There is no general agreement as to the meaning of long-term potentiation, but this cannot be resolved by using it to explain additional phenomena. Increased attention to recently experienced stimuli is a form of learning known to neuropsychologists as repetition priming. As more is learned about the neurochemistry of synaptic change, the term LTP will wither.

In the developing and adult cortex, just as in the adult hippocampus, LTP is unable to account for a variety of types of functional plasticity.

Although it is not their fault, Shors & Matzel's attempt to review the LTP and learning hypothesis suffers from there being no clear published statement of the idea. Their summary of relevant evidence is not without error, however, and it oversimplifies fundamental issues relating to NMDA receptor function. Their attentional hypothesis is intriguing but requires a better systems-level understanding of how attention contributes to cognitive function.

Shors & Matzel's target article is a thought-provoking attempt to reconceptualise long-term potentiation as an attentional or arousal mechanism rather than a memory storage mechanism. This is incompatible with the facts of the neurobiology of attention and of the behavioural neurophysiological properties of hippocampal neurons.

Preoccupation with LTP as a putative memory mechanism may have retarded the consideration of pathological modulation of synaptic plasticity in clinical disorders where memory dysfunction is not a primary feature. Encouraged by Shors & Matzel's review, we consider the relationship between stress, synaptic plasticity, and depressive disorder.

Controversy surrounds several experiments that have addressed whether selective synaptic strengthening occurs during learning. To date, the evidence suggests that widespread alterations in synaptic strength, through either kindling or electroconvulsive shock, can disrupt this hypothetical process. The lack of evidence for selective modification of learning through LTP stimulation, however, provides difficulties for both the prevailing hypothesis and the hypothesis advanced by Shors & Matzel. Subsequent experiments may indicate a role for LTP in both learning and arousal.

Recent evidence from our lab indicates that LTP shares an important property with memory consolidation: it is consolidated by natural reinforcement. Nevertheless, the hypothesis, that LTP-like mechanisms or other forms of enhanced synaptic efficacy are basic elements in learning is not unequivocally supported. Skepticism aside, LTP is an accessible experimental model that is optimally equipped for the investigation of the cellular and molecular machinery involved in synaptic weight changes.

Shors & Matzel's conclusion that LTP is not related to learning is similar to one we reached several years ago. We discuss some methodological advances that have relevance to the issue and applaud the authors for challenging existing dogma.

Shors & Matzel propose that hippocampal LTP increases the effective salience of discrete external stimuli and thereby facilitates the induction of memories at distant places. In line with this suggestion, a neural network model of associative learning and hippocampal function assumes that LTP increases hippocampal error signals to the cortex, thereby facilitating stimulus configuration in association cortex. Computer simulations show that under these assumptions the model correctly describes the effect of LTP induction and blockade in classical discriminations and place learning.

Shors & Matzel set up a straw man, that LTP is a memory storage mechanism, and knock him down without due consideration of the important relations among different levels of organization and analysis regarding LTP, learning, and memory. Assessing these relationships requires analysis and hypotheses linking specific brain regions, neural circuits, plasticity mechanisms, and task demands. The issue addressed by the authors is important, but their analysis is off target.

Shors & Matzel somewhat lightly dismiss the evidence that a process like LTP may underlie the learning-induced increase in neuronal activity in the hippocampus in eyeblink conditioning. I provide some 12 lines of evidence supporting this hypothesis and the further hypothesis that this learning-induced LTP-like hippocampal plasticity can play a critical role in certain aspects of learned behavior.

A fundamental assumption in Shors & Matzel's target article is that brain activity can be related to the traditional categories of mentalistic psychology. This has led them to make numerous further assumptions that are contradicted by the available evidence.

We agree with Shors & Matzel's general hypothesis that the proposed link between NMDA-dependent LTP and memory is weak. They suggest that NMDA-dependent LTP is important to arousal or attentional processes which influence learning in an anterograde manner. However, current evidence is also consistent with the view that NMDA receptors modulate memory consolidation retroactively, as occurs in several other receptor classes.

The neurophysiological phenomenon of LTP (long term potentiation) is considered by many to represent an adequate mechanism for acquiring or storing memories in the mammalian brain. In our target article, we reviewed the various arguments put forth in support of the LTP/memory hypothesis. We concluded that these arguments were inconsistent with the purported data base and proposed an alternative interpretation that we suggested was at least as compatible with the available data as the more widely held view. In doing so, we attempted to illustrate that the inadequacy of present experimental designs did not permit us to distinguish between equally viable hypotheses. In the four years since we wrote the first draft of our target article, hundreds of additional studies on LTP have been published and their results have been incorporated into current theories about memory. A diverse group of commentators responded to our target article with their own theories of how memories might be stored in the brain, some of which rely on LTP. Some commentators doubted whether memories can be stored through modifications of synaptic strength. Some assert that it will never be possible to understand the neural mechanisms of memory; still others remain hopeful that we will accomplish some semblance of a resolution, provided we appreciate LTP's role in a subset of seemingly amorphous memory systems. In summary, although it is commonly written that “LTP is a memory storage device,” the divergence of views among the commentators suggests, at least as strongly as our target article, that such conviction is unwarranted and fails to acknowledge both the lack of consensus regarding the role of LTP in memory and the complexity of the phenomenon of memory itself.

It is worthwhile to search for forms of coding, processing, and learning common to various cortical regions and cognitive functions. Local cortical processors may coordinate their activity by maximizing the transmission of information coherently related to the context in which it occurs, thus forming synchronized population codes. This coordination involves contextual field (CF) connections that link processors within and between cortical regions. The effects of CF connections are distinguished from those mediating receptive field (RF) input; it is shown how CFs can guide both learning and processing without becoming confused with the transmission of RF information. Simulations explore the capabilities of networks built from local processors with both RF and CF connections. Physiological evidence for synchronization, CFs, and plasticity of the RF and CF connections is described. Coordination via CFs is related to perceptual grouping, the effects of context on contrast sensitivity, amblyopia, implicit influences of color in achromotopsia, object and word perception, and the discovery of distal environmental variables and their interactions through self-organization. Cortical computation could thus involve the flexible evaluation of relations between input signals by locally specialized but adaptive processors whose activity is dynamically associated and coordinated within and between regions through specialized contextual connections.

The strong coupling of binding to cross-correlations is methodologically problematic. A completely unstructured network of neurons can produce cross-correlations very similar to the measured ones, and yet they have little dynamic effect.

Phillips & Singer (P&S) extend ideas derived from the observation eight years ago that the coherence (synchronization) of cortical oscillations can be modulated by the structure of visual stimuli. As described in the target article, a large part of the continued interest in this finding is related to independent theoretical work suggesting that synchronized cell firing could help solve the problem of binding together within cortex neuronal activity associated with different attributes of visual stimuli. The authors present an abstract “proof of concept” model describing how their cortical processing scheme could work, but our biologically realistic models of cortical relationships suggest that the proposal is biologically implausible. Our realistic models lead to a very different interpretation of the significance of cortical oscillations.

Binding needs to be task dependent, and cannot usefully be driven by properties of the stimulus alone. However, task dependent binding can only take place after the patterns in a stimulus have been identified. Thus pattern recognition needs to be done prior to binding. Synchronisation may be a consequence of pattern recognition and can be used to localise the pattern and tag its attributes at different levels of information processing.

Our work supports synchronization for binding within Phillips & Singer's “contextual field” (CF) as well as the type of its lateral interaction they propose. Both firmly agree with our “association field” (AF) and its modulatory influences (Eckhorn et al. 1990). However, the CF connections seem to produce anticorrelation among assemblies representing unrelated structures, whereas experimental evidence indicates decoupling. Finally, it is unclear how the cortex can have access to the logistic function used in the “coherent infomax” approach.

Contextual signals might supervise the discovery of coherently varying information between cortical modules computing different functions of their receptive field input. This hypothesis is explored in two sets of computational experiments, one studying the effects on learning of long-range unidirectional contextual signals mediated by intervening processors, and the other showing contextually supervised discovery of a high-order variable in a multilayer network.