Neurophysiology: The Brain at Work. Friston, K. Current Biology, 18(10):R418--R420, May, 2008.
Neurophysiology: The Brain at Work [link]Paper  doi  abstract   bibtex   
aWellcome Trust for Neuroimaging, Institute of Neurology, UCL, 12 Queen Square, London WC1N 3BG. UK A remarkable new study of conjoint electrophysiological and metabolic brain signals is the first to analyse trial-by-trial responses in awake-behaving primates and furnishes crucial constraints on the interpretation of functional brain mapping data. The basis of brain imaging signals has always been a preoccupation of imaging neuroscience and has been expressed in many guises. Initially, the focus was on quantitative metabolism and stoichiometrics, as exemplified by the pioneering work of Sokoloff [7]. With the advent of positron emission tomography (PET) and the opportunity to measure cerebral blood flow non-invasively, stoichiometric analyses focussed on the apparent uncoupling between blood flow and oxygen metabolism [8]; it seemed that the brain does not use all the oxygen delivered by increases in blood flow. This uncoupling is the basis of the blood-oxygenation-level-dependent (BOLD) signal in fMRI, established in the early 1990s [1]. Around this time, there was a shift in focus to distal mechanisms generating fMRI signals, such as balloon models [2] and [3] that emphasised hemodynamics per se. The past few years have seen a return to the stoichiometric analysis but finessed in terms of energy budgets that can be attributed to specific aspects of neuronal activity [9]. At the same time, a complementary approach [re]emerged by combining hemodynamic and electrophysiological measurements. This progressed at two scales; first, the study of correlations between non-invasive electroencephalographic (EEG) and fMRI signals that has been championed by epilepsy researchers [10]. The second was at a microscopic scale [11] and [12]. Logothetis and colleagues [11] led the way in correlating local field potentials and multi-unit activity with conjoint fMRI signals. This work was technically breathtaking in its sophistication and proficiency, and has provided some of the clearest insights into the link between neuronal dynamics and brain imaging signals to date. The [simplified] picture that emerges from the above research is that vasodilatory signals are elaborated in response to pre-synaptic release of neurotransmitters (principally glutamate). Their post-synaptic targets include glial cells, the secondary-messenger systems of which orchestrate the synthesis of signals (such as nitric oxide) that cause arteriolar muscle to relax and initiate well-characterised hemodynamics. Goense and Logothetis [6] suggest that the concomitant post-synaptic responses of neurons are fluctuations in transmembrane potential which give rise to local field potentials. Critically, these neuronal dynamics have characteristic frequencies in a range that corresponds to the time-constants of slow currents, mediating things like after-hyperpolarisation. This is important because these slower (10 to 100 milliseconds) post-synaptic dynamics are moderated by classical modulatory neurotransmitters and coincide with the longer time-constants associated with (nonlinear) NMDA receptors. Put simply, two streams of cellular processes are initiated by pre-synaptic glutamate release; the first (predominantly in glial cells) gives rise to the BOLD signal, while the second (in neurons) engenders local field potentials. This picture is supported by the new results [6] because it is the local field potentials and not multi-unit activity that predicts the BOLD signal. Goense and Logothetis [6] exploit the fact that multi-unit activity reflects action potentials in large cells (such as pyramidal or principal cells) to reach a fundamental conclusion: BOLD signals reflect the sequelae of pre-synaptic activity, not post-synaptic firing. This is important because one cannot interpret the results of a brain mapping activation in terms of the firing of principal or output neurons. In other words, one is not looking at the results of a computation or the neuronal code per se but at the somato-dendritic computations that furnish that representation. One can equate the firing of principal cells with a neural code, in the sense that these are the only signals that are transmitted to, or accessed by, other neurons. This means the BOLD signal is not a correlate of neuronal encoding; rather it reflects the effects of pre-synaptic activity, from which the code is elaborated. Goense and Logothetis [6] are careful to point out that this is not necessarily the input from another population or area, because pre-synaptic terminals are found on both extrinsic and intrinsic (or recurrent) connections from the same area. This conclusion is important and may explain why fMRI can see subtle effects normally associated with lateral or top-down inputs that elude single-unit electrode recording studies [13]. The work reported by Goense and Logothetis [6] is unique because it is the first conjoint recording study in awake-behaving monkeys that has examined trial-by-trial responses. It is also the first time that these data have been analysed to disclose the unique contributions of different local field potential frequencies. These analyses are based upon convolution models of the sort used to model conventional fMRI data in humans, but the stimulus functions (that usually encode surrogates for neural activity, such as stimulus presentation) comprised the power in different frequencies. Critically, this enabled the authors to examine the contribution of different frequencies, while explaining away the effects of others. Remarkably, their results suggest that all frequencies explain a significant part of the BOLD response. Having said this, the most potent frequency lies in the low-gamma range, in which the slow post-synaptic somato-dendritic responses to input are expressed [6]. This is important from a number of perspectives, particularly for those involved in EEG research. First, if it is true that the BOLD signal and local field potentials have the same cause (pre-synaptic activity), then there might be closer correspondence between EEG (a summation of local field potentials) and fMRI than previously thought. Second, these results may inform the study of frequency-specific correlates of fMRI signals in induced response, sleep and epilepsy research. The questions here are not about the relationship between BOLD signals and the neural code, but their relationship to macroscopic oscillations and transients. Here the questions pertain to the population dynamics, as opposed to the microscopic activity of single neurons. In this context, the findings of Goense and Logothetis [6] may substantiate the longstanding conjecture that faster (de-synchronised) EEG activity is associated with neuronal activation: Studies of EEG and fMRI suggest an increase in EEG frequency is associated with an increase in BOLD signal, at least in the cortical sources [14] and [15]. In other words, low-frequency (alpha) activity is usually associated with smaller fMRI signals, whereas fast (gamma) activity is associated with a hemodynamic activation. This relationship has been studied using computational modelling [16], which suggests the mechanism is quite simple; increased pre-synaptic activity opens ion channels, decreases effective membrane time-constants and reduces the integration time of post-synaptic neurons. This leads to faster oscillatory activity and an increase in the frequency of population activity. This is a nice example of circular causality, where the microscopic post-synaptic processes are enslaved by macroscopic variables (mean activity), which links the microscopic scale to observable macroscopic quantities. The precise form of the frequency dependency of BOLD signal reported in [1] may provide important constraints on synaptic time-constants and intrinsic connectivity of microcircuits [17] that support this self-organised behaviour.
@article{friston_neurophysiology:_2008,
	title = {Neurophysiology: {The} {Brain} at {Work}},
	volume = {18},
	shorttitle = {Neurophysiology},
	url = {http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6VRT-4SJ91G8-G&_user=5939061&_rdoc=1&_fmt=&_orig=search&_sort=d&view=c&_acct=C000009959&_version=1&_urlVersion=0&_userid=5939061&md5=9201aba4581a7300dbc3fb426cbb3e3d},
	doi = {10.1016/j.cub.2008.03.042},
	abstract = {aWellcome Trust for Neuroimaging, Institute of Neurology, UCL, 12 Queen Square, London WC1N 3BG. UK A remarkable new study of conjoint electrophysiological and metabolic brain signals is the first to analyse trial-by-trial responses in awake-behaving primates and furnishes crucial constraints on the interpretation of functional brain mapping data.   The basis of brain imaging signals has always been a preoccupation of imaging neuroscience and has been expressed in many guises. Initially, the focus was on quantitative metabolism and stoichiometrics, as exemplified by the pioneering work of Sokoloff [7]. With the advent of positron emission tomography (PET) and the opportunity to measure cerebral blood flow non-invasively, stoichiometric analyses focussed on the apparent uncoupling between blood flow and oxygen metabolism [8]; it seemed that the brain does not use all the oxygen delivered by increases in blood flow. This uncoupling is the basis of the blood-oxygenation-level-dependent (BOLD) signal in fMRI, established in the early 1990s [1]. Around this time, there was a shift in focus to distal mechanisms generating fMRI signals, such as balloon models [2] and [3] that emphasised hemodynamics per se. The past few years have seen a return to the stoichiometric analysis but finessed in terms of energy budgets that can be attributed to specific aspects of neuronal activity [9]. At the same time, a complementary approach [re]emerged by combining hemodynamic and electrophysiological measurements. This progressed at two scales; first, the study of correlations between non-invasive electroencephalographic (EEG) and fMRI signals that has been championed by epilepsy researchers [10]. The second was at a microscopic scale [11] and [12]. Logothetis and colleagues [11] led the way in correlating local field potentials and multi-unit activity with conjoint fMRI signals. This work was technically breathtaking in its sophistication and proficiency, and has provided some of the clearest insights into the link between neuronal dynamics and brain imaging signals to date. The [simplified] picture that emerges from the above research is that vasodilatory signals are elaborated in response to pre-synaptic release of neurotransmitters (principally glutamate). Their post-synaptic targets include glial cells, the secondary-messenger systems of which orchestrate the synthesis of signals (such as nitric oxide) that cause arteriolar muscle to relax and initiate well-characterised hemodynamics. Goense and Logothetis [6] suggest that the concomitant post-synaptic responses of neurons are fluctuations in transmembrane potential which give rise to local field potentials. Critically, these neuronal dynamics have characteristic frequencies in a range that corresponds to the time-constants of slow currents, mediating things like after-hyperpolarisation. This is important because these slower (10 to 100 milliseconds) post-synaptic dynamics are moderated by classical modulatory neurotransmitters and coincide with the longer time-constants associated with (nonlinear) NMDA receptors. Put simply, two streams of cellular processes are initiated by pre-synaptic glutamate release; the first (predominantly in glial cells) gives rise to the BOLD signal, while the second (in neurons) engenders local field potentials. This picture is supported by the new results [6] because it is the local field potentials and not multi-unit activity that predicts the BOLD signal. Goense and Logothetis [6] exploit the fact that multi-unit activity reflects action potentials in large cells (such as pyramidal or principal cells) to reach a fundamental conclusion: BOLD signals reflect the sequelae of pre-synaptic activity, not post-synaptic firing. This is important because one cannot interpret the results of a brain mapping activation in terms of the firing of principal or output neurons. In other words, one is not looking at the results of a computation or the neuronal code per se but at the somato-dendritic computations that furnish that representation. One can equate the firing of principal cells with a neural code, in the sense that these are the only signals that are transmitted to, or accessed by, other neurons. This means the BOLD signal is not a correlate of neuronal encoding; rather it reflects the effects of pre-synaptic activity, from which the code is elaborated. Goense and Logothetis [6] are careful to point out that this is not necessarily the input from another population or area, because pre-synaptic terminals are found on both extrinsic and intrinsic (or recurrent) connections from the same area. This conclusion is important and may explain why fMRI can see subtle effects normally associated with lateral or top-down inputs that elude single-unit electrode recording studies [13]. The work reported by Goense and Logothetis [6] is unique because it is the first conjoint recording study in awake-behaving monkeys that has examined trial-by-trial responses. It is also the first time that these data have been analysed to disclose the unique contributions of different local field potential frequencies. These analyses are based upon convolution models of the sort used to model conventional fMRI data in humans, but the stimulus functions (that usually encode surrogates for neural activity, such as stimulus presentation) comprised the power in different frequencies. Critically, this enabled the authors to examine the contribution of different frequencies, while explaining away the effects of others. Remarkably, their results suggest that all frequencies explain a significant part of the BOLD response. Having said this, the most potent frequency lies in the low-gamma range, in which the slow post-synaptic somato-dendritic responses to input are expressed [6]. This is important from a number of perspectives, particularly for those involved in EEG research. First, if it is true that the BOLD signal and local field potentials have the same cause (pre-synaptic activity), then there might be closer correspondence between EEG (a summation of local field potentials) and fMRI than previously thought. Second, these results may inform the study of frequency-specific correlates of fMRI signals in induced response, sleep and epilepsy research. The questions here are not about the relationship between BOLD signals and the neural code, but their relationship to macroscopic oscillations and transients. Here the questions pertain to the population dynamics, as opposed to the microscopic activity of single neurons. In this context, the findings of Goense and Logothetis [6] may substantiate the longstanding conjecture that faster (de-synchronised) EEG activity is associated with neuronal activation: Studies of EEG and fMRI suggest an increase in EEG frequency is associated with an increase in BOLD signal, at least in the cortical sources [14] and [15]. In other words, low-frequency (alpha) activity is usually associated with smaller fMRI signals, whereas fast (gamma) activity is associated with a hemodynamic activation. This relationship has been studied using computational modelling [16], which suggests the mechanism is quite simple; increased pre-synaptic activity opens ion channels, decreases effective membrane time-constants and reduces the integration time of post-synaptic neurons. This leads to faster oscillatory activity and an increase in the frequency of population activity. This is a nice example of circular causality, where the microscopic post-synaptic processes are enslaved by macroscopic variables (mean activity), which links the microscopic scale to observable macroscopic quantities. The precise form of the frequency dependency of BOLD signal reported in [1] may provide important constraints on synaptic time-constants and intrinsic connectivity of microcircuits [17] that support this self-organised behaviour.},
	number = {10},
	urldate = {2009-01-13},
	journal = {Current Biology},
	author = {Friston, Karl},
	month = may,
	year = {2008},
	keywords = {balloonmodel, haemodynamics, metabolism},
	pages = {R418--R420},
	file = {friston2008.pdf:/Users/nickb/Zotero/storage/HEWK65T9/friston2008.pdf:application/pdf;ScienceDirect Snapshot:/Users/nickb/Zotero/storage/WEMFH8TK/science.html:text/html}
}
Downloads: 0