Penelope would have been sad indeed had she realized that each time she was reminded of her beloved Odysseus — away from home for so many years after the Trojan war1 — she could entirely lose her precious memory of him. Fortunately this was also unknown to her pushy suitors. But now Nader and colleagues, writing on page 722 of this issue2, have made it public. Recollection, they claim, is a dangerous matter: whenever we bring a memory to mind, it may turn shaky and slip into oblivion.
Most memories, like humans and wines, do not mature instantly. Instead they are gradually stabilized in a process referred to as consolidation3, 4. Newly formed memory traces are sensitive to a variety of brain injuries and drugs, but after they have been consolidated they become more resistant to these treatments. Consolidation takes place at many levels of organization and complexity in the brain, and its overall kinetics depends on the type of memory involved. We know most about what happens in individual nerve cells and synapses — the points of communication between neurons — once they have been recruited to consolidate a memory.
The current textbook version, in a nutshell, goes like this. Training modifies proteins at synapses in the neuronal circuit that acquires the new memory. This alters synaptic efficacy and thus the encoding of information in that circuit. But protein molecules survive only for periods of minutes to weeks, whereas many memories are destined to live longer. It seems that at least part of the immunity of memory to this molecular turnover is achieved by training-induced modulation of gene expression in the modified neurons. The new gene products promote long-lasting remodelling of the activated synapses, in a process that involves crosstalk between the synapses and neuronal cell bodies5. It takes a few hours for the new pattern of gene expression and the synaptic change to be consolidated. During this time, the process can be halted by inhibitors of protein synthesis5, 6, 7.
This textbook version might tempt one to believe that, for every memorized item, consolidation starts and ends just once. But this view would be naive. Experimental psychologists told us long ago that memory traces are reconstructed with use, and that retrieving a memory involves mingling the representations of the past with the percepts of the present8. The study by Nader et al.2 echoes earlier reports that a consolidated memory can apparently be induced to vanish, provided that the memory is activated shortly before the use of the treatment leading to amnesia9, 10. The problem with these early studies was that, because the treatments were applied to the whole brain or even the whole body, and because little was known about the relevant neuronal circuits, the researchers could not target cellular mechanisms in identified memory traces. This has now changed.
Nader et al. took advantage of 'auditory fear conditioning' in rats. This works as follows. The rat hears a tone (the conditioned stimulus) in conjunction with a mild footshock (the unconditioned stimulus). The electric shock elicits fear (an unconditioned response). After one training session, the tone elicits fear responses, such as freezing, even in the absence of shock (a conditioned response). For readers who are not well versed in the emotional life of rats but do recall the story of Pavlov and his salivating dogs, suffice it to note that the situations are basically similar: both the dogs in Petrograd and the rats in Manhattan had to learn to associate conditioned and unconditioned stimuli. The protocol is therefore aptly dubbed 'pavlovian fear conditioning'. The neuronal circuit underlying pavlovian fear conditioning includes the lateral and basal nuclei of the amygdala. Inhibiting protein synthesis in this brain region immediately after fear conditioning, by infusing the antibiotic anisomycin into this region, blocks long-term fear memory (that existing more than 24 hours after the training), but not short-term memory2.
Knowing all this, Nader et al. trained rats in pavlovian fear conditioning, and tested them 24 hours later with the conditioned stimulus but without the unconditioned stimulus (test 1). The rats froze at the sound of the tone. At this point, when the long-term memory trace was expected to be already insensitive to anisomycin, Nader et al. injected the antibiotic into the amygdala. A day later, the authors tested the rats again with just the conditioned stimulus (test 2). Surprisingly, these rats showed a marked decrease in the time spent freezing in response to the tone. The same results were obtained even if test 1 took place 14 days after training, making it even more unlikely that the inhibition of protein synthesis in test 1 impaired a late phase of consolidation initiated by the original training. Omitting the conditioned stimulus before administering anisomycin in test 1 left memory intact. So the memory probably had to be retrieved for anisomycin to have its effect. The anisomycin was effective only if administered within a few hours after memory reactivation.
So it seems that fear-associated memories become temporarily labile on retrieval. Why should the brain invest so much energy in the original consolidation and then risk losing the trace by interference each time it is used? One can come up with teleological explanations — for example, that the brain prefers plasticity at the expense of stability — or mechanistic ones, suggesting in-built constraints on the synaptic machinery. But there is still much to do before we can jump to any sweeping conclusions about the cellular biology of memory retrieval. Some unanswered questions relate specifically to this experiment. Did tests 1 and 2 indeed tap the same memory trace? Did anisomycin abolish the original trace, or merely leave it dormant, waiting to be exposed by some smart behavioural protocol? Which cellular mechanisms are perturbed by anisomycin after retrieval, and are they are identical to those that produce the original consolidation?
More generally, might these results apply to different types of memory? Previous studies hinted that pavlovian fear conditioning may not be unique in being shaky on retrieval9, 10. But even if just a few types of memory must reconsolidate after use, the implications of the results of Nader et al.2 are remarkable. Consider, for example, the prospect of intentionally recalling the memory of a traumatic experience and then selectively erasing it. What such a possibility would mean for psychoanalysts on the one hand, and poets on the other, is quite a different matter.
