Cognitive neuroscience emphasises the use of non-invasive neuroimaging methods to better understand the neural bases of behaviour, cognition, and emotion. Of the many techniques employed, functional magnetic resonance imaging (fMRI) has emerged as the most widely used within the field. Maybe because of its relatively high spatial resolution (which naturally fits with cognitive neuroscience's goal of functional localisation within the brain), or the highly intuitive and coloured visuals it produces, fMRI is a mainstay in cognitive neuroscience's methodological toolkit, with some even arguing that it has defined the field since its introduction in 1992.
How is fMRI data normally interpreted? Generally, fMRI produces coloured maps of the brain, with different colours representing different levels of 'activation' across brain regions. For a particular region, differences in activation between task conditions are interpreted as differences in neural activity, with higher activation levels corresponding to increased neural activity. For example, if two conditions produced different activation intensities in the visual cortex, it wouldn't be uncommon for scientists to conclude the conditions led to different levels of neural activity in the visual cortex. Some might even make the stronger claim that the visual cortex is responsible for whatever functional difference exists between the conditions!
But can we actually draw such conclusions and inferences from fMRI data? While the above example is admittedly crude, I believe there are simply too many constraints surrounding the use of fMRI to warrant the claims that a) increased activation means increased neural activity and b) fMRI allows us to identify functional specialisation in the brain. As a mental heuristic, I think of these constraints as belonging to three categories: physiological, statistical/analytical, and conceptual/philosophical. In this blog, I focus on the physiological constraints on fMRI interpretation, and aim to cover the latter two in future posts.
What is measured in fMRI?
While fMRI data is widely interpreted as an indicator of neural activity, it remains an indirect measure of neural activity (note, this is an uncontroversial fact recognised by neuroscientists). fMRI data, or more specifically, the blood oxygenation level-dependent (BOLD) signal, is a measure of the deoxyhaemoglobin (deoxy-HB) concentration in blood. The higher the deoxy-HB concentration, the less oxygenated blood there is and the weaker the BOLD signal. Conversely, the lower the deoxy-HB concentration (and equivalently, the higher the oxy-HB concentration), the more oxygenated blood there is and the stronger the BOLD signal. The general logic is that when neural activity occurs in a particular brain region, there is an increased demand for glucose- and oxygen-rich blood in that region. While deoxy-HB concentration initially increases due to early oxygen metabolism, a subsequent increase in cerebral blood flow (CBF) rapidly overcompensates by delivering oxygen-rich blood to that brain region. Ultimately, deoxy-HB concentration decreases, and a stronger BOLD signal is observed. Given this view of the relation between neural activity and the BOLD signal, it is unsurprising that cognitive neuroscientists tend to view fMRI data as a surrogate for the amount and location of neural activity in the brain.
Physiological constraints on fMRI interpretation
The issue with this interpretation is that it takes neural activity as the primary (or worse, the only) contributor to changes in BOLD signal. This could not be further from the truth. To begin, let's consider the received view that neural activity drives changes in BOLD signal via increased CBF due to an increased demand for energy. Ideally, we'd want to see a proportional coupling between changes in neural activity, energy consumption, and CBF. This gives us confidence that the magnitude of BOLD change actually reflects that of changes in neural activity. That this coupling relationship is constant across brain regions allows for between-region comparisons of the amount of neural activity involved. Unfortunately, studies using positron emission tomography (PET) in human subjects have found differential couplings between CBF and cerebral metabolic rates for glucose (CMRglu) across the brain -- whereas some regions were hyperperfused (i.e., greater CBF relative to CMRglu), other regions were hypoperfused (i.e., reduced CBF relative to CMRglu) (Huettel et al., 2014). Assuming that aerobic (i.e., involving oxygen) metabolism is the dominant process by which ATP (i.e., the primary source of energy for cells) is derived from glucose, we might go on to infer that brain regions differ too in their coupling between CBF and oxygen metabolism. The immediate implication, then, is that observing a difference between brain regions in the magnitude of BOLD signal change does not allow us to automatically conclude that one brain region produced a larger increase in neural activity than the other brain region as a result of an experimental manipulation.
But evidence suggests that aerobic metabolism might not be the primary mechanism through which glucose is converted to ATP. For example, Fox et al. (1988) compared CBF, CMRglu, and cerebral metabolic rate for oxygen (CMRO2) during a visual stimulation task and found that while CBF and CMRglu increased by around 50%, CMRO2 only experienced a modest 5% increase. This strongly implicated anaerobic (i.e., not involving oxygen) metabolism, which prompted researchers to look for anaerobic-dominant processes in the brain that might accompany neural activity. This led to the astrocyte-neuron lactate shuttle hypothesis (Takano et al., 2006), which argues that the increased vasodilation that accompanies neural activity is a result not of increased oxygen or energy demands by neurons, but rather due to the cascade of neurotransmitter activity in the synaptic cleft that ultimately leads to the production of vasodilators in non-neuronal cells called astrocytes. Because CBF increases irrespective of any change in local oxygen concentration, the magnitude of the BOLD signal recorded is heavily influenced by the baseline oxy-Hb level in a particular region. For instance, veins have a lower baseline oxy-Hb level than arterioles and capillaries. Consequently, an increase in oxy-Hb-rich CBF leads to a disproportionately larger increase in oxy-Hb concentration (and BOLD signal) in the draining veins situated further downstream from the site of neural activation than in the capillaries at the site of activation. For all the praise fMRI receives for its high spatial resolution, this "great brain vs. vein debate" (Menon, 2012) places a nontrivial constraint on the spatial accuracy of fMRI data.
Regardless of the actual mechanism linking neural activity to increased CBF, there is a general consensus that the BOLD signal is correlated more with the postsynaptic potentials that reflect input from presynaptic neurons than with the firing of action potentials that reflect output from postsynaptic neurons (Logothetis et al., 2001). This suggests that fMRI is a measure of the integration between glutamate-driven excitatory postsynaptic potentials (EPSP) and GABA-driven inhibitory postsynaptic potentials (IPSP). It is important to note that a single EPSP or IPSP is a transient and local event that only temporarily changes the electrical potential of a neuron. On its own, the neuron's resting potential is quickly restored with the help of sodium-potassium pumps. However, when spatially and temporally combined with other postsynaptic potentials, the electrical potential of the neuron might exceed a certain threshold that triggers the firing of an action potential down the neuron's axonal fibre (i.e., an excitatory effect). On the other hand, the combination of multiple EPSPs and IPSPs might also hyperpolarise the neuron (i.e., an inhibitory effect), or lead to graded potentials that neither excite nor inhibit the postsynaptic neuron (i.e., a sub-modulatory effect). What the above illustrates is that, for a given BOLD activation observed (and assuming that it really does reflect neural activity), we don't actually know what kind of neural activity is on display. It might be an excitatory, inhibitory, or sub-modulatory effect (Page, 2006). Yet, I think the field of cognitive neuroscience is generally guilty of taking the claim of "increased fMRI activation reflects increased neural activity" as "increased fMRI activation reflects increased excitatory neural activity". But this is clearly not justified given the aforementioned physiological consideration.
A final point relates to the vascular system of the brain. Vascular richness can differ considerably across brain regions. For example, cortical areas typically have richer vasculature and greater microvascular density than subcortical regions. Additionally, similar to the earlier point on coupling, neurovascular coupling (i.e., the relationship between neural activity and CBF changes) can differ across populations, task conditions, and even within the same individual (Newman, 2019). The general point, then, is simply that differences in BOLD activation might not purely reflect changes in neural activity, but rather inter- and intra-group differences in vasculature and neurovascular coupling.
Concluding thoughts
Given the physiological constraints outlined, is fMRI doomed? Well, maybe not. There are many factors that mitigate the concerns above. fMRI experiments typically employ voxels 3 x 3 x 3 mm in size, which are typically large enough to accommodate both capillaries and draining veins, so the charge of spatial inaccuracy might end up being negligible after all. While we have acknowledged the ambiguity of the type of neural activity reflected in postsynaptic potentials, glutamate remains the most common neurotransmitter of the brain and is estimated to be released at around 90% of synapses, which might provide support to the view that BOLD activation likely reflects excitatory neural activity. Many methodological and statistical techniques also aim to mitigate the variability in haemodynamic response and the effect of baseline oxygen levels, including the use of model convolution and calibrated BOLD fMRI. Besides, despite these physiological constraints, fMRI research seems to corroborate results found from neuropsychology and other neuroimaging modalities.
On the sole consideration of the physiological constraints of fMRI interpretation, I don't believe that fMRI has nothing to contribute. The main takeaway is that the conclusions drawn from the BOLD signal should be more reserved and conservative than those currently made by neuroscientists. Who knows, with better experimental control and a clearer understanding of the physiological basis of the BOLD signal, one might posit that this might allow us to revert back to the rather strong claims made about brain activity based on fMRI data. Yet, I think this is still premature. Data interpretation is never theory-free. This should be clearer when we begin considering the conceptual and philosophical constraints on fMRI interpretation.
References
Fox, P. T., Raichle, M. E., Mintun, M. A., & Dence, C. (1988). Nonoxidative glucose consumption during focal physiologic neural activity. Science, 241(4864), 462–464. https://doi.org/10.1126/science.3260686
Huettel, S. A., Song, A. W., & McCarthy, G. (2014). Functional Magnetic Resonance Imaging. Sinauer Associates Incorporated.
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
Menon, R. S. (2012). The great brain versus vein debate. NeuroImage, 62(2), 970–974. https://doi.org/10.1016/j.neuroimage.2011.09.005
Newman, A. (2019). Research methods for Cognitive Neuroscience. SAGE Publications Limited.
Page, M. P. (2006). What can’t functional neuroimaging tell the cognitive psychologist? Cortex, 42(3), 428–443. https://doi.org/10.1016/s0010-9452(08)70375-7
Takano, T., Tian, G., Peng, W., Lou, N., Libionka, W., Han, X., & Nedergaard, M. (2006). Astrocyte-mediated control of cerebral blood flow. Nature Neuroscience, 9(2), 260–267. https://doi.org/10.1038/nn1623
No comments:
Post a Comment