Metabolic origin of BOLD signal fluctuations in theabsence of stimuli

Masaki Fukunaga1, Silvina G Horovitz1, Jacco A de Zwart1, Peter van Gelderen1,Thomas J Balkin2, Allen R Braun3 and Jeff H Duyn1

1Advanced MRI, LFMI, NINDS, National Institutes of Health, Bethesda, Maryland, USA; 2Departmentof Behavioral Biology, Walter Reed Army Institute of Research, Silver Spring, Maryland, USA;3Language Section, Voice Speech and Language Branch, NIDCD, National Institutes of Health, Bethesda,Maryland, USA

Blood oxygen level-dependent (BOLD) functional magnetic resonance imaging studies have shownthe existence of ongoing blood flow fluctuations in the absence of stimuli. Although this so-called‘resting-state activity’ appears to be correlated across brain regions with apparent functionalrelationship, its origin might be predominantly vascular and not directly representing neuronalsignaling. To investigate this, we simultaneously measured BOLD and perfusion signals on healthyhuman subjects (n = 11) and used their ratio (BOLD/perfusion ratio or BPR) as an indicator ofmetabolic demand. BPR during rest and sleep was compared with that during a visual task (VT) anda breath-holding task (BH), which are challenges with substantial and little metabolic involvement,respectively. Within the visual cortex, BPR was 3.76±1.23 during BH, which was significantly higherthan during the VT (1.76±0.27) and rest (1.56±0.41). Meanwhile, BPR values during VT and restwere not significantly different, suggesting a similar metabolic involvement. Eight subjects showedstage 1 and 2 sleep, during which temporally correlated BOLD and perfusion activity continued. Inthese subjects, there was no significant difference in BPR between the sleep and waking conditions(1.79±0.54 and 1.66±0.67, respectively), but both were lower than the BPR during BH. These datasuggest that resting-state activity, at least in part, represents a metabolic process.Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387; doi:10.1038/jcbfm.2008.25; published online2 April 2008

Keywords: BOLD; CMRO2; connectivity; fMRI; resting state; sleep

Introduction

Mapping of brain function with blood oxygen level-dependent (BOLD) functional magnetic resonanceimaging (fMRI) relies on the detection of focalchanges in cerebral blood flow (CBF) in responseto external stimuli. These changes are believed tosupport the changes in metabolic demand related totask-induced changes in electrical and chemicalsignaling (Logothetis et al, 2001) at neuronalsynapses.

In the absence of stimuli, the brain continues toshow temporal fluctuations in blood flow, asmeasured with BOLD and perfusion-based MRI(Biswal et al, 1995, 1997). This phenomenon ofso-called ‘resting-state activity’ resembles the verylow-frequency ( < 0.1 Hz) fluctuations of sponta-neous neuronal and metabolic activity, which wereobserved in animals that were deprived of externalstimuli (Allers et al, 2002; Dora and Kovach, 1981;Leopold et al, 2003; Ruskin et al, 2003). Theobserved synchronized fluctuation across distribu-ted regions is suggestive of the underlying func-tional relationship (Biswal et al, 1995; Cordes et al,2000; Greicius et al, 2003; Hampson et al, 2002;Lowe et al, 2000). In addition, the regions showingcorrelated fluctuations are highly consistent withinand across subjects (Beckmann et al, 2005; van deVen et al, 2004), suggesting that they might allowmapping and classification of the networks thatunderlie human brain function without the need forcarefully conditioned stimuli.

Received 23 November 2007; revised 5 March 2008; accepted 6March 2008; published online 2 April 2008

Correspondence: Dr M Fukunaga, Advanced MRI, LFMI, NINDS,National Institutes of Health, 9000 Rockville Pike, MSC 1065,Building 10, Room B1D-728, Bethesda, MD 20892-1065, USA.E-mail: fukunagm@mail.nih.gov

This study was supported by the Intramural Research Program of

the National Institute of Neurological Disorders and Stroke

(National Institutes of Health).

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387& 2008 ISCBFM All rights reserved 0271-678X/08 $30.00

www.jcbfm.com

Despite their potential functional significance forneuroscience research, the origin of blood flowfluctuations in the absence of stimuli has not yetbeen established. Potential contributors that involvemodulation in electrical and chemical signaling atneuronal synapses include cognitive processes,fluctuations in vigilance and conscious awareness,and homeostatic (restorative) and plasticity pro-cesses. An alternative possibility is that, rather thanhaving a neuronal origin, the fluctuations are causedby physiologic fluctuations (e.g., vasomotor effect(Johansson and Bohr, 1966) or fluctuations in end-tidal CO2 (Wise et al, 2004)), without a substantialmetabolic or neuronal component. As such, thesefluctuations might bear little relevance to themapping of functional networks.

Because of these confounding sources, the meta-bolic contribution to resting-state activity cannot bedetermined from BOLD or perfusion-based fMRImeasurements alone. Nevertheless, several previousstudies have shown that the analysis of simulta-neously acquired BOLD and perfusion signalsduring a variety of tasks might reveal the level ofthe underlying metabolic activity (Chiarelli et al,2007a; Davis et al, 1998; Hoge et al, 1999). In thecontext of this work, we designed an experiment tosimultaneously acquire perfusion and BOLD signalsduring resting-state activity and tasks with varyingmetabolic demand, to estimate relative levels ofmetabolic activity.

Methods

Theory

A number of studies have compared the relationshipbetween BOLD and perfusion signals during metabolicallydemanding tasks, such as stimulation-evoked activity, andtasks with reduced or negligible metabolic involvement,such as hypercapnia through controlled CO2 inhalation orbreath-holding (BH) (Kety and Schmidt, 1948; Novack etal, 1953). Hoge et al (1999) found that the BOLD perfusionrelationship could be accurately modeled by

DBOLD ¼ M � ð1 � ð1 þ DCMRO2Þb � ð1 þ DCBFÞa�bÞ ð1Þ

with DBOLD, DCBF, and DCMRO2normalized to baseline

levels; M, a scaling constant dependent on MRI acquisi-tion parameters and physiologic baseline state; and thephysiologic parameters a and b estimated at 0.38 and 1.5,respectively (Davis et al, 1998). The relationship betweenDCMRO2

and DCBF was found to be close to linear(Chiarelli et al, 2007b; Hoge et al, 1999) with a propor-tionality constant k dependent on metabolic demand,leading to

DBOLD ¼ M � ð1 � ð1 þ k � DCBFÞb � ð1 þ DCBFÞa�bÞ ð2ÞIn the absence of metabolic fluctuations, k = 0, whereasmetabolic processes evoked by a visual stimulation task(VT) or motor task are associated with nonzero, positive kvalues in the range of 0.24 to 0.51 (Ances et al, 2008;Chiarelli et al, 2007a; Hoge et al, 1999; Kastrup et al, 2002;

Stefanovic et al, 2005; Uludag et al, 2004). Equation (2)indicates that k directly affects the BOLD–perfusionrelationship, as illustrated in Figure 1, for a range of kvalues. Figure 1 suggests that for a certain level ofperfusion fluctuations, the expected BOLD fluctuationsare highest in the absence of a metabolic component(k = 0). Furthermore, the relative size of BOLD andperfusion fluctuations (henceforth referred to as BOLD/perfusion ratio or BPR) varies substantially in theobserved range of k = 0 to 0.51. For example, in theapproximation DCBF 51, equation (2) simplifies to

DBOLD � M � ðb � ðk þ 1Þ þ aÞ � DCBF ð3Þwith BPR = M (b (k + 1) + a), which, for a= 0.38 and b= 1.5,ranges from 1.12 to 0.36 M in the range of k = 0 to 0.51.According to these numbers, the proportionality constantbetween BOLD and perfusion signals can be up to 3.1times higher during BH than VT. This indicates that BPRvalues can be used to estimate the metabolic involvementin resting-state activity.

Experimental Design

To determine whether spontaneous fMRI activity duringthe resting state has a substantial metabolic component,we performed simultaneous BOLD and perfusion MRIstudies on 11 volunteers (7 males, 4 females, age: 22 to 38years, average: 28 years) during a 40-min long experiment.Informed consent was obtained from all subjects beforeconducting the study, which was approved by theInstitutional Review Board at the National Institutes ofHealth under protocol number 00-N-0082.

The experiment had three separate conditions (Figure2): a VT, a BH task, and a resting state (REST). Visual andBH tasks were designed to strongly modulate fMRI signalswith substantial and small or negligible fluctuation in

∆ CBF (%)

∆ B

OLD

(%

)

BOLD–CBF dependence

k = 00.1

0.2

0.3

0.4

0.5

0 10 20 30 40 500

1

2

3

4

5

6

Figure 1 Simulated BOLD–perfusion relationship based onequation (2) using a= 0.38, b= 1.5, M = 0.2, and a range ofphysiologic k values. Note that increasing k (increasingmetabolic contribution) shifts the relationship to a lower (lesssteep) curve. Perfusion changes are expressed as a fraction ofbaseline perfusion levels. BOLD, blood oxygen level-dependent;CBF, cerebral blood flow.

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1378

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

metabolic demand, respectively. A BH task was chosenover controlled CO2 inhalation for the ease of implementa-tion. If fluctuations during REST occur in the absence ofmetabolic fluctuations, their associated fluctuations incerebral metabolic rate of oxygen (CMRO2

) should besignificantly lower than CMRO2

fluctuations during VT,and similar to CMRO2

fluctuations during BH.

Stimulus Presentation

Before the start of each study, the subject was informed ofthe details of the experiment. In addition, subjects wereasked to pay attention during VT, avoid head motion, andwere encouraged to fall asleep during REST.

The visual stimulus, presented during VT at the initialpart of the experiment (see Figure 2), lasted 320 secs andconsisted of a contrast-reversing checkerboard (7.5 Hz)stimulating either the central (0 to 51) or peripheral (5 to151) part of the visual field in eight blocks of 40 secsduration each. To ensure alertness during this part of theexperiment, and minimize saccades, subjects were in-structed to focus on a small dot that was presented in thecenter of the checkerboard display, and push a button onan fORP MRI compatible response box (Current DesignInc., Philadelphia, PA, USA) when the color of the dotchanged. Color changes were randomized in time andoccurred on average at the rate of 0.05/sec. The visualstimuli were back-projected onto a translucent screen witha U2-1200DLP data projector (PLUS Vision Corp., Tokyo,Japan). Button presses were recorded using Presentation0.80 software (Neurobehavioral Systems Inc., Albany, CA,USA), which also recorded the timing of the MRIacquisition.

The BH task consisted of five cycles (120 secs each) of aBH paradigm, starting 1 min after the end of the VT. At thebeginning of each cycle, subjects were prompted to take adeep breath, followed by 40 secs of breath-hold. Duringthe remaining 80 secs of each cycle, the subjects wereinstructed to breathe freely.

The resting period that followed BH lasted 23 mins and40 secs. Room lights were dimmed and subjects wereinstructed to close their eyes and relax. They were allowedto sleep.

Magnetic Resonance Imaging Scanning

All experiments were performed on a SIGNA VH/i 3TMRI scanner (GE, Milwaukee, WI, USA), retrofit with a16-channel receiver system. For optimal sensitivity, a16-channel detector array was used, designed in-house,and constructed by Nova Medical Inc. (Wilmington, MA,USA). The MRI acquisition method was based on atechnique described by van Gelderen et al (2005), whichallowed simultaneous measurement of BOLD and perfu-sion signals. This method allows for minimizing thecontribution of stationary tissue and BOLD signals to theperfusion measurement. Suppression of stationary tissuesignal was achieved with a double-inversion labelingmethod. Five axial-oblique slices covering the primaryvisual cortex (VC) were acquired using the followingparameters: resolution, 3.75 mm� 3.75 mm� 5.0 mm; in-version delays, 1,320 ms (TI1) and 450 ms (TI2); repetitiontime, 4 secs; echo times, 20 and 35 ms for perfusion andBOLD acquisition, respectively. For optimal temporalresolution, perfusion measurements were performed with-out reference acquisition. On three subjects, additionaldata were acquired (using an additional VT condition)with perfusion reference to allow the estimation ofchanges relative to baseline perfusion. Signal-to-noiseratio in the individual BOLD images ranged from 200:1 indeep brain structures to 500:1 in cortical areas.

Physiologic Monitoring

To monitor the physiologic and cognitive state of thesubjects, a number of physiologic signals were recorded insynchrony with MRI acquisition. These signals included

0:40 2:00

breath-

holding

free

breathing

0:00

40:0016:205:200:00

VT BH REST

Figure 2 Timeline of simultaneous blood oxygen level-dependent and perfusion functional magnetic resonance imaging experiment.The experiment starts with a 5 mins and 20 secs task protocol, in which the central and peripheral portions of the visual field arealternatingly stimulated for 40-sec periods (visual task or VT). One minute after the VT, a breath-holding (BH) task follows, with fiverepeated cycles. Each cycle consists of taking in a single deep breath, followed by 40 secs of BH and 80 secs of normal breathing.The final part of the experiment consists of approximately 23 mins of rest, during which subjects close their eyes and are allowed tofall asleep (REST).

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1379

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

electroencephalography (EEG), cardiac and respiratorycycles, and button presses (subjects’ response during VT).

Electroencephalographic data, collected using a 64-channel Neuroscan Synamps2 system (CompumedicsUSA Ltd, El Paso, TX, USA), served primarily to measurethe subjects’ alertness and to detect sleep. Signals werecollected using 40 channels of a sintered Ag/AgClelectrode cap connected via carbon fibers and RF filtersto amplifiers and digitizers. Electrodes were placed tocover most of the scalp. Two of the channels were bipolarand were used to allow monitoring of the electrooculo-gram and electrocardiogram. Acquisition of EEG wasperformed continuously, in DC mode and with a digitiza-tion rate of 10 kHz and analog low-pass filtering at 2 kHz.At each MRI repetition interval (TR), a trigger pulse fromthe scanner was recorded and used for localization of thegradient artifact (see below).

Magnetic Resonance Imaging Data Analysis

All MRI data analyses were performed using IDL 6.2software (ITT Visual Information Solutions, Boulder, CO,USA). After EPI image reconstruction, to normalize theBOLD and perfusion signal time course to the restingbaseline, both BOLD and perfusion images were dividedby the first image in the BOLD-weighted time series.Subsequently, linear drift correction and band-pass filter-ing between 0.01 and 0.1 Hz were performed on bothBOLD and perfusion data. The specific choice of pass-band was made to reduce nonlinear drift and thermalnoise, while minimally affecting the predominant fre-quencies of BOLD and perfusion signals during VT, BH,and REST. After these preprocessing steps, VT, BH, andREST data segments were analyzed separately.

Based on the VT data, a functional (activated) region ofinterest (ROI) was selected by comparing the signal levelduring active stages to that during nonactive stages ofvisual stimuli. This was carried out for foveal andperipheral stimuli separately. This ROI roughly coveredmost of the lower visual areas. Pixels were consideredactivated when significant (t > 4.5) increases were ob-served in both BOLD and perfusion signals.

Subsequently, potential time shifts between BOLD andperfusion signals were determined by correlating thesesignals at different time shifts (lags). This was done tocompensate for the difference in acquisition time of thesetwo signals, and time shifts related to inherent contrastmechanisms (e.g., vascular delays). A perfusion lag of �1time point (�1.77 secs) was found to be optimal for allsubjects. This value was used to correct (time shift) allperfusion data (VT, BH, and REST) before further analysis.For the REST data, global intensity (average of whole-brain signal) fluctuations were removed by using thebrain-averaged signal as a regressor in a pixel-by-pixelmanner. This was done for BOLD and perfusion dataseparately.

Finally, to determine the proportionality constantbetween BOLD and perfusion signals (see equation (2)),linear regression was performed on the average signals inthe functional ROI in the VC for VT, BH, and REST

periods. For VT, foveal and peripheral signals wereaveraged after time-shifting to account for the differencesin stimulation paradigm between these regions. In addi-tion, to estimate the fluctuation level, root-mean-square(RMS) of the averaged signal time course in the VC wasmeasured for each period separately. Since this fluctua-tion is a spontaneous phenomenon, because of lack of thereference for regression analysis as in task-evoked activity,RMS value was employed in this study. To determine thedifferences between sleep and wake states, this analysiswas also performed separately on the eight subjects whofell asleep, as determined from EEG (see below).

To evaluate the presence of ongoing activity, thethrough-plane spatial correlation was determined on apixel-by-pixel basis. This was carried out for eachexperimental condition. Through-plane spatial correlationwas chosen rather than in-plane correlation to avoidpotential bias (spatial blurring) inherent to the imageacquisition process.

In addition, further analysis was performed using ROIselection based on spatial correlation during REST.Similar to the methodology used in earlier studies ofresting-state activity (Biswal et al, 1995; Fox and Raichle,2007), a ‘seed’ voxel was placed in a specific anatomiclocation, and an ROI was generated by selecting all pixelsthat had a correlation of 0.3 or higher with the time courseof the seed voxel in both BOLD and perfusion data.Regions of interest were placed in a number of locations,including VC, lateral temporo-parietal cortex, sensory-motor cortex, and prefrontal cortex based on the anatomicMRI data. Linear regression was performed on the averagesignals in each ROI.

Electroencephalographic Data Analysis

Electroencephalographic data were analyzed using Scan4.3.2 software (Neuroscan; Compumedics USA Ltd, ElPaso, TX, USA). After low-pass filtering of the data at250 Hz (IIR, 48 dB/oct), gradient artifacts were removedbased on template subtraction (Allen et al, 2000). Theparameters of template subtraction were selected asfollows: a window from �50 to 3,955 ms around thescanner’s trigger, phase shift of 99 points, and windowsize of 3 events. Data were then downsampled to 200 Hzand subsequently band-pass-filtered (IIR, 24 dB/oct) toselect a 0.5 to 28 Hz interval. A template of thecardioballistic artifact was constructed for each recordingand used to extract principal components. These compo-nents were removed from the filtered data set.

Electroencephalographic data sets were visually in-spected by a sleep expert (TJB) blinded to the fMRI data.A score (0: wake, 1: stage 1 sleep, 2: stage 2 sleep, 3/4:slow-wave sleep, 5: REM (rapid eye movement) sleep) wasassigned for each 30-sec interval (Rechtschaffen andKales, 1968). These values were then averaged over 2-min intervals and used to discriminate sleep from wakingconditions. Sleep was defined as an average score greaterthan 0 over a 2-min interval. The longest single sleepinterval was chosen for each (sleep) subject.

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1380

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

Results

All subjects actively performed both BH task and VT, asindicated from button responses and the signal recordedfrom the respiratory bellows. Eight of the 11 studiesshowed extended periods of sleep (stage 1 and 2 sleeponly) lasting at least 5 mins. No slow-wave (stage 3 or 4) orREM sleep was detected.

In all subjects, a significant amount of spatiallycorrelated activity was found outside the VC region thatwas active during the task. This was the case for both VTand REST periods, with correlation levels (averaged overcovered brain area) during REST (0.20±0.03) significantly(P < 0.05) higher than during VT (0.16±0.04). In addition,during REST, BOLD and perfusion signals were correlatedin most part of the brain (Figure 3). Significant correlationwas observed primarily in gray matter, with little involve-ment of white matter and cerebrospinal fluid. Within theVC of most subjects, large fluctuations in both BOLD andperfusion signals were observed. From the additionalperfusion reference data acquired on three subjects, it wasestimated that during VT, perfusion and BOLD signalsincreased about 39%±6.7% and 2.19%±0.34% abovebaseline values leading to M = 0.23 for a= 0.38, b= 1.5, andk = 0.51 (Hoge et al, 1999). The estimated M value was

higher than that found in a recent study (Chiarelli et al,2007a). This difference might relate to difference in pulsesequences and parameters, and the assumed k value. Anexample of the time course of BOLD and perfusion signalsin the VC is shown in Figure 4. The fluctuation levelduring REST approached the level of task-related signalchange observed during VT. Average RMS fluctuationlevels were 1.09±0.28, 1.31±0.22, and 0.54±0.14 forVT (RMS-BVT), BH (RMS-BBH), and REST (RMS-BREST),respectively, for BOLD, and 0.54±0.18 (RMS-PVT),0.31±0.10 (RMS-PBH), and 0.22±0.09 (RMS-PREST), re-spectively, for perfusion (all values are a percentage of theaverage BOLD-weighted image intensity; see also Figure5). RMS-BVT and RMS-BBH were significantly higher thanRMS-BREST (P = 5.2� 10�6 and P = 9.1� 10�9, t-test withBonferroni correction for multiple comparisons). RMS-PVT

was significantly higher than RMS-PBH and RMS-PREST

(P = 5.0� 10�3 and P = 5.3� 10�6). In most subjects, asubstantial correlation between BOLD and perfusionsignals was observed during all experimental stages,including the waking state as well as stage 1 and 2 sleep.

The average ratios of BOLD and perfusion signalsin VC were 1.76±0.27, 3.76±1.23, and 1.56±0.41for VT, BH, and REST periods, respectively (Figure 5).Normalized to the ratio during VT, the relative

1.0

-1.0

-0.40.4

VT

REST

BOLD:spatial correlation

perfusion:spatial correlation

BOLD–perfusioncorrelation

Figure 3 Example (single subject data) of correlated activity (see Methods) in one of the five slices studied. Images in each of thethree columns indicate (in color) spatial correlation of BOLD (blood oxygen level-dependent) signal (A, D), spatial correlation ofperfusion signal (B, E), and correlation between BOLD and perfusion signals (C, F). The top and bottom rows show correlation duringVT (visual task) (A to C) and REST (D to F), respectively. For all images, a correlation threshold of ±0.4 was used.

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1381

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

BOLD/perfusion ratios during BH and REST (BPRBH

and BPRREST) averaged 2.16±0.73 and 0.90±0.25,respectively. BPRBH was significantly higher than BPRVT

and BPRREST (P = 2.7� 10�6 and P = 5.0� 10�7), whereasBPRVT was not significantly different from BPRREST (Figure5). Normalized BPRBH was somewhat lower than thetheoretical value (equation (3)) of 3.1. This is possiblybecause of inaccurate model assumptions or parameterestimates.

Within VC of sleep subjects, BPRVT, BPRBH, and BPRREST

averaged 1.76±0.30, 3.75±1.44, and 1.60±0.47, respec-tively. Again, BPRBH was significantly higher than BPRVT

and BPRREST (P = 6.4� 10�4 and P = 2.7� 10�4), whereasBPRVT was not significantly different from BPRREST. Duringthe individual sleep and wake segments of REST, BPRSLEEP

and BPRWAKE averaged 1.79±0.54 and 1.66±0.67, respec-tively, which were significantly lower than BPRBH

(P = 8.6� 10�4 and P = 5.0� 10�3), whereas BPRSLEEP wasnot significantly different from BPRWAKE.

During REST, a substantial correlation between perfu-sion and BOLD signals was also observed in the ROIderived from seed-point analysis (see Methods). TheROIs in VC, lateral temporo-parietal cortex, sensory-motorcortex, and prefrontal cortex all showed BPRREST;in VC, BPRVT values were also shown, which weresignificantly higher than BPRBH (Figure 6 and Table 1).In general, for brain regions outside VC, BOLDand perfusion signals did not consistently show sufficientfluctuation above noise level to reliably determineBPR during VT. In addition, BPR in these regionsmight reflect a mixture of ongoing and task-evokedactivity. In fact, BPRVT was often unreliable because ofthe low level of correlation outside VC during this stage.This is attributed to a reduced level of signal fluctuation,

Sle

ep s

tage

Per

fusi

on (

%)

BO

LD (

%)

3

2

1

0

3

2

1

0

-1

0

2

4

-2

-4

-6

-8

Time (mins)10 200 30 40

EE

G fr

eque

ncy

(Hz)

VT BH REST

0

5

10

15

20

Figure 4 Example of time course EEG signals (parietal electrode), sleep stage, and BOLD and perfusion signals in the visual cortex(region activated during VT). Perfusion signals were normalized to baseline BOLD signal (see Methods). Image segments (top tobottom) show experimental timeline, EEG spectral amplitude, sleep stage (see Methods), and average fMRI signal evolution infunctional region of interest. BOLD (black line) and perfusion fMRI (red line) signals show substantial correlated activity duringeach of the experimental stages, including VT, BH, and REST. This activity persists during sleep periods (sleep stage > 0),which is characterized by a reduction in alpha power (8 to 12 Hz), and an increase in extended theta power (2 to 7 Hz). BH, breath-holding; BOLD, blood oxygen level-dependent; EEG, electroencephalography; fMRI, functional magnetic resonance imaging; VT,visual task.

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1382

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

possibly because of a reduction in ongoing activity owingto increased attention demands during VT (Nir et al,2006). For this reason, BPRVT values in these regions werenot included in the comparison between the variousconditions.

Discussion

Major Findings

This is the first human study to investigate meta-bolic involvement in the phenomenon of spatiallycorrelated BOLD signal fluctuations in the absenceof stimuli. Using the fluctuation level of BOLDsignal relative to simultaneously acquired perfusionsignal, an indicator was derived for relative CMRO2

during rest, and compared with values found duringtasks with substantial (VT) and insignificant (BH)metabolic involvement.

The main finding of this study is that in VC duringREST, BOLD fluctuations are accompanied by perfu-sion fluctuations to an extent that is very similar towhat occurs during VT, and significantly dissimilar toa BH task. The latter was true in the regions withstrong ongoing activity outside VC as well.

This finding suggests that ‘resting-state activity’ isnot a purely vascular effect, but has a metaboliccomponent, and possibly supports a functional role.Among the possible candidates are neuronal signal-ing, homeostatic processes, and plasticity processes.

Additionally, it was found that resting-state meta-bolic activity does not only occur during waking restbut also during light (stage 1 and 2) sleep. Thiscontinued metabolic activity during sleep indicatesthat the processes underlying BOLD resting-stateactivity neither exclusively support cognitive activ-ity nor activity underlying conscious awareness andmonitoring of the environment, all of which arepresumably reduced during sleep.

These findings are consistent with the presence ofslow fluctuations ( < 0.5 Hz) in cortical oxidativemetabolism, which was observed by reflectancespectrophotometry in animals (Vern et al, 1988).Furthermore, as observed in animal studies (Vern etal, 1997, 1988), the observed fluctuations weresynchronized over both hemispheres, and main-tained during both wake and sleep.

In addition to establishing a metabolic origin ofresting-state BOLD fluctuations, the current findingssupport the notion of a persistent level of brain

Perfusion (%)

BO

LD (

%)

VT BH REST

6

4

2

0

–2

–4

–6

3

2

1

0

6

5

4

3

2

1

0

6

5

4

210–1-2 210–1–2 210–1–2

∗∗

∗ ∗

∗∗

Sig

nal c

hang

e (%

RM

S)

BP

R

BPRPerfusionBOLD

VTBHREST

Figure 5 Perfusion and BOLD signals in visual cortex during three experimental stages. (A) Relationship between BOLD andperfusion signals (single subject data). The dashed line represents a linear fit to the data. The relative amplitude of BOLD withrespect to perfusion fluctuations (BPR) is similar during REST and VT, and increases during BH. (B) Mean fluctuation level of BOLDand perfusion signals (%RMS±s.d.) and BPR in visual cortex for all subjects. *P < 0.005 indicate significant differences. BH,breath-holding; BOLD, blood oxygen level-dependent; BPR, BOLD/perfusion ratio; RMS, root-mean-square; VT, visual task.

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1383

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

connectivity during sleep, as suggested previously(Fukunaga et al, 2006; Horovitz et al, 2007). To someextent, this appears contradictory to a recent studythat reports a reduced effect of transcranial magneticstimulation excitation on the spatial extent of EEGactivity during sleep as compared with waking restconditions (Massimini et al, 2005). However, it mustbe noted that the study reports on slow-wave sleep,and measured effective cortico-cortical connectivityon relatively fast time scales (hundreds of milli-seconds), which enabled the identification of acausal relationship (Massimini et al, 2005). Incontrast, the current MRI study reports on func-tional connectivity on a multisecond time scale thatmight not be causal and/or originate or be guided bysecondary nonneocortical regions, such as thethalamus (Steriade, 1999).

Limitations

Although the present study suggests the presence ofmetabolic processes with apparent functional rele-vance during rest and sleep, interpretation of theresults is subject to a number of limitations.

One of the main limitations of the current study isthat it does not allow the determination of absolutelevels of cerebral perfusion or CMRO2

. This isprimarily because of the lack of a perfusionreference scan (scan without perfusion label), whichwas omitted to improve temporal resolution andaccurately detect and capture perfusion fluctuationlevels. It is therefore difficult to relate the currentfindings to positron emission tomographic CBFmeasurements of sleep, which suggest minimal orno reduction in VC activity during light sleep, and a

VT6

4

2

0

–2

–4

–6–2 –1 0 1 2

BH REST

Perfusion (%)

BO

LD (

%)

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

6

4

2

0

–2

–4

–6–2 –1 0 1 2

Figure 6 Relationship between BOLD and perfusion signals during each of the experimental stages in a number of regions selectedbased on seed-point correlation analysis (single subject data). Regions include visual cortex (A), lateral temporo-parietal cortex (B),sensory-motor cortex (C), and prefrontal cortex (D). The dashed lines represent a linear fit to the data. No fit was performed on dataobtained during VT from regions outside the VC, as negligible task-evoked activity was expected. In all regions, the BPR increasesduring the BH task. BH, breath-holding; BOLD, blood oxygen level-dependent; BPR, BOLD/perfusion ratio; VT, visual task.

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1384

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

substantial reduction during stage 3 and 4 non-REMsleep (Braun et al, 1997; for review see Maquet andPhillips, 1998).

Another limitation is the low sensitivity of themethodology, which makes it difficult to derivereliable measures of relative CMRO2

in focal regionsoutside the VC under all conditions (including aVT). It is hoped that transition to a higher magneticfield (7 T), which is currently in progress in ourlaboratory, will alleviate this problem. Conversely, itwill remain difficult to assess CMRO2

on a pixel-by-pixel basis, because of the fact that MRI measures ofperfusion and BOLD do not originate from preciselythe same location. This is because BOLD signalsoriginate primarily from the venous side of thevasculature, whereas perfusion signals originateprimarily from the arterial side.

Lastly, there are limitations to the model assump-tions that might affect the comparison of relativelevel of metabolic fluctuations during the differentstages of the experiment. The calculation of BPRvalues used for this comparison was based on anassumed linearity between BOLD and perfusionsignals. With the actual size of perfusion changesoccurring in the experiments (estimated at about40% during VT; see Results), a substantial departurefrom linearity is expected (see Figure 1). Evidence ofthis nonlinearity can be seen in Figures 5A and 6,particularly with the BH challenge. In the presenceof these nonlinearities, the BOLD–perfusion rela-tionship cannot be accurately described with asingle BPR value. This might partly explain the factthat the measured average BPRBH value of 2.16(relative to BPRVT) was lower than the theoreticalestimate of 3.1 derived from equation (3) for smallperfusion changes. Other factors that might affectrelative BPR levels are the differences in actual andassumed values for a, b, and k. Lastly, the relativelylow value of BPRBH might indicate that this task

evokes a nonzero level of metabolic activity (Martinet al, 2006). Nevertheless, these limitations do notnegate the main finding of this work, the observedsimilarity in BOLD–perfusion relationship (as char-acterized from BPR values) between VT and RESTconditions, and the dissimilarity between theseconditions and the BH condition.

Implications

The suggestion that resting-state BOLD activityrepresents metabolic and, therefore, most likely,functionally relevant processes has potentially im-portant implications for the study of brain function.The findings suggest that the spatial clusters foundwith resting-state BOLD fMRI represent functionalconnectivity that is at least in part neuronal, ratherthan purely vascular in origin. However, they do notexclude the potential for a substantial nonmetaboliccontribution, the size of which will depend onexperimental conditions. The current findings vali-date the assumption that systematic identification ofcorrelation clusters might provide an insight intothe functional architecture under normal and patho-logic conditions. The study of functional connectiv-ity during sleep might provide new insights into therole and mechanistic aspects of sleep. In thiscontext, it appears particularly interesting to usefMRI to study deep (slow-wave) sleep, which isbelieved to be functionally relevant for recovery andmemory consolidation processes.

Acknowledgements

We thank Alan P Koretsky for helpful suggestionsand Susan C Fulton for her assistance with theexperiments.

Table 1 Mean fluctuation level (%RMS) of BOLD and perfusion signals and BPR in several brain regions

Visual (n = 11) Temporo-parietal (n = 7) Sensory-motor (n = 6) Prefrontal (n = 6)

Mean s.d. Mean s.d. Mean s.d. Mean s.d.

BOLDVT 0.841 0.147 0.276 0.115 0.225 0.006 0.254 0.088BH 1.431 0.277 1.108 0.239 1.064 0.191 0.922 0.182REST 0.505 0.142 0.418 0.088 0.416 0.123 0.335 0.078

PerfusionVT 0.431 0.100 0.134 0.051 0.127 0.063 0.123 0.034BH 0.299 0.121 0.300 0.070 0.316 0.098 0.281 0.076REST 0.175 0.076 0.215 0.050 0.211 0.056 0.170 0.025

BPRVT 1.684 0.300 — — — — — —BH 3.897 1.475 3.200 0.757 2.960 0.596 2.820 1.069REST 1.275 0.498 1.337 0.549 1.448 0.539 1.392 0.219

Abbreviations: BH, breath-holding; BOLD, blood oxygen level-dependent; BPR, BOLD/perfusion ratio; RMS, root-mean-square; VT, visual task.

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1385

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

References

Allen PJ, Josephs O, Turner R (2000) A method forremoving imaging artifact from continuous EEG re-corded during functional MRI. Neuroimage 12:230–9

Allers KA, Ruskin DN, Bergstrom DA, Freeman LE, GhaziLJ, Tierney PL, Walters JR (2002) Multisecond periodi-cities in basal ganglia firing rates correlate with thetabursts in transcortical and hippocampal EEG. J Neuro-physiol 87:1118–22

Ances BM, Leontiev O, Perthen JE, Liang C, Lansing AE,Buxton RB (2008) Regional differences in the couplingof cerebral blood flow and oxygen metabolism changesin response to activation: implications for BOLD-fMRI.Neuroimage 39:1510–21

Beckmann CF, DeLuca M, Devlin JT, Smith SM (2005)Investigations into resting-state connectivity usingindependent component analysis. Philos Trans R SocLond B Biol Sci 360:1001–13

Biswal B, Yetkin FZ, Haughton VM, Hyde JS (1995)Functional connectivity in the motor cortex of restinghuman brain using echo-planar MRI. Magn Reson Med34:537–41

Biswal BB, Van Kylen J, Hyde JS (1997) Simultaneousassessment of flow and BOLD signals in resting-statefunctional connectivity maps. NMR Biomed 10:165–170

Braun AR, Balkin TJ, Wesenten NJ, Carson RE, Varga M,Baldwin P, Selbie S, Belenky G, Herscovitch P (1997)Regional cerebral blood flow throughout the sleep–wake cycle. An H2(15)O PET study. Brain 120(Pt7):1173–97

Chiarelli PA, Bulte DP, Gallichan D, Piechnik SK, Wise R,Jezzard P (2007a) Flow-metabolism coupling in humanvisual, motor, and supplementary motor areas assessedby magnetic resonance imaging. Magn Reson Med57:538–47

Chiarelli PA, Bulte DP, Wise R, Gallichan D, Jezzard P(2007b) A calibration method for quantitative BOLDfMRI based on hyperoxia. Neuroimage 37:808–20

Cordes D, Haughton VM, Arfanakis K, Wendt GJ,Turski PA, Moritz CH, Quigley MA, Meyerand ME(2000) Mapping functionally related regions of brainwith functional connectivity MR imaging. AJNR AmJ Neuroradiol 21:1636–44

Davis TL, Kwong KK, Weisskoff RM, Rosen BR (1998)Calibrated functional MRI: mapping the dynamics ofoxidative metabolism. Proc Natl Acad Sci USA95:1834–9

Dora E, Kovach AG (1981) Metabolic and vascular volumeoscillations in the cat brain cortex. Acta Physiol AcadSci Hung 57:261–75

Fox MD, Raichle ME (2007) Spontaneous fluctuations inbrain activity observed with functional magneticresonance imaging. Nat Rev Neurosci 8:700–11

Fukunaga M, Horovitz SG, van Gelderen P, de Zwart JA,Jansma JM, Ikonomidou VN, Chu R, Deckers RH,Leopold DA, Duyn JH (2006) Large-amplitude, spatiallycorrelated fluctuations in BOLD fMRI signals duringextended rest and early sleep stages. Magn ResonImaging 24:979–92

Greicius MD, Krasnow B, Reiss AL, Menon V (2003)Functional connectivity in the resting brain: a networkanalysis of the default mode hypothesis. Proc NatlAcad Sci USA 100:253–8

Hampson M, Peterson BS, Skudlarski P, Gatenby JC, GoreJC (2002) Detection of functional connectivity using

temporal correlations in MR images. Hum Brain Mapp15:247–62

Hoge RD, Atkinson J, Gill B, Crelier GR, Marrett S, Pike GB(1999) Linear coupling between cerebral blood flowand oxygen consumption in activated human cortex.Proc Natl Acad Sci USA 96:9403–8

Horovitz SG, Fukunaga M, de Zwart JA, van Gelderen P,Fulton SC, Balkin TJ, Duyn JH (2007) Low frequencyBOLD fluctuations during resting wakefulness andlight sleep: a simultaneous EEG–fMRI study. HumBrain Mapp (in press)

Johansson B, Bohr DF (1966) Rhythmic activity in smoothmuscle from small subcutaneous arteries. Am J Physiol210:801–6

Kastrup A, Kruger G, Neumann-Haefelin T, Glover GH,Moseley ME (2002) Changes of cerebral blood flow,oxygenation, and oxidative metabolism during gradedmotor activation. Neuroimage 15:74–82

Kety SS, Schmidt CF (1948) The effects of altered arterialtensions of carbon dioxide and oxygen on cerebralblood flow and cerebral oxygen consumption of normalyoung men. J Clin Invest 27:484–92

Leopold DA, Murayama Y, Logothetis NK (2003) Very slowactivity fluctuations in monkey visual cortex: implica-tions for functional brain imaging. Cereb Cortex13:422–33

Logothetis NK, Pauls J, Augath M, Trinath T, OeltermannA (2001) Neurophysiological investigation of the basisof the fMRI signal. Nature 412:150–7

Lowe MJ, Dzemidzic M, Lurito JT, Mathews VP, PhillipsMD (2000) Correlations in low-frequency BOLD fluc-tuations reflect cortico-cortical connections. Neuro-image 12:582–7

Maquet P, Phillips C (1998) Functional brain imaging ofhuman sleep. J Sleep Res 7(Suppl 1):42–7

Martin C, Jones M, Martindale J, Mayhew J (2006)Haemodynamic and neural responses to hypercapniain the awake rat. Eur J Neurosci 24:2601–10

Massimini M, Ferrarelli F, Huber R, Esser SK, Singh H,Tononi G (2005) Breakdown of cortical effectiveconnectivity during sleep. Science 309:2228–32

Nir Y, Hasson U, Levy I, Yeshurun Y, Malach R (2006)Widespread functional connectivity and fMRI fluctua-tions in human visual cortex in the absence of visualstimulation. Neuroimage 30:1313–24

Novack P, Shenkin HA, Bortin L, Goluboff B, Soffe AM(1953) The effects of carbon dioxide inhalationupon the cerebral blood flow and cerebral oxygenconsumption in vascular disease. J Clin Invest 32:696–702

Rechtschaffen A, Kales A (1968) A manual of Standar-dized Terminology, Techniques and Scoring System forSLEEP Stages of Human Subjects. Bethesda, MD: USDepartment of Health, Education, and Welfare PublicHealth Service. National Institutes of Health. NationalInstitute of Neurological Disorders and Stroke. Neuro-logical Information Network

Ruskin DN, Bergstrom DA, Tierney PL, Walters JR(2003) Correlated multisecond oscillations infiring rate in the basal ganglia: modulation bydopamine and the subthalamic nucleus. Neuroscience117:427–38

Stefanovic B, Warnking JM, Kobayashi E, Bagshaw AP,Hawco C, Dubeau F, Gotman J, Pike GB (2005)Hemodynamic and metabolic responses to activation,deactivation and epileptic discharges. Neuroimage28:205–15

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1386

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387

Steriade M (1999) Coherent oscillations and short-termplasticity in corticothalamic networks. Trends Neurosci22:337–45

Uludag K, Dubowitz DJ, Yoder EJ, Restom K, Liu TT,Buxton RB (2004) Coupling of cerebral blood flow andoxygen consumption during physiological activationand deactivation measured with fMRI. Neuroimage23:148–55

van de Ven VG, Formisano E, Prvulovic D, Roeder CH,Linden DE (2004) Functional connectivity as revealedby spatial independent component analysis offMRI measurements during rest. Hum Brain Mapp22:165–78

van Gelderen P, Wu CWH, de Zwart JA, Cohen L, HallettM, Duyn JH (2005) Resolution and reproducibility of

BOLD and perfusion functional MRI at 3.0 Tesla. MagnReson Med 54:569–76

Vern BA, Leheta BJ, Juel VC, LaGuardia J, Graupe P,Schuette WH (1997) Interhemispheric synchrony ofslow oscillations of cortical blood volume and cyto-chrome aa3 redox state in unanesthetized rabbits. BrainRes 775:233–9

Vern BA, Schuette WH, Leheta B, Juel VC, Radulovacki M(1988) Low-frequency oscillations of cortical oxidativemetabolism in waking and sleep. J Cereb Blood FlowMetab 8:215–26

Wise RG, Ide K, Poulin MJ, Tracey I (2004) Restingfluctuations in arterial carbon dioxide induce signifi-cant low frequency variations in BOLD signal. Neuro-image 21:1652–64

Metabolic origin of resting-state BOLD signal fluctuationM Fukunaga et al

1387

Journal of Cerebral Blood Flow & Metabolism (2008) 28, 1377–1387