Selective Erasure of a Fear Memory

See allHide authors and affiliations

Science  13 Mar 2009:
Vol. 323, Issue 5920, pp. 1492-1496
DOI: 10.1126/science.1164139


Memories are thought to be encoded by sparsely distributed groups of neurons. However, identifying the precise neurons supporting a given memory (the memory trace) has been a long-standing challenge. We have shown previously that lateral amygdala (LA) neurons with increased cyclic adenosine monophosphate response element–binding protein (CREB) are preferentially activated by fear memory expression, which suggests that they are selectively recruited into the memory trace. We used an inducible diphtheria-toxin strategy to specifically ablate these neurons. Selectively deleting neurons overexpressing CREB (but not a similar portion of random LA neurons) after learning blocked expression of that fear memory. The resulting memory loss was robust and persistent, which suggests that the memory was permanently erased. These results establish a causal link between a specific neuronal subpopulation and memory expression, thereby identifying critical neurons within the memory trace.

Ensembles of neurons are thought to serve as the physical representation of memory (the memory trace) (1). However, identifying the precise neurons that constitute a memory trace is challenging because these neuronal ensembles are likely sparsely distributed (2). Previous studies detected neurons whose activity is correlated with memory encoding, expression, or both (37). However, correlative studies do not address whether these neurons are essential components of the memory trace. A direct test of this requires specifically disrupting only those activated neurons and determining whether subsequent memory expression is blocked. Establishing this causal role has been difficult, because of the limited ability of current techniques to target a specific subset of neurons within a brain region.

To target neurons whose activity is correlated with memory, we took advantage of our recent findings that LA neurons with relatively increased levels of the transcription factor CREB were preferentially activated by auditory fear memory training or testing (8). To manipulate CREB levels in a subpopulation (roughly 15%) of LA neurons, a region critical for auditory fear memory (912), we used replication-defective herpes simplex viral (HSV) vectors (13). Neurons overexpressing CREB (with CREB vector) were three times as likely to be activated as their noninfected neighbors after fear memory training or testing in wild-type (WT) mice and 10 times as likely to be activated in CREB-deficient mice. Conversely, WT neurons with a dominant-negative CREB vector were 1/12 as likely as their neighbors to be activated by fear training or testing. These findings suggest that neurons with relatively higher CREB function are preferentially recruited into the fear memory trace and that posttraining ablation of just these neurons should disrupt expression of the established fear memory.

To ablate neurons overexpressing CREB in the present study, we used transgenic mice in which cell death may be induced in a temporally and spatially restricted manner. Apoptosis is induced after diphtheria toxin (DT) binds to the DT receptor (DTR) (14). Because mice do not express functional DTRs (15, 16), we used transgenic mice that express simian DTR in a Cre-recombinase (cre)–inducible manner (iDTR mice) (15). A loxP-flanked STOP cassette that normally silences DTR expression is excised by cre, which allows constitutive expression of DTR. Injection of DT any time thereafter induces apoptosis only in cells expressing DTR.

We restricted DT-induced ablation to neurons overexpressing CREB by inserting cDNA encoding cre recombinase into our CREB vector (CREB-cre). DTR expression only occurs in neurons expressing cre, which allowed us to persistently tag infected neurons for subsequent ablation (Fig. 1A and fig. S1). As a control, we used Cntrl-cre vector to induce apoptosis in a similar portion of LA neurons that are not preferentially activated by fear testing. To examine cell death, we used two markers of apoptosis, caspase-3 activation and terminal deoxynucleotidyl transferase–mediated deoxyuridine triphosphate nick end labeling (TUNEL) (17). We microinjected CREB-cre, Cntrl-cre, and CREB alone (without cre) vectors into the LA of iDTR transgenic and WT littermate mice, then administered DT or vehicle. We observed substantial cell death only in LA neurons of iDTR mice microinjected with CREB-cre or Cntrl-cre vectors and administered DT (experimental groups, Fig. 1, B and C, and fig. S2) (15). Negligible apoptosis was observed in regions outside the LA (figs. S3 and S4) or in control groups that lacked a key component (iDTR, cre or DT) (Fig. 1, B and C). Note that Cntrl-cre and CREB-cre vectors produced equal levels of cell death (Fig. 1, B and C); however, the efficiency of cell death was not complete (see SOM text). Therefore, this system allows temporally specific ablation of tagged neurons.

Fig. 1.

Selective ablation of LA neurons. (A) iDTR mice express DTR under control of a floxed STOP cassette (no DTR expression, indicated by white). CREB-cre or Cntrl-cre microinjected into LA (green). Cre removes STOP cassette, which allows DTR expression (pink). DT (blue triangles) induces apoptosis (red) only in cells that have undergone recombination. (B) (Left) Low GFP (with cre vector, green) and high activated caspase-3 (aCas3, red) levels in experimental (iDTR/CREB-cre/DT, iDTR/Cntrl-cre/DT), but not control (WT/CREB-cre/DT), mice. DAPI (4′,6′-diamidino-2-phenylindole)–stained nuclei (blue). Scale bar, 100 μm. (Right) High activated caspase-3 levels in experimental [iDTR/CREB-cre/DT (n = 24), iDTR/Cntrl-cre/DT (n = 17)], but not control [WT/CREB-cre/DT (n = 4)], mice (F2,42 = 6.44, P < 0.001). (C) (Left) High TUNEL (red) levels in experimental (iDTR/CREB-cre/DT, iDTR/Cntrl-cre/DT), but not control (iDTR/CREB-cre/PBS), mice. Scale bar, 100 μm. (Right, top) Morphological indicators of apoptosis in experimental mice. Scale bar, 50 μm. (Right, bottom) High TUNEL levels in LA of experimental mice [iDTR/CREB-cre/DT (n = 5), iDTR/Cntrl-cre/DT (n = 6) versus control (n = 5); F2,13 = 22.31, P < 0.001].

To verify that LA neurons overexpressing CREB are selectivity activated by fear memory testing with our modified vector, we microinjected WT mice with CREB-cre or Cntrl-cre vector before auditory fear training. To visualize neurons specifically activated by memory expression, we examined Arc (activity-regulated cytoskeleton-associated protein; Arg3.1) RNA (18). Neuronal activity induces a rapid, but transient, burst of Arc RNA that is quickly transported to the cytoplasm, which allows nuclear-localized Arc RNA to serve as a molecular signature of a recently (5 to 15 min previously) active neuron (18). Five minutes after testing, we removed brains and examined Arc (activated by fear testing) and green fluorescent protein (GFP) (with vector) RNA. Neurons with CREB-cre vector preferentially expressed Arc after memory testing; neurons with CREB-cre were three times as likely to be activated by fear memory testing as their noninfected neighbors. In contrast, neurons with Cntrl-cre vector and their noninfected neighbors were equally likely to be activated by memory testing (Fig. 2).

Fig. 2.

Neurons overexpressing CREB preferentially activated by fear memory testing. (A) Double-labeled nuclei in LA of CREB-cre, but not Cntrl-cre, mice. Nuclei (blue), GFP+ (with CREB-cre or Cntrl-cre vector, green), Arc+ (pink), or double-labeled nuclei (GFP+ and Arc+; arrows). Scale bar, 20 μm. (B) In CREB-cre mice, Arc was preferentially localized in infected (GFP+), rather than noninfected (GFP), neurons. In Cntrl-cre mice, Arc was equally distributed in infected and noninfected neurons (Vector × GFP/Arc colocalization interaction, F1,5 = 18.74, P < 0.05).

We microinjected CREB-cre vector into the LA of iDTR mice before weak auditory fear training. Cntrl-cre vector was used to ablate a similar portion of random LA neurons (i.e., not preferentially activated by fear memory testing). We assessed memory before (test 1) and after (test 2) inducing cell death in tagged neurons by administering DT. The CREB-cre vector enhanced fear memory following weak training (Fig. 3, test 1), consistent with previous results (8, 1922). Selectively deleting neurons with CREB-cre vector completely reversed this enhancement (test 2). Note that CREB-enhanced memory was not blocked if either cre or DT was omitted, consistent with the absence of apoptosis in these control groups. The reversal of CREB-enhanced memory was not due to memory extinction with repeated testing because the control groups froze robustly on test 2. Therefore, increasing CREB in a subpopulation of LA neurons enhances a weak memory and specifically ablating just these neurons reverses this enhancement.

Fig. 3.

Overexpressing CREB in LA neurons enhances memory induced by weak training; subsequent ablation of these neurons reverses this enhancement. (Top) CREB-cre microinjection enhanced memory after weak training [test 1, CREB-cre/DT (n = 8), Cntrl-cre/DT (n = 9), P < 0.001]. DT administration reversed this memory enhancement (test 2, P < 0.001). CREB-enhanced memory was not blocked on test 2 if either cre [CREB/DT (n = 7), P > 0.05] or DT [CREB-cre/PBS (n = 6), P > 0.05] was omitted. Group × Test interaction F3,26 = 13.90, P < 0.001. (Bottom) Schematic of LA neurons after DT or PBS. Blue, DAPI-labeled neuronal nuclei; pink, neurons activated by memory; and white, ablated neurons.

Although increasing CREB in a subpopulation of LA neurons does not further enhance a strong memory, neurons overexpressing CREB are, nevertheless, preferentially activated by fear memory expression (8). This suggests that CREB levels dictate which neurons are recruited into a memory trace, even in the absence of behavioral change. To examine the effects of ablating neurons overexpressing CREB on a strong memory, we trained mice that received CREB-cre or Cntrl-cre microinjections using an intense protocol. Strong training produced robust auditory fear memory in both groups before DT administration (Fig. 4A). After DT, only CREB-cre mice showed a loss of auditory fear memory. To investigate whether memory in mice microinjected with CREB-cre is particularly susceptible to the ablation of a small number of neurons, we microinjected both Cntrl-cre and CREB (no cre) vectors, which allowed us to delete only Cntrl-cre neurons after training. Deleting this small portion of neurons (that were not overexpressing CREB) had no effect on memory (fig. S5). Therefore, memory loss was specific; it was not determined by the absolute number of deleted LA neurons but by whether these deleted neurons overexpressed CREB at the time of training.

Fig. 4.

Specifically ablating LA neurons overexpressing CREB, but not a similar portion of random neurons, blocks expression of strong memory. (A) Strong training produced equally robust and specific auditory fear memory (low baseline and high tone freezing) in CREB-cre (n = 6) and Cntrl-cre (n = 6) mice on test 1. Although Cntrl-cre mice showed robust memory after DT administration (test 2), CREB-cre mice showed impaired memory (low baseline and low tone freezing) (Vector × Test × Time significant three-way interaction, F1,10 = 5.55, P < 0.05). (B) Memory loss was not due to disruption of reconsolidation. Cntrl-cre (n = 5) mice showed robust auditory fear memory (low baseline but high tone freezing), whereas CREB-cre (n = 6) mice showed a loss of auditory fear memory (low baseline and tone freezing) (Vector × Time interaction, F1,10 = 6.20, P < 0.05). (C) Memory loss was persistent in CREB-cre mice, but they could relearn. Over repeated tests (tests 1 to 3), CREB-cre mice (n = 10) showed stable, low tone freezing, whereas Cntrl-cre mice (n = 7) showed robust tone freezing [Vector × Test analysis of variance (ANOVA), significant effect of Vector F1,15 = 67.81, P < 0.05 only]. (Right) After retraining CREB-cre mice showed an increase in tone freezing (F1,15 = 1.7, P > 0.05).

Administering DT after a fear memory test blocked expression of both CREB-enhanced memory produced by weak training and robust memory produced by strong training in iDTR mice microinjected with CREB-cre (but not Cntrl-cre) vector. Fear memory testing reactivates memory and may trigger a second wave of consolidation (reconsolidation) that, similar to initial consolidation, requires protein synthesis (23, 24). Because DT induces cell death by inhibiting protein synthesis, it is possible that impaired reconsolidation contributes to the memory loss. To assess this, we trained mice but omitted the memory reactivation induced by test 1. Consistent with our previous results, fear memory was blocked in CREB-cre, but not Cntrl-cre, mice (Fig. 4B). Therefore, memory loss was independent of memory reactivation, which ruled out the possibility that blocking reconsolidation accounts for the memory disruption.

If neurons overexpressing CREB during training are critically involved in the subsequent memory trace, then deleting them should permanently block memory expression. To examine the persistence of memory loss, we trained mice, administered DT, and assessed memory 2, 5, and 12 days later. Memory loss in CREB-cre mice was long-lasting, whereas memory remained robust in Cntrl-cre mice (Fig. 4C). Therefore, we found no evidence of memory recovery in mice in which neurons overexpressing CREB were deleted, which suggested that memory was not transiently suppressed. To rule out the possibility that the memory loss was due to a nonspecific impairment in LA function, we showed that CREB-cre mice could relearn (Fig. 4C). Similarly, pretraining deletion of neurons overexpressing CREB did not impair the acquisition, or stability, of a conditioned fear memory (fig. S6). Deleting neurons overexpressing CREB does not affect subsequent learning, presumably because the high portion of remaining (noninfected) neurons are sufficient to encode a new memory. Finally, ablating CREB-overexpressing neurons did not block expression of a memory acquired before surgery (fig. S8). Together, these findings indicate that ablating neurons that were overexpressing CREB at the time of memory encoding blocks memory for that particular learning event, highlighting the specificity of memory loss.

Our results show that neurons with increased CREB levels at the time of fear learning are critical to the stability of that memory, because selectively ablating these neurons after training blocks expression of this fear memory. This indicates that these neurons themselves are essential for memory expression in the days after fear conditioning; they are not simply creating a local environment that promotes memory formation (such as releasing trophic factors) (25). Deleting neurons whose activity is related to memory expression (overexpressing CREB at the time of training) produced memory loss, whereas ablating a similar number of random neurons (expressing Cntrl vector or overexpressing CREB well before or after training) did not. The observed amnesia was specific, robust, persistent and not due to a disruption in either reconsolidation or overall LA function. Together, these results suggest that ablating neurons overexpressing CREB permanently erases the fear memory (see supporting online material text). Fear learning may generate a broad memory trace that encompasses more LA neurons than affected by our treatment or multiple memory traces throughout the brain (26, 27). However, deleting just neurons overexpressing CREB at the time of training produces amnesia, which suggests that these neurons play an essential role within what is likely a broader fear neuronal network. These results establish a causal link between a defined subpopulation of neurons and expression of a fear memory and, thereby, identify a key component of the memory trace.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S9


Movie S1

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article