Multi-modal magnetic resonance imaging to assess cerebrovascular function in patients with atrial fibrillation

Introduction

Atrial fibrillation (AF) is the most prevalent sustained cardiac arrythmia, reportedly now affecting ~0.5–1% of the global population. ¹ This supraventricular arrythmia is projected to cause significant social and economic effects as demands on global healthcare systems intensify with improved diagnostics, ageing populations and increases in associated risk factors (e.g. obesity and hypertension [HTN]).^2–4 AF significantly increases the risk of thromboembolic events such as stroke, which often have high mortality, are more disabling and associated with poorer quality of life outcomes than non-AF related strokes. ⁵ Furthermore, AF patients with no previous history of stroke are at heightened risk of developing neurodegenerative disorders, including cognitive decline and dementia.^6,7 Candidate mechanisms include chronic cerebral hypoperfusion, ⁸ endothelial inflammation ⁹ and covert cerebral infarction, ¹⁰ however, rigorous characterisation of cerebrovascular function in AF using magnetic resonance imaging (MRI) has yet to be performed.

Cerebrovascular function is commonly evaluated with cerebrovascular carbon dioxide reactivity (CVR_CO2) and neurovascular coupling (NVC) assessments, which indicate cerebral vasodilatory reserve capacity and an ability to increase cerebral blood flow (CBF) to support neural metabolic demands, respectively. Importantly, a blunted CVR_CO2 is associated with an increased mortality risk, ¹¹ while impaired NVC is related to pathologies such as Alzheimer’s disease. ¹² Notably, CVR_CO2 and NVC have previously been shown to be blunted in AF patients when assessed using transcranial Doppler ultrasound (TCD).^13,14 Moreover, AF patients also exhibit elevated biomarkers of endothelial damage (e.g., plasma von Willebrand factor ¹⁵ ), supporting the notion of endothelial dysfunction both peripherally and within the cerebral vasculature. However, whilst TCD provides a non-invasive method of interrogating cerebral hemodynamics, it only provides information relating to blood flow velocity and is isolated to a single vessel of interest upstream of brain parenchyma. ¹⁶ MRI has previously been employed in AF patients to assess resting brain perfusion and has demonstrated lower global resting perfusion compared to healthy control (HC) participants. ⁸ Furthermore, rhythm correction procedures normalise resting perfusion back to values seen in HCs. ¹⁷ However, the use of physiological assessments (i.e. NVC and CVR_CO2) have not been implemented concurrently with MRI in AF patients, which can offer direct (arterial spin labelling [ASL]) and indirect (blood oxygen level-dependent [BOLD]) measurements of cerebral perfusion. These techniques offer excellent spatial specificity, the ability to interrogate the microvasculature downstream of the larger intracranial vessels and allow absolute quantification of CBF. Furthermore, the combined use of BOLD and ASL allows the estimation of cerebral metabolic rate of oxygen (CMRO₂) changes within the brain during neural activation (e.g., during NVC assessments), offering a closer indication of changes in underlying neural activity.

The aim of this study was to characterise the effects of AF on cerebrovascular function through assessments of NVC and CVR_CO2 using functional and perfusion MRI techniques. Considering existing evidence of cerebrovascular dysfunction in AF, we hypothesised that AF patients would exhibit worsened NVC and CVR_CO2 compared with HC.

Methods

Participant characteristics

19 AF patients and 20 HC participated in this study. AF patients were recruited through specialist clinics in Te Whatu Ora (Auckland), as well as community and online study advertisements. All patients had received a clinical diagnosis of AF and exhibited various AF temporality (i.e. paroxysmal AF vs persistent AF). Additionally, all patients underwent a comprehensive health screening and completed an MRI safety questionnaire with a cardiology nurse and study researcher prior to study participation. Exclusion criteria included valvular heart disease, severe left ventricular dysfunction, acute coronary syndrome within the previous 12 months, insulin-dependent diabetes, previous stroke and transient ischemic attack, uncontrolled thyroid disorders, current cancer diagnosis and severe respiratory, neurological, inflammatory, connective tissue, hepatic and renal disease. Additional exclusion criteria included contraindications for MRI (e.g. metallic implants), severe claustrophobia, current use of oral nitrates or recreational drugs, alcohol abuse and body mass index (BMI) < 18 kg/m². Anthropometric measurements (height and weight) were recorded, and office blood pressure (BP) was measured using an automated oscillometric device (M2, Omron, Kyoto, Japan) at the initial screening visit. Participant characteristics and medication use are shown in Table 1 and Supplemental Table 1, respectively.

OPEN IN VIEWER

Table 1.

Participant characteristics.

Variable	HC	AF	p
n	20	19
Age (years)	63.4 ± 9.5	66.6 ± 12.0	0.349
Women, n (%)	10 (50)	5 (27)	0.191
Weight (kg)	69.2 [60.1–79.0]	91.3 [78.6–102.4]	0.001
Height (m)	1.69 ± 0.12	1.76 ± 0.09	0.065
BMI (kg.m−2)	25.0 ± 3.6	29.2 ± 4.3	0.002
Systolic BP (mmHg)	124.0 ± 16.6	131.5 ± 13.6	0.128
Diastolic BP (mmHg)	74.6 ± 10.4	82.8 ± 10.6	0.020
MAP (mmHg)	88.2 [84.1–100.3]	95.3 [91.7–104.7]	0.041
HR (bpm)	65.9 ± 8.7	66.3 ± 9.5	0.901
PETCO2 (mmHg)	34.1 ± 4.2	31.5 ± 2.9	0.027
Hypertension, n (%)	1 (5)	4 (21)	0.182
Diabetes, n (%)	0 (0)	1 (5)	0.487
Previous smoker, n (%)	5 (25)	4 (22)	1.000
Alcohol use, n (%)	14 (70)	13 (68)	1.000
Activity status (1–5)	3 [2–4]	3 [3–4]	0.201
Fibrillating at time of study, n (%)	—	7 (37)	—
Persistent AF, n (%)	—	6 (32)	—
Previous rhythm correction, n (%)	—	7 (37)	—

AF: atrial fibrillation; BMI: body mass index; BP: blood pressure; HC: healthy control; HR: heart rate; MAP: mean arterial pressure; P_ETCO₂: partial pressure of end-tidal carbon dioxide.

Mean ± SD displayed for normally distributed and median [interquartile range] for non-normally distributed continuous variables. Frequency (percentage) displayed for categorical discrete variables. Independent samples t-test or Mann–Whitney U test used to infer statistical differences for continuous variables and Fisher’s Exact Probability test for categorical data. Patients in AF at the time of data acquisition are identified as ‘Fibrillating’. HC, n = 20; AF, n = 19.

Bold font denotes significance p ⩽ 0.05.

Data analysis

Respiratory

Temporal alignment of P_ETCO₂ data and BOLD and ASL data was required to accurately calculate CVR_CO2. The end-tidal PCO₂ (P_ETCO₂) was obtained on a breath-by-breath basis as the peak within-breath PCO₂ value. Temporal alignment was performed for each participant to account for the delay between the P_ETCO₂ trace and the BOLD signal. Firstly, P_ETCO₂ was down sampled to match the BOLD TR (2 s). Then a grey matter (GM) region of interest (ROI) was used to extract mean BOLD signal intensity across the entire time course, and the optimal time shift was found through an iterative process shifting the P_ETCO₂ trace by one TR (2 s) until the maximum cross-correlation between the P_ETCO₂ trace and the BOLD signal was obtained. ²¹ The same time shift was applied to P_ETCO₂ data obtained during ASL data acquisition. In addition, the P_ETCO₂ trace was shifted by 4 s to account for the delay induced by the gas sampling line length, calculated prior to the study with a known CO₂ concentration change between the sampling point and the gas analyser. For CVR_CO2 assessment, P_ETCO₂ data was averaged across a 64 s period preceding gas administration and 64 s across the gas administration period, respectively. For NVC assessment, data were averaged across the final 32 s of the stimulation off period and the first 32 s of the visual stimulation period, respectively. These timepoints were chosen so that the same period could be sampled relative to the experimental stimulus across BOLD and ASL data, as each ASL label/control pair required 16 s for data acquisition, meaning CBF data were acquired every 16 s. This is outlined in Supplemental Figure 1.

Brain magnetic resonance imaging

All DICOM MRI data were firstly converted to the Brain Imaging Data Structure (BIDS) format ²² using the ezBIDS BIDS-conversion tool. ²³ Tissue segmentation and anatomical volume assessments were performed using FreeSurfer (version 7.3.2)^24–26 on each participant’s T₁−w MPRAGE scan. Volumes for the following regions were extracted: total brain (excluding ventricles), total GM, whole-brain cortical ROI, cerebellar GM, hippocampal GM, frontal lobe GM, parietal lobe GM, occipital lobe GM and temporal lobe GM volume. Left and right volumes were summed for analysis, and all volumetric data was normalised to total intracranial volume.

Functional magnetic resonance imaging (fMRI) preprocessing was performed using fMRIPrep^27,28 and included tissue segmentation from FreeSurfer performed as described, slice timing correction, motion correction, susceptibility distortion correction, registration to the T1−w image, estimation of confounds and spatial smoothing with a 4 mm full-width at half maximum Gaussian kernel. A general linear model (GLM) was built containing eleven regressors (six motion parameters and five anatomical component-based noise correction [aCompCor] parameters to remove physiological noise from white matter (WM) and cerebrospinal fluid regions). For CVR_CO2 data, two additional regressors were included in the GLM (the shifted P_ETCO₂ time course and linear drift). For NVC data, the timing of visual stimulus presentations was modelled as a regressor in the GLM and convolved with the hemodynamic response function, and temporal filtering was performed using a high-pass filter of 0.0078 Hz to account for linear drift. ²⁹

ASL data was analysed using ExploreASL (version 1.11.0) ³⁰ with Matlab version 9.14.0 (R2023a). ³¹ This included motion correction, registration to the T₁−w image, pairwise subtraction of label/control images and quantification of CBF using the single compartment model ³² with partial volume correction.

Neurovascular coupling

An occipital pole ROI was extracted from the output of the FreeSurfer grey matter parcellation with the Destrieux atlas. This ROI was then resampled to the participants native BOLD and ASL images. The mean BOLD signal intensity and mean partial volume corrected CBF value within this ROI for each volume was then extracted across the whole-time course. Outliers were defined as the mean of the timeseries ± interquartile range (IQR)*2.2 and excluded from analyses. The timeseries was then split into individual epochs such that a mean epoch could be calculated for each participant, representing a mean BOLD response and a mean CBF response to visual stimulation. To ensure the same time period of visual stimulation was extracted from BOLD and ASL data, the stimulus “on” phase was defined as a 32 s period that ended 8 s prior to the end of visual stimulation, and the “baseline” phase as the 32 s preceding visual stimulation (see Supplemental Figure 1). Individual on and off epochs were averaged and used to calculate relative changes in BOLD and CBF due to the stimulus.

Cerebrovascular carbon dioxide reactivity

A whole-brain cortical mask was extracted from the output of the FreeSurfer grey matter parcellation with the Destrieux atlas and applied to the BOLD and ASL timeseries data to calculate regional CVR_CO2 responses. In addition, to perform calibrated BOLD, the same occipital pole ROI used for NVC analysis was used to extract the mean BOLD signal intensity and mean partial volume corrected CBF value for each volume across the whole-time course during CVR_CO2 assessment. Outliers were defined as the mean timeseries ± interquartile range (IQR)*2.2 and excluded from analyses, and the final 64 s of the baseline period and CO₂ stimulus, respectively, were used to calculate relative changes in BOLD and CBF. In addition, whole-brain voxel-wise CVR maps for BOLD and CBF data were computed for each individual and normalised to MNI space, before group averaging (see Supplemental Figure 2).

Calibrated blood oxygen level-dependent

The Davis model ³³ was used to estimate fractional changes in CMRO₂ evoked by visual stimulation using the following equation:

Δ S S 0 = M [ 1 − ( C B F C B F 0 ) α − β ( C M R O 2 C M R O 2 , 0 ) β ]

Where ∆S is BOLD signal change from baseline during visual stimulation, S₀ is baseline BOLD signal, M is the BOLD calibration factor, CBF is cerebral blood flow during visual stimulation, CBF₀ is baseline cerebral blood flow, α is the ratio of changes in CBF to changes in CMRO₂ (assumed a constant = 0.14), ³⁴ CMRO₂ is baseline cerebral metabolic rate of oxygen, CMRO_2,0 is cerebral metabolic rate of oxygen during visual stimulation and β is the ratio of changes in CMRO₂ to changes in BOLD signal (assumed a constant = 0.91). ³⁴

To estimate the calibration factor (M), the following equation was used:

M = = Δ S C O 2 S 0 1 − ( C B F C O 2 C B F 0 ) α

Where ∆S_CO2 is BOLD signal change from baseline during CO₂ administration, S₀ is baseline BOLD signal, CBF_CO2 is cerebral blood flow during CO₂ administration, CBF₀ is baseline cerebral blood flow and α is the ratio of changes in CBF to changes in CMRO₂ (as defined above).

CMRO₂ was then estimated by rearranging the BOLD signal change equation as follows:

C M R O 2 C M R O 2 , 0 = [ 1 − Δ S S 0 ⋅ 1 M ( C B F C B F 0 ) α − β ] 1 β

Cardiac magnetic resonance imaging

Cine MRI analysis was performed using Cardiac Image Modeller (v 9.1.0) as previously described. ³⁵ Briefly, this included definition of the apex and base points on the apical and basal short axis images, respectively. The two right ventricle insertion points to the left ventricle were then selected on a minimum of one short axis image. On each long axis image, two baseplane points were placed at the intersection of the mitral leaflet with the endocardium and interpolated across image frames. Participant images were then registered with the database model, allowing the generation of endocardial and epicardial contours on participant images. After correcting for any major breath-hold mis-registrations, the participant model was customised by the placement of guide-points on the short and long axis images, allowing deformation of the 3D model and subsequent acquisition of volumetric results.

Aortic flow analysis was performed using Syngo.via MR Cardiac Analysis Flow Quantification (Siemens Healthineers, Siemans Healthcare GmbH, Erlanger, Germany). This included automated vessel segmentation, automatic propagation and velocity encoding gradient correction.

Body surface area (BSA) was used to standardise cardiac measures and was calculated as follows: BSA = √((weight × height) ÷ 3600).

Results

Cognition

No differences were observed between groups across CANTAB assessments of executive function, visual memory and social cognitive domains (Table 2, all p > 0.05).

OPEN IN VIEWER

Table 2.

CANTAB outcome measures.

CANTAB variable	HC	AF	p
Paired associates learning
PALTEA	12.0 [7.25–18.50]	16.0 [6.0–39.0]	0.398
PALFAMS	12.25 ± 4.35	11.89 ± 4.45	0.802
Multi-tasking test
MTTLM	800.89 ± 100.49	838.40 ± 155.52	0.374
MTTTC	158.0 [152.0–159.0]	155.0 [150.0–158.0]	0.162
Spatial working memory
SWMTE	21.0 [8.25–24.0]	18.0 [9.0–23.0]	0.265
SWMS	10.0 [8.0–11.0]	9.0 [3.0–10.0]	0.105
Emotion recognition task
ERTCRT	1352.40 ± 348.98	1503.58 ± 349.21	0.185
ERTTH	27.30 ± 3.71	26.89 ± 5.34	0.784

AF: atrial fibrillation; ERTCRT: emotion recognition task median correct reaction time; ERTTH: emotion recognition task total correct emotion selections; HC: healthy control; MTTLM: multi-tasking test mean reaction latency; MTTTC: multi-tasking test total correct responses; PALFAMS: paired associates learning first attempt memory score; PALTEA: paired associates learning adjusted total errors; SWMS: spatial working memory strategy; SWMTE: spatial working memory total errors.

Mean ± SD displayed for normally distributed and median [interquartile range] for non-normally distributed continuous variables. Independent samples t-test or Mann–Whitney U test used to infer statistical differences. All HC, n = 20 except MTTTC (n = 19); AF, n = 19.

Significance p ⩽ 0.05.

Structural magnetic resonance imaging

Figure 1 displays relative volumes of major brain structures. Hippocampal GM volume (relative to total intracranial volume) was lower in AF (0.50 ± 0.06%) compared to HC (0.55 ± 0.06%; p = 0.020). No differences were observed between groups for relative measures of total brain volume (HC: 73.4 [68.5–74.8]; AF: 70.6 [66.3–72.1]%; p = 0.064), total GM volume (HC: 39.1 ± 3.11; AF: 37.6 ± 2.35%; p = 0.115), whole-brain cortical volume (HC: 28.6 ± 2.23; AF: 27.6 ± 1.64%; p = 0.117), cerebellar GM volume (HC: 6.81 [6.28–7.44]; AF: 6.58 [5.99–7.11]%; p = 0.187) and occipital lobe GM volume (HC: 2.90 ± 0.36; AF: 2.86 ± 0.32%; p = 0.769).

OPEN IN VIEWER

Figure 1.

Relative regional brain volumes: (a) total brain (excluding ventricles), (b) total GM, (c) whole brain cortical ROI, (d) cerebellar GM, (e) hippocampal GM, and (f) occipital lobe GM.

Cerebrovascular carbon dioxide reactivity

Respiratory data during CVR_CO2 assessment is presented in Supplemental Table 3. Hypercapnia evoked similar increases in P_ETCO₂ between groups during BOLD (HC: +8.9 [7.5–10.7]; AF: +9.8 [7.8–10.8]∆mmHg; p = 0.535) and ASL scans (HC: +9.1 ± 2.4; AF: +9.9 ± 2.3∆mmHg; p = 0.361). No differences were observed in P_ETCO₂ between groups, scans or their interaction (Supplemental Table 3 all p > 0.05).

%∆BOLD, absolute CBF changes, and indices of CVR_CO2 extracted from a whole-brain cortical ROI are shown in Figure 2. No differences were observed in %∆BOLD between groups (HC: 3.09 ± 0.50; AF: 3.23 ± 0.70%; p = 0.513). Hypercapnia evoked increases in CBF (Figure 2 main effect of condition; p < 0.001), whilst no differences were observed across group or interaction (Figure 2, both p > 0.05). %∆CBF was not different between groups (HC: 36.6 ± 18.8; AF: 43.6 ± 20.6%; p = 0.319). No differences were observed in CVR_CO2 between groups for BOLD (HC: 0.33 ± 0.07; AF: 0.35 ± 0.10 %∆BOLD/∆P_ETCO₂ [mmHg]; p = 0.651) or CBF data (HC: 4.19 ± 2.09; AF: 4.50 ± 2.17 %∆CBF/∆P_ETCO₂ [mmHg]; p = 0.686).

OPEN IN VIEWER

Figure 2.

Cortical BOLD (a, d) and CBF (b, c, e) responses during CVR_CO2 assessment: (a) %∆BOLD with hypercapnia, (b) absolute CBF at baseline and during hypercapnia, (c) BOLD CVR_CO2, and (d) CBF CVR_CO2.

Neurovascular coupling

P_ETCO₂ during NVC assessment is depicted in Supplemental Table 4. Visual stimulation did not evoke any changes in P_ETCO₂ during BOLD and ASL scans (both p > 0.05). P_ETCO₂ was lower in AF compared to HC during BOLD and ASL scans (both p < 0.05) and was higher during the BOLD scan compared to ASL in the HC group (p < 0.05).

%∆BOLD and absolute CBF changes evoked by visual stimulation are shown in Figure 3. No differences were observed in %∆BOLD between groups (HC: 1.60 ± 0.38; AF: 1.66 ± 0.79%; p = 0.762). Visual stimulation evoked increases in CBF (Figure 3, main effect of condition; p < 0.001), whilst no differences were observed across group or interaction (Figure 3, both p > 0.05). In addition, %∆CBF was not different between groups (HC: 49.4 [20.1–71.9]; AF: 51.7 [38.3–66.9]%; p = 0.850).

OPEN IN VIEWER

Figure 3.

BOLD and CBF responses during NVC assessment: (a) %∆BOLD to visual stimulation, (b) trace of group mean %∆BOLD during NVC assessment, and (c) absolute CBF at baseline and during visual stimulation.

Occipital cerebrovascular carbon dioxide reactivity

%∆BOLD, absolute CBF changes, and indices of CVR_CO2 extracted from an occipital ROI are shown in Figure 4. No differences were observed in %∆BOLD between groups (HC: 4.68 ± 2.33; AF: 5.10 ± 1.10%; p = 0.669). Hypercapnia evoked increases in CBF (Figure 4, main effect of condition; p < 0.001), whilst no differences were observed across group or interaction (Figure 4, both p > 0.05). %∆CBF was not different across groups (HC: 73.3 [47.3–97.9]; AF: 74.2 [38.2–220.5]%; p = 0.851]. No differences were observed in CVR_CO2 between groups for BOLD (HC: 0.42 [0.32–0.64]; AF: 0.51 [0.40–0.67] %∆BOLD/∆P_ETCO₂ (mmHg); p = 0.331) or CBF (HC: 7.46 [5.06–12.60]; AF: 7.56 [3.91–16.79] %∆CBF/∆P_ETCO₂ (mmHg); p = 1.000). Consistent with our CVR_CO2 ROI analyses, whole-brain CVR_CO2 maps depict notably higher CVR_CO2 responses within the occipital lobe (Supplemental Figure 2).

OPEN IN VIEWER

Figure 4.

Occipital BOLD and CBF responses during CVR_CO2 assessment: (a) %∆BOLD with hypercapnia, (b) absolute CBF at baseline and during hypercapnia, (c) BOLD CVR_CO2, and (d) CBF CVR_CO2.

Calibrated blood oxygen level-dependent

Visual stimulation evoked similar increases in CMRO₂ across groups (HC: 5.25 ± 1.84; AF: 4.36 ± 2.50%, p = 0.285). Within our occipital ROI, %∆CMRO₂ showed a strong positive correlation with %∆CBF in both our HC (r = 0.59, p = 0.043, 95% confidence interval [0.03–0.87]) and AF group (r = 0.56, p = 0.024, 95% confidence interval [0.09–0.83]). These results are displayed in Figure 5.

OPEN IN VIEWER

Figure 5.

CMRO₂ changes evoked by visual stimulation: (a) %∆CMRO₂ during visual stimulation and (b) relationship between %∆CMRO₂ and %∆CBF during visual stimulation.

Discussion

To the best of our knowledge, this is the first study to characterise NVC and CVR_CO2 responses in AF patients using MRI techniques. We have identified that, in contrast to TCD-derived evidence of cerebrovascular dysfunction in AF, using MRI, NVC and CVR_CO2 responses are not diminished in AF compared to HC.

The observed reduction in hippocampal volume in AF is in line with previous evidence. ¹⁸ The hippocampus is highly sensitive to reductions in perfusion, ³⁷ and in AF may be associated with intermittent microembolization, ³⁸ endothelial damage ¹⁵ and intermittent hypoxia ³⁹ associated with reduced perfusion during AF episodes. Interestingly, previous studies have reported conflicting results in terms of overall brain atrophy in AF, with both reductions compared to HC ¹⁹ and no differences ¹⁸ in brain volume reported. Despite demonstrating a trend for lower total brain volume, there was no difference between our HC and AF groups. This may be explained by the cumulative AF burden not being sufficient to elicit global negative effects on brain structure, as it is unlikely that the perfusion deficits elicited by AF exclusively affect the hippocampus.

Test scores for executive function, visual memory and social cognitive domains were not different between HC and AF groups, indicating no cognitive deficits in our AF cohort in the evaluated domains. Previous investigations have shown mixed cognitive outcomes in AF patients,^40,41 however, when cognitive deficits have been identified, executive function^18,20 and processing speeds^18,20,42 are typically the domains affected. Interestingly, previous work has identified associations between cognitive dysfunction and AF temporality, with permanent AF, but not paroxysmal AF patients, demonstrating worse performance on executive function tests compared to HC. ⁴³ Considering that 68% of AF patients in the present study had paroxysmal AF, it may be that the relatively lower AF burden was not great enough in our AF cohort to elicit cognitive deficits. However, in the context of our hippocampal results, the lack of deficits in cognition, especially memory, are surprising. Reductions in hippocampal volume and cognitive deficits in both memory and executive function domains have previously been associated in AF. ¹⁸ Indeed, the significant correlation found between lower hippocampal volume and worsened visual memory scores in the present study suggests we would expect worse visual memory performance in AF, particularly as memory performance is highly dependent on hippocampal integrity. ⁴⁴ Interestingly, patients with mild cognitive impairment who have worsened performance on tasks assessing visual memory also exhibit worsened hippocampal activation during high task load, likely due to decreases in hippocampal GM. ⁴⁵ This may suggest that AF exhibit some cognitive reserve, whereby structural deficits have not yet manifested as functional deficits in cognition, or that less pronounced hippocampal volume loss is present in AF compared to clinical populations with overt cognitive deficits.

Interestingly, CVR_CO2 responses in both cortical, and occipital ROIs were not different between HC and AF groups for BOLD and CBF measures. This contrasts with our previous work using TCD, whereby CVR_CO2 responses were reduced in AF compared to HC participants. ¹⁴ Importantly, in the present study, we directly quantified cerebral perfusion, as opposed to measuring velocity responses in a single isolated vessel of interest. Indeed, Burley et al. have demonstrated contradictory findings when implementing TCD and MRI methods concurrently to assess CVR_CO2, identifying no significant correlations between TCD and BOLD-derived CVR_CO2 indices when using a 5% CO₂ stimulus. ⁴⁶ One explanation for this may be the differential parts of the vascular network that the TCD and MRI methods assess. BOLD MRI measures relative change in oxygenation in the venules and veins, ASL MRI measures CBF in small vessels in GM tissue, whereas TCD measures blood velocity in arteries that ultimately reflects microvascular responses in the parenchyma. ⁴⁶ Additionally, vascular responses to CO₂ are different along the vascular tree. ⁴⁷ Thus, it may be that cerebrovasculature dysfunction is site-specific in AF. Pial vessels are known to be the primary drivers of CO₂-mediated dilatation responses, ⁴⁸ with subsequent dilatation of upstream intracranial ⁴⁹ and extracranial arteries. ⁵⁰ The haemodynamic and metabolic burden of AF may primarily be elicited on these larger vessels, whilst smaller vessels and venules downstream, with which we interrogated with MRI, may be protected to maintain adequate cerebral perfusion at the level of the brain parenchyma. In addition, CO was found to be a significant predictor of BOLD CVR_CO2 responses in the cortex. Thus, considering CO in AF was slightly elevated compared to HC, it may be unsurprising that CVR_CO2 responses were not blunted in AF. This is also in agreement with previous work in heart failure (HF) patients, which demonstrated a similar relationship between CO and resting CBF. ⁵¹ This could be an important consideration in AF patients who present with downstream HF as a function of their abnormal rhythm. ⁵² Methodological differences may also have contributed to the divergent CVR_CO2 responses reported. While in the current study a steady-state 5% hypercapnic stimulus was used, in our previous work a two-step 4% and 7% hypercapnic stimulus was implemented, whereby the slope of the relationship between P_ETCO₂ and blood flow velocity determined CVR_CO2. ¹⁴

BOLD and CBF responses during NVC assessment were also not different between HC and AF groups. Furthermore, neural activation (i.e. CMRO₂) evoked by visual stimulation was similar in HC and AF. Occipital lobe GM volume was not different between groups (Figure 1), thus ruling out the possibility of neurodegenerative blunting. Furthermore, our regression analysis revealed occipital lobe GM volume to be a predictor of %∆CBF and %∆CMRO₂ elicited by visual stimulation. Neurodegeneration in visual regions has previously been reported in AF, ⁵³ but this is yet to be coupled with NVC assessments. Moreover, CBF changes evoked by increases in CMRO₂ were well matched between groups, indicating that our AF cohort maintained robust NVC. NVC responses originate in the brain parenchyma, whereby neuronal activity results in local arteriole dilatation, ultimately spreading upstream to alter flow in larger arteries by retrograde propagation.^54–56 As aforementioned in the context of CVR_CO2 responses, our previous work identified NVC deficits in AF using TCD to assess PCA responses to visual stimulation. ¹³ Thus, it may be that progressive damage occurs along the vascular tree with increasing AF burden, with downstream vessels, as assessed in the present study, remaining protected as larger upstream arteries compensate to protect local NVC mechanisms. Importantly, despite demonstrating no functional deficits to visual stimulation in AF, it is also unclear whether other functional measures remain intact in AF, as no other methods of NVC assessment have been previously used (e.g. motor activation tasks).

Despite only a small number of AF patients fibrillating at the time of data acquisition, our subgroup analysis revealed no differences across NVC and CVR_CO2 responses, suggesting that AF per se does not impair NVC and CVR mechanisms. Turbulent flow causes the loss of normal SS patterns in AF,^57,58 which can reduce NO bioavailability as expression of eNOS is decreased. ⁵⁹ Due to the mechanistic redundancy associated with NVC and CVR_CO2,^56,60–62 it may be that in light of NO reductions, these responses can still occur similarly to those in normal sinus rhythm.

As expected, AF patients had significantly greater use of cardiovascular medications, including anticoagulants, β-inhibitors, Ca²⁺ channel inhibitors and statins (Supplemental Table 1). Whilst these may have the capacity to influence CBF and CVR through their respective mechanisms, such as the enhancement of eNOS, ⁶³ indirect effects on vascular health, ⁶⁴ BP reduction, autonomic effects, and changes to vascular tone, ⁶⁵ direct human evidence remains limited and incompletely understood in terms of their full influence on NVC and CVR. Moreover, given their preventative role and importance for control of AF symptoms, requesting AF patients to withhold these medications was deemed inappropriate after consultation with clinical collaborators.

Certain experimental limitations should be considered when interpreting the results of the present study. We acknowledge that the inclusion of AF patients with existing cognitive impairment in the present study may have strengthened our ability to examine underlying cerebrovascular function, though note that cognitive dysfunction is not a prerequisite for diminished NVC and CVRCO₂ responses.^66,67 BOLD and ASL data were acquired separately for NVC and CVR_CO2 assessments with standard MRI sequences. A custom BOLD and ASL sequence ⁶⁸ would have allowed both assessments to be acquired simultaneously but was not available. To mitigate order effects in our study, scan order was counterbalanced across participants. We acknowledge that the Davis model assumes constant α and β values when estimating CMRO₂ during the NVC task, however, recent empirical evidence suggests that these parameters show variability across cortical tissue. ⁶⁹ We cannot rule out the possibility that the cognitive tests used in the present study were insufficiently sensitive to detect subtle, domain-specific impairments in AF patients. However, the CANTAB assessments used have been validated and employed in numerous patient cohorts,^70–72 which gives us confidence in the approach taken. Moreover, due to the wide array of cognitive testing batteries available and the unique nature of their outputs, we are unable to compare absolute scores with previous studies. P_ETCO₂ is a valuable non-invasive measure of arterial pressure of CO₂ (P_aCO₂), ⁷³ but we acknowledge its possible underestimation of P_aCO₂ at rest. ⁷⁴ We used a fixed concentration of CO₂ (5%) in the present study to assess CVR_CO2 as the experimental set-up was MRI compatible, did not require a complex computerised gas delivery system, was simple to administer, demanded minimal participant engagement, and has good between-day test-retest reliability. ¹⁴ However, we appreciate that there is debate surrounding the optimal protocol for assessing CVR_CO2. ⁷⁵ The time of day in which experimental procedures were performed was matched as closely as possible between participants, however, due to MRI availability, this was not always possible. There is currently limited evidence of time-of-day effects ⁷⁶ associated with task-based functional and perfusion MRI, with most literature focused on resting-state spontaneous fluctuations in BOLD signals. ⁷⁷ Participants were requested to consume medications as normal prior to their experimental visit. We believe this approach allowed us to best assess participants in their typical medication state for a fairer assessment of their cerebrovascular function. We did not include medication use as a covariate during analysis due to the sample size and the wide range of medications used by AF patients.

In conclusion, this is the first study to implement MRI techniques to characterise NVC and CVR_CO2 responses in AF patients and suggests that these responses are not diminished in AF. Further studies are needed to fully elucidate the effects of AF and how they might vary along the cerebrovascular tree.

Supplemental Material

Supplemental Material

Supplemental Material