Neurophysiology: Dust Clearing on the Long-Term Potentiation Debate

DETECTING COINCIDENCE

Much current research is focused on NMDA-dependent LTP in the CA1 region of the hippocampus. Excitatory synaptic transmission depends on ionotropic receptors, primarily AMPA receptors (AMPARs) and NMDA receptors (NMDARs), which are triggered by transmitter release. Glutamate binds to AMPA, opening up an ion channel and producing postsynaptic depolarization.

When the cell is at rest, the pores of NMDARs are blocked by positively charged magnesium ions. For these voltage-dependent channels to open, two things have to happen almost simultaneously: glutamate release from the synapse, and the depolarization of the cell by other inputs (often back-propagating action potentials from their own somas). Thus, NMDARs work as coincidence detectors, at the conjunction of two different stimuli (postsynaptic depolarization and presynaptic glutamate release). In short, the associative process starts at the most basic biochemical level. Roger Nicoll of the University of California, San Francisco, calls NMDAR "a Hebbian molecule," referring to Donald O. Hebb, the McGill University psychologist whose 1949 book, The Organization of Behavior, famously described synaptic weighting.

Once the magnesium block is released, calcium ions move into the cell and activate kinases, most prominently calcium/calmodulin-dependent protein kinase II (CaMKII), leading through phosphorylation to AMPAR insertion into the postsynaptic membrane. Nicoll warns, "Most people are on board with that scenario up to activation of CaMKII. Then it's a total murky mess up to AMPAR insertion." The focus of much current research, including his own, is filling in that gap.

Malenka, Nicoll, and working separately, Roberto Malinow of Cold Spring Harbor Laboratory, reinforced the idea of NMDARs as an associative mechanism when they discovered so-called silent synapses, which have NMDARs but no AMPARs. Normally, the glutamate from cell A should make cell B depolarize, but here cell B (lacking AMPARs) does not. It is only by repeated activation of A and B firing together (for example, when B is activated by another pathway), that an ongoing association is made between these cells. After the repeated synchronous firing of A and B, AMPARs are finally recruited into B by Ca²⁺ influx through its NMDARs.

The idea is most easily understood in a developmental example: A baby's brain does not yet know how to "wire up" the world. It needs to wait for repetitive associations of stimuli. A silent synapse will eventually, through the process of LTP, get AMPAR inserted. But the elegance of that solution belies how the original question split the field in two: those who saw LTP as either presynaptic or postsynaptic.

A CONFOUNDING SWITCH

Despite all the familiar illustrations of synaptic clefts filled with neurotransmitter, it is actually released infrequently. "Synapses release transmitter in a probabilistic manner," says Nicoll, "and the probability is quite low." Typically only one-fourth of activated presynaptic termini will release transmitters.

At the time of the pre- and post- wars, the canonical idea was that if the synaptic failure rate went down, then surely something had changed on the presynaptic side to increase the probability of transmitter release. Then experiments uncovered synapses that confounded the field: synapses that have NMDARs but no AMPARs. Yet when LTP was induced, "You quickly got a full component of AMPA receptor current, without the NMDA current changing at all," says Nicoll. That meant a decrease in failure rate, but one that had to be postsynaptic in origin. If it were presynaptic, then both receptors would have been affected.

Theories were advanced, many of them focused on the idea that somehow something was going back to the presynaptic side, " [These theories were] all incredibly 'Rube Goldberg,"' says Nicoll, "The way everyone was tricked was that no one had considered that you could just in an all-or-none manner turn on synapses that were totally silent."

© 1999 AAAS

Glutamate release spurs Na⁺ inflow through the AMPAR. Only when the Mg²⁺ block is removed from the NMDAR during depolarization, can Na⁺ and Ca⁺ flow through the receptor. The rise in Ca²⁺ is a critical trigger for LTP. (From R.C. Malenka, R.A. Nicoll, Science, 285:1870–4, 1999.)

And so LTP was definitely found to involve a postsynaptic change. Well, maybe. "There's an emerging consensus that the Roger Nicoll view is correct. There's a postsynaptic induction and postsynaptic expression under many forms of LTP in the hippocampus in CA1. But with some forms produced at certain particular frequencies, there also is a presynaptic contribution," says Kandel, "In CA3 [in the hippocampus], there's consensus by almost everyone that it's almost certainly presynaptic and it's PKA- [protein kinase A] mediated." In the fine distinctions of the field, however, "almost" is an important operator.

LEARNING PROTOCOLS

NMDARs have been observed in invertebrates for more than 10 years; several labs have now reported NMDAR in everything from hydra to squid. ³⁴ Recently, Tim Tully found that NMDARs mediate learning in Drosophila. ⁵ These discoveries lend themselves to speculation about the ultimate role of NMDARs, says David Glanzman of UCLA, "Maybe NMDA receptors evolved as the mechanisms of associative learning ... or it could be that they evolved for something else and they were 'kidnapped' for learning. An animal has a mechanism whose essence is that it's associative, and now it can use it to learn."

Whatever their ultimate evolutionary origin, NMDARs appear to be an essential component of the kind of associative learning familiar from Pavlovian classical conditioning. In classical conditioning, an unconditioned stimulus (US), to which an animal would naturally react (e.g., a shock) is paired with a conditioned stimulus (CS) that normally would not evoke a reaction (e.g., a tone, or in the case of the sea slug Aplysia, a light touch).

Several experiments have shown that monoamines such as serotonin and dopamine might possibly cause AMPAR insertion, even without the activation of NMDARs. ⁶ Monoamines are also essential to a form of nonassociative learning, called sensitization. In sensitization, no pairing of stimuli occurs; the subject receives just one negative stimulus several times over, sensitizing it to even mild forms of the same stimulus. For example, an Aplysia that has had its tail shocked repeatedly will become sensitive to mild touch.

Robert Froemke, a postdoctoral researcher at UCSF with Christoph Schreiner and Michael Merzenich, offers an interpretation that intrigues Nicoll as well: "I would argue that why NMDA receptors are so critical is that they are association detectors. Dumping on serotonin or dopamine is not associative; there's no A and B firing, there's only A. With monoamines, A is telling B how to learn. With NMDAR, A and B are agreeing to learn something."

PAVLOV COMES OF AGE

© 1999 AAAS

In this model for LTP expression, Ca²⁺ within the dendritic spine binds to calmodulin (CaM) to activate CaMKII, which undergoes autophosphorylation to maintain activity. CaMKII phosphorylates AMPARs already present in the synaptic plasma membrane, increasing their single channel conductance. CaMKII is also postulated to influence subsynaptic localization of AMPARs, in effect, helping to deliver them to the synapse, but where these AMPARs come from is unclear. (From R.C. Malenka, R.A. Nicoll, Science, 285:1870–4, 1999.)

One of the important components of classical conditioning is timing: If the CS stimulus is presented first, the sense of association between the two stimuli is strengthened. If the US comes first, the sense of association is weakened. Hebb applied this idea to neural circuits to create the famous learning rule that Nicoll refers to above: If cell A leads to the firing of cell B in a consistent fashion, the synaptic connection is strengthened.

The classic rule assumes an oversimplified model, however, that provides no information about where the cells are synapsing, or even whether location makes any difference. Now a new generation of experimentalists with computational backgrounds has begun to look at the where as well as the when of synaptic firing, in a new specialty called spike timing-dependent plasticity (STDP). ⁷ "One of the reasons I believe in STDP is when I look at the behavioral electrophysiology: How are spikes produced in the animals running on tracks?" says Matt Wilson of the Massachusetts Institute of Technology. "What you see is that the timing of the spikes is produced in a way that preserves a lot of relative timing information,"

In a recent paper, ⁸ Froemke and his graduate advisor, Yang Dan of UC-Berkeley, looked at the role of NMDARs and STDP in a distal and proximal synapse in pyramidal neurons of rat visual cortex. "Depending on where the synapse is in the dendrites, the timing requirements are different. What that means is that different synapses have different rules for learning," Froemke says.

Courtesy of Genny Anderson

The sea slug, Aplysia, has been useful in deciphering the molecular mechanisms of conditioning.

In this scenario, the result of the back-propagating action potential is that the NMDAR sees a short, tight spike at the proximal dendrite. The distal dendrite has a much broader spike that goes on for a much longer amount of time. The difference means a much longer time window (on a millisecond scale) for coincidence detection the farther a synapse is from the soma.

"People thought ... who came first was key, and then the timing was key, and the combination of those two explained everything. What this paper says is there's another layer of complexity to all that, that it's not just order and time. It's also about location," says Tom O'Dell of UCLA. "You can't just say a synapse is a synapse in terms of plasticity; the rules that govern whether or not you're going to get a change vary along the length of the dendrite, which is amazing. It increases the computational power of the cell, because the synapses on the dendrites are all so different."

In a provocative interpretation, Froemke speculates that the results could also suggest that "NMDA receptors may be much more active, much more highly tuned, [and] extremely sensitive to small perturbations in synaptic transmission than previously thought." But he admits this is far removed from anything they showed in their paper. Froemke works in visual cortex, where NMDARs can naturally be much more dynamic, says Nicoll, illustrating once again the challenge of consensus in a field in which everyone works on the same process but in different areas. Indeed, some researchers consider STDP to be possibly the result of a change in induction protocols.

Kandel says that if one changes the frequencies of stimulation during a plasticity protocol, or the pattern of stimulation during a learning protocol, a set of different mechanisms will likely be recruited to a different form of learning. Kandel says: "We have gone through a complicated period. We now understand why some of these differences existed between groups, because they were using somewhat different protocols, and now we can begin to see which of these forms of LTP map onto different forms of learning." As per usual, not everyone agrees with this assessment of protocol disparity.

Perhaps the only thing everyone does agree on is that LTP remains enormously complicated. During the pre- versus post- debate, a theory called retrograde transmission suggested that nitric oxide might be a neurotransmitter involved in LTP. While Kandel says nitric oxide is still a "candidate molecule" for LTP-induced presynaptic transmission, Malenka and others say the theory's day has passed. When asked what he thought of the idea, Louis Ignarro, winner of the 1998 Nobel Prize for his joint discovery of nitric oxide as a signaling molecule in the cardiovascular system, deferred to the neuroscientists: "It's hard to work on the brain. I would never work in that area."