An interesting paper was recently published in PLoS ONE, which follows from a long line of experiments regarding how brain cells encode information. But before I get into it, and I want to get into it, let me provide a little context. One thing that makes neurons special among different cells is that they can generate electrical impulses called action potentials or spikes. Several other cell types also can generate electrical impulses such as muscle cells in the heart and skeletal muscle, and even some plant cells. In the canonical view, a neuron receives inputs from other neurons, and if these inputs are sufficiently large, they will cause the neuron to generate an action potential. The action potential then travels down a specialized protrusion called an axon and ends at what known as a synaptic terminal. The action potential causes the release of chemicals called neurotransmitters from the synaptic terminal. Neurotransmitters then activate other neurons and ultimately cause these to generate more action potentials and ultimately communicate with more downstream neurons. Furthermore, axons can be very long – the nerves that run throughout the body are bundles of many axons and can more than a meter long. In order for action potentials to travel down nerves, they are continually regenerated by a seers of proteins called ion channels.
One important aspect of how neurons compute information is that the strength of a stimulus is encoded by the rate of action potentials – a stronger stimulus will cause a neuron to fire action potentials more frequently. This is something known as rate coding and was first described by Edgar Adrian, also known as Lord Adrian (back in the days when science was often done by aristocratic types who could fund their own work), in the 1920’s. Adrian used a device, ( a triode thermionic valve amplifier and a capillary electrometer, for aficionados) in which he mounted an isolated muscle from a frog leg with the sensory nerve still attached. Muscles have all sorts of sensory receptors that help them detect things like stretch or pressure, and this information is conveyed to the brain via sensory nerves attached to these receptors. In Adrian’s experiment, he recorded electrical impulses from the sensory nerve while he hung different weights from one end of the muscle, creating increasing amounts of tension. Here’s his diagram of the device:
What he found was that as he stretched the muscle, he could record discrete electrical impulses in the sensory nerve, and that the rate of these impulses increased as he increased the size of the weight. He therefore concluded that information about tension was being conveyed by the rate of these impulses, and stronger stimuli resulted in faster generation of spikes. He also made a point to note that these spikes were discrete events, mostly about the same size, and were therefore likely to be all-or-nothing events. A few years later the experiments of Hodgkin and Huxley described the ionic basis of these action potentials, furthering the view that action potentials were discrete events and that neurons communicated with each other by varying the rate at which action potentials were generated. If a neuron fired more action potentials it would release more neurotransmitter, if it fired less it would release a smaller amount of neurotransmitter. This became one of the fundamental tenets of neuroscience.
However, nothing is ever as clear cut. If you ask anyone who performs electrophysiological recordings from individual neurons, they will tell you that although action potentials are discrete events, the can vary widely in size and shape within a single neuron. For example, in many neurons when you cause them to fire a train of action potentials, you will sometimes see that the size of the action potentials diminishes the further along you are in this train. This led people to speculate that it might not just be the rate of action potentials that affects the output of the neuron, but also their size and shape. In 2006 a pair of papers by independent labs came out in Science and Nature which showed that in fact smaller action potentials could result in less neurotransmitter release and bigger and longer-lasting action potentials could result in more neurotransmitter release. Thus coding of information was not just digital, but could have a graded analogue component. These papers received a lot of attention because the seemed to overturn one of the long-held dogmas in neuroscience.
OK, now back to the PLoS paper. In this paper, Chen and colleagues essentially reach a completely opposite conclusion than the Science and Nature articles. They show that even if they record action potentials in the cell body of a neuron that have different sizes and shapes, the output of the neuron, as measured by the release of neurotransmitter, does not correlate with action potential size. To test how this happens, they record electrical activity from within a neuron in the cell body and in the tip of the axon simultaneously. What they found was that although action potential size can vary significantly within the cell body, by the time the action potentials reach the end of the axons they are all roughly the same size. This suggests that there is something that the axon does to normalize the size of the action potentials. The authors go on to demonstrate that this is due to specific properties of the ion channels in the axon that allow action potentials to be regenerated as all or nothing events. If you look at the figure below, in panel A you see action potentials (the 2 bumps) recorded in the same cell over several trials and notice how the second one varies a lot in size. In panel b, notice that for the most part, the size and shape of the second action potential is the same for all the trials.
Thus, what the authors conclude is that neurons have a built in process by which they can amplify even little action potentials in the cell body and faithfully transmit them down the axon. This ensures not only that rate-coding can work efficiently, but eliminates some of the variability in the output of the cell that could be introduced by action potentials of different sized. As I said before, this is exactly the opposite conclusion than the 2006 papers. This doesn’t mean that those papers are wrong, what might vary is the type of cell they record from, the exact recording conditions or the age of the animals used. What surprises me is that the authors of the PLoS paper only mention the other papers in passing in the introduction even if their work is directly relevant. It also surprises me that they published this in such a low-impact journal. The experiments seem well done and there is a lot of work and novelty in it. And this may go to show that if a lab is not well known, their work will not get published in a high-visibility journal. Finally, I think that this set of papers illustrates how even concepts that are accepted as central tenets can change as different experimental techniques become available, and that the controversies that were relevant 90 years ago, are still alive and well, and like all good scientific theories, they evolve over time.
Chen, N., Yu, J., Qian, H., Ge, R., & Wang, J. (2010). Axons Amplify Somatic Incomplete Spikes into Uniform Amplitudes in Mouse Cortical Pyramidal Neurons PLoS ONE, 5 (7) DOI: 10.1371/journal.pone.0011868
Adrian, ED. The impulses produced by sensory nerve endings: Part I. J Physiol. 1926 Mar 18;61(1):49-72.