The blood that flows into our heads is obviously important for it provides nutrients and oxygen to that most energetically demanding of organs - the brain. But for neuroscientists, blood flow in the brain has a special significance; many have used it to measure brain activity using a technique called functional magnetic resonance imaging, or fMRI.
This scanning technology has become a common feature of modern neuroscience studies, where it's used to follow firing neurons and to identify parts of the brain that are active during common mental tasks. Its use rests on the assumption that the flow of blood ("haemodynamics" to those in the know) is a decent enough stand-in for the firing of neurons - the latter creates a shortage of nutrients and oxygen that is corrected by the former.
But Yevgeniy Sirotin and Aniruddha Das from Columbia University have found that this assumption might not be entirely valid. They used a new technique to independently measure and compare nerve activity and blood flow in the brains of live monkeys. Sure enough, they found a blood flow pattern that reliably matched the activity of the animals' neurons.
But they also spotted something that no one has seen before - a second haemodynamic signal, of equal strength to the first, that didn't correspond to any local brain activity. This second signal was not a sign of parts of the brain that are active, but those that may need to be active in the near future. It seems that if the brain expects a task in the future, it can anticipate which of its regions will be needed and flush them with blood in preparation.
Sirotin and Das worked with two rhesus monkeys that had transparent windows in their skulls just over their visual cortex, a part of the brain involved in processing images. With a camera pointed through this pane, the duo could literally watched the blood pumping through the monkeys' brains. Using two different wavelengths of light - one red and one green - they could work out how much blood was flowing through a specific blood vessel, and how much oxygen it carried. And with tiny electrodes, they could simultaneously record the electrical signals from nearby neurons.
The monkeys were oblivious to all this fuss - they were out to get a sip of juice. To do that, they had been trained to look at a small spot on a computer screen - when it shone in one colour, the monkeys had to focus their attention on it and when it switched to a second colour, they had to relax their eyes. They sat through a continuous series of these trials and soon picked up the recurring fixate-and-relax pattern.
It was during this task that Sirotin and Das noticed the strange second signal, so they tried to isolate it. They put their monkeys through the same exercise, but this time in almost total darkness. The tiny dot they had to look at was very faint and as it swapped between the "fixate" and "relax" colours, it would have looked like a "single, twinkling star in an otherwise black sky".
In this nigh-pitch blackness, the monkeys' nerves were silent. But their haemodynamic signals spoke volumes - they were still rising and falling in a steady rhythm even though there wasn't any nerve activity in the same area. The monkeys' pupils dilated in time with this signal, their arteries narrowed and widened, and their heart rate kept pace too. Their nerves? Nothing, save a steady background hum.
What was behind this mystery signal? It certainly kept the same timing as the alternating dot and sceptics might point out that there was some light, albeit very little. But different parts of the visual cortex respond to different parts of a monkey's field of view. And with the dot appearing only in the centre of their vision, Sirotin and Das could point their cameras at an area that they knew wouldn't respond to it. And indeed, the local nerves showed no sign of activity above their background levels.
Perhaps the signal was the result of some internal cycle? Unlikely - when Sirontin and Das changed the timing of the flickering dot, they found that the signal followed suit. As the length of each trial increased from six seconds to thirty, so the rhythm of the signal stretched to match it.
It's tempting to think that the signal represented the monkey's shifting attention, with every peak signifying blood flowing to the area during fixation and every trough corresponding to relaxation. But the signal's timing said otherwise - it showed that blood was starting to flow into the area before the start of each trial period, while the monkey was meant to be relaxing its gaze.
This new signal seemed to be pre-empting the monkey's actions. To confirm that, Sirontin and Das changed the timing of the trials after 10-20 cycles, when the monkeys had got into a rhythm. The animals quickly noticed the new pace and immediately picked it up. But the strange second signal was slower - it took a couple of rounds to adjust to the new tempo. It was still "anticipating" the previous timing, even though the animal itself had moved on.
Based on all of these observations, Sirontin and Das suggest that some higher part of the brain anticipates the demands of other regions and sends advance supplies of fresh blood to fuel the neural activity that it foresees. The exact mechanism still needs to be discovered.
For now, the study has an immediate and serious impact on the way that neuroscientists interpret the results of fMRI scans. It's a technique that is already facing a fair amount of controversy, from its technical limitations, to the way its results are analysed, to its popular facade as a mind-reading technology. These new results will surely only inflame the debate further.
Interpretations of fMRI experiments hinge on the idea that haemodynamic signals can predict the activity of neurons in specific parts of the brain. This new study shows that this is true to an extent. But it also reveals the existence of another group of signals that is just as strong and has absolutely nothing to do with local neurons.
Recent reports have suggested that the link between blood flow and neural activity is far from straightforward, but even allowing for that Sirontin and Das's results are something else. They're sure to cause a hefty amount of neural activity in the brains of the world's neuroscientists.
On a tangential and amusing note: The authors made me chuckle. When I googled Sirontin, the fourth link is this amusing video. And the paper makes it seem that Das is a member of every research department at Columbia (and some outside of it).
Reference: Yevgeniy B. Sirotin, Aniruddha Das (2009). Anticipatory haemodynamic signals in sensory cortex not predicted by local neuronal activity Nature, 457 (7228), 475-479 DOI: 10.1038/nature07664
More fMRI studies:
- Computer predicts brain activity associated with different objects
- The machine that identifies images from brain activity alone
- Why do people overbid in auctions?
- Unconscious brain activity shapes our decisions
- 'Talking face' simulations in the brain help us work out what's being said
I'd say that it's more accurate to say that Sirotin and Das have found the first case where neuronal activity and vascular response can be decoupled, and while it is not quite clear on how far reaching this finding is, that it probably doesn't effect most fMRI experiments. This finding fits more in the 'exception' category than the 'rule' category.
First off there's Logothetis' 2001 paper showing that in at least some other circumstances, BOLD is well correlated with local neural activity. This says that in at least some situations, fMRI analysis is perfectly valid.
Second is an fMRI method known as catch-design, I believe first described by Ollinger in 2001. With this method, you essentially do something very similar to what Sirotin and Das did: On some trials, maybe 30%, you end the trial early. The aim is to be able to differentiate the activity at the begining of the trial from the activity at the end of the trial. If Sirotin and Das' findings held over a wide array of fMRI studies, then this method would never work because the anticipatory vascular activity from the missing second half of the trial would look exactly like the vascular activity had the rest of the trial actually been presented.
However, I'm sure a lot of people will still think this invalidates all fMRI studies ever done, and that there will be active scientific study into exactly how far this finding stretches. I'd be interested in someone replicating Logothetis' 2001 paper's methodology with Sirotin's experimental design, so that we could get a direct measure of LFP and BOLD rather than optical imaging.
Also interesting is to see how regular and reliable the stimuli had to be to prompt this behavior. These monkeys were trained probably for at least 6 months on a single task. That type of repetitive training could exaggerate what would be a modest effect in a naive subject coming into an fMRI experiment. Or maybe even alternating trials between two different types would be enough to erase the effect, or maybe pseudo randomized trials would be required.
I hope people view this as the very beginnings of a line of scientific inquiry rather than another nail in the coffin of fMRI.
Thanks for the thoughtful analysis Kevin - I would certainly agree that this result doesn't wantonly invalidate previous fMRI studies, but it does raise concerns. Like you, I see it as the starting pistol for more work.
There are lots of simultaneous fMRI/EEG studies in humans. I wonder if there's any evidence of this "anticipatory" hemodynamic response in those studies.
I agree with Kevin, it doesn't necessarily invalidate previous fMRI studies, but it makes scrutinizing study design even more important.
Im thinking way back to my undergrad biology days so could certainly be wrong but i thought that counterintuitively, it was darkness that caused firing of the optic nerve neurons, whereas light prevented them from firing (something to do with the retinal cells and how they responded to the isomerisation of light-sensitive pigment).
I read your post and skimmed the article. I mostly agree with Kevin that this is more of an interesting exception case than something that tells us about typical fMRI studies. There is a growing literature of interesting setups that cause neural/hemodynamic uncoupling, but they aren't done like typical fMRI studies. In addition, this paper uses both neural spikes (action potentials) and local field potentials (LFPs). Figure 2 seems to show a weak LFP change and no spiking change. This is known to cause blood flow changes, though the magnitude of those changes is interesting. The fact that pupils change means that something is happening neurally (and probably with at least some visual cortex response).
On thing I couldn't locate in the article or supplemental was the location of the electrodes compared to the optical imaging. It sounds like they used a single depth electrode can compared it to the surface optical imaging. I'd need to know before about the locations before buying the results.
Finally, one finding you highlight here is fairly unimpressive. The fact that blood flow changes before task is rather old. When stimuli appear at a fixed rate, there is a predictive fMRI response (and electrical response too). Some of the early fMRI papers used fixed rates and you can see the response timing increase before the stimulus presentations. Good studies now have some timing jitters to remove the predictive problems (these jitters are used in other methods too like some EEG and MEG studies).
Although the idea of MRI's being inaccurate is pretty far reaching, I think the bit I'm most interested in here is about the brain anticipating which part may be needed, if in fact thats whats going on. I'd love to see more about that!
That said though, I think someone needs to inform Doctor House about this. I've played a drinking game where you have to finish your beer everytime you hear 'Get me an MRI!' and barely walked away from it
"When I googled Sirontin" - probably you spelled it correctly as "Sirotin" if you found all those links!
Fantastic report. Congratulations for this site!