Download Flow-metabolism coupling in human visual, motor, and

Magnetic Resonance in Medicine 57:538 –547 (2007)
Flow-Metabolism Coupling in Human Visual, Motor, and
Supplementary Motor Areas Assessed by Magnetic
Resonance Imaging
Peter A. Chiarelli, Daniel P. Bulte, Daniel Gallichan, Stefan K. Piechnik, Richard Wise,
and Peter Jezzard*
Combined blood oxygenation level-dependent (BOLD) and arterial spin labeling (ASL) functional MRI (fMRI) was performed
for simultaneous investigation of neurovascular coupling in the
primary visual cortex (PVC), primary motor cortex (PMC), and
supplementary motor area (SMA). The hypercapnia-calibrated
method was employed to estimate the fractional change in
cerebral metabolic rate of oxygen consumption (CMRO2) using
both a group-average and a per-subject calibration. The groupaveraged
calibration
showed
significantly
different
CMRO2ⴚCBF coupling ratios in the three regions (PVC: 0.34 ⴞ
0.03; PMC: 0.24 ⴞ 0.03; and SMA: 0.40 ⴞ 0.02). Part of this
difference emerges from the calculated values of the hypercapnic calibration constant M in each region (MPVC ⴝ 6.6 ⴞ 3.4,
MPMC ⴝ 4.3 ⴞ 3.5, and MSMA ⴝ 7.2 ⴞ 4.1), while a relatively minor
part comes from the spread and shape of the sensorimotor
BOLD–CBF responses. The averages of the per-subject calibrated CMRO2ⴚCBF slopes were 0.40 ⴞ 0.04 (PVC), 0.31 ⴞ 0.03
(PMC), and 0.44 ⴞ 0.03 (SMA). These results are 10 –30% higher
than group-calibrated values, and are potentially more useful
for quantifying individual differences in focal functional responses. The group-average calibrated motor coupling value is
increased to 0.28 ⴞ 0.03 when stimulus-correlated increases in
end-tidal CO2 are included. Our results support the existence of
regional differences in neurovascular coupling, and argue for
the importance of achieving optimal accuracy in hypercapnia
calibrations to resolve method-dependent variations in published results. Magn Reson Med 57:538 –547, 2007. © 2007
Wiley-Liss, Inc.
Key words: cerebrovascular coupling; functional MRI; oxygen
metabolism; BOLD; CBF
Over the past decade, blood oxygenation level-dependent
(BOLD) MRI methods (1– 4) have been developed to investigate sensorimotor and cognitive function via the hemodynamic correlates of neural activity (5,6). During a period
of increased neuronal response in the brain, a disproportional rise in the inflow of fully oxygenated arterial blood
dilutes the level of deoxygenated hemoglobin (dHb) in the
venous vasculature, resulting in a T2*-weighted signal increase (7,8). Parallel increases in the rate of local cerebral
oxygen metabolism (CMRO2) and the cerebral blood volume (CBV) partially counterbalance this cerebral blood
Oxford Centre for Functional Magnetic Resonance Imaging of the Brain,
Oxford University, John Radcliffe Hospital, Oxford, UK.
Grant sponsors: UK Medical Research Council; Rhodes Trust; UK EPSRC.
*Correspondence to: Peter Jezzard, Oxford Centre for Functional Magnetic
Resonance Imaging of the Brain, Oxford University, John Radcliffe Hospital,
Oxford, OX3 9DU, United Kingdom. E-mail: [email protected]
Received 5 June 2006; revised 20 November 2006; accepted 20 November
2006.
DOI 10.1002/mrm.21171
Published online in Wiley InterScience (www.interscience.wiley.com).
© 2007 Wiley-Liss, Inc.
flow (CBF)-induced signal increase by augmenting the
number of paramagnetic dHb molecules present. There is
no reason why such responses should be uniform across
brain regions. As many have already noted (9,10), the
ability to describe the quantitative relationships among
CBF, CBV, and CMRO2 in multiple regions of the brain is
crucial for attaining a more complete knowledge of the
fundamental neurophysiology reflected in BOLD measurements (11–14).
A recently developed model to extract CMRO2 estimates
from experimentally available BOLD and CBF measurements (15,16) was derived by combining Fick’s principle
of mass balance (17), Grubb’s relationship between fractional CBF and CBV changes (18), and the biophysical
description of R2* provided by Ogawa et al. (19) and
Boxerman et al. (20). Provided these relationships are accepted, the fractional change in BOLD signal can be expressed as a calibrated function of aerobic metabolism and
flow change as follows:
冋 冉
冊冉 冊 册
CMR O2
⌬BOLD
⫽M 1⫺
BOLD 0
共CMR O2兲 0
␤
CBF
CBF 0
␣⫺␤
,
[1]
where M represents the maximum BOLD signal change
(effectively in units of%) that can be attained by achieving
a theoretical 100% oxygen saturation in the venous vessels
(experimentally estimated using hypercapnia to provide
an assumed isometabolic CBF increase), ␣ is the Grubb
constant (assumed to be 0.38, accounting for an assumed
fixed relationship between changes in CBV and CBF) (18),
and the exponent ␤ describes the oxygenation and fieldstrength dependence of the BOLD effect. (All previous
studies with this model used ␤ ⫽ 1.5. However, these
studies predominantly were carried out at 1.5T, and at
higher fields it is expected to be closer to 1 as the extravascular components of the BOLD signal begin to dominate
(20,21)). To describe CMRO2 as a function of measurable
MR signal changes, one can reform the equation in terms of
the fractional CMRO2 change:
冢
冉
⌬BOLD
CMR O2
BOLD 0
⫽ 1⫺
共CMR O2兲 0
M
冊冣
1/␤
冉 冊
CBF
CBF 0
1⫺␣/␤
,
[2]
Table 1 summarizes recent research based on this calibrated model, which uses graded stimulation to measure
the relationship between changes in CMRO2 and CBF. The
CMRO2:CBF coupling ratios calculated to date span ranges
of 0.22– 0.51 for the visual cortex (16,22–24) and 0.30 –
538
Flow-Metabolism Coupling by MRI
539
Table 1
Human Neurovascular Coupling Results in Literature, Using the Hypercapnia Calibrated Model With Graded Stimulation, Compared
With This Study
Visual cortex
Study
Hoge et al. (16, 22)
Kastrup et al. (25)
Uludag et al. (24)
Stefanovic et al. (26)
Stefanovic et al. (23)
Current study (group
calibration)
Current study
(per-subject
calibration)
Current study (group
calibration with
ETCO2 correction)
Motor cortex
SMA
M (%)
R2*
CMRO2:
CBF
M (%)
R2*
CMRO2:
CBF
M (%)
R2*
CMRO2:
CBF
22
—
25
—
7.6 ⫾ 1.3
6.6 ⫾ 3.4
4.4
—
9.6
—
1.5 ⫾ 0.3
2.1 ⫾ 1.1
0.51 ⫾ 0.08
—
0.45 ⫾ 0.03
—
0.22 ⫾ 0.11
0.34 ⫾ 0.03
—
9⫾3
—
7.2 ⫾ 0.01
6.1 ⫾ 1.1
4.3 ⫾ 3.5
—
2.3 ⫾ 0.8
—
1.4 ⫾ 0.002
1.2 ⫾ 0.02
1.3 ⫾ 1.1
—
0.3 ⫾ .06
—
0.44 ⫾ 0.04
0.49 ⫾ 0.13
0.24 ⫾ 0.03
—
—
—
—
—
7.2 ⫾ 4.1
—
—
—
—
—
2.3 ⫾ 1.3
—
—
—
—
—
0.40 ⫾ 0.02
Range
3.9–9.4
Range
1.2–2.9
0.40 ⫾ 0.04
Range
3.5–7.2
Range
1.1–2.3
0.31 ⫾ 0.03
Range
4.7–10.1
Range
1.5–3.2
0.44 ⫾ 0.03
—
—
—
4.3 ⫾ 3.5
1.3 ⫾ 1.1
0.28 ⫾ 0.03
—
—
—
0.49 for the motor cortex (23,25,26). The spread in these
data obscure any possible significant difference in the
neurovascular coupling ratio between these two brain regions. However, Stefanovic et al. (23) recently obtained
data from both the visual and motor cortices within a
single session, and demonstrated a CMRO2:CBF coupling
ratio of 0.49 in the visual cortex and 0.22 in the motor
cortex. That study aimed to elucidate the effect of baseline
flow on the BOLD and CBF signals; hence both visual and
motor stimulations were presented simultaneously, overlapping in time with the presentation of longer-hypercapnia stimuli, which may have confounded the interpretation of the data.
In addition to the CMRO2⫺CBF coupling relationship, it
is also crucial to identify whether regional differences can
be observed in the calibration parameter M, which expresses baseline-CMRO2 BOLD⫺CBF coupling (27,28). In
FIG. 1. Schematic of the experimental design, with three levels
of visual stimulation and three
levels of motor stimulation. In this
experiment 36 min were devoted
to the alternating presentation of
visual and motor stimuli, and
21 min were devoted to the hypercapnic calibration.
contrast to other parameters of the model, this value (in
principle) can be estimated experimentally. In the present
work we extend the current understanding of neurovascular coupling in the human brain by employing a paradigm
that presents visual and motor stimulations separately followed by hypercapnia calibration (Fig. 1). We use the
experimental scheme to assess not only regional differences but also the impact of particular features of the
method on the experimental outcome.
MATERIALS AND METHODS
Graded Stimulation
Ten healthy volunteers (age range ⫽ 24 – 40 years) participated in this study after they provided informed consent. A
schematic of the experimental paradigm is provided in Fig. 1.
540
Chiarelli et al.
FIG. 2. a: BOLD activation data obtained
during the whole-brain pilot experiment,
used to plan oblique slices through the PVC,
PMC, and SMA. (b) BOLD and (c) CBF activations obtained from a single subject are
overlaid on oblique-orientation EPI images.
d: CBF activation map obtained from the
hypercapnia calibration for the same
subject.
We executed 45-s blocks of visual and motor stimulations in
an alternating fashion, with a pseudo-randomized level of
graded stimulation. Stimulation epochs were separated by
45-s blocks of rest, during which the fixation point was
displayed against a black background. The room was darkened to ensure minimum ambient light. The graded motor
and visual stimulation portion of the experiment lasted for a
total of 36 min, followed by 3 min of rest. Subsequently we
applied two 5-min epochs of 4% inspired CO2, separated by
4 min of rest. The subjects were instructed to remain alert
with their eyes directed toward the fixation point during the
whole experiment. Graded visual and motor stimulation
tasks were implemented as follows:
Visual Stimulation
For visual stimulation, a black-and-white oscillating square
checkerboard stimulus at 8 Hz (eight full cycles of light and
dark within a check per second) was back-projected onto a
screen covering the outer bore of the magnet using an In
Focus LP1000 projector providing 1024 ⫻ 768 pixels with a
⬃100 Hz refresh rate. The subjects viewed the screen via a
mirror. The visual stimulus covered ⬃30° of the visual field,
limited by the diameter of the magnet bore. We achieved
graded activation by adjusting the “white” squares in the
checkerboard to luminance fractions of 4%, 12%, and 41%
(representing the values available in Microsoft Powerpoint
software to adjust image brightness). The “black” squares
remained at maximum darkness. In this manner we were
able to simultaneously adjust total luminance and checkerboard contrast, which we found in pilot studies provided the
most controllable and widest range of stimulation intensities.
3) ring, (tap 4) little, (tap 5) ring, and (tap 6) middle. This
pattern was repeated in synchrony with a flashing fixation
point. Subjects were trained to cue tap 1 with the appearance of the fixation point, and to cue tap 4 with the disappearance of the fixation point. To obtain a graded stimulus, the tapping cue was modulated to yield individual
finger tapping rates of 1, 3, and 5 Hz.
Slice Localization
In a separate 165-s pilot session prior to the main experiment, each subject was presented with the oscillating
black and white checkerboard in a block design and instructed to perform bilateral finger-tapping coincident
with the visual stimulus. Whole-brain BOLD fMRI data
were acquired during this time, and were analyzed using
the in-line data analysis package provided with the scanner software to obtain functional activation maps and prescribe oblique slices that passed through the primary visual cortex (PVC), primary motor cortex (PMC), and supplementary motor area (SMA) (Fig. 2a).
Hypercapnia
Filtered air and 5% CO2 in air were mixed to deliver 4%
inspired CO2 through a tight-fitting face mask. Respiratory
composition within the mask was continuously sampled
with the use of equipment from Applied Electrochemistry
Inc. (Pittsburgh, PA, USA; CD-3A CO2 sensor, S-3A/I O2
sensor, and Flowcontrol R-2 vacuum pump). Data regarding the CO2 and O2 concentrations in the mask air were
logged at intervals of 10 ms with the use of Powerlab
software (ADInstruments, Colorado Springs, CO, USA).
Motor Stimulation
Motor stimulation was accomplished by voluntary cyclic
opposition between the thumb and each finger, according
to the following pattern: (tap 1) index, (tap 2) middle, (tap
MRI Parameters
Images were acquired on a Siemens Trio 3T MRI scanner
using an eight-channel head radiofrequency (RF) receive
Flow-Metabolism Coupling by MRI
541
FIG. 3. (a) Motor and (b) visual responses to 45-s
visual stimulation epochs, followed by 2-min and
15-s rest periods. Note the difference in undershoot amplitude between the two stimulus types.
coil. An interlaced BOLD/pulsed arterial spin-labeling
(ASL) sequence was developed to collect T2*-weighted
conventional EPI images and QUIPSS II with thin slice
periodic saturation (Q2TIPS) (29,30) cerebral perfusion
images in the following scheme: [ASL tag image/BOLD
image/ASL control image/BOLD image]n. The parameters
for the BOLD measurements included TR/TE ⫽ 4.5
s/32 ms, and for the ASL experiments they were TR/TE/
TI ⫽ 4.5 s/23 ms/1.4 s, TI1/TI2 ⫽ 700 ms/1400 ms, with a
TI1 stop time of 1200 ms. The width of the tagging band
was 10 cm with a gap of 1.5 cm between the tag and image
volumes. In both BOLD and ASL, five slices were acquired
per TR, with 4 ⫻ 4 ⫻ 6 mm3 voxel dimensions and a 64 ⫻
64 imaging matrix.
Data Analysis
All analyses were performed using the FMRIB Software
Library (FSL) package (31). Each fMRI acquisition yielded
a sequence of interleaved EPI BOLD and ASL images,
which were split and analyzed separately.
The processing steps for the BOLD data included 1)
2D-FT and EPI ghost removal, 2) motion correction (32), 3)
volume-by-volume extraction of brain matter from surrounding tissue and skull (33), 4) splitting images from the
sensorimotor stimulation portion of the experiment and
images from the hypercapnia portion into two separate
data sets, 5) spatial smoothing with a 5-mm FWHM Gaussian kernel on both data sets, 4) high-pass temporal filtering
with a cut-off of 180 s applied to the sensorimotor data and
a cutoff of 540 s applied to the hypercapnia data, 7) autocorrelation correction using a general linear model (GLM)based nonparametric estimation method (FMRIB’s Improved Linear Model) (34), and 8) voxelwise GLM-based
correlation of the signal time course with an appropriate
reference model constructed by convolution of a block
design boxcar function with a gamma-variate hemodynamic response model.
Analysis of the Q2TIPS perfusion data included steps
1–3 as described above. We performed temporal sinc-
interpolation on the set of tag images and the set of
control images, shifting the signal intensity of tag images to appear as if they were acquired 0.5 TRs later, and
control images to appear as if they were acquired 0.5
TRs earlier, followed by pairwise subtraction of image
intensity values. Steps 4 –7 as listed above were not
applied to perfusion data. GLM estimation of activation
(step 8) was performed in the same manner as for the
BOLD data. After the control and tag images were subtracted, the temporal resolution of the perfusion data
was half that of the BOLD data set.
To analyze the visual response, for which post-stimulus effects have been reported (35,36), we excluded
“rest” epochs that immediately followed visual stimulation from the estimate of baseline BOLD and CBF.
Only rest blocks corresponding to the intervals following the motor task were included in the estimate of
visual baseline. Based on a separate block-design pilot
study with alternating epochs of visual and motor stimulation (45 s “on”/135 s “off”), we observed a substantially larger post-stimulus BOLD undershoot following
flashing checkerboard stimulation compared to the finger tapping task (Fig. 3). After this correction was employed, inspection of the GLM fit to the visual data from
the main experiment revealed a significantly improved
estimation of the baseline, whereas a negligible difference was observed when a similar analysis was performed for the motor task.
Region of Interest (ROI) Definition
We calculated BOLD, CBF, and CMRO2 estimates within
an ROI defined for each individual subject by the overlap
of statistically thresholded BOLD (z-stat ⬎ 4, cluster
p-threshold ⬎ 0.05) and CBF (z-stat ⬎ 2.3, cluster p-threshold ⬎ 0.05) images. We obtained each ROI using data from
the most intense of the three graded stimulus levels, and
used the ROI to interrogate activation estimates at all three
BOLD and CBF activation levels.
542
Chiarelli et al.
RESULTS AND DISCUSSION
Regional CMRO2⫺CBF Coupling
BOLD and CBF images acquired at the appropriate oblique
orientation are displayed for a representative subject in
Fig. 2b and c, respectively, with data overlaid for the
separate motor (red) and visual (blue) tasks. The thresholded CBF image (CBF zstat ⬎ 2.3, cluster p-threshold ⬎
0.05) from the two-epoch 4% hypercapnia challenge is
supplied in Fig. 2d, with activated voxels consistent with
gray matter. In general the CBF maps tended to be noisier
than the BOLD maps, but had a higher spatial specificity to
the activated regions (see Fig. 2b and c). This is primarily
due to the significantly lower signal-to-noise ratio (SNR) of
the ASL data and the weighting of the BOLD signal toward
the larger venous vasculature. The significance thresholds
for the analyses of each data type were adjusted to account
for these inherent differences. The activation maps of the
two data sets compared very well with each other as a
result, and by using masks constructed from the overlap of
the two methods, we were able to remove the spurious
voxels from the ASL data along with the activated voxels
from the BOLD data set that were distal from the active
brain matter regions. Obviously, there are issues that are
poorly understood regarding the comparison of BOLD- and
ASL-based activation maps (37). Ultimately our goal was
to use a threshold criterion that yielded maps similar in
size to those used by others (the size of the maps is principally constrained by the CBF threshold), particularly
those based on retinotopic maps, which tend to be more
spatially restrictive.
Figure 4 displays GLM estimates of fractional change in
BOLD and CBF for all three stimulus levels from the ROI
for each individual in the PVC (Fig. 4a), PMC (Fig. 4b), and
SMA (Fig. 4c). Measurements from the three different
stimulation levels for a given subject are joined by lines,
and data points are consistently labeled so that somatosensory activation data from each individual can be compared
against the corresponding hypercapnia activation (red).
The changes seen in CBF are somewhat larger than expected, most likely due to inappropriate suppression of
the signal from large vessels.
BOLD and CBF values for the hypercapnia data were
obtained from a combined GLM fit to two consecutive
5-min blocks of 4% inspired CO2, which were separated
by 4 min of breathing air. We used 4% hypercapnia instead of 5% to avoid the potential stress response and
neuronal effects associated with higher levels of inspired
CO2 (38 – 40). 4% CO2 represented a compromise between
minimizing the stress response and maximizing the contrast for changes in BOLD and CBF. Average end-tidal CO2
levels were found to rise by ⬃7.4 mmHg during CO2 administration from a baseline value of 42.6 ⫾ 4.6 mmHg.
The calibration procedure indicated by Eq. [1] was performed such that the iso-CMRO2 BOLD⫺CBF contour
could be fitted to the group average BOLD and CBF values
derived from hypercapnia (thick line). We obtained an
estimated BOLD calibration amplitude of MPVC ⫽ 6.6 ⫾ 3.4
(similar to the value calculated by Stefanovic et al. (23)),
MPMC ⫽ 4.3 ⫾ 3.5 (similar to but significantly lower than
reported literature values), and MSMA ⫽ 7.2 ⫾ 4.1. The
FIG. 4. Functional activation (black) and hypercapnia data (red)
from the (a) PVC, (b) PMC, and (c) SMA. The resting-metabolism
iso-CMRO2 contour is shown fit to the hypercapnia data. Data points
are consistently labeled across the plots such that individual subject
data can be compared. [Color figure can be viewed in the online
issue, which is available at www.interscience.wiley.com.]
Flow-Metabolism Coupling by MRI
543
trend observed by Stefanovic et al. of a greater M value in
the PVC compared to the PMC is supported by our work.
Figure 5 displays plots of the estimated fractional CMRO2
change vs. fractional CBF change obtained in the PVC (a),
PMC (b), and SMA (c) regions using the single group average
value of M calculated in each respective brain region. A
linear fit is shown on each CMRO2⫺CBF plot. These fits were
constrained to pass through the point (0,0), since all BOLD
and CBF values were assumed to be calculated with respect
to well-defined coincident baseline epochs. The calculated
group-average CMRO2:CBF ratios were 0.34 ⫾ 0.03 (PVC),
0.24 ⫾ 0.03 (PMC), and 0.40 ⫾ 0.02 (SMA). CMRO2 data are
displayed for all 10 subjects in Fig. 5, although one subject
(⫹) did not complete the CO2 challenge, and therefore contributed no data to the average M value. The PMC data show
a substantially weaker linear correlation (R2 ⫽ 0.40) than the
other regions (R2 ⱖ 0.82). Three data sets in which the
CMRO2⫺CBF trend deviated significantly from the average
(based on a pooled variance t-test) are shaded gray to highlight the presence of cases in which graded stimulation did
not show concurrent changes between the CMRO2 and CBF.
The notion that a larger CMRO2⫺CBF slope in the PVC
compared to the PMC may be warranted is supported by a
comparison with other results reported by Hoge et al. (16),
Kastrup et al. (25), Uludag et al. (24), and Stefanovic et al.
(26). Those studies further suggest the possibility that the
resting-state BOLD:CBF coupling ratio (M) is larger in the
PVC compared to the PMC, a conclusion that is also supported in the present study. A second study by Stefanovic
et al. (23) found a trend that is opposite to that obtained in
the present study. Although the CMRO2⫺CBF slope obtained by Stefanovic et al. (23) in the PVC lies within the
range of the slope obtained here, their value for the PMC is
much higher than our data or previous literature data
would predict.
Systematic Sources of Variability
Group-Average Calibration Constant
An examination of the CMRO2⫺CBF relationship, as calculated using the group-average M values, hides a systematic source of variability originating from the calibration
constant itself. We previously investigated the relative impact of hypercapnia data and functional stimulation data
on the calculated measure of CMRO2⫺CBF coupling by the
calibrated model. We observed that on a theoretical basis,
the hypercapnia calibration constant determines the coupling slope and linearity of coupling for the functional
activation task, while the general shape and degree of
scatter in the functional BOLD⫺CBF data have a minor
impact on these values (41).
Figure 6 displays the variation in the experimental estimates of metabolic coupling due to uncertainty in the
isometabolic profiles described by M. The thin lines describe the impact of theoretical variation in M beyond the
experimentally estimated value on the calculated
CMRO2⫺CBF slope in the three brain regions. There is
relatively little difference between the coupling slope vs.
M behavior for the three data sets, although at any single
value of M the PVC data show the lowest coupling slope,
the SMA data show an intermediate coupling slope, and
the PMC data show the highest slope. The symbols corre-
FIG. 5. CMRO2–CBF coupling plots and linear fit (thick line) in the (a)
PVC, (b) PMC, and (c) SMA, as calculated using the calibrated
model. Labeling of data sets is consistent with Fig. 4, and data sets
that deviate significantly from the mean response are labeled in
gray.
544
FIG. 6. CMRO2–CBF coupling slope fit to functional activation data
from the PVC (␴), PMC (●), and SMA (■) over a range of possible M
values. Points along each curve corresponding to the calculated M
value are marked with (⫻).
spond to experimental M values calculated in this study,
for the PVC (␴), PMC (●), and SMA (■). The use of experimentally determined M values introduces a substantially
further difference between the CMRO2⫺CBF slopes for the
three brain regions. Note the large impact of transformed M
error on the coupling slope, especially for low values of M.
An M-independent measure of neurovascular coupling is
likely not feasible in this low M-value regime due to the
sensitivity of the model to this parameter. (See Ref. 27 for
a detailed analysis of the effect that the estimation of M has
on the determination of the coupling constant.)
Per-Subject Hypercapnia Calibration
In comparison to using a single group-average value of M
to calibrate sensorimotor BOLD⫺CBF data from all subjects (as has been done in all studies to date), the use of
Chiarelli et al.
subject-specific M values to calibrate the corresponding
sensorimotor data may more appropriately characterize
the inherent variability of CMRO2⫺CBF slope between
individuals, and provide a more realistic CMRO2⫺CBF
coupling estimate for individuals whose vascular physiology warrants a calibration constant different from the
group mean. Figure 7a– c show the range of iso-CMRO2
curves fitted on a subject-by-subject basis to hypercapnia
data from the PVC, PMC, and SMA, respectively. x-Axis
and y-axis error estimates represent the ⫾95% (confidence
interval to mean (CIm) as obtained from the variance of the
GLM fit to the underlying BOLD and perfusion hypercapnia data. Figure 7d–f display CMRO2⫺CBF plots that were
recalculated using these subject-specific M values for calibration of BOLD⫺CBF data. When a linear fit is applied to
the entire set of data, as in Fig. 5, the calculated
CMRO2⫺CBF slopes are 0.28 ⫾ 0.08 (PVC); 0.26 ⫾ 0.04
(PMC), and 0.35 ⫾ 0.06 (SMA). These values could be
interpreted as suggesting little difference between the regional CMRO2⫺CBF coupling values. However, in each
plot (Fig. 7d–f), there appears to be a principle direction
about which CMRO2⫺CBF data from the group are clustered. If data sets deemed as outliers (based on a 95%
confidence F-test, shaded gray) are excluded from the linear fit, the resulting slopes are 0.40 ⫾ 0.04, (PVC) 0.31 ⫾
0.03, (PMC), and 0.44 ⫾ 0.03, (SMA). The fits excluding
outliers are displayed in Fig. 7, in which coupling in the
SMA still reveals the highest slope, and coupling in the
PMC still produces the lowest slope. Values in the SMA
and PVC are more similar than for the data calculated with
a group-averaged M value, as shown in Fig. 5.
The recalculated values for the CMRO2⫺CBF slope in
the visual and motor regions correspond roughly to values
from Hoge et al. (16) (0.51 ⫾ 0.08) and Uludag et al. (24)
(0.45 ⫾ 0.03) in the PVC, and the value from Kastrup et al.
(25) (0.3 ⫾ 0.06) in the PMC, despite the large difference in
M used to calibrate the BOLD and CBF values.
Consistent labeling for the data from each subject allows
a comparison between Figs. 5 and 7. Note that the data set
labeled (●), whose slope deviates from the average value in
FIG. 7. Per-subject hypercapnia
calibration (a– c), and corresponding CMRO2–CBF coupling
plots (d–f), in the PVC (a and d),
PMC (b and e), and SMA (c and f).
BOLD (not visible) and CBF error
estimates on hypercapnia data
represent ⫾CIm, as obtained from
the residuals of the GLM fit. Data
sets deemed outliers in d–f are
labeled in gray and excluded
from the linear fit.
Flow-Metabolism Coupling by MRI
Fig. 5b (motor), no longer is an outlier when the subjectspecific M value is used (Fig. 7e). The calculated M value
from this subject is noticeably higher than the mean (Fig.
7b), suggesting a potential difference in vascular architecture. As another example, the data set labeled ⌬ reveals a
negligible change in metabolism in the PMC as a function
of different levels of graded stimulation when the subjectspecific M value is used. In this case, using the group
average M of 4.3 (Fig. 5b) makes the coupling estimates
appear closer to the group mean. It is difficult to completely exclude the possibility that the differences in the
estimated M-parameters from individual subjects are simply due to inherent experimental errors in the hypercapnia
calibration technique. However, a visual inspection of the
error margins in Fig. 7a– c suggests that the fitted M values
from many subjects overlap statistically, while a number
of subjects show fitted M values that do not overlap based
on the chosen 95% statistical criterion.
Previous research has illustrated the effect of factors
such as diet and physiological state on the neural-hemodynamic response to stimulation. It may be possible, therefore, that some of the nonlinear CMRO2⫺CBF coupling
relationships between individuals can be accounted for by
genuine physiological differences. Our analysis suggests
that certain subjects may show such behavior, which
would otherwise be masked by the use of single groupaveraged M values.
Systemic Effects of Functional Stimuli
Functional MRI (fMRI) experiments routinely involve a
number of potential confounds related to the effects of
stimulus delivery, such as stimulus-correlated motion or
modulation of attention (42). One potential confound that
is seldom monitored is the fluctuation of blood gasses
during the experiment (43). In the current experiment we
continuously measured end-tidal CO2 (ETCO2) and O2
(ETO2), which provided us with insight into respiratory
effects associated with cerebral stimulation. The average
baseline ETCO2 was 42.6 ⫾ 4.6 mmHg. There was a very
small tendency toward an increased respiratory rate during the CO2 blocks. Across all subjects, breaths-per-minute
averaged approximately 6.5 during baseline, and 7.5 at the
end of the hypercapnia blocks. The increase was nearly
linear over the duration of the block. Note that the total
minute volume was not measured, and this variable is
expected to rise with increased inspired CO2 concentration. ETO2 values were observed to rise very slightly from
⬃114 mmHg at baseline to ⬃122 mmHg at the end of the
CO2 block, likely induced by the increased minute volume. However, this was a very slow rise since the effect of
increased ventilation is partially counteracted by a diminished O2 concentration in the 5% CO2 in air mixture
compared to normal air (the inspired level of oxygen actually drops from ⬃157 mmHg to ⬃145 mmHg). We therefore expect the impact of change in O2 concentration to be
relatively minor. The issue of changing O2 concentration is
an effect that is present in all of the studies compared in
this work, and in the future we plan to incorporate dynamic forcing of end-tidal O2 or CO2 levels as a means to
avoid this potential effect.
Figure 8 displays the subject-average ETCO2 behavior
during epochs of the graded motor task. Despite prior
545
FIG. 8. End-tidal CO2 response to presentation of the motor stimulus (gray blocks) for low (L), medium (M), and high (H) tapping rates.
Dashed lines serve as a guide to the eye.
instruction of subjects to maintain a consistent rate of
breathing, we observed a significant involuntary alteration
in the respiratory cycle by all subjects during the motor
task (note that no difference in breathing or ETCO2 was
observed during presentation of the oscillating checkerboard for visual stimulation). None of the subjects reported
noticing this difference when questioned after the experiment. In some subjects the alteration in breathing was
manifested as shorter, shallower breaths, whereas in others the difference was slower breathing. The result was a
rise in ETCO2 levels by 0.45 mmHg during the steady-state
portion of the motor task, and a further increase in ETCO2
to 0.76 mmHg above baseline near the end of the task.
While these effects appear small, they constitute a sizeable
6% and 10% of the average ETCO2 change elicited by the
hypercapnia calibration used in this study.
There appeared to be no difference between ETCO2 responses elicited by the low (L), medium (M), and high (H)
tapping rates. While the observed change in ETCO2 is not
high enough to yield a statistically significant BOLD activation in non-motor areas, it likely does contribute to the
measured BOLD and CBF signal changes within the ROI
selected due to real PMC activation. Pilot data from lowlevel CO2 challenges suggest that the BOLD response to
CO2 is roughly linear from baseline up to the changes
elicited by 4% CO2. Using the hypercapnia-derived BOLD
signal change in the PMC from this study as a point of
reference, we estimate that the breathing-induced change
in ETCO2 will yield a stimulus-correlated BOLD signal
increase in the PMC between 0.08% and 0.14% (these
lower and upper estimates were obtained using ETCO2 data
from either of the two plateau regions in Fig. 8, marked by
dashed lines). Assuming an additive effect between the
apparent activation from this CO2 response and the signal
change due to functional activation, we applied a correc-
546
tion to the calculation of CMRO2 by the group-calibrated
approach and obtained a value for the CMRO2⫺CBF coupling slope between 0.26 and 0.28. The higher of these two
values represents a 16% change from the original calculated value of the coupling slope. These data illustrate the
potential importance of monitoring respiration during sensorimotor tasks, which require a large degree of concentration by the subject that may induce involuntary breathholding or respiratory changes.
CONCLUSIONS
In this study we performed the first simultaneous investigation of neurovascular coupling in three brain regions,
and found a larger proportional increase of CMRO2 to CBF
in the PVC region compared to the PMC region. An even
larger CMRO2⫺CBF coupling constant was observed in the
SMA. We observe that a major source of regional variability emerges from the hypercapnia calibration, while real
differences between sensorimotor stimulation data in the
three brain regions account for only a minor part of this
regional difference. Calibration on a per-subject basis revealed a group of tightly clustered CMRO2⫺CBF data in
each brain region, but with a number of noticeable outliers, which may reflect true differences in brain physiology
and vascular architecture. We uncovered a significant impact of involuntary respiratory changes during the motor
task, which has large potential implications for all fMRI
studies that involve any kind of exertion or exercise.
Chiarelli et al.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
ACKNOWLEDGMENTS
We gratefully acknowledge support from the UK Medical
Research Council (P.J., S.K.P., D.P.B., and R.W.), Rhodes
Trust (P.A.C.), and UK Engineering and Physical Sciences
Research Council (EPSRC) (D.G.).
REFERENCES
1. Ogawa S, Lee TM, Kay AR, Tank DW. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc Natl Acad Sci
USA 1990;87:9868 –9872.
2. Kwong KK, Belliveau JW, Chesler DA, Goldberg IE, Weisskoff RM,
Poncelet BP, Kennedy DN, Hoppel BE, Cohen MS, Turner R, Cheng
HM, Brady TJ, Rosen BR. Dynamic magnetic resonance imaging of
human brain activity during primary sensory stimulation. Proc Natl
Acad Sci USA 1992;89:5675–5679.
3. Bandettini PA, Wong EC, Hinks RS, Tikofsky RS, Hyde JS. Time course
EPI of human brain function during task activation. Magn Reson Med
1992;25:390 –397.
4. Ogawa S, Tank DW, Menon R, Ellermann JM, Kim SG, Merkle H,
Ugurbil K. Intrinsic signal changes accompanying sensory stimulation:
functional brain mapping with magnetic resonance imaging. Proc Natl
Acad Sci USA 1992;89:5951–5955.
5. Haacke EM, Lai S, Reichenbach JR, Kuppusamy K, Hoogenraad FGC,
Takeichi H, Lin W. In vivo measurement of blood oxygen saturation
using magnetic resonance imaging: a direct validation of the blood
oxygen level-dependent concept in functional brain imaging. Hum
Brain Mapp 1997;5:341–346.
6. Yablonskiy DA, Haacke EM. Theory of NMR signal behavior in magnetically inhomogeneous tissues: the static dephasing regime. Magn
Reson Med 1994;32:749 –763.
7. Buxton RB, Wong EC, Frank LR. Dynamics of blood flow and oxygenation changes during brain activation: the balloon model. Magn Reson
Med 1998;39:855– 864.
8. Lee SP, Duong TQ, Yang G, Iadecola C, Kim SG. Relative changes of
cerebral arterial and venous blood volumes during increased cerebral
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
blood flow: implications for bold fMRI. Magn Reson Med 2001;45:791–
800.
Li TQ, Haefelin TN, Chan B, Kastrup A, Jonsson T, Glover GH, Moseley
ME. Assessment of hemodynamic response during focal neural activity
in human using bolus tracking, arterial spin labeling and BOLD techniques. Neuroimage 2000;12:442– 451.
Hyder F, Rothman DL, Shulman RG. Total neuroenergetics support
localized brain activity: implications for the interpretation of fMRI.
Proc Natl Acad Sci USA 2002;99:10771–10776.
Rees G, Howseman A, Josephs O, Frith CD, Friston KJ, Frackowiak RSJ,
Turner R. Characterizing the relationship between BOLD contrast and
regional cerebral blood flow measurements by varying the stimulus
presentation rate. Neuroimage 1997;6:270 –278.
Nakai T, Matsuo K, Kato C, Takehara Y, Isoda H, Moriya T, Okada T,
Sakahara H. Post-stimulus response in hemodynamics observed by
functional magnetic resonance imaging— difference between the primary sensorimotor area and the supplementary motor area. Magn Reson Imaging 2000;18:1215–1219.
Kim SG. Progress in understanding functional imaging signals. Proc
Natl Acad Sci USA 2003;100:3550 –3552.
Obata T, Liu TT, Miller KL, Luh WM, Wong EC, Frank LR, Buxton RB.
Discrepancies between BOLD and flow dynamics in primary and supplementary motor areas: application of the balloon model to the interpretation of BOLD transients. Neuroimage 2004;21:144 –153.
Davis TL, Kwong KK, Weisskoff RM, Rosen BR. Calibrated functional
MRI: mapping the dynamics of oxidative metabolism. Proc Natl Acad
Sci USA 1998;95:1834 –1839.
Hoge RD, Atkinson J, Gill B, Crelier GR, Marrett S, Pike GB. Investigation of BOLD signal dependence on cerebral blood flow and oxygen
consumption: the deoxyhemoglobin dilution model. Magn Reson Med
1999;42:849 – 863.
Jezzard P, Ramsey NF. Functional MRI. In: Tofts P, editor. Quantitative
MRI of the brain: measuring changes caused by disease. West Sussex,
UK: John Wiley & Sons; 2003.
Grubb Jr RL, Raichle ME, Eichling JO, Ter Pogossian MM. The effects of
changes in PaCO2 on cerebral blood volume, blood flow, and vascular
mean transit time. Stroke 1974;5:630 – 639.
Ogawa S, Menon RS, Tank DW, Kim S-G, Merkle H, Ellermann JM,
Ugurbil K. Functional brain mapping by blood oxygenation level-dependent contrast magnetic resonance imaging. A comparison of signal
characteristics with a biophysical model. Biophys J 1993;64:803– 812.
Boxerman JL, Bandettini PA, Kwong KK, Baker JR, Davis TL, Rosen BR,
Weisskoff RM. The intravascular contribution to fMRI signal change:
Monte Carlo modeling and diffusion-weighted studies in vivo. Magn
Reson Med 1995;34:4 –10.
Buxton RB. Introduction to functional magnetic resonance imaging.
Cambridge, UK: Cambridge University Press; 2002. p 394 –395.
Hoge RD, Atkinson J, Gill B, Crelier GR, Marrett S, Pike GB. Linear
coupling between cerebral blood flow and oxygen consumption in
activated human cortex. Proc Natl Acad Sci USA 1999;96:9403–9408.
Stefanovic B, Warnking JM, Rylander KM, Pike GB. The effect of global
cerebral vasodilation on focal activation hemodynamics. Neuroimage
2006;30:726 –734.
Uludag K, Dubowitz DJ, Yoder EJ, Restom K, Liu TT, Buxton RB.
Coupling of cerebral blood flow and oxygen consumption during physiological activation and deactivation measured with fMRI. Neuroimage
2004;23:148 –155.
Kastrup A, Krüger G, Neumann-Haefelin T, Glover GH, Moseley ME.
Changes of cerebral blood flow, oxygenation, and oxidative metabolism
during graded motor activation. Neuroimage 2002;15:74 – 82.
Stefanovic B, Warnking JM, Pike GB. Hemodynamic and metabolic
responses to neuronal inhibition. Neuroimage 2004;22:771–778.
Chiarelli PA, Bulte DP, Piechnik S, Jezzard P. Sources of systematic
bias in hypercapnia-calibrated functional MRI estimation of oxygen
metabolism. Neuroimage 2007;34:35– 43.
Chiarelli PA, Bulte DP, Piechnik S, Jezzard P. Sources of systematic
bias in hypercapnia calibrated functional mri estimation of oxygen
metabolism. In: Proceedings of the Annual Meeting of ISMRM, Seattle,
WA, USA, 2006 (Abstract 2764).
Luh WM, Wong EC, Bandettini PA, Hyde JS. QUIPSS II with thin-slice
TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn Reson
Med 1999;41:1246 –1254.
Flow-Metabolism Coupling by MRI
30. Wong EC, Buxton RB, Frank LR. Quantitative imaging of perfusion
using a single subtraction (QUIPSS and QUIPSS II). Magn Reson Med
1998;39:702–708.
31. Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TEJ,
Johansen-Berg H, Bannister PR, De Luca M, Drobnjak I, Flitney DE, Niazy
RK, Saunders J, Vickers J, Zhang Y, De Stefano N, Brady JM, Matthews PM.
Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 2004;23(Suppl 1):S208 –S219.
32. Jenkinson M, Bannister P, Brady M, Smith S. Improved optimization
for the robust and accurate linear registration and motion correction of
brain images. Neuroimage 2002;17:825– 841.
33. Smith SM. Fast robust automated brain extraction. Hum Brain Mapp
2002;17:143–155.
34. Woolrich MW, Ripley BD, Brady M, Smith SM. Temporal autocorrelation in univariate linear modeling of FMRI data. Neuroimage 2001;14:
1370 –1386.
35. Schroeter ML, Kupka T, Mildner T, Uludag K, von Cramon DY.
Investigating the post-stimulus undershoot of the BOLD signal—
a simultaneous fMRI and fNIRS study. Neuroimage 2006;30:349 –
358.
36. Buxton RB, Wong EC, Frank LR. The post-stimulus undershoot of the
functional MRI signal. In: Moonen CTW, Bandettini PA, editors. Functional MRI. New York: Springer; 1999. p 253–262.
547
37. Tjandra T, Brooks JCW, Figueiredo P, Wise R, Matthews PM, Tracey I.
Quantitative assessment of the reproducibility of functional activation
measured with BOLD and MR perfusion imaging: implications for
clinical trial design. Neuroimage 2005;27:393– 401.
38. Kliefoth AB, Grubb Jr RL, Raichle ME. Depression of cerebral oxygen
utilization by hypercapnia in the rhesus monkey. J Neurochem 1979;
32:661– 663.
39. Stein SN, Pollock GH. Central inhibitory effects of carbon dioxide II.
Proc Soc Exp Biol Med 1949;70:290 –291.
40. Kogure K, Busto R, Scheinberg P, Reinmuth O. Dynamics of cerebral
metabolism during moderate hypercapnia. J Neurochem 1975;24:471–
478.
41. Chiarelli PA, Bulte DP, Piechnik S, Jezzard P. Sources of systematic
bias in hypercapnia-calibrated functional mri estimation of oxygen
metabolism. In: Proceedings of the 14th Annual Meeting of ISMRM,
Seattle, WA, USA, 2006 (Abstract 2764).
42. Lund TE, Norgaard MD, Rostrup E, Rowe JB, Paulson OB. Motion or
activity: their role in intra- and inter-subject variation in fMRI. Neuroimage 2005;26:960 –964.
43. Wise RG, Ide K, Poulin MJ, Tracey I. Resting fluctuations in arterial
carbon dioxide induce significant low frequency variations in BOLD
signal. Neuroimage 2004;21:1652–1664.