Early Tagging of Cortical Networks Is Required for the Formation of Enduring Associative Memory

See allHide authors and affiliations

Science  18 Feb 2011:
Vol. 331, Issue 6019, pp. 924-928
DOI: 10.1126/science.1196164


Although formation and stabilization of long-lasting associative memories are thought to require time-dependent coordinated hippocampal-cortical interactions, the underlying mechanisms remain unclear. Here, we present evidence that neurons in the rat cortex must undergo a “tagging process” upon encoding to ensure the progressive hippocampal-driven rewiring of cortical networks that support remote memory storage. This process was AMPA- and N-methyl-d-aspartate receptor–dependent, information-specific, and capable of modulating remote memory persistence by affecting the temporal dynamics of hippocampal-cortical interactions. Post-learning reinforcement of the tagging process via time-limited epigenetic modifications resulted in improved remote memory retrieval. Thus, early tagging of cortical networks is a crucial neurobiological process for remote memory formation whose functional properties fit the requirements imposed by the extended time scale of systems-level memory consolidation.

Memories for facts and events are not acquired in their definitive form but undergo a gradual process of stabilization over time (13). According to the so-called standard theory of systems-level memory consolidation, the hippocampus (HPC) is believed to integrate, in the form of an anatomical index, information transmitted from distributed cortical networks that support the various features of a whole experience (4). Upon encoding, the HPC rapidly fuses these different features into a coherent memory trace. Consolidation of this new memory trace at the cortical level would then occur slowly via repeated and coordinated reactivation of hippocampal-cortical networks in order to progressively increase the strength and stability of cortical-cortical connections that represent the original experience. Over days to weeks as memories mature, the role of the HPC would gradually diminish, presumably leaving cortical areas to become capable of sustaining permanent memories and mediating their retrieval independently (4, 5). Using brain imaging, we have provided evidence supporting the time-limited role of the HPC as a consolidation organizing device of remote memory in the cortex (5, 6). Yet the nature and dynamics of plasticity phenomena as well as the neuronal constraints within hippocampal-cortical networks responsible for the formation of remote memories have remained elusive.

To pinpoint the post-learning mechanisms underlying the hippocampal-cortical dialogue during the course of systems-level memory consolidation, we used the social transmission of food preference (STFP) paradigm, which involves an ethologically based form of associative olfactory memory (7). In this task, rats learn, within only one single interaction session of 30 min, about the safety of potential food sources by sampling the odor of those sources on the breath of littermates (8). After establishing the necessary role played by the HPC in acquisition of associative olfactory memory (9) (fig. S1), we trained independent groups of rats and tested them for memory retrieval either 1 day (recent memory) or 30 days (remote memory) later. Interaction of experimental observer rats with a demonstrator rat fed with cumin reversed the innate preference typically expressed by control observer rats (food preference), which interacted with a demonstrator rat fed with plain food (fig. S2A). The acquired memory for cumin was robust and long-lasting, which makes the STFP task particularly suitable to studying the processes underlying remote memory formation. Cellular imaging of the activity-dependent gene c-fos (6) coupled to region-specific pharmacological inactivation of HPC by using tetrodotoxin or the AMPA receptor antagonist 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) revealed a transitory role of this region in remote memory storage of associative olfactory information (figs. S2, B to D, and S3). At the cortical level, we focused our analyses on the orbitofrontal cortex (OFC) because of its privileged role in processing the relevance of associative olfactory information (fig. S2E) (10). The observed hippocampal disengagement was associated with a concomitant increase in Fos immunoreactivity in the OFC (fig. S2B). Accordingly, inactivation of the OFC selectively impaired remote memory retrieval (figs. S2, C and D, and S3). Although a broad cortical network is likely to be involved in the processing of remote associative olfactory memory, our results identify the OFC as a critical node within this network. Concurring with this, we found neuronal networks of the OFC to undergo time-dependent morphological changes at pre- and postsynaptic sites (fig. S4).

To determine whether such a progressive synaptic remodeling in the OFC during remote memory storage is driven by the HPC, we chronically inactivated the dorsal HPC during two critical post-learning periods: an early (from day 0 to day 12) and a late (from day 15 to day 27) time window (Fig. 1A). This reversible pharmacological approach enabled us to target selectively post-learning consolidation mechanisms without interfering with retrieval processes (fig. S3). Remote memory retrieval examined at day 30 was impaired when hippocampal activity was silenced during the early, but not the late, post-learning period (Fig. 1B and fig. S5). This early hippocampal dysfunction also completely abolished the late development of dendritic spine growth on OFC neurons (Fig. 1B). Delaying intrahippocampal infusion of CNQX by 1 day so as to prevent any confounding interference with cellular consolidation mechanisms (1, 2, 5) that were triggered immediately upon encoding yielded similar results (fig. S6). Therefore, these findings point to the crucial, but time-limited, involvement of the HPC in driving at least in part wiring plasticity in the OFC [supporting online material (SOM) text].

Fig. 1

Conjoint functional participation of HPC and OFC in the early, but not late, post-acquisition period is required for systems-level consolidation of remote associative olfactory memory. (A) Experimental design depicting region-specific pharmacological inactivation of HPC or OFC with CNQX during either the early or late post-acquisition periods after social interaction. Control rats received artificial cerebrospinal fluid (aCSF) as vehicle. (B) Early, but not late, hippocampal inactivation impaired remote memory retrieval at day 30 (left, treatment × delay interaction, F1,31 = 4.74; P < 0.05) and prevented the associated increase in spine density along basal dendrites of OFC neurons that is typically observed in aCSF rats (right, treatment × delay interaction, F1,15 = 5.56; P < 0.04). This interaction was close to reaching significance for apical dendrites (F1,15 = 3.65; P < 0.07). (C) Both early and late inactivation of OFC impaired remote memory retrieval (left, main treatment effect F1,36 = 15.26, P < 0.001) and prevented structural plasticity in the OFC compared with aCSF rats (right, main treatment effect, apical, F1,16 = 17.82, P < 0.01; basal, F1,16 = 11.25, P < 0.01). The dotted line represents performance or cortical spine density in food preference controls. *P < 0.05 versus aCSF, n = 5 to 12 rats per group.

Because the establishment of cortical memory is critically dependent on coordinated hippocampal-cortical interactions (15), we predicted that cortical disruption during the early post-learning period should preclude cortical processing of hippocampal inputs and therefore lead to memory degradation or erasure. We thus applied to the OFC the same experimental design as for the HPC (Fig. 1A). Silencing neuronal activity during the early post-learning period impaired remote memory retrieval and structural plasticity, indicating that early cortical activity is required for subsequent maturation and stabilization of the memory trace (Fig. 1C and figs. S5 and S6). More unexpected was the finding of similar alterations after OFC inactivation during the late post-learning period (Fig. 1C)—a time window reported above as being no longer hippocampal-dependent. Thus despite hippocampal disengagement, additional cortical-cortical interactions appear to be required to ensure stability of cortical networks. Alternatively, disrupting cortical activity during the late phase of the consolidation process may have interfered with neuronal mechanisms underlying memory maintenance (11, 12). To rule out the possibility that the observed memory disruption was due to a nonspecific impairment in hippocampal or OFC function, we established that the same chronically inactivated animals could relearn and consolidate when tested 1 day (HPC groups) or 30 days (OFC groups) after a novel interaction with a demonstrator rat fed with a different flavor (fig. S7).

Hippocampal-cortical projections are topographically organized, which makes the HPC ideally suited for organizing structural changes in the cortex (13, 14). How then does the HPC manage to activate the proper experience-relevant set of cortical neurons during the course of systems-level consolidation? Insights can be gained from a repertoire of cellular and molecular mechanisms that enables activated synapses to be “tagged” within the few hours that follow an event (15, 16)—a time course compatible with cellular consolidation (2, 5). To examine the relevance of the synaptic tagging framework to the process of systems-level memory consolidation that operates over much longer time windows, we carried out four complementary experiments (SOM text). Building on the hippocampal indexing theory (14), we hypothesized that tagged synapses should contribute to the formation of a hippocampal index that in turn would act as a coincidence regenerator for activating and strengthening over time the relevant set of distributed cortical cell assemblies (17). Thus, we predicted that hippocampal and cortical networks should be tagged simultaneously at the time of encoding, albeit possibly through different mechanisms. Because the HPC is crucial for acquisition of the STFP task, interfering with hippocampal function during encoding will inevitably result in memory retrieval impairment, which is a deleterious effect observed in most hippocampal-dependent tasks (fig. S1) (5). To circumvent this confounding factor, we focused on the OFC and first examined whether its inactivation immediately before social interaction interfered with memory retrieval tested 7 days later. In accordance with previous findings (18), memory retrieval at this early time point was unaffected, therefore indicating that the OFC is not necessary for acquisition of the STFP task (Fig. 2A and SOM text). Reliance only on the olfactory component of the STFP paradigm was insufficient to explain this lack of effect (fig. S8). In sharp contrast, a similar inactivation resulted in impaired memory retrieval probed at a longer time point (day 30) (Fig. 2A) and completely abolished the later occurrence of dendritic spine growth on OFC neurons (Fig. 2B). Thus, impairing the tagging of cortical networks at the time of encoding most likely interfered with the post-acquisition hippocampal-cortical dialogue, preventing the HPC to activate the adequate pattern of cortical neurons and synapses recruited during the initial learning episode. This finding identifies early cortical tagging as a potentially critical process responsible for the progressive embedding of enduring memories within cortical networks.

Fig. 2

Early tagging of OFC neurons is a prerequisite for the formation of remote associative olfactory memory. (A) Intra-OFC infusion of CNQX before social interaction (day 0) impaired remote memory formation assessed at day 30 while sparing retrieval of recent memory probed at day 7 (treatment × delay interaction, F1,31 = 5.67; P < 0.05). (B) OFC inactivation upon encoding prevented the late development of structural plasticity seen in this region in aCSF controls. OFC inactivation reduced cortical spine density along both apical (left, t9 = 2.66; P < 0.05) and basal (right, t9 = 2.31; P < 0.05) dendrites and resulted in a lower proportion of neurons exhibiting intermediate and high spine density (bottom). (C) Injecting AP-5 into OFC before social interaction resulted in impaired remote memory retrieval at day 30 (F1,14 = 17.39; P < 0.001). (D) Intra-OFC infusion of CNQX at the time of the second interaction (cumin) impaired selectively remote memory for this second flavor while sparing remote memory for the first one (cocoa) (treatment × flavor pair, F1,11 = 13.62; P < 0.01). (Bottom left) Schematic of OFC neurons presumably tagged upon encoding of the two olfactory associative informations. Blue, tagged neurons during cocoa interaction; crossed green, neurons whose tagging was prevented by CNQX during cumin interaction. (E and F) Although extensive OFC inactivation impaired memory retrieval at both day 15 and 30 (main treatment effect, F2,37 = 12.0; P < 0.01), partial inactivation affected remote memory only at day 30 (main treatment × delay interaction, F2,38 = 3.42; P < 0.05). (G and H) Extent of OFC inactivation upon encoding differentially affected the post-acquisition kinetics of hippocampal and cortical activation assessed by means of Fos counts relative to controls (dotted line indicates 100%). Persistence of hippocampal activity at day 15 after partial OFC inactivation [(H) top, main treatment, F1,24 = 13.32; P < 0.01] was not observed after extensive inactivation [(G) top, F1,22 = 0.87; P > 0.30, not significant]. Both types of OFC inactivation prevented the functional recruitment of this region seen over time in aCSF controls (main treatment, extensive, F1,22 = 30.65, P < 0.0001; partial, F1,24 = 12.79, P < 0.01). *P < 0.05 versus aCSF; n = 5 to 10 rats per group.

However, the validity of such a neurobiological process with respect to synaptic plasticity and remote memory formation rests in demonstrating additional key features, that is, (i) N-methyl-d-aspartate (NMDA) receptor dependency, (ii) information specificity, and (iii) modulation of memory persistence. As for AMPA receptors (Fig. 2A), selectively blocking NMDA receptors in the OFC led to a robust impairment in remote memory retrieval at day 30 (Fig. 2C). If cortical tagging is information-specific, then disrupting this process should predominantly prevent the formation and stabilization of one given memory trace but enable systems-level consolidation of another one. We submitted rats to two consecutive social interactions by using two different flavors and inactivated the OFC only before the second interaction, thus ensuring the occurrence of tagging of cortical cell assemblies coding for the first olfactory information (Fig. 2D). Remote memory was probed 30 days after each interaction. OFC inactivation selectively impaired retrieval for the second flavor while sparing retrieval for the first one (Fig. 2D). Therefore, cortical tagging appears as highly specific for a given memory trace. This property could contribute to minimizing interference with the processing of other information that is already engaged in the consolidation process and needs to be interleaved with existing cortical knowledge or mental schemas (19).

We next tested the ability of the cortical tagging process to modulate memory persistence by manipulating its efficacy. Because multiple cell assemblies are thought to act in concert during memory processing, we reasoned that the three-dimensional extent (partial versus extensive) of the OFC inactivation should differentially affect the number of interconnected neurons tagged upon encoding and potentially alter the efficacy of hippocampal-cortical interactions in the guidance of remote memory formation. To assess the time course of memory persistence, independent groups of rats were tested at 7, 15, and 30 days after social interaction (Fig. 2, E and F). Extensive OFC inactivation upon encoding impaired memory retrieval regardless of the two longer retention delays, indicating a rapid decay of the associative memory (Fig. 2E). In contrast, partial silencing of OFC activity was deleterious only at the 30-day interval (Fig. 2F). Manipulating cortical tagging affected the temporal dynamics of hippocampal-cortical interactions. The preserved memory at day 15 after partial OFC inactivation was indeed associated with an unusual maintenance of hippocampal activation at this time point, possibly to compensate for the absence of gradual recruitment of the OFC normally seen in control animals over the 30-day period (Fig. 2, G and H, and SOM text). Thus, despite the potential of the HPC to act as a temporary memory buffer, the alteration of cortical tagging led to an unstable memory trace that was more prone to degradation. This indicates that synaptic tags may serve as an early and persistent signature of activity in the cortex that is necessary to ensure the progressive rewiring of cortical networks that support remote memory storage.

Because chromatin remodeling events are crucial enduring regulatory mechanisms involved in gene transcription, we examined whether the setting of synaptic tags triggered specific signaling cascades leading to histone-tail acetylation in the OFC, a modification that modulates memory function (2022). Increased acetylation of histone H3 in this cortical region was memory-specific (Fig. 3A and fig. S9). We next blocked upstream the mitogen-activated protein kinase/extracellular-signal regular kinase (MAPK/ERK) cascade in the OFC immediately before social interaction. Downstream of MAPK/ERK, we also targeted the nuclear mitogen- and stress-activated protein kinase 1 (MSK1), which acts as a main regulating gateway of histone H3 modifications (23). Blocking activation of these targets upon acquisition prevented the associated increase in histone H3 acetylation (fig. S10), impaired remote memory assessed 30 days later (Fig. 3B), and abolished the late development of cortical structural plasticity (fig. S11). These effects are consistent with that observed after blockade of NMDA receptors whose activity is coupled with the activation of the MAPK cascade (Fig. 2C and fig. S11). Although confirming the important role played by the MAPK/ERK pathway in regulating gene expression associated with remote memory processing (24), these results identify MSK1 as one important nuclear mediator of the cortical tagging process. Because the observed STFP-induced acetylation in the OFC was transitory but pharmacologically modulable (Fig. 3C and figs. S10 and S12), we investigated whether maintenance of a higher level of acetylation would translate into a potentially easier-to-retrieve remote memory. Starting immediately upon completion of the social interaction, we performed repeated intracortical infusions of the histone deacetylase inhibitors sodium butyrate (NaB) or Trichostatin A during the early post-acquisition period. This resulted in improved remote memory retrieval probed 30 days later (Fig. 3D and fig. S12) and in a higher level of expression of synaptophysin in the OFC (fig. S13). Administering NaB during the late post-acquisition period was ineffective (Fig. 3D). Although cortical tags have to be set upon encoding and persist thereafter, these findings indicate that it may be possible to facilitate remote memory formation and retrieval by potentiating, in a time-dependent manner, the epigenetic mechanisms that underlie their formation and subsequent regulation (SOM text). Efficacy of this type of pharmacological manipulation matched the hippocampal-dependent phase of the memory consolidation process. Thus, cortical tags may act as the scaffolding to support the progressive hippocampal-driven embedding of permanent memories into cortical networks via “weight” (changes in the efficacy of synaptic transmission between existing synapses) and wiring plasticity (25), potentially making these memories more resistant over time when the initial scaffolding is reinforced (SOM text).

Fig. 3

The tagging process in the OFC triggers specific signaling cascades involved in the regulation of chromatin remodeling. (A) Experimental (EXP) rats showed increased histone H3 acetylation at lysine 9 in the OFC compared with food preference (FP) controls 1 hour after social interaction (F1,10 = 19.72; P < 0.01). (B) Intra-OFC infusion of the MAPK/ERK kinase inhibitor U0126 or the MSK1 inhibitor H89 before social interaction impaired remote memory retrieval assessed 30 days later (F2,18 = 4.03; P < 0.05). (C) Intra-OFC infusion of the histone deactelylase inhibitor NaB immediately upon completion of social interaction increased the level of histone H3 acetylation assessed 4 hours after acquisition compared with aCSF rats (F1,9 = 77.62; P < 0.0001). (D) Experimental design used to investigate the effects of maintaining in the OFC the level of histone H3 acetylation during the early or late STFP post-acquisition periods on remote memory retrieval at day 30 (left). Intra-OFC infusion of NaB improved remote memory only when delivered during the early, but not late, post-acquisition period (right, treatment × delay interaction, F1,35 = 4.27; P < 0.05). *P < 0.05 versus FP; *P < 0.05 versus aCSF; n = 5 to 12 rats per group.

Replay of encoding-related activity during phases of sleep has emerged as a core mechanism for driving structural changes within hippocampal-cortical neuronal networks (26). In this context, our findings identify early tagging of cortical networks as a crucial process for the formation of enduring memories and suggest that synaptic tags contribute actively, although not exclusively, to the development but also the maintenance of this form of plasticity in the cortex (SOM text) (11, 12, 27). The identity of synaptic tags needs to be further explored, although potential candidates have started to emerge (28, 29). Given the complex nature of memory traces that typically integrate the various modalities of an event, it is likely that a plethora of synaptic molecules act in concert to capture the various patterns of synaptic activity elicited upon encoding. These tags may then be recruited differentially during the course of systems-level consolidation depending on the status of existing knowledge in the cortex (2, 19).

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S13


References and Notes

  1. Materials and methods are available as supporting material on Science Online.
  2. We thank T. Durkin for comments on earlier drafts of the manuscript and T. Maviel, L. Restivo, M. Ammassari-Teule, C. Saugier, L. Decorte, and A. Faugère for their experimental contribution or technical advice. This work was supported by the CNRS, Fondation Simone et Cino del Duca (B.B. and O.L.G.), Fédération pour la Recherche sur le Cerveau (B.B.), and the Agence Nationale de la Recherche (ANR grant 09-MNPS-005-01 to B.B.).

Stay Connected to Science

Navigate This Article