Hemodynamic and metabolic changes during hypercapnia with normoxia and hyperoxia using pCASL and TRUST MRI in healthy adults

Abstract

Blood oxygenation level-dependent (BOLD) or arterial spin labeling (ASL) MRI with hypercapnic stimuli allow for measuring cerebrovascular reactivity (CVR). Hypercapnic stimuli are also employed in calibrated BOLD functional MRI for quantifying neuronally-evoked changes in cerebral oxygen metabolism (CMRO₂). It is often assumed that hypercapnic stimuli (with or without hyperoxia) are iso-metabolic; increasing arterial CO₂ or O₂ does not affect CMRO₂. We evaluated the null hypothesis that two common hypercapnic stimuli, ‘CO₂ in air’ and carbogen, are iso-metabolic. TRUST and ASL MRI were used to measure the cerebral venous oxygenation and cerebral blood flow (CBF), from which the oxygen extraction fraction (OEF) and CMRO₂ were calculated for room-air, ‘CO₂ in air’ and carbogen. As expected, CBF significantly increased (9.9% ± 9.3% and 12.1% ± 8.8% for ‘CO₂ in air’ and carbogen, respectively). CMRO₂ decreased for ‘CO₂ in air’ (−13.4% ± 13.0%, p < 0.01) compared to room-air, while the CMRO₂ during carbogen did not significantly change. Our findings indicate that ‘CO₂ in air’ is not iso-metabolic, while carbogen appears to elicit a mixed effect; the CMRO₂ reduction during hypercapnia is mitigated when including hyperoxia. These findings can be important for interpreting measurements using hypercapnic or hypercapnic-hyperoxic (carbogen) stimuli.

Introduction

Advanced MRI techniques provide an avenue to image functional parameters such as oxygen extraction fraction (OEF), cerebral blood flow (CBF), cerebral blood volume (CBV), and the cerebral metabolic rate of oxygen consumption (CMRO₂). Thereby, these MRI techniques can provide metabolic and hemodynamic metrics similar to those obtained using PET, but non-invasively and at higher spatial and temporal resolution. One such approach is calibrated fMRI (or calibrated BOLD) that can be used to scale task-related BOLD fMRI responses to the underlying CMRO₂ changes. ¹ This technique has found applications to examine metabolic demands associated with functional tasks but also ageing and brain disease.^2–5

The BOLD calibration models rely on measurements of both BOLD and CBF signal changes during either hypercapnic respiratory challenges or a combination of hypercapnic-hyperoxic challenges.^6–9 Modulations in arterial gas tensions evoke predictable physiological responses that are used to contextualize fMRI signal changes.^10–13 In short, hypercapnia is a potent vasodilator that leads to significant increases in CBV, CBF, and accompanying increases in venous blood oxygenation, while hyperoxia is thought to only modulate the blood oxygenation through plasma-dissolved O₂ and increased hemoglobin (Hb) bound O₂. This dual-action convolutes the net hemodynamic and metabolic response through its influence on the oxygen saturation curve via the Bohr effect (i.e. how CO₂ affects the binding affinity of Hb for O₂) and the Haldane effect (i.e. how O₂ affects the binding affinity of Hb for CO₂). An original assumption during the early development of the calibration model was that increasing arterial levels of CO₂ and O₂ had a negligible effect on neuronal metabolism and thus CMRO₂. ⁴ This notion has since been challenged, and correction methods have been devised to account for the possibility that changing arterial O₂ or CO₂ tensions do indeed modulate neuronal functioning and CMRO₂^9,11–18 Previous simulation work on calibrated BOLD models has reported that for non iso-metabolic hypercapnic challenges for calibration, one can find a significant error in estimating the basal OEF and activation-induced CMRO₂ changes in calibrated BOLD studies.^10,19,20 Correction methods for situations in which arterial O₂ and CO₂ tensions change simultaneously have not yet been adopted.

Quantitative assessment of venous oxygenation (Y_v), while considering the combined effects of changing CBF and OEF, provides an avenue to examine the iso-metabolic assumptions associated with respiratory stimuli via inferred changes in CMRO₂. This can be performed non-invasively using a technique known as T₂ -Relaxation-Under-Spin-Tagging (TRUST) MRI. ²¹ Direct knowledge of the effects that changing arterial O₂ and CO₂ tensions have on CMRO₂ has widespread implications. For instance, appropriate corrections can improve the accuracy of fMRI techniques for neuroscientific applications, refine BOLD signal models that simulate magnetic susceptibility effects under different physiological conditions,^22–24 and improve the interpretation of CVR for clinical applications via a more robust understanding of concomitant CMRO₂ changes.^25–28

The use of respiratory challenges during MRI in various patient populations is becoming more widespread. In many of these cases, the pathophysiology of neurological and cerebrovascular disease leads to altered cerebral metabolism. With this in mind, it becomes essential to identify co-factors that influence cerebral metabolism and are accounted for accordingly. Therefore, this study aimed to evaluate the assumption that hypercapnic stimuli applied in healthy adults are iso-metabolic. To achieve this, we used ASL and TRUST MRI for measures of global CBF and Y_v to compare measures of CMRO₂ under two commonly used hypercapnic conditions (‘CO₂ in air’: 5% CO₂ + 21% O₂ + 74% N₂; and carbogen: 5% CO₂ and 95% O₂). The CMRO₂ value for each condition was calculated and compared with the CMRO₂ value obtained during the breathing of medical air (referred to as ‘room-air’).

Methods

Volunteer demographics

Healthy volunteers (n = 10; 3 F/7 M; age = 29.4 ± 3.4 years) provided informed, written consent following the ethical standards of the Vanderbilt University Institutional Review Board, the Vanderbilt University Human Research Protection Program, as well as with the Helsinki Declaration of 1975 (and as revised in 1983). All components of this study were performed in compliance with the Health Insurance Portability and Accountability Act. All study components were reviewed and approved by the local Institutional Review Board (IRB Study 111116).

MRI experiment

All MRI measurements were performed at 3.0 T (Philips Healthcare, Best, The Netherlands). For a schematic overview of the MRI experiment see Figure 1. Participants were fitted with a nasal cannula (Salter Labs, Arvin, CA, USA, no. 4000) to monitor the end-tidal partial pressure of CO₂ (pEtCO₂) and a custom non-rebreathing face mask (Salter Labs, no. 8005) for gas stimulus administration. The masks were close-fitting and covered the nose and mouth. Elastic straps were used to reduce leakage. Gases were delivered from compressed cylinders outside the scan room and administered at 12 L/min. This flow rate was optimized in preliminary studies and was found to provide sufficient gas delivery in the presence of potential small leaks in the mask, while also maintaining comfort. Physiological monitoring was performed using an Invivo Research (Gainesville, FL, USA, 3150 MRI) monitor and a remote monitor (Millenia Vital System, Gainsville, FL, USA, 3115 MVS). Monitored parameters included partial pressure of end-tidal CO₂ (pEtCO₂ in mmHg), and peripheral arterial oxygenation (Y_a). The repeatability of this setup has been reported previously ²⁹ and a similar setup is used in a previously reported work. ²⁴

Figure 1.

Schematic overview of the experimental design. a) the subject in the MRI with the three different inspired gases and the physiological measurements of pEtCO₂ (purple) from the breathing mask, arterial oxygenation Y_a (red) from a pulse-oximeter, venous oxygenation Y_v (orange), and CBF (blue). Conceptual planning for cerebral venous oxygenation (Y_v) using T2-relaxation-underspin-tagging (TRUST) MRI is depicted by the orange region, where the dotted lines represent the measurement plane and the circle the location of the occipital part of the superior sagittal sinus. Whole-brain CBF values were acquired using a pseudo-continuous arterial spin labeling pCASL sequence (planning depicted in blue). b) The [O₂]_a, [O₂]_v, p_aO_2, and p_vO₂ values were calculated using a physiological blood oxygen content model (Dash et al.) and were applied for computing CMRO₂ and OEF using the formulas shown. c) Experimental design of the gas delivery paradigm. The dashed blue line represents the rest period (∼30 s) given to allow pEtCO₂ and Y_v to equilibrate. The order of ‘CO₂ in air’ and carbogen conditions and the TRUST and pCASL scan was randomized between subjects.

Venous oxygenation (Y_v) MRI measurements were performed on the occipital part of the superior sagittal sinus using TRUST MRI (TR = 3s, TI = 1.2s, voxel size = 3.4 × 3.4 × 5 mm³, four T₂ weightings (effective TEs: 0, 40, 80, and 160 ms), with a τ_CPMG = 10 ms, and 3 averages per T₂ weighting yielding a total scan duration of 1 min 12 s.^30,31 CBF measurements were performed using pCASL MRI with a multi-slice echo-planar imaging readout. Acquisition parameters were post-label delay (PLD) = 1.7 s, label duration =1.5 s, TR/TE = 3900/13.1 ms, measurements = 13, field-of-view = 240 × 240 × 119 mm³, spatial resolution = 3 × 3 × 7 mm³, slices = 17, SENSE factor = 1.8. The TRUST scan was repeated once for the hypercapnic breathing conditions but was only performed once for the room-air conditions. TRUST and pCASL data were sequentially acquired throughout the following paradigm: 3 min room-air (‘room-air 1’) – ∼4 min of 5% CO₂ balanced with medical-grade atmospheric air (5% CO₂/21% O₂/74% N₂; ‘CO₂ in air’: as described earlier ²⁴ ) – 3 min room-air (‘room-air 2’) – ∼4 min carbogen (5% CO₂, 95% CO₂; Figure 1). For each hypercapnic breathing condition, the two Y_v values were first estimated from the two TRUST measurements and subsequently averaged into a single Y_v value per condition. In addition, a calibration M₀ scan was acquired for both groups for CBF quantification, using identical acquisition geometry as the pCASL scan, but with TR = 15 s, and the spin labeling pulse train turned off. The order of ‘CO₂ in air’ and carbogen and the TRUST and pCASL scan was randomized between subjects and time (∼30 s) was allowed for pEtCO₂ and Y_v to equilibrate between the two hypercarbic conditions.

Analysis

CBF quantification

CBF quantification was performed using FSL BASIL; ³² pCASL label and control images were pair-wise subtracted, and a single-compartment kinetic model was applied to the data using tissue-blood partition coefficient of water λ = 0.9 ml/g and pCASL labeling efficiency α = 0.85. ³³ Slice time correction was incorporated by using slice-specific PLD values. For equilibrium blood water magnetization, the calibration M₀ image was used. An arterial blood water T₁ (T_1a) reduction from 1.65 s (used for the normoxic conditions) to 1.49 s (hyperoxic conditions) was used to account for hyperoxia-dependent changes related to the carbogen condition, as reported previously. ³⁴

For tissue T₁ and arterial (bolus) arrival time, gray matter values (GM) were used in BASIL FSL, 1.3 s ³⁵ and 1.3 s^29,36 for pCASL, respectively, including a GM arrival time reduction of 5% for the hypercapnic conditions.^29,36 Global CBF was estimated using subject-specific brain tissue segmentation masks and associated tissue probability maps of GM and white matter (WM). We focused on global CBF as opposed to solely GM CBF as the TRUST measurement for Y_v(%) also samples a global value. Tissue segmentation was performed on the M₀ image that exhibited adequate GM and WM contrast using FSL FAST. ³⁷ The global CBF estimation excluded subarachnoidal space and ventricular cerebral spinal fluid (CSF), cerebellum and brainstem regions. CSF regions were removed using the subject-specific segmentations masks from FSL FAST, the cerebellum and brainstem regions were removed using the MNI Structural Atlas and the Harvard-Oxford Structural Atlas ³⁸ MNI based (Montreal Neurological Institute) ROIs, respectively, available in FSL. ³⁹ Segmentation and atlas ROIs were constrained by the subject’s whole-brain mask obtained from the M₀ image using FSL BET. ⁴⁰ Note that GM values were used for the tissue T₁ and arterial arrival time for all voxels using FSL BASIL for the CBF quantification. However, these values are not appropriate for WM CBF included in this study to obtain global CBF and can lead to an underestimation for WM CBF (See Supplementary Material Figure 1). We computed a correction factor for WM CBF using the WM tissue T₁ and arterial arrival time, 0.84 s ³⁵ and 1.7 s, ³⁶ respectively, as recently reported by Juttakonda et al. ³⁶ (see Supplementary Material for computation). Average global CBF was recomputed by summing the GM and WM CBF maps weighted by the tissue probability maps and the WM correction factor for WM (∼1.20, see Supplementary Material and Supplementary Figure 1). Voxels with outlier CBF values were discarded for absolute CBF values larger than two standard deviations above the mean global CBF, i.e. |CBF_voxel| > mean (global CBF) ± 2 std (global CBF). To visualize the group average CBF map results, the subjects’ CBF maps were first spatially normalized to the standard MNI 2 mm³ stereotaxic space using FSL FLIRT with 12 degrees of freedom affine registration and sinc interpolation. ⁴¹ The subject averaged pCASL control image was used as an intermediate step.

CMRO2 and OEF estimation

Calculation of Y_v from the TRUST MRI data was done using the method described previously. ¹³ For the repeated scans, i.e. for the ‘CO₂ in air’ and ‘carbogen’ conditions, the fitted Y_v values were averaged. To account for the plasma dissolved O₂ present during the hyperoxic condition, we computed the total arterial O₂ content ([O₂]_a), venous O₂ content ([O₂]_v), and partial pressure of venous O₂ (p_vO₂) using the physiological model by Dash et al. ⁴² that models the blood O₂ and CO₂ content by generating the hemoglobin-O₂ and CO₂ dissociation (saturation) curve and computing the plasma-dissolved O₂^24,42 The model takes into account the Bohr effect for which O₂ binding to hemoglobin is inversely related to the presence of CO₂ by using the measured pEtCO₂ (see Table 1). This will yield subject-specific and breathing condition-specific hemoglobin O₂ saturation curves, dependent on blood pO₂ and pCO₂. We will show the O₂ saturation curve by Dash et al. ⁴² and the commonly used saturation curve by Severinghaus (1), that depends only on the pO₂: ⁴³

$$Y_{\left(\mathrm{\%}\right)}=\left(\frac{1}{\frac{1}{p{O}_{\mathit{2}}^{\mathit{3}}+150·p{O}_2}·23400+1}\right)·100\mathrm{\%}$$ (1)

Table 1.

Group average cerebral blood flow (CBF), venous oxygenation (Y_v), oxygen extraction fraction (OEF), and cerebral metabolic rate of oxygen (CMRO₂) results for the room-air, ‘CO₂ in air’ and carbogen breathing conditions. The group average change in end-tidal CO₂ (pEtCO₂), global CBF, Y_v, OEF, and CMRO₂ are also shown.

	end-tidal CO2	cerebral blood flow			venous oxygenation		oxygen extraction fraction		cerebral metabolic rate of O2
condition	ΔpEtCO2 (mmHg)	CBF (ml/100g/min)	ΔCBF (ml/100g/min)	ΔCBF (%)	Yv (%)	ΔYv (%)a	OEF	ΔOEF (%)	CMRO2 (µmol/100g/min)	ΔCMRO2 (%)b
room-air 1		54.9 ± 8.2			67.3 ± 3.6		0.3 ± 0.0		144.5 ± 25.5
room-air 2		52.3 ± 6.8			65.3 ± 5.1		0.3 ± 0.0		147.8 ± 33.3
CO2 in air	4.2 ± 1.7***	61.0 ± 8.0	6.1 ± 5.4*	9.9 ± 9.3*	73.9 ± 4.7	8.8 ± 3.2***	0.2 ± 0.1	-30.8 ± 15.2***	123.9 ± 24.8	−13.4 ± 13.0*
carbogen	4.5 ± 2.2***	60.0 ± 9.1	7.7 ± 5.4**	12.1 ± 8.8**	78.6 ± 7.3	16.3 ± 8.9***	0.3 ± 0.1	-39.6 ± 52.2*	139.3 ± 34.1	−2.0 ± 27.0n

Values represent the group average ± standard deviation.

^aThis is the fractional change in percentage in Y_v, i.e. not percentage points.

^bWhen ignoring the venous plasma dissolved O₂, we find a ΔCMRO₂(%) of −11.30% and 3.11% for ‘CO₂ in air’ and carbogen, respectively, i.e. showing a similar change. When also ignoring the arterial plasma dissolved O₂, we find a ΔCMRO₂(%) of −12.06% and −22.30% for ‘CO₂ in air’ and carbogen, respectively, showing a substantial difference for the carbogen condition. Note, when using the commonly used Severinghaus O₂ saturation curve instead of the model by Dash et al. ⁴² (see Figure 2(b)), we find a ΔCMRO₂(%) of -9.33% and 0.35% for ‘CO₂ in air’ and carbogen, respectively, also showing a similar change.

*p-value <0.05 significant change found (Student’s T-test) with respect to the preceding room-air condition, **p-value <0.005, ***p-value <0.001, ⁿno significant difference found.

The partial pressure of alveolar O₂ (p_AO₂) was assumed 104 mmHg respectively for the ‘CO₂ in air’ and room-air conditions. ⁴⁴ For these normoxic conditions, alveolar, p_AO₂, was converted to arterial O₂ pressure (p_aO₂) using the alveolar-arterial O₂ pressure gradient, which depends on the p_AO₂ and age. ¹² For the carbogen condition, we used the values from previous reports on direct measurements of arterial partial pressure PaO₂ for carbogen in healthy subjects that used an arterial line during PET examination:

p_aO_2,carbogen = 460 mmHg.^45,46 The venous partial pressure of CO₂ (p_VCO₂) was offset by +5 mmHg compared to the measured pEtCO₂. ⁴⁴ The p_vCO₂ estimates are needed to incorporate the Bohr effect; when locally CO₂ partial pressure is increased, the hemoglobin affinity for O₂ is decreased, which we can expect in peripheral and cerebral tissue and under hypercapnic conditions resulting in a right shift of the HbO₂ curve. The mathematical model put forward by Dash et al. ⁴² incorporates the p_vCO₂ in the K_HbO₂ factor, the apparent equilibrium constant for binding O₂ to hemoglobin. The K_HbO₂ factor itself is a complex equation describing the binding kinematics of O₂ to hemoglobin (see equations 1a,b and 3a,b in Dash et al. ⁴² ) Note that changes in pH, temperature, and 2,3 -DPG will also shift the HbO₂ curve and can in principle be incorporated into the physiological model. Hematocrit (Hct) values of 0.42 for males and 0.4 for females were assumed.

OEF and CMRO₂ were computed using formulas (2) and (3) using equations (4) and (5) for the calculation of [O₂]_a and [O₂]_v, i.e. the sum of the hemoglobin bound O₂ and plasma dissolved O₂ content.

$$\text{OEF}=\frac{{\left[O_2\right]}_a–{\left[O_2\right]}_v}{{\left[O_2\right]}_a}$$ (2)

$$\text{CMR}{\text{O}}_{\text{2}}=\text{CBF}·({\left[O_2\right]}_a–{\left[O_2\right]}_v)$$ (3)

with$${\left[O_2\right]}_a=Y_a·C_h+p_a{O}_2·C_d$$ (4)

and$${\left[O_2\right]}_v=Y_v·C_h+p_v{O}_2·C_d$$ (5)

Constants C_h (912 μmol O₂/100 ml blood for a Hct of 0.45, but adjusted for the assumed Hct values for male and female subjects ⁴² ) and C_d (0.138 μmol O₂/100 ml blood/mmHg p_aO₂) are the hemoglobin O₂ carrying capacity and blood plasma O₂ dissolving capacity, respectively. ¹³ Dash et al. 2016 model ⁴² can be downloaded (JSIM and Matlab code) from the NSR Physiome Project repository that contains integrative and descriptive models on human physiology: www.imagwiki.nibib.nih.gov/physiome/jsim/models/webmodel/NSR/SHbO2CO2Dash2016/.

Effect of arterial blood water T1a on CBF and CMRO₂ quantification

The choice of arterial blood water T_1a can have a considerable impact on the absolute CBF and CMRO₂ quantification, and especially during hyperoxic conditions where substantial T_1a changes can be expected. ³⁴ Therefore, we performed a sensitivity analysis to investigate the influence of varying T_1a on CBF and CMRO₂ quantification, caused by for instance hyperoxia but also Hct variations (see Supplementary Figure 2).

In addition, we investigated whether different T_1a scenarios impacted the CBF, ΔCBF, and importantly the CMRO₂ and ΔCMRO₂ quantification, using commonly used values at 3T for normoxic (T_1,NO) and hyperoxic (T_1,HO) conditions. For the normoxic conditions, T_1a values were used as reported by Lu et al. ⁴⁷ (T_1,NO = 1.65s, bovine blood, recommended ASL Whitepaper value ⁴⁸ ) by Pilkinton et al. ⁴⁹ (T_1,NO = 1.669s, rat blood) commonly used in calibrated BOLD studies, ⁵⁰ and the recently modelled and measured value by Li et al.^51,52 for human blood (T_1,NO = 1.898s, human blood).

For the hyperoxic condition (assuming a p_aO₂ = 460 mmHg for carbogen), we used the values as reported by Siero et al., ³⁴ T_1,HO = 1.49s based on T_1,NO = 1.65 s (Scenario I), the reported hyperoxic T_1a relativity by Ma et al., ⁵³ yielding a T_1,HO = 1.472s based on T_1,NO = 1.65 s (Scenario II), the report by Pilkinton et al., ⁴⁹ T_1,HO = 1.527s based on T_1,NO = 1.669 s (Scenario III), and the model by Li et al. ⁵¹ where we incorporated the hyperoxic relaxivity by Ma et al., ⁵³ yielding a T_1,HO = 1.743s based on T_1,NO = 1.898 s by Li et al. ⁵¹ (Scenario IV). In summary, four scenarios with different T_1,NO and T_1,HO combinations were used to assess the effect on the CBF, ΔCBF, CMRO₂ and ΔCMRO₂ quantification (see Supplementary Figure 3).

Statistical analysis

Statistical analysis was performed to assess significant differences in the change in hemodynamic parameters for each separate hypercapnic condition (‘CO₂ in air’ versus preceding room-air and carbogen versus preceding room-air) using Student’s T-test in SPSS Statistics version 26 (IBM Business Analytics, New York, USA). The distribution of the data was visually checked with histograms and Q-Q plots, and formally with the Kolmogorov-Smirnov test. A p-value <0.05 was considered significant.

Results

All ten subjects completed the study with no reported adverse effects. All data were normally distributed. The pEtCO₂ averaged across subjects increased significantly (p < 0.001) from the ‘room-air 1’ (45.0 ± 3.6 mmHg) to the ‘CO₂ in air’ conditions (49.2 ± 2.7 mmHg) and from the ‘room-air 2’ (43.7 ± 3.6 mmHg) to the carbogen (48.1 ± 3.0 mmHg) conditions. The change in pEtCO₂ (ΔpEtCO₂) is shown in Table 1 for all subjects. Arterial oxygenation (Y_a) for ‘room-air 1’, ‘room-air 2’, ‘CO₂ in air’ and ‘carbogen’ conditions were 98.2 ± 0.6%, 98.4 ± 0.4%, 98.5 ± 0.6% and 98.9 ± 0.4%, respectively. Relative to the corresponding room-air conditions, Y_a increased less for ‘CO₂ in air’ than for the carbogen (p = 0.01) condition.

Figure 2(a) shows the hemoglobin bound O₂ curve and the plasma dissolved O₂ curve as a function of the partial pressure of O₂ (pO₂). These curves were used to show the range of HbO₂ and plasma O₂, for both the arterial and venous networks, using the subject-specific Y_v and P_aO₂ values for all three breathing conditions. The curves were computed using the Dash et al. ⁴² physiological model, using the (assumed) Hct and measured pEtCO₂ as the input for each subject, and then averaged across subjects. Note the high p_aO₂ for the carbogen condition and the associated plasma dissolved O₂ content; venous plasma dissolved O₂ plays a negligible role in the arteriovenous difference to compute CMRO₂. Figure 2(b) depicts a selected portion of the O₂ saturation curve derived from the Severinghaus equation (dotted light-blue), and the model by Dash et al. ⁴² that incorporates the dependencies on pCO₂ and Hct, as well as the ‘right shift’ that occurs under hypercapnia due to the Bohr effect. Note that the effect of the latter is negligible for the arteriovenous O₂ difference for all hypercapnic breathing conditions. In addition, the effect of hyperoxia during carbogen inhalation did not notably increase the venous blood CO₂ content through the Haldane effect (<1%, data not shown). The blood CO₂ content is mostly in the form of bicarbonate and an order of magnitude less as dissolved CO₂ and hemoglobin bound CO₂. See Figure 3 for the values of the blood O₂ content and the arteriovenous difference for all breathing conditions.

Figure 2.

Oxygen saturation and content in blood for the different breathing conditions. The measured pEtCO₂ values and assumed hematocrit values were used as input for the Dash et al. model to generate the subject-specific curves on saturation, hemoglobin bound and plasma dissolved O₂ content over a range of pO₂ between 0 and 550 mmHg. a) Group average hemoglobin bound O₂ saturation curve (light blue) as computed using Dash et al. physiological model ⁴² and the plasma dissolved O₂ curve (magenta) as a function of the partial pressure of O₂ (pO₂). On the right y-axis, the measured ranges of hemoglobin blood oxygenation Y_a(%) and Y_v(%) for the different breathing conditions (room-air in blue, ‘CO₂ in air’ in orange, carbogen in red, and the corresponding O₂ content in μmol/ml blood on the left y-axis ([HbO₂]_a and [HbO₂]_v ranges). The associated partial pressure ranges of O₂ (p_vO₂ and p_aO₂) found via the O₂ saturation curve (light blue) are shown on the x-axis. Note the high p_aO₂ for the carbogen condition and the associated plasma dissolved O₂ content shown on the bottom left (red). b) A zoomed part of the O₂ saturation curve (light blue in a)) showing the traditional O₂ saturation curve by Severinghaus (dotted light-blue) compared to the revised model by Dash et al. with dependency on the subject’s pCO₂ and Hct. The hypercapnic conditions induce a right shift caused by the Bohr effect, shown by the arrow (‘CO₂ in air’ in orange, carbogen in red). The effect of this right shift, however, on the arteriovenous O₂ difference is negligible for all breathing conditions. See Figure 3 for the O₂ content and the arteriovenous difference values (boxplots) for all breathing conditions.

Figure 3.

Boxplots showing the group average O₂ content in μmol per ml blood for hemoglobin bound O₂ and plasma dissolved O₂ for the arterial and venous blood respectively, and the arteriovenous difference in O₂ content needed to compute the CMRO₂; [HbO₂]_a and [HbO₂]_v, [plasma O₂]_a, and [plasma O₂]_v, the total blood O₂ content [O₂]_a and [O₂]_v, and [O₂]_a-[O₂]_v for the different breathing conditions. Noticeable is the increased venous hemoglobin bound O₂ ([HbO₂]_v) for the hypercapnic conditions and the much-increased plasma dissolved O₂ (arterial, [plasma O₂]_a) content for the carbogen condition. Also, note the much smaller y-axis scale for the plasma dissolved O₂ content, showing that the venous plasma dissolved O₂ plays a negligible role in the arteriovenous difference to compute CMRO₂ for all breathing conditions. The boxplots show the minimum, maximum, median and interquartile range, open circles denote outliers.

The group average global CBF maps showed similar increases in CBF during the ‘CO₂ in air’ and carbogen conditions compared to room-air (Figure 4). Significant changes in global CBF, Y_v and OEF for ‘CO₂ in air’ and carbogen were observed (Figure 5, Table 1). The percentage CBF increase for ‘CO₂ in air’ and carbogen was 9.9 ± 9.3% and 12.1 ± 8.8%, respectively. The absolute CBF increase per mmHg pEtCO₂ for ‘CO₂ in air’ and carbogen were 1.7 ± 1.5 and 2.8 ± 3.2 ml/100g/min/mmHg, respectively. Relative and absolute CBF changes between each hypercapnic condition were not statistically significant (Table 1). Compared to the room-air condition, venous oxygenation Y_v increased by 8.8 ± 3.2% (p < 0.001) for the ‘CO₂ in air’ condition. As expected, a more considerable increase in Y_v was observed for the carbogen condition (16.3 ± 8.9%, p < 0.001) due to the hyperoxic gas mixture in addition to the increased blood flow. No significant differences in OEF were found between ‘CO₂ in air’ and carbogen conditions; the reduction in OEF was 30.8 ± 15.2% (p < 0.0001) and 39.6 ± 52.2% (p < 0.05) respectively compared to room-air (see Table 1). Note that the amount of plasma dissolved O₂ was included in the computation of the OEF and CMRO₂ for both conditions (see Methods).

Figure 4.

Group average CBF maps for the different conditions where a notable and similar increase in CBF is observed for ‘CO₂ in air’ and carbogen (see also Table 1). The individual maps were registered to MNI space before averaging and are overlaid on the 2 mm MNI brain template.

Figure 5.

Boxplots showing the group average global CBF, venous oxygenation (Y_v), computed oxygen extraction fraction (OEF), and cerebral metabolic rate of oxygen (CMRO₂) for the different conditions. Notable increases in CBF and Y_v are observed for both the ‘CO₂ in air’ and carbogen conditions, with a more considerable Y_v increase for the carbogen condition, as expected. A similar reduction in OEF is seen for both conditions. Only significant CMRO₂ changes are observed for the ‘CO₂ in air’ condition. The CBF increase for the ‘CO₂ in air’ and carbogen conditions was not significantly different. *p-value <0.05 significant change found (Student’s T-test) with respect to the preceding room-air condition, **p-value <0.005, ***p-value <0.001. The boxplots show the minimum, maximum, median and interquartile range, open circles denote outliers.

Significant CMRO₂ changes were only observed for the ‘CO₂ in air’ condition, showing a decrease of 13.4 ± 13.0% (p < 0.01) (see Table 1). ΔCMRO₂ did not significantly change for the carbogen condition, −2.0 ± 27.0%. Note, when using the commonly used Severinghaus O₂ saturation curve instead of the model by Dash et al. ⁴² (see Figure 2(b)), we find a ΔCMRO₂(%) of -10.4% and 0.3% for ‘CO₂ in air’ and carbogen, respectively, also showing a similar change. When ignoring the venous plasma dissolved O₂ in the CMRO₂ computation, the estimated ΔCMRO₂(%) were -12.7% and -1.3% for ‘CO₂ in air’ and carbogen, respectively. When also ignoring the arterial plasma dissolved O₂, ΔCMRO₂(%) was -13.8% and -20.6% for ‘CO₂ in air’ and carbogen, respectively. Values for all subjects are shown in Supplementary Tables 1 and 2. We performed a sensitivity analysis of variations in arterial blood water T_1a on CBF and CMRO₂ quantification caused by, for instance, hyperoxia and Hct variations. An increase in the hyperoxic T_1,HO value of 5% from the reference T_1,HO (=1.49 ms) leads to a decrease of CBF and CMRO₂ of ∼4.0%, while a T_1,HO decrease of 5% leads to an increase of CBF and CMRO₂ of ∼4.7% (Supplementary Figure 2). We also investigated whether different T_1a scenarios impacted the results using commonly used and recently reported T_1,NO and T_1,HO values. The absolute CBF and CMRO₂ results change in value as expected, however, the ΔCBF and notably the ΔCMRO₂ changes did not change significantly for the different T₁ scenarios (Supplementary Figure 3).

Discussion

Main findings

Using a combination of quantitative CBF and Y_v measurements along with physiological modeling based on measured arterial blood gas values, our results reinforce the notion that hypercapnia is not an iso-metabolic stimulus. Our main findings were threefold: 1) global CMRO₂ reduces with normoxic-hypercapnia; 2) correction for changes in the oxygen saturation curve due to hypercapnia did not significantly affect the calculation of global CMRO₂; 3) inhalation of carbogen gas appears to elicit a mixed effect where the reduction in CMRO₂ seen during normoxic-hypercapnia is mitigated with the inclusion of a hyperoxic stimulus.

Our results showed that the inspiration of 5% CO₂ in room-air led to an average decrease in global CMRO₂ of approximately 13.4%. This falls directly in line with the 13.4% reduction reported by Xu et al. ¹² Similar findings have been reported by Thesen et al., ¹⁸ who measured magnetoencephalogram responses while breathing air and a 5% CO₂ mixture and showed clear decreases in event-related field potentials under hypercapnia. ¹⁸ Also, for non-human primates, a clear reduction of CMRO₂ with hypercapnia has been reported.^14,16

The notion that CMRO₂ reductions follow a linear behavior for increasing levels of arterial CO₂ forms the basis of the updated BOLD calibration model reported by Driver et al. ¹¹ Several reports, including those by Jain et al. ⁵⁴ and Chen et al., ⁵⁵ showed no significant difference in CMRO₂ under hypercapnia. Our measured OEF values (0.32 for room-air, 0.25 for ‘CO₂ in air’, and 0.26 for carbogen) were in line with previous work reported in healthy subjects using ASL ⁵⁶ and PET. ⁴⁶ The decrease in OEF for the CO₂ condition can be explained by the decrease in CMRO₂ and increase in CBF, while the decrease in OEF for the carbogen condition may be (partly) attributed to the higher oxygen content in the arterial blood and the increase in CBF.

The changes in end-tidal CO₂ we observed during ‘CO₂ in air’ (∼4.2 mmHg) were lower than the 8-10mmHg changes typically observed in healthy subjects using a similar stimulus. Lower values may be attributed to potential gas leaking from the masks, leading to less efficient gas administration. The average baseline pEtCO₂ value measured in our subjects during the room-air condition was 44.3 ± 3.5 mmHg. These values are higher than the ∼40mmHg typically quoted in physiology textbooks and higher than what was reported by Chen et al., ⁵⁵ and Jain et al. ⁵⁴ in healthy subjects. The elevated baseline CO₂ in our experiments likely stems from the fact that we used a facemask and did not clamp end-tidal gas values. This may have effectively increased dead-space, leading to more rebreathing and increased pEtCO₂ as has been shown recently for the use of surgical facemasks during MRI acquisition. ⁵⁷ The facemask use and its effect on dead-space may also have contributed to higher room-air Y_v values in our subjects. Our reported Y_v during the two room-air conditions, 67.3 ± 3.6%, and 65.3 ± 5.1%, however, are in line with Jain et al. ⁵⁴ but are higher than Chen et al. ⁵⁵ It should be noted that both studies reported higher increases in CBF during hypercapnia, which may account for the differences in the estimated CMRO₂ change. The elevated room-air pEtCO₂ that we measured could have mitigated subsequent increases in pEtCO₂ due to a reduced alveolar-arterial CO₂ gradient during the CO₂ stimulus. For pre-dilated baseline states (due to CO₂ buildup in the mask), it is conceivable that further increases in arterial CO₂ may push both absolute CBF^58,59 and associated BOLD ²⁵ signals when considering calibrated MRI beyond the known linear response regime. We measured an increase in CBF of approximately 1.7 ± 1.5 and 2.8 ± 3.2 ml/100g/min per mmHg change in pEtCO₂ during the ‘CO₂ in air’ and carbogen conditions, respectively. These increases are in line with results observed for both carbogen (5% O₂ and 95% CO₂) and (5%) CO₂-enriched air inhalation in healthy volunteers. ⁶⁰

When measuring changes in physiological parameters in response to external stimuli, the accuracy of the stimulus delivery as well as the measurement of CBF and Y_v is an important concern. Particularly since the data interpretation relies on physiological modeling, our methods could have benefitted from tighter control and targeting of arterial blood gases via the use of a computer-controlled gas delivery system. This could provide more accurate measurements of end-tidal O₂ for more accurate estimates of blood O₂ content using physiological modeling by Dash et al. ⁴² Also, there is the notion of adequate sensitivity; the report by Xu et al. ¹² showed that ΔCMRO₂ is proportional to ΔpEtCO₂, suggesting a dose-dependent effect of CO₂ on CMRO₂. ¹² Therefore, a sufficiently high ΔpEtCO₂ should be attained to allow significant observations of CMRO₂ reductions during hypercapnia. In addition, the suggested dose-dependency of CO₂ on ΔCMRO₂ would mean that discrepant ΔpEtCO₂ values will lead to discrepant ΔCMRO₂ findings. In line with this, discrepant findings on ΔCMRO₂ during hypercapnia can also be caused by differences in the duration of the hypercapnic stimulus even though similar ΔpEtCO₂ was reached. Future work on modulation of the duration of hypercapnic stimulus would shed light on this.

Assuming iso-metabolic challenges when performing calibrated BOLD experiments may lead to a systemic bias, as shown by previous simulation and experimental studies.^10,19,20 A CO₂-dependent reduction in CMRO₂ (as has been shown in this work) would lead to an overestimation of the M-value and a concordant overestimation of the maximum possible BOLD signal change, which translates into an up to 50% overestimation in basal OEF. ¹⁹ For activation-induced ΔCMRO₂ estimates, one can expect a close to linear behavior between calibration bias and the overestimation of ΔCMRO₂. This translates to a bias in the estimated OEF and CMRO₂ changes (ΔCMRO₂) during task-evoked calibrated BOLD studies. For example, Griffeth et al. ²⁰ (see Figure 8A in their report) showed that the bias in estimated ΔCMRO₂ during activation was in the range of 5%–10% percentage point for 10%-15% reduction CMRO₂ during hypercapnic calibration.

Further work by Blockley et al. ¹⁰ and Merola et al. ¹⁹ provide a sense of how large potential errors might be due to violation of the iso-metabolic assumption.^10,19 The effect of a non-isometabolic stimulus on CVR experiments will depend on the type of technique used. As BOLD is sensitive to changes in deoxyhemoglobin; a decrease in CMRO₂ during hypercapnia could lead to an overestimation of the BOLD response and thus in an overestimation of the BOLD CVR amplitude. The size of this potential bias in BOLD-CVR needs to be investigated further to see how clinically relevant this bias is. An approach to mitigate potential metabolic contamination using a hypercapnic stimulus in BOLD-CVR could be to add a hypoxic component to the gas mixture, as suggested by Peng et al. ¹⁷ For ASL or ¹⁵O-H₂O PET-based CVR studies, we do not expect a bias in CVR since these techniques are dominated by CBF changes and not sensitive to changes in deoxyhemoglobin.

The CMRO₂ reported here under the room-air condition (144.5 ± 25.5 µmol/100 g/min) follows those previously reported based on TRUST-MRI, susceptibility-based Y_v measurements as well as the PET gold-standard.^46,54,61,62 When ignoring the venous plasma O₂ for the calculation of ΔCMRO_2, this results in a 0.7 percentage point increase for the difference in ΔCMRO₂ between room-air and ‘CO₂ in air’, the same increase was found for the difference between room-air and carbogen ΔCMRO₂. Based on these calculations, venous plasma dissolved O₂ played a negligible role in the arteriovenous oxygen difference, a key factor that determines CMRO₂ under the conditions assessed herein. Specifically, for the carbogen condition, arterial plasma dissolved O₂ was metabolized by the tissue (see Figure 3). Arterial plasma dissolved O₂ plays a negligible role in the transportation of oxygen under physiological conditions. This is reflected by the small change in ΔCMRO₂ when ignoring arterial and venous O₂ for the ‘CO₂ in air’ condition (−13.8% compared to −13.4% when including arterial and venous plasma dissolved O₂). However, for carbogen, the estimated ΔCMRO₂(%) drastically differed when ignoring the arterial and venous plasma dissolved O₂ (−20.6% instead of −2.0%). Given the very high arterial plasma O₂ values seen during carbogen, this difference was expected.

The setup we used for carbogen gas inhalation is a relatively simple way to evoke a vascular response and has seen use in numerous studies.^63–65 Furthermore, carbogen has been reported to be more comfortable and possibly safer for patients and participants compared to a normoxic CO₂ stimulus. ⁶⁶ Aside from its vasodilatory action, carbogen leads to a significant increase in arterial plasma dissolved O₂. Similar to the debate surrounding the effect of hypercapnia on CMRO₂, the question of whether changes in arterial O₂ content modulate CMRO₂ has also provided mixed conclusions. Studies looking at the potential neuroprotective effect of hyperoxia for acute treatment of traumatic brain injury have reported limited changes in CMRO₂ using normobaric hyperoxia. ⁶⁷ This finding is supported by a physiological study measuring CMRO₂ during prolonged apnea using trained free-divers performed by Ainslie et al. ⁶⁸ Interestingly, a significant post-treatment increase in CMRO₂ of 32% was observed in severe traumatic brain injury patients when applying hyperbaric hypercapnia. ⁶⁹ In contrast, the work of Xu et al., ¹³ reported an ∼16.9% reduction in CMRO₂ when breathing a normobaric fixed inspired O₂ of 98%. ¹³ In that study, a hypoxic stimulus was reported to increase CMRO₂ providing the notion that varying inspired O₂ content affects CMRO₂ in a dose-dependent manner. ¹³

In the context of our work, hyperoxia seems to affect the CMRO₂. If hyperoxia had a neutral effect on CMRO₂, the reduction in CMRO₂ seen during the ‘CO₂ in air’ condition would carry over to the carbogen condition, which was not observed. Similarly, a further decrease in CMRO₂ would have exaggerated the negative effect of hypercapnia on CMRO₂. Since neither of these responses was observed in our experiments, several hypotheses could explain our finding that CMRO₂ did not significantly change during the carbogen condition. The addition of the hyperoxic component seems to mitigate the hypercapnia-induced reduction in CMRO_2. Here, the addition of a hyperoxic component could have altered the brain arousal state and/or the neurophysiological reactions to the hypercapnic stimulus. The notion that an interplay between hypercapnic and hyperoxic states can modulate the brain’s response has been purported previously by Bain et al. ⁷⁰ It is important to note that while ‘CO₂ in air’ induces a physiological condition that can be achieved naturally (i.e. being out of breath after running up several stairs, or breath-holding after hyperventilation), the effect of carbogen inhalation on blood gas composition is impossible to reach under normal circumstances. It stands to reason that accurately predicting the effect of hypercapnic-hyperoxic stimuli on CMRO₂ is complex in a dynamic biological system as the human body, which has multiple mechanisms in place to actively maintain homeostasis.

Considerations

Our observation of no significant differences in CBF between the ‘CO₂ in air’ and carbogen conditions implies that the carbogen stimulus did not lead to significant O₂-mediated vasoconstriction. This is in line with previously reported MRI findings and also supports results reported using PET.^26,45,71 Hyperoxia is known to modulate arterial blood water T_1a due to hyperoxia; to account for this, we incorporated a global T_1,HO reduction in the CBF quantification as reported previously for the same carbogen stimulus, ³⁴ and investigated different T_1a scenarios (Supplementary Figure 3). It should be noted that intersubject variation will remain and will also depend on the subject's hematocrit. Any intersubject variation in CBF and arteriovenous O₂ difference will translate directly to variation in CMRO₂. From the sensitivity analysis on T₁, we observed that increasing or decreasing the hyperoxic blood water T₁ by 5%, CBF and CMRO₂ increased by 4.1% or decreased by 4.7%, respectively (Supplementary Figure 2). Notably, the ΔCBF and ΔCMRO₂ results did not significantly change for different T_1a scenarios using commonly used and recently reported T_1,NO and T_1,HO values (Supplementary Figure 3). It is our view that for this study, inter-subject variations were likely were ‘averaged out’, reducing the probability of systematic bias. It is important to emphasize that our results were derived from healthy subjects, using two commonly used stimuli. Based on the presented data, it would not be accurate to extrapolate results using different CO₂ with O₂ gas mixture concentrations, nor can we predict responses in patients with circulatory, pulmonary, or cerebrovascular pathologies. More research is needed to investigate whether changes in CMRO₂ vary with different combinations of CO₂ and O₂ concentrations when using carbogen-based designs. How the respective responses might change in pathological situations remains an open question.

With regards to the Y_v measurement, the TRUST technique may exhibit a bias from hyperoxia-mediated changes in venous plasma T₂. However, we observed that the venous plasma O₂ content was negligible for all conditions (Figure 3), and thus could not influence the Y_v measurement for the carbogen condition. Furthermore, the TRUST technique relies on a single measurement taken at the occipital part of the superior sagittal sinus. While the superior sagittal sinus drains most of the total cerebral blood, it also drains the periosteum, skull, meninges, and CSF. ⁴⁴ Given that most of the blood flowing through the sinus is from the brain and that earlier reports on TRUST Y_v measurements at different locations have shown no difference in Y_v, ⁷² it is safe to presume the superior sagittal sinus as a representative location for a global Y_v measurement. In addition, we were mainly interested in the relative change of Y_v in the ΔCMRO₂ comparison for the different conditions, for which the superior sagittal sinus is appropriate under the assumption that no redistribution of blood has occurred. There are other ways than pCASL to measure global CBF, such as phase-contrast MRI. These techniques have been previously compared and showed a close match.^55,73

For diagnostics, treatment, or follow-up in patients, possible regional differences in CMRO₂ might be of interest to clinicians. Our approach of measuring global Y_v excludes the possibility to examine regional differences in CMRO₂. Regional Y_v measurements can be obtained using QSM-based techniques ⁷⁴ or other T₂- based MRI acquisitions, like T₂-relaxation-under-phase-contrast MRI, ⁷⁵ Velocity Selective Excitation and Arterial Nulling ⁷⁶ or by using techniques based on the Asymmetric Spin Echo. ⁷⁷ Such approaches could further benefit from subject-specific measures of Hct or the incorporation of local Hct measures. ⁷⁸ We assumed Hct values (0.42 for males, and 0.40 for females) as both the T₁ and T₂ values of blood depend on Hct. However, in healthy subjects, the Hct generally only varies by around 10%; ⁵¹ it is unlikely that this would impact our Y_v results significantly. This could be different in disease though. An earlier reported sensitivity analysis on T₁ differences on TRUST Y_v measurements demonstrated a negligible effect of slight differences in Hct on Y_v. ⁶¹ For our study, the incorporation of subject-specific p_aCO₂ and (assumed) Hct in the model by Dash et al. ⁴² did not impact the results dramatically (as shown in Figure 1(b)). Future (clinical) studies using such a physiological model will allow extraction of subject-specific blood oxygen content where information such as p_aCO₂, Hct, pH, and temperature can in principle be incorporated, depending on the study design and disease of interest.

Conclusion

We found that a hypercapnic normoxic stimulus of 5% CO₂ is not necessarily iso-metabolic to room-air but leads to a decrease in CMRO₂ in healthy subjects. For a hypercapnic hyperoxic stimulus as carbogen, we demonstrate that it is more iso-metabolic to room-air. Although the oxygen saturation curve is dependent on p_aCO₂ (Bohr effect), amongst other parameters, correction for differences in blood p_aCO₂ did not significantly influence results. We believe these findings provide valuable insight into the global hemodynamic and metabolic effects of commonly used respiratory challenges. The reported findings can be useful for future experimental designs, BOLD signal modeling, and interpreting calibrated BOLD fMRI and CVR measurements using hypercapnic or hypercapnic-hyperoxic (carbogen) stimuli.