What's the biochemical basis for our learning and storing the names, events, sights, and sounds that stay with us for a lifetime? Can we, in fact, reduce and explain these bits of nostalgia in terms of the inner workings of cellular and molecular mechanisms, phenomena in the brain at the synapses between neurons?
Seth Grant and Richard Morris
"LTP has never been observed to occur during learning and memory, period."
Scientists who still have faith in LTP, including Bliss and Richard Morris, a professor of neuroscience at University of Edinburgh and coauthor of one of the first studies finding evidence in favor of the mechanism,3 point out three major characteristics that make LTP a good candidate building block of information-storing behavioral neural networks: input specificity, associativity, and persistence. LTP has specificity because only the active pathway, and not other pathways converging on the same cell, is potentiated. It has associativity because one active set of inputs can interact with a second set if both are activated simultaneously--a characteristic necessary for explaining associative learning such as classical conditioning.
According to Bliss, because LTP's got the right properties and because it's been found in the brain at identified synapses, most people believe that it's likely the neural basis of at least some forms of memory. "The problem has been trying to prove that assertion," says Bliss. "It has turned out to be not at all easy to really pin down." Scientists have had a tough time relating the neural phenomenon to the observed behavior.
Investigators test the validity of LTP as a memory substrate primarily by blocking it using drugs, genetic manipulations, or antisense, or by saturating the mechanism for inducing LTP until no further LTP can occur--and then checking to see if any further learning is possible. If LTP does indeed have a stake in long-term memory, then when LTP has been blocked or induced to the point of saturation, learning should theoretically all but cease.
The latest research criticizing LTP, reported in a December 1998 Proceedings of the National Academy of Sciences paper from investigators at the NIH's Laboratory of Adaptive Systems, uses antisense to block the phenomenon and suggests that the mechanism is not necessary for long-term memory.5 Antisense microinjected into the rat hippocampus blocked specific presynaptic potassium channel messenger RNA, "knocked down" the expression of the protein, and eliminated LTP--yet the rats still retained spatial memory, an ability tested for, as is the norm, by noting how well the affected animals navigated through a water maze.
He contends that his lab's recent PNAS paper--in conjunction with one his team published in April 1997, which, using antisense to block specific postsynaptic channels, suggested that LTP is not sufficient for long-term memory7--pokes serious holes in the LTP hypothesis.
"In a sense he's right," says Morris. "LTP has not been seen across a large number of neurons following an individual learning experience. But I personally take the view that looking for LTP after a learning experience is an incredibly difficult experiment to do because you'd be searching for a needle in a haystack ... so I'm less worried about that objection." Moreover, says Morris, researchers can't guarantee that blocking drugs or antisense oligonucleotides will infuse through 100 percent of the targeted tissue. Therefore, subsequent learning tests may or may not recruit the neurons that have been blocked and results could be skewed.
Bliss and Grant cite several potential problems with Alkon's study. Bliss notes that Alkon's lab only looked at the CA1 region of the hippocampus, that his lab's stimulus protocol is only one of several used by investigators (says Bliss, a different protocol might "rescue" LTP), and that the experiments need to be done in the living animal rather than in slices. Grant suggests that it may be necessary to test LTP using a range of frequencies, and that antisense may not cover the entire relevant region of the brain. Both Morris and Grant suggest that a paradigm other than the water maze, which Morris himself made famous, might be warranted in order to conclusively validate or invalidate LTP.
But Alkon maintains that these objections hold little water: He claims that only the CA1 region, an area of the hippocampus, is relevant since only that region has been strongly implicated in rat maze learning and memory based on numerous lesion experiments, and electrophysiological recording studies. And, he says, his lab protocols and targets are no different from those that investigators have been using and working on for decades; all investigators look for LTP slices of subject animals after disrupting neuron function and testing learning in live specimens. According to Alkon, if his methods were inadequate and his targets of investigation somehow incomplete or insufficient, then so too are those of all scientists investigating LTP.
"If you now wanted to study LTP as a memory model," says Alkon, "the likelihood is that you'd have to find another area [other than CA1]. In other words, you'd have to start all over. You'd have to [reestablish] all of the properties, and then all of the lesion work, and all of the relationship work."
Says Matzel, "The recent paper by Alkon ... is another good example of the discordance between LTP the mechanism, and behavior indicative of memory." Matzel and Alkon note that there are dozens of such papers.
Others are more optimistic about recent findings. Morris, in a September 1998 Science article, claims to have achieved learning impairment in rats after saturating LTP,8 a strategy that had previously been criticized as unreliable and inconclusive.
"I had a couple of objections," says Bliss, who wrote an analysis that accompanied the Science article, "but overall it's clearly a paper that supports the LTP hypothesis."
Matzel remains unconvinced by such studies, saying that it's no surprise that disrupting normal synaptic transmission disrupts memory. According to Matzel, most manipulations directed at LTP affect normal transmission or affect second messenger cascades efferent to the actual transmission. "You screw around with transmission cascades, you screw around with memory," he remarks. "I don't think these manipulations provide a specific insight into the actual memory mechanism."
Other findings suggest that focusing solely on a correlation between LTP and learning may be oversimplifying the learning mechanism. According to Seth Grant, director of the Centre for Neuroscience at the University of Edinburgh and a colleague of Morris, many investigators contend that long-term depression (LTD) may be part of the learning and memory substrate. Neural network modelers, says Grant, have shown that networks can actually store more information if they have synapses capable of both potentiation and depression, a condition known as a bidirectional modification.
In research reported in Nature last December,9 Grant, Morris, and colleagues hypothesize that they were able to disturb this bidirectional property by knocking out a protein called postsynaptic density-95 (PSD-95), which is thought to help regulate the signaling of the NMDA receptor to which it binds. The result of the knockout: enhanced LTP.
Alkon, noting that the Nature paper reported learning impairment despite LTP enhancement, and that currently there's no evidence to support the involvement of LTD, argues that, in fact, rather than introducing evidence in support of a bidirectional property, University of Edinburgh investigators have actually stretched their hypothesis to fit results that go toward invalidating LTP as the neural substrate of learning. Morris and Grant contend, however, that the implications of the PSD-95 knockout need to be explored further; conclusions are premature. Grant expects that genetics holds the key: He's working toward understanding the mechanisms of synaptic function by genetically engineering animals to have specific synapse characteristics and then, based on behavior, deciphering how those synapses are involved in learning.
Right now, alternatives to LTP, aside from LTD, are few. "It is the model for memory, and there's nothing that comes close, with the exception of long-term depression," says Bliss.
Matzel and Alkon disagree. "I hate to say it," says Matzel, "but the memory mechanism may be untenable given our present perspective." Matzel prefers to examine modulatory factors such as intracellular calcium, which is thought to play a role in memory induction. Elucidating the memory mechanism itself, he suggests, is much farther out of reach than most investigators think.
But LTP skeptics may be battling more than just what they believe to be irrelevant results. Even the most definitive of objections must go up against the momentum of a field in which LTP has been firmly entrenched for decades. That widespread perception, says Matzel, hinders the search for other potential mechanisms, and limits what people are willing to consider with regard to the role of LTP. Many LTP doubters welcome, for instance, the idea of LTP as a modulator of memory, rather than a memory storage device per se. But that's a tough sell for some since the actual memory mechanism itself is, as Matzel puts it, "the big prize."
"Both the strength and the trouble with the 'LTP equals learning' hypothesis is that, for too long, it has been the only show in town," maintains Morris. "We need other ideas ... by having them, we may be able to design better experiments that compare two rival hypotheses."