Tuesday, June 19, 2007

My brain is not a Christmas tree

There are few things that irk me more than the way in which fMRI studies are reported in the popular press. Invariably, the article refers to parts of the brain lighting up in a Christmas-tree-like fashion. There was no Christmas tree in my house when I was growing up. There certainly isn't one in my head. More importantly, this phraseology seems to imply that the brain regions in question were:
a) not active before the experimental manipulation which triggered the festive lighting, and
b) in some sense uniformly "activated" by the stimulation.
Even the revered journal Science has adopted this misleading nomenclature in its lay publications (don't you love ecclesiastic language used to describe academia? I'm a monk of science):
Ever flinch at the sight of an actor being punched in the face? The reason is that neurons in the brain light up when we watch others suffering.

No, no, no (and not just because the Mean Monkey doesn't care if you are sad).

The first implication (conveniently denoted (a) above) is completely wrong, but in a simple way. Neural activity occurs constantly throughout the brain. Urban legends would have you believe that only some small fraction of the brain is actively used. While it is true that many neurons only fire action potentials infrequently, information is carried in their silences as well as their action potentials. If a neuron spiked without pause, it would transmit no more information than if it were constantly inactive. Moreover, at any given moment, there are neurons active in every part of the brain. Even the parts that fail to convey their wishes for a Happy Holiday in fMRI pictures. If you close your eyes, neurons continue to fire in the primary visual cortex. The primary visual cortex is even active in blind people. fMRI measures relative increases and decreases in activity; the baseline is never zero.

The second implication ((b), if you've been following along) is more pernicious, because the underlying reality is a bit more complicated than the whiz-bang notion of "lighting up." fMRI's blood-oxygen-level dependent (BOLD) signal measures the slightly counter-intuitive increase in oxygenated hemoglobin when the metabolic requirements of a brain area increase. (Presumably, this is triggered by a homeostatic mechanism which senses the increased oxygen consumption and dilates blood vessels accordingly. The brain is all tricky like that. Don't even ask how it manages to maintain a reasonable connection strength at each synapse in the face of constant potentiation and depotentiation.) This signal is best correlated with the local field potential (the low-frequency component of electrode recordings due to the average synaptic activity over a span of hundreds of micrometers), rather than the actual spiking activity of the neurons in the area. The upshot of this is that fMRI represents the inputs to a brain region, not the local activity.

That a metabolic measure signals input rather than output is reasonable from a biophysical perspective, since relatively few ions move across the axonal membrane in the process of transmitting an action potential. Most of the axon is covered by a lipid-rich myelin sheath which blocks the flow of ions and decreases the capacitance, allowing the action potential to be transmitted quickly between the gaps in the myelin (known as the nodes of Ranvier, which is also the name of a metal band of questionable quality). In contrast, neurons have giant dendritic trees which are subject to a constant barrage of neurotransmitters, most of which cause ion channels to open. When ion channels open, ions flow through them passively along their electrochemical gradient, reducing the strength of the gradient. Thus, when the amount of input increases, more energy needs to be expended to move the ions back against the gradients, hence the increased need for oxygen.

Now that I write this, I'm not entirely convinced by this justification of the coupling between input strength and oxygen utilization, since although the total ionic flow is much greater in the dendrites than the axon, it's still very small compared to the total ionic content of the neuron. You could cut out the ionic pumps and the cell would be fine for hours or days, or so I'm told, in which case there's no need to immediately increase the amount of available oxygen so the ionic pumps can be run in overdrive. However, it's possible that while the cell as a whole would not lose its overall negative charge were the ionic pumps shut off briefly, everything would go out of equilibrium in the dendritic tree. The branches of the dendritic tree are really small, so the total number of ions in a dendritic spine or branch is not very large. Even the relatively insignificant ionic flow due to synaptic activity may be enough to knock around the ionic concentrations in such small volumes.

Anyway, my point is that the BOLD signal from fMRI measures input, not local activity. And it has absolutely atrocious spacial and temporal resolution. Something on the order of millimeters and seconds. But it makes pretty pictures and lets hack science journalists tell the doe-eyed public "this part of the brain is for when you feel sad and lonely; this part of the brain is for when you feel happy." The real action is in calcium imaging, which can track single spikes (almost) from hundreds of cells at a time (but only in layer 2/3 of the cortex of anaesthetised animals), and chronic multitetrode recordings (disclosure: my old lab used this technique; tetrodes are bundles of four electrodes, which allow the isolation of dozens of cells from a single such bundle through a variety of black-magic mathematical tricks), which can record from perhaps a hundred cells for days, weeks, or months at a time (depending upon the strength of your experimental mojo). But no one wants to see pictures of comatose cats with the backs of the heads lopped off, or rats running around with bundles of wires popping out of their skulls. And the experimental results, while useful and meaningful, rarely come with a five second sound-bite. Half the time even specialists in the field aren't sure what the ultimate implication of a study is. So fMRI gets the publicity and a disproportionate share of the funding.

Which is dumb, because very little useful science has come out of fMRI. Glorified phrenology. One of the most striking known facts about the brain is that most of it looks the same. So far as anyone can tell, aside from a few small and probably insignificant differences, cortex is cortex, regardless of whether it's processing visual data or planning an arm movement or contemplating the secrets of the universe. In fact, you can rewire things visual input to the auditory areas and everything works out just fine (von Melchner, Pallas, & Sur, 2000). In ferrets. Not fruit flies. Not frogs. Visual mammals like you and me.

This would seem to suggest that the same basic computation underlies most of what the brain is doing. Wouldn't it be nice to know what this computation is? Why would you waste your time attempting to pinpoint exactly how the computation is divided spatially, when all the evidence suggests that the computation is the same everywhere? People are strange...

No comments: