Nowadays, whenever we talk about studies measuring human brain activity, it’s usually by scanning participants using functional magnetic resonance imaging (fMRI), often considered the go-to method for brain imaging. However, unlike much older and traditional methods such as implanted electrodes or encephalography (EEG), fMRI doesn’t actually measure the brain’s electrical signals directly. Instead, it picks up the brain’s blood flow. But isn’t blood flow something totally different? How is it even related to brain activity?
This was the key question back when fMRI was developed in the late 20th century. At the time, there was one intuitive theory about the nature of the fMRI signal: When brain regions are active, their neurons fire (i.e. release action potentials). When neurons in a region fire, the region will use up more oxygen than less active regions. In response, the body delivers new oxygen to the active region via fresh blood to “fuel” its firing. In summary, the idea was that regions whose neurons fire, need to receive higher amounts of oxygen, and so more oxygenated blood flows towards this brain area. The fMRI scanner then detects this increase in oxygenated blood and software can make this change visible by highlighting the area in a 3D brain image.
The problem was how to test the theory of an area’s neural firing leading to more inflow of oxygenated blood. Ideally, one would collect both fMRI data (measuring blood flow) and electrodes (measuring action potentials) to see if there is an association between the two. But there was one roadblock: MRI scanners use magnetic fields that cause massive distortions in nearby electrical devices, including electrode recordings. Because of this, measuring both signals at the same time is very difficult.
Believe it or not, this didn’t stop researchers in 2001, who built the clever device depicted below, custom engineered specifically to answer the question of how blood flow and neural activity are related. The bulky-looking pillar in the image is a vertical MRI scanner designed to have monkeys sit on a chair. The researchers presented visual stimuli to the monkeys in order to elicit neuronal firing in their visual cortex. Without going into the details of the engineering, this unique setup allowed the researchers to shield the electrodes from the magnetic distortion of the MRI scanner. Through this, the research team became the first to simultaneously measure both the neural activity and the blood flow in the same brain area.
(Logothetis Neurophysiology Lab, Max Planck Institute Tuebingen; see Logothetis et al., 2001)
The findings of this experiment were revolutionary: Opposite to the intuitive theory, the blood flow as measured by the fMRI signal didn’t seem to be caused by neurons’ action potentials. Instead, the blood flow seemed to be more closely related to something called the local field potential, which is another signal that’s picked up by electrodes just like action potentials. In very simplified terms, if action potentials measure the electrical signals going out of a neuron, then local field potentials measure the electrical signals going into a neuron. This suggested that the fMRI signal doesn’t reflect a region’s output but rather its input!
In fact, in a follow-up experiment the research team injected serotonin into the visual cortex, which has the effect of artificially reducing action potentials. They wanted to check and see if the fMRI signal would be in any way affected by less neuronal firing. As expected, although action potentials were absent, local field potentials as well as the fMRI signal remained unaffected! This showed that fMRI activity doesn’t depend on the strength of neuronal firing. Instead, it’s more closely linked to how much input a brain area receives. In short: The fMRI response is related more to input than to output!
The exact reason for this unintuitive result is still not really clear and to this day scientists are debating between different explanations for how neural processes bring about blood flow in the brain. However, regardless of the reason, this result fundamentally shaped how one should interpret fMRI data. Still, even today some neuroscientists occasionally make the mistake of equating a region’s fMRI activity with neural firing. Of course, in some cases they would be right: after all, if a neuron gets a lot of input, it’s going to sum up and bring the neuron closer to the threshold for releasing an action potential. So, the more input, the likelier it is that a neuron will eventually fire. But the important point is that we can’t say this for sure, since the fMRI signal doesn’t contain information on whether neurons actually fired or not.
An illustrative example of this is a stroke victim losing the ability to move their hand. When the patient is lying in an fMRI scanner while the affected hand is stimulated by a brush, this will cause activity in the somatosensory cortex, as expected. However, there are also cases where there is additional fMRI activity in the primary motor cortex. This is strange, because the neurons in the motor cortex are connected to muscles. If the neurons really were firing, they would surely cause movement in the hand. How can we explain this? What we see in the fMRI data is likely a result of the following process: The touch of the brush is being carried through the peripheral nervous system up into somatosensory cortex, which is connected with primary motor cortex. Hence, the signal is carried forward and slightly excites neurons in primary motor cortex. Because there is input, oxygenated blood will flow to primary motor cortex and show up as a signal in the fMRI data. However, as a result of the stroke, the neurons will not actually fire. In other words, we may observe a signal when there is input but no output.
It’s attractive to conclude from a fMRI signal that an area is producing action potentials and plays an active role in the studied process, even though it’s just passively receiving inputs. Hence, fMRI should always be interpreted with caution.
The goal of this article was to highlight the simple but crucial caveat that fMRI activity is not the same as neural firing. Nevertheless, fMRI remains an invaluable tool to neuroscience and hopefully this small journey into the past of neuroscience helped you appreciate the great lengths that researchers went to in order to study the accuracy of our measurement instruments. So, the next time you see a colourful fMRI image: remember that the bright patches don’t necessarily mean that a signal is being outputted. Like in the case of the stroke patient’s motor cortex, always check if any activity you see is in line with what we know and makes logical sense. It could always just be an artifact of inputs.
Author: Emil Stroecker
References & Further Reading
Heeger, D. J., & Ress, D. (2002). What does fMRI tell us about neuronal activity? Nature Reviews. Neuroscience, 3(2), 142–151. https://doi.org/10.1038/nrn730
Logothetis, N. K., Pauls, J., Augath, M., Trinath, T., & Oeltermann, A. (2001). Neurophysiological investigation of the basis of the fMRI signal. Nature, 412(6843), 150–157. https://doi.org/10.1038/35084005
Rauch, A., Rainer, G., & Logothetis, N. K. (2008). The effect of a serotonin-induced dissociation between spiking and perisynaptic activity on BOLD functional MRI. Proceedings of the National Academy of Sciences, 105(18), 6759–6764. https://doi.org/10.1073/pnas.0800312105
