Association of cerebral small vessel disease burden with brain structure and cognitive and vascular risk trajectories in mid-to-late life

We characterize the associations of total cerebral small vessel disease (SVD) burden with brain structure, trajectories of vascular risk factors, and cognitive functions in mid-to-late life. Participants were 623 community-dwelling adults from the Whitehall II Imaging Sub-study with multi-modal MRI (mean age 69.96, SD = 5.18, 79% men). We used linear mixed-effects models to investigate associations of SVD burden with up to 25-year retrospective trajectories of vascular risk and cognitive performance. General linear modelling was used to investigate concurrent associations with grey matter (GM) density and white matter (WM) microstructure, and whether these associations were modified by cognitive status (Montreal Cognitive Asessment [MoCA] scores of < 26 vs. ≥ 26). Severe SVD burden in older age was associated with higher mean arterial pressure throughout midlife (β = 3.36, 95% CI [0.42-6.30]), and faster cognitive decline in letter fluency (β = −0.07, 95% CI [−0.13–−0.01]), and verbal reasoning (β = −0.05, 95% CI [−0.11–−0.001]). Moreover, SVD burden was related to lower GM volumes in 9.7% of total GM, and widespread WM microstructural decline (FWE-corrected p < 0.05). The latter association was most pronounced in individuals who demonstrated cognitive impairments on MoCA (MoCA < 26; F3,608 = 2.14, p = 0.007). These findings highlight the importance of managing midlife vascular health to preserve brain structure and cognitive function in old age.


Description of cohort profile
The Whitehall II study targeted all civil servants that worked in the London offices of 20 Whitehall departments between 1985-1988, established by University College London. The initial Wave included 10 308 British Civil servants (6895 men), aged 35-55 years. Whitehall II Study participants have received detailed clinical followups for up to 30 years at 5-year intervals (1991( -19941997-19992002-20042007-2009, Wave 9; 2012-2013, Wave 11); 2015-2016, Wave 12). Since the inception of the Whitehall II study, the retention rate for this cohort has been relatively high; about 87% of Wave 9 participants returned for the follow-up at Wave 11. The Whitehall II Imaging-Sub study randomly selected 774 participants aged 60-85 years from the Whitehall II Wave 11 cohort for multi-modal brain MRI work-up and cognitive tests at the University of Oxford. For the Imaging Sub-study, participants were included with contraindications to MRI scanning (e.g., particular metallic implants) or who were unable to travel to Oxford without assistance. 1, 2

Description of SVD ratings on MRI
Periventricular and deep white matter hyperintensities (WMH) were rated by trained raters (C.L.A., A.G.T., V.V.; see acknowledgments) on FLAIR images using the Fazekas scale, providing a score between 0-3 depending on the severity of WMH in the corresponding brain areas. 3 Enlarged perivascular spaces (EPVS) and lacunes were rated by an experienced rater (M.G.J.) following extensive training, and in consensus with other experienced raters (S.S., L.M.). Lacunes were rated using both T1-weighted and FLAIR images, following established criteria to distinguish lacunes from EPVS. 4,5 The intra-rater reliability for lacune ratings indicated high similarity, as reflected by an intraclass correlation (ICC) of 0.91, based on a random sample of 25 participants. EPVS were assessed in the basal ganglia on T1-weighted images using the validated qualitative EPVS rating scale, as T2weighted images were not acquired in this cohort. 6 The ICC for EPVS was 0.85, based on a random sample of 30 participants, indicating good intra-rater reliability. We used a semi-automatic detection method to identify possible cerebral microbleeds (CMBs), based on the radial symmetry transform. 7 Subsequently, one experienced rater (S.S.) evaluated all possible CMBs using previously established criteria 7,8 , and in consensus with a clinical psychiatrist (K.P.E.). The intra-rater reliability yielded excellent results (ICC = 0.92, based on a random sample of 100 participants).
MRI pre-processing steps MRI scans were analysed using FMRIB Software Library v6.0 (FSL; https://fsl.fmrib.ox.ac.uk/). 9 T1 images were bias corrected, brain extracted using FSL-ANAT and segmented using FSL-FAST to provide estimates of grey matter (GM, white matter (WM) and total brain volume (TBV). 10 All segmentations were visually inspected to ensure quality. Diffusion-weighted images were pre-processed using FMRIB's diffusion toolbox. 11 Briefly, after applying motion and eddy current corrections with FSL-TOPUP, diffusivity maps for each metric was extracted using DTIFit and aligned into standard space using FMRIB's Nonlinear Registration Tool (FNIRT).
FLAIR scans were used to extract WMH using the Brain Intensity AbNormality Classification Algorithm (BIANCA). 12 This algorithm uses both intensity features (provided from FLAIR, T1 images, and fractional anisotropy) and spatial features to classify all voxels. BIANCA was initially trained on manually segmented WMH masks of participants who were scanned on the Prisma scanner (N = 24), Verio scanner (N=24), and an independent sample from the UK Biobank Study (N = 12) to avoid scanner-dependent bias effects. This approach was specifically designed to improve the consistency of BIANCA output, and has been shown to result in better intra-and inter-rater reliability relatively to manual segmentations. [12][13][14] Additional information on pre-processing was mentioned previously. 1, 15 With regards to harmonization between scanners, several of the aforementioned pre-processing steps were specifically incorporated to minimize scanner effects (i.e. bias correction on T1 images, and training BIANCA on the Prisma, Verio, and an independent sample of the UK Biobank). In addition, we have compared volumetric measures between the Prisma and Verio scanner in previous work, where measures of GM and CSF were relatively increased at the Prisma scanner, and measures of WM were relatively increased at the Verio scanner. 16 To ensure that our results were not affected by scanner effects, we also included scanner as confounding variable in all of our analyses.

Study variables
The FSRS was based on age, sex, systolic blood pressure, use of antihypertensive medications, diabetes mellitus, current smoking, current or history of atrial fibrillation, left ventricular hypertrophy, and current or history of cardiovascular disease. 17 These measurements were obtained using both questionnaires and clinical assessments, using standard operating protocols, as described previously. 18, 19 Blood pressure measurements were obtained in sitting position after five minutes rest; the average of two measurements was used for further analysis. The use of antihypertensive medication was self-reported (e.g., diuretics, beta blockers, angiotensin-converting enzyme inhibitors, and calcium channel blockers). Diabetes was defined by having a fasting glucose level of ≥ 7.0 mml/L or a 2hr post-load glucose level of ≥ 11.1 mml/L, based on glucose measurements obtained from venous blood; self-reported diabetes diagnosed by a doctor or use of diabetes medication. 20 Smoking behaviour was self-reported (current, past/no smoking). A standard electrocardiogram analysis combined with manual review and Minnesota code classification system for electrocardiographic findings was used to identify atrial fibrillation and left ventricular hypertrophy. 21 Cardiovascular disease was evaluated using corroborated records from the general practitioner, hospital, and electrocardiogram and angiogram examinations at Wave 1, 3 and 5. Subsequently, the FSRS was computed using the beta coefficients of the Cox proportional hazards regression model in the Framingham Study, to indicate an individual's 10-year risk of stroke. 22

5
The longitudinal test battery of the Whitehall II cohort includes several cognitive tests, proven to be sensitive to detect changes in cognitive functions in this study population. 23 The complete test battery took 30 minutes to complete. To measure letter fluency, participants were instructed to recall as many words beginning with an "S" within one minute. For semantic fluency, participants were given similar instructions, but instead needed to recall as many animal names. Short-term memory was evaluated by initially presenting a list of 20 one or two syllable words at two seconds intervals. Subsequently, participants were asked to recall as many of the word list, within two minutes. The Alice Heim 4-I test composes 65 verbal and mathematical reasoning items with increasing difficulty (e.g., where participants had to identify certain patterns or rules), covering verbal and numerical reasoning. 24 Participants were given 10 minutes to complete the test. Besides this, the test battery also included the Mill Hill vocabulary test 25 and the Mini Mental State Examination 26 , however these tests were not included in the present study due to the observed ceiling effects.
Additional information on the vascular and cognitive study variables was mentioned previously. 16,23,27,28

Description of linear mixed effect models
To investigate the association between the cerebral small vessel disease (SVD) MRI score and trajectories of vascular risk and cognitive performance over 25 years, we employed the following equation for each dependent variable of interest (V): = 0 + 1 + 2 2 + 3 + 4 + 5 2 + 6 1 + 7 2 + 8 2 + 9 1 2 + 0 + 1 + Vij is the dependent variable of interest of the i th participant at the j th occasion, timeij is years since the baseline measurement between 1995-1999 for the i th participant at the j th occasion, time 2 ij is the orthogonalized polynomial quadratic time term for the i th participant at the j th occasion, SVDi is the total SVD score obtained during the Whitehall Imaging Sub-study of the i th participant, X1i is the covariate for scanner model (Prisma vs. Verio) for the i th participant, X2i is a vector of covariates (age at baseline, sex, education) for the i th participant, U0i is the random intercept, U1i is the random slope, and eij is the residual.
The dependent variables of interest focused on vascular risk and cognitive performance.
Vascular risk was defined using the Framingham Stroke Risk Score (FSRS) and mean arterial pressure (MAP).
Due to the skewed distribution of the residuals, FSRS was log-transformed.
Cognitive performance was covered for the following domains: letter fluency, semantic fluency, verbal reasoning, numerical reasoning, and global cognition. We implemented a continuous autoregressive moving-average correlation structure to consider repeated measures for each individual.