Neuroscience Methods Tutorial (44:08)
September 20, 2019
August 11, 2019
All Captioned Videos Brains, Minds and Machines Summer Course 2019
Diego Mendoza-Halliday, MIT
Introduction to methods for recording, analyzing, and visualizing neural signals, focused on electrophysiology methods for acquiring signals that the brain uses to transmit information. This tutorial describes the capabilities and limitations of methods such as the patch clamp technique, intracellular and extracellular electrode recordings, local field potentials, electrocorticography (ECoG), and electroencephalography (EEG).
Download the slides (
DIEGO MENDOZA-HALLIDAY: So how many of you have actually acquired neuroscience data, like brain data of any kind? All right. And how many of you have not acquired data, but have analyzed data?
AUDIENCE: Just neuroscience?
DIEGO MENDOZA-HALLIDAY: Neuroscience data, yeah. All right. And how many of you have not acquired nor analyzed neuroscience data? All right. So it's a mix of multiple things. And so one of the reasons why we have this tutorial is because of that diversity of backgrounds.
So the idea is that as we saw, there's a fraction of you guys who might have acquired neuroscience data, who know that you have to train animals, and they poop, and you have to deal with all these intricacies and complexities. Some others are luckier and they just get the clean data-- or sometimes dirty data, as you'll see, but they don't have to deal with that acquisition.
And sometimes they know the complexity of those signals, or how they have to be acquired, but sometimes they don't. They just get the package, and they just basically analyze it, and then they conclude something. And they kind of lack the knowledge of the complexity of the signals they're analyzing.
And some people who are not in any of those fields, have not even analyzed real data, might be reading papers and kind of expecting certain things about the data-- certain cleanness or other things about the data-- and might misinterpret what the results in the papers are, or might maybe underestimate what the papers are. They maybe think that it's actually-- that this means nothing because it's not at the level at which I expect it to be.
And so what I'm what I'm trying to explain here is the complexity of the basic-- of the principles underlying neural signals, and some of the kind of most used methods to acquire them, and try to explain why this is complex, and why the signals don't necessarily look like a lot of people would expect them once they learn about the principles of neuroscience from a book.
And the second goal would be for you to develop a critical perspective of these different methods, to understand their capabilities and their limitations. So some tools are good for some things, but they're terrible for others. And there's a lot of misusage of tools all over the place, all the time. And so you should build that critical perspective as well.
So I think I have gone through that. So let me start off with the simplest neuroscience tool that we've had for hundreds of years. And that is psychophysics, which I call the black box method-- or many people call the black box method. It's very simple. You don't even need to see what's inside the brain.
You present an organism-- a human, an animal-- with a particular sensory input-- set of sensory inputs. And then you measure all of its behavioral outputs, what comes out of the black box. And you do that with different inputs and you get different outputs. And then you can make conclusions based on the correspondence of inputs and outputs about what might be happening in that black box.
And so there is fascinating things that can be done with that, especially if you know already about the brain, which defeats the purpose. But there's a lot of limitations to that, obviously. And what we want to do is really go inside that black box. Open the black box and see what is going on in there. And that opening the black box means a lot of things, as you'll see.
So how do we open that black box? What can we do? How can we interrogate the brain, or understand what's going on inside that computer? So let me start off with explaining what a signal is, first of all.
So it's a-- in general terms. Is a fluctuating quantity in a medium whose variations represent information. So that medium can be light, sound, electricity, magnetism, heat, even materials whose kind of states fluctuate. And those fluctuations can carry information. So the brain uses many of those-- several of those.
And there is a difference between a signal in terms of neural function and a signal in terms of experimental acquisition. That is a really important thing to understand when you're understanding neuroscience methods. Why do I say that? There is a huge difference between talking about a signal like spiking activity or talking about signal like a neuromagnetic signal.
What is the main difference between those two? Spikes as a signal are a signal that the brain itself uses to carry information, right? So it's a signal that is part of the mechanisms of brain function, whereas a neuromagnetic signal is not. Some people might want to claim it-- I'll stay skeptical. But I might say, most likely not. So the brain does not rely on magnetism to transmit information.
But neuromagnetic signals are also signals, and they come out of the brain, and they tell us a lot because they are proxies or indicators of neural activity. But importantly, if you do acquire a signal that's like that, like neuromagnetic or even BOLD signals, you must not interpret them as brain related-- or brain information, right? It's not the brain itself carrying information to itself, but just a proxy for activity that we can use as scientists. So very important to understand the mechanisms of the brain.
So to understand what signals I can acquire-- or different signals I can acquire from the brain, let me bring back, in a nutshell, to what you already saw in a previous lecture, I think yesterday or the day before-- probably yesterday-- which is basically how neurons activate and transmit information between each other. And so I won't explain again, but the idea is you have a particular neuron that has a particular level of activity. And at some points, it wants to transmit information.
It sends action potentials down its axon. And they end up in the synaptic terminals at the synaptic-- so the axon itself-- sorry-- the action potential itself based on its physiological mechanisms has a particular property or frequency contents that makes it very sharp, especially compared to many other signals. And that, as you'll see, is important to separate it from others.
After the spike reaches the synaptic terminals, the currents that the axon itself is representing will cause calcium influx into the synaptic terminal, which itself is a signal that the brain is using. And that will cause these synaptic vesicles to release transmitted molecules into the synaptic cleft.
And that neurotransmitter will then be-- that signal-- in itself, a chemical signal-- would be captured by transmitter receptors, which will open ion channels and allow certain currents to flow into the postsynaptic neuron, and causing postsynaptic currents, which will then flow and in a complex mechanism, join together to cause fluctuations in the overall potential of that postsynaptic cell, which themselves might cross a threshold and spike again, right?
And so that's kind of in a nutshell-- and I won't go into more and more details because we don't have time. But in a nutshell, that's the whole process. So in that entire process, there are several signals that we can acquire. Yes?
AUDIENCE: So I wonder what the definition here for neural signals. Are you saying so based on your definition, do you think that MEG's data is not neural signal? Would you call two-photon imaging also not neural signal?
DIEGO MENDOZA-HALLIDAY: Two-photon imaging is a technique.
AUDIENCE: So the data it acquires--
DIEGO MENDOZA-HALLIDAY: The data it acquires.
AUDIENCE: You can [INAUDIBLE]
DIEGO MENDOZA-HALLIDAY: Yeah. So we'll actually-- if I have enough time, I'll try to go through two-photon imaging and calcium imaging. But actually, if you look at all these signals, then maybe the answer can be here, right? So the idea is that all of these signals are neural signals-- are signals that the brain is using for its own purpose, right?
AUDIENCE: [INAUDIBLE] We don't call a magnetic signal--
DIEGO MENDOZA-HALLIDAY: MEG is a signal. It's just a signal that is not part of the actual mechanism of neural transmission. It is a signal that-- it is an epiphenomenon of this mechanism-- right-- that then can be acquired by neuroscientists. And if it correlates with these mechanisms, can be used as a proxy for neural activity.
But what's important is for you to understand which signals that you can acquire from brain are actually part of these mechanisms, and which are just proxies. And you cannot conclude mechanisms directly from the proxies. You have to know what they're proxies of, and then deal at that level. That's what I wanted you to understand, the difference between those two things. Does that kind of clarify it a little?
AUDIENCE: Yeah. I see your logic. Magnetic field is like [INAUDIBLE] the relationship--
DIEGO MENDOZA-HALLIDAY: Yeah. It's not about the precision itself. It's about the purpose. Is the brain using magnetic fields to transmit information? Is that the question you want to-- you want to go to. Are magnetic fields used by the brain for any purpose or are they just an epiphenomenon? That's kind of at the idea.
All right. So then in all these mechanisms, you get several signals-- oh and I forgot to mention, in those postsynaptic currents that you get after synaptic transmission, those currents themselves generate fields around the neuron. And obviously, neurons are not isolated, they have neighbors. And all of those neighbors together will have an addition of those fields. And those are what we call then Local Field Potentials, LFPs.
And so just to summarize from all that mechanism, we have a handful of signals that we can acquire. These are not all of them. We have, for example, neuromodulators and other few things, but these are the most important that neuroscience use. And those are the action potentials, the local field potentials, and that's the list of electrical signals. And then the chemical signals, which is calcium influx and neurotransmitter release.
So those four signals, if we can acquire them, we can know about what's going on locally in the brain, and even at a larger scale. Everything else will likely be a proxy or an epiphenomenon that we can also benefit from. But it's not part of the mechanisms of the brain.
OK. So let me start with electrophysiology, which is the acquisition of those electrical signals that were in the first part of the list of biological origin. There are different scales at which we can acquire them. So here there is a list of different methods by spatial scale, starting from patch clamp, which deals with individual channels, up to electroencephalography, which deals with the gathered activity of millions of neurons.
So let me just go through it really quickly. I'm just giving you an overview of all of those scales. And the first one, you have patch clamp, where you have a glass pipette that can actually kiss the membrane of a single neuron. By suction, you kind of capture the cell, you lock to it. And then you can record the flux of individual ions through individual receptors and measure what the physiological properties of these individual receptors are, which is really interesting.
Then at a different scale, you can actually-- instead of kissing the cell, you can actually go into the cell-- intracellular space and record the electrical potentials that are happening inside the cell with respect to the outside. And those fluctuations tell you about the computations that are going on in that individual neuron as a computer. And that there's a lot of complexity in there that some of the other methods actually miss. And there's a lot of information there how a particular neuron will gather inputs from other different neurons and create-- and compute an output, and then output those signals. Yes?
AUDIENCE: So would the data collected from these procedures be sort of subject to that uncertainty that you were talking about with proxies? Or would these be legitimate signals you could make certain claims with?
DIEGO MENDOZA-HALLIDAY: These are purely legitimate signals. For example here, ions actually cross there, and that they are essential for brain function here. All those-- excuse me-- you can actually capture action potentials with very, very high precision, and actually intracellular potentials that are basically occurring when that particular neuron receives inputs from different neurons. So it tells you a lot of information. And all of that being directly relevant to what the neuron is actually doing. So none of those are actually proxies. Yeah?
AUDIENCE: What's the length scale for the patch clamp relative to the size of being the neuron
DIEGO MENDOZA-HALLIDAY: Oh, pretty tiny, because you have to capture individual-- sometimes they can capture individual receptors. I actually forget the-- I have it in my notes-- but since this is an overview, maybe I can find that out for you. But it's actually really, really-- really small compared to the other techniques for the reason that I am saying.
You want to capture the properties of individual receptors. And so of course, if you're too large, you're actually going to catch too many of those. And then you miss the purpose. But it's quite amazing that they can do this. And some people are trying to do this like awake behaving animals, which I-- blows my mind.
They also are trying to do this-- have tried to do this in many behaving animals. I know that in monkeys, it's pretty impossible. I think people have been able to catch neurons being inside a nerve for about 10 minutes or so. And that's already quite amazing. It tells us a lot of information. But obviously, you can't have a task with several trials occur during those 10 minutes, so it's limiting.
But what is much less limiting is extracellular recordings. And that's basically one level up. And in that scale, you can have a microelectrode-- I'll tell you about them in a second. And if you introduce it into the brain-- those are very tiny-- you can go between cells. And digging into the extracellular space, you can get close enough to a neuron that you start to acquire some of the electrical signals that are coming out of that individual cell or neighboring cells.
And those include action potentials and local field potentials. And that is because those electrical currents, including action potentials and local field potentials, don't just occur right around the cell, but in fact, propagate through the extracellular space. And what's really cool about this is that you can perform it-- because you don't require that exact precision, you can more readily perform it in behaving animals, including macaque monkeys that I work with.
And so what do those electrodes look like? There are many different kinds. This was used for decades. This is kind of the classical electrode, a tungsten tip that has a recording contact at the very tip and gives you one signal at a time. And that is, for the most part, in many labs, people have tried to kind of move on to different, to more complex techniques that allow you more acquisition, because we're in an era in which single cells are not enough to tell us about the brain.
We want to know about populations of cells that are interconnected, are occurring at the same time. We want to acquire signals at the same time from multiple cells and multiple areas of a brain. So then we move on to other different electrodes. Yes?
AUDIENCE: Is this the Neuropixel that we see right here?
DIEGO MENDOZA-HALLIDAY: That's the last one, yes. I will go through it. So the first thing you can do is you can actually have a single-- a probe, that instead of having one contact, has four contacts. That's what we call a tetrode. And that gives us the ability to then capture the signals from four different locations. And much like the two eyes can actually discriminate between depth of objects, these tetrodes allow to discriminate the activity of multiple cells happening around the vicinity and separate them, which is really-- which is a huge advantage.
And then after that, you can actually have arrays of those. So linear arrays of individual contacts or by tetrodes. And that allows you to record the activity of neurons along depth of the brain. You can also have arrays of-- 2D matrix electrode arrays that capture the signals from the surface of the brain across a 2D space of mostly cortex in mammals. And the 3D matrix array, which is the same-- has the same organization, except it also has depth. So you have kind of a 3D array of recordings.
And more recently, high-density probes, including the Neuropixels that you can see over there, in which the contacts are aligned or arrayed in high density, in such a way that a neuron will be heard by multiple contacts. And these contacts will almost kind of create an electrical image, if you may, of all the neurons that are gathered around the array.
And you can have now thousands of channels recorded at the same time. And obviously, if you want, you can implant multiple of those in different parts of a brain, and you have signals in the 2,000, 3,000, 4,000, 5,000 channels, which is what we're going to see in neuroscience pretty soon. Any questions over there?
AUDIENCE: Yeah. Does the [INAUDIBLE] like if you have two electrodes in that electrode array that are right next to each other, are you going to be getting the same signal twice because it's--
DIEGO MENDOZA-HALLIDAY: Exactly. And that's the principle that tetrodes use, right? They are acquiring a particular set of signals that are partially redundant, but not completely. Much like your two eyes, right? Your two eyes get a little bit of redundancy, but any difference between them gives you information about the location of objects and depth. So a little bit like that. You can use that redundancy and that divergence to kind of distinguish, disambiguate between multiple neurons being captured by all of the tetrodes.
And so what does a signal look, when you acquire it? This is an example of a snippet of what could be a signal. It's made of multiple components. One of it is just noise, electrical noise, that comes from multiple sources-- could be your electrical outputs in the room, it could be actually a radio. I've listened to the through the brain of my animals-- because those get captured also through the probes and some of the wiring.
But also on top of that noise, you get these kind of fluctuations, some of which are the local field potentials I was talking about. And also these sharp spikes that are the action potentials of particular neurons. They could be different [AUDIO OUT] tall one, and then some little ones.
What you can do is if you're interested in a single neuron, can you can then-- well, actually before that, you can filter the data to extract some of the noise that you know is not biological. So you filter out everything below 1 hertz and above 9,000 hertz. And that gives you some of the local field potentials and spikes.
And then to really narrow it down to extracting local field potentials alone and spikes alone, what you do is you low-pass filter your data at around 250 hertz so that those low frequencies can get rid of all the spikes. So you basically would end up with this signal without the little spikes in there, and that's your local field potential. And if you do the opposite, you high-pass filter it at around 300 hertz-- so everything above that kind of gets rid of those lower frequencies and flattens the signal a lot, so that you can see those spikes all gathered together.
And at that level, you can threshold the data. So you say, every point at which the data crosses a particular threshold, that's going to be a spike, or a waveform-- a potential waveform. Then you align all of those waveforms to threshold crossing. And you can see that if all of those waveforms look very similar, they're likely from the same neuron. That's kind of a principle that we rely on in electrophysiology, which is that a particular neuron will tend to spike with a waveform that is a signature of that neuron.
So all waveforms will look very similar to each other, and much like your fingerprints. You do a fingerprint here, and there, and there, and there, and there, they all tend to be very similar, and different from those of other people. So those are the fingerprints of a neuron. Yes?
AUDIENCE: Do you have any idea how much the waveform would vary across [INAUDIBLE] Is it pretty consistent on longer-term scales as well?
DIEGO MENDOZA-HALLIDAY: Yeah. Some people have studied that. I only know of one paper, I think, John Reynolds, which was actually looking at the variations in those spikes based on different cognitive areas-- I think attention, in that case. I think they discovered that there's an effect on the spike amplitude by attention. I'm not sure about the long term.
And I believe that based on the physiology of the cells, that shouldn't change too much because it's basically-- it depends on the cell type, on the cell size, shape, they arboring-- dendritic arboring, et cetera. And those things-- obviously, there's plasticity, so one would expect changes. But in general, don't those don't change from one day to the next. That would be my guess.
So because they are quite similar, you can actually use it as a signature of a particular neuron. And if you capture multiple neurons from the same electrode-- like in this case, you see these tiny little spikes as well, that might be actually present here from another neuron, or another pair of neurons-- you might have three neurons in there. What you can do is you-- if you're not interested in that one cell-- one big cell-- but all of the cells in there, you can increase your threshold a little bit until it captures all of the spikes trying to get the noise-- to keep away from the noise.
And then you can capture spikes from multiple cells. If you align them all, you'll see that some spikes go like this altogether, and they kind of look alike. And then others do this, and they all look alike. Others do this, and they all look alike.
And you can use a tool like this software, the Plexon Offline Sorter, Blackrock has some-- a lot of them-- that can-- a combination of automatic and manual methods can allow you to do principal components analysis of those spike waveforms, a cluster analysis, and then you can gather and see that spikes of similar shape will tend to cluster, and those clusters will tend to represent the different cells or neurons that are present in your signal. And so you can sort them out. Yeah?
AUDIENCE: So depending on your clustering algorithm, you may say [INAUDIBLE] that there are two different neurons, or you may say there are one type of neuron. How do you differentiate one [INAUDIBLE] years of experience. This is this one kind of algorithm.
DIEGO MENDOZA-HALLIDAY: Yeah. The only way to do it is really with a ground truth experiment. Those are very hard to do. So you basically record-- at the same time-- intracellular recordings and extracellular recordings. And the intracellular recordings tell you the truth, right? Because you're inside a cell, whenever that cell spikes, you're really able to know what it is.
And then you can compare it with the extracellular recording. And you can vary the distance of the extracellular recording. They've done some of those experiments. And some of the spike sorting algorithms depend on that ground truth data to see how well they performed the role of spike sorting. But that's really hard-- it's hard to get the data. Yeah?
AUDIENCE: I wonder if I want to get sharp wave ripples, which frequencies might be [INAUDIBLE]
DIEGO MENDOZA-HALLIDAY: If you get what?
AUDIENCE: If I want to get sharp wave ripples, it's very high-frequency data. Which kind of low frequency might you low-pass?
DIEGO MENDOZA-HALLIDAY: Sorry, I'm not able to hear. If you want to get what?
AUDIENCE: If you want to get sharp wave ripples--
DIEGO MENDOZA-HALLIDAY: Oh, sharp wave ripples. I mean, I'm not into hippocampus data or data related to other kind of data like that. I haven't dealt with it. But I am assuming that extracellular recordings will be able to capture that.
DIEGO MENDOZA-HALLIDAY: I'm not sure what kind of frequency you would deal with. I've never dealt with sharp wave ripples, actually. So I'm not sure. But I believe that they would they would end up around the local field potentials, somewhere around there. So in those frequencies. But I'm not I'm not too familiar.
AUDIENCE: Well, I guess my question is [INAUDIBLE] of the-- so let's say [INAUDIBLE] signal at 100 hertz. And then if I wanted to low-pass filter, would you recommend I do 200 hertz, so 300 hertz? Like what is the [INAUDIBLE] you would recommend to do low-pass filter.
DIEGO MENDOZA-HALLIDAY: Oh, it depends on what your signals of interest are.
AUDIENCE: Yeah. That is what I'm saying. [INAUDIBLE]
AUDIENCE: If you want the sharp wave, you want to band pass filter around 15 to 20 hertz, I think. But definitely it's the sharp wave reflection. Because the sharp wave and the ripple [INAUDIBLE]
AUDIENCE: The sharp wave is not definitely, not over 100 hertz, right?
AUDIENCE: It's a very fast downward spike, but [INAUDIBLE] so there's a lot of noise [INAUDIBLE] and typically, if you want to see a ripple, you're going to want to be somewhere around 1.5--
DIEGO MENDOZA-HALLIDAY: All right. So in the interest of time, let me move on to-- so this is basically offline spike sorting after you've acquired your data. And why is that so important? So many people in the field will not offline sort their data. They will not even sort their data.
What they do is they just threshold their data. And they get a bunch of spikes, and they gather them together, they call it multi-unit activity, and they just analyze that. I'm a strong believer in spike sorting for multiple reasons. And I'll try to convince you of that with one figure that I created from one of what my recordings. And there are many, many, many of those. This is just an example.
Consider those two neurons there. This is the firing rate over time of each of those two neurons to two conditions. Forget what the conditions are-- doesn't mean anything at this point. What you can see is that the selectivity for those two conditions in one neuron actually sustains over time. The other one reverses.
Those are neighboring neurons. They're just a couple of microns away from each other. And they have completely different profiles of activity. And I can bet you that there is a reason why the brain has those two neurons, and they are connected in a particular way, and they're connected to other neighbors in a particular way. They form a complex microcircuitry that requires this kind of reversal activity, et cetera.
If I were to do what many of my colleagues do and forget about sorting, I would have ended up with the neuron-- the unit down below, from which I would have concluded that the neuron presents selectivity for the conditions, and then loses it altogether. And has no selectivity during the second period of a task. And that would have completely missed the point.
Why else to do it? Well, not only to understand the complexity of the individual unit of signal processing, which is the individual neuron, but also because we can. Many people cannot. They do EEG, MEG, acquire BOLD signals. And their signals include the activity of millions of neurons. And we are putting electrodes in there going, through the trouble, and we have the ability to have single neuron data.
And just because we want to be lazy about one little step, we miss out on that complexity. I say, please don't do that. Take advantage of that. One more reason is, in fact, the spike waveform of individual neurons even tells us about the neural types.
You heard a lecture yesterday from Christof Koch about all the complexities of neuron types. This cannot capture that complexity, but it can tell you some of those-- about some of those differences. And one of them is that it's been shown that the spike waveform width is an indicator-- plus or minus, an indicator of the likelihood of a neuron being putatively excitatory and putatively inhibitory.
By sorting your units, and having this spiking activity from individual units, you can then come up with a template spike from each neuron. You can plot at them all, and you can see that they don't show waveform widths along the entire temporal dimension, but they gather into two. And that histogram shows it.
So short spiking, or narrow spiking, and broad spiking units, which we know are indicators of what inhibitory neurons and excitatory neurons tend to spike. So this can allow us to separate neurons into putatively inhibitory and putatively excitatory neurons. And then allow us to ask questions regarding the microcircuitry of the area we're recording. Or do inhibitory neurons and excitatory neurons respond in a very different way to our task? And how are they connected?
All right. So once you have the spikes, the way to reduce your data is to create timestamps. So basically, all that you need to get is the timestamp of each of the spikes in your recording. And once you get that, then you can graphically represent your spiking activity in many ways. So you can create the simplest representation, which is a raster plot in which you put a little bar at the timestamp of each spike. And you can actually align them to the onset of many trials that you can have in your task.
So typically, you have an experiment where one particular task is repeated multiple times, in multiple trials, and you align all of the rasters to the onset of the trials. You can have many trials, and that gives you an indication of the replicability of your data, and how variable it is across repetitions.
Another way to gather more averaged version of it is to create kind of a frequency histogram of that spiking activity over time, which we call a Peristimulus Time Histogram, or PSTH, which will look like that. I don't tend to like them so much because they're spiky. I like to-- in fact, smooth my data to create what we call a spike density function.
So each spike, instead of being a single millisecond of activity, gets transformed into a Gaussian-- a little Gaussian with an amplitude. And when you add up all the Gaussians, they give you a curve like that. And you can then plot those curves for many conditions, and then compare them in one single plot, which is very nice. And you can even show the standard error across trials or repetitions of each condition.
Moving onto local field potentials then, you have these fluctuations. One of the best ways to analyze the data is to look into the oscillatory activity of these potentials. So we know that the brain often communicates-- or actually, represents information using oscillatory patterns, or communicates between areas with oscillatory patterns. And those oscillations can be captured at the level of the local field potentials.
And so the way to analyze the oscillatory activity is to do spectral analysis, or a Fourier transform, that takes your actual signal or your trace and extracts the power of each frequency in a particular range. And so one way to represent that over time is with a spectrogram like this, where you can have frequency in one axis, time in another axis, and your color axis will represent the power at each frequency.
So in 80 hertz, you have low power. And at some point, the power at that frequency increases. You see that that particular frequency, 80 hertz, starts to increase in amplitude. And that tells us about the different components that might be occurring in that signal at the same time. So you may be able to capture two different frequency components co-occurring in the same signal because you're separating all of the components.
And you can represent the data by averaging time, and just presenting the power as a function of frequency in a particular time of your task, and compare different conditions to see whether you're experimental variables are represented in the power of different frequencies. Yeah?
AUDIENCE: [INAUDIBLE] range that it could be used for [INAUDIBLE] because maybe the spectrogram can be done by [INAUDIBLE]
DIEGO MENDOZA-HALLIDAY: When you say parameters, what kind of parameters are you talking about?
AUDIENCE: [INAUDIBLE] spectrogram like size of the--
DIEGO MENDOZA-HALLIDAY: Right, your bin size and all these things. I don't think there is one way to do it. I think it depends a lot on the timescale of interest. So if you are, for example, looking into signals that you know occur in timescales of several hundreds of milliseconds, you might not need to take bins of 20 milliseconds. That might be overdoing your data.
But some data might require bins that are smaller because the task includes a lot of complexity in the timescale. So I think it really depends on the timescale of the function that you're looking into. I've used something around 100 to 150 milliseconds because my tasks are cognitive tasks, where-- delay period where there's a maintenance of information over 3.2 seconds does not require that much specificity, but it depends.
AUDIENCE: Difficult frequencies [INAUDIBLE]
DIEGO MENDOZA-HALLIDAY: Yeah. So classically, you have your frequencies, including delta, theta, alpha, beta, and gamma, which go from about perhaps 3 hertz up to 100, 150 hertz in the high gamma. Above that, it starts to get a little tricky, and spiking activity starts to kick in. So those would be the best. So the lower frequencies, from 3 to about 20 hertz, from 20 on upwards, you get higher frequencies, including gamma.
All right. So let me just look at the time. OK. So let me just finish. Unfortunately, I won't have time, because we have a bunch of other really interesting things to talk about. So I will I will probably stop at the electrophysiological signals. These are some of the most important. And if you have questions about other techniques, other signals, like BOLD, EMG-- sorry-- MEG, magnetoencephalography, calcium imaging, et cetera, you can come to me and ask me.
But just to wrap up, you get then two more scales in those electrophysiological recordings. The next one is based on surface electrodes, like those that you see there, that could be implanted in patients. So the advantage is that they are not brain invasive. You can put them on the surface of the cortex, and then capture field potentials that are the aggregate of-- a large-scale aggregate of many local field potentials, if you may.
And so they look a lot like local field potentials in the nature of the data. But the spatial scale is much broader, so the resolution is much less. And therefore, you can't really tell apart between the activity of fine-tuned populations as you can do with local field potentials. But you can do many really interesting things, and capture a lot of features of brain processing and selectivity in those spikes. But the analysis is very similar to local field potentials. You're basically looking at the power spectra like you do with local field potentials.
And most interestingly, it doesn't stop there. You can actually leave this the skull in there-- leave the scalp-- the skin-- and even hair in there, and record from above, and you can still get those signals, which still blows my mind. I would have never bet for something like this to work, but it does work. So people put these electrodes over the scalp, and they can also see these kind of fluctuations in the electrical potentials.
And obviously, these are very low resolution. So they gather basically the potentials-- the field potentials happening in a massive area around each of those electrodes. But by putting several of them, you can actually help the resolution by kind of fine-tuning with the inverse problem to where things might be occurring, which is similar to what the magnetoencephalography can do. But with the advantage that you don't lose too much in temporal scale. So the timing of the signals is not that different from your local field potentials with an electrode, because it's electrical signals that transmit very, very quickly. Yes?
AUDIENCE: Is it the same, for example, [INAUDIBLE]
DIEGO MENDOZA-HALLIDAY: What do you mean by the same?
AUDIENCE: Does it represent [INAUDIBLE]
DIEGO MENDOZA-HALLIDAY: Yes, but at different spatial scales. So actually this, this is nicely explained in this paper by Buszaki and colleagues, from 2012, where they recorded from different levels-- different parts of the cortex. And they could show you what intracellular recording looks like with all its action potentials, and then intracellular potentials, and what that would look like in local field potentials recorded from a deep electrode. And then if you were to put an electrode in the surface, what the same thing would look like.
And so what they see is that, obviously, the first thing that you lose is from the [INAUDIBLE] you don't see the spikes, but you still see those field potentials kind of fluctuating in a very correlated fashion. And mainly, you see a loss in the amplitude of the signal, and some buffing of the signal because obviously here, you're capturing the local field potentials, not only from this region, from this one, and this one, and this one, and this one. And those may differ a little bit. So you lose a little bit of that resolution.
But there is some correspondence. And that is then the end of electrophysiological signals. We have a whole array of other techniques to measure other signals, like the chemical signals I was telling you about. Calcium influx can be measured with calcium imaging. You can measure neurotransmitter release with many other techniques that are sensitive to changes in the amount of neurotransmitter present locally in certain regions.
And then you have all those techniques that use the epiphenomena that I was talking about, like BOLD signals and neuromagnetic signals that unfortunately, are beyond the scope of the tutorial this point, because we want to include more about experimental design or other things that are going to be very useful for you for the summer school. So if you want to learn about any others that are of interest to you, just come talk to me, all right? And I'll pass it on.