Global Cerebral Oxidative Metabolism during Hypercapnia and Hypocapnia in Humans: Implications for BOLD fMRI

Abstract

The effect of carbon dioxide (CO₂) on cerebral metabolism is of tremendous interest to functional imaging. In particular, mild-to-moderate hypercapnia is routinely used in calibrated blood oxygen-level dependent (BOLD)-functional magnetic resonance imaging (fMRI)-based quantification of cerebral oxidative metabolism changes (ΔCMRO₂), and relies on the assumption of a stable CMRO₂ during CO₂ challenges. However, this assumption has been challenged by certain animal studies, necessitating its verification in humans and under conditions customary to fMRI. We report, for the first time, on global ΔCMRO₂ measurements made noninvasively in humans during graded hypercapnia and hypocapnia. We used computerized end-tidal CO₂ modulation to minimize undesired concurrent changes in oxygen pressure, and our findings suggest that no significant change in global CMRO₂ is expected at the levels of end-tidal CO₂ changes customary to calibrated BOLD.

Introduction

Carbon dioxide (CO₂) is a potent vasodilator, and its effect on cerebral metabolism is of great interest. Notably, hypercapnia is routinely used in calibrated blood oxygen-level dependent (BOLD) functional magnetic resonance imaging (fMRI) for estimating changes in cerebral oxygen metabolism (ΔCMRO₂), but this methodology is contingent upon the assumption of negligible ΔCMRO₂ during hypercapnia-induced cerebral blood flow (CBF) increases (Davis et al, 1998; Hoge et al, 1999). Although CMRO₂ behavior under manipulations of end-tidal partial pressure of CO₂ (PETCO₂) has been investigated previously in animals (Hino et al, 2000; Horvath et al, 1994; Jones et al, 2005; Martin et al, 2006; McPherson et al, 1991; Yang and Krasney, 1995; Zappe et al, 2008), the wide range of anesthetic agents used, compounded by the diverging levels of PETCO₂ perturbations involved, has made it difficult to draw definitive conclusions regarding the CMRO₂ response during hypercapnia and hypocapnia with regard to human fMRI. Furthermore, the sampling intervals in these experiments were largely inadequate to capture the dynamic CMRO₂ changes related to transient metabolic responses, and the PETCO₂ manipulations were too severe to be directly interpreted for human fMRI.

In spite of this need for verification, CMRO₂ invariability has yet to be established for the range of PETCO₂ values relevant for functional imaging, as the corresponding experiments are technically demanding. Conventionally, human cerebral oxidative metabolism has been assessed invasively through arteriovenous blood oxygenation differences, obtained through internal jugular catheterization. In this study, we build on this well-established concept, and estimate CMRO₂ changes from flow and oxygenation data using a magnetic resonance-equivalent of the Kety–Schmidt technique. In particular, we use pulsed arterial-spin labeling (PASL), phase-contrast (PC) flow imaging, as well as in vivo magnetic resonance oximetry to measure whole-brain CBF and CMRO₂ changes noninvasively in awake humans during graded hypercapnia and hypocapnia. We show that at the levels of PETCO₂ perturbations routinely used in calibrated BOLD, there is no significant global change in CMRO₂.

Materials and methods

All acquisitions were performed using a Siemens Trio 3-T system (Siemens, Erlangen, Germany) on 10 healthy adult subjects (age = 25.3 ± 3.6 years, 5 women) who gave informed consent according to the protocols specified by our Institutional Review Board. The body coil was used for radio frequency transmission and a phased-array neurovascular coil for reception. A three-dimensional 1-mm isotropic resolution T₁-weighted (FLASH (fast low-angle shot)) scan served as the anatomic reference for each subject. Venous blood T₂ was obtained in vivo for each subject in both jugular veins (at the neck level, immediately above the facial vein drainage) using a magnetizationprepared segmented echo-planar imaging sequence (Chen and Pike, 2009; Stefanovic and Pike, 2004). From these T₂ measurements, venous oxygenation (Y) was estimated using the calibration relationship (T₂ dependence on blood oxygenation and Carr-Purcell-Meiboom-Gill (CPMG) refocusing interval) obtained by Chen and Pike (2009) based on in vitro whole human blood relaxometry (Chen and Pike, 2009) and the weak field-inhomogeneity diffusion model (Jensen and Chandra, 2001). Each oximetry measurement lasted 4.5 minutes. In addition, ΔCBF changes were measured using PICORE (proximal inversion with a control for off-resonance effects)-QUIPSS (quantitative imaging of perfusion using a single subtraction) II PASL with asymmetric BASSI (bandwidth-modulated adiabatic selective saturation and inversion) saturation/inversion (Warnking and Pike, 2006) and ASSIST (attenuating the static signal in arterial spin tagging) background suppression (Ye et al, 2000). The imaged slab was positioned approximately parallel to the AC-PC (anterior commissure-posterior commissure) line. It covered 8 slices of 5 mm (separated by 1-mm gaps), with field of view = 256 mm, matrix size = 64 × 64, repetition time = 5 seconds, TI₁ (inversion time) = 700 milliseconds, TI₂ = 1,300 milliseconds, echo time = 25 milliseconds, labeling thickness = 150 mm, and labeling gap = 5 mm. Using the ‘Head’ and ‘Neck’ modes of the neurovascular coil, we optimized the signal-to-noise characteristics for both whole-brain and neck-level acquisitions.

Mild and moderate levels of hypercapnia and hypocapnia were induced through the administration of mixtures of CO₂ and medical air using the RespirAct breathing circuit (Thornhill Research, Toronto, ON, Canada), which provided computerized and independent targeting of PETCO₂ and PETO₂ (partial pressure of end-tidal oxygen) based on sequential gas delivery (Slessarev et al, 2007). This device provided high stability in PETCO₂ targeting, while maintaining significantly more constant PETO₂ levels than manual methods, both of which are crucial to the accurate quantification of CMRO₂ changes due solely to PETCO₂ changes. Each 12-minute PASL measurement consisted of three (60 seconds off:180 seconds on:120 seconds off) paradigms. In addition, two-dimensional cine PC flow data were obtained from five subjects in the internal carotid and vertebral arteries. These vessels account for the majority of cerebral inflow, and the corresponding PC acquisition (2.5 minutes) was used for cross-validation with the PASL data as a measure of global CBF. A finger pulse oximeter provided cardiac triggering and arterial oxygen data.

For each functional run of each subject, subject-specific brain masks were derived by applying an intensity threshold to the PASL control images. The resultant masks were manually verified to ensure exclusion of extracerebral structures, and applied to the PASL data to assess global flow changes. Using the venous Y measurements, ΔCMRO₂ and the neurovascular coupling ratio were estimated on the basis of equation (1), which was derived from the Fick principle (Kety and Schmidt, 1948),

$Y=\frac{\Delta \text{CBF}\cdot (1-\Psi )(Y_{\mathrm{a}}-Y_0)}{(\Delta \text{CBF}+1)}+Y_0$

(1)

where Ψ = ΔCMRO₂/ΔCBF, Y_a and Y₀ are arterial and baseline venous oxygenation levels, respectively. Unconstrained weighted nonlinear least-square curve-fitting (trust-region method) was used to fit the data to the model, with weightings proportional to the inverse s.d. of the individual measurements. All statistical analyses were performed after outlier removal.

Results and Discussion

Typical PASL, internal jugular venous relaxometry, and internal-carotid PC flowmetry images are shown in Figure 1. The average estimated baseline venous oxygenation (Y₀) was 58.2% ± 8.2% (mean ± s.d.), corresponding to an average normocapnic PETCO₂ of 39.9 ± 1.4 mm Hg. The mean measured Y_a as assessed using pulse oximetry was B98%, in line with literature values (Valdez-Lowe et al, 2009; Wagener and Hendricker, 1987). The ΔPETCO₂ measurements associated with all four levels of challenge, namely mild and moderate hypercapnia and hypocapnia, as categorized by fuzzy-cluster classification of the measured ΔPETCO₂ values, are summarized in Table 1. Stability in PETO₂ was maintained during PETCO₂ manipulations in all cases, as exemplified in Figure 2. The corresponding ΔCBF and ΔY measurements are also listed in Table 1. A significant difference in ΔPETCO₂ was found between mild and moderate hypercapnia (P = 3.42 × 10⁻¹⁰), as well as between mild and moderate hypocapnia (P = 9.74 × 10⁻⁸). The same was true for ΔCBF (P = 7.92 × 10⁻⁷) and ΔY (P = 5.01 × 10⁻⁴) for hypercapnia, but not for hypocapnia (mild versus moderate hypocapnia, P = 0.11 and P = 0.43 for ΔCBF and ΔY differences, respectively). In general, the vascular reactivity to hypercapnia was higher than to hypocapnia, reflected by the ΔCBF: ΔPETCO₂ ratios (P = 0.004). However, the ΔY: ΔCBF ratios were higher for hypocapnia than for hypercapnia (P = 0.01). These two ratios were not significantly modulated by the severity of the challenge.

Figure 1

(A) The multislice pulsed arterial-spin labeling (PASL) measurements of cerebral blood flow (CBF) changes (in top box) were compared with phase-contrast flow measurements obtained at the neck level (the bottom box), wherein in vivo jugular oximetry was also performed. A typical CBF activation t-map obtained from PASL and a brain mask derived through intensity thresholding are shown in B and C, respectively. The PASL measurements were validated using phase-contrast magnetic resonance imaging (MRI) measurements in the internal carotid (ICAs) and vertebral arteries (VAs), labeled in D. Finally, venous blood oxygenation measurements were performed using internal jugular vein (IJV) MR oximetry (E).

Figure 2

(A) Moderate hypercapnia and (B) moderate hypocapnia end-tidal partial pressure of CO₂ (PETCO₂) and PETO₂ tracings (top two plots) for a representative subject show that PETO₂ remained largely unchanged during PETCO₂ manipulations, an important aspect in delivering CO₂ challenges in functional magnetic resonance imaging (fMRI). The bottom plots are the corresponding group-averaged global pulsed arterial-spin labeling (PASL) ΔCBF time courses.

Table 1

Physiologic measurements at all four levels of PETCO₂ manipulation (mean ± s.d.); statistically significant differences are denoted by asterisks (*)

Challenge	Level	N	ΔPETCO2 (mm Hg)	ΔCBF (%)	ΔY (%)	ΔCBF:ΔPETCO2 (%/mm Hg)	ΔY:ΔCBF (%/%)	Ψ (%/%)	ΔCMRO2 (%)
Hypercapnia	Mild	10	4.7 ± 0.3	23.6 ± 3.0	4.9 ± 2.0	5.2 ± 0.7	0.26 ± 0.05	0.30 ± 0.17	5.0 ± 3.0
	Moderate	11	9.0 ± 0.3	53.3 ± 3.1	16.9 ± 2.1	5.9 ± 0.3	0.36 ± 0.05	−0.21 ± 0.16	–6.8 ± 8.0
Hypocapnia	Mild	10	−2.9 ± 0.2	−13.2 ± 2.1	−7.5 ± 3.0	4.7 ± 0.8	1.40 ± 0.57	0.18 ± 0.60	10.2 ± 3.0
	Moderate	9	−5.8 ± 0.3	−16.9 ± 2.0	−12.7 ± 2.0	5.2 ± 0.7	0.76 ± 0.25	−0.24 ± 0.60	12.5 ± 9.1
Hypocapnia versus hypercapnia						P = 0.004*	P = 0.01*	P > 0.99	P = 0.04*
Mild versus moderate challenges						P = 0.29	P = 0.18	P > 0.99	P = 0.21

CBF, cerebral blood flow; ΔCMRO₂, cerebral oxidative metabolism change; PETCO₂, end-tidal partial pressure of CO₂.

The means were weighted by the inverse of the measurement uncertainties. No significant difference in Ψ was found between mild and moderate hypercapnia (P > 0.99), or between mild and moderate hypocapnia (P > 0.99). Moreover, the Ψ estimates computed from all hypercapnic (Ψ = −0.12 ± 0.11, mean ± s.d.) and from all hypocapnic (Ψ = 0.06 ± 0.30) cases did not differ significantly (P > 0.99).

Whole-brain PASL ΔCBF in the whole-brain masks and PC flow measurements were found to be equivalent, as detailed in Table 2. Using PASL-based global ΔCBF, global ΔCMRO₂ and Ψ values were estimated for each challenge. The ΔCMRO₂ values computed on the basis of equation (1), did not differ significantly between mild and moderate levels of hypercapnia (P = 0.11) or hypocapnia (P = 0.40). Neither differences were observed when comparing between mild and moderate (hypercapnia + hypocapnia) challenges (P = 0.21), but ΔCMRO₂ during hypercapnic and hypocapnic (mild + moderate) challenges were found to differ (P = 0.04). This trend remained when comparing hypercapnic and hypocapnic conditions solely at the moderate level, but did not reach statistical significance (P = 0.06). With regard to the extent of CBF–CMRO₂ coupling, no significant difference in Ψ was found between mild and moderate hypercapnia (P > 0.99); similar findings apply to mild and moderate hypocapnia P > 0.99). Similarly, the values of Ψ computed from all hypercapnic (Ψ = −0.12 ± 0.11) and from all hypocapnic (Ψ = 0.06 ± 0.30, mean ± s.d.) cases did not differ significantly (P > 0.99). Thus, a single fit was performed on data across all subjects and challenges, yielding Ψ = −0.12 ± 0.14, with a correlation of 0.78 and a root-mean-squared error of 1.37% (Figure 3), which is indistinguishable from zero (P > 0.99). We did not observe a significant difference (P > 0.99) between Ψ estimated using PASL and PC flowmetry (Ψ = 0.02 ± 0.17). Finally, assuming Ψ=0, and using the measured ΔCBF, the expected venous oxygenation was calculated on the basis of equation (1) for all conditions. The resulting Y values were not significantly different from the corresponding measured values (P = 0.48). We further noted that although we adopted the group mean arterial oxygenation of 98% for these calculations, coinciding with commonly assumed values (Hoogenraad et al, 1998), our observations were robust against physiologically plausible variations in arterial oxygenation. For Y_a varying between 96 and 100%, the resultant fractional uncertainties ΔCMRO₂ and Ψ were both within ∼10%.

Figure 3

The relationship derived from the nonlinear fit of the data (closed dots) to the model (mesh) are shown, and the model fit was characterized by a correlation coefficient of R=0.78 and a root-mean-square error of 1.37%. The finding of a small Ψ (−0.12 ± 0.14) shows that global CMRO₂ is largely invariant during these challenges.

Table 2

Comparison of hypercapnic CBF modulation measured using PASL and PC techniques

Level	PASL				PC
	ΔCBF (%)	ΔY (%)	ΔY:ΔCBF (%/%)	ΔCMRO2 (%)	ΔCBF (%)	ΔY (%)	ΔY:ΔCBF (%/%)	ΔCMRO2 (%)
Mild	23.6 ± 3.0	4.9 ± 2.0	0.26 ± 0.05	5.0 ± 3.0	22.0 ± 6.9	5.4 ± 1.2	0.3 ± 0.1	5.3 ± 3.4
Moderate	53.3 ± 3.1	16.9 ± 2.1	0.36 ± 0.05	−6.8 ± 8.0	30.7 ± 8.7	11.0 ± 1.7	0.5 ± 0.2	−4.3 ± 7.0
Both	N/A	N/A	0.34 ± 0.04	−1.5 ± 4.7	N/A	N/A	0.4 ± 0.1	0.5 ± 0.9
Mild versus moderate	N/A	N/A	P = 0.23	P = 0.11	N/A	N/A	P = 0.09	P = 0.15

CBF, cerebral blood flow; ΔCMRO₂, cerebral oxidative metabolism change; N/A, not applicable; PASL, pulsed arterial-spin labeling; PC, phase contrast; PETCO₂, end-tidal partial pressure of CO₂.

The PASL and PC experiments used similar levels of PETCO₂ perturbations, and the corresponding derived parameters, ΔY:ΔCBF and ΔCMRO₂, did not differ significantly.

Our measurement of cerebral venous oxygenation (Y) by T₂-based magnetic resonance oximetry was in good agreement with jugular bulb measurements (McCormick et al, 1991), and covaried with CBF. Moreover, the comparability between PASL and PC measurements of the CBF response to graded hypercapnia and hypocapnia supports the use of PASL for measuring whole-brain CBF changes. In addition, our observations of cerebral vascular reactivity to hypercapnia are in line with previous reports (McPherson et al, 1991; Noth et al, 2006). Our flow and oximetry methodology derives from principles similar to those shown recently by Xu et al (2009), although venous flow was used in the latter case and is customarily more challenging to assess than arterial flow. Notwithstanding, a unique advantage of our methodology is our strict control of PETO₂ during CO₂ manipulations. It is well known that conventional approaches to PETCO₂ manipulation can introduce nonnegligible variations in PETO₂; increases in PETO₂, consistently observed during hypercapnic challenges in the literature, can effectively undermine the effect of CO₂ (Losert et al, 2002) and hence can introduce biases in ΔCBF. In this study, we continuously monitored and minimized changes in PETO₂ through the use of computerized targeting facilitated by RespirAct, and were therefore able to achieve the isolation of CO₂ effects, which is a key requirement for the experiment at hand, and constitutes a basic assumption in numerous fMRI experiments using CO₂ challenges. Our results suggest that there are no significant alterations in global CMRO₂ due to these challenges, consistent with predictions obtained using the deoxyhemoglobin dilution model (Hoge et al, 1999).

Although the current measurements show an absence of substantial global CMRO₂ modulation by low levels of PETCO₂ modulation, they do not preclude the possibility of spatial heterogeneities in the CMRO₂ response to capnic challenges. For instance, a slight increase in neuronal activity modulations by fluctuating inhaled CO₂ in the pons and medulla may occur, while decreases in motor synaptic synchrony (Biswal et al, 1997) have also been observed, potentially linked to CO₂-mediated suspension in oxidative metabolism fluctuations (Hudetz et al, 1998) and slight alterations in the intrathalamic network (Steriade et al, 1996). On the basis of our experimental observations, we expect the spatial distribution and the magnitude of this response to also exhibit sizeable subject dependence. In view of the sources of variation, we sought to assess group trends, and to verify whether the assumption of negligibility of ΔCMRO₂ globally is reasonable for the purpose of calibrated BOLD.

Several animal studies, invaluable precursors to similar experiments in humans, document the CMRO₂ response during hypercapnic challenges, but thus far, the conclusions have been inconsistent. The CMRO₂ invariability was reported by Hino et al (2000), McPherson et al (1991), and Yang and Krasney (1995), in anesthetized lambs and dogs, as well as in awake sheep, respectively, at ΔPETCO₂ values of up to 30 mm Hg. In contrast, Horvath et al (1994) showed a ∼30% decrease in CMRO₂ in anesthetized rats (ΔPETCO₂≈50 mm Hg), whereas Jones et al (2005) found a 30% CMRO₂ increase during 10% hypercapnia in anesthetized rats (8 minutes, ΔPETCO₂≈50 mm Hg), and negligible ΔCMRO₂ at 5% CO₂ (ΔPETCO₂≈25 mm Hg). The latter result was corroborated by findings in awake rats by Martin et al (2006) (5% CO₂), but questioned by data obtained from the study by Zappe et al (2008), which showed a ∼15% CMRO₂ reduction during 6% CO₂ (ΔPETCO₂ = 24 mm Hg) in anesthetized macaques. These variations in outcome can in part be attributed to the substantial methodological differences, because: (1) different types of anesthesia and states of alertness can alter the neuronal response significantly and (2) the levels and durations of PETCO₂ manipulations across these studies varied substantially (the duration varying from 60 seconds to 96 hour). More importantly, the comparatively high severity of the above PETCO₂ perturbations preclude comparability with commonly used human fMRI paradigms. Further complications stem from evidence in the literature that the general response to inhaled CO₂ may be highly species- and subject-dependent, as a 5% CO₂ challenge in awake humans rarely produces as much as a 25 mm Hg PETCO₂ increase. Given these considerations, we regard the widely varying experimental protocols to be a key potential contributor to the discrepancies among animal and human literature, which must be interpreted with care.

In summary, in this study, we sought to address accurate ΔCMRO₂ measurement during hypercapnia and hypocapnia in awake humans, which has traditionally been hindered by nontrivial methodological challenges, particularly, the reproducible targeting of PETCO₂ values and in the effective minimization of concurrent PETO₂ variations. We explored the persisting question of whether the levels of end-tidal CO₂ manipulations commonly used in human calibrated fMRI experiments can be expected to elicit significant changes in global CMRO₂. We circumvented some of the key traditional challenges by using computerized PETCO₂ and PETO₂ targeting, and performed global ΔCMRO₂ measurements in awake humans during graded hypercapnia and hypocapnia within a range chosen to address fMRI applications (more specifically, ranging from −6 to + 9 mm Hg). Under such conditions, we found no evidence of significant changes in global CMRO₂ during PETCO₂ manipulations. Therefore, our findings support the assumption of negligible ΔCMRO₂ under the conditions widely used in calibrated fMRI studies of awake humans.