about being bold - homepage -...

Review

About being BOLD

Dinesh G. Nair*

Palmer 127, Department of Neurology, Beth Israel Deaconess Medical Center/Harvard Medical School, Boston, MA 02215, USA

Accepted 13 July 2005

Available online 5 October 2005

Abstract

The last decade has seen an unprecedented increase in the use of functional magnetic resonance imaging (fMRI) to understand the neuralbasis of cognition and behavior. Being non-invasive and relatively easy to use, most studies relied on changes in the blood oxygenation leveldependent (BOLD) contrast as an indirect marker of variations in brain activity. However, the fact that BOLD fMRI is dependent on the

blood flow response that follows neural activity and does not measure neural activity per se is seen as an inherent cause for concern whileinterpreting data from these studies. In order to characterize the BOLD signal correctly, it is imperative that we have a better understanding ofneural events that lead to the BOLD response. A review of recent studies that addressed several aspects of BOLD fMRI including events at

the level of the synapse, the nature of the neurovascular coupling, and some parameters of the BOLD signal is provided. This is intended toserve as background information for the interpretation of fMRI data in normal subjects and in patients with compromised neurovascularcoupling. One of the aims is also to encourage researchers to interpret the results of functional imaging studies in light of the dynamicinteractions between different brain regions, something that often is neglected.

D 2005 Elsevier B.V. All rights reserved.

Theme: Neural basis of behavior

Topic: Cognition

Keywords: BOLD; Functional magnetic resonance imaging; Neurovascular coupling; Cerebral blood flow; Neural activation

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2302. At the site of neural activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230

2.1. Neurovascular coupling and metabolic events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230

2.2. Electrical activity and the BOLD signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2333. Spatial resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2344. Temporal resolution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2355. Negative BOLD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236

6. Resting state fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2387. Other influences on cerebral blood flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2388. Implications for the study of clinical disorders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

9. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240

0165-0173/$ - see front matter D 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.brainresrev.2005.07.001

* Fax: +1 617 632 8920.

E-mail address: [email protected].

Brain Research Reviews 50 (2005) 229 – 243

www.elsevier.com/locate/brainresrev

1. Introduction

Identifying active brain areas while executing differenttasks—motor, sensory (somatic, visual, auditory), andcognitive (memory, attention, and other intellectual abil-ities), has been an important domain of neuroscienceresearch since the introduction of functional brain imagingtechniques such as Positron Emission Tomography (PET)and functional magnetic resonance imaging (fMRI). Themost common method of fMRI is blood oxygenation leveldependent (BOLD) contrast imaging, in which hemoglobinis used as an endogenous contrast agent, relying on thedifference in magnetic properties of oxyhemoglobin (dia-magnetic) and deoxyhemoglobin (paramagnetic) [103].Functional MRI BOLD thus measures a correlate of neuralactivity, the hemodynamic response. The hemodynamicresponse, as the name implies is dynamic and depends oncerebral oxygenation, blood flow, and blood volume (forreviews see [2,5,19,58,60,81,82,95,128]). An importantissue regarding interpretation of functional MRI signals(especially when using block designs) is that the functionalbrain maps that we see are static representations of thisdynamic activity averaged over a long period of time. So,the accurate interpretation of the BOLD signal depends onhow effectively one characterizes the nature of the under-lying neural activity that gives rise to the hemodynamicresponse and how these two (neural activity and blood flowresponse) are coupled. This in turn warrants a betterunderstanding of what happens in the brain at the level ofneurons and their immediate microvasculature and also ofother factors that modulate the BOLD signal. Heeger andRess [58] grouped factors other than neurovascular couplingessentially into three

1. fMRI acquisition technique: Since BOLD fMRI providesa complex signal, which depends on blood flow, bloodvolume, and oxygenation, the signal is highly dependenton the details of local vasculature—i.e., the relation ofdraining veins, venules, and capillaries to the neuralstructure under consideration. Variations of the acquis-ition technique can be used to emphasize or suppresssignals from these sources.

2. Behavioral and stimulation protocols and data analysistechniques: These determine the spatial and temporalresolution of the signal. For example, the traveling wavevisual stimulation protocol [37], sparse temporal sam-pling [50,53].

3. How the neuronal activity is measured and quantified:The signal depends on whether one measures the activityof a whole population of neurons, or of a sub-population,the local field potential or the current source density.

Given the dynamic interconnectivity of neurons in thehuman brain and the complex nature of the fMRI BOLDresponse, it is easy to see why the interpretation of thissignal is not straightforward. Previous work on BOLD fMRI

focused mainly on one aspect of the signal—events at thesynapse and neuroenergetics, neurovascular coupling, rela-tion of the signal to neural spikes, or the basis of some partof the BOLD signal such as the initial dip or the temporaland spatial attributes. The aim of this article is to integrateresults from most of these domains and give the reader abroader perspective on the origin and interpretation of theBOLD signal. First, a review of the neurometabolic eventsat the level of the synapses is given, followed by how theyrelate to neurovascular coupling. Later, some parameters ofthe BOLD signal are described along with their clinicalsignificance. The paper concludes with a note of caution onthe interpretation of BOLD signals and establishingstructure–function relationships in the brain based onBOLD fMRI.

2. At the site of neural activity

Neuronal action potentials result in (1) the release ofneurotransmitter into the synaptic cleft—glutamate beingthe most common excitatory neurotransmitter in the cerebralcortex and (2) changes in ionic gradients. Continuous neuralactivity and maintenance of homeostasis are dependent onactive processes requiring energy such as restoration ofionic gradients and repacking of neurotransmitter molecules.Cerebral blood flow not only delivers glucose and O2, butalso carries away CO2 and heat. Although Roy andSherrington [114] postulated that vascular supply in thebrain changes with local variations in functional activity, thedemonstration of neurometabolic coupling was first possibleby the development of the 2-deoxyglucose autoradiographictechnique by Sokoloff and colleagues [125] and its laterextension 2-fluorodeoxyglucose PET [123]. It is importantto understand the mechanisms that bring about this neuro-metabolic coupling at the molecular level (for an excellentreview see [85]).

2.1. Neurovascular coupling and metabolic events

Under normal resting conditions, the brain’s energydemands are met (ATP production) almost exclusively byglucose oxidation. More than 90% of resting state glucoseconsumption is oxidative. Since the energy yield of glucoseoxidation is much more than that of glycolysis (at least 15times more ATP), more than 99% of the ATP production inthe resting stage is by glucose oxidation [41], oxidizingglucose to water and CO2. Fox et al. [41] reported that themean whole brain cerebral metabolic rate for oxygen(CMRO2) and that for glucose (CMRGlu) is in a 4.1:1molar ratio during the resting state. This increased cerebralmetabolic rate for O2 verifies the finding of greater glucoseoxidation and hence oxidative mechanisms for ATPproduction during rest. Functional activation increasescerebral metabolic rate for glucose and cerebral blood flowby about 50%. However, cerebral metabolic rate for O2

D.G. Nair / Brain Research Reviews 50 (2005) 229–243230

increased only by 5% [40,41]. Thus, about 90% of theactivity-induced increase in glucose uptake is not oxidized.The non-oxidative, alternative pathway for glucose metab-olism is glycolysis which results in an increase in lactateproduction. Moreover, with an increase in cerebral bloodflow that far exceeds cerebral metabolic rate for O2, a highlysignificant reduction in oxygen extraction fraction occurs.Creutzfeldt [29] estimated that only a maximum of 3%cortical energy consumption was accounted for by spikeactivity of cortical nerve cells. In order to understand howthe relationship between CBF and CMRO2 varies across thebrain, Davis et al. [31] used CO2 breathing as a physio-logical means of manipulating CBF independent ofCMRO2. BOLD signal in the human visual cortex followingphotic stimulation (before and after CO2 inhalation) wascalibrated against perfusion-sensitive MRI. The derivedcalibration parameter was used to compute dynamicoxidative metabolism maps (CMRO2) relative to baseline.This provided them a quantitative method not only to assessthe sensitivity of the BOLD signal to changes in cellularactivity of the brain, but also to compare these measure-ments between regions and subjects. The authors concludedthat the observed increase in CMRO2, which was about 1/3of the increase in CBF, supports a partial coupling of bloodflow to O2 demand. This is against previous evidence forlittle change in oxidative metabolism during task activation[40,41].

How does neural activity result in an increase in cerebralblood flow? What are the factors that are crucial to thecontrol of brain microcirculation? A recent review byIadecola [65] highlights the fact that activity-inducedhemodynamic responses require complex signaling mecha-nisms that involve not only neurons but also astrocytes andvascular cells. These cells are involved in the regulation of asequence of chemical signals, which result in functionalhyperemia. How the need for an increase in blood flow iscommunicated to the local blood vessels is quite intriguing.Interneurons and astrocytes, which are integrated structur-ally into the local circuitry, are considered to play a crucialrole in this regard. It has been shown in rat cortical slicesthat, dilation of arterioles triggered by neural activity isdependent on glutamate-mediated Ca2+ oscillations inastrocytes [134]. In addition, Iadecola [65] discusses thepossibility of retrograde intramural signaling throughendothelial and smooth muscle cells to cause vasodilatationof upstream vessels. This would allow more blood to reachthe site of neural activation downstream. Astrocytes releaseseveral vasoactive factors such as NO, K+, adenosine,arachdonic acid metabolites, while endothelial cells releaseNO, prostacyclin, CO, and endothelium-derived hyper-polarizing factor. Metabolites such as lactate, K+, H+, oradenosine and neurotransmitters such as vasoactive intesti-nal peptide (VIP), acetylcholine (ACh), and noradrenaline(for details see [65,75,85]) also act as vasoactive substances.These chemical signals act on smooth muscle cells andpericytes and result in vasodilatation, through changes in

intracellular concentration of Ca2+. The details of howneurons and other supporting cells in the brain arestructurally organized to facilitate neurovascular couplingand how they result in the metabolic events following neuralactivation has previously been studied [85,86]. Astrocytesare stellate cells that have processes around the synapse andalso around intraparenchymal capillaries (end feet). Theenormous number of astrocytes in the brain (astrocyte/neuron = 10:1) and their anatomical proximity to thesynapse and capillaries make them ideal candidates forcoupling neuronal activity with metabolism. It has beenshown that a glucose transporter (GLUT1 type) is expressedon astrocytic end feet [100] and the synaptic contacts (ofastrocytes) possess receptors for various neurotransmitters,especially glutamate. Activated synaptic terminals releaseglutamate, which acts on target neurons mediated byglutamate receptors. This action is terminated by a reuptakesystem, present in the astrocytes (Fig. 1). How astrocytescouple this glutamate reuptake system with glucose uptakefrom capillaries is the key to the metabolic processes thatfollow neurotransmitter release into the synapse. Removalof glutamate from the synaptic cleft takes place throughspecific glutamate receptors GLT-1 and GLAST, which areglial specific [112]. Glutamate uptake is driven by theelectrochemical gradient of sodium ions, three Na+ beingco-transported along with one glutamate into the astrocyte.Inside the astrocyte, glutamate is converted into glutamine,which is taken back into the neuron to replenish theglutamate reserve. Glutamate uptake into the astrocyte alsostimulates glucose uptake into the astrocyte from thecapillaries and glycolysis. This is mediated by Na+-K+-ATPase since oubain, a Na+-K+-ATPase inhibitor com-pletely inhibits the glutamate-evoked 2DG uptake byastrocytes [104]. Activity-induced glycolysis results in anincrease in lactate concentration [106] inside astrocytes,which is further transported into neurons to meet the energydemands of active neurons. Lactate dehydrogenase (LDH),the enzyme that catalyzes the conversion of lactate topyruvate has been found in neurons (specifically the LDH1type, which is the form found in lactate-consuming tissues)[15]. This suggests that the lactate that enters neurons fromastrocytes could be converted to pyruvate, which enters theTCA cycle to serve as energy molecules (see Fig. 1). Atwelland Laughlin [6] estimated the metabolic costs of brainactivity and showed that most of the energy is used byneurons depending on their firing rate; only a smallpercentage of energy is used for neurotransmitter recyclingby astrocytes. Shulman and colleagues [116,119] providedsupport for this finding in their studies using magneticresonance spectroscopy (MRS) by showing that a largefraction (about 80%) of the energy use in the brain iscorrelated with glutamate cycling and hence active signalingprocesses.

Several models have been proposed to account for thechanges in blood flow, blood volume, oxygenation, and theBOLD signal following neural activation. Predominant

D.G. Nair / Brain Research Reviews 50 (2005) 229–243 231

among them, is the balloon model of Buxton and colleagues[21] and the delayed compliance model of Mandeville et al.[88] which are biomechanical models that incorporate theeffects of changes in both blood oxygenation and bloodvolume to explain the dynamic changes in deoxyhemoglo-bin content during brain activation. These models explainthe post-stimulus undershoot of the BOLD response interms of a CBV response that recovers to a baseline moreslowly than the CBF response. In Buxton’s model, it isassumed that the vascular bed (CBV) within a small volumeof tissue can be modeled as an expandable venouscompartment or balloon that expands with increasing CBFand the outflow is modeled as a function of the balloonvolume. The dynamics of the BOLD response changesdepending on the outflow function. The model predictstransient changes in deoxyhemoglobin and the BOLDresponse, including the initial dip, overshoot, and aprolonged post-stimulus undershoot of the BOLD signal.Using a generalization of their model, Buxton andcolleagues [102] recently tried to determine the origin ofhemodynamic transients such as the post-stimulus under-shoot. They used arterial spin labeling to simultaneouslymeasure BOLD and CBF in the primary motor cortex (M1)and the supplementary motor area (SMA) during a finger-tapping experiment. Since the CBF to these areas were notsignificantly different but the BOLD responses were, theyfurther assumed that the transients arose out of the differenttime courses of CBF, CBV, and CMRO2. In order to

investigate how different these time courses have to be, toproduce such different BOLD responses for similar flowresponses, the authors used the balloon model to computecurves of blood volume and total deoxyhemoglobin fordifferent assumed curves for CBF, CBV, and CMRO2.These curves of blood volume and deoxyhemoglobin werein turn used to calculate the BOLD response. Such studiesunderscore the relevance of theoretical models in under-standing the dynamics of the BOLD signal. Several suchmodels exist [7], the details of which are beyond the scopeof this review.

The mechanisms of neural activity-induced increase inblood flow and metabolism can be summarized as follows.

(i) Increase in neural activity is followed by a relativelylarge increase in blood flow and cerebral metabolicrate for glucose with a slight increase in cerebralmetabolic rate for O2.

(ii) Glutamate release induces glucose transport intoastrocytes aided by Na+-K+-ATPase activity.

(iii) Astrocytes rely on non-oxidative glycolysis as sug-gested by the increase in lactate levels. Neuronsdepend on oxidative metabolism of lactate.

The issue of activity-dependent glycolytic processing ofglucose is currently under debate. Using 13C-glucosemagnetic resonance spectroscopy, Hyder et al. [63] reporteddata that support an increase in oxygen use during neural

Fig. 1. Schematic for the mechanism of glutamate-induced glycolysis in astrocytes during physiological activation (arrow labeled A). At glutamatergic

synapses, glutamate depolarizes neurons by acting at specific receptor subtypes. The action of glutamate is terminated by an efficient reuptake system located

primarily in the astrocytes. Glutamate is co-transported with Na+, resulting in the activation of Na+-K+-ATPase. Lactate, which results from glycolysis in the

astrocytes, is taken up by the pre-synaptic terminal. Direct glucose uptake into the neuron under basal conditions (B) is also shown. Pyr, pyruvate; Lac, lactate;

Gln, glutamine; G, guanine nucleotide binding protein; PGK, phosphoglycerate kinase. Adapted from Pellerin and Magistretti, Proc. Natl. Acad. Sci. USA. 91

(1994) 10628. Copyright, Proc. Natl. Acad. Sci. USA. 1994.


activation. Based on Magistretti and Pellerin’s model, it ispossible that an initial glycolytic processing of glucose inastrocytes is followed by oxidation of lactate in neurons.But how much oxygen is used is not known. Thompson etal. [126] studied the relation between single neuron activityand tissue oxygenation in cat’s visual cortex. They simulta-neously measured tissue oxygenation (using an oxygenmicroelectrode) and single-cell neural activity (using aplatinum microelectrode housed within the pipette contain-ing the oxygen electrode) and found that increases inneuronal spike rate were accompanied by immediatedecreases in tissue oxygenation. This initial decrease inoxygenation is believed to be responsible for what has cometo be known as the ‘‘initial dip’’ in BOLD response.Thompson et al. [126] found that like hemodynamicresponses, the time course of the oxygen response exhibitedthe initial dip followed by a positive peak. Under optimalorientation conditions of the visual stimulus, large neuralresponses (spikes) occurred in area 17 of the cat visualcortex that were accompanied by immediate decreases inoxygenation as revealed by the largest initial dip. This resulthighlights the initial coupling between neural activity andoxidative metabolism.

During neural activation, the initial dip is thought torepresent an increase in deoxyhemoglobin due to shortuncoupling between blood flow and oxidative metabolism[20,40,62,85,126]. This phenomenon was first reported byGrinvald and colleagues [49,87] using optical techniques tomeasure dynamic changes in oxy- and deoxyHb in cats. Thefirst human fMRI studies to demonstrate the initial response[38,61,97] used visual paradigms. These studies demon-strated that some focal areas of the visual cortex (columnsthat are highly selective to the visual stimulus) displayed aninitial negative response while some others including thosewhere draining veins are visible, showed only the positivesignal change [62]. The initial dip is highly controversialbecause very few studies were able to show this earlyresponse. Jones et al. [66] and Lindauer et al. [80] usedoptical measurements in a rat model with stimulation of thewhisker pad to look for the initial dip. Jones et al. were ableto detect the dip while Lindauer et al. were not. Theapparent discrepancy in results between the two studiescould be a result of the differences in experimentalprocedures such as surgery and anesthesia [130]. It ispossible that the initial dip is also very sensitive to imagingparameters. Among them, two are mentioned below. First,majority of imaging experiments that demonstrated the earlyresponse were conducted at high field strengths (¨4 T).However, the dip has also been observed at low fieldstrengths [113], albeit in patients with disturbed cerebralautoregulation. Second, Kim and his colleagues, in twostudies with different end tidal CO2 levels [56,73] found theinitial dip only in the study with a larger end-tidal CO2.Larger end-tidal CO2 presumably leads to more oxygenextraction immediately after neuronal firing and hence moredeoxyhemoglobin, leading to a pronounced initial dip.

Current data thus indicate that the initial dip is not a robustphenomenon.

2.2. Electrical activity and the BOLD signal

In the human brain, glucose forms a major, if notexclusive energy substrate [124]. When a particular brainarea is recruited in carrying out a task, a group of neurons inthis area fires action potentials. However, there may begroups of neurons which do not fire action potentials, butnevertheless utilize oxygen—for instance, neurons at sub-threshold levels of activation, neurons with varying levels ofsimultaneous excitation and inhibition, and feedback fromlocal and distant sites. In addition, the fMRI signal may alsoreflect changes in neuronal synchrony without a concom-itant rise in mean firing rate [43]. However, neuronaldynamics as assessed by event-related, dynamic correlationsusing joint peristimulus time histograms and the mutualinformation between stimulus-induced transients in twoneuronal populations suggest that synchronization betweentwo neuronal populations and the mean firing rate arethemselves tightly coupled [25]. If true, this would simplifythe interpretation of BOLD responses, which should indexan increase in mean firing, and an increase in synchronizedneuronal interactions. Without a measure of the neuronalelectrical activity, it is difficult to say whether fMRI signalchanges occur due to the neuronal firing, vascular response,or both. Logothetis et al. [83] simultaneously measuredneuronal activity and the hemodynamic response to studyhow these factors are related. It was found that the BOLDresponse in primates directly reflects an increase in neuralactivity, correlating particularly with local field potentials(LFP—represent synchronized synaptic inputs of a givenneuronal population). The Logothetis study [83] examinedhow well fMRI measurements could be predicted from LFPand multi-unit activity (MUA—reflects spiking activity ofneurons near the electrode tip) and found that on average,LFP was better than MUA in predicting fMRI responses.Interestingly, they found that the fMRI BOLD responseincreased much more rapidly than LFP at low stimuluscontrasts, but less rapidly at higher contrasts. In other words,they observed a non-linear relationship between LFP andBOLD signal.

The relationship between brain energy metabolism andcellular activity was also studied by combining magneticResonance Spectroscopy (MRS) with extracellular record-ing of neuronal activity in rats [64,121]. It was observedusing MRS that the change in oxygen consumptionproduced by somatosensory afferent stimulation was pro-portional to the change in excitatory glutamatergic neuro-transmitter flux. This in turn was proportional to the changein neuronal spiking as measured by intracellular recording.Other studies have compared human fMRI and monkeysingle-unit data to infer that fMRI signals are directlyproportional to average neuronal firing rates [59,60,109].Rees et al. [109] found that the average neuronal firing rates


calculated from single-unit data in macaque MT and fMRImeasurements from human MT complex (V5) increasedapproximately linearly with motion coherence of visualstimuli. Heeger et al. [60] also did similar comparisonsbetween fMRI measurements from human primary visualcortex (V1) and single unit data from monkey V1 usingstimulus contrast as the independent variable and reported aproportional relationship between average neuronal firingrate and the fMRI signal. Results from the above studiesrevealed that in two different cortical areas (V1 and MT),fMRI responses are proportional to average firing rates, butwith different proportionality constants. Several reasonscould account for the discrepancy in the two estimates of theproportionality constant. These include differences inmethodology, physiology, and functional connectivitybetween the two regions [60]. In addition, it must be notedthat the range of the fMRI signal measured in the two labsdiffered and that the response of these two visual areas tothe two types of stimuli may in fact, be different. The factthat Rees et al. [109] used a 2 T magnet compared to the 1.5T magnet by the other group may also have contributed tothese differences. Nevertheless, the important result of thesestudies is that the hemodynamic response shows a roughlylinear relation to underlying neuronal activity (see also[16,17]). Similar conclusions were reported by Arthurs andcolleagues [3,4] when they compared the amplitude ofcortical evoked potentials secondary to median nervestimulation with the BOLD response. Non-linearities inthe fMRI time series have been shown to be related toseveral parameters of the stimulus, especially the stimulusduration [13,16,30,99,105,131]. The nature of the non-linearity [19] is that the response to a brief stimulus (<4 s)appears stronger than would be expected, given the responseto a longer stimulus. A likely source of this non-linearity isthe nature of the spiking activity in neurons itself, where wesee an initial peak before a sustained plateau. This couldexplain a stronger BOLD signal for a brief stimulus [16].Another is that any BOLD signal increase results from aceiling effect in CBF increase. Also, as mentioned earlier,metabolic changes follow different time courses with theCBV change lagging behind the change in CBF. Non-linearities have been extensively modeled by Friston andcolleagues [44,46–48,93] and have helped us better under-stand the relationship of the BOLD response to stimulusparameters such as amplitude, duration, prior stimulus,stimulus onset asynchrony and also to early and lateneuronal components. It is evident from the abovediscussion that several factors, intrinsic and extrinsic,determine neurovascular coupling and the complex natureof the BOLD signal. Other factors such as the localdistribution of blood flow within the vascular network[57] also play a role. In addition, it must be remembered thatthe signal we observe may include at least two differentsources of activity, viz., the mean rate of generation ofaction potentials within a region of the brain (neural output)and also the synaptic activity responsible for neurotrans-

mitter recycling (synaptic input). These two types of activitywork together for excitatory synapses, since a greater rate ofsynaptic activity will result in a greater rate of actionpotentials. It may seem that these two neural phenomenawork in different directions and the associated energydemands cancel out for inhibitory synapses, since greatersynaptic activity results in decreased firing of neurons.However, this assumes that inhibitory activity does notrequire (or requires lesser) energy than excitatory activity.Strong inhibitory interactions in fact, result in similar fMRImeasurements and mean firing rates as excitatory phenom-ena [59]. Moreover, for an inhibitory neuron to increasefiring rate, it should receive more excitatory input from theneighboring cortex [60]. Hence, for excitation as well asinhibition, one would expect the BOLD signal to increase.Lauritzen [78] stressed the importance of distinguishingbetween effective synaptic inhibition and deactivation thatincrease and decrease CBF and glucose consumption,respectively. Recently, Lauritzen and colleagues [22]reported that combined electrical stimulation of two neuro-nal networks, one excitatory and one inhibitory, in thecerebellar cortex projecting to the same target cell producedchanges in CBF and local field potentials (LFP) that werecontext sensitive. Their results suggest that the balancebetween synaptic excitation and inhibition, controlled theamplitude of the vascular signal. By topical application ofthe GABAA agonist, muscimol, they [23] also demonstrateda dissociation of cerebellar Purkinje cell spiking and CBFunder nonstimulated (basal) conditions, and of synapticactivity and CBF under stimulated conditions. Overall theirresults implied that changes in CBF to excitatory input arestrongly influenced by synaptic inhibition and that it isdifficult to establish whether neural activity of a brain regionis increased based on maps that use blood-flow changes asmarkers. That the fMRI signal is presumably proportional tothe local firing rate averaged over a small region of cortexand a short period of time raises two issues of the fMRIsignal, namely the spatial and temporal resolution.

3. Spatial resolution

Typical in-plane spatial resolution in human fMRIstudies is 3–5 mm, which is better than PET and singlephoton emission computerized tomography (SPECT) [73].Comparing single and multi-unit neuronal activity with theBOLD signal, Kim et al. [72] recently concluded that theBOLD signal is a robust predictor of neuronal activity onlyat the supra-millimeter scales. The localization of the BOLDsignal depends on the spatial relation of voxels to adjacentvessels. We know that the BOLD effect is manifested in thecapillary beds, venules, and draining veins, which are only60–70% saturated with oxygen at rest, and hence have thecapacity to become more oxygenated. The size of thevessels giving rise to the fMRI signal thus varies from thecapillary bed (< 10 Am) to draining veins (a few millimeters)


[77,79,96] and the signal from the latter can be displaced byseveral millimeters from the site of neuronal firing[19,42,70]. Given the complex nature of the BOLD signaland the influence of large draining veins on the location ofthe observed signal, it is desirable to obtain higher spatialresolution for the signal. Improving spatial resolution of thefMRI signal means localizing neuronal activity moreprecisely. Malonek and Grinvald [87] using optical imagingtechniques, exploited the initial dip to achieve high spatialresolution and delineate even the columnar structure of thecat visual cortex. The initial dip, as mentioned above, occursdue to immediate oxygen utilization after neural activity,and hence is quite localized compared to the ensuing bloodflow, which occurs at a much coarser spatial scale. Malonekand Grinvald [87] referred to this delayed increase incerebral blood flow as ‘‘watering the garden for the sake ofone thirsty flower’’. Evidence that tissue oxygen utilizationalso followed this pattern came from Thompson et al. [126].They found that an initial dip occurred only in corticalcolumns, which responded maximally to the stimulus whilethe neighboring columns exhibited a robust positive peak inoxygen response without an initial dip. Hence the initial dipnot only provides evidence for uncoupling between neuralactivity and blood flow, but also offers a temporal windowfor mapping signals at higher spatial resolution. Althoughthe dip is not consistently observed, this potential of theinitial dip for mapping columnar structure was exploited bya few fMRI studies. It is not surprising that most of thesefMRI studies used high field strength magnets in the orderof 4.7 T [33,69,71,94,98], as higher spatial resolutionappears to derive from higher intrinsic signal, which isobtained by using high field magnets.

Factors other than physiological, contribute to improvedspatial resolution of the fMRI signal (for a detaileddiscussion, see [73]). Issues to be considered include task-induced signal change to signal fluctuation ratio, andminimization of large vascular contributions using differentimaging parameters. In addition to technical limitations thatrestrict spatial resolution, biological variables may play arole. For instance, vascular supply is not regulated on thescale of individual neurons and the neural-hemodynamiccoupling varies between species, people, and even betweendifferent areas of the brain [95]. A combination of thesefactors may prescribe the limits to which this technologycan be pushed to achieve better spatial resolution.

Several techniques such as Spin Echo (SE) imaging,diffusion-weighted imaging (DWI) and arterial spin labeling(ASL) [1,19,21,33,34,36,68,76,132] have been introduced toquantitatively and qualitatively characterize cerebral bloodflow. Some of these such as perfusion-based fMRI improvethe spatial specificity of the signal [33], especially whenlarge-vessel contributions are minimized or eliminated. In anSE-BOLD experiment, the echo time (TE) is longer than for agradient echo (GE) experiment, to maximize the signalchange due to a small change in transverse relaxation rate.Compared to larger vessels, diffusion in smaller vessels

results in a larger change in transverse relaxation rate, andhence the SE-BOLD effect is greater in smaller vessels. SinceSE-BOLD is more sensitive to the smallest vessels (capil-laries and small venules), it is considered to improve thespatial resolution of the BOLD signal [19]. In diffusion-weighted imaging, essentially a bipolar symmetric gradient isadded to a SE sequence, which enhances dephasing of spinsand results in attenuation of the signal. Since flowing bloodhas larger displacements than diffusion effects, the bloodsignal can be selectively suppressed with only a small effecton the extravascular signal. This would enhance the spatialselectivity of the SE experiment. Arterial spin labeling (ASL)is essentially a non-invasive tracer method, where arterialwater is magnetically labeled to act as the diffusible tracer.After the labeled protons reach the capillaries, they pass intothe tissue and alter the total magnetization. In an ASLactivation experiment, a series of alternating control and tagimages are collected and can be used to obtain quantitativemeasurements of flow change. Since the ASL activation is achange in arterial flow to the region, whereas the BOLDsignal is primarily a change in venous oxygenation, theformer is free from the draining vein problem and isconsidered spatially a more accurate signal. However, itmust be noted that sensitivity is reduced with SE, DWI, andASL-based fMRI and hence these techniques are not quitepopular as routine functional imaging techniques.

By and large, the optimal spatial resolution that we needfor the type of cognitive and motor experiments that aretypically done using fMRI, is determined by the questionswe ask in the study. If the goal of the experiment is to testwhether a brain region in recruited in executing a particulartask, then the displacements due to draining veins may notbe critical. However, if one embarks on a detailed brainmapping study requiring precise anatomical information, anASL experiment may be more appropriate than a BOLDexperiment.

4. Temporal resolution

Temporal resolution has been defined in several ways inthe literature which includes the image acquisition rate, thetime it takes for activation-induced response to rise or fall agiven amount, the maximum rate at which activation can beturned on and off, and the smallest detectable activationduration (for details of these definitions, refer to [10]).Better temporal resolution is desired in order to resolve thedifference in the onset of activity between two regions of thebrain, involved in executing a task. Determining the serialorder of activation is important for several reasons. First, ittells us how different brain areas work together to carry outa task. Second, this is crucial in identifying the functionalconnectivity of a brain area and how different connectionsare exploited for say, different cognitive demands of thetask. Third, it throws light on how different brain regions arerecruited into and removed from the functional network


over time. This is especially relevant in disease states whereneural and vascular adaptive changes occur to compensatefor the functional deficit. For instance, in motor recoveryfollowing cerebral infarction, one observes widespreadactivation in the initial post-stroke period, which changesto a more localized pattern of activation during the later partof recovery [24,28,39,89]. Although the order of activationof brain areas that comprise serial pathways is veryinteresting, it is difficult to infer differences in neuronallatencies among areas based upon the hemodynamicresponse, because this inference assumes the hemodynamicresponse function is exactly the same in all areas. Therefore,most people look at the effects of an experimentalmanipulation on hemodynamic latencies within the samearea.

Early fMRI studies used the block design and collectedminute long epochs of averaged brain activity. Then cameattempts at improving the temporal resolution of the fMRIsignal. One of the notable advances was event-related fMRI[18,45,91], which allowed identification of brain signalsinduced by individual cognitive events—for instance todistinguish brain activation during go and no-go trials of ago/no-go paradigm. Another manipulation to improvetemporal resolution was the introduction of a time-resolvedfMRI technique known as sparse temporal sampling withjittered volume acquisition [50,53], which is mainly done inauditory experiments in order to minimize the effects of thescanner noise on the auditory cortex. Analysis techniquessuch as selective averaging [30] and multiple regression[26], offered the possibility of separating brain responses byusing rapidly presented, mixed trials. However, it transpiredthat these approaches were special cases of the generallinear convolution model that had already been establishedfor block designs. Nowadays, fMRI data are usuallyanalyzed by convolving some stimulus function with ahemodynamic response function to provide regressors in alinear model. This hemodynamic response function modelsthe neurovascular coupling discussed above [45]. Boyntonet al. [16] used manipulations of contrast and timing ofvisual stimuli to show that the duration and amplitude of thehemodynamic response were respectively proportional tochanges in duration and amplitude of neural activity. Thishad important implications in finding solutions to thehemodynamic inverse problem [17], in that temporallyshifted hemodynamic responses could represent delays inunderlying neuronal activity, especially given the linearrelation between the two [16,58]. Bellgowan et al. [11]tested this idea in a 3 T fMRI study using a lexical decisiontask, and found that the hemodynamic timing estimates ofcertain brain regions correlated with timing delays inducedby the cognitive demands of the task. For instance,prefrontal cortex showed an onset delay proportionate tothe rotation of letter strings and also during comparison ofnon-words with words. Kim and colleagues [110,111]introduced what they call ‘‘time-resolved fMRI’’ to identifythe temporal sequence of neuronal activation when subjects

performed certain tasks. The essential idea behind this wasthat, a variable behavioral parameter (such as reaction time)could be correlated with a measure of the time series (onsettime or width of the response) to give more information onthe time at which a particular brain area was involved in atask. It must be noted here that these studies recorded singletrial data at 4 T, because at high magnetic fields it is possibleto monitor fMRI signal evolution in a single execution of atask without averaging over many trials. If averagingbecomes necessary, it is usually done by aligning theresponses with respect to a behavioral correlate, such as thereaction time. This method of improving the temporalresolution of the fMRI signal is based on the tenet that brainareas activate in a temporal order. Evidently, this appliesonly to hierarchical networks that use a sequential mode ofinformation processing (as may be the case of simple tasks,without involving higher order semantic processing) andmay not be applicable for many other neural networks,where information processing occurs in parallel. In additionto manipulations of imaging parameters, use of fMRI withother imaging modalities such as magnetoencephalography(MEG) and electroencephalography (EEG) will improve ourunderstanding of the temporal nature of activation andhence aid in identification of neural loops involved in thetask.

5. Negative BOLD

Brain regions showing positive BOLD signals aretypically interpreted as being ‘‘involved’’ in carrying outthe task. Not infrequently, a negative BOLD response isseen in fMRI studies, which is seldom reported, or reportedwithout adequate functional interpretation. Given ourcurrent understanding of the mechanisms that operate inthe neural and microvascular levels that give rise to theBOLD signal, we know at least that a negative BOLDmeans that there is a reduction in local venous oxygenationlevel. Considering the sequence of events that follow neuralexcitation, one of the following possibly contributes to anegative BOLD—(a) larger fractional increase in cerebralmetabolic rate of O2 compared to cerebral blood flow; (b) alarger fractional decrease in blood flow compared tocerebral metabolic rate of O2; (c) a delay in blood flow tothe region causing the initial dip to be prolonged. As wehave seen earlier, the first possibility, which is a largerfractional increase in cerebral metabolic rate of O2

compared to cerebral blood flow, is not known to occur inthe brain with neural activity. Harel et al. [56] investigatedthe second possibility, which is a decrease in cerebral bloodflow following neural activity, using T2* weighted BOLDand cerebral blood volume techniques in an anesthetized cat.For blood volume weighted imaging they used injections ofmonocrystalline iron oxide nanoparticle (MION) as acontrast agent, which imparts sensitivity to changes incerebral blood volume. An inherent assumption in this case


is that blood volume changes are monotonically related toblood flow changes. An increase in the amount of MION ina voxel causes a decrease in T2* weighted MRI signal. Inother words, following the injection of MION, brain areasthat show an increase in T2* weighted signal have less ofMION and hence decreased cerebral blood volume. Theyfound that during visual stimulation, the primary visualcortex with a positive BOLD signal showed an increase incerebral blood volume (negative signal after MION) and thesuprasylvian gyrus with a negative BOLD, showed adecrease in cerebral blood volume (positive signal changeafter MION). A decrease in cerebral blood volume andcerebral blood flow hence contributed to the observednegative BOLD signal in the suprasylvian gyrus. Harel et al.[56] explained this phenomenon by two possible neuro-physiologic mechanisms—(a) a reduction in neuronalactivity causing a decreased blood flow or (b) a vascular‘‘steal’’ effect in which blood is diverted or allocated to themost active areas while adjacent areas exhibit a decrease inblood flow. An interesting fact in this study was that thesuprasylvian gyrus, which is known to have an increasedfiring rate following similar visual stimuli, showed adecrease in blood flow. So the authors ruled out the firstpossibility of a reduction in neuronal activity as a plausibleexplanation for a reduction in cerebral blood flow. However,it must be noted that they did not have an independentmeasure of neuronal activity to validate their claim.Considering the second possibility, Harel et al. concludedthat the reason for a decrease in cerebral blood flow andcerebral blood volume to the suprasylvian gyrus despite itsincreased firing rate was related to the diversion of blood tomore active or energy demanding areas such as the V1. Theproximity of the suprasylvian gyrus to V1 and thepossibility of both areas depending on the same vascularresources, support this proposal. Nevertheless, it seemsunlikely that redistribution of cerebral blood flow is theunderlying cause due to the following reasons. First, thecerebral vascular reserve is so large compared to changes inblood flow associated with the task, that there is little reasonto ‘‘steal’’ from other adjacent areas. Second, the decreasescan occur remote from the site of increases [118] making alocal vascular steal almost impossible. It is important toremember here that there may be neurons at the subthres-hold level, neurons having simultaneous excitation andinhibition, or neurons with feedback from local or distantsites, which result in different levels of energy expenditureand hence different CBF demands. In addition, increasedfiring of a small number of neurons, as recorded using unitelectrode techniques, may actually suppress the averagefiring rate of neurons within a voxel through lateralinteractions and ‘‘winner-takes-all-like’’ competition. Thispossibility was not considered in early interpretations ofthese results.

Rother et al. [113] examined the third possibility of adelay in blood flow to active neural regions in a patient withimpaired cerebrovascular reserve capacity due to transient

ischemic attack. They observed a prolonged negative BOLDresponse prior to the positive response and argued that thisprolonged negative response was due to persistent deoxy-hemoglobin arising out of the absence of an immediatehemodynamic response. Hence, in conditions of impairedcerebral blood flow or impaired autoregulation, cerebralblood flow and metabolism may be uncoupled not only for afew seconds, but for longer periods of time, till thehemodynamic coupling has taken place.

Several other studies have reported instances where thenegative BOLD signal was observed. Human subjects,who showed a positive BOLD response while awake,exhibited a negative signal following anesthesia [90],suggesting that anesthesia may contribute to the genesisof the negative BOLD signal. Similarly, Logothetis [81]demonstrated the negative BOLD signal in V1 and V2 ofanesthetized monkeys in response to rotating checkers ateccentricities of 0-–10-. Shulman et al. [118] reportedseveral foci in the brain where there is a decrease in CBFfollowing some language and visual tasks. Visual attentionwas shown to induce a negative BOLD response in theoccipital cortex in other studies [120,127]. By showingthat a visual stimulus that stimulates primary visual cortexin one hemisphere can cause extensive suppression in theother hemisphere, Smith et al. [122] recently claimed thatthe negative BOLD response is not a local phenomenon. Inaddition to the possibility that this signal arises out of adecrease in neural activity, they also considered the optionthat it results from blood flow reduction caused by activeconstriction of vessels under neural control. Besides theabove results, it must be remembered that the negativeBOLD response may also arise from correlated noise orhead motion or as an EPI phase encoding artifact [19].Hence, it is necessary to rule out these possibilities beforeattributing functional significance [117].

Previous studies have also suggested that the negativesignal change is due to inhibition or suppression of neuronalfiring [107]. Such task-induced deactivation has beenobserved in many neuroimaging studies [12,118] andimplies that there is more blood flow during ‘‘rest’’ thanduring the task. Some of these task-induced deactivationsare not specific to the task, stimuli, or imaging modality[12,51,92,118]. Common regions in the brain includeposterior cingulate cortex, precuneus, rostral anterior cingu-late, and orbitofrontal cortex. These task-independentdecreases indicate that there is a structured pattern ofactivation in several brain areas during rest and that thisactivity is attenuated during various goal-directed behaviors[51,108]. The crucial idea here is that ‘‘rest’’ does notnecessarily mean a period of inactivity, but a cognitivelyrich state characterized by a variety of possible attention-dependent processes. These attention demanding processesinclude verbal and visual imagery, planning and problemsolving, monitoring the external environment, monitoringthe internal sensory state and body image, monitoringemotional state, episodic memory encoding and retrieval


and working memory (see [51] for a review and [92]). Theevolutionary implications and the need for such processes inthe resting state are quite evident. Areas that deactivateduring a task do so because other actively involved areas(specific to the task) require attentional input to stay active.Thus, when attentional resources are relocated from areas,which are active during rest, they become deactivated.McKiernan et al. [92] tested this hypothesis by parametri-cally manipulating task difficulty and found that taskdifficulty indeed determined the extent of resources to beallocated and hence, the amount of deactivation seen.

To summarize, current evidence suggests that thenegative BOLD response arises out of a decrease in neuralactivity and a consequent decrease in CBF. In most cases,the decrease in activity seems to be coupled with neuralactivity in the positive BOLD regions, such that the decreasein activity follows inhibition from those active regions. Ininterpreting an observed negative BOLD signal, researchersshould first rule out circumstances that alter or affect thehemodynamic response such as anesthesia and drugs, headmotion, or other artifacts. One should then look for co-activations (positive and negative BOLD) of different brainregions and see how the activation pattern fits with therealm of the cognitive demands of the task and thecytoarchitecture/connectivity of these regions. Such anapproach may offer us broader and holistic views ofstructure–function relationship of different brain regions.

6. Resting state fluctuations

Knowledge of the functional connectivity of differentbrain regions is crucial to the understanding of informationprocessing in the brain and the associated neural dynamics.In fact, functional connectivity serves as a gateway inlinking human brain to behavior. Recently, there has beengrowing interest in assessing functional connectivity notonly during active cognitive tasks, but also during theresting state [27,129,133] when subjects perform no specificcognitive task. The essential idea is that even during theresting state, voxels of spatially separate regions that areinterconnected show a high degree of correlated, sponta-neous activity over time. The traditional method to assessfunctional connectivity maps for a specific region of interestin fMRI is the ‘‘seed voxel’’ approach in which a smallcluster of voxels is chosen, whose averaged time courseserves as a reference model for cross-correlation analysiswith the remaining brain voxels, to yield a spatial zero-lagcross-correlation map [14]. The seed voxel is usuallyselected using anatomical landmarks or functional knowl-edge (e.g., center of mass of significantly active clusters).Using this approach, studies have provided functionalconnectivity maps for bilateral auditory, visual and motorcortices [27,84,133]. In their study, Cordes et al. [27] lookedat the frequency composition of the averaged time course invoxels and reported that low-frequency fluctuations (<0.1

Hz) contributed to the temporal structure of maps in allcortical regions studied (including the motor, auditory, andvisual cortices). A recent study by van de Ven et al. [129]used spatial independent component analysis (sICA), whichessentially allows separation of the voxel time series intoindependent spatial components with a unique time course,to create functional connectivity maps during the restingstate. One of the limitations of these studies however, is thatthese maps do not provide any information on the directionof connectivity, as the maps are based on zero-lagcovariation over time. Useful interpretation is also limitedby the fact that cortical analysis ignores the role of sub-cortical structures [129]. Nevertheless, results of thesestudies coupled with techniques such as diffusion tensorimaging (DTI) that offer details on the trajectories of whitematter tracts, may provide better insights into the functionalrelevance of such connectivity maps.

Above, I have discussed resting-state functional con-nectivity. It should be noted that functional connectivity isdefined as ‘‘the statistical dependencies among remoteneurophysiological measurements’’. Therefore, functionalconnectivity is a much broader concept than covered byresting state analyses. Although resting-state functionalconnectivity is an interesting, if poorly understood, phe-nomena, most analyses of functional or effective connec-tivity use correlations that are induced by experimentaldesign.

7. Other influences on cerebral blood flow

Although the relation between neural metabolism andcerebral blood flow has been established, what exactlycauses blood to flow to the active neural tissue is stillunclear. Under normal physiological conditions, a varietyof factors are responsible for maintenance of cerebral bloodflow. In fact, at certain ranges of systemic perfusionpressures (50–150 mm Hg), CBF remains relativelyconstant and this is called autoregulation. The factorsinfluencing autoregulation [65,75] include metabolic (H+,K+, adenosine), myogenic (stretch-dependent constrictionand dilatation), endothelial, perivascular (sympathetic andparasympathetic), and other vasoactive substances (seroto-nin, histamine, NO). As can be seen, the regulation ofcerebral blood flow is much more complex than the effectof O2 and CO2 on cerebral vessels. So, during neuralactivity, which is to a large extent modulated by the natureof the task, isolated or combined effect of some of thesefactors plays a role in determining the blood flow response.For instance, tasks that involve perception of emotionspossibly activate the sympathetic system. It is as yetunclear whether activation of the sympathetic systemresults in a local or generalized effect on the brainvasculature. This makes identification of active neuralareas quite challenging and warrants careful interpretationof BOLD signals.


8. Implications for the study of clinical disorders

Over the last couple of years, fMRI has been increasinglyused in clinics especially to study neurological and psychi-atric diseases and also for pre-operative evaluation ofpatients with intractable epilepsy to determine the lateraliza-tion of language and memory and localizing the epilepticfocus [54]. With an ever-expanding use of fMRI in severaldomains of Cognitive Neuroscience, Clinical Neurology, andPsychiatry, an exhaustive list of clinical applications willalways be incomplete. Having reviewed the basis of theBOLD signal, I would like to leave a note of cautionregarding the use of fMRI in Neurological and Psychiatricdisorders and its interpretation, especially when used in casesof compromised cerebrovascular reserve capacity and bloodflow. A good example is the use of BOLD fMRI to studystroke patients. It is known that during the acute phase ofstroke, there is profound mismatch between cerebralmetabolism and blood flow, with varying degrees of relativeor absolute hyperemia [32]. This means that cerebral bloodflow does not reflect cerebral function in the early phase ofinjury. Moreover, there are associated initial vascular adjust-ments such as luxury perfusion, altered vasoreactivity, peri-infarct edema, and diaschisis [8,9,101,115]. Hence, BOLDfMRI studies in stroke are typically done a few months afterthe initial vascular changes settle down. Similar issues are ofconcern when BOLD fMRI is used in patients on medi-cations that affect blood flow and perfusion. Hamzei et al.[55] recently verified in a group of patients with intra- andextracranial artery disease that they not only had reducedcerebrovascular reserve capacity but also showed a reducedor negative BOLD response, when compared to healthyadults. Overall, when vascular responses are compromised,the difficulties in interpreting the BOLD signal arise mainlywhen comparing between patients. The differential responsesin any brain region within one patient, for example, whichdefine lateralization of language, are still interpretable in astatistical if not quantitative sense.

9. Concluding remarks

A review of the basic mechanisms involved in generationof the BOLD signal, tells us that this neural correlate is verycomplex and several factors should be borne in mind wheninterpreting its functional significance. First, it is notsufficient that one localizes the brain areas ‘‘involved’’ ina task. Localizing neural activity alone does not give us thecomplete picture of the neural processes involved in the task[74]. To improve the information yield of a study, oneshould first focus on the design of the experiment so that thekey hypotheses under consideration could be verified. Theexperimental protocol, including the scanning method andthe spatiotemporal resolution of the signal, is decided by thenature of the task and the questions addressed in theexperiment. Also of relevance are issues of data analysis.

That the results and functional interpretation are sensitive tothese differences can be attested by anyone familiar withfMRI data analysis. Although the analysis of fMRI data isquite specialized and researchers use different analysissoftware, methods, and statistics, there is a fair degree ofconsistency in laboratories around the world. Almostuniversally, people use some form of statistical parametricmapping. Statistical parametric maps are images of statisticsthat test for activations at each voxel. Usually, the statisticalmodel is a convolution model that assumes a linear mappingbetween neuronal activity and the hemodynamic response.Since subtraction of activity is the most widely used methodto identify brain areas active in one condition relative toanother, the task pairs should ideally differ in only oneparameter. This however assumes that changing oneparameter of the task alters only one aspect of process-ing—an assumption that is not necessarily true in majorityof situations. A related issue is the choice of the controlcondition. Especially during the early days of functionalimaging, researchers compared an active task condition with‘‘rest’’ where the subject does nothing. However, one flawthat is inherent in this approach is that although subjects donot execute an overt task during rest, their neural activity isfar from the baseline required for the subtraction analysis tobe meaningful. Brain activity in different participants differdepending on their definition of rest and is hence highlysubjective. One alternative would be to engage subjects in atask that occupies them, but does not influence the task/mental construct under study.

Finally, how we interpret fMRI data holds the key to ourunderstanding of the function of different brain areas. Areview of the current fMRI literature does show that severalstudies adopt an unhealthy practice of attributing functionalsignificance to brain areas piecewise. If one knows theprocesses implemented in a particular brain area for sure,one can argue that activation of this area in response to atask, is evidence that these processes were employed whileperforming the task. But there may not exist a one to onerelationship between a task and a brain region, even forspecific areas of the cortex such as the primary motorcortex. More and more evidence points to the brain as ahighly interconnected, spatiotemporal dynamic system thatuses distributed representational schemes [35,44,52,67].Hence, it is likely that for every task, there are brain areasunique to the task as well as neural resources shared bydifferent task components. Different brain areas may also beengaged and disengaged over time. It is quite reasonable toinfer that the activated brain areas form a neural system thatis sufficient to carry out the task. Nonetheless, this may notbe the only system that subserves this function; severalparallel systems may be involved. Several researchers hencebelieve that while lesion studies reveal brain areas necessaryfor the task, functional imaging studies point out the areassufficient for executing the task. The overall context inwhich the task is administered, the strategies that subjectsuse and the cognitive demands of the task should be borne


in mind when we interpret the ‘‘hot spots’’. A holisticapproach based on the tenet that the whole is more than thesum of the parts may be necessary. Furthermore, one gainsmore information when fMRI results are interpreted inrelation to behavioral measures and results of other imagingtechniques such as EEG and MEG.

Although fMRI has been criticized as a million dollarrelative of Phrenology, the last decade has seen severaladvances in imaging techniques and protocols, whichyielded more useful information than was initially thoughtpossible. The main contribution of fMRI is the wealth ofinformation this technology provides about the structure–function relationship of brain areas and pathways. This maygo well beyond the immediate implications of such resultsas identifying brain areas responsible for executing certainfunctions and indeed help us in our pursuit for improvementof diagnosis and therapy of diseases. As with any inves-tigative tool in its infancy, there are limitations in its use andambiguities in interpretation of data. But this thought shouldnot deter us from having hypotheses about functions of thehuman brain and testing them—we need to go forward.

Acknowledgments

Major part of this review was written while I was at theCenter for Complex Systems and Brain Sciences at FloridaAtlantic University and was supported by NIMH grantsMH42900 and MH01386 to Dr. J.A. Scott Kelso. I thankDrs. Kelso, Armin Fuchs and K.J. Jantzen for theirencouragement and suggestions on the earlier versions ofthe paper. Comments from Dr. Gottfried Schlaug at BethIsrael Deaconess Medical Center, Harvard Medical Schoolalso helped me improve this paper.

References

[1] D.C. Alsop, J.A. Detre, Reduced transit-time sensitivity in non-

invasive magnetic resonance imaging of human cerebral blood flow,

J. Cereb. Blood Flow Metab. 16 (1996) 1236–1249.

[2] O.J. Arthurs, S. Boniface, How well do we understand the neural

origins of the fMRI BOLD signal? Trends Neurosci. 25 (2002)

27–31.

[3] O.J. Arthurs, S. Boniface, What aspect of the fMRI BOLD signal

best reflects the underlying electrophysiology in human somatosen-

sory cortex? Clin. Neurophysiol. 114 (2003) 1203–1209.

[4] O.J. Arthurs, E.J. Williams, T.A. Carpenter, J.D. Pickard, S.J.

Boniface, Linear coupling between functional magnetic resonance

imaging and evoked potential amplitude in human somatosensory

cortex, Neuroscience 101 (2000) 803–806.

[5] D. Attwell, C. Iadecola, The neural basis of functional brain imaging

signals, Trends Neurosci. 25 (2002) 621–625.

[6] D. Attwell, S.B. Laughlin, An energy budget for signaling in the

gray matter of the brain, J. Cereb. Blood Flow Metab. 21 (2001)

1133–1145.

[7] A. Aubert, R. Costalat, A model of the coupling between brain

electrical activity, metabolism, and hemodynamics: application to the

interpretation of functional neuroimaging, NeuroImage 17 (2002)

1162–1181.

[8] M.M. Bahn, A.B. Oser, D.T. Cross, CT and MRI of stroke, J. Magn.

Res. Imaging 6 (1996) 833–845.

[9] A.E. Baird, S. Warach, Magnetic resonance imaging of acute stroke,

J. Cereb. Blood Flow Metab. 18 (1996) 583–609.

[10] P.A. Bandettini, The temporal resolution of functional MRI, in:

C.T.W. Moonen, P.A. Bandettini (Eds.), Functional MRI, Springer

Verlag, 2000, pp. 205–220.

[11] P.S. Bellgowan, Z.S. Saad, P.A. Bandettini, Understanding neural

system dynamics through task modulation and measurement of

functional MRI amplitude, latency, and width, Proc. Natl. Acad. Sci.

U. S. A. 100 (2003) 1415–1419.

[12] J.R. Binder, J.A. Frost, T.A. Hammeke, P.S.F. Bellgowan, S.M.

Rao, R.W. Cox, Conceptual processing during the conscious

resting state, a functional MRI study, J. Cogn. Neurosci. 11

(1999) 80–93.

[13] R.M. Birn, Z.S. Saad, P.A. Bandettini, Spatial heterogeneity of the

nonlinear dynamics in the FMRI BOLD response, NeuroImage 14

(2001) 817–826.

[14] B.B. Biswal, F.Z. Yetkin, V.M. Haughton, J.S. Hyde, Functional

connectivity in the motor cortex of resting human brain using echo-

planar MRI, Magn. Reson. Med. 34 (1995) 537–541.

[15] P.G. Bittar, Y. Charnay, L. Pellerin, C. Bouras, P.J. Magistretti,

Selective distribution of lactate dehydrogenase isoenzymes in

neurons and astrocytes of human brain, J. Cereb. Blood Flow Metab.

16 (1996) 1079–1089.

[16] G.M. Boynton, S.A. Engel, G.H. Glover, D.J. Heeger, Linear

systems analysis of functional magnetic resonance imaging in

human V1, J. Neurosci. 16 (1996) 4207–4221.

[17] R.L. Buckner, The hemodynamic inverse problem: making infer-

ences about neural activity from measured MRI signals, Proc. Natl.

Acad. Sci. U. S. A. 100 (2003) 2177–2179.

[18] R.L. Buckner, P.A. Bandettini, K.M. O’Craven, R.L. Savoy, S.E.

Petersen, M.E. Raichle, B.R. Rosen, Detection of cortical activation

during averaged single trials of a cognitive task using functional

magnetic resonance imaging, Proc. Natl. Acad. Sci. U. S. A. 93

(1996) 14878–14883.

[19] R.B. Buxton, Introduction to Functional Magnetic Resonance

Imaging: Principles and Techniques, Cambridge Univ. Press, 2001.

[20] R.B. Buxton, The elusive initial dip, NeuroImage 13 (2001)

953–958.

[21] R.B. Buxton, E.C. Wong, L.R. Frank, Dynamics of blood flow and

oxygenation changes during brain activation: the balloon model,

Magn. Reson. Med. 39 (1998) 855–864.

[22] K. Caesar, L. Gold, M. Lauritzen, Context sensitivity of activity-

dependent increases in cerebral blood flow, Proc. Natl. Acad. Sci.

U. S. A. 100 (2003) 4239–4244.

[23] K. Caesar, K. Thomsen,M. Lauritzen, Dissociation of spikes, synaptic

activity, and activity-dependent increments in rat cerebellar blood

flow by tonic synaptic inhibition, Proc. Natl. Acad. Sci. U. S. A. 26

(2003) 16000–16005.

[24] C. Calautti, F. Leroy, J.Y. Guincestre, J.C. Baron, Dynamics of motor

network overactivation after striatocapsular stroke: a longitudinal

PET study using a fixed-performance paradigm, Stroke 32 (2001)

2534–2542.

[25] D. Chawla, E.D. Lumer, K.J. Friston, Relating macroscopic measures

of brain activity to fast, dynamic neuronal interactions, Neural

Comput. 12 (2000) 2805–2821.

[26] V.P. Clark, J.M. Maisogm, J.V. Haxby, fMRI study of face perception

and memory using random stimulus sequences, J. Neurophysiol. 79

(1998) 3257–3265.

[27] D. Cordes, V.M. Haughton, K. Arfanakis, J.D. Carew, P.A. Turski,

C.H. Moritz, M.A. Quigley, M.E. Meyerand, Frequencies contribu-

ting to functional connectivity in the cerebral cortex in ‘‘resting-

state’’ data, Am. J. Neuroradiol. 22 (2001) 1326–1333.

[28] S.C. Cramer, Stroke recovery: lessons from functional MR imaging

and other methods of human brain mapping, Phys. Med. Rehabil.

Clin. North Am. 10 (1999) 875–886.


[29] O. Creutzfeldt, Brain work: the coupling of function, metabolism and

blood flow in the brain, in: D. Ingvar, N.A. Lassen (Eds.), Alfred

Benzon Symposium VIII, Munksgaard, Academic Press, Copenha-

gen, 1975, pp. 22–47.

[30] A.M. Dale, R.L. Buckner, Selective averaging of rapidly presented

individual trials using fMRI, Hum. Brain Mapp. 5 (1997) 329–340.

[31] T.L. Davis, K.K. Kwong, R.M. Weisskoff, B.R. Rosen, Calibrated

functional MRI: mapping the dynamics of oxidative metabolism,

Proc. Natl. Acad. Sci. U. S. A. 95 (1998) 1834–1839.

[32] V. DiPiero, F.M. Chollet, P. MacCarthy, G.L. Lenzi, R.S.J. Frack-

owiak, Motor recovery after acute ischemic stroke, a metabolic study,

J. Neurol., Neurosurg. Psychiatry 55 (1992) 990–996.

[33] T.Q. Duong, D.S. Kim, K. Ugurbil, S.-G. Kim, Localized cerebral

blood flow response at submillimeter columnar resolution, Proc.

Natl. Acad. Sci. U. S. A. 98 (2001) 10904–10909.

[34] T.Q. Duong, E. Yacoub, G. Adriany, X. Hu, K. Ugurbil, J.T.

Vaughan, H. Merkle, S.-G. Kim, High-resolution, spin-echo BOLD,

and CBF fMRI at 4 and 7 T, Magn. Reson. Med. 48 (2002) 589–593.

[35] G.M. Edelman, G. Tononi, A Universe of Consciousness, How

Matter Becomes Imagination, Basic Books, 2000.

[36] R.R. Edelman, B. Siewert, D.G. Darby, V. Thangaraj, A.C. Nobre,

M.M. Mesulam, S. Warach, Qualitative mapping of cerebral blood

flow and functional localization with echo-planar MR imaging and

signal targeting with alternating radio frequency, Radiology 192

(1994) 513–520.

[37] S.A. Engel, D.E. Rumelhart, B.A. Wandell, A.T. Lee, G.H. Glover,

E.J. Chichilnisky, M.N. Shadlen, fMRI of human visual cortex,

Nature 369 (1994) 525.

[38] T. Ernst, J. Hennig, Observation of a fast response in functional MR,

Magn. Reson. Med. 32 (1994) 146–149.

[39] A. Feydy, R. Carlier, A. Roby-Brami, B. Bussel, F. Cazalis, L. Pierot,

Y. Burnod, M.A. Maier, Longitudinal study of motor recovery after

stroke: recruitment and focusing of brain activation, Stroke 33 (2002)

1610–1617.

[40] P.T. Fox, M.E. Raichle, Focal physiologic uncoupling of cerebral

blood flow and oxidative metabolism during somatosensory stim-

ulation in human subjects, Proc. Natl. Acad. Sci. U. S. A. 83 (1986)

1140–1144.

[41] P.T. Fox, M.E. Raichle, M.A. Mintun, C. Dence, Non-oxidative

glucose consumption during focal physiologic neural activity,

Science 241 (1988) 462–464.

[42] J. Frahm, K.D. Merboldt, W. Hanicke, A. Kleinschmidt, H. Boecker,

Brain or vein-oxygenation or flow? On signal physiology in

functional MRI of human brain activation, NMR Biomed. 7 (1994)

45–53.

[43] P. Fries, J.H. Reynolds, A.E. Rorie, R. Desimone, Modulation of

oscillatory neuronal synchronization by selective visual attention,

Science 291 (2001) 1560–1563.

[44] K.J. Friston, The labile brain: I. Neuronal transients and nonlinear

coupling, Philos. Trans. R. Soc. Lond., B Biol. Sci. 355 (2000)

215–236.

[45] K.J. Friston, P. Fletcher, O. Josephs, A. Holmes, M.D. Rugg, R.

Turner, Event-related fMRI, characterizing differential responses,

NeuroImage 7 (1998) 30–40.

[46] K.J. Friston, O. Josephs, G. Rees, R. Turner, Nonlinear event-related

responses in fMRI, Magn. Reson. Med. 39 (1998) 41–52.

[47] K.J. Friston, A. Mechelli, R. Turner, C.J. Price, Nonlinear responses

in fMRI: the balloon model, Volterra kernels, and other hemody-

namics, NeuroImage 12 (2000) 466–477.

[48] K.J. Friston, L. Harrison, W. Penny, Dynamic causal modeling,

NeuroImage 19 (2003) 1273–1302.

[49] R.D. Frostig, E.E. Lieke, D.Y. Ts’o, A. Grinvald, Cortical func-

tional architecture and local coupling between neuronal activity and

the microcirculation revealed by in vivo high-resolution optical

imaging of intrinsic signals, Proc. Natl. Acad. Sci. U. S. A. 87

(1990) 6082–6086.

[50] N. Gaab, C. Gaser, T. Zaehle, L. Jancke, G. Schlaug, Functional

anatomy of pitch memory—An fMRI study with sparse temporal

sampling, NeuroImage 19 (2003) 1417–1426.

[51] D.A. Gusnard, M.E. Raichle, Searching for a baseline, functional

imaging and the resting human brain, Nat. Rev., Neurosci. 2 (2001)

685–694.

[52] H. Haken, Principles of Brain Functioning, a Synergetic Approach to

Brain Activity, Behavior and Cognition, Springer Verlag, 1996.

[53] D.A. Hall, M.P. Haggard, M.A. Akeroyd, A.R. Palmer, A.Q.

Summerfield, M.R. Elliott, E.M. Gurney, R.W. Bowtell, Sparse

temporal sampling in auditory fMRI, Hum. Brain Mapp. 7 (1999)

213–223.

[54] T.A. Hammeke, Functional MRI in neurology, in: C.T.W. Moonen,

P.A. Bandettini (Eds.), Functional MRI, Springer Verlag, 2000.

[55] F. Hamzei, R. Knab, C. Weiller, J. Rother, The influence of extra- and

intracranial artery disease on the BOLD signal in FMRI, NeuroImage

20 (2003) 1393–1399.

[56] N. Harel, S.P. Lee, T. Nagaoka, D.S. Kim, S.-G. Kim, Origin of

negative blood oxygenation level-dependent fMRI signals, J. Cereb.

Blood Flow Metab. 22 (2002) 908–917.

[57] R.V. Harrison, N. Harel, J. Panesar, R.J. Mount, Blood capillary

distribution correlates with hemodynamic-based functional imaging

in the cerebral cortex, Cereb. Cortex 3 (2002) 225–233.

[58] D.J. Heeger, D. Ress, What does fMRI tell us about neuronal

activity? Nat. Rev., Neurosci. 3 (2002) 142–151.

[59] D.J. Heeger, G.M. Boynton, J.B. Demb, E. Seidemann, W.T.

Newsome, Motion opponency in visual cortex, J. Neurosci. 19

(1999) 7162–7174.

[60] D.J. Heeger, A.C. Huk, W.S. Geisler, D.G. Albrecht, Spikes versus

BOLD, what does neuroimaging tell us about neuronal activity? Nat.

Neurosci. 3 (2000) 631–633.

[61] X. Hu, T.H. Le, K. Ugurbil, Evaluation of the early response in fMRI

in individual subjects using short stimulus duration, Magn. Reson.

Med. 37 (1997) 877–884.

[62] X. Hu, E. Yacoub, T.H. Le, E.R. Cohen, K. Ugurbil, Functional MRI

signal decrease at the onset of stimulation, in: C.T.W. Moonen, P.A.

Bandettini (Eds.), Functional MRI, Springer Verlag, 2000.

[63] F. Hyder, J.R. Chase, K.L. Behar, G.F. Mason, M. Siddeek, D.L.

Rothman, R.G. Shulman, Increased tricarboxylic acid cycle flux in

rat brain during forepaw stimulation detected with 1H[13C]NMR,


[64] F. Hyder, D.L. Rothman, R.G. Shulman, Total neuroenergetics

support localized brain activity: implications for the interpreta-

tion of FMRI, Proc. Natl. Acad. Sci. U. S. A. 99 (2002)

10771–10776.

[65] C. Iadecola, Neurovascular regulation in the normal brain and in

Alzheimer’s disease, Nat. Rev., Neurosci. 5 (2004) 347–360.

[66] M. Jones, J. Berwick, D. Johnston, J. Mayhew, Concurrent optical

imaging spectroscopy and laser-Doppler flowmetry, the relationship

between blood flow, oxygenation, and volume in rodent barrel

cortex, NeuroImage 13 (2001) 1002–1015.

[67] J.A.S. Kelso, Dynamic Patterns, the Self-Organization of Brain and

Behavior, MIT Press, Cambridge, 1995.

[68] S.-G. Kim, Quantification of regional cerebral blood flow change by

flow-sensitivity alternating inversion recovery (FAIR) technique:

application to functional mapping, Magn. Reson. Med. 34 (1995)

293–301.

[69] S.-G. Kim, T.Q. Duong, Mapping cortical columnar structures using

fMRI, Physiol. Behav. 77 (2002) 641–644.

[70] S.-G. Kim, K. Hendrich, X. Hu, H. Merkle, K. Ugurbil, Potential

pitfalls of functional MRI using conventional gradient-recalled echo

techniques, NMR Biomed. 7 (1994) 69–74.

[71] D.S. Kim, T.Q. Duong, S.-G. Kim, High-resolution mapping of iso-

orientation columns by fMRI, Nat. Neurosci. 3 (2000) 164–199.

[72] D.S. Kim, I. Ronen, C. Olman, S.-G. Kim, K. Ugurbil, L.J. Toth,

Spatial relationship between neuronal activity and BOLD functional

MRI, NeuroImage 21 (2004) 876–885.

[73] S.-G. Kim, S.P. Lee, G. Goodyear, A.C. Dilva, Spatial resolution of


BOLD and other fMRI techniques, in: C.T.W. Moonen, P.A.

Bandettini (Eds.), Functional MRI, Springer Verlag, 2000.

[74] S.M. Kosslyn, If neuroimaging is the answer, what is the

question? Philos. Trans. R. Soc. Lond., B Biol. Sci. 354 (1999)

1283–1294.

[75] W. Kuschinsky, Regulation of cerebral blood flow, in: C.T.W.

Moonen, P.A. Bandettini (Eds.), Functional MRI, Springer Verlag,

2000.

[76] K.K. Kwong, J.W. Belliveau, D.A. Chesler, I.E. Goldberg, R.M.

Weisskoff, B.P. Poncelet, D.N. Kennedy, B.E. Hoppel, M.S. Cohen,

R. Turner, H.-M. Cheng, T.J. Brady, B.R. Rosen, Dynamic magnetic

resonance imaging of human brain activity during primary sensory

stimulation, Proc. Natl. Acad. Sci. U. S. A. 89 (1992) 5675–5679.

[77] S. Lai, A.L. Hopkins, E.M. Haacke, D. Li, B.A. Wasserman, P.

Buckley, L. Friedman, H. Meltzer, P. Hedera, R. Friedland,

Identification of vascular structures as a major source of signal

contrast in high resolution 2D and 3D functional activation imaging

of the motor cortex at 1.5 T, preliminary results, Magn. Reson. Med.

30 (1993) 387–392.

[78] M. Lauritzen, Relationship of spikes, synaptic activity, and local

changes of cerebral blood flow, J. Cereb. Blood Flow Metab. 21

(2001) 1367–1383.

[79] A.T. Lee, G.H. Glover, C.H. Meyer, Discrimination of large venous

vessels in time-course spiral blood-oxygen-level-dependent mag-

netic-resonance functional neuroimaging, Magn. Reson. Med. 33

(1995) 745–754.

[80] U. Lindauer, G. Royl, C. Leithner, M. Kuhl, L. Gold, J. Gethmann,

M. Kohl-Bareis, A. Villringer, U. Dirnagl, No evidence for early

decrease in blood oxygenation in rat whisker cortex in response to

functional activation, NeuroImage 13 (2001) 988–1001.

[81] N.K. Logothetis, The neural basis of the blood-oxygen-level-depend-

ent functional magnetic resonance imaging signal, Philos. Trans. R.

Soc. Lond., B Biol. Sci. 357 (2002) 1003–1037.

[82] N.K. Logothetis, B.A. Wandell, Interpreting the BOLD signal, Annu.

Rev. Physiol. 66 (2004) 735–769.

[83] N.K. Logothetis, J. Pauls, M. Augath, T. Trinath, A. Oeltermann,

Neurophysiological investigation of the basis of the fMRI signal,

Nature 412 (2001) 150–157.

[84] M.J. Lowe, B.J. Mock, J.A. Sorenson, Functional connectivity in

single multislice echoplanar imaging using resting-state fluctuations,

NeuroImage 7 (1998) 119–132.

[85] P.J. Magistretti, L. Pellerin, Cellular mechanisms of brain energy

metabolism and their relevance to functional brain imaging, Philos.

Trans. R. Soc. Lond., B Biol. Sci. 354 (1999) 1155–1163.

[86] P.J. Magistretti, L. Pellerin, Regulation of cerebral energy metabo-

lism, in: C.T.W. Moonen, P.A. Bandettini (Eds.), Functional MRI,

Springer Verlag, 2000.

[87] D. Malonek, A. Grinvald, Interactions between electrical activity

and cortical microcirculation revealed by imaging spectroscopy,

implications for functional brain mapping, Science 272 (1996)

551–554.

[88] J.B. Mandeville, J.J. Marota, C. Ayata, G. Zaharchuk, M.A.

Moskowitz, B.R. Rosen, R.M. Weisskoff, Evidence of a cerebrovas-

cular postarteriole windkessel with delayed compliance, J. Cereb.

Blood Flow Metab. 19 (1999) 679–689.

[89] R.S. Marshall, G.M. Perera, R.M. Lazar, J.W. Krakauer, R.C.

Constantine, R.L. DeLaPaz, Evolution of cortical activation during

recovery from corticospinal tract infarction, Stroke 31 (2000)

656–661.

[90] E. Martin, T. Thiel, P. Joeri, T. Loenneker, D. Ekatodramis, T.

Huisman, J. Hennig, V.L. Marcar, Effect of pentobarbital on visual

processing in man, Hum. Brain Mapp. 10 (2000) 132–139.

[91] G. McCarthy, M. Luby, J. Gore, P. Goldman-Rakic, Infrequent events

transiently activate human prefrontal and parietal cortex as measured

by functional MRI, J. Neurophysiol. 77 (1997) 1630–1634.

[92] K.A. McKiernan, J.N. Kaufman, J.K. Thompson, J.R. Binder, A

parametric manipulation of factors affecting task-induced deactiva-

tion in functional neuroimaging, J. Cogn. Neurosci. 15 (2003)

394–408.

[93] A. Mechelli, C.J. Price, K.J. Friston, Nonlinear coupling between

evoked rCBF and BOLD signals: a simulation study of hemody-

namic responses, NeuroImage 14 (2001) 862–872.

[94] R.S. Menon, B.G. Goodyear, Submillimeter functional localization in

human striate cortex using BOLD contrast at 4 Tesla, implications for

the vascular point-spread function, Magn. Reson. Med. 41 (1999)

230–235.

[95] R.S. Menon, S.-G. Kim, Spatial and temporal limits in cognitive

neuroimaging with FMRI, Trends Cogn. Sci. 3 (1999) 207–216.

[96] R.S. Menon, S. Ogawa, D.W. Tank, K. Ugurbil, 4Tesla gradient

recalled echo characteristics of photic stimulation-induced signal

changes in the human primary visual cortex, Magn. Reson. Med. 30

(1993) 380–386.

[97] R.S. Menon, S. Ogawa, X. Hu, J.P. Strupp, P. Anderson, K.

Ugurbil, BOLD based functional MRI at 4 Tesla includes a

capillary bed contribution, echo-planar imaging correlates with

previous optical imaging using intrinsic signals, Magn. Reson. Med.

33 (1995) 453–459.

[98] R.S. Menon, S. Ogawa, J.P. Strupp, K. Ugurbil, Ocular dominance in

human V1 demonstrated by functional magnetic resonance imaging,

J. Neurophysiol. 77 (1997) 2780–2787.

[99] K.L. Miller, W.M. Luh, T.T. Liu, A. Martinez, T. Obata, E.C. Wong,

L.R. Frank, R.B. Buxton, Nonlinear temporal dynamics of the

cerebral blood flow response, Hum. Brain Mapp. 13 (2001) 1–12.

[100] S. Morgello, R.R. Uson, E.J. Schwartz, R.S. Haber, The human

blood–brain barrier glucose transporter (GLUT1) is a glucose

transporter of gray matter astrocytes, Glia 14 (1995) 43–54.

[101] R.J. Nudo, E.J. Plautz, S.B. Frost, Role of adaptive plasticity in

recovery of function after damage to motor cortex, Muscle Nerve 24

(2001) 1000–1019.

[102] T. Obata, T.T. Liu, K.L. Miller, W.M. Luh, E.C. Wong, L.R. Frank,

R.B. Buxton, Discrepancies between BOLD and flow dynamics in

primary and supplementary motor areas: application of the balloon

model to the interpretation of BOLD transients, NeuroImage 21

(2004) 144–153.

[103] S. Ogawa, T.M. Lee, A.R. Kay, D.W. Tank, Brain magnetic

resonance imaging with contrast dependent on blood oxygenation,


[104] L. Pellerin, P.J. Magistretti, Glutamate uptake in astrocytes stimulates

aerobic glycolysis, a mechanism coupling neuronal activity to

glucose utilization, Proc. Natl. Acad. Sci. U. S. A. 91 (1994)

10625–10629.

[105] J. Pfeuffer, J.C. McCullough, P.F. Van de Moortele, K. Ugurbil, X.

Hu, Spatial dependence of the nonlinear BOLD response at short

stimulus duration, NeuroImage 18 (2003) 990–1000.

[106] D. Prichard, E. Rothman, E. Novotny, T. Petroff, M. Kuwabara, A.

Avison, C. Howseman, S.R. Hanstock, Lactate rise detected by 1H

NMR in human visual cortex during physiologic stimulation, Med.

Sci. 88 (1991) 5829–5831.

[107] M.E. Raichle, Behind the scenes of functional brain imaging, a

historical and physiological perspective, Proc. Natl. Acad. Sci.

U. S. A. 95 (1998) 765–772.

[108] M.E. Raichle, A.M. MacLeod, A.Z. Snyder, W.J. Powers, D.A.

Gusnard, A default mode of brain function, Proc. Natl. Acad. Sci.

U. S. A. 98 (2001) 676–682.

[109] G. Rees, K. Friston, C. Koch, A direct quantitative relationship

between the functional properties of human and macaque V5, Nat.

Neurosci. 3 (2000) 716–723.

[110] W. Richter, P.M. Andersen, A.P. Georgopoulos, S.-G. Kim, Sequen-

tial activity in human motor areas during a delayed cued finger

movement task studied by time-resolved fMRI, NeuroReport 8

(1997) 1257–1261.

[111] W. Richter, K. Ugurbil, A. Georgopoulos, S.-G. Kim, Time-resolved

fMRI of mental rotation, NeuroReport 8 (1997) 3697–3702.

[112] M.D. Robinson, L.A. Dowd, Heterogeneity and functional proper-


ties of subtypes of sodium dependent glutamate transporters in the

mammalian central nervous system, Adv. Pharmacol. 37 (1997)

69–115.

[113] J. Rother, R. Knab, F. Hamzei, J. Fiehler, J.R. Reichenbach, C.

Buchel, C. Weiller, Negative dip in BOLD fMRI is caused by blood

flow-oxygen consumption uncoupling in humans, NeuroImage 15

(2002) 98–102.

[114] C.S. Roy, C.S. Sherrington, On the regulation of the blood supply of

the brain, J. Physiol. 11 (1890) 85–108.

[115] G. Schlaug, A. Benfield, A.E. Baird, B. Siewert, K.O. Lovblad, R.A.

Parker, R.R. Edelman, S. Warach, The ischemic penumbra: opera-

tionally defined by diffusion and perfusion MRI, Neurology 53

(1999) 1528–1537.

[116] J. Shen, K.F. Petersen, K.L. Behar, P. Brown, T.W. Nixon, G.F.

Mason, O.A. Petroff, G.I. Shulman, R.G. Shulman, D.L. Rothman,

Determination of the rate of the glutamate/glutamine cycle in the

human brain by in vivo 13 C NMR, Proc. Natl. Acad. Sci. U. S. A. 96

(1999) 8235–8240.

[117] A. Shmuel, E. Yacoub, J. Pfeuffer, P.F. Van de Moortele, G. Adriany,

X. Hu, K. Ugurbil, Sustained negative BOLD, blood flow and

oxygen consumption response and its coupling to the positive

response in the human brain, Neuron 36 (2002) 993–995.

[118] G.L. Shulman, J.A. Fiez, M. Corbetta, R.L. Buckner, F.M. Miezin,

M.E. Raichle, S.E. Peterson, Common blood flow changes across

visual tasks: II. Decreases in cerebral cortex, J. Cogn. Neurosci. 9

(1997) 648–663.

[119] N.R. Sibson, A. Dhankhar, G.F. Mason, D.L. Rothman, K.L. Behar,

R.G. Shulman, Stoichiometric coupling of brain glucose metabolism

and glutamatergic neuronal activity, Proc. Natl. Acad. Sci. U. S. A.

95 (1998) 316–321.

[120] A.T. Smith, K.D. Singh, M.W. Greenlee, Attentional suppression

of activity in the human visual cortex, NeuroReport 11 (2000)

271–277.

[121] A.J. Smith, H. Blumenfeld, K.L. Behar, D.L. Rothman, R.G.

Shulman, F. Hyder, Cerebral energetics and spiking frequency: the

neurophysiological basis of fMRI, Proc. Natl. Acad. Sci. U. S. A. 99

(2002) 10765–10770.

[122] A.T. Smith, A.L. Williams, K.D. Singh, Negative BOLD in the visual

cortex: evidence against blood stealing, Hum. Brain Mapp. 21 (2004)

213–220.

[123] L. Sokoloff, Localization of functional activity in the central nervous

system by measurement of glucose utilization with radioactive

deoxyglucose, J. Cereb. Blood Flow Metab. 1 (1981) 7–36.

[124] L. Sokoloff, Circulation and energy metabolism of the brain, in: G.

Siegel, B. Agranoff, R.W. Albers, P. Molinoff (Eds.), Basic Neuro-

chemistry, Molecular, Cellular and Medical Aspects, 4th edR, Raven,New York, 1989.

[125] L. Sokoloff, M. Reivich, C. Kennedy, M.H. Des Roisers, C.S. Patlak,

K.D. Pettigrew, O. Sakurada, M. Shinohara, The 14C deoxyglucose

method for the measurement of local cerebral glucose utilization,

theory, procedure and normal values in the conscious and anesthe-

tized albino rat, J. Neurochem. 28 (1977) 897–916.

[126] J.K. Thompson, M.R. Peterson, R.D. Freeman, Single-neuron

activity and tissue oxygenation in the cerebral cortex, Science 299

(2003) 1070–1072.

[127] R.B.H. Tootell, N. Hadjikhani, E.K. Hall, S. Marret, W. Vanduffel,

J.T. Vaughan, A.M. Dale, The retinotopy of visual spatial attention,

Neuron 21 (1998) 1409–1422.

[128] K. Ugurbil, L. Toth, D.S. Kim, How accurate is magnetic resonance

imaging of brain function? Trends Neurosci. 26 (2003) 108–114.

[129] V.G. van de Ven, E. Formisano, D. Prvulovic, C.H. Roeder, D.E.

Linden, Functional connectivity as revealed by spatial independent

component analysis of fMRI measurements during rest, Hum. Brain

Mapp. 22 (2004) 165–178.

[130] I. Vanzetta, A. Grinvald, Evidence and lack of evidence for the initial

dip in the anesthetized rat: implications for human functional brain

imaging, NeuroImage 13 (2001) 959–967.

[131] A.L. Vazquez, D.C. Noll, Nonlinear aspects of the BOLD response in

functional MRI, NeuroImage 7 (1998) 108–118.

[132] E.C. Wong, R.B. Buxton, L.R. Frank, A theoretical and experimental

comparison of continuous and pulsed arterial spin labeling techni-

ques for quantitative perfusion imaging, Magn. Reson. Med. 40

(1998) 348–355.

[133] J. Xiong, L.M. Parsons, J.H. Gao, P.T. Fox, Interregional connectiv-

ity to primary motor cortex revealed using MRI resting state images,

Hum. Brain Mapp. 8 (1999) 151–156.

[134] M. Zonta, M.C. Angulo, S. Gobbo, B. Rosengarten, K.-A.

Hossmann, T. Pozzan, G. Carmignoto, Neuron to astrocyte signaling

is central to the dynamic control of brain microcirculation, Nat.

Neurosci. 6 (2003) 43–50.


about being bold - homepage -...

Documents