Spatial distribution and cognitive impact of cerebrovascular risk-related white matter hyperintensities

Abstract

Objectives: White matter hyperintensities (WMHs) are considered macroscale markers of cerebrovascular burden and are associated with increased risk of vascular cognitive impairment and dementia. However, the spatial location of WMHs has typically been considered in broad categories of periventricular versus deep white matter. The spatial distribution of WHMs associated with individual cerebrovascular risk factors (CVR), controlling for frequently comorbid risk factors, has not been systematically investigated at the population level in a healthy ageing cohort. Furthermore, there is an inconsistent relationship between total white matter hyperintensity load and cognition, which may be due to the confounding of several simultaneous risk factors in models based on smaller cohorts.

Methods: We examined trends in individual CVR factors on total WMH burden in 13,680 individuals (aged 45-80) using data from the UK Biobank. We estimated the spatial distribution of white matter hyperintensities associated with each risk factor and their contribution to explaining total WMH load using voxel-wise probit regression and univariate linear regression. Finally, we explored the impact of CVR-related WMHs on speed of processing using regression and mediation analysis.

Results: Contrary to the assumed dominance of hypertension as the biggest predictor of WMH burden, we show associations with a number of risk factors including diabetes, heavy smoking, APOE ε4/ε4 status and high waist-to-hip ratio of similar, or greater magnitude to hypertension. The spatial distribution of WMHs varied considerably with individual cerebrovascular risk factors. There were independent effects of visceral adiposity, as measured by waist-to-hip ratio, and carriage of the APOE ε4 allele in terms of the unique spatial distribution of CVR-related WMHs. Importantly, the relationship between total WMH load and speed of processing was mediated by waist-to-hip ratio suggesting cognitive consequences to WMHs associated with excessive visceral fat deposition.

Conclusion: Waist-to-hip ratio, diabetes, heavy smoking, hypercholesterolemia and homozygous APOE ε4 status are important risk factors, beyond hypertension, associated with WMH total burden and warrant careful control across ageing. The spatial distribution associated with different risk factors may provide important clues as to the pathogenesis and cognitive consequences of WMHs. High waist-to-hip ratio is a key risk factor associated with slowing in speed of processing. With global obesity levels rising, focused management of visceral adiposity may present a useful strategy for the mitigation of cognitive decline in ageing.

Keywords: APOE; Cardiovascular risk; Structural MRI; Visceral adiposity; White matter hyperintensities.

Fig. 1

Diagram demonstrating the flow of gradually refining participants starting from all UK Biobank participants with available T1 structural brain images. Most common neuropsychiatric/neurological conditions in decreasing number of participants: stroke (185), transient ischaemic attack (115), epilepsy (77), etc. (see Table A.1 for a full list). Missingness by risk factor: hypertension risk (2,075), hypercholesterolemia (211), diabetes (165), smoking in pack years (3,146), waist-to-hip ratio (413), and APOE-$\epsilon$ status (514). Some individuals have more than one of the variables missing; for characteristics of individuals excluded due to missingness, see Table A.2.

Fig. 2

Sex-specific trends in CVR score (left) and in log(WMH load) (right) across age. Solid lines represent the loess-smoothed curve with a span of 20% and the points are the observed data points. Males have higher CVR burden than females across all ages and log(WMH load) increases across age for both sexes but potentially at different speed for males and females. *uniform noise U(0, 1) added to the CVR scores to disperse the values in the y-axis direction (left plot)..

Fig. 3

Cerebrovascular risk factor specific trends in log(WMH load) across age. Solid lines represent the loess-smoothed curve with a span of 20% and the points are the observed data points. The presence of any of the risk factors suggests higher log(WMH load). Crossing fitted curves would suggest a potential risk factor by age interaction and parallel line its absence, respectively. WMH: white matter hyperintensity; WHR: waist-to-hip ratio; APOE: apolypoprotein-E..

Fig. 4

Square root transformed empirical WMH probability based on binary WMH masks of 13,680 UK Biobank individuals; axial slices $z=\{35,40,45,50\}$ shown (from left to right). Square root transformation leads to more dispersed values allowing for better visualisation. Voxels with three or fewer individuals having a WMH are plotted as transparent to show a standard anatomical MRI for reference.

Fig. 5

Significance maps (z-scores based on mean bias-reduced estimates) for model S.1, which includes age, sex (baseline men), age-sex interaction and head size and cerebrovascular risk (CVR) score as explanatory variables. Data on 13,680 UK Biobank individuals, and voxels with at least four individuals having a WMH explored (i.e. 0.03% WMH incidence); 5% FDR correction applied; axial slice $z=45$ shown.

Fig. 6

Significance maps (z-scores based on mean bias-reduced estimates) for (a) model S.2 (joint) and (b) models S.3.1 – S.3.6 (marginal). All models include the same confounding variables as models S.1 (age, sex (baseline men), age-sex interaction and head size). Data on 13,680 UK Biobank individuals, and voxels with at least four individuals having a WMH explored (i.e. 0.03% WMH incidence); 5% FDR correction applied; axial slice $z=45$ shown.

Fig. A.1

Significance maps (z-scores based on mean bias-reduced estimates) for APOE $\epsilon 4/\epsilon 4$ effect compared to no $\epsilon 4$ alleles (marginal model). Data on 13,680 UK Biobank individuals, and voxels with at least four individuals having a WMH explored (i.e. 0.03% WMH incidence); 5% FDR correction applied; Slices $x=36,y=25,z=40$ shown.

Fig. A.2

Significance maps (z-scores based on mean bias-reduced estimates) across five axial slices $z=\{35,40,45,50,55\}$ in a model including age, sex, age-sex interaction, head size, systolic blood pressure and waist-to-hip ratio; data on 13,680 UK Biobank individuals and voxels with at least four individuals having a WMH explored; 5% FDR correction applied. From top to bottom each row shows z-scores for (a) Systolic BP, (b) WHR and their ‘unique’ effect, respectively, (c) ‘significant systolic BP/not significant WHR’ and (d) ‘not significant systolic BP/significant WHR’. BP: blood pressure; WHR: waist-to-hip ratio..

Fig. A.3

Mediation analysis diagram. The mediators explored are the six cerebrovascular risk factors. All models are controlled for age, sex, age-sex interaction, and head size. The main interest is in whether the indirect effect is significant, i.e. whether the effect of the predictor X on an outcome Y operates through a mediator variable M (fully or partially).