Recently, a paper from the Tonegawa lab showed how memory can be formed in a single neuron. The paper was published in the journal Neuron and garnered a lot of interest in the general community. Although I have not read the paper, it was recently presented as part of a seminar series by Nathan, and I am going to summarize what I learned from his talk (and, more importantly, what is yet to be learned).
The mechanism that most scientists use to describe the mechanism of memory formation is called Long Term Potentiation (LTP). That mass of letters describes a phenomenon whereby there is a permanent change in the "attention" paid by one neuron to communications from others. Neurons communicate using electrical signals which are transmitted between them at synapses. When a synapse undergoes LTP, there is a greater change in the voltage of a postsynaptic (i.e. listening) neuron when a presynaptic (i.e. talking) neuron fires action potential (i.e. communicates). This change is at least long-lasting, if not permanent, and therefore scientists believe this to be the basis of memory, which is also long-lasting, if not permanent.
LTP, however, has problems describing memory. The main issue is that LTP only occurs at one synapse whereas memory requires association. For instance, for me to learn that depositing a quarter in the vending machine causes candy to fall out, there needs to be an association between the neuron that communicates "I deposited a quarter," and another that communicates, "Candy is tasty." In addition, these signals have to be delayed, since there is a time gap between depositing a quarter and enjoying a delicious sugary snack. How does a neuron accomplish this ?
The authors found that there were two phases of LTP, called early and late LTP. Early LTP is not long-lasting, and therefore is either not part of memory, or is part of short-term memory. Late LTP is probably what we think about as memory, in that it is long lasting, if not permanent. The authors found that if they induce late LTP at one synapse and early LTP at nearby synapses, the early LTP is converted into late LTP (i.e. it becomes long lasting). This conversion depends on many things, including how far away the synapses are, how quickly the early LTP follows (or even comes before) the late LTP, and if the synapses are on the same or different dendrites.
To explain my earlier example, when I deposit a quarter into the vending machine, the neuron that communicates that fires action potentials. The postsyanptic neuron records this in its "short-term" memory. When I recieve the candy a few minutes later, another neuron comunicates that "candy is delicious." Since the second communication is critically important to my brain, it is converted into "long-term" memory, as is the "short-term" record of depositing a quarter in the machine. If I had taken the candy and eaten it an hour or two later, it would not have been converted, because the deposit of a quarter does not mean a candy is about to be eaten.
The paper was remarkable because, in addition to explaining this mechanism, the authors also provide testable predictions. For instance, according to their model, "coin deposit" and "candy" neurons should form synapses on the same dendrite and near one another. More generally, stimuli that occur together must be represented by synapses that are near one another. Also, since one can pair a candy with almost anything and the paired item must be remembered as a candy predictor, "candy" encoding neurons (whose identities are known) must be represented on all dendrites at regular intervals.
In conclusion, I think this paper is a landmark paper in understanding how molecules and cells of the brain change permanently to form the basis of memory. As was discussed at the presentation, there are scientific holes that must be plugged, but I feel like the underlying theory is appealing because it is a simple mechanism that predicts a lot of properties of memory.
Cheers,
Rohit.
References:
The dendritic branch is the preferred integrative unit for protein synthesis-dependent LTP. Arvind Govindarajan, Inbal Israely, Shu-Ying Huang, and Susumu Tonegawa
Neuron, 2011
