Single Unit Data and Spike Trains#

As you know, neurons communicate by generating action potentials. Action potentials are also called “spikes”, because when they are recorded, they manifest as spikes in the electrical recording - transient changes from a baseline level of voltage to a different level, then a return to baseline. Neurons have a negative resting potential, meaning that their intracellular space has a negative voltage (more negative ions, like chloride) relative to the extracellular space (which has more positive ions, like sodium). When an action potential occurs, ion channels in the neuron’s cell wall open, allowing the intracellular space to become more positive. Thus the neuron depolarizes when it fires, and its voltage moves closer to zero. The sodium-potassium pumps on the cell wall then re-polarize the neuron to its resting (negative) voltage.

Source: *Image by reproduced from [Wikipedia]( and used under a [Creative Commons Attribution-Share Alike 3.0 Unported license](*

Spike trains are recordings of action potentials, from electrodes inserted in the brain. Often these are from intracellular recordings, meaning that an electrode penetrated the membrane of a neuron and records its action potentials. In other cases, spike train data may be from extracellular recordings, in which the electrode is inserted into the brain but does not penetrate a neuron. Instead, the electrode is located in extracellular space (between cells). In this case, the electrode will likely pick up action potentials from multiple nearby neurons. Extracellular recordings are commonly done with arrays of closely-spaced electrodes, such that multiple electrodes pick up the spiking of each neuron but — based on the proximity of the neurons to the electrodes — each electrode likely picks up action potentials from only a subset of the neurons that another electrode does. A process called spike sorting is applied to the data after it is recorded to attempt to identify spikes from different neurons, based on which electrodes detect which spikes.

Spike train data are binary — either a neuron is spiking (“on”) or not (“off”). Even though, technically, the membrane potential of the neuron rises continuously from its resting state potential to a peak value, then declines again (as shown in the figure above), the data as we work with it at binarized as either spiking (1) or not (0). You can make an analogy from this to Boolean types in Python: a Boolean variable is either True (1) or False (0), just as a neuron is either spiking or not. This is distinct from many other types of neural data, such as EEG or fMRI, or even local field potentials recorded from implanted electrodes, which record a continuously-varying range of data values (not just on/off).

Spike data, or “spike trains”, ultimately consist of information concerning the times at which spikes occurred. Broadly speaking, there are two forms for this sort of data. In what we’ll call the “continuous” form, we have data at every time point over a time period (e.g., a 2 s period covering one trial in an experiment). At each time point, the data is either 0 (no spike) or 1 (spike). In the “discontinuous” form, the data instead are represented as the time points at which spikes occurred. The assumption is that nothing happened, and we only encode “news of a difference” from that assumption. This is a much more efficient way of representing the data, since we don’t encode all those time points where nothing happened. But regardless, both are valid ways of representing the data that you’ll see in this unit of the textbook.

Here’s a great article by neuroscientist Mark Humphries on spike trains in a data science context, that will give you more background information (thanks to student Meg South for the link!). In this textbook we focus on the first of the three approaches he talks about: determining whether there is structure in the data.

If you want to understand more about the process and lab techniques involved in recording spikes, there is a great (30 min) video by Dr. Carl Petersen.