A biological perspective on memory

Memory enables information to be stored and retrieved after seconds to years and is essential for daily life. This issue of Current Biology takes a broad look at the biology of memory. There is no universal definition of memory, but we consider the term to refer to enduring changes in the mechanisms of behavior based on prior experience with environmental input; the focus here is on specific plasticity systems, methods organisms have evolved to retain information that may be useful at a later time. In practice, memory is in evidence when some observed output at time point B can be attributed to an earlier input experience at time point A. The usual design needed to place that attribution on solid ground is as follows. One group of individuals receives input X at time point A, whereas another group does not (or better, receives some unrelated input Y). Next, after a delay, performance at time point B is said to depend on memory if individuals in the first group perform differently from those in the other group. Notice that other potential explanations for the change in the performance of the first group, such as injury and disease, must first be ruled out, with additional control groups if necessary; furthermore, if behavioral change does not occur in the first group, it does not necessarily indicate the absence of learning — the subjects must be shown to be attending to input X. An everyday example: not buying flowers for a spouse on the occasion of a wedding anniversary can lead to unpleasant associated consequences, so buying flowers each year becomes something to remember. (Mercifully, some situations can yield single-trial learning!) One of the grand challenges of science is to understand the biological mechanisms that support memory.

This issue contains an expansive purview of the biology of memory. Coverage includes a broad range of phyla and species — from bacteria (Escherichia coli) to plants (Arabidopsis) to animals, including invertebrates — Drosophila, stink bugs, wasps, and ants — and vertebrates — mice, rats, monkeys and people; of biological levels of analyses — gene regulation, signal transduction, cellular/synaptic changes, formation of neural networks (biological and synthetic) and alterations in regional brain activity; and of embodiment — bacteria encoding changes in their biochemical environment [1], plants recording yearly differences in the length of seasonal cold and warm episodes [2], the construction of synthetic memory circuits [3], adaptations of immune systems that permit enhanced responsiveness to antigens upon subsequent exposure, thereby targeting pathogens without attacking the rest of the body [4], and, of course, information storage by central nervous systems. Moreover, the issue illustrates the striking convergence of knowledge about fundamental cognitive processes in human and nonhuman animals that has taken place during the past four decades. This convergence should facilitate the development of animal models of memory that can help to solve fundamental mysteries about the biology of memory. Ultimately, our ability to harness the full potential of such animal models rests on the rock of evolution, the certain knowledge that the brain, not excepting the human brain, evolved, just as did our other organs.

Progress toward understanding the biology of memory can be aided by first disentangling some basic distinctions. One classic distinction is that of learning and performance. Many factors may influence performance on a test, including some of the same factors that influence learning. A common solution to this classic problem is to vary the inputs at the early time point and employ a common test to assess learning. Performance factors are equated by employing the common test, so it is a safe bet to attribute differences in performance on the test to the learning that occurred in consequence to the different inputs. Another common distinction is between learning and memory. Learning is the process of acquiring new information, whereas memory involves retaining specific information over a delay. Interestingly, although this distinction appears to be graded, mechanistically, in central nervous systems, Stock and Zhang [1] provide evidence that learning and memory are molecularly separate processes in bacteria. This raises the intriguing question of the functional purpose underlying the evolution of graded learning and memory processes in nervous systems.

Memory comes in many forms. Early approaches to the study of memory focused on retention of information that was about fixed elements. For example, learning that element X is associated with element Y represents a classic form of learning. The classic approach continues to draw interest. Nevertheless, newer approaches focus on item-specific memory. For example, a series of items may be presented sequentially or simultaneously, followed by a delay and a subsequent assessment of memory for any item in the series.

A central question in the biology of memory is the extent to which underlying mechanisms are shared across species. The answer to this question appears to be that, to a remarkable degree, the basic molecular and cellular mechanisms of memory have been conserved during evolution. A fascinating example of such conservation is described by Song et al. [2] in their review on vernalization, a memory-like phenomenon observed in plants in which exposure of a plant to prolonged cold accelerates its flowering during its subsequent exposure to warm conditions, for example during the spring. Vernalization in the plant Arabidopsis thaliana is accomplished via modulation of the expression of the floral repressor gene FLOWERING LOCUS C (FLC). During a cold period, the expression of FLC is gradually downregulated via epigenetic repression, and this repression persists when the plants are returned to warmer temperatures. Interestingly, with longer periods of cold, the epigenetic repressive changes in the FLC chromatin progressively accumulate, so that, during a subsequent period of warmer temperatures, flowering in the plants is accelerated, and this acceleration is quantitatively proportional to the accumulation of the epigenetic changes. Song et al. [2] point out that the specific mechanisms that underlie vernalization in Arabidopsis have parallels in Drosophila and mammals, and speculate that accumulation of epigenetic memory may play a general role in memory. Given the increasing appreciation of the importance of epigenetic mechanisms in memory formation and maintenance in animals, this idea is likely to be correct.

Another striking example of conservation of memory mechanisms is the ubiquity of N-methyl-D-aspartate (NMDA) receptors in the animal kingdom. The major candidate for a synaptic mechanism of learning and memory in mammals is long-term potentiation (LTP), which is mediated by activation of postsynaptic NMDA receptors (see [5]). However, NMDA receptors are not unique to mammals; the nervous systems in animals ranging from nematode worms to slugs to flies to fish all possess NMDA receptors, and non-mammalian animals also exhibit NMDA receptor-dependent forms of learning and memory [6].

Whereas all animals appear to share a common cell biology of memory, the extent to which basic mnemonic processes that underlie memory are shared among widely diverse species is uncertain. The phenomenon of memory consolidation illustrates this point. Consolidation of memory in the mammalian brain occurs on two levels, the cellular/synaptic level and the systems level [7]. The mechanisms of cellular/synaptic level memory consolidation are fairly well understood; these include activation of various protein kinases or protein phosphatases, which in turn can trigger protein synthesis and gene transcription or repression. The molecular products of this protein synthesis and gene transcription/repression mediate the strengthening and growth, or the weakening and retraction, of synapses; the end result is the persistent modification of neural circuits in an animal’s nervous system that constitutes memory.

The mechanisms of cellular/synaptic level consolidation appear to be universal among animals; for example, activation of the transcription factor cyclic AMP response element binding protein (CREB) is a necessary step in the cell/systems-level consolidation of many forms of invertebrate and vertebrate memory [8]. As discussed by Preston and Eichenbaum [9], however, the consolidation of some memories in the mammalian brain involves, in addition, a time-dependent transfer of information from one brain region, the hippocampus, to another, the medial prefrontal cortex. The functional reason for this information transfer is unclear, as is whether the transfer is permanent, as proposed by some [7], or, whether instead, as Preston and Eichenbaum [9] argue, memories can reside permanently in both regions, thereby allowing the two memory representations to interact under some circumstances. The purpose of this post-learning interaction between the hippocampus and medial prefrontal cortex, according to Preston and Eichenbaum [9], is the formation of memory ‘schemas’, which give an animal the ability, for example, to resolve conflicts between new events and old memories.

Regardless, at present the evidence for systems level consolidation in invertebrate memory is sparse (but see [10]). Another potential disjunction between vertebrate and invertebrate mnemonic processes concerns the role of sleep in memory consolidation. As reviewed by Abel and colleagues [5], sleep is critical for the consolidation of many forms of memory in mammals. Strikingly, electrophysiological recordings from single ‘place cell’ neurons in the hippocampus of rats during a spatial learning experience and during non-REM sleep immediately after such learning have shown that the neurons exhibit similar patterns of firing during learning and sleep. This finding has led to the idea that learning-induced patterns of hippocampal activity are ‘replayed’ during non-REM sleep and that this hippocampal reactivation plays a role in memory consolidation. (Replay of learning-related neuronal activity during sleep has also been reported for vocal learning in songbirds [11].) Whether a similar process occurs in invertebrates is not known. Sleep-like behavior has been observed in invertebrates, particularly Caenorhabditis elegans and Drosophila. Moreover, sleep has recently been reported to be crucial for a form of one-day memory in the fly. However, the reactivation of specific patterns of learning-induced neural activity during sleep has not yet been documented in an invertebrate. Also, the evidence for the presence of a sleep state in some invertebrates that are unambiguously capable of learning, such as mollusks, is equivocal.

Besides its intrinsic intellectual fascination, the issue of how evolutionarily conserved are the neural processes of memory is interesting to neurobiologists for two other major reasons, one practical, the other ethical. Neurobiologists tend to be reductionistic in their approach to behavior and cognition. If it can be shown that a neurobiologically simple, experimentally tractable organism, such as C. elegans, exhibits the identical form of memory — say, habituation — as a monkey, most neurobiologists interested in that form of memory would probably choose to work on the simpler animal. (Simpler animals also tend to be cheaper, a not inconsequential advantage in these times of reduced extramural funding for research.) With respect to ethical considerations, it is difficult to justify taking the life of a monkey or a mouse if one can use a snail, for example, to study a given memory-related phenomenon.

Many types of higher order learning and memory can only be studied in mammals, however, and in some cases, perhaps, only in humans. Thus, Collett et al. [12] conclude that insects do not use cognitive maps, despite the impressive displays of spatial navigation by some insect species. By contrast, Templer and Hampton [13] review evidence that critical elements of episodic memory, the memory system that stores unique personal past experiences, are shared by humans and nonhumans such as rats and monkeys. The development of convincing animal models of episodic memory is valuable; from the perspective of an experimental neurobiologist, humans are perhaps the least attractive of all subjects, both because of the unsurpassed complexity of their brains and because of the relative crudeness of the experimental tools available for studying human brains. (Despite these significant scientific impediments, some of the most important intellectual advances in understanding memory (e.g., [14]) have come from studies of people.)

Development of valid animal models of memory is important because such models have significant potential for translational research to improve outcomes, for example, the impairments in memory that occur as we age and as a consequence of disease. A range of amnesic syndromes in humans include prominent deficits in episodic memory. People with Alzheimer’s disease (AD), for example, exhibit profound impairments in episodic memory. Eventually and inevitably, patients suffering from AD experience a profound loss of cognitive function, including the inability to recognize even close friends and family members. In addition to AD, episodic memory is also impaired in a range of disorders, including frontal lobe lesions, Huntington’s disease, mild cognitive impairment, normal aging, schizophrenia, and stroke. The societal impact of memory disorders is staggering. In addition to the enormous personal and emotional costs such disorders incur, they cost the US economy approximately $200 billion annually [15]. The financial and societal consequences of memory impairment disorders are expected to increase as the population of elderly increases. At present, there are approximately 5.4 million Americans with AD; an estimated 6.7 million will have the disease by 2025 and 11–16 million by 2050 [15]. A better understanding of mechanisms of memory and memory impairments may ultimately reduce both escalating health care costs and unnecessary suffering in AD. Notice that even small improvements in retention of cognitive function can have enormous impacts on wellbeing, social engagement, and productivity by decreasing healthcare and long-term care costs.

Most research using animal models of AD assesses only general aspects of learning and memory, and thus the translational relevance to episodic memory impairments in AD is uncertain. This is a significant and widespread problem because a variety of approaches to modeling AD have appeared promising at early stages of preclinical testing, only to fail in subsequent clinical trials [16]. For example, at least 20 compounds have provided preliminary evidence for benefits in AD preclinical studies and phase II clinical trials, yet failed to show consistent success in phase III clinical trials, which occurs in 40–50% of tested compounds. Recent examples include drug candidates that have failed for lack of efficacy at phase II (AZD-103, bapineuzumab) and at phase III (atorvastatin, phenserine, rosiglitiazone, tarenflurbil, tramiprostate) clinical trials. This problem is further compounded because unsuccessful preclinical and clinical trials are often not published. Importantly, our understanding of the molecular underpinnings of AD, for example, has greatly outpaced our ability to model the types of cognitive impairments observed clinically. The ability to translate successfully from animals to humans will be improved by development of approaches that include modeling of the specific memory impairments observed in clinical populations rather than general memory assessments (for example, spatial memory) that are not specifically impaired in AD.

In addition to impairments of memory due to molecular abnormalities — such as the amyloid plaques and neurofibrillary tangles of AD — and brain injury, people suffer from disorders, particularly posttraumatic stress disorder (PTSD) and drug addiction, that are caused by abnormal hyperactivation of memory-related processes. Two recent developments in our understanding of the mechanisms of memory maintenance hold particular promise for the treatment of PTSD and other disorders of overstimulated memory. The first, described in the primer of Alberini and LeDoux [17], is the recognition that reactivation of a consolidated memory by a stimulus that reminds the animal of the original learning experience (this is typically the conditioned stimulus in a classical conditioning paradigm) can trigger a new round of consolidation (‘reconsolidation’); reactivation-induced reconsolidation of a memory depends on many of the same processes, particularly protein synthesis, that are required for original consolidation of the memory. (However, the mechanisms of original consolidation and reconsolidation differ in certain respects.) The evidence indicates that when memories undergo successful reconsolidation, they become strengthened. Conversely, if reconsolidation is disrupted (through, for example, administering a protein synthesis inhibitor to an animal soon after a reminder stimulus), the memory is weakened or eliminated entirely.

Thus, consolidated memories are not unchanging as previously believed; rather, they are dynamic and potentially quite labile. The function of reconsolidation appears to be to provide the ability to respond flexibly to an ever-changing environment; reconsolidation permits an organism to update its memories, either strengthening or weakening them, without having to undergo re-exposure to the original learning situation. The recent discovery (or more correctly, re-discovery) of reconsolidation has led to an attempt to put reconsolidative processes to clinical use to treat PTSD. Two drugs that have been used on human patients in reconsolidation protocols in attempts to weaken traumatic memories are propranolol, a β-adrenergic receptor antagonist (noradrenaline has been implicated in memory reconsolidation in rats) and rapamycin (or sirolimus), an inhibitor of protein synthesis. Unfortunately, neither pharmacological intervention has proved successful, perhaps because highly traumatic events may have consequences in humans that are not mimicked in laboratory studies of rats and mice.

A second major advance in our knowledge of memory maintenance, not represented in the present issue, has been evidence that a constitutively active isoform of protein kinase C (PKC) known as PKMζ may play a critical role in maintaining memories. PKMζ mRNA is formed from alternative splicing of the gene for the atypical PKCζ; the PKMζ mRNA is then transported to dendrites, where it can be locally translated by learning-related synaptic stimulation, particularly stimulation that induces LTP [18]. The PKMζ protein lacks a regulatory domain and so its activity normally cannot be inhibited (hence its attraction as a memory maintenance molecule); however, pharmacological inhibitors of PKMζ are available. Many studies have now shown that inhibition of PKMζ appears to erase consolidated memories, as well as established LTP [18]. But not all forms of consolidated memories are susceptible to disruption by inhibition of PKMζ [19]. Furthermore, the specificity of the inhibitors that have been used to block the activity of PKMζ has recently been questioned (discussed in [19]). Finally, at present there is no way to ensure the precision of the memory-weakening actions of PKMζ; in principal, non-traumatic and traumatic memories would be erased indiscriminately by inhibiting PKMζ’s activity in the brain. These facts suggest that manipulation of the activity of PKMζ is unlikely to prove clinically useful in the near future.

The grand challenge to understand the biological mechanisms that support memory is unfolding during a golden age of neuroscience research. One prospect for the future is the goal of integrating a deep understanding of biological mechanisms with sophisticated models of human cognition. For example, there is growing evidence that specific aspects of human memory can be modeled in non-human animals, including such processes as episodic memory, declarative memory, and prospective memory (‘remembering to remember’). Combining these approaches with new insights about the biology of memory has the potential not only to illuminate some profound mysteries of the mind, but also to advance translational research that may ultimately foster the development of therapeutic approaches to severe human cognitive disorders [20]. Another reason for optimism is the rapid progress in experimental methodologies available for studying memory. For example, optogenetic tools now permit the targeted expression of calcium indicators, or light-gated ion channels, neurotransmitter receptors and ion pumps, in specific types of neurons; investigators can thereby optically monitor, or remotely manipulate, the activity of the neurons in intact animals while the animals are actually learning or recalling a learned experience (see for example [10]). These and other developments point to a bright future for research into how brains store and retrieve information about the past.