused to a ﬁnal volume of 1.5 μL per hemisphere at the same rate. Injection needles remained in place for two minutes post-injection to allow the virus to diffuse. The injectors were then raised 0.1 mm and allowed to sit for another minute before being removed at a rate of 0.1 mm per 15 s. Viral infusions were performed 2 weeks before behavioral analysis whereas siRNA knockdown was performed 2d before training13. For all injection experiments, animals were randomly assigned to the different injection conditions (with the exception of HDAC3+/+ and HDAC3ﬂox/ﬂox mice, which were all injected with AAV-CaMKII-Cre). For all behavioral experiments, animals within each viral condition were randomly assigned to homecage/trained groups and all conditions (objects, boxes, etc.) were counterbalanced between groups.

Cistanche-improve memory1

AAV production. AAV2.1-CaMKII-Cre was purchased from Penn Vector Core(titer: 1.81 × 1013 GC/ml). For AAV2.1-HDAC3(Y298H)-v5, we ampliﬁed wildtype HDAC3 from hippocampal cDNA and cloned the product into a modiﬁed pAAV- IRES-hrGFP (Agilent) under control of the CMV promoter and β-globin intron. The 3×-FLAG tag, IRES element, and hrGFP were removed from the vector and replaced with a V5 tag, allowing for a C-terminal fusion to HDAC3 (plasmid MW91). To create the point mutation, we changed the nucleotides to code for a histidine residue in place of tyrosine at amino acid 298 (plasmid MW92). For the empty vector control, the HDAC3 coding sequence was not present, but all other elements remain (plasmid MW87). AAV was made by the Penn Vector Core and the ﬁnal titers were determined by qPCR (AAV-HDAC3(Y298H): 6.48 × 1012 GC/ ml; AAV-EV: 1.35 × 1013 GC/ml).

Lentivirus production. For the CRISPR/dCas9 Synergistic Activation Mediator (SAM) system, lentiviral plasmids were purchased from Addgene for the dCas9- VP64-GFP (#61422-LVC) and MS2-P65-HSF1_Hygro (#61426-LVC) constructs (titers ≥ 8× 106 TU/ml). For the Per1 sgRNA, we cloned and inserted an antisense guide sequence corresponding to the CRE element in the Per1 promoter (AGAGGGAGGTGACGTCAAAG) into the Addgene sgRNA(MS2) cloning backbone (#61427). The control sgRNA was identical, except that no guide sequence was cloned into the plasmid. Lentiviruses for both the Per1 sgRNA (titer: 6.8 × 107 IFU/ml) and control sgRNA (3.5 × 107 IFU/ml) were produced by the USC School of Pharmacy Lentiviral Laboratory.

For the pLVX-V5Per1 overexpression construct, we ampliﬁed full-length wildtype Per1 from the Addgene pCMV-Sport2-mPer1 plasmid (#16203) and cloned the product into a modiﬁed pLVX-EF1α-IRES-mCherry backbone (Takara, #631987). The IRES and mCherry elements were removed and were replaced with a V5 tag, allowing for an N-terminal fusion to PER1 (plasmid MW206). For the empty vector (EV) control, the PER1 coding sequence was not present but all other elements remained (Plasmid MW93). Lentiviruses for the pLVX-v5Per1 (titer: 1.3 × 108 IFU/ml) and pLVX-EV (titer: 1.5 × 108 IFU/ml) were produced by the USC School of Pharmacy Lentiviral Laboratory. All lentiviral constructs were expressed under the EF1α promoter.

Cell Culture Veriﬁcation of pLVX-v5Per1 and CRISPR-SAM. To verify that pLVX-v5Per1 produces overexpression of Per1 mRNA, mouse HT22 cells (Salk Institute, La Jolla, CA, #T09031) were transfected with pLVX-v5Per1 or pLVX- EV (Fig. 5a) using Lipofectaine LTX (Invitrogen). Similarly, to verify that the CRISPR-SAM system can effectively drive Per1 transcription, HT22 cells were transfected with dCas9-VP64 (lenti MS2-P65_HSF1_Blast was a gift from Feng Zhang Addgene plasmid #61425), MS2-P65-HSF1 (lenti dCAS-VP64_Blast was a gift from Feng Zhang, Addgene plasmid #61426), and either Per1 sgRNA (Per1 sgRNA) or the non-targeting control sgRNA (ctrl sgRNA) (lenti sgRNA (MS2)_zeo backbone was a gift from Feng Zhang, Addgene plasmid #61427) using Lipofectamine LTX (Invitrogen). Cells were harvested after 24 or 48 h, lysed, and mRNA was isolated as described above. qRT-PCR was performed as described above using the Per1, Per2, and Hes7 primers and probe listed in Table S4.

siRNA. For the Per1 knockdown experiment, Accell SMARTpool small interfering RNAs (siRNAs; Dharmacon, GE) targeting Per1 were diluted to a ﬁnal total concentration of 10 μM in ddH20 and infused into the DH (1.0 μl/side). Accell non-targeting pool siRNA was used as the control (total concentration, 10 μM) and was infused in the same manner. For siRNA experiments, mice were handled and habituated as described above and surgery was performed the day after the ﬁnal day of habituation. Mice were given a full day of recovery after surgery and were trained the following day (~48 h after surgery) to ensure maximal target knock-down. To ensure knockdown, mice were sacriﬁced ~1 h after test and punches from the dorsal hippocampus were processed with western blots to ensure knockdown of PER1 protein.

Object location and object recognition memory tasks. For object location and object recognition memory tasks, mice were handled for 2 min/day for 4 day and then habituated to the context for 5 min/day for six consecutive days in the absence of objects. During training, mice were exposed to two identical objects (100 ml beakers, spice tins, or glass candle holders) and allowed to explore for 10 min. During the retention test (24 h later for long-term memory or 60 m later for short-term memory), mice were allowed to explore for 5 min. For object location memory, one of the two familiar objects was moved to a new location. For object recognition memory, the object locations remained constant but one of the objects was replaced with a new item. Habituation for object recognition memory began at least one week after the completion of OLM testing and a new context and unfamiliar objects were used48. Exploration was scored when the mouse head oriented toward the object and came within 1 cm or when the nose touched the object. Total exploration time was recorded (t) and preference for the novel item was expressed as a discrimination index (DI = (tnovel -tfamiliar) / (tnovel + tfamiliar) x 100%). For training sessions, the object designated to be moved at test was used as the novel object to allow the training and testing DI to be directly compared. Mice that explored both objects for less than 2 s during testing or 3 s during training were removed from further analysis. Mice that showed a preference for one object during training (DI > ±20) were also removed. Habituation sessions were analyzed (to determine the distance traveled and speed) using ANY-maze behavioral analysis software (Stoelting Co). All habituation, training, testing, and scoring were performed by experimenters blinded to the experimental groups.

Cistanche-improve memory20

Context fear to condition. For contextual fear conditioning, mice were ﬁrst handled for 5 days. During acquisition, mice were exposed to the context for 2 min and 28 s followed by a 2 s (0.75 mA) shock, a protocol that typically produces robust long-term memory12,20. Mice remained in the context for an additional 30 s before being removed. Mice were sacriﬁced 60 m after training along with homecage controls that were handled but not trained. Freezing behavior was measured using Ethovision 11 software (Noldus)12.

Elevated plus-maze. The plus-maze was conducted by an experimenter blind to the experimental groups. One week after the completion of ORM, a subset of mice were tested on the plus-maze. Two arms of the maze were open (30 × 5 cm) and two arms were enclosed (30 × 5 × 15 cm), connected by a central platform (5 × 5 cm). The maze was elevated 40 cm above the ﬂoor. Mice were tested for 5 min on the apparatus, which consisted of placing each mouse onto the central platform facing one of the open arms. Between subjects, the maze was cleaned with 70% ethanol. The percentage of time spent in the closed and open arms was scored using ANY-maze software.

Circadian rhythm analysis. Young (3-m.o) and aging (18-m.o.) HDAC3+/+ and HDAC3ﬂox/ﬂox mice were bred and housed under a 12 h light/dark (LD) cycle. Two weeks after AAV-CaMKII-Cre infusion (described above), mice were transferred to an isolated 12 h LD entrainment room for 7 days. Mice were then transferred into a light-protected activity analysis room, where locomotor activity was analyzed using optical beam motion detectors (Philips Respironics). The activity was monitored during 2 weeks of LD cycle entrainment in the light-protected room before mice were switched to constant darkness (DD) for an additional 3 weeks. Activity monitoring continued throughout the DD phase to determine whether HDAC3 deletion in the DH affected endogenous circadian rhythms. Data were collected using Minimitter VitalView 5.0 and Clocklab software (Actimetrics) was used to determine the onset of free activity. Tau values were calculated by obtaining the slope of this onset and calculating the least-squares ﬁt with Clocklab software49,50.

Immunohistochemistry. After the completion of behavior, mice were euthanized by cervical dislocation and their brains were removed and ﬂash-frozen in ice-cold isopentane. Twenty-micrometer slices were collected throughout the dorsal hip- campus, thaw-mounted on slides and stored at −80 °C. Slides were ﬁxed with 4% paraformaldehyde (10-min), permeabilized in 0.01% Triton X-100 in 0.1 M PBS (5-min), and blocked for 1 h with 8% normal goat serum (Jackson). Slides were incubated overnight (4 °C) in rabbit antibody to HDAC3 (1:250, Abcam, clone Y415, ab32369) or V5 (1:1000, Abcam, ab9116), or chicken antibody to GFP (1:250, Aves Labs, #GFP1010). The following day, slides were washed and incubated for 1 h at room temperature with goat anti-rabbit Alexa 488 (HDAC3 and V5; 1:1000, ThermoFisher, #A-11008) or goat anti-chicken Alexa 488 (GFP; 1:1000, ThermoFisher, #A-11039) in the dark. Slides were then washed with PBST and incubated for 50 m in NeuroTrace 530/615 (1:50; ThermoFisher, #N21482), a ﬂuorescent nissl stain. To quench nonspeciﬁc autoﬂuorescence51, slices were then washed in PBS with 0.01% Triton, rinsed in water, and incubated for 10-min in 10 mM CuSO4 in 50 mM ammonium acetate buffer. Slices were again rinsed in water, washed in PBS, and coverslipped with VectaShield Antifade mounting medium (Vector Laboratories).

All images were acquired with an Olympus Scanner VS110 with a 20x apochromatic objective (numerical aperture 0.75) with VS110 scanner software. All treatment groups were represented on each slide and all images on a slide were captured with the same exposure time under nonsaturating conditions. Immunolabeling intensity was quantiﬁed with ImageJ by sampling the optical density of the cell layer in CA1 and subtracting a sample of background ﬂuorescence in the same image. For all AAV experiments, animals that failed to express the virus in area CA1 of the dorsal hippocampus were excluded from analyses. Imaging and quantiﬁcation were performed by experimenters blind to the experimental conditions.

Western blot. To verify PER1 knockdown, Per1 siRNA and control siRNA mice were sacriﬁced 2 h after testing and brains were ﬂash-frozen. Brains were coronally sectioned and 1 mm DH punches were collected from 500 μm slices. Punches were homogenized in T-PER buffer (Thermo Fisher) with Halt protease and phosphatase inhibitor (Thermo Fisher) using a Dounce homogenizer. Protein lysates were quantiﬁed using a modiﬁed Bradford assay (BioRad) and 50 μg total protein lysate was loaded into each lane of a 7.5% NuPAGE Bis-Tris gel (Thermo Fisher). Gels ran for 50 min at 200 volts and blots were transferred overnight at 15 V at 4 °C onto nitrocellulose membranes (Novexi, LC2001). The following day, membranes were incubated in blocking buffer for 1 h (5% nonfat milk in Tris-buffered saline with 0.01% Tween 20), washed (0.1% Tween 20 in TBS) and then incubated in primary antibody (1:500, rabbit anti-Per1, Thermo Fisher #PA1-524) in primary antibody buffer (3% BSA in TBS with 0.1% Tween) overnight at 4 °C. The membranes were then washed and incubated in HRP- conjugated mouse antibody to rabbit (1:10,000, Jackson Laboratories, light chain-speciﬁc, #211-032-171) for 1 h. Membranes were washed and developed using Pierce SuperSignal West Pico Chemiluminescent Substrate (Pierce, 34077). Multiple ﬁlm exposures were used to verify linearity. Blots were washed and stripped for 10 m with Restore Western Blot Stripping Buffer (Thermo Fisher #21059), washed again, and then re-probed overnight (4 °C) with a rabbit antibody to GAPDH (1:1000, Santa Cruz Biotechnology, SC25778). Full-length PER1 and GAPDH blots are shown in Supplementary Figure 5. Relative optical densities were calculated from scanned ﬁlm using ImageJ (US National Institutes of Health) by an experimenter blind to the experimental conditions. All values were normalized to GAPDH expression levels.

qRT-PCR. Tissue was collected from DH punches (described above) and frozen at −80 °C until processing. RNA was isolated using an RNeasy Minikit (Qiagen, #74104) and cDNA was created using the Transcriptor First Strand cDNA Synthesis kit (Roche, 04379012001). Primers and probes were derived from the Roche Universal ProbeLibrary (Table S4) and were used for multiplexing in the Roche Light-Cycle 480 II machine (Roche). All values were normalized to Gapdh. Analyses and statistics were performed using the Roche proprietary algorithms and REST 2009 software based on the Pfafﬂ method52,53.

RNA sequencing. RNA was isolated from dorsal hippocampus punches as described above, using the RNeasy Minikit (Qiagen, 74104). RNA quality was assessed by Bioanalyzer and samples with an RNA integrity number >9 were included in our analysis. cDNA libraries for each group were prepared according to the TruSeq RNA Sample Preparation Guide (Illumina). Two hundred and ﬁfty nanograms of total RNA from each mouse was puriﬁed with poly-T oligo-attached magnetic beads and heat fragmented. The ﬁrst and second-strand cDNA was then synthesized and puriﬁed. After the ends were blunted, the 3′ end was adenylated to prevent concatenation of the template during adapter ligation. For each group, a unique adapter set was added to the ends of the cDNA and the libraries were ampliﬁed by PCR. The quality of the library was assessed by Bioanalyzer and quantiﬁed using qRT-PCR with a standard curve prepared from a commercial sequencing library (Illumina). Samples were multiplexed, with each behavioral group represented in each ﬂow cell of the sequencer. 10 nM of each library was pooled in four multiplex libraries and sequenced on an Illumina HiSeq 2500 instrument during a single-read 50 bp sequencing, run by the Genomic High- Throughput Facility at the University of California, Irvine. The resulting sequencing data for each library were post-processed to produce FastQ ﬁles. The data were then demultiplexed and ﬁltered using Illumina CASAVA 1.8.2 software as well as in-house software. Poor-quality reads (failing Illumina’s standard quality tests) and control reads successfully aligned to the PhiX control genome were removed from analyses. The quality of the remaining sequences was further assessed using PHRED quality scores produced in real-time during the base-calling step of the sequencing run (Supplementary Fig. 3a).

Alignment to the reference genome and transcriptome. The reads from each experiment were separately aligned to the reference genome and corresponding transcriptome using short-read aligners ELAND v2e (Illumina) and Bowtie24. Reads uniquely aligned by both tools to known exons or splice junctions with no more than two mismatches on any 25 bp fragment of the read were included in the transcriptome. Reads uniquely aligned, but with more than two mismatches on any 25 bp fragment of the read were removed from analyses. Similarly, reads matching several locations in the reference genome were removed from the analysis. The percentage of reads assigned to the reference genome and transcriptome using this protocol is reported for each group of replicates (Supplementary Fig. 3b). This resulted in an average of approximately 30 million reads per sample.

To verify the presence of the pLVX and CRISPR-SAM plasmids following lentiviral infusion, we ran RNA sequencing on hippocampal samples from a subset of three animals per group following behavior. As the number of cells infected with the lentiviruses in vivo was too small to reliably detect the presence of the appropriate transcripts using RT-qPCR (Supplementary Fig. 6a, d), we used RNA-seq as a more sensitive method to detect the presence of these transcripts. An updated version of genome assembly and genome annotation were built by appending the sequences and annotation of the plasmid constructs to the original mm10 mouse genome and to the mm10 mouse genome annotation, respectively. Using the reads alignment tool TopHat in the the Tuxedo Suite54, reads were mapped to the genome and only hits that were unique to the plasmid transcripts in comparison with the endogenous mouse reference genome were considered “matched reads” to the plasmid constructs. The number of those matched reads was then quantiﬁed for each construct for each sample.

Gene expression and differential analysis. Gene expression levels were directly computed from the read alignment results for each replicate. Standard RPKM values55, reads per kilobase of exon model per million mapped reads) were extracted for each gene covered by the sequencing data and each replicate used in this study.

Differential transcriptional analyses were performed using Cyber T56,57 across each pair of groups (homecage versus 60 m after training) to identify genes up-or down-regulated after OLM. In addition to the 18-month-old HDAC3+/+ and HDAC3ﬂox/ﬂox mice trained for the current study, identically processed RNA-sequencing data from 3-month old C57 wildtype mice (homecage and OLM-trained in the same manner as the current study) from a previous study20 was used for differential analyses. The number of animals (biological replicates) for each group was: Young HDAC+/+ HC: 6, Young HDAC3+/+ OLM: 6, Old HDAC3+/+ HC: 6, Old HDAC3+/+ OLM: 6, Old HDAC3ﬂox/ﬂox HC: 8, Old HDAC3ﬂox/ﬂox OLM: 8. No technical replicates were used. The p-value threshold used for determining differential expression is 0.05 for all groups. False Discovery Rate (FDR) as measured by the Benjamini & Hochberg (BH) q-value was used to correct for multiple testing, with a FDR threshold of 0.15. The sets of genes upregulated after the experience (compared to homecage) for these 3 groups (young wildtype, aging HDAC+/+, or aging HDAC3ﬂox/ﬂox) were intersected to determine genes common to two or more groups. Enrichment of each group for tissue-speciﬁc expression, Gene Ontology terms58, and KEGG pathways59,60 was assessed using DAVID61, based on differentially expressed genes after behavior. Data visualization was performed using “matplotlib” for python and “ggplot” for R.

Chromatin immunoprecipitation. ChIP was performed on DH punches based on the protocol from the Millipore ChIP kit11,12,62. Tissue was cross-linked with 1% formaldehyde (Sigma), lysed and sonicated, and chromatin was immunoprecipitated overnight with 2 μl of anti-H4K8Ac (Millipore #17-10099) or 4 μl of anti- HDAC3 (Millipore #17-10238) or an equivalent amount of Normal Rabbit Serum (H4K8Ac negative control, Millipore) or anti-mouse IgG (HDAC3 negative control, Millipore). After washing, chromatin was eluted from the beads and reverse cross-linked in the presence of proteinase K before column pur- iﬁcation of DNA. Primer sequences for the promoters of each gene were designed by the Primer 3 program (Table S4). Five μl of input, anti-H4K8Ac IgG/anti-HDAC3 IgG, or anti-rabbit/mouse IgG immunoprecipitate from each animal were examined in duplicate. To normalize ChIP-qPCR data, we used the percent input method. The input sample was adjusted to 100% and both the IP and IgG samples were calculated as a percent of this input using the formula 100*AE^(adjusted input – Ct(IP)). Fold enrichment was then calculated as a ratio of the ChIP to the average IgG. An in-plate standard curve determined ampliﬁcation efﬁciency (AE).

Slice preparation and recording. Young (approximately 3-m.o.) and aging (18-m.o.) mice were stereotaxically infused with the virus. For the ﬁrst experiment, young and old HDAC3+/+ and HDAC3ﬂox/ﬂox mice were infused with AAV-CaMKII- Cre bilaterally into the DH. For the second experiment, young and old wildtype C57 mice were infused with AAV-HDAC3(Y298H) into the DH of one hemisphere and AAV-EV control into the other hemisphere. Two weeks after infusion (to allow for optimal virus expression)2, transverse hippocampal slices (320 μm) were placed in an interface recording chamber with preheated (31 ± 1 °C) artiﬁcial cerebrospinal ﬂuid (124 mM NaCl, 3 mM KCl, 1.25 mM KH2PO4, 1.5 mM MgSO4, 2.5 mM CaCl2, 26 mM NaHCO3, and 10 mM D-glucose). Slices were continuously perfused at a rate of 1.75–2 ml per min while the surface of the slices was exposed to warm, humidiﬁed 95% O2/5% CO2. Recordings began after at least 2 h of incubation.

Field excitatory postsynaptic potentials (fEPSPs) were recorded from CA1b stratum radiatum using a single glass pipette (2–3MΩ) ﬁlled with 2 M NaCl. Stimulation pulses (0.05 Hz) were delivered to Schaffer collateral-commissural projections using a bipolar stimulating electrode (twisted nichrome wire, 65 μm) positioned in CA1c. The current intensity was adjusted to obtain 50% of maximal fEPSP response. After a stable baseline was established, LTP was induced with a single train of ﬁve theta bursts, in which each burst (four pulses at 100 Hz) was delivered 200 ms apart (i.e., at theta frequency). The stimulation intensity was not increased during TBS. Data were collected and digitized by NAC 2.0 Neurodata Acquisition System (Theta Burst).

Statistical analysis. Statistical analyses were conducted as indicated in the text and ﬁgure legends using Prism 6 (GraphPad). Our analytic approaches are based on previously published work12,20,63,64. No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those generally used in the ﬁeld, including those reported in the previous publications12,20,64,65. Data distribution was assumed to be normal, with similar variance observed among groups, but this was not formally tested. When an experiment had two groups to compare, two-tailed Student’s t-tests were used. When two factors where compared, (such as age and genotype or session and group), data were analyzed with two-way ANOVAs followed by Sidak’s multiple comparisons post hoc tests. All analyses are two-tailed, with an α value of 0.05 required for signiﬁcance. Error bars in all ﬁgures represent SEM. For all experiments, values ± 2SD from the group mean were considered outliers and were removed from analyses.

Data availability. The data supporting the ﬁndings of this study are available from the corresponding author upon reasonable request. RNA sequencing data have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE94832.

References

1. Deibel, S. H., Zelinski, E. L., Keeley, R. J., Kovalchuk, O. & McDonald, R. J. Epigenetic alterations in the suprachiasmatic nucleus and hippocampus contribute to age-related cognitive decline. Oncotarget 6, 23181–23203 (2015).

2. Gerstner, J. R. & Yin, J. C. Circadian rhythms and memory formation. Nat. Rev. Neurosci. 11, 577–588 (2010).

3. Eckel-Mahan, K. L. et al. Circadian oscillation of hippocampal MAPK activity and cAmp: implications for memory persistence. Nat. Neurosci. 11, 1074– 1082 (2008).

4. Chaudhury, D. & Colwell, C. S. Circadian modulation of learning and memory in fear-conditioned mice. Behav. Brain Res. 133, 95– 108 (2002).

5. Kondratova, A. A. & Kondratov, R. V. The circadian clock and pathology of the ageing brain. Nat. Rev. Neurosci. 13, 325–335 (2012).

6. Rawashdeh, O., Jilg, A., Maronde, E., Fahrenkrug, J. & Stehle, J. H. Period1 gates the circadian modulation of memory-relevant signaling in mouse hippocampus by regulating the nuclear shuttling of the CREB kinase pP90RSK. J. Neurochem. 138, 731–745 (2016).

7. Penner, M. R., Roth, T. L., Barnes, C. A. & Sweatt, J. D. An epigenetic hypothesis of aging-related cognitive dysfunction. Front Aging Neurosci. 2, 9 (2010).

8. Peleg, S. et al. Altered histone acetylation is associated with age-dependent memory impairment in mice. Science 328, 753–756 (2010).

9. Snigdha, S. et al. H3K9me3 inhibition improves memory, promotes spine formation, and increases BDNF levels in the aged hippocampus. J. Neurosci. 36, 3611–3622 (2016).

10. Spiegel, A. M., Sewal, A. S. & Rapp, P. R. Epigenetic contributions to cognitive aging: disentangling mindspan and lifespan. Learn Mem. 21, 569–574 (2014).

11. Malvaez, M. et al. HDAC3-selective inhibitor enhance extinction of cocaine-seeking behavior in a persistent manner. Proc. Natl Acad. Sci. USA 110, 2647–2652 (2013).

12. Kwapis, J. L. et al. Context and auditory fear are differentially regulated by HDAC3 activity in the lateral and basal subnuclei of the amygdala. Neuropsychopharmacology 42, 1284– 1294 (2017).

13. McQuown, S. C. et al. HDAC3 is a critical negative regulator of long-term memory formation. J. Neurosci. 31, 764–774 (2011).

14. Bieszczad, K. M. et al. Histone deacetylase inhibition via RGFP966 releases the brakes on sensory cortical plasticity and the speciﬁcity of memory formation.J. Neurosci. 35, 13124– 13132 (2015).

15. Lahm, A. et al. Unraveling the hidden catalytic activity of vertebrate class IIa histone deacetylases. Proc. Natl Acad. Sci. USA 104, 17335– 17340 (2007).

16. Sun, Z. et al. Deacetylase-independent function of HDAC3 in transcription and metabolism requires nuclear receptor corepressor. Mol. Cell 52, 769–782(2013).

17. Alaghband, Y. et al. Distinct roles for the deacetylase domain of HDAC3 in the hippocampus and medial prefrontal cortex in the formation and extinction of memory. Neurobiol. Learn. Mem. 145, 94– 104 (2017).

18. Nott, A. et al. Histone deacetylase 3 associates with MeCP2 to regulate FOXO and social behavior. Nat. Neurosci. 19, 1497– 1505 (2016).

19. Norwood, J., Franklin, J. M., Sharma, D. & D’Mello, S. R. Histone deacetylase 3 is necessary for proper brain development. J. Biol. Chem. 289, 34569–34582(2014).

20. Vogel-Ciernia, A. et al. The neuron-speciﬁc chromatin regulatory subunit BAF53b is necessary for synaptic plasticity and memory. Nat. Neurosci. 16, 552–561 (2013).

21. Alberini, C. M. Transcription factors in long-term memory and synaptic plasticity. Physiol. Rev. 89, 121– 145 (2009).

22. Barnes, C. A. Long-term potentiation and the ageing brain. Philos. Trans. R. Soc. Lond. B 358, 765–772 (2003).

23. Speir, M. L. et al. The UCSC Genome Browser database: 2016 update. Nucleic Acids Res. 44, D717–D725 (2016).

24. Langmead, B., Trapnell, C., Pop, M. & Salzberg, S. L. Ultrafast and memory- efﬁcient alignment of short DNA sequences to the human genome. Genome Biol. 10, R25 (2009).

25. Rowe, W. B. et al. Hippocampal expression analyses reveal selective association of immediate-early, neuroenergetic, and myelinogenic pathways with cognitive impairment in aged rats. J. Neurosci. 27, 3098–3110 (2007).

26. McNulty, S. E. et al. Differential roles for Nr4a1 and Nr4a2 in object location vs. object recognition long-term memory. Learn Mem. 19, 588–592 (2012).

27. Vecsey, C. G., Park, A. J., Khatib, N. & Abel, T. Effects of sleep deprivation and aging on long-term and remote memory in mice. Learn Mem. 22, 197–202 (2015).

28. Rawashdeh, O. et al. PERIOD1 coordinates hippocampal rhythms and memory processing with daytime. Hippocampus 24, 712–723 (2014).

29. Jilg, A. et al. Temporal dynamics of mouse hippocampal clock gene expression support memory processing. Hippocampus 20, 377–388 (2010).

30. Eckel-Mahan, K. L. Circadian oscillations within the hippocampus support memory formation and persistence. Front Mol. Neurosci. 5, 46 (2012).

31. Guzowski, J. F., Setlow, B., Wagner, E. K. & McGaugh, J. L. Experience- dependent gene expression in the rat hippocampus after spatial learning: a comparison of the immediate-early genes Arc, c-fos, and zif268. J. Neurosci. 21, 5089–5098 (2001).

32. Kouzarides, T. Chromatin modiﬁcations and their function. Cell 128, 693–705(2007).

33. Konermann, S. et al. Genome-scale transcriptional activation by an engineered CRISPR-Cas9 complex. Nature 517, 583–588 (2015).

34. Sharma, M., Shetty, M. S., Arumugam, T. V. & Sajikumar, S. Histone deacetylase 3 inhibition re-establishes synaptic tagging and capture in aging through the activation of nuclear factor kappa B. Sci. Rep. 5, 16616 (2015).

35. Kramar, E. A. et al. Synaptic evidence for the efﬁcacy of spaced learning. Proc. Natl Acad. Sci. USA 109, 5121–5126 (2012).

36. Barrett, R. M. et al. Hippocampal focal knockout of CBP affects speciﬁc histone modiﬁcations, long-term potentiation, and long-term memory. Neuropsychopharmacology 36, 1545– 1556 (2011).

37. Franklin, A. V., Rusche, J. R. & McMahon, L. L. Increased long-term potentiation at medial-perforant path-dentate granule cell synapses induced by selective inhibition of histone deacetylase 3 requires Fragile X mental retardation protein. Neurobiol. Learn. Mem. 114, 193– 197 (2014).

38. Silva, A. J., Kogan, J. H., Frankland, P. W. & Kida, S. CREB and memory. Annu. Rev. Neurosci. 21, 127– 148 (1998).

39. Kida, S. & Serita, T. Functional roles of CREB as a positive regulator in the formation and enhancement of memory. Brain Res. Bull. 105, 17–24 (2014).

40. Smith, K. T., Nicholls, R. D. & Reines, D. The gene encoding the fragile X RNA-binding protein is controlled by nuclear respiratory factor 2 and the CREB family of transcription factors. Nucleic Acids Res. 34, 1205– 1215 (2006).

41. Wang, H. et al. Roles of CREB in the regulation of FMRP by group I metabotropic glutamate receptors in cingulate cortex. Mol. Brain 5, 27 (2012).

42. Cermakian, N., Monaco, L., Pando, M. P., Dierich, A. & Sassone-Corsi, P. Altered behavioral rhythms and clock gene expression in mice with a targeted mutation in the Period1 gene. EMBO J. 20, 3967–3974 (2001).

43. Halder, R. et al. DNA methylation changes in plasticity genes accompany the formation and maintenance of memory. Nat. Neurosci. 19, 102– 110 (2016).

44. Duke, C. G., Kennedy, A. J., Gavin, C. F., Day, J. J. & Sweatt, J. D. Experience- dependent epigenomic reorganization in the hippocampus. Learn Mem. 24, 278–288 (2017).

45. Sakai, T., Tamura, T., Kitamoto, T. & Kidokoro, Y. A clock gene, period, plays a key role in long-term memory formation in Drosophila. Proc. Natl Acad. Sci. USA 101, 16058– 16063 (2004).

46. Abarca, C., Albrecht, U. & Spanagel, R. Cocaine sensitization and reward are under the inﬂuence of circadian genes and rhythm. Proc. Natl Acad. Sci. USA 99, 9026–9030 (2002).

47. Phan, T. X., Chan, G. C., Sindreu, C. B., Eckel-Mahan, K. L. & Storm, D. R. The diurnal oscillation of MAP (mitogen-activated protein) kinase and adenylyl cyclase activities in the hippocampus depends on the suprachiasmatic nucleus. J. Neurosci. 31, 10640– 10647 (2011).

48. Vogel-Ciernia, A. & Wood, M. A. Examining object location and object recognition memory in mice. Curr. Protoc. Neurosci. 69, 1– 17 (2014).

49. Orozco-Solis, R. et al. The circadian clock in the ventromedial hypothalamus controls cyclic energy expenditure. Cell Metab. 23, 467–478 (2016).

50. Eckel-Mahan, K. & Sassone-Corsi, P. Phenotyping circadian rhythms in mice. Curr. Protoc. Mouse Biol. 5, 271–281 (2015).

51. Schnell, S. A., Staines, W. A. & Wessendorf, M. W. Reduction of lipofuscin- like autoﬂuorescence in ﬂuorescently labeled tissue. J. Histochem. Cytochem. 47, 719–730 (1999).

52. Pfafﬂ, M. W. A new mathematical model for relative quantiﬁcation in real- time RT-PCR. Nucleic Acids Res. 29, e45 (2001).

53. Pfafﬂ, M. W., Horgan, G. W. & Dempﬂe, L. Relative expression software tool (REST) for group-wise comparison and statistical analysis of relative expression results in real-time PCR. Nucleic Acids Res. 30, e36 (2002).

54. Trapnell, C. et al. Differential gene and transcript expression analysis of RNA- seq experiments with TopHat and Cufﬂinks. Nat. Protoc. 7, 562–578 (2012).

55. Mortazavi, A., Williams, B. A., McCue, K., Schaeffer, L. & Wold, B. Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat. Methods 5, 621–628 (2008).

56. Kayala, M. A. & Baldi, P. Cyber-T web server: differential analysis of high- throughput data. Nucleic Acids Res. 40, W553–W559 (2012).

57. Baldi, P. & Long, A. D. A Bayesian framework for the analysis of microarray expression data: regularized t -test and statistical inferences of gene changes. Bioinformatics 17, 509–519 (2001).

Part 1:A Dopamine Gradient Controls Access To Distributed Working Memory in The Large-scale Monkey Cortex

Targeted Memory Reactivation in REM But Not SWS Selectively Reduces Arousal Responses