What does that MRI signal MEAN, anyway?

Jun 16 2010 Published by under Behavioral Neuro, Neuroanatomy, Neuroscience

Sci was incredibly excited to see this paper come out. It's got lots of stuff going for it, and all its powers combined were enough to send Sci bouncing around in her seat and sending emails to Ed Yong saying "OMG COOL PAPER!!".
What's it got, you say? It's got the meaning of life, the universe, and that pesky MRI signal.
ResearchBlogging.org Lee et al. "Global and local fMRI signals driven by neurons defined optogenetically by type and wiring" Nature, 2010.
fMRI1.jpg
Ah, the pretty brain picture. But what does it MEAN?


By now, I'm sure you've all heard of BOLD fMRI. fMRI is functional magnetic resonance imaging, and BOLD stands for "blood oxygen level-dependent". So what that breaks down to (basically) is that fMRI uses magnetic resonance to form images of the anatomy of your brain. When you make MRI functional by adding a BOLD signal, it not only reveals the anatomy of your brain, it also shows where blood oxygen is being heavily used, and thus which regions of your brain show "activity" during certain tasks. For instance, your occipital cortex, which is very important in the processing of vision, "lights up" with a BOLD signal when you are viewing something. Blood rich in oxygen is being recruited to the area and oxygen is being used.
But for a long time now (well, a long time since the invention of MRI), scientists have been skeptical. MRI is great in that it's very non-invasive (Sci has done time in many MRIs are a graduate student, and though the noise is annoying, it's certainly not invasive), and in that it can show us the anatomy of the brain (or any other area, for that matter, or at least whatever area you can fit into the scanner). But the BOLD signal has been a thorn in the side of neuroscientists for some time. We know that there is "activity" and that blood, with oxygen, is going to the area and being used, but what does this MEAN?!
Now, you might say "well, that's dumb, of course it means that brain area is involved in whatever thing you're doing". Well, yes, but HOW. There are excitatory, inhibitory, and modulatory groups of neurons in the brain. You might have activity in one area when you're performing a task, but is that excitatory activity? Or is that inhibitory activity that is inhibiting something else to allow excitation? Or is it even more complicated than that? Scientists don't know. So all they can really do is say that a certain area of the brain which gives off a BOLD signal in response to a certain task is "activated". They can't say HOW.
And of course, the media doesn't like this. Heck, science doesn't like this. It's like being given a random assortment of exactly half of the pieces of a puzzle. And so this has led to a tendency to over-interpret fMRI data. Scientists often try not to do this, but we're only human, and we love our pet hypotheses. But even if the scientists don't do this, the media often does, saying that "OMG activity in this area means you're psycho!" or something. When in reality, all we have is activity in an area, which corresponds to a task, and with no real idea as to what it DOES.
Until now.
The scientists in this paper have found a way to interpret the fMRI signal. They figured it out using optogenetics, which is the new hotness in neuroscience techniques these days, and really does seem like something out of a Sci-fi novel.
Here's how it works: There are certain channels in cells (the ones we are concerned with are from green algae) which can respond to LIGHT. These are called channelrhodopsins. When they are hit by light of a specific wavelength (in the case of this paper, 473nM, which is a blue, the channels will open and ions will move in and out. In the case of some of these channels, the ions moving in and out will cause the cell the channel is on to fire.
Looks kind of like this:
fMRI2.png
A few years ago, Karl Deisseroth (who may win some BIG PRIZES for this discovery, he's already a rockstar of neuroscience) discovered that you could take the gene that encodes this light-responding channel, and place it into a virus. You could then use carefully targeted viral infection to infect a local area of neurons with this virus, implanting the channel into the cells.
And guess what happened then?
fMRI3.png
Looks cool, yeah? Above you can see these viruses placed into the primary motor cortex of a rat. When you stimulate with light, the now-light responding cells will FIRE in response to the light! When you put this into the motor cortex, you will get a motor response, like in this video:

Here you can see a mouse, who has a viral mediated infection of light responding channels in his motor cortex. He's just sniffin around and hanging out, but when he gets a jolt of blue light (which you can see), he starts to run, because his cortex has been activated.
(Is that the freakin' coolest thing or what?!? We're on the way to having our own little mouse armies that you can control with LASERS ATTACHED TO THEIR HEADS!)
dr-evil.jpg
This technique is called optogenetics, and is already the next big thing. Using optogenetics, you can stimulate very specific parts of the brain using light, and see what happens. You can also link these light-responding channels to very specific receptors, and see what happens when ONLY those receptors in specific areas of the brain are activated. It's really massively cool, and you're probably going to be seeing huge numbers of papers coming out about this (well, tons already HAVE, but a lot more WILL).
Now, on to this paper.
So we know that the problem with the fMRI BOLD signal is that we know where the blood is going, but we don't know what neurons specifically are being activated. So what you really need to do is be able to specifically activate a bunch of neurons in a specific area, and then check the BOLD signal.
Well, with these light controlled channels and viruses to insert them into the brain, that isn't so hard, now is it?!
fMRI4.png
Here you can see what happened in rats that had light-responsive channels in their motor cortex. The rats were anesthetized and lying asleep in the scanner (getting a rat to hold still for this sort of thing is a significant challenge). Then, they gave the rat pulses of light. And the BOLD signal leaped up (Figure E) each time they did.
So, what's so great about that? Well it shows definitively that the BOLD signal (and thus blood flow and oxygen usage) are directly correlated with actual neuron firing, because every time these neurons fired, the BOLD signal went up.
What is also really cool about this is that the virus infected only excitatory neurons with the light-responsive channels, showing that the BOLD signal also correlates with excitatory neuron firing in the brain. To drive this home, the authors ALSO infected a bunch of inhibitory cells with this same light-responsive channels. This time, when they shone a light, they got DECREASES in BOLD signal which corresponded with increased stimulation of inhibitory neurons. This means that not only does the BOLD signal correlate with excitatory neuron activity, is also negatively correlates with inhibitory neuron activity.
So this is massively cool enough. It implies that when we are looking at a BOLD signal in a patient, we are looking at primarily excitatory neuron activity. This means, for the first time, we can begin to interpret the BOLD fMRI signal. Of course, there could be other neurons activated as well which are neither excitatory nor inhibitory, but it's a start.
BUT, the scientists for this paper didn't stop there! They also looked at where the neurons they were stimulating WENT. They could do this because the virus that infected the neurons with the light-responsive channels allowed the channels to be expressed all over the neurons. Thus, when the cells were stimulated with light, the whole thing was activated, and if you attach a GFP (green fluorescent protein) to your light channels, you get something like this:
fMRI5.png
You can see that light stimulation activates certain neurons, and causes the neurons that are activated to glow all along their length. This means that, not only do you know which neurons you've activated, but you also know WHERE they go! They confirmed this by checking the electrical signals of the neurons, checking the signals where it started, as well as where it projected to. This could be really important in neuroanatomy, allowing us to look very carefully and very specifically at neuronal projections in the brain, where exactly they are going, and, with optogenetics, what exactly they are doing when they get there.
But the really cool finding here is that, for the first time, we may have come a little closer to understanding what those BOLD fMRI signals mean. We can't put virus-delivered light-activated channels into the human brain, and we don't have the specificity to do it for every cell type, but we do have a start. And by carefully mapping the specific regions of animal brains using this technique, right down to the specific cell types, we may one day be able to determine what exactly your brain scans are showing you.
Lee, J., Durand, R., Gradinaru, V., Zhang, F., Goshen, I., Kim, D., Fenno, L., Ramakrishnan, C., & Deisseroth, K. (2010). Global and local fMRI signals driven by neurons defined optogenetically by type and wiring Nature, 465 (7299), 788-792 DOI: 10.1038/nature09108

13 responses so far

  • Janne says:

    In the text above, I suspect you don't mean "inhibitory cells". You perhaps mean that neurons getting excitatory inputs get a positive BOLD signal, and neurons getting inhibitory inputs get a negative BOLD signal?
    If not, why would inhibitory neurons decrease energy use when they become active? Inhibitory neurons aren't fundamentally different from excitatory ones after all; just the destination synapse (on another neuron) differs.

  • Michael says:

    Awesome! *fistbump*

  • Scicurious says:

    Janne: actually, to get a negative BOLD signal WOULD mean that decreased energy is going to the area. And some inhibitory neurons (in this case, the ones containing parvalbumin which they were working on) are indeed different from excitatory neurons. It's not just where they are going, it's what is being released from them. For the excitatory signals, they infected only principle cortical neurons containing CAM Kinase II, but not GABA or glial cells, with the virus containing the channel rhodopsin. These cells when stimulated produced a positve BOLD signal. When they infected parvalbumin positive cells which have GABA (an inhibitory neurotransmitter) they got a negative BOLD signal.
    But it is interesting, and kind of odd, that inhibitory neurons would DECREASE the BOLD signal, implying that they decrease blood oxygen flow to the area. Unfortunately I don't know enough to speculate on that one. In this particular case, the inhibitory cells they were looking at were possibly connected to vasculature, which could explain it, but the principle behind it is definitely confusing.

  • daedalus2u says:

    Janne, I think that Sci has it right. They are triggering excitatory and inhibitory nerves and it is these nerves that are affecting the fMRI BOLD signal. I think that vasodilatation is all mediated through NO, and this is how the brain metaprograms itself, by sending NO to regions via excitatory neurons and the increased NO then increases the activity of that specific neuronal tissue so that action potentials can propagate into it and do computation.
    There are some misconception here. FMRI BOLD does not measure O2 consumption, it does not measure neuron activation, it measures local relative oxyhemoblobin and deoxyhemoglobin concentrations. These are not necessarily related to O2 concentrations, they are related to hemoglobin concentrations. More blood, even with the same PAO2 will give a fMRI BOLD signal. If all the blood were at equilibrium, then O2 concentration could be inferred from the oxy/deoxy ratio, but that is not the case. The arterial blood has high oxyHb, venous blood has low oxyHb, what changes is the ratio of arterial to venous blood.
    The fMRI signal comes from vasodilatation. The vessels in one region of the brain dilate, the vessels in another part of the brain constrict and the relative amounts of oxyhemoglobin in the two regions changes resulting in the fMRI signal. Because the skull is rigid, the total volume of the brain and all the fluid in it stays constant. More blood somewhere observed by fMRI means less elsewhere because the total quantity of blood in the brain stays constant.
    What regulates vasodilatation in the brain and everywhere else is nitric oxide. It is little puffs of NO that cause the vasodilatation observed in fMRI. The puffs of NO come before the neural activity and (my hypothesis) it is this NO that makes nerve cells more sensitive to firing, so that the action potentials propagate into the region made more active by the puff of NO. You can get a fMRI BOLD signal without follow on neuronal activation. This was shown in monkeys where by using repetitive visual signals, the monkey brain could be “tricked” into sending a puff of NO to a region in anticipation of a signal from the eye needing to be processed. When the visual signal was anticipated by not provided, there was a BOLD fMRI signal but no neuronal activation.
    Even in very highly active regions, most nerve cells are not firing. The details of which cells are firing and which axon are conducting action potentials is obviously important and is what is doing what ever computation that neural region is doing.
    The usual case is NO causing vasodilatation before there is neuronal activation. The NO release comes from the excitatory neurons. Because the total blood flow in the brain is constant, for every region with more blood flow there have to be equivalent blood flow regions with less blood flow. For every excitatory region there has to be an equivalent blood flow inhibitory region such that the total blood flow in the brain remains constant.
    What is very exciting for me about this result is that it shows that some of the basal vasodilatation in the brain is caused by local basal NO levels, and that by stimulating the inhibitory neurons, there is a reduction in vasodilatation, implying a reduction in local NO levels showing that basal NO level is directly coupled to basal vasodilatation and any change in the basal NO level will change the vasodilatation. This implies that all disorders of insufficient brain blood flow are characterized by decreases in basal NO levels.
    This is a very cool technique, and yes you could use it to program and control brains with it; mouse, rat, horse, human. Because it doesn't take implanted electrodes, this should be easier and more reliable.

  • bsci says:

    I might post a longer comment on the article a bit later, after I read it, but I wanted to correct some things in daedalus2u's comment. There are several pathways that cause vasodilation. Changes in NO, Ca2+, and CO2 concentrations can all alter vessel diameter. There are several mechanisms through which these act and interact, and they occur after different thresholds and timing of neuronal energy usage. To make this purely a story of NO is a vast oversimplification. The reason this is relevant here is that this method can possibly be used to isolate and study each of these mechanisms in detail! I've be very surprised if the truth is merely that all excitation increases the BOLD signal and all inhibition decreases the BOLD signal. This may be true for most common cases, but the exceptions are going to be interesting.
    daedalus2u's description of what the BOLD signal is also slightly flawed. Using standard methods, the more deoxyhemoglobin in a measurement volume (a voxel) the lower the BOLD signal. The measurement had nothing to do with arterial to venous blood. In fact, since arteries have primarily oxyhemoglobin, standard BOLD doesn't show much change in arteries. It's in the capillaries and veins where deoxyHb concentrations might change that generates BOLD signal changes. There are three main ways that BOLD changes are observed in blood:
    1. The total blood volume in a voxel changes. Even if the blood retains a constant x% deoxyHb, if there is more blood, there is more deoxyHb in a voxel and decrease BOLD signal.
    2. Neurons extract more oxygen from blood. This will increase the % of deoxyHb in the same volume of blood and decrease the BOLD signal.
    3. The rate of blood flow increases. Given no or small blood volume changes, this will bring in more oxyHB thus decreasing the relative amount of deoxyHB and increase the BOLD signal.
    Since we see a BOLD signal increase during excitation and most standard fMRI tasks, it's actually #3 that is dominating the response. This means that your brain is generating excess oxygenation (#3) in response to region oxygen needs (#2). I definitely took forward to ways we can use these two methods to break down and examine this slightly strange vascular response.

  • Bold Man says:

    Pure excitatory cell activity produces an increase in the bold signal. Pure inhibitory cell activity produces a decrease in the bold signal. This is what that paper shows. Can we interpret the bold signal? NO.
    Imagine that we observe an increase in bold signal in visual cortex when you imagine flowers.
    This could mean
    a) More excitatory cells are firing
    b) The same excitatory cells are firing, but they are firing more
    c) Fewer cells are responding, but they are responding much more stronger
    d) Both inhibition and excitation increase, but excitation increased more
    e) Inhibition is reduced
    f) Many inhibitory cells decrease their firing rate, but a few increased their firing rate by a lot
    h) .....
    Basically, before and after this paper, in terms of interpreting BOLD, we are where we started.

  • CHCH says:

    Lots of good points raised here; I would like to clarify however that the stimulation of inhibitory interneurons did not *really* reduce the BOLD response in any meaningful way. That is, they were imaging at a higher field strength and spatial resolution than most human studies. This is important because the volume of tissue showing BOLD increases following stimulation of interneurons was nearly three times as large as the volume of surrounding tissue showing decreases in BOLD (1.7mm^3 vs 0.6mm^3, respectively). In addition, even in this smaller volume, the observed % decrease in BOLD was less than half as large as the observed % increase in BOLD (~5% vs ~2%). So activation of inhibitory interneurons in a particular area should be expected to primarily increase the local BOLD response, both in terms of the volume of tissue affected and the magnitude of that effect.
    The effects of stimulating interneurons with long range targets would likely yield a different picture, but that's just speculation.

  • daedalus2u says:

    Bsci, you are right, Ca+ and CO2 can cause vasodilatation, but those things do not change on the time and spacial scales that changes in the fMRI BOLD signal are observe to occur in.
    You are also right that it is primarily the reduction in deoxyhemoglobin that produces the BOLD fMRI signal; deoxyhemoglobin is paramagnetic, oxyhemoglobin is diamagnetic and BOLD fMRI looks at magnetic susceptibility. However other imaging techniques (NIRS for example) show a greater increase in oxyhemoglobin than the decrease in deoxyhemoglobin.
    http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0006881
    You left out an option 4. Neurons extract less oxygen from blood, decreasing the deoxyhemoglobin level. NO inhibits cytochrome c oxidase. Cytochrome c oxidase is what takes up O2 from the blood. If cytochrome c oxidase is inhibited, it will take up less O2, and the quantity of deoxyhemoglobin will decrease and the fMRI BOLD signal will go up.
    The vasodilatation results in increased blood flow by increasing the cross section available for flow. This vasodilatation occurs upstream of the capillary beds in blood that is primarily oxygenated before the O2 has been taken out in the capillary beds. In order for there to be more flow, there has to be more volume (because the cross section has to increase). So in order for #3 to happen, #1 has to happen too, and first.
    If you look at the fMRI BOLD signal in detail, there is often a decrease in the signal before there is the increase, that is there is an apparent and slight increase in deoxyhemoglobin before there is the larger decrease that is the main signal. I suspect that first increase is due to O2 being used to make NO.
    The quantity of increased blood flow is such that the metabolic demand of the activated brain region is more than satisfied (except when things like seizures happen). This is important because then a separate mechanism to regulate blood flow so as to provide enough O2, glucose, albumin, and all the other nutrients that are required in sufficient quantities and sufficient CO2 removal capacity isn't necessary.
    The concept that the skull is rigid and so the volume of blood in the brain at any moment in time is constant is quite important. For vasodilatation to occur in one place, vasoconstriction has to occur in another. This is what causes brain damage in things like a subdural hematoma. Blood leaks out of vessels and fills up a space and that increased volume forces the volume of the rest of the brain to decrease. The only way that can happen is by the vasculature being constricted. If the vasculature volume goes down by too much, then enough blood can't flow through it.
    I think that dividing neurons into “excitatory” and “inhibitory” is too simplistic. I have zero doubt that there is lots of cross-talk between them and that many of them do some of both, probably to different extents at different times.
    I haven't seen the paper yet, but that the decrease in the signal due to activation of the inhibitory nerves is less than the increase due to activation of the excitatory nerves is what I would expect if the excitatory nerves are a source of an excitatory neurotransmitter and the inhibitory nerves are a sink of that same excitatory neurotransmitter.

  • AK says:

    I haven't dug into the paper yet, but I suspect the majority of activity in the brain regions studied is excitatory, so that the general effect of stimulating a bunch of inhibitory neurons would be to reduce overall activity. Remember that energy usage in (unmyelinated) axons is roughly proportional to length as well as firing rate. IIRC the overall proportion of excitatory axons in the cortex is much higher than inhibitory, except perhaps in layer IV in some regions.
    In normal operation, AFAIK, pyramidal cells in a specific layer tend to inhibit their neighbors via local collaterals and inhibitory neurons with short axons. When many pyramidal cells in a local area are forced to fire despite this inhibition it's hardly surprising that it would be accompanied by vasodilation. OTOH when the short-axoned interneurons are forced to fire they would tend to suppress almost all activity by pyramidal cells, reducing the need for oxygen.
    @daedalus2u

    If you look at the fMRI BOLD signal in detail, there is often a decrease in the signal before there is the increase, that is there is an apparent and slight increase in deoxyhemoglobin before there is the larger decrease that is the main signal. I suspect that first increase is due to O2 being used to make NO.

    Seems to me a more parsimonious explanation would be that oxygen takeup by mitochondria has lowered the pO2 before the NO signal can produce vasodilatation. (Not that I don't have issues with the overuse of parsimony.)

    The concept that the skull is rigid and so the volume of blood in the brain at any moment in time is constant is quite important. For vasodilatation to occur in one place, vasoconstriction has to occur in another.

    AFAIK there's a serious flaw in this notion: in addition to blood and incompressible cell matter, the brain (like the rest of the body) also contains lymph, both intracellularly and in lymph veins. Vasodilatation in one place could be balanced by lymph outflow and reduced plasma flow across capillary walls in that one place. The various flows (blood and lymph) are presumably controlled by a complex balance of pressures which is badly disturbed by large global volume changed from subdural hematoma but already evolved to handle the local volume changes from vasodilation.

  • daedalus2u says:

    This post didn't seem to go through because it had a couple of links.
    AK, you are missing the most important fact about B OLD. The correlation between the BOLD signal and neuron firing is not absolute. You can have a positive BOLD signal without nerve activity.
    (if you do a google scholar search on "Pre-emptive blood flow", the first two hits are a commentary on the third hit which shows that there can be a BOLD signal without neuronal activation. These were the two links I have taken out.)
    The vasodilatation that results in the BOLD signal comes before and is somewhat independent of neuronal activity.
    The skull is essentially rigid. There isn't any place for fluid to go except through the penetrations in the skull for nerves and blood vessels. Lymph is also incompressible. Only gases are compressible and there aren't any inside the skull. The pressure equalizes as the speed of sound. On that time scale the skull is rigid and quantities of fluid can't leave the skull through the vessels. The peed of sound in sea water is ~1500 m/s. The pressure in the skull equalizes on a time scale of 0.1/1500 ~ 70 microseconds.
    Yes, physiology has adapted to accommodate normal vasodilatation. It does so by causing vasoconstriction elsewhere inside the rigid skull. I suspect that is one role of the venous sinuses in the brain, to act as a buffer for the volume changes that directing blood flow to different regions must produce.

  • bsci says:

    daedalus2u,
    I should have written my list in comment five to account for increases or decreases in blood volume, blood flow, and oxygen metabolism. Those are still the only three things that can change that would cause a BOLD signal change. The mechanisms for how these change and interact are obviously more complex. I know that oxyhemoglobin can also change, but those changes don't directly affect the BOLD signal.
    Incidentally, the initial decrease of the BOLD signal is not reliably observed in BOLD data (though I think it is more reliably observed with optical or directly tissue oxygen sensing method). While there might just be signal magnitude issues, it is not clear that this is a universal finding.
    I'm not sure why you think NO is the only factor that can cause vasculature changes on the time-scale we observe. There's a respectably sized literature on all the other mechanisms I've listed with regards to BOLD signal. I can try to pull up a few representative links if they'd be useful.
    Your discussion on how blood volume changes interact with a rigid skull is incorrect. First, if this were the case, a large BOLD signal increase would need to be paired with a decrease in another region and that's not observed. What you are forgetting is that an increase is blood flow into and out of the brain would keep the total volume of blood in the brain the same but would increase local blood flow and alter the BOLD signal. The simplest way to demonstrate this is to have someone breathe air with an increased carbon dioxide concentration. While increased CO2 doesn't directly affect the BOLD signal, it increases heart rate and cerebral blood flow. Thus most oxyhemoglobin is getting into the veins across the entire brain and the BOLD signal globally increases.
    Finally you keep referencing the 2009 article on pre-emptive blood flow. I'm not doing an in depth critique of that here, but it's worth noting that it is a single study with data from 2 monkeys. The results might be supported by more studies in the future, but I wouldn't take such a study as ground truth until there are replications.

  • Very cool paper Sci, thanks for pointing me to it. Question -- where in the paper does it talk about infecting inhibitory neurons? I couldn't find it there, and I talked with one of the authors, who said they didn't do that in this experiment...

  • […] more complicated than that. But the basic technique is rather simple: fMRI utilizes what is called a BOLD signal (I didn’t even let you guess the acronym this time!). BOLD stands for “blood oxygen […]

Leave a Reply