Cell, Vol. 79, 59-68, October 7, 1994, Copyright 0 1994 by Cell Press

Deficient Long-Term Memory in Mice with a Targeted Mutation of the CAMP-Responsive Element-Binding Protein Roussoudan Bourtchuladze, l Bruno Frengueili,’ Julie Blendy,t Diana Cioffi,* Gunther Schutz,t and Alcino J. Silva” *Cold Spring Harbor Laboratory Cold Spring Harbor, New York 11724 tGerman Cancer Research Center University of Heidelberg D-69120 Heidelberg Federal Republic of Germany

Summary

The cAMP-responsive element-binding protein (WEB) has been implicated in the activation of protein synthe- sis required for long-term facilitation, a cellular model of memory in Aplysia. Our studies with fear condition- ing and with the water maze show that mice with a targeted disruption of the a and 6 isoforms of CREB are profoundly deficient in long-term memory. In con- trast, short-term memory, lasting between 30 and 60 min, is normal. Consistent with models claiming a role for long-term potentiation (LTP) in memory, LTP in hippocampal slices from CREB mutants decayed to baseline 90 min after tetanic stimulation. However, paired-pulse facilitation and posttetanic potentiation are normal. These results implicate CREB-dependent transcription in mammalian long-term memory.

Introduction The synthesis of new proteins is a pivotal requirement of long-term memory (reviewed by Davis and Squire, 1984; Matthies, 1989). A variety of inhibitors of protein and RNA synthesis have been shown to block effectively long-term, but not short-term, memory. Since these inhibitors have a general effect on all protein synthesis, they cannot be used to identify and study the specific components and mechanisms involved in memory consolidation.

Members of the CAMP-responsive element-binding pro- tein (CREB) family of transcription factors activate the tran- scription of genes with CRE sequences in response to increases in the intracellular concentration of CAMP and Ca2+ (reviewed by Lee and Masson, 1993). interestingly, trapping of CREB-like proteins by oligonucleotides with CRE sequences injected into Aplysia cultured neurons selectively blocked long-term, but not short-term, facilita- tion of neurotransmitter release (Dash et al., 1990), a model for a nonassociative learning response in Aplysia. Additionally, genetic studies in Drosophila have demon- strated the involvement of CAMP-dependent processes in associative learning and early memory (Tully, 1991). These data and other preliminary results (T. Tully, per- sonal communication) suggested the involvement of CAMP-dependent transcription and CREB-like proteins in learning and memory.

To determine whether CREB was required for memory

consolidation in mammals, we have studied mice with a targeted mutation of the CREB gene (CREB- mice; Hum- mler et al., 1994). Previous molecular analysis found that these mice lack the a and the 6 isoforms of CREB (Hum- mler et al., 1994). Studies with fear conditioning and with the water maze reported in this manuscript show that CREB- mice have normal short-term memory but deficient long-term memory. Parallel electrophysiological studies in hippocampal slices revealed that LTP in CREB mutants is small and decays to baseline within 1.5 hr of induction even though two other forms of synaptic plasticity, paired- pulse facilitation (PPF) and posttetanic potentiation (PTP), are normal. Interestingly, an accompanying manuscript shows that disrupting CREB function in Drosophila blocks long-term memory (Yin et al., 1994 [this issue of CeIj).

Results

General Observations CREB is ubiquitously expressed, but its expression is greater in some tissues, such as brain, than in others (Hummler et al., 1994). Thus, the loss of the a and 6 soforms could conceivably lead to generalized neu- roanatomical and neurophysiological deficits. The CREB- mutants, however, appear healthy and groomed, they show no hints of ataxiaor of any other motor disorders, and they show no overt abnormalities. Consistent with these observations, light microscopy analysis of thionin-stained coronal sections obtained throughout the rostral-caudal axis of the CREB- forebrain did not detect any gross ana- tomical abnormalities (Figure 1).

Fear Conditioning Studies Fear conditioning is a form of associative learning found in many species (Dickinson and Mackintosh, 1978). We first tested whether the CREB- mutation altered nocicep- tive reactions to shock in the mutants since changes in pain sensitivity affect conditioning (Fanselow and Bolies, 1979a, 1979b). We determined the minimal amount of cur- rent required to elicit three common reactions to shock: flinch, run/jump, and vocalization. Figure 2 demonstrates that for both CREB- mutants and controls, each of these three stereotypical behaviors was elicited with similar cur- rent intensities (p = 0.42), indicating that pain sensitivity is normal in CREB- mutants. This data also provides evi- dence that the mutation caused neither sensory nor motor performance deficits; that is, the mutants could jump, flinch, and vocalize normally to sensory stimuli.

Animals learn to ‘fear” a previously neutral stimulus (con- ditioned stimulus [CS]) simply because of its temporal as- sociation with an aversive stimulus (unconditioned stimu- lus [US]), such as afoot shock. Conditioned animals, when exposed to the CS, tend to refrain from all but respiratory movements by freezing (Fanselow and Belies, 1979a; Miller and Weiss, 1969). Freezing responses can be trig- gered with two different types of CS, each with distinct

lmm lmm

Figure 1. Coronal Forebrain Section of a CREB Mutant and Control Mice

Coronal brain sections (15 urn) from a CREB mutant (A) and control littermate (8) stained with 1% thionin. The rostral-caudal level of the mutant and control slices was matched for the hippocampus. Abbreviations: Am, amygdala; CA1 and CA3, pyramidal cell fields of the hippocampus; DG, dentate gyrus; NC, neocortex; Th, thalamus; and Cp, . caudate putamen.

neuroanatomical substrates (Kim and Fanselow, 1992; Phillips and LeDoux, 1992). In cued conditioning, the CS is simply a tone, and lesions of the amygdala (but not the hippocampus) disrupt this type of conditioning (Kim et al., 1993; Phillips and LeDoux, 1992). Rodents also can be conditioned to the context in which they were exposed to the US (contextual conditioning), and this type of condi- tioning is known to be dependent on both hippocampal and amygdala function (Kim and Fanselow, 1992; Phillips and LeDoux, 1992). These tasks are particularly appro- priate for the study of memory because robust learning can be triggered with a single 3 min trial.

Figure 3A indicates that WEB- mutants and controls showed identical levels of freezing immediately after train- ing (p = 0.98). A similar result was also obtained when contextual conditioning was tested 30 min after training (p = 0.58; Figure 38). Both groups spent approximately 40% [controls (41% f 4%) mutants (39% f 40/o)] of a 5 min testing interval without any perceptible movement. Previous studies (Fanselow, 1988; Rudy and Morledge, 1994) and our own experiments with mice (data not shown) indicate that freezing responses measured immediately after training are the result of associative processes. All together these results demonstrate that the CREB- mice can be contextually conditioned, and they also indicate that the CREB- mutation did not prevent learning or inter- fere with contextual memory for at least 30 min.

In striking contrast with the group tested 30 min after training, a second group of CREB- mutants displayed little freezing (12% f 5%) when tested 1 hr after training (Fig- ure 3C) even though the controls showed extensive freez- ing (42% f 8%). Figure 3D shows the results of contex-

2 0.2s E c 0.2- ;; 5 0.15-

Fi 0. I -

0.05

0 I

n Controls El 0 Mutant

Il..- Flinch Jump Vocalize

Figure 2. Sensitivity to Shock in CREB Mutants and Controls To evaluate the sensitivity of the mice to shock, we determined the minimal amount of current required to elicit three stereotypical behav- iors in CREB mutants (n = 9) and controls (n = 7): flinching, running/ jumping, and vocalizing. ANOVA showed that there was no difference between the two genotypes (F(1,14) = 0.707, p = 0.421.

Deficient Long-Term Memory in CREB- Mice 61

0.5 1.0 1.5 2.0 2.5 30 1 2 3 4 5 TIME (min)

T

I 2 3 4 5 TIME (min)

Figure 3. Contextual Conditioning with a Sin- gle Training Trial (A) CREB mice and controls from groups 1, 2, and 3 were given a single training trial. A solid line indicates the duration of the tone (CS), while a closed box indicates the 2 s shock (US). Since experiments with all three groups were done in a balanced manner and since the train- ing procedure was identical for all groups, (A) represents pooled training data (n = 28 wild- type and n = 28 mutant mice) from all three groups. ANOVA showed no significant differ- ence between mutants and controls in freez- ing responses immediately following the US (F[1,54] = 0.0002, p = 0.98). (B) Group 1 was tested for contextual condition- ing 30 min after training. ANOVA showed that freezing in CREB mutants (n = 12) was not statistically different from that of controls (n = 12) (F[l, 22) = 0.343, p = 0.56). (C)Group 2 was tested for contextual condition- ing 1 hr after training. At this time, ANOVA showed that controls (n = 9) expressed sig nificantly more freezing responses than did CREE mutants (n = 9) (F[l, 161 = 50.279, p < 0.001). Planned comparisons between CREB mutants and controls showed significant differ- ences at 1 min (F[l,lS] = 8.43, p < O.Ol), 2 min (F[1,16] = 12.85, p < O.Ol), 3 min (F[1,16] = 13.49, p < O.Ol), 4 min (F[1,16] = 9.14, p < O.Ol), and 5 min (F[1,16] = 7.14, p < 0.002). (D) Group 3 was tested for contextual condition-

ing 24 hr after training. ANOVA showed that controls (n = 7) expressed significantly more freezing responses than did CREB mutants (n = 7); (F(1,12] = 23.936, p < 0.001). Planned comparisons between CREB mutants and controls showed significant differences at 2 min (F[1,21] = 5.11, p < 0.04), 4 min (F[1,21] = 7.06, p < 0.02), and 5 min (F[1,21] = 5.38, p < 0.04).

tual conditioning experiments with another group of mice tested 24 hr after training: controls once again showed significantly (p < 0.001) more freezing (48% f 7%) than did mutants (12% f 6%). These results indicate that the CREB- mutation resulted in an abrupt disruption of mem- ory between 30 and 60 min posttraining.

It is possible that freezing responses measured 30 min after training are not specific to the training context and instead are the result of general indiscriminate fear of any novel context. To test this hypothesis, we trained another group of CREB- mice and controls and tested them 30 min later in a novel context (Figure 48, Pre-CS). Unlike the training context, this novel context triggered little or no freezing (5% f 1% and 5% f 1% for mutants and controls, respectively).

For tests of cued conditioning, we measured freezing in response to the tone that had previously been paired with the foot shock during training (Kim and Fanselow, 1992; Phillips and LeDoux, 1992). To avoid the confound- ing effects of contextual conditioning, cued conditioning was tested in a novel context. The mice were placed in a novel chamber for 3 min, and then they were exposed to the CS for another 3 min. Cued conditioning was found to be normal 30 and 60 min after training in CREB- mu- tants, but it was abnormally low 2 and 24 hr after training (Figure 4). These results are consistent with the hypothe- sis that long-term memory, but not short-term memory, is disrupted in the CREB- mutants. The fact that cued conditioning produced nearly normal freezing responses

in CREB- mutants tested 1 hr after training indicates that the time course of the amnesia in CREB- mutants differs for different tasks involving different neuroanatomical structures(Kim and Fanselow, 1992; Kim et al., 1993; Phil- lips and LeDoux, 1992). Previous studies had shown that the time course of the amnesia caused by protein synthe- sis inhibitors also differed for different tasks, training proto- cols, and species (Davis and Squire, 1984; Flood et al., 1975). Nevertheless, our results (Figures 3 and 4) suggest that memory for contextual conditioning is more affected by the CREB mutation than by memory for cued condition- ing, It is noteworthy that in our cued conditioning (2 hr and 24 hr) experiments, two mutants tested showed freezing similar to that of controls (35% and 37%). In our contextual conditioning experiments, the performance of one CREB mutant tested 24 hr after training was also similar to that of controls (39%).

We also tested whether additional training might com- pensate for the memory deficits of the CREB- mutants. Thus, we gave CREB- mice and controls three additional CS/US pairings 3 days after training with one CSIUS. In- deed, CREB- mutants trained with a total of four foot shocks showed contextual conditioning (42% f 4%) 24 hr after training. However, even with four shocks, the CREB- mice were still worse than controls (66% f 5%; F(1,21) = 27.36, p < 0.001; n = 13 and n = 10, respectively) trained the same way, underscoring the impact of the CREB muta- tion on memory. Nevertheless, these results show that the mutation of CREB did not block all long-term memory.

70 A

60-

0, 03 I .o 1.5 2.0 2.5 3.0

TIME (min)

60

1 2 3 4 5 6 1 2 3 4 5 6

TIME (min)

60

cs

E cs -

f---h

1 2 3 4 5 6 1 2 3

TIME (min)

Our results are consistent with the model (Kim et al., 1992; Rudy and Morledge, 1994) that fear conditioning is dependent on at least two temporally distinct memory processes since we show that the CREB mutation did not affect an early memory phase lasting up to 30 min; it only affected a later phase.

Water Maze Studies We also tested the CREB- mutants in the Morris water maze tasks (Morris, 1981; reviewed by Brandeis et al., 1989). In this test, animals are placed in a round pool, and they escape the water by learning to swim to a platform. In the “spatial” version of the water maze, a test known to require hippocampal function (Morris et al., 1982; Suth- erland et al., 1982) mice learn to navigate to a submerged platform using distal cues in the room. A disruption of memory, as was found in the fear conditioning studies, should affect performance in this task.

4 5 6

Figure 4. Cued Conditioning with a Single Trial (A) CREB mice and controls from groups 1,2, 3, and 4 were given a single training trial. A solid line indicates the duration of the tone (CS), while a closed box indicates the 2 s shock (US). Since experiments with all four groups were done in a balanced manner and since the training procedure was identical for all groups, (A) represents pooled training data (n = 34 wild-type and n = 35 mutant mice) from all four groups. (B) Group 1 was tested for cued conditioning 30 min after training (CREB, n = 7 and controls, n = 6). ANOVA shows that the CS significantly increased freezing (CS test) for controls [F(l) 10) = 46.662, p < 0.001) and CREB mutants [F(l, 12) = 10.516, p < 0.007, and it showed that the amount of freezing exhibited by con- trols was indistinguishable from that exhibited by CREB mutants (F[l ,I I] = 0.746, p = 0.40). (C) Group 2 was tested for cued conditioning 1 hr after training. ANOVA showed that freezing in CREB mutants (n = 12) was indistinguish- able from that in controls (n = 12) (F[l,22] = 2.515, p = 0.12). (D) Group 3 was tested for cued conditioning 2 hr after training. ANOVA showed that CREB mutants (n = 9) expressed significantly less freezing responses than did controls (n = 9) (F[1,16] = 216.62, p < 0.001). Planned comparisons between CREB mutants and con- trols showed a significant difference at 4 min (F[1,16] = 64.79, p < 0.001), 5 min (F[l,l6] = 60.5, p < O.OOl), and 6 min (F(1,16] = 73.75, p < 0.001). (E) Group 4 was tested for cued conditioning 24 hraftertraining.CREBmutants(n = 7)showed significantly less freezing than controls (n = 7) (F[i ,121 = 8.232, p < 0.02). Planned com- parisons between CREB mutants and con- trols showed a significant difference at 5 min (F[l,l4] = 5.34, p < 0.04) and 6 min (FIl.141 = 5.61, p < 0.04).

We started by testing mice on the spatial version of the water maze with a single training trial each day. Figure 5A shows that the overall performance of control mice was significantly better than that of CREB- mutants during training in this task (p < 0.01). It is noteworthy that the performance of the mutants did not improve during train- ing (p = 0.13). Time to reach the platform, however, is only one of a number of different measures of spatial learning in the water maze(Brandeis et al., 1989). Therefore, we also tested the mice in a probe trial in which the platform was removed, and the trained mice were allowed to search for it for 60 s. Figure 58 shows that, while controls searched selectively for the absent platform (spending 48% f 3%

of their time swimming in the quadrant in which the plat- form had been during training [p < O.OOl]), CREB- mice searched randomly in all four quadrants (p = 0.33). Con- trols also crossed the position of the platform more fre- quently than did CREB- mutants (Figure 5C). Taken to-

Deficient Long-Term Memory in CREB- Mice 63

60

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

T

Trial #

I OP AL AR

Quadrant

C 10

9-

g 8- ‘G; 7-

“0 6-

6 5-

E 4- CE: t;l

3-

z 2 l-

n Controls

cl q Mutants

O- I I I 1 T AR AL OP

Quadrant

Figure 5. One-Trial-a-Day Training in the Morris Water Maze

(A) CREB mutants (n = 10) and controls (n = 6) were trained with one trial per day for 15 days. The average time (in seconds) to reach the submerged platform was plotted versus trial number. ANOVA with repeated measures showed that the overall performance of controls was significantly better than that of CREB mutants (F[1,16] = 11.14, p < 0.01). However, planned comparisons between controls and CREB

gether, these data demonstrate that the CREB mutation dramatically impairs the performance of the mutants in this spatial learning task.

As in our fear conditioning experiments, we next deter- mined whether more intensive training could partially alle- viate the performance deficit of CREB- mice in the water maze. We tested a new group of mutants and controls with three blocks of four trials per day (total of twelve trials per day; with a 1 min intertrial interval) for three days. Figure 6A shows that both mutants and controls improved (p < 0.05) and that with intensive training, the performance of the CREB- mice in acquisition trials was indistinguish- able from that of controls (p = 0.45). After 3 days of train- ing, the mutants spend 40% f 6% of the 60 s probe trial searching for the missing platform in the training quadrant. Nevertheless, analysis of the probe trials indicate that the searches of controls were significantly (p < 0.01) more selective than were those of CREB- mutants (Figure 6B and 6C). Therefore, with this training schedule, the deficit in the CREB- mutants was too subtle to be detected during acquisition but not too subtle to be detected during probe trials. It is noteworthy that the performances of CREB- mice and controls were identical in the first day of training both in the one trial per day (p = 0.92) and in the twelve trials per day experiments (p = 0.38) demonstrating that initial differences in motivation, perception, swimming speed, or coordination could not account for the impair- ment of the CREB- mutants.

These results show that CREB- mutants are impaired in the water maze tasks, which corroborates and extends our fear conditioning experiments. Interestingly, intensive training partially compensates for the deficits shown in both cases. Other CAMP-responsive transcription factors might have compensated partially for the loss of the a and 6 isoforms of CREB in the brain (Hummler et al., 1994) CAMP-independent mechanisms (Bliss and Collingridge, 1993) might contribute to long-term memory, or both (Mat- thies, 1989). Interestingly, mice with a targeted mutation of a Ca2+/calmodulin-dependent kinase II (aCaMKII) re- vealed a more dramatic deficit in the water maze even when trained with 12 trials per day for 5 days (Silva et al., 1992). However, analysis of that mutation implicated the

mutants for each of the 15 trials revealed no significant differences, (6) The graph shows the results of a probe trial given after 15 days of training. ANOVA with dependent measures showed that controls searched selectively in the quadrant in which the platform was located during training (F/l ,641 = 68.56, p < 0.001) while CREB mutants did not (F[1,64] = 0.96. p = 0.33). Planned comparisons between the time that controls spent in the training quadrant and the time spent in all other quadrants confirmed significant differences (T versus AL, p < 0.001; T versus AR, p < 0.001; and T versus OP, p < 0.001). (C)The control mice (F[3, 30) = 90.636, p < 0.001) but not the CREB mutants (F[3,36] = 2.044, p = 0.13) crossed the exact site where the platform was located during training more often than they crossed equivalent sites in the other three quadrants. A planned comparison between control and mutants revealed a significant difference in the number of platform crossings in the training quadrant (91,641 = 64.69, p < 0.001). Abbreviations: T, training quadrant; AR, the adjacent quadrant to the right of T; AL, the adjacent quadrant to the left of T; and OP, the quadrant opposite to T.

Cdl 64

A 60

3 40-

P f 3 30- Q 4 20-

10-

0 I I I 1 I I I I I 1 2 3 4 5 6 7 8 9

I3 60

3 50

5 40

& E 30 .I 2 20 .II b

I$ 10

0

T

I ~ T

Blocks of 4 Trials

T

i AR

WControls

q Mutants

.r:ln AL ’ OP

Quadrant

I I

Quadratt OP

WControls

q Mutants

1

Figure 6. Intensive Training in the Morris Water Maze

(A) CREB mutants (n = 7) and controls (n = 7) were trained with three blocks of four trials per day (for 3 days). The average time to reach the submerged platform was plotted versus trial block. ANOVA with repeated measures revealed that there was no significant difference between the overall performance of the controls and mutant mice (F[1,12] = 0.6046, p = 0.45). Post-hoc analysis indicated that both improved from day 1 to day 3 (p < 0.05). (B) The graph shows the results of a probe trial given after 3 days of training. ANOVA with dependent measures revealed that both controls

ccCaMKll in the induction of LTP and in learning (Silva et al., 1992) while the CREB mutation does not seem to affect learning but only memory. Additionally, intensive training is capable of partially compensating for the mem- ory deficit in the CREB mutants but not for the learning deficit in the aCaMKll mutants tested with the water maze (Silva et al., 1992) or with fear conditioning tasks (Ft. B. and A. J. S., unpublished data).

Hippocampal Electrophysiological Analysis Since current models suggest that hippocampal LTP may be a mechanism underlying learning and memory (Bliss and Collingridge, 1993; Eichenbaum and Otto, 1993) and since our behavioral analysis of the CREB- mice focused primarily on hippocampal-dependent tasks, we tested hip- pocampal LTP in CREB- mutants. LTP was measured in hippocampal slices of adult mice (- 3-4 months) by stimu- lating Schaffer collaterals and recording field excitatory postsynaptic potentials (fEPSPs) in stratum radiatum of CA1 . After collecting stable baseline responses for at least 30 min, LTP was induced by a train of 100 pulses at 100 Hz. We followed LTP for at least 2 hr after tetanic stimula- tion since the amnesia of CREB- mutants was detected within the first hour after conditioning. The experiments with mutant mice were interleaved with the experiments with normal controls.

Figure 7A shows data from wild-type and mutant slices in which LTP was followed for 2 hr. Wild-type slices (6 slices from 6 animals) showed stable LTP at 2 hr, whereas the LTPof mutantslices(10slicesfrom9animals)decayed to baseline 90 min after the tetanus and remained at that level until the end of the experiment (Figure 7A). Interest- ingly, previousstudies showed that asingle tetanus similar to the one used in this study does not result in a measur- able increase in CAMP in hippocampal slices (Frey et al., 1993). This suggests that if there is an increase in CAMP triggered by this tetanus, it is limited perhaps only to the tetanized synapses and might therefore be difficult to mea- sure in slice homogenates. Although in the CREB- mu- tants both LTP and memory decay within the first hour, it is unclear whether the decay in LTP is related to the decay

(F[1,46] = 53.99, p < 0.001) and mutants (FI1.48) = 11.41, p < 0.01) searched selectively for the platform in the training quadrant. Planned comparisons between the time that controls spent in the training quad- rant and the time spent in all other quadrants showed significant differ- ences (T versus AL, p < 0.001; T versus AR, p < 0.001; and T versus OP, p < 0.001). Similarly, planned comparisons for the mutants re- vealed that they also spent more time in the training quadrant (Tversus AL, p < 0.001; T versus AR, p < 0.04; and T versus OP, p < 0.001). However, a planned comparison analysis between controls and mu- tants revealed that mutants spent significantly less time in the training quadrant than did controls (F[i ,481 = 7.88, p < 0.01). (C)The control mice (F[3,26] = 20.812, p < 0.01) as well as the CREB mutants (F[3,26] = 10.97, p < 0.01) crossed the exact site where the platform was located during training more often than they crossed equivalent sites in the other three quadrants. However, a planned comparison between genotypes revealed that mutants crossed the platform site in the training quadrant significantly fewer times than did controls (F[1,48] = 9.44, p < 0.01). Abbreviations: T, training quadrant; AR, the adjacent quadrant to the right of T; AL, the adjacent quadrant to the left of T; and OP. the quadrant opposite to T.

Deficient Long-Term Memory in CREE- Mice 65

1

ALL

SO

rime (mins) w SLICES ANIMALS

G 180 r

__~~_---__~-----------

I I 100 150 200 250 300

Inter-Pulse Interval (ms)

Figure 7. Electrophysiological Analysis of CREB- Mutants and Wild Type (A) FEPSPswere recorded in the stratum radiatum of the hippocampal CA1 region. Baseline measurements were taken for at least 30 min to confirm their stability. A 100 Hz 1 s tetanus was given at time zero in the graphs, and then fEPSPs were collected for 2 hr. The curves represent averages of IO slices from nine CREB- mutants, and 6 slices from six wild-type mice. Each point is a 5 min average of 10 fEPSP slope measurements. The inset graph shows peak PTP, the measure- ments of which were started 15-20 s after the tetanus. No significant difference was seen between wild types and mutants in the first 2 min after the tetanus. The inset (averages of 10 consecutive traces) shows LTP still present at 90 min in a wild-type slice (trace I), but no LTP in a mutant slice (trace 2). Closed symbols represent controls, and open symbols represent mutants throughout. Scale bar, 300 pV/lO ms. (B) The figure summarizes the analysis of basal synaptic transmission in CREB homozygotes and wild-type hippocampal slices. The ratio of the EPSP slope to the fv amplitude, over a range of stimulus strengths, is taken as a measure of basal synaptic transmission. The rate of stimulation was either one stimulus per 30 s or one stimulus per 60 s. The slope& ratio is approximately 1.6 times larger in the mutants than in wild types. Thisanalysis issignificantlydifferent if all the individ- ual ratios from all slices are compared (n = 63 for mutant, n = 141 for controls; p < O.OOS), if the average ratio from each slice is compared (n = 13 for mutants, n = 32 for controls; p < O.OOS), or if the average ratio per animal is compared (n = 11 for mutants, n = 21 for controls; p < 0.01). (C)A plot of the percent facilitation (the slope of the second response expressed as a percentage of the slope of the first response; PPF) versus interpulse intervals for CREB mutants (nine slices from eight animals) and controls (sixteen slices from fifteen animals).

in memory or whether these two phenomena are unrelated consequences of the CREB- mutation.

We also found that even as early as 10 min after tetanic conditioning, the level of synaptic potentiation in CREB- mutants (124% f 3%) was significantly lower (p < 0.05) than that in controls (144% -c 6%). However, peak PTP, measured during the first 2 min after the tetanus, was no different (p > 0.05) in mutants and controls (236% -c 6% and 253% -c 1 O%, respectively; Figure 7A).

An analysis of input-output curves in untetanized slices revealed an approximately 1.6-fold increase in the ratio of fEPSP slope to fiber volley amplitude (fEPSP/fv) in the mutants (Figure 78). This indicates that for a given fiber volley (i.e., presynaptic axonal activation), there is a larger postsynaptic response in untetanized mutant neurons as if some of the synapses were already in a potentiated state in the mutants. Alternatively, the disruption of the CREB transcription factor might have prevented the expression of a protein required for the down-regulation of neurotrans- mitter release.

To establish a possible locus for the fEPSP/fv increase in CREB mutants, PPF was analyzed. PPF is an enhance- ment of synaptic responses believed to be the result of presynaptic mechanisms (McNaughton, 1962; Muller and Lynch, 1989; Hess et al., 1967; Zalutsky and Nicoll, 1990). Figure 7C shows that, at all interpulse intervals measured, PPF was indistinguishable in mutant (n = 9 slices from 6 mice) and wild-type slices (n = 16 slices from 15 mice). This argues against a possible increased probability of re- lease in mutants accounting for their greater basal fEPSP/ fv ratio. If the initial probability of neurotransmitter release was higher, PPF would be expected to be lower in the mutants. Alternative explanations include differences in the number of release sites for transmitter, in the postsyn- aptic regulation of membrane potential, or in postsynaptic receptor responsiveness. Future studies with single-cell recordings such as whole-cell patch clamping might distin- guish among these possibilities. Nevertheless, finding nor- mal PPF and PTP indicates that the CREB mutation does not affect all aspects of synaptic function.

Discussion

Studies with Aplysia cultured neurons suggest the involve ment of CAMP and CREB in the expression of genes re- quired for long-term synaptic plasticity (Dash et al., 1990; Kaang et al., 1993; Alberini et al., 1994). In remarkable parallel with these findings, a late stage of hippocampal LTP appears to be dependent both on protein synthesis and on an increase in PKA activity triggered by CAMP (Stanton and Sarvey, 1964; Frey et al., 1988, 1993; Rey- mann, 1993; Huang and Kandel, 1994). Importantly, for- ward and reverse genetic studies in Drosophila showed that mutations in genes involved in CAMP signaling path- ways (CAMP phosphodiesterase, adenylyl cyclase, and protein kinase A) affected olfactory learning (reviewed by Tully, 1991). All of these observations suggest the involve- ment of CAMP-dependent processes in memory formation. In this manuscript, we evaluate the role of CREB-dependent

mechanisms in memory by studying mice with a genetic disruption of the a and 6 isoforms of CREB.

A key aspect of our results is that the loss of CREB function disrupts long-term memory for cued and contex- tual conditioning without affecting initial memory. Conse- quently, it is unlikely that either motor, sensory, motiva- tional, or attention abnormalitiescould explain the memory deficits of the CREB- mutants. These results are consis- tent with the hypothesis that CREB-dependent transcrip- tion is required for memory consolidation (Dash et al., 1990). An accompanying manuscript (Yin et al., 1994) also reports compelling evidence for the involvement of CREB in Drosophila memory formation. Induction of a dominant negative CREB transgene, which blocks CAMP-respon- sive transcriptional activation, disrupts long-term memory but not learning, early memory, or several performance measures (Yin et al., 1994). Other behavioral studies in different species have also demonstrated that formation of long-term, but not short-term, memories can be blocked by agents that interfere with transcription or translation (Davis and Squire, 1984).

It is possible that the CREB- mutation in mice has subtle effects on developmental processes in brain structure(s), such as the hippocampus, that are involved in memory formation (Eichenbaum and Otto, 1992). Hippocampal le- sions, however, are probably not the cause for the memory deficits in CREB mutants since CREB mutants show nor- mal retention of hippocampal-dependent memories for 30-80 min, while animals with hippocampal lesions show normal recall only for a few seconds to minutes after train- ing in a variety of hippocampal-dependent tasks (Squire, 1992).

Our studies with the water maze suggest that the mem- ory deficits observed in the mutant mice are not restricted to Pavlovian tasks. The water maze studies also confirmed that extended training can partly overcome the profound memory deficits of the mutants, suggesting that other mechanisms can compensate for the mutation of CREB. Consistent with the hypothesis that the CREB mutation has a selective effect on memory, we found that the CREB- mutants were normal in most other aspects studied. They appear groomed and healthy, and they gain weight nor- mally. We found no evidence of ataxia or of any other movement and posture disorders. We did not observe any tremors, shakes, head nods, unusual sudden bursts of activity, or any other hints of possible seizure activity. Their nociceptive reactions to a range of foot shocks are normal, and they show normal freezing responses. Our cued con- ditioning studies demonstrate that they can hear the tone used for cued conditioning and that they show normal memory 80 min after training in this task. Contextual condi- tioning studies indicate that the CREB- mutants can distin- guish between contexts, suggesting that the mutation did not disrupt perception. That hippocampal- and amygdala- dependent learning are normal in CREB- mutants was suggested by observing normal memory for both cued (80 min) and contextual (30 min) conditioning. Additionally, our water maze studies indicated that they also have the required motivation and motor coordination to search for the hidden platform in the Morris water maze.

Despite the ubiquitous expression of the CREB gene, the CREB- mutation has a surprisingly restricted impact on developmental and behavioral processes. Previous studies of the CREB- mutants found a 2- to 3-fold increase in the expression of the CAMP-responsive modulator, an- other CREElike transcription factor (Hummler et al., 1994). The expression of both the activator (a and p) and the repressor (7) isoforms of CAMP-responsive modulator was increased. Disruption of the a and 6 isoforms of CREB, however, did not affect another closely related transcrip- tion factor, activating factor 1 (Hummler et al., 1994). Fur- thermore, a new CREB splice variant has recently been identified. This isoform lacks pat-f of the transactivation domain but is still capable of transcriptional activation. This isoform is present in wild-type mice and in the CREB- mutants (J. B. and G. S., unpublished data). The presence of a CREB isoform as well as the up-regulation of the CAMP-responsive modulator could explain the unexpected restricted impact of the disruption of the CREB mutation.

To address the physiological basis of the long-term memory deficits of the mutants, we studied LTP, a candi- date memory mechanism (Bliss and Collingridge, 1993; Eichenbaum and Otto, 1993). Consistent with Hebbian models of learning and memory, our electrophysiological studies found abnormal LTP in CREB- mutants. The LTP of these mutants is smaller and decays to baseline within 90 min of induction. In contrast, other forms of synaptic plasticity, such as PPF and PTP, are normal. Future stud- ies will determine whether the decay in LTP is correlated with the decay in memory or whether these two phenom- ena are unrelated consequences of the CREB- mutation.

The behavioral and electrophysiological findings re- ported in this manuscript identify CREB as an important component of memory consolidation in mammals. These findings, together with studies in Aplysia and Drosophila, strongly suggest that CREB is an evolutionary conserved component of the molecular cascade of events leading to memory consolidation.

Experlmental Procedures

Raising and Oanotyping the Mice In the experiments described, controls always included littermates of the mutants. At 4-5 weeks postnatally, the animals were weaned, and their genotypes were determined with Southern blot analysis of tail DNA samples. All experiments were done with mice that were 3-6 months old. The mice were kept on a 12:12 light-dark cycle, and the experiments were always conducted during the light phase of the cy cle. With the exception of testing times, the mice had ad lib access to food and water. The Cold Spring Harbor Laboratory animal facility is fully accredited by the American Association for the Accreditation of Laboratory Animal Care, and animals are maintained in accordance with the Animal Welfare Act and the Department of Health and Human Services guide.

Noclception Tests We measured the sensitivity of mutants and controls to foot shock. In this test, the mice were placed in the conditioning chamber (described below) and were then given 1 s shocks of increasing intensity. The interval between shocks was 10 s, and the sequence of currents used was as follows: ,075 mA, .l mA, .25 mA, .35 mA, .45 mA, .55 mA, .65 mA. and .75 mA. We determined the level of current required to elicit the following behaviors: flinching, running/jumping, and vocalization. These experiments were performed blindly.

Deficient Long-Term Memory in CREB- Mice 67

Fear Conditioning Our conditioning chamber was in a soundproof box (72 cm x 51 cm x 48 cm). With surrounding noise measuring 75 dB, noise within the box registered 88 dB. A clear Plexiglas window (2 cm thick, 12 cm x 20 cm) allowed the experimenter to observe the mice. To provide background white noise (72 dB), a single computer fan was installed in one of the sides of the isolation chamber. The conditioning chamber (33 cm x 20 cm x 22 cm) is made of transparent Plexiglas on two sides and metal on the other two. Each of the metal sides has a speaker and a 24 V light. The chamber has a 36-bar insulated shock grid floor. The floor is removable, and after each experimental subject, we cleaned it with 75% ethanol and then with water. Each bar (1.5 cm in diameter) is connected through a harness to a Master Shocker (model 82404SS), a device that delivers scrambled shocks. The Son- alert speaker used to deliver the conditioned stimulus is connected to a power supply with an adjustable current output that was kept constant throughout the experiments. Only one subject at a time was present in the experimentation room. The other subjects remained in their home cages. Each subject was carried to the behavioral room (5 R away) in a cage with shavings from its home cage, and the proceed- ings were filmed.

For the cued and contextual conditioning experiments, mice were placed in the conditioning chamber for 2 min before the onset of the discrete CS (lasted 30 s at 2800 Hz and 85 dB of sound). In the last 2 s of the CS, they were exposed to the US (at 0.75 mA for 2 s of continuous foot shock). After the CSAJS pairing, the mice were left in the conditioning chamber for another 30 sand were then placed back in their home cages. Conditioning was assessed by measuring “freez- ing”: the animals were judged as either completely immobile or not completely immobile (respiratory movements are not counted) in inter- vals of 5 s. If the animal did not move in that interval, we scored it positively. For contextual conditioning, freezing was measured for 5 min (consecutive) in the chamber in which the mice were trained. Groups 1,2, and 3 were tested at 0.5 hr, 1 hr, and 24 hr after training, respectively. In cued conditioning, the mice were placed in a novel context (triangular cage with smooth flat floor and with lily odorant) for 3 min (pm-CS test), after which they were exposed to the CS for 3 min (CS test). Groups 1, 2, 3, and 4 were tested 0.5 hr, 1 hr, 2 hr, and 24 hr after training, respectively. For both the cued and contextual conditioning experiments, the mice were trained in three sessions. Mice from each session were tested at each of the different time points tested. Since the experiments were done in a balance manner and since the training procedure was identical, we pooled all of the training data for each of the cued and contextual conditioning experiments. In the experiments with additional training trials, three CSAJS pairings were given, with a 1 min interval between shocks. The experimenter was blind to the genotype of the subjects.

Water Maze Tests Mice tested in the water maze were extensively handled (2 min every day for 10 days). We tested approximately equal numbers of male and female mice 12-15 weeks old. Only mice housed in groups of 3 (those heavier than 25 g per mouse) or 4 (those lighter than 25 g per mouse) per standard mouse cage were used. Before each experiment (2 hr). the mice were brought to the water maze room to allow them to acclima- tize to it. Our pool is 1.2 m in diameter and is made of polypropylene white plastic. The test subjects were kept on shelves underneath the pool to eliminate directional olfactory and auditory cues. Underneath the pool, a thermoregulated spiral coil keeps the water temperature at 28°C -c 1°C. The temperature of the room is also kept at 28OC. The water in the pool is made opaque with nontoxic white paint to hide the escape platform. The Plexiglas platform is IO cm x 10 cm, and its borders are slanted and scarred to help the mice climb onto it. The water surface is 10 cm from the rim of the pool, and the white plastic inner wall is always carefully wiped to eliminate any local cues. The rim of the pool is 1.5 m from the nearest visual cue. The walls of the room are painted white, and each wall has a single salient cluster of cues (e.g., a dark soccer ball). The room has adjustable indirect illumination. A camera is fixed to the ceiling of the room, 1.5 m from the water surface. The camera is connected to a digital tracking device (VP118, HVS Image). The tracking information is then processed by an IBM computer with the HVS Water Maze software.

On the first day of training, the mice are placed on the platform for

30 s. Then they were allowed one 30 s practice swim as well as three platform climbs. They are allowed to rest for another 30 s on the plat- form, and then testing was initiated. During training in the hidden platform test of the Morris maze, the platform was not marked by any cue, and it was kept in the same place throughout training. The mice were divided into groups of two or three animals. Each of these groups was trained in a different platform site to avoid quadrant biases. In a block of trials, the starting position of the mice was varied pseudoran- domly so as to ensure that all four starting positions were used. The mice were either trained with twelve trials per day for 3 days or with one trial per day for 15 days. Each trial started with the mice facing the wall of the pool and ended when they climbed the platform. The mice were not allowed to search for it for more than 60 s, after which they were placed on the platform. At the end of each trial, they always stayed on the platform for 60 s. The mice were tested at approximately the same time every day.

In the probe test used, we removed the platform and measured the time the mice spent in the quadrant in which the platform was located during training, and we counted the number of times the mice crossed the platform site during searching. All experiments were done blind to the genotype of the subjects.

Studies of LTP Hippocampal slices (400 urn thick) were prepared by standard meth- ods. The mice tested were used in behavioral studies 2-3 weeks be- fore. Experiments were conducted in medium containing 125 mM NaCI, 2 mM KCI, 1.25 mM NaH2P0,, 26 mM NaHC03, 10 mM D-glucose, 1.5 mM MgS04, and 2.5 mM CaCI,. Stimulating bipolar concentric and recording (1-5 MD filled with cerebrospinal fluid) elec- trodes were placed in stratum radiatum in CA1 to record fEPSPs. Stimulus intensity was determined by first using a range of stimulus strengths to define a synaptic input-output curve and then by choosing an intensity that produced a response that was 25%-35% of the maxi- mum evoked response with a 250 ms pulse width (stimulation rate, 0.033 Hz or 0.067 Hz). This intensity was then used for the baseline, PPF, tetanic stimulation, PTP, and LTP. Initial slope of EPSPs was used in all analysis. In the LTP studies, only those slices that had a stable baseline for 30 min were included in the analysis. Experiments were carried out in a submerged slice chamber at room temperature (20°C-230C). In the majority of the studies, the experimenter was blind to the genotype of the animals.

Data Analysls Results from male and female mice were combined since we did not find any significant differences between them. For the analysis of ex- periments, we used analysis of variance (ANOVA) with repeated mea- sures and ANOVA with dependent measures. Planned comparisons were used for post-hoc analysis. All values in the text and figure leg- ends are expressed as mean f SEM.

Acknowledgments

All correspondence should be addressed to A. J. S. We thank K. P. Giese, H. Nawa, S. Rose, J. Sarvey, and T. Tully for invaluable discus- sions that helped to shape this manuscript, and we thank 2. Marowitz for help with data analysis. This work was funded by grants from the Cold Spring Harbor Laboratory Association, the Whitehall Foundation, the Klingenstein Foundation, and the Beckman Foundation to A. J. S., the Wellcome Trust (U. K.) to B. G. F., and the Deutsche Forschungs- gemeinschaft, Leibniz Programm and the Fonds der Chemischen In- dustrie to G. S.

References

Alberini, C. M., Ghirardi, M., Metz, Ft., and Kandel, E. R. (1994). ClEBP is an immediate-early gene required for the consolidation of long-term facilitation in Aplysia. Cell 76, 1099-I 114.

Bliss, T. V. P., and Collingridge, G. L. (1993). A synaptic model of memory: long-term-potentiation. Nature 367, 31-39.

Brandeis, R., Brandys, Y., andYehuda, S. (1989). TheuseoftheMorris water maze in the study of memory and learning. Int. J. Neurosci. 48, 29-69.

Cell 88

Dash, P. K., Hochner, B., and Kandel, E. R. (1990). Injection of the CAMP-responsive element into the nucleus of Ap/ysia sensory neurons blocks long-term facilitation. Nature 345, 718-721.

Davis, t-l. P.. and Squire, L. R. (1984). Protein synthesis and memory. Psychol. Bull. 96, 518-559.

Dickinson, A., and Mackintosh, N. J. (1978). Classical conditioning in animals. Annu. Rev. Psychol. 29, 587-612.

Eichenbaum, H., and Otto, T. (1992). The hippocampus -what does it do? Behav. Neural Biol. 57, 2-36.

Eichenbaum, l-l., and Otto, T. (1993). LTP and memory: can we en- hance the connection? Trends Neurosci. 16, 163-164. Fanselow, M. F. (1986). Associative versus topographical accounts of the immediate shock-freezing deficit in rats: implications for the response selection rules governing species-specific defensive reac- lions. Learning Motiv. 77, 16-39.

Fanselow, M. S., and Belles, R. C. (1979a). Naloxone and shock- elicited freezing. J. Comp. Physiol. Psychol. 93, 738-744. Fanselow, M. S., and Bolles, R. C. (1979b). Triggering the endorphin analgesic reaction by a cue previously associates with shock: reversal by naloxone. Bull. Psychon. Sot. 14, 88-90.

Flood, J. F., Bennet, E. L., and Orme, A. E. (1975). Relation of memory formation to controlled amounts of brain protein synthesis. Physiol. Behav. 15, 97-102.

Frey, U., Krug, M., Reymann, K. G., and Matthies, H. (1988). Aniso- mycin, an inhibitor of protein synthesis, blocks late phases of LTP phenomena in the hippocampal CA1 region in vitro. Brain Res. 452, 57-65.

Frey, U., Huang, Y., andKandel, E. R.(1993). Effectsof cAMPsimulate a late stage of LTP in hippocampal CA1 neurons. Science 260,1681- 1684. Hess, G., Kuhnt, U., and Voronin, L. L. (1987). Quanta1 analysis of paired-pulse facilitation in guinea pig hippocampal slices. Neurosci. Lett. 77, 187-192.

Huang, Y.-Y., and Kandel, E. R. (1994). Recruitment of long-lasting and protein kinase A-dependent long-term potentiation in the CA1 region of hippocampus requires repeated tetanization. Learning Mem- ory 7, 74-82.

Hummler, E., Cole, T. J., Blendy, J. A., Ganss, R., Aguzzi, A., Schmid, W., Beermann, F., and Schutz, G. (1994). Targeted mutation of the CAMP response element binding protein (CREB) gene: compensation within the CREB/ATF family of transcr’@tion factors. Proc. Natl. Acad. Sci. USA 91, 5847-5851.

Kaang, B.-K., Kandel, E. R., and Grant, S. G. N. (1993). Acitivalion of cAMP-responsive genes by stimuli that produce long-term facilita- tion in Aplysia sensory neurons. Neuron IO, 427-435. Kim, J. J., and Fanselow, M. S. (1992). Modality-specific retrograde amnesia of fear. Science 256, 675-677. Kim, J. J., Fanselow, M. S., DeCola, J. P., and Fernandez, J.-L. (1992). Selective impairment of long-term but not short-term conditional fear by the N-Methyl-o-aspartate antagonist APV. Behav. Neurosci. 106, 591-598.

Kim, J. K., Risen, R. A., and Fanselow, M. S. (1993). Effects of amyg- dala, hippocampus, and periaqueductal gray lesions on short- and long-term contextual fear. Behav. Neurosci. 707, 1093-1098. Lee, K. A., and Masson, N. (1993). Transcriptional regulation by CREB and its relatives. Biochim Biophys Acta 7 7 74, 221-233.

Matthies, H. (1989). In search of cellular mechanisms of memory. Prog. Neurobiol. 32, 277-349.

McNaughton, B. L. (1982). Long-term synaptic enhancement and short-term potentiation in rat fascia dentataact through different mech- anisms. J. Physiol. 324, 249-262. Miller, N. E., and Weiss, J. M. (1969). Effects of the somatic or visceral responses to punishment. In Punishment and Aversive Behavior, B. A. Campbell and R. M. Church, eds., (New York: Appleton-Century- Crofts).

Morris, R. G. M. (1981). Spatial localization does not require the pres- ence of local cues. Learning Motiv. 12, 239-260. Morris, R. G. M., Garrud, P., Rawlins, J. N. P., and C’Keefe, J. (1982).

Place navigation impaired in rats with hippocampal lesions. Nature 297,681-883. Muller, D., and Lynch, G. (1989). Evidence that changes in presynaptic calcium currents are not responsible for long-term potentiation in hip- pocampus. Brain Res. 479,290-299.

Phillips, R. G., and LeDoux, J. E. (1992). Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav. Neurosci. 706, 274-285.

Reymann, K. G. (1993). Mechanisms underlying synaptic long-term potentiation in the hippocampus: focus on postsynaptic glutamate re- ceptors and protein kinases. Funct. Neurol. (Suppl.) 5, 7-32.

Rudy, J., and Morledge. P. (1994). Ontogeny of contextual fear condi- tioning in rats: implications for consolidation, infantile amnesia, and hippocampal system function. Behav. Neurosci. 106, 227-234.

Silva, A. J., Paylon, R., Wehner, J. M., and Tonegawa, S. (1992). Impaired spatial learning in a-calcium-calmodulin kinase II mutant mice. Science 257, 206-21 I.

Squire, L. R. (1992). Memory and the hippocampus: a synthesis from findings with rat, monkeys, and humans. Psychol. Rev. 99, 195-231.

Stanton, P. K., and Sarvey, J. M. (1984). Blockageof long-term potenti- ation in rat hippocampal CA1 region by inhibitors of protein synthesis. J. Neurosci. 4, 3080308.8.

Sutherland, R.J., Kolb, B., andwhishaw, l.Q(l982). Spatialmapping: definitive disruption by hippocampal or medial frontal cortical damage in the rat. Neurosci. Lett. 31, 271-276.

Tully, T. (1991). Genetic dissection of learning and memory in Drosoph- ila melanogaster. In Neurobiology of Learning, Emotion and Affect, J. Madden IV, ed., (New York: Raven Press), pp. 30-66. Yin, J., Wallach, J. S., Del Vecchio, M., Wilder, E. L., Quinn, W. G., and Tully, T. (1994). Induction of a dominant negative CREB transgene specifically blocks long-term memory in Drosophila. Cell, this issue.

Zalutsky, R. A., and Nicoll, R. A. (1990). Comparison of two forms of long-term potentiation in single hippocampal neurons. Science 246, 1619-1624.