Introduction

Memory traces of episodic-like events are encoded in parallel by the hippocampus and neocortex throughout the day, but their retention over time is often transient. Traces subject to consolidation are retained, whereas later memory retrieval is unsuccessful when consolidation fails or is insufficient. Consolidation in both the hippocampus and neocortex is, however, now recognised as a complex set of processes involving both “cellular” mechanisms that operate largely within individual neurons and “systems” mechanisms that include network interactions across brain areas [1–4]. An additional mechanism called “reconsolidation” enables consolidated traces to be updated, indicating that stabilization need not imply fixation [5–7]. The distinction between cellular and systems consolidation is therefore not a sharp one, for the enactment of systems consolidation (involving interactions between hippocampus and neocortex) will necessarily involve the mechanisms of cellular consolidation as well. This overlap of mechanisms contributes to the challenge of studying of how hippocampal and cortical consolidation interact.

The overarching aim of this study was to investigate the interaction of hippocampal and cortical consolidation with respect to the retention of two potentially incompatible associations. Consider the following hypothetical situation. An experimental subject, be it human or an animal model, is required to learn first one thing and then later something different that may even contradict the first thing. In the procedural domain, it is important that the new skill overrides the first one and is then expressed in isolation (e.g., learning new balancing skills when riding a bicycle). However, in the episodic domain, it can be beneficial for the subject to remember both things even when they contradict one another (as in, “I used to think that John liked Mary but I now know it is only Mary that likes John”). This raises the conceptually deep issue of when new knowledge should interfere with and so “overwrite” earlier knowledge and when two items of ostensibly contradictory knowledge should both be retained.

Morris and Doyle [8] trained rats in a hippocampal-dependent watermaze task over many days to find a hidden escape platform in the northeast corner of the pool (in practice, this location was geometrically counterbalanced). Once this memory was well established, a “reversal” procedure was instituted such that, over eight trials, the platform was hidden in the opposite southwest corner. The key variable manipulated in the experiment was the interval of time between these eight trials (30 s or 24 h). In the 24-h condition, the animals learned the reversal and thereafter always searched for the platform in the southwest corner in successive memory tests over several weeks until the memory was lost. The animals also learned the reversal in the 30 s condition and first searched persistently in the southwest corner during an initial posttraining memory test, but, without any further training, they switched to searching preferentially in the northeast corner during a memory test conducted 12 d later. The amount of training on the reversal (eight trials) was exactly matched, but arguably the opportunity for engaging hippocampal (fast) and cortical (slow) consolidation mechanisms differed as a function of the short versus long intertrial intervals, respectively. In the former case, two incompatible traces were retained; in the latter, the first memory was overwritten.

The present study builds on this preliminary observation as a first step towards a systematic analysis of the intriguing issue of interference versus mutual retention. In the hippocampus, protein synthesis–dependent cellular consolidation acts soon after encoding as a selective filter to enable traces to be retained for longer [9–12], a process now known to be enhanced by postencoding novelty [13], possibly via a ventral tegmental area-hippocampal formation feedback loop [14]. Separately, cortical consolidation can occur (especially during overnight sleep) to guide and stabilize network interactions between the hippocampus and neocortex [15–22]. This likely operates using overlapping mechanisms (i.e., both cellular and systems mechanisms) but with the passage of sleep activating distinct neural mechanisms to enable either stable, episodic-like memory traces in the cortex, the potential loss of contextual associations, and/or the successful assimilation of new information with prior knowledge [23–27]. Cortical consolidation is widely thought to be a slower process [15], but there is growing evidence that it can sometimes be initiated soon after learning and act relatively quickly, such as during sleep [2,20,27,28].

We therefore sought to create two conflicting memory traces and then identify manipulations that would favour interference and loss or dominant behavioural expression of one contradictory memory without loss of the other. Postencoding novelty in the waking state or the opportunity for sleep soon after training were two distinct behavioural manipulations used to potentiate hippocampal or neocortical consolidation, respectively. These were supplemented by pretraining to assist assimilation with prior knowledge (cortical) or an interference protocol (likely to operate preferentially in the hippocampal domain).

One complication was that the novelty condition necessitated the simultaneous use of brief sleep deprivation to distinguish it from the sleep condition, and this necessitated an additional control study to check that brief sleep deprivation itself did not alter memory performance. Another unavoidable concern was that it is unlikely that novelty or sleep act exclusively on the hippocampus or cortex, respectively. Nonetheless, while less clear-cut than would be optimal, there are grounds for believing that postencoding novelty will have a preferential impact in the hippocampus [13,29], whereas sleep has a preferential impact on the interactions between the hippocampus and neocortex [16,17]. In the context of our experimental design, we can think of the two competing memory traces as occupying each side of a children’s “seesaw.” The relative dominance of one or the other trace is then “flipped” by changing behavioural parameters of training that likely affect hippocampal and cortical consolidation preferentially [30].

These manipulations being behavioural, it was incumbent upon us to identify whether potential neural markers of consolidation, such as immediate early gene (IEG) expression, were activated differentially at these two relevant anatomical sites. Our aim here was not to compare detailed patterns of expression across hippocampal subregions or cortical brain regions, nor to conduct a comprehensive comparison of expression patterns as a function of time [31], but rather to secure preliminary measures of the impact of these manipulations in the hippocampus and a specific region of the cortex. We predicted that novelty would lead to a learning-independent increase in immediate early gene (IEG) mRNA expression in the hippocampus related to the production of plasticity-related proteins implicated in synaptic tagging and capture [32] and the consequent consolidation of hippocampal traces [29]. In contrast, sleep should trigger a relatively selective increase in cortical consolidation after learning but against the background of a time-dependent down-regulation of IEG expression unrelated to memory consolidation, resembling findings recently reported for firing-rate changes [16,17,33–35].

Results

In separate experiments, rats (n = 337) were randomly assigned for training in a spatial memory task, with brain tissue from a subset of animals analysed (blind) with respect to the expression of IEGs. An initial cohort (n = 32) was given brief spatial learning (four blocks of two trials per block) in a watermaze to one escape location, followed—after 7.5 h—by equivalent training to an opposite location (Fig 1A). These two sessions of training deliberately set up two competing memories such that memory tested much later could of be of one memory, the other, or of both. The animals learned each location in a comparable manner across the two training sessions within a day (Fig 1B) and showed, during a memory test 7 d later, significantly above chance swim time in predefined zones centred on the two platform locations (t = 2.45, df 26, p = 0.022; Fig 1C). While there was a trend favouring the more recently trained location, there was no significant difference in acquisition of memory associated with the two sessions (p > 0.7; Fig 1C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. A watermaze protocol to examine competition of two memory traces.

(A) Rats learned two opposite hidden platform locations in a watermaze over two successive sessions (four blocks of two trials per block, separated by 7.5 h) with a probe trial (no platform) conducted 7 d later. (B) Over both sessions, the animals decreased their escape latency (F = 21.5, df 1.7/51, p < 0.001). (C) At the 7 d memory test, the animals swam on average above chance level across zones (striped bar; t = 2.45, df 26, p = 0.022), but the trend favouring recency was not significant (p > 0.7). (D) Each session was followed by either sleep or novelty + sleep deprivation (N + SD) in a counterbalanced design. (E) Example swim path at test, with platform location followed by sleep NW (black zone) and N + SD SE (white zone) and starting location NE (green arrow). Note: based on extensive observation of swim patterns in the watermaze, the zones were deliberately designed to include an area by the side walls adjacent to one or other platform. (F) Swim time in zone followed by N + SD dominates over that followed by sleep (t = 1.97, df 52, p = 0.054, Cohen’s d = 0.54, N + SD to chance t = 2.31, df 26, p = 0.03). N + SD = novelty with sleep deprivation, NW = northwest, SE = southeast, *p < 0.05 t-test to chance. Means +/- 1 standard error of the mean (SEM). All raw data available in S1 Data.

https://doi.org/10.1371/journal.pbio.2000531.g001

Two conditions, (1) sleep and (2) novelty + sleep deprivation (N + SD), were scheduled in a counterbalanced and within-subjects manner after the first and second (competing) sessions of spatial learning. In the N + SD condition, the animals were placed in a novel environment with the repeated presentation of novel objects and other items and repeated gentle handling to prevent the animals from going to sleep. To control that the effects seen here were due to novelty and not sleep deprivation, we repeated the experiments but only with sleep deprivation by gentle handling and excluded novelty (see S11 Fig). We used a doubly counterbalanced design (with respect to both the order of platform location and of the N + SD versus sleep conditions; Fig 1D; S1 Fig). Accumulating data across animals and conditions required us to “rotate” the data matrix of half of the datasets by 180° for averaging, statistical, and graphical purposes. The swim paths in the 7 d probe test (memory retrieval) showed swim paths that moved back and forth between the two trained locations but revealed preferential search in the trained location that was followed by N + SD (representative animal in Fig 1E). The time searching in a virtual zone around the escape location was above chance for the N + SD condition but did not differ from chance in the sleep condition (N + SD t = 2.31, df 26, p = 0.03; sleep t = –0.13, df 26, p > 0.8, Fig 1F, separated for sequence [see S8 Fig]). Thus, under “baseline” conditions, and despite the opportunity for 6 further days of the animals’ routine sleep/waking cycle in the absence of further training, weak spatial learning followed by conditions favouring cellular consolidation in the hippocampus dominates the expression of memory in behaviour.

We then examined two further cohorts (n = 32 in each) in which we sought to flip the “seesaw” in one direction or the other. Experiment 1 can be thought of as having established a “default” situation in which the competing memory trace followed by N + SD is dominant over the memory trace followed by sleep (Fig 2A, Base condition). One behavioural method of flipping the seesaw was extensive preexposure of the animals to the context. Context preexposure would create prior knowledge of the extramaze cues of later watermaze learning and, we predicted, should enhance the speed and effectiveness of its cortical consolidation with relatively little effect on the trace, followed by N + SD (Fig 2A Pre-E). Context preexposure consisted of 3 d of 8 min exploration of the training context, achieved by placing the animals on a solid floor located within the watermaze (without water, but at the same height as the water would normally be, and with all extramaze cues visible). This allowed the animals to explore the environment and should have enabled them to learn about the relative location of extramaze cues. Subsequent training in the watermaze might then trigger learning in which the location of the hidden platform is rapidly assimilated within a previously established context representation. Specifically, we predicted “fast” systems consolidation [36], in much the same manner as can happen when animals have previously learned schematic knowledge [27].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Distinct protocols favour cellular or systems consolidation.

(A) The experimental design was, metaphorically, like a children’s seesaw. In addition to the baseline experiment (Fig 1), a further group was preexposed to the watermaze extramaze cues with a dry-land inlay for 3 d prior to the experiment (Pre-E) and a third group to an interference trial 24 h after training (Int). Sleep and N + SD followed in a counterbalanced manner. The “dominant” trace is the lightest, rising above the other conflicting trace. (B) In the zone analysis, a group x condition interaction was seen (F = 3.3, df 2/82, p = 0.043, with post hoc linear contrast p < 0.05, both controlling for sequence of consolidation-type, d = 3.37 *p = 0.03, p = 0.025 and **p = 0.01 t-test to chance). (C) Spatial dwell time maps of the watermaze at 7 d test with warm colours indicating higher average dwell time. Note systematic shift of the “hot” area across the different group protocols. Example paths from individual animals are shown in S18 Fig. (D) The cluster analysis showed two clusters in the baseline and preexposure groups but only one cluster in the interference group. Red cross indicates platform position, grey cross indicates cluster centre. Each point represents a local maxima derived from the dwell time maps. For panels C and D, the location followed by sleep is graphically presented at NW (black) and followed by N + SD at SE (white) but was counterbalanced. (E) Shows the number of animals that had above 20% swim time in the zone surrounding the platform location for each condition and experiment. Only in Int was a significant effect seen, with more sleep animals being above and more N + SD animals below chance level (Fisher’s exact test p = 0.019). (F) To ensure a specific effect of novelty, we ran two further cohorts (n = 32 in each) repeating the Base and Int experiments but this time with sleep deprivation by gentle handling instead of novelty (SD). There was a significant novelty x experiment x condition interaction (F = 3.7, df 1/116, p = 0.033). N + SD = novelty with sleep deprivation, NW = northwest, SE = southeast, Means +/- 1 SEM; all raw data available in S1 Data.

https://doi.org/10.1371/journal.pbio.2000531.g002

The other method of influencing trace dominance was interference by removing access to the hidden platform during a test of memory 24 h after training. This involved a 120 s swim trial in the watermaze with no platform present, a procedure that would be likely to have the greatest effect on the “dominant” trace [30]. The consequences of such additional postacquisition learning should serve to diminish the capacity of the earlier trained N + SD trace to dominate the sleep memory with which it is competing (Fig 2A Int) and might even alter permanently the hippocampal representation of where escape may be possible.

The results confirmed these predictions (Fig 2B). Context preexposure enhanced performance after watermaze training to a level at which postencoding sleep enabled rapid assimilation of new information about the escape location during the four blocks of swim trials (Fig 2B). Conversely, postlearning interference had a greater impact on information that had been subject to strengthening by the prior N + SD condition (Fig 2B). Analysis of all three conditions—baseline, pre-exposure and interference—revealed a significant conditions x consolidation type interaction (F = 3.2, df 2/83, p = 0.043, d = 3.37 controlling for sequence of consolidation type).

This statistical interaction justified deeper analyses of the data—specifically, to compare the impact of masking (through trace dominance) and erasure (through interference or new learning). From both the baseline study and the additional experiments, we derived “dwell time” (heat) maps (Fig 2C) and a related but statistically distinct “cluster analysis” (Fig 2D). Dwell time maps show a summated occupancy of locations across the pool, with the “hot” colours reflecting greatest time. The cluster analysis, in contrast, identifies local maxima of occupancy even when these occur at levels well below the absolute maximum, signified by the hottest colour in the dwell time map. Using a gap-statistic method [37], the optimal number of clusters of occupancy could be calculated from the spread of all identified local maxima locations (see Methods). The cluster analysis is important as it has the potential to reveal the existence of a spatial memory (a focused cluster) even in circumstances in which its behavioural expression is masked by a separate dominant memory; similarly, it can reveal its absence. For graphical purposes, the sleep condition is displayed as northwest while the N + SD condition is shown as southeast (but this was fully counterbalanced, S1 Fig). For the baseline condition, the dwell time map (Fig 2C left) showed most searching in the N + SD location, complementing the previous zone-analysis, but the cluster-analysis (Fig 2D left) revealed memory for two separate locations. This indicated that the dominant behavioural expression of the memory consolidated by posttraining N + SD only masked the other memory with which it was in direct competition with respect to the control of behaviour. The presence of a significant negative correlation between swim time in the respective zones for N + SD and sleep offers further evidence for memory competition rather than erasure (S9 Fig). The preexposure condition revealed a more symmetrically balanced heat map and detectable clusters at or near both escape platform locations (Fig 2C and 2D, middle). In contrast, giving the animals 24 h interference trial, in which they could learn that neither platform was available, reduced the intensity of “hot” colours in the dwell time map and resulted in a complete loss of any clustering around the platform location followed by N + SD (Fig 2C and 2D, right). In this case, however, there are grounds for suspecting a loss of the N + SD location rather than masking because the cluster analysis identifies only one cluster centred on the sleep location with no local maxima for the N + SD being detectable. Our manipulations have had differential effects.

The distinction between “masking” and “erasure” is subtle but important. To further substantiate this putative dissociation, a third analysis was performed based on the swim time in the zones surrounding the platform locations (Fig 2E). For each experiment and group, we divided the animals into good performers above 20% swim time in zone and poor performers below it. The number of good and poor performers did not differ between the sleep and N + SD conditions for the Base and Pre-E experiment. In contrast, after interference, significantly more good performers were present in the sleep condition, and more poor performers were present in the N + SD condition (Fisher’s exact test p = 0.019). This supports the idea that the masking of memory traces takes place in the baseline experiment, but when the opportunity is given to learn that no escape is possible, memory erasure can both occur and occur selectively.

N + SD animals were placed in a novel environment and subjected to the repeated presentation of novel objects and items, coupled with gentle handling to prevent them going to sleep. To check whether the effects of N + SD were due to novelty, rather than sleep deprivation, two further cohorts (n = 32 in each) were run that repeated the Base and Int experiments but with sleep deprivation by gentle handling and explicitly excluding novelty (see Fig 2F and S11 Fig). In contrast to the N + SD experiments, sleep deprivation by gentle handling did not lead to the interaction effects seen across experiments and conditions. In fact, when compared with the main experimental condition, there was a significant novelty (N + SD/SD) x experiment (Base/Int) x condition (sleep/sleep deprivation) interaction (F = 3.7, df 1/116, p = 0.033).

The next important step was to complement these behavioural observations with measures of likely markers of consolidation, namely immediate early genes. If the account we have offered so far in terms of differential impact of hippocampal versus cortical consolidation is valid, there should be analogous changes at the IEG level following these same manipulations. We therefore sought to observe the impact of N + SD and sleep on genes likely relevant to neuronal plasticity. We focused on selected markers to represent activity-related and plasticity-related processes: immediate early gene expression of cFos mRNA as an indicator of activity [38] and Zif-268 and Arc mRNA as an indicator of plasticity [1]. These were all monitored in both the hippocampus and medial prefrontal cortex (mPFC).

There were two key decisions about the experimental design. First, the experimental method involved in measuring mRNA expression in association with memory encoding and consolidation was real-time quantitative PCR analysis. We chose qPCR in contrast to immunocytochemistry as we sought to achieve an exact and quantitative measure of the extent and time course of gene transcription. This enabled more complex statistical models with multiple within- and between-subject comparisons at the cost of being unable to compare and contrast different subregions within the hippocampus or areas of the cortex. Second, behavioural training was necessarily to a single escape location in the watermaze because the “trace-competition” design could not be used unambiguously. Additionally, it was important to measure IEG activation at defined time points soon after encoding, and this also precluded the use of training to two escape platforms several hours apart. The consolidation-specific effects of the two conditions (sleep and N + SD) were investigated as well as general effects of the condition. To achieve this aim, specific contrasts in our analysis were chosen. For consolidation-specific effects, the contrast was between animals that did or did not experience the watermaze, but we controlled for their condition with both groups having either sleep or N + SD (Fig 3). In contrast, for general condition effects, the contrast was between animals that experienced the watermaze (and had either sleep or N + SD) with awake, home cage control animals (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. RT-qPCR analysis consolidation.

(A) To elucidate learning specific effects, we directly compared animals 0.5 h and 5 h after having a learning session in the watermaze (WM) or remaining in the home cage (no exposure to the watermaze [NoWM]) with either sleep or N + SD in the 5 h period. Positive values reflect higher gene expression in WM animals, and negative values reflect higher gene expression in NoWM. At encoding (0.5 h), a significant gene effect was seen, with cFos showing higher gene expression changes than both other genes across both brain areas (F = 19.8, df 1.1/6.6, p = 0.01, post hoc simple, contrasts cFos versus Arc F = 10.3, df 1/4, p = 0.033 and cFos versus Zif F = 56.5, df 1/4, p = 0.002). 5 h later (Sleep and N + SD), a significant brain area x condition interaction was seen (F = 6.9, df 1/24, p = 0.015, post hoc linear contrast p = 0.015), with WM showing higher gene expression than NoWM in the HPC after N + SD, but lower in sleep. Thus, time (0.5 and 5 h) showed a differential effect across brain areas and condition (B and C). (B) In the hippocampus, only sleep and not N + SD showed a significant decrease in gene expression from 0.5 to 5 h (Sleep: F = 12.2, df 1/8, p = 0.008, d = 2.48; N + SD: F = 1.6, df 1/8, p > 0.2, d = 0.90). (C) In contrast, in the mPFC, the opposite pattern was seen, with a significant decrease in gene expression only in N + SD (sleep: F = 0.3, df 1/8, p > 0.5, d = 0.42; N + SD: F = 5.6, df 1/8, p = 0.046, d = 1.67). HPC = hippocampus, mPFC = medial prefrontal cortex, IEG = immediate early gene, N + SD = novelty with sleep deprivation. Means +/- 1 SEM. All raw data available in S1 Data.

https://doi.org/10.1371/journal.pbio.2000531.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. RT-qPCR analysis conditions.

(A) Left: Rats learned one platform location in the watermaze followed by either sleep or N + SD. The animals were humanely killed (arrows) at different time points throughout the procedure, capturing consolidation steps across the two conditions (Con) to allow for qPCR analysis of IEG expression in the mPFC and HPC. Right: To exclude learning-specific effects, we directly compared animals at 5 h after having a learning session in the watermaze (WM) or remaining in the home cage (NoWM) with either sleep or N + SD in the 5 h period. All animals are compared to home cage controls (HCC). (B) In comparison to a neutral wake condition (HCC), N + SD showed elevated gene expression that was sustained throughout the N + SD period in HPC (yellow background). In contrast, sleep showed a decrease. There was both a condition x brain area and condition x time interaction seen during the consolidation period (F = 13.1, df 1/24, p = 0.001 with post hoc linear contrast p < 0.001; F = 6.1, df 2/24, p = 0.007, respectively). These effects were seen independent from any experiences in the watermaze (right). (C) In the mPFC, sleep was associated with a decrease in IEG, which was monotonic with respect to time in the mPFC (grey background; 2, 4, and 6 h from left to right). In contrast, N + SD showed an up-regulation of expression. And again, these effects were seen independent from any experiences in the watermaze (right). HPC = hippocampus, mPFC = medial prefrontal cortex, IEG = immediate early gene, N + SD = novelty with sleep deprivation, HCC = home cage controls. Means +/- 1 SEM. All raw data available in S1 Data.

https://doi.org/10.1371/journal.pbio.2000531.g004

The study design focused on comparing IEG expression during the course of the respective postencoding N + SD or sleep manipulations in brain tissue from animals that had all learned the watermaze in a single session with tissue from animals that had not been subject to training (Fig 3). For all experiments, the brains were extracted with the hippocampus (HPC) (yellow) and mPFC (grey) immediately dissected and then snap-frozen for later analysis (Fig 3A). We compared brain tissue from animals that had experienced a learning session in the watermaze (WM) with animals that did not (NoWM; Fig 3B and 3C). This was done either directly after encoding (0.5 h) or 5 h into the consolidation period in association with postencoding sleep or N + SD (n = 30). The graphs are plotted such that positive values indicate higher gene expression in WM than in NoWM (negative values, vice versa). We chose a neutral wake control condition (NoWM) because possible alternative control conditions such as swimming in the watermaze without a platform can display alterations in IEG expression in association with stress or with incidental learning about the environment through exploration [39–41], and these confounding factors can hinder interpretation of results [42]. Furthermore, for present purposes, the critical results are the contrasts between N + SD and sleep with respect to spatial learning or its absence.

For the 0.5 h encoding condition, without subsequent behavioural manipulations, there was a significant and equivalent increase gene expression in both the HPC and mPFC relative to NoWM (Fig 3B and 3C, left). There was a significantly larger increase in cFos compared to the plasticity-associated genes (F = 19.8, df 1.1/6.6, p = 0.01, post hoc simple, contrasts cFos versus Arc F = 10.3, df 1/4, p = 0.033 and cFos versus Zif F = 56.5, df 1/4, p = 0.002). This is an important finding, as it points to substantial and rapid gene up-regulation in both brain regions at or soon after learning.

The comparison of WM and NoWM after 5 h of N + SD or sleep revealed different patterns of gene expression in hippocampus versus cortex (Fig 3B and 3C, right). In the hippocampus, the pattern was for higher gene expression with WM after N + SD and lower after sleep; in the cortex, the opposite pattern prevailed. The ANOVA showed a significant brain area x condition interaction at the 5 h time point (F = 6.9, df 1/24, p = 0.015, post hoc linear contrast p = 0.015). This interaction is important and most easily illustrated by focusing on the effect of time (0.5 and 5 h) on cFos. Note that the relative level of gene expression in HPC after 5 h of N + SD (at approximately 20% greater for the animals had been trained) was not significantly less than that observed at 0.5 h (t = 2.0, df = 8, p > 0.08); a much larger and now significant decrease in relative gene expression over time was observed for sleep with gene expression being approximately 35% less in animals that had been trained (t = 4.0, df = 8, p = 0.004; Fig 3B). In contrast, if one looks at the relative impact of N + SD and sleep in cortex, the decrease in relative cFos expression across training conditions from +120% to –60% for N + SD was significant (t = 3.1, df = 8, p = 0.015), whereas for sleep the corresponding difference did not reach significance (t = 1.1, df = 8, p > 0.3; Fig 3C). The same pattern was also seen across all genes (HPC: sleep: F = 12.2, df 1/8, p = 0.008, d = 2.48; N + SD: F = 1.6, df 1/8, p > 0.2, d = 0.90; mPFC: sleep: F = 0.3, df 1/8, p > 0.5, d = 0.42; N + SD: F = 5.6, df 1/8, p = 0.046, d = 1.67).

The differential effects of N + SD and sleep on gene expression, offset by consolidation-associated changes, warranted further study into the more general effects of N + SD and sleep. Specifically, in a separate cohort of 35 animals, IEG expression was measured in hippocampus (yellow) or prefrontal cortex (grey) at 2 h, 4 h, or 6 h after encoding (Fig 4A), during which postencoding N + SD (white) or sleep (black) were occurring in different subgroups of animals (conditions). This time, these were referenced to a neutral, awake home cage control (HCC) to enable comparison of the general change in gene expression across both behaviours (n = 5 for each “con” subgroup and n = 5 for the “HCC” control). In the separate cohort of trained and untrained animals (n = 25), the time-point 5 h was examined only.

Animals allowed to sleep for varying time periods showed a monotonic, time-dependent decrease in gene expression relative to HCCs at the neocortical but not the hippocampal site (for an explanation of the “fold-change” measure, see Methods RT-qPCR). In contrast, the impact of recurring exposure to novelty coupled to sleep deprivation was associated with positive changes in gene expression in both brain regions that did not change over time. An overall ANOVA of gene expression—including condition (Sleep, N + SD), time (2, 4, and 6 h), brain area (HPC, mPFC), and gene (cFos, Zif-268 and Arc)—revealed significant interactions for condition x time and condition x brain area (F = 13.1, df 1/24, p = 0.001 with post hoc linear contrast p < 0.001; F = 6.1, df 2/24, p = 0.007, respectively). The same general pattern was seen for all genes, but Arc expression drove the result in the cortex as indicated by a significant condition x time x brain area x gene interaction. Ongoing novelty for 2, 4, or 6 h was associated with an apparently stable up-regulation of IEG expression with no relevant interactions.

In an additional control analysis, we also compared the 5 h WM/NoWM and sleep/N + SD conditions to a neutral wake home cage control and could replicate the general decrease in gene expression in sleep and increase in N + SD, which was thus shown to be learning independent and mainly reflected the current experience (sleep or novelty) of the animal (Fig 4 right).

Discussion

The overarching aim of this study was to investigate the interaction of hippocampal and cortical consolidation with respect to the retention of two potentially incompatible associations. The key findings are that (1) rats can learn and retain two incompatible spatial memories without interference, with one memory dominating expression in certain circumstances; (2) postencoding sleep or novelty plus sleep deprivation have differential effects on memory consolidation and the resulting pattern of behavioural expression; this differential effect is genuinely because of sleep versus novelty, as a brief period of sleep deprivation alone was shown to have minimal effects on memory consolidation; and (3) novelty and sleep are associated with differential patterns of up-regulation of immediate early genes in the hippocampus and the medial prefrontal cortex. We shall argue that these findings support the idea that the stabilization of memory traces within cell assemblies reflects a time-overlapping interaction of two interdependent processes—hippocampal and cortical consolidation.

Methods

Supporting Information

Acknowledgments

We are grateful to Richard Fitzpatrick, Olivia Macquire, Chloe Stephenson-Wright, and Jane Tulloch for technical assistance and Tomonori Takeuchi, Vassilis Beglopoulos, and Antonis Asiminas for advice on the qPCR analysis. We would also like to thank Atsuko Takashima and Jennifer Tudor for helpful comments on the manuscript.