From simple synaptic rearrangements to complex neural networks, researchers are wrestling mightily with memory, one of the greatest biological mysteries. Writing in the April 5 Cell, Nahum Sonenberg and colleagues report that phosphorylation/dephosphorylation of a translational regulator acts as a simple toggle for short- and long-term memory. Tackling memory in its broadest sense, Richard (aka water maze) Morris and colleagues report in the April 6 Science that mice use mental schemas, or frameworks, to rapidly consolidate new memories. And Gordon Winocur and colleagues report in the April 1 Nature Neuroscience that such consolidation is not simply the transfer of memories from the hippocampus to the neocortex, but rather a transformation of the original memory into something that is more schematic and independent of context. Together, the three papers typify the molecular and organismal approaches that scientists are taking to understand various intricacies of memory—not a moment too soon, as the number of people with memory disorders is set to rise alarmingly over the next decades.
Sonenberg, together with colleagues at McGill University, Montreal, and other labs in Canada, the U.S., and Israel, examined the role of eukaryotic initiation factor 2α (eIF2α) in long-term potentiation (LTP). This translation factor has been implicated in memory before. While dephosphorylation of eIF2α generally stimulates translation, it has the opposite effect on the transcription factor ATF4, suppressing its translation. ATF4 blocks transcription of genes driven by cAMP responsive element promoters, many of which are required for LTP, and therefore dephosphorylation of eIF2α should, in theory, help learning and memory.
To examine the role of eIF2α phosphorylation in memory in detail, first author Mauro Costa-Mattioli and colleagues turned to mice expressing mutant forms of the protein, which cannot be phosphorylated at serine 51, a key regulatory site for the protein’s translational activity. The researchers generated mice that are either homozygous or heterozygous for eIF2αS51A, where serine is substituted with an alanine residue. The homozygous animals were not viable, but the heterozygotes appeared normal, with no detectable aberrant morphology in the hippocampus or the brain in general.
The researchers tested hippocampal slices from these animals for LTP. While a short burst of high-frequency stimulation elicited short-lasting, or early, LTP (E-LTP) from wild-type tissue slices, the same stimulation resulted in sustained, or long, LTP (L-LTP) in tissue from the eIF2αS51A+/- animals. L-LTP is known to require protein synthesis, and indeed the researchers found that the L-LTP in the mutant tissue could be blocked by inhibitors of protein synthesis.
The finding suggests that phosphorylation of eIF2α is sufficient to suppress L-LTP, but does that affect memory? To test this, Costa-Mattioli and colleagues put mutant and wild-type mice through their paces in the Morris water maze, a widely used test of spatial learning and memory. In this paradigm, mice typically undergo three to four training sessions per day in the maze, and the time they need to find a hidden platform gets progressively shorter as they learn its position. But in the case of the mutant animals, the researchers found that once-a-day training was sufficient. After six days, the mutant mice were performing almost as well as wild-type mice that had been trained three times a day. Furthermore, when the mutant animals were trained three times a day, then by the second day they were already making improvements over their wild-type cousins. All told, the research suggests that simply inhibiting phosphorylation of eIF2α could substantially boost long-term memory and learning. The same was true in other learning paradigms, such as contextual fear conditioning and a taste aversion test. In both cases, the eIF2αS51A+/- animals performed significantly better than wild-type animals.
“These findings also raise the interesting possibility that regulators of translation could serve as therapeutic targets for the improvement of memory, for instance, in human disorders associated with memory loss,” write the authors. Such an approach could have dramatic systemic consequences. For example, work in yeast shows that GCN kinases have evolved to help regulate basic metabolism in response to nutrient availability, and recent evidence indicates that GCN2 is a key regulator of fatty acid homeostasis in the liver (see Guo and Cavener, 2007). Nonetheless, the authors showed that, in principle, it might work. They were able to suppress memory using Sal003, a derivative of the phosphatase inhibitor salubrinal. Sal003 inhibits the phosphatase that dephosphorylates eIF2α so it should, prevent LTP and suppress memory. This is what the researchers found. The drug significantly impaired LTP in normal hippocampal slices and, when infused into the hippocampus, it suppressed contextual fear conditioning and impaired learning in the water maze. The data imply that if the opposite, that is, eIF2α dephosphorylation, could be achieved, then memory could be improved.
The Bigger Scheme of Things
Let’s face it: it is easier to assimilate new information when one already knows something about the subject matter. This writer would be hard pressed to remember the details of a lecture on quantum physics, or the latest and greatest in civil engineering. Biologically, this type of context-aided memory is related to the concept of mental schemas, or frameworks we build upon as we acquire new memories. But whether animals build similar schemas is unclear, partly because researchers, for the most part, use experimentally naïve animals to test learning and memory.
Work conducted in the laboratory of Richard Morris, University of Edinburgh, Scotland, suggests that rats not only construct such schemas but that memories built upon them are also consolidated in as little as 48 hours. “This is unexpectedly rapid for a process that, on the basis of as many as 20 studies in experimental animals, ordinarily takes at least a month,” writes Larry Squire, University of California, San Diego, in an accompanying Science perspective.
The evidence for rat memory schemas comes from intricate paired-association (PA) studies. First author Dorothy Tse and colleagues trained rats in a large arena in which sand wells can be placed in any of 49 locations and landmarks are added for orientation. Over a period of six weeks, the researchers trained the animals to associate six set locations with six different food smells. Beginning in a start box, the animals were given a whiff of a tasty treat, then allowed to explore until they found the identically flavored treat in one of the wells. The trials were repeated hourly with different flavors until the rats had located all six types of treats. Over 21 such sessions, the animals improved significantly and went from spending only about 18 percent of the time digging in the right spot in probe trials (no treat available) to spending about 35 percent of the time there. To test whether this improvement is related to the formation of some kind of schema or merely due to memorizing simple facts (e.g., bacon equals the well farthest to the left), Tse introduced two new flavor-location pairs on session 21. The animals were given one shot at finding the new treats, then tested 24 hours later to see if they remembered where they were. Surprisingly, they spent almost 50 percent of the time digging in the correct location. “The rapid acquisition of new PAs in a single trial and their retention over 24 hours are indications that the prior learning of an associative schema may aid the encoding, storage, and/or consolidation of new PAs,” write the authors.
So what’s this got to do with consolidation? There is overwhelming evidence to suggest that memories first form in the hippocampus, then somehow transfer over to the neocortex where they are retained in the long term. But when the researchers lesioned the hippocampi 48 hours after presenting the two new flavor-location pairs, in a subsequent probe trial the animals remembered the locations just as well as sham-operated rats. The findings suggest that “(i) the memory traces for these PAs must be stored outside the hippocampus, probably in the neocortex; and (ii) consolidation of new associates whose acquisition is mediated by the hippocampus takes place within 48 hours,” write the authors. That the hippocampal-lesioned animals failed to remember a third new flavor-location pair, introduced post-operation, supports this idea that the hippocampus is involved in learning the new locations.
To be sure the animals are laying down a schema and not just learning a methodology, the researchers continued testing the same animals but this time in a new arena with new landmarks. Here the animals were back to square one, and their rate of learning was no greater than when they were first trained. This suggests that they are not learning a methodology, but rather learning the geography of the arena and putting new information into that context. The hippocampus-lesioned animals could not learn this new arena, as would be expected, but remarkably, when they were put back into the original arena in which they had been trained 4-5 months earlier, they could remember the locations and even trended toward doing better than sham-operated animals, possibly because they were not confused by memories of the new arena. The authors note that the rats’ ability to remember the original arena is similar to the now famous patient E.P., who has helped Squire and others study the biology of human memory. E.P. suffered damage to the hippocampus, and while he readily remembers how to navigate the town he grew up in, he cannot navigate new locations.
In one last test of the schema idea, Tse and colleagues compared animals trained in a fixed arena with those trained in an arena where the flavor-place combinations were switched every two sessions (the inconsistent arena). Those trained in the latter did improve slightly over time, but when presented with the single-trial chance to remember a new flavor-place pair, they failed, “presumably because they had not established a stable schema that could guide rapid learning,” wrote Squire.
“These findings indicate that animals—no less than people—can bring activated mental schemas to bear in a [paired association] learning task and thereby encode, assimilate, and rapidly consolidate relevant new information after a single trial,” write the authors. As for the implications for consolidation, Tse and colleagues acknowledge the large number of studies that suggest consolidation is a slow process, but suggest that they may not “quite capture the potential that the neocortex has for rapid consolidation when newly acquired information is compatible with previously acquired knowledge.”
Squire also points out that the study speaks to our fundamental understanding of memory itself. “The fact that storage and recall of spatial memory can occur independently of the hippocampus runs counter to the proposal that the hippocampus forms and stores cognitive maps,” he writes. Indeed recent studies suggest that, as time passes, the hippocampus becomes less important for given memories and the cortical regions take over (for review see Frankland and Bontempi, 2005). “The idea is not that memory is literally transferred from hippocampus to neocortex, but rather that the hippocampus guides gradual changes in the neocortex that increase the complexity, distribution, and interconnectivity of memory storage sites,” Squire writes, and this brings us to the third paper. Gordon Winocur and colleagues at the University of Toronto, Canada, addressed the question of whether consolidation involves transferring memories in their original, detailed form from the hippocampus to the neocortex, or whether the memory is transformed into something more generic and schematic.
The scientists come down in favor of the latter. They tackled the question by testing rats in two learning paradigms: contextual fear conditioning and a food preference task. In the latter, test rats are allowed to smell the breath of a cage-mate that has just eaten a tasty and aromatic morsel. When presented with a food choice, the test rat then usually goes for the food with the familiar smell rather than something with an unfamiliar aroma.
To test if the memories are consolidated in some sort of generic form, Winocur and colleagues trained the rats in the two learning paradigms, then tested them at either short (1 day) or long intervals (8 days for the smell test and 28 days for the contextual fear conditioning test). The crux of the paper comes down to the following: in addition to repeating the test verbatim, the researchers also measured how well the rats performed when given the task in a new context. If consolidation depends on context, then the animals should perform less well in the new context than in the familiar context at both short or long intervals, but if consolidation involves building a generic schema, then the context may become less important at long intervals—after consolidation has occurred. That is exactly what the researchers found.
In the food preference test, rats did poorly in the new context when tested there one day after training, but 8 days later they were doing even better than those tested in familiar contexts. Similarly, one day after fear conditioning the rats “froze” only about one-third as often as when tested in the familiar context, but 28 days later, the context had no effect whatsoever. The hippocampus was required for these effects because lesioned animals failed to show any context-dependent differences at either interval.
“The finding that memory was context-sensitive at relatively short delays and that this sensitivity was mediated by the hippocampus is consistent with the view that, initially, memories are inextricably linked to details of the environment in which they were formed,” write the authors. “With the passage of time, contextual sensitivity is diminished, and the learned response can be elicited by new environments that share general, but not specific, features of the original learning situation.” Winocur and colleagues also note that this does not imply context-dependent memory is lost, merely that the “gist” version, represented by the transformed memory, also will come into play. “The extent to which one or the other influences behavior likely depends on several factors,” they write.—Tom Fagan
- Guo F, Cavener DR. The GCN2 eIF2alpha kinase regulates fatty-acid homeostasis in the liver during deprivation of an essential amino acid. Cell Metab. 2007 Feb;5(2):103-14. PubMed.
- Frankland PW, Bontempi B. The organization of recent and remote memories. Nat Rev Neurosci. 2005 Feb;6(2):119-30. PubMed.
- Arendash GW, King DL, Gordon MN, Morgan D, Hatcher JM, Hope CE, Diamond DM. Progressive, age-related behavioral impairments in transgenic mice carrying both mutant amyloid precursor protein and presenilin-1 transgenes. Brain Res. 2001 Feb 9;891(1-2):42-53. PubMed.
- Tse D, Langston RF, Kakeyama M, Bethus I, Spooner PA, Wood ER, Witter MP, Morris RG. Schemas and memory consolidation. Science. 2007 Apr 6;316(5821):76-82. PubMed.
- Winocur G, Moscovitch M, Sekeres M. Memory consolidation or transformation: context manipulation and hippocampal representations of memory. Nat Neurosci. 2007 May;10(5):555-7. PubMed.
- Costa-Mattioli M, Gobert D, Stern E, Gamache K, Colina R, Cuello C, Sossin W, Kaufman R, Pelletier J, Rosenblum K, Krnjević K, Lacaille JC, Nader K, Sonenberg N. eIF2alpha phosphorylation bidirectionally regulates the switch from short- to long-term synaptic plasticity and memory. Cell. 2007 Apr 6;129(1):195-206. PubMed.
- Squire LR. Neuroscience. Rapid consolidation. Science. 2007 Apr 6;316(5821):57-8. PubMed.