can be crudely explained in this way: A presynaptic
neuron creates, stores, and releases a chemical neurotransmitter
, which crosses a very small gap (the synapse
) and is taken up by the postsynaptic
neuron. While the axons and dendrites of neurons branch out in three dimensions, the act of neurotransmission may be thought of as two-dimensional -- that is, no branching happens between transmission and reception over a given synapse
during the period of transmission. Hence, if this were the only type of neurotransmission going on, discrete paths of all of the system's information could be understood by looking only at neurons and their connections.
As we all know, Nature is never quite so easy to grasp, and she has a great deal of subtle tricks up her sleeve. One trick that we've recently discovered is a way of overriding this connective idiom of brain function with one that is purely spatial, nitric oxide neurotransmission.
Some neurons, when inundated with enough calcium ions, activate an enzyme named NOS (nitric oxide synthase). This enzyme metabolizes the amino acid L-arginine to form, I bet you could have guessed, nitric oxide (NO) gas. Now, many gasses diffuse in the brain naturally, as they're tiny little non-polar molecules that can slip through membranes without even slowing down. Since the NOS enzymes aren't localized to any particular area, the newly generated NO gas produces a "cloud" that spreads through all of the surrounding neurons, and modifies their behavior. And what is it called when the release of a chemical by one neuron changes another's activity? Neurotransmission.
Reasons why this is important, in no particular order:
- Dimensionality: By moving in four dimensions (all directions for a given period of time), gaseous diffusion neurotransmission may affect cell assemblies and even whole networks that would ordinarily be inaccessible to the transmitting neuron. This is incredibly important, because it means the firing of a neuron within the cloud isn't controlled only by the dendritic input it receives from neurons it synapses with, but from neurons in possibly very distant parts of the network. To put it another way, the locality of the transmitting neuron simply skyrockets, since it has an effective fanout of billions of cells instead of thousands.
- Long term potentiation: LTP happens when a group of neurons that fires together becomes sensitized, so they fire together more easily. Not only are the postsynaptic neurons sensitized with NMDA activation, but presynaptic neurons also release more neurotransmitter after potentiation, leading to an altogether more sensitive synapse. This has been something of a mystery to science, since neurotransmission has always been thought to be two-dimensional. Because of NO diffusion, this may not be the case.
Here's how it works, to the best of my understanding. Upon repeated firing, the postsynaptic neuron's NMDA channels pop open, and they let a bunch of calcium ions into the neuron. These calcium ions sensitize that neuron to further incoming signals, which has been known for a while; they also kick NOS into action, so the postsynaptic neuron begins to diffuse NO into the surrounding area. Theoretically, the NO has a special effect on the presynaptic neuron because it recently fired, an effect which causes it to fire with greater strength in the future. Thus, both parts of the synapse are strengthened, so long term potentiation is complete.
- Additional temporal sensitivity: Regular neurotransmission happens in a relatively short period of time, a matter of milliseconds between when the vesicle is released and when the neurotransmitter is broken down postsynaptically. NO diffusion happens at a much slower rate, and can continue for up to a second (!) after a neuron's NOS enzymes have stopped producing NO. It's possible that natural networks use this lengthy communication for medium-term storage of information, since it lasts longer than a synaptic firing but shorter than semi-permanent changes of LTP or other potentiation. Research is incomplete and ongoing, so don't take this as gospel.
- Strokes, Parkinson's and Alzheimer's: All of these damage neurons, though each act in different areas. Until recently the mass neuron death that accompanies each of these wasn't clearly understood, since there didn't seem to be any reason for it to be as widespread as it is. With a gas diffusion model of damage, this damage makes more sense.
One current theory of the damage is caused is this: First, trauma occurs to neurons for various reasons, either lack of oxygen from a stroke or aging processes for the diseases. Because of this, they flood the neurotransmitter glutamate into the area surrounding them. NMDA receptors exposed to this flood open up, and fill their own neurons with calcium ions, which in turn heavily activate NOS enzymes. Finally, huge amounts of NO diffuse through the surrounding area. In the usual amounts NO is harmless, but in large amounts it becomes quite toxic -- concentrated NO is used by microphages to kill invading bacteria. So, all of this NO that's suddenly floating around kills more neurons, causing them to release additional glutamate into the system. You get the picture, I'm sure, and it's not a pretty one.
This theory has been partially proven by studies in which rats were denied NOS and exposed to MPTP. MPTP causes rapid onset of Parkinsonian symptoms, but rats that either had NOS inhibited or their NOS gene removed were much less affected by it. Drug companies are researching NOS inhibitors for this reason right now.
Artificial Intelligence: Finally, this is the one I'm sort of interested in, being something of a CogSci geek. Artificial neural networks (ANN's) have traditionally used two-dimensional synapses, in keeping with the neurological model. GasNets, a new form of ANN, have been developed to use all of these new gaseous diffusion neurotransmission findings, and early experimental results look promising. Here are a few ways a GasNet can be useful to an ANN:
- As a low-pass filter. That is, long-term diffusion can make noise less of a factor to the system, by only transmitting those signals which cause the neuron to be activated for a long enough period of time to release a large amount of gas. In other words, high frequency inputs can't stimulate the neuron to release enough NO to reach outlying areas, whereas low frequency (and thus high period) inputs can.
- For rhythmic output with or without input. If a diffusive neuron is in the same cluster as non-diffusive ones and is being stimulated at regulated neurotransmission speeds, it will ramp up or down gas release based on the overall rhythm of the cluster. Since this affects surrounding neurons, a complex rhythm (probably best described by an evil differential equation) may arise.
- Causing processes that build up effect over time. Since the simulated gas can stay present for much longer than its producer neuron stays active, it can build up concentration in an area. Given the right wiring, this slowly changing concentration can cause all kinds of interesting emergent effects.
Chings and respect out to anyone who feels competent enough to fill GasNet; I certainly do not yet.