Brain Activation: Does 2 + 2 = 4?

By Neuroskeptic | March 26, 2013 2:36 pm

An interesting Journal of Neuroscience paper just out argues that Spontaneous and Task-Evoked Brain Activity Negatively Interact.

If true, this could be explosive, because a lot of neuroscience is built on the assumption that those two things don’t interact. So what’s going on?

We know that the brain is active all of the time. Even if you’re not doing anything in particular, there is ‘spontaneous’ (or resting-state) activation, which varies over time. But what if something in particular does happen, causing brain activation? Is that evoked activity a constant that simply gets added on to the pre-existing activity, or do they interact in some more complex way?

The new paper’s author, Biyu He, argues as follows. If evoked activity is independent of spontaneous activity and they sum additively, then the variance of the brain activation ought to increase after a stimulus. That’s because whatever variability was there spontaneously would still be there, and then the variability of the stimulus-evoked activity would be added on to that. Even if the evoked response was completely fixed (zero variance), the total signal variance would stay the same.

But is that how the brain works? He considered data from a simple fMRI scanning study, in which participants had to press a button in response to a simple visual stimulus. Essentially, He claims that variance decreases following stimulus presentation. This just can’t happen, if evoked and spontaneous activity sum additively. Rather, these data imply that they interact.

Here’s the key graphs. After the onset of a stimulus (time=0), the signal (averaged over all trials) increases in most of the brain regions of interest; each coloured line is a region, and the black line’s the average. Nothing new there. But He shows that the variance of the signal across all trials decreases at the same time. This is a new finding (I think) and pretty counterintuitive.

What’s more, areas (voxels) that show a larger increase in signal, also show a larger decrease in variance:


If it’s real, non-additivity would be a very interesting phenomena in itself. But it could also have big implications, because it would undermine the assumptions behind conventional fMRI data analysis. Conventional analysis averages across all of the events of interest (e.g. every time a stimulus appears) and looks for areas where activity increases, or decreases, on average.

However, He’s results raise the possibility that the same stimulus could cause both increases and decreases, depending on the spontaneous activity baseline preceding each event. Average them all together, and they might cancel out, so you couldn’t see them. Conventional fMRI studies might, in this way, be missing interesting activations.

This is very interesting stuff, and I wonder what it means. Suppose that when there’s no stimulus, brain activity is dominated by internally generated events (day-dreams, memories, whatever). These are highly variable, because at any given moment, you could be thinking about anything. But then, at the moment an important external event occurs, the brain’s activity becomes dominated by that.

In which case, is it any surprise that activity is less variable just at the time of an external event?

ResearchBlogging.orgHe BJ (2013). Spontaneous and task-evoked brain activity negatively interact. The Journal of Neuroscience, 33 (11), 4672-82 PMID: 23486941

CATEGORIZED UNDER: fMRI, methods, papers, select, Top Posts
  • anonymus

    check out

    Stimulus onset quenches neural variability: a widespread cortical phenomenonMark M Churchland et al.
    Nature Neuroscience 2010

    • Neuroskeptic

      Thanks. It was referenced in the paper in fact. However, the Churchland study was about intracellular and extracellular recordings i.e. it was about small numbers of neurons; for the variance-quenching to hold on such a large scale that it shows up on fMRI is another thing entirely (hence I suppose why He got a J Neurosci paper.)

      • Grace Lindsay

        I immediately thought of the Churchland et al (et al et al et al…) results as well. I understand that those results don’t necessarily predict these, but is it really “another thing entirely”? If cells all across cortex are reducing their variability, and fMRI is in some way measuring neural activity on a large scale, then it seems reasonable to expect a decrease in BOLD variability as well. I think the huge difference in timescales between ephys and fMRI data is what makes this finding something else entirely, but not so much the difference in the population size.

  • DS

    For a long time (on the fMRI time-scale) it has been known that BOLD signals do not simply add. You can take a look at early literature on the visual cortex in which responses of the visual cortex to two stimuli separately and together were studied. So not seeing linearity of response is nothing new.

    But even for the visual cortex there is some loosely defined experimental parameter space in which linearity of response is approximately true. So what this paper is adding to fMRI science (if the results hold) is the elucidation of the parameter space in which resting state and task based BOLD has a linear response – and, with this one study as guide, presently that space appears empty.

    I had a quick look at the methods section of the paper. I see that a number of things are done with the data that make me suspicious. For example slice-timing correction was applied as well as motion correction (realignment). There are problems with separating these two operations. Also, although I in general put little stock in retrospective rigid body motion measurement, it would be nice to see the motion data or some summary of it so that one could be somewhat confident that motion and variance was not, unintuitively, being suppressed by the task performance.

  • disqus_seUDH2vULq


    this reduction should be greater in early visual areas, right?

  • Bradley Voytek

    Just for clarification, Biyu He is female.

    She also has one of the best Acknowledgement sections of any paper I’ve seen:

    “B.J.H. would also like to thank the U.S. Immigration Service under the Bush administration, whose visa background security check forced her to spend two months (following an international conference) in a third country, free of routine obligations—it was during this time that the hypothesis presented herein was initially conjectured.”

    * He, B. J. & Raichle, M. E. The fMRI signal, slow cortical potential and consciousness. Trends in Cognitive Sciences 13, 302–309 (2009).

    • Neuroskeptic

      He’s she. Got it.

      Luckily, my post is gender neutral – because all of the “he”‘s are actually “He”‘s.

      • Bradley Voytek

        Ha! Language. “He considered data from…” Even when I *know* what is said, I have a very difficult time parsing it correctly.

  • mua ve xem phao hoa da nang

    thank you admin very much, information is very good

  • Aaron Kucyi

    Interesting paper. I agree though that maybe it’s no surprise that activity is less variable during controlled external events. Would have been cool to see if this trial-to-trial variability predicted reaction time for the button press.

  • Jeff Stout

    I am wondering if this statement is accurate: “Essentially, He claims that variance decreases following stimulus presentation. This just can’t happen, if evoked and spontaneous activity sum additively. Rather, these data imply that they interact.” Evoked and spontaneous brain activity still sum additively, it just means that they are not independent (covariance is equal to zero). The conclusion of the third sentence remains correct.

  • Timothée Masquelier

    I couldn’t agree more with your last sentence:
    Is it really surprising that sensory neurons are less variable when involved in sensory processing of a controlled stimulus, than when we do not know nor control what they are doing (daydreaming, etc.)?

    To push the argument further: does it really make sense to pool together data points in which nothing is controlled and compute mean/variance?
    I have developed these ideas in this recent opinion paper:

    Also, is it really surprising that evoked and spontaneous activity interact, given that they (presumably) involve the same neurons?

    Your blog is very thought-provoking, thanks a lot!



No brain. No gain.

About Neuroskeptic

Neuroskeptic is a British neuroscientist who takes a skeptical look at his own field, and beyond. His blog offers a look at the latest developments in neuroscience, psychiatry and psychology through a critical lens.


See More

@Neuro_Skeptic on Twitter


Discover's Newsletter

Sign up to get the latest science news delivered weekly right to your inbox!

Collapse bottom bar