Regional and interindividual relationships between cerebral perfusion and oxygen metabolism.

Quantitative measurements of resting cerebral blood flow (CBF) and metabolic rate of oxygen (CMRO2) show large between-subject and regional variability, but the relationships between CBF and CMRO2 measurements regionally and globally are not fully established. Here, we investigated the between-subject and regional associations between CBF and CMRO2 measures with independent and quantitative PET techniques. We included resting CBF and CMRO2 measurements from 50 healthy volunteers (aged 22-81 years), and calculated the regional and global values of oxygen delivery (DO2) and oxygen extraction fraction (OEF). Linear mixed model analysis showed that CBF and CMRO2 measurements were closely associated regionally, but no significant between-subject association could be demonstrated, even when adjusting for arterial PCO2 and hemoglobin concentration. The analysis also showed regional differences of OEF, reflecting variable relationship between DO2 and CMRO2, resulting in lower estimates of OEF in thalami, brainstem, and mesial temporal cortices, and higher estimates of OEF in occipital cortex. In the present study, we demonstrated no between-subject association of quantitative measurements of CBF and CMRO2 in healthy subjects. Thus, quantitative measurements of CBF did not reflect the underlying between-subject variability of oxygen metabolism measures, mainly because of large interindividual OEF variability not accounted for by PCO2 and hemoglobin concentration.


INTRODUCTION
Extensive quantitative measurements of cerebral blood flow (CBF) have been made in the past 70 years with a wide range of techniques. Measurements of global CBF (gCBF) in the resting human brain exhibit large between-subject variability with upper and lower normal limits differing by a factor of 2 between individuals within groups of healthy subjects (1)(2)(3)(4)(5). Classically, CBF is held to be adjusted to the metabolic demands of the brain, and thus is considered to be an index of neuronal function and integrity under normal circumstances (6). Regional CBF (rCBF) measurements confirm similar regional distributions for cerebral metabolic rate of oxygen (rCMRO 2 ) and of glucose (rCMR glc ), in healthy (1,7) and diseased brains (8)(9)(10), and gCBF values vary with global CMR glc (gCMR glc ) and CMRO 2 (gCMRO 2 ) estimates across altered states of consciousness (11,12). The constant relationships among these measures agree with the oxidative metabolism of glucose in neurons as the predominant source of ATP in the brain.
It is well known that measurements of gCMRO 2 show similar large between-subject differences as measurements of gCBF (1,4,13). The oxygen extraction fraction (OEF) relates estimates of CBF and oxygen delivery (DO 2 , i.e., the total transport of oxygen to the brain by blood flow, of which a fraction is extracted due to brain metabolism) to estimates of CMRO 2 . Thus, assuming that a primary function of perfusion is the delivery of oxygen to the brain, and that extraction of oxygen occurs by metabolism, then OEF should be constant both among regions and among subjects. Indeed, different organ systems can be characterized by very different and organ-specific OEF values (14), and a uniform OEF has been suggested to be a defining feature of the resting brain (7,15). If OEF is constant, the relationship between DO 2 (CBF scaled by oxygen content) and CMRO 2 must be linear. If this hypothesis does not hold, OEF cannot be constant. This would suggest that factors other than oxygen delivery are contributing significantly to variation of CBF across subjects and regions.
Invasive measurements of global OEF (gOEF) show considerable between-subject variability (16). Variations of arterial PCO 2 or O 2 content are known to be associated with changes of CBF, but generally are not thought to influence CMRO 2 (17)(18)(19), and therefore would be expected to be major causes of deviation from the predicted linear relationship between CBF and CMRO 2 . Also, plasma caffeine and hemoglobin (Hgb) levels are known to be related to changes of resting CBF and the observed association of CBF with metabolism (20,21).
A recent analysis of same-day measurements of rCBF (and rDO 2 ), gCMRO 2 , and rCMR glc using independent techniques (in the sense that each quantity is derived from a separate measurement) failed to confirm direct between-subject associations of these measures, also with attention to arterial PCO 2 and caffeine levels (22). The analysis applied MRI-based measurements of gOEF and gCMRO 2 that have not been validated against accepted reference techniques in humans. MRI-based measurements of gOEF provide much simpler noninvasive access to quantification of oxygen metabolism than classically more invasive techniques, but provide no information on regional differences of OEF. Previous studies also indicated smaller regional (7,23) and age-related differences of regional OEF (rOEF) (24) and also regional differences of the ratio of rCBF to rCMR glc , primarily in phylogenetically older infratentorial and mesial temporal structures (25). For measurements of gOEF to be used to assess regional oxygen metabolism, increased insight into regional OEF variations relative to global variations is required. Thus, it is not clear how to interpret variations of absolute quantitative CBF measurements in the resting healthy brain, or to determine to what extent differences of resting CBF regionally and between-subjects do correspond to differences in CMRO 2 .
In previous reports of CMRO 2 and CBF measurements in healthy subjects in general, authors did not distinguish between-subject from within-subject (regional) associations (1), and have applied quantitative methods by which values of CBF were used to calculate values of CMRO 2 , e.g., the Kety-Schmidt technique where CBF is multiplied with arteriovenous oxygen difference (26) (28,29), respectively, such that the values may allow a more unbiased assessment of the relationship between the two measures. Although the one-step methodology has been used less frequently than the three-step method, the results have been shown to be in good agreement with those of the three-step method (29) and have been applied in a number of publications over the past three decades (as more suitable for potential non steady states intervening in hour-long procedures).
Typically, single-study data sets include only a smaller number of subjects, rendering it difficult to include more than one or two covariates in the analysis. We analyzed a pooled data set of sequentially obtained resting rCBF and rCMRO 2 measurements from previously published studies (30)(31)(32)(33)(34)(35)(36), obtained by application of PET techniques for quantification of the two variables that are independent (28,29). This unique data set permitted unbiased assessment of the relationship between the two measures, and hence let us test the following hypotheses: 1) Global CBF and CMRO 2 are positively correlated in subjects, and deviations from this relationship are mainly attributed to variations of arterial oxygen content and PCO 2 .
2) The relationship between rCBF and rCMRO 2 is constant across different brain regions, i.e., there are no significant differences in rOEF between brain regions.

Data
We reanalyzed resting rCBF and rCMRO 2 measurements performed in seven previously published studies (30)(31)(32)(33)(34)(35)(36). The analyzed data set partially overlaps with a previously published pooled data set (24). We included only subjects in whom complete sets of data (CBF, CMRO 2 , MRI, arterial blood gases, and Hgb) were available, constituting a total of 50 healthy subjects (16 females and 34 males, median age 47 yr, range 21-81 yr). All participants were free of neurological diseases and did not take any medication that may have interfered with CBF or CMRO 2 measurements. The individual studies were approved by the official Science Ethics Committee of Central Region Denmark, and all studies were conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.
In each subject, a single measurement of rCBF with [ 15 O] H 2 O and one or two measurements of rCMRO 2 with gaseous [ 15 O]O 2 were obtained during a single session with few minutes between measurements. All experiments were performed in daytime between 9 AM and 3 PM. Except for data from 11 participants from one study (36), participants were instructed to keep eyes open during imaging sessions. Details of data acquisition and generation of parameter maps have been published previously (24). All PET images were acquired using the same ECAT EXACT HR 47 PET system (CTI/Siemens, Knoxville, TN) with an isotropic resolution of 4.6 mm and were fitted with a 68 Ge source for attenuation correction. A 3-min dynamic acquisition was obtained following intravenous injection of $500 MBq of [ 15 2 ). Images were reconstructed into 21 frames: 12 Â 5 s, 6 Â 10 s, 3 Â 20 s (128 Â 128 Â 47 matrix, voxel size 2 Â 2 Â 3.1 mm 3 using filtered back projection, and applying a 0.5 per cycle Hahn filter).
Arterial blood samples were obtained by automatic (Allogg AB, Mariefred, Sweden) or manual sampling (temporal resolution of 5 s) and corrected for delay using the software Fitdelay (Turku PET Centre, www.turkupetcentre.net/petanalysis/ tpcclib/doc/fitdelay.html). Arterial blood samples obtained at every measurement were analyzed for arterial blood gases and Hgb (ABL-555 or ABL-300, Radiometer, Copenhagen, Denmark), and the average of at least two measurements from each participant was used in the analysis.
A three-dimensional (3D) T1-weighted MRI was obtained on a 1.0 T GE Signa using a 3D-SPGR sequence or a 3.0 T Signa Excite GE Magnet using a 3D-IR-fSPGR sequence.

Generation of CBF and CMRO 2 Parametric Maps
All original data were reprocessed for the pooled analysis, as previously described (24). A 6-mm filter was applied for smoothing, and parametric images of the unidirectional blood-brain clearance of radiolabeled water (K 1 H 2 O) and oxygen (K 1 O 2 ) of either tracer were calculated using a one-tissue compartment model, as described by Ohta et al. (28,29

Volume of Interest Analysis
CBF and CMRO 2 parameter maps and T1 MRI estimates from each subject were coregistered and normalized to MNI standard space, as previously described (24). All further analyses were performed using FSL BET, and FAST for tissue segmentation (FMRIB, www.fsl.fmrib.ox.ac.uk) and in-house software for Matlab (Mathworks, Natick, MA). T1 MRI was segmented using FSL FAST to obtain a subject-specific gray matter (GM) mask.
First, mean volume of interest (VOI) values of CBF and CMRO 2 from standard AAL (automated anatomic labeling) regions (rCBF and rCMRO 2 ) from MNI structural atlas and Harvard-Oxford anatomical atlas supplied with FSL ( Fig.   1) were extracted using FSL and applying a 25% probability threshold. For the whole brain region (gCBF and gCMRO 2 ), we used the standard atlas MNI brain mask. For cortical regions, a combined subject-specific gray matter (GM) mask (applying 50% probability threshold) þ standard AAL region mask was applied to minimize partial volume effects. For white matter regions, mean values of standard ellipsoid VOIs in the centrum semioval were used.
For each region DO 2 was calculated as and assuming an arterial oxygen saturation of 98%, OEF was calculated as Relative rOEF (rOEF rel ) was calculated as the ratio of rOEF to gOEF.
To take into account a variable anatomical basis of CBF and CMRO 2 parameter maps, in particular for the cerebellum, mean VOI values from each map were obtained from identical VOIs. We calculated the actual coverage, i.e., the fraction of positive voxels within anatomical VOI, for each VOI in each subject, allowing the possible influence of incomplete coverage to be assessed. Exploratory analysis revealed a possible interaction between low cerebellar coverage and the calculated OEF. Cerebellum values in subjects with <40% coverage (n = 6) were consequently not included in the analysis.

Statistics
Descriptive statistics are summarized as means ± SD. We calculated between-region coefficient of variation (CV) from the average regional values (i.e., one value per VOI), and interindividual CV from the global values of each subject. CV was calculated as the standard deviations divided by the population mean.
The crude correlations between regional CMRO 2 and CBF estimates were initially evaluated in scatterplots and subsequently by simple linear regression for each region. When assessing univariate correlation of CBF and CMRO 2 , from multiple regions, the marginal coefficient of determination (R 2 ), taking into account multiple data points in each subject, was calculated by comparing the sum of variance components from random component models (with random effects similar to the mixed models below) with and without CMRO 2 as fixed effect.
To further assess the effect of covariates on regional CBF estimates, we applied a two-level linear mixed model with rCBF as dependent variable and CMRO 2 , brain region, age, Hgb, and PCO 2 as fixed effects, and data set and subject as random effects. To separate global and regional associations of rCMRO 2 with rCBF, both gCMRO 2 and the deviation of rCMRO 2 (DrCMRO 2 ) from gCMRO 2 were entered as fixed effects.
Thus, we expressed regional CBF in the j'th region of the i'th subject from the h'th data set as, where n and ɛ denote between-study and between-subject residual error terms, respectively, and the prefixes r and g refer to regional and global values, respectively. The linear mixed model thus provided estimates of regional (within-subject) and between-subject effects in a single model. Explorative analysis of the data included showed that no effects of sex other than that on CBF related to lower Hgb in women (2) could be demonstrated, and sex consequently was not included in any of the models. An identical model with DO 2 as dependent variable (and omitting Hgb as fixed effect) was also investigated. To account for variance inhomogeneity and the fact that some regions were more strongly correlated than others, P values and confidence intervals from the mixed-model analyses were based on robust standard errors. For analysis of regional differences of rOEF and rOEF rel , we tested similar two-level models (omitting CMRO 2 as fixed effect). When comparing regional average values, a linear mixed model with region as fixed effect and subject and study as random effects was applied, with parietal cortex as reference. Distributions of residuals were assessed by inspection of histograms and residual plots for model control.
To assess the risk of false-positive findings, all P values in the main mixed-model analyses were in addition adjusted for multiple testing using the method of Benjamini and Hochberg (37) that controls the false discovery rate, such that the reported significance is 5% likely to be a false positive.
Goodness of fit was evaluated by residual plots. Analyses were performed with SAS version 9.4 (SAS Institute Inc., Cary, NC). Mixed-model analysis was performed using PROC MIXED module. Robust standard errors were estimated using the sandwich covariance estimator. Adjusted P values were computed using PROC MULTTEST.
Linear mixed models showed that regional CBF was positively associated with the regional deviation of rCMRO 2 from gCMRO 2 (DCMRO 2 ) but not with gCMRO 2, and also no association of global CBF with whole brain CMRO 2 , thus indicating only significant within-subject, but no betweensubject association of CBF with CMRO 2 ( Table 2 and Fig.  4). In contrast, rCBF was positively associated with PCO 2 and inversely associated with both age and Hgb, indicating significant between-subject associations.
For individual regions, we found a positive significant between-subject association between rCBF and rCMRO 2 in the caudate (R 2 = 0.34, P < 0.001), but no significant association for any other region (Fig. 5). Distinct and statistically significant regional variations of OEF were observed not only with lower rOEF values in particular in brainstem, thalamus, and mesial temporal cortex, but also with slightly lower values in frontal cortex and putamen, whereas higher values were found in occipital cortex and less pronounced in temporal cortex compared with parietal cortex as reference ( Fig. 2 and Table 2). Occipital rOEF and rOEF rel estimates were not significantly different between participants studied with eyes open and eyes closed.

DISCUSSION
Here, we tested interindividual and regional associations of CBF and CMRO 2 by means of independent methods applied to a large data set. The main findings include 1) high between-subject OEF variability that obscured direct between-subject associations of CBF with CMRO 2 , even when accounted for interindividual differences of PCO 2 and Hgb and 2) values of rOEF that appeared to vary in a distinct regional pattern.
The average whole brain, white matter, gray matter, and cerebellar values of CBF, CMRO 2 , and OEF generally were within the range of previously reported normal values for PET (1,38,39), despite differences of study populations, applied techniques, and data analysis. Some prior PET studies revealed between-subject associations of rCBF with rCMRO 2 (1, 5, 39), but the method used was not optimal for this purpose, with rCMRO 2 calculated from values of rOEF scaled by the oxygen content and rCBF. Analyses of data from studies with the Kety-Schmidt technique (19) suggested not only a linear relationship between gCBF and gCMRO 2 values, but also in this case, the association was biased by calculating gCMRO 2 from arterialvenous O 2 differences scaled by gCBF. The finding of high between-subject variability in gOEF [normal range (means ± 2SD) = 0.28-0.60] is in good agreement with similar large normal variability in gOEF [normal range (means ± 2 SD) = 0.29-0.54] reported using invasive measurements (16) and also with a number of studies using PET (1,38) or MRI susceptometry (22). As suggested by the present analysis and in a recent study (22), the variability exceeds what can be explained by factors known to alter the relationship between gCBF and gCMRO 2 , e.g., arterial PCO 2 , Hgb, and caffeine.
Although regional differences in CMRO 2 and CBF probably mainly are determined by tissue composition and to a lesser degree by partial volume effects, the potential effects would not influence estimation of OEF. The present analysis adds to previous reports of regional deviations in OEF in the occipital lobe (7) and primary sensory-motor cortex (23) by also demonstrating significantly lower values in thalamus, brainstem, and mesial temporal cortex, and confirming similar OEF in cerebellum and cerebral cortex (40,41). Raichle et al. (7) suggested that increased rOEF in occipital cortex may be the result of deactivation related to subjects imaged with eyes closed. We also found increased values of rOEF in occipital cortex, but most subjects were imaged with eyes open, and thus did not support the hypothesis of deactivation. The average parametric rOEF maps are very similar to those reported by Raichle et al. (35), showing lower rOEF values in thalami and brain stem, but these regions were not investigated statistically by the authors. Here, we find that the regional differences largely agree with previous studies, and the present analysis adds to the existing knowledge by a more rigorous statistical analysis of the cortical differences.
A number of factors potentially can alter the relationship between delivery and consumption of substrates of energy metabolism globally and regionally, e.g., variable efficiency due to uncoupling (13), use of alternate energy substrates such as lactate (42) or ketone bodies (43), presence of aerobic glycolysis (44), or variable capillary density or capillary transit time heterogeneity that both could require higher perfusion to maintain tissue O 2 pressure (45). In particular, the possibility of regionally variable contributions of aerobic glycolysis has been the subject of recent debate (46)(47)(48). Previous studies have reported higher rCBF/rCMR glc ratios in infratentorial structures, thalami, and mesial temporal cortex (22,25,49), and the lower rOEF in the same regions thus appears to outbalance the relatively increased perfusion in these areas. Still, higher rCBF/rCMR glc ratios despite similar rOEF in cerebellum compared with cerebral cortex may support a specific absence of aerobic glycolysis in the cerebellum (47). As quantitative measurements of glucose metabolism were not available, the basis of the residual variable OEF cannot be assessed from the present study and inferences as to its relationship with glucose metabolism will be speculative. Multiple factors may also influence oxygen transport and reactions at the cellular and subcellular levels, such as variation of dendrite abundance and expression of specific transport proteins, but these contributions are difficult to assess in experiments with humans as in the present case, where the resolution of PET does not allow us to study the underlying processes. In addition, a number of environmental factors and the constitution of the research subject may also influence CBF, as extensively reviewed recently (50), but as we analyzed a pooled data set in which only the key information was available in all studies, the present analysis is limited to the data presented. Also, method bias likely may vary regionally, and such method Â region interaction potentially may induce regional bias. However, it is difficult to make inferences of how such effects may have contributed to our findings, neither from our data nor from the literature.
We confirmed a much lower average regional (within-subject) variation of rOEF than the between-subject variation. Measurements of gOEF, e.g., by means of the technically less demanding MRI oximetry (51), thus may account for the larger part of total OEF variability. Still, these smaller regional variations should be remembered when regional cerebral metabolism is studied. Importantly, the average regional deviations reported in this study refer to resting healthy brain, but do not necessarily apply to altered states, e.g., sleep or activation, nor diseased brains.
It should also be stressed that the observed lack of int-erindividual association between quantitative measurements CBF 1.00 ± 0.00 P values' regional differences from linear mixed model with region as fixed effect, and subject and study as random effects (see statistics) using parietal cortex as reference.
Adj. P value corrected for false discovery rate, † regional OEF/whole brain OEF. CBF, cerebral blood flow; CMRO 2 , cerebral metabolic rate of oxygen; OEF, oxygen extraction fraction; VOI, volume of interest. and CMRO 2 (and consequently the association with neuronal function) is the result of measurements performed in healthy subjects and does not relate to the diseased brain. Indeed, reduced (or increased) CBF regionally or globally is a common feature of many neurological diseases, as a cause or as a consequence of brain dysfunction. Thus, among diseased subjects, a between-subject association of CBF with CMRO 2 reflecting disease severity may exist. High between-subject variabilities of blood flow and DO 2 measures are not unique to the brain. Resting measurements of function or perfusion of most organ systems probably shows considerable variability among healthy subjects, as reflected by the use of confidence intervals for laboratory tests. Although similar methodological issues also apply to investigations of other organ systems, the links perfusion, energy metabolism, and function, may be highly organ specific, and the findings of the present study cannot necessarily be generalized.
For correlations of absolute physiological measurements, method accuracy is of great importance. Here, for CBF, we applied a one-tissue compartment model with correction for arterial blood volume. We found the expected associations of values of CBF with both arterial PCO 2 and Hgb that support the validity of our [ 15 O]H 2 O PET measurements for quantitation of cerebral perfusion. Limited water extraction at high flow rates may cause underestimation of rCBF in cortical regions, but the data do not indicate regional or global ceiling effects. Also, from stoichiometric considerations, a linear relationship is expected for the resting brain within normal physiological range, and neither our data nor the statistical analysis further suggested that the linear model used is not an appropriate approximation.
Due to the lack of a proper reference, it is difficult to assess the accuracy of in vivo rCMRO 2 , measurements in general. A fair agreement of the one-step approach with the longer duration three-step approach was found in the original paper describing the method (29), and we regard the applied method for quantification of CMRO 2 as ideal for the main purpose of the present analysis, i.e., assessment of the association of CBF to CMRO 2 without potential bias resulting from the use of one parameter to calculate another or from departures from physiological steady state. As only single measurements of rCBF and rCMRO 2 were available, method imprecision could not be accounted for in the analysis, and the potential effect of attenuation bias cannot be assessed. To our knowledge, data on test-retest variability of the specific techniques applied have not been published. Preferably, duplicate measurements should have been performed, which would allow us to estimate and account for method imprecisions in the statistical analysis, but this was not standard practice in all studies to reduce radiation exposure and the duration of the experiment (minimizing cost and risk of clotting of arterial cannula). Instead, we used averages of duplicate measurements when performed, and the linear model analysis further will tend to account for measurement imprecision in the outcome parameter, whereas imprecision in explanatory variables will induce attenuation bias (and thus underestimation of the effect). Here, we can only assess the potential influence of method precision on the results presented using data from test-retest studies applying different approaches, although unpublished analysis from a prior study (24) has shown similar precision as for the three-step approach (1), and the CBF approach applied is not expected to differ from published data in terms of precision. To get a crude estimate of the  influence of method precision, attenuation factors for the two parameters were first calculated from within-and between-subject variance estimates based on previously published test-rest studies of rCBF and rCMRO 2 measurements (using the 3-step approach) (1,52,53). Then, a post hoc power calculation was performed showing that a study with same sample size had a power of ! 0.98 for detecting a significant (2-sided P 0.05, Pearson correlation) true correlation of 0.8 (true R 2 = 0.64 or observed R = 0.27-0.44 depending on the variance estimates used) between gCBF and gCMRO 2 . Assuming similar method precision of the methods applied in the present study, the calculation indicates that the study had sufficient power to detect a close between-subject association of CBF and CMRO 2 . Errors related to analytical precision of blood sample analysis were expected to be very low compared with precision of PET measurements and were not included in the power calculation. Unlike the conventional three-step PET approach, where rOEF values as calculated directly from the blood and brain activities of the three ([ 15  PET studies, we here proceeded indirectly from VOI values of parametric rCBF and rCMRO 2 maps. This procedure is likely to lower the precision of rOEF estimates due to added method imprecision. In accordance with this, the CVs of rOEF values in the present study are slightly higher than reported in studies using the three-step approach (range 17%-22% vs. 15%-18%) (23,40). Still, we find the determination of average rOEF values by this indirect approach not necessarily inferior to other methods of measurement of rOEF. A general limitation of [ 15 O]H 2 O studies for CBF measurements is that extraction is not 100%, but is flow dependent and may possibly vary regionally (54), and potentially could contribute to the observed variable relationships between CBF and CMRO 2 . However, the values are relatively constant within the physiological range of CBF values, and the reported regional differences in extraction are not consistent in the literature (54)(55)(56) and do not appear to correlate well with the regional OEF difference reported in the present and previous studies.
In addition to the above methodological considerations, the analysis has some limitations related to the study design. First of all, we analyzed data obtained over a 10-yr period by some different and some overlapping authors of seven different studies. To obtain sufficient statistical power, the number of required independent samples exceeds what is usually performed in single studies, and as for previous studies, we therefore pooled data from smaller individual studies using identical imaging protocols (2,7,57). Pooling inevitably adds some heterogeneity to the data. To minimize any study or time-related effects, all PET data were reanalyzed by the same investigator. Still, some study-related factors potentially may have influenced the results. MR images for brain segmentation were acquired on two different MRI systems at different field strengths, which could introduce bias in brain segmentation. Using FSL for segmentation, bias related to field strength has been reported to be low for whole brain structures (1%-3%), and as identical regions were used for regional analysis in each subject, we believe that the influence of scanner-related differences have only minimal influence on the findings reported. Similarly, both manual and automatic blood sampling were used, which can influence estimate of CBF and CMRO 2 , but as data are paired, these errors are expected to cancel out each other within each study. Finally, study-related offsets are further accounted for in the statistical analysis by including data set as a random effect.
Although measurements of CBF and CMRO 2 were not obtained simultaneously, they were performed in close sequence only minutes apart, and we find that the lack of association is unlikely to be attributable to changes of physiological state. Physiological covariates included Hgb and arterial PCO 2 values, but not caffeine, which may also influence CBF and alter the relationship between CBF and CMRO 2 (21). According to laboratory practices at the time of studies, no particular instructions of participants as to intake of caffeine before the experiment were given, and we cannot rule out variable caffeine levels that may have added noise to the interindividual association of CBF with CMRO 2 .
In conclusion, the present study failed to confirm an interindividual association of cerebral perfusion and oxygen consumption in the healthy resting brain due to very high interindividual variability of oxygen extraction fraction not accounted for by standard physiological covariates. Thus, the findings suggest that resting CBF is regulated by other mechanisms than CMRO 2 alone, and that quantitative measurements of resting CBF cannot be used a surrogate measure of absolute cerebral metabolism or function in healthy subjects, although from the present analysis we cannot determine the basis of the apparent disassociation. The study further shows smaller, but significant and distinct regional differences in the relationship between perfusion and oxygen metabolism resulting in similar regional variations in OEF in the resting brain. , is driven by a strong regional association (middle: n = 12 regions), whereas no between-subject association is found when including only whole brain values (right: n = 50 subjects). R 2 and P values are calculated for simple linear correlation and regression line shown with 95% confidence intervals.  Figure 5. Correlation of cerebral blood flow (CBF) and metabolic rate of oxygen (CMRO 2 ) by volume of interest. Number of observations in each region, n = 50 (except for cerebellum, n = 44). R 2 , P values, and regression line shown with 95% confidence intervals are shown.