Mapping the interplay of atrial fibrillation, brain structure, and cognitive dysfunction

Abstract INTRODUCTION Atrial fibrillation (AF) is associated with an elevated risk of cognitive impairment and dementia. Understanding the cognitive sequelae and brain structural changes associated with AF is vital for addressing ensuing health care needs. METHODS AND RESULTS We examined 1335 stroke‐free individuals with AF and 2683 matched controls using neuropsychological assessments and multimodal neuroimaging. The analysis revealed that individuals with AF exhibited deficits in executive function, processing speed, and reasoning, accompanied by reduced cortical thickness, elevated extracellular free‐water content, and widespread white matter abnormalities, indicative of small vessel pathology. Notably, brain structural differences statistically mediated the relationship between AF and cognitive performance. DISCUSSION Integrating a comprehensive analysis approach with extensive clinical and magnetic resonance imaging data, our study highlights small vessel pathology as a possible unifying link among AF, cognitive decline, and abnormal brain structure. These insights can inform diagnostic approaches and motivate the ongoing implementation of effective therapeutic strategies. Highlights We investigated neuropsychological and multimodal neuroimaging data of 1335 individuals with atrial fibrillation (AF) and 2683 matched controls. Our analysis revealed AF‐associated deficits in cognitive domains of attention, executive function, processing speed, and reasoning. Cognitive deficits in the AF group were accompanied by structural brain alterations including reduced cortical thickness and gray matter volume, alongside increased extracellular free‐water content as well as widespread differences of white matter integrity. Structural brain changes statistically mediated the link between AF and cognitive performance, emphasizing the potential of structural imaging markers as a diagnostic tool in AF‐related cognitive decline.


Methods
Together, SBM and VBM are imaging methods that enable the identification and characterization of macrostructural brain abnormalities.

Surface-based morphometry
SBM results in vertex-level measures of cortical thickness and cortical folding, i.e., point-wise estimates of cortical geometry.Surface-based morphometry was performed in two stages: (1)   surface creation and registration and (2) computation of morphometric estimates.In the surface creation phase, CAT12 utilized a projection-based thickness method to derive the initial cortical thickness and central surface.Subsequently, the central surface underwent refinement and topological defect correction, yielding final central, pial, and white matter surface meshes.These refined surfaces were then employed to re-estimate cortical thickness using the FreeSurfer thickness method, which assesses the width of the gray matter ribbon by measuring the distance between its inner and outer boundaries. 2 Furthermore, the final central surface served for the computation of three cortical folding metrics.The gyrification index was computed as the "smoothed absolute mean curvature", which involved averaging curvature values from each vertex of a spherical surface mesh within a 3 mm radius. 3Larger negative gyrification index values signify sulci, while positive values represent gyri.Sulcal depth was computed as the distance between a point on the sulcal surface and the nearest point on the brain's convex hull, which envelopes the cortex's outermost (pial) surface. 4Deeper sulcal depths indicate more pronounced sulci, while shallower depths suggest less pronounced ones.
Fractal dimension was calculated as the slope of a logarithmic plot comparing surface area to the maximum l-value of the surface reconstruction (a measure of the bandwidth used to reconstruct the surface shape), with a higher FD indicating greater complexity and irregularity of cortical folding, and a lower FD denoting a smoother surface. 5Cortical thickness was smoothed using a 12mm kernel, while the cortical folding metrics were smoothed with a 20mm kernel.

Voxel-based morphometry
Voxel-based morphometry (VBM) results in voxel-wise estimates of gray matter volume.In CAT12, the VBM protocol include a tissue segmentation and spatial registration procedure.
First, the T1w was denoised by applying a spatially adaptive non-local means (SANLM) filter.
After the denoising, SPM's unified segmentation was performed, producing segmentation maps of multiple brain tissues, including a gray matter map.These gray matter segmentations capture the voxel-wise gray matter content.The tissue segmentations were registered to standardized MNI space templates using Geodesic Shooting. 6The registration information was then incorporated into the gray matter segmentation by multiplying the deformation maps with the gray matter segmentation maps in a "modulation step."This adjustment accounts for volume alterations during spatial normalization, thereby estimating local gray matter volume in native space. 7Finally, modulated gray matter volume maps were smoothed using a 6mm Gaussian kernel.

Preprocessing
Analysis of diffusion-weighted MR images (DWI) was based on data already preprocessed by the UK Biobank.Details on UK Biobank preprocessing procedures can be found online (https://biobank.ctsu.ox.ac.uk/crystal/crystal/docs/brain_mri.pdf).In brief, DWI preprocessing included correction of eddy currents, head motion and outlier-slice removal using FSL's eddy as well as gradient distortion correction. 8

Diffusion tensor imaging and Neurite Orientation Dispersion and Density Imaging
Fractional anisotropy (FA) and mean diffusivity (MD) were derived from diffusion tensors which were modelled based on preprocessed DWI using a least-squares fit.

Tract-based spatial statistics
In order to derive skeletonized maps of each of the estimated diffusion parameters, we conducted tract-based spatial statistics (TBSS) utilizing the standardized FA template from FSL as the registration target. 13,14Put briefly, individual FA images in template space got eroded to exclude non-brain voxels on the outer edge of the image.Next, a valid mask containing only the intersection of all subjects' brains was derived and used to mask the average of all previously eroded FA images.This mean FA image was subsequently used to derive a white matter skeleton.Next, all individual FA images were projected onto the mean FA skeleton.The resultant projection vectors were used to skeletonize all of the remaining diffusion metrics, i.e.MD and NODDI markers.

Peak width of skeletonized mean diffusivity
Peak width of skeletonized mean diffusivity (PSMD) was calculated based on standard procedures. 15PSMD was calculated as the difference between the 95 th and 5 th percentile of MD values on the white matter skeleton in standard (MNI) space.A mask supplied by the developers was used to exclude white matter areas susceptible to partial volume effects of cerebrospinal fluid (https://github.com/miacresearch/psmd/blob/main/skeleton_mask_2019.nii.gz).

White matter hyperintensity segmentation
Precomputed WMH segmentation masks of the UK Biobank were used.To obtain them, FSL's Brain Intensity AbNormality Classification Algorithm (BIANCA) 16 was applied on FLAIR images and T1w images for white matter hyperintensity (WMH) segmentation.The WMH load was calculated as the WMH volume divided by the total intracranial volume calculated by CAT12.

Quality assurance
In order to minimize the influence of noise and artifacts, quality assurance (QA) of MRI data was conducted both quantitatively and qualitatively.

Figure S3 -Mediation analysis results
Mediation analysis results.Mediation effects of global imaging markers on the relationship between AF and attention/executive function as well as reasoning.Path plots display standardized effects and p-values: (a) AF to imaging marker, (b) imaging marker to cognitive score, (ab) indirect effect (c') direct effect and (c) total effect.Significant associations are in blue; non-significant in light gray.If a relationship is significantly mediated, i.e., the indirect effect ab was significant and the direct effect c' was reduced or non-significant compared to the total effect c, the text for ab is highlighted in blue.The left path plots show results regarding attention/executive function, while the right path plots depict those of reasoning.

Figure S4 -Regression analysis: cognition ~ cortical macrostructural imaging markers
Linear regression analysis of the association between macrostructural cortical markers and cognitive outcomes.Analysis incorporated cognitive domain scores as the dependent variable, global imaging markers as the independent variable, and included age, sex, education and cardiovascular risk factors as covariates.The interaction effect of an atrial fibrillation diagnosis (global imaging marker * atrial fibrillation) was also tested.Significant associations between imaging and cognitive scores are denoted by blue dots.Non-significant associations are shown in gray.Abbreviations: pFDR -false discovery rate corrected p-value, adjusted for age, sex, education, and cardiovascular risk; pinteraction -p-value for the interaction term (imaging marker * atrial fibrillation) with false discovery rate correction; rsp -Spearman correlation.

Figure S7 -Cortical effect maps comparison
The correlation matrix displays spatial correlations of Schaefer400-parcellated Cohen's d maps resulting from the comparison of regional imaging markers between AF and controls.The lower triangle of the matrix displays scatter plots and the upper triangle the Pearson correlation coefficients of the corresponding correlations.The color and size of the dots on the upper triangle encode the effect directionality and effect size, respectively.The asterisks indicate statistical significance.To account for spatial smoothness, p-values were determined based on spin permutations (pspin). 17The matrix diagonal displays histograms.

Figure S8 -Group comparison of tract-level fractional anisotropy (FA)
Radar plots represent the group differences for fractional anisotropy in each tract, with blue dots indicating significant differences and red dots marking non-significant differences.

Figure S9 -Group comparison of tract-level mean diffusivity (MD)
Radar plots represent the group differences for mean diffusivity in each tract, with blue dots indicating significant differences and red dots marking non-significant differences.

Figure S10 -Group comparison of tract-level neurite density index
Radar plots represent the group differences of neurite density in each tract, with blue dots indicating significant differences and red dots marking non-significant differences.

Figure S11 -Group comparison of tract-level orientation dispersion
Radar plots represent the group differences for orientation dispersion in each tract, with blue dots indicating significant differences and red dots marking non-significant differences.

Discussion
Text S2 -Microstructural abnormalities map to specific cortical networks Our study pinpointed AF-related variations in gray matter microstructure, with symmetrically impacted regions including the anterior cingulate, insula, medial prefrontal cortex, medial temporal lobe and precuneus.This observation becomes especially notable when contextualized with two pathomechanistic theories explaining its origin.
First, the pattern might be a manifestation of advancing AF-related cerebral pathophysiology.
The anterior cingulate, medial temporal lobe and precuneus have a particular role in Alzheimer's and vascular dementia, directly influencing associations with cognitive symptoms. 18- 21 t is plausible that pathology in these areas might accelerate cognitive decline and the onset of dementia in individuals with AF.As there is an observable disparity between cortical patterns of macro-and microstructural changes, we propose that these differences could represent varying local stages of tissue changes all tethered to a shared disease mechanism.An increase in extracellular free-water might act as a precursor to more pronounced neurodegenerative processes.Thus, the regions showing microstructural deviations in our current crosssectional analysis might manifest macrostructural changes as the disease progresses.This interpretation finds support in studies assessing cortical thickness in diverse stages of vascular cognitive impairment.These studies demonstrate that thickness differences between mild and severe cases of vascular cognitive impairment coincide with the cortical regions highlighted by altered microstructure in our analysis which might be due to common disease mechanisms. 22,23rther support comes from observations of microstructural indices being demonstrably highly sensitive to subtle disease alterations in vascular cognitive impairment and preceding morphological effects and lesion occurence. 24,25cond, on a more speculative note, the observed associative effects could also be attributed to neurodegenerative processes increasing the risk of AF.The regions we identified are all components of the central autonomic network (CAN), which plays a pivotal role in regulating heart rate and contractility by transmitting autonomic signals. 26,27Chronic ischemic and degenerative changes within the central autonomic network, might heighten the propensity for AF development.This perspective is supported by preclinical studies, which have demonstrated neurogenically triggered cardiac arrhythmias and changes in the left atrial myocardium poststroke in rodent models. 28,29Furthermore, in cases of AF detected after stroke (AFDAS), stroke lesions predominantly appear within the central autonomic network, pointing towards a potential neurogenic origin. 26Notably, lesions in the right anterior insula correlate with post-stroke acute myocardial injury, underscoring the region's significance in brain-heart interactions. 30In fact, the anterior insula is the only area showing significant group differences of neurite density in our analysis, implying greater intracellular volume loss possibly due to cellular and axonal damage.Longitudinal neuroimaging studies are needed to further substantiate these hypotheses in AF.
Figure S1 -Sample selection flowchart

Table S1 -
Excluded diagnoses based on non-cancer illness information 11,12thub.com/daducci/AMICO).11,12NODDI offers an assessment of the brain's microstructure beyond conventional DTI markers by capitalizing on a multi-compartment model.This model dissects the intricate features of neural tissue into distinct compartments.
9,10FA and MD provide insights into the directionality and magnitude of water diffusion within tissues, respectively.In addition, Neurite Orientation Dispersion and Density Imaging (NODDI) was performed using httpsrepresented by measurement of isotropic volume fraction (ISOVF), which reflects the volume occupied by the space outside the cellular structures.Neurite orientations, a key aspect of neural tissue microstructure and organization, are described by orientation dispersion.This metric provides insights into the variability in the alignment of axons and dendrites within a voxel.Together, DTI and NODDI measures provide a detailed understanding of the underlying tissue microarchitecture, aiding in deciphering the intricate complexities of neural structures.

Table S2 -
AF comorbidities c uncorrected P-values of χ2 tests

Table S3 -
Group comparisons controlling for AF comorbidities a Presented as mean ± SD (N) b Uncorrected P values of analyses of covariance, adjusted for age, sex, education, cardiovascular risk factors and AF comorbidities c False discovery rate-corrected P values of analyses of covariance, adjusted for age, sex, education, cardiovascular risk factors and AF comorbidities (*P <0.05, **P <0.01, ***P <0.001)

Table S4 -
Group comparisons of individual cognitive tests