Research Article

Independent Cellular Processes for Hippocampal Memory Consolidation and Reconsolidation

See allHide authors and affiliations

Science  07 May 2004:
Vol. 304, Issue 5672, pp. 839-843
DOI: 10.1126/science.1095760

Abstract

The idea that new memories undergo a time-dependent consolidation process after acquisition has received considerable experimental support. More controversial has been the demonstration that established memories, once recalled, become labile and sensitive to disruption, requiring “reconsolidation” to become permanent. By infusing antisense oligodeoxynucleotides into the hippocampus of rats, we show that consolidation and reconsolidation are doubly dissociable component processes of memory. Consolidation involves brain-derived neurotrophic factor (BDNF) but not the transcription factor Zif268, whereas reconsolidation recruits Zif268 but not BDNF. These findings confirm a requirement for BDNF specifically in memory consolidation and also resolve the role of Zif268 in brain plasticity, learning, and memory.

The long-standing hypothesis of memory consolidation proposes that fragile, dynamic memory traces are converted into stable, long-term memory (LTM) gradually over time (13). Thus, internal representations of the external world eventually become encoded and stored as persistent molecular and/or structural modifications (4). The central dogma of the permanence of LTM has been challenged by evidence showing the disruption of what are apparently fully consolidated memories when the memory is retrieved, or reactivated, immediately before treatment with various amnestic agents or to behavioral manipulations (5). The recall of a memory thus appears to place it into an active and labile state, from which it is reconsolidated back into an inactive and stable state [(5, 6), but see (7, 8)]. This reconsolidation has been suggested, but not proven, to involve mechanisms similar to the consolidation process for the memory trace to be restabilized (9) and its retrievability to be enhanced (10), or for new information to be incorporated into the old memory trace (11). This view is supported by the similarities in some of the cellular processes of consolidation and reconsolidation. Both are blocked by the protein synthesis inhibitor anisomycin (9, 1216), and both require the activation of the transcription factors CREB (17) and Zif268 (18, 19). However, there are also differences in the temporal profile and susceptibility of memories to disruption after acquisition and retrieval (12, 13, 16, 2023), and reconsolidation does not engage all of the molecular mechanisms involved in consolidation (24). These data are consistent with reconsolidation being at most a partial recapitulation of consolidation processes. However, direct comparisons between consolidation and reconsolidation are difficult because of the different stimuli present at each, namely with a reinforcing, unconditioned stimulus (US) present during the former but not the latter.

The expression of brain-derived neurotrophic factor (BDNF) in the hippocampus is correlated with contextual fear conditioning (25), whereas hippocampal levels of Zif268 are up-regulated after the retrieval of contextual fear memories (26). Hippocampal BDNF is required for the late phase of LTP (L-LTP) (27, 28) as well as for mnemonic processes (29), although its specific involvement in acquisition, consolidation, and reconsolidation remains unclear. Hippocampal Zif268, although similarly implicated in L-LTP (18, 30), has not been strongly correlated with hippocampal-dependent learning (18, 25, 3133). Global inactivation of brain Zif268 in mutant mice prevents the consolidation of several forms of LTM under normal (but not extensive) training conditions; the reconsolidation of the overtrained memory after reactivation is impaired (18, 19). Thus, the role of Zif268 in the acquisition, consolidation, and reconsolidation component processes of LTM remains ambiguous.

We used antisense oligodeoxynucleotides (ODN) that inhibit protein expression locally in the dorsal hippocampus to show that BDNF is required for consolidation, but not reconsolidation, of contextual fear memory, whereas Zif268 is only necessary for reconsolidation, but not consolidation, of such memory. This double dissociation in the roles of BDNF and Zif268 not only clarifies the requirement for these proteins in distinct phases of hippocampal-dependent memory, but shows that cellular reconsolidation of fear memory does not engage the same processes as consolidation.

Spatial and temporal localization of acute ODN infusion. Rats implanted with chronic indwelling cannulae, targeting the dorsal hippocampus (Fig. 1, A and B), received unilateral infusions of biotinylated ODN and were perfused 90 min or 24 hours later (34) to analyze the localization and approximate stability of ODNs in the hippocampus (Fig. 1, C and D). At 90 min, the infused ODN had diffused throughout the dorsal hippocampus but had not spread into the ventral hippocampus or striatum. Only a small area of overlying cortex surrounding the cannula tract was also stained for biotinylated ODN. After 24 hours, however, the presence of biotinylated ODN in the dorsal hippocampus was greatly reduced.

Fig. 1.

Dorsal hippocampus cannulations and spread of biotinylated ODN (A) Schematic representation showing that all cannulae were located close to or within the CA1 region of the dorsal hippocampus (shaded area). (B) Section from the brain of a typical cannulated rat. The placements of the injectors are clearly visible as glial scar tracts terminating within the dorsal hippocampus. (C and D) Spread of biotinylated ODN 90 min (C) and 24 hours (D) after infusion into the dorsal hippocampus. By 90 min, the ODN diffused throughout the dorsal hippocampus and only slightly into the overlying cortex, and after 24 hours, the ODN was cleared from the hippocampus.

Consolidation is dependent on BDNF but not on Zif268. Rats infused with BDNF antisense ODN 90 min before contextual fear conditioning (34) were subsequently impaired in conditioned freezing to the training context in an LTM test 24 hours after conditioning (Fig. 2). Infusion of BDNF antisense ODN had no effect on short-term, protein synthesis–independent memory (STM) (Fig. 2). Intact STM after infusion of BDNF antisense ODN also indicates that contextual fear was acquired normally in these rats, and further that hippocampal BDNF is specifically required for the consolidation of contextual fear conditioning. LTM-impaired rats retrained normally 7 days later, and there was no evidence of spontaneous recovery of freezing behavior before footshock delivery (fig. S1). This showed that pretraining intrahippocampal infusion of BDNF antisense ODN resulted in a lasting impairment in LTM for contextual fear conditioning and, together with histological assessment (Fig. 1B), demonstrated that BDNF antisense ODN did not cause permanent damage to the hippocampus—nor, therefore, a permanent impairment in contextual fear conditioning. The impairment in consolidation was seen only when the BDNF antisense ODN was infused 90 min before conditioning, and not when infused immediately after training (fig. S2). There were no effects on locomotor activity after intrahippocampal infusion of BDNF antisense or missense ODN (fig. S3).

Fig. 2.

Contextual freezing in rats infused before conditioning with BDNF antisense ODN. Impaired LTM (24 hours after conditioning) freezing, with intact STM at 3 hours, in rats infused before conditioning with BDNF antisense ODN (1 nmol/μl) (black, n = 7) relative to BDNF missense ODN (white, n = 7). Repeated-measures analysis of variance (ANOVA) revealed a group × test interaction (F(1,12) = 5.820, P < 0.05). BDNF antisense ODN (black, n = 15) administration reduced freezing levels at the LTM test (ANOVA, *P < 0.01) relative to the BDNF missense ODN (white, n = 14) group.

The selective loss of LTM, but not STM, after infusion of BDNF antisense ODN may be a result of a mismatch between the central state present during learning and that during retrieval (8, 35), such that the deficits in LTM may be attributed to a failure of the context to retrieve adequately the contextual fear memory in the absence of BDNF antisense ODN during the recall test. However, intrahippocampal infusion of BDNF antisense ODN, 90 min before both conditioning and test, resulted in the same impairment in LTM as was observed with infusions given only before conditioning (fig. S4). Therefore, state dependency cannot explain the absence of freezing during the retrieval test. Rats infused with BDNF antisense ODN thus have a specific impairment in the consolidation of contextual fear memories.

To confirm that the impairment in contextual fear memory after intrahippocampal infusion of BDNF antisense ODN is causally related to a reduction in hippocampal BDNF, we performed two further experiments. In the first, levels of Arc, the product of a plasticity-associated immediate-early gene (36), which is regulated by BDNF signaling in the hippocampus (3739), were measured. There were no detectable levels of Arc in the dentate gyrus before conditioning. Contextual fear conditioning caused a large (factor of 10) increase in Arc protein levels in the CA1 region 6 hours after conditioning (Fig. 3A), which was substantially (66%) reduced by infusion of BDNF antisense ODN 90 min before conditioning (Fig. 3B). Direct assessment by Western blot analysis of BDNF levels in microdissected hippocampal CA1 and dentate gyrus/CA3 regions was not possible because the levels of BDNF were too low for adequate quantification. The second experiment provided direct evidence that an impairment in hippocampal BDNF signaling produced by the BDNF antisense ODN caused an impairment in consolidation. Rats infused with BDNF antisense ODN 90 min before conditioning, and with a phosphate-buffered saline (PBS) vehicle 15 min before conditioning, showed a selective loss of LTM (Fig. 3C). This impairment was completely rescued by infusion of human recombinant BDNF protein (hr-BDNF), but not PBS vehicle, with no effect on STM. Furthermore, the co-infusion of BDNF missense ODN and hr-BDNF did not elevate further the levels of freezing at either test. Therefore, the deficit in LTM does indeed reflect a loss of BDNF protein in the hippocampus and indicates that hippocampal BDNF is specifically required for the consolidation of contextual fear conditioning.

Fig. 3.

Assessment of the requirement for BDNF and Zif268 in contextual fear conditioning. (A) Contextual fear conditioning increased CA1 Arc levels 6 hours after conditioning (F(3,8) = 36.478, P < 0.001, n = 3at each time point). *P < 0.05 compared to 0 hours, †P < 0.05 compared to 2 hours and 4 hours. (B) BDNF antisense ODN (black, n = 3) attenuated the learning-related increase in Arc 6 hours after conditioning (*P < 0.01, n = 3). (C) hr-BDNF infusion completely rescued the BDNF antisense ODN-induced impairment in LTM freezing. There was a group × test interaction (F(2,17) = 18.453, P < 0.001). *P < 0. 05, comparison of rats infused with BDNF antisense ODN and PBS (black, n = 6) with rats infused with BDNF antisense ODN and hr-BDNF (gray, n = 7), or BDNF missense ODN and hr-BDNF (white, n = 8). (D) Zif268 antisense ODN (2 nmol/ μl) (black, n = 15) did not affect the acquisition of contextual fear conditioning relative to a control group infused with Zif268 missense ODN (2 nmol/μl) (white, n = 15).

Relative to infusion of Zif268 missense ODN, infusion of Zif268 antisense ODN into the hippocampus 90 min before conditioning had no effect on contextual freezing 24 hours later (Fig. 3D).

Reconsolidation requires Zif268 but not BDNF. Dorsal hippocampal protein synthesis is known to be necessary for the reconsolidation of contextual fear memories (16), but the involvement of hippocampal gene transcription has not yet been assessed. A requirement for gene transcription indicates that new mRNA coding for proteins not expressed at the time must be synthesized when reconsolidation occurs. This is a minimum requirement for the synaptic changes believed to underlie new associations (40). To test the hypothesis that reconsolidation, like consolidation (15), requires the synthesis of mRNA, we infused the transcriptional inhibitor actinomycin D into the dorsal hippocampus immediately after a retention (LTM) test, 24 hours after conditioning, which served as a reexposure trial to reactivate the memory. Infusion of either the protein synthesis inhibitor anisomycin or the transcriptional inhibitor actinomycin D produced amnesia for contextual fear in a subsequent post-reactivation retention (PR-LTM) test a further 24 hours later (Fig. 4A). This disruption of reconsolidation was critically dependent upon reexposure to the training context, an important criterion for the reconsolidation process (6), as demonstrated by the lack of amnesia when the reactivation trial was omitted (fig. S5).

Fig. 4.

Reconsolidation of contextual fear. (A) Both anisomycin (ANI, light gray, n = 7) and actinomycin D (Act D, dark gray, n = 8) impaired the reconsolidation of contextual fear. ANOVA revealed a significant group × test interaction (F(2,21) = 8.086, P < 0.005) at the PR-LTM test, with no effect of group at the LTM test. *P < 0.05 impaired relative to the PBS group (white, n = 9). (B) Zif268 antisense ODN (2 nmol/μl) (black, n = 7) impaired reconsolidation. There was a significant group × test interaction (F(1,12) = 25.314, P < 0.001) but no effect of group at the LTM test; *P < 0.05 impaired relative to the group receiving PBS vehicle infusion (white, n = 7). (C) The freezing impairment after Zif268 antisense ODN infusion was dependent on memory reactivation. When the LTM test was omitted, there was no effect of Zif268 antisense ODN (black, n = 5) or Zif268 missense ODN (white, n = 5). (D) BDNF antisense ODN (black, n = 15) or BDNF missense ODN (white, n = 15) had no effect on the reconsolidation of contextual fear memories.

To test the hypothesis that hippocampal Zif268 is involved in the reconsolidation of contextual fear memories after their reactivation (26), we infused rats with Zif268 antisense ODN 90 min before the reactivation of a contextual fear memory. Relative to infusion of control Zif268 missense ODN, infusion of Zif268 antisense ODN (2 nmol/μl) resulted in amnesia for contextual fear conditioning at the PR-LTM test, but not at the LTM test (Fig. 4B). Of critical importance was the observation that omission of the reexposure trial (LTM test) eliminated the Zif268 antisense ODN-mediated impairment in conditioned freezing (Fig. 4C). The observed amnesia was thus dependent on reactivation of the memory, which discounts the possibility that Zif268 antisense ODN impaired a late phase of consolidation of the original memory trace rather than its reconsolidation.

In contrast to its disruptive effect on consolidation, the infusion of BDNF antisense ODN had no effect on the reconsolidation of contextual fear memories, even at twice the effective concentration (1 nmol/μl) at which BDNF antisense ODN prevented consolidation (Fig. 4D).

The absence of a sequence-specific effect of Zif268 antisense ODN on conditioned freezing during the LTM test indicates that Zif268 was not important for the retrieval of contextual fear memories, nor for the expression of contextual freezing. The similar levels of conditioned freezing during the LTM test also suggest that the Zif268 antisense ODN-induced impairment at the PR-LTM test cannot be attributed to differences in the degree of fear conditioning. A nonspecific impairment of the freezing response is also unlikely, as Zif268 antisense ODN did not impair the expression of freezing 90 min after infusion (at the LTM test), nor at 24 hours after infusion when the LTM test was omitted. Intact freezing behavior was observed during a post-reactivation STM (PR-STM) test (Fig. 5), which shows that the hippocampus was functioning normally 3 hours after Zif268 antisense ODN infusion. The amnestic effect of Zif268 antisense ODN was persistent because there was no spontaneous recovery of the freezing behavior 7 days later during a second PR-LTM (PR-LTM2) test (Fig. 5). Therefore, intrahippocampal Zif268 antisense ODN infusion specifically impaired the reconsolidation of contextual fear memories.

Fig. 5.

Specificity and persistence of Zif268 antisense ODN effects. Animals infused with Zif268 antisense ODN (black, n = 5) showed contextual freezing deficits relative to animals infused with missense ODN (white, n = 7). ANOVA revealed significant overall effects of group (F(1,10) = 7.873, P < 0.02) and test (F(3,30) = 4. 000, P < 0.018). Reconsolidation was impaired in Zif268 antisense ODN infused animals. Comparing tests LTM and PR-LTM, there was a significant group × test interaction (F(1,10) = 6.690, P < 0.028), with no effect of group at the LTM test. However, Zif268 antisense ODN had no effect on PR-STM at 3 hours, which did not differ from the LTM test. Furthermore, the PR-LTM deficit in Zif268 antisense ODN infused animals was persistent, with contextual freezing impairments remaining 7 days after conditioning (PR-LTM2). Comparing tests LTM and PR-LTM2, there was a significant group × test interaction (F(1,10) = 22.800, P < 0.002), with no effect of group at the LTM test. *P < 0.05 compared to group receiving Zif268 missense ODN.

Western blot analysis revealed that the retrieval of a contextual fear memory caused an increase in Zif268 protein levels in the both CA1 (200%) and dentate gyrus/CA3 (50%) regions, 2 hours after retrieval, which returned to baseline levels 4 and 6 hours after retrieval (Fig. 6A). Moreover, relative to infusion of Zif268 missense ODN, infusion of Zif268 antisense ODN into the dorsal hippocampus 90 min before retrieval produced a marked reduction in levels of Zif268 protein in CA1 (66%, Fig. 6B) but not in dentate gyrus/CA3 (Fig. 6C), 2 hours after retrieval. Therefore, infusion of Zif268 antisense ODN into the dorsal hippocampus resulted in an estimated complete knockdown of induced Zif268 protein levels in the hippocampal CA1 field and impaired the reconsolidation of contextual fear. This indicates that hippocampal Zif268, specifically in the CA1 field, is required for the reconsolidation of contextual fear memories.

Fig. 6.

Knockdown of Zif268 protein. (A) Zif268 protein levels were increased in CA1 (F(3,8) = 29.534, P < 0.001, n = 3 at each time point) and dentate gyrus/CA3 (F(3,8) = 5.424, P < 0.03, n = 3 at each time point) preparations 2 hours after retrieval, returning to baseline at 4 and 6 hours after retrieval, *P < 0.05 compared to 0 hours. (B) Zif268 antisense ODN (black, n = 3) attenuated the retrieval-related increase in Zif268 in CA1 (*P < 0.001) 2 hours later, but not in the dentate gyrus/CA3 (C), relative to Zif268 missense ODN (white, n = 3).

Discussion. Our experiments revealed a double dissociation in the molecular processes in the hippocampus that are necessary for the consolidation and reconsolidation of long-term contextual fear memory. BDNF is required for the initial consolidation of long-term fear memory but does not play a causal role in the reconsolidation of LTM after its retrieval. Conversely, Zif268 is necessary specifically for the reconsolidation, but not consolidation, of hippocampus-dependent fear memory. Reconsolidation takes less time to establish than consolidation, is differentially sensitive to amnestic agents (12, 13, 16, 2023), and shows a requirement for some, but not all, of the cellular processes needed for consolidation (17, 24). Despite the difference in experiential conditions between fear memory conditioning and recall at the behavioral level, because the footshock US is not given during the retrieval test, the double dissociation at the molecular level provides clear evidence that consolidation and reconsolidation cannot be simply reduced to a quantitative effect. Although reconsolidation and consolidation may be processes with qualitatively equivalent outcomes (5), reconsolidation is at best dependent on only a subset of those cellular processes required for consolidation. Our data show that the molecular processes that support reconsolidation are neither a simple nor a partial recapitulation of the molecular processes that underlie the consolidation of long-term fear memory. In addition, we show that antisense ODN inhibition of Zif268 expression after memory reactivation results in a stable amnesia. This fundamental finding replicates and extends that from a previous study (16), which showed that the reconsolidation of fear memory does not create a new copy of the original memory.

The antisense approach used here has also established that BDNF is required selectively for the consolidation of long-term fear memories, confirming recent findings that endogenous BDNF is required for fear-motivated learning in rats (41). BDNF antisense ODN had no effect on the encoding of associative memory, because STM was normal 3 hours after training. BDNF antisense ODN also did not produce a time-dependent performance deficit, because amnesia for the fearful event was evident a week later. Moreover, the deleterious effects of BDNF antisense ODN on LTM 24 hours after training were reversed by the intrahippocampal administration of exogenous BDNF protein around the time of memory formation. The failure of the exogenous protein to affect LTM per se (BDNF missense ODN + hr-BDNF group, Fig. 3C) further suggests that the action of BDNF is confined to cellular consolidation processes engaged during acquisition.

We also provide direct evidence that reconsolidation of hippocampal-dependent memories requires de novo gene transcription, because actinomycin D inhibited fear memory in a retrieval-dependent manner. Indirect evidence for transcription-dependent reconsolidation in the hippocampus has been provided in CREB (17) and Zif268 (19) null mutant mice, which have brain-wide reduction of these transcription factors. The necessity for transcription suggests the engagement of a gene expression program for the restabilization of previously acquired memories after their recall. Similar to new memories, the dependence on transcription satisfies a critical requirement for the synaptic modifications suggested to underlie LTM (4). CREB and Zif268 are necessary for the consolidation of newly acquired LTM with standard training procedures, but the deficiency in the formation of LTM can be overcome in both instances with overtraining (1719); this finding suggests that despite global deficits in brain CREB and Zif268, LTM can still be consolidated. The highly selective and regionally specific targeting of mRNA for Zif268 in wild-type rats by antisense ODN to erase established contextual fear memory once retrieved, but not freshly encoded memories, identifies specifically and exclusively a causal role for Zif268 in memory reconsolidation. Moreover, our data revealing a critical role for Zif268 specifically in reconsolidation, although in accord with one correlative study (26), resolve a discrepancy in the literature, which had previously shown Zif268 to be correlated with the stabilization of long-term synaptic plasticity (18, 31) but not with hippocampal learning (25, 32, 33).

The double dissociation in the roles of BDNF and Zif268 indicates that after retrieval, a new memory trace is not formed; rather, distinct cellular processes are engaged to maintain an existing fear memory after recall. These processes are collectively unique to reconsolidation and differ from those used after memory acquisition. Indeed, the proteins, including BDNF (42), that contribute to or sustain the presumed synaptic growth underlying consolidation may be different from those that restabilize already existing but reactivated synapses after memory recall (9, 43) or that maintain access to the memory for retrieval (10). At the cellular level, consolidation and reconsolidation may both require gene transcription, but each is orchestrated by distinct, although perhaps overlapping, intracellular signaling pathways (44). There is an essential, but reciprocal, requirement for functional levels of BDNF and Zif268 in hippocampal CA1 neurons for the two memory processes. The requirement for Zif268 in CA1 neurons for reconsolidation reflects the role in the initial consolidation of memory of synaptic plasticity in this hippocampal subregion (45, 46). The brain regions in which plasticity contributes to the stabilization of new versus reactivated memories have recently been dissociated in other learning tasks (44). However, for contextual fear memory, there is an apparent degree of structural homology, with plasticity mechanisms in CA1 supporting consolidation and reconsolidation, but through collectively distinct cellular processes.

The identification of the nonoverlapping mechanisms of consolidation or reconsolidation of fear memory may provide novel targets for therapeutic approaches to disorders such as post-traumatic stress disorder, phobias, and drug addiction, where persistent memories underlie maladaptive behavior and cognitive responses. In addition, newly encoded and reactivated memories may be manipulated separately, without causing a universal amnesia.

Supporting Online Material

www.sciencemag.org/cgi/content/full/1095760/DC1

Materials and Methods

SOM Text

Figs. S1 to S5

References

References and Notes

View Abstract

Navigate This Article