Action Potential: Impulse of Neural Networks

The birth of voltage-gated ion channels conceived one of the most important phenomena in the evolution of intelligence and intelligence. That is "Action Potential". As I mentioned already in the previous post, the voltage-gated ion channel is a cluster of proteins embedded inside the cell membrane. It is capable of assuming more than one conformation and some of them can create a channel through the membrane for specific types of ions. Its conformation is influenced by membrane potential. Then, what is membrane potential?


Membrane Potential

All cells in the animal are electrically polarized. This is called the membrane potential, which occurs through a voltage difference between the inside and the outside of the cell membrane. This electrical polarization is due to interactions between ion pumps and ion channels in the cell membrane. In neurons, these ion channels are very diverse and distributed throughout the cell. That is why some parts of the neurons are excitable to trigger action potentials and others are not. The most excitable area is the axon attached to the cell body called the "axon hillock" of the neuron.

There are two important levels of membrane potential: resting potential and threshold potential. Resting potential is a membrane potential that is maintained when there is no interference to the cell, and threshold potential is a higher level that can cause some changes. In the case of a typical neuron, the resting potential is around -70 millivolts (mV) and the threshold potential is around -55 mV when measured in the axon hillock. Synaptic inputs in neurons can depolarize the membrane potential of the membrane or cause hyperpolarization. The subject of this post, action potentials, occurs when a sufficient number of depolarizations accumulate and the membrane potential exceeds the threshold. Once an action potential occurs, the membrane potential rises sharply and then falls sharply again, sometimes even lower than the resting potential. Many neurons produce action potentials of about 10 to 100 times per second. Of course, there are neurons that occur once every few minutes or even less.


Shape of a typical action potential from Wikipedia.org


Positive Feedback and Backpropagation

Membrane potentials control the state of ion channels, and ion channels affect membrane potential. Therefore, in some cases, the rise of the membrane potential causes the ion channels to open, which leads to a higher membrane potential. This feedback is called positive feedback. An action potential occurs when a positive feedback cycle occurs explosively. There are several types of ion channels involved in positive feedback causing action potentials. Voltage-gated sodium channels are involved in the fast action potentials involved in nerve conduction, and voltage-gated calcium channels are involved in slower action potentials that can be observed in muscle cells or in particular types of neurons.

Among these, voltage-gated sodium channels that cause fast nerve conduction have been elucidated by Alan Hodgkin and Andrew Huxley in detail, which is a biophysics study that reveals the secret of action potential, which won the Nobel Prize for recognition. Their Hodgkin-Huxley model is very important for neurophysiology, which will be discussed in more detail later in another post.

Although action potentials occur locally in areas with excitable membranes, the resulting currents can lead to action potentials' propagation like domino. When the action potentials propagated to axon terminals, depolarized axon terminals release the neurotransmitter to the synaptic cleft. Now, really important phenomenon was observed. Neocortex studies have found that the backpropagation of action potentials in dendrites of pyramidal neurons. This phenomenon is widespread throughout the neocortex, which is presumed to make the plasticity of our brain. This process is a spike-timing dependent phenomenon different from the backpropagation used in deep learning. But, it may inspire the scientists like Geoff Hinton who introduced backpropagation in deep neural networks. Before the discoveries of these neurophysiological researches, there were very first neurophysiologists who invented "artificial neuron", McCulloch and Pitts. I will post their story in next post.

Here is the video-clip for explaining action potential. Although its mechanism is quite different from artificial neural networks, it is worth to understand since somebody can be inspired by the nature's instructions.




References

Action Potentials in Wikipedia
Dichotomy of Action-Potential Backpropagation in CA1 Pyramidal Neuron Dendrites

Comments

Popular Posts