Hyperoxic cerebral vasoconstriction in healthy humans: global, lobe and regions of interest

Abstract

Hyperoxia is used for hypoxia/ischemia-related conditions, but is associated with poorer clinical outcomes, potentially related to reductions in cerebral blood flow (CBF). Global CBF hyperoxia responses are variable; limited evidence suggests heightened sensitivity in anterior brain regions, leaving important gaps in understanding spatial resolution. Using 3-Tesla magnetic resonance imaging (MRI), we tested the hypotheses that 1) isocapnic hyperoxia would reduce CBF at whole brain, grey matter (GM), and white matter (WM), and 2) This reduction would differ across all lobes. CBF was quantified using pseudo-continuous arterial spin labeling (pcASL) in 15 young adults (7 females) during normoxia followed by hyperoxia (~80% O₂). Data presented as mean ± SD. End-tidal CO₂ remained unchanged between conditions (p=0.127) and blood pressure increased by 5±6 mmHg (p=0.013). Hyperoxia did not significantly reduce CBF at whole brain, lobar, or any of 65 brain regions of interest (ROIs) (all p>0.05). In contrast, whole brain cerebrovascular conductance (CVC) decreased by 12±15%, (75±20 to 66±21 mL/100 g/min/100mmHg; p=0.011), with similar relative reductions in GM (13±18%, 90±24 to 78±27 mL/100 g/min/100mmHg; p=0.021) and WM (8±11%, 53±14 to 49±14 mL/100 g/min/100mmHg; p=0.013). CVC decreased across all lobes (9–12%; all p<0.05), with no inter-lobar differences (p=0.866). Exploratory analyses showed reduced CVC in all 65 ROIs (3–35%; all p<0.026), but no regional heterogeneity were found within a given lobe (all p>0.401). These new findings demonstrate widespread cerebral vasoconstriction without lobe- or region-specific effects during acute hyperoxia, providing a foundation for future studies involving greater hyperoxic exposure and clinically relevant populations.

NEW & NOTEWORTHY

This study demonstrates acute isocapnic hyperoxia induces widespread cerebral vasoconstriction, reflected by reduced cerebrovascular conductance. Contrary to hypothesis the vasoconstrictive responses appear spatially uniform across lobes and regions of interest (ROIs), although the ROI conclusions are preliminary. These findings refine regional understanding of hyperoxia-induced cerebrovascular regulation and underscore the importance of considering vascular conductance, rather than CBF alone, when evaluating the cerebral effects of supplemental oxygen.

Graphical Abstract

INTRODUCTION

Hyperoxia is commonplace in various medical care settings as a means of treating or preventing hypoxemia (1) in conditions including respiratory failure (2), during wound care and surgery (3,4), ischemic stroke (5), myocardial infarction (6,7), and traumatic brain injury (TBI) (8). In addition to clinical uses of hyperoxia, military and other extreme environmental occupations commonly experience hyperoxia as well. To combat hypobaric hypoxemia, both aviators and astronauts breathe hyperoxic gas when operating in high-altitude scenarios (9).

While hyperoxic breathing is useful as noted above, it may impart negative effects. Specifically, in hospital intensive care units, hyperoxia is associated with higher mortality (10–12). It may also augment brain damage and reduce favorable outcomes secondary to TBIs due to cerebral vasoconstriction, among other mechanisms (13–16). One aspect of hyperoxia is the formation of unstable and highly reactive molecules known as reactive oxygen species (ROS) (17,18), which is linked to central nervous system oxygen toxicity (19,20).

Even without the clinical challenges noted above, one of the ways hyperoxia may lead to poor brain health outcomes is through cerebral vasoconstriction and the consequent decrements in cerebral blood flow (CBF) and cerebral oxygen delivery (CDO₂), which may not overcome the increased arterial oxygen content (CaO₂) (18,21–23). Over 75 years ago, Kety and Schmidt (22), reported an acute 13% reduction in CBF in healthy, young men during inhalation of 85–100% oxygen. Since then, newer studies, using primarily ultrasound, but also magnetic resonance imaging (MRI), have assessed hyperoxic cerebral vasoconstriction, reporting conflicting results with CBF reductions ranging from 7–37% (24–29,21) or no effect (30–33).

Methodological differences in CBF measures and arterial pressure of carbon dioxide (PaCO₂) control might be the two important contributors to some of these discrepancies. For example, Transcranial Doppler (TCD) ultrasound, which measures blood velocity rather than volumetric flow has recognized limitations (34–36). In contrast, MRI offers a more comprehensive view of cerebral perfusion with excellent spatial resolution with the ability to distinguish between tissue types, including grey matter (GM) and white matter (WM), as well as, brain regions of interest (ROIs). Another key parameter in hyperoxia studies is PaCO₂ control, as hyperoxia can induce paradoxical hyperventilation and therefore hypocapnia (37). Some studies maintained isocapnia during hyperoxia (18,25,27,30) whereas others allowed poikilocapnia to reflect real-world conditions (21,23,24), emphasizing the difficulty of disentangling hyperoxia’s direct effects on CBF from those of hypocapnia, considering the effect of PaCO₂ on CBF regulation (38).

Although GM and WM differ in metabolic and CBF demands, few studies have examined the regional effects of hyperoxia. Existing pseudo-continuous arterial spin labeling (pcASL) MRI studies (21,24–29) report significant CBF reductions in both GM and WM but vary in magnitude and make no statistical comparisons, leaving the question of region-specific hyperoxic vasoconstriction unresolved. Such regional variability in CBF control might hold important implications for cognitive function if certain brain regions are more or less affected by hyperoxia. Given the region-specific activation patterns of the brain during cognitive tasks, as well as the regional manifestation of neurocognitive and cerebrovascular diseases, both linked to altered CBF regulation, investigating the regional effects of hyperoxia may provide critical insight into its impact on brain health and function.

In this context the cerebral circulation demonstrates region-specific responses to other blood gas manipulations. Specifically, CBF reductions to hypocapnia are greater in the posterior over anterior regions (39). Conversely, hypoxia-induced vasodilation appears greater in the anterior brain circulation (internal carotid artery) rather than the posterior (40–42). Region-specific responses in CBF to hyperoxia are limited. Mattos and colleagues (18) suggests hyperoxia reduces CBF in the anterior circulation, while the posterior is unaffected. In contrast, Kolbitsch and colleagues (27) reported CBF changes in 10 of 12 regions ranging from −15% to +2%; but no statistical comparisons were made between regions. In addition, both human and animal studies demonstrate that antioxidant defenses and oxidative stress markers are unevenly distributed across brain regions (43–45). This suggests that certain regions may be more vulnerable to oxygen-induced alterations in CBF. Taken together, the available evidence supports the concept that hyperoxia might exert regionally heterogeneous effects on CBF.

The purpose of this study was to use MRI to identify global, lobar, and region-specific CBF responses to acute isocapnic hyperoxia. We tested the hypotheses that 1) isocapnic hyperoxia would reduce CBF at whole brain, GM and WM, and 2) this reduction would differ across lobes. We also tested the exploratory hypothesis that these reductions would exhibit non-uniform patterns between ROIs within a given lobe.

METHODS

Subjects

Fifteen young, healthy adults (7 females) voluntarily participated in this study. All participants provided written informed consent prior to participation. Study procedures were approved by the University of Wisconsin–Madison Health Sciences Institutional Review Board and were conducted in accordance with the Declaration of Helsinki [ClinicalTrials.gov ID: NCT04265053].

Eligibility was determined at an initial screening visit, which included a medical history questionnaire, MRI safety screening, venipuncture, a pregnancy test (females only), and anthropometric measurements (height and weight). Brachial artery blood pressure was measured using an automated sphygmomanometer (Datex Ohmeda), and the lowest of three readings was used. Inclusion criteria were: age 18–30 years, systolic blood pressure (SBP) ≤ 125 mmHg, diastolic blood pressure (DBP) ≤ 80 mmHg, body mass index (BMI) ≤ 25 kg/m², blood glucose < 100 mg/dL, low-density lipoprotein (LDL) < 130 mg/dL, and triglycerides <150 mg/dL. Exclusion criteria included smoking; use of cardiovascular, hormonal, or metabolic medications; history of cardiovascular, neurological, psychiatric, autoimmune, or reproductive disorders; and MRI contraindications. Female participants were not pregnant or lactating, had regular menstrual cycles, no known reproductive disorders, and had not used hormonal contraceptives for at least 6 months.

MRI Study Visit

Pre-visit Procedures.

Participants fasted for ≥8 hours and abstained from vigorous exercise, alcohol, caffeine, and nonsteroidal anti-inflammatory drugs for ≥ 24 hours. Female participants were tested during the early follicular phase (days 1–5) of their menstrual cycle, and negative pregnancy status was reconfirmed on the study day.

Instrumentation.

MRI scans were conducted using a 3-Tesla system with a 48-channel head coil (Discovery MR750; GE Healthcare, Waukesha, WI). Participants were fitted with a respiratory gating belt, pulse oximeter, and blood pressure cuff (automated sphygmomanometer). The study team continuously monitored subject vital signs during the study visit (Medrad Veris MR Vital Signs Patient Monitor; Bayer Healthcare, Whippany, NJ). End-tidal CO₂ (ETCO₂) was used as a non-invasive surrogate for PaCO₂, given their strong correlation (46–48). ETCO₂ waveforms were visually inspected in real time using capnography to confirm stable breathing patterns and consistent ETCO₂ levels. During MRI scans, heart rate (HR), ETCO₂, and arterial oxygen saturation (SpO₂) were recorded at one-minute intervals, and values were averaged across the duration of each pcASL acquisition. Mean arterial pressure (MAP) and respiratory rate (RR) were measured during the final minute of each scan.

Experimental Design.

Participants were instructed to remain awake and still with their eyes closed during MRI acquisition. As this study was conducted as part of a larger experimental protocol, the hyperoxic condition was preceded by ~10 minutes of hypercapnia. All hemodynamic and respiratory variables had returned to baseline levels prior to the start of the hyperoxia ramp and were confirmed to be comparable at the start of the present study (HR, p = 0.458; MAP, p = 0.068; RR, p = 0.998; ETCO₂, p = 0.774; see Supplemental Table 1).

CBF was measured using a T1-weighted structural scan and two pcASL scans (each ~4.5 minutes) during normoxic and hyperoxic breathing. The hyperoxia challenge involved participants breathing oxygen through a Hans Rudolph mask with a two-way non-rebreathing valve connected to a 100-liter reservoir bag. To achieve isocapnic hyperoxia, a gas mixture containing 5% CO₂ balanced with nitrogen was titrated into the breathing circuit. In our pilot experiments conducted in a separate cohort of six participants, inspired O₂ (FiO₂) increased from ~21% during room air to ~80% during hyperoxia. For comparison, clinical practice typically involves 30–40% FiO₂ to maintain arterial saturations >92%, whereas our protocol produced about twice the usual clinical exposure and similar to the levels encountered in military aviation (e.g., fighter pilots). This elevated exposure was selected to magnify potential regional effects on CBF and to align with prior studies that have used 50–100% oxygen. As discussed above, ETCO₂ waveforms were continuously visually monitored in real time using capnography, and manual adjustments were made to maintain ETCO₂ at baseline (normoxic) levels (See supplemental Figure 1).

The hyperoxia pcASL scan began once ETCO₂ was stable at baseline levels and SpO₂ reached 100% for at least three minutes (~3–5 minutes after starting hyperoxia). This timing was selected to ensure isocapnia and stable oxygenation before CBF measurement. The 4.5-minute scan duration reflects the imaging time available within a larger ongoing protocol, in which approximately 15 minutes were allocated for the entire hyperoxia phase.

MRI Parameters

T1-Weighted Imaging.

A T1-weighted structural scan was acquired based on the ADNI 3 protocol (49) using an accelerated sagittal 3D inversion recovery fast spoiled gradient echo (IR-FSPGR) sequence was used (flip angle = 11°, repetition time = 7.2 ms, echo time = 2.9 ms, inversion time = 400 ms, field of view = 27 cm, slice thickness = 1 mm, matrix size= 256×256, number of excitations = 1). Scans were reviewed for motion artifacts that could compromise CBF quantification.

pcASL Imaging.

Perfusion imaging was performed using a 3D fast spin-echo spiral pcASL sequence with Hadamard time encoding, enabling acquisition at multiple post-labeling delays (PLDs: 1.0, 1.8, and 2.7 s). Imaging parameters included a 5-ms readout spiral with eight interleaves (refocusing flip angle = 111°, repetition time = 6031 ms, echo time = 64.8 ms, field of view = 240×240×176 mm³, 4mm isotropic spatial resolution, labeling radiofrequency amplitude = 0.24 mG). A fluid-suppressed proton density (PD) image was acquired in the same slab without labeling for proton density referencing.

Data analysis and calculations

T1-weighted and pcASL images were processed in MATLAB (MathWorks, Natick, MA) using the Computational Anatomy Toolbox (CAT12) for Statistical Parametric Mapping version 12 (SPM12). CAT12 utilizes data from the MICCAI 2012 Grand Challenge on Multi-Atlas Labeling (B. Landman and S. Warfield, MICCAI 2012 workshop on multiatlas labeling, in MICCAI Grand Challenge and Workshop on Multi-Atlas Labeling, CreateSpace Independent Publishing Platform, Nice, France, 2012), released under the creative commons attribution-noncommercial license (CCBYNC) with no end date. Structural templates were derived from the OASIS dataset (https://www.oasis-brains.org/), and regional labels were provided by Neuromorphometrics, Inc. under academic subscription (50). pcASL images were converted from DICOM to NIfTI format and corrected for motion and transit time. Arterial transit time (ATT) was estimated using a multi-PLD kinetic model based on the Buxton framework, fitting the ASL signal across PLDs to jointly estimate CBF and arterial arrival time. Transit-time–corrected CBF estimates were used for all analyses. CBF quantification assumed a constant labeling efficiency (α = 0.9). Although flow velocity at the labeling plane was not directly measured, labeling efficiency was assumed to remain stable across conditions within physiological flow ranges. Normoxic data were processed assuming the standard arterial blood T1 (T1a:1600 ms). To account for the increase in dissolved oxygen, the assumed T1a in the CBF model during hyperoxia was reduced to 1400 ms consistent with reported measures at 3T (26). T1-weighted images were resampled, denoised, bias-corrected, and segmented into GM and WM using CAT12 (threshold = 0.8). GM masks were coregistered to ASL space using affine transformation and smoothed with a 7-mm full-width half-maximum Gaussian kernel. Global GM and WM CBF were quantified in native space, and regional CBF was quantified using the Neuromorphometrics atlas in MNI152NLin2009cAsym space. Scan quality was assessed using CAT12 quality metrics and independent visual inspection. Data were excluded if excessive motion, improper spatial normalization, or image inhomogeneity artifacts were present.

Cerebrovascular conductance (CVC), which accounts for variations in blood pressure by normalizing CBF, was calculated using the formula: CVC (mL/100g/min/100 mmHg) = (CBF / MAP) × 100. A positive change in CVC indicates vasodilation, whereas a negative change indicates vasoconstriction.

Arterial oxygen (PaO₂) was not measured directly but estimated using the alveolar gas equation: PAO₂ = FiO₂ × (P_ATM – P_H2O) – PCO₂ / RQ, where PAO₂ is the alveolar partial pressure of oxygen, P_ATM is the barometric pressure (742 mmHg in Madison, WI, elevation 267 m), P_H2O is the partial pressure of airway water vapor (47 mmHg), individual ETCO₂ was used as a surrogate for PaCO₂, and the respiratory quotient (RQ) was set to 0.8. Estimated PaO₂ was obtained by subtracting 5 mmHg from PAO₂ to account for the alveolar–arterial gradient (29).

CaO₂ was computed using CaO₂ (mL O₂/dL) = (1.36 mL O₂/g Hb × [Hb] (g/dL) × SpO₂ ) + (PO₂ (mmHg) × 0.003 mL O₂/dL/mmHg), where [Hb] is hemoglobin concentration and SpO₂ is arterial oxygen saturation. CDO₂ was calculated as: CDO₂ (mL O₂/min) = BF (mL/min) × CaO₂ (mL O₂/dL) × (1 dL / 100 mL), where blood flow (BF, in mL/min) = tissue volume (cm³) × tissue density (g/cm³) × CBF (in mL/100 g/min). Tissue density was assumed to be 1.05 g/cm³ for GM and 1.04 g/cm³ for WM.

Statistical Analysis

Data were analyzed using GraphPad Prism (version 10.6.1; GraphPad Software, San Diego, CA). Statistical significance was defined as p < 0.05. Data normality was assessed using the Shapiro–Wilk test, and homogeneity of variances was evaluated with the Brown–Forsythe test. CBF and CVC were compared between normoxia and isocapnic hyperoxia using paired two-tailed T tests for normally distributed data and Wilcoxon signed-rank tests for non-normally distributed data. Because MAP increased during hyperoxia, additional analyses were performed with blood pressure included as a covariate to account for its influence on CBF. Specifically, mixed-effects models were used to assess condition-related differences in CBF while adjusting for MAP, with condition entered as a fixed effect and subject as a random effect. To compare the magnitude of changes across lobes, a one-way repeated-measures ANOVA was performed on relative (%) changes in CBF and CVC across the six lobe-level regions (occipital, parietal, frontal, temporal, subcortical regions, and brainstem/cerebellum), with lobe as the within-subject factor. All regional CBF and CVC values represent the average of the left and right hemispheres; no significant hemispheric differences were observed (p > 0.05). 65 brain ROIs were defined and anatomically grouped into six lobe-level summary measures by averaging across constituent ROIs. To control the familywise error rate for multiple comparisons, the Shaffer sequentially rejective procedure was applied (51,52) as previously described (53). This approach was selected because it controls the familywise error rate while preserving statistical power for interdependent comparisons, offering greater sensitivity than the more conservative and commonly used Bonferroni correction. Data are presented as mean ± standard deviation (ranges are provided for relative changes in CBF and CVC).

RESULTS

Subject Characteristics

Subject characteristics and whole brain volume are summarized in Table 1. As intended by the study design, all participants had a healthy BMI, normal blood pressure, and healthy glucose and lipid levels, consistent with the absence of cardiovascular and cerebrovascular disease.

Table 1.

Subject characteristics

	n = 15 (7F)
Age, yr	22 ± 4
BMI, kg/m2	22 ± 2
SBP, mmHg	110 ± 9
DBP, mmHg	69 ± 6
MAP, mmHg	83 ± 6
Glucose, mg/dL	74 ± 9
Cholesterol, mg/dL
Total	140 ± 28
HDL	57 ± 12
LDL	64 ± 23
Triglycerides, mg/dL	72 ± 15
Hemoglobin, g/dL	13 ± 2
Grey matter volume, cm3	745 ± 61
White matter volume, cm3	512 ± 49

Values are means ± SD. BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; MAP, mean arterial blood pressure; HDL, high-density lipoprotein; LDL, low-density lipoprotein.

Respiratory and Hemodynamic values

Respiratory and hemodynamic variables relevant to this study are summarized in Table 2. Average HR (p=0.602) and ETCO₂ (p=0.127) during the pcASL scan were not different between conditions. However, average MAP was significantly higher during hyperoxia (p=0.013).

Table 2.

Hemodynamic and Respiratory Variables

	Normoxia	Hyperoxia	p-value
HR, beats/min	59 ± 10	58 ±12	0.602
MAP, mmHg	73 ±6	78 ± 8	0.013
ETCO2, mmHg	35 ± 3	35 ± 3	0.127

Values are means ± SD. Comparisons between baseline and pre-ramp were performed using paired t-tests. HR, heart rate ; MAP, mean arterial pressure ; ETCO₂, end-tidal carbon dioxide .

Global CBF and CVC response

Representative cerebral perfusion maps from a subject whose values closely approximated the group mean are shown in Figure 1. Isocapnic hyperoxia did not result in statistically significant reductions in whole brain, GM, or WM CBF. Specifically, whole brain CBF decreased by 5 ± 18% (range: −51% to +16%), from 54 ± 13 to 51 ± 17 mL/100 g/min during normoxia and hyperoxia, respectively, but this change did not reach statistical significance (p=0.153). Similarly, GM CBF declined by 7 ± 22% (−67% to +17%), from 65 ± 16 to 61 ± 22 mL/100 g/min (p=0.143), and WM CBF declined by 3 ± 13% (−26% to +14%), from 39 ± 9 to 38 ± 11 mL/100 g/min (p=0.246; Figure 2A–D).

Figure 1.

Whole-brain magnetic resonance imaging (MRI) perfusion images from one representative subject, whose data closely reflect the group mean, showing coregistered pseudo-continuous arterial spin labeling (pcASL) during normoxia (A) and hyperoxia (B). Yellow-to-red coloring represents greater blood flow to a region, while green-to-blue indicates lower blood flow. Note that surface-based visualizations are limited to cortical regions and do not display infratentorial structures (e.g., brainstem, cerebellum), although these regions were included in quantitative analyses.

Figure 2. Effects of hyperoxia on cerebral blood flow (CBF) and cerebrovascular conductance (CVC).

Average and individual data (bars and dots). Absolute whole brain CBF during normoxia and hyperoxia (A), grey matter (B), and white matter (C). Percent change (Δ%) in CBF from normoxia (D). Absolute whole brain CVC during normoxia and hyperoxia (E), grey matter (F), and white matter (G). Percent change (Δ%) in CVC from normoxia (H). Bars represent mean ± SD. *P < 0.05 vs. normoxia using paired t test.

Interestingly, when accounting for changes in blood pressure on cerebral perfusion, hyperoxia led to a significant reduction in CVC. Whole brain CVC decreased by 12 ± 15% (range: −43% to +10%), from 75 ± 20 to 66 ± 21 mL/100 g/min/100 mmHg (p=0.011). GM CVC decreased by 13 ± 18% (−62% to +11%), from 90 ± 24 to 78 ± 27 mL/100 g/min/100 mmHg (p=0.021), while WM CVC declined by 8 ± 11% (−29% to +8%), from 53 ± 14 to 49 ± 14 mL/100 g/min/100 mmHg (p=0.013; Figure 2E–H). These findings indicate substantial hyperoxic vasoconstriction despite largely preserved CBF.

CVC calculations assume a linear relationship between CBF and MAP, an assumption that may not hold when MAP varies substantially. To avoid relying on this assumption, we performed an additional analysis that directly modeled CBF while statistically accounting for MAP using a linear mixed-effects model with subject included as a random effect. There was no significant interaction between condition and MAP for whole brain CBF (p = 0.452), GM CBF (p = 0.582), or WM CBF (p = 0.268), indicating that the relationship between MAP and CBF was similar under normoxic and hyperoxic conditions. After adjusting for MAP, whole brain CBF was significantly lower during hyperoxia compared with normoxia (ΔCBF = 7.1 mL/100 g/min, p = 0.009). A similar pattern was observed in GM, where CBF was significantly lower during hyperoxia after accounting for MAP (ΔGM CBF = 10.1 mL/100 g/min, p = 0.009). In WM, CBF was also lower during hyperoxia after adjustment for MAP, although this effect did not reach statistical significance (ΔWM CBF = 2.5 mL/100 g/min, p = 0.080). Across models, MAP was positively associated with CBF in whole brain CBF and GM (both p ≤ 0.011) and showed a trend-level association in WM (p = 0.080). At the lobar level, linear mixed-effects models showed a consistent positive association between MAP and CBF across all six lobes (all p < 0.046). There was no significant interaction between condition and MAP in any lobe (all p > 0.195). After adjusting for MAP, hyperoxia was associated with lower CBF across lobes (all p < 0.021; see Supplemental Table 2).

Lobar CBF and CVC response

Lobar analyses demonstrated consistent reductions in CBF across all six lobes; however, none of these changes reached statistical significance. CBF decreased by 1 ± 2% (range across ROIs: −4% to +1%; p=0.203) in the occipital lobe, 1 ± 4% (−8% to +10%; p=0.220) in the parietal lobe, 4 ± 15% (−38% to +23%; p=0.262) in the frontal lobe, 6 ± 13% (−35% to +16%; p=0.131) in the temporal lobe, 3 ± 14% (−28% to +26%; p=0.474) in subcortical regions, and 5 ± 16% (−23% to +19%; p=0.348) in the brainstem/cerebellum. Repeated-measures ANOVA confirmed that the magnitude of CBF reduction did not differ between lobes (p=0.935; Figure 3A–B).

Figure 3. Regional effects of hyperoxia on cerebral blood flow (CBF) and cerebrovascular conductance (CVC).

Average and individual data (bars and dots). Absolute CBF during normoxia and hyperoxia across lobes (occipital, parietal, frontal, temporal, subcortical regions, and brainstem/cerebellum) (A). Percent change (Δ%) in CBF from normoxia across lobes (occipital, parietal, frontal, temporal, subcortical regions, and brainstem/cerebellum) (B). Absolute CVC during normoxia and hyperoxia across lobes (C). Percent change (Δ%) in CVC from normoxia across lobes (D). Bars represent mean ± SD. *P < 0.05 vs. normoxia using paired t tests and the Shaffer sequentially rejective method to control the familywise error rate for multiple comparisons. n=15.

In parallel with global findings, lobar CVC was significantly reduced during hyperoxia across all six lobes. CVC decreased by 9 ± 10% (−29% to 4%; p=0.006) in the occipital lobe, 12 ± 7% (−26% to 6%; p<0.001) in the parietal lobe, 10 ± 10% (−26% to 11%; p=0.001) in the frontal lobe, 12 ± 9% (−24% to 3%; p<0.001) in the temporal lobe, 9 ± 10% (−24% to +14%; p=0.004) in subcortical regions, and 10 ± 12% (−34% to 8%; p=0.007) in the brainstem/cerebellum. Repeated-measures ANOVA confirmed that the magnitude of CVC reduction did not differ between lobes (p=0.866; Figure 3C–D).

Regional (ROI level analysis) CBF and CVC response

Exploratory ROI-level analyses demonstrated spatially uniform responses to hyperoxia. ROI comparisons were performed separately within each of the six predefined lobes (ranging from 6 to 18 ROIs per lobe) to minimize irrelevant comparisons (e.g., between anatomically or vascularly unrelated regions such as the hippocampus and the cuneus) and to focus on physiologically meaningful relationships. Using Shaffer’s sequentially rejective method to correct for multiple comparisons, no individual ROI exhibited a statistically significant reduction in CBF (all p > 0.05; Table 3), and the magnitude of CBF change did not differ among ROIs within any lobe (all p > 0.607). In contrast, CVC was significantly reduced across all examined ROIs during hyperoxia (range 3-35%, all p < 0.026; Table 3) without any significant differences between ROIs within a given lobe (all p > 0.401), indicating widespread cerebrovascular vasoconstriction at the ROI level.

Table 3.

Regional Cerebral Blood Flow and Cerebrovascular Conductance During Normoxia and Hyperoxia

Cerebral Blood Flow (mL/100 g/min)		Brain region of interests	Cerebrovascular Conductance (mL/100 g/min/100 mmHg)
Normoxia	Hyperoxia		Normoxia	Hyperoxia
Brainstem/Cerebellum
37 ± 9	35 ± 9	Brain Stem	51 ± 12	45 ± 12 *
55 ± 16	54 ± 19	Cerebellum Exterior	76 ± 23	70 ± 23 *
38 ± 10	36 ± 13	Cerebellum White Matter	52 ± 15	46 ± 15 *
37 ± 15	37 ± 17	Cerebellar Vermal Lobules I–V	51 ± 20	48 ± 21 *
42 ± 17	40 ± 21	Cerebellar Vermal Lobules VI–VII	57 ± 24	51 ± 26 *
50 ± 16	48 ± 18	Cerebellar Vermal Lobules VIII–X	68 ± 22	62 ± 23 *
Subcortical regions
68 ± 11	68 ± 17	Accumbens Area	94 ± 19	88 ± 22 *
50 ± 10	48 ± 13	Amygdala	69 ± 18	62 ± 16 *
54 ± 9	53 ± 13	Caudate	75 ± 16	69 ± 16 *
40 ± 7	41 ± 9	Pallidum	56 ± 11	53 ± 12 *
59 ± 10	60 ± 14	Putamen	82 ± 17	77 ± 19 *
48 ± 13	46 ± 16	Thalamus Proper	67 ± 20	59 ± 20 *
41 ± 10	38 ± 12	Ventral DC	56 ± 16	49 ± 15 *
41 ± 11	39 ± 11	Optic Chiasm	57 ± 18	51 ± 16 *
81 ± 16	79 ± 16	Anterior cingulate gyrus	111 ± 26	102 ± 22 *
75 ± 17	72 ± 18	Middle cingulate gyrus	103 ± 26	93 ± 23 *
68 ± 20	65 ± 23	Posterior cingulate gyrus	93 ± 29	84 ± 28 *
63 ± 11	61 ± 14	Subcallosal area	87 ± 20	79 ± 21 *
Frontal lobe
60 ± 11	57 ± 12	Basal Forebrain	82 ± 19	73 ± 16 *
75 ± 17	70 ± 19	Anterior orbital gyrus	103 ± 26	90 ± 24 *
81 ± 16	78 ± 16	Frontal operculum	112 ± 26	100 ± 21 *
80 ± 20	74 ± 23	Frontal pole	110 ± 30	95 ± 28 *
80 ± 16	78 ± 21	Gyrus rectus	110 ± 25	101 ± 28 *
77 ± 17	75 ± 19	Lateral orbital gyrus	107 ± 24	96 ± 22 *
83 ± 17	72 ± 18	Medial frontal cortex	115 ± 28	93 ± 23 *
74 ± 19	81 ± 21	Middle frontal gyrus	102 ± 28	105 ± 26 *
74 ± 13	70 ± 22	Medial orbital gyrus	102 ± 22	89 ± 26 *
62 ± 17	73 ± 18	Precentral gyrus (medial segment)	85 ± 23	94 ± 23 *
83 ± 18	58 ± 18	Superior frontal gyrus (medial segment)	114 ± 27	74 ± 22 *
80 ± 18	78 ± 19	Opercular part of the inferior frontal gyrus	110 ± 26	100 ± 23 *
80 ± 17	77 ± 15	Orbital part of the inferior frontal gyrus	110 ± 26	99 ± 19 *
79 ± 15	79 ± 17	Posterior orbital gyrus	109 ± 23	102 ± 23 *
61 ± 16	59 ± 19	Precentral gyrus	84 ± 23	77 ± 24 *
68 ± 18	63 ± 20	Superior frontal gyrus	93 ± 25	82 ± 24 *
77 ± 18	73 ± 18	Supplementary motor cortex	105 ± 26	94 ± 22 *
81 ± 19	76 ± 19	Triangular part of the inferior frontal gyrus	111 ± 27	97 ± 23 *
Temporal lobe
56 ± 13	54 ± 16	Hippocampus	78 ± 20	69 ± 20 *
74 ± 13	73 ± 15	Anterior insula	102 ± 20	95 ± 20 *
51 ± 10	48 ± 12	Entorhinal area	71 ± 16	62 ± 15 *
53 ± 15	52 ± 18	Fusiform gyrus	73 ± 21	68 ± 23 *
59 ± 17	55 ± 19	Inferior temporal gyrus	81 ± 25	71 ± 24 *
72 ± 19	68 ± 21	Middle temporal gyrus	100 ± 29	88 ± 26 *
50 ± 10	48 ± 13	Parahippocampal gyrus	68 ± 15	61 ± 15 *
71 ± 12	70 ± 17	Posterior insula	97 ± 18	90 ± 23 *
74 ± 14	67 ± 16	Planum polare	102 ± 20	86 ± 19 *
81 ± 15	73 ± 18	Planum temporale	111 ± 24	94 ± 23 *
74 ± 17	68 ± 18	Superior temporal gyrus	103 ± 26	87 ± 23 *
58 ± 14	55 ± 15	Temporal pole	80 ± 21	71 ± 19 *
84 ± 15	77 ± 18	Transverse temporal gyrus	116 ± 23	99 ± 23 *
Parietal lobe
74 ± 13	72 ± 16	Central operculum	102 ± 21	92 ± 21 *
71 ± 20	69 ± 19	Angular gyrus	98 ± 30	89 ± 26 *
54 ± 18	51 ± 20	Postcentral gyrus (medial)	74 ± 25	66 ± 25 *
66 ± 20	62 ± 25	Precuneus	91 ± 29	80 ± 30 *
76 ± 15	72 ± 18	Parietal operculum	105 ± 23	93 ± 23 *
53 ± 15	53 ± 18	Postcentral gyrus	73 ± 22	69 ± 24 *
71 ± 17	70 ± 20	Supramarginal gyrus	98 ± 25	90 ± 25 *
50 ± 17	49 ± 19	Superior parietal lobule	68 ± 24	64 ± 25 *
Occipital lobe
59 ± 17	60 ± 23	Calcarine cortex	81 ± 24	77 ± 28 *
59 ± 20	57 ± 24	Cuneus	82 ± 28	74 ± 30 *
59 ± 20	58 ± 26	Inferior occipital gyrus	81 ± 30	75 ± 31 *
57 ± 18	56 ± 22	Lingual gyrus	78 ± 25	72 ± 27 *
64 ± 19	62 ± 23	Middle occipital gyrus	88 ± 29	80 ± 29 *
57 ± 23	56 ± 27	Occipital pole	78 ± 33	73 ± 33 *
53 ± 18	53 ± 22	Occipital fusiform gyrus	73 ± 25	68 ± 27 *
51 ± 18	50 ± 22	Superior occipital gyrus	70 ± 27	65 ± 28 *

Values are means ± SD. Cerebral blood flow and cerebrovascular conductance

^*P < 0.05 vs. normoxia using paired t tests and the Shaffer sequentially rejective method to control the familywise error rate for multiple comparisons. n=15.

Importantly, a sensitivity analysis excluding the two participants with the largest reductions in CBF and CVC indicated that the primary findings, including the absence of regional heterogeneity, were not materially altered.

Cerebral oxygen delivery

Despite the absence of statistically significant changes in CBF, indices of oxygen transport exhibited modest but directionally consistent increases during hyperoxia. CaO₂ increased significantly from 18 ± 3 to 19 ± 3 mL O₂/100 mL blood (p < 0.001). In contrast, CDO₂ demonstrated only modest, nonsignificant changes globally, such that; whole brain CDO₂ increased by 3 ± 20% (range: −47% to +27%), from 122 ± 27 to 124 ± 31 mL O₂/min (p=0.787; for GM CDO₂ and WM CDO₂ see supplemental table 3).

DISCUSSION

The primary purpose of this study was to quantify the effects of isocapnic hyperoxia on CBF using high-resolution MRI, with a specific focus on lobar and regional responses and exploring ROI patterns, a spatial resolution question that remains largely unanswered. Contrary to our hypotheses, we observed no significant reductions in CBF during breathing ~ 80% FiO₂, but CVC data clearly indicated: 1) cerebral vasoconstriction to hyperoxia occurring at the global, lobe, and ROI levels and 2) relative reductions didn’t differ among six brain lobes, and 3) exploratory analyses suggests a lack of heterogeneous ROI responses within specific lobes. Together, these findings confirm that under isocapnic hyperoxia the human brain actively vasoconstricts and provide the first all-inclusive analysis of lobe and regional control of CBF during hyperoxia which appears largely uniform across lobes. This lobe and ROI level vasoconstriction warrants future studies to dive deeply and more definitively into ROI hyperoxia responses and sets the stage for studies with longer hyperoxia exposures, including in clinically important populations.

Global Response to Hyperoxia

The present CBF findings are consistent with five previous studies reporting no significant change in CBF using MRI (30,32,33) or Doppler ultrasound (27,31). This finding does not support our hypotheses, which was based on the remaining 12 hyperoxia human cerebral literature demonstrating some level of CBF reduction (18,21–26,29,54,55) along with physiologic rationale that hyperoxia can induce ROS-mediated vasoconstriction (18,56,57), and the pro-antioxidant brain signaling is heterogeneously expressed in the brain (43–45). Present data lands at the intersection of these 17 studies, as CBF does not change (Figure 2A–D), but CVC declines ~ 12% globally (Figure 2E–H) and ~ 9 −12% across six lobes (Figure 3C–D). What is similar and different between studies reporting no change in CBF versus up to 37% reductions?

Participant characteristics is the first key point to focus on. While the vast majority of studies use largely young healthy subjects, some studies lack information for specifically targeting “disease-free’ subjects with healthy blood and cardiovascular parameters (Table 1) and do not consistently report ruling out comorbidities and pathologies. Subject age and obesity could be influential, as Domato and colleagues (21) found the greatest reported vasoconstriction (37%) in a cohort spanning wide ranges in both age (19–59 years) and BMI (20–39 kg/m²). Both age and body composition are well known to influence vascular function (58,59). Similarly, middle-aged males (32–42 years) exhibited a 27% reduction in CBF, compared to a 16% reduction in a younger subjects (23–26 years) (24). Also, Floyd and colleagues (25) studying a broad age range (21–62 years) reported a 33% CBF reduction, although this study did not specifically isolate the effect of age. In sum, participant characteristics including age and body composition varied across studies and may represent a possible explanation for the discrepancies between those findings and ours obtained in healthy young subjects. However, we acknowledge neither the present study nor prior published work have systematically tested the impact of all these factors on reactivity individually or collectively. It is noteworthy that future patients most in need of supplemental oxygen (e.g. older obese) may experience robust hyperoxic vasoconstriction with potential reductions in O₂ delivery. When focusing on younger healthy adults, present CVC data indicates hyperoxic vasoconstriction at a level comparable to previous studies (18,23,24,26–28).

A second critically important consideration when using pcASL is the influence of T1a on CBF quantification, especially under non-normoxic gas conditions. In this study, we measured CBF with whole-brain pcASL, consistent with previous work (21,25,26,32,33). Under normoxia at 3T, T1a is typically assumed to be ~1.65 s (26,60,61), but several studies have shown that T1a decreases under hyperoxia to ~1.38–1.49 s (26,60,61). If this reduction is not accounted for, ASL-based CBF measures are underestimated which overestimates the effect of hyperoxia on CBF. Indeed, analyzing present ASL scans without T1a correction indicated CBF decreased 20 ± 10% and whole brain CVC reduced 25 ± 11% (p both <0.05) (supplemental Figure 2 and 3). underscoring the importance of adjusting arterial blood T1 under hyperoxic conditions (26,60,61). In fact, MRI studies that use the lower T1a report no change (32,33) or relatively small change in CBF (5%; (26)) similar to present findings (Fig 2A–D). In contrast, studies using ASL MRI that did not report using a corrected T1a suggested the greatest levels of hyperoxic vasoconstriction (21,25,29). If we apply a similar ~50% reduction in hyperoxic constriction (i.e. 25% to ~12% in CVC, Supplemental Figure 2), their findings may be reduced from 37% to 18.5% (21), from 30% to 15% (25), and from 26% to 13% (29). If this is the case, then our data and the collective 12 prior human studies using MRI indicate that the overall global reduction in CBF (or CVC) is on the order of zero to 18% (with most in the 5–15%), substantially lower than previous reports of >30% (21,25,29).

While the above arguments seem most relevant to the present hypotheses, there are other more obvious design discrepancies range from CBF methodology, gas delivery, and level of control of ETCO₂ (i.e. arterial CO₂) that may contribute to varied hyperoxic responses. First, studies vary in methodology for CBF measures, including nitrous oxide (22), TCD (55), Doppler ultrasound of extracranial (i.e., Carotid) arteries (18,23,31,56), or MRI (21,24–26,28–30,32,33) and one even including both MRI and TCD (27). Collectively each CBF method has inherent assumptions and limitations, but all are validated measures and therefore unlikely to explain the variability in hyperoxic responses. A second consideration is the hyperoxia dose, as studies have varied from 30–100% O₂ and durations ranging from 5–30 minutes. Most studies last ~10 minutes with ~80% O₂ (22,26,29), as in the present design. The range of hyperoxia levels does not appear to explain variability in results, but some data suggest a constriction threshold around 50–60% O₂ (26,29). A third relevant consideration is ETCO₂ control, given its association with arterial CO₂ (47). Arterial CO₂ is a well-established modulator of CBF (37,38). In particular, hyperventilation-induced hypocapnia during hyperoxia is a known vasoconstrictor that can exaggerate CBF reductions (23,25,56). Specifically, greater reductions in CBF were observed during poikilocapnic hyperoxia (~26%) compared to isocapnic hyperoxia (~16%), supporting a key role for CO₂ in modulating this response (23). A wide range of methods have been used to control ETCO₂ including premixed gases with added CO₂ to clamp PCO₂ (25), verbal coaching (27), rebreathing circuits (18), or mathematical correction (30), whereas several studies did not control (21,24), or directly tested poikilocapnic responses (23,56). Taken together, differences in CO₂ control likely contribute meaningfully to variability in reported CBF responses across studies. In the present study, ETCO₂ was continuously monitored and maintained at baseline levels using titrated CO₂ to achieve isocapnic hyperoxia, thereby minimizing the confounding influence of CO₂ on CBF and allowing the effects of hyperoxia to be evaluated more directly. Perhaps data variability is the product of physiological variance, or it is small contributions of all three of these that collectively contribute. All of these deserve close attention to control and/or measure.

Cerebrovascular Conductance

The significant decrease in CVC indicates that cerebral vasoconstriction was indeed occurring at global, lobar, and ROI levels. When CBF data were analyzed with blood pressure as a covariate, a significant decrease to hyperoxia was revealed (see results and supplemental Table 2), suggesting that the rise in MAP masked the hyperoxia-induced vasoconstriction. Taken together, these two analyses increase confidence in concluding that hyperoxia leads to cerebral vasoconstriction, a finding consistent with the majority of hyperoxia CBF literature. Some studies did not report blood pressure responses to hyperoxia (21,24,25,54), whereas others reported no change in MAP (18,23,27,56), and one reported a decrease in MAP (22), complicating interpretation of hyperoxia’s vasoconstrictive effects, and emphasize the importance of measuring the blood pressure response to hyperoxia in order to fully interrogate hyperoxic cerebrovascular control.

Lobar and Regional Responses to hyperoxia

CBF findings at the lobe (Figure 3A–B) and ROI (Table 3) levels did not support our second hypothesis and Our hypothesis formed in the context of literature on ROS and regional antioxidant capacity, which indicate that antioxidant defenses (e.g., glutathione) and pro-oxidant markers are not uniformly distributed across the brain, with regions such as the hippocampus, parietal, and frontal cortex (43–45), as well as by a single Doppler ultrasound CBF study reporting greater vasoconstriction in the internal carotid versus vertebral arteries (18). CVC data in all six lobes indicate a broadly uniform level of vasoconstriction such that no significant differences in the magnitude of these reductions were observed between lobes (Figure 3C–D). Present findings fit with prior MRI work reporting CBF reductions in majority of brain with the exception of the frontoparietal area (27).

This study presents the first complete analysis of hyperoxia across 65 bilateral ROIs. All ROIs demonstrated significant reductions in CVC to hyperoxia (Table 3). These findings as preliminary and exploratory, as the relative changes in CVC ranged from 3-35%, suggesting we are underpowered to detect changes in so many brain regions, despite using Shaffer method to increase power (Table 3). Future studies with larger sample sizes are warranted to more definitively assess the spatial specificity of hyperoxia-induced CBF changes.

Limitations and experimental considerations

Several limitations should be acknowledged. First, our sample size, while demographically balanced, was modest and may limit the ability to detect sex differences or subtle regional heterogeneity in CBF responses. Second, participants were exposed to approximately 9-10 minutes of hyperoxia consistent with much of the current hyperoxia literature (18,27); longer exposures may reveal different spatiotemporal dynamics of perfusion. Third, PaO₂ was not directly measured but estimated from inspired oxygen fraction, which may not capture individual variability in oxygenation. Similarly, PaCO₂ was inferred from ETCO₂, a reliable but indirect surrogate (46–48). Fourth, we did not measure oxidative stress, or antioxidant capacity, which could influence cerebrovascular responses and interindividual variability. Fifth, CMRO₂ was not directly measured, which limits our ability to distinguish whether the observed reductions in CVC during hyperoxia reflect metabolic downregulation or primarily vascular effects. Prior work ranges from no effect of hyperoxia (31), to increased CMRO₂ (30) and even decreased CMRO₂ (23), suggesting this outcome is sensitive to study design factors, like CO₂ regulation, and are in need of further delineating the integration of vascular responses and oxygen delivery in supporting CMRO₂. Sixth, prior work suggest cerebral CO₂ may be ~ 2 mmHg higher than arterial CO₂ (18), all CBF hyperoxia studies using ETCO₂ may underestimate the true extent of hyperoxia-induced cerebral vasoconstriction. In addition , pcASL MRI provides model-based quantitative estimates of cerebral perfusion rather than direct flow measurements, however it relies on well-established physical principles that provide robust and physiologically meaningful estimates of cerebral perfusion. While it provides high spatial resolution, it has limited sensitivity in deep brain regions and lower temporal resolution potentially missing transient or localized perfusion changes. CBF quantification assumes constant labeling efficiency (α = 0.9), and flow velocity at the labeling plane was not directly measured. Although labeling efficiency is generally stable within physiological flow ranges, condition-related or interindividual variability cannot be completely excluded. Furthermore, our pcASL acquisition used only three PLDs (1.0, 1.8, and 2.7 s), which precluded robust estimation of ATT. As a result, we were unable to assess vascular transit dynamics, which may provide complementary mechanistic information beyond conventional CBF and CVC measures. Consequently, potential regional heterogeneity in ATT responses to hyperoxia could not be evaluated. Finally, we did not assess large-vessel diameter responses using pcVIPR imaging of the Circle of Willis. As a result, the contribution of hyperoxia-induced changes in conduit artery caliber to the observed reductions in CBF and CVC could not be determined, limiting mechanistic insight into the relative roles of large- and small-vessel contributions.

Implications and Future Directions

Previous work (18) demonstrated that hyperoxia-induced reductions in CBF were abolished by intravenous ascorbic acid, suggesting that antioxidant capacity may influence microvascular perfusion. Future work should directly assess global and regional ROS production and test whether targeted antioxidant interventions elicit either globally or region-specific effects on blood flow to better define mechanistic links between oxidative stress and hyperoxia-induced vascular responses. Investigating these mechanisms in populations with altered cerebrovascular or metabolic status—such as older adults or patients with cardiovascular or neurodegenerative disease where ROS may be elevated—will also be critical to applying these findings to clinically relevant populations. Future investigations should examine the effects of hyperoxia across a range of oxygen doses, including graded and clinically typical FiO₂ levels (0.3–0.4) to determine whether hyperoxic vasoconstriction observed at higher oxygen fractions persist under standard clinical practice.

It is noteworthy that, although CaO₂ increased significantly during hyperoxia, there was no corresponding change in CDO₂ in either GM or WM. This indicates that the rise in blood oxygen content did not translate into enhanced tissue oxygen delivery, suggesting that hyperoxic therapy may not provide the intended benefit. This is true, at least, in the young, healthy adult population measured in the present study; findings may vary considerably across age or clinical conditions. In studies showing absolute CBF reductions, hyperoxia leads to real decrement in CDO₂, which may be counterproductive to maintaining oxygen delivery.

Conclusion

In conclusion, isocapnic hyperoxia in young, healthy adults evokes global, lobar and regional reductions in cerebrovascular conductance. The new finding of broad lack of differences between brain lobe responses suggests a broadly uniform constrictor response at lobe level, and exploratory ROI constriction deserves more study for potential of variable constriction within lobes. These findings provide a detailed characterization of cerebrovascular responses under hyperoxic conditions and offer a framework for future research into the mechanisms underlying hyperoxia-induced vasoconstriction, the role of oxidative stress, and regional vascular sensitivity with larger sample size and/or longer hyperoxia exposures. Understanding these dynamics is essential for optimizing the safe use of supplemental oxygen in both healthy and clinical populations.

Supplementary Material

Supplemental figures S1-S3 (https://doi.org/10.6084/m9.figshare.31337596 )

Supplemental tables S1-S3 (https://doi.org/10.6084/m9.figshare.31337608 )

DATA AVAILABILITY

Data will be made available upon reasonable request, upon ethical approval for data sharing.