Regulation of gene expression and its role in long-term memory and synaptic plasticity.

Histories of science yet to be written will view the latter half of this century as the Age of Molecular Genetics. From a flash of insight that yielded the double helix (1) to the first genetic clone of a mammal (2), molecular genetics has invaded every aspect of biological research. Initially, this molecular-genetic onslaught was limited to species, such as bacteria, yeast, nematodes, and fruit f lies, whose size and life cycle constituted an economy of scale that was advantageous to breeding (3). With the introduction of gene-knockout techniques to mice (4), however, molecular genetics now is storming mammals (5, 6). In the broadest sense, the recent paper by Guzowski and McGaugh (7) represents a vanguard of this invasion. By using antisense oligonucleotides as ‘‘pharmaceutical’’ disruptors of gene expression, they have liberated molecular genetics from breeding. Endogenous regulation of gene expression has been outflanked by exogenous control. In a narrower sense, Guzowski and McGaugh’s work comprises the latest installment in an emerging theme in molecular biology. Basic molecular and cellular processes appear to have evolved early in the animal kingdom, and they underlie more complex functions from development to behavioral plasticity. Time and again, biologists working on similar problems in diverse species have stumbled upon homologous genetic mechanisms—leading neurogeneticist J. C. Hall at Brandeis University to remark, ‘‘We’re all working on the same genes, we just don’t know it yet.’’ Molecular–genetic studies of behavioral plasticity began more than 25 years ago with Seymour Benzer and coworkers (8), who identified dunce and rutabaga from forward-genetic screens for single-gene mutants with defective associative learning. Biochemical experiments established, and molecular cloning later confirmed, that these two genes encoded a cAMP-specific phosphodiesterase and adenylyl cyclase, respectively. Subsequent reverse-genetic experiments then demonstrated similar behavioral defects from gene disruptions of an a subunit of G protein, a catalytic subunit of cAMPdependent kinase (PKA), and a regulatory subunit of PKA. Thus, cAMP signaling was implicated in Drosophila learning, as it also was in Aplysia (7). Across the animal kingdom, long-lasting memory is dependent on protein synthesis and, for most tasks, is stronger and longer lasting after spaced training (multiple sessions with a rest interval between each) rather than massed training (multiple sessions with no rest interval between each; ref. 9). In flies, spaced training, in fact, is required to induce protein synthesis-dependent long-term memory (LTM). Conversely, inhibition of protein synthesis completely blocks LTM after spaced training without affecting learning, early memory, or long-lasting memory after massed training (10). From a molecular perspective, protein synthesis often can involve regulation of gene expression by nuclear transcription factors—some of which are known to be cAMP-responsive (11, 12). Thus, Yin et al. (13) cloned dCREB2, a f ly homolog of rat CREB that is alternatively spliced to yield one protein isoform that acts as a cAMP-responsive activator of (CRE-mediated) gene expression and another isoform that acts as a repressor of the activator. Transgenic flies then were bred that inducibly express either CREB repressor or CREB activator. As they had hoped, Yin et al. (14) found that expression of CREB repressor blocks the protein synthesis-dependent LTM induced by spaced training, without affecting learning or early memory formation. This result prompted Bourtchuladze et al. (15) to look at memory formation in mutant mice homozygous for a partial knockout of CREB (16, 17). Here too, LTM of cued or contextual footshock conditioning was found to be disrupted, with no apparent effect on learning or early memory formation. These behavior–genetic studies thus established a conserved role for CREB and the regulation of gene expression in the formation of long-term memory. The integrative use of psychological manipulation (training protocols), pharmacology, and genetics in the study of fruit f ly memory formation led to proper interpretation of results from experiments on transgenic flies carrying CREB activator. Memory formation after spaced training was normal in these flies; induced expression of CREB activator neither enhanced nor suppressed LTM. Instead, maximal LTMwas formed after only one training session—the functional equivalent of a ‘‘photographic memory’’ (18). In molecular-genetic terms, these opposing effects of CREB activator and repressor indicated that CREB acts as a ‘‘molecular switch’’ during the induction of LTM. This notion led Yin et al. (18) to propose a (DC) model of LTM formation, in which the ratio of CREB activators to repressors sets the switch to one of three functional states (Fig. 1). If activator levels predominate, then the switch is ‘‘on’’ and LTM ensues after one training session. If repressor levels predominate, then the switch is ‘‘off’’ and LTM is blocked. If activator and repressor levels are equal, then a single training session or massed training produces only a transient increase in CREB activator, which is insufficient to induce (maximal) LTM. Consequently, spaced training is required to induce LTM. In this state, then, the CREB switch acts as an information filter; the only new experience stored in LTM is that which recurs at discrete intervals (19). Generalization of this DC model of LTM formation to mammals led to the speculation that the activatoryrepressor ratio might reflect the net action of protein isoforms from CREB, CREM, and possibly ATF1, all of which are cAMPresponsive CREB family members (20, 21). In this context, the CREB2 knockout in mice might simply have reduced the activatoryrepressor ratio rather than blocked CREB-mediated gene expression altogether—leading to the prediction that spaced training (but not massed training) would ‘‘rescue’’ the LTM deficit in CREB2 mutant mice. Kogan et al. (22) largely have confirmed this prediction for three different tasks (con-