Report

Localization of a Stable Neural Correlate of Associative Memory

See allHide authors and affiliations

Science  31 Aug 2007:
Vol. 317, Issue 5842, pp. 1230-1233
DOI: 10.1126/science.1143839

Abstract

Do learning and retrieval of a memory activate the same neurons? Does the number of reactivated neurons correlate with memory strength? We developed a transgenic mouse that enables the long-lasting genetic tagging of c-fos–active neurons. We found neurons in the basolateral amygdala that are activated during Pavlovian fear conditioning and are reactivated during memory retrieval. The number of reactivated neurons correlated positively with the behavioral expression of the fear memory, indicating a stable neural correlate of associative memory. The ability to manipulate these neurons genetically should allow a more precise dissection of the molecular mechanisms of memory encoding within a distributed neuronal network.

Memories are presumably stored in subgroups of neurons that are activated in response to a given conjunction of sensory inputs. Sparse encoding of memories in complex neuronal networks has been identified with electrophysiological recordings of large groups of single units (14). Similar approaches have identified neurons with firing properties temporally linked to various aspects of task performance (5), and imaging techniques have enabled the physical localization of neurons that are activated during either learning or retrieval of a memory (69). In an associative task, the sensory stimuli and their temporal relationships differ during the learning and retrieval trial, and it is unclear to what extent the neuronal representations of these two events overlap.

We generated a transgenic mouse (TetTag mouse) that allows the persistent tagging of neurons that are activated during a given time window (Fig. 1A). The tag can be used for the direct comparison of neuronal ensemble activity at two widely spaced time points. The TetTag mouse combines elements of the tetracycline-transactivator (tTA) system for transgene regulation with neuronal activity–induced activation of the c-fos promoter to tag activated neurons (1013). We used seizure-induced neuronal activation to test various aspects of regulation in the TetTag mouse. As shown in Fig. 1B, tau-LacZ (LAC) expression had a low baseline, could be induced broadly in the brain by seizure in a doxycycline (Dox)–regulated manner (figs. S1 and S2), and could be maintained for at least 5 days in the presence of Dox after the initial activation.

Fig. 1.

(A) Tagging of activated neurons is achieved by two transgenes present in the TetTag mouse. TetTag mice are raised on food containing Dox (left yellow block). During this time, neuronal activation (indicated by lightning-bolt symbols) that leads to expression of tTA through c-fos–promoter activation will not trigger tagging, because Dox blocks activation (indicated by the red “x”) of the tetO promoter (see “neuron A”). The time window for tagging is opened by switching mice to food without Dox (middle white block). Neuronal activation will now activate the transcriptional feedback loop and start expression of tau-LacZ (see “neuron B”). The time window is closed by putting mice back on Dox food (right yellow block) to block further feedback loop activation (see “neuron C”). However, neurons that were activated during the “no Dox” time window will continue to express tau-LacZ (see neuron B), because the feedback loop can maintain its own activation through the Dox-insensitive tTAH100Y (tTA*), where H100Y represents His100→Tyr100. (B) Kainic acid–induced seizures were used to test whether Dox can be used to open a time window for tagging activated neurons. In the presence of Dox, no neurons are tagged after induction of seizure, as indicated by X-Gal (5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside) staining of a coronal section (“mouse 1”). When a TetTag mouse is left in the home cage in the absence of Dox, only a limited number of neurons are tagged (“mouse 2”). Seizure in the absence of Dox triggers the tagging of a large number of neurons throughout the forebrain, as can be seen at 7 hours (“mouse 3”) and 5 days (“mouse 4”) after seizure. “Mouse 4” was put on Dox food immediately after seizure.

The TetTag mouse was used to determine whether neurons that are activated during learning are reactivated during retrieval. We used expression of LAC as an indicator of neuronal activity during learning and expression of the immediate-early gene Zif/Egr (ZIF) as an indicator of neuronal activity during retrieval (Fig. 2A) (6, 7, 9). We assessed associative memory using fear conditioning, with a focus on the learning and retrieval of a context-shock association (14, 15). The basolateral amygdala (BLA) has been implicated as a potential storage site for context-shock associations (1618). Using the TetTag mice, we were able to identify neurons in the BLA that express both LAC and ZIF after learning and retrieval of conditioned fear (Fig. 2B). These reactivated neurons are probably projection neurons, because LAC neurons in the BLA did not overlap with γ-aminobutyric acid–containing neurons (fig. S3).

Fig. 2.

(A) A protocol was designed to detect repeated activation of neurons during learning and retrieval of conditioned fear. Learning takes place in the absence of Dox, when activation of neurons triggers long-lasting expression of LAC and short-lasting expression of immediate-early genes like Zif/Egr (ZIF). Retrieval takes place 3 days after learning in the presence of Dox, which prevents activation of the molecular feedback loop. Dissection of brains is done 1 hour after retrieval, when LAC and ZIF can be used as indicators of learning-induced and retrieval-induced activation, respectively. (B) Example of LAC and ZIF expression in BLA neurons of a mouse that was subjected to the protocol described in (A). The yellow square in the left darkfield picture marks the area shown in the three immunostaining pictures. Yellow arrows mark neurons that express both LAC and ZIF. Scale bar, 100 μm. CA, central amygdala.

The experiment, outlined in Fig. 3A, used four groups of mice: The home cage (HC) group remained in the home cage throughout the experiment, the fear-conditioning (FC) group underwent fear-conditioning training and was tested for retrieval 3 days later, the fear-conditioning no-retrieval (FC-NR) group was not subjected to a retrieval test, and the no-shock (NS) group was treated identically to the FC group except that no shocks were administered. The FC group had higher freezing (lack of complete movement) scores during context retrieval than the NS group (Fig. 3B). BLA neurons activated during learning of conditioned fear were tagged, as was shown by the higher number of LAC-positive neurons in the combined FC groups as compared with those in both the HC and NS groups (Fig. 3C). This increase in LAC was not caused by retrieval-induced activation, because there was no difference between the FC and FC-NR groups [t(16) = 0.77]. We next asked whether neurons activated during fear conditioning were reactivated during retrieval. No significant differences in the number of ZIF neurons were found (F2,29 = 0.28); however, the FC group had a higher number of LAC+ZIF–positive neurons than the HC and NS groups (Fig. 3D). Only the FC group showed overlapping expression that was significantly higher than expected by chance (Fig. 3D). After normalizing for chance overlap, the FC group had more LAC+ZIF neurons than the HC and FC-NR groups (Fig. 3E).

Fig. 3.

Reactivated neurons in the BLA during fear conditioning provide a stable neural correlate of associative memory. (A) Diagram of the experimental design. HC, n = 21 mice; NS, n = 10 mice; FC, n = 10 mice; FC-NR, n = 8 mice. (B) The FC group showed more freezing than the NS group during context retrieval [t(18) = 5.9, P = 0.00001]. (C) The combined FC andFC-NR group had an increased number of LAC-positive neurons as compared with both the HC and NS groups (F2,46 = 6.2, P = 0.004). (D) The FC group had a higher number of LAC+ZIF–positive neurons than the HC and NS groups (F3,45 = 7.7, P = 0.0003). In the FC group, but not the other three groups, the number of LAC+ZIF neurons was higher than chance level [FC: t(9) = 4.1, P = 0.003]. (E) The FC group had more LAC+ZIF neurons than the HC and FC-NR groups after subtracting chance level (F3,45 = 5.7, P = 0.002). Data in (C) to (E) are percentages of total neurons as determined by 4′,6′-diamidino-2-phenylindole (DAPI) staining. Means ± SEM are shown in (B) to (E). *P < 0.05, **P <0.01.

Does the strength of neuronal reactivation in the BLA correlate with the strength of the memory retrieval? We used extinction to weaken the expression of the fear memory (19). Two groups of mice were used: The FC group was treated identically to the FC group from the previous experiment, whereas the extinction (EX) group was subjected to extinction trials between learning and retrieval (Fig. 4A). Extinction resulted in a significant decrease in fearmemory expression, as indicated by freezing during context retrieval [EX: 13 ± 2% freezing due to context retrieval; FC: 53 ± 9%; t(20) = 6.4, P = 0.000003]. The EX and FC groups had similar numbers of LAC, ZIF, or LAC+ZIF neurons in the BLA (F1,20 = 0.00, F1,20 = 0.34, and F1,20 = 0.61, respectively). However, extinction showed variable effectiveness in individual mice, with freezing scores ranging from 2 to 32%. This result allowed us to look for a correlation, within a group of similarly treated mice, between the strength of (behaviorally expressed) conditioned fear and the number of reactivated neurons. Within the EX group, there was a strong correlation between the number of LAC+ZIF neurons and the freezing score measured during context retrieval (Fig. 4B). There was no correlation with LAC alone (R = 0.26, P = 0.3) or ZIF alone (R = 0.23, P = 0.4). Training, extinction, and retrieval used both the context and a tone as conditioned stimuli (CS) so that neuronal correlates of both memories could be examined. Freezing scores shown in Fig. 4B were obtained during the beginning of the retrieval test before the first tone and thus indicated contextual fear memory. We obtained an estimate of the strength of the tone fear memory by subtracting the freezing before the first tone from the freezing during the first tone. There was no correlation between the estimated freezing during tone retrieval and the number of reactivated neurons in the BLA (Fig. 4C). However, the number of reactivated neurons in the lateral amygdala (LA) correlated with the estimated freezing during tone retrieval but did not correlate with the freezing during context retrieval (Fig. 4, D and E).

Fig. 4.

The number of reactivated neurons in the BLA and LA correlates with the strength of the conditioned fear memory. (A) Diagram of the experimental design. EX, n = 16 mice; FC, n = 6 mice. (B) Within the EX group, the number of LAC+ZIF neurons in the BLA correlated with freezing during context retrieval (R = 0.73, P = 0.001). (C) The number of BLA LAC+ZIF neurons did not correlate with the estimated freezing during tone retrieval (R = 0.16, P = 0.6). N.S., not significant. Freezing during tone retrieval was estimated by subtracting the freezing before the first tone from the freezing during the first tone. (D) In the LA, the number of LAC+ZIF neurons did not correlate with freezing during context retrieval (R = 0.10, P = 0.7). (E) In the LA, there was a significant correlation between the number of LAC+ZIF neurons and the estimated freezing during tone retrieval (R = 0.65, P = 0.006).

We localized a group of neurons in the amygdala that was activated during learning and reactivated during retrieval of a fear memory. Reactivation of neurons in the BLA and LA correlated with contextual and tone fear memory, respectively, which is in accordance with previous studies (17, 18). The reactivated neurons did not simply respond to the repeated pattern of CS (i.e., context and tone), because the unconditioned stimulus (US) (i.e., shock) was required during learning-induced activation but not during retrieval-induced reactivation. Hebbian models of plasticity postulate that synaptic weight changes should occur at synapses that are active concurrently with neuronal firing. In fear conditioning, this should occur at neurons that receive both CS and US inputs, such that the enhanced strength of the CS inputs produced by Hebbian plasticity would allow firing (or increased rates of firing) of conditioning-activated neurons in response to the CS presentation alone during the retrieval trial. This reactivation would, in effect, recapitulate aspects of the original learning episode. In the FC group, 12% of the neurons tagged with LAC were reactivated during retrieval (Fig. 3, C and D). The nonreactivated LAC cells would represent either neurons activated in the home cage (as observed in the HC group in Fig. 3C) or US-stimulated neurons that did not receive CS inputs during learning and were therefore not reactivated during retrieval. Previous studies found neurons in the BLA and LA that receive convergent projections from CS and US pathways and that can undergo fear conditioning–induced Hebbian plasticity that lasts for several hours (4, 16, 20, 21). The current result suggests that this plasticity evolves into stable synaptic changes that confer on these neurons a capacity for reactivation by the CS that lasts for at least 3 days. The reactivated neurons seem to be a likely component of a stable engram or memory trace for conditioned fear.

The TetTag mouse provides a distinctive tool for localizing memories in a complex neuronal network. It combines the ability to image all the neurons in the region of interest with the ability to assess activation of the same neuron at widely separate time points. Because reactivation of neurons takes place in freely moving mice, it can be correlated with the behavioral expression of the memory. Use of the TetTag mouse only requires basic immunohistochemistry and microscopy tools, which are available to most researchers. The TetTag mouse not only provides a tag through the expression of a reporter gene, but can also drive the expression of additional transgenes that can be used to both follow and manipulate the physiology of the tagged neurons (22).

Supporting Online Material

www.sciencemag.org/cgi/content/full/317/5842/1230/DC1

Materials and Methods

Figs. S1 to S3

References

References and Notes

View Abstract

Navigate This Article